From 11d1578f9454159c43499d1d8fe8a7d728c176a3 Mon Sep 17 00:00:00 2001 From: Vince Weaver <vince@deater.net> Date: Wed, 8 Jul 2009 17:46:14 -0400 Subject: perf_counter: Add P6 PMU support Add basic P6 PMU support. The P6 uses the EVNTSEL0 EN bit to enable/disable both its counters. We use this for the global enable/disable, and clear all config bits (except EN) to disable individual counters. Actual ia32 hardware doesn't support lfence, so use a locked op without side-effect to implement a full barrier. perf stat and perf record seem to function correctly. [a.p.zijlstra@chello.nl: cleanups and complete the enable/disable code] Signed-off-by: Vince Weaver <vince@deater.net> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <Pine.LNX.4.64.0907081718450.2715@pianoman.cluster.toy> Signed-off-by: Ingo Molnar <mingo@elte.hu> --- arch/x86/kernel/cpu/perf_counter.c | 227 ++++++++++++++++++++++++++++++++++--- 1 file changed, 213 insertions(+), 14 deletions(-) (limited to 'arch/x86') diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c index 36c3dc7b8991..1910f39ff19a 100644 --- a/arch/x86/kernel/cpu/perf_counter.c +++ b/arch/x86/kernel/cpu/perf_counter.c @@ -65,6 +65,44 @@ static DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters) = { .enabled = 1, }; +/* + * Not sure about some of these + */ +static const u64 p6_perfmon_event_map[] = +{ + [PERF_COUNT_HW_CPU_CYCLES] = 0x0079, + [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, + [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0000, + [PERF_COUNT_HW_CACHE_MISSES] = 0x0000, + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, + [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, + [PERF_COUNT_HW_BUS_CYCLES] = 0x0062, +}; + +static u64 p6_pmu_event_map(int event) +{ + return p6_perfmon_event_map[event]; +} + +static u64 p6_pmu_raw_event(u64 event) +{ +#define P6_EVNTSEL_EVENT_MASK 0x000000FFULL +#define P6_EVNTSEL_UNIT_MASK 0x0000FF00ULL +#define P6_EVNTSEL_EDGE_MASK 0x00040000ULL +#define P6_EVNTSEL_INV_MASK 0x00800000ULL +#define P6_EVNTSEL_COUNTER_MASK 0xFF000000ULL + +#define P6_EVNTSEL_MASK \ + (P6_EVNTSEL_EVENT_MASK | \ + P6_EVNTSEL_UNIT_MASK | \ + P6_EVNTSEL_EDGE_MASK | \ + P6_EVNTSEL_INV_MASK | \ + P6_EVNTSEL_COUNTER_MASK) + + return event & P6_EVNTSEL_MASK; +} + + /* * Intel PerfMon v3. Used on Core2 and later. */ @@ -726,6 +764,23 @@ static int __hw_perf_counter_init(struct perf_counter *counter) return 0; } +static void p6_pmu_disable_all(void) +{ + struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); + unsigned long val; + + if (!cpuc->enabled) + return; + + cpuc->enabled = 0; + barrier(); + + /* p6 only has one enable register */ + rdmsrl(MSR_P6_EVNTSEL0, val); + val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE; + wrmsrl(MSR_P6_EVNTSEL0, val); +} + static void intel_pmu_disable_all(void) { wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0); @@ -767,6 +822,23 @@ void hw_perf_disable(void) return x86_pmu.disable_all(); } +static void p6_pmu_enable_all(void) +{ + struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); + unsigned long val; + + if (cpuc->enabled) + return; + + cpuc->enabled = 1; + barrier(); + + /* p6 only has one enable register */ + rdmsrl(MSR_P6_EVNTSEL0, val); + val |= ARCH_PERFMON_EVENTSEL0_ENABLE; + wrmsrl(MSR_P6_EVNTSEL0, val); +} + static void intel_pmu_enable_all(void) { wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl); @@ -819,16 +891,13 @@ static inline void intel_pmu_ack_status(u64 ack) static inline void x86_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) { - int err; - err = checking_wrmsrl(hwc->config_base + idx, + (void)checking_wrmsrl(hwc->config_base + idx, hwc->config | ARCH_PERFMON_EVENTSEL0_ENABLE); } static inline void x86_pmu_disable_counter(struct hw_perf_counter *hwc, int idx) { - int err; - err = checking_wrmsrl(hwc->config_base + idx, - hwc->config); + (void)checking_wrmsrl(hwc->config_base + idx, hwc->config); } static inline void @@ -836,13 +905,24 @@ intel_pmu_disable_fixed(struct hw_perf_counter *hwc, 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); + (void)checking_wrmsrl(hwc->config_base, ctrl_val); +} + +static inline void +p6_pmu_disable_counter(struct hw_perf_counter *hwc, int idx) +{ + struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); + unsigned long val = ARCH_PERFMON_EVENTSEL0_ENABLE; + + if (!cpuc->enabled) + val = 0; + + (void)checking_wrmsrl(hwc->config_base + idx, val); } static inline void @@ -943,6 +1023,17 @@ intel_pmu_enable_fixed(struct hw_perf_counter *hwc, int __idx) err = checking_wrmsrl(hwc->config_base, ctrl_val); } +static void p6_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) +{ + struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); + + if (cpuc->enabled) + x86_pmu_enable_counter(hwc, idx); + else + x86_pmu_disable_counter(hwc, idx); +} + + static void intel_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) { if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) { @@ -1176,6 +1267,49 @@ static void intel_pmu_reset(void) local_irq_restore(flags); } +static int p6_pmu_handle_irq(struct pt_regs *regs) +{ + struct perf_sample_data data; + struct cpu_hw_counters *cpuc; + struct perf_counter *counter; + struct hw_perf_counter *hwc; + int idx, handled = 0; + u64 val; + + data.regs = regs; + data.addr = 0; + + cpuc = &__get_cpu_var(cpu_hw_counters); + + for (idx = 0; idx < x86_pmu.num_counters; idx++) { + if (!test_bit(idx, cpuc->active_mask)) + continue; + + counter = cpuc->counters[idx]; + hwc = &counter->hw; + + val = x86_perf_counter_update(counter, hwc, idx); + if (val & (1ULL << (x86_pmu.counter_bits - 1))) + continue; + + /* + * counter overflow + */ + handled = 1; + data.period = counter->hw.last_period; + + if (!x86_perf_counter_set_period(counter, hwc, idx)) + continue; + + if (perf_counter_overflow(counter, 1, &data)) + p6_pmu_disable_counter(hwc, idx); + } + + if (handled) + inc_irq_stat(apic_perf_irqs); + + return handled; +} /* * This handler is triggered by the local APIC, so the APIC IRQ handling @@ -1185,14 +1319,13 @@ static int intel_pmu_handle_irq(struct pt_regs *regs) { struct perf_sample_data data; struct cpu_hw_counters *cpuc; - int bit, cpu, loops; + int bit, loops; u64 ack, status; data.regs = regs; data.addr = 0; - cpu = smp_processor_id(); - cpuc = &per_cpu(cpu_hw_counters, cpu); + cpuc = &__get_cpu_var(cpu_hw_counters); perf_disable(); status = intel_pmu_get_status(); @@ -1249,14 +1382,13 @@ static int amd_pmu_handle_irq(struct pt_regs *regs) struct cpu_hw_counters *cpuc; struct perf_counter *counter; struct hw_perf_counter *hwc; - int cpu, idx, handled = 0; + int idx, handled = 0; u64 val; data.regs = regs; data.addr = 0; - cpu = smp_processor_id(); - cpuc = &per_cpu(cpu_hw_counters, cpu); + cpuc = &__get_cpu_var(cpu_hw_counters); for (idx = 0; idx < x86_pmu.num_counters; idx++) { if (!test_bit(idx, cpuc->active_mask)) @@ -1353,6 +1485,32 @@ static __read_mostly struct notifier_block perf_counter_nmi_notifier = { .priority = 1 }; +static struct x86_pmu p6_pmu = { + .name = "p6", + .handle_irq = p6_pmu_handle_irq, + .disable_all = p6_pmu_disable_all, + .enable_all = p6_pmu_enable_all, + .enable = p6_pmu_enable_counter, + .disable = p6_pmu_disable_counter, + .eventsel = MSR_P6_EVNTSEL0, + .perfctr = MSR_P6_PERFCTR0, + .event_map = p6_pmu_event_map, + .raw_event = p6_pmu_raw_event, + .max_events = ARRAY_SIZE(p6_perfmon_event_map), + .max_period = (1ULL << 31) - 1, + .version = 0, + .num_counters = 2, + /* + * Counters have 40 bits implemented. However they are designed such + * that bits [32-39] are sign extensions of bit 31. As such the + * effective width of a counter for P6-like PMU is 32 bits only. + * + * See IA-32 Intel Architecture Software developer manual Vol 3B + */ + .counter_bits = 32, + .counter_mask = (1ULL << 32) - 1, +}; + static struct x86_pmu intel_pmu = { .name = "Intel", .handle_irq = intel_pmu_handle_irq, @@ -1392,6 +1550,41 @@ static struct x86_pmu amd_pmu = { .max_period = (1ULL << 47) - 1, }; +static int p6_pmu_init(void) +{ + int high, low; + + switch (boot_cpu_data.x86_model) { + case 1: + case 3: /* Pentium Pro */ + case 5: + case 6: /* Pentium II */ + case 7: + case 8: + case 11: /* Pentium III */ + break; + case 9: + case 13: + /* for Pentium M, we need to check if PMU exist */ + rdmsr(MSR_IA32_MISC_ENABLE, low, high); + if (low & MSR_IA32_MISC_ENABLE_EMON) + break; + default: + pr_cont("unsupported p6 CPU model %d ", + boot_cpu_data.x86_model); + return -ENODEV; + } + + if (!cpu_has_apic) { + pr_info("no Local APIC, try rebooting with lapic"); + return -ENODEV; + } + + x86_pmu = p6_pmu; + + return 0; +} + static int intel_pmu_init(void) { union cpuid10_edx edx; @@ -1400,8 +1593,14 @@ static int intel_pmu_init(void) unsigned int ebx; int version; - if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) + if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) { + /* check for P6 processor family */ + if (boot_cpu_data.x86 == 6) { + return p6_pmu_init(); + } else { return -ENODEV; + } + } /* * Check whether the Architectural PerfMon supports -- cgit v1.2.3 From 9c74fb50867e8fb5f3be3be06716492c0f79309e Mon Sep 17 00:00:00 2001 From: Peter Zijlstra <a.p.zijlstra@chello.nl> Date: Wed, 8 Jul 2009 10:21:41 +0200 Subject: perf_counter: Fix up P6 PMU details The P6 doesn't seem to support cache ref/hit/miss counts, so we extend the generic hardware event codes to have 0 and -1 mean the same thing as for the generic cache events. Furthermore, it turns out the 0 event does not count (that is, its reported that on PPro it actually does count something), therefore use a event configuration that's specified not to count to disable the counters. Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu> --- arch/x86/kernel/cpu/perf_counter.c | 28 +++++++++++++++++++++++----- 1 file changed, 23 insertions(+), 5 deletions(-) (limited to 'arch/x86') diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c index 1910f39ff19a..c7cc6eac14ec 100644 --- a/arch/x86/kernel/cpu/perf_counter.c +++ b/arch/x86/kernel/cpu/perf_counter.c @@ -84,6 +84,14 @@ static u64 p6_pmu_event_map(int event) return p6_perfmon_event_map[event]; } +/* + * Counter setting that is specified not to count anything. + * We use this to effectively disable a counter. + * + * L2_RQSTS with 0 MESI unit mask. + */ +#define P6_NOP_COUNTER 0x0000002EULL + static u64 p6_pmu_raw_event(u64 event) { #define P6_EVNTSEL_EVENT_MASK 0x000000FFULL @@ -704,6 +712,7 @@ static int __hw_perf_counter_init(struct perf_counter *counter) { struct perf_counter_attr *attr = &counter->attr; struct hw_perf_counter *hwc = &counter->hw; + u64 config; int err; if (!x86_pmu_initialized()) @@ -756,10 +765,19 @@ static int __hw_perf_counter_init(struct perf_counter *counter) if (attr->config >= x86_pmu.max_events) return -EINVAL; + /* * The generic map: */ - hwc->config |= x86_pmu.event_map(attr->config); + config = x86_pmu.event_map(attr->config); + + if (config == 0) + return -ENOENT; + + if (config == -1LL) + return -EINVAL; + + hwc->config |= config; return 0; } @@ -767,7 +785,7 @@ static int __hw_perf_counter_init(struct perf_counter *counter) static void p6_pmu_disable_all(void) { struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - unsigned long val; + u64 val; if (!cpuc->enabled) return; @@ -917,10 +935,10 @@ static inline void p6_pmu_disable_counter(struct hw_perf_counter *hwc, int idx) { struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - unsigned long val = ARCH_PERFMON_EVENTSEL0_ENABLE; + u64 val = P6_NOP_COUNTER; - if (!cpuc->enabled) - val = 0; + if (cpuc->enabled) + val |= ARCH_PERFMON_EVENTSEL0_ENABLE; (void)checking_wrmsrl(hwc->config_base + idx, val); } -- cgit v1.2.3 From 984b838ce69c063a91b87550598ab7f3439dd94a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra <a.p.zijlstra@chello.nl> Date: Fri, 10 Jul 2009 09:59:56 +0200 Subject: perf_counter: Clean up global vs counter enable Ingo noticed that both AMD and P6 call x86_pmu_disable_counter() on *_pmu_enable_counter(). This is because we rely on the side effect of that call to program the event config but not touch the EN bit. We change that for AMD by having enable_all() simply write the full config in, and for P6 by explicitly coding it. Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu> --- arch/x86/kernel/cpu/perf_counter.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) (limited to 'arch/x86') diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c index c7cc6eac14ec..bed1c4c2f251 100644 --- a/arch/x86/kernel/cpu/perf_counter.c +++ b/arch/x86/kernel/cpu/perf_counter.c @@ -874,13 +874,13 @@ static void amd_pmu_enable_all(void) barrier(); for (idx = 0; idx < x86_pmu.num_counters; idx++) { + struct perf_counter *counter = cpuc->counters[idx]; u64 val; if (!test_bit(idx, cpuc->active_mask)) continue; - rdmsrl(MSR_K7_EVNTSEL0 + idx, val); - if (val & ARCH_PERFMON_EVENTSEL0_ENABLE) - continue; + + val = counter->hw.config; val |= ARCH_PERFMON_EVENTSEL0_ENABLE; wrmsrl(MSR_K7_EVNTSEL0 + idx, val); } @@ -1044,11 +1044,13 @@ intel_pmu_enable_fixed(struct hw_perf_counter *hwc, int __idx) static void p6_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) { struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); + u64 val; + val = hwc->config; if (cpuc->enabled) - x86_pmu_enable_counter(hwc, idx); - else - x86_pmu_disable_counter(hwc, idx); + val |= ARCH_PERFMON_EVENTSEL0_ENABLE; + + (void)checking_wrmsrl(hwc->config_base + idx, val); } @@ -1068,8 +1070,6 @@ static void amd_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) if (cpuc->enabled) x86_pmu_enable_counter(hwc, idx); - else - x86_pmu_disable_counter(hwc, idx); } static int -- cgit v1.2.3 From f1c6a58121f9846ac665b0fbd3cbab90ce8bcbac Mon Sep 17 00:00:00 2001 From: Daniel Qarras <dqarras@yahoo.com> Date: Sun, 12 Jul 2009 04:32:40 -0700 Subject: perf_counter, x86: Extend perf_counter Pentium M support I've attached a patch to remove the Pentium M special casing of EMON and as noticed at least with my Pentium M the hardware PMU now works: Performance counter stats for '/bin/ls /var/tmp': 1.809988 task-clock-msecs # 0.125 CPUs 1 context-switches # 0.001 M/sec 0 CPU-migrations # 0.000 M/sec 224 page-faults # 0.124 M/sec 1425648 cycles # 787.656 M/sec 912755 instructions # 0.640 IPC Vince suggested that this code was trying to address erratum Y17 in Pentium-M's: http://download.intel.com/support/processors/mobile/pm/sb/25266532.pdf But that erratum (related to IA32_MISC_ENABLES.7) does not affect perfcounters as we dont use this toggle to disable RDPMC and WRMSR/RDMSR access to performance counters. We keep cr4's bit 8 (X86_CR4_PCE) clear so unprivileged RDPMC access is not allowed anyway. Cc: Vince Weaver <vince@deater.net> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@googlemail.com> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu> --- arch/x86/kernel/cpu/perf_counter.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'arch/x86') diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c index bed1c4c2f251..7e346d4bc0fb 100644 --- a/arch/x86/kernel/cpu/perf_counter.c +++ b/arch/x86/kernel/cpu/perf_counter.c @@ -1583,10 +1583,8 @@ static int p6_pmu_init(void) break; case 9: case 13: - /* for Pentium M, we need to check if PMU exist */ - rdmsr(MSR_IA32_MISC_ENABLE, low, high); - if (low & MSR_IA32_MISC_ENABLE_EMON) - break; + /* Pentium M */ + break; default: pr_cont("unsupported p6 CPU model %d ", boot_cpu_data.x86_model); -- cgit v1.2.3 From 9b7019ae6a542a801a80df6694c66e240e2c3e39 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra <a.p.zijlstra@chello.nl> Date: Wed, 22 Jul 2009 16:31:36 +0200 Subject: perf_counter: Remove unused variables Fix a gcc unused variables warning. Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> --- arch/x86/kernel/cpu/perf_counter.c | 2 -- 1 file changed, 2 deletions(-) (limited to 'arch/x86') diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c index 7e346d4bc0fb..a7aa8f900954 100644 --- a/arch/x86/kernel/cpu/perf_counter.c +++ b/arch/x86/kernel/cpu/perf_counter.c @@ -1570,8 +1570,6 @@ static struct x86_pmu amd_pmu = { static int p6_pmu_init(void) { - int high, low; - switch (boot_cpu_data.x86_model) { case 1: case 3: /* Pentium Pro */ -- cgit v1.2.3