From 35edc2a5095efb189e60dc32bbb9d2663aec6d24 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sun, 20 Nov 2011 20:36:02 +0100 Subject: perf, arch: Rework perf_event_index() Put the logic to compute the event index into a per pmu method. This is required because the x86 rules are weird and wonderful and don't match the capabilities of the current scheme. AFAIK only powerpc actually has a usable userspace read of the PMCs but I'm not at all sure anybody actually used that. ARM is restored to the default since it currently does not support userspace access at all. And all software events are provided with a method that reports their index as 0 (disabled). Signed-off-by: Peter Zijlstra Cc: Michael Cree Cc: Will Deacon Cc: Deng-Cheng Zhu Cc: Anton Blanchard Cc: Eric B Munson Cc: Heiko Carstens Cc: Paul Mundt Cc: David S. Miller Cc: Richard Kuo Cc: Stephane Eranian Cc: Arun Sharma Link: http://lkml.kernel.org/n/tip-dfydxodki16lylkt3gl2j7cw@git.kernel.org Signed-off-by: Ingo Molnar --- arch/arm/include/asm/perf_event.h | 4 ---- arch/frv/include/asm/perf_event.h | 2 -- arch/hexagon/include/asm/perf_event.h | 2 -- arch/powerpc/include/asm/perf_event_server.h | 2 -- arch/powerpc/kernel/perf_event.c | 6 ++++++ arch/s390/include/asm/perf_event.h | 1 - arch/x86/include/asm/perf_event.h | 2 -- 7 files changed, 6 insertions(+), 13 deletions(-) (limited to 'arch') diff --git a/arch/arm/include/asm/perf_event.h b/arch/arm/include/asm/perf_event.h index 0f8e3827a89b..08f94d8fc04c 100644 --- a/arch/arm/include/asm/perf_event.h +++ b/arch/arm/include/asm/perf_event.h @@ -12,10 +12,6 @@ #ifndef __ARM_PERF_EVENT_H__ #define __ARM_PERF_EVENT_H__ -/* ARM performance counters start from 1 (in the cp15 accesses) so use the - * same indexes here for consistency. */ -#define PERF_EVENT_INDEX_OFFSET 1 - /* ARM perf PMU IDs for use by internal perf clients. */ enum arm_perf_pmu_ids { ARM_PERF_PMU_ID_XSCALE1 = 0, diff --git a/arch/frv/include/asm/perf_event.h b/arch/frv/include/asm/perf_event.h index a69e0155d146..c52ea5546b5b 100644 --- a/arch/frv/include/asm/perf_event.h +++ b/arch/frv/include/asm/perf_event.h @@ -12,6 +12,4 @@ #ifndef _ASM_PERF_EVENT_H #define _ASM_PERF_EVENT_H -#define PERF_EVENT_INDEX_OFFSET 0 - #endif /* _ASM_PERF_EVENT_H */ diff --git a/arch/hexagon/include/asm/perf_event.h b/arch/hexagon/include/asm/perf_event.h index 6c2910f91180..8b8526b491c7 100644 --- a/arch/hexagon/include/asm/perf_event.h +++ b/arch/hexagon/include/asm/perf_event.h @@ -19,6 +19,4 @@ #ifndef _ASM_PERF_EVENT_H #define _ASM_PERF_EVENT_H -#define PERF_EVENT_INDEX_OFFSET 0 - #endif /* _ASM_PERF_EVENT_H */ diff --git a/arch/powerpc/include/asm/perf_event_server.h b/arch/powerpc/include/asm/perf_event_server.h index 8f1df1208d23..1a8093fa8f71 100644 --- a/arch/powerpc/include/asm/perf_event_server.h +++ b/arch/powerpc/include/asm/perf_event_server.h @@ -61,8 +61,6 @@ struct pt_regs; extern unsigned long perf_misc_flags(struct pt_regs *regs); extern unsigned long perf_instruction_pointer(struct pt_regs *regs); -#define PERF_EVENT_INDEX_OFFSET 1 - /* * Only override the default definitions in include/linux/perf_event.h * if we have hardware PMU support. diff --git a/arch/powerpc/kernel/perf_event.c b/arch/powerpc/kernel/perf_event.c index 10a140f82cb8..d614ab57ccca 100644 --- a/arch/powerpc/kernel/perf_event.c +++ b/arch/powerpc/kernel/perf_event.c @@ -1187,6 +1187,11 @@ static int power_pmu_event_init(struct perf_event *event) return err; } +static int power_pmu_event_idx(struct perf_event *event) +{ + return event->hw.idx; +} + struct pmu power_pmu = { .pmu_enable = power_pmu_enable, .pmu_disable = power_pmu_disable, @@ -1199,6 +1204,7 @@ struct pmu power_pmu = { .start_txn = power_pmu_start_txn, .cancel_txn = power_pmu_cancel_txn, .commit_txn = power_pmu_commit_txn, + .event_idx = power_pmu_event_idx, }; /* diff --git a/arch/s390/include/asm/perf_event.h b/arch/s390/include/asm/perf_event.h index a75f168d2718..4eb444edbe49 100644 --- a/arch/s390/include/asm/perf_event.h +++ b/arch/s390/include/asm/perf_event.h @@ -6,4 +6,3 @@ /* Empty, just to avoid compiling error */ -#define PERF_EVENT_INDEX_OFFSET 0 diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h index 096c975e099f..9b922c136254 100644 --- a/arch/x86/include/asm/perf_event.h +++ b/arch/x86/include/asm/perf_event.h @@ -188,8 +188,6 @@ extern u32 get_ibs_caps(void); #ifdef CONFIG_PERF_EVENTS extern void perf_events_lapic_init(void); -#define PERF_EVENT_INDEX_OFFSET 0 - /* * Abuse bit 3 of the cpu eflags register to indicate proper PEBS IP fixups. * This flag is otherwise unused and ABI specified to be 0, so nobody should -- cgit v1.2.3 From fe4a330885aee20f233de36085fb15c38094e635 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sun, 20 Nov 2011 20:44:06 +0100 Subject: perf, x86: Implement user-space RDPMC support, to allow fast, user-space access to self-monitoring counters Implement a correct pmu::event_idx for the x86 counter index rules and set CR4.PCE on CPU_STARTING. Signed-off-by: Peter Zijlstra Cc: Stephane Eranian Cc: Arun Sharma Cc: Thomas Gleixner Cc: "H. Peter Anvin" Link: http://lkml.kernel.org/n/tip-mwxab34dibqgzk5zywutfnha@git.kernel.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 5adce1040b11..53b569910175 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -1210,6 +1210,7 @@ x86_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu) break; case CPU_STARTING: + set_in_cr4(X86_CR4_PCE); if (x86_pmu.cpu_starting) x86_pmu.cpu_starting(cpu); break; @@ -1542,6 +1543,18 @@ static int x86_pmu_event_init(struct perf_event *event) return err; } +static int x86_pmu_event_idx(struct perf_event *event) +{ + int idx = event->hw.idx; + + if (x86_pmu.num_counters_fixed && idx >= X86_PMC_IDX_FIXED) { + idx -= X86_PMC_IDX_FIXED; + idx |= 1 << 30; + } + + return idx + 1; +} + static struct pmu pmu = { .pmu_enable = x86_pmu_enable, .pmu_disable = x86_pmu_disable, @@ -1557,6 +1570,8 @@ static struct pmu pmu = { .start_txn = x86_pmu_start_txn, .cancel_txn = x86_pmu_cancel_txn, .commit_txn = x86_pmu_commit_txn, + + .event_idx = x86_pmu_event_idx, }; /* -- cgit v1.2.3 From 0c9d42ed4cee2aa1dfc3a260b741baae8615744f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sun, 20 Nov 2011 23:30:47 +0100 Subject: perf, x86: Provide means for disabling userspace RDPMC Allow the disabling of RDPMC via a pmu specific attribute: echo 0 > /sys/bus/event_source/devices/cpu/rdpmc Signed-off-by: Peter Zijlstra Cc: Stephane Eranian Cc: Arun Sharma Link: http://lkml.kernel.org/n/tip-pqeog465zo5hsimtkfz73f27@git.kernel.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.c | 55 +++++++++++++++++++++++++++++++++++++++- arch/x86/kernel/cpu/perf_event.h | 8 ++++++ include/linux/perf_event.h | 1 + kernel/events/core.c | 1 + 4 files changed, 64 insertions(+), 1 deletion(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 53b569910175..116b040a73a8 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -24,6 +24,7 @@ #include #include #include +#include #include #include @@ -1210,7 +1211,8 @@ x86_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu) break; case CPU_STARTING: - set_in_cr4(X86_CR4_PCE); + if (x86_pmu.attr_rdpmc) + set_in_cr4(X86_CR4_PCE); if (x86_pmu.cpu_starting) x86_pmu.cpu_starting(cpu); break; @@ -1320,6 +1322,8 @@ static int __init init_hw_perf_events(void) } } + x86_pmu.attr_rdpmc = 1; /* enable userspace RDPMC usage by default */ + pr_info("... version: %d\n", x86_pmu.version); pr_info("... bit width: %d\n", x86_pmu.cntval_bits); pr_info("... generic registers: %d\n", x86_pmu.num_counters); @@ -1555,10 +1559,59 @@ static int x86_pmu_event_idx(struct perf_event *event) return idx + 1; } +static ssize_t get_attr_rdpmc(struct device *cdev, + struct device_attribute *attr, + char *buf) +{ + return snprintf(buf, 40, "%d\n", x86_pmu.attr_rdpmc); +} + +static void change_rdpmc(void *info) +{ + bool enable = !!(unsigned long)info; + + if (enable) + set_in_cr4(X86_CR4_PCE); + else + clear_in_cr4(X86_CR4_PCE); +} + +static ssize_t set_attr_rdpmc(struct device *cdev, + struct device_attribute *attr, + const char *buf, size_t count) +{ + unsigned long val = simple_strtoul(buf, NULL, 0); + + if (!!val != !!x86_pmu.attr_rdpmc) { + x86_pmu.attr_rdpmc = !!val; + smp_call_function(change_rdpmc, (void *)val, 1); + } + + return count; +} + +static DEVICE_ATTR(rdpmc, S_IRUSR | S_IWUSR, get_attr_rdpmc, set_attr_rdpmc); + +static struct attribute *x86_pmu_attrs[] = { + &dev_attr_rdpmc.attr, + NULL, +}; + +static struct attribute_group x86_pmu_attr_group = { + .attrs = x86_pmu_attrs, +}; + +static const struct attribute_group *x86_pmu_attr_groups[] = { + &x86_pmu_attr_group, + NULL, +}; + static struct pmu pmu = { .pmu_enable = x86_pmu_enable, .pmu_disable = x86_pmu_disable, + .attr_groups = x86_pmu_attr_groups, + .event_init = x86_pmu_event_init, .add = x86_pmu_add, diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 8944062f46e2..513d617b93c4 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -307,6 +307,14 @@ struct x86_pmu { struct x86_pmu_quirk *quirks; int perfctr_second_write; + /* + * sysfs attrs + */ + int attr_rdpmc; + + /* + * CPU Hotplug hooks + */ int (*cpu_prepare)(int cpu); void (*cpu_starting)(int cpu); void (*cpu_dying)(int cpu); diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 02545e6df95b..5311b79fe62c 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -615,6 +615,7 @@ struct pmu { struct list_head entry; struct device *dev; + const struct attribute_group **attr_groups; char *name; int type; diff --git a/kernel/events/core.c b/kernel/events/core.c index 05affc3878ff..dcd4049e92fc 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -5505,6 +5505,7 @@ static int pmu_dev_alloc(struct pmu *pmu) if (!pmu->dev) goto out; + pmu->dev->groups = pmu->attr_groups; device_initialize(pmu->dev); ret = dev_set_name(pmu->dev, "%s", pmu->name); if (ret) -- cgit v1.2.3 From e3f3541c19c89a4daae39300defba68943301949 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 21 Nov 2011 11:43:53 +0100 Subject: perf: Extend the mmap control page with time (TSC) fields Extend the mmap control page with fields so that userspace can compute time deltas relative to the provided time fields. Currently only implemented for x86 with constant and nonstop TSC. Signed-off-by: Peter Zijlstra Cc: Stephane Eranian Cc: Arun Sharma Link: http://lkml.kernel.org/n/tip-3u1jucza77j3wuvs0x2bic0f@git.kernel.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.c | 14 ++++++++++++++ include/linux/perf_event.h | 4 +++- kernel/events/core.c | 21 ++++++++++++++------- 3 files changed, 31 insertions(+), 8 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 116b040a73a8..f8bddb5b0600 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -32,6 +32,7 @@ #include #include #include +#include #include "perf_event.h" @@ -1627,6 +1628,19 @@ static struct pmu pmu = { .event_idx = x86_pmu_event_idx, }; +void perf_update_user_clock(struct perf_event_mmap_page *userpg, u64 now) +{ + if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) + return; + + if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) + return; + + userpg->time_mult = this_cpu_read(cyc2ns); + userpg->time_shift = CYC2NS_SCALE_FACTOR; + userpg->time_offset = this_cpu_read(cyc2ns_offset) - now; +} + /* * callchain support */ diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 5311b79fe62c..0b91db2522cc 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -291,12 +291,14 @@ struct perf_event_mmap_page { __s64 offset; /* add to hardware event value */ __u64 time_enabled; /* time event active */ __u64 time_running; /* time event on cpu */ + __u32 time_mult, time_shift; + __u64 time_offset; /* * Hole for extension of the self monitor capabilities */ - __u64 __reserved[123]; /* align to 1k */ + __u64 __reserved[121]; /* align to 1k */ /* * Control data for the mmap() data buffer. diff --git a/kernel/events/core.c b/kernel/events/core.c index dcd4049e92fc..3a9c7d81afbf 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -3220,17 +3220,22 @@ static int perf_event_index(struct perf_event *event) } static void calc_timer_values(struct perf_event *event, + u64 *now, u64 *enabled, u64 *running) { - u64 now, ctx_time; + u64 ctx_time; - now = perf_clock(); - ctx_time = event->shadow_ctx_time + now; + *now = perf_clock(); + ctx_time = event->shadow_ctx_time + *now; *enabled = ctx_time - event->tstamp_enabled; *running = ctx_time - event->tstamp_running; } +void __weak perf_update_user_clock(struct perf_event_mmap_page *userpg, u64 now) +{ +} + /* * Callers need to ensure there can be no nesting of this function, otherwise * the seqlock logic goes bad. We can not serialize this because the arch @@ -3240,7 +3245,7 @@ void perf_event_update_userpage(struct perf_event *event) { struct perf_event_mmap_page *userpg; struct ring_buffer *rb; - u64 enabled, running; + u64 enabled, running, now; rcu_read_lock(); /* @@ -3252,7 +3257,7 @@ void perf_event_update_userpage(struct perf_event *event) * because of locking issue as we can be called in * NMI context */ - calc_timer_values(event, &enabled, &running); + calc_timer_values(event, &now, &enabled, &running); rb = rcu_dereference(event->rb); if (!rb) goto unlock; @@ -3277,6 +3282,8 @@ void perf_event_update_userpage(struct perf_event *event) userpg->time_running = running + atomic64_read(&event->child_total_time_running); + perf_update_user_clock(userpg, now); + barrier(); ++userpg->lock; preempt_enable(); @@ -3763,7 +3770,7 @@ static void perf_output_read_group(struct perf_output_handle *handle, static void perf_output_read(struct perf_output_handle *handle, struct perf_event *event) { - u64 enabled = 0, running = 0; + u64 enabled = 0, running = 0, now; u64 read_format = event->attr.read_format; /* @@ -3776,7 +3783,7 @@ static void perf_output_read(struct perf_output_handle *handle, * NMI context */ if (read_format & PERF_FORMAT_TOTAL_TIMES) - calc_timer_values(event, &enabled, &running); + calc_timer_values(event, &now, &enabled, &running); if (event->attr.read_format & PERF_FORMAT_GROUP) perf_output_read_group(handle, event, enabled, running); -- cgit v1.2.3