diff options
Diffstat (limited to 'arch/mips/sibyte/common')
| -rw-r--r-- | arch/mips/sibyte/common/Makefile | 6 | ||||
| -rw-r--r-- | arch/mips/sibyte/common/bus_watcher.c | 228 | ||||
| -rw-r--r-- | arch/mips/sibyte/common/cfe.c | 330 | ||||
| -rw-r--r-- | arch/mips/sibyte/common/cfe_console.c | 81 | ||||
| -rw-r--r-- | arch/mips/sibyte/common/dma.c | 14 | ||||
| -rw-r--r-- | arch/mips/sibyte/common/sb_tbprof.c | 595 |
6 files changed, 1254 insertions, 0 deletions
diff --git a/arch/mips/sibyte/common/Makefile b/arch/mips/sibyte/common/Makefile new file mode 100644 index 000000000..57f670aa1 --- /dev/null +++ b/arch/mips/sibyte/common/Makefile @@ -0,0 +1,6 @@ +# SPDX-License-Identifier: GPL-2.0-only +obj-y := cfe.o +obj-$(CONFIG_SWIOTLB) += dma.o +obj-$(CONFIG_SIBYTE_BUS_WATCHER) += bus_watcher.o +obj-$(CONFIG_SIBYTE_CFE_CONSOLE) += cfe_console.o +obj-$(CONFIG_SIBYTE_TBPROF) += sb_tbprof.o diff --git a/arch/mips/sibyte/common/bus_watcher.c b/arch/mips/sibyte/common/bus_watcher.c new file mode 100644 index 000000000..d43291473 --- /dev/null +++ b/arch/mips/sibyte/common/bus_watcher.c @@ -0,0 +1,228 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * Copyright (C) 2002,2003 Broadcom Corporation + */ + +/* + * The Bus Watcher monitors internal bus transactions and maintains + * counts of transactions with error status, logging details and + * causing one of several interrupts. This driver provides a handler + * for those interrupts which aggregates the counts (to avoid + * saturating the 8-bit counters) and provides a presence in + * /proc/bus_watcher if PROC_FS is on. + */ + +#include <linux/init.h> +#include <linux/kernel.h> +#include <linux/interrupt.h> +#include <linux/sched.h> +#include <linux/proc_fs.h> +#include <linux/seq_file.h> +#include <asm/io.h> + +#include <asm/sibyte/sb1250.h> +#include <asm/sibyte/sb1250_regs.h> +#include <asm/sibyte/sb1250_int.h> +#include <asm/sibyte/sb1250_scd.h> +#if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80) +#include <asm/sibyte/bcm1480_regs.h> +#endif + + +struct bw_stats_struct { + uint64_t status; + uint32_t l2_err; + uint32_t memio_err; + int status_printed; + unsigned long l2_cor_d; + unsigned long l2_bad_d; + unsigned long l2_cor_t; + unsigned long l2_bad_t; + unsigned long mem_cor_d; + unsigned long mem_bad_d; + unsigned long bus_error; +} bw_stats; + + +static void print_summary(uint32_t status, uint32_t l2_err, + uint32_t memio_err) +{ + printk("Bus watcher error counters: %08x %08x\n", l2_err, memio_err); + printk("\nLast recorded signature:\n"); + printk("Request %02x from %d, answered by %d with Dcode %d\n", + (unsigned int)(G_SCD_BERR_TID(status) & 0x3f), + (int)(G_SCD_BERR_TID(status) >> 6), + (int)G_SCD_BERR_RID(status), + (int)G_SCD_BERR_DCODE(status)); +} + +/* + * check_bus_watcher is exported for use in situations where we want + * to see the most recent status of the bus watcher, which might have + * already been destructively read out of the registers. + * + * notes: this is currently used by the cache error handler + * should provide locking against the interrupt handler + */ +void check_bus_watcher(void) +{ + u32 status, l2_err, memio_err; + +#if defined(CONFIG_SIBYTE_BCM112X) || defined(CONFIG_SIBYTE_SB1250) + /* Use non-destructive register */ + status = csr_in32(IOADDR(A_SCD_BUS_ERR_STATUS_DEBUG)); +#elif defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80) + /* Use non-destructive register */ + /* Same as 1250 except BUS_ERR_STATUS_DEBUG is in a different place. */ + status = csr_in32(IOADDR(A_BCM1480_BUS_ERR_STATUS_DEBUG)); +#else +#error bus watcher being built for unknown Sibyte SOC! +#endif + if (!(status & 0x7fffffff)) { + printk("Using last values reaped by bus watcher driver\n"); + status = bw_stats.status; + l2_err = bw_stats.l2_err; + memio_err = bw_stats.memio_err; + } else { + l2_err = csr_in32(IOADDR(A_BUS_L2_ERRORS)); + memio_err = csr_in32(IOADDR(A_BUS_MEM_IO_ERRORS)); + } + if (status & ~(1UL << 31)) + print_summary(status, l2_err, memio_err); + else + printk("Bus watcher indicates no error\n"); +} + +#ifdef CONFIG_PROC_FS + +/* For simplicity, I want to assume a single read is required each + time */ +static int bw_proc_show(struct seq_file *m, void *v) +{ + struct bw_stats_struct *stats = m->private; + + seq_puts(m, "SiByte Bus Watcher statistics\n"); + seq_puts(m, "-----------------------------\n"); + seq_printf(m, "L2-d-cor %8ld\nL2-d-bad %8ld\n", + stats->l2_cor_d, stats->l2_bad_d); + seq_printf(m, "L2-t-cor %8ld\nL2-t-bad %8ld\n", + stats->l2_cor_t, stats->l2_bad_t); + seq_printf(m, "MC-d-cor %8ld\nMC-d-bad %8ld\n", + stats->mem_cor_d, stats->mem_bad_d); + seq_printf(m, "IO-err %8ld\n", stats->bus_error); + seq_puts(m, "\nLast recorded signature:\n"); + seq_printf(m, "Request %02x from %d, answered by %d with Dcode %d\n", + (unsigned int)(G_SCD_BERR_TID(stats->status) & 0x3f), + (int)(G_SCD_BERR_TID(stats->status) >> 6), + (int)G_SCD_BERR_RID(stats->status), + (int)G_SCD_BERR_DCODE(stats->status)); + /* XXXKW indicate multiple errors between printings, or stats + collection (or both)? */ + if (stats->status & M_SCD_BERR_MULTERRS) + seq_puts(m, "Multiple errors observed since last check.\n"); + if (stats->status_printed) { + seq_puts(m, "(no change since last printing)\n"); + } else { + stats->status_printed = 1; + } + + return 0; +} + +static void create_proc_decoder(struct bw_stats_struct *stats) +{ + struct proc_dir_entry *ent; + + ent = proc_create_single_data("bus_watcher", S_IWUSR | S_IRUGO, NULL, + bw_proc_show, stats); + if (!ent) { + printk(KERN_INFO "Unable to initialize bus_watcher /proc entry\n"); + return; + } +} + +#endif /* CONFIG_PROC_FS */ + +/* + * sibyte_bw_int - handle bus watcher interrupts and accumulate counts + * + * notes: possible re-entry due to multiple sources + * should check/indicate saturation + */ +static irqreturn_t sibyte_bw_int(int irq, void *data) +{ + struct bw_stats_struct *stats = data; + unsigned long cntr; +#ifdef CONFIG_SIBYTE_BW_TRACE + int i; +#endif + +#ifdef CONFIG_SIBYTE_BW_TRACE + csr_out32(M_SCD_TRACE_CFG_FREEZE, IOADDR(A_SCD_TRACE_CFG)); + csr_out32(M_SCD_TRACE_CFG_START_READ, IOADDR(A_SCD_TRACE_CFG)); + + for (i=0; i<256*6; i++) + printk("%016llx\n", + (long long)__raw_readq(IOADDR(A_SCD_TRACE_READ))); + + csr_out32(M_SCD_TRACE_CFG_RESET, IOADDR(A_SCD_TRACE_CFG)); + csr_out32(M_SCD_TRACE_CFG_START, IOADDR(A_SCD_TRACE_CFG)); +#endif + + /* Destructive read, clears register and interrupt */ + stats->status = csr_in32(IOADDR(A_SCD_BUS_ERR_STATUS)); + stats->status_printed = 0; + + stats->l2_err = cntr = csr_in32(IOADDR(A_BUS_L2_ERRORS)); + stats->l2_cor_d += G_SCD_L2ECC_CORR_D(cntr); + stats->l2_bad_d += G_SCD_L2ECC_BAD_D(cntr); + stats->l2_cor_t += G_SCD_L2ECC_CORR_T(cntr); + stats->l2_bad_t += G_SCD_L2ECC_BAD_T(cntr); + csr_out32(0, IOADDR(A_BUS_L2_ERRORS)); + + stats->memio_err = cntr = csr_in32(IOADDR(A_BUS_MEM_IO_ERRORS)); + stats->mem_cor_d += G_SCD_MEM_ECC_CORR(cntr); + stats->mem_bad_d += G_SCD_MEM_ECC_BAD(cntr); + stats->bus_error += G_SCD_MEM_BUSERR(cntr); + csr_out32(0, IOADDR(A_BUS_MEM_IO_ERRORS)); + + return IRQ_HANDLED; +} + +int __init sibyte_bus_watcher(void) +{ + memset(&bw_stats, 0, sizeof(struct bw_stats_struct)); + bw_stats.status_printed = 1; + + if (request_irq(K_INT_BAD_ECC, sibyte_bw_int, 0, "Bus watcher", &bw_stats)) { + printk("Failed to register bus watcher BAD_ECC irq\n"); + return -1; + } + if (request_irq(K_INT_COR_ECC, sibyte_bw_int, 0, "Bus watcher", &bw_stats)) { + free_irq(K_INT_BAD_ECC, &bw_stats); + printk("Failed to register bus watcher COR_ECC irq\n"); + return -1; + } + if (request_irq(K_INT_IO_BUS, sibyte_bw_int, 0, "Bus watcher", &bw_stats)) { + free_irq(K_INT_BAD_ECC, &bw_stats); + free_irq(K_INT_COR_ECC, &bw_stats); + printk("Failed to register bus watcher IO_BUS irq\n"); + return -1; + } + +#ifdef CONFIG_PROC_FS + create_proc_decoder(&bw_stats); +#endif + +#ifdef CONFIG_SIBYTE_BW_TRACE + csr_out32((M_SCD_TRSEQ_ASAMPLE | M_SCD_TRSEQ_DSAMPLE | + K_SCD_TRSEQ_TRIGGER_ALL), + IOADDR(A_SCD_TRACE_SEQUENCE_0)); + csr_out32(M_SCD_TRACE_CFG_RESET, IOADDR(A_SCD_TRACE_CFG)); + csr_out32(M_SCD_TRACE_CFG_START, IOADDR(A_SCD_TRACE_CFG)); +#endif + + return 0; +} + +device_initcall(sibyte_bus_watcher); diff --git a/arch/mips/sibyte/common/cfe.c b/arch/mips/sibyte/common/cfe.c new file mode 100644 index 000000000..89f7fca45 --- /dev/null +++ b/arch/mips/sibyte/common/cfe.c @@ -0,0 +1,330 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * Copyright (C) 2000, 2001, 2002, 2003 Broadcom Corporation + */ + +#include <linux/init.h> +#include <linux/kernel.h> +#include <linux/linkage.h> +#include <linux/mm.h> +#include <linux/blkdev.h> +#include <linux/memblock.h> +#include <linux/pm.h> +#include <linux/smp.h> + +#include <asm/bootinfo.h> +#include <asm/reboot.h> +#include <asm/setup.h> +#include <asm/sibyte/board.h> +#include <asm/smp-ops.h> + +#include <asm/fw/cfe/cfe_api.h> +#include <asm/fw/cfe/cfe_error.h> + +/* Max ram addressable in 32-bit segments */ +#ifdef CONFIG_64BIT +#define MAX_RAM_SIZE (~0ULL) +#else +#ifdef CONFIG_HIGHMEM +#ifdef CONFIG_PHYS_ADDR_T_64BIT +#define MAX_RAM_SIZE (~0ULL) +#else +#define MAX_RAM_SIZE (0xffffffffULL) +#endif +#else +#define MAX_RAM_SIZE (0x1fffffffULL) +#endif +#endif + +#define SIBYTE_MAX_MEM_REGIONS 8 +phys_addr_t board_mem_region_addrs[SIBYTE_MAX_MEM_REGIONS]; +phys_addr_t board_mem_region_sizes[SIBYTE_MAX_MEM_REGIONS]; +unsigned int board_mem_region_count; + +int cfe_cons_handle; + +#ifdef CONFIG_BLK_DEV_INITRD +extern unsigned long initrd_start, initrd_end; +#endif + +static void __noreturn cfe_linux_exit(void *arg) +{ + int warm = *(int *)arg; + + if (smp_processor_id()) { + static int reboot_smp; + + /* Don't repeat the process from another CPU */ + if (!reboot_smp) { + /* Get CPU 0 to do the cfe_exit */ + reboot_smp = 1; + smp_call_function(cfe_linux_exit, arg, 0); + } + } else { + printk("Passing control back to CFE...\n"); + cfe_exit(warm, 0); + printk("cfe_exit returned??\n"); + } + while (1); +} + +static void __noreturn cfe_linux_restart(char *command) +{ + static const int zero; + + cfe_linux_exit((void *)&zero); +} + +static void __noreturn cfe_linux_halt(void) +{ + static const int one = 1; + + cfe_linux_exit((void *)&one); +} + +static __init void prom_meminit(void) +{ + u64 addr, size, type; /* regardless of PHYS_ADDR_T_64BIT */ + int mem_flags = 0; + unsigned int idx; + int rd_flag; +#ifdef CONFIG_BLK_DEV_INITRD + unsigned long initrd_pstart; + unsigned long initrd_pend; + + initrd_pstart = CPHYSADDR(initrd_start); + initrd_pend = CPHYSADDR(initrd_end); + if (initrd_start && + ((initrd_pstart > MAX_RAM_SIZE) + || (initrd_pend > MAX_RAM_SIZE))) { + panic("initrd out of addressable memory"); + } + +#endif /* INITRD */ + + for (idx = 0; cfe_enummem(idx, mem_flags, &addr, &size, &type) != CFE_ERR_NOMORE; + idx++) { + rd_flag = 0; + if (type == CFE_MI_AVAILABLE) { + /* + * See if this block contains (any portion of) the + * ramdisk + */ +#ifdef CONFIG_BLK_DEV_INITRD + if (initrd_start) { + if ((initrd_pstart > addr) && + (initrd_pstart < (addr + size))) { + memblock_add(addr, + initrd_pstart - addr); + rd_flag = 1; + } + if ((initrd_pend > addr) && + (initrd_pend < (addr + size))) { + memblock_add(initrd_pend, + (addr + size) - initrd_pend); + rd_flag = 1; + } + } +#endif + if (!rd_flag) { + if (addr > MAX_RAM_SIZE) + continue; + if (addr+size > MAX_RAM_SIZE) + size = MAX_RAM_SIZE - (addr+size) + 1; + /* + * memcpy/__copy_user prefetch, which + * will cause a bus error for + * KSEG/KUSEG addrs not backed by RAM. + * Hence, reserve some padding for the + * prefetch distance. + */ + if (size > 512) + size -= 512; + memblock_add(addr, size); + } + board_mem_region_addrs[board_mem_region_count] = addr; + board_mem_region_sizes[board_mem_region_count] = size; + board_mem_region_count++; + if (board_mem_region_count == + SIBYTE_MAX_MEM_REGIONS) { + /* + * Too many regions. Need to configure more + */ + while(1); + } + } + } +#ifdef CONFIG_BLK_DEV_INITRD + if (initrd_start) { + memblock_add(initrd_pstart, initrd_pend - initrd_pstart); + memblock_reserve(initrd_pstart, initrd_pend - initrd_pstart); + } +#endif +} + +#ifdef CONFIG_BLK_DEV_INITRD +static int __init initrd_setup(char *str) +{ + char rdarg[64]; + int idx; + char *tmp, *endptr; + unsigned long initrd_size; + + /* Make a copy of the initrd argument so we can smash it up here */ + for (idx = 0; idx < sizeof(rdarg)-1; idx++) { + if (!str[idx] || (str[idx] == ' ')) break; + rdarg[idx] = str[idx]; + } + + rdarg[idx] = 0; + str = rdarg; + + /* + *Initrd location comes in the form "<hex size of ramdisk in bytes>@<location in memory>" + * e.g. initrd=3abfd@80010000. This is set up by the loader. + */ + for (tmp = str; *tmp != '@'; tmp++) { + if (!*tmp) { + goto fail; + } + } + *tmp = 0; + tmp++; + if (!*tmp) { + goto fail; + } + initrd_size = simple_strtoul(str, &endptr, 16); + if (*endptr) { + *(tmp-1) = '@'; + goto fail; + } + *(tmp-1) = '@'; + initrd_start = simple_strtoul(tmp, &endptr, 16); + if (*endptr) { + goto fail; + } + initrd_end = initrd_start + initrd_size; + printk("Found initrd of %lx@%lx\n", initrd_size, initrd_start); + return 1; + fail: + printk("Bad initrd argument. Disabling initrd\n"); + initrd_start = 0; + initrd_end = 0; + return 1; +} + +#endif + +extern const struct plat_smp_ops sb_smp_ops; +extern const struct plat_smp_ops bcm1480_smp_ops; + +/* + * prom_init is called just after the cpu type is determined, from setup_arch() + */ +void __init prom_init(void) +{ + uint64_t cfe_ept, cfe_handle; + unsigned int cfe_eptseal; + int argc = fw_arg0; + char **envp = (char **) fw_arg2; + int *prom_vec = (int *) fw_arg3; + + _machine_restart = cfe_linux_restart; + _machine_halt = cfe_linux_halt; + pm_power_off = cfe_linux_halt; + + /* + * Check if a loader was used; if NOT, the 4 arguments are + * what CFE gives us (handle, 0, EPT and EPTSEAL) + */ + if (argc < 0) { + cfe_handle = (uint64_t)(long)argc; + cfe_ept = (long)envp; + cfe_eptseal = (uint32_t)(unsigned long)prom_vec; + } else { + if ((int32_t)(long)prom_vec < 0) { + /* + * Old loader; all it gives us is the handle, + * so use the "known" entrypoint and assume + * the seal. + */ + cfe_handle = (uint64_t)(long)prom_vec; + cfe_ept = (uint64_t)((int32_t)0x9fc00500); + cfe_eptseal = CFE_EPTSEAL; + } else { + /* + * Newer loaders bundle the handle/ept/eptseal + * Note: prom_vec is in the loader's useg + * which is still alive in the TLB. + */ + cfe_handle = (uint64_t)((int32_t *)prom_vec)[0]; + cfe_ept = (uint64_t)((int32_t *)prom_vec)[2]; + cfe_eptseal = (unsigned int)((uint32_t *)prom_vec)[3]; + } + } + if (cfe_eptseal != CFE_EPTSEAL) { + /* too early for panic to do any good */ + printk("CFE's entrypoint seal doesn't match. Spinning."); + while (1) ; + } + cfe_init(cfe_handle, cfe_ept); + /* + * Get the handle for (at least) prom_putchar, possibly for + * boot console + */ + cfe_cons_handle = cfe_getstdhandle(CFE_STDHANDLE_CONSOLE); + if (cfe_getenv("LINUX_CMDLINE", arcs_cmdline, COMMAND_LINE_SIZE) < 0) { + if (argc >= 0) { + /* The loader should have set the command line */ + /* too early for panic to do any good */ + printk("LINUX_CMDLINE not defined in cfe."); + while (1) ; + } + } + +#ifdef CONFIG_BLK_DEV_INITRD + { + char *ptr; + /* Need to find out early whether we've got an initrd. So scan + the list looking now */ + for (ptr = arcs_cmdline; *ptr; ptr++) { + while (*ptr == ' ') { + ptr++; + } + if (!strncmp(ptr, "initrd=", 7)) { + initrd_setup(ptr+7); + break; + } else { + while (*ptr && (*ptr != ' ')) { + ptr++; + } + } + } + } +#endif /* CONFIG_BLK_DEV_INITRD */ + + /* Not sure this is needed, but it's the safe way. */ + arcs_cmdline[COMMAND_LINE_SIZE-1] = 0; + + prom_meminit(); + +#if defined(CONFIG_SIBYTE_BCM112X) || defined(CONFIG_SIBYTE_SB1250) + register_smp_ops(&sb_smp_ops); +#endif +#if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80) + register_smp_ops(&bcm1480_smp_ops); +#endif +} + +void __init prom_free_prom_memory(void) +{ + /* Not sure what I'm supposed to do here. Nothing, I think */ +} + +void prom_putchar(char c) +{ + int ret; + + while ((ret = cfe_write(cfe_cons_handle, &c, 1)) == 0) + ; +} diff --git a/arch/mips/sibyte/common/cfe_console.c b/arch/mips/sibyte/common/cfe_console.c new file mode 100644 index 000000000..8af7b41f7 --- /dev/null +++ b/arch/mips/sibyte/common/cfe_console.c @@ -0,0 +1,81 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <linux/init.h> +#include <linux/errno.h> +#include <linux/console.h> + +#include <asm/sibyte/board.h> + +#include <asm/fw/cfe/cfe_api.h> +#include <asm/fw/cfe/cfe_error.h> + +extern int cfe_cons_handle; + +static void cfe_console_write(struct console *cons, const char *str, + unsigned int count) +{ + int i, last, written; + + for (i=0, last=0; i<count; i++) { + if (!str[i]) + /* XXXKW can/should this ever happen? */ + return; + if (str[i] == '\n') { + do { + written = cfe_write(cfe_cons_handle, &str[last], i-last); + if (written < 0) + ; + last += written; + } while (last < i); + while (cfe_write(cfe_cons_handle, "\r", 1) <= 0) + ; + } + } + if (last != count) { + do { + written = cfe_write(cfe_cons_handle, &str[last], count-last); + if (written < 0) + ; + last += written; + } while (last < count); + } + +} + +static int cfe_console_setup(struct console *cons, char *str) +{ + char consdev[32]; + /* XXXKW think about interaction with 'console=' cmdline arg */ + /* If none of the console options are configured, the build will break. */ + if (cfe_getenv("BOOT_CONSOLE", consdev, 32) >= 0) { +#ifdef CONFIG_SERIAL_SB1250_DUART + if (!strcmp(consdev, "uart0")) { + setleds("u0cn"); + } else if (!strcmp(consdev, "uart1")) { + setleds("u1cn"); + } else +#endif +#ifdef CONFIG_VGA_CONSOLE + if (!strcmp(consdev, "pcconsole0")) { + setleds("pccn"); + } else +#endif + return -ENODEV; + } + return 0; +} + +static struct console sb1250_cfe_cons = { + .name = "cfe", + .write = cfe_console_write, + .setup = cfe_console_setup, + .flags = CON_PRINTBUFFER, + .index = -1, +}; + +static int __init sb1250_cfe_console_init(void) +{ + register_console(&sb1250_cfe_cons); + return 0; +} + +console_initcall(sb1250_cfe_console_init); diff --git a/arch/mips/sibyte/common/dma.c b/arch/mips/sibyte/common/dma.c new file mode 100644 index 000000000..eb47a94f3 --- /dev/null +++ b/arch/mips/sibyte/common/dma.c @@ -0,0 +1,14 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* + * DMA support for Broadcom SiByte platforms. + * + * Copyright (c) 2018 Maciej W. Rozycki + */ + +#include <linux/swiotlb.h> +#include <asm/bootinfo.h> + +void __init plat_swiotlb_setup(void) +{ + swiotlb_init(1); +} diff --git a/arch/mips/sibyte/common/sb_tbprof.c b/arch/mips/sibyte/common/sb_tbprof.c new file mode 100644 index 000000000..f80d7a710 --- /dev/null +++ b/arch/mips/sibyte/common/sb_tbprof.c @@ -0,0 +1,595 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * + * Copyright (C) 2001, 2002, 2003 Broadcom Corporation + * Copyright (C) 2007 Ralf Baechle <ralf@linux-mips.org> + * Copyright (C) 2007 MIPS Technologies, Inc. + * written by Ralf Baechle <ralf@linux-mips.org> + */ + +#undef DEBUG + +#include <linux/device.h> +#include <linux/module.h> +#include <linux/kernel.h> +#include <linux/types.h> +#include <linux/init.h> +#include <linux/interrupt.h> +#include <linux/sched.h> +#include <linux/vmalloc.h> +#include <linux/fs.h> +#include <linux/errno.h> +#include <linux/wait.h> +#include <asm/io.h> +#include <asm/sibyte/sb1250.h> + +#if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80) +#include <asm/sibyte/bcm1480_regs.h> +#include <asm/sibyte/bcm1480_scd.h> +#include <asm/sibyte/bcm1480_int.h> +#elif defined(CONFIG_SIBYTE_SB1250) || defined(CONFIG_SIBYTE_BCM112X) +#include <asm/sibyte/sb1250_regs.h> +#include <asm/sibyte/sb1250_scd.h> +#include <asm/sibyte/sb1250_int.h> +#else +#error invalid SiByte UART configuration +#endif + +#if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80) +#undef K_INT_TRACE_FREEZE +#define K_INT_TRACE_FREEZE K_BCM1480_INT_TRACE_FREEZE +#undef K_INT_PERF_CNT +#define K_INT_PERF_CNT K_BCM1480_INT_PERF_CNT +#endif + +#include <linux/uaccess.h> + +#define SBPROF_TB_MAJOR 240 + +typedef u64 tb_sample_t[6*256]; + +enum open_status { + SB_CLOSED, + SB_OPENING, + SB_OPEN +}; + +struct sbprof_tb { + wait_queue_head_t tb_sync; + wait_queue_head_t tb_read; + struct mutex lock; + enum open_status open; + tb_sample_t *sbprof_tbbuf; + int next_tb_sample; + + volatile int tb_enable; + volatile int tb_armed; + +}; + +static struct sbprof_tb sbp; + +#define MAX_SAMPLE_BYTES (24*1024*1024) +#define MAX_TBSAMPLE_BYTES (12*1024*1024) + +#define MAX_SAMPLES (MAX_SAMPLE_BYTES/sizeof(u_int32_t)) +#define TB_SAMPLE_SIZE (sizeof(tb_sample_t)) +#define MAX_TB_SAMPLES (MAX_TBSAMPLE_BYTES/TB_SAMPLE_SIZE) + +/* ioctls */ +#define SBPROF_ZBSTART _IOW('s', 0, int) +#define SBPROF_ZBSTOP _IOW('s', 1, int) +#define SBPROF_ZBWAITFULL _IOW('s', 2, int) + +/* + * Routines for using 40-bit SCD cycle counter + * + * Client responsible for either handling interrupts or making sure + * the cycles counter never saturates, e.g., by doing + * zclk_timer_init(0) at least every 2^40 - 1 ZCLKs. + */ + +/* + * Configures SCD counter 0 to count ZCLKs starting from val; + * Configures SCD counters1,2,3 to count nothing. + * Must not be called while gathering ZBbus profiles. + */ + +#define zclk_timer_init(val) \ + __asm__ __volatile__ (".set push;" \ + ".set mips64;" \ + "la $8, 0xb00204c0;" /* SCD perf_cnt_cfg */ \ + "sd %0, 0x10($8);" /* write val to counter0 */ \ + "sd %1, 0($8);" /* config counter0 for zclks*/ \ + ".set pop" \ + : /* no outputs */ \ + /* enable, counter0 */ \ + : /* inputs */ "r"(val), "r" ((1ULL << 33) | 1ULL) \ + : /* modifies */ "$8" ) + + +/* Reads SCD counter 0 and puts result in value + unsigned long long val; */ +#define zclk_get(val) \ + __asm__ __volatile__ (".set push;" \ + ".set mips64;" \ + "la $8, 0xb00204c0;" /* SCD perf_cnt_cfg */ \ + "ld %0, 0x10($8);" /* write val to counter0 */ \ + ".set pop" \ + : /* outputs */ "=r"(val) \ + : /* inputs */ \ + : /* modifies */ "$8" ) + +#define DEVNAME "sb_tbprof" + +#define TB_FULL (sbp.next_tb_sample == MAX_TB_SAMPLES) + +/* + * Support for ZBbus sampling using the trace buffer + * + * We use the SCD performance counter interrupt, caused by a Zclk counter + * overflow, to trigger the start of tracing. + * + * We set the trace buffer to sample everything and freeze on + * overflow. + * + * We map the interrupt for trace_buffer_freeze to handle it on CPU 0. + * + */ + +static u64 tb_period; + +static void arm_tb(void) +{ + u64 scdperfcnt; + u64 next = (1ULL << 40) - tb_period; + u64 tb_options = M_SCD_TRACE_CFG_FREEZE_FULL; + + /* + * Generate an SCD_PERFCNT interrupt in TB_PERIOD Zclks to + * trigger start of trace. XXX vary sampling period + */ + __raw_writeq(0, IOADDR(A_SCD_PERF_CNT_1)); + scdperfcnt = __raw_readq(IOADDR(A_SCD_PERF_CNT_CFG)); + + /* + * Unfortunately, in Pass 2 we must clear all counters to knock down + * a previous interrupt request. This means that bus profiling + * requires ALL of the SCD perf counters. + */ +#if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80) + __raw_writeq((scdperfcnt & ~M_SPC_CFG_SRC1) | + /* keep counters 0,2,3,4,5,6,7 as is */ + V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */ + IOADDR(A_BCM1480_SCD_PERF_CNT_CFG0)); + __raw_writeq( + M_SPC_CFG_ENABLE | /* enable counting */ + M_SPC_CFG_CLEAR | /* clear all counters */ + V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */ + IOADDR(A_BCM1480_SCD_PERF_CNT_CFG1)); +#else + __raw_writeq((scdperfcnt & ~M_SPC_CFG_SRC1) | + /* keep counters 0,2,3 as is */ + M_SPC_CFG_ENABLE | /* enable counting */ + M_SPC_CFG_CLEAR | /* clear all counters */ + V_SPC_CFG_SRC1(1), /* counter 1 counts cycles */ + IOADDR(A_SCD_PERF_CNT_CFG)); +#endif + __raw_writeq(next, IOADDR(A_SCD_PERF_CNT_1)); + /* Reset the trace buffer */ + __raw_writeq(M_SCD_TRACE_CFG_RESET, IOADDR(A_SCD_TRACE_CFG)); +#if 0 && defined(M_SCD_TRACE_CFG_FORCECNT) + /* XXXKW may want to expose control to the data-collector */ + tb_options |= M_SCD_TRACE_CFG_FORCECNT; +#endif + __raw_writeq(tb_options, IOADDR(A_SCD_TRACE_CFG)); + sbp.tb_armed = 1; +} + +static irqreturn_t sbprof_tb_intr(int irq, void *dev_id) +{ + int i; + + pr_debug(DEVNAME ": tb_intr\n"); + + if (sbp.next_tb_sample < MAX_TB_SAMPLES) { + /* XXX should use XKPHYS to make writes bypass L2 */ + u64 *p = sbp.sbprof_tbbuf[sbp.next_tb_sample++]; + /* Read out trace */ + __raw_writeq(M_SCD_TRACE_CFG_START_READ, + IOADDR(A_SCD_TRACE_CFG)); + __asm__ __volatile__ ("sync" : : : "memory"); + /* Loop runs backwards because bundles are read out in reverse order */ + for (i = 256 * 6; i > 0; i -= 6) { + /* Subscripts decrease to put bundle in the order */ + /* t0 lo, t0 hi, t1 lo, t1 hi, t2 lo, t2 hi */ + p[i - 1] = __raw_readq(IOADDR(A_SCD_TRACE_READ)); + /* read t2 hi */ + p[i - 2] = __raw_readq(IOADDR(A_SCD_TRACE_READ)); + /* read t2 lo */ + p[i - 3] = __raw_readq(IOADDR(A_SCD_TRACE_READ)); + /* read t1 hi */ + p[i - 4] = __raw_readq(IOADDR(A_SCD_TRACE_READ)); + /* read t1 lo */ + p[i - 5] = __raw_readq(IOADDR(A_SCD_TRACE_READ)); + /* read t0 hi */ + p[i - 6] = __raw_readq(IOADDR(A_SCD_TRACE_READ)); + /* read t0 lo */ + } + if (!sbp.tb_enable) { + pr_debug(DEVNAME ": tb_intr shutdown\n"); + __raw_writeq(M_SCD_TRACE_CFG_RESET, + IOADDR(A_SCD_TRACE_CFG)); + sbp.tb_armed = 0; + wake_up_interruptible(&sbp.tb_sync); + } else { + /* knock down current interrupt and get another one later */ + arm_tb(); + } + } else { + /* No more trace buffer samples */ + pr_debug(DEVNAME ": tb_intr full\n"); + __raw_writeq(M_SCD_TRACE_CFG_RESET, IOADDR(A_SCD_TRACE_CFG)); + sbp.tb_armed = 0; + if (!sbp.tb_enable) + wake_up_interruptible(&sbp.tb_sync); + wake_up_interruptible(&sbp.tb_read); + } + return IRQ_HANDLED; +} + +static irqreturn_t sbprof_pc_intr(int irq, void *dev_id) +{ + printk(DEVNAME ": unexpected pc_intr"); + return IRQ_NONE; +} + +/* + * Requires: Already called zclk_timer_init with a value that won't + * saturate 40 bits. No subsequent use of SCD performance counters + * or trace buffer. + */ + +static int sbprof_zbprof_start(struct file *filp) +{ + u64 scdperfcnt; + int err; + + if (xchg(&sbp.tb_enable, 1)) + return -EBUSY; + + pr_debug(DEVNAME ": starting\n"); + + sbp.next_tb_sample = 0; + filp->f_pos = 0; + + err = request_irq(K_INT_TRACE_FREEZE, sbprof_tb_intr, 0, + DEVNAME " trace freeze", &sbp); + if (err) + return -EBUSY; + + /* Make sure there isn't a perf-cnt interrupt waiting */ + scdperfcnt = __raw_readq(IOADDR(A_SCD_PERF_CNT_CFG)); + /* Disable and clear counters, override SRC_1 */ + __raw_writeq((scdperfcnt & ~(M_SPC_CFG_SRC1 | M_SPC_CFG_ENABLE)) | + M_SPC_CFG_ENABLE | M_SPC_CFG_CLEAR | V_SPC_CFG_SRC1(1), + IOADDR(A_SCD_PERF_CNT_CFG)); + + /* + * We grab this interrupt to prevent others from trying to use + * it, even though we don't want to service the interrupts + * (they only feed into the trace-on-interrupt mechanism) + */ + if (request_irq(K_INT_PERF_CNT, sbprof_pc_intr, 0, DEVNAME " scd perfcnt", &sbp)) { + free_irq(K_INT_TRACE_FREEZE, &sbp); + return -EBUSY; + } + + /* + * I need the core to mask these, but the interrupt mapper to + * pass them through. I am exploiting my knowledge that + * cp0_status masks out IP[5]. krw + */ +#if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80) + __raw_writeq(K_BCM1480_INT_MAP_I3, + IOADDR(A_BCM1480_IMR_REGISTER(0, R_BCM1480_IMR_INTERRUPT_MAP_BASE_L) + + ((K_BCM1480_INT_PERF_CNT & 0x3f) << 3))); +#else + __raw_writeq(K_INT_MAP_I3, + IOADDR(A_IMR_REGISTER(0, R_IMR_INTERRUPT_MAP_BASE) + + (K_INT_PERF_CNT << 3))); +#endif + + /* Initialize address traps */ + __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_0)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_1)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_2)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_UP_3)); + + __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_0)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_1)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_2)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_DOWN_3)); + + __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_0)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_1)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_2)); + __raw_writeq(0, IOADDR(A_ADDR_TRAP_CFG_3)); + + /* Initialize Trace Event 0-7 */ + /* when interrupt */ + __raw_writeq(M_SCD_TREVT_INTERRUPT, IOADDR(A_SCD_TRACE_EVENT_0)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_1)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_2)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_3)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_4)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_5)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_6)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_EVENT_7)); + + /* Initialize Trace Sequence 0-7 */ + /* Start on event 0 (interrupt) */ + __raw_writeq(V_SCD_TRSEQ_FUNC_START | 0x0fff, + IOADDR(A_SCD_TRACE_SEQUENCE_0)); + /* dsamp when d used | asamp when a used */ + __raw_writeq(M_SCD_TRSEQ_ASAMPLE | M_SCD_TRSEQ_DSAMPLE | + K_SCD_TRSEQ_TRIGGER_ALL, + IOADDR(A_SCD_TRACE_SEQUENCE_1)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_2)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_3)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_4)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_5)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_6)); + __raw_writeq(0, IOADDR(A_SCD_TRACE_SEQUENCE_7)); + + /* Now indicate the PERF_CNT interrupt as a trace-relevant interrupt */ +#if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80) + __raw_writeq(1ULL << (K_BCM1480_INT_PERF_CNT & 0x3f), + IOADDR(A_BCM1480_IMR_REGISTER(0, R_BCM1480_IMR_INTERRUPT_TRACE_L))); +#else + __raw_writeq(1ULL << K_INT_PERF_CNT, + IOADDR(A_IMR_REGISTER(0, R_IMR_INTERRUPT_TRACE))); +#endif + arm_tb(); + + pr_debug(DEVNAME ": done starting\n"); + + return 0; +} + +static int sbprof_zbprof_stop(void) +{ + int err = 0; + + pr_debug(DEVNAME ": stopping\n"); + + if (sbp.tb_enable) { + /* + * XXXKW there is a window here where the intr handler may run, + * see the disable, and do the wake_up before this sleep + * happens. + */ + pr_debug(DEVNAME ": wait for disarm\n"); + err = wait_event_interruptible(sbp.tb_sync, !sbp.tb_armed); + pr_debug(DEVNAME ": disarm complete, stat %d\n", err); + + if (err) + return err; + + sbp.tb_enable = 0; + free_irq(K_INT_TRACE_FREEZE, &sbp); + free_irq(K_INT_PERF_CNT, &sbp); + } + + pr_debug(DEVNAME ": done stopping\n"); + + return err; +} + +static int sbprof_tb_open(struct inode *inode, struct file *filp) +{ + int minor; + + minor = iminor(inode); + if (minor != 0) + return -ENODEV; + + if (xchg(&sbp.open, SB_OPENING) != SB_CLOSED) + return -EBUSY; + + memset(&sbp, 0, sizeof(struct sbprof_tb)); + sbp.sbprof_tbbuf = vzalloc(MAX_TBSAMPLE_BYTES); + if (!sbp.sbprof_tbbuf) { + sbp.open = SB_CLOSED; + wmb(); + return -ENOMEM; + } + + init_waitqueue_head(&sbp.tb_sync); + init_waitqueue_head(&sbp.tb_read); + mutex_init(&sbp.lock); + + sbp.open = SB_OPEN; + wmb(); + + return 0; +} + +static int sbprof_tb_release(struct inode *inode, struct file *filp) +{ + int minor; + + minor = iminor(inode); + if (minor != 0 || sbp.open != SB_CLOSED) + return -ENODEV; + + mutex_lock(&sbp.lock); + + if (sbp.tb_armed || sbp.tb_enable) + sbprof_zbprof_stop(); + + vfree(sbp.sbprof_tbbuf); + sbp.open = SB_CLOSED; + wmb(); + + mutex_unlock(&sbp.lock); + + return 0; +} + +static ssize_t sbprof_tb_read(struct file *filp, char *buf, + size_t size, loff_t *offp) +{ + int cur_sample, sample_off, cur_count, sample_left; + char *src; + int count = 0; + char *dest = buf; + long cur_off = *offp; + + if (!access_ok(buf, size)) + return -EFAULT; + + mutex_lock(&sbp.lock); + + count = 0; + cur_sample = cur_off / TB_SAMPLE_SIZE; + sample_off = cur_off % TB_SAMPLE_SIZE; + sample_left = TB_SAMPLE_SIZE - sample_off; + + while (size && (cur_sample < sbp.next_tb_sample)) { + int err; + + cur_count = size < sample_left ? size : sample_left; + src = (char *)(((long)sbp.sbprof_tbbuf[cur_sample])+sample_off); + err = __copy_to_user(dest, src, cur_count); + if (err) { + *offp = cur_off + cur_count - err; + mutex_unlock(&sbp.lock); + return err; + } + pr_debug(DEVNAME ": read from sample %d, %d bytes\n", + cur_sample, cur_count); + size -= cur_count; + sample_left -= cur_count; + if (!sample_left) { + cur_sample++; + sample_off = 0; + sample_left = TB_SAMPLE_SIZE; + } else { + sample_off += cur_count; + } + cur_off += cur_count; + dest += cur_count; + count += cur_count; + } + *offp = cur_off; + mutex_unlock(&sbp.lock); + + return count; +} + +static long sbprof_tb_ioctl(struct file *filp, + unsigned int command, + unsigned long arg) +{ + int err = 0; + + switch (command) { + case SBPROF_ZBSTART: + mutex_lock(&sbp.lock); + err = sbprof_zbprof_start(filp); + mutex_unlock(&sbp.lock); + break; + + case SBPROF_ZBSTOP: + mutex_lock(&sbp.lock); + err = sbprof_zbprof_stop(); + mutex_unlock(&sbp.lock); + break; + + case SBPROF_ZBWAITFULL: { + err = wait_event_interruptible(sbp.tb_read, TB_FULL); + if (err) + break; + + err = put_user(TB_FULL, (int *) arg); + break; + } + + default: + err = -EINVAL; + break; + } + + return err; +} + +static const struct file_operations sbprof_tb_fops = { + .owner = THIS_MODULE, + .open = sbprof_tb_open, + .release = sbprof_tb_release, + .read = sbprof_tb_read, + .unlocked_ioctl = sbprof_tb_ioctl, + .compat_ioctl = sbprof_tb_ioctl, + .mmap = NULL, + .llseek = default_llseek, +}; + +static struct class *tb_class; +static struct device *tb_dev; + +static int __init sbprof_tb_init(void) +{ + struct device *dev; + struct class *tbc; + int err; + + if (register_chrdev(SBPROF_TB_MAJOR, DEVNAME, &sbprof_tb_fops)) { + printk(KERN_WARNING DEVNAME ": initialization failed (dev %d)\n", + SBPROF_TB_MAJOR); + return -EIO; + } + + tbc = class_create(THIS_MODULE, "sb_tracebuffer"); + if (IS_ERR(tbc)) { + err = PTR_ERR(tbc); + goto out_chrdev; + } + + tb_class = tbc; + + dev = device_create(tbc, NULL, MKDEV(SBPROF_TB_MAJOR, 0), NULL, "tb"); + if (IS_ERR(dev)) { + err = PTR_ERR(dev); + goto out_class; + } + tb_dev = dev; + + sbp.open = SB_CLOSED; + wmb(); + tb_period = zbbus_mhz * 10000LL; + pr_info(DEVNAME ": initialized - tb_period = %lld\n", + (long long) tb_period); + return 0; + +out_class: + class_destroy(tb_class); +out_chrdev: + unregister_chrdev(SBPROF_TB_MAJOR, DEVNAME); + + return err; +} + +static void __exit sbprof_tb_cleanup(void) +{ + device_destroy(tb_class, MKDEV(SBPROF_TB_MAJOR, 0)); + unregister_chrdev(SBPROF_TB_MAJOR, DEVNAME); + class_destroy(tb_class); +} + +module_init(sbprof_tb_init); +module_exit(sbprof_tb_cleanup); + +MODULE_ALIAS_CHARDEV_MAJOR(SBPROF_TB_MAJOR); +MODULE_AUTHOR("Ralf Baechle <ralf@linux-mips.org>"); +MODULE_LICENSE("GPL"); |