From 8d91c1e411f55d7ea91b1183a2e9f8088fb4d5be Mon Sep 17 00:00:00 2001 From: André Fabian Silva Delgado Date: Tue, 15 Dec 2015 14:52:16 -0300 Subject: Linux-libre 4.3.2-gnu --- arch/xtensa/kernel/Makefile | 10 +- arch/xtensa/kernel/entry.S | 177 +++++++++++----- arch/xtensa/kernel/irq.c | 27 ++- arch/xtensa/kernel/pci-dma.c | 203 +++++++++++++++--- arch/xtensa/kernel/perf_event.c | 454 ++++++++++++++++++++++++++++++++++++++++ arch/xtensa/kernel/stacktrace.c | 167 ++++++++++++++- arch/xtensa/kernel/time.c | 53 +++-- arch/xtensa/kernel/traps.c | 31 ++- arch/xtensa/kernel/vectors.S | 10 +- 9 files changed, 1003 insertions(+), 129 deletions(-) create mode 100644 arch/xtensa/kernel/perf_event.c (limited to 'arch/xtensa/kernel') diff --git a/arch/xtensa/kernel/Makefile b/arch/xtensa/kernel/Makefile index d3a0f0fd5..50137bc9e 100644 --- a/arch/xtensa/kernel/Makefile +++ b/arch/xtensa/kernel/Makefile @@ -13,6 +13,7 @@ obj-$(CONFIG_PCI) += pci.o obj-$(CONFIG_MODULES) += xtensa_ksyms.o module.o obj-$(CONFIG_FUNCTION_TRACER) += mcount.o obj-$(CONFIG_SMP) += smp.o mxhead.o +obj-$(CONFIG_XTENSA_VARIANT_HAVE_PERF_EVENTS) += perf_event.o AFLAGS_head.o += -mtext-section-literals @@ -27,10 +28,11 @@ AFLAGS_head.o += -mtext-section-literals # # Replicate rules in scripts/Makefile.build -sed-y = -e 's/\*(\(\.[a-z]*it\|\.ref\|\)\.text)/*(\1.literal \1.text)/g' \ - -e 's/\.text\.unlikely/.literal.unlikely .text.unlikely/g' \ - -e 's/\*(\(\.text .*\))/*(.literal \1)/g' \ - -e 's/\*(\(\.text\.[a-z]*\))/*(\1.literal \1)/g' +sed-y = -e ':a; s/\*(\([^)]*\)\.text\.unlikely/*(\1.literal.unlikely .{text}.unlikely/; ta; ' \ + -e ':b; s/\*(\([^)]*\)\.text\(\.[a-z]*\)/*(\1.{text}\2.literal .{text}\2/; tb; ' \ + -e ':c; s/\*(\([^)]*\)\(\.[a-z]*it\|\.ref\)\.text/*(\1\2.literal \2.{text}/; tc; ' \ + -e ':d; s/\*(\([^)]\+ \|\)\.text/*(\1.literal .{text}/; td; ' \ + -e 's/\.{text}/.text/g' quiet_cmd__cpp_lds_S = LDS $@ cmd__cpp_lds_S = $(CPP) $(cpp_flags) -P -C -Uxtensa -D__ASSEMBLY__ $< \ diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S index a2a902140..504130357 100644 --- a/arch/xtensa/kernel/entry.S +++ b/arch/xtensa/kernel/entry.S @@ -1,6 +1,4 @@ /* - * arch/xtensa/kernel/entry.S - * * Low-level exception handling * * This file is subject to the terms and conditions of the GNU General Public @@ -8,6 +6,7 @@ * for more details. * * Copyright (C) 2004 - 2008 by Tensilica Inc. + * Copyright (C) 2015 Cadence Design Systems Inc. * * Chris Zankel * @@ -75,6 +74,27 @@ #endif .endm + + .macro irq_save flags tmp +#if XTENSA_FAKE_NMI +#if defined(CONFIG_DEBUG_KERNEL) && (LOCKLEVEL | TOPLEVEL) >= XCHAL_DEBUGLEVEL + rsr \flags, ps + extui \tmp, \flags, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH + bgei \tmp, LOCKLEVEL, 99f + rsil \tmp, LOCKLEVEL +99: +#else + movi \tmp, LOCKLEVEL + rsr \flags, ps + or \flags, \flags, \tmp + xsr \flags, ps + rsync +#endif +#else + rsil \flags, LOCKLEVEL +#endif + .endm + /* ----------------- DEFAULT FIRST LEVEL EXCEPTION HANDLERS ----------------- */ /* @@ -122,6 +142,7 @@ _user_exception: /* Save SAR and turn off single stepping */ movi a2, 0 + wsr a2, depc # terminate user stack trace with 0 rsr a3, sar xsr a2, icountlevel s32i a3, a1, PT_SAR @@ -301,7 +322,18 @@ _kernel_exception: s32i a14, a1, PT_AREG14 s32i a15, a1, PT_AREG15 + _bnei a2, 1, 1f + + /* Copy spill slots of a0 and a1 to imitate movsp + * in order to keep exception stack continuous + */ + l32i a3, a1, PT_SIZE + l32i a0, a1, PT_SIZE + 4 + s32e a3, a1, -16 + s32e a0, a1, -12 1: + l32i a0, a1, PT_AREG0 # restore saved a0 + wsr a0, depc #ifdef KERNEL_STACK_OVERFLOW_CHECK @@ -340,75 +372,88 @@ common_exception: /* It is now save to restore the EXC_TABLE_FIXUP variable. */ - rsr a0, exccause + rsr a2, exccause movi a3, 0 - rsr a2, excsave1 - s32i a0, a1, PT_EXCCAUSE - s32i a3, a2, EXC_TABLE_FIXUP - - /* All unrecoverable states are saved on stack, now, and a1 is valid, - * so we can allow exceptions and interrupts (*) again. - * Set PS(EXCM = 0, UM = 0, RING = 0, OWB = 0, WOE = 1, INTLEVEL = X) + rsr a0, excsave1 + s32i a2, a1, PT_EXCCAUSE + s32i a3, a0, EXC_TABLE_FIXUP + + /* All unrecoverable states are saved on stack, now, and a1 is valid. + * Now we can allow exceptions again. In case we've got an interrupt + * PS.INTLEVEL is set to LOCKLEVEL disabling furhter interrupts, + * otherwise it's left unchanged. * - * (*) We only allow interrupts if they were previously enabled and - * we're not handling an IRQ + * Set PS(EXCM = 0, UM = 0, RING = 0, OWB = 0, WOE = 1, INTLEVEL = X) */ rsr a3, ps - addi a0, a0, -EXCCAUSE_LEVEL1_INTERRUPT - movi a2, LOCKLEVEL + s32i a3, a1, PT_PS # save ps + +#if XTENSA_FAKE_NMI + /* Correct PS needs to be saved in the PT_PS: + * - in case of exception or level-1 interrupt it's in the PS, + * and is already saved. + * - in case of medium level interrupt it's in the excsave2. + */ + movi a0, EXCCAUSE_MAPPED_NMI + extui a3, a3, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH + beq a2, a0, .Lmedium_level_irq + bnei a2, EXCCAUSE_LEVEL1_INTERRUPT, .Lexception + beqz a3, .Llevel1_irq # level-1 IRQ sets ps.intlevel to 0 + +.Lmedium_level_irq: + rsr a0, excsave2 + s32i a0, a1, PT_PS # save medium-level interrupt ps + bgei a3, LOCKLEVEL, .Lexception + +.Llevel1_irq: + movi a3, LOCKLEVEL + +.Lexception: + movi a0, 1 << PS_WOE_BIT + or a3, a3, a0 +#else + addi a2, a2, -EXCCAUSE_LEVEL1_INTERRUPT + movi a0, LOCKLEVEL extui a3, a3, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH # a3 = PS.INTLEVEL - moveqz a3, a2, a0 # a3 = LOCKLEVEL iff interrupt + moveqz a3, a0, a2 # a3 = LOCKLEVEL iff interrupt movi a2, 1 << PS_WOE_BIT or a3, a3, a2 - rsr a0, exccause - xsr a3, ps + rsr a2, exccause +#endif - s32i a3, a1, PT_PS # save ps + /* restore return address (or 0 if return to userspace) */ + rsr a0, depc + wsr a3, ps + rsync # PS.WOE => rsync => overflow /* Save lbeg, lend */ - rsr a2, lbeg + rsr a4, lbeg rsr a3, lend - s32i a2, a1, PT_LBEG + s32i a4, a1, PT_LBEG s32i a3, a1, PT_LEND /* Save SCOMPARE1 */ #if XCHAL_HAVE_S32C1I - rsr a2, scompare1 - s32i a2, a1, PT_SCOMPARE1 + rsr a3, scompare1 + s32i a3, a1, PT_SCOMPARE1 #endif /* Save optional registers. */ - save_xtregs_opt a1 a2 a4 a5 a6 a7 PT_XTREGS_OPT + save_xtregs_opt a1 a3 a4 a5 a6 a7 PT_XTREGS_OPT -#ifdef CONFIG_TRACE_IRQFLAGS - l32i a4, a1, PT_DEPC - /* Double exception means we came here with an exception - * while PS.EXCM was set, i.e. interrupts disabled. - */ - bgeui a4, VALID_DOUBLE_EXCEPTION_ADDRESS, 1f - l32i a4, a1, PT_EXCCAUSE - bnei a4, EXCCAUSE_LEVEL1_INTERRUPT, 1f - /* We came here with an interrupt means interrupts were enabled - * and we've just disabled them. - */ - movi a4, trace_hardirqs_off - callx4 a4 -1: -#endif - /* Go to second-level dispatcher. Set up parameters to pass to the * exception handler and call the exception handler. */ rsr a4, excsave1 mov a6, a1 # pass stack frame - mov a7, a0 # pass EXCCAUSE - addx4 a4, a0, a4 + mov a7, a2 # pass EXCCAUSE + addx4 a4, a2, a4 l32i a4, a4, EXC_TABLE_DEFAULT # load handler /* Call the second-level handler */ @@ -419,8 +464,17 @@ common_exception: .global common_exception_return common_exception_return: +#if XTENSA_FAKE_NMI + l32i a2, a1, PT_EXCCAUSE + movi a3, EXCCAUSE_MAPPED_NMI + beq a2, a3, .LNMIexit +#endif 1: - rsil a2, LOCKLEVEL + irq_save a2, a3 +#ifdef CONFIG_TRACE_IRQFLAGS + movi a4, trace_hardirqs_off + callx4 a4 +#endif /* Jump if we are returning from kernel exceptions. */ @@ -445,6 +499,10 @@ common_exception_return: /* Call do_signal() */ +#ifdef CONFIG_TRACE_IRQFLAGS + movi a4, trace_hardirqs_on + callx4 a4 +#endif rsil a2, 0 movi a4, do_notify_resume # int do_notify_resume(struct pt_regs*) mov a6, a1 @@ -453,6 +511,10 @@ common_exception_return: 3: /* Reschedule */ +#ifdef CONFIG_TRACE_IRQFLAGS + movi a4, trace_hardirqs_on + callx4 a4 +#endif rsil a2, 0 movi a4, schedule # void schedule (void) callx4 a4 @@ -471,6 +533,12 @@ common_exception_return: j 1b #endif +#if XTENSA_FAKE_NMI +.LNMIexit: + l32i a3, a1, PT_PS + _bbci.l a3, PS_UM_BIT, 4f +#endif + 5: #ifdef CONFIG_DEBUG_TLB_SANITY l32i a4, a1, PT_DEPC @@ -481,16 +549,8 @@ common_exception_return: 6: 4: #ifdef CONFIG_TRACE_IRQFLAGS - l32i a4, a1, PT_DEPC - /* Double exception means we came here with an exception - * while PS.EXCM was set, i.e. interrupts disabled. - */ - bgeui a4, VALID_DOUBLE_EXCEPTION_ADDRESS, 1f - l32i a4, a1, PT_EXCCAUSE - bnei a4, EXCCAUSE_LEVEL1_INTERRUPT, 1f - /* We came here with an interrupt means interrupts were enabled - * and we'll reenable them on return. - */ + extui a4, a3, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH + bgei a4, LOCKLEVEL, 1f movi a4, trace_hardirqs_on callx4 a4 1: @@ -1562,6 +1622,13 @@ ENTRY(fast_second_level_miss) rfde 9: l32i a0, a1, TASK_ACTIVE_MM # unlikely case mm == 0 + bnez a0, 8b + + /* Even more unlikely case active_mm == 0. + * We can get here with NMI in the middle of context_switch that + * touches vmalloc area. + */ + movi a0, init_mm j 8b #if (DCACHE_WAY_SIZE > PAGE_SIZE) @@ -1865,10 +1932,8 @@ ENTRY(_switch_to) /* Disable ints while we manipulate the stack pointer. */ - rsil a14, LOCKLEVEL - rsr a3, excsave1 + irq_save a14, a3 rsync - s32i a3, a3, EXC_TABLE_FIXUP /* enter critical section */ /* Switch CPENABLE */ @@ -1889,9 +1954,7 @@ ENTRY(_switch_to) */ rsr a3, excsave1 # exc_table - movi a6, 0 addi a7, a5, PT_REGS_OFFSET - s32i a6, a3, EXC_TABLE_FIXUP s32i a7, a3, EXC_TABLE_KSTK /* restore context of the task 'next' */ diff --git a/arch/xtensa/kernel/irq.c b/arch/xtensa/kernel/irq.c index 3eee94f62..4ac3d2316 100644 --- a/arch/xtensa/kernel/irq.c +++ b/arch/xtensa/kernel/irq.c @@ -28,7 +28,7 @@ #include #include -atomic_t irq_err_count; +DECLARE_PER_CPU(unsigned long, nmi_count); asmlinkage void do_IRQ(int hwirq, struct pt_regs *regs) { @@ -57,11 +57,16 @@ asmlinkage void do_IRQ(int hwirq, struct pt_regs *regs) int arch_show_interrupts(struct seq_file *p, int prec) { + unsigned cpu __maybe_unused; #ifdef CONFIG_SMP show_ipi_list(p, prec); #endif - seq_printf(p, "%*s: ", prec, "ERR"); - seq_printf(p, "%10u\n", atomic_read(&irq_err_count)); +#if XTENSA_FAKE_NMI + seq_printf(p, "%*s:", prec, "NMI"); + for_each_online_cpu(cpu) + seq_printf(p, " %10lu", per_cpu(nmi_count, cpu)); + seq_puts(p, " Non-maskable interrupts\n"); +#endif return 0; } @@ -106,6 +111,12 @@ int xtensa_irq_map(struct irq_domain *d, unsigned int irq, irq_set_chip_and_handler_name(irq, irq_chip, handle_percpu_irq, "timer"); irq_clear_status_flags(irq, IRQ_LEVEL); +#ifdef XCHAL_INTTYPE_MASK_PROFILING + } else if (mask & XCHAL_INTTYPE_MASK_PROFILING) { + irq_set_chip_and_handler_name(irq, irq_chip, + handle_percpu_irq, "profiling"); + irq_set_status_flags(irq, IRQ_LEVEL); +#endif } else {/* XCHAL_INTTYPE_MASK_WRITE_ERROR */ /* XCHAL_INTTYPE_MASK_NMI */ irq_set_chip_and_handler_name(irq, irq_chip, @@ -166,23 +177,25 @@ void migrate_irqs(void) for_each_active_irq(i) { struct irq_data *data = irq_get_irq_data(i); + struct cpumask *mask; unsigned int newcpu; if (irqd_is_per_cpu(data)) continue; - if (!cpumask_test_cpu(cpu, data->affinity)) + mask = irq_data_get_affinity_mask(data); + if (!cpumask_test_cpu(cpu, mask)) continue; - newcpu = cpumask_any_and(data->affinity, cpu_online_mask); + newcpu = cpumask_any_and(mask, cpu_online_mask); if (newcpu >= nr_cpu_ids) { pr_info_ratelimited("IRQ%u no longer affine to CPU%u\n", i, cpu); - cpumask_setall(data->affinity); + cpumask_setall(mask); } - irq_set_affinity(i, data->affinity); + irq_set_affinity(i, mask); } } #endif /* CONFIG_HOTPLUG_CPU */ diff --git a/arch/xtensa/kernel/pci-dma.c b/arch/xtensa/kernel/pci-dma.c index e8b76b8e4..fb75ebf14 100644 --- a/arch/xtensa/kernel/pci-dma.c +++ b/arch/xtensa/kernel/pci-dma.c @@ -1,6 +1,4 @@ /* - * arch/xtensa/kernel/pci-dma.c - * * DMA coherent memory allocation. * * This program is free software; you can redistribute it and/or modify it @@ -9,6 +7,7 @@ * option) any later version. * * Copyright (C) 2002 - 2005 Tensilica Inc. + * Copyright (C) 2015 Cadence Design Systems Inc. * * Based on version for i386. * @@ -25,13 +24,107 @@ #include #include +void dma_cache_sync(struct device *dev, void *vaddr, size_t size, + enum dma_data_direction dir) +{ + switch (dir) { + case DMA_BIDIRECTIONAL: + __flush_invalidate_dcache_range((unsigned long)vaddr, size); + break; + + case DMA_FROM_DEVICE: + __invalidate_dcache_range((unsigned long)vaddr, size); + break; + + case DMA_TO_DEVICE: + __flush_dcache_range((unsigned long)vaddr, size); + break; + + case DMA_NONE: + BUG(); + break; + } +} +EXPORT_SYMBOL(dma_cache_sync); + +static void xtensa_sync_single_for_cpu(struct device *dev, + dma_addr_t dma_handle, size_t size, + enum dma_data_direction dir) +{ + void *vaddr; + + switch (dir) { + case DMA_BIDIRECTIONAL: + case DMA_FROM_DEVICE: + vaddr = bus_to_virt(dma_handle); + __invalidate_dcache_range((unsigned long)vaddr, size); + break; + + case DMA_NONE: + BUG(); + break; + + default: + break; + } +} + +static void xtensa_sync_single_for_device(struct device *dev, + dma_addr_t dma_handle, size_t size, + enum dma_data_direction dir) +{ + void *vaddr; + + switch (dir) { + case DMA_BIDIRECTIONAL: + case DMA_TO_DEVICE: + vaddr = bus_to_virt(dma_handle); + __flush_dcache_range((unsigned long)vaddr, size); + break; + + case DMA_NONE: + BUG(); + break; + + default: + break; + } +} + +static void xtensa_sync_sg_for_cpu(struct device *dev, + struct scatterlist *sg, int nents, + enum dma_data_direction dir) +{ + struct scatterlist *s; + int i; + + for_each_sg(sg, s, nents, i) { + xtensa_sync_single_for_cpu(dev, sg_dma_address(s), + sg_dma_len(s), dir); + } +} + +static void xtensa_sync_sg_for_device(struct device *dev, + struct scatterlist *sg, int nents, + enum dma_data_direction dir) +{ + struct scatterlist *s; + int i; + + for_each_sg(sg, s, nents, i) { + xtensa_sync_single_for_device(dev, sg_dma_address(s), + sg_dma_len(s), dir); + } +} + /* * Note: We assume that the full memory space is always mapped to 'kseg' * Otherwise we have to use page attributes (not implemented). */ -void * -dma_alloc_coherent(struct device *dev,size_t size,dma_addr_t *handle,gfp_t flag) +static void *xtensa_dma_alloc(struct device *dev, size_t size, + dma_addr_t *handle, gfp_t flag, + struct dma_attrs *attrs) { unsigned long ret; unsigned long uncached = 0; @@ -52,20 +145,15 @@ dma_alloc_coherent(struct device *dev,size_t size,dma_addr_t *handle,gfp_t flag) BUG_ON(ret < XCHAL_KSEG_CACHED_VADDR || ret > XCHAL_KSEG_CACHED_VADDR + XCHAL_KSEG_SIZE - 1); + uncached = ret + XCHAL_KSEG_BYPASS_VADDR - XCHAL_KSEG_CACHED_VADDR; + *handle = virt_to_bus((void *)ret); + __invalidate_dcache_range(ret, size); - if (ret != 0) { - memset((void*) ret, 0, size); - uncached = ret+XCHAL_KSEG_BYPASS_VADDR-XCHAL_KSEG_CACHED_VADDR; - *handle = virt_to_bus((void*)ret); - __flush_invalidate_dcache_range(ret, size); - } - - return (void*)uncached; + return (void *)uncached; } -EXPORT_SYMBOL(dma_alloc_coherent); -void dma_free_coherent(struct device *hwdev, size_t size, - void *vaddr, dma_addr_t dma_handle) +static void xtensa_dma_free(struct device *hwdev, size_t size, void *vaddr, + dma_addr_t dma_handle, struct dma_attrs *attrs) { unsigned long addr = (unsigned long)vaddr + XCHAL_KSEG_CACHED_VADDR - XCHAL_KSEG_BYPASS_VADDR; @@ -75,24 +163,79 @@ void dma_free_coherent(struct device *hwdev, size_t size, free_pages(addr, get_order(size)); } -EXPORT_SYMBOL(dma_free_coherent); +static dma_addr_t xtensa_map_page(struct device *dev, struct page *page, + unsigned long offset, size_t size, + enum dma_data_direction dir, + struct dma_attrs *attrs) +{ + dma_addr_t dma_handle = page_to_phys(page) + offset; + + BUG_ON(PageHighMem(page)); + xtensa_sync_single_for_device(dev, dma_handle, size, dir); + return dma_handle; +} -void consistent_sync(void *vaddr, size_t size, int direction) +static void xtensa_unmap_page(struct device *dev, dma_addr_t dma_handle, + size_t size, enum dma_data_direction dir, + struct dma_attrs *attrs) { - switch (direction) { - case PCI_DMA_NONE: - BUG(); - case PCI_DMA_FROMDEVICE: /* invalidate only */ - __invalidate_dcache_range((unsigned long)vaddr, - (unsigned long)size); - break; + xtensa_sync_single_for_cpu(dev, dma_handle, size, dir); +} - case PCI_DMA_TODEVICE: /* writeback only */ - case PCI_DMA_BIDIRECTIONAL: /* writeback and invalidate */ - __flush_invalidate_dcache_range((unsigned long)vaddr, - (unsigned long)size); - break; +static int xtensa_map_sg(struct device *dev, struct scatterlist *sg, + int nents, enum dma_data_direction dir, + struct dma_attrs *attrs) +{ + struct scatterlist *s; + int i; + + for_each_sg(sg, s, nents, i) { + s->dma_address = xtensa_map_page(dev, sg_page(s), s->offset, + s->length, dir, attrs); + } + return nents; +} + +static void xtensa_unmap_sg(struct device *dev, + struct scatterlist *sg, int nents, + enum dma_data_direction dir, + struct dma_attrs *attrs) +{ + struct scatterlist *s; + int i; + + for_each_sg(sg, s, nents, i) { + xtensa_unmap_page(dev, sg_dma_address(s), + sg_dma_len(s), dir, attrs); } } -EXPORT_SYMBOL(consistent_sync); + +int xtensa_dma_mapping_error(struct device *dev, dma_addr_t dma_addr) +{ + return 0; +} + +struct dma_map_ops xtensa_dma_map_ops = { + .alloc = xtensa_dma_alloc, + .free = xtensa_dma_free, + .map_page = xtensa_map_page, + .unmap_page = xtensa_unmap_page, + .map_sg = xtensa_map_sg, + .unmap_sg = xtensa_unmap_sg, + .sync_single_for_cpu = xtensa_sync_single_for_cpu, + .sync_single_for_device = xtensa_sync_single_for_device, + .sync_sg_for_cpu = xtensa_sync_sg_for_cpu, + .sync_sg_for_device = xtensa_sync_sg_for_device, + .mapping_error = xtensa_dma_mapping_error, +}; +EXPORT_SYMBOL(xtensa_dma_map_ops); + +#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16) + +static int __init xtensa_dma_init(void) +{ + dma_debug_init(PREALLOC_DMA_DEBUG_ENTRIES); + return 0; +} +fs_initcall(xtensa_dma_init); diff --git a/arch/xtensa/kernel/perf_event.c b/arch/xtensa/kernel/perf_event.c new file mode 100644 index 000000000..54f01188c --- /dev/null +++ b/arch/xtensa/kernel/perf_event.c @@ -0,0 +1,454 @@ +/* + * Xtensa Performance Monitor Module driver + * See Tensilica Debug User's Guide for PMU registers documentation. + * + * Copyright (C) 2015 Cadence Design Systems Inc. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + */ + +#include +#include +#include +#include +#include +#include + +#include +#include + +/* Global control/status for all perf counters */ +#define XTENSA_PMU_PMG 0x1000 +/* Perf counter values */ +#define XTENSA_PMU_PM(i) (0x1080 + (i) * 4) +/* Perf counter control registers */ +#define XTENSA_PMU_PMCTRL(i) (0x1100 + (i) * 4) +/* Perf counter status registers */ +#define XTENSA_PMU_PMSTAT(i) (0x1180 + (i) * 4) + +#define XTENSA_PMU_PMG_PMEN 0x1 + +#define XTENSA_PMU_COUNTER_MASK 0xffffffffULL +#define XTENSA_PMU_COUNTER_MAX 0x7fffffff + +#define XTENSA_PMU_PMCTRL_INTEN 0x00000001 +#define XTENSA_PMU_PMCTRL_KRNLCNT 0x00000008 +#define XTENSA_PMU_PMCTRL_TRACELEVEL 0x000000f0 +#define XTENSA_PMU_PMCTRL_SELECT_SHIFT 8 +#define XTENSA_PMU_PMCTRL_SELECT 0x00001f00 +#define XTENSA_PMU_PMCTRL_MASK_SHIFT 16 +#define XTENSA_PMU_PMCTRL_MASK 0xffff0000 + +#define XTENSA_PMU_MASK(select, mask) \ + (((select) << XTENSA_PMU_PMCTRL_SELECT_SHIFT) | \ + ((mask) << XTENSA_PMU_PMCTRL_MASK_SHIFT) | \ + XTENSA_PMU_PMCTRL_TRACELEVEL | \ + XTENSA_PMU_PMCTRL_INTEN) + +#define XTENSA_PMU_PMSTAT_OVFL 0x00000001 +#define XTENSA_PMU_PMSTAT_INTASRT 0x00000010 + +struct xtensa_pmu_events { + /* Array of events currently on this core */ + struct perf_event *event[XCHAL_NUM_PERF_COUNTERS]; + /* Bitmap of used hardware counters */ + unsigned long used_mask[BITS_TO_LONGS(XCHAL_NUM_PERF_COUNTERS)]; +}; +static DEFINE_PER_CPU(struct xtensa_pmu_events, xtensa_pmu_events); + +static const u32 xtensa_hw_ctl[] = { + [PERF_COUNT_HW_CPU_CYCLES] = XTENSA_PMU_MASK(0, 0x1), + [PERF_COUNT_HW_INSTRUCTIONS] = XTENSA_PMU_MASK(2, 0xffff), + [PERF_COUNT_HW_CACHE_REFERENCES] = XTENSA_PMU_MASK(10, 0x1), + [PERF_COUNT_HW_CACHE_MISSES] = XTENSA_PMU_MASK(12, 0x1), + /* Taken and non-taken branches + taken loop ends */ + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = XTENSA_PMU_MASK(2, 0x490), + /* Instruction-related + other global stall cycles */ + [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = XTENSA_PMU_MASK(4, 0x1ff), + /* Data-related global stall cycles */ + [PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = XTENSA_PMU_MASK(3, 0x1ff), +}; + +#define C(_x) PERF_COUNT_HW_CACHE_##_x + +static const u32 xtensa_cache_ctl[][C(OP_MAX)][C(RESULT_MAX)] = { + [C(L1D)] = { + [C(OP_READ)] = { + [C(RESULT_ACCESS)] = XTENSA_PMU_MASK(10, 0x1), + [C(RESULT_MISS)] = XTENSA_PMU_MASK(10, 0x2), + }, + [C(OP_WRITE)] = { + [C(RESULT_ACCESS)] = XTENSA_PMU_MASK(11, 0x1), + [C(RESULT_MISS)] = XTENSA_PMU_MASK(11, 0x2), + }, + }, + [C(L1I)] = { + [C(OP_READ)] = { + [C(RESULT_ACCESS)] = XTENSA_PMU_MASK(8, 0x1), + [C(RESULT_MISS)] = XTENSA_PMU_MASK(8, 0x2), + }, + }, + [C(DTLB)] = { + [C(OP_READ)] = { + [C(RESULT_ACCESS)] = XTENSA_PMU_MASK(9, 0x1), + [C(RESULT_MISS)] = XTENSA_PMU_MASK(9, 0x8), + }, + }, + [C(ITLB)] = { + [C(OP_READ)] = { + [C(RESULT_ACCESS)] = XTENSA_PMU_MASK(7, 0x1), + [C(RESULT_MISS)] = XTENSA_PMU_MASK(7, 0x8), + }, + }, +}; + +static int xtensa_pmu_cache_event(u64 config) +{ + unsigned int cache_type, cache_op, cache_result; + int ret; + + cache_type = (config >> 0) & 0xff; + cache_op = (config >> 8) & 0xff; + cache_result = (config >> 16) & 0xff; + + if (cache_type >= ARRAY_SIZE(xtensa_cache_ctl) || + cache_op >= C(OP_MAX) || + cache_result >= C(RESULT_MAX)) + return -EINVAL; + + ret = xtensa_cache_ctl[cache_type][cache_op][cache_result]; + + if (ret == 0) + return -EINVAL; + + return ret; +} + +static inline uint32_t xtensa_pmu_read_counter(int idx) +{ + return get_er(XTENSA_PMU_PM(idx)); +} + +static inline void xtensa_pmu_write_counter(int idx, uint32_t v) +{ + set_er(v, XTENSA_PMU_PM(idx)); +} + +static void xtensa_perf_event_update(struct perf_event *event, + struct hw_perf_event *hwc, int idx) +{ + uint64_t prev_raw_count, new_raw_count; + int64_t delta; + + do { + prev_raw_count = local64_read(&hwc->prev_count); + new_raw_count = xtensa_pmu_read_counter(event->hw.idx); + } while (local64_cmpxchg(&hwc->prev_count, prev_raw_count, + new_raw_count) != prev_raw_count); + + delta = (new_raw_count - prev_raw_count) & XTENSA_PMU_COUNTER_MASK; + + local64_add(delta, &event->count); + local64_sub(delta, &hwc->period_left); +} + +static bool xtensa_perf_event_set_period(struct perf_event *event, + struct hw_perf_event *hwc, int idx) +{ + bool rc = false; + s64 left; + + if (!is_sampling_event(event)) { + left = XTENSA_PMU_COUNTER_MAX; + } else { + s64 period = hwc->sample_period; + + left = local64_read(&hwc->period_left); + if (left <= -period) { + left = period; + local64_set(&hwc->period_left, left); + hwc->last_period = period; + rc = true; + } else if (left <= 0) { + left += period; + local64_set(&hwc->period_left, left); + hwc->last_period = period; + rc = true; + } + if (left > XTENSA_PMU_COUNTER_MAX) + left = XTENSA_PMU_COUNTER_MAX; + } + + local64_set(&hwc->prev_count, -left); + xtensa_pmu_write_counter(idx, -left); + perf_event_update_userpage(event); + + return rc; +} + +static void xtensa_pmu_enable(struct pmu *pmu) +{ + set_er(get_er(XTENSA_PMU_PMG) | XTENSA_PMU_PMG_PMEN, XTENSA_PMU_PMG); +} + +static void xtensa_pmu_disable(struct pmu *pmu) +{ + set_er(get_er(XTENSA_PMU_PMG) & ~XTENSA_PMU_PMG_PMEN, XTENSA_PMU_PMG); +} + +static int xtensa_pmu_event_init(struct perf_event *event) +{ + int ret; + + switch (event->attr.type) { + case PERF_TYPE_HARDWARE: + if (event->attr.config >= ARRAY_SIZE(xtensa_hw_ctl) || + xtensa_hw_ctl[event->attr.config] == 0) + return -EINVAL; + event->hw.config = xtensa_hw_ctl[event->attr.config]; + return 0; + + case PERF_TYPE_HW_CACHE: + ret = xtensa_pmu_cache_event(event->attr.config); + if (ret < 0) + return ret; + event->hw.config = ret; + return 0; + + case PERF_TYPE_RAW: + /* Not 'previous counter' select */ + if ((event->attr.config & XTENSA_PMU_PMCTRL_SELECT) == + (1 << XTENSA_PMU_PMCTRL_SELECT_SHIFT)) + return -EINVAL; + event->hw.config = (event->attr.config & + (XTENSA_PMU_PMCTRL_KRNLCNT | + XTENSA_PMU_PMCTRL_TRACELEVEL | + XTENSA_PMU_PMCTRL_SELECT | + XTENSA_PMU_PMCTRL_MASK)) | + XTENSA_PMU_PMCTRL_INTEN; + return 0; + + default: + return -ENOENT; + } +} + +/* + * Starts/Stops a counter present on the PMU. The PMI handler + * should stop the counter when perf_event_overflow() returns + * !0. ->start() will be used to continue. + */ +static void xtensa_pmu_start(struct perf_event *event, int flags) +{ + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + + if (WARN_ON_ONCE(idx == -1)) + return; + + if (flags & PERF_EF_RELOAD) { + WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE)); + xtensa_perf_event_set_period(event, hwc, idx); + } + + hwc->state = 0; + + set_er(hwc->config, XTENSA_PMU_PMCTRL(idx)); +} + +static void xtensa_pmu_stop(struct perf_event *event, int flags) +{ + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + + if (!(hwc->state & PERF_HES_STOPPED)) { + set_er(0, XTENSA_PMU_PMCTRL(idx)); + set_er(get_er(XTENSA_PMU_PMSTAT(idx)), + XTENSA_PMU_PMSTAT(idx)); + hwc->state |= PERF_HES_STOPPED; + } + + if ((flags & PERF_EF_UPDATE) && + !(event->hw.state & PERF_HES_UPTODATE)) { + xtensa_perf_event_update(event, &event->hw, idx); + event->hw.state |= PERF_HES_UPTODATE; + } +} + +/* + * Adds/Removes a counter to/from the PMU, can be done inside + * a transaction, see the ->*_txn() methods. + */ +static int xtensa_pmu_add(struct perf_event *event, int flags) +{ + struct xtensa_pmu_events *ev = this_cpu_ptr(&xtensa_pmu_events); + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + + if (__test_and_set_bit(idx, ev->used_mask)) { + idx = find_first_zero_bit(ev->used_mask, + XCHAL_NUM_PERF_COUNTERS); + if (idx == XCHAL_NUM_PERF_COUNTERS) + return -EAGAIN; + + __set_bit(idx, ev->used_mask); + hwc->idx = idx; + } + ev->event[idx] = event; + + hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED; + + if (flags & PERF_EF_START) + xtensa_pmu_start(event, PERF_EF_RELOAD); + + perf_event_update_userpage(event); + return 0; +} + +static void xtensa_pmu_del(struct perf_event *event, int flags) +{ + struct xtensa_pmu_events *ev = this_cpu_ptr(&xtensa_pmu_events); + + xtensa_pmu_stop(event, PERF_EF_UPDATE); + __clear_bit(event->hw.idx, ev->used_mask); + perf_event_update_userpage(event); +} + +static void xtensa_pmu_read(struct perf_event *event) +{ + xtensa_perf_event_update(event, &event->hw, event->hw.idx); +} + +static int callchain_trace(struct stackframe *frame, void *data) +{ + struct perf_callchain_entry *entry = data; + + perf_callchain_store(entry, frame->pc); + return 0; +} + +void perf_callchain_kernel(struct perf_callchain_entry *entry, + struct pt_regs *regs) +{ + xtensa_backtrace_kernel(regs, PERF_MAX_STACK_DEPTH, + callchain_trace, NULL, entry); +} + +void perf_callchain_user(struct perf_callchain_entry *entry, + struct pt_regs *regs) +{ + xtensa_backtrace_user(regs, PERF_MAX_STACK_DEPTH, + callchain_trace, entry); +} + +void perf_event_print_debug(void) +{ + unsigned long flags; + unsigned i; + + local_irq_save(flags); + pr_info("CPU#%d: PMG: 0x%08lx\n", smp_processor_id(), + get_er(XTENSA_PMU_PMG)); + for (i = 0; i < XCHAL_NUM_PERF_COUNTERS; ++i) + pr_info("PM%d: 0x%08lx, PMCTRL%d: 0x%08lx, PMSTAT%d: 0x%08lx\n", + i, get_er(XTENSA_PMU_PM(i)), + i, get_er(XTENSA_PMU_PMCTRL(i)), + i, get_er(XTENSA_PMU_PMSTAT(i))); + local_irq_restore(flags); +} + +irqreturn_t xtensa_pmu_irq_handler(int irq, void *dev_id) +{ + irqreturn_t rc = IRQ_NONE; + struct xtensa_pmu_events *ev = this_cpu_ptr(&xtensa_pmu_events); + unsigned i; + + for (i = find_first_bit(ev->used_mask, XCHAL_NUM_PERF_COUNTERS); + i < XCHAL_NUM_PERF_COUNTERS; + i = find_next_bit(ev->used_mask, XCHAL_NUM_PERF_COUNTERS, i + 1)) { + uint32_t v = get_er(XTENSA_PMU_PMSTAT(i)); + struct perf_event *event = ev->event[i]; + struct hw_perf_event *hwc = &event->hw; + u64 last_period; + + if (!(v & XTENSA_PMU_PMSTAT_OVFL)) + continue; + + set_er(v, XTENSA_PMU_PMSTAT(i)); + xtensa_perf_event_update(event, hwc, i); + last_period = hwc->last_period; + if (xtensa_perf_event_set_period(event, hwc, i)) { + struct perf_sample_data data; + struct pt_regs *regs = get_irq_regs(); + + perf_sample_data_init(&data, 0, last_period); + if (perf_event_overflow(event, &data, regs)) + xtensa_pmu_stop(event, 0); + } + + rc = IRQ_HANDLED; + } + return rc; +} + +static struct pmu xtensa_pmu = { + .pmu_enable = xtensa_pmu_enable, + .pmu_disable = xtensa_pmu_disable, + .event_init = xtensa_pmu_event_init, + .add = xtensa_pmu_add, + .del = xtensa_pmu_del, + .start = xtensa_pmu_start, + .stop = xtensa_pmu_stop, + .read = xtensa_pmu_read, +}; + +static void xtensa_pmu_setup(void) +{ + unsigned i; + + set_er(0, XTENSA_PMU_PMG); + for (i = 0; i < XCHAL_NUM_PERF_COUNTERS; ++i) { + set_er(0, XTENSA_PMU_PMCTRL(i)); + set_er(get_er(XTENSA_PMU_PMSTAT(i)), XTENSA_PMU_PMSTAT(i)); + } +} + +static int xtensa_pmu_notifier(struct notifier_block *self, + unsigned long action, void *data) +{ + switch (action & ~CPU_TASKS_FROZEN) { + case CPU_STARTING: + xtensa_pmu_setup(); + break; + + default: + break; + } + + return NOTIFY_OK; +} + +static int __init xtensa_pmu_init(void) +{ + int ret; + int irq = irq_create_mapping(NULL, XCHAL_PROFILING_INTERRUPT); + + perf_cpu_notifier(xtensa_pmu_notifier); +#if XTENSA_FAKE_NMI + enable_irq(irq); +#else + ret = request_irq(irq, xtensa_pmu_irq_handler, IRQF_PERCPU, + "pmu", NULL); + if (ret < 0) + return ret; +#endif + + ret = perf_pmu_register(&xtensa_pmu, "cpu", PERF_TYPE_RAW); + if (ret) + free_irq(irq, NULL); + + return ret; +} +early_initcall(xtensa_pmu_init); diff --git a/arch/xtensa/kernel/stacktrace.c b/arch/xtensa/kernel/stacktrace.c index 7d2c317bd..7538d802b 100644 --- a/arch/xtensa/kernel/stacktrace.c +++ b/arch/xtensa/kernel/stacktrace.c @@ -1,11 +1,12 @@ /* - * arch/xtensa/kernel/stacktrace.c + * Kernel and userspace stack tracing. * * This file is subject to the terms and conditions of the GNU General Public * License. See the file "COPYING" in the main directory of this archive * for more details. * * Copyright (C) 2001 - 2013 Tensilica Inc. + * Copyright (C) 2015 Cadence Design Systems Inc. */ #include #include @@ -13,6 +14,170 @@ #include #include +#include + +#if IS_ENABLED(CONFIG_OPROFILE) || IS_ENABLED(CONFIG_PERF_EVENTS) + +/* Address of common_exception_return, used to check the + * transition from kernel to user space. + */ +extern int common_exception_return; + +/* A struct that maps to the part of the frame containing the a0 and + * a1 registers. + */ +struct frame_start { + unsigned long a0; + unsigned long a1; +}; + +void xtensa_backtrace_user(struct pt_regs *regs, unsigned int depth, + int (*ufn)(struct stackframe *frame, void *data), + void *data) +{ + unsigned long windowstart = regs->windowstart; + unsigned long windowbase = regs->windowbase; + unsigned long a0 = regs->areg[0]; + unsigned long a1 = regs->areg[1]; + unsigned long pc = regs->pc; + struct stackframe frame; + int index; + + if (!depth--) + return; + + frame.pc = pc; + frame.sp = a1; + + if (pc == 0 || pc >= TASK_SIZE || ufn(&frame, data)) + return; + + /* Two steps: + * + * 1. Look through the register window for the + * previous PCs in the call trace. + * + * 2. Look on the stack. + */ + + /* Step 1. */ + /* Rotate WINDOWSTART to move the bit corresponding to + * the current window to the bit #0. + */ + windowstart = (windowstart << WSBITS | windowstart) >> windowbase; + + /* Look for bits that are set, they correspond to + * valid windows. + */ + for (index = WSBITS - 1; (index > 0) && depth; depth--, index--) + if (windowstart & (1 << index)) { + /* Get the PC from a0 and a1. */ + pc = MAKE_PC_FROM_RA(a0, pc); + /* Read a0 and a1 from the + * corresponding position in AREGs. + */ + a0 = regs->areg[index * 4]; + a1 = regs->areg[index * 4 + 1]; + + frame.pc = pc; + frame.sp = a1; + + if (pc == 0 || pc >= TASK_SIZE || ufn(&frame, data)) + return; + } + + /* Step 2. */ + /* We are done with the register window, we need to + * look through the stack. + */ + if (!depth) + return; + + /* Start from the a1 register. */ + /* a1 = regs->areg[1]; */ + while (a0 != 0 && depth--) { + struct frame_start frame_start; + /* Get the location for a1, a0 for the + * previous frame from the current a1. + */ + unsigned long *psp = (unsigned long *)a1; + + psp -= 4; + + /* Check if the region is OK to access. */ + if (!access_ok(VERIFY_READ, psp, sizeof(frame_start))) + return; + /* Copy a1, a0 from user space stack frame. */ + if (__copy_from_user_inatomic(&frame_start, psp, + sizeof(frame_start))) + return; + + pc = MAKE_PC_FROM_RA(a0, pc); + a0 = frame_start.a0; + a1 = frame_start.a1; + + frame.pc = pc; + frame.sp = a1; + + if (pc == 0 || pc >= TASK_SIZE || ufn(&frame, data)) + return; + } +} +EXPORT_SYMBOL(xtensa_backtrace_user); + +void xtensa_backtrace_kernel(struct pt_regs *regs, unsigned int depth, + int (*kfn)(struct stackframe *frame, void *data), + int (*ufn)(struct stackframe *frame, void *data), + void *data) +{ + unsigned long pc = regs->depc > VALID_DOUBLE_EXCEPTION_ADDRESS ? + regs->depc : regs->pc; + unsigned long sp_start, sp_end; + unsigned long a0 = regs->areg[0]; + unsigned long a1 = regs->areg[1]; + + sp_start = a1 & ~(THREAD_SIZE - 1); + sp_end = sp_start + THREAD_SIZE; + + /* Spill the register window to the stack first. */ + spill_registers(); + + /* Read the stack frames one by one and create the PC + * from the a0 and a1 registers saved there. + */ + while (a1 > sp_start && a1 < sp_end && depth--) { + struct stackframe frame; + unsigned long *psp = (unsigned long *)a1; + + frame.pc = pc; + frame.sp = a1; + + if (kernel_text_address(pc) && kfn(&frame, data)) + return; + + if (pc == (unsigned long)&common_exception_return) { + regs = (struct pt_regs *)a1; + if (user_mode(regs)) { + if (ufn == NULL) + return; + xtensa_backtrace_user(regs, depth, ufn, data); + return; + } + a0 = regs->areg[0]; + a1 = regs->areg[1]; + continue; + } + + sp_start = a1; + + pc = MAKE_PC_FROM_RA(a0, pc); + a0 = *(psp - 4); + a1 = *(psp - 3); + } +} +EXPORT_SYMBOL(xtensa_backtrace_kernel); + +#endif void walk_stackframe(unsigned long *sp, int (*fn)(struct stackframe *frame, void *data), diff --git a/arch/xtensa/kernel/time.c b/arch/xtensa/kernel/time.c index 2a1823de6..b97767dbc 100644 --- a/arch/xtensa/kernel/time.c +++ b/arch/xtensa/kernel/time.c @@ -52,8 +52,6 @@ static struct clocksource ccount_clocksource = { static int ccount_timer_set_next_event(unsigned long delta, struct clock_event_device *dev); -static void ccount_timer_set_mode(enum clock_event_mode mode, - struct clock_event_device *evt); struct ccount_timer { struct clock_event_device evt; int irq_enabled; @@ -77,35 +75,34 @@ static int ccount_timer_set_next_event(unsigned long delta, return ret; } -static void ccount_timer_set_mode(enum clock_event_mode mode, - struct clock_event_device *evt) +/* + * There is no way to disable the timer interrupt at the device level, + * only at the intenable register itself. Since enable_irq/disable_irq + * calls are nested, we need to make sure that these calls are + * balanced. + */ +static int ccount_timer_shutdown(struct clock_event_device *evt) +{ + struct ccount_timer *timer = + container_of(evt, struct ccount_timer, evt); + + if (timer->irq_enabled) { + disable_irq(evt->irq); + timer->irq_enabled = 0; + } + return 0; +} + +static int ccount_timer_set_oneshot(struct clock_event_device *evt) { struct ccount_timer *timer = container_of(evt, struct ccount_timer, evt); - /* - * There is no way to disable the timer interrupt at the device level, - * only at the intenable register itself. Since enable_irq/disable_irq - * calls are nested, we need to make sure that these calls are - * balanced. - */ - switch (mode) { - case CLOCK_EVT_MODE_SHUTDOWN: - case CLOCK_EVT_MODE_UNUSED: - if (timer->irq_enabled) { - disable_irq(evt->irq); - timer->irq_enabled = 0; - } - break; - case CLOCK_EVT_MODE_RESUME: - case CLOCK_EVT_MODE_ONESHOT: - if (!timer->irq_enabled) { - enable_irq(evt->irq); - timer->irq_enabled = 1; - } - default: - break; + if (!timer->irq_enabled) { + enable_irq(evt->irq); + timer->irq_enabled = 1; } + return 0; } static irqreturn_t timer_interrupt(int irq, void *dev_id); @@ -126,7 +123,9 @@ void local_timer_setup(unsigned cpu) clockevent->features = CLOCK_EVT_FEAT_ONESHOT; clockevent->rating = 300; clockevent->set_next_event = ccount_timer_set_next_event; - clockevent->set_mode = ccount_timer_set_mode; + clockevent->set_state_shutdown = ccount_timer_shutdown; + clockevent->set_state_oneshot = ccount_timer_set_oneshot; + clockevent->tick_resume = ccount_timer_set_oneshot; clockevent->cpumask = cpumask_of(cpu); clockevent->irq = irq_create_mapping(NULL, LINUX_TIMER_INT); if (WARN(!clockevent->irq, "error: can't map timer irq")) diff --git a/arch/xtensa/kernel/traps.c b/arch/xtensa/kernel/traps.c index 9d2f45f01..42d441f78 100644 --- a/arch/xtensa/kernel/traps.c +++ b/arch/xtensa/kernel/traps.c @@ -62,6 +62,7 @@ extern void fast_coprocessor(void); extern void do_illegal_instruction (struct pt_regs*); extern void do_interrupt (struct pt_regs*); +extern void do_nmi(struct pt_regs *); extern void do_unaligned_user (struct pt_regs*); extern void do_multihit (struct pt_regs*, unsigned long); extern void do_page_fault (struct pt_regs*, unsigned long); @@ -146,6 +147,9 @@ COPROCESSOR(6), #if XTENSA_HAVE_COPROCESSOR(7) COPROCESSOR(7), #endif +#if XTENSA_FAKE_NMI +{ EXCCAUSE_MAPPED_NMI, 0, do_nmi }, +#endif { EXCCAUSE_MAPPED_DEBUG, 0, do_debug }, { -1, -1, 0 } @@ -199,6 +203,28 @@ void do_multihit(struct pt_regs *regs, unsigned long exccause) extern void do_IRQ(int, struct pt_regs *); +#if XTENSA_FAKE_NMI + +irqreturn_t xtensa_pmu_irq_handler(int irq, void *dev_id); + +DEFINE_PER_CPU(unsigned long, nmi_count); + +void do_nmi(struct pt_regs *regs) +{ + struct pt_regs *old_regs; + + if ((regs->ps & PS_INTLEVEL_MASK) < LOCKLEVEL) + trace_hardirqs_off(); + + old_regs = set_irq_regs(regs); + nmi_enter(); + ++*this_cpu_ptr(&nmi_count); + xtensa_pmu_irq_handler(0, NULL); + nmi_exit(); + set_irq_regs(old_regs); +} +#endif + void do_interrupt(struct pt_regs *regs) { static const unsigned int_level_mask[] = { @@ -211,8 +237,11 @@ void do_interrupt(struct pt_regs *regs) XCHAL_INTLEVEL6_MASK, XCHAL_INTLEVEL7_MASK, }; - struct pt_regs *old_regs = set_irq_regs(regs); + struct pt_regs *old_regs; + + trace_hardirqs_off(); + old_regs = set_irq_regs(regs); irq_enter(); for (;;) { diff --git a/arch/xtensa/kernel/vectors.S b/arch/xtensa/kernel/vectors.S index 1b397a902..abcdb527f 100644 --- a/arch/xtensa/kernel/vectors.S +++ b/arch/xtensa/kernel/vectors.S @@ -627,7 +627,11 @@ ENTRY(_Level\level\()InterruptVector) wsr a0, excsave2 rsr a0, epc\level wsr a0, epc1 + .if \level <= LOCKLEVEL movi a0, EXCCAUSE_LEVEL1_INTERRUPT + .else + movi a0, EXCCAUSE_MAPPED_NMI + .endif wsr a0, exccause rsr a0, eps\level # branch to user or kernel vector @@ -682,11 +686,13 @@ ENDPROC(_WindowOverflow4) .align 4 _SimulateUserKernelVectorException: addi a0, a0, (1 << PS_EXCM_BIT) +#if !XTENSA_FAKE_NMI wsr a0, ps +#endif bbsi.l a0, PS_UM_BIT, 1f # branch if user mode - rsr a0, excsave2 # restore a0 + xsr a0, excsave2 # restore a0 j _KernelExceptionVector # simulate kernel vector exception -1: rsr a0, excsave2 # restore a0 +1: xsr a0, excsave2 # restore a0 j _UserExceptionVector # simulate user vector exception #endif -- cgit v1.2.3-54-g00ecf