perf: Add support for supplementary event registers
authorAndi Kleen <ak@linux.intel.com>
Thu, 3 Mar 2011 02:34:47 +0000 (10:34 +0800)
committerIngo Molnar <mingo@elte.hu>
Fri, 4 Mar 2011 10:32:53 +0000 (11:32 +0100)
Change logs against Andi's original version:

- Extends perf_event_attr:config to config{,1,2} (Peter Zijlstra)
- Fixed a major event scheduling issue. There cannot be a ref++ on an
  event that has already done ref++ once and without calling
  put_constraint() in between. (Stephane Eranian)
- Use thread_cpumask for percore allocation. (Lin Ming)
- Use MSR names in the extra reg lists. (Lin Ming)
- Remove redundant "c = NULL" in intel_percore_constraints
- Fix comment of perf_event_attr::config1

Intel Nehalem/Westmere have a special OFFCORE_RESPONSE event
that can be used to monitor any offcore accesses from a core.
This is a very useful event for various tunings, and it's
also needed to implement the generic LLC-* events correctly.

Unfortunately this event requires programming a mask in a separate
register. And worse this separate register is per core, not per
CPU thread.

This patch:

- Teaches perf_events that OFFCORE_RESPONSE needs extra parameters.
  The extra parameters are passed by user space in the
  perf_event_attr::config1 field.

- Adds support to the Intel perf_event core to schedule per
  core resources. This adds fairly generic infrastructure that
  can be also used for other per core resources.
  The basic code has is patterned after the similar AMD northbridge
  constraints code.

Thanks to Stephane Eranian who pointed out some problems
in the original version and suggested improvements.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1299119690-13991-2-git-send-email-ming.m.lin@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/include/asm/msr-index.h
arch/x86/kernel/cpu/perf_event.c
arch/x86/kernel/cpu/perf_event_intel.c
include/linux/perf_event.h

index 4d0dfa0d998e9f80ce244d86e1fd583513aaaaca..d25e74cc1a509ba67c6cf765c27e438af742b007 100644 (file)
@@ -47,6 +47,9 @@
 #define MSR_IA32_MCG_STATUS            0x0000017a
 #define MSR_IA32_MCG_CTL               0x0000017b
 
+#define MSR_OFFCORE_RSP_0              0x000001a6
+#define MSR_OFFCORE_RSP_1              0x000001a7
+
 #define MSR_IA32_PEBS_ENABLE           0x000003f1
 #define MSR_IA32_DS_AREA               0x00000600
 #define MSR_IA32_PERF_CAPABILITIES     0x00000345
index ea03c725e465cc82b601c8a423a7d5bea05b7f10..ec6a6db0733282690af7c9a5e5f600b8d0e717e9 100644 (file)
@@ -93,6 +93,8 @@ struct amd_nb {
        struct event_constraint event_constraints[X86_PMC_IDX_MAX];
 };
 
+struct intel_percore;
+
 #define MAX_LBR_ENTRIES                16
 
 struct cpu_hw_events {
@@ -127,6 +129,13 @@ struct cpu_hw_events {
        struct perf_branch_stack        lbr_stack;
        struct perf_branch_entry        lbr_entries[MAX_LBR_ENTRIES];
 
+       /*
+        * Intel percore register state.
+        * Coordinate shared resources between HT threads.
+        */
+       int                             percore_used; /* Used by this CPU? */
+       struct intel_percore            *per_core;
+
        /*
         * AMD specific bits
         */
@@ -177,6 +186,28 @@ struct cpu_hw_events {
 #define for_each_event_constraint(e, c)        \
        for ((e) = (c); (e)->weight; (e)++)
 
+/*
+ * Extra registers for specific events.
+ * Some events need large masks and require external MSRs.
+ * Define a mapping to these extra registers.
+ */
+struct extra_reg {
+       unsigned int            event;
+       unsigned int            msr;
+       u64                     config_mask;
+       u64                     valid_mask;
+};
+
+#define EVENT_EXTRA_REG(e, ms, m, vm) {        \
+       .event = (e),           \
+       .msr = (ms),            \
+       .config_mask = (m),     \
+       .valid_mask = (vm),     \
+       }
+#define INTEL_EVENT_EXTRA_REG(event, msr, vm)  \
+       EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm)
+#define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0)
+
 union perf_capabilities {
        struct {
                u64     lbr_format    : 6;
@@ -221,6 +252,7 @@ struct x86_pmu {
        void            (*put_event_constraints)(struct cpu_hw_events *cpuc,
                                                 struct perf_event *event);
        struct event_constraint *event_constraints;
+       struct event_constraint *percore_constraints;
        void            (*quirks)(void);
        int             perfctr_second_write;
 
@@ -249,6 +281,11 @@ struct x86_pmu {
         */
        unsigned long   lbr_tos, lbr_from, lbr_to; /* MSR base regs       */
        int             lbr_nr;                    /* hardware stack size */
+
+       /*
+        * Extra registers for events
+        */
+       struct extra_reg *extra_regs;
 };
 
 static struct x86_pmu x86_pmu __read_mostly;
@@ -341,6 +378,31 @@ static inline unsigned int x86_pmu_event_addr(int index)
        return x86_pmu.perfctr + x86_pmu_addr_offset(index);
 }
 
+/*
+ * Find and validate any extra registers to set up.
+ */
+static int x86_pmu_extra_regs(u64 config, struct perf_event *event)
+{
+       struct extra_reg *er;
+
+       event->hw.extra_reg = 0;
+       event->hw.extra_config = 0;
+
+       if (!x86_pmu.extra_regs)
+               return 0;
+
+       for (er = x86_pmu.extra_regs; er->msr; er++) {
+               if (er->event != (config & er->config_mask))
+                       continue;
+               if (event->attr.config1 & ~er->valid_mask)
+                       return -EINVAL;
+               event->hw.extra_reg = er->msr;
+               event->hw.extra_config = event->attr.config1;
+               break;
+       }
+       return 0;
+}
+
 static atomic_t active_events;
 static DEFINE_MUTEX(pmc_reserve_mutex);
 
@@ -665,6 +727,8 @@ static void x86_pmu_disable(struct pmu *pmu)
 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
                                          u64 enable_mask)
 {
+       if (hwc->extra_reg)
+               wrmsrl(hwc->extra_reg, hwc->extra_config);
        wrmsrl(hwc->config_base, hwc->config | enable_mask);
 }
 
index c3ce053ecb46122f388cd6933b73a2af01f7637e..13cb6cf013f64b80418d94473a2441ae3033021f 100644 (file)
@@ -1,5 +1,27 @@
 #ifdef CONFIG_CPU_SUP_INTEL
 
+#define MAX_EXTRA_REGS 2
+
+/*
+ * Per register state.
+ */
+struct er_account {
+       int                     ref;            /* reference count */
+       unsigned int            extra_reg;      /* extra MSR number */
+       u64                     extra_config;   /* extra MSR config */
+};
+
+/*
+ * Per core state
+ * This used to coordinate shared registers for HT threads.
+ */
+struct intel_percore {
+       raw_spinlock_t          lock;           /* protect structure */
+       struct er_account       regs[MAX_EXTRA_REGS];
+       int                     refcnt;         /* number of threads */
+       unsigned                core_id;
+};
+
 /*
  * Intel PerfMon, used on Core and later.
  */
@@ -64,6 +86,18 @@ static struct event_constraint intel_nehalem_event_constraints[] =
        EVENT_CONSTRAINT_END
 };
 
+static struct extra_reg intel_nehalem_extra_regs[] =
+{
+       INTEL_EVENT_EXTRA_REG(0xb7, MSR_OFFCORE_RSP_0, 0xffff),
+       EVENT_EXTRA_END
+};
+
+static struct event_constraint intel_nehalem_percore_constraints[] =
+{
+       INTEL_EVENT_CONSTRAINT(0xb7, 0),
+       EVENT_CONSTRAINT_END
+};
+
 static struct event_constraint intel_westmere_event_constraints[] =
 {
        FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
@@ -89,6 +123,20 @@ static struct event_constraint intel_snb_event_constraints[] =
        EVENT_CONSTRAINT_END
 };
 
+static struct extra_reg intel_westmere_extra_regs[] =
+{
+       INTEL_EVENT_EXTRA_REG(0xb7, MSR_OFFCORE_RSP_0, 0xffff),
+       INTEL_EVENT_EXTRA_REG(0xbb, MSR_OFFCORE_RSP_1, 0xffff),
+       EVENT_EXTRA_END
+};
+
+static struct event_constraint intel_westmere_percore_constraints[] =
+{
+       INTEL_EVENT_CONSTRAINT(0xb7, 0),
+       INTEL_EVENT_CONSTRAINT(0xbb, 0),
+       EVENT_CONSTRAINT_END
+};
+
 static struct event_constraint intel_gen_event_constraints[] =
 {
        FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
@@ -906,6 +954,67 @@ intel_bts_constraints(struct perf_event *event)
        return NULL;
 }
 
+static struct event_constraint *
+intel_percore_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
+{
+       struct hw_perf_event *hwc = &event->hw;
+       unsigned int e = hwc->config & ARCH_PERFMON_EVENTSEL_EVENT;
+       struct event_constraint *c;
+       struct intel_percore *pc;
+       struct er_account *era;
+       int i;
+       int free_slot;
+       int found;
+
+       if (!x86_pmu.percore_constraints || hwc->extra_alloc)
+               return NULL;
+
+       for (c = x86_pmu.percore_constraints; c->cmask; c++) {
+               if (e != c->code)
+                       continue;
+
+               /*
+                * Allocate resource per core.
+                */
+               pc = cpuc->per_core;
+               if (!pc)
+                       break;
+               c = &emptyconstraint;
+               raw_spin_lock(&pc->lock);
+               free_slot = -1;
+               found = 0;
+               for (i = 0; i < MAX_EXTRA_REGS; i++) {
+                       era = &pc->regs[i];
+                       if (era->ref > 0 && hwc->extra_reg == era->extra_reg) {
+                               /* Allow sharing same config */
+                               if (hwc->extra_config == era->extra_config) {
+                                       era->ref++;
+                                       cpuc->percore_used = 1;
+                                       hwc->extra_alloc = 1;
+                                       c = NULL;
+                               }
+                               /* else conflict */
+                               found = 1;
+                               break;
+                       } else if (era->ref == 0 && free_slot == -1)
+                               free_slot = i;
+               }
+               if (!found && free_slot != -1) {
+                       era = &pc->regs[free_slot];
+                       era->ref = 1;
+                       era->extra_reg = hwc->extra_reg;
+                       era->extra_config = hwc->extra_config;
+                       cpuc->percore_used = 1;
+                       hwc->extra_alloc = 1;
+                       c = NULL;
+               }
+               raw_spin_unlock(&pc->lock);
+               return c;
+       }
+
+       return NULL;
+}
+
 static struct event_constraint *
 intel_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
 {
@@ -919,9 +1028,51 @@ intel_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event
        if (c)
                return c;
 
+       c = intel_percore_constraints(cpuc, event);
+       if (c)
+               return c;
+
        return x86_get_event_constraints(cpuc, event);
 }
 
+static void intel_put_event_constraints(struct cpu_hw_events *cpuc,
+                                       struct perf_event *event)
+{
+       struct extra_reg *er;
+       struct intel_percore *pc;
+       struct er_account *era;
+       struct hw_perf_event *hwc = &event->hw;
+       int i, allref;
+
+       if (!cpuc->percore_used)
+               return;
+
+       for (er = x86_pmu.extra_regs; er->msr; er++) {
+               if (er->event != (hwc->config & er->config_mask))
+                       continue;
+
+               pc = cpuc->per_core;
+               raw_spin_lock(&pc->lock);
+               for (i = 0; i < MAX_EXTRA_REGS; i++) {
+                       era = &pc->regs[i];
+                       if (era->ref > 0 &&
+                           era->extra_config == hwc->extra_config &&
+                           era->extra_reg == er->msr) {
+                               era->ref--;
+                               hwc->extra_alloc = 0;
+                               break;
+                       }
+               }
+               allref = 0;
+               for (i = 0; i < MAX_EXTRA_REGS; i++)
+                       allref += pc->regs[i].ref;
+               if (allref == 0)
+                       cpuc->percore_used = 0;
+               raw_spin_unlock(&pc->lock);
+               break;
+       }
+}
+
 static int intel_pmu_hw_config(struct perf_event *event)
 {
        int ret = x86_pmu_hw_config(event);
@@ -993,11 +1144,43 @@ static __initconst const struct x86_pmu core_pmu = {
         */
        .max_period             = (1ULL << 31) - 1,
        .get_event_constraints  = intel_get_event_constraints,
+       .put_event_constraints  = intel_put_event_constraints,
        .event_constraints      = intel_core_event_constraints,
 };
 
+static int intel_pmu_cpu_prepare(int cpu)
+{
+       struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu);
+
+       cpuc->per_core = kzalloc_node(sizeof(struct intel_percore),
+                                     GFP_KERNEL, cpu_to_node(cpu));
+       if (!cpuc->per_core)
+               return NOTIFY_BAD;
+
+       raw_spin_lock_init(&cpuc->per_core->lock);
+       cpuc->per_core->core_id = -1;
+       return NOTIFY_OK;
+}
+
 static void intel_pmu_cpu_starting(int cpu)
 {
+       struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu);
+       int core_id = topology_core_id(cpu);
+       int i;
+
+       for_each_cpu(i, topology_thread_cpumask(cpu)) {
+               struct intel_percore *pc = per_cpu(cpu_hw_events, i).per_core;
+
+               if (pc && pc->core_id == core_id) {
+                       kfree(cpuc->per_core);
+                       cpuc->per_core = pc;
+                       break;
+               }
+       }
+
+       cpuc->per_core->core_id = core_id;
+       cpuc->per_core->refcnt++;
+
        init_debug_store_on_cpu(cpu);
        /*
         * Deal with CPUs that don't clear their LBRs on power-up.
@@ -1007,6 +1190,15 @@ static void intel_pmu_cpu_starting(int cpu)
 
 static void intel_pmu_cpu_dying(int cpu)
 {
+       struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu);
+       struct intel_percore *pc = cpuc->per_core;
+
+       if (pc) {
+               if (pc->core_id == -1 || --pc->refcnt == 0)
+                       kfree(pc);
+               cpuc->per_core = NULL;
+       }
+
        fini_debug_store_on_cpu(cpu);
 }
 
@@ -1031,7 +1223,9 @@ static __initconst const struct x86_pmu intel_pmu = {
         */
        .max_period             = (1ULL << 31) - 1,
        .get_event_constraints  = intel_get_event_constraints,
+       .put_event_constraints  = intel_put_event_constraints,
 
+       .cpu_prepare            = intel_pmu_cpu_prepare,
        .cpu_starting           = intel_pmu_cpu_starting,
        .cpu_dying              = intel_pmu_cpu_dying,
 };
@@ -1151,7 +1345,9 @@ static __init int intel_pmu_init(void)
 
                x86_pmu.event_constraints = intel_nehalem_event_constraints;
                x86_pmu.pebs_constraints = intel_nehalem_pebs_event_constraints;
+               x86_pmu.percore_constraints = intel_nehalem_percore_constraints;
                x86_pmu.enable_all = intel_pmu_nhm_enable_all;
+               x86_pmu.extra_regs = intel_nehalem_extra_regs;
                pr_cont("Nehalem events, ");
                break;
 
@@ -1174,8 +1370,10 @@ static __init int intel_pmu_init(void)
                intel_pmu_lbr_init_nhm();
 
                x86_pmu.event_constraints = intel_westmere_event_constraints;
+               x86_pmu.percore_constraints = intel_westmere_percore_constraints;
                x86_pmu.enable_all = intel_pmu_nhm_enable_all;
                x86_pmu.pebs_constraints = intel_westmere_pebs_event_constraints;
+               x86_pmu.extra_regs = intel_westmere_extra_regs;
                pr_cont("Westmere events, ");
                break;
 
index 8ceb5a6fd9c91ca2fbbcbacecdddd04b406bd69b..614615b8d42b851481a624aeabcb110e7e27078e 100644 (file)
@@ -225,8 +225,14 @@ struct perf_event_attr {
        };
 
        __u32                   bp_type;
-       __u64                   bp_addr;
-       __u64                   bp_len;
+       union {
+               __u64           bp_addr;
+               __u64           config1; /* extension of config */
+       };
+       union {
+               __u64           bp_len;
+               __u64           config2; /* extension of config1 */
+       };
 };
 
 /*
@@ -541,6 +547,9 @@ struct hw_perf_event {
                        unsigned long   event_base;
                        int             idx;
                        int             last_cpu;
+                       unsigned int    extra_reg;
+                       u64             extra_config;
+                       int             extra_alloc;
                };
                struct { /* software */
                        struct hrtimer  hrtimer;