diff options
Diffstat (limited to 'arch/x86/kernel/cpu')
-rw-r--r-- | arch/x86/kernel/cpu/Makefile | 12 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/common.c | 116 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/intel_cacheinfo.c | 63 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/mcheck/mce_amd_64.c | 21 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/mcheck/mce_intel_64.c | 1 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/perf_counter.c | 733 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/perfctr-watchdog.c | 2 |
7 files changed, 845 insertions, 103 deletions
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile index 82db7f45e2d..c3813306e0b 100644 --- a/arch/x86/kernel/cpu/Makefile +++ b/arch/x86/kernel/cpu/Makefile @@ -1,5 +1,5 @@ # -# Makefile for x86-compatible CPU details and quirks +# Makefile for x86-compatible CPU details, features and quirks # # Don't trace early stages of a secondary CPU boot @@ -22,11 +22,13 @@ obj-$(CONFIG_CPU_SUP_CENTAUR_64) += centaur_64.o obj-$(CONFIG_CPU_SUP_TRANSMETA_32) += transmeta.o obj-$(CONFIG_CPU_SUP_UMC_32) += umc.o -obj-$(CONFIG_X86_MCE) += mcheck/ -obj-$(CONFIG_MTRR) += mtrr/ -obj-$(CONFIG_CPU_FREQ) += cpufreq/ +obj-$(CONFIG_PERF_COUNTERS) += perf_counter.o -obj-$(CONFIG_X86_LOCAL_APIC) += perfctr-watchdog.o +obj-$(CONFIG_X86_MCE) += mcheck/ +obj-$(CONFIG_MTRR) += mtrr/ +obj-$(CONFIG_CPU_FREQ) += cpufreq/ + +obj-$(CONFIG_X86_LOCAL_APIC) += perfctr-watchdog.o quiet_cmd_mkcapflags = MKCAP $@ cmd_mkcapflags = $(PERL) $(srctree)/$(src)/mkcapflags.pl $< $@ diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c index 83492b1f93b..6fd316689c4 100644 --- a/arch/x86/kernel/cpu/common.c +++ b/arch/x86/kernel/cpu/common.c @@ -17,18 +17,21 @@ #include <asm/mmu_context.h> #include <asm/mtrr.h> #include <asm/mce.h> +#include <asm/perf_counter.h> #include <asm/pat.h> #include <asm/asm.h> #include <asm/numa.h> #include <asm/smp.h> +#include <asm/cpu.h> +#include <asm/cpumask.h> #ifdef CONFIG_X86_LOCAL_APIC #include <asm/mpspec.h> #include <asm/apic.h> #include <mach_apic.h> #include <asm/genapic.h> +#include <asm/uv/uv.h> #endif -#include <asm/pda.h> #include <asm/pgtable.h> #include <asm/processor.h> #include <asm/desc.h> @@ -62,23 +65,23 @@ cpumask_t cpu_sibling_setup_map; static struct cpu_dev *this_cpu __cpuinitdata; +DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { #ifdef CONFIG_X86_64 -/* We need valid kernel segments for data and code in long mode too - * IRET will check the segment types kkeil 2000/10/28 - * Also sysret mandates a special GDT layout - */ -/* The TLS descriptors are currently at a different place compared to i386. - Hopefully nobody expects them at a fixed place (Wine?) */ -DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = { + /* + * We need valid kernel segments for data and code in long mode too + * IRET will check the segment types kkeil 2000/10/28 + * Also sysret mandates a special GDT layout + * + * The TLS descriptors are currently at a different place compared to i386. + * Hopefully nobody expects them at a fixed place (Wine?) + */ [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } }, [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } }, [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } }, [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } }, [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } }, [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } }, -} }; #else -DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } }, [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } }, [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } }, @@ -110,9 +113,9 @@ DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } }, [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } }, - [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } }, -} }; + [GDT_ENTRY_PERCPU] = { { { 0x0000ffff, 0x00cf9200 } } }, #endif +} }; EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); #ifdef CONFIG_X86_32 @@ -772,6 +775,7 @@ void __init identify_boot_cpu(void) #else vgetcpu_set_mode(); #endif + init_hw_perf_counters(); } void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) @@ -877,54 +881,26 @@ static __init int setup_disablecpuid(char *arg) __setup("clearcpuid=", setup_disablecpuid); #ifdef CONFIG_X86_64 -struct x8664_pda **_cpu_pda __read_mostly; -EXPORT_SYMBOL(_cpu_pda); - struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table }; -static char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss; - -void __cpuinit pda_init(int cpu) -{ - struct x8664_pda *pda = cpu_pda(cpu); +DEFINE_PER_CPU_FIRST(union irq_stack_union, + irq_stack_union) __aligned(PAGE_SIZE); +#ifdef CONFIG_SMP +DEFINE_PER_CPU(char *, irq_stack_ptr); /* will be set during per cpu init */ +#else +DEFINE_PER_CPU(char *, irq_stack_ptr) = + per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64; +#endif - /* Setup up data that may be needed in __get_free_pages early */ - loadsegment(fs, 0); - loadsegment(gs, 0); - /* Memory clobbers used to order PDA accessed */ - mb(); - wrmsrl(MSR_GS_BASE, pda); - mb(); - - pda->cpunumber = cpu; - pda->irqcount = -1; - pda->kernelstack = (unsigned long)stack_thread_info() - - PDA_STACKOFFSET + THREAD_SIZE; - pda->active_mm = &init_mm; - pda->mmu_state = 0; - - if (cpu == 0) { - /* others are initialized in smpboot.c */ - pda->pcurrent = &init_task; - pda->irqstackptr = boot_cpu_stack; - pda->irqstackptr += IRQSTACKSIZE - 64; - } else { - if (!pda->irqstackptr) { - pda->irqstackptr = (char *) - __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER); - if (!pda->irqstackptr) - panic("cannot allocate irqstack for cpu %d", - cpu); - pda->irqstackptr += IRQSTACKSIZE - 64; - } +DEFINE_PER_CPU(unsigned long, kernel_stack) = + (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE; +EXPORT_PER_CPU_SYMBOL(kernel_stack); - if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE) - pda->nodenumber = cpu_to_node(cpu); - } -} +DEFINE_PER_CPU(unsigned int, irq_count) = -1; -static char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + - DEBUG_STKSZ] __page_aligned_bss; +static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks + [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]) + __aligned(PAGE_SIZE); extern asmlinkage void ignore_sysret(void); @@ -982,15 +958,18 @@ void __cpuinit cpu_init(void) struct tss_struct *t = &per_cpu(init_tss, cpu); struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu); unsigned long v; - char *estacks = NULL; struct task_struct *me; int i; - /* CPU 0 is initialised in head64.c */ - if (cpu != 0) - pda_init(cpu); - else - estacks = boot_exception_stacks; + loadsegment(fs, 0); + loadsegment(gs, 0); + load_gs_base(cpu); + +#ifdef CONFIG_NUMA + if (cpu != 0 && percpu_read(node_number) == 0 && + cpu_to_node(cpu) != NUMA_NO_NODE) + percpu_write(node_number, cpu_to_node(cpu)); +#endif me = current; @@ -1024,18 +1003,13 @@ void __cpuinit cpu_init(void) * set up and load the per-CPU TSS */ if (!orig_ist->ist[0]) { - static const unsigned int order[N_EXCEPTION_STACKS] = { - [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER, - [DEBUG_STACK - 1] = DEBUG_STACK_ORDER + static const unsigned int sizes[N_EXCEPTION_STACKS] = { + [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ, + [DEBUG_STACK - 1] = DEBUG_STKSZ }; + char *estacks = per_cpu(exception_stacks, cpu); for (v = 0; v < N_EXCEPTION_STACKS; v++) { - if (cpu) { - estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]); - if (!estacks) - panic("Cannot allocate exception " - "stack %ld %d\n", v, cpu); - } - estacks += PAGE_SIZE << order[v]; + estacks += sizes[v]; orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks; } diff --git a/arch/x86/kernel/cpu/intel_cacheinfo.c b/arch/x86/kernel/cpu/intel_cacheinfo.c index da299eb85fc..7293508d8f5 100644 --- a/arch/x86/kernel/cpu/intel_cacheinfo.c +++ b/arch/x86/kernel/cpu/intel_cacheinfo.c @@ -147,7 +147,16 @@ struct _cpuid4_info { union _cpuid4_leaf_ecx ecx; unsigned long size; unsigned long can_disable; - cpumask_t shared_cpu_map; /* future?: only cpus/node is needed */ + DECLARE_BITMAP(shared_cpu_map, NR_CPUS); +}; + +/* subset of above _cpuid4_info w/o shared_cpu_map */ +struct _cpuid4_info_regs { + union _cpuid4_leaf_eax eax; + union _cpuid4_leaf_ebx ebx; + union _cpuid4_leaf_ecx ecx; + unsigned long size; + unsigned long can_disable; }; #ifdef CONFIG_PCI @@ -278,7 +287,7 @@ amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax, } static void __cpuinit -amd_check_l3_disable(int index, struct _cpuid4_info *this_leaf) +amd_check_l3_disable(int index, struct _cpuid4_info_regs *this_leaf) { if (index < 3) return; @@ -286,7 +295,8 @@ amd_check_l3_disable(int index, struct _cpuid4_info *this_leaf) } static int -__cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf) +__cpuinit cpuid4_cache_lookup_regs(int index, + struct _cpuid4_info_regs *this_leaf) { union _cpuid4_leaf_eax eax; union _cpuid4_leaf_ebx ebx; @@ -314,6 +324,15 @@ __cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf) return 0; } +static int +__cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf) +{ + struct _cpuid4_info_regs *leaf_regs = + (struct _cpuid4_info_regs *)this_leaf; + + return cpuid4_cache_lookup_regs(index, leaf_regs); +} + static int __cpuinit find_num_cache_leaves(void) { unsigned int eax, ebx, ecx, edx; @@ -353,11 +372,10 @@ unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c) * parameters cpuid leaf to find the cache details */ for (i = 0; i < num_cache_leaves; i++) { - struct _cpuid4_info this_leaf; - + struct _cpuid4_info_regs this_leaf; int retval; - retval = cpuid4_cache_lookup(i, &this_leaf); + retval = cpuid4_cache_lookup_regs(i, &this_leaf); if (retval >= 0) { switch(this_leaf.eax.split.level) { case 1: @@ -506,17 +524,20 @@ static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index) num_threads_sharing = 1 + this_leaf->eax.split.num_threads_sharing; if (num_threads_sharing == 1) - cpu_set(cpu, this_leaf->shared_cpu_map); + cpumask_set_cpu(cpu, to_cpumask(this_leaf->shared_cpu_map)); else { index_msb = get_count_order(num_threads_sharing); for_each_online_cpu(i) { if (cpu_data(i).apicid >> index_msb == c->apicid >> index_msb) { - cpu_set(i, this_leaf->shared_cpu_map); + cpumask_set_cpu(i, + to_cpumask(this_leaf->shared_cpu_map)); if (i != cpu && per_cpu(cpuid4_info, i)) { - sibling_leaf = CPUID4_INFO_IDX(i, index); - cpu_set(cpu, sibling_leaf->shared_cpu_map); + sibling_leaf = + CPUID4_INFO_IDX(i, index); + cpumask_set_cpu(cpu, to_cpumask( + sibling_leaf->shared_cpu_map)); } } } @@ -528,9 +549,10 @@ static void __cpuinit cache_remove_shared_cpu_map(unsigned int cpu, int index) int sibling; this_leaf = CPUID4_INFO_IDX(cpu, index); - for_each_cpu_mask_nr(sibling, this_leaf->shared_cpu_map) { + for_each_cpu(sibling, to_cpumask(this_leaf->shared_cpu_map)) { sibling_leaf = CPUID4_INFO_IDX(sibling, index); - cpu_clear(cpu, sibling_leaf->shared_cpu_map); + cpumask_clear_cpu(cpu, + to_cpumask(sibling_leaf->shared_cpu_map)); } } #else @@ -635,8 +657,9 @@ static ssize_t show_shared_cpu_map_func(struct _cpuid4_info *this_leaf, int n = 0; if (len > 1) { - cpumask_t *mask = &this_leaf->shared_cpu_map; + const struct cpumask *mask; + mask = to_cpumask(this_leaf->shared_cpu_map); n = type? cpulist_scnprintf(buf, len-2, mask) : cpumask_scnprintf(buf, len-2, mask); @@ -699,7 +722,8 @@ static struct pci_dev *get_k8_northbridge(int node) static ssize_t show_cache_disable(struct _cpuid4_info *this_leaf, char *buf) { - int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map)); + const struct cpumask *mask = to_cpumask(this_leaf->shared_cpu_map); + int node = cpu_to_node(cpumask_first(mask)); struct pci_dev *dev = NULL; ssize_t ret = 0; int i; @@ -733,7 +757,8 @@ static ssize_t store_cache_disable(struct _cpuid4_info *this_leaf, const char *buf, size_t count) { - int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map)); + const struct cpumask *mask = to_cpumask(this_leaf->shared_cpu_map); + int node = cpu_to_node(cpumask_first(mask)); struct pci_dev *dev = NULL; unsigned int ret, index, val; @@ -878,7 +903,7 @@ err_out: return -ENOMEM; } -static cpumask_t cache_dev_map = CPU_MASK_NONE; +static DECLARE_BITMAP(cache_dev_map, NR_CPUS); /* Add/Remove cache interface for CPU device */ static int __cpuinit cache_add_dev(struct sys_device * sys_dev) @@ -918,7 +943,7 @@ static int __cpuinit cache_add_dev(struct sys_device * sys_dev) } kobject_uevent(&(this_object->kobj), KOBJ_ADD); } - cpu_set(cpu, cache_dev_map); + cpumask_set_cpu(cpu, to_cpumask(cache_dev_map)); kobject_uevent(per_cpu(cache_kobject, cpu), KOBJ_ADD); return 0; @@ -931,9 +956,9 @@ static void __cpuinit cache_remove_dev(struct sys_device * sys_dev) if (per_cpu(cpuid4_info, cpu) == NULL) return; - if (!cpu_isset(cpu, cache_dev_map)) + if (!cpumask_test_cpu(cpu, to_cpumask(cache_dev_map))) return; - cpu_clear(cpu, cache_dev_map); + cpumask_clear_cpu(cpu, to_cpumask(cache_dev_map)); for (i = 0; i < num_cache_leaves; i++) kobject_put(&(INDEX_KOBJECT_PTR(cpu,i)->kobj)); diff --git a/arch/x86/kernel/cpu/mcheck/mce_amd_64.c b/arch/x86/kernel/cpu/mcheck/mce_amd_64.c index 8ae8c4ff094..4772e91e824 100644 --- a/arch/x86/kernel/cpu/mcheck/mce_amd_64.c +++ b/arch/x86/kernel/cpu/mcheck/mce_amd_64.c @@ -67,7 +67,7 @@ static struct threshold_block threshold_defaults = { struct threshold_bank { struct kobject *kobj; struct threshold_block *blocks; - cpumask_t cpus; + cpumask_var_t cpus; }; static DEFINE_PER_CPU(struct threshold_bank *, threshold_banks[NR_BANKS]); @@ -481,7 +481,7 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank) #ifdef CONFIG_SMP if (cpu_data(cpu).cpu_core_id && shared_bank[bank]) { /* symlink */ - i = first_cpu(per_cpu(cpu_core_map, cpu)); + i = cpumask_first(&per_cpu(cpu_core_map, cpu)); /* first core not up yet */ if (cpu_data(i).cpu_core_id) @@ -501,7 +501,7 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank) if (err) goto out; - b->cpus = per_cpu(cpu_core_map, cpu); + cpumask_copy(b->cpus, &per_cpu(cpu_core_map, cpu)); per_cpu(threshold_banks, cpu)[bank] = b; goto out; } @@ -512,15 +512,20 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank) err = -ENOMEM; goto out; } + if (!alloc_cpumask_var(&b->cpus, GFP_KERNEL)) { + kfree(b); + err = -ENOMEM; + goto out; + } b->kobj = kobject_create_and_add(name, &per_cpu(device_mce, cpu).kobj); if (!b->kobj) goto out_free; #ifndef CONFIG_SMP - b->cpus = CPU_MASK_ALL; + cpumask_setall(b->cpus); #else - b->cpus = per_cpu(cpu_core_map, cpu); + cpumask_copy(b->cpus, &per_cpu(cpu_core_map, cpu)); #endif per_cpu(threshold_banks, cpu)[bank] = b; @@ -529,7 +534,7 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank) if (err) goto out_free; - for_each_cpu_mask_nr(i, b->cpus) { + for_each_cpu(i, b->cpus) { if (i == cpu) continue; @@ -545,6 +550,7 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank) out_free: per_cpu(threshold_banks, cpu)[bank] = NULL; + free_cpumask_var(b->cpus); kfree(b); out: return err; @@ -619,7 +625,7 @@ static void threshold_remove_bank(unsigned int cpu, int bank) #endif /* remove all sibling symlinks before unregistering */ - for_each_cpu_mask_nr(i, b->cpus) { + for_each_cpu(i, b->cpus) { if (i == cpu) continue; @@ -632,6 +638,7 @@ static void threshold_remove_bank(unsigned int cpu, int bank) free_out: kobject_del(b->kobj); kobject_put(b->kobj); + free_cpumask_var(b->cpus); kfree(b); per_cpu(threshold_banks, cpu)[bank] = NULL; } diff --git a/arch/x86/kernel/cpu/mcheck/mce_intel_64.c b/arch/x86/kernel/cpu/mcheck/mce_intel_64.c index 4b48f251fd3..5e8c79e748a 100644 --- a/arch/x86/kernel/cpu/mcheck/mce_intel_64.c +++ b/arch/x86/kernel/cpu/mcheck/mce_intel_64.c @@ -7,6 +7,7 @@ #include <linux/interrupt.h> #include <linux/percpu.h> #include <asm/processor.h> +#include <asm/apic.h> #include <asm/msr.h> #include <asm/mce.h> #include <asm/hw_irq.h> diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c new file mode 100644 index 00000000000..383d4c6423a --- /dev/null +++ b/arch/x86/kernel/cpu/perf_counter.c @@ -0,0 +1,733 @@ +/* + * Performance counter x86 architecture code + * + * Copyright(C) 2008 Thomas Gleixner <tglx@linutronix.de> + * Copyright(C) 2008 Red Hat, Inc., Ingo Molnar + * + * For licencing details see kernel-base/COPYING + */ + +#include <linux/perf_counter.h> +#include <linux/capability.h> +#include <linux/notifier.h> +#include <linux/hardirq.h> +#include <linux/kprobes.h> +#include <linux/module.h> +#include <linux/kdebug.h> +#include <linux/sched.h> + +#include <asm/perf_counter.h> +#include <asm/apic.h> + +static bool perf_counters_initialized __read_mostly; + +/* + * Number of (generic) HW counters: + */ +static int nr_counters_generic __read_mostly; +static u64 perf_counter_mask __read_mostly; +static u64 counter_value_mask __read_mostly; + +static int nr_counters_fixed __read_mostly; + +struct cpu_hw_counters { + struct perf_counter *counters[X86_PMC_IDX_MAX]; + unsigned long used[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; + unsigned long interrupts; + u64 global_enable; +}; + +/* + * Intel PerfMon v3. Used on Core2 and later. + */ +static DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters); + +static const int intel_perfmon_event_map[] = +{ + [PERF_COUNT_CPU_CYCLES] = 0x003c, + [PERF_COUNT_INSTRUCTIONS] = 0x00c0, + [PERF_COUNT_CACHE_REFERENCES] = 0x4f2e, + [PERF_COUNT_CACHE_MISSES] = 0x412e, + [PERF_COUNT_BRANCH_INSTRUCTIONS] = 0x00c4, + [PERF_COUNT_BRANCH_MISSES] = 0x00c5, + [PERF_COUNT_BUS_CYCLES] = 0x013c, +}; + +static const int max_intel_perfmon_events = ARRAY_SIZE(intel_perfmon_event_map); + +/* + * Propagate counter elapsed time into the generic counter. + * Can only be executed on the CPU where the counter is active. + * Returns the delta events processed. + */ +static void +x86_perf_counter_update(struct perf_counter *counter, + struct hw_perf_counter *hwc, int idx) +{ + u64 prev_raw_count, new_raw_count, delta; + + /* + * Careful: an NMI might modify the previous counter value. + * + * Our tactic to handle this is to first atomically read and + * exchange a new raw count - then add that new-prev delta + * count to the generic counter atomically: + */ +again: + prev_raw_count = atomic64_read(&hwc->prev_count); + rdmsrl(hwc->counter_base + idx, new_raw_count); + + if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, + new_raw_count) != prev_raw_count) + goto again; + + /* + * Now we have the new raw value and have updated the prev + * timestamp already. We can now calculate the elapsed delta + * (counter-)time and add that to the generic counter. + * + * Careful, not all hw sign-extends above the physical width + * of the count, so we do that by clipping the delta to 32 bits: + */ + delta = (u64)(u32)((s32)new_raw_count - (s32)prev_raw_count); + + atomic64_add(delta, &counter->count); + atomic64_sub(delta, &hwc->period_left); +} + +/* + * Setup the hardware configuration for a given hw_event_type + */ +static int __hw_perf_counter_init(struct perf_counter *counter) +{ + struct perf_counter_hw_event *hw_event = &counter->hw_event; + struct hw_perf_counter *hwc = &counter->hw; + + if (unlikely(!perf_counters_initialized)) + return -EINVAL; + + /* + * Generate PMC IRQs: + * (keep 'enabled' bit clear for now) + */ + hwc->config = ARCH_PERFMON_EVENTSEL_INT; + + /* + * Count user and OS events unless requested not to. + */ + if (!hw_event->exclude_user) + hwc->config |= ARCH_PERFMON_EVENTSEL_USR; + if (!hw_event->exclude_kernel) + hwc->config |= ARCH_PERFMON_EVENTSEL_OS; + + /* + * If privileged enough, allow NMI events: + */ + hwc->nmi = 0; + if (capable(CAP_SYS_ADMIN) && hw_event->nmi) + hwc->nmi = 1; + + hwc->irq_period = hw_event->irq_period; + /* + * Intel PMCs cannot be accessed sanely above 32 bit width, + * so we install an artificial 1<<31 period regardless of + * the generic counter period: + */ + if ((s64)hwc->irq_period <= 0 || hwc->irq_period > 0x7FFFFFFF) + hwc->irq_period = 0x7FFFFFFF; + + atomic64_set(&hwc->period_left, hwc->irq_period); + + /* + * Raw event type provide the config in the event structure + */ + if (hw_event->raw) { + hwc->config |= hw_event->type; + } else { + if (hw_event->type >= max_intel_perfmon_events) + return -EINVAL; + /* + * The generic map: + */ + hwc->config |= intel_perfmon_event_map[hw_event->type]; + } + counter->wakeup_pending = 0; + + return 0; +} + +u64 hw_perf_save_disable(void) +{ + u64 ctrl; + + if (unlikely(!perf_counters_initialized)) + return 0; + + rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl); + wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0); + + return ctrl; +} +EXPORT_SYMBOL_GPL(hw_perf_save_disable); + +void hw_perf_restore(u64 ctrl) +{ + if (unlikely(!perf_counters_initialized)) + return; + + wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl); +} +EXPORT_SYMBOL_GPL(hw_perf_restore); + +static inline void +__pmc_fixed_disable(struct perf_counter *counter, + struct hw_perf_counter *hwc, unsigned int __idx) +{ + int idx = __idx - X86_PMC_IDX_FIXED; + u64 ctrl_val, mask; + int err; + + mask = 0xfULL << (idx * 4); + + rdmsrl(hwc->config_base, ctrl_val); + ctrl_val &= ~mask; + err = checking_wrmsrl(hwc->config_base, ctrl_val); +} + +static inline void +__pmc_generic_disable(struct perf_counter *counter, + struct hw_perf_counter *hwc, unsigned int idx) +{ + if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) + __pmc_fixed_disable(counter, hwc, idx); + else + wrmsr_safe(hwc->config_base + idx, hwc->config, 0); +} + +static DEFINE_PER_CPU(u64, prev_left[X86_PMC_IDX_MAX]); + +/* + * Set the next IRQ period, based on the hwc->period_left value. + * To be called with the counter disabled in hw: + */ +static void +__hw_perf_counter_set_period(struct perf_counter *counter, + struct hw_perf_counter *hwc, int idx) +{ + s64 left = atomic64_read(&hwc->period_left); + s32 period = hwc->irq_period; + int err; + + /* + * If we are way outside a reasoable range then just skip forward: + */ + if (unlikely(left <= -period)) { + left = period; + atomic64_set(&hwc->period_left, left); + } + + if (unlikely(left <= 0)) { + left += period; + atomic64_set(&hwc->period_left, left); + } + + per_cpu(prev_left[idx], smp_processor_id()) = left; + + /* + * The hw counter starts counting from this counter offset, + * mark it to be able to extra future deltas: + */ + atomic64_set(&hwc->prev_count, (u64)-left); + + err = checking_wrmsrl(hwc->counter_base + idx, + (u64)(-left) & counter_value_mask); +} + +static inline void +__pmc_fixed_enable(struct perf_counter *counter, + struct hw_perf_counter *hwc, unsigned int __idx) +{ + int idx = __idx - X86_PMC_IDX_FIXED; + u64 ctrl_val, bits, mask; + int err; + + /* + * Enable IRQ generation (0x8), + * and enable ring-3 counting (0x2) and ring-0 counting (0x1) + * if requested: + */ + bits = 0x8ULL; + if (hwc->config & ARCH_PERFMON_EVENTSEL_USR) + bits |= 0x2; + if (hwc->config & ARCH_PERFMON_EVENTSEL_OS) + bits |= 0x1; + bits <<= (idx * 4); + mask = 0xfULL << (idx * 4); + + rdmsrl(hwc->config_base, ctrl_val); + ctrl_val &= ~mask; + ctrl_val |= bits; + err = checking_wrmsrl(hwc->config_base, ctrl_val); +} + +static void +__pmc_generic_enable(struct perf_counter *counter, + struct hw_perf_counter *hwc, int idx) +{ + if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) + __pmc_fixed_enable(counter, hwc, idx); + else + wrmsr(hwc->config_base + idx, + hwc->config | ARCH_PERFMON_EVENTSEL0_ENABLE, 0); +} + +static int +fixed_mode_idx(struct perf_counter *counter, struct hw_perf_counter *hwc) +{ + unsigned int event; + + if (unlikely(hwc->nmi)) + return -1; + + event = hwc->config & ARCH_PERFMON_EVENT_MASK; + + if (unlikely(event == intel_perfmon_event_map[PERF_COUNT_INSTRUCTIONS])) + return X86_PMC_IDX_FIXED_INSTRUCTIONS; + if (unlikely(event == intel_perfmon_event_map[PERF_COUNT_CPU_CYCLES])) + return X86_PMC_IDX_FIXED_CPU_CYCLES; + if (unlikely(event == intel_perfmon_event_map[PERF_COUNT_BUS_CYCLES])) + return X86_PMC_IDX_FIXED_BUS_CYCLES; + + return -1; +} + +/* + * Find a PMC slot for the freshly enabled / scheduled in counter: + */ +static int pmc_generic_enable(struct perf_counter *counter) +{ + struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); + struct hw_perf_counter *hwc = &counter->hw; + int idx; + + idx = fixed_mode_idx(counter, hwc); + if (idx >= 0) { + /* + * Try to get the fixed counter, if that is already taken + * then try to get a generic counter: + */ + if (test_and_set_bit(idx, cpuc->used)) + goto try_generic; + + hwc->config_base = MSR_ARCH_PERFMON_FIXED_CTR_CTRL; + /* + * We set it so that counter_base + idx in wrmsr/rdmsr maps to + * MSR_ARCH_PERFMON_FIXED_CTR0 ... CTR2: + */ + hwc->counter_base = + MSR_ARCH_PERFMON_FIXED_CTR0 - X86_PMC_IDX_FIXED; + hwc->idx = idx; + } else { + idx = hwc->idx; + /* Try to get the previous generic counter again */ + if (test_and_set_bit(idx, cpuc->used)) { +try_generic: + idx = find_first_zero_bit(cpuc->used, nr_counters_generic); + if (idx == nr_counters_generic) + return -EAGAIN; + + set_bit(idx, cpuc->used); + hwc->idx = idx; + } + hwc->config_base = MSR_ARCH_PERFMON_EVENTSEL0; + hwc->counter_base = MSR_ARCH_PERFMON_PERFCTR0; + } + + perf_counters_lapic_init(hwc->nmi); + + __pmc_generic_disable(counter, hwc, idx); + + cpuc->counters[idx] = counter; + /* + * Make it visible before enabling the hw: + */ + smp_wmb(); + + __hw_perf_counter_set_period(counter, hwc, idx); + __pmc_generic_enable(counter, hwc, idx); + + return 0; +} + +void perf_counter_print_debug(void) +{ + u64 ctrl, status, overflow, pmc_ctrl, pmc_count, prev_left, fixed; + struct cpu_hw_counters *cpuc; + int cpu, idx; + + if (!nr_counters_generic) + return; + + local_irq_disable(); + + cpu = smp_processor_id(); + cpuc = &per_cpu(cpu_hw_counters, cpu); + + rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl); + rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status); + rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow); + rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR_CTRL, fixed); + + printk(KERN_INFO "\n"); + printk(KERN_INFO "CPU#%d: ctrl: %016llx\n", cpu, ctrl); + printk(KERN_INFO "CPU#%d: status: %016llx\n", cpu, status); + printk(KERN_INFO "CPU#%d: overflow: %016llx\n", cpu, overflow); + printk(KERN_INFO "CPU#%d: fixed: %016llx\n", cpu, fixed); + printk(KERN_INFO "CPU#%d: used: %016llx\n", cpu, *(u64 *)cpuc->used); + + for (idx = 0; idx < nr_counters_generic; idx++) { + rdmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx, pmc_ctrl); + rdmsrl(MSR_ARCH_PERFMON_PERFCTR0 + idx, pmc_count); + + prev_left = per_cpu(prev_left[idx], cpu); + + printk(KERN_INFO "CPU#%d: gen-PMC%d ctrl: %016llx\n", + cpu, idx, pmc_ctrl); + printk(KERN_INFO "CPU#%d: gen-PMC%d count: %016llx\n", + cpu, idx, pmc_count); + printk(KERN_INFO "CPU#%d: gen-PMC%d left: %016llx\n", + cpu, idx, prev_left); + } + for (idx = 0; idx < nr_counters_fixed; idx++) { + rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, pmc_count); + + printk(KERN_INFO "CPU#%d: fixed-PMC%d count: %016llx\n", + cpu, idx, pmc_count); + } + local_irq_enable(); +} + +static void pmc_generic_disable(struct perf_counter *counter) +{ + struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); + struct hw_perf_counter *hwc = &counter->hw; + unsigned int idx = hwc->idx; + + __pmc_generic_disable(counter, hwc, idx); + + clear_bit(idx, cpuc->used); + cpuc->counters[idx] = NULL; + /* + * Make sure the cleared pointer becomes visible before we + * (potentially) free the counter: + */ + smp_wmb(); + + /* + * Drain the remaining delta count out of a counter + * that we are disabling: + */ + x86_perf_counter_update(counter, hwc, idx); +} + +static void perf_store_irq_data(struct perf_counter *counter, u64 data) +{ + struct perf_data *irqdata = counter->irqdata; + + if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) { + irqdata->overrun++; + } else { + u64 *p = (u64 *) &irqdata->data[irqdata->len]; + + *p = data; + irqdata->len += sizeof(u64); + } +} + +/* + * Save and restart an expired counter. Called by NMI contexts, + * so it has to be careful about preempting normal counter ops: + */ +static void perf_save_and_restart(struct perf_counter *counter) +{ + struct hw_perf_counter *hwc = &counter->hw; + int idx = hwc->idx; + + x86_perf_counter_update(counter, hwc, idx); + __hw_perf_counter_set_period(counter, hwc, idx); + + if (counter->state == PERF_COUNTER_STATE_ACTIVE) + __pmc_generic_enable(counter, hwc, idx); +} + +static void +perf_handle_group(struct perf_counter *sibling, u64 *status, u64 *overflown) +{ + struct perf_counter *counter, *group_leader = sibling->group_leader; + + /* + * Store sibling timestamps (if any): + */ + list_for_each_entry(counter, &group_leader->sibling_list, list_entry) { + + x86_perf_counter_update(counter, &counter->hw, counter->hw.idx); + perf_store_irq_data(sibling, counter->hw_event.type); + perf_store_irq_data(sibling, atomic64_read(&counter->count)); + } +} + +/* + * Maximum interrupt frequency of 100KHz per CPU + */ +#define PERFMON_MAX_INTERRUPTS 100000/HZ + +/* + * This handler is triggered by the local APIC, so the APIC IRQ handling + * rules apply: + */ +static void __smp_perf_counter_interrupt(struct pt_regs *regs, int nmi) +{ + int bit, cpu = smp_processor_id(); + u64 ack, status; + struct cpu_hw_counters *cpuc = &per_cpu(cpu_hw_counters, cpu); + + rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, cpuc->global_enable); + + /* Disable counters globally */ + wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0); + ack_APIC_irq(); + + rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status); + if (!status) + goto out; + +again: + inc_irq_stat(apic_perf_irqs); + ack = status; + for_each_bit(bit, (unsigned long *)&status, X86_PMC_IDX_MAX) { + struct perf_counter *counter = cpuc->counters[bit]; + + clear_bit(bit, (unsigned long *) &status); + if (!counter) + continue; + + perf_save_and_restart(counter); + + switch (counter->hw_event.record_type) { + case PERF_RECORD_SIMPLE: + continue; + case PERF_RECORD_IRQ: + perf_store_irq_data(counter, instruction_pointer(regs)); + break; + case PERF_RECORD_GROUP: + perf_handle_group(counter, &status, &ack); + break; + } + /* + * From NMI context we cannot call into the scheduler to + * do a task wakeup - but we mark these generic as + * wakeup_pending and initate a wakeup callback: + */ + if (nmi) { + counter->wakeup_pending = 1; + set_tsk_thread_flag(current, TIF_PERF_COUNTERS); + } else { + wake_up(&counter->waitq); + } + } + + wrmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack); + + /* + * Repeat if there is more work to be done: + */ + rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status); + if (status) + goto again; +out: + /* + * Restore - do not reenable when global enable is off or throttled: + */ + if (++cpuc->interrupts < PERFMON_MAX_INTERRUPTS) + wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, cpuc->global_enable); +} + +void perf_counter_unthrottle(void) +{ + struct cpu_hw_counters *cpuc; + u64 global_enable; + + if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) + return; + + if (unlikely(!perf_counters_initialized)) + return; + + cpuc = &per_cpu(cpu_hw_counters, smp_processor_id()); + if (cpuc->interrupts >= PERFMON_MAX_INTERRUPTS) { + if (printk_ratelimit()) + printk(KERN_WARNING "PERFMON: max interrupts exceeded!\n"); + wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, cpuc->global_enable); + } + rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, global_enable); + if (unlikely(cpuc->global_enable && !global_enable)) + wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, cpuc->global_enable); + cpuc->interrupts = 0; +} + +void smp_perf_counter_interrupt(struct pt_regs *regs) +{ + irq_enter(); + apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR); + __smp_perf_counter_interrupt(regs, 0); + + irq_exit(); +} + +/* + * This handler is triggered by NMI contexts: + */ +void perf_counter_notify(struct pt_regs *regs) +{ + struct cpu_hw_counters *cpuc; + unsigned long flags; + int bit, cpu; + + local_irq_save(flags); + cpu = smp_processor_id(); + cpuc = &per_cpu(cpu_hw_counters, cpu); + + for_each_bit(bit, cpuc->used, X86_PMC_IDX_MAX) { + struct perf_counter *counter = cpuc->counters[bit]; + + if (!counter) + continue; + + if (counter->wakeup_pending) { + counter->wakeup_pending = 0; + wake_up(&counter->waitq); + } + } + + local_irq_restore(flags); +} + +void perf_counters_lapic_init(int nmi) +{ + u32 apic_val; + + if (!perf_counters_initialized) + return; + /* + * Enable the performance counter vector in the APIC LVT: + */ + apic_val = apic_read(APIC_LVTERR); + + apic_write(APIC_LVTERR, apic_val | APIC_LVT_MASKED); + if (nmi) + apic_write(APIC_LVTPC, APIC_DM_NMI); + else + apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR); + apic_write(APIC_LVTERR, apic_val); +} + +static int __kprobes +perf_counter_nmi_handler(struct notifier_block *self, + unsigned long cmd, void *__args) +{ + struct die_args *args = __args; + struct pt_regs *regs; + + if (likely(cmd != DIE_NMI_IPI)) + return NOTIFY_DONE; + + regs = args->regs; + + apic_write(APIC_LVTPC, APIC_DM_NMI); + __smp_perf_counter_interrupt(regs, 1); + + return NOTIFY_STOP; +} + +static __read_mostly struct notifier_block perf_counter_nmi_notifier = { + .notifier_call = perf_counter_nmi_handler, + .next = NULL, + .priority = 1 +}; + +void __init init_hw_perf_counters(void) +{ + union cpuid10_eax eax; + unsigned int ebx; + unsigned int unused; + union cpuid10_edx edx; + + if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) + return; + + /* + * Check whether the Architectural PerfMon supports + * Branch Misses Retired Event or not. + */ + cpuid(10, &eax.full, &ebx, &unused, &edx.full); + if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED) + return; + + printk(KERN_INFO "Intel Performance Monitoring support detected.\n"); + + printk(KERN_INFO "... version: %d\n", eax.split.version_id); + printk(KERN_INFO "... num counters: %d\n", eax.split.num_counters); + nr_counters_generic = eax.split.num_counters; + if (nr_counters_generic > X86_PMC_MAX_GENERIC) { + nr_counters_generic = X86_PMC_MAX_GENERIC; + WARN(1, KERN_ERR "hw perf counters %d > max(%d), clipping!", + nr_counters_generic, X86_PMC_MAX_GENERIC); + } + perf_counter_mask = (1 << nr_counters_generic) - 1; + perf_max_counters = nr_counters_generic; + + printk(KERN_INFO "... bit width: %d\n", eax.split.bit_width); + counter_value_mask = (1ULL << eax.split.bit_width) - 1; + printk(KERN_INFO "... value mask: %016Lx\n", counter_value_mask); + + printk(KERN_INFO "... mask length: %d\n", eax.split.mask_length); + + nr_counters_fixed = edx.split.num_counters_fixed; + if (nr_counters_fixed > X86_PMC_MAX_FIXED) { + nr_counters_fixed = X86_PMC_MAX_FIXED; + WARN(1, KERN_ERR "hw perf counters fixed %d > max(%d), clipping!", + nr_counters_fixed, X86_PMC_MAX_FIXED); + } + printk(KERN_INFO "... fixed counters: %d\n", nr_counters_fixed); + + perf_counter_mask |= ((1LL << nr_counters_fixed)-1) << X86_PMC_IDX_FIXED; + + printk(KERN_INFO "... counter mask: %016Lx\n", perf_counter_mask); + perf_counters_initialized = true; + + perf_counters_lapic_init(0); + register_die_notifier(&perf_counter_nmi_notifier); +} + +static void pmc_generic_read(struct perf_counter *counter) +{ + x86_perf_counter_update(counter, &counter->hw, counter->hw.idx); +} + +static const struct hw_perf_counter_ops x86_perf_counter_ops = { + .enable = pmc_generic_enable, + .disable = pmc_generic_disable, + .read = pmc_generic_read, +}; + +const struct hw_perf_counter_ops * +hw_perf_counter_init(struct perf_counter *counter) +{ + int err; + + err = __hw_perf_counter_init(counter); + if (err) + return NULL; + + return &x86_perf_counter_ops; +} diff --git a/arch/x86/kernel/cpu/perfctr-watchdog.c b/arch/x86/kernel/cpu/perfctr-watchdog.c index 9abd48b2267..d6f5b9fbde3 100644 --- a/arch/x86/kernel/cpu/perfctr-watchdog.c +++ b/arch/x86/kernel/cpu/perfctr-watchdog.c @@ -20,7 +20,7 @@ #include <linux/kprobes.h> #include <asm/apic.h> -#include <asm/intel_arch_perfmon.h> +#include <asm/perf_counter.h> struct nmi_watchdog_ctlblk { unsigned int cccr_msr; |