diff options
Diffstat (limited to 'src/spdk/dpdk/drivers/bus/dpaa')
31 files changed, 12457 insertions, 0 deletions
diff --git a/src/spdk/dpdk/drivers/bus/dpaa/Makefile b/src/spdk/dpdk/drivers/bus/dpaa/Makefile new file mode 100644 index 000000000..ba40b2116 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/Makefile @@ -0,0 +1,48 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright 2016 NXP + +include $(RTE_SDK)/mk/rte.vars.mk +RTE_BUS_DPAA=$(RTE_SDK)/drivers/bus/dpaa + +# +# library name +# +LIB = librte_bus_dpaa.a + +CFLAGS := -I$(SRCDIR) $(CFLAGS) +CFLAGS += -O3 $(WERROR_FLAGS) +CFLAGS += -Wno-pointer-arith +CFLAGS += -Wno-cast-qual +CFLAGS += -I$(RTE_BUS_DPAA)/ +CFLAGS += -I$(RTE_BUS_DPAA)/include +CFLAGS += -I$(RTE_BUS_DPAA)/base/qbman +CFLAGS += -I$(RTE_SDK)/drivers/common/dpaax +CFLAGS += -I$(RTE_SDK)/lib/librte_eal/include + +# versioning export map +EXPORT_MAP := rte_bus_dpaa_version.map + +# all source are stored in SRCS-y +# +SRCS-$(CONFIG_RTE_LIBRTE_DPAA_BUS) += \ + dpaa_bus.c + +SRCS-$(CONFIG_RTE_LIBRTE_DPAA_BUS) += \ + base/fman/fman.c \ + base/fman/fman_hw.c \ + base/fman/netcfg_layer.c \ + base/qbman/process.c \ + base/qbman/bman.c \ + base/qbman/bman_driver.c \ + base/qbman/qman.c \ + base/qbman/qman_driver.c \ + base/qbman/dpaa_alloc.c \ + base/qbman/dpaa_sys.c + +# Link Pthread +LDLIBS += -lpthread +LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring +LDLIBS += -lrte_ethdev +LDLIBS += -lrte_common_dpaax + +include $(RTE_SDK)/mk/rte.lib.mk diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/fman/fman.c b/src/spdk/dpdk/drivers/bus/dpaa/base/fman/fman.c new file mode 100644 index 000000000..6d77a7e39 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/fman/fman.c @@ -0,0 +1,583 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2010-2016 Freescale Semiconductor Inc. + * Copyright 2017-2019 NXP + * + */ + +#include <sys/types.h> +#include <sys/ioctl.h> +#include <ifaddrs.h> + +/* This header declares the driver interface we implement */ +#include <fman.h> +#include <dpaa_of.h> +#include <rte_malloc.h> +#include <rte_dpaa_logs.h> +#include <rte_string_fns.h> + +#define QMI_PORT_REGS_OFFSET 0x400 + +/* CCSR map address to access ccsr based register */ +void *fman_ccsr_map; +/* fman version info */ +u16 fman_ip_rev; +static int get_once; +u32 fman_dealloc_bufs_mask_hi; +u32 fman_dealloc_bufs_mask_lo; + +int fman_ccsr_map_fd = -1; +static COMPAT_LIST_HEAD(__ifs); + +/* This is the (const) global variable that callers have read-only access to. + * Internally, we have read-write access directly to __ifs. + */ +const struct list_head *fman_if_list = &__ifs; + +static void +if_destructor(struct __fman_if *__if) +{ + struct fman_if_bpool *bp, *tmpbp; + + if (!__if) + return; + + if (__if->__if.mac_type == fman_offline) + goto cleanup; + + list_for_each_entry_safe(bp, tmpbp, &__if->__if.bpool_list, node) { + list_del(&bp->node); + free(bp); + } +cleanup: + free(__if); +} + +static int +fman_get_ip_rev(const struct device_node *fman_node) +{ + const uint32_t *fman_addr; + uint64_t phys_addr; + uint64_t regs_size; + uint32_t ip_rev_1; + int _errno; + + fman_addr = of_get_address(fman_node, 0, ®s_size, NULL); + if (!fman_addr) { + pr_err("of_get_address cannot return fman address\n"); + return -EINVAL; + } + phys_addr = of_translate_address(fman_node, fman_addr); + if (!phys_addr) { + pr_err("of_translate_address failed\n"); + return -EINVAL; + } + fman_ccsr_map = mmap(NULL, regs_size, PROT_READ | PROT_WRITE, + MAP_SHARED, fman_ccsr_map_fd, phys_addr); + if (fman_ccsr_map == MAP_FAILED) { + pr_err("Can not map FMan ccsr base"); + return -EINVAL; + } + + ip_rev_1 = in_be32(fman_ccsr_map + FMAN_IP_REV_1); + fman_ip_rev = (ip_rev_1 & FMAN_IP_REV_1_MAJOR_MASK) >> + FMAN_IP_REV_1_MAJOR_SHIFT; + + _errno = munmap(fman_ccsr_map, regs_size); + if (_errno) + pr_err("munmap() of FMan ccsr failed"); + + return 0; +} + +static int +fman_get_mac_index(uint64_t regs_addr_host, uint8_t *mac_idx) +{ + int ret = 0; + + /* + * MAC1 : E_0000h + * MAC2 : E_2000h + * MAC3 : E_4000h + * MAC4 : E_6000h + * MAC5 : E_8000h + * MAC6 : E_A000h + * MAC7 : E_C000h + * MAC8 : E_E000h + * MAC9 : F_0000h + * MAC10: F_2000h + */ + switch (regs_addr_host) { + case 0xE0000: + *mac_idx = 1; + break; + case 0xE2000: + *mac_idx = 2; + break; + case 0xE4000: + *mac_idx = 3; + break; + case 0xE6000: + *mac_idx = 4; + break; + case 0xE8000: + *mac_idx = 5; + break; + case 0xEA000: + *mac_idx = 6; + break; + case 0xEC000: + *mac_idx = 7; + break; + case 0xEE000: + *mac_idx = 8; + break; + case 0xF0000: + *mac_idx = 9; + break; + case 0xF2000: + *mac_idx = 10; + break; + default: + ret = -EINVAL; + } + + return ret; +} + +static int +fman_if_init(const struct device_node *dpa_node) +{ + const char *rprop, *mprop; + uint64_t phys_addr; + struct __fman_if *__if; + struct fman_if_bpool *bpool; + + const phandle *mac_phandle, *ports_phandle, *pools_phandle; + const phandle *tx_channel_id = NULL, *mac_addr, *cell_idx; + const phandle *rx_phandle, *tx_phandle; + uint64_t tx_phandle_host[4] = {0}; + uint64_t rx_phandle_host[4] = {0}; + uint64_t regs_addr_host = 0; + uint64_t cell_idx_host = 0; + + const struct device_node *mac_node = NULL, *tx_node; + const struct device_node *pool_node, *fman_node, *rx_node; + const uint32_t *regs_addr = NULL; + const char *mname, *fname; + const char *dname = dpa_node->full_name; + size_t lenp; + int _errno; + const char *char_prop; + uint32_t na; + + if (of_device_is_available(dpa_node) == false) + return 0; + + rprop = "fsl,qman-frame-queues-rx"; + mprop = "fsl,fman-mac"; + + /* Allocate an object for this network interface */ + __if = rte_malloc(NULL, sizeof(*__if), RTE_CACHE_LINE_SIZE); + if (!__if) { + FMAN_ERR(-ENOMEM, "malloc(%zu)\n", sizeof(*__if)); + goto err; + } + memset(__if, 0, sizeof(*__if)); + INIT_LIST_HEAD(&__if->__if.bpool_list); + strlcpy(__if->node_path, dpa_node->full_name, PATH_MAX - 1); + __if->node_path[PATH_MAX - 1] = '\0'; + + /* Obtain the MAC node used by this interface except macless */ + mac_phandle = of_get_property(dpa_node, mprop, &lenp); + if (!mac_phandle) { + FMAN_ERR(-EINVAL, "%s: no %s\n", dname, mprop); + goto err; + } + assert(lenp == sizeof(phandle)); + mac_node = of_find_node_by_phandle(*mac_phandle); + if (!mac_node) { + FMAN_ERR(-ENXIO, "%s: bad 'fsl,fman-mac\n", dname); + goto err; + } + mname = mac_node->full_name; + + /* Map the CCSR regs for the MAC node */ + regs_addr = of_get_address(mac_node, 0, &__if->regs_size, NULL); + if (!regs_addr) { + FMAN_ERR(-EINVAL, "of_get_address(%s)\n", mname); + goto err; + } + phys_addr = of_translate_address(mac_node, regs_addr); + if (!phys_addr) { + FMAN_ERR(-EINVAL, "of_translate_address(%s, %p)\n", + mname, regs_addr); + goto err; + } + __if->ccsr_map = mmap(NULL, __if->regs_size, + PROT_READ | PROT_WRITE, MAP_SHARED, + fman_ccsr_map_fd, phys_addr); + if (__if->ccsr_map == MAP_FAILED) { + FMAN_ERR(-errno, "mmap(0x%"PRIx64")\n", phys_addr); + goto err; + } + na = of_n_addr_cells(mac_node); + /* Get rid of endianness (issues). Convert to host byte order */ + regs_addr_host = of_read_number(regs_addr, na); + + + /* Get the index of the Fman this i/f belongs to */ + fman_node = of_get_parent(mac_node); + na = of_n_addr_cells(mac_node); + if (!fman_node) { + FMAN_ERR(-ENXIO, "of_get_parent(%s)\n", mname); + goto err; + } + fname = fman_node->full_name; + cell_idx = of_get_property(fman_node, "cell-index", &lenp); + if (!cell_idx) { + FMAN_ERR(-ENXIO, "%s: no cell-index)\n", fname); + goto err; + } + assert(lenp == sizeof(*cell_idx)); + cell_idx_host = of_read_number(cell_idx, lenp / sizeof(phandle)); + __if->__if.fman_idx = cell_idx_host; + if (!get_once) { + _errno = fman_get_ip_rev(fman_node); + if (_errno) { + FMAN_ERR(-ENXIO, "%s: ip_rev is not available\n", + fname); + goto err; + } + } + + if (fman_ip_rev >= FMAN_V3) { + /* + * Set A2V, OVOM, EBD bits in contextA to allow external + * buffer deallocation by fman. + */ + fman_dealloc_bufs_mask_hi = FMAN_V3_CONTEXTA_EN_A2V | + FMAN_V3_CONTEXTA_EN_OVOM; + fman_dealloc_bufs_mask_lo = FMAN_V3_CONTEXTA_EN_EBD; + } else { + fman_dealloc_bufs_mask_hi = 0; + fman_dealloc_bufs_mask_lo = 0; + } + /* Is the MAC node 1G, 10G? */ + __if->__if.is_memac = 0; + + if (of_device_is_compatible(mac_node, "fsl,fman-1g-mac")) + __if->__if.mac_type = fman_mac_1g; + else if (of_device_is_compatible(mac_node, "fsl,fman-10g-mac")) + __if->__if.mac_type = fman_mac_10g; + else if (of_device_is_compatible(mac_node, "fsl,fman-memac")) { + __if->__if.is_memac = 1; + char_prop = of_get_property(mac_node, "phy-connection-type", + NULL); + if (!char_prop) { + printf("memac: unknown MII type assuming 1G\n"); + /* Right now forcing memac to 1g in case of error*/ + __if->__if.mac_type = fman_mac_1g; + } else { + if (strstr(char_prop, "sgmii")) + __if->__if.mac_type = fman_mac_1g; + else if (strstr(char_prop, "rgmii")) { + __if->__if.mac_type = fman_mac_1g; + __if->__if.is_rgmii = 1; + } else if (strstr(char_prop, "xgmii")) + __if->__if.mac_type = fman_mac_10g; + } + } else { + FMAN_ERR(-EINVAL, "%s: unknown MAC type\n", mname); + goto err; + } + + /* + * For MAC ports, we cannot rely on cell-index. In + * T2080, two of the 10G ports on single FMAN have same + * duplicate cell-indexes as the other two 10G ports on + * same FMAN. Hence, we now rely upon addresses of the + * ports from device tree to deduce the index. + */ + + _errno = fman_get_mac_index(regs_addr_host, &__if->__if.mac_idx); + if (_errno) { + FMAN_ERR(-EINVAL, "Invalid register address: %" PRIx64, + regs_addr_host); + goto err; + } + + /* Extract the MAC address for private and shared interfaces */ + mac_addr = of_get_property(mac_node, "local-mac-address", + &lenp); + if (!mac_addr) { + FMAN_ERR(-EINVAL, "%s: no local-mac-address\n", + mname); + goto err; + } + memcpy(&__if->__if.mac_addr, mac_addr, ETHER_ADDR_LEN); + + /* Extract the Tx port (it's the second of the two port handles) + * and get its channel ID + */ + ports_phandle = of_get_property(mac_node, "fsl,port-handles", + &lenp); + if (!ports_phandle) + ports_phandle = of_get_property(mac_node, "fsl,fman-ports", + &lenp); + if (!ports_phandle) { + FMAN_ERR(-EINVAL, "%s: no fsl,port-handles\n", + mname); + goto err; + } + assert(lenp == (2 * sizeof(phandle))); + tx_node = of_find_node_by_phandle(ports_phandle[1]); + if (!tx_node) { + FMAN_ERR(-ENXIO, "%s: bad fsl,port-handle[1]\n", mname); + goto err; + } + /* Extract the channel ID (from tx-port-handle) */ + tx_channel_id = of_get_property(tx_node, "fsl,qman-channel-id", + &lenp); + if (!tx_channel_id) { + FMAN_ERR(-EINVAL, "%s: no fsl-qman-channel-id\n", + tx_node->full_name); + goto err; + } + + rx_node = of_find_node_by_phandle(ports_phandle[0]); + if (!rx_node) { + FMAN_ERR(-ENXIO, "%s: bad fsl,port-handle[0]\n", mname); + goto err; + } + regs_addr = of_get_address(rx_node, 0, &__if->regs_size, NULL); + if (!regs_addr) { + FMAN_ERR(-EINVAL, "of_get_address(%s)\n", mname); + goto err; + } + phys_addr = of_translate_address(rx_node, regs_addr); + if (!phys_addr) { + FMAN_ERR(-EINVAL, "of_translate_address(%s, %p)\n", + mname, regs_addr); + goto err; + } + __if->bmi_map = mmap(NULL, __if->regs_size, + PROT_READ | PROT_WRITE, MAP_SHARED, + fman_ccsr_map_fd, phys_addr); + if (__if->bmi_map == MAP_FAILED) { + FMAN_ERR(-errno, "mmap(0x%"PRIx64")\n", phys_addr); + goto err; + } + + /* No channel ID for MAC-less */ + assert(lenp == sizeof(*tx_channel_id)); + na = of_n_addr_cells(mac_node); + __if->__if.tx_channel_id = of_read_number(tx_channel_id, na); + + /* Extract the Rx FQIDs. (Note, the device representation is silly, + * there are "counts" that must always be 1.) + */ + rx_phandle = of_get_property(dpa_node, rprop, &lenp); + if (!rx_phandle) { + FMAN_ERR(-EINVAL, "%s: no fsl,qman-frame-queues-rx\n", dname); + goto err; + } + + assert(lenp == (4 * sizeof(phandle))); + + na = of_n_addr_cells(mac_node); + /* Get rid of endianness (issues). Convert to host byte order */ + rx_phandle_host[0] = of_read_number(&rx_phandle[0], na); + rx_phandle_host[1] = of_read_number(&rx_phandle[1], na); + rx_phandle_host[2] = of_read_number(&rx_phandle[2], na); + rx_phandle_host[3] = of_read_number(&rx_phandle[3], na); + + assert((rx_phandle_host[1] == 1) && (rx_phandle_host[3] == 1)); + __if->__if.fqid_rx_err = rx_phandle_host[0]; + __if->__if.fqid_rx_def = rx_phandle_host[2]; + + /* Extract the Tx FQIDs */ + tx_phandle = of_get_property(dpa_node, + "fsl,qman-frame-queues-tx", &lenp); + if (!tx_phandle) { + FMAN_ERR(-EINVAL, "%s: no fsl,qman-frame-queues-tx\n", dname); + goto err; + } + + assert(lenp == (4 * sizeof(phandle))); + /*TODO: Fix for other cases also */ + na = of_n_addr_cells(mac_node); + /* Get rid of endianness (issues). Convert to host byte order */ + tx_phandle_host[0] = of_read_number(&tx_phandle[0], na); + tx_phandle_host[1] = of_read_number(&tx_phandle[1], na); + tx_phandle_host[2] = of_read_number(&tx_phandle[2], na); + tx_phandle_host[3] = of_read_number(&tx_phandle[3], na); + assert((tx_phandle_host[1] == 1) && (tx_phandle_host[3] == 1)); + __if->__if.fqid_tx_err = tx_phandle_host[0]; + __if->__if.fqid_tx_confirm = tx_phandle_host[2]; + + /* Obtain the buffer pool nodes used by this interface */ + pools_phandle = of_get_property(dpa_node, "fsl,bman-buffer-pools", + &lenp); + if (!pools_phandle) { + FMAN_ERR(-EINVAL, "%s: no fsl,bman-buffer-pools\n", dname); + goto err; + } + /* For each pool, parse the corresponding node and add a pool object + * to the interface's "bpool_list" + */ + assert(lenp && !(lenp % sizeof(phandle))); + while (lenp) { + size_t proplen; + const phandle *prop; + uint64_t bpid_host = 0; + uint64_t bpool_host[6] = {0}; + const char *pname; + /* Allocate an object for the pool */ + bpool = rte_malloc(NULL, sizeof(*bpool), RTE_CACHE_LINE_SIZE); + if (!bpool) { + FMAN_ERR(-ENOMEM, "malloc(%zu)\n", sizeof(*bpool)); + goto err; + } + /* Find the pool node */ + pool_node = of_find_node_by_phandle(*pools_phandle); + if (!pool_node) { + FMAN_ERR(-ENXIO, "%s: bad fsl,bman-buffer-pools\n", + dname); + rte_free(bpool); + goto err; + } + pname = pool_node->full_name; + /* Extract the BPID property */ + prop = of_get_property(pool_node, "fsl,bpid", &proplen); + if (!prop) { + FMAN_ERR(-EINVAL, "%s: no fsl,bpid\n", pname); + rte_free(bpool); + goto err; + } + assert(proplen == sizeof(*prop)); + na = of_n_addr_cells(mac_node); + /* Get rid of endianness (issues). + * Convert to host byte-order + */ + bpid_host = of_read_number(prop, na); + bpool->bpid = bpid_host; + /* Extract the cfg property (count/size/addr). "fsl,bpool-cfg" + * indicates for the Bman driver to seed the pool. + * "fsl,bpool-ethernet-cfg" is used by the network driver. The + * two are mutually exclusive, so check for either of them. + */ + prop = of_get_property(pool_node, "fsl,bpool-cfg", + &proplen); + if (!prop) + prop = of_get_property(pool_node, + "fsl,bpool-ethernet-cfg", + &proplen); + if (!prop) { + /* It's OK for there to be no bpool-cfg */ + bpool->count = bpool->size = bpool->addr = 0; + } else { + assert(proplen == (6 * sizeof(*prop))); + na = of_n_addr_cells(mac_node); + /* Get rid of endianness (issues). + * Convert to host byte order + */ + bpool_host[0] = of_read_number(&prop[0], na); + bpool_host[1] = of_read_number(&prop[1], na); + bpool_host[2] = of_read_number(&prop[2], na); + bpool_host[3] = of_read_number(&prop[3], na); + bpool_host[4] = of_read_number(&prop[4], na); + bpool_host[5] = of_read_number(&prop[5], na); + + bpool->count = ((uint64_t)bpool_host[0] << 32) | + bpool_host[1]; + bpool->size = ((uint64_t)bpool_host[2] << 32) | + bpool_host[3]; + bpool->addr = ((uint64_t)bpool_host[4] << 32) | + bpool_host[5]; + } + /* Parsing of the pool is complete, add it to the interface + * list. + */ + list_add_tail(&bpool->node, &__if->__if.bpool_list); + lenp -= sizeof(phandle); + pools_phandle++; + } + + /* Parsing of the network interface is complete, add it to the list */ + DPAA_BUS_LOG(DEBUG, "Found %s, Tx Channel = %x, FMAN = %x," + "Port ID = %x", + dname, __if->__if.tx_channel_id, __if->__if.fman_idx, + __if->__if.mac_idx); + + list_add_tail(&__if->__if.node, &__ifs); + return 0; +err: + if_destructor(__if); + return _errno; +} + +int +fman_init(void) +{ + const struct device_node *dpa_node; + int _errno; + + /* If multiple dependencies try to initialise the Fman driver, don't + * panic. + */ + if (fman_ccsr_map_fd != -1) + return 0; + + fman_ccsr_map_fd = open(FMAN_DEVICE_PATH, O_RDWR); + if (unlikely(fman_ccsr_map_fd < 0)) { + DPAA_BUS_LOG(ERR, "Unable to open (/dev/mem)"); + return fman_ccsr_map_fd; + } + + for_each_compatible_node(dpa_node, NULL, "fsl,dpa-ethernet-init") { + _errno = fman_if_init(dpa_node); + if (_errno) { + FMAN_ERR(_errno, "if_init(%s)\n", dpa_node->full_name); + goto err; + } + } + + return 0; +err: + fman_finish(); + return _errno; +} + +void +fman_finish(void) +{ + struct __fman_if *__if, *tmpif; + + assert(fman_ccsr_map_fd != -1); + + list_for_each_entry_safe(__if, tmpif, &__ifs, __if.node) { + int _errno; + + /* disable Rx and Tx */ + if ((__if->__if.mac_type == fman_mac_1g) && + (!__if->__if.is_memac)) + out_be32(__if->ccsr_map + 0x100, + in_be32(__if->ccsr_map + 0x100) & ~(u32)0x5); + else + out_be32(__if->ccsr_map + 8, + in_be32(__if->ccsr_map + 8) & ~(u32)3); + /* release the mapping */ + _errno = munmap(__if->ccsr_map, __if->regs_size); + if (unlikely(_errno < 0)) + fprintf(stderr, "%s:%d:%s(): munmap() = %d (%s)\n", + __FILE__, __LINE__, __func__, + -errno, strerror(errno)); + printf("Tearing down %s\n", __if->node_path); + list_del(&__if->__if.node); + rte_free(__if); + } + + close(fman_ccsr_map_fd); + fman_ccsr_map_fd = -1; +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/fman/fman_hw.c b/src/spdk/dpdk/drivers/bus/dpaa/base/fman/fman_hw.c new file mode 100644 index 000000000..9ab8e835d --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/fman/fman_hw.c @@ -0,0 +1,611 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * + * Copyright 2017 NXP + * + */ + +#include <sys/types.h> +#include <sys/ioctl.h> +#include <ifaddrs.h> +#include <fman.h> +/* This header declares things about Fman hardware itself (the format of status + * words and an inline implementation of CRC64). We include it only in order to + * instantiate the one global variable it depends on. + */ +#include <fsl_fman.h> +#include <fsl_fman_crc64.h> +#include <fsl_bman.h> + +#define FMAN_SP_SG_DISABLE 0x80000000 +#define FMAN_SP_EXT_BUF_MARG_START_SHIFT 16 + +/* Instantiate the global variable that the inline CRC64 implementation (in + * <fsl_fman.h>) depends on. + */ +DECLARE_FMAN_CRC64_TABLE(); + +#define ETH_ADDR_TO_UINT64(eth_addr) \ + (uint64_t)(((uint64_t)(eth_addr)[0] << 40) | \ + ((uint64_t)(eth_addr)[1] << 32) | \ + ((uint64_t)(eth_addr)[2] << 24) | \ + ((uint64_t)(eth_addr)[3] << 16) | \ + ((uint64_t)(eth_addr)[4] << 8) | \ + ((uint64_t)(eth_addr)[5])) + +void +fman_if_set_mcast_filter_table(struct fman_if *p) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + void *hashtable_ctrl; + uint32_t i; + + hashtable_ctrl = &((struct memac_regs *)__if->ccsr_map)->hashtable_ctrl; + for (i = 0; i < 64; i++) + out_be32(hashtable_ctrl, i|HASH_CTRL_MCAST_EN); +} + +void +fman_if_reset_mcast_filter_table(struct fman_if *p) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + void *hashtable_ctrl; + uint32_t i; + + hashtable_ctrl = &((struct memac_regs *)__if->ccsr_map)->hashtable_ctrl; + for (i = 0; i < 64; i++) + out_be32(hashtable_ctrl, i & ~HASH_CTRL_MCAST_EN); +} + +static +uint32_t get_mac_hash_code(uint64_t eth_addr) +{ + uint64_t mask1, mask2; + uint32_t xorVal = 0; + uint8_t i, j; + + for (i = 0; i < 6; i++) { + mask1 = eth_addr & (uint64_t)0x01; + eth_addr >>= 1; + + for (j = 0; j < 7; j++) { + mask2 = eth_addr & (uint64_t)0x01; + mask1 ^= mask2; + eth_addr >>= 1; + } + + xorVal |= (mask1 << (5 - i)); + } + + return xorVal; +} + +int +fman_if_add_hash_mac_addr(struct fman_if *p, uint8_t *eth) +{ + uint64_t eth_addr; + void *hashtable_ctrl; + uint32_t hash; + + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + + eth_addr = ETH_ADDR_TO_UINT64(eth); + + if (!(eth_addr & GROUP_ADDRESS)) + return -1; + + hash = get_mac_hash_code(eth_addr) & HASH_CTRL_ADDR_MASK; + hash = hash | HASH_CTRL_MCAST_EN; + + hashtable_ctrl = &((struct memac_regs *)__if->ccsr_map)->hashtable_ctrl; + out_be32(hashtable_ctrl, hash); + + return 0; +} + +int +fman_if_get_primary_mac_addr(struct fman_if *p, uint8_t *eth) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + void *mac_reg = + &((struct memac_regs *)__if->ccsr_map)->mac_addr0.mac_addr_l; + u32 val = in_be32(mac_reg); + + eth[0] = (val & 0x000000ff) >> 0; + eth[1] = (val & 0x0000ff00) >> 8; + eth[2] = (val & 0x00ff0000) >> 16; + eth[3] = (val & 0xff000000) >> 24; + + mac_reg = &((struct memac_regs *)__if->ccsr_map)->mac_addr0.mac_addr_u; + val = in_be32(mac_reg); + + eth[4] = (val & 0x000000ff) >> 0; + eth[5] = (val & 0x0000ff00) >> 8; + + return 0; +} + +void +fman_if_clear_mac_addr(struct fman_if *p, uint8_t addr_num) +{ + struct __fman_if *m = container_of(p, struct __fman_if, __if); + void *reg; + + if (addr_num) { + reg = &((struct memac_regs *)m->ccsr_map)-> + mac_addr[addr_num-1].mac_addr_l; + out_be32(reg, 0x0); + reg = &((struct memac_regs *)m->ccsr_map)-> + mac_addr[addr_num-1].mac_addr_u; + out_be32(reg, 0x0); + } else { + reg = &((struct memac_regs *)m->ccsr_map)->mac_addr0.mac_addr_l; + out_be32(reg, 0x0); + reg = &((struct memac_regs *)m->ccsr_map)->mac_addr0.mac_addr_u; + out_be32(reg, 0x0); + } +} + +int +fman_if_add_mac_addr(struct fman_if *p, uint8_t *eth, uint8_t addr_num) +{ + struct __fman_if *m = container_of(p, struct __fman_if, __if); + + void *reg; + u32 val; + + memcpy(&m->__if.mac_addr, eth, ETHER_ADDR_LEN); + + if (addr_num) + reg = &((struct memac_regs *)m->ccsr_map)-> + mac_addr[addr_num-1].mac_addr_l; + else + reg = &((struct memac_regs *)m->ccsr_map)->mac_addr0.mac_addr_l; + + val = (m->__if.mac_addr.addr_bytes[0] | + (m->__if.mac_addr.addr_bytes[1] << 8) | + (m->__if.mac_addr.addr_bytes[2] << 16) | + (m->__if.mac_addr.addr_bytes[3] << 24)); + out_be32(reg, val); + + if (addr_num) + reg = &((struct memac_regs *)m->ccsr_map)-> + mac_addr[addr_num-1].mac_addr_u; + else + reg = &((struct memac_regs *)m->ccsr_map)->mac_addr0.mac_addr_u; + + val = ((m->__if.mac_addr.addr_bytes[4] << 0) | + (m->__if.mac_addr.addr_bytes[5] << 8)); + out_be32(reg, val); + + return 0; +} + +void +fman_if_set_rx_ignore_pause_frames(struct fman_if *p, bool enable) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + u32 value = 0; + void *cmdcfg; + + assert(fman_ccsr_map_fd != -1); + + /* Set Rx Ignore Pause Frames */ + cmdcfg = &((struct memac_regs *)__if->ccsr_map)->command_config; + if (enable) + value = in_be32(cmdcfg) | CMD_CFG_PAUSE_IGNORE; + else + value = in_be32(cmdcfg) & ~CMD_CFG_PAUSE_IGNORE; + + out_be32(cmdcfg, value); +} + +void +fman_if_conf_max_frame_len(struct fman_if *p, unsigned int max_frame_len) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + unsigned int *maxfrm; + + assert(fman_ccsr_map_fd != -1); + + /* Set Max frame length */ + maxfrm = &((struct memac_regs *)__if->ccsr_map)->maxfrm; + out_be32(maxfrm, (MAXFRM_RX_MASK & max_frame_len)); +} + +void +fman_if_stats_get(struct fman_if *p, struct rte_eth_stats *stats) +{ + struct __fman_if *m = container_of(p, struct __fman_if, __if); + struct memac_regs *regs = m->ccsr_map; + + /* read recved packet count */ + stats->ipackets = ((u64)in_be32(®s->rfrm_u)) << 32 | + in_be32(®s->rfrm_l); + stats->ibytes = ((u64)in_be32(®s->roct_u)) << 32 | + in_be32(®s->roct_l); + stats->ierrors = ((u64)in_be32(®s->rerr_u)) << 32 | + in_be32(®s->rerr_l); + + /* read xmited packet count */ + stats->opackets = ((u64)in_be32(®s->tfrm_u)) << 32 | + in_be32(®s->tfrm_l); + stats->obytes = ((u64)in_be32(®s->toct_u)) << 32 | + in_be32(®s->toct_l); + stats->oerrors = ((u64)in_be32(®s->terr_u)) << 32 | + in_be32(®s->terr_l); +} + +void +fman_if_stats_get_all(struct fman_if *p, uint64_t *value, int n) +{ + struct __fman_if *m = container_of(p, struct __fman_if, __if); + struct memac_regs *regs = m->ccsr_map; + int i; + uint64_t base_offset = offsetof(struct memac_regs, reoct_l); + + for (i = 0; i < n; i++) + value[i] = ((u64)in_be32((char *)regs + + base_offset + 8 * i + 4)) << 32 | + ((u64)in_be32((char *)regs + + base_offset + 8 * i)); +} + +void +fman_if_stats_reset(struct fman_if *p) +{ + struct __fman_if *m = container_of(p, struct __fman_if, __if); + struct memac_regs *regs = m->ccsr_map; + uint32_t tmp; + + tmp = in_be32(®s->statn_config); + + tmp |= STATS_CFG_CLR; + + out_be32(®s->statn_config, tmp); + + while (in_be32(®s->statn_config) & STATS_CFG_CLR) + ; +} + +void +fman_if_promiscuous_enable(struct fman_if *p) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + void *cmdcfg; + + assert(fman_ccsr_map_fd != -1); + + /* Enable Rx promiscuous mode */ + cmdcfg = &((struct memac_regs *)__if->ccsr_map)->command_config; + out_be32(cmdcfg, in_be32(cmdcfg) | CMD_CFG_PROMIS_EN); +} + +void +fman_if_promiscuous_disable(struct fman_if *p) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + void *cmdcfg; + + assert(fman_ccsr_map_fd != -1); + + /* Disable Rx promiscuous mode */ + cmdcfg = &((struct memac_regs *)__if->ccsr_map)->command_config; + out_be32(cmdcfg, in_be32(cmdcfg) & (~CMD_CFG_PROMIS_EN)); +} + +void +fman_if_enable_rx(struct fman_if *p) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + /* enable Rx and Tx */ + out_be32(__if->ccsr_map + 8, in_be32(__if->ccsr_map + 8) | 3); +} + +void +fman_if_disable_rx(struct fman_if *p) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + /* only disable Rx, not Tx */ + out_be32(__if->ccsr_map + 8, in_be32(__if->ccsr_map + 8) & ~(u32)2); +} + +void +fman_if_loopback_enable(struct fman_if *p) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + /* Enable loopback mode */ + if ((__if->__if.is_memac) && (__if->__if.is_rgmii)) { + unsigned int *ifmode = + &((struct memac_regs *)__if->ccsr_map)->if_mode; + out_be32(ifmode, in_be32(ifmode) | IF_MODE_RLP); + } else{ + unsigned int *cmdcfg = + &((struct memac_regs *)__if->ccsr_map)->command_config; + out_be32(cmdcfg, in_be32(cmdcfg) | CMD_CFG_LOOPBACK_EN); + } +} + +void +fman_if_loopback_disable(struct fman_if *p) +{ + struct __fman_if *__if = container_of(p, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + /* Disable loopback mode */ + if ((__if->__if.is_memac) && (__if->__if.is_rgmii)) { + unsigned int *ifmode = + &((struct memac_regs *)__if->ccsr_map)->if_mode; + out_be32(ifmode, in_be32(ifmode) & ~IF_MODE_RLP); + } else { + unsigned int *cmdcfg = + &((struct memac_regs *)__if->ccsr_map)->command_config; + out_be32(cmdcfg, in_be32(cmdcfg) & ~CMD_CFG_LOOPBACK_EN); + } +} + +void +fman_if_set_bp(struct fman_if *fm_if, unsigned num __always_unused, + int bpid, size_t bufsize) +{ + u32 fmbm_ebmpi; + u32 ebmpi_val_ace = 0xc0000000; + u32 ebmpi_mask = 0xffc00000; + + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + fmbm_ebmpi = + in_be32(&((struct rx_bmi_regs *)__if->bmi_map)->fmbm_ebmpi[0]); + fmbm_ebmpi = ebmpi_val_ace | (fmbm_ebmpi & ebmpi_mask) | (bpid << 16) | + (bufsize); + + out_be32(&((struct rx_bmi_regs *)__if->bmi_map)->fmbm_ebmpi[0], + fmbm_ebmpi); +} + +int +fman_if_get_fc_threshold(struct fman_if *fm_if) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + unsigned int *fmbm_mpd; + + assert(fman_ccsr_map_fd != -1); + + fmbm_mpd = &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_mpd; + return in_be32(fmbm_mpd); +} + +int +fman_if_set_fc_threshold(struct fman_if *fm_if, u32 high_water, + u32 low_water, u32 bpid) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + unsigned int *fmbm_mpd; + + assert(fman_ccsr_map_fd != -1); + + fmbm_mpd = &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_mpd; + out_be32(fmbm_mpd, FMAN_ENABLE_BPOOL_DEPLETION); + return bm_pool_set_hw_threshold(bpid, low_water, high_water); + +} + +int +fman_if_get_fc_quanta(struct fman_if *fm_if) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + return in_be32(&((struct memac_regs *)__if->ccsr_map)->pause_quanta[0]); +} + +int +fman_if_set_fc_quanta(struct fman_if *fm_if, u16 pause_quanta) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + out_be32(&((struct memac_regs *)__if->ccsr_map)->pause_quanta[0], + pause_quanta); + return 0; +} + +int +fman_if_get_fdoff(struct fman_if *fm_if) +{ + u32 fmbm_rebm; + int fdoff; + + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + fmbm_rebm = in_be32(&((struct rx_bmi_regs *)__if->bmi_map)->fmbm_rebm); + + fdoff = (fmbm_rebm >> FMAN_SP_EXT_BUF_MARG_START_SHIFT) & 0x1ff; + + return fdoff; +} + +void +fman_if_set_err_fqid(struct fman_if *fm_if, uint32_t err_fqid) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + unsigned int *fmbm_refqid = + &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_refqid; + out_be32(fmbm_refqid, err_fqid); +} + +int +fman_if_get_ic_params(struct fman_if *fm_if, struct fman_if_ic_params *icp) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + int val = 0; + int iceof_mask = 0x001f0000; + int icsz_mask = 0x0000001f; + int iciof_mask = 0x00000f00; + + assert(fman_ccsr_map_fd != -1); + + unsigned int *fmbm_ricp = + &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_ricp; + val = in_be32(fmbm_ricp); + + icp->iceof = (val & iceof_mask) >> 12; + icp->iciof = (val & iciof_mask) >> 4; + icp->icsz = (val & icsz_mask) << 4; + + return 0; +} + +int +fman_if_set_ic_params(struct fman_if *fm_if, + const struct fman_if_ic_params *icp) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + int val = 0; + int iceof_mask = 0x001f0000; + int icsz_mask = 0x0000001f; + int iciof_mask = 0x00000f00; + + assert(fman_ccsr_map_fd != -1); + + val |= (icp->iceof << 12) & iceof_mask; + val |= (icp->iciof << 4) & iciof_mask; + val |= (icp->icsz >> 4) & icsz_mask; + + unsigned int *fmbm_ricp = + &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_ricp; + out_be32(fmbm_ricp, val); + + return 0; +} + +void +fman_if_set_fdoff(struct fman_if *fm_if, uint32_t fd_offset) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + unsigned int *fmbm_rebm; + int val = 0; + int fmbm_mask = 0x01ff0000; + + val = fd_offset << FMAN_SP_EXT_BUF_MARG_START_SHIFT; + + assert(fman_ccsr_map_fd != -1); + + fmbm_rebm = &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_rebm; + + out_be32(fmbm_rebm, (in_be32(fmbm_rebm) & ~fmbm_mask) | val); +} + +void +fman_if_set_maxfrm(struct fman_if *fm_if, uint16_t max_frm) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + unsigned int *reg_maxfrm; + + assert(fman_ccsr_map_fd != -1); + + reg_maxfrm = &((struct memac_regs *)__if->ccsr_map)->maxfrm; + + out_be32(reg_maxfrm, (in_be32(reg_maxfrm) & 0xFFFF0000) | max_frm); +} + +uint16_t +fman_if_get_maxfrm(struct fman_if *fm_if) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + unsigned int *reg_maxfrm; + + assert(fman_ccsr_map_fd != -1); + + reg_maxfrm = &((struct memac_regs *)__if->ccsr_map)->maxfrm; + + return (in_be32(reg_maxfrm) | 0x0000FFFF); +} + +/* MSB in fmbm_rebm register + * 0 - If BMI cannot store the frame in a single buffer it may select a buffer + * of smaller size and store the frame in scatter gather (S/G) buffers + * 1 - Scatter gather format is not enabled for frame storage. If BMI cannot + * store the frame in a single buffer, the frame is discarded. + */ + +int +fman_if_get_sg_enable(struct fman_if *fm_if) +{ + u32 fmbm_rebm; + + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + + assert(fman_ccsr_map_fd != -1); + + fmbm_rebm = in_be32(&((struct rx_bmi_regs *)__if->bmi_map)->fmbm_rebm); + + return (fmbm_rebm & FMAN_SP_SG_DISABLE) ? 0 : 1; +} + +void +fman_if_set_sg(struct fman_if *fm_if, int enable) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + unsigned int *fmbm_rebm; + int val; + int fmbm_mask = FMAN_SP_SG_DISABLE; + + if (enable) + val = 0; + else + val = FMAN_SP_SG_DISABLE; + + assert(fman_ccsr_map_fd != -1); + + fmbm_rebm = &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_rebm; + + out_be32(fmbm_rebm, (in_be32(fmbm_rebm) & ~fmbm_mask) | val); +} + +void +fman_if_set_dnia(struct fman_if *fm_if, uint32_t nia) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + unsigned int *fmqm_pndn; + + assert(fman_ccsr_map_fd != -1); + + fmqm_pndn = &((struct fman_port_qmi_regs *)__if->qmi_map)->fmqm_pndn; + + out_be32(fmqm_pndn, nia); +} + +void +fman_if_discard_rx_errors(struct fman_if *fm_if) +{ + struct __fman_if *__if = container_of(fm_if, struct __fman_if, __if); + unsigned int *fmbm_rfsdm, *fmbm_rfsem; + + fmbm_rfsem = &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_rfsem; + out_be32(fmbm_rfsem, 0); + + /* Configure the discard mask to discard the error packets which have + * DMA errors, Frame size error, Header error etc. The mask 0x010EE3F0 + * is to configured discard all the errors which come in the FD[STATUS] + */ + fmbm_rfsdm = &((struct rx_bmi_regs *)__if->bmi_map)->fmbm_rfsdm; + out_be32(fmbm_rfsdm, 0x010EE3F0); +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/fman/netcfg_layer.c b/src/spdk/dpdk/drivers/bus/dpaa/base/fman/netcfg_layer.c new file mode 100644 index 000000000..36eca88cd --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/fman/netcfg_layer.c @@ -0,0 +1,160 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2010-2016 Freescale Semiconductor Inc. + * Copyright 2017-2019 NXP + * + */ +#include <inttypes.h> +#include <dpaa_of.h> +#include <net/if.h> +#include <sys/ioctl.h> +#include <error.h> +#include <net/if_arp.h> +#include <assert.h> +#include <unistd.h> + +#include <rte_malloc.h> + +#include <rte_dpaa_logs.h> +#include <netcfg.h> + +/* This data structure contaings all configurations information + * related to usages of DPA devices. + */ +static struct netcfg_info *netcfg; +/* fd to open a socket for making ioctl request to disable/enable shared + * interfaces. + */ +static int skfd = -1; + +#ifdef RTE_LIBRTE_DPAA_DEBUG_DRIVER +void +dump_netcfg(struct netcfg_info *cfg_ptr) +{ + int i; + + printf(".......... DPAA Configuration ..........\n\n"); + + /* Network interfaces */ + printf("Network interfaces: %d\n", cfg_ptr->num_ethports); + for (i = 0; i < cfg_ptr->num_ethports; i++) { + struct fman_if_bpool *bpool; + struct fm_eth_port_cfg *p_cfg = &cfg_ptr->port_cfg[i]; + struct fman_if *__if = p_cfg->fman_if; + + printf("\n+ Fman %d, MAC %d (%s);\n", + __if->fman_idx, __if->mac_idx, + (__if->mac_type == fman_mac_1g) ? "1G" : "10G"); + + printf("\tmac_addr: %02x:%02x:%02x:%02x:%02x:%02x\n", + (&__if->mac_addr)->addr_bytes[0], + (&__if->mac_addr)->addr_bytes[1], + (&__if->mac_addr)->addr_bytes[2], + (&__if->mac_addr)->addr_bytes[3], + (&__if->mac_addr)->addr_bytes[4], + (&__if->mac_addr)->addr_bytes[5]); + + printf("\ttx_channel_id: 0x%02x\n", + __if->tx_channel_id); + + printf("\tfqid_rx_def: 0x%x\n", p_cfg->rx_def); + printf("\tfqid_rx_err: 0x%x\n", __if->fqid_rx_err); + + printf("\tfqid_tx_err: 0x%x\n", __if->fqid_tx_err); + printf("\tfqid_tx_confirm: 0x%x\n", __if->fqid_tx_confirm); + fman_if_for_each_bpool(bpool, __if) + printf("\tbuffer pool: (bpid=%d, count=%"PRId64 + " size=%"PRId64", addr=0x%"PRIx64")\n", + bpool->bpid, bpool->count, bpool->size, + bpool->addr); + } +} +#endif /* RTE_LIBRTE_DPAA_DEBUG_DRIVER */ + +struct netcfg_info * +netcfg_acquire(void) +{ + struct fman_if *__if; + int _errno, idx = 0; + uint8_t num_ports = 0; + uint8_t num_cfg_ports = 0; + size_t size; + + /* Extract dpa configuration from fman driver and FMC configuration + * for command-line interfaces. + */ + + /* Open a basic socket to enable/disable shared + * interfaces. + */ + skfd = socket(AF_PACKET, SOCK_RAW, 0); + if (unlikely(skfd < 0)) { + error(0, errno, "%s(): open(SOCK_RAW)", __func__); + return NULL; + } + + /* Initialise the Fman driver */ + _errno = fman_init(); + if (_errno) { + DPAA_BUS_LOG(ERR, "FMAN driver init failed (%d)", errno); + close(skfd); + skfd = -1; + return NULL; + } + + /* Number of MAC ports */ + list_for_each_entry(__if, fman_if_list, node) + num_ports++; + + if (!num_ports) { + DPAA_BUS_LOG(ERR, "FMAN ports not available"); + return NULL; + } + /* Allocate space for all enabled mac ports */ + size = sizeof(*netcfg) + + (num_ports * sizeof(struct fm_eth_port_cfg)); + + netcfg = rte_calloc(NULL, 1, size, 0); + if (unlikely(netcfg == NULL)) { + DPAA_BUS_LOG(ERR, "Unable to allocat mem for netcfg"); + goto error; + } + + netcfg->num_ethports = num_ports; + + list_for_each_entry(__if, fman_if_list, node) { + struct fm_eth_port_cfg *cfg = &netcfg->port_cfg[idx]; + /* Hook in the fman driver interface */ + cfg->fman_if = __if; + cfg->rx_def = __if->fqid_rx_def; + num_cfg_ports++; + idx++; + } + + if (!num_cfg_ports) { + DPAA_BUS_LOG(ERR, "No FMAN ports found"); + goto error; + } else if (num_ports != num_cfg_ports) + netcfg->num_ethports = num_cfg_ports; + + return netcfg; + +error: + if (netcfg) { + rte_free(netcfg); + netcfg = NULL; + } + + return NULL; +} + +void +netcfg_release(struct netcfg_info *cfg_ptr) +{ + rte_free(cfg_ptr); + /* Close socket for shared interfaces */ + if (skfd >= 0) { + close(skfd); + skfd = -1; + } +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman.c b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman.c new file mode 100644 index 000000000..8a6290734 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman.c @@ -0,0 +1,361 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ + +#include "bman.h" +#include <rte_branch_prediction.h> + +/* Compilation constants */ +#define RCR_THRESH 2 /* reread h/w CI when running out of space */ +#define IRQNAME "BMan portal %d" +#define MAX_IRQNAME 16 /* big enough for "BMan portal %d" */ + +struct bman_portal { + struct bm_portal p; + /* 2-element array. pools[0] is mask, pools[1] is snapshot. */ + struct bman_depletion *pools; + int thresh_set; + unsigned long irq_sources; + u32 slowpoll; /* only used when interrupts are off */ + /* When the cpu-affine portal is activated, this is non-NULL */ + const struct bm_portal_config *config; + char irqname[MAX_IRQNAME]; +}; + +static cpumask_t affine_mask; +static DEFINE_SPINLOCK(affine_mask_lock); +static RTE_DEFINE_PER_LCORE(struct bman_portal, bman_affine_portal); + +static inline struct bman_portal *get_affine_portal(void) +{ + return &RTE_PER_LCORE(bman_affine_portal); +} + +/* + * This object type refers to a pool, it isn't *the* pool. There may be + * more than one such object per BMan buffer pool, eg. if different users of + * the pool are operating via different portals. + */ +struct bman_pool { + struct bman_pool_params params; + /* Used for hash-table admin when using depletion notifications. */ + struct bman_portal *portal; + struct bman_pool *next; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + atomic_t in_use; +#endif +}; + +static inline +struct bman_portal *bman_create_portal(struct bman_portal *portal, + const struct bm_portal_config *c) +{ + struct bm_portal *p; + const struct bman_depletion *pools = &c->mask; + int ret; + u8 bpid = 0; + + p = &portal->p; + /* + * prep the low-level portal struct with the mapped addresses from the + * config, everything that follows depends on it and "config" is more + * for (de)reference... + */ + p->addr.ce = c->addr_virt[DPAA_PORTAL_CE]; + p->addr.ci = c->addr_virt[DPAA_PORTAL_CI]; + if (bm_rcr_init(p, bm_rcr_pvb, bm_rcr_cce)) { + pr_err("Bman RCR initialisation failed\n"); + return NULL; + } + if (bm_mc_init(p)) { + pr_err("Bman MC initialisation failed\n"); + goto fail_mc; + } + portal->pools = kmalloc(2 * sizeof(*pools), GFP_KERNEL); + if (!portal->pools) + goto fail_pools; + portal->pools[0] = *pools; + bman_depletion_init(portal->pools + 1); + while (bpid < bman_pool_max) { + /* + * Default to all BPIDs disabled, we enable as required at + * run-time. + */ + bm_isr_bscn_mask(p, bpid, 0); + bpid++; + } + portal->slowpoll = 0; + /* Write-to-clear any stale interrupt status bits */ + bm_isr_disable_write(p, 0xffffffff); + portal->irq_sources = 0; + bm_isr_enable_write(p, portal->irq_sources); + bm_isr_status_clear(p, 0xffffffff); + snprintf(portal->irqname, MAX_IRQNAME, IRQNAME, c->cpu); + if (request_irq(c->irq, NULL, 0, portal->irqname, + portal)) { + pr_err("request_irq() failed\n"); + goto fail_irq; + } + + /* Need RCR to be empty before continuing */ + ret = bm_rcr_get_fill(p); + if (ret) { + pr_err("Bman RCR unclean\n"); + goto fail_rcr_empty; + } + /* Success */ + portal->config = c; + + bm_isr_disable_write(p, 0); + bm_isr_uninhibit(p); + return portal; +fail_rcr_empty: + free_irq(c->irq, portal); +fail_irq: + kfree(portal->pools); +fail_pools: + bm_mc_finish(p); +fail_mc: + bm_rcr_finish(p); + return NULL; +} + +struct bman_portal * +bman_create_affine_portal(const struct bm_portal_config *c) +{ + struct bman_portal *portal = get_affine_portal(); + + /*This function is called from the context which is already affine to + *CPU or in other words this in non-migratable to other CPUs. + */ + portal = bman_create_portal(portal, c); + if (portal) { + spin_lock(&affine_mask_lock); + CPU_SET(c->cpu, &affine_mask); + spin_unlock(&affine_mask_lock); + } + return portal; +} + +static inline +void bman_destroy_portal(struct bman_portal *bm) +{ + const struct bm_portal_config *pcfg; + + pcfg = bm->config; + bm_rcr_cce_update(&bm->p); + bm_rcr_cce_update(&bm->p); + + free_irq(pcfg->irq, bm); + + kfree(bm->pools); + bm_mc_finish(&bm->p); + bm_rcr_finish(&bm->p); + bm->config = NULL; +} + +const struct +bm_portal_config *bman_destroy_affine_portal(void) +{ + struct bman_portal *bm = get_affine_portal(); + const struct bm_portal_config *pcfg; + + pcfg = bm->config; + bman_destroy_portal(bm); + spin_lock(&affine_mask_lock); + CPU_CLR(pcfg->cpu, &affine_mask); + spin_unlock(&affine_mask_lock); + return pcfg; +} + +int +bman_get_portal_index(void) +{ + struct bman_portal *p = get_affine_portal(); + return p->config->index; +} + +static const u32 zero_thresholds[4] = {0, 0, 0, 0}; + +struct bman_pool *bman_new_pool(const struct bman_pool_params *params) +{ + struct bman_pool *pool = NULL; + u32 bpid; + + if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID) { + int ret = bman_alloc_bpid(&bpid); + + if (ret) + return NULL; + } else { + if (params->bpid >= bman_pool_max) + return NULL; + bpid = params->bpid; + } + if (params->flags & BMAN_POOL_FLAG_THRESH) { + int ret = bm_pool_set(bpid, params->thresholds); + + if (ret) + goto err; + } + + pool = kmalloc(sizeof(*pool), GFP_KERNEL); + if (!pool) + goto err; + pool->params = *params; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + atomic_set(&pool->in_use, 1); +#endif + if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID) + pool->params.bpid = bpid; + + return pool; +err: + if (params->flags & BMAN_POOL_FLAG_THRESH) + bm_pool_set(bpid, zero_thresholds); + + if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID) + bman_release_bpid(bpid); + kfree(pool); + + return NULL; +} + +void bman_free_pool(struct bman_pool *pool) +{ + if (pool->params.flags & BMAN_POOL_FLAG_THRESH) + bm_pool_set(pool->params.bpid, zero_thresholds); + if (pool->params.flags & BMAN_POOL_FLAG_DYNAMIC_BPID) + bman_release_bpid(pool->params.bpid); + kfree(pool); +} + +const struct bman_pool_params *bman_get_params(const struct bman_pool *pool) +{ + return &pool->params; +} + +static void update_rcr_ci(struct bman_portal *p, int avail) +{ + if (avail) + bm_rcr_cce_prefetch(&p->p); + else + bm_rcr_cce_update(&p->p); +} + +#define BMAN_BUF_MASK 0x0000fffffffffffful +int bman_release(struct bman_pool *pool, const struct bm_buffer *bufs, u8 num, + u32 flags __maybe_unused) +{ + struct bman_portal *p; + struct bm_rcr_entry *r; + u32 i = num - 1; + u8 avail; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (!num || (num > 8)) + return -EINVAL; + if (pool->params.flags & BMAN_POOL_FLAG_NO_RELEASE) + return -EINVAL; +#endif + + p = get_affine_portal(); + avail = bm_rcr_get_avail(&p->p); + if (avail < 2) + update_rcr_ci(p, avail); + r = bm_rcr_start(&p->p); + if (unlikely(!r)) + return -EBUSY; + + /* + * we can copy all but the first entry, as this can trigger badness + * with the valid-bit + */ + r->bufs[0].opaque = + cpu_to_be64(((u64)pool->params.bpid << 48) | + (bufs[0].opaque & BMAN_BUF_MASK)); + if (i) { + for (i = 1; i < num; i++) + r->bufs[i].opaque = + cpu_to_be64(bufs[i].opaque & BMAN_BUF_MASK); + } + + bm_rcr_pvb_commit(&p->p, BM_RCR_VERB_CMD_BPID_SINGLE | + (num & BM_RCR_VERB_BUFCOUNT_MASK)); + + return 0; +} + +int bman_acquire(struct bman_pool *pool, struct bm_buffer *bufs, u8 num, + u32 flags __maybe_unused) +{ + struct bman_portal *p = get_affine_portal(); + struct bm_mc_command *mcc; + struct bm_mc_result *mcr; + int ret, i; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (!num || (num > 8)) + return -EINVAL; + if (pool->params.flags & BMAN_POOL_FLAG_ONLY_RELEASE) + return -EINVAL; +#endif + + mcc = bm_mc_start(&p->p); + mcc->acquire.bpid = pool->params.bpid; + bm_mc_commit(&p->p, BM_MCC_VERB_CMD_ACQUIRE | + (num & BM_MCC_VERB_ACQUIRE_BUFCOUNT)); + while (!(mcr = bm_mc_result(&p->p))) + cpu_relax(); + ret = mcr->verb & BM_MCR_VERB_ACQUIRE_BUFCOUNT; + if (bufs) { + for (i = 0; i < num; i++) + bufs[i].opaque = + be64_to_cpu(mcr->acquire.bufs[i].opaque); + } + if (ret != num) + ret = -ENOMEM; + return ret; +} + +int bman_query_pools(struct bm_pool_state *state) +{ + struct bman_portal *p = get_affine_portal(); + struct bm_mc_result *mcr; + + bm_mc_start(&p->p); + bm_mc_commit(&p->p, BM_MCC_VERB_CMD_QUERY); + while (!(mcr = bm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & BM_MCR_VERB_CMD_MASK) == + BM_MCR_VERB_CMD_QUERY); + *state = mcr->query; + state->as.state.state[0] = be32_to_cpu(state->as.state.state[0]); + state->as.state.state[1] = be32_to_cpu(state->as.state.state[1]); + state->ds.state.state[0] = be32_to_cpu(state->ds.state.state[0]); + state->ds.state.state[1] = be32_to_cpu(state->ds.state.state[1]); + return 0; +} + +u32 bman_query_free_buffers(struct bman_pool *pool) +{ + return bm_pool_free_buffers(pool->params.bpid); +} + +int bman_update_pool_thresholds(struct bman_pool *pool, const u32 *thresholds) +{ + u32 bpid; + + bpid = bman_get_params(pool)->bpid; + + return bm_pool_set(bpid, thresholds); +} + +int bman_shutdown_pool(u32 bpid) +{ + struct bman_portal *p = get_affine_portal(); + return bm_shutdown_pool(&p->p, bpid); +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman.h b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman.h new file mode 100644 index 000000000..21a6bee77 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman.h @@ -0,0 +1,541 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2010-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ + +#ifndef __BMAN_H +#define __BMAN_H + +#include "bman_priv.h" + +/* Cache-inhibited register offsets */ +#define BM_REG_RCR_PI_CINH 0x3000 +#define BM_REG_RCR_CI_CINH 0x3100 +#define BM_REG_RCR_ITR 0x3200 +#define BM_REG_CFG 0x3300 +#define BM_REG_SCN(n) (0x3400 + ((n) << 6)) +#define BM_REG_ISR 0x3e00 +#define BM_REG_IIR 0x3ec0 + +/* Cache-enabled register offsets */ +#define BM_CL_CR 0x0000 +#define BM_CL_RR0 0x0100 +#define BM_CL_RR1 0x0140 +#define BM_CL_RCR 0x1000 +#define BM_CL_RCR_PI_CENA 0x3000 +#define BM_CL_RCR_CI_CENA 0x3100 + +/* BTW, the drivers (and h/w programming model) already obtain the required + * synchronisation for portal accesses via lwsync(), hwsync(), and + * data-dependencies. Use of barrier()s or other order-preserving primitives + * simply degrade performance. Hence the use of the __raw_*() interfaces, which + * simply ensure that the compiler treats the portal registers as volatile (ie. + * non-coherent). + */ + +/* Cache-inhibited register access. */ +#define __bm_in(bm, o) be32_to_cpu(__raw_readl((bm)->ci + (o))) +#define __bm_out(bm, o, val) __raw_writel(cpu_to_be32(val), \ + (bm)->ci + (o)) +#define bm_in(reg) __bm_in(&portal->addr, BM_REG_##reg) +#define bm_out(reg, val) __bm_out(&portal->addr, BM_REG_##reg, val) + +/* Cache-enabled (index) register access */ +#define __bm_cl_touch_ro(bm, o) dcbt_ro((bm)->ce + (o)) +#define __bm_cl_touch_rw(bm, o) dcbt_rw((bm)->ce + (o)) +#define __bm_cl_in(bm, o) be32_to_cpu(__raw_readl((bm)->ce + (o))) +#define __bm_cl_out(bm, o, val) \ + do { \ + u32 *__tmpclout = (bm)->ce + (o); \ + __raw_writel(cpu_to_be32(val), __tmpclout); \ + dcbf(__tmpclout); \ + } while (0) +#define __bm_cl_invalidate(bm, o) dccivac((bm)->ce + (o)) +#define bm_cl_touch_ro(reg) __bm_cl_touch_ro(&portal->addr, BM_CL_##reg##_CENA) +#define bm_cl_touch_rw(reg) __bm_cl_touch_rw(&portal->addr, BM_CL_##reg##_CENA) +#define bm_cl_in(reg) __bm_cl_in(&portal->addr, BM_CL_##reg##_CENA) +#define bm_cl_out(reg, val) __bm_cl_out(&portal->addr, BM_CL_##reg##_CENA, val) +#define bm_cl_invalidate(reg)\ + __bm_cl_invalidate(&portal->addr, BM_CL_##reg##_CENA) + +/* Cyclic helper for rings. FIXME: once we are able to do fine-grain perf + * analysis, look at using the "extra" bit in the ring index registers to avoid + * cyclic issues. + */ +static inline u8 bm_cyc_diff(u8 ringsize, u8 first, u8 last) +{ + /* 'first' is included, 'last' is excluded */ + if (first <= last) + return last - first; + return ringsize + last - first; +} + +/* Portal modes. + * Enum types; + * pmode == production mode + * cmode == consumption mode, + * Enum values use 3 letter codes. First letter matches the portal mode, + * remaining two letters indicate; + * ci == cache-inhibited portal register + * ce == cache-enabled portal register + * vb == in-band valid-bit (cache-enabled) + */ +enum bm_rcr_pmode { /* matches BCSP_CFG::RPM */ + bm_rcr_pci = 0, /* PI index, cache-inhibited */ + bm_rcr_pce = 1, /* PI index, cache-enabled */ + bm_rcr_pvb = 2 /* valid-bit */ +}; + +enum bm_rcr_cmode { /* s/w-only */ + bm_rcr_cci, /* CI index, cache-inhibited */ + bm_rcr_cce /* CI index, cache-enabled */ +}; + +/* --- Portal structures --- */ + +#define BM_RCR_SIZE 8 + +struct bm_rcr { + struct bm_rcr_entry *ring, *cursor; + u8 ci, available, ithresh, vbit; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + u32 busy; + enum bm_rcr_pmode pmode; + enum bm_rcr_cmode cmode; +#endif +}; + +struct bm_mc { + struct bm_mc_command *cr; + struct bm_mc_result *rr; + u8 rridx, vbit; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + enum { + /* Can only be _mc_start()ed */ + mc_idle, + /* Can only be _mc_commit()ed or _mc_abort()ed */ + mc_user, + /* Can only be _mc_retry()ed */ + mc_hw + } state; +#endif +}; + +struct bm_addr { + void __iomem *ce; /* cache-enabled */ + void __iomem *ci; /* cache-inhibited */ +}; + +struct bm_portal { + struct bm_addr addr; + struct bm_rcr rcr; + struct bm_mc mc; + struct bm_portal_config config; +} ____cacheline_aligned; + +/* Bit-wise logic to wrap a ring pointer by clearing the "carry bit" */ +#define RCR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(BM_RCR_SIZE << 6))) + +/* Bit-wise logic to convert a ring pointer to a ring index */ +static inline u8 RCR_PTR2IDX(struct bm_rcr_entry *e) +{ + return ((uintptr_t)e >> 6) & (BM_RCR_SIZE - 1); +} + +/* Increment the 'cursor' ring pointer, taking 'vbit' into account */ +static inline void RCR_INC(struct bm_rcr *rcr) +{ + /* NB: this is odd-looking, but experiments show that it generates + * fast code with essentially no branching overheads. We increment to + * the next RCR pointer and handle overflow and 'vbit'. + */ + struct bm_rcr_entry *partial = rcr->cursor + 1; + + rcr->cursor = RCR_CARRYCLEAR(partial); + if (partial != rcr->cursor) + rcr->vbit ^= BM_RCR_VERB_VBIT; +} + +static inline int bm_rcr_init(struct bm_portal *portal, enum bm_rcr_pmode pmode, + __maybe_unused enum bm_rcr_cmode cmode) +{ + /* This use of 'register', as well as all other occurrences, is because + * it has been observed to generate much faster code with gcc than is + * otherwise the case. + */ + register struct bm_rcr *rcr = &portal->rcr; + u32 cfg; + u8 pi; + + rcr->ring = portal->addr.ce + BM_CL_RCR; + rcr->ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1); + + pi = bm_in(RCR_PI_CINH) & (BM_RCR_SIZE - 1); + rcr->cursor = rcr->ring + pi; + rcr->vbit = (bm_in(RCR_PI_CINH) & BM_RCR_SIZE) ? BM_RCR_VERB_VBIT : 0; + rcr->available = BM_RCR_SIZE - 1 + - bm_cyc_diff(BM_RCR_SIZE, rcr->ci, pi); + rcr->ithresh = bm_in(RCR_ITR); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + rcr->busy = 0; + rcr->pmode = pmode; + rcr->cmode = cmode; +#endif + cfg = (bm_in(CFG) & 0xffffffe0) | (pmode & 0x3); /* BCSP_CFG::RPM */ + bm_out(CFG, cfg); + return 0; +} + +static inline void bm_rcr_finish(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + u8 pi = bm_in(RCR_PI_CINH) & (BM_RCR_SIZE - 1); + u8 ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1); + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(!rcr->busy); +#endif + if (pi != RCR_PTR2IDX(rcr->cursor)) + pr_crit("losing uncommitted RCR entries\n"); + if (ci != rcr->ci) + pr_crit("missing existing RCR completions\n"); + if (rcr->ci != RCR_PTR2IDX(rcr->cursor)) + pr_crit("RCR destroyed unquiesced\n"); +} + +static inline struct bm_rcr_entry *bm_rcr_start(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(!rcr->busy); +#endif + if (!rcr->available) + return NULL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + rcr->busy = 1; +#endif + dcbz_64(rcr->cursor); + return rcr->cursor; +} + +static inline void bm_rcr_abort(struct bm_portal *portal) +{ + __maybe_unused register struct bm_rcr *rcr = &portal->rcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->busy); + rcr->busy = 0; +#endif +} + +static inline struct bm_rcr_entry *bm_rcr_pend_and_next( + struct bm_portal *portal, u8 myverb) +{ + register struct bm_rcr *rcr = &portal->rcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->busy); + DPAA_ASSERT(rcr->pmode != bm_rcr_pvb); +#endif + if (rcr->available == 1) + return NULL; + rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit; + dcbf_64(rcr->cursor); + RCR_INC(rcr); + rcr->available--; + dcbz_64(rcr->cursor); + return rcr->cursor; +} + +static inline void bm_rcr_pci_commit(struct bm_portal *portal, u8 myverb) +{ + register struct bm_rcr *rcr = &portal->rcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->busy); + DPAA_ASSERT(rcr->pmode == bm_rcr_pci); +#endif + rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit; + RCR_INC(rcr); + rcr->available--; + hwsync(); + bm_out(RCR_PI_CINH, RCR_PTR2IDX(rcr->cursor)); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + rcr->busy = 0; +#endif +} + +static inline void bm_rcr_pce_prefetch(struct bm_portal *portal) +{ + __maybe_unused register struct bm_rcr *rcr = &portal->rcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->pmode == bm_rcr_pce); +#endif + bm_cl_invalidate(RCR_PI); + bm_cl_touch_rw(RCR_PI); +} + +static inline void bm_rcr_pce_commit(struct bm_portal *portal, u8 myverb) +{ + register struct bm_rcr *rcr = &portal->rcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->busy); + DPAA_ASSERT(rcr->pmode == bm_rcr_pce); +#endif + rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit; + RCR_INC(rcr); + rcr->available--; + lwsync(); + bm_cl_out(RCR_PI, RCR_PTR2IDX(rcr->cursor)); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + rcr->busy = 0; +#endif +} + +static inline void bm_rcr_pvb_commit(struct bm_portal *portal, u8 myverb) +{ + register struct bm_rcr *rcr = &portal->rcr; + struct bm_rcr_entry *rcursor; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->busy); + DPAA_ASSERT(rcr->pmode == bm_rcr_pvb); +#endif + lwsync(); + rcursor = rcr->cursor; + rcursor->__dont_write_directly__verb = myverb | rcr->vbit; + dcbf_64(rcursor); + RCR_INC(rcr); + rcr->available--; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + rcr->busy = 0; +#endif +} + +static inline u8 bm_rcr_cci_update(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + u8 diff, old_ci = rcr->ci; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->cmode == bm_rcr_cci); +#endif + rcr->ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1); + diff = bm_cyc_diff(BM_RCR_SIZE, old_ci, rcr->ci); + rcr->available += diff; + return diff; +} + +static inline void bm_rcr_cce_prefetch(struct bm_portal *portal) +{ + __maybe_unused register struct bm_rcr *rcr = &portal->rcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->cmode == bm_rcr_cce); +#endif + bm_cl_touch_ro(RCR_CI); +} + +static inline u8 bm_rcr_cce_update(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + u8 diff, old_ci = rcr->ci; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(rcr->cmode == bm_rcr_cce); +#endif + rcr->ci = bm_cl_in(RCR_CI) & (BM_RCR_SIZE - 1); + bm_cl_invalidate(RCR_CI); + diff = bm_cyc_diff(BM_RCR_SIZE, old_ci, rcr->ci); + rcr->available += diff; + return diff; +} + +static inline u8 bm_rcr_get_ithresh(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + + return rcr->ithresh; +} + +static inline void bm_rcr_set_ithresh(struct bm_portal *portal, u8 ithresh) +{ + register struct bm_rcr *rcr = &portal->rcr; + + rcr->ithresh = ithresh; + bm_out(RCR_ITR, ithresh); +} + +static inline u8 bm_rcr_get_avail(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + + return rcr->available; +} + +static inline u8 bm_rcr_get_fill(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + + return BM_RCR_SIZE - 1 - rcr->available; +} + +/* --- Management command API --- */ + +static inline int bm_mc_init(struct bm_portal *portal) +{ + register struct bm_mc *mc = &portal->mc; + + mc->cr = portal->addr.ce + BM_CL_CR; + mc->rr = portal->addr.ce + BM_CL_RR0; + mc->rridx = (__raw_readb(&mc->cr->__dont_write_directly__verb) & + BM_MCC_VERB_VBIT) ? 0 : 1; + mc->vbit = mc->rridx ? BM_MCC_VERB_VBIT : 0; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + mc->state = mc_idle; +#endif + return 0; +} + +static inline void bm_mc_finish(struct bm_portal *portal) +{ + __maybe_unused register struct bm_mc *mc = &portal->mc; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == mc_idle); + if (mc->state != mc_idle) + pr_crit("Losing incomplete MC command\n"); +#endif +} + +static inline struct bm_mc_command *bm_mc_start(struct bm_portal *portal) +{ + register struct bm_mc *mc = &portal->mc; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == mc_idle); + mc->state = mc_user; +#endif + dcbz_64(mc->cr); + return mc->cr; +} + +static inline void bm_mc_abort(struct bm_portal *portal) +{ + __maybe_unused register struct bm_mc *mc = &portal->mc; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == mc_user); + mc->state = mc_idle; +#endif +} + +static inline void bm_mc_commit(struct bm_portal *portal, u8 myverb) +{ + register struct bm_mc *mc = &portal->mc; + struct bm_mc_result *rr = mc->rr + mc->rridx; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == mc_user); +#endif + lwsync(); + mc->cr->__dont_write_directly__verb = myverb | mc->vbit; + dcbf(mc->cr); + dcbit_ro(rr); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + mc->state = mc_hw; +#endif +} + +static inline struct bm_mc_result *bm_mc_result(struct bm_portal *portal) +{ + register struct bm_mc *mc = &portal->mc; + struct bm_mc_result *rr = mc->rr + mc->rridx; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == mc_hw); +#endif + /* The inactive response register's verb byte always returns zero until + * its command is submitted and completed. This includes the valid-bit, + * in case you were wondering. + */ + if (!__raw_readb(&rr->verb)) { + dcbit_ro(rr); + return NULL; + } + mc->rridx ^= 1; + mc->vbit ^= BM_MCC_VERB_VBIT; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + mc->state = mc_idle; +#endif + return rr; +} + +#define SCN_REG(bpid) BM_REG_SCN((bpid) / 32) +#define SCN_BIT(bpid) (0x80000000 >> (bpid & 31)) +static inline void bm_isr_bscn_mask(struct bm_portal *portal, u8 bpid, + int enable) +{ + u32 val; + + DPAA_ASSERT(bpid < bman_pool_max); + /* REG_SCN for bpid=0..31, REG_SCN+4 for bpid=32..63 */ + val = __bm_in(&portal->addr, SCN_REG(bpid)); + if (enable) + val |= SCN_BIT(bpid); + else + val &= ~SCN_BIT(bpid); + __bm_out(&portal->addr, SCN_REG(bpid), val); +} + +static inline u32 __bm_isr_read(struct bm_portal *portal, enum bm_isr_reg n) +{ +#if defined(RTE_ARCH_ARM64) + return __bm_in(&portal->addr, BM_REG_ISR + (n << 6)); +#else + return __bm_in(&portal->addr, BM_REG_ISR + (n << 2)); +#endif +} + +static inline void __bm_isr_write(struct bm_portal *portal, enum bm_isr_reg n, + u32 val) +{ +#if defined(RTE_ARCH_ARM64) + __bm_out(&portal->addr, BM_REG_ISR + (n << 6), val); +#else + __bm_out(&portal->addr, BM_REG_ISR + (n << 2), val); +#endif +} + +/* Buffer Pool Cleanup */ +static inline int bm_shutdown_pool(struct bm_portal *p, u32 bpid) +{ + struct bm_mc_command *bm_cmd; + struct bm_mc_result *bm_res; + + int aq_count = 0; + bool stop = false; + + while (!stop) { + /* Acquire buffers until empty */ + bm_cmd = bm_mc_start(p); + bm_cmd->acquire.bpid = bpid; + bm_mc_commit(p, BM_MCC_VERB_CMD_ACQUIRE | 1); + while (!(bm_res = bm_mc_result(p))) + cpu_relax(); + if (!(bm_res->verb & BM_MCR_VERB_ACQUIRE_BUFCOUNT)) { + /* Pool is empty */ + stop = true; + } else + ++aq_count; + }; + return 0; +} + +#endif /* __BMAN_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman_driver.c b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman_driver.c new file mode 100644 index 000000000..750b756b9 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman_driver.c @@ -0,0 +1,295 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ + +#include <rte_branch_prediction.h> + +#include <fsl_usd.h> +#include <process.h> +#include "bman_priv.h" +#include <sys/ioctl.h> + +/* + * Global variables of the max portal/pool number this bman version supported + */ +static u16 bman_ip_rev; +u16 bman_pool_max; +static void *bman_ccsr_map; + +/*****************/ +/* Portal driver */ +/*****************/ + +static __thread int bmfd = -1; +static __thread struct bm_portal_config pcfg; +static __thread struct dpaa_ioctl_portal_map map = { + .type = dpaa_portal_bman +}; + +static int fsl_bman_portal_init(uint32_t idx, int is_shared) +{ + cpu_set_t cpuset; + struct bman_portal *portal; + int loop, ret; + struct dpaa_ioctl_irq_map irq_map; + + /* Verify the thread's cpu-affinity */ + ret = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t), + &cpuset); + if (ret) { + error(0, ret, "pthread_getaffinity_np()"); + return ret; + } + pcfg.cpu = -1; + for (loop = 0; loop < CPU_SETSIZE; loop++) + if (CPU_ISSET(loop, &cpuset)) { + if (pcfg.cpu != -1) { + pr_err("Thread is not affine to 1 cpu"); + return -EINVAL; + } + pcfg.cpu = loop; + } + if (pcfg.cpu == -1) { + pr_err("Bug in getaffinity handling!"); + return -EINVAL; + } + /* Allocate and map a bman portal */ + map.index = idx; + ret = process_portal_map(&map); + if (ret) { + error(0, ret, "process_portal_map()"); + return ret; + } + /* Make the portal's cache-[enabled|inhibited] regions */ + pcfg.addr_virt[DPAA_PORTAL_CE] = map.addr.cena; + pcfg.addr_virt[DPAA_PORTAL_CI] = map.addr.cinh; + pcfg.is_shared = is_shared; + pcfg.index = map.index; + bman_depletion_fill(&pcfg.mask); + + bmfd = open(BMAN_PORTAL_IRQ_PATH, O_RDONLY); + if (bmfd == -1) { + pr_err("BMan irq init failed"); + process_portal_unmap(&map.addr); + return -EBUSY; + } + /* Use the IRQ FD as a unique IRQ number */ + pcfg.irq = bmfd; + + portal = bman_create_affine_portal(&pcfg); + if (!portal) { + pr_err("Bman portal initialisation failed (%d)", + pcfg.cpu); + process_portal_unmap(&map.addr); + return -EBUSY; + } + + /* Set the IRQ number */ + irq_map.type = dpaa_portal_bman; + irq_map.portal_cinh = map.addr.cinh; + process_portal_irq_map(bmfd, &irq_map); + return 0; +} + +static int fsl_bman_portal_finish(void) +{ + __maybe_unused const struct bm_portal_config *cfg; + int ret; + + process_portal_irq_unmap(bmfd); + + cfg = bman_destroy_affine_portal(); + DPAA_BUG_ON(cfg != &pcfg); + ret = process_portal_unmap(&map.addr); + if (ret) + error(0, ret, "process_portal_unmap()"); + return ret; +} + +int bman_thread_fd(void) +{ + return bmfd; +} + +int bman_thread_init(void) +{ + /* Convert from contiguous/virtual cpu numbering to real cpu when + * calling into the code that is dependent on the device naming. + */ + return fsl_bman_portal_init(QBMAN_ANY_PORTAL_IDX, 0); +} + +int bman_thread_finish(void) +{ + return fsl_bman_portal_finish(); +} + +void bman_thread_irq(void) +{ + qbman_invoke_irq(pcfg.irq); + /* Now we need to uninhibit interrupts. This is the only code outside + * the regular portal driver that manipulates any portal register, so + * rather than breaking that encapsulation I am simply hard-coding the + * offset to the inhibit register here. + */ + out_be32(pcfg.addr_virt[DPAA_PORTAL_CI] + 0xe0c, 0); +} + +int bman_init_ccsr(const struct device_node *node) +{ + static int ccsr_map_fd; + uint64_t phys_addr; + const uint32_t *bman_addr; + uint64_t regs_size; + + bman_addr = of_get_address(node, 0, ®s_size, NULL); + if (!bman_addr) { + pr_err("of_get_address cannot return BMan address"); + return -EINVAL; + } + phys_addr = of_translate_address(node, bman_addr); + if (!phys_addr) { + pr_err("of_translate_address failed"); + return -EINVAL; + } + + ccsr_map_fd = open(BMAN_CCSR_MAP, O_RDWR); + if (unlikely(ccsr_map_fd < 0)) { + pr_err("Can not open /dev/mem for BMan CCSR map"); + return ccsr_map_fd; + } + + bman_ccsr_map = mmap(NULL, regs_size, PROT_READ | + PROT_WRITE, MAP_SHARED, ccsr_map_fd, phys_addr); + if (bman_ccsr_map == MAP_FAILED) { + pr_err("Can not map BMan CCSR base Bman: " + "0x%x Phys: 0x%" PRIx64 " size 0x%" PRIu64, + *bman_addr, phys_addr, regs_size); + return -EINVAL; + } + + return 0; +} + +int bman_global_init(void) +{ + const struct device_node *dt_node; + static int done; + + if (done) + return -EBUSY; + /* Use the device-tree to determine IP revision until something better + * is devised. + */ + dt_node = of_find_compatible_node(NULL, NULL, "fsl,bman-portal"); + if (!dt_node) { + pr_err("No bman portals available for any CPU\n"); + return -ENODEV; + } + if (of_device_is_compatible(dt_node, "fsl,bman-portal-1.0") || + of_device_is_compatible(dt_node, "fsl,bman-portal-1.0.0")) { + bman_ip_rev = BMAN_REV10; + bman_pool_max = 64; + } else if (of_device_is_compatible(dt_node, "fsl,bman-portal-2.0") || + of_device_is_compatible(dt_node, "fsl,bman-portal-2.0.8")) { + bman_ip_rev = BMAN_REV20; + bman_pool_max = 8; + } else if (of_device_is_compatible(dt_node, "fsl,bman-portal-2.1.0") || + of_device_is_compatible(dt_node, "fsl,bman-portal-2.1.1") || + of_device_is_compatible(dt_node, "fsl,bman-portal-2.1.2") || + of_device_is_compatible(dt_node, "fsl,bman-portal-2.1.3")) { + bman_ip_rev = BMAN_REV21; + bman_pool_max = 64; + } else { + pr_warn("unknown BMan version in portal node,default " + "to rev1.0"); + bman_ip_rev = BMAN_REV10; + bman_pool_max = 64; + } + + if (!bman_ip_rev) { + pr_err("Unknown bman portal version\n"); + return -ENODEV; + } + { + const struct device_node *dn = of_find_compatible_node(NULL, + NULL, "fsl,bman"); + if (!dn) + pr_err("No bman device node available"); + + if (bman_init_ccsr(dn)) + pr_err("BMan CCSR map failed."); + } + + done = 1; + return 0; +} + +#define BMAN_POOL_CONTENT(n) (0x0600 + ((n) * 0x04)) +u32 bm_pool_free_buffers(u32 bpid) +{ + return in_be32(bman_ccsr_map + BMAN_POOL_CONTENT(bpid)); +} + +static u32 __generate_thresh(u32 val, int roundup) +{ + u32 e = 0; /* co-efficient, exponent */ + int oddbit = 0; + + while (val > 0xff) { + oddbit = val & 1; + val >>= 1; + e++; + if (roundup && oddbit) + val++; + } + DPAA_ASSERT(e < 0x10); + return (val | (e << 8)); +} + +#define POOL_SWDET(n) (0x0000 + ((n) * 0x04)) +#define POOL_HWDET(n) (0x0100 + ((n) * 0x04)) +#define POOL_SWDXT(n) (0x0200 + ((n) * 0x04)) +#define POOL_HWDXT(n) (0x0300 + ((n) * 0x04)) +int bm_pool_set(u32 bpid, const u32 *thresholds) +{ + if (!bman_ccsr_map) + return -ENODEV; + if (bpid >= bman_pool_max) + return -EINVAL; + out_be32(bman_ccsr_map + POOL_SWDET(bpid), + __generate_thresh(thresholds[0], 0)); + out_be32(bman_ccsr_map + POOL_SWDXT(bpid), + __generate_thresh(thresholds[1], 1)); + out_be32(bman_ccsr_map + POOL_HWDET(bpid), + __generate_thresh(thresholds[2], 0)); + out_be32(bman_ccsr_map + POOL_HWDXT(bpid), + __generate_thresh(thresholds[3], 1)); + return 0; +} + +#define BMAN_LOW_DEFAULT_THRESH 0x40 +#define BMAN_HIGH_DEFAULT_THRESH 0x80 +int bm_pool_set_hw_threshold(u32 bpid, const u32 low_thresh, + const u32 high_thresh) +{ + if (!bman_ccsr_map) + return -ENODEV; + if (bpid >= bman_pool_max) + return -EINVAL; + if (low_thresh && high_thresh) { + out_be32(bman_ccsr_map + POOL_HWDET(bpid), + __generate_thresh(low_thresh, 0)); + out_be32(bman_ccsr_map + POOL_HWDXT(bpid), + __generate_thresh(high_thresh, 1)); + } else { + out_be32(bman_ccsr_map + POOL_HWDET(bpid), + __generate_thresh(BMAN_LOW_DEFAULT_THRESH, 0)); + out_be32(bman_ccsr_map + POOL_HWDXT(bpid), + __generate_thresh(BMAN_HIGH_DEFAULT_THRESH, 1)); + } + return 0; +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman_priv.h b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman_priv.h new file mode 100644 index 000000000..5a3e330d9 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/bman_priv.h @@ -0,0 +1,92 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ + +#ifndef __BMAN_PRIV_H +#define __BMAN_PRIV_H + +#include "dpaa_sys.h" +#include <fsl_bman.h> + +/* Revision info (for errata and feature handling) */ +#define BMAN_REV10 0x0100 +#define BMAN_REV20 0x0200 +#define BMAN_REV21 0x0201 + +#define BMAN_PORTAL_IRQ_PATH "/dev/fsl-usdpaa-irq" +#define BMAN_CCSR_MAP "/dev/mem" + +/* This mask contains all the "irqsource" bits visible to API users */ +#define BM_PIRQ_VISIBLE (BM_PIRQ_RCRI | BM_PIRQ_BSCN) + +/* These are bm_<reg>_<verb>(). So for example, bm_disable_write() means "write + * the disable register" rather than "disable the ability to write". + */ +#define bm_isr_status_read(bm) __bm_isr_read(bm, bm_isr_status) +#define bm_isr_status_clear(bm, m) __bm_isr_write(bm, bm_isr_status, m) +#define bm_isr_enable_read(bm) __bm_isr_read(bm, bm_isr_enable) +#define bm_isr_enable_write(bm, v) __bm_isr_write(bm, bm_isr_enable, v) +#define bm_isr_disable_read(bm) __bm_isr_read(bm, bm_isr_disable) +#define bm_isr_disable_write(bm, v) __bm_isr_write(bm, bm_isr_disable, v) +#define bm_isr_inhibit(bm) __bm_isr_write(bm, bm_isr_inhibit, 1) +#define bm_isr_uninhibit(bm) __bm_isr_write(bm, bm_isr_inhibit, 0) + +/* + * Global variables of the max portal/pool number this bman version supported + */ +extern u16 bman_pool_max; + +/* used by CCSR and portal interrupt code */ +enum bm_isr_reg { + bm_isr_status = 0, + bm_isr_enable = 1, + bm_isr_disable = 2, + bm_isr_inhibit = 3 +}; + +struct bm_portal_config { + /* + * Corenet portal addresses; + * [0]==cache-enabled, [1]==cache-inhibited. + */ + void __iomem *addr_virt[2]; + /* Allow these to be joined in lists */ + struct list_head list; + /* User-visible portal configuration settings */ + /* This is used for any "core-affine" portals, ie. default portals + * associated to the corresponding cpu. -1 implies that there is no + * core affinity configured. + */ + int cpu; + /* portal interrupt line */ + int irq; + /* the unique index of this portal */ + u32 index; + /* Is this portal shared? (If so, it has coarser locking and demuxes + * processing on behalf of other CPUs.). + */ + int is_shared; + /* These are the buffer pool IDs that may be used via this portal. */ + struct bman_depletion mask; + +}; + +int bman_init_ccsr(const struct device_node *node); + +struct bman_portal *bman_create_affine_portal( + const struct bm_portal_config *config); +const struct bm_portal_config *bman_destroy_affine_portal(void); + +/* Set depletion thresholds associated with a buffer pool. Requires that the + * operating system have access to Bman CCSR (ie. compiled in support and + * run-time access courtesy of the device-tree). + */ +int bm_pool_set(u32 bpid, const u32 *thresholds); + +/* Read the free buffer count for a given buffer */ +u32 bm_pool_free_buffers(u32 bpid); + +#endif /* __BMAN_PRIV_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_alloc.c b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_alloc.c new file mode 100644 index 000000000..a05803c23 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_alloc.c @@ -0,0 +1,71 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2009-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ + +#include "dpaa_sys.h" +#include <process.h> +#include <fsl_qman.h> +#include <fsl_bman.h> + +int bman_alloc_bpid_range(u32 *result, u32 count, u32 align, int partial) +{ + return process_alloc(dpaa_id_bpid, result, count, align, partial); +} + +void bman_release_bpid_range(u32 bpid, u32 count) +{ + process_release(dpaa_id_bpid, bpid, count); +} + +int bman_reserve_bpid_range(u32 bpid, u32 count) +{ + return process_reserve(dpaa_id_bpid, bpid, count); +} + +int qman_alloc_fqid_range(u32 *result, u32 count, u32 align, int partial) +{ + return process_alloc(dpaa_id_fqid, result, count, align, partial); +} + +void qman_release_fqid_range(u32 fqid, u32 count) +{ + process_release(dpaa_id_fqid, fqid, count); +} + +int qman_reserve_fqid_range(u32 fqid, unsigned int count) +{ + return process_reserve(dpaa_id_fqid, fqid, count); +} + +int qman_alloc_pool_range(u32 *result, u32 count, u32 align, int partial) +{ + return process_alloc(dpaa_id_qpool, result, count, align, partial); +} + +void qman_release_pool_range(u32 pool, u32 count) +{ + process_release(dpaa_id_qpool, pool, count); +} + +int qman_reserve_pool_range(u32 pool, u32 count) +{ + return process_reserve(dpaa_id_qpool, pool, count); +} + +int qman_alloc_cgrid_range(u32 *result, u32 count, u32 align, int partial) +{ + return process_alloc(dpaa_id_cgrid, result, count, align, partial); +} + +void qman_release_cgrid_range(u32 cgrid, u32 count) +{ + process_release(dpaa_id_cgrid, cgrid, count); +} + +int qman_reserve_cgrid_range(u32 cgrid, u32 count) +{ + return process_reserve(dpaa_id_cgrid, cgrid, count); +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_sys.c b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_sys.c new file mode 100644 index 000000000..9d6bfd40a --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_sys.c @@ -0,0 +1,103 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2013-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ + +#include <process.h> +#include "dpaa_sys.h" + +struct process_interrupt { + int irq; + irqreturn_t (*isr)(int irq, void *arg); + unsigned long flags; + const char *name; + void *arg; + struct list_head node; +}; + +static COMPAT_LIST_HEAD(process_irq_list); +static pthread_mutex_t process_irq_lock = PTHREAD_MUTEX_INITIALIZER; + +static void process_interrupt_install(struct process_interrupt *irq) +{ + int ret; + /* Add the irq to the end of the list */ + ret = pthread_mutex_lock(&process_irq_lock); + assert(!ret); + list_add_tail(&irq->node, &process_irq_list); + ret = pthread_mutex_unlock(&process_irq_lock); + assert(!ret); +} + +static void process_interrupt_remove(struct process_interrupt *irq) +{ + int ret; + + ret = pthread_mutex_lock(&process_irq_lock); + assert(!ret); + list_del(&irq->node); + ret = pthread_mutex_unlock(&process_irq_lock); + assert(!ret); +} + +static struct process_interrupt *process_interrupt_find(int irq_num) +{ + int ret; + struct process_interrupt *i = NULL; + + ret = pthread_mutex_lock(&process_irq_lock); + assert(!ret); + list_for_each_entry(i, &process_irq_list, node) { + if (i->irq == irq_num) + goto done; + } +done: + ret = pthread_mutex_unlock(&process_irq_lock); + assert(!ret); + return i; +} + +/* This is the interface from the platform-agnostic driver code to (de)register + * interrupt handlers. We simply create/destroy corresponding structs. + */ +int qbman_request_irq(int irq, irqreturn_t (*isr)(int irq, void *arg), + unsigned long flags, const char *name, + void *arg __maybe_unused) +{ + struct process_interrupt *irq_node = + kmalloc(sizeof(*irq_node), GFP_KERNEL); + + if (!irq_node) + return -ENOMEM; + irq_node->irq = irq; + irq_node->isr = isr; + irq_node->flags = flags; + irq_node->name = name; + irq_node->arg = arg; + process_interrupt_install(irq_node); + return 0; +} + +int qbman_free_irq(int irq, __maybe_unused void *arg) +{ + struct process_interrupt *irq_node = process_interrupt_find(irq); + + if (!irq_node) + return -EINVAL; + process_interrupt_remove(irq_node); + kfree(irq_node); + return 0; +} + +/* This is the interface from the platform-specific driver code to obtain + * interrupt handlers that have been registered. + */ +void qbman_invoke_irq(int irq) +{ + struct process_interrupt *irq_node = process_interrupt_find(irq); + + if (irq_node) + irq_node->isr(irq, irq_node->arg); +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_sys.h b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_sys.h new file mode 100644 index 000000000..9377738df --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/dpaa_sys.h @@ -0,0 +1,29 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ + +#ifndef __DPAA_SYS_H +#define __DPAA_SYS_H + +#include <compat.h> +#include <dpaa_of.h> + +/* For 2-element tables related to cache-inhibited and cache-enabled mappings */ +#define DPAA_PORTAL_CE 0 +#define DPAA_PORTAL_CI 1 + +#define DPAA_ASSERT(x) RTE_ASSERT(x) + +/* This is the interface from the platform-agnostic driver code to (de)register + * interrupt handlers. We simply create/destroy corresponding structs. + */ +int qbman_request_irq(int irq, irqreturn_t (*isr)(int irq, void *arg), + unsigned long flags, const char *name, void *arg); +int qbman_free_irq(int irq, void *arg); + +void qbman_invoke_irq(int irq); + +#endif /* __DPAA_SYS_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/process.c b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/process.c new file mode 100644 index 000000000..2c23c98df --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/process.c @@ -0,0 +1,298 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2011-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ +#include <assert.h> +#include <fcntl.h> +#include <unistd.h> +#include <sys/ioctl.h> + +#include "process.h" + +#include <fsl_usd.h> + +/* As higher-level drivers will be built on top of this (dma_mem, qbman, ...), + * it's preferable that the process driver itself not provide any exported API. + * As such, combined with the fact that none of these operations are + * performance critical, it is justified to use lazy initialisation, so that's + * what the lock is for. + */ +static int fd = -1; +static pthread_mutex_t fd_init_lock = PTHREAD_MUTEX_INITIALIZER; + +static int check_fd(void) +{ + int ret; + + if (fd >= 0) + return 0; + ret = pthread_mutex_lock(&fd_init_lock); + assert(!ret); + /* check again with the lock held */ + if (fd < 0) + fd = open(PROCESS_PATH, O_RDWR); + ret = pthread_mutex_unlock(&fd_init_lock); + assert(!ret); + return (fd >= 0) ? 0 : -ENODEV; +} + +#define DPAA_IOCTL_MAGIC 'u' +struct dpaa_ioctl_id_alloc { + uint32_t base; /* Return value, the start of the allocated range */ + enum dpaa_id_type id_type; /* what kind of resource(s) to allocate */ + uint32_t num; /* how many IDs to allocate (and return value) */ + uint32_t align; /* must be a power of 2, 0 is treated like 1 */ + int partial; /* whether to allow less than 'num' */ +}; + +struct dpaa_ioctl_id_release { + /* Input; */ + enum dpaa_id_type id_type; + uint32_t base; + uint32_t num; +}; + +struct dpaa_ioctl_id_reserve { + enum dpaa_id_type id_type; + uint32_t base; + uint32_t num; +}; + +#define DPAA_IOCTL_ID_ALLOC \ + _IOWR(DPAA_IOCTL_MAGIC, 0x01, struct dpaa_ioctl_id_alloc) +#define DPAA_IOCTL_ID_RELEASE \ + _IOW(DPAA_IOCTL_MAGIC, 0x02, struct dpaa_ioctl_id_release) +#define DPAA_IOCTL_ID_RESERVE \ + _IOW(DPAA_IOCTL_MAGIC, 0x0A, struct dpaa_ioctl_id_reserve) + +int process_alloc(enum dpaa_id_type id_type, uint32_t *base, uint32_t num, + uint32_t align, int partial) +{ + struct dpaa_ioctl_id_alloc id = { + .id_type = id_type, + .num = num, + .align = align, + .partial = partial + }; + int ret = check_fd(); + + if (ret) + return ret; + ret = ioctl(fd, DPAA_IOCTL_ID_ALLOC, &id); + if (ret) + return ret; + for (ret = 0; ret < (int)id.num; ret++) + base[ret] = id.base + ret; + return id.num; +} + +void process_release(enum dpaa_id_type id_type, uint32_t base, uint32_t num) +{ + struct dpaa_ioctl_id_release id = { + .id_type = id_type, + .base = base, + .num = num + }; + int ret = check_fd(); + + if (ret) { + fprintf(stderr, "Process FD failure\n"); + return; + } + ret = ioctl(fd, DPAA_IOCTL_ID_RELEASE, &id); + if (ret) + fprintf(stderr, "Process FD ioctl failure type %d base 0x%x num %d\n", + id_type, base, num); +} + +int process_reserve(enum dpaa_id_type id_type, uint32_t base, uint32_t num) +{ + struct dpaa_ioctl_id_reserve id = { + .id_type = id_type, + .base = base, + .num = num + }; + int ret = check_fd(); + + if (ret) + return ret; + return ioctl(fd, DPAA_IOCTL_ID_RESERVE, &id); +} + +/***************************************/ +/* Mapping and using QMan/BMan portals */ +/***************************************/ + +#define DPAA_IOCTL_PORTAL_MAP \ + _IOWR(DPAA_IOCTL_MAGIC, 0x07, struct dpaa_ioctl_portal_map) +#define DPAA_IOCTL_PORTAL_UNMAP \ + _IOW(DPAA_IOCTL_MAGIC, 0x08, struct dpaa_portal_map) + +int process_portal_map(struct dpaa_ioctl_portal_map *params) +{ + int ret = check_fd(); + + if (ret) + return ret; + + ret = ioctl(fd, DPAA_IOCTL_PORTAL_MAP, params); + if (ret) { + perror("ioctl(DPAA_IOCTL_PORTAL_MAP)"); + return ret; + } + return 0; +} + +int process_portal_unmap(struct dpaa_portal_map *map) +{ + int ret = check_fd(); + + if (ret) + return ret; + + ret = ioctl(fd, DPAA_IOCTL_PORTAL_UNMAP, map); + if (ret) { + perror("ioctl(DPAA_IOCTL_PORTAL_UNMAP)"); + return ret; + } + return 0; +} + +#define DPAA_IOCTL_PORTAL_IRQ_MAP \ + _IOW(DPAA_IOCTL_MAGIC, 0x09, struct dpaa_ioctl_irq_map) + +int process_portal_irq_map(int ifd, struct dpaa_ioctl_irq_map *map) +{ + map->fd = fd; + return ioctl(ifd, DPAA_IOCTL_PORTAL_IRQ_MAP, map); +} + +int process_portal_irq_unmap(int ifd) +{ + return close(ifd); +} + +struct dpaa_ioctl_raw_portal { + /* inputs */ + enum dpaa_portal_type type; /* Type of portal to allocate */ + + uint8_t enable_stash; /* set to non zero to turn on stashing */ + /* Stashing attributes for the portal */ + uint32_t cpu; + uint32_t cache; + uint32_t window; + /* Specifies the stash request queue this portal should use */ + uint8_t sdest; + + /* Specifes a specific portal index to map or QBMAN_ANY_PORTAL_IDX + * for don't care. The portal index will be populated by the + * driver when the ioctl() successfully completes. + */ + uint32_t index; + + /* outputs */ + uint64_t cinh; + uint64_t cena; +}; + +#define DPAA_IOCTL_ALLOC_RAW_PORTAL \ + _IOWR(DPAA_IOCTL_MAGIC, 0x0C, struct dpaa_ioctl_raw_portal) + +#define DPAA_IOCTL_FREE_RAW_PORTAL \ + _IOR(DPAA_IOCTL_MAGIC, 0x0D, struct dpaa_ioctl_raw_portal) + +static int process_portal_allocate(struct dpaa_ioctl_raw_portal *portal) +{ + int ret = check_fd(); + + if (ret) + return ret; + + ret = ioctl(fd, DPAA_IOCTL_ALLOC_RAW_PORTAL, portal); + if (ret) { + perror("ioctl(DPAA_IOCTL_ALLOC_RAW_PORTAL)"); + return ret; + } + return 0; +} + +static int process_portal_free(struct dpaa_ioctl_raw_portal *portal) +{ + int ret = check_fd(); + + if (ret) + return ret; + + ret = ioctl(fd, DPAA_IOCTL_FREE_RAW_PORTAL, portal); + if (ret) { + perror("ioctl(DPAA_IOCTL_FREE_RAW_PORTAL)"); + return ret; + } + return 0; +} + +int qman_allocate_raw_portal(struct dpaa_raw_portal *portal) +{ + struct dpaa_ioctl_raw_portal input; + int ret; + + input.type = dpaa_portal_qman; + input.index = portal->index; + input.enable_stash = portal->enable_stash; + input.cpu = portal->cpu; + input.cache = portal->cache; + input.window = portal->window; + input.sdest = portal->sdest; + + ret = process_portal_allocate(&input); + if (ret) + return ret; + portal->index = input.index; + portal->cinh = input.cinh; + portal->cena = input.cena; + return 0; +} + +int qman_free_raw_portal(struct dpaa_raw_portal *portal) +{ + struct dpaa_ioctl_raw_portal input; + + input.type = dpaa_portal_qman; + input.index = portal->index; + input.cinh = portal->cinh; + input.cena = portal->cena; + + return process_portal_free(&input); +} + +int bman_allocate_raw_portal(struct dpaa_raw_portal *portal) +{ + struct dpaa_ioctl_raw_portal input; + int ret; + + input.type = dpaa_portal_bman; + input.index = portal->index; + input.enable_stash = 0; + + ret = process_portal_allocate(&input); + if (ret) + return ret; + portal->index = input.index; + portal->cinh = input.cinh; + portal->cena = input.cena; + return 0; +} + +int bman_free_raw_portal(struct dpaa_raw_portal *portal) +{ + struct dpaa_ioctl_raw_portal input; + + input.type = dpaa_portal_bman; + input.index = portal->index; + input.cinh = portal->cinh; + input.cena = portal->cena; + + return process_portal_free(&input); +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman.c b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman.c new file mode 100644 index 000000000..b596e79c2 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman.c @@ -0,0 +1,2852 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017,2019 NXP + * + */ + +#include "qman.h" +#include <rte_branch_prediction.h> +#include <rte_dpaa_bus.h> +#include <rte_eventdev.h> +#include <rte_byteorder.h> + +#include <dpaa_bits.h> + +/* Compilation constants */ +#define DQRR_MAXFILL 15 +#define EQCR_ITHRESH 4 /* if EQCR congests, interrupt threshold */ +#define IRQNAME "QMan portal %d" +#define MAX_IRQNAME 16 /* big enough for "QMan portal %d" */ +/* maximum number of DQRR entries to process in qman_poll() */ +#define FSL_QMAN_POLL_LIMIT 8 + +/* Lock/unlock frame queues, subject to the "LOCKED" flag. This is about + * inter-processor locking only. Note, FQLOCK() is always called either under a + * local_irq_save() or from interrupt context - hence there's no need for irq + * protection (and indeed, attempting to nest irq-protection doesn't work, as + * the "irq en/disable" machinery isn't recursive...). + */ +#define FQLOCK(fq) \ + do { \ + struct qman_fq *__fq478 = (fq); \ + if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \ + spin_lock(&__fq478->fqlock); \ + } while (0) +#define FQUNLOCK(fq) \ + do { \ + struct qman_fq *__fq478 = (fq); \ + if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \ + spin_unlock(&__fq478->fqlock); \ + } while (0) + +static inline void fq_set(struct qman_fq *fq, u32 mask) +{ + dpaa_set_bits(mask, &fq->flags); +} + +static inline void fq_clear(struct qman_fq *fq, u32 mask) +{ + dpaa_clear_bits(mask, &fq->flags); +} + +static inline int fq_isset(struct qman_fq *fq, u32 mask) +{ + return fq->flags & mask; +} + +static inline int fq_isclear(struct qman_fq *fq, u32 mask) +{ + return !(fq->flags & mask); +} + +struct qman_portal { + struct qm_portal p; + /* PORTAL_BITS_*** - dynamic, strictly internal */ + unsigned long bits; + /* interrupt sources processed by portal_isr(), configurable */ + unsigned long irq_sources; + u32 use_eqcr_ci_stashing; + u32 slowpoll; /* only used when interrupts are off */ + /* only 1 volatile dequeue at a time */ + struct qman_fq *vdqcr_owned; + u32 sdqcr; + int dqrr_disable_ref; + /* A portal-specific handler for DCP ERNs. If this is NULL, the global + * handler is called instead. + */ + qman_cb_dc_ern cb_dc_ern; + /* When the cpu-affine portal is activated, this is non-NULL */ + const struct qm_portal_config *config; + struct dpa_rbtree retire_table; + char irqname[MAX_IRQNAME]; + /* 2-element array. cgrs[0] is mask, cgrs[1] is snapshot. */ + struct qman_cgrs *cgrs; + /* linked-list of CSCN handlers. */ + struct list_head cgr_cbs; + /* list lock */ + spinlock_t cgr_lock; + /* track if memory was allocated by the driver */ +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + /* Keep a shadow copy of the DQRR on LE systems as the SW needs to + * do byte swaps of DQRR read only memory. First entry must be aligned + * to 2 ** 10 to ensure DQRR index calculations based shadow copy + * address (6 bits for address shift + 4 bits for the DQRR size). + */ + struct qm_dqrr_entry shadow_dqrr[QM_DQRR_SIZE] + __rte_aligned(1024); +#endif +}; + +/* Global handler for DCP ERNs. Used when the portal receiving the message does + * not have a portal-specific handler. + */ +static qman_cb_dc_ern cb_dc_ern; + +static cpumask_t affine_mask; +static DEFINE_SPINLOCK(affine_mask_lock); +static u16 affine_channels[NR_CPUS]; +static RTE_DEFINE_PER_LCORE(struct qman_portal, qman_affine_portal); + +static inline struct qman_portal *get_affine_portal(void) +{ + return &RTE_PER_LCORE(qman_affine_portal); +} + +/* This gives a FQID->FQ lookup to cover the fact that we can't directly demux + * retirement notifications (the fact they are sometimes h/w-consumed means that + * contextB isn't always a s/w demux - and as we can't know which case it is + * when looking at the notification, we have to use the slow lookup for all of + * them). NB, it's possible to have multiple FQ objects refer to the same FQID + * (though at most one of them should be the consumer), so this table isn't for + * all FQs - FQs are added when retirement commands are issued, and removed when + * they complete, which also massively reduces the size of this table. + */ +IMPLEMENT_DPAA_RBTREE(fqtree, struct qman_fq, node, fqid); +/* + * This is what everything can wait on, even if it migrates to a different cpu + * to the one whose affine portal it is waiting on. + */ +static DECLARE_WAIT_QUEUE_HEAD(affine_queue); + +static inline int table_push_fq(struct qman_portal *p, struct qman_fq *fq) +{ + int ret = fqtree_push(&p->retire_table, fq); + + if (ret) + pr_err("ERROR: double FQ-retirement %d\n", fq->fqid); + return ret; +} + +static inline void table_del_fq(struct qman_portal *p, struct qman_fq *fq) +{ + fqtree_del(&p->retire_table, fq); +} + +static inline struct qman_fq *table_find_fq(struct qman_portal *p, u32 fqid) +{ + return fqtree_find(&p->retire_table, fqid); +} + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +static void **qman_fq_lookup_table; +static size_t qman_fq_lookup_table_size; + +int qman_setup_fq_lookup_table(size_t num_entries) +{ + num_entries++; + /* Allocate 1 more entry since the first entry is not used */ + qman_fq_lookup_table = vmalloc((num_entries * sizeof(void *))); + if (!qman_fq_lookup_table) { + pr_err("QMan: Could not allocate fq lookup table\n"); + return -ENOMEM; + } + memset(qman_fq_lookup_table, 0, num_entries * sizeof(void *)); + qman_fq_lookup_table_size = num_entries; + pr_debug("QMan: Allocated lookup table at %p, entry count %lu\n", + qman_fq_lookup_table, + (unsigned long)qman_fq_lookup_table_size); + return 0; +} + +void qman_set_fq_lookup_table(void **fq_table) +{ + qman_fq_lookup_table = fq_table; +} + +/* global structure that maintains fq object mapping */ +static DEFINE_SPINLOCK(fq_hash_table_lock); + +static int find_empty_fq_table_entry(u32 *entry, struct qman_fq *fq) +{ + u32 i; + + spin_lock(&fq_hash_table_lock); + /* Can't use index zero because this has special meaning + * in context_b field. + */ + for (i = 1; i < qman_fq_lookup_table_size; i++) { + if (qman_fq_lookup_table[i] == NULL) { + *entry = i; + qman_fq_lookup_table[i] = fq; + spin_unlock(&fq_hash_table_lock); + return 0; + } + } + spin_unlock(&fq_hash_table_lock); + return -ENOMEM; +} + +static void clear_fq_table_entry(u32 entry) +{ + spin_lock(&fq_hash_table_lock); + DPAA_BUG_ON(entry >= qman_fq_lookup_table_size); + qman_fq_lookup_table[entry] = NULL; + spin_unlock(&fq_hash_table_lock); +} + +static inline struct qman_fq *get_fq_table_entry(u32 entry) +{ + DPAA_BUG_ON(entry >= qman_fq_lookup_table_size); + return qman_fq_lookup_table[entry]; +} +#endif + +static inline void cpu_to_hw_fqd(struct qm_fqd *fqd) +{ + /* Byteswap the FQD to HW format */ + fqd->fq_ctrl = cpu_to_be16(fqd->fq_ctrl); + fqd->dest_wq = cpu_to_be16(fqd->dest_wq); + fqd->ics_cred = cpu_to_be16(fqd->ics_cred); + fqd->context_b = cpu_to_be32(fqd->context_b); + fqd->context_a.opaque = cpu_to_be64(fqd->context_a.opaque); + fqd->opaque_td = cpu_to_be16(fqd->opaque_td); +} + +static inline void hw_fqd_to_cpu(struct qm_fqd *fqd) +{ + /* Byteswap the FQD to CPU format */ + fqd->fq_ctrl = be16_to_cpu(fqd->fq_ctrl); + fqd->dest_wq = be16_to_cpu(fqd->dest_wq); + fqd->ics_cred = be16_to_cpu(fqd->ics_cred); + fqd->context_b = be32_to_cpu(fqd->context_b); + fqd->context_a.opaque = be64_to_cpu(fqd->context_a.opaque); +} + +static inline void cpu_to_hw_fd(struct qm_fd *fd) +{ + fd->addr = cpu_to_be40(fd->addr); + fd->status = cpu_to_be32(fd->status); + fd->opaque = cpu_to_be32(fd->opaque); +} + +static inline void hw_fd_to_cpu(struct qm_fd *fd) +{ + fd->addr = be40_to_cpu(fd->addr); + fd->status = be32_to_cpu(fd->status); + fd->opaque = be32_to_cpu(fd->opaque); +} + +/* In the case that slow- and fast-path handling are both done by qman_poll() + * (ie. because there is no interrupt handling), we ought to balance how often + * we do the fast-path poll versus the slow-path poll. We'll use two decrementer + * sources, so we call the fast poll 'n' times before calling the slow poll + * once. The idle decrementer constant is used when the last slow-poll detected + * no work to do, and the busy decrementer constant when the last slow-poll had + * work to do. + */ +#define SLOW_POLL_IDLE 1000 +#define SLOW_POLL_BUSY 10 +static u32 __poll_portal_slow(struct qman_portal *p, u32 is); +static inline unsigned int __poll_portal_fast(struct qman_portal *p, + unsigned int poll_limit); + +/* Portal interrupt handler */ +static irqreturn_t portal_isr(__always_unused int irq, void *ptr) +{ + struct qman_portal *p = ptr; + /* + * The CSCI/CCSCI source is cleared inside __poll_portal_slow(), because + * it could race against a Query Congestion State command also given + * as part of the handling of this interrupt source. We mustn't + * clear it a second time in this top-level function. + */ + u32 clear = QM_DQAVAIL_MASK | (p->irq_sources & + ~(QM_PIRQ_CSCI | QM_PIRQ_CCSCI)); + u32 is = qm_isr_status_read(&p->p) & p->irq_sources; + /* DQRR-handling if it's interrupt-driven */ + if (is & QM_PIRQ_DQRI) + __poll_portal_fast(p, FSL_QMAN_POLL_LIMIT); + /* Handling of anything else that's interrupt-driven */ + clear |= __poll_portal_slow(p, is); + qm_isr_status_clear(&p->p, clear); + return IRQ_HANDLED; +} + +/* This inner version is used privately by qman_create_affine_portal(), as well + * as by the exported qman_stop_dequeues(). + */ +static inline void qman_stop_dequeues_ex(struct qman_portal *p) +{ + if (!(p->dqrr_disable_ref++)) + qm_dqrr_set_maxfill(&p->p, 0); +} + +static int drain_mr_fqrni(struct qm_portal *p) +{ + const struct qm_mr_entry *msg; +loop: + msg = qm_mr_current(p); + if (!msg) { + /* + * if MR was full and h/w had other FQRNI entries to produce, we + * need to allow it time to produce those entries once the + * existing entries are consumed. A worst-case situation + * (fully-loaded system) means h/w sequencers may have to do 3-4 + * other things before servicing the portal's MR pump, each of + * which (if slow) may take ~50 qman cycles (which is ~200 + * processor cycles). So rounding up and then multiplying this + * worst-case estimate by a factor of 10, just to be + * ultra-paranoid, goes as high as 10,000 cycles. NB, we consume + * one entry at a time, so h/w has an opportunity to produce new + * entries well before the ring has been fully consumed, so + * we're being *really* paranoid here. + */ + u64 now, then = mfatb(); + + do { + now = mfatb(); + } while ((then + 10000) > now); + msg = qm_mr_current(p); + if (!msg) + return 0; + } + if ((msg->ern.verb & QM_MR_VERB_TYPE_MASK) != QM_MR_VERB_FQRNI) { + /* We aren't draining anything but FQRNIs */ + pr_err("Found verb 0x%x in MR\n", msg->ern.verb); + return -1; + } + qm_mr_next(p); + qm_mr_cci_consume(p, 1); + goto loop; +} + +static inline int qm_eqcr_init(struct qm_portal *portal, + enum qm_eqcr_pmode pmode, + unsigned int eq_stash_thresh, + int eq_stash_prio) +{ + /* This use of 'register', as well as all other occurrences, is because + * it has been observed to generate much faster code with gcc than is + * otherwise the case. + */ + register struct qm_eqcr *eqcr = &portal->eqcr; + u32 cfg; + u8 pi; + + eqcr->ring = portal->addr.ce + QM_CL_EQCR; + eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + qm_cl_invalidate(EQCR_CI); + pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1); + eqcr->cursor = eqcr->ring + pi; + eqcr->vbit = (qm_in(EQCR_PI_CINH) & QM_EQCR_SIZE) ? + QM_EQCR_VERB_VBIT : 0; + eqcr->available = QM_EQCR_SIZE - 1 - + qm_cyc_diff(QM_EQCR_SIZE, eqcr->ci, pi); + eqcr->ithresh = qm_in(EQCR_ITR); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + eqcr->busy = 0; + eqcr->pmode = pmode; +#endif + cfg = (qm_in(CFG) & 0x00ffffff) | + (eq_stash_thresh << 28) | /* QCSP_CFG: EST */ + (eq_stash_prio << 26) | /* QCSP_CFG: EP */ + ((pmode & 0x3) << 24); /* QCSP_CFG::EPM */ + qm_out(CFG, cfg); + return 0; +} + +static inline void qm_eqcr_finish(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 pi, ci; + u32 cfg; + + /* + * Disable EQCI stashing because the QMan only + * presents the value it previously stashed to + * maintain coherency. Setting the stash threshold + * to 1 then 0 ensures that QMan has resyncronized + * its internal copy so that the portal is clean + * when it is reinitialized in the future + */ + cfg = (qm_in(CFG) & 0x0fffffff) | + (1 << 28); /* QCSP_CFG: EST */ + qm_out(CFG, cfg); + cfg &= 0x0fffffff; /* stash threshold = 0 */ + qm_out(CFG, cfg); + + pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1); + ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + + /* Refresh EQCR CI cache value */ + qm_cl_invalidate(EQCR_CI); + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(!eqcr->busy); +#endif + if (pi != EQCR_PTR2IDX(eqcr->cursor)) + pr_crit("losing uncommitted EQCR entries\n"); + if (ci != eqcr->ci) + pr_crit("missing existing EQCR completions\n"); + if (eqcr->ci != EQCR_PTR2IDX(eqcr->cursor)) + pr_crit("EQCR destroyed unquiesced\n"); +} + +static inline int qm_dqrr_init(struct qm_portal *portal, + __maybe_unused const struct qm_portal_config *config, + enum qm_dqrr_dmode dmode, + __maybe_unused enum qm_dqrr_pmode pmode, + enum qm_dqrr_cmode cmode, u8 max_fill) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u32 cfg; + + /* Make sure the DQRR will be idle when we enable */ + qm_out(DQRR_SDQCR, 0); + qm_out(DQRR_VDQCR, 0); + qm_out(DQRR_PDQCR, 0); + dqrr->ring = portal->addr.ce + QM_CL_DQRR; + dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->ci = qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->cursor = dqrr->ring + dqrr->ci; + dqrr->fill = qm_cyc_diff(QM_DQRR_SIZE, dqrr->ci, dqrr->pi); + dqrr->vbit = (qm_in(DQRR_PI_CINH) & QM_DQRR_SIZE) ? + QM_DQRR_VERB_VBIT : 0; + dqrr->ithresh = qm_in(DQRR_ITR); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + dqrr->dmode = dmode; + dqrr->pmode = pmode; + dqrr->cmode = cmode; +#endif + /* Invalidate every ring entry before beginning */ + for (cfg = 0; cfg < QM_DQRR_SIZE; cfg++) + dccivac(qm_cl(dqrr->ring, cfg)); + cfg = (qm_in(CFG) & 0xff000f00) | + ((max_fill & (QM_DQRR_SIZE - 1)) << 20) | /* DQRR_MF */ + ((dmode & 1) << 18) | /* DP */ + ((cmode & 3) << 16) | /* DCM */ + 0xa0 | /* RE+SE */ + (0 ? 0x40 : 0) | /* Ignore RP */ + (0 ? 0x10 : 0); /* Ignore SP */ + qm_out(CFG, cfg); + qm_dqrr_set_maxfill(portal, max_fill); + return 0; +} + +static inline void qm_dqrr_finish(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if ((dqrr->cmode != qm_dqrr_cdc) && + (dqrr->ci != DQRR_PTR2IDX(dqrr->cursor))) + pr_crit("Ignoring completed DQRR entries\n"); +#endif +} + +static inline int qm_mr_init(struct qm_portal *portal, + __maybe_unused enum qm_mr_pmode pmode, + enum qm_mr_cmode cmode) +{ + register struct qm_mr *mr = &portal->mr; + u32 cfg; + + mr->ring = portal->addr.ce + QM_CL_MR; + mr->pi = qm_in(MR_PI_CINH) & (QM_MR_SIZE - 1); + mr->ci = qm_in(MR_CI_CINH) & (QM_MR_SIZE - 1); + mr->cursor = mr->ring + mr->ci; + mr->fill = qm_cyc_diff(QM_MR_SIZE, mr->ci, mr->pi); + mr->vbit = (qm_in(MR_PI_CINH) & QM_MR_SIZE) ? QM_MR_VERB_VBIT : 0; + mr->ithresh = qm_in(MR_ITR); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + mr->pmode = pmode; + mr->cmode = cmode; +#endif + cfg = (qm_in(CFG) & 0xfffff0ff) | + ((cmode & 1) << 8); /* QCSP_CFG:MM */ + qm_out(CFG, cfg); + return 0; +} + +static inline void qm_mr_pvb_update(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + const struct qm_mr_entry *res = qm_cl(mr->ring, mr->pi); + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mr->pmode == qm_mr_pvb); +#endif + /* when accessing 'verb', use __raw_readb() to ensure that compiler + * inlining doesn't try to optimise out "excess reads". + */ + if ((__raw_readb(&res->ern.verb) & QM_MR_VERB_VBIT) == mr->vbit) { + mr->pi = (mr->pi + 1) & (QM_MR_SIZE - 1); + if (!mr->pi) + mr->vbit ^= QM_MR_VERB_VBIT; + mr->fill++; + res = MR_INC(res); + } + dcbit_ro(res); +} + +struct qman_portal * +qman_init_portal(struct qman_portal *portal, + const struct qm_portal_config *c, + const struct qman_cgrs *cgrs) +{ + struct qm_portal *p; + char buf[16]; + int ret; + u32 isdr; + + p = &portal->p; + + if (!c) + c = portal->config; + + if (dpaa_svr_family == SVR_LS1043A_FAMILY) + portal->use_eqcr_ci_stashing = 3; + else + portal->use_eqcr_ci_stashing = + ((qman_ip_rev >= QMAN_REV30) ? 1 : 0); + + /* + * prep the low-level portal struct with the mapped addresses from the + * config, everything that follows depends on it and "config" is more + * for (de)reference + */ + p->addr.ce = c->addr_virt[DPAA_PORTAL_CE]; + p->addr.ci = c->addr_virt[DPAA_PORTAL_CI]; + /* + * If CI-stashing is used, the current defaults use a threshold of 3, + * and stash with high-than-DQRR priority. + */ + if (qm_eqcr_init(p, qm_eqcr_pvb, + portal->use_eqcr_ci_stashing, 1)) { + pr_err("Qman EQCR initialisation failed\n"); + goto fail_eqcr; + } + if (qm_dqrr_init(p, c, qm_dqrr_dpush, qm_dqrr_pvb, + qm_dqrr_cdc, DQRR_MAXFILL)) { + pr_err("Qman DQRR initialisation failed\n"); + goto fail_dqrr; + } + if (qm_mr_init(p, qm_mr_pvb, qm_mr_cci)) { + pr_err("Qman MR initialisation failed\n"); + goto fail_mr; + } + if (qm_mc_init(p)) { + pr_err("Qman MC initialisation failed\n"); + goto fail_mc; + } + + /* static interrupt-gating controls */ + qm_dqrr_set_ithresh(p, 0); + qm_mr_set_ithresh(p, 0); + qm_isr_set_iperiod(p, 0); + portal->cgrs = kmalloc(2 * sizeof(*cgrs), GFP_KERNEL); + if (!portal->cgrs) + goto fail_cgrs; + /* initial snapshot is no-depletion */ + qman_cgrs_init(&portal->cgrs[1]); + if (cgrs) + portal->cgrs[0] = *cgrs; + else + /* if the given mask is NULL, assume all CGRs can be seen */ + qman_cgrs_fill(&portal->cgrs[0]); + INIT_LIST_HEAD(&portal->cgr_cbs); + spin_lock_init(&portal->cgr_lock); + portal->bits = 0; + portal->slowpoll = 0; + portal->sdqcr = QM_SDQCR_SOURCE_CHANNELS | QM_SDQCR_COUNT_UPTO3 | + QM_SDQCR_DEDICATED_PRECEDENCE | QM_SDQCR_TYPE_PRIO_QOS | + QM_SDQCR_TOKEN_SET(0xab) | QM_SDQCR_CHANNELS_DEDICATED; + portal->dqrr_disable_ref = 0; + portal->cb_dc_ern = NULL; + sprintf(buf, "qportal-%d", c->channel); + dpa_rbtree_init(&portal->retire_table); + isdr = 0xffffffff; + qm_isr_disable_write(p, isdr); + portal->irq_sources = 0; + qm_isr_enable_write(p, portal->irq_sources); + qm_isr_status_clear(p, 0xffffffff); + snprintf(portal->irqname, MAX_IRQNAME, IRQNAME, c->cpu); + if (request_irq(c->irq, portal_isr, 0, portal->irqname, + portal)) { + pr_err("request_irq() failed\n"); + goto fail_irq; + } + + /* Need EQCR to be empty before continuing */ + isdr &= ~QM_PIRQ_EQCI; + qm_isr_disable_write(p, isdr); + ret = qm_eqcr_get_fill(p); + if (ret) { + pr_err("Qman EQCR unclean\n"); + goto fail_eqcr_empty; + } + isdr &= ~(QM_PIRQ_DQRI | QM_PIRQ_MRI); + qm_isr_disable_write(p, isdr); + if (qm_dqrr_current(p)) { + pr_err("Qman DQRR unclean\n"); + qm_dqrr_cdc_consume_n(p, 0xffff); + } + if (qm_mr_current(p) && drain_mr_fqrni(p)) { + /* special handling, drain just in case it's a few FQRNIs */ + if (drain_mr_fqrni(p)) + goto fail_dqrr_mr_empty; + } + /* Success */ + portal->config = c; + qm_isr_disable_write(p, 0); + qm_isr_uninhibit(p); + /* Write a sane SDQCR */ + qm_dqrr_sdqcr_set(p, portal->sdqcr); + return portal; +fail_dqrr_mr_empty: +fail_eqcr_empty: + free_irq(c->irq, portal); +fail_irq: + kfree(portal->cgrs); + spin_lock_destroy(&portal->cgr_lock); +fail_cgrs: + qm_mc_finish(p); +fail_mc: + qm_mr_finish(p); +fail_mr: + qm_dqrr_finish(p); +fail_dqrr: + qm_eqcr_finish(p); +fail_eqcr: + return NULL; +} + +#define MAX_GLOBAL_PORTALS 8 +static struct qman_portal global_portals[MAX_GLOBAL_PORTALS]; +static rte_atomic16_t global_portals_used[MAX_GLOBAL_PORTALS]; + +struct qman_portal * +qman_alloc_global_portal(struct qm_portal_config *q_pcfg) +{ + unsigned int i; + + for (i = 0; i < MAX_GLOBAL_PORTALS; i++) { + if (rte_atomic16_test_and_set(&global_portals_used[i])) { + global_portals[i].config = q_pcfg; + return &global_portals[i]; + } + } + pr_err("No portal available (%x)\n", MAX_GLOBAL_PORTALS); + + return NULL; +} + +int +qman_free_global_portal(struct qman_portal *portal) +{ + unsigned int i; + + for (i = 0; i < MAX_GLOBAL_PORTALS; i++) { + if (&global_portals[i] == portal) { + rte_atomic16_clear(&global_portals_used[i]); + return 0; + } + } + return -1; +} + +void +qman_portal_uninhibit_isr(struct qman_portal *portal) +{ + qm_isr_uninhibit(&portal->p); +} + +struct qman_portal *qman_create_affine_portal(const struct qm_portal_config *c, + const struct qman_cgrs *cgrs) +{ + struct qman_portal *res; + struct qman_portal *portal = get_affine_portal(); + + /* A criteria for calling this function (from qman_driver.c) is that + * we're already affine to the cpu and won't schedule onto another cpu. + */ + res = qman_init_portal(portal, c, cgrs); + if (res) { + spin_lock(&affine_mask_lock); + CPU_SET(c->cpu, &affine_mask); + affine_channels[c->cpu] = + c->channel; + spin_unlock(&affine_mask_lock); + } + return res; +} + +static inline +void qman_destroy_portal(struct qman_portal *qm) +{ + const struct qm_portal_config *pcfg; + + /* Stop dequeues on the portal */ + qm_dqrr_sdqcr_set(&qm->p, 0); + + /* + * NB we do this to "quiesce" EQCR. If we add enqueue-completions or + * something related to QM_PIRQ_EQCI, this may need fixing. + * Also, due to the prefetching model used for CI updates in the enqueue + * path, this update will only invalidate the CI cacheline *after* + * working on it, so we need to call this twice to ensure a full update + * irrespective of where the enqueue processing was at when the teardown + * began. + */ + qm_eqcr_cce_update(&qm->p); + qm_eqcr_cce_update(&qm->p); + pcfg = qm->config; + + free_irq(pcfg->irq, qm); + + kfree(qm->cgrs); + qm_mc_finish(&qm->p); + qm_mr_finish(&qm->p); + qm_dqrr_finish(&qm->p); + qm_eqcr_finish(&qm->p); + + qm->config = NULL; + + spin_lock_destroy(&qm->cgr_lock); +} + +const struct qm_portal_config * +qman_destroy_affine_portal(struct qman_portal *qp) +{ + /* We don't want to redirect if we're a slave, use "raw" */ + struct qman_portal *qm; + const struct qm_portal_config *pcfg; + int cpu; + + if (qp == NULL) + qm = get_affine_portal(); + else + qm = qp; + pcfg = qm->config; + cpu = pcfg->cpu; + + qman_destroy_portal(qm); + + spin_lock(&affine_mask_lock); + CPU_CLR(cpu, &affine_mask); + spin_unlock(&affine_mask_lock); + + qman_free_global_portal(qm); + + return pcfg; +} + +int qman_get_portal_index(void) +{ + struct qman_portal *p = get_affine_portal(); + return p->config->index; +} + +/* Inline helper to reduce nesting in __poll_portal_slow() */ +static inline void fq_state_change(struct qman_portal *p, struct qman_fq *fq, + const struct qm_mr_entry *msg, u8 verb) +{ + FQLOCK(fq); + switch (verb) { + case QM_MR_VERB_FQRL: + DPAA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_ORL)); + fq_clear(fq, QMAN_FQ_STATE_ORL); + table_del_fq(p, fq); + break; + case QM_MR_VERB_FQRN: + DPAA_ASSERT((fq->state == qman_fq_state_parked) || + (fq->state == qman_fq_state_sched)); + DPAA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_CHANGING)); + fq_clear(fq, QMAN_FQ_STATE_CHANGING); + if (msg->fq.fqs & QM_MR_FQS_NOTEMPTY) + fq_set(fq, QMAN_FQ_STATE_NE); + if (msg->fq.fqs & QM_MR_FQS_ORLPRESENT) + fq_set(fq, QMAN_FQ_STATE_ORL); + else + table_del_fq(p, fq); + fq->state = qman_fq_state_retired; + break; + case QM_MR_VERB_FQPN: + DPAA_ASSERT(fq->state == qman_fq_state_sched); + DPAA_ASSERT(fq_isclear(fq, QMAN_FQ_STATE_CHANGING)); + fq->state = qman_fq_state_parked; + } + FQUNLOCK(fq); +} + +static u32 __poll_portal_slow(struct qman_portal *p, u32 is) +{ + const struct qm_mr_entry *msg; + struct qm_mr_entry swapped_msg; + + if (is & QM_PIRQ_CSCI) { + struct qman_cgrs rr, c; + struct qm_mc_result *mcr; + struct qman_cgr *cgr; + + spin_lock(&p->cgr_lock); + /* + * The CSCI bit must be cleared _before_ issuing the + * Query Congestion State command, to ensure that a long + * CGR State Change callback cannot miss an intervening + * state change. + */ + qm_isr_status_clear(&p->p, QM_PIRQ_CSCI); + qm_mc_start(&p->p); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + /* mask out the ones I'm not interested in */ + qman_cgrs_and(&rr, (const struct qman_cgrs *) + &mcr->querycongestion.state, &p->cgrs[0]); + /* check previous snapshot for delta, enter/exit congestion */ + qman_cgrs_xor(&c, &rr, &p->cgrs[1]); + /* update snapshot */ + qman_cgrs_cp(&p->cgrs[1], &rr); + /* Invoke callback */ + list_for_each_entry(cgr, &p->cgr_cbs, node) + if (cgr->cb && qman_cgrs_get(&c, cgr->cgrid)) + cgr->cb(p, cgr, qman_cgrs_get(&rr, cgr->cgrid)); + spin_unlock(&p->cgr_lock); + } + + if (is & QM_PIRQ_EQRI) { + qm_eqcr_cce_update(&p->p); + qm_eqcr_set_ithresh(&p->p, 0); + wake_up(&affine_queue); + } + + if (is & QM_PIRQ_MRI) { + struct qman_fq *fq; + u8 verb, num = 0; +mr_loop: + qm_mr_pvb_update(&p->p); + msg = qm_mr_current(&p->p); + if (!msg) + goto mr_done; + swapped_msg = *msg; + hw_fd_to_cpu(&swapped_msg.ern.fd); + verb = msg->ern.verb & QM_MR_VERB_TYPE_MASK; + /* The message is a software ERN iff the 0x20 bit is set */ + if (verb & 0x20) { + switch (verb) { + case QM_MR_VERB_FQRNI: + /* nada, we drop FQRNIs on the floor */ + break; + case QM_MR_VERB_FQRN: + case QM_MR_VERB_FQRL: + /* Lookup in the retirement table */ + fq = table_find_fq(p, + be32_to_cpu(msg->fq.fqid)); + DPAA_BUG_ON(!fq); + fq_state_change(p, fq, &swapped_msg, verb); + if (fq->cb.fqs) + fq->cb.fqs(p, fq, &swapped_msg); + break; + case QM_MR_VERB_FQPN: + /* Parked */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(msg->fq.contextB); +#else + fq = (void *)(uintptr_t)msg->fq.contextB; +#endif + fq_state_change(p, fq, msg, verb); + if (fq->cb.fqs) + fq->cb.fqs(p, fq, &swapped_msg); + break; + case QM_MR_VERB_DC_ERN: + /* DCP ERN */ + if (p->cb_dc_ern) + p->cb_dc_ern(p, msg); + else if (cb_dc_ern) + cb_dc_ern(p, msg); + else { + static int warn_once; + + if (!warn_once) { + pr_crit("Leaking DCP ERNs!\n"); + warn_once = 1; + } + } + break; + default: + pr_crit("Invalid MR verb 0x%02x\n", verb); + } + } else { + /* Its a software ERN */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(be32_to_cpu(msg->ern.tag)); +#else + fq = (void *)(uintptr_t)be32_to_cpu(msg->ern.tag); +#endif + fq->cb.ern(p, fq, &swapped_msg); + } + num++; + qm_mr_next(&p->p); + goto mr_loop; +mr_done: + qm_mr_cci_consume(&p->p, num); + } + /* + * QM_PIRQ_CSCI/CCSCI has already been cleared, as part of its specific + * processing. If that interrupt source has meanwhile been re-asserted, + * we mustn't clear it here (or in the top-level interrupt handler). + */ + return is & (QM_PIRQ_EQCI | QM_PIRQ_EQRI | QM_PIRQ_MRI); +} + +/* + * remove some slowish-path stuff from the "fast path" and make sure it isn't + * inlined. + */ +static noinline void clear_vdqcr(struct qman_portal *p, struct qman_fq *fq) +{ + p->vdqcr_owned = NULL; + FQLOCK(fq); + fq_clear(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + wake_up(&affine_queue); +} + +/* + * The only states that would conflict with other things if they ran at the + * same time on the same cpu are: + * + * (i) setting/clearing vdqcr_owned, and + * (ii) clearing the NE (Not Empty) flag. + * + * Both are safe. Because; + * + * (i) this clearing can only occur after qman_set_vdq() has set the + * vdqcr_owned field (which it does before setting VDQCR), and + * qman_volatile_dequeue() blocks interrupts and preemption while this is + * done so that we can't interfere. + * (ii) the NE flag is only cleared after qman_retire_fq() has set it, and as + * with (i) that API prevents us from interfering until it's safe. + * + * The good thing is that qman_set_vdq() and qman_retire_fq() run far + * less frequently (ie. per-FQ) than __poll_portal_fast() does, so the nett + * advantage comes from this function not having to "lock" anything at all. + * + * Note also that the callbacks are invoked at points which are safe against the + * above potential conflicts, but that this function itself is not re-entrant + * (this is because the function tracks one end of each FIFO in the portal and + * we do *not* want to lock that). So the consequence is that it is safe for + * user callbacks to call into any QMan API. + */ +static inline unsigned int __poll_portal_fast(struct qman_portal *p, + unsigned int poll_limit) +{ + const struct qm_dqrr_entry *dq; + struct qman_fq *fq; + enum qman_cb_dqrr_result res; + unsigned int limit = 0; +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + struct qm_dqrr_entry *shadow; +#endif + do { + qm_dqrr_pvb_update(&p->p); + dq = qm_dqrr_current(&p->p); + if (unlikely(!dq)) + break; +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + /* If running on an LE system the fields of the + * dequeue entry must be swapper. Because the + * QMan HW will ignore writes the DQRR entry is + * copied and the index stored within the copy + */ + shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)]; + *shadow = *dq; + dq = shadow; + shadow->fqid = be32_to_cpu(shadow->fqid); + shadow->seqnum = be16_to_cpu(shadow->seqnum); + hw_fd_to_cpu(&shadow->fd); +#endif + + if (dq->stat & QM_DQRR_STAT_UNSCHEDULED) { + /* + * VDQCR: don't trust context_b as the FQ may have + * been configured for h/w consumption and we're + * draining it post-retirement. + */ + fq = p->vdqcr_owned; + /* + * We only set QMAN_FQ_STATE_NE when retiring, so we + * only need to check for clearing it when doing + * volatile dequeues. It's one less thing to check + * in the critical path (SDQCR). + */ + if (dq->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_clear(fq, QMAN_FQ_STATE_NE); + /* + * This is duplicated from the SDQCR code, but we + * have stuff to do before *and* after this callback, + * and we don't want multiple if()s in the critical + * path (SDQCR). + */ + res = fq->cb.dqrr(p, fq, dq); + if (res == qman_cb_dqrr_stop) + break; + /* Check for VDQCR completion */ + if (dq->stat & QM_DQRR_STAT_DQCR_EXPIRED) + clear_vdqcr(p, fq); + } else { + /* SDQCR: context_b points to the FQ */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(dq->contextB); +#else + fq = (void *)(uintptr_t)dq->contextB; +#endif + /* Now let the callback do its stuff */ + res = fq->cb.dqrr(p, fq, dq); + /* + * The callback can request that we exit without + * consuming this entry nor advancing; + */ + if (res == qman_cb_dqrr_stop) + break; + } + /* Interpret 'dq' from a driver perspective. */ + /* + * Parking isn't possible unless HELDACTIVE was set. NB, + * FORCEELIGIBLE implies HELDACTIVE, so we only need to + * check for HELDACTIVE to cover both. + */ + DPAA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) || + (res != qman_cb_dqrr_park)); + /* just means "skip it, I'll consume it myself later on" */ + if (res != qman_cb_dqrr_defer) + qm_dqrr_cdc_consume_1ptr(&p->p, dq, + res == qman_cb_dqrr_park); + /* Move forward */ + qm_dqrr_next(&p->p); + /* + * Entry processed and consumed, increment our counter. The + * callback can request that we exit after consuming the + * entry, and we also exit if we reach our processing limit, + * so loop back only if neither of these conditions is met. + */ + } while (++limit < poll_limit && res != qman_cb_dqrr_consume_stop); + + return limit; +} + +int qman_irqsource_add(u32 bits) +{ + struct qman_portal *p = get_affine_portal(); + + bits = bits & QM_PIRQ_VISIBLE; + + /* Clear any previously remaining interrupt conditions in + * QCSP_ISR. This prevents raising a false interrupt when + * interrupt conditions are enabled in QCSP_IER. + */ + qm_isr_status_clear(&p->p, bits); + dpaa_set_bits(bits, &p->irq_sources); + qm_isr_enable_write(&p->p, p->irq_sources); + + return 0; +} + +int qman_fq_portal_irqsource_add(struct qman_portal *p, u32 bits) +{ + bits = bits & QM_PIRQ_VISIBLE; + + /* Clear any previously remaining interrupt conditions in + * QCSP_ISR. This prevents raising a false interrupt when + * interrupt conditions are enabled in QCSP_IER. + */ + qm_isr_status_clear(&p->p, bits); + dpaa_set_bits(bits, &p->irq_sources); + qm_isr_enable_write(&p->p, p->irq_sources); + + return 0; +} + +int qman_irqsource_remove(u32 bits) +{ + struct qman_portal *p = get_affine_portal(); + u32 ier; + + /* Our interrupt handler only processes+clears status register bits that + * are in p->irq_sources. As we're trimming that mask, if one of them + * were to assert in the status register just before we remove it from + * the enable register, there would be an interrupt-storm when we + * release the IRQ lock. So we wait for the enable register update to + * take effect in h/w (by reading it back) and then clear all other bits + * in the status register. Ie. we clear them from ISR once it's certain + * IER won't allow them to reassert. + */ + + bits &= QM_PIRQ_VISIBLE; + dpaa_clear_bits(bits, &p->irq_sources); + qm_isr_enable_write(&p->p, p->irq_sources); + ier = qm_isr_enable_read(&p->p); + /* Using "~ier" (rather than "bits" or "~p->irq_sources") creates a + * data-dependency, ie. to protect against re-ordering. + */ + qm_isr_status_clear(&p->p, ~ier); + return 0; +} + +int qman_fq_portal_irqsource_remove(struct qman_portal *p, u32 bits) +{ + u32 ier; + + /* Our interrupt handler only processes+clears status register bits that + * are in p->irq_sources. As we're trimming that mask, if one of them + * were to assert in the status register just before we remove it from + * the enable register, there would be an interrupt-storm when we + * release the IRQ lock. So we wait for the enable register update to + * take effect in h/w (by reading it back) and then clear all other bits + * in the status register. Ie. we clear them from ISR once it's certain + * IER won't allow them to reassert. + */ + + bits &= QM_PIRQ_VISIBLE; + dpaa_clear_bits(bits, &p->irq_sources); + qm_isr_enable_write(&p->p, p->irq_sources); + ier = qm_isr_enable_read(&p->p); + /* Using "~ier" (rather than "bits" or "~p->irq_sources") creates a + * data-dependency, ie. to protect against re-ordering. + */ + qm_isr_status_clear(&p->p, ~ier); + return 0; +} + +u16 qman_affine_channel(int cpu) +{ + if (cpu < 0) { + struct qman_portal *portal = get_affine_portal(); + + cpu = portal->config->cpu; + } + DPAA_BUG_ON(!CPU_ISSET(cpu, &affine_mask)); + return affine_channels[cpu]; +} + +unsigned int qman_portal_poll_rx(unsigned int poll_limit, + void **bufs, + struct qman_portal *p) +{ + struct qm_portal *portal = &p->p; + register struct qm_dqrr *dqrr = &portal->dqrr; + struct qm_dqrr_entry *dq[QM_DQRR_SIZE], *shadow[QM_DQRR_SIZE]; + struct qman_fq *fq; + unsigned int limit = 0, rx_number = 0; + uint32_t consume = 0; + + do { + qm_dqrr_pvb_update(&p->p); + if (!dqrr->fill) + break; + + dq[rx_number] = dqrr->cursor; + dqrr->cursor = DQRR_CARRYCLEAR(dqrr->cursor + 1); + /* Prefetch the next DQRR entry */ + rte_prefetch0(dqrr->cursor); + +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + /* If running on an LE system the fields of the + * dequeue entry must be swapper. Because the + * QMan HW will ignore writes the DQRR entry is + * copied and the index stored within the copy + */ + shadow[rx_number] = + &p->shadow_dqrr[DQRR_PTR2IDX(dq[rx_number])]; + shadow[rx_number]->fd.opaque_addr = + dq[rx_number]->fd.opaque_addr; + shadow[rx_number]->fd.addr = + be40_to_cpu(dq[rx_number]->fd.addr); + shadow[rx_number]->fd.opaque = + be32_to_cpu(dq[rx_number]->fd.opaque); +#else + shadow[rx_number] = dq[rx_number]; +#endif + + /* SDQCR: context_b points to the FQ */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = qman_fq_lookup_table[dq[rx_number]->contextB]; +#else + fq = (void *)dq[rx_number]->contextB; +#endif + if (fq->cb.dqrr_prepare) + fq->cb.dqrr_prepare(shadow[rx_number], + &bufs[rx_number]); + + consume |= (1 << (31 - DQRR_PTR2IDX(shadow[rx_number]))); + rx_number++; + --dqrr->fill; + } while (++limit < poll_limit); + + if (rx_number) + fq->cb.dqrr_dpdk_pull_cb(&fq, shadow, bufs, rx_number); + + /* Consume all the DQRR enries together */ + qm_out(DQRR_DCAP, (1 << 8) | consume); + + return rx_number; +} + +void qman_clear_irq(void) +{ + struct qman_portal *p = get_affine_portal(); + u32 clear = QM_DQAVAIL_MASK | (p->irq_sources & + ~(QM_PIRQ_CSCI | QM_PIRQ_CCSCI)); + qm_isr_status_clear(&p->p, clear); +} + +u32 qman_portal_dequeue(struct rte_event ev[], unsigned int poll_limit, + void **bufs) +{ + const struct qm_dqrr_entry *dq; + struct qman_fq *fq; + enum qman_cb_dqrr_result res; + unsigned int limit = 0; + struct qman_portal *p = get_affine_portal(); +#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN + struct qm_dqrr_entry *shadow; +#endif + unsigned int rx_number = 0; + + do { + qm_dqrr_pvb_update(&p->p); + dq = qm_dqrr_current(&p->p); + if (!dq) + break; +#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN + /* + * If running on an LE system the fields of the + * dequeue entry must be swapper. Because the + * QMan HW will ignore writes the DQRR entry is + * copied and the index stored within the copy + */ + shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)]; + *shadow = *dq; + dq = shadow; + shadow->fqid = be32_to_cpu(shadow->fqid); + shadow->seqnum = be16_to_cpu(shadow->seqnum); + hw_fd_to_cpu(&shadow->fd); +#endif + + /* SDQCR: context_b points to the FQ */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(dq->contextB); +#else + fq = (void *)(uintptr_t)dq->contextB; +#endif + /* Now let the callback do its stuff */ + res = fq->cb.dqrr_dpdk_cb(&ev[rx_number], p, fq, + dq, &bufs[rx_number]); + rx_number++; + /* Interpret 'dq' from a driver perspective. */ + /* + * Parking isn't possible unless HELDACTIVE was set. NB, + * FORCEELIGIBLE implies HELDACTIVE, so we only need to + * check for HELDACTIVE to cover both. + */ + DPAA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) || + (res != qman_cb_dqrr_park)); + if (res != qman_cb_dqrr_defer) + qm_dqrr_cdc_consume_1ptr(&p->p, dq, + res == qman_cb_dqrr_park); + /* Move forward */ + qm_dqrr_next(&p->p); + /* + * Entry processed and consumed, increment our counter. The + * callback can request that we exit after consuming the + * entry, and we also exit if we reach our processing limit, + * so loop back only if neither of these conditions is met. + */ + } while (++limit < poll_limit); + + return limit; +} + +struct qm_dqrr_entry *qman_dequeue(struct qman_fq *fq) +{ + struct qman_portal *p = get_affine_portal(); + const struct qm_dqrr_entry *dq; +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + struct qm_dqrr_entry *shadow; +#endif + + qm_dqrr_pvb_update(&p->p); + dq = qm_dqrr_current(&p->p); + if (!dq) + return NULL; + + if (!(dq->stat & QM_DQRR_STAT_FD_VALID)) { + /* Invalid DQRR - put the portal and consume the DQRR. + * Return NULL to user as no packet is seen. + */ + qman_dqrr_consume(fq, (struct qm_dqrr_entry *)dq); + return NULL; + } + +#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)]; + *shadow = *dq; + dq = shadow; + shadow->fqid = be32_to_cpu(shadow->fqid); + shadow->seqnum = be16_to_cpu(shadow->seqnum); + hw_fd_to_cpu(&shadow->fd); +#endif + + if (dq->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_clear(fq, QMAN_FQ_STATE_NE); + + return (struct qm_dqrr_entry *)dq; +} + +void qman_dqrr_consume(struct qman_fq *fq, + struct qm_dqrr_entry *dq) +{ + struct qman_portal *p = get_affine_portal(); + + if (dq->stat & QM_DQRR_STAT_DQCR_EXPIRED) + clear_vdqcr(p, fq); + + qm_dqrr_cdc_consume_1ptr(&p->p, dq, 0); + qm_dqrr_next(&p->p); +} + +int qman_poll_dqrr(unsigned int limit) +{ + struct qman_portal *p = get_affine_portal(); + int ret; + + ret = __poll_portal_fast(p, limit); + return ret; +} + +void qman_poll(void) +{ + struct qman_portal *p = get_affine_portal(); + + if ((~p->irq_sources) & QM_PIRQ_SLOW) { + if (!(p->slowpoll--)) { + u32 is = qm_isr_status_read(&p->p) & ~p->irq_sources; + u32 active = __poll_portal_slow(p, is); + + if (active) { + qm_isr_status_clear(&p->p, active); + p->slowpoll = SLOW_POLL_BUSY; + } else + p->slowpoll = SLOW_POLL_IDLE; + } + } + if ((~p->irq_sources) & QM_PIRQ_DQRI) + __poll_portal_fast(p, FSL_QMAN_POLL_LIMIT); +} + +void qman_stop_dequeues(void) +{ + struct qman_portal *p = get_affine_portal(); + + qman_stop_dequeues_ex(p); +} + +void qman_start_dequeues(void) +{ + struct qman_portal *p = get_affine_portal(); + + DPAA_ASSERT(p->dqrr_disable_ref > 0); + if (!(--p->dqrr_disable_ref)) + qm_dqrr_set_maxfill(&p->p, DQRR_MAXFILL); +} + +void qman_static_dequeue_add(u32 pools, struct qman_portal *qp) +{ + struct qman_portal *p = qp ? qp : get_affine_portal(); + + pools &= p->config->pools; + p->sdqcr |= pools; + qm_dqrr_sdqcr_set(&p->p, p->sdqcr); +} + +void qman_static_dequeue_del(u32 pools, struct qman_portal *qp) +{ + struct qman_portal *p = qp ? qp : get_affine_portal(); + + pools &= p->config->pools; + p->sdqcr &= ~pools; + qm_dqrr_sdqcr_set(&p->p, p->sdqcr); +} + +u32 qman_static_dequeue_get(struct qman_portal *qp) +{ + struct qman_portal *p = qp ? qp : get_affine_portal(); + return p->sdqcr; +} + +void qman_dca(const struct qm_dqrr_entry *dq, int park_request) +{ + struct qman_portal *p = get_affine_portal(); + + qm_dqrr_cdc_consume_1ptr(&p->p, dq, park_request); +} + +void qman_dca_index(u8 index, int park_request) +{ + struct qman_portal *p = get_affine_portal(); + + qm_dqrr_cdc_consume_1(&p->p, index, park_request); +} + +/* Frame queue API */ +static const char *mcr_result_str(u8 result) +{ + switch (result) { + case QM_MCR_RESULT_NULL: + return "QM_MCR_RESULT_NULL"; + case QM_MCR_RESULT_OK: + return "QM_MCR_RESULT_OK"; + case QM_MCR_RESULT_ERR_FQID: + return "QM_MCR_RESULT_ERR_FQID"; + case QM_MCR_RESULT_ERR_FQSTATE: + return "QM_MCR_RESULT_ERR_FQSTATE"; + case QM_MCR_RESULT_ERR_NOTEMPTY: + return "QM_MCR_RESULT_ERR_NOTEMPTY"; + case QM_MCR_RESULT_PENDING: + return "QM_MCR_RESULT_PENDING"; + case QM_MCR_RESULT_ERR_BADCOMMAND: + return "QM_MCR_RESULT_ERR_BADCOMMAND"; + } + return "<unknown MCR result>"; +} + +int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq) +{ + struct qm_fqd fqd; + struct qm_mcr_queryfq_np np; + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) { + int ret = qman_alloc_fqid(&fqid); + + if (ret) + return ret; + } + spin_lock_init(&fq->fqlock); + fq->fqid = fqid; + fq->fqid_le = cpu_to_be32(fqid); + fq->flags = flags; + fq->state = qman_fq_state_oos; + fq->cgr_groupid = 0; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + if (unlikely(find_empty_fq_table_entry(&fq->key, fq))) { + pr_info("Find empty table entry failed\n"); + return -ENOMEM; + } + fq->qman_fq_lookup_table = qman_fq_lookup_table; +#endif + if (!(flags & QMAN_FQ_FLAG_AS_IS) || (flags & QMAN_FQ_FLAG_NO_MODIFY)) + return 0; + /* Everything else is AS_IS support */ + p = get_affine_portal(); + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("QUERYFQ failed: %s\n", mcr_result_str(mcr->result)); + goto err; + } + fqd = mcr->queryfq.fqd; + hw_fqd_to_cpu(&fqd); + mcc = qm_mc_start(&p->p); + mcc->queryfq_np.fqid = cpu_to_be32(fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ_NP); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("QUERYFQ_NP failed: %s\n", mcr_result_str(mcr->result)); + goto err; + } + np = mcr->queryfq_np; + /* Phew, have queryfq and queryfq_np results, stitch together + * the FQ object from those. + */ + fq->cgr_groupid = fqd.cgid; + switch (np.state & QM_MCR_NP_STATE_MASK) { + case QM_MCR_NP_STATE_OOS: + break; + case QM_MCR_NP_STATE_RETIRED: + fq->state = qman_fq_state_retired; + if (np.frm_cnt) + fq_set(fq, QMAN_FQ_STATE_NE); + break; + case QM_MCR_NP_STATE_TEN_SCHED: + case QM_MCR_NP_STATE_TRU_SCHED: + case QM_MCR_NP_STATE_ACTIVE: + fq->state = qman_fq_state_sched; + if (np.state & QM_MCR_NP_STATE_R) + fq_set(fq, QMAN_FQ_STATE_CHANGING); + break; + case QM_MCR_NP_STATE_PARKED: + fq->state = qman_fq_state_parked; + break; + default: + DPAA_ASSERT(NULL == "invalid FQ state"); + } + if (fqd.fq_ctrl & QM_FQCTRL_CGE) + fq->state |= QMAN_FQ_STATE_CGR_EN; + return 0; +err: + if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) + qman_release_fqid(fqid); + return -EIO; +} + +void qman_destroy_fq(struct qman_fq *fq, u32 flags __maybe_unused) +{ + /* + * We don't need to lock the FQ as it is a pre-condition that the FQ be + * quiesced. Instead, run some checks. + */ + switch (fq->state) { + case qman_fq_state_parked: + DPAA_ASSERT(flags & QMAN_FQ_DESTROY_PARKED); + /* Fallthrough */ + case qman_fq_state_oos: + if (fq_isset(fq, QMAN_FQ_FLAG_DYNAMIC_FQID)) + qman_release_fqid(fq->fqid); +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + clear_fq_table_entry(fq->key); +#endif + return; + default: + break; + } + DPAA_ASSERT(NULL == "qman_free_fq() on unquiesced FQ!"); +} + +u32 qman_fq_fqid(struct qman_fq *fq) +{ + return fq->fqid; +} + +void qman_fq_state(struct qman_fq *fq, enum qman_fq_state *state, u32 *flags) +{ + if (state) + *state = fq->state; + if (flags) + *flags = fq->flags; +} + +int qman_init_fq(struct qman_fq *fq, u32 flags, struct qm_mcc_initfq *opts) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + u8 res, myverb = (flags & QMAN_INITFQ_FLAG_SCHED) ? + QM_MCC_VERB_INITFQ_SCHED : QM_MCC_VERB_INITFQ_PARKED; + + if ((fq->state != qman_fq_state_oos) && + (fq->state != qman_fq_state_parked)) + return -EINVAL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + if (opts && (opts->we_mask & QM_INITFQ_WE_OAC)) { + /* And can't be set at the same time as TDTHRESH */ + if (opts->we_mask & QM_INITFQ_WE_TDTHRESH) + return -EINVAL; + } + /* Issue an INITFQ_[PARKED|SCHED] management command */ + p = get_affine_portal(); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + ((fq->state != qman_fq_state_oos) && + (fq->state != qman_fq_state_parked)))) { + FQUNLOCK(fq); + return -EBUSY; + } + mcc = qm_mc_start(&p->p); + if (opts) + mcc->initfq = *opts; + mcc->initfq.fqid = cpu_to_be32(fq->fqid); + mcc->initfq.count = 0; + /* + * If the FQ does *not* have the TO_DCPORTAL flag, context_b is set as a + * demux pointer. Otherwise, the caller-provided value is allowed to + * stand, don't overwrite it. + */ + if (fq_isclear(fq, QMAN_FQ_FLAG_TO_DCPORTAL)) { + dma_addr_t phys_fq; + + mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTB; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + mcc->initfq.fqd.context_b = cpu_to_be32(fq->key); +#else + mcc->initfq.fqd.context_b = (u32)(uintptr_t)fq; +#endif + /* + * and the physical address - NB, if the user wasn't trying to + * set CONTEXTA, clear the stashing settings. + */ + if (!(mcc->initfq.we_mask & QM_INITFQ_WE_CONTEXTA)) { + mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTA; + memset(&mcc->initfq.fqd.context_a, 0, + sizeof(mcc->initfq.fqd.context_a)); + } else { + phys_fq = rte_mem_virt2iova(fq); + qm_fqd_stashing_set64(&mcc->initfq.fqd, phys_fq); + } + } + if (flags & QMAN_INITFQ_FLAG_LOCAL) { + mcc->initfq.fqd.dest.channel = p->config->channel; + if (!(mcc->initfq.we_mask & QM_INITFQ_WE_DESTWQ)) { + mcc->initfq.we_mask |= QM_INITFQ_WE_DESTWQ; + mcc->initfq.fqd.dest.wq = 4; + } + } + mcc->initfq.we_mask = cpu_to_be16(mcc->initfq.we_mask); + cpu_to_hw_fqd(&mcc->initfq.fqd); + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + FQUNLOCK(fq); + return -EIO; + } + if (opts) { + if (opts->we_mask & QM_INITFQ_WE_FQCTRL) { + if (opts->fqd.fq_ctrl & QM_FQCTRL_CGE) + fq_set(fq, QMAN_FQ_STATE_CGR_EN); + else + fq_clear(fq, QMAN_FQ_STATE_CGR_EN); + } + if (opts->we_mask & QM_INITFQ_WE_CGID) + fq->cgr_groupid = opts->fqd.cgid; + } + fq->state = (flags & QMAN_INITFQ_FLAG_SCHED) ? + qman_fq_state_sched : qman_fq_state_parked; + FQUNLOCK(fq); + return 0; +} + +int qman_schedule_fq(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + int ret = 0; + u8 res; + + if (fq->state != qman_fq_state_parked) + return -EINVAL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + /* Issue a ALTERFQ_SCHED management command */ + p = get_affine_portal(); + + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state != qman_fq_state_parked))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_SCHED); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_SCHED); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } + fq->state = qman_fq_state_sched; +out: + FQUNLOCK(fq); + + return ret; +} + +int qman_retire_fq(struct qman_fq *fq, u32 *flags) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + int rval; + u8 res; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_sched)) + return -EINVAL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + p = get_affine_portal(); + + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state == qman_fq_state_retired) || + (fq->state == qman_fq_state_oos))) { + rval = -EBUSY; + goto out; + } + rval = table_push_fq(p, fq); + if (rval) + goto out; + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_RETIRE); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_RETIRE); + res = mcr->result; + /* + * "Elegant" would be to treat OK/PENDING the same way; set CHANGING, + * and defer the flags until FQRNI or FQRN (respectively) show up. But + * "Friendly" is to process OK immediately, and not set CHANGING. We do + * friendly, otherwise the caller doesn't necessarily have a fully + * "retired" FQ on return even if the retirement was immediate. However + * this does mean some code duplication between here and + * fq_state_change(). + */ + if (likely(res == QM_MCR_RESULT_OK)) { + rval = 0; + /* Process 'fq' right away, we'll ignore FQRNI */ + if (mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY) + fq_set(fq, QMAN_FQ_STATE_NE); + if (mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT) + fq_set(fq, QMAN_FQ_STATE_ORL); + else + table_del_fq(p, fq); + if (flags) + *flags = fq->flags; + fq->state = qman_fq_state_retired; + if (fq->cb.fqs) { + /* + * Another issue with supporting "immediate" retirement + * is that we're forced to drop FQRNIs, because by the + * time they're seen it may already be "too late" (the + * fq may have been OOS'd and free()'d already). But if + * the upper layer wants a callback whether it's + * immediate or not, we have to fake a "MR" entry to + * look like an FQRNI... + */ + struct qm_mr_entry msg; + + msg.ern.verb = QM_MR_VERB_FQRNI; + msg.fq.fqs = mcr->alterfq.fqs; + msg.fq.fqid = fq->fqid; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + msg.fq.contextB = fq->key; +#else + msg.fq.contextB = (u32)(uintptr_t)fq; +#endif + fq->cb.fqs(p, fq, &msg); + } + } else if (res == QM_MCR_RESULT_PENDING) { + rval = 1; + fq_set(fq, QMAN_FQ_STATE_CHANGING); + } else { + rval = -EIO; + table_del_fq(p, fq); + } +out: + FQUNLOCK(fq); + return rval; +} + +int qman_oos_fq(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + int ret = 0; + u8 res; + + if (fq->state != qman_fq_state_retired) + return -EINVAL; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + p = get_affine_portal(); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_BLOCKOOS)) || + (fq->state != qman_fq_state_retired))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_OOS); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } + fq->state = qman_fq_state_oos; +out: + FQUNLOCK(fq); + return ret; +} + +int qman_fq_flow_control(struct qman_fq *fq, int xon) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + + int ret = 0; + u8 res; + u8 myverb; + + if ((fq->state == qman_fq_state_oos) || + (fq->state == qman_fq_state_retired) || + (fq->state == qman_fq_state_parked)) + return -EINVAL; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + /* Issue a ALTER_FQXON or ALTER_FQXOFF management command */ + p = get_affine_portal(); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state == qman_fq_state_parked) || + (fq->state == qman_fq_state_oos) || + (fq->state == qman_fq_state_retired))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + mcc->alterfq.count = 0; + myverb = xon ? QM_MCC_VERB_ALTER_FQXON : QM_MCC_VERB_ALTER_FQXOFF; + + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } +out: + FQUNLOCK(fq); + return ret; +} + +int qman_query_fq(struct qman_fq *fq, struct qm_fqd *fqd) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res; + + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *fqd = mcr->queryfq.fqd; + hw_fqd_to_cpu(fqd); + if (res != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} + +int qman_query_fq_has_pkts(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + int ret = 0; + u8 res; + + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + ret = !!mcr->queryfq_np.frm_cnt; + return ret; +} + +int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res; + + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) { + *np = mcr->queryfq_np; + np->fqd_link = be24_to_cpu(np->fqd_link); + np->odp_seq = be16_to_cpu(np->odp_seq); + np->orp_nesn = be16_to_cpu(np->orp_nesn); + np->orp_ea_hseq = be16_to_cpu(np->orp_ea_hseq); + np->orp_ea_tseq = be16_to_cpu(np->orp_ea_tseq); + np->orp_ea_hptr = be24_to_cpu(np->orp_ea_hptr); + np->orp_ea_tptr = be24_to_cpu(np->orp_ea_tptr); + np->pfdr_hptr = be24_to_cpu(np->pfdr_hptr); + np->pfdr_tptr = be24_to_cpu(np->pfdr_tptr); + np->ics_surp = be16_to_cpu(np->ics_surp); + np->byte_cnt = be32_to_cpu(np->byte_cnt); + np->frm_cnt = be24_to_cpu(np->frm_cnt); + np->ra1_sfdr = be16_to_cpu(np->ra1_sfdr); + np->ra2_sfdr = be16_to_cpu(np->ra2_sfdr); + np->od1_sfdr = be16_to_cpu(np->od1_sfdr); + np->od2_sfdr = be16_to_cpu(np->od2_sfdr); + np->od3_sfdr = be16_to_cpu(np->od3_sfdr); + } + if (res == QM_MCR_RESULT_ERR_FQID) + return -ERANGE; + else if (res != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} + +int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = cpu_to_be32(fq->fqid); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + + if (mcr->result == QM_MCR_RESULT_OK) + *frm_cnt = be24_to_cpu(mcr->queryfq_np.frm_cnt); + else if (mcr->result == QM_MCR_RESULT_ERR_FQID) + return -ERANGE; + else if (mcr->result != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} + +int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res, myverb; + + myverb = (query_dedicated) ? QM_MCR_VERB_QUERYWQ_DEDICATED : + QM_MCR_VERB_QUERYWQ; + mcc = qm_mc_start(&p->p); + mcc->querywq.channel.id = cpu_to_be16(wq->channel.id); + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) { + int i, array_len; + + wq->channel.id = be16_to_cpu(mcr->querywq.channel.id); + array_len = ARRAY_SIZE(mcr->querywq.wq_len); + for (i = 0; i < array_len; i++) + wq->wq_len[i] = be32_to_cpu(mcr->querywq.wq_len[i]); + } + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERYWQ failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} + +int qman_testwrite_cgr(struct qman_cgr *cgr, u64 i_bcnt, + struct qm_mcr_cgrtestwrite *result) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res; + + mcc = qm_mc_start(&p->p); + mcc->cgrtestwrite.cgid = cgr->cgrid; + mcc->cgrtestwrite.i_bcnt_hi = (u8)(i_bcnt >> 32); + mcc->cgrtestwrite.i_bcnt_lo = (u32)i_bcnt; + qm_mc_commit(&p->p, QM_MCC_VERB_CGRTESTWRITE); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_CGRTESTWRITE); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *result = mcr->cgrtestwrite; + if (res != QM_MCR_RESULT_OK) { + pr_err("CGR TEST WRITE failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} + +int qman_query_cgr(struct qman_cgr *cgr, struct qm_mcr_querycgr *cgrd) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + u8 res; + unsigned int i; + + mcc = qm_mc_start(&p->p); + mcc->querycgr.cgid = cgr->cgrid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCGR); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYCGR); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *cgrd = mcr->querycgr; + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERY_CGR failed: %s\n", mcr_result_str(res)); + return -EIO; + } + cgrd->cgr.wr_parm_g.word = + be32_to_cpu(cgrd->cgr.wr_parm_g.word); + cgrd->cgr.wr_parm_y.word = + be32_to_cpu(cgrd->cgr.wr_parm_y.word); + cgrd->cgr.wr_parm_r.word = + be32_to_cpu(cgrd->cgr.wr_parm_r.word); + cgrd->cgr.cscn_targ = be32_to_cpu(cgrd->cgr.cscn_targ); + cgrd->cgr.__cs_thres = be16_to_cpu(cgrd->cgr.__cs_thres); + for (i = 0; i < ARRAY_SIZE(cgrd->cscn_targ_swp); i++) + cgrd->cscn_targ_swp[i] = + be32_to_cpu(cgrd->cscn_targ_swp[i]); + return 0; +} + +int qman_query_congestion(struct qm_mcr_querycongestion *congestion) +{ + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + u8 res; + unsigned int i; + + qm_mc_start(&p->p); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCC_VERB_QUERYCONGESTION); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *congestion = mcr->querycongestion; + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERY_CONGESTION failed: %s\n", mcr_result_str(res)); + return -EIO; + } + for (i = 0; i < ARRAY_SIZE(congestion->state.state); i++) + congestion->state.state[i] = + be32_to_cpu(congestion->state.state[i]); + return 0; +} + +int qman_set_vdq(struct qman_fq *fq, u16 num, uint32_t vdqcr_flags) +{ + struct qman_portal *p = get_affine_portal(); + uint32_t vdqcr; + int ret = -EBUSY; + + vdqcr = vdqcr_flags; + vdqcr |= QM_VDQCR_NUMFRAMES_SET(num); + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_retired)) { + ret = -EINVAL; + goto out; + } + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) { + ret = -EBUSY; + goto out; + } + vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid; + + if (!p->vdqcr_owned) { + FQLOCK(fq); + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + goto escape; + fq_set(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + p->vdqcr_owned = fq; + ret = 0; + } +escape: + if (!ret) + qm_dqrr_vdqcr_set(&p->p, vdqcr); + +out: + return ret; +} + +int qman_volatile_dequeue(struct qman_fq *fq, u32 flags __maybe_unused, + u32 vdqcr) +{ + struct qman_portal *p; + int ret = -EBUSY; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_retired)) + return -EINVAL; + if (vdqcr & QM_VDQCR_FQID_MASK) + return -EINVAL; + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + return -EBUSY; + vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid; + + p = get_affine_portal(); + + if (!p->vdqcr_owned) { + FQLOCK(fq); + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + goto escape; + fq_set(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + p->vdqcr_owned = fq; + ret = 0; + } +escape: + if (ret) + return ret; + + /* VDQCR is set */ + qm_dqrr_vdqcr_set(&p->p, vdqcr); + return 0; +} + +static noinline void update_eqcr_ci(struct qman_portal *p, u8 avail) +{ + if (avail) + qm_eqcr_cce_prefetch(&p->p); + else + qm_eqcr_cce_update(&p->p); +} + +int qman_eqcr_is_empty(void) +{ + struct qman_portal *p = get_affine_portal(); + u8 avail; + + update_eqcr_ci(p, 0); + avail = qm_eqcr_get_fill(&p->p); + return (avail == 0); +} + +void qman_set_dc_ern(qman_cb_dc_ern handler, int affine) +{ + if (affine) { + struct qman_portal *p = get_affine_portal(); + + p->cb_dc_ern = handler; + } else + cb_dc_ern = handler; +} + +static inline struct qm_eqcr_entry *try_p_eq_start(struct qman_portal *p, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + u8 avail; + + if (p->use_eqcr_ci_stashing) { + /* + * The stashing case is easy, only update if we need to in + * order to try and liberate ring entries. + */ + eq = qm_eqcr_start_stash(&p->p); + } else { + /* + * The non-stashing case is harder, need to prefetch ahead of + * time. + */ + avail = qm_eqcr_get_avail(&p->p); + if (avail < 2) + update_eqcr_ci(p, avail); + eq = qm_eqcr_start_no_stash(&p->p); + } + + if (unlikely(!eq)) + return NULL; + + if (flags & QMAN_ENQUEUE_FLAG_DCA) + eq->dca = QM_EQCR_DCA_ENABLE | + ((flags & QMAN_ENQUEUE_FLAG_DCA_PARK) ? + QM_EQCR_DCA_PARK : 0) | + ((flags >> 8) & QM_EQCR_DCA_IDXMASK); + eq->fqid = cpu_to_be32(fq->fqid); +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + eq->tag = cpu_to_be32(fq->key); +#else + eq->tag = cpu_to_be32((u32)(uintptr_t)fq); +#endif + eq->fd = *fd; + cpu_to_hw_fd(&eq->fd); + return eq; +} + +int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags) +{ + struct qman_portal *p = get_affine_portal(); + struct qm_eqcr_entry *eq; + + eq = try_p_eq_start(p, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + return 0; +} + +int qman_enqueue_multi(struct qman_fq *fq, + const struct qm_fd *fd, u32 *flags, + int frames_to_send) +{ + struct qman_portal *p = get_affine_portal(); + struct qm_portal *portal = &p->p; + + register struct qm_eqcr *eqcr = &portal->eqcr; + struct qm_eqcr_entry *eq = eqcr->cursor, *prev_eq; + + u8 i = 0, diff, old_ci, sent = 0; + + /* Update the available entries if no entry is free */ + if (!eqcr->available) { + old_ci = eqcr->ci; + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + if (!diff) + return 0; + } + + /* try to send as many frames as possible */ + while (eqcr->available && frames_to_send--) { + eq->fqid = fq->fqid_le; + eq->fd.opaque_addr = fd->opaque_addr; + eq->fd.addr = cpu_to_be40(fd->addr); + eq->fd.status = cpu_to_be32(fd->status); + eq->fd.opaque = cpu_to_be32(fd->opaque); + if (flags && (flags[i] & QMAN_ENQUEUE_FLAG_DCA)) { + eq->dca = QM_EQCR_DCA_ENABLE | + ((flags[i] >> 8) & QM_EQCR_DCA_IDXMASK); + } + i++; + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + eqcr->available--; + sent++; + fd++; + } + lwsync(); + + /* In order for flushes to complete faster, all lines are recorded in + * 32 bit word. + */ + eq = eqcr->cursor; + for (i = 0; i < sent; i++) { + eq->__dont_write_directly__verb = + QM_EQCR_VERB_CMD_ENQUEUE | eqcr->vbit; + prev_eq = eq; + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + if (unlikely((prev_eq + 1) != eq)) + eqcr->vbit ^= QM_EQCR_VERB_VBIT; + } + + /* We need to flush all the lines but without load/store operations + * between them + */ + eq = eqcr->cursor; + for (i = 0; i < sent; i++) { + dcbf(eq); + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + } + /* Update cursor for the next call */ + eqcr->cursor = eq; + return sent; +} + +int +qman_enqueue_multi_fq(struct qman_fq *fq[], const struct qm_fd *fd, + u32 *flags, int frames_to_send) +{ + struct qman_portal *p = get_affine_portal(); + struct qm_portal *portal = &p->p; + + register struct qm_eqcr *eqcr = &portal->eqcr; + struct qm_eqcr_entry *eq = eqcr->cursor, *prev_eq; + + u8 i = 0, diff, old_ci, sent = 0; + + /* Update the available entries if no entry is free */ + if (!eqcr->available) { + old_ci = eqcr->ci; + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + if (!diff) + return 0; + } + + /* try to send as many frames as possible */ + while (eqcr->available && frames_to_send--) { + eq->fqid = fq[sent]->fqid_le; + eq->fd.opaque_addr = fd->opaque_addr; + eq->fd.addr = cpu_to_be40(fd->addr); + eq->fd.status = cpu_to_be32(fd->status); + eq->fd.opaque = cpu_to_be32(fd->opaque); + if (flags && (flags[i] & QMAN_ENQUEUE_FLAG_DCA)) { + eq->dca = QM_EQCR_DCA_ENABLE | + ((flags[i] >> 8) & QM_EQCR_DCA_IDXMASK); + } + i++; + + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + eqcr->available--; + sent++; + fd++; + } + lwsync(); + + /* In order for flushes to complete faster, all lines are recorded in + * 32 bit word. + */ + eq = eqcr->cursor; + for (i = 0; i < sent; i++) { + eq->__dont_write_directly__verb = + QM_EQCR_VERB_CMD_ENQUEUE | eqcr->vbit; + prev_eq = eq; + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + if (unlikely((prev_eq + 1) != eq)) + eqcr->vbit ^= QM_EQCR_VERB_VBIT; + } + + /* We need to flush all the lines but without load/store operations + * between them + */ + eq = eqcr->cursor; + for (i = 0; i < sent; i++) { + dcbf(eq); + eq = (void *)((unsigned long)(eq + 1) & + (~(unsigned long)(QM_EQCR_SIZE << 6))); + } + /* Update cursor for the next call */ + eqcr->cursor = eq; + return sent; +} + +int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum) +{ + struct qman_portal *p = get_affine_portal(); + struct qm_eqcr_entry *eq; + + eq = try_p_eq_start(p, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Process ORP-specifics here */ + if (flags & QMAN_ENQUEUE_FLAG_NLIS) + orp_seqnum |= QM_EQCR_SEQNUM_NLIS; + else { + orp_seqnum &= ~QM_EQCR_SEQNUM_NLIS; + if (flags & QMAN_ENQUEUE_FLAG_NESN) + orp_seqnum |= QM_EQCR_SEQNUM_NESN; + else + /* No need to check 4 QMAN_ENQUEUE_FLAG_HOLE */ + orp_seqnum &= ~QM_EQCR_SEQNUM_NESN; + } + eq->seqnum = cpu_to_be16(orp_seqnum); + eq->orp = cpu_to_be32(orp->fqid); + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_ORP | + ((flags & (QMAN_ENQUEUE_FLAG_HOLE | QMAN_ENQUEUE_FLAG_NESN)) ? + 0 : QM_EQCR_VERB_CMD_ENQUEUE) | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + + return 0; +} + +int qman_modify_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + + u8 res; + u8 verb = QM_MCC_VERB_MODIFYCGR; + + mcc = qm_mc_start(&p->p); + if (opts) + mcc->initcgr = *opts; + mcc->initcgr.we_mask = cpu_to_be16(mcc->initcgr.we_mask); + mcc->initcgr.cgr.wr_parm_g.word = + cpu_to_be32(mcc->initcgr.cgr.wr_parm_g.word); + mcc->initcgr.cgr.wr_parm_y.word = + cpu_to_be32(mcc->initcgr.cgr.wr_parm_y.word); + mcc->initcgr.cgr.wr_parm_r.word = + cpu_to_be32(mcc->initcgr.cgr.wr_parm_r.word); + mcc->initcgr.cgr.cscn_targ = cpu_to_be32(mcc->initcgr.cgr.cscn_targ); + mcc->initcgr.cgr.__cs_thres = cpu_to_be16(mcc->initcgr.cgr.__cs_thres); + + mcc->initcgr.cgid = cgr->cgrid; + if (flags & QMAN_CGR_FLAG_USE_INIT) + verb = QM_MCC_VERB_INITCGR; + qm_mc_commit(&p->p, verb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == verb); + res = mcr->result; + return (res == QM_MCR_RESULT_OK) ? 0 : -EIO; +} + +#define TARG_MASK(n) (0x80000000 >> (n->config->channel - \ + QM_CHANNEL_SWPORTAL0)) +#define TARG_DCP_MASK(n) (0x80000000 >> (10 + n)) +#define PORTAL_IDX(n) (n->config->channel - QM_CHANNEL_SWPORTAL0) + +int qman_create_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts) +{ + struct qm_mcr_querycgr cgr_state; + struct qm_mcc_initcgr local_opts; + int ret; + struct qman_portal *p; + + /* We have to check that the provided CGRID is within the limits of the + * data-structures, for obvious reasons. However we'll let h/w take + * care of determining whether it's within the limits of what exists on + * the SoC. + */ + if (cgr->cgrid >= __CGR_NUM) + return -EINVAL; + + p = get_affine_portal(); + + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + cgr->chan = p->config->channel; + spin_lock(&p->cgr_lock); + + /* if no opts specified, just add it to the list */ + if (!opts) + goto add_list; + + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) + goto release_lock; + if (opts) + local_opts = *opts; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = + QM_CGR_TARG_UDP_CTRL_WRITE_BIT | PORTAL_IDX(p); + else + /* Overwrite TARG */ + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ | + TARG_MASK(p); + local_opts.we_mask |= QM_CGR_WE_CSCN_TARG; + + /* send init if flags indicate so */ + if (opts && (flags & QMAN_CGR_FLAG_USE_INIT)) + ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, &local_opts); + else + ret = qman_modify_cgr(cgr, 0, &local_opts); + if (ret) + goto release_lock; +add_list: + list_add(&cgr->node, &p->cgr_cbs); + + /* Determine if newly added object requires its callback to be called */ + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) { + /* we can't go back, so proceed and return success, but screen + * and wail to the log file. + */ + pr_crit("CGR HW state partially modified\n"); + ret = 0; + goto release_lock; + } + if (cgr->cb && cgr_state.cgr.cscn_en && qman_cgrs_get(&p->cgrs[1], + cgr->cgrid)) + cgr->cb(p, cgr, 1); +release_lock: + spin_unlock(&p->cgr_lock); + return ret; +} + +int qman_create_cgr_to_dcp(struct qman_cgr *cgr, u32 flags, u16 dcp_portal, + struct qm_mcc_initcgr *opts) +{ + struct qm_mcc_initcgr local_opts; + struct qm_mcr_querycgr cgr_state; + int ret; + + if ((qman_ip_rev & 0xFF00) < QMAN_REV30) { + pr_warn("QMan version doesn't support CSCN => DCP portal\n"); + return -EINVAL; + } + /* We have to check that the provided CGRID is within the limits of the + * data-structures, for obvious reasons. However we'll let h/w take + * care of determining whether it's within the limits of what exists on + * the SoC. + */ + if (cgr->cgrid >= __CGR_NUM) + return -EINVAL; + + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) + return ret; + + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + if (opts) + local_opts = *opts; + + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = + QM_CGR_TARG_UDP_CTRL_WRITE_BIT | + QM_CGR_TARG_UDP_CTRL_DCP | dcp_portal; + else + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ | + TARG_DCP_MASK(dcp_portal); + local_opts.we_mask |= QM_CGR_WE_CSCN_TARG; + + /* send init if flags indicate so */ + if (opts && (flags & QMAN_CGR_FLAG_USE_INIT)) + ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, + &local_opts); + else + ret = qman_modify_cgr(cgr, 0, &local_opts); + + return ret; +} + +int qman_delete_cgr(struct qman_cgr *cgr) +{ + struct qm_mcr_querycgr cgr_state; + struct qm_mcc_initcgr local_opts; + int ret = 0; + struct qman_cgr *i; + struct qman_portal *p = get_affine_portal(); + + if (cgr->chan != p->config->channel) { + pr_crit("Attempting to delete cgr from different portal than" + " it was create: create 0x%x, delete 0x%x\n", + cgr->chan, p->config->channel); + ret = -EINVAL; + goto put_portal; + } + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + spin_lock(&p->cgr_lock); + list_del(&cgr->node); + /* + * If there are no other CGR objects for this CGRID in the list, + * update CSCN_TARG accordingly + */ + list_for_each_entry(i, &p->cgr_cbs, node) + if ((i->cgrid == cgr->cgrid) && i->cb) + goto release_lock; + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) { + /* add back to the list */ + list_add(&cgr->node, &p->cgr_cbs); + goto release_lock; + } + /* Overwrite TARG */ + local_opts.we_mask = QM_CGR_WE_CSCN_TARG; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = PORTAL_IDX(p); + else + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ & + ~(TARG_MASK(p)); + ret = qman_modify_cgr(cgr, 0, &local_opts); + if (ret) + /* add back to the list */ + list_add(&cgr->node, &p->cgr_cbs); +release_lock: + spin_unlock(&p->cgr_lock); +put_portal: + return ret; +} + +int qman_shutdown_fq(u32 fqid) +{ + struct qman_portal *p; + struct qm_portal *low_p; + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + u8 state; + int orl_empty, fq_empty, drain = 0; + u32 result; + u32 channel, wq; + u16 dest_wq; + + p = get_affine_portal(); + low_p = &p->p; + + /* Determine the state of the FQID */ + mcc = qm_mc_start(low_p); + mcc->queryfq_np.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + state = mcr->queryfq_np.state & QM_MCR_NP_STATE_MASK; + if (state == QM_MCR_NP_STATE_OOS) + return 0; /* Already OOS, no need to do anymore checks */ + + /* Query which channel the FQ is using */ + mcc = qm_mc_start(low_p); + mcc->queryfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ); + + /* Need to store these since the MCR gets reused */ + dest_wq = be16_to_cpu(mcr->queryfq.fqd.dest_wq); + channel = dest_wq & 0x7; + wq = dest_wq >> 3; + + switch (state) { + case QM_MCR_NP_STATE_TEN_SCHED: + case QM_MCR_NP_STATE_TRU_SCHED: + case QM_MCR_NP_STATE_ACTIVE: + case QM_MCR_NP_STATE_PARKED: + orl_empty = 0; + mcc = qm_mc_start(low_p); + mcc->alterfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_ALTER_RETIRE); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_RETIRE); + result = mcr->result; /* Make a copy as we reuse MCR below */ + + if (result == QM_MCR_RESULT_PENDING) { + /* Need to wait for the FQRN in the message ring, which + * will only occur once the FQ has been drained. In + * order for the FQ to drain the portal needs to be set + * to dequeue from the channel the FQ is scheduled on + */ + const struct qm_mr_entry *msg; + const struct qm_dqrr_entry *dqrr = NULL; + int found_fqrn = 0; + __maybe_unused u16 dequeue_wq = 0; + + /* Flag that we need to drain FQ */ + drain = 1; + + if (channel >= qm_channel_pool1 && + channel < (u16)(qm_channel_pool1 + 15)) { + /* Pool channel, enable the bit in the portal */ + dequeue_wq = (channel - + qm_channel_pool1 + 1) << 4 | wq; + } else if (channel < qm_channel_pool1) { + /* Dedicated channel */ + dequeue_wq = wq; + } else { + pr_info("Cannot recover FQ 0x%x," + " it is scheduled on channel 0x%x", + fqid, channel); + return -EBUSY; + } + /* Set the sdqcr to drain this channel */ + if (channel < qm_channel_pool1) + qm_dqrr_sdqcr_set(low_p, + QM_SDQCR_TYPE_ACTIVE | + QM_SDQCR_CHANNELS_DEDICATED); + else + qm_dqrr_sdqcr_set(low_p, + QM_SDQCR_TYPE_ACTIVE | + QM_SDQCR_CHANNELS_POOL_CONV + (channel)); + while (!found_fqrn) { + /* Keep draining DQRR while checking the MR*/ + qm_dqrr_pvb_update(low_p); + dqrr = qm_dqrr_current(low_p); + while (dqrr) { + qm_dqrr_cdc_consume_1ptr( + low_p, dqrr, 0); + qm_dqrr_pvb_update(low_p); + qm_dqrr_next(low_p); + dqrr = qm_dqrr_current(low_p); + } + /* Process message ring too */ + qm_mr_pvb_update(low_p); + msg = qm_mr_current(low_p); + while (msg) { + if ((msg->ern.verb & + QM_MR_VERB_TYPE_MASK) + == QM_MR_VERB_FQRN) + found_fqrn = 1; + qm_mr_next(low_p); + qm_mr_cci_consume_to_current(low_p); + qm_mr_pvb_update(low_p); + msg = qm_mr_current(low_p); + } + cpu_relax(); + } + } + if (result != QM_MCR_RESULT_OK && + result != QM_MCR_RESULT_PENDING) { + /* error */ + pr_err("qman_retire_fq failed on FQ 0x%x," + " result=0x%x\n", fqid, result); + return -1; + } + if (!(mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT)) { + /* ORL had no entries, no need to wait until the + * ERNs come in. + */ + orl_empty = 1; + } + /* Retirement succeeded, check to see if FQ needs + * to be drained. + */ + if (drain || mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY) { + /* FQ is Not Empty, drain using volatile DQ commands */ + fq_empty = 0; + do { + const struct qm_dqrr_entry *dqrr = NULL; + u32 vdqcr = fqid | QM_VDQCR_NUMFRAMES_SET(3); + + qm_dqrr_vdqcr_set(low_p, vdqcr); + + /* Wait for a dequeue to occur */ + while (dqrr == NULL) { + qm_dqrr_pvb_update(low_p); + dqrr = qm_dqrr_current(low_p); + if (!dqrr) + cpu_relax(); + } + /* Process the dequeues, making sure to + * empty the ring completely. + */ + while (dqrr) { + if (dqrr->fqid == fqid && + dqrr->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_empty = 1; + qm_dqrr_cdc_consume_1ptr(low_p, + dqrr, 0); + qm_dqrr_pvb_update(low_p); + qm_dqrr_next(low_p); + dqrr = qm_dqrr_current(low_p); + } + } while (fq_empty == 0); + } + qm_dqrr_sdqcr_set(low_p, 0); + + /* Wait for the ORL to have been completely drained */ + while (orl_empty == 0) { + const struct qm_mr_entry *msg; + + qm_mr_pvb_update(low_p); + msg = qm_mr_current(low_p); + while (msg) { + if ((msg->ern.verb & QM_MR_VERB_TYPE_MASK) == + QM_MR_VERB_FQRL) + orl_empty = 1; + qm_mr_next(low_p); + qm_mr_cci_consume_to_current(low_p); + qm_mr_pvb_update(low_p); + msg = qm_mr_current(low_p); + } + cpu_relax(); + } + mcc = qm_mc_start(low_p); + mcc->alterfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_OOS); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err( + "OOS after drain Failed on FQID 0x%x, result 0x%x\n", + fqid, mcr->result); + return -1; + } + return 0; + + case QM_MCR_NP_STATE_RETIRED: + /* Send OOS Command */ + mcc = qm_mc_start(low_p); + mcc->alterfq.fqid = cpu_to_be32(fqid); + qm_mc_commit(low_p, QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(low_p))) + cpu_relax(); + DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_OOS); + if (mcr->result) { + pr_err("OOS Failed on FQID 0x%x\n", fqid); + return -1; + } + return 0; + + } + return -1; +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman.h b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman.h new file mode 100644 index 000000000..4346d8653 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman.h @@ -0,0 +1,913 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017 NXP + * + */ + +#include "qman_priv.h" + +/***************************/ +/* Portal register assists */ +/***************************/ +#define QM_REG_EQCR_PI_CINH 0x3000 +#define QM_REG_EQCR_CI_CINH 0x3040 +#define QM_REG_EQCR_ITR 0x3080 +#define QM_REG_DQRR_PI_CINH 0x3100 +#define QM_REG_DQRR_CI_CINH 0x3140 +#define QM_REG_DQRR_ITR 0x3180 +#define QM_REG_DQRR_DCAP 0x31C0 +#define QM_REG_DQRR_SDQCR 0x3200 +#define QM_REG_DQRR_VDQCR 0x3240 +#define QM_REG_DQRR_PDQCR 0x3280 +#define QM_REG_MR_PI_CINH 0x3300 +#define QM_REG_MR_CI_CINH 0x3340 +#define QM_REG_MR_ITR 0x3380 +#define QM_REG_CFG 0x3500 +#define QM_REG_ISR 0x3600 +#define QM_REG_IIR 0x36C0 +#define QM_REG_ITPR 0x3740 + +/* Cache-enabled register offsets */ +#define QM_CL_EQCR 0x0000 +#define QM_CL_DQRR 0x1000 +#define QM_CL_MR 0x2000 +#define QM_CL_EQCR_PI_CENA 0x3000 +#define QM_CL_EQCR_CI_CENA 0x3040 +#define QM_CL_DQRR_PI_CENA 0x3100 +#define QM_CL_DQRR_CI_CENA 0x3140 +#define QM_CL_MR_PI_CENA 0x3300 +#define QM_CL_MR_CI_CENA 0x3340 +#define QM_CL_CR 0x3800 +#define QM_CL_RR0 0x3900 +#define QM_CL_RR1 0x3940 + +/* BTW, the drivers (and h/w programming model) already obtain the required + * synchronisation for portal accesses via lwsync(), hwsync(), and + * data-dependencies. Use of barrier()s or other order-preserving primitives + * simply degrade performance. Hence the use of the __raw_*() interfaces, which + * simply ensure that the compiler treats the portal registers as volatile (ie. + * non-coherent). + */ + +/* Cache-inhibited register access. */ +#define __qm_in(qm, o) be32_to_cpu(__raw_readl((qm)->ci + (o))) +#define __qm_out(qm, o, val) __raw_writel((cpu_to_be32(val)), \ + (qm)->ci + (o)) +#define qm_in(reg) __qm_in(&portal->addr, QM_REG_##reg) +#define qm_out(reg, val) __qm_out(&portal->addr, QM_REG_##reg, val) + +/* Cache-enabled (index) register access */ +#define __qm_cl_touch_ro(qm, o) dcbt_ro((qm)->ce + (o)) +#define __qm_cl_touch_rw(qm, o) dcbt_rw((qm)->ce + (o)) +#define __qm_cl_in(qm, o) be32_to_cpu(__raw_readl((qm)->ce + (o))) +#define __qm_cl_out(qm, o, val) \ + do { \ + u32 *__tmpclout = (qm)->ce + (o); \ + __raw_writel(cpu_to_be32(val), __tmpclout); \ + dcbf(__tmpclout); \ + } while (0) +#define __qm_cl_invalidate(qm, o) dccivac((qm)->ce + (o)) +#define qm_cl_touch_ro(reg) __qm_cl_touch_ro(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_touch_rw(reg) __qm_cl_touch_rw(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_in(reg) __qm_cl_in(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_out(reg, val) __qm_cl_out(&portal->addr, QM_CL_##reg##_CENA, val) +#define qm_cl_invalidate(reg)\ + __qm_cl_invalidate(&portal->addr, QM_CL_##reg##_CENA) + +/* Cache-enabled ring access */ +#define qm_cl(base, idx) ((void *)base + ((idx) << 6)) + +/* Cyclic helper for rings. FIXME: once we are able to do fine-grain perf + * analysis, look at using the "extra" bit in the ring index registers to avoid + * cyclic issues. + */ +static inline u8 qm_cyc_diff(u8 ringsize, u8 first, u8 last) +{ + /* 'first' is included, 'last' is excluded */ + if (first <= last) + return last - first; + return ringsize + last - first; +} + +/* Portal modes. + * Enum types; + * pmode == production mode + * cmode == consumption mode, + * dmode == h/w dequeue mode. + * Enum values use 3 letter codes. First letter matches the portal mode, + * remaining two letters indicate; + * ci == cache-inhibited portal register + * ce == cache-enabled portal register + * vb == in-band valid-bit (cache-enabled) + * dc == DCA (Discrete Consumption Acknowledgment), DQRR-only + * As for "enum qm_dqrr_dmode", it should be self-explanatory. + */ +enum qm_eqcr_pmode { /* matches QCSP_CFG::EPM */ + qm_eqcr_pci = 0, /* PI index, cache-inhibited */ + qm_eqcr_pce = 1, /* PI index, cache-enabled */ + qm_eqcr_pvb = 2 /* valid-bit */ +}; + +enum qm_dqrr_dmode { /* matches QCSP_CFG::DP */ + qm_dqrr_dpush = 0, /* SDQCR + VDQCR */ + qm_dqrr_dpull = 1 /* PDQCR */ +}; + +enum qm_dqrr_pmode { /* s/w-only */ + qm_dqrr_pci, /* reads DQRR_PI_CINH */ + qm_dqrr_pce, /* reads DQRR_PI_CENA */ + qm_dqrr_pvb /* reads valid-bit */ +}; + +enum qm_dqrr_cmode { /* matches QCSP_CFG::DCM */ + qm_dqrr_cci = 0, /* CI index, cache-inhibited */ + qm_dqrr_cce = 1, /* CI index, cache-enabled */ + qm_dqrr_cdc = 2 /* Discrete Consumption Acknowledgment */ +}; + +enum qm_mr_pmode { /* s/w-only */ + qm_mr_pci, /* reads MR_PI_CINH */ + qm_mr_pce, /* reads MR_PI_CENA */ + qm_mr_pvb /* reads valid-bit */ +}; + +enum qm_mr_cmode { /* matches QCSP_CFG::MM */ + qm_mr_cci = 0, /* CI index, cache-inhibited */ + qm_mr_cce = 1 /* CI index, cache-enabled */ +}; + +/* ------------------------- */ +/* --- Portal structures --- */ + +#define QM_EQCR_SIZE 8 +#define QM_DQRR_SIZE 16 +#define QM_MR_SIZE 8 + +struct qm_eqcr { + struct qm_eqcr_entry *ring, *cursor; + u8 ci, available, ithresh, vbit; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + u32 busy; + enum qm_eqcr_pmode pmode; +#endif +}; + +struct qm_dqrr { + struct qm_dqrr_entry *ring, *cursor; + u8 pi, ci, fill, ithresh, vbit; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + enum qm_dqrr_dmode dmode; + enum qm_dqrr_pmode pmode; + enum qm_dqrr_cmode cmode; +#endif +}; + +struct qm_mr { + const struct qm_mr_entry *ring, *cursor; + u8 pi, ci, fill, ithresh, vbit; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + enum qm_mr_pmode pmode; + enum qm_mr_cmode cmode; +#endif +}; + +struct qm_mc { + struct qm_mc_command *cr; + struct qm_mc_result *rr; + u8 rridx, vbit; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + enum { + /* Can be _mc_start()ed */ + qman_mc_idle, + /* Can be _mc_commit()ed or _mc_abort()ed */ + qman_mc_user, + /* Can only be _mc_retry()ed */ + qman_mc_hw + } state; +#endif +}; + +#define QM_PORTAL_ALIGNMENT ____cacheline_aligned + +struct qm_addr { + void __iomem *ce; /* cache-enabled */ + void __iomem *ci; /* cache-inhibited */ +}; + +struct qm_portal { + struct qm_addr addr; + struct qm_eqcr eqcr; + struct qm_dqrr dqrr; + struct qm_mr mr; + struct qm_mc mc; +} QM_PORTAL_ALIGNMENT; + +/* Bit-wise logic to wrap a ring pointer by clearing the "carry bit" */ +#define EQCR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_EQCR_SIZE << 6))) + +extern dma_addr_t rte_mem_virt2iova(const void *addr); + +/* Bit-wise logic to convert a ring pointer to a ring index */ +static inline u8 EQCR_PTR2IDX(struct qm_eqcr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_EQCR_SIZE - 1); +} + +/* Increment the 'cursor' ring pointer, taking 'vbit' into account */ +static inline void EQCR_INC(struct qm_eqcr *eqcr) +{ + /* NB: this is odd-looking, but experiments show that it generates fast + * code with essentially no branching overheads. We increment to the + * next EQCR pointer and handle overflow and 'vbit'. + */ + struct qm_eqcr_entry *partial = eqcr->cursor + 1; + + eqcr->cursor = EQCR_CARRYCLEAR(partial); + if (partial != eqcr->cursor) + eqcr->vbit ^= QM_EQCR_VERB_VBIT; +} + +static inline struct qm_eqcr_entry *qm_eqcr_start_no_stash(struct qm_portal + *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(!eqcr->busy); +#endif + if (!eqcr->available) + return NULL; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + eqcr->busy = 1; +#endif + + return eqcr->cursor; +} + +static inline struct qm_eqcr_entry *qm_eqcr_start_stash(struct qm_portal + *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(!eqcr->busy); +#endif + if (!eqcr->available) { + old_ci = eqcr->ci; + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + if (!diff) + return NULL; + } +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + eqcr->busy = 1; +#endif + return eqcr->cursor; +} + +static inline void qm_eqcr_abort(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(eqcr->busy); + eqcr->busy = 0; +#endif +} + +static inline struct qm_eqcr_entry *qm_eqcr_pend_and_next( + struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(eqcr->busy); + DPAA_ASSERT(eqcr->pmode != qm_eqcr_pvb); +#endif + if (eqcr->available == 1) + return NULL; + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + dcbf(eqcr->cursor); + EQCR_INC(eqcr); + eqcr->available--; + return eqcr->cursor; +} + +#define EQCR_COMMIT_CHECKS(eqcr) \ +do { \ + DPAA_ASSERT(eqcr->busy); \ + DPAA_ASSERT(eqcr->cursor->orp == (eqcr->cursor->orp & 0x00ffffff)); \ + DPAA_ASSERT(eqcr->cursor->fqid == (eqcr->cursor->fqid & 0x00ffffff)); \ +} while (0) + +static inline void qm_eqcr_pci_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + EQCR_COMMIT_CHECKS(eqcr); + DPAA_ASSERT(eqcr->pmode == qm_eqcr_pci); +#endif + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + EQCR_INC(eqcr); + eqcr->available--; + dcbf(eqcr->cursor); + hwsync(); + qm_out(EQCR_PI_CINH, EQCR_PTR2IDX(eqcr->cursor)); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + eqcr->busy = 0; +#endif +} + +static inline void qm_eqcr_pce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(eqcr->pmode == qm_eqcr_pce); +#endif + qm_cl_invalidate(EQCR_PI); + qm_cl_touch_rw(EQCR_PI); +} + +static inline void qm_eqcr_pce_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + EQCR_COMMIT_CHECKS(eqcr); + DPAA_ASSERT(eqcr->pmode == qm_eqcr_pce); +#endif + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + EQCR_INC(eqcr); + eqcr->available--; + dcbf(eqcr->cursor); + lwsync(); + qm_cl_out(EQCR_PI, EQCR_PTR2IDX(eqcr->cursor)); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + eqcr->busy = 0; +#endif +} + +static inline void qm_eqcr_pvb_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + struct qm_eqcr_entry *eqcursor; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + EQCR_COMMIT_CHECKS(eqcr); + DPAA_ASSERT(eqcr->pmode == qm_eqcr_pvb); +#endif + lwsync(); + eqcursor = eqcr->cursor; + eqcursor->__dont_write_directly__verb = myverb | eqcr->vbit; + dcbf(eqcursor); + EQCR_INC(eqcr); + eqcr->available--; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + eqcr->busy = 0; +#endif +} + +static inline u8 qm_eqcr_cci_update(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci = eqcr->ci; + + eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + return diff; +} + +static inline void qm_eqcr_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + + qm_cl_touch_ro(EQCR_CI); +} + +static inline u8 qm_eqcr_cce_update(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci = eqcr->ci; + + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + qm_cl_invalidate(EQCR_CI); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + return diff; +} + +static inline u8 qm_eqcr_get_ithresh(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + return eqcr->ithresh; +} + +static inline void qm_eqcr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + eqcr->ithresh = ithresh; + qm_out(EQCR_ITR, ithresh); +} + +static inline u8 qm_eqcr_get_avail(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + return eqcr->available; +} + +static inline u8 qm_eqcr_get_fill(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + return QM_EQCR_SIZE - 1 - eqcr->available; +} + +#define DQRR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_DQRR_SIZE << 6))) + +static inline u8 DQRR_PTR2IDX(const struct qm_dqrr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_DQRR_SIZE - 1); +} + +static inline struct qm_dqrr_entry *DQRR_INC( + const struct qm_dqrr_entry *e) +{ + return DQRR_CARRYCLEAR(e + 1); +} + +static inline void qm_dqrr_set_maxfill(struct qm_portal *portal, u8 mf) +{ + qm_out(CFG, (qm_in(CFG) & 0xff0fffff) | + ((mf & (QM_DQRR_SIZE - 1)) << 20)); +} + +static inline const struct qm_dqrr_entry *qm_dqrr_current( + struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + if (!dqrr->fill) + return NULL; + return dqrr->cursor; +} + +static inline u8 qm_dqrr_cursor(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + return DQRR_PTR2IDX(dqrr->cursor); +} + +static inline u8 qm_dqrr_next(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + DPAA_ASSERT(dqrr->fill); + dqrr->cursor = DQRR_INC(dqrr->cursor); + return --dqrr->fill; +} + +static inline u8 qm_dqrr_pci_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u8 diff, old_pi = dqrr->pi; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->pmode == qm_dqrr_pci); +#endif + dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1); + diff = qm_cyc_diff(QM_DQRR_SIZE, old_pi, dqrr->pi); + dqrr->fill += diff; + return diff; +} + +static inline void qm_dqrr_pce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->pmode == qm_dqrr_pce); +#endif + qm_cl_invalidate(DQRR_PI); + qm_cl_touch_ro(DQRR_PI); +} + +static inline u8 qm_dqrr_pce_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u8 diff, old_pi = dqrr->pi; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->pmode == qm_dqrr_pce); +#endif + dqrr->pi = qm_cl_in(DQRR_PI) & (QM_DQRR_SIZE - 1); + diff = qm_cyc_diff(QM_DQRR_SIZE, old_pi, dqrr->pi); + dqrr->fill += diff; + return diff; +} + +static inline void qm_dqrr_pvb_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + const struct qm_dqrr_entry *res = qm_cl(dqrr->ring, dqrr->pi); + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->pmode == qm_dqrr_pvb); +#endif + /* when accessing 'verb', use __raw_readb() to ensure that compiler + * inlining doesn't try to optimise out "excess reads". + */ + if ((__raw_readb(&res->verb) & QM_DQRR_VERB_VBIT) == dqrr->vbit) { + dqrr->pi = (dqrr->pi + 1) & (QM_DQRR_SIZE - 1); + if (!dqrr->pi) + dqrr->vbit ^= QM_DQRR_VERB_VBIT; + dqrr->fill++; + } +} + +static inline void qm_dqrr_cci_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cci); +#endif + dqrr->ci = (dqrr->ci + num) & (QM_DQRR_SIZE - 1); + qm_out(DQRR_CI_CINH, dqrr->ci); +} + +static inline void qm_dqrr_cci_consume_to_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cci); +#endif + dqrr->ci = DQRR_PTR2IDX(dqrr->cursor); + qm_out(DQRR_CI_CINH, dqrr->ci); +} + +static inline void qm_dqrr_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cce); +#endif + qm_cl_invalidate(DQRR_CI); + qm_cl_touch_rw(DQRR_CI); +} + +static inline void qm_dqrr_cce_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cce); +#endif + dqrr->ci = (dqrr->ci + num) & (QM_DQRR_SIZE - 1); + qm_cl_out(DQRR_CI, dqrr->ci); +} + +static inline void qm_dqrr_cce_consume_to_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cce); +#endif + dqrr->ci = DQRR_PTR2IDX(dqrr->cursor); + qm_cl_out(DQRR_CI, dqrr->ci); +} + +static inline void qm_dqrr_cdc_consume_1(struct qm_portal *portal, u8 idx, + int park) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cdc); +#endif + DPAA_ASSERT(idx < QM_DQRR_SIZE); + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + ((park ? 1 : 0) << 6) | /* PK */ + idx); /* DCAP_CI */ +} + +static inline void qm_dqrr_cdc_consume_1ptr(struct qm_portal *portal, + const struct qm_dqrr_entry *dq, + int park) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + u8 idx = DQRR_PTR2IDX(dq); + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cdc); +#endif + DPAA_ASSERT(idx < QM_DQRR_SIZE); + qm_out(DQRR_DCAP, (0 << 8) | /* DQRR_DCAP::S */ + ((park ? 1 : 0) << 6) | /* DQRR_DCAP::PK */ + idx); /* DQRR_DCAP::DCAP_CI */ +} + +static inline void qm_dqrr_cdc_consume_n(struct qm_portal *portal, u16 bitmask) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cdc); +#endif + qm_out(DQRR_DCAP, (1 << 8) | /* DQRR_DCAP::S */ + ((u32)bitmask << 16)); /* DQRR_DCAP::DCAP_CI */ + dqrr->ci = qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->fill = qm_cyc_diff(QM_DQRR_SIZE, dqrr->ci, dqrr->pi); +} + +static inline u8 qm_dqrr_cdc_cci(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cdc); +#endif + return qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1); +} + +static inline void qm_dqrr_cdc_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cdc); +#endif + qm_cl_invalidate(DQRR_CI); + qm_cl_touch_ro(DQRR_CI); +} + +static inline u8 qm_dqrr_cdc_cce(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode == qm_dqrr_cdc); +#endif + return qm_cl_in(DQRR_CI) & (QM_DQRR_SIZE - 1); +} + +static inline u8 qm_dqrr_get_ci(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode != qm_dqrr_cdc); +#endif + return dqrr->ci; +} + +static inline void qm_dqrr_park(struct qm_portal *portal, u8 idx) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode != qm_dqrr_cdc); +#endif + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + (1 << 6) | /* PK */ + (idx & (QM_DQRR_SIZE - 1))); /* DCAP_CI */ +} + +static inline void qm_dqrr_park_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(dqrr->cmode != qm_dqrr_cdc); +#endif + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + (1 << 6) | /* PK */ + DQRR_PTR2IDX(dqrr->cursor)); /* DCAP_CI */ +} + +static inline void qm_dqrr_sdqcr_set(struct qm_portal *portal, u32 sdqcr) +{ + qm_out(DQRR_SDQCR, sdqcr); +} + +static inline u32 qm_dqrr_sdqcr_get(struct qm_portal *portal) +{ + return qm_in(DQRR_SDQCR); +} + +static inline void qm_dqrr_vdqcr_set(struct qm_portal *portal, u32 vdqcr) +{ + qm_out(DQRR_VDQCR, vdqcr); +} + +static inline u32 qm_dqrr_vdqcr_get(struct qm_portal *portal) +{ + return qm_in(DQRR_VDQCR); +} + +static inline u8 qm_dqrr_get_ithresh(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + return dqrr->ithresh; +} + +static inline void qm_dqrr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + qm_out(DQRR_ITR, ithresh); +} + +static inline u8 qm_dqrr_get_maxfill(struct qm_portal *portal) +{ + return (qm_in(CFG) & 0x00f00000) >> 20; +} + +/* -------------- */ +/* --- MR API --- */ + +#define MR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_MR_SIZE << 6))) + +static inline u8 MR_PTR2IDX(const struct qm_mr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_MR_SIZE - 1); +} + +static inline const struct qm_mr_entry *MR_INC(const struct qm_mr_entry *e) +{ + return MR_CARRYCLEAR(e + 1); +} + +static inline void qm_mr_finish(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + if (mr->ci != MR_PTR2IDX(mr->cursor)) + pr_crit("Ignoring completed MR entries\n"); +} + +static inline const struct qm_mr_entry *qm_mr_current(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + if (!mr->fill) + return NULL; + return mr->cursor; +} + +static inline u8 qm_mr_next(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + DPAA_ASSERT(mr->fill); + mr->cursor = MR_INC(mr->cursor); + return --mr->fill; +} + +static inline void qm_mr_cci_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_mr *mr = &portal->mr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mr->cmode == qm_mr_cci); +#endif + mr->ci = (mr->ci + num) & (QM_MR_SIZE - 1); + qm_out(MR_CI_CINH, mr->ci); +} + +static inline void qm_mr_cci_consume_to_current(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mr->cmode == qm_mr_cci); +#endif + mr->ci = MR_PTR2IDX(mr->cursor); + qm_out(MR_CI_CINH, mr->ci); +} + +static inline void qm_mr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + qm_out(MR_ITR, ithresh); +} + +/* ------------------------------ */ +/* --- Management command API --- */ +static inline int qm_mc_init(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + + mc->cr = portal->addr.ce + QM_CL_CR; + mc->rr = portal->addr.ce + QM_CL_RR0; + mc->rridx = (__raw_readb(&mc->cr->__dont_write_directly__verb) & + QM_MCC_VERB_VBIT) ? 0 : 1; + mc->vbit = mc->rridx ? QM_MCC_VERB_VBIT : 0; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + mc->state = qman_mc_idle; +#endif + return 0; +} + +static inline void qm_mc_finish(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mc *mc = &portal->mc; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == qman_mc_idle); + if (mc->state != qman_mc_idle) + pr_crit("Losing incomplete MC command\n"); +#endif +} + +static inline struct qm_mc_command *qm_mc_start(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == qman_mc_idle); + mc->state = qman_mc_user; +#endif + dcbz_64(mc->cr); + return mc->cr; +} + +static inline void qm_mc_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_mc *mc = &portal->mc; + struct qm_mc_result *rr = mc->rr + mc->rridx; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == qman_mc_user); +#endif + lwsync(); + mc->cr->__dont_write_directly__verb = myverb | mc->vbit; + dcbf(mc->cr); + dcbit_ro(rr); +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + mc->state = qman_mc_hw; +#endif +} + +static inline struct qm_mc_result *qm_mc_result(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + struct qm_mc_result *rr = mc->rr + mc->rridx; + +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + DPAA_ASSERT(mc->state == qman_mc_hw); +#endif + /* The inactive response register's verb byte always returns zero until + * its command is submitted and completed. This includes the valid-bit, + * in case you were wondering. + */ + if (!__raw_readb(&rr->verb)) { + dcbit_ro(rr); + return NULL; + } + mc->rridx ^= 1; + mc->vbit ^= QM_MCC_VERB_VBIT; +#ifdef RTE_LIBRTE_DPAA_HWDEBUG + mc->state = qman_mc_idle; +#endif + return rr; +} + +/* Portal interrupt register API */ +static inline void qm_isr_set_iperiod(struct qm_portal *portal, u16 iperiod) +{ + qm_out(ITPR, iperiod); +} + +static inline u32 __qm_isr_read(struct qm_portal *portal, enum qm_isr_reg n) +{ +#if defined(RTE_ARCH_ARM64) + return __qm_in(&portal->addr, QM_REG_ISR + (n << 6)); +#else + return __qm_in(&portal->addr, QM_REG_ISR + (n << 2)); +#endif +} + +static inline void __qm_isr_write(struct qm_portal *portal, enum qm_isr_reg n, + u32 val) +{ +#if defined(RTE_ARCH_ARM64) + __qm_out(&portal->addr, QM_REG_ISR + (n << 6), val); +#else + __qm_out(&portal->addr, QM_REG_ISR + (n << 2), val); +#endif +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman_driver.c b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman_driver.c new file mode 100644 index 000000000..1166d68e2 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman_driver.c @@ -0,0 +1,353 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017,2019 NXP + * + */ + +#include <fsl_usd.h> +#include <process.h> +#include "qman_priv.h" +#include <sys/ioctl.h> +#include <rte_branch_prediction.h> + +/* Global variable containing revision id (even on non-control plane systems + * where CCSR isn't available). + */ +u16 qman_ip_rev; +u16 qm_channel_pool1 = QMAN_CHANNEL_POOL1; +u16 qm_channel_caam = QMAN_CHANNEL_CAAM; +u16 qm_channel_pme = QMAN_CHANNEL_PME; + +/* Ccsr map address to access ccsrbased register */ +static void *qman_ccsr_map; +/* The qman clock frequency */ +static u32 qman_clk; + +static __thread int qmfd = -1; +static __thread struct qm_portal_config qpcfg; +static __thread struct dpaa_ioctl_portal_map map = { + .type = dpaa_portal_qman +}; + +u16 dpaa_get_qm_channel_caam(void) +{ + return qm_channel_caam; +} + +u16 dpaa_get_qm_channel_pool(void) +{ + return qm_channel_pool1; +} + +static int fsl_qman_portal_init(uint32_t index, int is_shared) +{ + struct qman_portal *portal; + struct dpaa_ioctl_irq_map irq_map; + int ret; + + /* Allocate and map a qman portal */ + map.index = index; + ret = process_portal_map(&map); + if (ret) { + error(0, ret, "process_portal_map()"); + return ret; + } + qpcfg.channel = map.channel; + qpcfg.pools = map.pools; + qpcfg.index = map.index; + + /* Make the portal's cache-[enabled|inhibited] regions */ + qpcfg.addr_virt[DPAA_PORTAL_CE] = map.addr.cena; + qpcfg.addr_virt[DPAA_PORTAL_CI] = map.addr.cinh; + + qmfd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY); + if (qmfd == -1) { + pr_err("QMan irq init failed\n"); + process_portal_unmap(&map.addr); + return -EBUSY; + } + + qpcfg.is_shared = is_shared; + qpcfg.node = NULL; + qpcfg.irq = qmfd; + + portal = qman_create_affine_portal(&qpcfg, NULL); + if (!portal) { + pr_err("Qman portal initialisation failed (%d)\n", + qpcfg.cpu); + process_portal_unmap(&map.addr); + return -EBUSY; + } + + irq_map.type = dpaa_portal_qman; + irq_map.portal_cinh = map.addr.cinh; + process_portal_irq_map(qmfd, &irq_map); + return 0; +} + +static int fsl_qman_portal_finish(void) +{ + __maybe_unused const struct qm_portal_config *cfg; + int ret; + + process_portal_irq_unmap(qmfd); + + cfg = qman_destroy_affine_portal(NULL); + DPAA_BUG_ON(cfg != &qpcfg); + ret = process_portal_unmap(&map.addr); + if (ret) + error(0, ret, "process_portal_unmap()"); + return ret; +} + +int qman_thread_fd(void) +{ + return qmfd; +} + +int qman_thread_init(void) +{ + /* Convert from contiguous/virtual cpu numbering to real cpu when + * calling into the code that is dependent on the device naming. + */ + return fsl_qman_portal_init(QBMAN_ANY_PORTAL_IDX, 0); +} + +int qman_thread_finish(void) +{ + return fsl_qman_portal_finish(); +} + +void qman_thread_irq(void) +{ + qbman_invoke_irq(qpcfg.irq); + + /* Now we need to uninhibit interrupts. This is the only code outside + * the regular portal driver that manipulates any portal register, so + * rather than breaking that encapsulation I am simply hard-coding the + * offset to the inhibit register here. + */ + out_be32(qpcfg.addr_virt[DPAA_PORTAL_CI] + 0x36C0, 0); +} + +void qman_fq_portal_thread_irq(struct qman_portal *qp) +{ + qman_portal_uninhibit_isr(qp); +} + +struct qman_portal *fsl_qman_fq_portal_create(int *fd) +{ + struct qman_portal *portal = NULL; + struct qm_portal_config *q_pcfg; + struct dpaa_ioctl_irq_map irq_map; + struct dpaa_ioctl_portal_map q_map = {0}; + int q_fd = 0, ret; + + q_pcfg = kzalloc((sizeof(struct qm_portal_config)), 0); + if (!q_pcfg) { + error(0, -1, "q_pcfg kzalloc failed"); + return NULL; + } + + /* Allocate and map a qman portal */ + q_map.type = dpaa_portal_qman; + q_map.index = QBMAN_ANY_PORTAL_IDX; + ret = process_portal_map(&q_map); + if (ret) { + error(0, ret, "process_portal_map()"); + kfree(q_pcfg); + return NULL; + } + q_pcfg->channel = q_map.channel; + q_pcfg->pools = q_map.pools; + q_pcfg->index = q_map.index; + + /* Make the portal's cache-[enabled|inhibited] regions */ + q_pcfg->addr_virt[DPAA_PORTAL_CE] = q_map.addr.cena; + q_pcfg->addr_virt[DPAA_PORTAL_CI] = q_map.addr.cinh; + + q_fd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY); + if (q_fd == -1) { + pr_err("QMan irq init failed\n"); + goto err; + } + + q_pcfg->irq = q_fd; + + portal = qman_alloc_global_portal(q_pcfg); + if (!portal) { + pr_err("Qman portal initialisation failed (%d)\n", + q_pcfg->cpu); + goto err; + } + + irq_map.type = dpaa_portal_qman; + irq_map.portal_cinh = q_map.addr.cinh; + process_portal_irq_map(q_fd, &irq_map); + + *fd = q_fd; + return portal; +err: + if (portal) + qman_free_global_portal(portal); + if (q_fd) + close(q_fd); + process_portal_unmap(&q_map.addr); + kfree(q_pcfg); + return NULL; +} + +int fsl_qman_fq_portal_init(struct qman_portal *qp) +{ + struct qman_portal *res; + + res = qman_init_portal(qp, NULL, NULL); + if (!res) { + pr_err("Qman portal initialisation failed\n"); + return -1; + } + + return 0; +} + +int fsl_qman_fq_portal_destroy(struct qman_portal *qp) +{ + const struct qm_portal_config *cfg; + struct dpaa_portal_map addr; + int ret; + + cfg = qman_destroy_affine_portal(qp); + + ret = qman_free_global_portal(qp); + if (ret) + pr_err("qman_free_global_portal() (%d)\n", ret); + + kfree(qp); + + process_portal_irq_unmap(cfg->irq); + + addr.cena = cfg->addr_virt[DPAA_PORTAL_CE]; + addr.cinh = cfg->addr_virt[DPAA_PORTAL_CI]; + + ret = process_portal_unmap(&addr); + if (ret) + pr_err("process_portal_unmap() (%d)\n", ret); + + kfree((void *)cfg); + + return ret; +} + +int qman_global_init(void) +{ + const struct device_node *dt_node; + size_t lenp; + const u32 *chanid; + static int ccsr_map_fd; + const uint32_t *qman_addr; + uint64_t phys_addr; + uint64_t regs_size; + const u32 *clk; + + static int done; + + if (done) + return -EBUSY; + + /* Use the device-tree to determine IP revision until something better + * is devised. + */ + dt_node = of_find_compatible_node(NULL, NULL, "fsl,qman-portal"); + if (!dt_node) { + pr_err("No qman portals available for any CPU\n"); + return -ENODEV; + } + if (of_device_is_compatible(dt_node, "fsl,qman-portal-1.0") || + of_device_is_compatible(dt_node, "fsl,qman-portal-1.0.0")) + pr_err("QMan rev1.0 on P4080 rev1 is not supported!\n"); + else if (of_device_is_compatible(dt_node, "fsl,qman-portal-1.1") || + of_device_is_compatible(dt_node, "fsl,qman-portal-1.1.0")) + qman_ip_rev = QMAN_REV11; + else if (of_device_is_compatible(dt_node, "fsl,qman-portal-1.2") || + of_device_is_compatible(dt_node, "fsl,qman-portal-1.2.0")) + qman_ip_rev = QMAN_REV12; + else if (of_device_is_compatible(dt_node, "fsl,qman-portal-2.0") || + of_device_is_compatible(dt_node, "fsl,qman-portal-2.0.0")) + qman_ip_rev = QMAN_REV20; + else if (of_device_is_compatible(dt_node, "fsl,qman-portal-3.0.0") || + of_device_is_compatible(dt_node, "fsl,qman-portal-3.0.1")) + qman_ip_rev = QMAN_REV30; + else if (of_device_is_compatible(dt_node, "fsl,qman-portal-3.1.0") || + of_device_is_compatible(dt_node, "fsl,qman-portal-3.1.1") || + of_device_is_compatible(dt_node, "fsl,qman-portal-3.1.2") || + of_device_is_compatible(dt_node, "fsl,qman-portal-3.1.3")) + qman_ip_rev = QMAN_REV31; + else if (of_device_is_compatible(dt_node, "fsl,qman-portal-3.2.0") || + of_device_is_compatible(dt_node, "fsl,qman-portal-3.2.1")) + qman_ip_rev = QMAN_REV32; + else + qman_ip_rev = QMAN_REV11; + + if (!qman_ip_rev) { + pr_err("Unknown qman portal version\n"); + return -ENODEV; + } + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) { + qm_channel_pool1 = QMAN_CHANNEL_POOL1_REV3; + qm_channel_caam = QMAN_CHANNEL_CAAM_REV3; + qm_channel_pme = QMAN_CHANNEL_PME_REV3; + } + + dt_node = of_find_compatible_node(NULL, NULL, "fsl,pool-channel-range"); + if (!dt_node) { + pr_err("No qman pool channel range available\n"); + return -ENODEV; + } + chanid = of_get_property(dt_node, "fsl,pool-channel-range", &lenp); + if (!chanid) { + pr_err("Can not get pool-channel-range property\n"); + return -EINVAL; + } + + /* get ccsr base */ + dt_node = of_find_compatible_node(NULL, NULL, "fsl,qman"); + if (!dt_node) { + pr_err("No qman device node available\n"); + return -ENODEV; + } + qman_addr = of_get_address(dt_node, 0, ®s_size, NULL); + if (!qman_addr) { + pr_err("of_get_address cannot return qman address\n"); + return -EINVAL; + } + phys_addr = of_translate_address(dt_node, qman_addr); + if (!phys_addr) { + pr_err("of_translate_address failed\n"); + return -EINVAL; + } + + ccsr_map_fd = open("/dev/mem", O_RDWR); + if (unlikely(ccsr_map_fd < 0)) { + pr_err("Can not open /dev/mem for qman ccsr map\n"); + return ccsr_map_fd; + } + + qman_ccsr_map = mmap(NULL, regs_size, PROT_READ | PROT_WRITE, + MAP_SHARED, ccsr_map_fd, phys_addr); + if (qman_ccsr_map == MAP_FAILED) { + pr_err("Can not map qman ccsr base\n"); + return -EINVAL; + } + + clk = of_get_property(dt_node, "clock-frequency", NULL); + if (!clk) + pr_warn("Can't find Qman clock frequency\n"); + else + qman_clk = be32_to_cpu(*clk); + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + return qman_setup_fq_lookup_table(CONFIG_FSL_QMAN_FQ_LOOKUP_MAX); +#endif + return 0; +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman_priv.h b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman_priv.h new file mode 100644 index 000000000..8254729e6 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/base/qbman/qman_priv.h @@ -0,0 +1,287 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2016 Freescale Semiconductor Inc. + * Copyright 2017,2019 NXP + * + */ + +#ifndef __QMAN_PRIV_H +#define __QMAN_PRIV_H + +#include "dpaa_sys.h" +#include <fsl_qman.h> + +/* Congestion Groups */ +/* + * This wrapper represents a bit-array for the state of the 256 QMan congestion + * groups. Is also used as a *mask* for congestion groups, eg. so we ignore + * those that don't concern us. We harness the structure and accessor details + * already used in the management command to query congestion groups. + */ +struct qman_cgrs { + struct __qm_mcr_querycongestion q; +}; + +static inline void qman_cgrs_init(struct qman_cgrs *c) +{ + memset(c, 0, sizeof(*c)); +} + +static inline void qman_cgrs_fill(struct qman_cgrs *c) +{ + memset(c, 0xff, sizeof(*c)); +} + +static inline int qman_cgrs_get(struct qman_cgrs *c, int num) +{ + return QM_MCR_QUERYCONGESTION(&c->q, num); +} + +static inline void qman_cgrs_set(struct qman_cgrs *c, int num) +{ + c->q.state[__CGR_WORD(num)] |= (0x80000000 >> __CGR_SHIFT(num)); +} + +static inline void qman_cgrs_unset(struct qman_cgrs *c, int num) +{ + c->q.state[__CGR_WORD(num)] &= ~(0x80000000 >> __CGR_SHIFT(num)); +} + +static inline int qman_cgrs_next(struct qman_cgrs *c, int num) +{ + while ((++num < (int)__CGR_NUM) && !qman_cgrs_get(c, num)) + ; + return num; +} + +static inline void qman_cgrs_cp(struct qman_cgrs *dest, + const struct qman_cgrs *src) +{ + memcpy(dest, src, sizeof(*dest)); +} + +static inline void qman_cgrs_and(struct qman_cgrs *dest, + const struct qman_cgrs *a, + const struct qman_cgrs *b) +{ + int ret; + u32 *_d = dest->q.state; + const u32 *_a = a->q.state; + const u32 *_b = b->q.state; + + for (ret = 0; ret < 8; ret++) + *(_d++) = *(_a++) & *(_b++); +} + +static inline void qman_cgrs_xor(struct qman_cgrs *dest, + const struct qman_cgrs *a, + const struct qman_cgrs *b) +{ + int ret; + u32 *_d = dest->q.state; + const u32 *_a = a->q.state; + const u32 *_b = b->q.state; + + for (ret = 0; ret < 8; ret++) + *(_d++) = *(_a++) ^ *(_b++); +} + +/* used by CCSR and portal interrupt code */ +enum qm_isr_reg { + qm_isr_status = 0, + qm_isr_enable = 1, + qm_isr_disable = 2, + qm_isr_inhibit = 3 +}; + +struct qm_portal_config { + /* + * Corenet portal addresses; + * [0]==cache-enabled, [1]==cache-inhibited. + */ + void __iomem *addr_virt[2]; + struct device_node *node; + /* Allow these to be joined in lists */ + struct list_head list; + /* User-visible portal configuration settings */ + /* If the caller enables DQRR stashing (and thus wishes to operate the + * portal from only one cpu), this is the logical CPU that the portal + * will stash to. Whether stashing is enabled or not, this setting is + * also used for any "core-affine" portals, ie. default portals + * associated to the corresponding cpu. -1 implies that there is no + * core affinity configured. + */ + int cpu; + /* portal interrupt line */ + int irq; + /* the unique index of this portal */ + u32 index; + /* Is this portal shared? (If so, it has coarser locking and demuxes + * processing on behalf of other CPUs.). + */ + int is_shared; + /* The portal's dedicated channel id, use this value for initialising + * frame queues to target this portal when scheduled. + */ + u16 channel; + /* A mask of which pool channels this portal has dequeue access to + * (using QM_SDQCR_CHANNELS_POOL(n) for the bitmask). + */ + u32 pools; + +}; + +/* Revision info (for errata and feature handling) */ +#define QMAN_REV11 0x0101 +#define QMAN_REV12 0x0102 +#define QMAN_REV20 0x0200 +#define QMAN_REV30 0x0300 +#define QMAN_REV31 0x0301 +#define QMAN_REV32 0x0302 +extern u16 qman_ip_rev; /* 0 if uninitialised, otherwise QMAN_REVx */ + +int qm_set_wpm(int wpm); +int qm_get_wpm(int *wpm); + +struct qman_portal *qman_create_affine_portal( + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs); +const struct qm_portal_config * +qman_destroy_affine_portal(struct qman_portal *q); + +struct qman_portal * +qman_init_portal(struct qman_portal *portal, + const struct qm_portal_config *c, + const struct qman_cgrs *cgrs); + +struct qman_portal *qman_alloc_global_portal(struct qm_portal_config *q_pcfg); +int qman_free_global_portal(struct qman_portal *portal); + +void qman_portal_uninhibit_isr(struct qman_portal *portal); + +struct qm_portal_config *qm_get_unused_portal(void); +struct qm_portal_config *qm_get_unused_portal_idx(uint32_t idx); + +void qm_put_unused_portal(struct qm_portal_config *pcfg); +void qm_set_liodns(struct qm_portal_config *pcfg); + +/* This CGR feature is supported by h/w and required by unit-tests and the + * debugfs hooks, so is implemented in the driver. However it allows an explicit + * corruption of h/w fields by s/w that are usually incorruptible (because the + * counters are usually maintained entirely within h/w). As such, we declare + * this API internally. + */ +int qman_testwrite_cgr(struct qman_cgr *cgr, u64 i_bcnt, + struct qm_mcr_cgrtestwrite *result); + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +/* If the fq object pointer is greater than the size of context_b field, + * than a lookup table is required. + */ +int qman_setup_fq_lookup_table(size_t num_entries); +#endif + +/* QMan s/w corenet portal, low-level i/face */ + +/* + * For Choose one SOURCE. Choose one COUNT. Choose one + * dequeue TYPE. Choose TOKEN (8-bit). + * If SOURCE == CHANNELS, + * Choose CHANNELS_DEDICATED and/or CHANNELS_POOL(n). + * You can choose DEDICATED_PRECEDENCE if the portal channel should have + * priority. + * If SOURCE == SPECIFICWQ, + * Either select the work-queue ID with SPECIFICWQ_WQ(), or select the + * channel (SPECIFICWQ_DEDICATED or SPECIFICWQ_POOL()) and specify the + * work-queue priority (0-7) with SPECIFICWQ_WQ() - either way, you get the + * same value. + */ +#define QM_SDQCR_SOURCE_CHANNELS 0x0 +#define QM_SDQCR_SOURCE_SPECIFICWQ 0x40000000 +#define QM_SDQCR_COUNT_EXACT1 0x0 +#define QM_SDQCR_COUNT_UPTO3 0x20000000 +#define QM_SDQCR_DEDICATED_PRECEDENCE 0x10000000 +#define QM_SDQCR_TYPE_MASK 0x03000000 +#define QM_SDQCR_TYPE_NULL 0x0 +#define QM_SDQCR_TYPE_PRIO_QOS 0x01000000 +#define QM_SDQCR_TYPE_ACTIVE_QOS 0x02000000 +#define QM_SDQCR_TYPE_ACTIVE 0x03000000 +#define QM_SDQCR_TOKEN_MASK 0x00ff0000 +#define QM_SDQCR_TOKEN_SET(v) (((v) & 0xff) << 16) +#define QM_SDQCR_TOKEN_GET(v) (((v) >> 16) & 0xff) +#define QM_SDQCR_CHANNELS_DEDICATED 0x00008000 +#define QM_SDQCR_SPECIFICWQ_MASK 0x000000f7 +#define QM_SDQCR_SPECIFICWQ_DEDICATED 0x00000000 +#define QM_SDQCR_SPECIFICWQ_POOL(n) ((n) << 4) +#define QM_SDQCR_SPECIFICWQ_WQ(n) (n) + +#define QM_VDQCR_FQID_MASK 0x00ffffff +#define QM_VDQCR_FQID(n) ((n) & QM_VDQCR_FQID_MASK) + +#define QM_EQCR_VERB_VBIT 0x80 +#define QM_EQCR_VERB_CMD_MASK 0x61 /* but only one value; */ +#define QM_EQCR_VERB_CMD_ENQUEUE 0x01 +#define QM_EQCR_VERB_COLOUR_MASK 0x18 /* 4 possible values; */ +#define QM_EQCR_VERB_COLOUR_GREEN 0x00 +#define QM_EQCR_VERB_COLOUR_YELLOW 0x08 +#define QM_EQCR_VERB_COLOUR_RED 0x10 +#define QM_EQCR_VERB_COLOUR_OVERRIDE 0x18 +#define QM_EQCR_VERB_INTERRUPT 0x04 /* on command consumption */ +#define QM_EQCR_VERB_ORP 0x02 /* enable order restoration */ +#define QM_EQCR_DCA_ENABLE 0x80 +#define QM_EQCR_DCA_PARK 0x40 +#define QM_EQCR_DCA_IDXMASK 0x0f /* "DQRR::idx" goes here */ +#define QM_EQCR_SEQNUM_NESN 0x8000 /* Advance NESN */ +#define QM_EQCR_SEQNUM_NLIS 0x4000 /* More fragments to come */ +#define QM_EQCR_SEQNUM_SEQMASK 0x3fff /* sequence number goes here */ +#define QM_EQCR_FQID_NULL 0 /* eg. for an ORP seqnum hole */ + +#define QM_MCC_VERB_VBIT 0x80 +#define QM_MCC_VERB_MASK 0x7f /* where the verb contains; */ +#define QM_MCC_VERB_INITFQ_PARKED 0x40 +#define QM_MCC_VERB_INITFQ_SCHED 0x41 +#define QM_MCC_VERB_QUERYFQ 0x44 +#define QM_MCC_VERB_QUERYFQ_NP 0x45 /* "non-programmable" fields */ +#define QM_MCC_VERB_QUERYWQ 0x46 +#define QM_MCC_VERB_QUERYWQ_DEDICATED 0x47 +#define QM_MCC_VERB_ALTER_SCHED 0x48 /* Schedule FQ */ +#define QM_MCC_VERB_ALTER_FE 0x49 /* Force Eligible FQ */ +#define QM_MCC_VERB_ALTER_RETIRE 0x4a /* Retire FQ */ +#define QM_MCC_VERB_ALTER_OOS 0x4b /* Take FQ out of service */ +#define QM_MCC_VERB_ALTER_FQXON 0x4d /* FQ XON */ +#define QM_MCC_VERB_ALTER_FQXOFF 0x4e /* FQ XOFF */ +#define QM_MCC_VERB_INITCGR 0x50 +#define QM_MCC_VERB_MODIFYCGR 0x51 +#define QM_MCC_VERB_CGRTESTWRITE 0x52 +#define QM_MCC_VERB_QUERYCGR 0x58 +#define QM_MCC_VERB_QUERYCONGESTION 0x59 + +/* + * Used by all portal interrupt registers except 'inhibit' + * Channels with frame availability + */ +#define QM_PIRQ_DQAVAIL 0x0000ffff + +/* The DQAVAIL interrupt fields break down into these bits; */ +#define QM_DQAVAIL_PORTAL 0x8000 /* Portal channel */ +#define QM_DQAVAIL_POOL(n) (0x8000 >> (n)) /* Pool channel, n==[1..15] */ +#define QM_DQAVAIL_MASK 0xffff +/* This mask contains all the "irqsource" bits visible to API users */ +#define QM_PIRQ_VISIBLE (QM_PIRQ_SLOW | QM_PIRQ_DQRI) + +/* These are qm_<reg>_<verb>(). So for example, qm_disable_write() means "write + * the disable register" rather than "disable the ability to write". + */ +#define qm_isr_status_read(qm) __qm_isr_read(qm, qm_isr_status) +#define qm_isr_status_clear(qm, m) __qm_isr_write(qm, qm_isr_status, m) +#define qm_isr_enable_read(qm) __qm_isr_read(qm, qm_isr_enable) +#define qm_isr_enable_write(qm, v) __qm_isr_write(qm, qm_isr_enable, v) +#define qm_isr_disable_read(qm) __qm_isr_read(qm, qm_isr_disable) +#define qm_isr_disable_write(qm, v) __qm_isr_write(qm, qm_isr_disable, v) +/* TODO: unfortunate name-clash here, reword? */ +#define qm_isr_inhibit(qm) __qm_isr_write(qm, qm_isr_inhibit, 1) +#define qm_isr_uninhibit(qm) __qm_isr_write(qm, qm_isr_inhibit, 0) + +#define QMAN_PORTAL_IRQ_PATH "/dev/fsl-usdpaa-irq" + +#endif /* _QMAN_PRIV_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/dpaa_bus.c b/src/spdk/dpdk/drivers/bus/dpaa/dpaa_bus.c new file mode 100644 index 000000000..d53fe6083 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/dpaa_bus.c @@ -0,0 +1,758 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * + * Copyright 2017-2019 NXP + * + */ +/* System headers */ +#include <stdio.h> +#include <inttypes.h> +#include <unistd.h> +#include <limits.h> +#include <sched.h> +#include <signal.h> +#include <pthread.h> +#include <sys/types.h> +#include <sys/syscall.h> + +#include <rte_byteorder.h> +#include <rte_common.h> +#include <rte_interrupts.h> +#include <rte_log.h> +#include <rte_debug.h> +#include <rte_atomic.h> +#include <rte_branch_prediction.h> +#include <rte_memory.h> +#include <rte_tailq.h> +#include <rte_eal.h> +#include <rte_alarm.h> +#include <rte_ether.h> +#include <rte_ethdev_driver.h> +#include <rte_malloc.h> +#include <rte_ring.h> +#include <rte_bus.h> +#include <rte_mbuf_pool_ops.h> + +#include <dpaa_of.h> +#include <rte_dpaa_bus.h> +#include <rte_dpaa_logs.h> +#include <dpaax_iova_table.h> + +#include <fsl_usd.h> +#include <fsl_qman.h> +#include <fsl_bman.h> +#include <netcfg.h> + +int dpaa_logtype_bus; + +static struct rte_dpaa_bus rte_dpaa_bus; +struct netcfg_info *dpaa_netcfg; + +/* define a variable to hold the portal_key, once created.*/ +static pthread_key_t dpaa_portal_key; + +unsigned int dpaa_svr_family; + +#define FSL_DPAA_BUS_NAME dpaa_bus + +RTE_DEFINE_PER_LCORE(bool, dpaa_io); +RTE_DEFINE_PER_LCORE(struct dpaa_portal_dqrr, held_bufs); + +struct fm_eth_port_cfg * +dpaa_get_eth_port_cfg(int dev_id) +{ + return &dpaa_netcfg->port_cfg[dev_id]; +} + +static int +compare_dpaa_devices(struct rte_dpaa_device *dev1, + struct rte_dpaa_device *dev2) +{ + int comp = 0; + + /* Segragating ETH from SEC devices */ + if (dev1->device_type > dev2->device_type) + comp = 1; + else if (dev1->device_type < dev2->device_type) + comp = -1; + else + comp = 0; + + if ((comp != 0) || (dev1->device_type != FSL_DPAA_ETH)) + return comp; + + if (dev1->id.fman_id > dev2->id.fman_id) { + comp = 1; + } else if (dev1->id.fman_id < dev2->id.fman_id) { + comp = -1; + } else { + /* FMAN ids match, check for mac_id */ + if (dev1->id.mac_id > dev2->id.mac_id) + comp = 1; + else if (dev1->id.mac_id < dev2->id.mac_id) + comp = -1; + else + comp = 0; + } + + return comp; +} + +static inline void +dpaa_add_to_device_list(struct rte_dpaa_device *newdev) +{ + int comp, inserted = 0; + struct rte_dpaa_device *dev = NULL; + struct rte_dpaa_device *tdev = NULL; + + TAILQ_FOREACH_SAFE(dev, &rte_dpaa_bus.device_list, next, tdev) { + comp = compare_dpaa_devices(newdev, dev); + if (comp < 0) { + TAILQ_INSERT_BEFORE(dev, newdev, next); + inserted = 1; + break; + } + } + + if (!inserted) + TAILQ_INSERT_TAIL(&rte_dpaa_bus.device_list, newdev, next); +} + +/* + * Reads the SEC device from DTS + * Returns -1 if SEC devices not available, 0 otherwise + */ +static inline int +dpaa_sec_available(void) +{ + const struct device_node *caam_node; + + for_each_compatible_node(caam_node, NULL, "fsl,sec-v4.0") { + return 0; + } + + return -1; +} + +static void dpaa_clean_device_list(void); + +static struct rte_devargs * +dpaa_devargs_lookup(struct rte_dpaa_device *dev) +{ + struct rte_devargs *devargs; + char dev_name[32]; + + RTE_EAL_DEVARGS_FOREACH("dpaa_bus", devargs) { + devargs->bus->parse(devargs->name, &dev_name); + if (strcmp(dev_name, dev->device.name) == 0) { + DPAA_BUS_INFO("**Devargs matched %s", dev_name); + return devargs; + } + } + return NULL; +} + +static int +dpaa_create_device_list(void) +{ + int i; + int ret; + struct rte_dpaa_device *dev; + struct fm_eth_port_cfg *cfg; + struct fman_if *fman_intf; + + /* Creating Ethernet Devices */ + for (i = 0; i < dpaa_netcfg->num_ethports; i++) { + dev = calloc(1, sizeof(struct rte_dpaa_device)); + if (!dev) { + DPAA_BUS_LOG(ERR, "Failed to allocate ETH devices"); + ret = -ENOMEM; + goto cleanup; + } + + dev->device.bus = &rte_dpaa_bus.bus; + + cfg = &dpaa_netcfg->port_cfg[i]; + fman_intf = cfg->fman_if; + + /* Device identifiers */ + dev->id.fman_id = fman_intf->fman_idx + 1; + dev->id.mac_id = fman_intf->mac_idx; + dev->device_type = FSL_DPAA_ETH; + dev->id.dev_id = i; + + /* Create device name */ + memset(dev->name, 0, RTE_ETH_NAME_MAX_LEN); + sprintf(dev->name, "fm%d-mac%d", (fman_intf->fman_idx + 1), + fman_intf->mac_idx); + DPAA_BUS_LOG(INFO, "%s netdev added", dev->name); + dev->device.name = dev->name; + dev->device.devargs = dpaa_devargs_lookup(dev); + + dpaa_add_to_device_list(dev); + } + + rte_dpaa_bus.device_count = i; + + /* Unlike case of ETH, RTE_LIBRTE_DPAA_MAX_CRYPTODEV SEC devices are + * constantly created only if "sec" property is found in the device + * tree. Logically there is no limit for number of devices (QI + * interfaces) that can be created. + */ + + if (dpaa_sec_available()) { + DPAA_BUS_LOG(INFO, "DPAA SEC devices are not available"); + return 0; + } + + /* Creating SEC Devices */ + for (i = 0; i < RTE_LIBRTE_DPAA_MAX_CRYPTODEV; i++) { + dev = calloc(1, sizeof(struct rte_dpaa_device)); + if (!dev) { + DPAA_BUS_LOG(ERR, "Failed to allocate SEC devices"); + ret = -1; + goto cleanup; + } + + dev->device_type = FSL_DPAA_CRYPTO; + dev->id.dev_id = rte_dpaa_bus.device_count + i; + + /* Even though RTE_CRYPTODEV_NAME_MAX_LEN is valid length of + * crypto PMD, using RTE_ETH_NAME_MAX_LEN as that is the size + * allocated for dev->name/ + */ + memset(dev->name, 0, RTE_ETH_NAME_MAX_LEN); + sprintf(dev->name, "dpaa_sec-%d", i+1); + DPAA_BUS_LOG(INFO, "%s cryptodev added", dev->name); + dev->device.name = dev->name; + dev->device.devargs = dpaa_devargs_lookup(dev); + + dpaa_add_to_device_list(dev); + } + + rte_dpaa_bus.device_count += i; + + return 0; + +cleanup: + dpaa_clean_device_list(); + return ret; +} + +static void +dpaa_clean_device_list(void) +{ + struct rte_dpaa_device *dev = NULL; + struct rte_dpaa_device *tdev = NULL; + + TAILQ_FOREACH_SAFE(dev, &rte_dpaa_bus.device_list, next, tdev) { + TAILQ_REMOVE(&rte_dpaa_bus.device_list, dev, next); + free(dev); + dev = NULL; + } +} + +int rte_dpaa_portal_init(void *arg) +{ + unsigned int cpu, lcore = rte_lcore_id(); + int ret; + struct dpaa_portal *dpaa_io_portal; + + BUS_INIT_FUNC_TRACE(); + + if ((size_t)arg == 1 || lcore == LCORE_ID_ANY) + lcore = rte_get_master_lcore(); + else + if (lcore >= RTE_MAX_LCORE) + return -1; + + cpu = rte_lcore_to_cpu_id(lcore); + + /* Initialise bman thread portals */ + ret = bman_thread_init(); + if (ret) { + DPAA_BUS_LOG(ERR, "bman_thread_init failed on core %u" + " (lcore=%u) with ret: %d", cpu, lcore, ret); + return ret; + } + + DPAA_BUS_LOG(DEBUG, "BMAN thread initialized - CPU=%d lcore=%d", + cpu, lcore); + + /* Initialise qman thread portals */ + ret = qman_thread_init(); + if (ret) { + DPAA_BUS_LOG(ERR, "qman_thread_init failed on core %u" + " (lcore=%u) with ret: %d", cpu, lcore, ret); + bman_thread_finish(); + return ret; + } + + DPAA_BUS_LOG(DEBUG, "QMAN thread initialized - CPU=%d lcore=%d", + cpu, lcore); + + dpaa_io_portal = rte_malloc(NULL, sizeof(struct dpaa_portal), + RTE_CACHE_LINE_SIZE); + if (!dpaa_io_portal) { + DPAA_BUS_LOG(ERR, "Unable to allocate memory"); + bman_thread_finish(); + qman_thread_finish(); + return -ENOMEM; + } + + dpaa_io_portal->qman_idx = qman_get_portal_index(); + dpaa_io_portal->bman_idx = bman_get_portal_index(); + dpaa_io_portal->tid = syscall(SYS_gettid); + + ret = pthread_setspecific(dpaa_portal_key, (void *)dpaa_io_portal); + if (ret) { + DPAA_BUS_LOG(ERR, "pthread_setspecific failed on core %u" + " (lcore=%u) with ret: %d", cpu, lcore, ret); + dpaa_portal_finish(NULL); + + return ret; + } + + RTE_PER_LCORE(dpaa_io) = true; + + DPAA_BUS_LOG(DEBUG, "QMAN thread initialized"); + + return 0; +} + +int +rte_dpaa_portal_fq_init(void *arg, struct qman_fq *fq) +{ + /* Affine above created portal with channel*/ + u32 sdqcr; + int ret; + + if (unlikely(!RTE_PER_LCORE(dpaa_io))) { + ret = rte_dpaa_portal_init(arg); + if (ret < 0) { + DPAA_BUS_LOG(ERR, "portal initialization failure"); + return ret; + } + } + + /* Initialise qman specific portals */ + ret = fsl_qman_fq_portal_init(fq->qp); + if (ret) { + DPAA_BUS_LOG(ERR, "Unable to init fq portal"); + return -1; + } + + sdqcr = QM_SDQCR_CHANNELS_POOL_CONV(fq->ch_id); + qman_static_dequeue_add(sdqcr, fq->qp); + + return 0; +} + +int rte_dpaa_portal_fq_close(struct qman_fq *fq) +{ + return fsl_qman_fq_portal_destroy(fq->qp); +} + +void +dpaa_portal_finish(void *arg) +{ + struct dpaa_portal *dpaa_io_portal = (struct dpaa_portal *)arg; + + if (!dpaa_io_portal) { + DPAA_BUS_LOG(DEBUG, "Portal already cleaned"); + return; + } + + bman_thread_finish(); + qman_thread_finish(); + + pthread_setspecific(dpaa_portal_key, NULL); + + rte_free(dpaa_io_portal); + dpaa_io_portal = NULL; + + RTE_PER_LCORE(dpaa_io) = false; +} + +static int +rte_dpaa_bus_parse(const char *name, void *out_name) +{ + int i, j; + int max_fman = 2, max_macs = 16; + char *dup_name; + char *sep = NULL; + + /* There are two ways of passing device name, with and without + * separator. "dpaa_bus:fm1-mac3" with separator, and "fm1-mac3" + * without separator. Both need to be handled. + * It is also possible that "name=fm1-mac3" is passed along. + */ + DPAA_BUS_DEBUG("Parse device name (%s)", name); + + /* Check for dpaa_bus:fm1-mac3 style */ + dup_name = strdup(name); + sep = strchr(dup_name, ':'); + if (!sep) + /* If not, check for name=fm1-mac3 style */ + sep = strchr(dup_name, '='); + + if (sep) + /* jump over the seprator */ + sep = (char *) (sep + 1); + else + sep = dup_name; + + for (i = 0; i < max_fman; i++) { + for (j = 0; j < max_macs; j++) { + char fm_name[16]; + snprintf(fm_name, 16, "fm%d-mac%d", i, j); + if (strcmp(fm_name, sep) == 0) { + if (out_name) + strcpy(out_name, sep); + free(dup_name); + return 0; + } + } + } + + for (i = 0; i < RTE_LIBRTE_DPAA_MAX_CRYPTODEV; i++) { + char sec_name[16]; + + snprintf(sec_name, 16, "dpaa_sec-%d", i+1); + if (strcmp(sec_name, sep) == 0) { + if (out_name) + strcpy(out_name, sep); + free(dup_name); + return 0; + } + } + + free(dup_name); + return -EINVAL; +} + +#define DPAA_DEV_PATH1 "/sys/devices/platform/soc/soc:fsl,dpaa" +#define DPAA_DEV_PATH2 "/sys/devices/platform/fsl,dpaa" + +static int +rte_dpaa_bus_scan(void) +{ + int ret; + + BUS_INIT_FUNC_TRACE(); + + if ((access(DPAA_DEV_PATH1, F_OK) != 0) && + (access(DPAA_DEV_PATH2, F_OK) != 0)) { + RTE_LOG(DEBUG, EAL, "DPAA Bus not present. Skipping.\n"); + return 0; + } + + if (rte_dpaa_bus.detected) + return 0; + + rte_dpaa_bus.detected = 1; + + /* create the key, supplying a function that'll be invoked + * when a portal affined thread will be deleted. + */ + ret = pthread_key_create(&dpaa_portal_key, dpaa_portal_finish); + if (ret) { + DPAA_BUS_LOG(DEBUG, "Unable to create pthread key. (%d)", ret); + dpaa_clean_device_list(); + return ret; + } + + return 0; +} + +/* register a dpaa bus based dpaa driver */ +void +rte_dpaa_driver_register(struct rte_dpaa_driver *driver) +{ + RTE_VERIFY(driver); + + BUS_INIT_FUNC_TRACE(); + + TAILQ_INSERT_TAIL(&rte_dpaa_bus.driver_list, driver, next); + /* Update Bus references */ + driver->dpaa_bus = &rte_dpaa_bus; +} + +/* un-register a dpaa bus based dpaa driver */ +void +rte_dpaa_driver_unregister(struct rte_dpaa_driver *driver) +{ + struct rte_dpaa_bus *dpaa_bus; + + BUS_INIT_FUNC_TRACE(); + + dpaa_bus = driver->dpaa_bus; + + TAILQ_REMOVE(&dpaa_bus->driver_list, driver, next); + /* Update Bus references */ + driver->dpaa_bus = NULL; +} + +static int +rte_dpaa_device_match(struct rte_dpaa_driver *drv, + struct rte_dpaa_device *dev) +{ + if (!drv || !dev) { + DPAA_BUS_DEBUG("Invalid drv or dev received."); + return -1; + } + + if (drv->drv_type == dev->device_type) + return 0; + + return -1; +} + +static int +rte_dpaa_bus_dev_build(void) +{ + int ret; + + /* Load the device-tree driver */ + ret = of_init(); + if (ret) { + DPAA_BUS_LOG(ERR, "of_init failed with ret: %d", ret); + return -1; + } + + /* Get the interface configurations from device-tree */ + dpaa_netcfg = netcfg_acquire(); + if (!dpaa_netcfg) { + DPAA_BUS_LOG(ERR, "netcfg_acquire failed"); + return -EINVAL; + } + + RTE_LOG(NOTICE, EAL, "DPAA Bus Detected\n"); + + if (!dpaa_netcfg->num_ethports) { + DPAA_BUS_LOG(INFO, "no network interfaces available"); + /* This is not an error */ + return 0; + } + +#ifdef RTE_LIBRTE_DPAA_DEBUG_DRIVER + dump_netcfg(dpaa_netcfg); +#endif + + DPAA_BUS_LOG(DEBUG, "Number of ethernet devices = %d", + dpaa_netcfg->num_ethports); + ret = dpaa_create_device_list(); + if (ret) { + DPAA_BUS_LOG(ERR, "Unable to create device list. (%d)", ret); + return ret; + } + return 0; +} + +static int +rte_dpaa_bus_probe(void) +{ + int ret = -1; + struct rte_dpaa_device *dev; + struct rte_dpaa_driver *drv; + FILE *svr_file = NULL; + unsigned int svr_ver; + int probe_all = rte_dpaa_bus.bus.conf.scan_mode != RTE_BUS_SCAN_WHITELIST; + static int process_once; + + /* If DPAA bus is not present nothing needs to be done */ + if (!rte_dpaa_bus.detected) + return 0; + + /* Device list creation is only done once */ + if (!process_once) { + rte_dpaa_bus_dev_build(); + if (rte_eal_process_type() == RTE_PROC_PRIMARY) { + /* One time load of Qman/Bman drivers */ + ret = qman_global_init(); + if (ret) { + DPAA_BUS_ERR("QMAN initialization failed: %d", + ret); + return ret; + } + ret = bman_global_init(); + if (ret) { + DPAA_BUS_ERR("BMAN initialization failed: %d", + ret); + return ret; + } + } + } + process_once = 1; + + /* If no device present on DPAA bus nothing needs to be done */ + if (TAILQ_EMPTY(&rte_dpaa_bus.device_list)) + return 0; + + svr_file = fopen(DPAA_SOC_ID_FILE, "r"); + if (svr_file) { + if (fscanf(svr_file, "svr:%x", &svr_ver) > 0) + dpaa_svr_family = svr_ver & SVR_MASK; + fclose(svr_file); + } + + /* And initialize the PA->VA translation table */ + dpaax_iova_table_populate(); + + /* For each registered driver, and device, call the driver->probe */ + TAILQ_FOREACH(dev, &rte_dpaa_bus.device_list, next) { + TAILQ_FOREACH(drv, &rte_dpaa_bus.driver_list, next) { + ret = rte_dpaa_device_match(drv, dev); + if (ret) + continue; + + if (rte_dev_is_probed(&dev->device)) + continue; + + if (!drv->probe || + (dev->device.devargs && + dev->device.devargs->policy == RTE_DEV_BLACKLISTED)) + continue; + + if (probe_all || + (dev->device.devargs && + dev->device.devargs->policy == + RTE_DEV_WHITELISTED)) { + ret = drv->probe(drv, dev); + if (ret) { + DPAA_BUS_ERR("unable to probe:%s", + dev->name); + } else { + dev->driver = drv; + dev->device.driver = &drv->driver; + } + } + break; + } + } + + /* Register DPAA mempool ops only if any DPAA device has + * been detected. + */ + rte_mbuf_set_platform_mempool_ops(DPAA_MEMPOOL_OPS_NAME); + + return 0; +} + +static struct rte_device * +rte_dpaa_find_device(const struct rte_device *start, rte_dev_cmp_t cmp, + const void *data) +{ + struct rte_dpaa_device *dev; + const struct rte_dpaa_device *dstart; + + /* find_device is called with 'data' as an opaque object - just call + * cmp with this and each device object on bus. + */ + + if (start != NULL) { + dstart = RTE_DEV_TO_DPAA_CONST(start); + dev = TAILQ_NEXT(dstart, next); + } else { + dev = TAILQ_FIRST(&rte_dpaa_bus.device_list); + } + + while (dev != NULL) { + if (cmp(&dev->device, data) == 0) { + DPAA_BUS_DEBUG("Found dev=(%s)\n", dev->device.name); + return &dev->device; + } + dev = TAILQ_NEXT(dev, next); + } + + DPAA_BUS_DEBUG("Unable to find any device\n"); + return NULL; +} + +/* + * Get iommu class of DPAA2 devices on the bus. + */ +static enum rte_iova_mode +rte_dpaa_get_iommu_class(void) +{ + if ((access(DPAA_DEV_PATH1, F_OK) != 0) && + (access(DPAA_DEV_PATH2, F_OK) != 0)) { + return RTE_IOVA_DC; + } + return RTE_IOVA_PA; +} + +static int +dpaa_bus_plug(struct rte_device *dev __rte_unused) +{ + /* No operation is performed while plugging the device */ + return 0; +} + +static int +dpaa_bus_unplug(struct rte_device *dev __rte_unused) +{ + /* No operation is performed while unplugging the device */ + return 0; +} + +static void * +dpaa_bus_dev_iterate(const void *start, const char *str, + const struct rte_dev_iterator *it __rte_unused) +{ + const struct rte_dpaa_device *dstart; + struct rte_dpaa_device *dev; + char *dup, *dev_name = NULL; + + /* Expectation is that device would be name=device_name */ + if (strncmp(str, "name=", 5) != 0) { + DPAA_BUS_DEBUG("Invalid device string (%s)\n", str); + return NULL; + } + + /* Now that name=device_name format is available, split */ + dup = strdup(str); + dev_name = dup + strlen("name="); + + if (start != NULL) { + dstart = RTE_DEV_TO_DPAA_CONST(start); + dev = TAILQ_NEXT(dstart, next); + } else { + dev = TAILQ_FIRST(&rte_dpaa_bus.device_list); + } + + while (dev != NULL) { + if (strcmp(dev->device.name, dev_name) == 0) { + free(dup); + return &dev->device; + } + dev = TAILQ_NEXT(dev, next); + } + + free(dup); + return NULL; +} + +static struct rte_dpaa_bus rte_dpaa_bus = { + .bus = { + .scan = rte_dpaa_bus_scan, + .probe = rte_dpaa_bus_probe, + .parse = rte_dpaa_bus_parse, + .find_device = rte_dpaa_find_device, + .get_iommu_class = rte_dpaa_get_iommu_class, + .plug = dpaa_bus_plug, + .unplug = dpaa_bus_unplug, + .dev_iterate = dpaa_bus_dev_iterate, + }, + .device_list = TAILQ_HEAD_INITIALIZER(rte_dpaa_bus.device_list), + .driver_list = TAILQ_HEAD_INITIALIZER(rte_dpaa_bus.driver_list), + .device_count = 0, +}; + +RTE_REGISTER_BUS(FSL_DPAA_BUS_NAME, rte_dpaa_bus.bus); + +RTE_INIT(dpaa_init_log) +{ + dpaa_logtype_bus = rte_log_register("bus.dpaa"); + if (dpaa_logtype_bus >= 0) + rte_log_set_level(dpaa_logtype_bus, RTE_LOG_NOTICE); +} diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/dpaa_bits.h b/src/spdk/dpdk/drivers/bus/dpaa/include/dpaa_bits.h new file mode 100644 index 000000000..9bc14d0c3 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/dpaa_bits.h @@ -0,0 +1,39 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * + * Copyright 2017 NXP + * + */ + +#ifndef __DPAA_BITS_H +#define __DPAA_BITS_H + +/* Bitfield stuff. */ +#define BITS_PER_ULONG (sizeof(unsigned long) << 3) +#define SHIFT_PER_ULONG (((1 << 5) == BITS_PER_ULONG) ? 5 : 6) +#define BITS_MASK(idx) (1UL << ((idx) & (BITS_PER_ULONG - 1))) +#define BITS_IDX(idx) ((idx) >> SHIFT_PER_ULONG) + +static inline void dpaa_set_bits(unsigned long mask, + volatile unsigned long *p) +{ + *p |= mask; +} + +static inline void dpaa_set_bit(int idx, volatile unsigned long *bits) +{ + dpaa_set_bits(BITS_MASK(idx), bits + BITS_IDX(idx)); +} + +static inline void dpaa_clear_bits(unsigned long mask, + volatile unsigned long *p) +{ + *p &= ~mask; +} + +static inline void dpaa_clear_bit(int idx, + volatile unsigned long *bits) +{ + dpaa_clear_bits(BITS_MASK(idx), bits + BITS_IDX(idx)); +} + +#endif /* __DPAA_BITS_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/dpaa_rbtree.h b/src/spdk/dpdk/drivers/bus/dpaa/include/dpaa_rbtree.h new file mode 100644 index 000000000..6c237e705 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/dpaa_rbtree.h @@ -0,0 +1,117 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * + * Copyright 2017 NXP + * + */ + +#ifndef __DPAA_RBTREE_H +#define __DPAA_RBTREE_H + +#include <rte_common.h> +/************/ +/* RB-trees */ +/************/ + +/* Linux has a good RB-tree implementation, that we can't use (GPL). It also has + * a flat/hooked-in interface that virtually requires license-contamination in + * order to write a caller-compatible implementation. Instead, I've created an + * RB-tree encapsulation on top of linux's primitives (it does some of the work + * the client logic would normally do), and this gives us something we can + * reimplement on LWE. Unfortunately there's no good+free RB-tree + * implementations out there that are license-compatible and "flat" (ie. no + * dynamic allocation). I did find a malloc-based one that I could convert, but + * that will be a task for later on. For now, LWE's RB-tree is implemented using + * an ordered linked-list. + * + * Note, the only linux-esque type is "struct rb_node", because it's used + * statically in the exported header, so it can't be opaque. Our version doesn't + * include a "rb_parent_color" field because we're doing linked-list instead of + * a true rb-tree. + */ + +struct rb_node { + struct rb_node *prev, *next; +}; + +struct dpa_rbtree { + struct rb_node *head, *tail; +}; + +#define DPAA_RBTREE { NULL, NULL } +static inline void dpa_rbtree_init(struct dpa_rbtree *tree) +{ + tree->head = tree->tail = NULL; +} + +#define QMAN_NODE2OBJ(ptr, type, node_field) \ + (type *)((char *)ptr - offsetof(type, node_field)) + +#define IMPLEMENT_DPAA_RBTREE(name, type, node_field, val_field) \ +static inline int name##_push(struct dpa_rbtree *tree, type *obj) \ +{ \ + struct rb_node *node = tree->head; \ + if (!node) { \ + tree->head = tree->tail = &obj->node_field; \ + obj->node_field.prev = obj->node_field.next = NULL; \ + return 0; \ + } \ + while (node) { \ + type *item = QMAN_NODE2OBJ(node, type, node_field); \ + if (obj->val_field == item->val_field) \ + return -EBUSY; \ + if (obj->val_field < item->val_field) { \ + if (tree->head == node) \ + tree->head = &obj->node_field; \ + else \ + node->prev->next = &obj->node_field; \ + obj->node_field.prev = node->prev; \ + obj->node_field.next = node; \ + node->prev = &obj->node_field; \ + return 0; \ + } \ + node = node->next; \ + } \ + obj->node_field.prev = tree->tail; \ + obj->node_field.next = NULL; \ + tree->tail->next = &obj->node_field; \ + tree->tail = &obj->node_field; \ + return 0; \ +} \ +static inline void name##_del(struct dpa_rbtree *tree, type *obj) \ +{ \ + if (tree->head == &obj->node_field) { \ + if (tree->tail == &obj->node_field) \ + /* Only item in the list */ \ + tree->head = tree->tail = NULL; \ + else { \ + /* Is the head, next != NULL */ \ + tree->head = tree->head->next; \ + tree->head->prev = NULL; \ + } \ + } else { \ + if (tree->tail == &obj->node_field) { \ + /* Is the tail, prev != NULL */ \ + tree->tail = tree->tail->prev; \ + tree->tail->next = NULL; \ + } else { \ + /* Is neither the head nor the tail */ \ + obj->node_field.prev->next = obj->node_field.next; \ + obj->node_field.next->prev = obj->node_field.prev; \ + } \ + } \ +} \ +static inline type *name##_find(struct dpa_rbtree *tree, u32 val) \ +{ \ + struct rb_node *node = tree->head; \ + while (node) { \ + type *item = QMAN_NODE2OBJ(node, type, node_field); \ + if (val == item->val_field) \ + return item; \ + if (val < item->val_field) \ + return NULL; \ + node = node->next; \ + } \ + return NULL; \ +} + +#endif /* __DPAA_RBTREE_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/fman.h b/src/spdk/dpdk/drivers/bus/dpaa/include/fman.h new file mode 100644 index 000000000..c02d32d22 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/fman.h @@ -0,0 +1,426 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2010-2012 Freescale Semiconductor, Inc. + * All rights reserved. + * + */ + +#ifndef __FMAN_H +#define __FMAN_H + +#include <stdbool.h> +#include <net/if.h> + +#include <rte_ethdev_driver.h> +#include <rte_ether.h> + +#include <compat.h> +#include <dpaa_list.h> + +#ifndef FMAN_DEVICE_PATH +#define FMAN_DEVICE_PATH "/dev/mem" +#endif + +#define MEMAC_NUM_OF_PADDRS 7 /* Num of additional exact match MAC adr regs */ + +/* Control and Configuration Register (COMMAND_CONFIG) for MEMAC */ +#define CMD_CFG_LOOPBACK_EN 0x00000400 +/**< 21 XGMII/GMII loopback enable */ +#define CMD_CFG_PROMIS_EN 0x00000010 +/**< 27 Promiscuous operation enable */ +#define CMD_CFG_PAUSE_IGNORE 0x00000100 +/**< 23 Ignore Pause frame quanta */ + +/* Statistics Configuration Register (STATN_CONFIG) */ +#define STATS_CFG_CLR 0x00000004 +/**< 29 Reset all counters */ +#define STATS_CFG_CLR_ON_RD 0x00000002 +/**< 30 Clear on read */ +#define STATS_CFG_SATURATE 0x00000001 +/**< 31 Saturate at the maximum val */ + +/**< Max receive frame length mask */ +#define MAXFRM_SIZE_MEMAC 0x00007fe0 +#define MAXFRM_RX_MASK 0x0000ffff + +/**< Interface Mode Register Register for MEMAC */ +#define IF_MODE_RLP 0x00000820 + +/**< Pool Limits */ +#define FMAN_PORT_MAX_EXT_POOLS_NUM 8 +#define FMAN_PORT_OBS_EXT_POOLS_NUM 2 + +#define FMAN_PORT_CG_MAP_NUM 8 +#define FMAN_PORT_PRS_RESULT_WORDS_NUM 8 +#define FMAN_PORT_BMI_FIFO_UNITS 0x100 +#define FMAN_PORT_IC_OFFSET_UNITS 0x10 + +#define FMAN_ENABLE_BPOOL_DEPLETION 0xF00000F0 + +#define HASH_CTRL_MCAST_EN 0x00000100 +#define GROUP_ADDRESS 0x0000010000000000LL +#define HASH_CTRL_ADDR_MASK 0x0000003F + +/* Pre definitions of FMAN interface and Bpool structures */ +struct __fman_if; +struct fman_if_bpool; +/* Lists of fman interfaces and bpools */ +TAILQ_HEAD(rte_fman_if_list, __fman_if); + +/* Represents the different flavour of network interface */ +enum fman_mac_type { + fman_offline = 0, + fman_mac_1g, + fman_mac_10g, +}; + +struct mac_addr { + uint32_t mac_addr_l; /**< Lower 32 bits of 48-bit MAC address */ + uint32_t mac_addr_u; /**< Upper 16 bits of 48-bit MAC address */ +}; + +struct memac_regs { + /* General Control and Status */ + uint32_t res0000[2]; + uint32_t command_config; /**< 0x008 Ctrl and cfg */ + struct mac_addr mac_addr0; /**< 0x00C-0x010 MAC_ADDR_0...1 */ + uint32_t maxfrm; /**< 0x014 Max frame length */ + uint32_t res0018[5]; + uint32_t hashtable_ctrl; /**< 0x02C Hash table control */ + uint32_t res0030[4]; + uint32_t ievent; /**< 0x040 Interrupt event */ + uint32_t tx_ipg_length; + /**< 0x044 Transmitter inter-packet-gap */ + uint32_t res0048; + uint32_t imask; /**< 0x04C Interrupt mask */ + uint32_t res0050; + uint32_t pause_quanta[4]; /**< 0x054 Pause quanta */ + uint32_t pause_thresh[4]; /**< 0x064 Pause quanta threshold */ + uint32_t rx_pause_status; /**< 0x074 Receive pause status */ + uint32_t res0078[2]; + struct mac_addr mac_addr[MEMAC_NUM_OF_PADDRS]; + /**< 0x80-0x0B4 mac padr */ + uint32_t lpwake_timer; + /**< 0x0B8 Low Power Wakeup Timer */ + uint32_t sleep_timer; + /**< 0x0BC Transmit EEE Low Power Timer */ + uint32_t res00c0[8]; + uint32_t statn_config; + /**< 0x0E0 Statistics configuration */ + uint32_t res00e4[7]; + /* Rx Statistics Counter */ + uint32_t reoct_l; /**<Rx Eth Octets Counter */ + uint32_t reoct_u; + uint32_t roct_l; /**<Rx Octet Counters */ + uint32_t roct_u; + uint32_t raln_l; /**<Rx Alignment Error Counter */ + uint32_t raln_u; + uint32_t rxpf_l; /**<Rx valid Pause Frame */ + uint32_t rxpf_u; + uint32_t rfrm_l; /**<Rx Frame counter */ + uint32_t rfrm_u; + uint32_t rfcs_l; /**<Rx frame check seq error */ + uint32_t rfcs_u; + uint32_t rvlan_l; /**<Rx Vlan Frame Counter */ + uint32_t rvlan_u; + uint32_t rerr_l; /**<Rx Frame error */ + uint32_t rerr_u; + uint32_t ruca_l; /**<Rx Unicast */ + uint32_t ruca_u; + uint32_t rmca_l; /**<Rx Multicast */ + uint32_t rmca_u; + uint32_t rbca_l; /**<Rx Broadcast */ + uint32_t rbca_u; + uint32_t rdrp_l; /**<Rx Dropper Packet */ + uint32_t rdrp_u; + uint32_t rpkt_l; /**<Rx packet */ + uint32_t rpkt_u; + uint32_t rund_l; /**<Rx undersized packets */ + uint32_t rund_u; + uint32_t r64_l; /**<Rx 64 byte */ + uint32_t r64_u; + uint32_t r127_l; + uint32_t r127_u; + uint32_t r255_l; + uint32_t r255_u; + uint32_t r511_l; + uint32_t r511_u; + uint32_t r1023_l; + uint32_t r1023_u; + uint32_t r1518_l; + uint32_t r1518_u; + uint32_t r1519x_l; + uint32_t r1519x_u; + uint32_t rovr_l; /**<Rx oversized but good */ + uint32_t rovr_u; + uint32_t rjbr_l; /**<Rx oversized with bad csum */ + uint32_t rjbr_u; + uint32_t rfrg_l; /**<Rx fragment Packet */ + uint32_t rfrg_u; + uint32_t rcnp_l; /**<Rx control packets (0x8808 */ + uint32_t rcnp_u; + uint32_t rdrntp_l; /**<Rx dropped due to FIFO overflow */ + uint32_t rdrntp_u; + uint32_t res01d0[12]; + /* Tx Statistics Counter */ + uint32_t teoct_l; /**<Tx eth octets */ + uint32_t teoct_u; + uint32_t toct_l; /**<Tx Octets */ + uint32_t toct_u; + uint32_t res0210[2]; + uint32_t txpf_l; /**<Tx valid pause frame */ + uint32_t txpf_u; + uint32_t tfrm_l; /**<Tx frame counter */ + uint32_t tfrm_u; + uint32_t tfcs_l; /**<Tx FCS error */ + uint32_t tfcs_u; + uint32_t tvlan_l; /**<Tx Vlan Frame */ + uint32_t tvlan_u; + uint32_t terr_l; /**<Tx frame error */ + uint32_t terr_u; + uint32_t tuca_l; /**<Tx Unicast */ + uint32_t tuca_u; + uint32_t tmca_l; /**<Tx Multicast */ + uint32_t tmca_u; + uint32_t tbca_l; /**<Tx Broadcast */ + uint32_t tbca_u; + uint32_t res0258[2]; + uint32_t tpkt_l; /**<Tx Packet */ + uint32_t tpkt_u; + uint32_t tund_l; /**<Tx Undersized */ + uint32_t tund_u; + uint32_t t64_l; + uint32_t t64_u; + uint32_t t127_l; + uint32_t t127_u; + uint32_t t255_l; + uint32_t t255_u; + uint32_t t511_l; + uint32_t t511_u; + uint32_t t1023_l; + uint32_t t1023_u; + uint32_t t1518_l; + uint32_t t1518_u; + uint32_t t1519x_l; + uint32_t t1519x_u; + uint32_t res02a8[6]; + uint32_t tcnp_l; /**<Tx Control Packet type - 0x8808 */ + uint32_t tcnp_u; + uint32_t res02c8[14]; + /* Line Interface Control */ + uint32_t if_mode; /**< 0x300 Interface Mode Control */ + uint32_t if_status; /**< 0x304 Interface Status */ + uint32_t res0308[14]; + /* HiGig/2 */ + uint32_t hg_config; /**< 0x340 Control and cfg */ + uint32_t res0344[3]; + uint32_t hg_pause_quanta; /**< 0x350 Pause quanta */ + uint32_t res0354[3]; + uint32_t hg_pause_thresh; /**< 0x360 Pause quanta threshold */ + uint32_t res0364[3]; + uint32_t hgrx_pause_status; /**< 0x370 Receive pause status */ + uint32_t hg_fifos_status; /**< 0x374 fifos status */ + uint32_t rhm; /**< 0x378 rx messages counter */ + uint32_t thm; /**< 0x37C tx messages counter */ +}; + +struct rx_bmi_regs { + uint32_t fmbm_rcfg; /**< Rx Configuration */ + uint32_t fmbm_rst; /**< Rx Status */ + uint32_t fmbm_rda; /**< Rx DMA attributes*/ + uint32_t fmbm_rfp; /**< Rx FIFO Parameters*/ + uint32_t fmbm_rfed; /**< Rx Frame End Data*/ + uint32_t fmbm_ricp; /**< Rx Internal Context Parameters*/ + uint32_t fmbm_rim; /**< Rx Internal Buffer Margins*/ + uint32_t fmbm_rebm; /**< Rx External Buffer Margins*/ + uint32_t fmbm_rfne; /**< Rx Frame Next Engine*/ + uint32_t fmbm_rfca; /**< Rx Frame Command Attributes.*/ + uint32_t fmbm_rfpne; /**< Rx Frame Parser Next Engine*/ + uint32_t fmbm_rpso; /**< Rx Parse Start Offset*/ + uint32_t fmbm_rpp; /**< Rx Policer Profile */ + uint32_t fmbm_rccb; /**< Rx Coarse Classification Base */ + uint32_t fmbm_reth; /**< Rx Excessive Threshold */ + uint32_t reserved003c[1]; /**< (0x03C 0x03F) */ + uint32_t fmbm_rprai[FMAN_PORT_PRS_RESULT_WORDS_NUM]; + /**< Rx Parse Results Array Init*/ + uint32_t fmbm_rfqid; /**< Rx Frame Queue ID*/ + uint32_t fmbm_refqid; /**< Rx Error Frame Queue ID*/ + uint32_t fmbm_rfsdm; /**< Rx Frame Status Discard Mask*/ + uint32_t fmbm_rfsem; /**< Rx Frame Status Error Mask*/ + uint32_t fmbm_rfene; /**< Rx Frame Enqueue Next Engine */ + uint32_t reserved0074[0x2]; /**< (0x074-0x07C) */ + uint32_t fmbm_rcmne; + /**< Rx Frame Continuous Mode Next Engine */ + uint32_t reserved0080[0x20];/**< (0x080 0x0FF) */ + uint32_t fmbm_ebmpi[FMAN_PORT_MAX_EXT_POOLS_NUM]; + /**< Buffer Manager pool Information-*/ + uint32_t fmbm_acnt[FMAN_PORT_MAX_EXT_POOLS_NUM]; + /**< Allocate Counter-*/ + uint32_t reserved0130[8]; + /**< 0x130/0x140 - 0x15F reserved -*/ + uint32_t fmbm_rcgm[FMAN_PORT_CG_MAP_NUM]; + /**< Congestion Group Map*/ + uint32_t fmbm_mpd; /**< BM Pool Depletion */ + uint32_t reserved0184[0x1F]; /**< (0x184 0x1FF) */ + uint32_t fmbm_rstc; /**< Rx Statistics Counters*/ + uint32_t fmbm_rfrc; /**< Rx Frame Counter*/ + uint32_t fmbm_rfbc; /**< Rx Bad Frames Counter*/ + uint32_t fmbm_rlfc; /**< Rx Large Frames Counter*/ + uint32_t fmbm_rffc; /**< Rx Filter Frames Counter*/ + uint32_t fmbm_rfdc; /**< Rx Frame Discard Counter*/ + uint32_t fmbm_rfldec; /**< Rx Frames List DMA Error Counter*/ + uint32_t fmbm_rodc; /**< Rx Out of Buffers Discard nntr*/ + uint32_t fmbm_rbdc; /**< Rx Buffers Deallocate Counter*/ + uint32_t reserved0224[0x17]; /**< (0x224 0x27F) */ + uint32_t fmbm_rpc; /**< Rx Performance Counters*/ + uint32_t fmbm_rpcp; /**< Rx Performance Count Parameters*/ + uint32_t fmbm_rccn; /**< Rx Cycle Counter*/ + uint32_t fmbm_rtuc; /**< Rx Tasks Utilization Counter*/ + uint32_t fmbm_rrquc; + /**< Rx Receive Queue Utilization cntr*/ + uint32_t fmbm_rduc; /**< Rx DMA Utilization Counter*/ + uint32_t fmbm_rfuc; /**< Rx FIFO Utilization Counter*/ + uint32_t fmbm_rpac; /**< Rx Pause Activation Counter*/ + uint32_t reserved02a0[0x18]; /**< (0x2A0 0x2FF) */ + uint32_t fmbm_rdbg; /**< Rx Debug-*/ +}; + +struct fman_port_qmi_regs { + uint32_t fmqm_pnc; /**< PortID n Configuration Register */ + uint32_t fmqm_pns; /**< PortID n Status Register */ + uint32_t fmqm_pnts; /**< PortID n Task Status Register */ + uint32_t reserved00c[4]; /**< 0xn00C - 0xn01B */ + uint32_t fmqm_pnen; /**< PortID n Enqueue NIA Register */ + uint32_t fmqm_pnetfc; /**< PortID n Enq Total Frame Counter */ + uint32_t reserved024[2]; /**< 0xn024 - 0x02B */ + uint32_t fmqm_pndn; /**< PortID n Dequeue NIA Register */ + uint32_t fmqm_pndc; /**< PortID n Dequeue Config Register */ + uint32_t fmqm_pndtfc; /**< PortID n Dequeue tot Frame cntr */ + uint32_t fmqm_pndfdc; /**< PortID n Dequeue FQID Dflt Cntr */ + uint32_t fmqm_pndcc; /**< PortID n Dequeue Confirm Counter */ +}; + +/* This struct exports parameters about an Fman network interface, determined + * from the device-tree. + */ +struct fman_if { + /* Which Fman this interface belongs to */ + uint8_t fman_idx; + /* The type/speed of the interface */ + enum fman_mac_type mac_type; + /* Boolean, set when mac type is memac */ + uint8_t is_memac; + /* Boolean, set when PHY is RGMII */ + uint8_t is_rgmii; + /* The index of this MAC (within the Fman it belongs to) */ + uint8_t mac_idx; + /* The MAC address */ + struct rte_ether_addr mac_addr; + /* The Qman channel to schedule Tx FQs to */ + u16 tx_channel_id; + /* The hard-coded FQIDs for this interface. Note: this doesn't cover + * the PCD nor the "Rx default" FQIDs, which are configured via FMC + * and its XML-based configuration. + */ + uint32_t fqid_rx_def; + uint32_t fqid_rx_err; + uint32_t fqid_tx_err; + uint32_t fqid_tx_confirm; + + struct list_head bpool_list; + /* The node for linking this interface into "fman_if_list" */ + struct list_head node; +}; + +/* This struct exposes parameters for buffer pools, extracted from the network + * interface settings in the device tree. + */ +struct fman_if_bpool { + uint32_t bpid; + uint64_t count; + uint64_t size; + uint64_t addr; + /* The node for linking this bpool into fman_if::bpool_list */ + struct list_head node; +}; + +/* Internal Context transfer params - FMBM_RICP*/ +struct fman_if_ic_params { + /*IC offset in the packet buffer */ + uint16_t iceof; + /*IC internal offset */ + uint16_t iciof; + /*IC size to copy */ + uint16_t icsz; +}; + +/* The exported "struct fman_if" type contains the subset of fields we want + * exposed. This struct is embedded in a larger "struct __fman_if" which + * contains the extra bits we *don't* want exposed. + */ +struct __fman_if { + struct fman_if __if; + char node_path[PATH_MAX]; + uint64_t regs_size; + void *ccsr_map; + void *bmi_map; + void *qmi_map; + struct list_head node; +}; + +/* And this is the base list node that the interfaces are added to. (See + * fman_if_enable_all_rx() below for an example of its use.) + */ +extern const struct list_head *fman_if_list; + +extern int fman_ccsr_map_fd; + +/* To iterate the "bpool_list" for an interface. Eg; + * struct fman_if *p = get_ptr_to_some_interface(); + * struct fman_if_bpool *bp; + * printf("Interface uses following BPIDs;\n"); + * fman_if_for_each_bpool(bp, p) { + * printf(" %d\n", bp->bpid); + * [...] + * } + */ +#define fman_if_for_each_bpool(bp, __if) \ + list_for_each_entry(bp, &(__if)->bpool_list, node) + +#define FMAN_ERR(rc, fmt, args...) \ + do { \ + _errno = (rc); \ + DPAA_BUS_LOG(ERR, fmt "(%d)", ##args, errno); \ + } while (0) + +#define FMAN_IP_REV_1 0xC30C4 +#define FMAN_IP_REV_1_MAJOR_MASK 0x0000FF00 +#define FMAN_IP_REV_1_MAJOR_SHIFT 8 +#define FMAN_V3 0x06 +#define FMAN_V3_CONTEXTA_EN_A2V 0x10000000 +#define FMAN_V3_CONTEXTA_EN_OVOM 0x02000000 +#define FMAN_V3_CONTEXTA_EN_EBD 0x80000000 +#define FMAN_CONTEXTA_DIS_CHECKSUM 0x7ull +#define FMAN_CONTEXTA_SET_OPCODE11 0x2000000b00000000 +extern u16 fman_ip_rev; +extern u32 fman_dealloc_bufs_mask_hi; +extern u32 fman_dealloc_bufs_mask_lo; + +/** + * Initialize the FMAN driver + * + * @args void + * @return + * 0 for success; error OTHERWISE + */ +int fman_init(void); + +/** + * Teardown the FMAN driver + * + * @args void + * @return void + */ +void fman_finish(void); + +#endif /* __FMAN_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_bman.h b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_bman.h new file mode 100644 index 000000000..82da2fcfe --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_bman.h @@ -0,0 +1,348 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2012 Freescale Semiconductor, Inc. + * + */ + +#ifndef __FSL_BMAN_H +#define __FSL_BMAN_H + +#ifdef __cplusplus +extern "C" { +#endif + +/* This wrapper represents a bit-array for the depletion state of the 64 Bman + * buffer pools. + */ +struct bman_depletion { + u32 state[2]; +}; + +static inline void bman_depletion_init(struct bman_depletion *c) +{ + c->state[0] = c->state[1] = 0; +} + +static inline void bman_depletion_fill(struct bman_depletion *c) +{ + c->state[0] = c->state[1] = ~0; +} + +/* --- Bman data structures (and associated constants) --- */ + +/* Represents s/w corenet portal mapped data structures */ +struct bm_rcr_entry; /* RCR (Release Command Ring) entries */ +struct bm_mc_command; /* MC (Management Command) command */ +struct bm_mc_result; /* MC result */ + +/* Code-reduction, define a wrapper for 48-bit buffers. In cases where a buffer + * pool id specific to this buffer is needed (BM_RCR_VERB_CMD_BPID_MULTI, + * BM_MCC_VERB_ACQUIRE), the 'bpid' field is used. + */ +struct bm_buffer { + union { + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u8 __reserved1; + u8 bpid; + u16 hi; /* High 16-bits of 48-bit address */ + u32 lo; /* Low 32-bits of 48-bit address */ +#else + u32 lo; + u16 hi; + u8 bpid; + u8 __reserved; +#endif + }; + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u64 __notaddress:16; + u64 addr:48; +#else + u64 addr:48; + u64 __notaddress:16; +#endif + }; + u64 opaque; + }; +} __rte_aligned(8); +static inline u64 bm_buffer_get64(const struct bm_buffer *buf) +{ + return buf->addr; +} + +static inline dma_addr_t bm_buf_addr(const struct bm_buffer *buf) +{ + return (dma_addr_t)buf->addr; +} + +#define bm_buffer_set64(buf, v) \ + do { \ + struct bm_buffer *__buf931 = (buf); \ + __buf931->hi = upper_32_bits(v); \ + __buf931->lo = lower_32_bits(v); \ + } while (0) + +/* See 1.5.3.5.4: "Release Command" */ +struct bm_rcr_entry { + union { + struct { + u8 __dont_write_directly__verb; + u8 bpid; /* used with BM_RCR_VERB_CMD_BPID_SINGLE */ + u8 __reserved1[62]; + }; + struct bm_buffer bufs[8]; + }; +} __packed; +#define BM_RCR_VERB_VBIT 0x80 +#define BM_RCR_VERB_CMD_MASK 0x70 /* one of two values; */ +#define BM_RCR_VERB_CMD_BPID_SINGLE 0x20 +#define BM_RCR_VERB_CMD_BPID_MULTI 0x30 +#define BM_RCR_VERB_BUFCOUNT_MASK 0x0f /* values 1..8 */ + +/* See 1.5.3.1: "Acquire Command" */ +/* See 1.5.3.2: "Query Command" */ +struct bm_mcc_acquire { + u8 bpid; + u8 __reserved1[62]; +} __packed; +struct bm_mcc_query { + u8 __reserved2[63]; +} __packed; +struct bm_mc_command { + u8 __dont_write_directly__verb; + union { + struct bm_mcc_acquire acquire; + struct bm_mcc_query query; + }; +} __packed; +#define BM_MCC_VERB_VBIT 0x80 +#define BM_MCC_VERB_CMD_MASK 0x70 /* where the verb contains; */ +#define BM_MCC_VERB_CMD_ACQUIRE 0x10 +#define BM_MCC_VERB_CMD_QUERY 0x40 +#define BM_MCC_VERB_ACQUIRE_BUFCOUNT 0x0f /* values 1..8 go here */ + +/* See 1.5.3.3: "Acquire Response" */ +/* See 1.5.3.4: "Query Response" */ +struct bm_pool_state { + u8 __reserved1[32]; + /* "availability state" and "depletion state" */ + struct { + u8 __reserved1[8]; + /* Access using bman_depletion_***() */ + struct bman_depletion state; + } as, ds; +}; + +struct bm_mc_result { + union { + struct { + u8 verb; + u8 __reserved1[63]; + }; + union { + struct { + u8 __reserved1; + u8 bpid; + u8 __reserved2[62]; + }; + struct bm_buffer bufs[8]; + } acquire; + struct bm_pool_state query; + }; +} __packed; +#define BM_MCR_VERB_VBIT 0x80 +#define BM_MCR_VERB_CMD_MASK BM_MCC_VERB_CMD_MASK +#define BM_MCR_VERB_CMD_ACQUIRE BM_MCC_VERB_CMD_ACQUIRE +#define BM_MCR_VERB_CMD_QUERY BM_MCC_VERB_CMD_QUERY +#define BM_MCR_VERB_CMD_ERR_INVALID 0x60 +#define BM_MCR_VERB_CMD_ERR_ECC 0x70 +#define BM_MCR_VERB_ACQUIRE_BUFCOUNT BM_MCC_VERB_ACQUIRE_BUFCOUNT /* 0..8 */ + +/* Portal and Buffer Pools */ +/* Represents a managed portal */ +struct bman_portal; + +/* This object type represents Bman buffer pools. */ +struct bman_pool; + +/* This struct specifies parameters for a bman_pool object. */ +struct bman_pool_params { + /* index of the buffer pool to encapsulate (0-63), ignored if + * BMAN_POOL_FLAG_DYNAMIC_BPID is set. + */ + u32 bpid; + /* bit-mask of BMAN_POOL_FLAG_*** options */ + u32 flags; + /* depletion-entry/exit thresholds, if BMAN_POOL_FLAG_THRESH is set. NB: + * this is only allowed if BMAN_POOL_FLAG_DYNAMIC_BPID is used *and* + * when run in the control plane (which controls Bman CCSR). This array + * matches the definition of bm_pool_set(). + */ + u32 thresholds[4]; +}; + +/* Flags to bman_new_pool() */ +#define BMAN_POOL_FLAG_NO_RELEASE 0x00000001 /* can't release to pool */ +#define BMAN_POOL_FLAG_ONLY_RELEASE 0x00000002 /* can only release to pool */ +#define BMAN_POOL_FLAG_DYNAMIC_BPID 0x00000008 /* (de)allocate bpid */ +#define BMAN_POOL_FLAG_THRESH 0x00000010 /* set depletion thresholds */ + +/* Flags to bman_release() */ +#define BMAN_RELEASE_FLAG_NOW 0x00000008 /* issue immediate release */ + + +/** + * bman_get_portal_index - get portal configuration index + */ +int bman_get_portal_index(void); + +/** + * bman_rcr_is_empty - Determine if portal's RCR is empty + * + * For use in situations where a cpu-affine caller needs to determine when all + * releases for the local portal have been processed by Bman but can't use the + * BMAN_RELEASE_FLAG_WAIT_SYNC flag to do this from the final bman_release(). + * The function forces tracking of RCR consumption (which normally doesn't + * happen until release processing needs to find space to put new release + * commands), and returns zero if the ring still has unprocessed entries, + * non-zero if it is empty. + */ +int bman_rcr_is_empty(void); + +/** + * bman_alloc_bpid_range - Allocate a contiguous range of BPIDs + * @result: is set by the API to the base BPID of the allocated range + * @count: the number of BPIDs required + * @align: required alignment of the allocated range + * @partial: non-zero if the API can return fewer than @count BPIDs + * + * Returns the number of buffer pools allocated, or a negative error code. If + * @partial is non zero, the allocation request may return a smaller range of + * BPs than requested (though alignment will be as requested). If @partial is + * zero, the return value will either be 'count' or negative. + */ +int bman_alloc_bpid_range(u32 *result, u32 count, u32 align, int partial); +static inline int bman_alloc_bpid(u32 *result) +{ + int ret = bman_alloc_bpid_range(result, 1, 0, 0); + + return (ret > 0) ? 0 : ret; +} + +/** + * bman_release_bpid_range - Release the specified range of buffer pool IDs + * @bpid: the base BPID of the range to deallocate + * @count: the number of BPIDs in the range + * + * This function can also be used to seed the allocator with ranges of BPIDs + * that it can subsequently allocate from. + */ +void bman_release_bpid_range(u32 bpid, unsigned int count); +static inline void bman_release_bpid(u32 bpid) +{ + bman_release_bpid_range(bpid, 1); +} + +int bman_reserve_bpid_range(u32 bpid, unsigned int count); +static inline int bman_reserve_bpid(u32 bpid) +{ + return bman_reserve_bpid_range(bpid, 1); +} + +void bman_seed_bpid_range(u32 bpid, unsigned int count); + +int bman_shutdown_pool(u32 bpid); + +/** + * bman_new_pool - Allocates a Buffer Pool object + * @params: parameters specifying the buffer pool ID and behaviour + * + * Creates a pool object for the given @params. A portal and the depletion + * callback field of @params are only used if the BMAN_POOL_FLAG_DEPLETION flag + * is set. NB, the fields from @params are copied into the new pool object, so + * the structure provided by the caller can be released or reused after the + * function returns. + */ +__rte_internal +struct bman_pool *bman_new_pool(const struct bman_pool_params *params); + +/** + * bman_free_pool - Deallocates a Buffer Pool object + * @pool: the pool object to release + */ +__rte_internal +void bman_free_pool(struct bman_pool *pool); + +/** + * bman_get_params - Returns a pool object's parameters. + * @pool: the pool object + * + * The returned pointer refers to state within the pool object so must not be + * modified and can no longer be read once the pool object is destroyed. + */ +__rte_internal +const struct bman_pool_params *bman_get_params(const struct bman_pool *pool); + +/** + * bman_release - Release buffer(s) to the buffer pool + * @pool: the buffer pool object to release to + * @bufs: an array of buffers to release + * @num: the number of buffers in @bufs (1-8) + * @flags: bit-mask of BMAN_RELEASE_FLAG_*** options + * + */ +__rte_internal +int bman_release(struct bman_pool *pool, const struct bm_buffer *bufs, u8 num, + u32 flags); + +/** + * bman_acquire - Acquire buffer(s) from a buffer pool + * @pool: the buffer pool object to acquire from + * @bufs: array for storing the acquired buffers + * @num: the number of buffers desired (@bufs is at least this big) + * + * Issues an "Acquire" command via the portal's management command interface. + * The return value will be the number of buffers obtained from the pool, or a + * negative error code if a h/w error or pool starvation was encountered. + */ +__rte_internal +int bman_acquire(struct bman_pool *pool, struct bm_buffer *bufs, u8 num, + u32 flags); + +/** + * bman_query_pools - Query all buffer pool states + * @state: storage for the queried availability and depletion states + */ +int bman_query_pools(struct bm_pool_state *state); + +/** + * bman_query_free_buffers - Query how many free buffers are in buffer pool + * @pool: the buffer pool object to query + * + * Return the number of the free buffers + */ +__rte_internal +u32 bman_query_free_buffers(struct bman_pool *pool); + +/** + * bman_update_pool_thresholds - Change the buffer pool's depletion thresholds + * @pool: the buffer pool object to which the thresholds will be set + * @thresholds: the new thresholds + */ +int bman_update_pool_thresholds(struct bman_pool *pool, const u32 *thresholds); + +/** + * bm_pool_set_hw_threshold - Change the buffer pool's thresholds + * @pool: Pool id + * @low_thresh: low threshold + * @high_thresh: high threshold + */ +int bm_pool_set_hw_threshold(u32 bpid, const u32 low_thresh, + const u32 high_thresh); + +#ifdef __cplusplus +} +#endif + +#endif /* __FSL_BMAN_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_fman.h b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_fman.h new file mode 100644 index 000000000..6c87c8db0 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_fman.h @@ -0,0 +1,181 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2017 NXP + * + */ + +#ifndef __FSL_FMAN_H +#define __FSL_FMAN_H + +#include <rte_compat.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* Status field in FD is updated on Rx side by FMAN with following information. + * Refer to field description in FM BG. + */ +struct fm_status_t { + unsigned int reserved0:3; + unsigned int dcl4c:1; /* Don't Check L4 Checksum */ + unsigned int reserved1:1; + unsigned int ufd:1; /* Unsupported Format */ + unsigned int lge:1; /* Length Error */ + unsigned int dme:1; /* DMA Error */ + + unsigned int reserved2:4; + unsigned int fpe:1; /* Frame physical Error */ + unsigned int fse:1; /* Frame Size Error */ + unsigned int dis:1; /* Discard by Classification */ + unsigned int reserved3:1; + + unsigned int eof:1; /* Key Extraction goes out of frame */ + unsigned int nss:1; /* No Scheme selected */ + unsigned int kso:1; /* Key Size Overflow */ + unsigned int reserved4:1; + unsigned int fcl:2; /* Frame Color */ + unsigned int ipp:1; /* Illegal Policer Profile Selected */ + unsigned int flm:1; /* Frame Length Mismatch */ + unsigned int pte:1; /* Parser Timeout */ + unsigned int isp:1; /* Invalid Soft Parser Instruction */ + unsigned int phe:1; /* Header Error during parsing */ + unsigned int frdr:1; /* Frame Dropped by disabled port */ + unsigned int reserved5:4; +} __rte_packed; + +/* Set MAC address for a particular interface */ +__rte_internal +int fman_if_add_mac_addr(struct fman_if *p, uint8_t *eth, uint8_t addr_num); + +/* Remove a MAC address for a particular interface */ +__rte_internal +void fman_if_clear_mac_addr(struct fman_if *p, uint8_t addr_num); + +/* Get the FMAN statistics */ +__rte_internal +void fman_if_stats_get(struct fman_if *p, struct rte_eth_stats *stats); + +/* Reset the FMAN statistics */ +__rte_internal +void fman_if_stats_reset(struct fman_if *p); + +/* Get all of the FMAN statistics */ +__rte_internal +void fman_if_stats_get_all(struct fman_if *p, uint64_t *value, int n); + +/* Set ignore pause option for a specific interface */ +void fman_if_set_rx_ignore_pause_frames(struct fman_if *p, bool enable); + +/* Set max frame length */ +void fman_if_conf_max_frame_len(struct fman_if *p, unsigned int max_frame_len); + +/* Enable/disable Rx promiscuous mode on specified interface */ +__rte_internal +void fman_if_promiscuous_enable(struct fman_if *p); +__rte_internal +void fman_if_promiscuous_disable(struct fman_if *p); + +/* Enable/disable Rx on specific interfaces */ +__rte_internal +void fman_if_enable_rx(struct fman_if *p); +__rte_internal +void fman_if_disable_rx(struct fman_if *p); + +/* Enable/disable loopback on specific interfaces */ +__rte_internal +void fman_if_loopback_enable(struct fman_if *p); +__rte_internal +void fman_if_loopback_disable(struct fman_if *p); + +/* Set buffer pool on specific interface */ +__rte_internal +void fman_if_set_bp(struct fman_if *fm_if, unsigned int num, int bpid, + size_t bufsize); + +/* Get Flow Control threshold parameters on specific interface */ +__rte_internal +int fman_if_get_fc_threshold(struct fman_if *fm_if); + +/* Enable and Set Flow Control threshold parameters on specific interface */ +__rte_internal +int fman_if_set_fc_threshold(struct fman_if *fm_if, + u32 high_water, u32 low_water, u32 bpid); + +/* Get Flow Control pause quanta on specific interface */ +__rte_internal +int fman_if_get_fc_quanta(struct fman_if *fm_if); + +/* Set Flow Control pause quanta on specific interface */ +__rte_internal +int fman_if_set_fc_quanta(struct fman_if *fm_if, u16 pause_quanta); + +/* Set default error fqid on specific interface */ +void fman_if_set_err_fqid(struct fman_if *fm_if, uint32_t err_fqid); + +/* Get IC transfer params */ +int fman_if_get_ic_params(struct fman_if *fm_if, struct fman_if_ic_params *icp); + +/* Set IC transfer params */ +__rte_internal +int fman_if_set_ic_params(struct fman_if *fm_if, + const struct fman_if_ic_params *icp); + +/* Get interface fd->offset value */ +__rte_internal +int fman_if_get_fdoff(struct fman_if *fm_if); + +/* Set interface fd->offset value */ +__rte_internal +void fman_if_set_fdoff(struct fman_if *fm_if, uint32_t fd_offset); + +/* Get interface SG enable status value */ +__rte_internal +int fman_if_get_sg_enable(struct fman_if *fm_if); + +/* Set interface SG support mode */ +__rte_internal +void fman_if_set_sg(struct fman_if *fm_if, int enable); + +/* Get interface Max Frame length (MTU) */ +uint16_t fman_if_get_maxfrm(struct fman_if *fm_if); + +/* Set interface Max Frame length (MTU) */ +__rte_internal +void fman_if_set_maxfrm(struct fman_if *fm_if, uint16_t max_frm); + +/* Set interface next invoked action for dequeue operation */ +void fman_if_set_dnia(struct fman_if *fm_if, uint32_t nia); + +/* discard error packets on rx */ +__rte_internal +void fman_if_discard_rx_errors(struct fman_if *fm_if); + +__rte_internal +void fman_if_set_mcast_filter_table(struct fman_if *p); + +__rte_internal +void fman_if_reset_mcast_filter_table(struct fman_if *p); + +int fman_if_add_hash_mac_addr(struct fman_if *p, uint8_t *eth); + +int fman_if_get_primary_mac_addr(struct fman_if *p, uint8_t *eth); + + +/* Enable/disable Rx on all interfaces */ +static inline void fman_if_enable_all_rx(void) +{ + struct fman_if *__if; + + list_for_each_entry(__if, fman_if_list, node) + fman_if_enable_rx(__if); +} + +static inline void fman_if_disable_all_rx(void) +{ + struct fman_if *__if; + + list_for_each_entry(__if, fman_if_list, node) + fman_if_disable_rx(__if); +} +#endif /* __FSL_FMAN_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_fman_crc64.h b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_fman_crc64.h new file mode 100644 index 000000000..08ad63042 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_fman_crc64.h @@ -0,0 +1,230 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2011 Freescale Semiconductor, Inc. + * + */ + +#ifndef __FSL_FMAN_CRC64_H +#define __FSL_FMAN_CRC64_H + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * This following definitions provide a software implementation of the CRC64 + * algorithm implemented within Fman. + * + * The following example shows how to compute a CRC64 hash value based on + * SRC_IP, DST_IP and ESP_SPI values + * + * #define compute_hash(saddr,daddr,spi) \ + * do { \ + * uint64_t result; \ + * result = fman_crc64_init(); \ + * result = fman_crc64_compute_32bit(saddr, result); \ + * result = fman_crc64_compute_32bit(daddr, result); \ + * result = fman_crc64_compute_32bit(spi, result); \ + * return (uint32_t) result & RC_HASH_MASK; \ + * } while (0); + * + * If hashing over a different number of fields (or of different types) is + * required, this can be implemented using the following primitives. + */ + +/* The following table provides the constants used by the Fman CRC64 + * implementation. The table is instantiated within the DPAA fman driver. + * However if the application is not going to be linked against the DPAA fman + * driver but will use this Fman CRC64 implementation, then it will need to + * instantiate this table by using the DECLARE_FMAN_CRC64_TABLE() macro. + */ +struct fman_crc64_t { + uint64_t initial; + uint64_t table[1 << 8]; +}; +extern struct fman_crc64_t fman_crc64_ecma_182; +#define DECLARE_FMAN_CRC64_TABLE() \ +struct fman_crc64_t fman_crc64_ecma_182 = { \ + 0xFFFFFFFFFFFFFFFFULL, \ + { \ + 0x0000000000000000ULL, 0xb32e4cbe03a75f6fULL, \ + 0xf4843657a840a05bULL, 0x47aa7ae9abe7ff34ULL, \ + 0x7bd0c384ff8f5e33ULL, 0xc8fe8f3afc28015cULL, \ + 0x8f54f5d357cffe68ULL, 0x3c7ab96d5468a107ULL, \ + 0xf7a18709ff1ebc66ULL, 0x448fcbb7fcb9e309ULL, \ + 0x0325b15e575e1c3dULL, 0xb00bfde054f94352ULL, \ + 0x8c71448d0091e255ULL, 0x3f5f08330336bd3aULL, \ + 0x78f572daa8d1420eULL, 0xcbdb3e64ab761d61ULL, \ + 0x7d9ba13851336649ULL, 0xceb5ed8652943926ULL, \ + 0x891f976ff973c612ULL, 0x3a31dbd1fad4997dULL, \ + 0x064b62bcaebc387aULL, 0xb5652e02ad1b6715ULL, \ + 0xf2cf54eb06fc9821ULL, 0x41e11855055bc74eULL, \ + 0x8a3a2631ae2dda2fULL, 0x39146a8fad8a8540ULL, \ + 0x7ebe1066066d7a74ULL, 0xcd905cd805ca251bULL, \ + 0xf1eae5b551a2841cULL, 0x42c4a90b5205db73ULL, \ + 0x056ed3e2f9e22447ULL, 0xb6409f5cfa457b28ULL, \ + 0xfb374270a266cc92ULL, 0x48190ecea1c193fdULL, \ + 0x0fb374270a266cc9ULL, 0xbc9d3899098133a6ULL, \ + 0x80e781f45de992a1ULL, 0x33c9cd4a5e4ecdceULL, \ + 0x7463b7a3f5a932faULL, 0xc74dfb1df60e6d95ULL, \ + 0x0c96c5795d7870f4ULL, 0xbfb889c75edf2f9bULL, \ + 0xf812f32ef538d0afULL, 0x4b3cbf90f69f8fc0ULL, \ + 0x774606fda2f72ec7ULL, 0xc4684a43a15071a8ULL, \ + 0x83c230aa0ab78e9cULL, 0x30ec7c140910d1f3ULL, \ + 0x86ace348f355aadbULL, 0x3582aff6f0f2f5b4ULL, \ + 0x7228d51f5b150a80ULL, 0xc10699a158b255efULL, \ + 0xfd7c20cc0cdaf4e8ULL, 0x4e526c720f7dab87ULL, \ + 0x09f8169ba49a54b3ULL, 0xbad65a25a73d0bdcULL, \ + 0x710d64410c4b16bdULL, 0xc22328ff0fec49d2ULL, \ + 0x85895216a40bb6e6ULL, 0x36a71ea8a7ace989ULL, \ + 0x0adda7c5f3c4488eULL, 0xb9f3eb7bf06317e1ULL, \ + 0xfe5991925b84e8d5ULL, 0x4d77dd2c5823b7baULL, \ + 0x64b62bcaebc387a1ULL, 0xd7986774e864d8ceULL, \ + 0x90321d9d438327faULL, 0x231c512340247895ULL, \ + 0x1f66e84e144cd992ULL, 0xac48a4f017eb86fdULL, \ + 0xebe2de19bc0c79c9ULL, 0x58cc92a7bfab26a6ULL, \ + 0x9317acc314dd3bc7ULL, 0x2039e07d177a64a8ULL, \ + 0x67939a94bc9d9b9cULL, 0xd4bdd62abf3ac4f3ULL, \ + 0xe8c76f47eb5265f4ULL, 0x5be923f9e8f53a9bULL, \ + 0x1c4359104312c5afULL, 0xaf6d15ae40b59ac0ULL, \ + 0x192d8af2baf0e1e8ULL, 0xaa03c64cb957be87ULL, \ + 0xeda9bca512b041b3ULL, 0x5e87f01b11171edcULL, \ + 0x62fd4976457fbfdbULL, 0xd1d305c846d8e0b4ULL, \ + 0x96797f21ed3f1f80ULL, 0x2557339fee9840efULL, \ + 0xee8c0dfb45ee5d8eULL, 0x5da24145464902e1ULL, \ + 0x1a083bacedaefdd5ULL, 0xa9267712ee09a2baULL, \ + 0x955cce7fba6103bdULL, 0x267282c1b9c65cd2ULL, \ + 0x61d8f8281221a3e6ULL, 0xd2f6b4961186fc89ULL, \ + 0x9f8169ba49a54b33ULL, 0x2caf25044a02145cULL, \ + 0x6b055fede1e5eb68ULL, 0xd82b1353e242b407ULL, \ + 0xe451aa3eb62a1500ULL, 0x577fe680b58d4a6fULL, \ + 0x10d59c691e6ab55bULL, 0xa3fbd0d71dcdea34ULL, \ + 0x6820eeb3b6bbf755ULL, 0xdb0ea20db51ca83aULL, \ + 0x9ca4d8e41efb570eULL, 0x2f8a945a1d5c0861ULL, \ + 0x13f02d374934a966ULL, 0xa0de61894a93f609ULL, \ + 0xe7741b60e174093dULL, 0x545a57dee2d35652ULL, \ + 0xe21ac88218962d7aULL, 0x5134843c1b317215ULL, \ + 0x169efed5b0d68d21ULL, 0xa5b0b26bb371d24eULL, \ + 0x99ca0b06e7197349ULL, 0x2ae447b8e4be2c26ULL, \ + 0x6d4e3d514f59d312ULL, 0xde6071ef4cfe8c7dULL, \ + 0x15bb4f8be788911cULL, 0xa6950335e42fce73ULL, \ + 0xe13f79dc4fc83147ULL, 0x521135624c6f6e28ULL, \ + 0x6e6b8c0f1807cf2fULL, 0xdd45c0b11ba09040ULL, \ + 0x9aefba58b0476f74ULL, 0x29c1f6e6b3e0301bULL, \ + 0xc96c5795d7870f42ULL, 0x7a421b2bd420502dULL, \ + 0x3de861c27fc7af19ULL, 0x8ec62d7c7c60f076ULL, \ + 0xb2bc941128085171ULL, 0x0192d8af2baf0e1eULL, \ + 0x4638a2468048f12aULL, 0xf516eef883efae45ULL, \ + 0x3ecdd09c2899b324ULL, 0x8de39c222b3eec4bULL, \ + 0xca49e6cb80d9137fULL, 0x7967aa75837e4c10ULL, \ + 0x451d1318d716ed17ULL, 0xf6335fa6d4b1b278ULL, \ + 0xb199254f7f564d4cULL, 0x02b769f17cf11223ULL, \ + 0xb4f7f6ad86b4690bULL, 0x07d9ba1385133664ULL, \ + 0x4073c0fa2ef4c950ULL, 0xf35d8c442d53963fULL, \ + 0xcf273529793b3738ULL, 0x7c0979977a9c6857ULL, \ + 0x3ba3037ed17b9763ULL, 0x888d4fc0d2dcc80cULL, \ + 0x435671a479aad56dULL, 0xf0783d1a7a0d8a02ULL, \ + 0xb7d247f3d1ea7536ULL, 0x04fc0b4dd24d2a59ULL, \ + 0x3886b22086258b5eULL, 0x8ba8fe9e8582d431ULL, \ + 0xcc0284772e652b05ULL, 0x7f2cc8c92dc2746aULL, \ + 0x325b15e575e1c3d0ULL, 0x8175595b76469cbfULL, \ + 0xc6df23b2dda1638bULL, 0x75f16f0cde063ce4ULL, \ + 0x498bd6618a6e9de3ULL, 0xfaa59adf89c9c28cULL, \ + 0xbd0fe036222e3db8ULL, 0x0e21ac88218962d7ULL, \ + 0xc5fa92ec8aff7fb6ULL, 0x76d4de52895820d9ULL, \ + 0x317ea4bb22bfdfedULL, 0x8250e80521188082ULL, \ + 0xbe2a516875702185ULL, 0x0d041dd676d77eeaULL, \ + 0x4aae673fdd3081deULL, 0xf9802b81de97deb1ULL, \ + 0x4fc0b4dd24d2a599ULL, 0xfceef8632775faf6ULL, \ + 0xbb44828a8c9205c2ULL, 0x086ace348f355aadULL, \ + 0x34107759db5dfbaaULL, 0x873e3be7d8faa4c5ULL, \ + 0xc094410e731d5bf1ULL, 0x73ba0db070ba049eULL, \ + 0xb86133d4dbcc19ffULL, 0x0b4f7f6ad86b4690ULL, \ + 0x4ce50583738cb9a4ULL, 0xffcb493d702be6cbULL, \ + 0xc3b1f050244347ccULL, 0x709fbcee27e418a3ULL, \ + 0x3735c6078c03e797ULL, 0x841b8ab98fa4b8f8ULL, \ + 0xadda7c5f3c4488e3ULL, 0x1ef430e13fe3d78cULL, \ + 0x595e4a08940428b8ULL, 0xea7006b697a377d7ULL, \ + 0xd60abfdbc3cbd6d0ULL, 0x6524f365c06c89bfULL, \ + 0x228e898c6b8b768bULL, 0x91a0c532682c29e4ULL, \ + 0x5a7bfb56c35a3485ULL, 0xe955b7e8c0fd6beaULL, \ + 0xaeffcd016b1a94deULL, 0x1dd181bf68bdcbb1ULL, \ + 0x21ab38d23cd56ab6ULL, 0x9285746c3f7235d9ULL, \ + 0xd52f0e859495caedULL, 0x6601423b97329582ULL, \ + 0xd041dd676d77eeaaULL, 0x636f91d96ed0b1c5ULL, \ + 0x24c5eb30c5374ef1ULL, 0x97eba78ec690119eULL, \ + 0xab911ee392f8b099ULL, 0x18bf525d915feff6ULL, \ + 0x5f1528b43ab810c2ULL, 0xec3b640a391f4fadULL, \ + 0x27e05a6e926952ccULL, 0x94ce16d091ce0da3ULL, \ + 0xd3646c393a29f297ULL, 0x604a2087398eadf8ULL, \ + 0x5c3099ea6de60cffULL, 0xef1ed5546e415390ULL, \ + 0xa8b4afbdc5a6aca4ULL, 0x1b9ae303c601f3cbULL, \ + 0x56ed3e2f9e224471ULL, 0xe5c372919d851b1eULL, \ + 0xa26908783662e42aULL, 0x114744c635c5bb45ULL, \ + 0x2d3dfdab61ad1a42ULL, 0x9e13b115620a452dULL, \ + 0xd9b9cbfcc9edba19ULL, 0x6a978742ca4ae576ULL, \ + 0xa14cb926613cf817ULL, 0x1262f598629ba778ULL, \ + 0x55c88f71c97c584cULL, 0xe6e6c3cfcadb0723ULL, \ + 0xda9c7aa29eb3a624ULL, 0x69b2361c9d14f94bULL, \ + 0x2e184cf536f3067fULL, 0x9d36004b35545910ULL, \ + 0x2b769f17cf112238ULL, 0x9858d3a9ccb67d57ULL, \ + 0xdff2a94067518263ULL, 0x6cdce5fe64f6dd0cULL, \ + 0x50a65c93309e7c0bULL, 0xe388102d33392364ULL, \ + 0xa4226ac498dedc50ULL, 0x170c267a9b79833fULL, \ + 0xdcd7181e300f9e5eULL, 0x6ff954a033a8c131ULL, \ + 0x28532e49984f3e05ULL, 0x9b7d62f79be8616aULL, \ + 0xa707db9acf80c06dULL, 0x14299724cc279f02ULL, \ + 0x5383edcd67c06036ULL, 0xe0ada17364673f59ULL} \ +} + +/* + * Return the initial CRC seed. Use the value returned from this API as the + * "crc" parameter to the first call to add data. + */ +static inline uint64_t fman_crc64_init(void) +{ + return fman_crc64_ecma_182.initial; +} + +/* Updates the CRC with arbitrary data */ +static inline uint64_t fman_crc64_update(uint64_t crc, + void *data, unsigned int len) +{ + uint8_t *p = data; + while (len--) + crc = fman_crc64_ecma_182.table[(crc ^ *(p++)) & 0xff] ^ + (crc >> 8); + return crc; +} + +/* Shorthands for updating the CRC with 8/16/32 bits of data. + * IMPORTANT NOTE: the typed "data" arguments should not be mistaken for + * host-endian numerical values, the assumption is that these values contain + * big-endian (ie. network byte order) data. + */ +static inline uint64_t fman_crc64_compute_32bit(uint32_t data, uint64_t crc) +{ + return fman_crc64_update(crc, &data, sizeof(data)); +} +static inline uint64_t fman_crc64_compute_16bit(uint16_t data, uint64_t crc) +{ + return fman_crc64_update(crc, &data, sizeof(data)); +} +static inline uint64_t fman_crc64_compute_8bit(uint8_t data, uint64_t crc) +{ + return fman_crc64_update(crc, &data, sizeof(data)); +} + +/* + * Finalise the CRC (using 2's complement) + */ +static inline uint64_t fman_crc64_finish(uint64_t seed) +{ + return ~seed; +} + +#ifdef __cplusplus +} +#endif + +#endif /* __FSL_FMAN_CRC64_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_qman.h b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_qman.h new file mode 100644 index 000000000..78b698f39 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_qman.h @@ -0,0 +1,2139 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2008-2012 Freescale Semiconductor, Inc. + * Copyright 2019 NXP + * + */ + +#ifndef __FSL_QMAN_H +#define __FSL_QMAN_H + +#ifdef __cplusplus +extern "C" { +#endif + +#include <dpaa_rbtree.h> +#include <rte_eventdev.h> + +/* FQ lookups (turn this on for 64bit user-space) */ +#if (__WORDSIZE == 64) +#define CONFIG_FSL_QMAN_FQ_LOOKUP +/* if FQ lookups are supported, this controls the number of initialised, + * s/w-consumed FQs that can be supported at any one time. + */ +#define CONFIG_FSL_QMAN_FQ_LOOKUP_MAX (32 * 1024) +#endif + +/* Last updated for v00.800 of the BG */ + +/* Hardware constants */ +#define QM_CHANNEL_SWPORTAL0 0 +#define QMAN_CHANNEL_POOL1 0x21 +#define QMAN_CHANNEL_CAAM 0x80 +#define QMAN_CHANNEL_PME 0xa0 +#define QMAN_CHANNEL_POOL1_REV3 0x401 +#define QMAN_CHANNEL_CAAM_REV3 0x840 +#define QMAN_CHANNEL_PME_REV3 0x860 +extern u16 qm_channel_pool1; +extern u16 qm_channel_caam; +extern u16 qm_channel_pme; +enum qm_dc_portal { + qm_dc_portal_fman0 = 0, + qm_dc_portal_fman1 = 1, + qm_dc_portal_caam = 2, + qm_dc_portal_pme = 3 +}; + +__rte_internal +u16 dpaa_get_qm_channel_caam(void); + +__rte_internal +u16 dpaa_get_qm_channel_pool(void); + +/* Portal processing (interrupt) sources */ +#define QM_PIRQ_CCSCI 0x00200000 /* CEETM Congestion State Change */ +#define QM_PIRQ_CSCI 0x00100000 /* Congestion State Change */ +#define QM_PIRQ_EQCI 0x00080000 /* Enqueue Command Committed */ +#define QM_PIRQ_EQRI 0x00040000 /* EQCR Ring (below threshold) */ +#define QM_PIRQ_DQRI 0x00020000 /* DQRR Ring (non-empty) */ +#define QM_PIRQ_MRI 0x00010000 /* MR Ring (non-empty) */ +/* + * This mask contains all the interrupt sources that need handling except DQRI, + * ie. that if present should trigger slow-path processing. + */ +#define QM_PIRQ_SLOW (QM_PIRQ_CSCI | QM_PIRQ_EQCI | QM_PIRQ_EQRI | \ + QM_PIRQ_MRI | QM_PIRQ_CCSCI) + +/* For qman_static_dequeue_*** APIs */ +#define QM_SDQCR_CHANNELS_POOL_MASK 0x00007fff +/* for n in [1,15] */ +#define QM_SDQCR_CHANNELS_POOL(n) (0x00008000 >> (n)) +/* for conversion from n of qm_channel */ +static inline u32 QM_SDQCR_CHANNELS_POOL_CONV(u16 channel) +{ + return QM_SDQCR_CHANNELS_POOL(channel + 1 - dpaa_get_qm_channel_pool()); +} + +/* For qman_volatile_dequeue(); Choose one PRECEDENCE. EXACT is optional. Use + * NUMFRAMES(n) (6-bit) or NUMFRAMES_TILLEMPTY to fill in the frame-count. Use + * FQID(n) to fill in the frame queue ID. + */ +#define QM_VDQCR_PRECEDENCE_VDQCR 0x0 +#define QM_VDQCR_PRECEDENCE_SDQCR 0x80000000 +#define QM_VDQCR_EXACT 0x40000000 +#define QM_VDQCR_NUMFRAMES_MASK 0x3f000000 +#define QM_VDQCR_NUMFRAMES_SET(n) (((n) & 0x3f) << 24) +#define QM_VDQCR_NUMFRAMES_GET(n) (((n) >> 24) & 0x3f) +#define QM_VDQCR_NUMFRAMES_TILLEMPTY QM_VDQCR_NUMFRAMES_SET(0) + +/* --- QMan data structures (and associated constants) --- */ + +/* Represents s/w corenet portal mapped data structures */ +struct qm_eqcr_entry; /* EQCR (EnQueue Command Ring) entries */ +struct qm_dqrr_entry; /* DQRR (DeQueue Response Ring) entries */ +struct qm_mr_entry; /* MR (Message Ring) entries */ +struct qm_mc_command; /* MC (Management Command) command */ +struct qm_mc_result; /* MC result */ + +#define QM_FD_FORMAT_SG 0x4 +#define QM_FD_FORMAT_LONG 0x2 +#define QM_FD_FORMAT_COMPOUND 0x1 +enum qm_fd_format { + /* + * 'contig' implies a contiguous buffer, whereas 'sg' implies a + * scatter-gather table. 'big' implies a 29-bit length with no offset + * field, otherwise length is 20-bit and offset is 9-bit. 'compound' + * implies a s/g-like table, where each entry itself represents a frame + * (contiguous or scatter-gather) and the 29-bit "length" is + * interpreted purely for congestion calculations, ie. a "congestion + * weight". + */ + qm_fd_contig = 0, + qm_fd_contig_big = QM_FD_FORMAT_LONG, + qm_fd_sg = QM_FD_FORMAT_SG, + qm_fd_sg_big = QM_FD_FORMAT_SG | QM_FD_FORMAT_LONG, + qm_fd_compound = QM_FD_FORMAT_COMPOUND +}; + +/* Capitalised versions are un-typed but can be used in static expressions */ +#define QM_FD_CONTIG 0 +#define QM_FD_CONTIG_BIG QM_FD_FORMAT_LONG +#define QM_FD_SG QM_FD_FORMAT_SG +#define QM_FD_SG_BIG (QM_FD_FORMAT_SG | QM_FD_FORMAT_LONG) +#define QM_FD_COMPOUND QM_FD_FORMAT_COMPOUND + +/* "Frame Descriptor (FD)" */ +struct qm_fd { + union { + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u8 dd:2; /* dynamic debug */ + u8 liodn_offset:6; + u8 bpid:8; /* Buffer Pool ID */ + u8 eliodn_offset:4; + u8 __reserved:4; + u8 addr_hi; /* high 8-bits of 40-bit address */ + u32 addr_lo; /* low 32-bits of 40-bit address */ +#else + u8 liodn_offset:6; + u8 dd:2; /* dynamic debug */ + u8 bpid:8; /* Buffer Pool ID */ + u8 __reserved:4; + u8 eliodn_offset:4; + u8 addr_hi; /* high 8-bits of 40-bit address */ + u32 addr_lo; /* low 32-bits of 40-bit address */ +#endif + }; + struct { + u64 __notaddress:24; + /* More efficient address accessor */ + u64 addr:40; + }; + u64 opaque_addr; + }; + /* The 'format' field indicates the interpretation of the remaining 29 + * bits of the 32-bit word. For packing reasons, it is duplicated in the + * other union elements. Note, union'd structs are difficult to use with + * static initialisation under gcc, in which case use the "opaque" form + * with one of the macros. + */ + union { + /* For easier/faster copying of this part of the fd (eg. from a + * DQRR entry to an EQCR entry) copy 'opaque' + */ + u32 opaque; + /* If 'format' is _contig or _sg, 20b length and 9b offset */ + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + enum qm_fd_format format:3; + u16 offset:9; + u32 length20:20; +#else + u32 length20:20; + u16 offset:9; + enum qm_fd_format format:3; +#endif + }; + /* If 'format' is _contig_big or _sg_big, 29b length */ + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + enum qm_fd_format _format1:3; + u32 length29:29; +#else + u32 length29:29; + enum qm_fd_format _format1:3; +#endif + }; + /* If 'format' is _compound, 29b "congestion weight" */ + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + enum qm_fd_format _format2:3; + u32 cong_weight:29; +#else + u32 cong_weight:29; + enum qm_fd_format _format2:3; +#endif + }; + }; + union { + u32 cmd; + u32 status; + }; +} __rte_aligned(8); +#define QM_FD_DD_NULL 0x00 +#define QM_FD_PID_MASK 0x3f +static inline u64 qm_fd_addr_get64(const struct qm_fd *fd) +{ + return fd->addr; +} + +static inline dma_addr_t qm_fd_addr(const struct qm_fd *fd) +{ + return (dma_addr_t)fd->addr; +} + +/* Macro, so we compile better if 'v' isn't always 64-bit */ +#define qm_fd_addr_set64(fd, v) \ + do { \ + struct qm_fd *__fd931 = (fd); \ + __fd931->addr = v; \ + } while (0) + +/* Scatter/Gather table entry */ +struct qm_sg_entry { + union { + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u8 __reserved1[3]; + u8 addr_hi; /* high 8-bits of 40-bit address */ + u32 addr_lo; /* low 32-bits of 40-bit address */ +#else + u32 addr_lo; /* low 32-bits of 40-bit address */ + u8 addr_hi; /* high 8-bits of 40-bit address */ + u8 __reserved1[3]; +#endif + }; + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u64 __notaddress:24; + u64 addr:40; +#else + u64 addr:40; + u64 __notaddress:24; +#endif + }; + u64 opaque; + }; + union { + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u32 extension:1; /* Extension bit */ + u32 final:1; /* Final bit */ + u32 length:30; +#else + u32 length:30; + u32 final:1; /* Final bit */ + u32 extension:1; /* Extension bit */ +#endif + }; + u32 val; + }; + u8 __reserved2; + u8 bpid; + union { + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u16 __reserved3:3; + u16 offset:13; +#else + u16 offset:13; + u16 __reserved3:3; +#endif + }; + u16 val_off; + }; +} __packed; +static inline u64 qm_sg_entry_get64(const struct qm_sg_entry *sg) +{ + return sg->addr; +} + +static inline dma_addr_t qm_sg_addr(const struct qm_sg_entry *sg) +{ + return (dma_addr_t)sg->addr; +} + +/* Macro, so we compile better if 'v' isn't always 64-bit */ +#define qm_sg_entry_set64(sg, v) \ + do { \ + struct qm_sg_entry *__sg931 = (sg); \ + __sg931->addr = v; \ + } while (0) + +/* See 1.5.8.1: "Enqueue Command" */ +struct __rte_aligned(8) qm_eqcr_entry { + u8 __dont_write_directly__verb; + u8 dca; + u16 seqnum; + u32 orp; /* 24-bit */ + u32 fqid; /* 24-bit */ + u32 tag; + struct qm_fd fd; /* this has alignment 8 */ + u8 __reserved3[32]; +} __packed; + + +/* "Frame Dequeue Response" */ +struct __rte_aligned(8) qm_dqrr_entry { + u8 verb; + u8 stat; + u16 seqnum; /* 15-bit */ + u8 tok; + u8 __reserved2[3]; + u32 fqid; /* 24-bit */ + u32 contextB; + struct qm_fd fd; /* this has alignment 8 */ + u8 __reserved4[32]; +}; + +#define QM_DQRR_VERB_VBIT 0x80 +#define QM_DQRR_VERB_MASK 0x7f /* where the verb contains; */ +#define QM_DQRR_VERB_FRAME_DEQUEUE 0x60 /* "this format" */ +#define QM_DQRR_STAT_FQ_EMPTY 0x80 /* FQ empty */ +#define QM_DQRR_STAT_FQ_HELDACTIVE 0x40 /* FQ held active */ +#define QM_DQRR_STAT_FQ_FORCEELIGIBLE 0x20 /* FQ was force-eligible'd */ +#define QM_DQRR_STAT_FD_VALID 0x10 /* has a non-NULL FD */ +#define QM_DQRR_STAT_UNSCHEDULED 0x02 /* Unscheduled dequeue */ +#define QM_DQRR_STAT_DQCR_EXPIRED 0x01 /* VDQCR or PDQCR expired*/ + + +/* "ERN Message Response" */ +/* "FQ State Change Notification" */ +struct qm_mr_entry { + union { + struct { + u8 verb; + u8 dca; + u16 seqnum; + u8 rc; /* Rejection Code */ + u32 orp:24; + u32 fqid; /* 24-bit */ + u32 tag; + struct qm_fd fd; /* this has alignment 8 */ + } __packed __rte_aligned(8) ern; + struct { + u8 verb; +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u8 colour:2; /* See QM_MR_DCERN_COLOUR_* */ + u8 __reserved1:4; + enum qm_dc_portal portal:2; +#else + enum qm_dc_portal portal:3; + u8 __reserved1:3; + u8 colour:2; /* See QM_MR_DCERN_COLOUR_* */ +#endif + u16 __reserved2; + u8 rc; /* Rejection Code */ + u32 __reserved3:24; + u32 fqid; /* 24-bit */ + u32 tag; + struct qm_fd fd; /* this has alignment 8 */ + } __packed __rte_aligned(8) dcern; + struct { + u8 verb; + u8 fqs; /* Frame Queue Status */ + u8 __reserved1[6]; + u32 fqid; /* 24-bit */ + u32 contextB; + u8 __reserved2[16]; + } __packed __rte_aligned(8) fq; /* FQRN/FQRNI/FQRL/FQPN */ + }; + u8 __reserved2[32]; +} __packed __rte_aligned(8); +#define QM_MR_VERB_VBIT 0x80 +/* + * ERNs originating from direct-connect portals ("dcern") use 0x20 as a verb + * which would be invalid as a s/w enqueue verb. A s/w ERN can be distinguished + * from the other MR types by noting if the 0x20 bit is unset. + */ +#define QM_MR_VERB_TYPE_MASK 0x27 +#define QM_MR_VERB_DC_ERN 0x20 +#define QM_MR_VERB_FQRN 0x21 +#define QM_MR_VERB_FQRNI 0x22 +#define QM_MR_VERB_FQRL 0x23 +#define QM_MR_VERB_FQPN 0x24 +#define QM_MR_RC_MASK 0xf0 /* contains one of; */ +#define QM_MR_RC_CGR_TAILDROP 0x00 +#define QM_MR_RC_WRED 0x10 +#define QM_MR_RC_ERROR 0x20 +#define QM_MR_RC_ORPWINDOW_EARLY 0x30 +#define QM_MR_RC_ORPWINDOW_LATE 0x40 +#define QM_MR_RC_FQ_TAILDROP 0x50 +#define QM_MR_RC_ORPWINDOW_RETIRED 0x60 +#define QM_MR_RC_ORP_ZERO 0x70 +#define QM_MR_FQS_ORLPRESENT 0x02 /* ORL fragments to come */ +#define QM_MR_FQS_NOTEMPTY 0x01 /* FQ has enqueued frames */ +#define QM_MR_DCERN_COLOUR_GREEN 0x00 +#define QM_MR_DCERN_COLOUR_YELLOW 0x01 +#define QM_MR_DCERN_COLOUR_RED 0x02 +#define QM_MR_DCERN_COLOUR_OVERRIDE 0x03 +/* + * An identical structure of FQD fields is present in the "Init FQ" command and + * the "Query FQ" result, it's suctioned out into the "struct qm_fqd" type. + * Within that, the 'stashing' and 'taildrop' pieces are also factored out, the + * latter has two inlines to assist with converting to/from the mant+exp + * representation. + */ +struct qm_fqd_stashing { + /* See QM_STASHING_EXCL_<...> */ +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u8 exclusive; + u8 __reserved1:2; + /* Numbers of cachelines */ + u8 annotation_cl:2; + u8 data_cl:2; + u8 context_cl:2; +#else + u8 context_cl:2; + u8 data_cl:2; + u8 annotation_cl:2; + u8 __reserved1:2; + u8 exclusive; +#endif +} __packed; +struct qm_fqd_taildrop { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u16 __reserved1:3; + u16 mant:8; + u16 exp:5; +#else + u16 exp:5; + u16 mant:8; + u16 __reserved1:3; +#endif +} __packed; +struct qm_fqd_oac { + /* "Overhead Accounting Control", see QM_OAC_<...> */ +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u8 oac:2; /* "Overhead Accounting Control" */ + u8 __reserved1:6; +#else + u8 __reserved1:6; + u8 oac:2; /* "Overhead Accounting Control" */ +#endif + /* Two's-complement value (-128 to +127) */ + signed char oal; /* "Overhead Accounting Length" */ +} __packed; +struct qm_fqd { + union { + u8 orpc; + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u8 __reserved1:2; + u8 orprws:3; + u8 oa:1; + u8 olws:2; +#else + u8 olws:2; + u8 oa:1; + u8 orprws:3; + u8 __reserved1:2; +#endif + } __packed; + }; + u8 cgid; + u16 fq_ctrl; /* See QM_FQCTRL_<...> */ + union { + u16 dest_wq; + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u16 channel:13; /* qm_channel */ + u16 wq:3; +#else + u16 wq:3; + u16 channel:13; /* qm_channel */ +#endif + } __packed dest; + }; +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u16 __reserved2:1; + u16 ics_cred:15; +#else + u16 __reserved2:1; + u16 ics_cred:15; +#endif + /* + * For "Initialize Frame Queue" commands, the write-enable mask + * determines whether 'td' or 'oac_init' is observed. For query + * commands, this field is always 'td', and 'oac_query' (below) reflects + * the Overhead ACcounting values. + */ + union { + uint16_t opaque_td; + struct qm_fqd_taildrop td; + struct qm_fqd_oac oac_init; + }; + u32 context_b; + union { + /* Treat it as 64-bit opaque */ + u64 opaque; + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u32 hi; + u32 lo; +#else + u32 lo; + u32 hi; +#endif + }; + /* Treat it as s/w portal stashing config */ + /* see "FQD Context_A field used for [...]" */ + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + struct qm_fqd_stashing stashing; + /* + * 48-bit address of FQ context to + * stash, must be cacheline-aligned + */ + u16 context_hi; + u32 context_lo; +#else + u32 context_lo; + u16 context_hi; + struct qm_fqd_stashing stashing; +#endif + } __packed; + } context_a; + struct qm_fqd_oac oac_query; +} __packed; +/* 64-bit converters for context_hi/lo */ +static inline u64 qm_fqd_stashing_get64(const struct qm_fqd *fqd) +{ + return ((u64)fqd->context_a.context_hi << 32) | + (u64)fqd->context_a.context_lo; +} + +static inline dma_addr_t qm_fqd_stashing_addr(const struct qm_fqd *fqd) +{ + return (dma_addr_t)qm_fqd_stashing_get64(fqd); +} + +static inline u64 qm_fqd_context_a_get64(const struct qm_fqd *fqd) +{ + return ((u64)fqd->context_a.hi << 32) | + (u64)fqd->context_a.lo; +} + +static inline void qm_fqd_stashing_set64(struct qm_fqd *fqd, u64 addr) +{ + fqd->context_a.context_hi = upper_32_bits(addr); + fqd->context_a.context_lo = lower_32_bits(addr); +} + +static inline void qm_fqd_context_a_set64(struct qm_fqd *fqd, u64 addr) +{ + fqd->context_a.hi = upper_32_bits(addr); + fqd->context_a.lo = lower_32_bits(addr); +} + +/* convert a threshold value into mant+exp representation */ +static inline int qm_fqd_taildrop_set(struct qm_fqd_taildrop *td, u32 val, + int roundup) +{ + u32 e = 0; + int oddbit = 0; + + if (val > 0xe0000000) + return -ERANGE; + while (val > 0xff) { + oddbit = val & 1; + val >>= 1; + e++; + if (roundup && oddbit) + val++; + } + td->exp = e; + td->mant = val; + return 0; +} + +/* and the other direction */ +static inline u32 qm_fqd_taildrop_get(const struct qm_fqd_taildrop *td) +{ + return (u32)td->mant << td->exp; +} + + +/* See "Frame Queue Descriptor (FQD)" */ +/* Frame Queue Descriptor (FQD) field 'fq_ctrl' uses these constants */ +#define QM_FQCTRL_MASK 0x07ff /* 'fq_ctrl' flags; */ +#define QM_FQCTRL_CGE 0x0400 /* Congestion Group Enable */ +#define QM_FQCTRL_TDE 0x0200 /* Tail-Drop Enable */ +#define QM_FQCTRL_ORP 0x0100 /* ORP Enable */ +#define QM_FQCTRL_CTXASTASHING 0x0080 /* Context-A stashing */ +#define QM_FQCTRL_CPCSTASH 0x0040 /* CPC Stash Enable */ +#define QM_FQCTRL_FORCESFDR 0x0008 /* High-priority SFDRs */ +#define QM_FQCTRL_AVOIDBLOCK 0x0004 /* Don't block active */ +#define QM_FQCTRL_HOLDACTIVE 0x0002 /* Hold active in portal */ +#define QM_FQCTRL_PREFERINCACHE 0x0001 /* Aggressively cache FQD */ +#define QM_FQCTRL_LOCKINCACHE QM_FQCTRL_PREFERINCACHE /* older naming */ + +/* See "FQD Context_A field used for [...] */ +/* Frame Queue Descriptor (FQD) field 'CONTEXT_A' uses these constants */ +#define QM_STASHING_EXCL_ANNOTATION 0x04 +#define QM_STASHING_EXCL_DATA 0x02 +#define QM_STASHING_EXCL_CTX 0x01 + +/* See "Intra Class Scheduling" */ +/* FQD field 'OAC' (Overhead ACcounting) uses these constants */ +#define QM_OAC_ICS 0x2 /* Accounting for Intra-Class Scheduling */ +#define QM_OAC_CG 0x1 /* Accounting for Congestion Groups */ + +/* + * This struct represents the 32-bit "WR_PARM_[GYR]" parameters in CGR fields + * and associated commands/responses. The WRED parameters are calculated from + * these fields as follows; + * MaxTH = MA * (2 ^ Mn) + * Slope = SA / (2 ^ Sn) + * MaxP = 4 * (Pn + 1) + */ +struct qm_cgr_wr_parm { + union { + u32 word; + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u32 MA:8; + u32 Mn:5; + u32 SA:7; /* must be between 64-127 */ + u32 Sn:6; + u32 Pn:6; +#else + u32 Pn:6; + u32 Sn:6; + u32 SA:7; /* must be between 64-127 */ + u32 Mn:5; + u32 MA:8; +#endif + } __packed; + }; +} __packed; +/* + * This struct represents the 13-bit "CS_THRES" CGR field. In the corresponding + * management commands, this is padded to a 16-bit structure field, so that's + * how we represent it here. The congestion state threshold is calculated from + * these fields as follows; + * CS threshold = TA * (2 ^ Tn) + */ +struct qm_cgr_cs_thres { + union { + u16 hword; + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u16 __reserved:3; + u16 TA:8; + u16 Tn:5; +#else + u16 Tn:5; + u16 TA:8; + u16 __reserved:3; +#endif + } __packed; + }; +} __packed; +/* + * This identical structure of CGR fields is present in the "Init/Modify CGR" + * commands and the "Query CGR" result. It's suctioned out here into its own + * struct. + */ +struct __qm_mc_cgr { + struct qm_cgr_wr_parm wr_parm_g; + struct qm_cgr_wr_parm wr_parm_y; + struct qm_cgr_wr_parm wr_parm_r; + u8 wr_en_g; /* boolean, use QM_CGR_EN */ + u8 wr_en_y; /* boolean, use QM_CGR_EN */ + u8 wr_en_r; /* boolean, use QM_CGR_EN */ + u8 cscn_en; /* boolean, use QM_CGR_EN */ + union { + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u16 cscn_targ_upd_ctrl; /* use QM_CSCN_TARG_UDP_ */ + u16 cscn_targ_dcp_low; /* CSCN_TARG_DCP low-16bits */ +#else + u16 cscn_targ_dcp_low; /* CSCN_TARG_DCP low-16bits */ + u16 cscn_targ_upd_ctrl; /* use QM_CSCN_TARG_UDP_ */ +#endif + }; + u32 cscn_targ; /* use QM_CGR_TARG_* */ + }; + u8 cstd_en; /* boolean, use QM_CGR_EN */ + u8 cs; /* boolean, only used in query response */ + union { + struct qm_cgr_cs_thres cs_thres; + /* use qm_cgr_cs_thres_set64() */ + u16 __cs_thres; + }; + u8 mode; /* QMAN_CGR_MODE_FRAME not supported in rev1.0 */ +} __packed; +#define QM_CGR_EN 0x01 /* For wr_en_*, cscn_en, cstd_en */ +#define QM_CGR_TARG_UDP_CTRL_WRITE_BIT 0x8000 /* value written to portal bit*/ +#define QM_CGR_TARG_UDP_CTRL_DCP 0x4000 /* 0: SWP, 1: DCP */ +#define QM_CGR_TARG_PORTAL(n) (0x80000000 >> (n)) /* s/w portal, 0-9 */ +#define QM_CGR_TARG_FMAN0 0x00200000 /* direct-connect portal: fman0 */ +#define QM_CGR_TARG_FMAN1 0x00100000 /* : fman1 */ +/* Convert CGR thresholds to/from "cs_thres" format */ +static inline u64 qm_cgr_cs_thres_get64(const struct qm_cgr_cs_thres *th) +{ + return (u64)th->TA << th->Tn; +} + +static inline int qm_cgr_cs_thres_set64(struct qm_cgr_cs_thres *th, u64 val, + int roundup) +{ + u32 e = 0; + int oddbit = 0; + + while (val > 0xff) { + oddbit = val & 1; + val >>= 1; + e++; + if (roundup && oddbit) + val++; + } + th->Tn = e; + th->TA = val; + return 0; +} + +/* See 1.5.8.5.1: "Initialize FQ" */ +/* See 1.5.8.5.2: "Query FQ" */ +/* See 1.5.8.5.3: "Query FQ Non-Programmable Fields" */ +/* See 1.5.8.5.4: "Alter FQ State Commands " */ +/* See 1.5.8.6.1: "Initialize/Modify CGR" */ +/* See 1.5.8.6.2: "CGR Test Write" */ +/* See 1.5.8.6.3: "Query CGR" */ +/* See 1.5.8.6.4: "Query Congestion Group State" */ +struct qm_mcc_initfq { + u8 __reserved1; + u16 we_mask; /* Write Enable Mask */ + u32 fqid; /* 24-bit */ + u16 count; /* Initialises 'count+1' FQDs */ + struct qm_fqd fqd; /* the FQD fields go here */ + u8 __reserved3[30]; +} __packed; +struct qm_mcc_queryfq { + u8 __reserved1[3]; + u32 fqid; /* 24-bit */ + u8 __reserved2[56]; +} __packed; +struct qm_mcc_queryfq_np { + u8 __reserved1[3]; + u32 fqid; /* 24-bit */ + u8 __reserved2[56]; +} __packed; +struct qm_mcc_alterfq { + u8 __reserved1[3]; + u32 fqid; /* 24-bit */ + u8 __reserved2; + u8 count; /* number of consecutive FQID */ + u8 __reserved3[10]; + u32 context_b; /* frame queue context b */ + u8 __reserved4[40]; +} __packed; +struct qm_mcc_initcgr { + u8 __reserved1; + u16 we_mask; /* Write Enable Mask */ + struct __qm_mc_cgr cgr; /* CGR fields */ + u8 __reserved2[2]; + u8 cgid; + u8 __reserved4[32]; +} __packed; +struct qm_mcc_cgrtestwrite { + u8 __reserved1[2]; + u8 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */ + u32 i_bcnt_lo; /* low 32-bits of 40-bit */ + u8 __reserved2[23]; + u8 cgid; + u8 __reserved3[32]; +} __packed; +struct qm_mcc_querycgr { + u8 __reserved1[30]; + u8 cgid; + u8 __reserved2[32]; +} __packed; +struct qm_mcc_querycongestion { + u8 __reserved[63]; +} __packed; +struct qm_mcc_querywq { + u8 __reserved; + /* select channel if verb != QUERYWQ_DEDICATED */ + union { + u16 channel_wq; /* ignores wq (3 lsbits) */ + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u16 id:13; /* qm_channel */ + u16 __reserved1:3; +#else + u16 __reserved1:3; + u16 id:13; /* qm_channel */ +#endif + } __packed channel; + }; + u8 __reserved2[60]; +} __packed; + +struct qm_mc_command { + u8 __dont_write_directly__verb; + union { + struct qm_mcc_initfq initfq; + struct qm_mcc_queryfq queryfq; + struct qm_mcc_queryfq_np queryfq_np; + struct qm_mcc_alterfq alterfq; + struct qm_mcc_initcgr initcgr; + struct qm_mcc_cgrtestwrite cgrtestwrite; + struct qm_mcc_querycgr querycgr; + struct qm_mcc_querycongestion querycongestion; + struct qm_mcc_querywq querywq; + }; +} __packed; + +/* INITFQ-specific flags */ +#define QM_INITFQ_WE_MASK 0x01ff /* 'Write Enable' flags; */ +#define QM_INITFQ_WE_OAC 0x0100 +#define QM_INITFQ_WE_ORPC 0x0080 +#define QM_INITFQ_WE_CGID 0x0040 +#define QM_INITFQ_WE_FQCTRL 0x0020 +#define QM_INITFQ_WE_DESTWQ 0x0010 +#define QM_INITFQ_WE_ICSCRED 0x0008 +#define QM_INITFQ_WE_TDTHRESH 0x0004 +#define QM_INITFQ_WE_CONTEXTB 0x0002 +#define QM_INITFQ_WE_CONTEXTA 0x0001 +/* INITCGR/MODIFYCGR-specific flags */ +#define QM_CGR_WE_MASK 0x07ff /* 'Write Enable Mask'; */ +#define QM_CGR_WE_WR_PARM_G 0x0400 +#define QM_CGR_WE_WR_PARM_Y 0x0200 +#define QM_CGR_WE_WR_PARM_R 0x0100 +#define QM_CGR_WE_WR_EN_G 0x0080 +#define QM_CGR_WE_WR_EN_Y 0x0040 +#define QM_CGR_WE_WR_EN_R 0x0020 +#define QM_CGR_WE_CSCN_EN 0x0010 +#define QM_CGR_WE_CSCN_TARG 0x0008 +#define QM_CGR_WE_CSTD_EN 0x0004 +#define QM_CGR_WE_CS_THRES 0x0002 +#define QM_CGR_WE_MODE 0x0001 + +struct qm_mcr_initfq { + u8 __reserved1[62]; +} __packed; +struct qm_mcr_queryfq { + u8 __reserved1[8]; + struct qm_fqd fqd; /* the FQD fields are here */ + u8 __reserved2[30]; +} __packed; +struct qm_mcr_queryfq_np { + u8 __reserved1; + u8 state; /* QM_MCR_NP_STATE_*** */ +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u8 __reserved2; + u32 fqd_link:24; + u16 __reserved3:2; + u16 odp_seq:14; + u16 __reserved4:2; + u16 orp_nesn:14; + u16 __reserved5:1; + u16 orp_ea_hseq:15; + u16 __reserved6:1; + u16 orp_ea_tseq:15; + u8 __reserved7; + u32 orp_ea_hptr:24; + u8 __reserved8; + u32 orp_ea_tptr:24; + u8 __reserved9; + u32 pfdr_hptr:24; + u8 __reserved10; + u32 pfdr_tptr:24; + u8 __reserved11[5]; + u8 __reserved12:7; + u8 is:1; + u16 ics_surp; + u32 byte_cnt; + u8 __reserved13; + u32 frm_cnt:24; + u32 __reserved14; + u16 ra1_sfdr; /* QM_MCR_NP_RA1_*** */ + u16 ra2_sfdr; /* QM_MCR_NP_RA2_*** */ + u16 __reserved15; + u16 od1_sfdr; /* QM_MCR_NP_OD1_*** */ + u16 od2_sfdr; /* QM_MCR_NP_OD2_*** */ + u16 od3_sfdr; /* QM_MCR_NP_OD3_*** */ +#else + u8 __reserved2; + u32 fqd_link:24; + + u16 odp_seq:14; + u16 __reserved3:2; + + u16 orp_nesn:14; + u16 __reserved4:2; + + u16 orp_ea_hseq:15; + u16 __reserved5:1; + + u16 orp_ea_tseq:15; + u16 __reserved6:1; + + u8 __reserved7; + u32 orp_ea_hptr:24; + + u8 __reserved8; + u32 orp_ea_tptr:24; + + u8 __reserved9; + u32 pfdr_hptr:24; + + u8 __reserved10; + u32 pfdr_tptr:24; + + u8 __reserved11[5]; + u8 is:1; + u8 __reserved12:7; + u16 ics_surp; + u32 byte_cnt; + u8 __reserved13; + u32 frm_cnt:24; + u32 __reserved14; + u16 ra1_sfdr; /* QM_MCR_NP_RA1_*** */ + u16 ra2_sfdr; /* QM_MCR_NP_RA2_*** */ + u16 __reserved15; + u16 od1_sfdr; /* QM_MCR_NP_OD1_*** */ + u16 od2_sfdr; /* QM_MCR_NP_OD2_*** */ + u16 od3_sfdr; /* QM_MCR_NP_OD3_*** */ +#endif +} __packed; + +struct qm_mcr_alterfq { + u8 fqs; /* Frame Queue Status */ + u8 __reserved1[61]; +} __packed; +struct qm_mcr_initcgr { + u8 __reserved1[62]; +} __packed; +struct qm_mcr_cgrtestwrite { + u16 __reserved1; + struct __qm_mc_cgr cgr; /* CGR fields */ + u8 __reserved2[3]; + u32 __reserved3:24; + u32 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */ + u32 i_bcnt_lo; /* low 32-bits of 40-bit */ + u32 __reserved4:24; + u32 a_bcnt_hi:8;/* high 8-bits of 40-bit "Average" */ + u32 a_bcnt_lo; /* low 32-bits of 40-bit */ + u16 lgt; /* Last Group Tick */ + u16 wr_prob_g; + u16 wr_prob_y; + u16 wr_prob_r; + u8 __reserved5[8]; +} __packed; +struct qm_mcr_querycgr { + u16 __reserved1; + struct __qm_mc_cgr cgr; /* CGR fields */ + u8 __reserved2[3]; + union { + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u32 __reserved3:24; + u32 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */ + u32 i_bcnt_lo; /* low 32-bits of 40-bit */ +#else + u32 i_bcnt_lo; /* low 32-bits of 40-bit */ + u32 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */ + u32 __reserved3:24; +#endif + }; + u64 i_bcnt; + }; + union { + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u32 __reserved4:24; + u32 a_bcnt_hi:8;/* high 8-bits of 40-bit "Average" */ + u32 a_bcnt_lo; /* low 32-bits of 40-bit */ +#else + u32 a_bcnt_lo; /* low 32-bits of 40-bit */ + u32 a_bcnt_hi:8;/* high 8-bits of 40-bit "Average" */ + u32 __reserved4:24; +#endif + }; + u64 a_bcnt; + }; + union { + u32 cscn_targ_swp[4]; + u8 __reserved5[16]; + }; +} __packed; + +struct __qm_mcr_querycongestion { + u32 state[8]; +}; + +struct qm_mcr_querycongestion { + u8 __reserved[30]; + /* Access this struct using QM_MCR_QUERYCONGESTION() */ + struct __qm_mcr_querycongestion state; +} __packed; +struct qm_mcr_querywq { + union { + u16 channel_wq; /* ignores wq (3 lsbits) */ + struct { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + u16 id:13; /* qm_channel */ + u16 __reserved:3; +#else + u16 __reserved:3; + u16 id:13; /* qm_channel */ +#endif + } __packed channel; + }; + u8 __reserved[28]; + u32 wq_len[8]; +} __packed; + +struct qm_mc_result { + u8 verb; + u8 result; + union { + struct qm_mcr_initfq initfq; + struct qm_mcr_queryfq queryfq; + struct qm_mcr_queryfq_np queryfq_np; + struct qm_mcr_alterfq alterfq; + struct qm_mcr_initcgr initcgr; + struct qm_mcr_cgrtestwrite cgrtestwrite; + struct qm_mcr_querycgr querycgr; + struct qm_mcr_querycongestion querycongestion; + struct qm_mcr_querywq querywq; + }; +} __packed; + +#define QM_MCR_VERB_RRID 0x80 +#define QM_MCR_VERB_MASK QM_MCC_VERB_MASK +#define QM_MCR_VERB_INITFQ_PARKED QM_MCC_VERB_INITFQ_PARKED +#define QM_MCR_VERB_INITFQ_SCHED QM_MCC_VERB_INITFQ_SCHED +#define QM_MCR_VERB_QUERYFQ QM_MCC_VERB_QUERYFQ +#define QM_MCR_VERB_QUERYFQ_NP QM_MCC_VERB_QUERYFQ_NP +#define QM_MCR_VERB_QUERYWQ QM_MCC_VERB_QUERYWQ +#define QM_MCR_VERB_QUERYWQ_DEDICATED QM_MCC_VERB_QUERYWQ_DEDICATED +#define QM_MCR_VERB_ALTER_SCHED QM_MCC_VERB_ALTER_SCHED +#define QM_MCR_VERB_ALTER_FE QM_MCC_VERB_ALTER_FE +#define QM_MCR_VERB_ALTER_RETIRE QM_MCC_VERB_ALTER_RETIRE +#define QM_MCR_VERB_ALTER_OOS QM_MCC_VERB_ALTER_OOS +#define QM_MCR_RESULT_NULL 0x00 +#define QM_MCR_RESULT_OK 0xf0 +#define QM_MCR_RESULT_ERR_FQID 0xf1 +#define QM_MCR_RESULT_ERR_FQSTATE 0xf2 +#define QM_MCR_RESULT_ERR_NOTEMPTY 0xf3 /* OOS fails if FQ is !empty */ +#define QM_MCR_RESULT_ERR_BADCHANNEL 0xf4 +#define QM_MCR_RESULT_PENDING 0xf8 +#define QM_MCR_RESULT_ERR_BADCOMMAND 0xff +#define QM_MCR_NP_STATE_FE 0x10 +#define QM_MCR_NP_STATE_R 0x08 +#define QM_MCR_NP_STATE_MASK 0x07 /* Reads FQD::STATE; */ +#define QM_MCR_NP_STATE_OOS 0x00 +#define QM_MCR_NP_STATE_RETIRED 0x01 +#define QM_MCR_NP_STATE_TEN_SCHED 0x02 +#define QM_MCR_NP_STATE_TRU_SCHED 0x03 +#define QM_MCR_NP_STATE_PARKED 0x04 +#define QM_MCR_NP_STATE_ACTIVE 0x05 +#define QM_MCR_NP_PTR_MASK 0x07ff /* for RA[12] & OD[123] */ +#define QM_MCR_NP_RA1_NRA(v) (((v) >> 14) & 0x3) /* FQD::NRA */ +#define QM_MCR_NP_RA2_IT(v) (((v) >> 14) & 0x1) /* FQD::IT */ +#define QM_MCR_NP_OD1_NOD(v) (((v) >> 14) & 0x3) /* FQD::NOD */ +#define QM_MCR_NP_OD3_NPC(v) (((v) >> 14) & 0x3) /* FQD::NPC */ +#define QM_MCR_FQS_ORLPRESENT 0x02 /* ORL fragments to come */ +#define QM_MCR_FQS_NOTEMPTY 0x01 /* FQ has enqueued frames */ +/* This extracts the state for congestion group 'n' from a query response. + * Eg. + * u8 cgr = [...]; + * struct qm_mc_result *res = [...]; + * printf("congestion group %d congestion state: %d\n", cgr, + * QM_MCR_QUERYCONGESTION(&res->querycongestion.state, cgr)); + */ +#define __CGR_WORD(num) (num >> 5) +#define __CGR_SHIFT(num) (num & 0x1f) +#define __CGR_NUM (sizeof(struct __qm_mcr_querycongestion) << 3) +static inline int QM_MCR_QUERYCONGESTION(struct __qm_mcr_querycongestion *p, + u8 cgr) +{ + return p->state[__CGR_WORD(cgr)] & (0x80000000 >> __CGR_SHIFT(cgr)); +} + + /* Portal and Frame Queues */ +/* Represents a managed portal */ +struct qman_portal; + +/* + * This object type represents QMan frame queue descriptors (FQD), it is + * cacheline-aligned, and initialised by qman_create_fq(). The structure is + * defined further down. + */ +struct qman_fq; + +/* + * This object type represents a QMan congestion group, it is defined further + * down. + */ +struct qman_cgr; + +/* + * This enum, and the callback type that returns it, are used when handling + * dequeued frames via DQRR. Note that for "null" callbacks registered with the + * portal object (for handling dequeues that do not demux because context_b is + * NULL), the return value *MUST* be qman_cb_dqrr_consume. + */ +enum qman_cb_dqrr_result { + /* DQRR entry can be consumed */ + qman_cb_dqrr_consume, + /* Like _consume, but requests parking - FQ must be held-active */ + qman_cb_dqrr_park, + /* Does not consume, for DCA mode only. This allows out-of-order + * consumes by explicit calls to qman_dca() and/or the use of implicit + * DCA via EQCR entries. + */ + qman_cb_dqrr_defer, + /* + * Stop processing without consuming this ring entry. Exits the current + * qman_p_poll_dqrr() or interrupt-handling, as appropriate. If within + * an interrupt handler, the callback would typically call + * qman_irqsource_remove(QM_PIRQ_DQRI) before returning this value, + * otherwise the interrupt will reassert immediately. + */ + qman_cb_dqrr_stop, + /* Like qman_cb_dqrr_stop, but consumes the current entry. */ + qman_cb_dqrr_consume_stop +}; + +typedef enum qman_cb_dqrr_result (*qman_cb_dqrr)(struct qman_portal *qm, + struct qman_fq *fq, + const struct qm_dqrr_entry *dqrr); + +typedef enum qman_cb_dqrr_result (*qman_dpdk_cb_dqrr)(void *event, + struct qman_portal *qm, + struct qman_fq *fq, + const struct qm_dqrr_entry *dqrr, + void **bd); + +/* This callback type is used when handling buffers in dpdk pull mode */ +typedef void (*qman_dpdk_pull_cb_dqrr)(struct qman_fq **fq, + struct qm_dqrr_entry **dqrr, + void **bufs, + int num_bufs); + +typedef void (*qman_dpdk_cb_prepare)(struct qm_dqrr_entry *dq, void **bufs); + +/* + * This callback type is used when handling ERNs, FQRNs and FQRLs via MR. They + * are always consumed after the callback returns. + */ +typedef void (*qman_cb_mr)(struct qman_portal *qm, struct qman_fq *fq, + const struct qm_mr_entry *msg); + +/* This callback type is used when handling DCP ERNs */ +typedef void (*qman_cb_dc_ern)(struct qman_portal *qm, + const struct qm_mr_entry *msg); +/* + * s/w-visible states. Ie. tentatively scheduled + truly scheduled + active + + * held-active + held-suspended are just "sched". Things like "retired" will not + * be assumed until it is complete (ie. QMAN_FQ_STATE_CHANGING is set until + * then, to indicate it's completing and to gate attempts to retry the retire + * command). Note, park commands do not set QMAN_FQ_STATE_CHANGING because it's + * technically impossible in the case of enqueue DCAs (which refer to DQRR ring + * index rather than the FQ that ring entry corresponds to), so repeated park + * commands are allowed (if you're silly enough to try) but won't change FQ + * state, and the resulting park notifications move FQs from "sched" to + * "parked". + */ +enum qman_fq_state { + qman_fq_state_oos, + qman_fq_state_parked, + qman_fq_state_sched, + qman_fq_state_retired +}; + + +/* + * Frame queue objects (struct qman_fq) are stored within memory passed to + * qman_create_fq(), as this allows stashing of caller-provided demux callback + * pointers at no extra cost to stashing of (driver-internal) FQ state. If the + * caller wishes to add per-FQ state and have it benefit from dequeue-stashing, + * they should; + * + * (a) extend the qman_fq structure with their state; eg. + * + * // myfq is allocated and driver_fq callbacks filled in; + * struct my_fq { + * struct qman_fq base; + * int an_extra_field; + * [ ... add other fields to be associated with each FQ ...] + * } *myfq = some_my_fq_allocator(); + * struct qman_fq *fq = qman_create_fq(fqid, flags, &myfq->base); + * + * // in a dequeue callback, access extra fields from 'fq' via a cast; + * struct my_fq *myfq = (struct my_fq *)fq; + * do_something_with(myfq->an_extra_field); + * [...] + * + * (b) when and if configuring the FQ for context stashing, specify how ever + * many cachelines are required to stash 'struct my_fq', to accelerate not + * only the QMan driver but the callback as well. + */ + +struct qman_fq_cb { + union { /* for dequeued frames */ + qman_dpdk_cb_dqrr dqrr_dpdk_cb; + qman_dpdk_pull_cb_dqrr dqrr_dpdk_pull_cb; + qman_cb_dqrr dqrr; + }; + qman_dpdk_cb_prepare dqrr_prepare; + qman_cb_mr ern; /* for s/w ERNs */ + qman_cb_mr fqs; /* frame-queue state changes*/ +}; + +struct qman_fq { + /* Caller of qman_create_fq() provides these demux callbacks */ + struct qman_fq_cb cb; + + u32 fqid_le; + u32 fqid; + + int q_fd; + u16 ch_id; + u8 cgr_groupid; + u8 is_static:4; + u8 qp_initialized:4; + + /* DPDK Interface */ + void *dpaa_intf; + + struct rte_event ev; + /* affined portal in case of static queue */ + struct qman_portal *qp; + struct dpaa_bp_info *bp_array; + + volatile unsigned long flags; + + enum qman_fq_state state; + spinlock_t fqlock; + + struct rb_node node; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + void **qman_fq_lookup_table; + u32 key; +#endif +}; + +/* + * This callback type is used when handling congestion group entry/exit. + * 'congested' is non-zero on congestion-entry, and zero on congestion-exit. + */ +typedef void (*qman_cb_cgr)(struct qman_portal *qm, + struct qman_cgr *cgr, int congested); + +struct qman_cgr { + /* Set these prior to qman_create_cgr() */ + u32 cgrid; /* 0..255, but u32 to allow specials like -1, 256, etc.*/ + qman_cb_cgr cb; + /* These are private to the driver */ + u16 chan; /* portal channel this object is created on */ + struct list_head node; +}; + +/* Flags to qman_create_fq() */ +#define QMAN_FQ_FLAG_NO_ENQUEUE 0x00000001 /* can't enqueue */ +#define QMAN_FQ_FLAG_NO_MODIFY 0x00000002 /* can only enqueue */ +#define QMAN_FQ_FLAG_TO_DCPORTAL 0x00000004 /* consumed by CAAM/PME/Fman */ +#define QMAN_FQ_FLAG_LOCKED 0x00000008 /* multi-core locking */ +#define QMAN_FQ_FLAG_AS_IS 0x00000010 /* query h/w state */ +#define QMAN_FQ_FLAG_DYNAMIC_FQID 0x00000020 /* (de)allocate fqid */ + +/* Flags to qman_destroy_fq() */ +#define QMAN_FQ_DESTROY_PARKED 0x00000001 /* FQ can be parked or OOS */ + +/* Flags from qman_fq_state() */ +#define QMAN_FQ_STATE_CHANGING 0x80000000 /* 'state' is changing */ +#define QMAN_FQ_STATE_NE 0x40000000 /* retired FQ isn't empty */ +#define QMAN_FQ_STATE_ORL 0x20000000 /* retired FQ has ORL */ +#define QMAN_FQ_STATE_BLOCKOOS 0xe0000000 /* if any are set, no OOS */ +#define QMAN_FQ_STATE_CGR_EN 0x10000000 /* CGR enabled */ +#define QMAN_FQ_STATE_VDQCR 0x08000000 /* being volatile dequeued */ + +/* Flags to qman_init_fq() */ +#define QMAN_INITFQ_FLAG_SCHED 0x00000001 /* schedule rather than park */ +#define QMAN_INITFQ_FLAG_LOCAL 0x00000004 /* set dest portal */ + +/* Flags to qman_enqueue(). NB, the strange numbering is to align with hardware, + * bit-wise. (NB: the PME API is sensitive to these precise numberings too, so + * any change here should be audited in PME.) + */ +#define QMAN_ENQUEUE_FLAG_WATCH_CGR 0x00080000 /* watch congestion state */ +#define QMAN_ENQUEUE_FLAG_DCA 0x00008000 /* perform enqueue-DCA */ +#define QMAN_ENQUEUE_FLAG_DCA_PARK 0x00004000 /* If DCA, requests park */ +#define QMAN_ENQUEUE_FLAG_DCA_PTR(p) /* If DCA, p is DQRR entry */ \ + (((u32)(p) << 2) & 0x00000f00) +#define QMAN_ENQUEUE_FLAG_C_GREEN 0x00000000 /* choose one C_*** flag */ +#define QMAN_ENQUEUE_FLAG_C_YELLOW 0x00000008 +#define QMAN_ENQUEUE_FLAG_C_RED 0x00000010 +#define QMAN_ENQUEUE_FLAG_C_OVERRIDE 0x00000018 +/* For the ORP-specific qman_enqueue_orp() variant; + * - this flag indicates "Not Last In Sequence", ie. all but the final fragment + * of a frame. + */ +#define QMAN_ENQUEUE_FLAG_NLIS 0x01000000 +/* - this flag performs no enqueue but fills in an ORP sequence number that + * would otherwise block it (eg. if a frame has been dropped). + */ +#define QMAN_ENQUEUE_FLAG_HOLE 0x02000000 +/* - this flag performs no enqueue but advances NESN to the given sequence + * number. + */ +#define QMAN_ENQUEUE_FLAG_NESN 0x04000000 + +/* Flags to qman_modify_cgr() */ +#define QMAN_CGR_FLAG_USE_INIT 0x00000001 +#define QMAN_CGR_MODE_FRAME 0x00000001 + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +__rte_internal +void qman_set_fq_lookup_table(void **table); +#endif + +/** + * qman_get_portal_index - get portal configuration index + */ +int qman_get_portal_index(void); + +__rte_internal +u32 qman_portal_dequeue(struct rte_event ev[], unsigned int poll_limit, + void **bufs); + +/** + * qman_irqsource_add - add processing sources to be interrupt-driven + * @bits: bitmask of QM_PIRQ_**I processing sources + * + * Adds processing sources that should be interrupt-driven (rather than + * processed via qman_poll_***() functions). Returns zero for success, or + * -EINVAL if the current CPU is sharing a portal hosted on another CPU. + */ +__rte_internal +int qman_irqsource_add(u32 bits); + +/** + * qman_fq_portal_irqsource_add - samilar to qman_irqsource_add, but it + * takes portal (fq specific) as input rather than using the thread affined + * portal. + */ +__rte_internal +int qman_fq_portal_irqsource_add(struct qman_portal *p, u32 bits); + +/** + * qman_irqsource_remove - remove processing sources from being interrupt-driven + * @bits: bitmask of QM_PIRQ_**I processing sources + * + * Removes processing sources from being interrupt-driven, so that they will + * instead be processed via qman_poll_***() functions. Returns zero for success, + * or -EINVAL if the current CPU is sharing a portal hosted on another CPU. + */ +__rte_internal +int qman_irqsource_remove(u32 bits); + +/** + * qman_fq_portal_irqsource_remove - similar to qman_irqsource_remove, but it + * takes portal (fq specific) as input rather than using the thread affined + * portal. + */ +__rte_internal +int qman_fq_portal_irqsource_remove(struct qman_portal *p, u32 bits); + +/** + * qman_affine_channel - return the channel ID of an portal + * @cpu: the cpu whose affine portal is the subject of the query + * + * If @cpu is -1, the affine portal for the current CPU will be used. It is a + * bug to call this function for any value of @cpu (other than -1) that is not a + * member of the cpu mask. + */ +u16 qman_affine_channel(int cpu); + +__rte_internal +unsigned int qman_portal_poll_rx(unsigned int poll_limit, + void **bufs, struct qman_portal *q); + +/** + * qman_set_vdq - Issue a volatile dequeue command + * @fq: Frame Queue on which the volatile dequeue command is issued + * @num: Number of Frames requested for volatile dequeue + * @vdqcr_flags: QM_VDQCR_EXACT flag to for VDQCR command + * + * This function will issue a volatile dequeue command to the QMAN. + */ +__rte_internal +int qman_set_vdq(struct qman_fq *fq, u16 num, uint32_t vdqcr_flags); + +/** + * qman_dequeue - Get the DQRR entry after volatile dequeue command + * @fq: Frame Queue on which the volatile dequeue command is issued + * + * This function will return the DQRR entry after a volatile dequeue command + * is issued. It will keep returning NULL until there is no packet available on + * the DQRR. + */ +__rte_internal +struct qm_dqrr_entry *qman_dequeue(struct qman_fq *fq); + +/** + * qman_dqrr_consume - Consume the DQRR entriy after volatile dequeue + * @fq: Frame Queue on which the volatile dequeue command is issued + * @dq: DQRR entry to consume. This is the one which is provided by the + * 'qbman_dequeue' command. + * + * This will consume the DQRR enrey and make it available for next volatile + * dequeue. + */ +__rte_internal +void qman_dqrr_consume(struct qman_fq *fq, + struct qm_dqrr_entry *dq); + +/** + * qman_poll_dqrr - process DQRR (fast-path) entries + * @limit: the maximum number of DQRR entries to process + * + * Use of this function requires that DQRR processing not be interrupt-driven. + * Ie. the value returned by qman_irqsource_get() should not include + * QM_PIRQ_DQRI. If the current CPU is sharing a portal hosted on another CPU, + * this function will return -EINVAL, otherwise the return value is >=0 and + * represents the number of DQRR entries processed. + */ +__rte_internal +int qman_poll_dqrr(unsigned int limit); + +/** + * qman_poll + * + * Dispatcher logic on a cpu can use this to trigger any maintenance of the + * affine portal. There are two classes of portal processing in question; + * fast-path (which involves demuxing dequeue ring (DQRR) entries and tracking + * enqueue ring (EQCR) consumption), and slow-path (which involves EQCR + * thresholds, congestion state changes, etc). This function does whatever + * processing is not triggered by interrupts. + * + * Note, if DQRR and some slow-path processing are poll-driven (rather than + * interrupt-driven) then this function uses a heuristic to determine how often + * to run slow-path processing - as slow-path processing introduces at least a + * minimum latency each time it is run, whereas fast-path (DQRR) processing is + * close to zero-cost if there is no work to be done. + */ +void qman_poll(void); + +/** + * qman_stop_dequeues - Stop h/w dequeuing to the s/w portal + * + * Disables DQRR processing of the portal. This is reference-counted, so + * qman_start_dequeues() must be called as many times as qman_stop_dequeues() to + * truly re-enable dequeuing. + */ +void qman_stop_dequeues(void); + +/** + * qman_start_dequeues - (Re)start h/w dequeuing to the s/w portal + * + * Enables DQRR processing of the portal. This is reference-counted, so + * qman_start_dequeues() must be called as many times as qman_stop_dequeues() to + * truly re-enable dequeuing. + */ +void qman_start_dequeues(void); + +/** + * qman_static_dequeue_add - Add pool channels to the portal SDQCR + * @pools: bit-mask of pool channels, using QM_SDQCR_CHANNELS_POOL(n) + * + * Adds a set of pool channels to the portal's static dequeue command register + * (SDQCR). The requested pools are limited to those the portal has dequeue + * access to. + */ +__rte_internal +void qman_static_dequeue_add(u32 pools, struct qman_portal *qm); + +/** + * qman_static_dequeue_del - Remove pool channels from the portal SDQCR + * @pools: bit-mask of pool channels, using QM_SDQCR_CHANNELS_POOL(n) + * + * Removes a set of pool channels from the portal's static dequeue command + * register (SDQCR). The requested pools are limited to those the portal has + * dequeue access to. + */ +void qman_static_dequeue_del(u32 pools, struct qman_portal *qp); + +/** + * qman_static_dequeue_get - return the portal's current SDQCR + * + * Returns the portal's current static dequeue command register (SDQCR). The + * entire register is returned, so if only the currently-enabled pool channels + * are desired, mask the return value with QM_SDQCR_CHANNELS_POOL_MASK. + */ +u32 qman_static_dequeue_get(struct qman_portal *qp); + +/** + * qman_dca - Perform a Discrete Consumption Acknowledgment + * @dq: the DQRR entry to be consumed + * @park_request: indicates whether the held-active @fq should be parked + * + * Only allowed in DCA-mode portals, for DQRR entries whose handler callback had + * previously returned 'qman_cb_dqrr_defer'. NB, as with the other APIs, this + * does not take a 'portal' argument but implies the core affine portal from the + * cpu that is currently executing the function. For reasons of locking, this + * function must be called from the same CPU as that which processed the DQRR + * entry in the first place. + */ +void qman_dca(const struct qm_dqrr_entry *dq, int park_request); + +/** + * qman_dca_index - Perform a Discrete Consumption Acknowledgment + * @index: the DQRR index to be consumed + * @park_request: indicates whether the held-active @fq should be parked + * + * Only allowed in DCA-mode portals, for DQRR entries whose handler callback had + * previously returned 'qman_cb_dqrr_defer'. NB, as with the other APIs, this + * does not take a 'portal' argument but implies the core affine portal from the + * cpu that is currently executing the function. For reasons of locking, this + * function must be called from the same CPU as that which processed the DQRR + * entry in the first place. + */ +__rte_internal +void qman_dca_index(u8 index, int park_request); + +/** + * qman_eqcr_is_empty - Determine if portal's EQCR is empty + * + * For use in situations where a cpu-affine caller needs to determine when all + * enqueues for the local portal have been processed by Qman but can't use the + * QMAN_ENQUEUE_FLAG_WAIT_SYNC flag to do this from the final qman_enqueue(). + * The function forces tracking of EQCR consumption (which normally doesn't + * happen until enqueue processing needs to find space to put new enqueue + * commands), and returns zero if the ring still has unprocessed entries, + * non-zero if it is empty. + */ +int qman_eqcr_is_empty(void); + +/** + * qman_set_dc_ern - Set the handler for DCP enqueue rejection notifications + * @handler: callback for processing DCP ERNs + * @affine: whether this handler is specific to the locally affine portal + * + * If a hardware block's interface to Qman (ie. its direct-connect portal, or + * DCP) is configured not to receive enqueue rejections, then any enqueues + * through that DCP that are rejected will be sent to a given software portal. + * If @affine is non-zero, then this handler will only be used for DCP ERNs + * received on the portal affine to the current CPU. If multiple CPUs share a + * portal and they all call this function, they will be setting the handler for + * the same portal! If @affine is zero, then this handler will be global to all + * portals handled by this instance of the driver. Only those portals that do + * not have their own affine handler will use the global handler. + */ +void qman_set_dc_ern(qman_cb_dc_ern handler, int affine); + + /* FQ management */ + /* ------------- */ +/** + * qman_create_fq - Allocates a FQ + * @fqid: the index of the FQD to encapsulate, must be "Out of Service" + * @flags: bit-mask of QMAN_FQ_FLAG_*** options + * @fq: memory for storing the 'fq', with callbacks filled in + * + * Creates a frame queue object for the given @fqid, unless the + * QMAN_FQ_FLAG_DYNAMIC_FQID flag is set in @flags, in which case a FQID is + * dynamically allocated (or the function fails if none are available). Once + * created, the caller should not touch the memory at 'fq' except as extended to + * adjacent memory for user-defined fields (see the definition of "struct + * qman_fq" for more info). NO_MODIFY is only intended for enqueuing to + * pre-existing frame-queues that aren't to be otherwise interfered with, it + * prevents all other modifications to the frame queue. The TO_DCPORTAL flag + * causes the driver to honour any contextB modifications requested in the + * qm_init_fq() API, as this indicates the frame queue will be consumed by a + * direct-connect portal (PME, CAAM, or Fman). When frame queues are consumed by + * software portals, the contextB field is controlled by the driver and can't be + * modified by the caller. If the AS_IS flag is specified, management commands + * will be used on portal @p to query state for frame queue @fqid and construct + * a frame queue object based on that, rather than assuming/requiring that it be + * Out of Service. + */ +__rte_internal +int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq); + +/** + * qman_destroy_fq - Deallocates a FQ + * @fq: the frame queue object to release + * @flags: bit-mask of QMAN_FQ_FREE_*** options + * + * The memory for this frame queue object ('fq' provided in qman_create_fq()) is + * not deallocated but the caller regains ownership, to do with as desired. The + * FQ must be in the 'out-of-service' state unless the QMAN_FQ_FREE_PARKED flag + * is specified, in which case it may also be in the 'parked' state. + */ +void qman_destroy_fq(struct qman_fq *fq, u32 flags); + +/** + * qman_fq_fqid - Queries the frame queue ID of a FQ object + * @fq: the frame queue object to query + */ +__rte_internal +u32 qman_fq_fqid(struct qman_fq *fq); + +/** + * qman_fq_state - Queries the state of a FQ object + * @fq: the frame queue object to query + * @state: pointer to state enum to return the FQ scheduling state + * @flags: pointer to state flags to receive QMAN_FQ_STATE_*** bitmask + * + * Queries the state of the FQ object, without performing any h/w commands. + * This captures the state, as seen by the driver, at the time the function + * executes. + */ +__rte_internal +void qman_fq_state(struct qman_fq *fq, enum qman_fq_state *state, u32 *flags); + +/** + * qman_init_fq - Initialises FQ fields, leaves the FQ "parked" or "scheduled" + * @fq: the frame queue object to modify, must be 'parked' or new. + * @flags: bit-mask of QMAN_INITFQ_FLAG_*** options + * @opts: the FQ-modification settings, as defined in the low-level API + * + * The @opts parameter comes from the low-level portal API. Select + * QMAN_INITFQ_FLAG_SCHED in @flags to cause the frame queue to be scheduled + * rather than parked. NB, @opts can be NULL. + * + * Note that some fields and options within @opts may be ignored or overwritten + * by the driver; + * 1. the 'count' and 'fqid' fields are always ignored (this operation only + * affects one frame queue: @fq). + * 2. the QM_INITFQ_WE_CONTEXTB option of the 'we_mask' field and the associated + * 'fqd' structure's 'context_b' field are sometimes overwritten; + * - if @fq was not created with QMAN_FQ_FLAG_TO_DCPORTAL, then context_b is + * initialised to a value used by the driver for demux. + * - if context_b is initialised for demux, so is context_a in case stashing + * is requested (see item 4). + * (So caller control of context_b is only possible for TO_DCPORTAL frame queue + * objects.) + * 3. if @flags contains QMAN_INITFQ_FLAG_LOCAL, the 'fqd' structure's + * 'dest::channel' field will be overwritten to match the portal used to issue + * the command. If the WE_DESTWQ write-enable bit had already been set by the + * caller, the channel workqueue will be left as-is, otherwise the write-enable + * bit is set and the workqueue is set to a default of 4. If the "LOCAL" flag + * isn't set, the destination channel/workqueue fields and the write-enable bit + * are left as-is. + * 4. if the driver overwrites context_a/b for demux, then if + * QM_INITFQ_WE_CONTEXTA is set, the driver will only overwrite + * context_a.address fields and will leave the stashing fields provided by the + * user alone, otherwise it will zero out the context_a.stashing fields. + */ +__rte_internal +int qman_init_fq(struct qman_fq *fq, u32 flags, struct qm_mcc_initfq *opts); + +/** + * qman_schedule_fq - Schedules a FQ + * @fq: the frame queue object to schedule, must be 'parked' + * + * Schedules the frame queue, which must be Parked, which takes it to + * Tentatively-Scheduled or Truly-Scheduled depending on its fill-level. + */ +int qman_schedule_fq(struct qman_fq *fq); + +/** + * qman_retire_fq - Retires a FQ + * @fq: the frame queue object to retire + * @flags: FQ flags (as per qman_fq_state) if retirement completes immediately + * + * Retires the frame queue. This returns zero if it succeeds immediately, +1 if + * the retirement was started asynchronously, otherwise it returns negative for + * failure. When this function returns zero, @flags is set to indicate whether + * the retired FQ is empty and/or whether it has any ORL fragments (to show up + * as ERNs). Otherwise the corresponding flags will be known when a subsequent + * FQRN message shows up on the portal's message ring. + * + * NB, if the retirement is asynchronous (the FQ was in the Truly Scheduled or + * Active state), the completion will be via the message ring as a FQRN - but + * the corresponding callback may occur before this function returns!! Ie. the + * caller should be prepared to accept the callback as the function is called, + * not only once it has returned. + */ +__rte_internal +int qman_retire_fq(struct qman_fq *fq, u32 *flags); + +/** + * qman_oos_fq - Puts a FQ "out of service" + * @fq: the frame queue object to be put out-of-service, must be 'retired' + * + * The frame queue must be retired and empty, and if any order restoration list + * was released as ERNs at the time of retirement, they must all be consumed. + */ +__rte_internal +int qman_oos_fq(struct qman_fq *fq); + +/** + * qman_fq_flow_control - Set the XON/XOFF state of a FQ + * @fq: the frame queue object to be set to XON/XOFF state, must not be 'oos', + * or 'retired' or 'parked' state + * @xon: boolean to set fq in XON or XOFF state + * + * The frame should be in Tentatively Scheduled state or Truly Schedule sate, + * otherwise the IFSI interrupt will be asserted. + */ +int qman_fq_flow_control(struct qman_fq *fq, int xon); + +/** + * qman_query_fq - Queries FQD fields (via h/w query command) + * @fq: the frame queue object to be queried + * @fqd: storage for the queried FQD fields + */ +int qman_query_fq(struct qman_fq *fq, struct qm_fqd *fqd); + +/** + * qman_query_fq_has_pkts - Queries non-programmable FQD fields and returns '1' + * if packets are in the frame queue. If there are no packets on frame + * queue '0' is returned. + * @fq: the frame queue object to be queried + */ +int qman_query_fq_has_pkts(struct qman_fq *fq); + +/** + * qman_query_fq_np - Queries non-programmable FQD fields + * @fq: the frame queue object to be queried + * @np: storage for the queried FQD fields + */ +__rte_internal +int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np); + +/** + * qman_query_fq_frmcnt - Queries fq frame count + * @fq: the frame queue object to be queried + * @frm_cnt: number of frames in the queue + */ +__rte_internal +int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt); + +/** + * qman_query_wq - Queries work queue lengths + * @query_dedicated: If non-zero, query length of WQs in the channel dedicated + * to this software portal. Otherwise, query length of WQs in a + * channel specified in wq. + * @wq: storage for the queried WQs lengths. Also specified the channel to + * to query if query_dedicated is zero. + */ +int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq); + +/** + * qman_volatile_dequeue - Issue a volatile dequeue command + * @fq: the frame queue object to dequeue from + * @flags: a bit-mask of QMAN_VOLATILE_FLAG_*** options + * @vdqcr: bit mask of QM_VDQCR_*** options, as per qm_dqrr_vdqcr_set() + * + * Attempts to lock access to the portal's VDQCR volatile dequeue functionality. + * The function will block and sleep if QMAN_VOLATILE_FLAG_WAIT is specified and + * the VDQCR is already in use, otherwise returns non-zero for failure. If + * QMAN_VOLATILE_FLAG_FINISH is specified, the function will only return once + * the VDQCR command has finished executing (ie. once the callback for the last + * DQRR entry resulting from the VDQCR command has been called). If not using + * the FINISH flag, completion can be determined either by detecting the + * presence of the QM_DQRR_STAT_UNSCHEDULED and QM_DQRR_STAT_DQCR_EXPIRED bits + * in the "stat" field of the "struct qm_dqrr_entry" passed to the FQ's dequeue + * callback, or by waiting for the QMAN_FQ_STATE_VDQCR bit to disappear from the + * "flags" retrieved from qman_fq_state(). + */ +__rte_internal +int qman_volatile_dequeue(struct qman_fq *fq, u32 flags, u32 vdqcr); + +/** + * qman_enqueue - Enqueue a frame to a frame queue + * @fq: the frame queue object to enqueue to + * @fd: a descriptor of the frame to be enqueued + * @flags: bit-mask of QMAN_ENQUEUE_FLAG_*** options + * + * Fills an entry in the EQCR of portal @qm to enqueue the frame described by + * @fd. The descriptor details are copied from @fd to the EQCR entry, the 'pid' + * field is ignored. The return value is non-zero on error, such as ring full + * (and FLAG_WAIT not specified), congestion avoidance (FLAG_WATCH_CGR + * specified), etc. If the ring is full and FLAG_WAIT is specified, this + * function will block. If FLAG_INTERRUPT is set, the EQCI bit of the portal + * interrupt will assert when Qman consumes the EQCR entry (subject to "status + * disable", "enable", and "inhibit" registers). If FLAG_DCA is set, Qman will + * perform an implied "discrete consumption acknowledgment" on the dequeue + * ring's (DQRR) entry, at the ring index specified by the FLAG_DCA_IDX(x) + * macro. (As an alternative to issuing explicit DCA actions on DQRR entries, + * this implicit DCA can delay the release of a "held active" frame queue + * corresponding to a DQRR entry until Qman consumes the EQCR entry - providing + * order-preservation semantics in packet-forwarding scenarios.) If FLAG_DCA is + * set, then FLAG_DCA_PARK can also be set to imply that the DQRR consumption + * acknowledgment should "park request" the "held active" frame queue. Ie. + * when the portal eventually releases that frame queue, it will be left in the + * Parked state rather than Tentatively Scheduled or Truly Scheduled. If the + * portal is watching congestion groups, the QMAN_ENQUEUE_FLAG_WATCH_CGR flag + * is requested, and the FQ is a member of a congestion group, then this + * function returns -EAGAIN if the congestion group is currently congested. + * Note, this does not eliminate ERNs, as the async interface means we can be + * sending enqueue commands to an un-congested FQ that becomes congested before + * the enqueue commands are processed, but it does minimise needless thrashing + * of an already busy hardware resource by throttling many of the to-be-dropped + * enqueues "at the source". + */ +__rte_internal +int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags); + +__rte_internal +int qman_enqueue_multi(struct qman_fq *fq, const struct qm_fd *fd, u32 *flags, + int frames_to_send); + +/** + * qman_enqueue_multi_fq - Enqueue multiple frames to their respective frame + * queues. + * @fq[]: Array of frame queue objects to enqueue to + * @fd: pointer to first descriptor of frame to be enqueued + * @frames_to_send: number of frames to be sent. + * + * This API is similar to qman_enqueue_multi(), but it takes fd which needs + * to be processed by different frame queues. + */ +__rte_internal +int +qman_enqueue_multi_fq(struct qman_fq *fq[], const struct qm_fd *fd, + u32 *flags, int frames_to_send); + +typedef int (*qman_cb_precommit) (void *arg); + +/** + * qman_enqueue_orp - Enqueue a frame to a frame queue using an ORP + * @fq: the frame queue object to enqueue to + * @fd: a descriptor of the frame to be enqueued + * @flags: bit-mask of QMAN_ENQUEUE_FLAG_*** options + * @orp: the frame queue object used as an order restoration point. + * @orp_seqnum: the sequence number of this frame in the order restoration path + * + * Similar to qman_enqueue(), but with the addition of an Order Restoration + * Point (@orp) and corresponding sequence number (@orp_seqnum) for this + * enqueue operation to employ order restoration. Each frame queue object acts + * as an Order Definition Point (ODP) by providing each frame dequeued from it + * with an incrementing sequence number, this value is generally ignored unless + * that sequence of dequeued frames will need order restoration later. Each + * frame queue object also encapsulates an Order Restoration Point (ORP), which + * is a re-assembly context for re-ordering frames relative to their sequence + * numbers as they are enqueued. The ORP does not have to be within the frame + * queue that receives the enqueued frame, in fact it is usually the frame + * queue from which the frames were originally dequeued. For the purposes of + * order restoration, multiple frames (or "fragments") can be enqueued for a + * single sequence number by setting the QMAN_ENQUEUE_FLAG_NLIS flag for all + * enqueues except the final fragment of a given sequence number. Ordering + * between sequence numbers is guaranteed, even if fragments of different + * sequence numbers are interlaced with one another. Fragments of the same + * sequence number will retain the order in which they are enqueued. If no + * enqueue is to performed, QMAN_ENQUEUE_FLAG_HOLE indicates that the given + * sequence number is to be "skipped" by the ORP logic (eg. if a frame has been + * dropped from a sequence), or QMAN_ENQUEUE_FLAG_NESN indicates that the given + * sequence number should become the ORP's "Next Expected Sequence Number". + * + * Side note: a frame queue object can be used purely as an ORP, without + * carrying any frames at all. Care should be taken not to deallocate a frame + * queue object that is being actively used as an ORP, as a future allocation + * of the frame queue object may start using the internal ORP before the + * previous use has finished. + */ +int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum); + +/** + * qman_alloc_fqid_range - Allocate a contiguous range of FQIDs + * @result: is set by the API to the base FQID of the allocated range + * @count: the number of FQIDs required + * @align: required alignment of the allocated range + * @partial: non-zero if the API can return fewer than @count FQIDs + * + * Returns the number of frame queues allocated, or a negative error code. If + * @partial is non zero, the allocation request may return a smaller range of + * FQs than requested (though alignment will be as requested). If @partial is + * zero, the return value will either be 'count' or negative. + */ +int qman_alloc_fqid_range(u32 *result, u32 count, u32 align, int partial); +static inline int qman_alloc_fqid(u32 *result) +{ + int ret = qman_alloc_fqid_range(result, 1, 0, 0); + + return (ret > 0) ? 0 : ret; +} + +/** + * qman_release_fqid_range - Release the specified range of frame queue IDs + * @fqid: the base FQID of the range to deallocate + * @count: the number of FQIDs in the range + * + * This function can also be used to seed the allocator with ranges of FQIDs + * that it can subsequently allocate from. + */ +void qman_release_fqid_range(u32 fqid, unsigned int count); +static inline void qman_release_fqid(u32 fqid) +{ + qman_release_fqid_range(fqid, 1); +} + +void qman_seed_fqid_range(u32 fqid, unsigned int count); + +int qman_shutdown_fq(u32 fqid); + +/** + * qman_reserve_fqid_range - Reserve the specified range of frame queue IDs + * @fqid: the base FQID of the range to deallocate + * @count: the number of FQIDs in the range + */ +__rte_internal +int qman_reserve_fqid_range(u32 fqid, unsigned int count); +static inline int qman_reserve_fqid(u32 fqid) +{ + return qman_reserve_fqid_range(fqid, 1); +} + +/* Pool-channel management */ +/** + * qman_alloc_pool_range - Allocate a contiguous range of pool-channel IDs + * @result: is set by the API to the base pool-channel ID of the allocated range + * @count: the number of pool-channel IDs required + * @align: required alignment of the allocated range + * @partial: non-zero if the API can return fewer than @count + * + * Returns the number of pool-channel IDs allocated, or a negative error code. + * If @partial is non zero, the allocation request may return a smaller range of + * than requested (though alignment will be as requested). If @partial is zero, + * the return value will either be 'count' or negative. + */ +__rte_internal +int qman_alloc_pool_range(u32 *result, u32 count, u32 align, int partial); +static inline int qman_alloc_pool(u32 *result) +{ + int ret = qman_alloc_pool_range(result, 1, 0, 0); + + return (ret > 0) ? 0 : ret; +} + +/** + * qman_release_pool_range - Release the specified range of pool-channel IDs + * @id: the base pool-channel ID of the range to deallocate + * @count: the number of pool-channel IDs in the range + */ +void qman_release_pool_range(u32 id, unsigned int count); +static inline void qman_release_pool(u32 id) +{ + qman_release_pool_range(id, 1); +} + +/** + * qman_reserve_pool_range - Reserve the specified range of pool-channel IDs + * @id: the base pool-channel ID of the range to reserve + * @count: the number of pool-channel IDs in the range + */ +int qman_reserve_pool_range(u32 id, unsigned int count); +static inline int qman_reserve_pool(u32 id) +{ + return qman_reserve_pool_range(id, 1); +} + +void qman_seed_pool_range(u32 id, unsigned int count); + + /* CGR management */ + /* -------------- */ +/** + * qman_create_cgr - Register a congestion group object + * @cgr: the 'cgr' object, with fields filled in + * @flags: QMAN_CGR_FLAG_* values + * @opts: optional state of CGR settings + * + * Registers this object to receiving congestion entry/exit callbacks on the + * portal affine to the cpu portal on which this API is executed. If opts is + * NULL then only the callback (cgr->cb) function is registered. If @flags + * contains QMAN_CGR_FLAG_USE_INIT, then an init hw command (which will reset + * any unspecified parameters) will be used rather than a modify hw hardware + * (which only modifies the specified parameters). + */ +__rte_internal +int qman_create_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts); + +/** + * qman_create_cgr_to_dcp - Register a congestion group object to DCP portal + * @cgr: the 'cgr' object, with fields filled in + * @flags: QMAN_CGR_FLAG_* values + * @dcp_portal: the DCP portal to which the cgr object is registered. + * @opts: optional state of CGR settings + * + */ +int qman_create_cgr_to_dcp(struct qman_cgr *cgr, u32 flags, u16 dcp_portal, + struct qm_mcc_initcgr *opts); + +/** + * qman_delete_cgr - Deregisters a congestion group object + * @cgr: the 'cgr' object to deregister + * + * "Unplugs" this CGR object from the portal affine to the cpu on which this API + * is executed. This must be excuted on the same affine portal on which it was + * created. + */ +__rte_internal +int qman_delete_cgr(struct qman_cgr *cgr); + +/** + * qman_modify_cgr - Modify CGR fields + * @cgr: the 'cgr' object to modify + * @flags: QMAN_CGR_FLAG_* values + * @opts: the CGR-modification settings + * + * The @opts parameter comes from the low-level portal API, and can be NULL. + * Note that some fields and options within @opts may be ignored or overwritten + * by the driver, in particular the 'cgrid' field is ignored (this operation + * only affects the given CGR object). If @flags contains + * QMAN_CGR_FLAG_USE_INIT, then an init hw command (which will reset any + * unspecified parameters) will be used rather than a modify hw hardware (which + * only modifies the specified parameters). + */ +__rte_internal +int qman_modify_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts); + +/** + * qman_query_cgr - Queries CGR fields + * @cgr: the 'cgr' object to query + * @result: storage for the queried congestion group record + */ +int qman_query_cgr(struct qman_cgr *cgr, struct qm_mcr_querycgr *result); + +/** + * qman_query_congestion - Queries the state of all congestion groups + * @congestion: storage for the queried state of all congestion groups + */ +int qman_query_congestion(struct qm_mcr_querycongestion *congestion); + +/** + * qman_alloc_cgrid_range - Allocate a contiguous range of CGR IDs + * @result: is set by the API to the base CGR ID of the allocated range + * @count: the number of CGR IDs required + * @align: required alignment of the allocated range + * @partial: non-zero if the API can return fewer than @count + * + * Returns the number of CGR IDs allocated, or a negative error code. + * If @partial is non zero, the allocation request may return a smaller range of + * than requested (though alignment will be as requested). If @partial is zero, + * the return value will either be 'count' or negative. + */ +__rte_internal +int qman_alloc_cgrid_range(u32 *result, u32 count, u32 align, int partial); +static inline int qman_alloc_cgrid(u32 *result) +{ + int ret = qman_alloc_cgrid_range(result, 1, 0, 0); + + return (ret > 0) ? 0 : ret; +} + +/** + * qman_release_cgrid_range - Release the specified range of CGR IDs + * @id: the base CGR ID of the range to deallocate + * @count: the number of CGR IDs in the range + */ +__rte_internal +void qman_release_cgrid_range(u32 id, unsigned int count); +static inline void qman_release_cgrid(u32 id) +{ + qman_release_cgrid_range(id, 1); +} + +/** + * qman_reserve_cgrid_range - Reserve the specified range of CGR ID + * @id: the base CGR ID of the range to reserve + * @count: the number of CGR IDs in the range + */ +int qman_reserve_cgrid_range(u32 id, unsigned int count); +static inline int qman_reserve_cgrid(u32 id) +{ + return qman_reserve_cgrid_range(id, 1); +} + +void qman_seed_cgrid_range(u32 id, unsigned int count); + + /* Helpers */ + /* ------- */ +/** + * qman_poll_fq_for_init - Check if an FQ has been initialised from OOS + * @fqid: the FQID that will be initialised by other s/w + * + * In many situations, a FQID is provided for communication between s/w + * entities, and whilst the consumer is responsible for initialising and + * scheduling the FQ, the producer(s) generally create a wrapper FQ object using + * and only call qman_enqueue() (no FQ initialisation, scheduling, etc). Ie; + * qman_create_fq(..., QMAN_FQ_FLAG_NO_MODIFY, ...); + * However, data can not be enqueued to the FQ until it is initialised out of + * the OOS state - this function polls for that condition. It is particularly + * useful for users of IPC functions - each endpoint's Rx FQ is the other + * endpoint's Tx FQ, so each side can initialise and schedule their Rx FQ object + * and then use this API on the (NO_MODIFY) Tx FQ object in order to + * synchronise. The function returns zero for success, +1 if the FQ is still in + * the OOS state, or negative if there was an error. + */ +static inline int qman_poll_fq_for_init(struct qman_fq *fq) +{ + struct qm_mcr_queryfq_np np; + int err; + + err = qman_query_fq_np(fq, &np); + if (err) + return err; + if ((np.state & QM_MCR_NP_STATE_MASK) == QM_MCR_NP_STATE_OOS) + return 1; + return 0; +} + +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ +#define cpu_to_hw_sg(x) +#define hw_sg_to_cpu(x) +#else +#define cpu_to_hw_sg(x) __cpu_to_hw_sg(x) +#define hw_sg_to_cpu(x) __hw_sg_to_cpu(x) + +static inline void __cpu_to_hw_sg(struct qm_sg_entry *sgentry) +{ + sgentry->opaque = cpu_to_be64(sgentry->opaque); + sgentry->val = cpu_to_be32(sgentry->val); + sgentry->val_off = cpu_to_be16(sgentry->val_off); +} + +static inline void __hw_sg_to_cpu(struct qm_sg_entry *sgentry) +{ + sgentry->opaque = be64_to_cpu(sgentry->opaque); + sgentry->val = be32_to_cpu(sgentry->val); + sgentry->val_off = be16_to_cpu(sgentry->val_off); +} +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* __FSL_QMAN_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_usd.h b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_usd.h new file mode 100644 index 000000000..dcf35e4ad --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/fsl_usd.h @@ -0,0 +1,94 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2010-2011 Freescale Semiconductor, Inc. + * All rights reserved. + * Copyright 2019 NXP + * + */ + +#ifndef __FSL_USD_H +#define __FSL_USD_H + +#include <compat.h> +#include <dpaa_list.h> +#include <fsl_qman.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* Thread-entry/exit hooks; */ +int qman_thread_init(void); +int bman_thread_init(void); +int qman_thread_finish(void); +int bman_thread_finish(void); + +#define QBMAN_ANY_PORTAL_IDX 0xffffffff + +/* Obtain and free raw (unitialized) portals */ + +struct dpaa_raw_portal { + /* inputs */ + + /* set to non zero to turn on stashing */ + uint8_t enable_stash; + /* Stashing attributes for the portal */ + uint32_t cpu; + uint32_t cache; + uint32_t window; + + /* Specifies the stash request queue this portal should use */ + uint8_t sdest; + + /* Specifes a specific portal index to map or QBMAN_ANY_PORTAL_IDX + * for don't care. The portal index will be populated by the + * driver when the ioctl() successfully completes. + */ + uint32_t index; + + /* outputs */ + uint64_t cinh; + uint64_t cena; +}; + +int qman_allocate_raw_portal(struct dpaa_raw_portal *portal); +int qman_free_raw_portal(struct dpaa_raw_portal *portal); + +int bman_allocate_raw_portal(struct dpaa_raw_portal *portal); +int bman_free_raw_portal(struct dpaa_raw_portal *portal); + +/* Obtain thread-local UIO file-descriptors */ +__rte_internal +int qman_thread_fd(void); +int bman_thread_fd(void); + +/* Post-process interrupts. NB, the kernel IRQ handler disables the interrupt + * line before notifying us, and this post-processing re-enables it once + * processing is complete. As such, it is essential to call this before going + * into another blocking read/select/poll. + */ +__rte_internal +void qman_thread_irq(void); + +__rte_internal +void bman_thread_irq(void); +__rte_internal +void qman_fq_portal_thread_irq(struct qman_portal *qp); +__rte_internal +void qman_clear_irq(void); + +/* Global setup */ +int qman_global_init(void); +int bman_global_init(void); + +/* Direct portal create and destroy */ +__rte_internal +struct qman_portal *fsl_qman_fq_portal_create(int *fd); +int fsl_qman_fq_portal_destroy(struct qman_portal *qp); +int fsl_qman_fq_portal_init(struct qman_portal *qp); + +#ifdef __cplusplus +} +#endif + +#endif /* __FSL_USD_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/netcfg.h b/src/spdk/dpdk/drivers/bus/dpaa/include/netcfg.h new file mode 100644 index 000000000..d7d1befd2 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/netcfg.h @@ -0,0 +1,65 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2010-2012 Freescale Semiconductor, Inc. + * All rights reserved. + * + */ + +#ifndef __NETCFG_H +#define __NETCFG_H + +#include <fman.h> +#include <argp.h> + +/* Configuration information related to a specific ethernet port */ +struct fm_eth_port_cfg { + /**< A list of PCD FQ ranges, obtained from FMC configuration */ + struct list_head *list; + /**< The "Rx default" FQID, obtained from FMC configuration */ + uint32_t rx_def; + /**< Other interface details are in the fman driver interface */ + struct fman_if *fman_if; +}; + +struct netcfg_info { + uint8_t num_ethports; + /**< Number of ports */ + struct fm_eth_port_cfg port_cfg[0]; + /**< Variable structure array of size num_ethports */ +}; + +struct interface_info { + char *name; + struct rte_ether_addr mac_addr; + struct rte_ether_addr peer_mac; + int mac_present; + int fman_enabled_mac_interface; +}; + +struct netcfg_interface { + uint8_t numof_netcfg_interface; + uint8_t numof_fman_enabled_macless; + struct interface_info interface_info[0]; +}; + +/* pcd_file: FMC netpcd XML ("policy") file, that contains PCD information. + * cfg_file: FMC config XML file + * Returns the configuration information in newly allocated memory. + */ +__rte_internal +struct netcfg_info *netcfg_acquire(void); + +/* cfg_ptr: configuration information pointer. + * Frees the resources allocated by the configuration layer. + */ +__rte_internal +void netcfg_release(struct netcfg_info *cfg_ptr); + +#ifdef RTE_LIBRTE_DPAA_DEBUG_DRIVER +/* cfg_ptr: configuration information pointer. + * This function dumps configuration data to stdout. + */ +void dump_netcfg(struct netcfg_info *cfg_ptr); +#endif + +#endif /* __NETCFG_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/include/process.h b/src/spdk/dpdk/drivers/bus/dpaa/include/process.h new file mode 100644 index 000000000..d9ec94ee2 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/include/process.h @@ -0,0 +1,77 @@ +/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0) + * + * Copyright 2010-2011 Freescale Semiconductor, Inc. + * All rights reserved. + * + */ + +#ifndef __PROCESS_H +#define __PROCESS_H + +#include <compat.h> + +/* The process device underlies process-wide user/kernel interactions, such as + * mapping dma_mem memory and providing accompanying ioctl()s. (This isn't used + * for portals, which use one UIO device each.). + */ +#define PROCESS_PATH "/dev/fsl-usdpaa" + +/* Allocation of resource IDs uses a generic interface. This enum is used to + * distinguish between the type of underlying object being manipulated. + */ +enum dpaa_id_type { + dpaa_id_fqid, + dpaa_id_bpid, + dpaa_id_qpool, + dpaa_id_cgrid, + dpaa_id_max /* <-- not a valid type, represents the number of types */ +}; + +int process_alloc(enum dpaa_id_type id_type, uint32_t *base, uint32_t num, + uint32_t align, int partial); +void process_release(enum dpaa_id_type id_type, uint32_t base, uint32_t num); + +int process_reserve(enum dpaa_id_type id_type, uint32_t base, uint32_t num); + +/* Mapping and using QMan/BMan portals */ +enum dpaa_portal_type { + dpaa_portal_qman, + dpaa_portal_bman, +}; + +struct dpaa_portal_map { + void *cinh; + void *cena; +}; + +struct dpaa_ioctl_portal_map { + /* Input parameter, is a qman or bman portal required. */ + enum dpaa_portal_type type; + /* Specifes a specific portal index to map or 0xffffffff + * for don't care. + */ + uint32_t index; + + /* Return value if the map succeeds, this gives the mapped + * cache-inhibited (cinh) and cache-enabled (cena) addresses. + */ + struct dpaa_portal_map addr; + + /* Qman-specific return values */ + u16 channel; + uint32_t pools; +}; + +int process_portal_map(struct dpaa_ioctl_portal_map *params); +int process_portal_unmap(struct dpaa_portal_map *map); + +struct dpaa_ioctl_irq_map { + enum dpaa_portal_type type; /* Type of portal to map */ + int fd; /* File descriptor that contains the portal */ + void *portal_cinh; /* Cache inhibited area to identify the portal */ +}; + +int process_portal_irq_map(int fd, struct dpaa_ioctl_irq_map *irq); +int process_portal_irq_unmap(int fd); + +#endif /* __PROCESS_H */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/meson.build b/src/spdk/dpdk/drivers/bus/dpaa/meson.build new file mode 100644 index 000000000..c2cffa494 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/meson.build @@ -0,0 +1,29 @@ +# SPDX-License-Identifier: BSD-3-Clause +# Copyright 2018 NXP + +if not is_linux + build = false + reason = 'only supported on linux' +endif + +deps += ['common_dpaax', 'eventdev'] +sources = files('base/fman/fman.c', + 'base/fman/fman_hw.c', + 'base/fman/netcfg_layer.c', + 'base/qbman/bman.c', + 'base/qbman/bman_driver.c', + 'base/qbman/dpaa_alloc.c', + 'base/qbman/dpaa_sys.c', + 'base/qbman/process.c', + 'base/qbman/qman.c', + 'base/qbman/qman_driver.c', + 'dpaa_bus.c') + +if cc.has_argument('-Wno-cast-qual') + cflags += '-Wno-cast-qual' +endif +if cc.has_argument('-Wno-pointer-arith') + cflags += '-Wno-pointer-arith' +endif + +includes += include_directories('include', 'base/qbman') diff --git a/src/spdk/dpdk/drivers/bus/dpaa/rte_bus_dpaa_version.map b/src/spdk/dpdk/drivers/bus/dpaa/rte_bus_dpaa_version.map new file mode 100644 index 000000000..46d42f7d6 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/rte_bus_dpaa_version.map @@ -0,0 +1,93 @@ +DPDK_20.0 { + local: *; +}; + +INTERNAL { + global: + + bman_acquire; + bman_free_pool; + bman_get_params; + bman_new_pool; + bman_query_free_buffers; + bman_release; + bman_thread_irq; + dpaa_get_eth_port_cfg; + dpaa_get_qm_channel_caam; + dpaa_get_qm_channel_pool; + dpaa_svr_family; + fman_dealloc_bufs_mask_hi; + fman_dealloc_bufs_mask_lo; + fman_if_add_mac_addr; + fman_if_clear_mac_addr; + fman_if_disable_rx; + fman_if_discard_rx_errors; + fman_if_enable_rx; + fman_if_get_fc_quanta; + fman_if_get_fc_threshold; + fman_if_get_fdoff; + fman_if_get_sg_enable; + fman_if_loopback_disable; + fman_if_loopback_enable; + fman_if_promiscuous_disable; + fman_if_promiscuous_enable; + fman_if_reset_mcast_filter_table; + fman_if_set_bp; + fman_if_set_fc_quanta; + fman_if_set_fc_threshold; + fman_if_set_fdoff; + fman_if_set_ic_params; + fman_if_set_maxfrm; + fman_if_set_mcast_filter_table; + fman_if_set_sg; + fman_if_stats_get; + fman_if_stats_get_all; + fman_if_stats_reset; + fman_ip_rev; + fsl_qman_fq_portal_create; + netcfg_acquire; + netcfg_release; + per_lcore_dpaa_io; + per_lcore_held_bufs; + qman_alloc_cgrid_range; + qman_alloc_pool_range; + qman_clear_irq; + qman_create_cgr; + qman_create_fq; + qman_dca_index; + qman_delete_cgr; + qman_dequeue; + qman_dqrr_consume; + qman_enqueue; + qman_enqueue_multi; + qman_enqueue_multi_fq; + qman_fq_fqid; + qman_fq_portal_irqsource_add; + qman_fq_portal_irqsource_remove; + qman_fq_portal_thread_irq; + qman_fq_state; + qman_init_fq; + qman_irqsource_add; + qman_irqsource_remove; + qman_modify_cgr; + qman_oos_fq; + qman_poll_dqrr; + qman_portal_dequeue; + qman_portal_poll_rx; + qman_query_fq_frm_cnt; + qman_query_fq_np; + qman_release_cgrid_range; + qman_reserve_fqid_range; + qman_retire_fq; + qman_set_fq_lookup_table; + qman_set_vdq; + qman_static_dequeue_add; + qman_thread_fd; + qman_thread_irq; + qman_volatile_dequeue; + rte_dpaa_driver_register; + rte_dpaa_driver_unregister; + rte_dpaa_portal_fq_close; + rte_dpaa_portal_fq_init; + rte_dpaa_portal_init; +}; diff --git a/src/spdk/dpdk/drivers/bus/dpaa/rte_dpaa_bus.h b/src/spdk/dpdk/drivers/bus/dpaa/rte_dpaa_bus.h new file mode 100644 index 000000000..2a186d83f --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/rte_dpaa_bus.h @@ -0,0 +1,224 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * + * Copyright 2017-2019 NXP + * + */ +#ifndef __RTE_DPAA_BUS_H__ +#define __RTE_DPAA_BUS_H__ + +#include <rte_bus.h> +#include <rte_mempool.h> +#include <dpaax_iova_table.h> + +#include <dpaa_of.h> +#include <fsl_usd.h> +#include <fsl_qman.h> +#include <fsl_bman.h> +#include <netcfg.h> + +#define DPAA_MEMPOOL_OPS_NAME "dpaa" + +#define DEV_TO_DPAA_DEVICE(ptr) \ + container_of(ptr, struct rte_dpaa_device, device) + +/* DPAA SoC identifier; If this is not available, it can be concluded + * that board is non-DPAA. Single slot is currently supported. + */ +#define DPAA_SOC_ID_FILE "/sys/devices/soc0/soc_id" + +#define SVR_LS1043A_FAMILY 0x87920000 +#define SVR_LS1046A_FAMILY 0x87070000 +#define SVR_MASK 0xffff0000 + +#define RTE_DEV_TO_DPAA_CONST(ptr) \ + container_of(ptr, const struct rte_dpaa_device, device) + +extern unsigned int dpaa_svr_family; + +extern RTE_DEFINE_PER_LCORE(bool, dpaa_io); + +struct rte_dpaa_device; +struct rte_dpaa_driver; + +/* DPAA Device and Driver lists for DPAA bus */ +TAILQ_HEAD(rte_dpaa_device_list, rte_dpaa_device); +TAILQ_HEAD(rte_dpaa_driver_list, rte_dpaa_driver); + +enum rte_dpaa_type { + FSL_DPAA_ETH = 1, + FSL_DPAA_CRYPTO, +}; + +struct rte_dpaa_bus { + struct rte_bus bus; + struct rte_dpaa_device_list device_list; + struct rte_dpaa_driver_list driver_list; + int device_count; + int detected; +}; + +struct dpaa_device_id { + uint8_t fman_id; /**< Fman interface ID, for ETH type device */ + uint8_t mac_id; /**< Fman MAC interface ID, for ETH type device */ + uint16_t dev_id; /**< Device Identifier from DPDK */ +}; + +struct rte_dpaa_device { + TAILQ_ENTRY(rte_dpaa_device) next; + struct rte_device device; + union { + struct rte_eth_dev *eth_dev; + struct rte_cryptodev *crypto_dev; + }; + struct rte_dpaa_driver *driver; + struct dpaa_device_id id; + struct rte_intr_handle intr_handle; + enum rte_dpaa_type device_type; /**< Ethernet or crypto type device */ + char name[RTE_ETH_NAME_MAX_LEN]; +}; + +typedef int (*rte_dpaa_probe_t)(struct rte_dpaa_driver *dpaa_drv, + struct rte_dpaa_device *dpaa_dev); +typedef int (*rte_dpaa_remove_t)(struct rte_dpaa_device *dpaa_dev); + +struct rte_dpaa_driver { + TAILQ_ENTRY(rte_dpaa_driver) next; + struct rte_driver driver; + struct rte_dpaa_bus *dpaa_bus; + enum rte_dpaa_type drv_type; + rte_dpaa_probe_t probe; + rte_dpaa_remove_t remove; +}; + +struct dpaa_portal { + uint32_t bman_idx; /**< BMAN Portal ID*/ + uint32_t qman_idx; /**< QMAN Portal ID*/ + uint64_t tid;/**< Parent Thread id for this portal */ +}; + +/* Various structures representing contiguous memory maps */ +struct dpaa_memseg { + TAILQ_ENTRY(dpaa_memseg) next; + char *vaddr; + rte_iova_t iova; + size_t len; +}; + +TAILQ_HEAD(dpaa_memseg_list, dpaa_memseg); +extern struct dpaa_memseg_list rte_dpaa_memsegs; + +/* Either iterate over the list of internal memseg references or fallback to + * EAL memseg based iova2virt. + */ +static inline void *rte_dpaa_mem_ptov(phys_addr_t paddr) +{ + struct dpaa_memseg *ms; + void *va; + + va = dpaax_iova_table_get_va(paddr); + if (likely(va != NULL)) + return va; + + /* Check if the address is already part of the memseg list internally + * maintained by the dpaa driver. + */ + TAILQ_FOREACH(ms, &rte_dpaa_memsegs, next) { + if (paddr >= ms->iova && paddr < + ms->iova + ms->len) + return RTE_PTR_ADD(ms->vaddr, (uintptr_t)(paddr - ms->iova)); + } + + /* If not, Fallback to full memseg list searching */ + va = rte_mem_iova2virt(paddr); + + dpaax_iova_table_update(paddr, va, RTE_CACHE_LINE_SIZE); + + return va; +} + +static inline rte_iova_t +rte_dpaa_mem_vtop(void *vaddr) +{ + const struct rte_memseg *ms; + + ms = rte_mem_virt2memseg(vaddr, NULL); + if (ms) + return ms->iova + RTE_PTR_DIFF(vaddr, ms->addr); + + return (size_t)NULL; +} + +/** + * Register a DPAA driver. + * + * @param driver + * A pointer to a rte_dpaa_driver structure describing the driver + * to be registered. + */ +__rte_internal +void rte_dpaa_driver_register(struct rte_dpaa_driver *driver); + +/** + * Unregister a DPAA driver. + * + * @param driver + * A pointer to a rte_dpaa_driver structure describing the driver + * to be unregistered. + */ +__rte_internal +void rte_dpaa_driver_unregister(struct rte_dpaa_driver *driver); + +/** + * Initialize a DPAA portal + * + * @param arg + * Per thread ID + * + * @return + * 0 in case of success, error otherwise + */ +__rte_internal +int rte_dpaa_portal_init(void *arg); + +__rte_internal +int rte_dpaa_portal_fq_init(void *arg, struct qman_fq *fq); + +__rte_internal +int rte_dpaa_portal_fq_close(struct qman_fq *fq); + +/** + * Cleanup a DPAA Portal + */ +void dpaa_portal_finish(void *arg); + +/** Helper for DPAA device registration from driver (eth, crypto) instance */ +#define RTE_PMD_REGISTER_DPAA(nm, dpaa_drv) \ +RTE_INIT(dpaainitfn_ ##nm) \ +{\ + (dpaa_drv).driver.name = RTE_STR(nm);\ + rte_dpaa_driver_register(&dpaa_drv); \ +} \ +RTE_PMD_EXPORT_NAME(nm, __COUNTER__) + +/* Create storage for dqrr entries per lcore */ +#define DPAA_PORTAL_DEQUEUE_DEPTH 16 +struct dpaa_portal_dqrr { + void *mbuf[DPAA_PORTAL_DEQUEUE_DEPTH]; + uint64_t dqrr_held; + uint8_t dqrr_size; +}; + +RTE_DECLARE_PER_LCORE(struct dpaa_portal_dqrr, held_bufs); + +#define DPAA_PER_LCORE_DQRR_SIZE RTE_PER_LCORE(held_bufs).dqrr_size +#define DPAA_PER_LCORE_DQRR_HELD RTE_PER_LCORE(held_bufs).dqrr_held +#define DPAA_PER_LCORE_DQRR_MBUF(i) RTE_PER_LCORE(held_bufs).mbuf[i] + +__rte_internal +struct fm_eth_port_cfg *dpaa_get_eth_port_cfg(int dev_id); + +#ifdef __cplusplus +} +#endif + +#endif /* __RTE_DPAA_BUS_H__ */ diff --git a/src/spdk/dpdk/drivers/bus/dpaa/rte_dpaa_logs.h b/src/spdk/dpdk/drivers/bus/dpaa/rte_dpaa_logs.h new file mode 100644 index 000000000..149b12743 --- /dev/null +++ b/src/spdk/dpdk/drivers/bus/dpaa/rte_dpaa_logs.h @@ -0,0 +1,40 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * + * Copyright 2017 NXP + * + */ + +#ifndef _DPAA_LOGS_H_ +#define _DPAA_LOGS_H_ + +#include <rte_log.h> + +extern int dpaa_logtype_bus; + +#define DPAA_BUS_LOG(level, fmt, args...) \ + rte_log(RTE_LOG_ ## level, dpaa_logtype_bus, "dpaa: " fmt "\n", ##args) + +#ifdef RTE_LIBRTE_DPAA_DEBUG_BUS +#define DPAA_BUS_HWWARN(cond, fmt, args...) \ + do {\ + if (cond) \ + DPAA_BUS_LOG(DEBUG, "WARN: " fmt, ##args); \ + } while (0) +#else +#define DPAA_BUS_HWWARN(cond, fmt, args...) do { } while (0) +#endif + +#define DPAA_BUS_DEBUG(fmt, args...) \ + rte_log(RTE_LOG_DEBUG, dpaa_logtype_bus, "dpaa: %s(): " fmt "\n", \ + __func__, ##args) + +#define BUS_INIT_FUNC_TRACE() DPAA_BUS_DEBUG(" >>") + +#define DPAA_BUS_INFO(fmt, args...) \ + DPAA_BUS_LOG(INFO, fmt, ## args) +#define DPAA_BUS_ERR(fmt, args...) \ + DPAA_BUS_LOG(ERR, fmt, ## args) +#define DPAA_BUS_WARN(fmt, args...) \ + DPAA_BUS_LOG(WARNING, fmt, ## args) + +#endif /* _DPAA_LOGS_H_ */ |