diff options
Diffstat (limited to 'src/seastar/dpdk/drivers/net/xenvirt')
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/Makefile | 57 | ||||
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt.c | 767 | ||||
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt.h | 61 | ||||
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt_version.map | 7 | ||||
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/rte_mempool_gntalloc.c | 295 | ||||
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/rte_xen_lib.c | 454 | ||||
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/rte_xen_lib.h | 116 | ||||
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/virtio_logs.h | 70 | ||||
| -rw-r--r-- | src/seastar/dpdk/drivers/net/xenvirt/virtqueue.h | 273 |
9 files changed, 2100 insertions, 0 deletions
diff --git a/src/seastar/dpdk/drivers/net/xenvirt/Makefile b/src/seastar/dpdk/drivers/net/xenvirt/Makefile new file mode 100644 index 00000000..8b4b8f03 --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/Makefile @@ -0,0 +1,57 @@ +# BSD LICENSE +# +# Copyright(c) 2010-2014 Intel Corporation. All rights reserved. +# All rights reserved. +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions +# are met: +# +# * Redistributions of source code must retain the above copyright +# notice, this list of conditions and the following disclaimer. +# * Redistributions in binary form must reproduce the above copyright +# notice, this list of conditions and the following disclaimer in +# the documentation and/or other materials provided with the +# distribution. +# * Neither the name of Intel Corporation nor the names of its +# contributors may be used to endorse or promote products derived +# from this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +include $(RTE_SDK)/mk/rte.vars.mk + +# +# library name +# +LIB = librte_pmd_xenvirt.a + +CFLAGS += -O3 +CFLAGS += $(WERROR_FLAGS) +LDLIBS += -lxenstore + +EXPORT_MAP := rte_eth_xenvirt_version.map + +LIBABIVER := 1 + +# +# all source are stored in SRCS-y +# +SRCS-$(CONFIG_RTE_LIBRTE_PMD_XENVIRT) += rte_eth_xenvirt.c rte_mempool_gntalloc.c rte_xen_lib.c + +# +# Export include files +# +SYMLINK-y-include += rte_eth_xenvirt.h + +include $(RTE_SDK)/mk/rte.lib.mk diff --git a/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt.c b/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt.c new file mode 100644 index 00000000..7bd29fae --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt.c @@ -0,0 +1,767 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2015 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include <stdint.h> +#include <unistd.h> +#include <stdlib.h> +#include <string.h> +#include <sys/types.h> +#include <sys/mman.h> +#include <errno.h> +#include <sys/user.h> +#ifndef PAGE_SIZE +#define PAGE_SIZE sysconf(_SC_PAGE_SIZE) +#endif +#include <linux/binfmts.h> +#include <xen/xen-compat.h> +#if __XEN_LATEST_INTERFACE_VERSION__ < 0x00040200 +#include <xs.h> +#else +#include <xenstore.h> +#endif +#include <linux/virtio_ring.h> + +#include <rte_mbuf.h> +#include <rte_ethdev.h> +#include <rte_malloc.h> +#include <rte_memcpy.h> +#include <rte_string_fns.h> +#include <rte_vdev.h> +#include <cmdline_parse.h> +#include <cmdline_parse_etheraddr.h> + +#include "rte_xen_lib.h" +#include "virtqueue.h" +#include "rte_eth_xenvirt.h" + +#define VQ_DESC_NUM 256 +#define VIRTIO_MBUF_BURST_SZ 64 + +/* virtio_idx is increased after new device is created.*/ +static int virtio_idx = 0; + +static struct rte_eth_link pmd_link = { + .link_speed = ETH_SPEED_NUM_10G, + .link_duplex = ETH_LINK_FULL_DUPLEX, + .link_status = ETH_LINK_DOWN, + .link_autoneg = ETH_LINK_SPEED_FIXED +}; + +static void +eth_xenvirt_free_queues(struct rte_eth_dev *dev); + +static uint16_t +eth_xenvirt_rx(void *q, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) +{ + struct virtqueue *rxvq = q; + struct rte_mbuf *rxm, *new_mbuf; + uint16_t nb_used, num; + uint32_t len[VIRTIO_MBUF_BURST_SZ]; + uint32_t i; + struct pmd_internals *pi = rxvq->internals; + + nb_used = VIRTQUEUE_NUSED(rxvq); + + rte_smp_rmb(); + num = (uint16_t)(likely(nb_used <= nb_pkts) ? nb_used : nb_pkts); + num = (uint16_t)(likely(num <= VIRTIO_MBUF_BURST_SZ) ? num : VIRTIO_MBUF_BURST_SZ); + if (unlikely(num == 0)) return 0; + + num = virtqueue_dequeue_burst(rxvq, rx_pkts, len, num); + PMD_RX_LOG(DEBUG, "used:%d dequeue:%d\n", nb_used, num); + for (i = 0; i < num ; i ++) { + rxm = rx_pkts[i]; + PMD_RX_LOG(DEBUG, "packet len:%d\n", len[i]); + rxm->next = NULL; + rxm->data_off = RTE_PKTMBUF_HEADROOM; + rxm->data_len = (uint16_t)(len[i] - sizeof(struct virtio_net_hdr)); + rxm->nb_segs = 1; + rxm->port = pi->port_id; + rxm->pkt_len = (uint32_t)(len[i] - sizeof(struct virtio_net_hdr)); + } + /* allocate new mbuf for the used descriptor */ + while (likely(!virtqueue_full(rxvq))) { + new_mbuf = rte_mbuf_raw_alloc(rxvq->mpool); + if (unlikely(new_mbuf == NULL)) { + break; + } + if (unlikely(virtqueue_enqueue_recv_refill(rxvq, new_mbuf))) { + rte_pktmbuf_free_seg(new_mbuf); + break; + } + } + pi->eth_stats.ipackets += num; + return num; +} + +static uint16_t +eth_xenvirt_tx(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) +{ + struct virtqueue *txvq = tx_queue; + struct rte_mbuf *txm; + uint16_t nb_used, nb_tx, num, i; + int error; + uint32_t len[VIRTIO_MBUF_BURST_SZ]; + struct rte_mbuf *snd_pkts[VIRTIO_MBUF_BURST_SZ]; + struct pmd_internals *pi = txvq->internals; + + nb_tx = 0; + + if (unlikely(nb_pkts == 0)) + return 0; + + PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts); + nb_used = VIRTQUEUE_NUSED(txvq); + + rte_smp_rmb(); + + num = (uint16_t)(likely(nb_used <= VIRTIO_MBUF_BURST_SZ) ? nb_used : VIRTIO_MBUF_BURST_SZ); + num = virtqueue_dequeue_burst(txvq, snd_pkts, len, num); + + for (i = 0; i < num ; i ++) { + /* mergable not supported, one segment only */ + rte_pktmbuf_free_seg(snd_pkts[i]); + } + + while (nb_tx < nb_pkts) { + if (likely(!virtqueue_full(txvq))) { + /* TODO drop tx_pkts if it contains multiple segments */ + txm = tx_pkts[nb_tx]; + error = virtqueue_enqueue_xmit(txvq, txm); + if (unlikely(error)) { + if (error == ENOSPC) + PMD_TX_LOG(ERR, "virtqueue_enqueue Free count = 0\n"); + else if (error == EMSGSIZE) + PMD_TX_LOG(ERR, "virtqueue_enqueue Free count < 1\n"); + else + PMD_TX_LOG(ERR, "virtqueue_enqueue error: %d\n", error); + break; + } + nb_tx++; + } else { + PMD_TX_LOG(ERR, "No free tx descriptors to transmit\n"); + /* virtqueue_notify not needed in our para-virt solution */ + break; + } + } + pi->eth_stats.opackets += nb_tx; + return nb_tx; +} + +static int +eth_dev_configure(struct rte_eth_dev *dev __rte_unused) +{ + RTE_LOG(ERR, PMD, "%s\n", __func__); + return 0; +} + +/* + * Create a shared page between guest and host. + * Host monitors this page if it is cleared on unmap, and then + * do necessary clean up. + */ +static void +gntalloc_vring_flag(int vtidx) +{ + char key_str[PATH_MAX]; + char val_str[PATH_MAX]; + uint32_t gref_tmp; + void *ptr; + + if (grefwatch_from_alloc(&gref_tmp, &ptr)) { + RTE_LOG(ERR, PMD, "grefwatch_from_alloc error\n"); + exit(0); + } + + *(uint8_t *)ptr = MAP_FLAG; + snprintf(val_str, sizeof(val_str), "%u", gref_tmp); + snprintf(key_str, sizeof(key_str), + DPDK_XENSTORE_PATH"%d"VRING_FLAG_STR, vtidx); + xenstore_write(key_str, val_str); +} + +/* + * Notify host this virtio device is started. + * Host could start polling this device. + */ +static void +dev_start_notify(int vtidx) +{ + char key_str[PATH_MAX]; + char val_str[PATH_MAX]; + + RTE_LOG(INFO, PMD, "%s: virtio %d is started\n", __func__, vtidx); + gntalloc_vring_flag(vtidx); + + snprintf(key_str, sizeof(key_str), "%s%s%d", + DPDK_XENSTORE_PATH, EVENT_TYPE_START_STR, + vtidx); + snprintf(val_str, sizeof(val_str), "1"); + xenstore_write(key_str, val_str); +} + +/* + * Notify host this virtio device is stopped. + * Host could stop polling this device. + */ +static void +dev_stop_notify(int vtidx) +{ + RTE_SET_USED(vtidx); +} + + +static int +update_mac_address(struct ether_addr *mac_addrs, int vtidx) +{ + char key_str[PATH_MAX]; + char val_str[PATH_MAX]; + int rv; + + if (mac_addrs == NULL) { + RTE_LOG(ERR, PMD, "%s: NULL pointer mac specified\n", __func__); + return -1; + } + rv = snprintf(key_str, sizeof(key_str), + DPDK_XENSTORE_PATH"%d_ether_addr", vtidx); + if (rv == -1) + return rv; + rv = snprintf(val_str, sizeof(val_str), "%02x:%02x:%02x:%02x:%02x:%02x", + mac_addrs->addr_bytes[0], + mac_addrs->addr_bytes[1], + mac_addrs->addr_bytes[2], + mac_addrs->addr_bytes[3], + mac_addrs->addr_bytes[4], + mac_addrs->addr_bytes[5]); + if (rv == -1) + return rv; + if (xenstore_write(key_str, val_str)) + return rv; + return 0; +} + + +static int +eth_dev_start(struct rte_eth_dev *dev) +{ + struct virtqueue *rxvq = dev->data->rx_queues[0]; + struct virtqueue *txvq = dev->data->tx_queues[0]; + struct rte_mbuf *m; + struct pmd_internals *pi = (struct pmd_internals *)dev->data->dev_private; + int rv; + + dev->data->dev_link.link_status = ETH_LINK_UP; + while (!virtqueue_full(rxvq)) { + m = rte_mbuf_raw_alloc(rxvq->mpool); + if (m == NULL) + break; + /* Enqueue allocated buffers. */ + if (virtqueue_enqueue_recv_refill(rxvq, m)) { + rte_pktmbuf_free_seg(m); + break; + } + } + + rxvq->internals = pi; + txvq->internals = pi; + + rv = update_mac_address(dev->data->mac_addrs, pi->virtio_idx); + if (rv) + return -1; + dev_start_notify(pi->virtio_idx); + + return 0; +} + +static void +eth_dev_stop(struct rte_eth_dev *dev) +{ + struct pmd_internals *pi = (struct pmd_internals *)dev->data->dev_private; + + dev->data->dev_link.link_status = ETH_LINK_DOWN; + dev_stop_notify(pi->virtio_idx); +} + +/* + * Notify host this virtio device is closed. + * Host could do necessary clean up to this device. + */ +static void +eth_dev_close(struct rte_eth_dev *dev) +{ + eth_xenvirt_free_queues(dev); +} + +static void +eth_dev_info(struct rte_eth_dev *dev, + struct rte_eth_dev_info *dev_info) +{ + struct pmd_internals *internals = dev->data->dev_private; + + RTE_SET_USED(internals); + dev_info->max_mac_addrs = 1; + dev_info->max_rx_pktlen = (uint32_t)2048; + dev_info->max_rx_queues = (uint16_t)1; + dev_info->max_tx_queues = (uint16_t)1; + dev_info->min_rx_bufsize = 0; +} + +static void +eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) +{ + struct pmd_internals *internals = dev->data->dev_private; + if(stats) + rte_memcpy(stats, &internals->eth_stats, sizeof(*stats)); +} + +static void +eth_stats_reset(struct rte_eth_dev *dev) +{ + struct pmd_internals *internals = dev->data->dev_private; + /* Reset software totals */ + memset(&internals->eth_stats, 0, sizeof(internals->eth_stats)); +} + +static void +eth_queue_release(void *q) +{ + rte_free(q); +} + +static int +eth_link_update(struct rte_eth_dev *dev __rte_unused, + int wait_to_complete __rte_unused) +{ + return 0; +} + +/* + * Create shared vring between guest and host. + * Memory is allocated through grant alloc driver, so it is not physical continuous. + */ +static void * +gntalloc_vring_create(int queue_type, uint32_t size, int vtidx) +{ + char key_str[PATH_MAX] = {0}; + char val_str[PATH_MAX] = {0}; + void *va = NULL; + int pg_size; + uint32_t pg_num; + uint32_t *gref_arr = NULL; + phys_addr_t *pa_arr = NULL; + uint64_t start_index; + int rv; + + pg_size = getpagesize(); + size = RTE_ALIGN_CEIL(size, pg_size); + pg_num = size / pg_size; + + gref_arr = calloc(pg_num, sizeof(gref_arr[0])); + pa_arr = calloc(pg_num, sizeof(pa_arr[0])); + + if (gref_arr == NULL || pa_arr == NULL) { + RTE_LOG(ERR, PMD, "%s: calloc failed\n", __func__); + goto out; + } + + va = gntalloc(size, gref_arr, &start_index); + if (va == NULL) { + RTE_LOG(ERR, PMD, "%s: gntalloc failed\n", __func__); + goto out; + } + + if (get_phys_map(va, pa_arr, pg_num, pg_size)) + goto out; + + /* write in xenstore gref and pfn for each page of vring */ + if (grant_node_create(pg_num, gref_arr, pa_arr, val_str, sizeof(val_str))) { + gntfree(va, size, start_index); + va = NULL; + goto out; + } + + if (queue_type == VTNET_RQ) + rv = snprintf(key_str, sizeof(key_str), DPDK_XENSTORE_PATH"%d"RXVRING_XENSTORE_STR, vtidx); + else + rv = snprintf(key_str, sizeof(key_str), DPDK_XENSTORE_PATH"%d"TXVRING_XENSTORE_STR, vtidx); + if (rv == -1 || xenstore_write(key_str, val_str) == -1) { + gntfree(va, size, start_index); + va = NULL; + } +out: + free(pa_arr); + free(gref_arr); + + return va; +} + + + +static struct virtqueue * +virtio_queue_setup(struct rte_eth_dev *dev, int queue_type) +{ + struct virtqueue *vq = NULL; + uint16_t vq_size = VQ_DESC_NUM; + int i = 0; + char vq_name[VIRTQUEUE_MAX_NAME_SZ]; + size_t size; + struct vring *vr; + + /* Allocate memory for virtqueue. */ + if (queue_type == VTNET_RQ) { + snprintf(vq_name, sizeof(vq_name), "port%d_rvq", + dev->data->port_id); + vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) + + vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE); + if (vq == NULL) { + RTE_LOG(ERR, PMD, "%s: unabled to allocate virtqueue\n", __func__); + return NULL; + } + memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name)); + } else if(queue_type == VTNET_TQ) { + snprintf(vq_name, sizeof(vq_name), "port%d_tvq", + dev->data->port_id); + vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) + + vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE); + if (vq == NULL) { + RTE_LOG(ERR, PMD, "%s: unabled to allocate virtqueue\n", __func__); + return NULL; + } + memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name)); + } + + memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name)); + + vq->vq_alignment = VIRTIO_PCI_VRING_ALIGN; + vq->vq_nentries = vq_size; + vq->vq_free_cnt = vq_size; + /* Calcuate vring size according to virtio spec */ + size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN); + vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN); + /* Allocate memory for virtio vring through gntalloc driver*/ + vq->vq_ring_virt_mem = gntalloc_vring_create(queue_type, vq->vq_ring_size, + ((struct pmd_internals *)dev->data->dev_private)->virtio_idx); + memset(vq->vq_ring_virt_mem, 0, vq->vq_ring_size); + vr = &vq->vq_ring; + vring_init(vr, vq_size, vq->vq_ring_virt_mem, vq->vq_alignment); + /* + * Locally maintained last consumed index, this idex trails + * vq_ring.used->idx. + */ + vq->vq_used_cons_idx = 0; + vq->vq_desc_head_idx = 0; + vq->vq_free_cnt = vq->vq_nentries; + memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries); + + /* Chain all the descriptors in the ring with an END */ + for (i = 0; i < vq_size - 1; i++) + vr->desc[i].next = (uint16_t)(i + 1); + vr->desc[i].next = VQ_RING_DESC_CHAIN_END; + + return vq; +} + +static int +eth_rx_queue_setup(struct rte_eth_dev *dev,uint16_t rx_queue_id, + uint16_t nb_rx_desc __rte_unused, + unsigned int socket_id __rte_unused, + const struct rte_eth_rxconf *rx_conf __rte_unused, + struct rte_mempool *mb_pool) +{ + struct virtqueue *vq; + vq = dev->data->rx_queues[rx_queue_id] = virtio_queue_setup(dev, VTNET_RQ); + vq->mpool = mb_pool; + return 0; +} + +static int +eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id, + uint16_t nb_tx_desc __rte_unused, + unsigned int socket_id __rte_unused, + const struct rte_eth_txconf *tx_conf __rte_unused) +{ + dev->data->tx_queues[tx_queue_id] = virtio_queue_setup(dev, VTNET_TQ); + return 0; +} + +static void +eth_xenvirt_free_queues(struct rte_eth_dev *dev) +{ + int i; + + for (i = 0; i < dev->data->nb_rx_queues; i++) { + eth_queue_release(dev->data->rx_queues[i]); + dev->data->rx_queues[i] = NULL; + } + dev->data->nb_rx_queues = 0; + + for (i = 0; i < dev->data->nb_tx_queues; i++) { + eth_queue_release(dev->data->tx_queues[i]); + dev->data->tx_queues[i] = NULL; + } + dev->data->nb_tx_queues = 0; +} + +static const struct eth_dev_ops ops = { + .dev_start = eth_dev_start, + .dev_stop = eth_dev_stop, + .dev_close = eth_dev_close, + .dev_configure = eth_dev_configure, + .dev_infos_get = eth_dev_info, + .rx_queue_setup = eth_rx_queue_setup, + .tx_queue_setup = eth_tx_queue_setup, + .rx_queue_release = eth_queue_release, + .tx_queue_release = eth_queue_release, + .link_update = eth_link_update, + .stats_get = eth_stats_get, + .stats_reset = eth_stats_reset, +}; + + +static int +rte_eth_xenvirt_parse_args(struct xenvirt_dict *dict, + const char *name, const char *params) +{ + int i; + char *pairs[RTE_ETH_XENVIRT_MAX_ARGS]; + int num_of_pairs; + char *pair[2]; + char *args; + int ret = -1; + + if (params == NULL) + return 0; + + args = rte_zmalloc(NULL, strlen(params) + 1, RTE_CACHE_LINE_SIZE); + if (args == NULL) { + RTE_LOG(ERR, PMD, "Couldn't parse %s device \n", name); + return -1; + } + rte_memcpy(args, params, strlen(params)); + + num_of_pairs = rte_strsplit(args, strnlen(args, MAX_ARG_STRLEN), + pairs, + RTE_ETH_XENVIRT_MAX_ARGS , + RTE_ETH_XENVIRT_PAIRS_DELIM); + + for (i = 0; i < num_of_pairs; i++) { + pair[0] = NULL; + pair[1] = NULL; + rte_strsplit(pairs[i], strnlen(pairs[i], MAX_ARG_STRLEN), + pair, 2, + RTE_ETH_XENVIRT_KEY_VALUE_DELIM); + + if (pair[0] == NULL || pair[1] == NULL || pair[0][0] == 0 + || pair[1][0] == 0) { + RTE_LOG(ERR, PMD, + "Couldn't parse %s device," + "wrong key or value \n", name); + goto err; + } + + if (!strncmp(pair[0], RTE_ETH_XENVIRT_MAC_PARAM, + sizeof(RTE_ETH_XENVIRT_MAC_PARAM))) { + if (cmdline_parse_etheraddr(NULL, + pair[1], + &dict->addr, + sizeof(dict->addr)) < 0) { + RTE_LOG(ERR, PMD, + "Invalid %s device ether address\n", + name); + goto err; + } + + dict->addr_valid = 1; + } + } + + ret = 0; +err: + rte_free(args); + return ret; +} + +enum dev_action { + DEV_CREATE, + DEV_ATTACH +}; + +static struct rte_vdev_driver pmd_xenvirt_drv; + +static int +eth_dev_xenvirt_create(const char *name, const char *params, + const unsigned numa_node, + enum dev_action action) +{ + struct rte_eth_dev_data *data = NULL; + struct pmd_internals *internals = NULL; + struct rte_eth_dev *eth_dev = NULL; + struct xenvirt_dict dict; + + memset(&dict, 0, sizeof(struct xenvirt_dict)); + + RTE_LOG(INFO, PMD, "Creating virtio rings backed ethdev on numa socket %u\n", + numa_node); + RTE_SET_USED(action); + + if (rte_eth_xenvirt_parse_args(&dict, name, params) < 0) { + RTE_LOG(ERR, PMD, "%s: Failed to parse ethdev parameters\n", __func__); + return -1; + } + + /* now do all data allocation - for eth_dev structure, dummy pci driver + * and internal (private) data + */ + data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node); + if (data == NULL) + goto err; + + internals = rte_zmalloc_socket(name, sizeof(*internals), 0, numa_node); + if (internals == NULL) + goto err; + + /* reserve an ethdev entry */ + eth_dev = rte_eth_dev_allocate(name); + if (eth_dev == NULL) + goto err; + + data->dev_private = internals; + data->port_id = eth_dev->data->port_id; + data->nb_rx_queues = (uint16_t)1; + data->nb_tx_queues = (uint16_t)1; + data->dev_link = pmd_link; + data->mac_addrs = rte_zmalloc("xen_virtio", ETHER_ADDR_LEN, 0); + + if(dict.addr_valid) + memcpy(&data->mac_addrs->addr_bytes, &dict.addr, sizeof(struct ether_addr)); + else + eth_random_addr(&data->mac_addrs->addr_bytes[0]); + + eth_dev->data = data; + eth_dev->dev_ops = &ops; + + eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE; + eth_dev->data->kdrv = RTE_KDRV_NONE; + eth_dev->data->drv_name = pmd_xenvirt_drv.driver.name; + eth_dev->data->numa_node = numa_node; + + eth_dev->rx_pkt_burst = eth_xenvirt_rx; + eth_dev->tx_pkt_burst = eth_xenvirt_tx; + + internals->virtio_idx = virtio_idx++; + internals->port_id = eth_dev->data->port_id; + + return 0; + +err: + rte_free(data); + rte_free(internals); + + return -1; +} + + +static int +eth_dev_xenvirt_free(const char *name, const unsigned numa_node) +{ + struct rte_eth_dev *eth_dev = NULL; + + RTE_LOG(DEBUG, PMD, + "Free virtio rings backed ethdev on numa socket %u\n", + numa_node); + + /* find an ethdev entry */ + eth_dev = rte_eth_dev_allocated(name); + if (eth_dev == NULL) + return -1; + + if (eth_dev->data->dev_started == 1) { + eth_dev_stop(eth_dev); + eth_dev_close(eth_dev); + } + + eth_dev->rx_pkt_burst = NULL; + eth_dev->tx_pkt_burst = NULL; + eth_dev->dev_ops = NULL; + + rte_free(eth_dev->data); + rte_free(eth_dev->data->dev_private); + rte_free(eth_dev->data->mac_addrs); + + virtio_idx--; + + return 0; +} + +/*TODO: Support multiple process model */ +static int +rte_pmd_xenvirt_probe(struct rte_vdev_device *dev) +{ + if (virtio_idx == 0) { + if (xenstore_init() != 0) { + RTE_LOG(ERR, PMD, "%s: xenstore init failed\n", __func__); + return -1; + } + if (gntalloc_open() != 0) { + RTE_LOG(ERR, PMD, "%s: grant init failed\n", __func__); + return -1; + } + } + eth_dev_xenvirt_create(rte_vdev_device_name(dev), + rte_vdev_device_args(dev), rte_socket_id(), DEV_CREATE); + return 0; +} + +static int +rte_pmd_xenvirt_remove(struct rte_vdev_device *dev) +{ + eth_dev_xenvirt_free(rte_vdev_device_name(dev), rte_socket_id()); + + if (virtio_idx == 0) { + if (xenstore_uninit() != 0) + RTE_LOG(ERR, PMD, "%s: xenstore uninit failed\n", __func__); + + gntalloc_close(); + } + return 0; +} + +static struct rte_vdev_driver pmd_xenvirt_drv = { + .probe = rte_pmd_xenvirt_probe, + .remove = rte_pmd_xenvirt_remove, +}; + +RTE_PMD_REGISTER_VDEV(net_xenvirt, pmd_xenvirt_drv); +RTE_PMD_REGISTER_ALIAS(net_xenvirt, eth_xenvirt); +RTE_PMD_REGISTER_PARAM_STRING(net_xenvirt, + "mac=<mac addr>"); diff --git a/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt.h b/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt.h new file mode 100644 index 00000000..598adc6f --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt.h @@ -0,0 +1,61 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef _RTE_ETH_XENVIRT_H_ +#define _RTE_ETH_XENVIRT_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include <rte_mempool.h> + +/** + * Creates mempool for xen virtio PMD. + * This function uses memzone_reserve to allocate memory for meta data, + * and uses grant alloc driver to allocate memory for data area. + * The input parameters are exactly the same as rte_mempool_create. + */ +struct rte_mempool * +rte_mempool_gntalloc_create(const char *name, unsigned elt_num, unsigned elt_size, + unsigned cache_size, unsigned private_data_size, + rte_mempool_ctor_t *mp_init, void *mp_init_arg, + rte_mempool_obj_cb_t *obj_init, void *obj_init_arg, + int socket_id, unsigned flags); + + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt_version.map b/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt_version.map new file mode 100644 index 00000000..dd636f72 --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/rte_eth_xenvirt_version.map @@ -0,0 +1,7 @@ +DPDK_2.0 { + global: + + rte_mempool_gntalloc_create; + + local: *; +}; diff --git a/src/seastar/dpdk/drivers/net/xenvirt/rte_mempool_gntalloc.c b/src/seastar/dpdk/drivers/net/xenvirt/rte_mempool_gntalloc.c new file mode 100644 index 00000000..73e82f80 --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/rte_mempool_gntalloc.c @@ -0,0 +1,295 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include <stdint.h> +#include <unistd.h> +#include <stdlib.h> +#include <sys/mman.h> +#include <sys/ioctl.h> +#include <string.h> +#include <xen/sys/gntalloc.h> + +#include <rte_common.h> +#include <rte_mempool.h> +#include <rte_memory.h> +#include <rte_errno.h> + +#include "rte_xen_lib.h" +#include "rte_eth_xenvirt.h" + +struct _gntarr { + uint32_t gref; + phys_addr_t pa; + uint64_t index; + void *va; +}; + +struct _mempool_gntalloc_info { + struct rte_mempool *mp; + uint32_t pg_num; + uint32_t *gref_arr; + phys_addr_t *pa_arr; + void *va; + uint32_t mempool_idx; + uint64_t start_index; +}; + + +static rte_atomic32_t global_xenvirt_mempool_idx = RTE_ATOMIC32_INIT(-1); + +static int +compare(const void *p1, const void *p2) +{ + return ((const struct _gntarr *)p1)->pa - ((const struct _gntarr *)p2)->pa; +} + + +static struct _mempool_gntalloc_info +_create_mempool(const char *name, unsigned elt_num, unsigned elt_size, + unsigned cache_size, unsigned private_data_size, + rte_mempool_ctor_t *mp_init, void *mp_init_arg, + rte_mempool_obj_cb_t *obj_init, void *obj_init_arg, + int socket_id, unsigned flags) +{ + struct _mempool_gntalloc_info mgi; + struct rte_mempool *mp = NULL; + struct rte_mempool_objsz objsz; + uint32_t pg_num, rpg_num, pg_shift, pg_sz; + char *va, *orig_va, *uv; /* uv: from which, the pages could be freed */ + ssize_t sz, usz; /* usz: unused size */ + /* + * for each page allocated through xen_gntalloc driver, + * gref_arr:stores grant references, + * pa_arr: stores physical address, + * gnt_arr: stores all meta dat + */ + uint32_t *gref_arr = NULL; + phys_addr_t *pa_arr = NULL; + struct _gntarr *gnt_arr = NULL; + /* start index of the grant referances, used for dealloc*/ + uint64_t start_index; + uint32_t i, j; + int rv = 0; + struct ioctl_gntalloc_dealloc_gref arg; + + mgi.mp = NULL; + va = orig_va = uv = NULL; + pg_num = rpg_num = 0; + sz = 0; + + pg_sz = getpagesize(); + if (rte_is_power_of_2(pg_sz) == 0) { + goto out; + } + pg_shift = rte_bsf32(pg_sz); + + rte_mempool_calc_obj_size(elt_size, flags, &objsz); + sz = rte_mempool_xmem_size(elt_num, objsz.total_size, pg_shift); + pg_num = sz >> pg_shift; + + pa_arr = calloc(pg_num, sizeof(pa_arr[0])); + gref_arr = calloc(pg_num, sizeof(gref_arr[0])); + gnt_arr = calloc(pg_num, sizeof(gnt_arr[0])); + if ((gnt_arr == NULL) || (gref_arr == NULL) || (pa_arr == NULL)) + goto out; + + /* grant index is continuous in ascending order */ + orig_va = gntalloc(sz, gref_arr, &start_index); + if (orig_va == NULL) + goto out; + + get_phys_map(orig_va, pa_arr, pg_num, pg_sz); + for (i = 0; i < pg_num; i++) { + gnt_arr[i].index = start_index + i * pg_sz; + gnt_arr[i].gref = gref_arr[i]; + gnt_arr[i].pa = pa_arr[i]; + gnt_arr[i].va = RTE_PTR_ADD(orig_va, i * pg_sz); + } + qsort(gnt_arr, pg_num, sizeof(struct _gntarr), compare); + + va = get_xen_virtual(sz, pg_sz); + if (va == NULL) { + goto out; + } + + /* + * map one by one, as index isn't continuous now. + * pg_num VMAs, doesn't linux has a limitation on this? + */ + for (i = 0; i < pg_num; i++) { + /* update gref_arr and pa_arr after sort */ + gref_arr[i] = gnt_arr[i].gref; + pa_arr[i] = gnt_arr[i].pa; + gnt_arr[i].va = mmap(va + i * pg_sz, pg_sz, PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_FIXED, gntalloc_fd, gnt_arr[i].index); + if ((gnt_arr[i].va == MAP_FAILED) || (gnt_arr[i].va != (va + i * pg_sz))) { + RTE_LOG(ERR, PMD, "failed to map %d pages\n", i); + goto mmap_failed; + } + } + + /* + * Check that allocated size is big enough to hold elt_num + * objects and a calcualte how many bytes are actually required. + */ + usz = rte_mempool_xmem_usage(va, elt_num, objsz.total_size, pa_arr, pg_num, pg_shift); + if (usz < 0) { + mp = NULL; + i = pg_num; + goto mmap_failed; + } else { + /* unmap unused pages if any */ + uv = RTE_PTR_ADD(va, usz); + if ((usz = va + sz - uv) > 0) { + + RTE_LOG(ERR, PMD, + "%s(%s): unmap unused %zu of %zu " + "mmaped bytes @%p orig:%p\n", + __func__, name, usz, sz, uv, va); + munmap(uv, usz); + i = (sz - usz) / pg_sz; + for (; i < pg_num; i++) { + arg.count = 1; + arg.index = gnt_arr[i].index; + rv = ioctl(gntalloc_fd, IOCTL_GNTALLOC_DEALLOC_GREF, &arg); + if (rv) { + /* shouldn't fail here */ + RTE_LOG(ERR, PMD, "va=%p pa=%"PRIu64"x index=%"PRIu64" %s\n", + gnt_arr[i].va, + gnt_arr[i].pa, + arg.index, strerror(errno)); + rte_panic("gntdealloc failed when freeing pages\n"); + } + } + + rpg_num = (sz - usz) >> pg_shift; + } else + rpg_num = pg_num; + + mp = rte_mempool_xmem_create(name, elt_num, elt_size, + cache_size, private_data_size, + mp_init, mp_init_arg, + obj_init, obj_init_arg, + socket_id, flags, va, pa_arr, rpg_num, pg_shift); + + RTE_ASSERT(elt_num == mp->size); + } + mgi.mp = mp; + mgi.pg_num = rpg_num; + mgi.gref_arr = gref_arr; + mgi.pa_arr = pa_arr; + if (mp) + mgi.mempool_idx = rte_atomic32_add_return(&global_xenvirt_mempool_idx, 1); + mgi.start_index = start_index; + mgi.va = va; + + if (mp == NULL) { + i = pg_num; + goto mmap_failed; + } + +/* + * unmap only, without deallocate grant reference. + * unused pages have already been unmaped, + * unmap twice will fail, but it is safe. + */ +mmap_failed: + for (j = 0; j < i; j++) { + if (gnt_arr[i].va) + munmap(gnt_arr[i].va, pg_sz); + } +out: + free(gnt_arr); + if (orig_va) + munmap(orig_va, sz); + if (mp == NULL) { + free(gref_arr); + free(pa_arr); + + /* some gref has already been de-allocated from the list in the driver, + * so dealloc one by one, and it is safe to deallocate twice + */ + if (orig_va) { + for (i = 0; i < pg_num; i++) { + arg.index = start_index + i * pg_sz; + rv = ioctl(gntalloc_fd, IOCTL_GNTALLOC_DEALLOC_GREF, arg); + } + } + } + return mgi; +} + +struct rte_mempool * +rte_mempool_gntalloc_create(const char *name, unsigned elt_num, unsigned elt_size, + unsigned cache_size, unsigned private_data_size, + rte_mempool_ctor_t *mp_init, void *mp_init_arg, + rte_mempool_obj_cb_t *obj_init, void *obj_init_arg, + int socket_id, unsigned flags) +{ + int rv; + uint32_t i; + struct _mempool_gntalloc_info mgi; + struct ioctl_gntalloc_dealloc_gref arg; + int pg_sz = getpagesize(); + + mgi = _create_mempool(name, elt_num, elt_size, + cache_size, private_data_size, + mp_init, mp_init_arg, + obj_init, obj_init_arg, + socket_id, flags); + if (mgi.mp) { + rv = grant_gntalloc_mbuf_pool(mgi.mp, + mgi.pg_num, + mgi.gref_arr, + mgi.pa_arr, + mgi.mempool_idx); + free(mgi.gref_arr); + free(mgi.pa_arr); + if (rv == 0) + return mgi.mp; + /* + * in _create_mempool, unused pages have already been unmapped, deallocagted + * unmap and dealloc the remained ones here. + */ + munmap(mgi.va, pg_sz * mgi.pg_num); + for (i = 0; i < mgi.pg_num; i++) { + arg.index = mgi.start_index + i * pg_sz; + rv = ioctl(gntalloc_fd, IOCTL_GNTALLOC_DEALLOC_GREF, arg); + } + return NULL; + } + return NULL; + + + +} diff --git a/src/seastar/dpdk/drivers/net/xenvirt/rte_xen_lib.c b/src/seastar/dpdk/drivers/net/xenvirt/rte_xen_lib.c new file mode 100644 index 00000000..6c9a1d49 --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/rte_xen_lib.c @@ -0,0 +1,454 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2015 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include <stdint.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <sys/types.h> +#include <fcntl.h> +#include <sys/mman.h> +#include <sys/ioctl.h> +#include <xen/xen-compat.h> +#if __XEN_LATEST_INTERFACE_VERSION__ < 0x00040200 +#include <xs.h> +#else +#include <xenstore.h> +#endif +#include <xen/sys/gntalloc.h> + +#include <rte_common.h> +#include <rte_string_fns.h> +#include <rte_malloc.h> + +#include "rte_xen_lib.h" + +/* + * The grant node format in xenstore for vring/mpool is: + * 0_rx_vring_gref = "gref1#, gref2#, gref3#" + * 0_mempool_gref = "gref1#, gref2#, gref3#" + * each gref# is a grant reference for a shared page. + * In each shared page, we store the grant_node_item items. + */ +struct grant_node_item { + uint32_t gref; + uint32_t pfn; +} __attribute__((packed)); + +/* fd for xen_gntalloc driver, used to allocate grant pages*/ +int gntalloc_fd = -1; + +/* xenstore path for local domain, now it is '/local/domain/domid/' */ +static char *dompath = NULL; +/* handle to xenstore read/write operations */ +static struct xs_handle *xs = NULL; +/* flag to indicate if xenstore cleanup is required */ +static bool is_xenstore_cleaned_up; + +/* + * Reserve a virtual address space. + * On success, returns the pointer. On failure, returns NULL. + */ +void * +get_xen_virtual(size_t size, size_t page_sz) +{ + void *addr; + uintptr_t aligned_addr; + + addr = mmap(NULL, size + page_sz, PROT_READ, MAP_SHARED | MAP_ANONYMOUS, -1, 0); + if (addr == MAP_FAILED) { + RTE_LOG(ERR, PMD, "failed get a virtual area\n"); + return NULL; + } + + aligned_addr = RTE_ALIGN_CEIL((uintptr_t)addr, page_sz); + addr = (void *)(aligned_addr); + + return addr; +} + +/* + * Get the physical address for virtual memory starting at va. + */ +int +get_phys_map(void *va, phys_addr_t pa[], uint32_t pg_num, uint32_t pg_sz) +{ + int32_t fd, rc = 0; + uint32_t i, nb; + off_t ofs; + + ofs = (uintptr_t)va / pg_sz * sizeof(*pa); + nb = pg_num * sizeof(*pa); + + if ((fd = open(PAGEMAP_FNAME, O_RDONLY)) < 0 || + (rc = pread(fd, pa, nb, ofs)) < 0 || + (rc -= nb) != 0) { + RTE_LOG(ERR, PMD, "%s: failed read of %u bytes from \'%s\' " + "at offset %lu, error code: %d\n", + __func__, nb, PAGEMAP_FNAME, (unsigned long)ofs, errno); + rc = ENOENT; + } + + close(fd); + for (i = 0; i != pg_num; i++) + pa[i] = (pa[i] & PAGEMAP_PFN_MASK) * pg_sz; + + return rc; +} + +int +gntalloc_open(void) +{ + gntalloc_fd = open(XEN_GNTALLOC_FNAME, O_RDWR); + return (gntalloc_fd != -1) ? 0 : -1; +} + +void +gntalloc_close(void) +{ + if (gntalloc_fd != -1) + close(gntalloc_fd); + gntalloc_fd = -1; +} + +void * +gntalloc(size_t size, uint32_t *gref, uint64_t *start_index) +{ + int page_size = getpagesize(); + uint32_t i, pg_num; + void *va; + int rv; + struct ioctl_gntalloc_alloc_gref *arg; + struct ioctl_gntalloc_dealloc_gref arg_d; + + if (size % page_size) { + RTE_LOG(ERR, PMD, "%s: %zu isn't multiple of page size\n", + __func__, size); + return NULL; + } + + pg_num = size / page_size; + arg = malloc(sizeof(*arg) + (pg_num - 1) * sizeof(uint32_t)); + if (arg == NULL) + return NULL; + arg->domid = DOM0_DOMID; + arg->flags = GNTALLOC_FLAG_WRITABLE; + arg->count = pg_num; + + rv = ioctl(gntalloc_fd, IOCTL_GNTALLOC_ALLOC_GREF, arg); + if (rv) { + RTE_LOG(ERR, PMD, "%s: ioctl error\n", __func__); + free(arg); + return NULL; + } + + va = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, gntalloc_fd, arg->index); + if (va == MAP_FAILED) { + RTE_LOG(ERR, PMD, "%s: mmap failed\n", __func__); + arg_d.count = pg_num; + arg_d.index = arg->index; + ioctl(gntalloc_fd, IOCTL_GNTALLOC_DEALLOC_GREF, arg_d); + free(arg); + return NULL; + } + + if (gref) { + for (i = 0; i < pg_num; i++) { + gref[i] = arg->gref_ids[i]; + } + } + if (start_index) + *start_index = arg->index; + + free(arg); + + return va; +} + +int +grefwatch_from_alloc(uint32_t *gref, void **pptr) +{ + int rv; + void *ptr; + int pg_size = getpagesize(); + struct ioctl_gntalloc_alloc_gref arg = { + .domid = DOM0_DOMID, + .flags = GNTALLOC_FLAG_WRITABLE, + .count = 1 + }; + struct ioctl_gntalloc_dealloc_gref arg_d; + struct ioctl_gntalloc_unmap_notify notify = { + .action = UNMAP_NOTIFY_CLEAR_BYTE + }; + + rv = ioctl(gntalloc_fd, IOCTL_GNTALLOC_ALLOC_GREF, &arg); + if (rv) { + RTE_LOG(ERR, PMD, "%s: ioctl error\n", __func__); + return -1; + } + + ptr = (void *)mmap(NULL, pg_size, PROT_READ|PROT_WRITE, MAP_SHARED, gntalloc_fd, arg.index); + arg_d.index = arg.index; + arg_d.count = 1; + if (ptr == MAP_FAILED) { + RTE_LOG(ERR, PMD, "%s: mmap failed\n", __func__); + ioctl(gntalloc_fd, IOCTL_GNTALLOC_DEALLOC_GREF, &arg_d); + return -1; + } + if (pptr) + *pptr = ptr; + if (gref) + *gref = arg.gref_ids[0]; + + notify.index = arg.index; + rv = ioctl(gntalloc_fd, IOCTL_GNTALLOC_SET_UNMAP_NOTIFY, ¬ify); + if (rv) { + RTE_LOG(ERR, PMD, "%s: unmap notify failed\n", __func__); + munmap(ptr, pg_size); + ioctl(gntalloc_fd, IOCTL_GNTALLOC_DEALLOC_GREF, &arg_d); + return -1; + } + + return 0; +} + +void +gntfree(void *va, size_t sz, uint64_t start_index) +{ + struct ioctl_gntalloc_dealloc_gref arg_d; + + if (va && sz) { + munmap(va, sz); + arg_d.count = sz / getpagesize(); + arg_d.index = start_index; + ioctl(gntalloc_fd, IOCTL_GNTALLOC_DEALLOC_GREF, &arg_d); + } +} + +static int +xenstore_cleanup(void) +{ + char store_path[PATH_MAX] = {0}; + + if (snprintf(store_path, sizeof(store_path), + "%s%s", dompath, DPDK_XENSTORE_NODE) == -1) + return -1; + + if (xs_rm(xs, XBT_NULL, store_path) == false) { + RTE_LOG(ERR, PMD, "%s: failed cleanup node\n", __func__); + return -1; + } + + return 0; +} + +int +xenstore_init(void) +{ + unsigned int len, domid; + char *buf; + char *end; + + xs = xs_domain_open(); + if (xs == NULL) { + RTE_LOG(ERR, PMD,"%s: xs_domain_open failed\n", __func__); + return -1; + } + buf = xs_read(xs, XBT_NULL, "domid", &len); + if (buf == NULL) { + RTE_LOG(ERR, PMD, "%s: failed read domid\n", __func__); + return -1; + } + errno = 0; + domid = strtoul(buf, &end, 0); + if (errno != 0 || end == NULL || end == buf || domid == 0) + return -1; + + RTE_LOG(INFO, PMD, "retrieved dom ID = %d\n", domid); + + dompath = xs_get_domain_path(xs, domid); + if (dompath == NULL) + return -1; + + xs_transaction_start(xs); /* When to stop transaction */ + + if (is_xenstore_cleaned_up == 0) { + if (xenstore_cleanup()) + return -1; + is_xenstore_cleaned_up = 1; + } + + return 0; +} + +int +xenstore_uninit(void) +{ + xs_close(xs); + + if (is_xenstore_cleaned_up == 0) { + if (xenstore_cleanup()) + return -1; + is_xenstore_cleaned_up = 1; + } + free(dompath); + dompath = NULL; + + return 0; +} + +int +xenstore_write(const char *key_str, const char *val_str) +{ + char grant_path[PATH_MAX]; + int rv, len; + + if (xs == NULL) { + RTE_LOG(ERR, PMD, "%s: xenstore init failed\n", __func__); + return -1; + } + rv = snprintf(grant_path, sizeof(grant_path), "%s%s", dompath, key_str); + if (rv == -1) { + RTE_LOG(ERR, PMD, "%s: snprintf %s %s failed\n", + __func__, dompath, key_str); + return -1; + } + len = strnlen(val_str, PATH_MAX); + + if (xs_write(xs, XBT_NULL, grant_path, val_str, len) == false) { + RTE_LOG(ERR, PMD, "%s: xs_write failed\n", __func__); + return -1; + } + + return 0; +} + +int +grant_node_create(uint32_t pg_num, uint32_t *gref_arr, phys_addr_t *pa_arr, char *val_str, size_t str_size) +{ + uint64_t start_index; + int pg_size; + uint32_t pg_shift; + void *ptr = NULL; + uint32_t count, entries_per_pg; + uint32_t i, j = 0, k = 0; + uint32_t *gref_tmp; + int first = 1; + char tmp_str[PATH_MAX] = {0}; + int rv = -1; + + pg_size = getpagesize(); + if (rte_is_power_of_2(pg_size) == 0) { + return -1; + } + pg_shift = rte_bsf32(pg_size); + if (pg_size % sizeof(struct grant_node_item)) { + RTE_LOG(ERR, PMD, "pg_size isn't a multiple of grant node item\n"); + return -1; + } + + entries_per_pg = pg_size / sizeof(struct grant_node_item); + count = (pg_num + entries_per_pg - 1 ) / entries_per_pg; + gref_tmp = malloc(count * sizeof(uint32_t)); + if (gref_tmp == NULL) + return -1; + ptr = gntalloc(pg_size * count, gref_tmp, &start_index); + if (ptr == NULL) { + RTE_LOG(ERR, PMD, "%s: gntalloc error of %d pages\n", __func__, count); + free(gref_tmp); + return -1; + } + + while (j < pg_num) { + if (first) { + rv = snprintf(val_str, str_size, "%u", gref_tmp[k]); + first = 0; + } else { + snprintf(tmp_str, PATH_MAX, "%s", val_str); + rv = snprintf(val_str, str_size, "%s,%u", tmp_str, gref_tmp[k]); + } + k++; + if (rv == -1) + break; + + for (i = 0; i < entries_per_pg && j < pg_num ; i++) { + ((struct grant_node_item *)ptr)->gref = gref_arr[j]; + ((struct grant_node_item *)ptr)->pfn = pa_arr[j] >> pg_shift; + ptr = RTE_PTR_ADD(ptr, sizeof(struct grant_node_item)); + j++; + } + } + if (rv == -1) { + gntfree(ptr, pg_size * count, start_index); + } else + rv = 0; + free(gref_tmp); + return rv; +} + + +int +grant_gntalloc_mbuf_pool(struct rte_mempool *mpool, uint32_t pg_num, uint32_t *gref_arr, phys_addr_t *pa_arr, int mempool_idx) +{ + char key_str[PATH_MAX] = {0}; + char val_str[PATH_MAX] = {0}; + void *mempool_obj_va; + + if (grant_node_create(pg_num, gref_arr, pa_arr, val_str, sizeof(val_str))) { + return -1; + } + + if (snprintf(key_str, sizeof(key_str), + DPDK_XENSTORE_PATH"%d"MEMPOOL_XENSTORE_STR, mempool_idx) == -1) + return -1; + if (xenstore_write(key_str, val_str) == -1) + return -1; + + if (snprintf(key_str, sizeof(key_str), + DPDK_XENSTORE_PATH"%d"MEMPOOL_VA_XENSTORE_STR, mempool_idx) == -1) + return -1; + if (mpool->nb_mem_chunks != 1) { + RTE_LOG(ERR, PMD, + "mempool with more than 1 chunk is not supported\n"); + return -1; + } + mempool_obj_va = STAILQ_FIRST(&mpool->mem_list)->addr; + if (snprintf(val_str, sizeof(val_str), "%"PRIxPTR, + (uintptr_t)mempool_obj_va) == -1) + return -1; + if (xenstore_write(key_str, val_str) == -1) + return -1; + + return 0; +} diff --git a/src/seastar/dpdk/drivers/net/xenvirt/rte_xen_lib.h b/src/seastar/dpdk/drivers/net/xenvirt/rte_xen_lib.h new file mode 100644 index 00000000..d973eacb --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/rte_xen_lib.h @@ -0,0 +1,116 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2015 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef _RTE_XEN_DUMMY_PMD_H +#define _RTE_XEN_DUMMY_PMD_H + +#include <stdint.h> + +#include <rte_common.h> +#include <rte_mempool.h> +#include <rte_ether.h> + +#define PAGEMAP_FNAME "/proc/self/pagemap" +#define XEN_GNTALLOC_FNAME "/dev/xen/gntalloc" +#define DPDK_XENSTORE_PATH "/control/dpdk/" +#define DPDK_XENSTORE_NODE "/control/dpdk" +/*format 0_mempool_gref = "1537,1524,1533" */ +#define MEMPOOL_XENSTORE_STR "_mempool_gref" +/*format 0_mempool_va = 0x80340000 */ +#define MEMPOOL_VA_XENSTORE_STR "_mempool_va" +/*format 0_rx_vring_gref = "1537,1524,1533" */ +#define RXVRING_XENSTORE_STR "_rx_vring_gref" +/*format 0_tx_vring_gref = "1537,1524,1533" */ +#define TXVRING_XENSTORE_STR "_tx_vring_gref" +#define VRING_FLAG_STR "_vring_flag" +/*format: event_type_start_0 = 1*/ +#define EVENT_TYPE_START_STR "event_type_start_" + +#define DOM0_DOMID 0 +/* + * the pfn (page frame number) are bits 0-54 (see pagemap.txt in linux + * Documentation). + */ +#define PAGEMAP_PFN_BITS 54 +#define PAGEMAP_PFN_MASK RTE_LEN2MASK(PAGEMAP_PFN_BITS, phys_addr_t) + +#define MAP_FLAG 0xA5 + +#define RTE_ETH_XENVIRT_PAIRS_DELIM ';' +#define RTE_ETH_XENVIRT_KEY_VALUE_DELIM '=' +#define RTE_ETH_XENVIRT_MAX_ARGS 1 +#define RTE_ETH_XENVIRT_MAC_PARAM "mac" +struct xenvirt_dict { + uint8_t addr_valid; + struct ether_addr addr; +}; + +extern int gntalloc_fd; + +int +gntalloc_open(void); + +void +gntalloc_close(void); + +void * +gntalloc(size_t sz, uint32_t *gref, uint64_t *start_index); + +void +gntfree(void *va, size_t sz, uint64_t start_index); + +int +xenstore_init(void); + +int +xenstore_uninit(void); + +int +xenstore_write(const char *key_str, const char *val_str); + +int +get_phys_map(void *va, phys_addr_t pa[], uint32_t pg_num, uint32_t pg_sz); + +void * +get_xen_virtual(size_t size, size_t page_sz); + +int +grefwatch_from_alloc(uint32_t *gref, void **pptr); + + +int grant_node_create(uint32_t pg_num, uint32_t *gref_arr, phys_addr_t *pa_arr, char *val_str, size_t str_size); + +int +grant_gntalloc_mbuf_pool(struct rte_mempool *mpool, uint32_t pg_num, uint32_t *gref_arr, phys_addr_t *pa_arr, int mempool_idx); + +#endif diff --git a/src/seastar/dpdk/drivers/net/xenvirt/virtio_logs.h b/src/seastar/dpdk/drivers/net/xenvirt/virtio_logs.h new file mode 100644 index 00000000..d6c33f7b --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/virtio_logs.h @@ -0,0 +1,70 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef _VIRTIO_LOGS_H_ +#define _VIRTIO_LOGS_H_ + +#include <rte_log.h> + +#ifdef RTE_LIBRTE_VIRTIO_DEBUG_INIT +#define PMD_INIT_LOG(level, fmt, args...) \ + RTE_LOG(level, PMD, "%s(): " fmt "\n", __func__, ## args) +#define PMD_INIT_FUNC_TRACE() PMD_INIT_LOG(DEBUG, " >>") +#else +#define PMD_INIT_LOG(level, fmt, args...) do { } while(0) +#define PMD_INIT_FUNC_TRACE() do { } while(0) +#endif + +#ifdef RTE_LIBRTE_VIRTIO_DEBUG_RX +#define PMD_RX_LOG(level, fmt, args...) \ + RTE_LOG(level, PMD, "%s() rx: " fmt , __func__, ## args) +#else +#define PMD_RX_LOG(level, fmt, args...) do { } while(0) +#endif + +#ifdef RTE_LIBRTE_VIRTIO_DEBUG_TX +#define PMD_TX_LOG(level, fmt, args...) \ + RTE_LOG(level, PMD, "%s() tx: " fmt , __func__, ## args) +#else +#define PMD_TX_LOG(level, fmt, args...) do { } while(0) +#endif + + +#ifdef RTE_LIBRTE_VIRTIO_DEBUG_DRIVER +#define PMD_DRV_LOG(level, fmt, args...) \ + RTE_LOG(level, PMD, "%s(): " fmt , __func__, ## args) +#else +#define PMD_DRV_LOG(level, fmt, args...) do { } while(0) +#endif + +#endif /* _VIRTIO_LOGS_H_ */ diff --git a/src/seastar/dpdk/drivers/net/xenvirt/virtqueue.h b/src/seastar/dpdk/drivers/net/xenvirt/virtqueue.h new file mode 100644 index 00000000..350eae3e --- /dev/null +++ b/src/seastar/dpdk/drivers/net/xenvirt/virtqueue.h @@ -0,0 +1,273 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef _VIRTQUEUE_H_ +#define _VIRTQUEUE_H_ + +#include <stdint.h> +#include <linux/virtio_ring.h> +#include <linux/virtio_net.h> + +#include <rte_atomic.h> +#include <rte_memory.h> +#include <rte_memzone.h> +#include <rte_mempool.h> + +#include "virtio_logs.h" + +struct rte_mbuf; + +/* The alignment to use between consumer and producer parts of vring. */ +#define VIRTIO_PCI_VRING_ALIGN 4096 + +enum { VTNET_RQ = 0, VTNET_TQ = 1, VTNET_CQ = 2 }; + +/** + * The maximum virtqueue size is 2^15. Use that value as the end of + * descriptor chain terminator since it will never be a valid index + * in the descriptor table. This is used to verify we are correctly + * handling vq_free_cnt. + */ +#define VQ_RING_DESC_CHAIN_END 32768 + +#define VIRTQUEUE_MAX_NAME_SZ 32 + +struct pmd_internals { + struct rte_eth_stats eth_stats; + int port_id; + int virtio_idx; +}; + + +struct virtqueue { + char vq_name[VIRTQUEUE_MAX_NAME_SZ]; + struct rte_mempool *mpool; /**< mempool for mbuf allocation */ + uint16_t queue_id; /**< DPDK queue index. */ + uint16_t vq_queue_index; /**< PCI queue index */ + uint8_t port_id; /**< Device port identifier. */ + + void *vq_ring_virt_mem; /**< virtual address of vring*/ + int vq_alignment; + int vq_ring_size; + + struct vring vq_ring; /**< vring keeping desc, used and avail */ + struct pmd_internals *internals; /**< virtio device internal info. */ + uint16_t vq_nentries; /**< vring desc numbers */ + uint16_t vq_desc_head_idx; + uint16_t vq_free_cnt; /**< num of desc available */ + uint16_t vq_used_cons_idx; /**< Last consumed desc in used table, trails vq_ring.used->idx*/ + + struct vq_desc_extra { + void *cookie; + uint16_t ndescs; + } vq_descx[0] __rte_cache_aligned; +}; + + +#ifdef RTE_LIBRTE_XENVIRT_DEBUG_DUMP +#define VIRTQUEUE_DUMP(vq) do { \ + uint16_t used_idx, nused; \ + used_idx = (vq)->vq_ring.used->idx; \ + nused = (uint16_t)(used_idx - (vq)->vq_used_cons_idx); \ + PMD_INIT_LOG(DEBUG, \ + "VQ: %s - size=%d; free=%d; used=%d; desc_head_idx=%d;" \ + " avail.idx=%d; used_cons_idx=%d; used.idx=%d;" \ + " avail.flags=0x%x; used.flags=0x%x\n", \ + (vq)->vq_name, (vq)->vq_nentries, (vq)->vq_free_cnt, nused, \ + (vq)->vq_desc_head_idx, (vq)->vq_ring.avail->idx, \ + (vq)->vq_used_cons_idx, (vq)->vq_ring.used->idx, \ + (vq)->vq_ring.avail->flags, (vq)->vq_ring.used->flags); \ +} while (0) +#else +#define VIRTQUEUE_DUMP(vq) do { } while (0) +#endif + + +/** + * Dump virtqueue internal structures, for debug purpose only. + */ +void virtqueue_dump(struct virtqueue *vq); + +/** + * Get all mbufs to be freed. + */ +struct rte_mbuf * virtqueue_detatch_unused(struct virtqueue *vq); + +static inline int __attribute__((always_inline)) +virtqueue_full(const struct virtqueue *vq) +{ + return vq->vq_free_cnt == 0; +} + +#define VIRTQUEUE_NUSED(vq) ((uint16_t)((vq)->vq_ring.used->idx - (vq)->vq_used_cons_idx)) + +static inline void __attribute__((always_inline)) +vq_ring_update_avail(struct virtqueue *vq, uint16_t desc_idx) +{ + uint16_t avail_idx; + /* + * Place the head of the descriptor chain into the next slot and make + * it usable to the host. The chain is made available now rather than + * deferring to virtqueue_notify() in the hopes that if the host is + * currently running on another CPU, we can keep it processing the new + * descriptor. + */ + avail_idx = (uint16_t)(vq->vq_ring.avail->idx & (vq->vq_nentries - 1)); + vq->vq_ring.avail->ring[avail_idx] = desc_idx; + rte_smp_wmb(); + vq->vq_ring.avail->idx++; +} + +static inline void __attribute__((always_inline)) +vq_ring_free_chain(struct virtqueue *vq, uint16_t desc_idx) +{ + struct vring_desc *dp; + struct vq_desc_extra *dxp; + + dp = &vq->vq_ring.desc[desc_idx]; + dxp = &vq->vq_descx[desc_idx]; + vq->vq_free_cnt = (uint16_t)(vq->vq_free_cnt + dxp->ndescs); + while (dp->flags & VRING_DESC_F_NEXT) { + dp = &vq->vq_ring.desc[dp->next]; + } + dxp->ndescs = 0; + + /* + * We must append the existing free chain, if any, to the end of + * newly freed chain. If the virtqueue was completely used, then + * head would be VQ_RING_DESC_CHAIN_END (ASSERTed above). + */ + dp->next = vq->vq_desc_head_idx; + vq->vq_desc_head_idx = desc_idx; +} + +static inline int __attribute__((always_inline)) +virtqueue_enqueue_recv_refill(struct virtqueue *rxvq, struct rte_mbuf *cookie) +{ + const uint16_t needed = 1; + const uint16_t head_idx = rxvq->vq_desc_head_idx; + struct vring_desc *start_dp = rxvq->vq_ring.desc; + struct vq_desc_extra *dxp; + + if (unlikely(rxvq->vq_free_cnt == 0)) + return -ENOSPC; + if (unlikely(rxvq->vq_free_cnt < needed)) + return -EMSGSIZE; + if (unlikely(head_idx >= rxvq->vq_nentries)) + return -EFAULT; + + dxp = &rxvq->vq_descx[head_idx]; + dxp->cookie = (void *)cookie; + dxp->ndescs = needed; + + start_dp[head_idx].addr = + (uint64_t) ((uintptr_t)cookie->buf_addr + RTE_PKTMBUF_HEADROOM - sizeof(struct virtio_net_hdr)); + start_dp[head_idx].len = cookie->buf_len - RTE_PKTMBUF_HEADROOM + sizeof(struct virtio_net_hdr); + start_dp[head_idx].flags = VRING_DESC_F_WRITE; + rxvq->vq_desc_head_idx = start_dp[head_idx].next; + rxvq->vq_free_cnt = (uint16_t)(rxvq->vq_free_cnt - needed); + vq_ring_update_avail(rxvq, head_idx); + + return 0; +} + +static inline int __attribute__((always_inline)) +virtqueue_enqueue_xmit(struct virtqueue *txvq, struct rte_mbuf *cookie) +{ + + const uint16_t needed = 2; + struct vring_desc *start_dp = txvq->vq_ring.desc; + uint16_t head_idx = txvq->vq_desc_head_idx; + uint16_t idx = head_idx; + struct vq_desc_extra *dxp; + + if (unlikely(txvq->vq_free_cnt == 0)) + return -ENOSPC; + if (unlikely(txvq->vq_free_cnt < needed)) + return -EMSGSIZE; + if (unlikely(head_idx >= txvq->vq_nentries)) + return -EFAULT; + + dxp = &txvq->vq_descx[idx]; + dxp->cookie = (void *)cookie; + dxp->ndescs = needed; + + start_dp = txvq->vq_ring.desc; + start_dp[idx].addr = 0; +/* + * TODO: save one desc here? + */ + start_dp[idx].len = sizeof(struct virtio_net_hdr); + start_dp[idx].flags = VRING_DESC_F_NEXT; + start_dp[idx].addr = (uintptr_t)NULL; + idx = start_dp[idx].next; + start_dp[idx].addr = (uint64_t)rte_pktmbuf_mtod(cookie, uintptr_t); + start_dp[idx].len = cookie->data_len; + start_dp[idx].flags = 0; + idx = start_dp[idx].next; + txvq->vq_desc_head_idx = idx; + txvq->vq_free_cnt = (uint16_t)(txvq->vq_free_cnt - needed); + vq_ring_update_avail(txvq, head_idx); + + return 0; +} + +static inline uint16_t __attribute__((always_inline)) +virtqueue_dequeue_burst(struct virtqueue *vq, struct rte_mbuf **rx_pkts, uint32_t *len, uint16_t num) +{ + struct vring_used_elem *uep; + struct rte_mbuf *cookie; + uint16_t used_idx, desc_idx; + uint16_t i; + /* Caller does the check */ + for (i = 0; i < num ; i ++) { + used_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1)); + uep = &vq->vq_ring.used->ring[used_idx]; + desc_idx = (uint16_t) uep->id; + cookie = (struct rte_mbuf *)vq->vq_descx[desc_idx].cookie; + if (unlikely(cookie == NULL)) { + PMD_DRV_LOG(ERR, "vring descriptor with no mbuf cookie at %u\n", + vq->vq_used_cons_idx); + RTE_LOG(ERR, PMD, "%s: inconsistent (%u, %u)\n", __func__, used_idx , desc_idx); + break; + } + len[i] = uep->len; + rx_pkts[i] = cookie; + vq->vq_used_cons_idx++; + vq_ring_free_chain(vq, desc_idx); + vq->vq_descx[desc_idx].cookie = NULL; + } + return i; +} + +#endif /* _VIRTQUEUE_H_ */ |