perf-events: Add support for supplementary event registers v5
Subsystems (2)
PATCH 1/6 perf-events: Add support for supplementary event registers v5
From: Andi Kleen <[email protected]>
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 adds:
- Teaches perf_events that OFFCORE_RESPONSE needs extra parameters. The extra parameters are passed by user space in the perf_event_attr::event_extra field.
- Add 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.
--- arch/x86/kernel/cpu/perf_event.c | 64 ++++++++++ arch/x86/kernel/cpu/perf_event_intel.c | 198 ++++++++++++++++++++++++++++++++ include/linux/perf_event.h | 7 +- 3 files changed, 268 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event.c +64 -0
@@ -93,6 +93,8 @@ struct amd_nb {struct event_constraint event_constraints[X86_PMC_IDX_MAX ]; }; +struct intel_percore ; +#define MAX_LBR_ENTRIES 16struct cpu_hw_events {
@@ -128,6 +130,13 @@ struct cpu_hw_events {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 */ struct amd_nb *amd_nb;
@@ -175,6 +184,28 @@ 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;
@@ -219,6 +250,7 @@ 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;
@@ -247,6 +279,11 @@ 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 ;
@@ -339,6 +376,31 @@ 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.event_extra & ~er->valid_mask) +return -EINVAL ; + event->hw.extra_reg = er->msr; + event->hw.extra_config = event->attr.event_extra; +break ; + } +return 0 ; +} +static atomic_t active_events;static DEFINE_MUTEX (pmc_reserve_mutex);
@@ -663,6 +725,8 @@ 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); }
arch/x86/kernel/cpu/perf_event_intel.c +198 -0
268 lines not on this page · raw message
include/linux/perf_event.h +6 -1
@@ -225,7 +225,10 @@ struct perf_event_attr { };__u32 bp_type; -__u64 bp_addr; +union { +__u64 bp_addr; +__u64 event_extra;/* Extra for some events */ + };__u64 bp_len; };
@@ -541,6 +544,8 @@ struct hw_perf_event {unsigned long event_base;int idx;int last_cpu; +unsigned int extra_reg; +u64 extra_config; };struct {/* software */ struct hrtimer hrtimer;
Signature · 1 line
1.7.3
Review struct perf_event_attr { · line 233
Quoted 11 lines · Lin Ming
(quoted diff) +4
-1@@ -225,7 +225,10 @@ struct perf_event_attr { }; __u32 bp_type; - __u64 bp_addr; + union { + __u64 bp_addr; + __u64 event_extra; /* Extra for some events */ + }; __u64 bp_len; };
I did s/event_extra/config1/g and:
include/linux/perf_event.h +5 -2
@@ -227,9 +230,12 @@ struct perf_event_attr {__u32 bp_type;union {__u64 bp_addr; -__u64 config1;/* Extra for some events */ +__u64 config1;/* extension of config0 */ + }; +union { +__u64 bp_len; +__u64 config2;/* extension of config1 */ }; -__u64 bp_len; };/*
Giving us basically 192 bits of config{,1,2} space and the possibility of eventually deprecating the bp_ stuff.
Quoted 35 lines · Peter Zijlstra
Review struct perf_event_attr { · line 233
Quoted 11 lines · Lin Ming
(quoted diff) +4
-1@@ -225,7 +225,10 @@ struct perf_event_attr { }; __u32 bp_type; - __u64 bp_addr; + union { + __u64 bp_addr; + __u64 event_extra; /* Extra for some events */ + }; __u64 bp_len; };I did s/event_extra/config1/g and:
Index:
linux-2.6/include/linux/perf_event.h
===================================================================include/linux/perf_event.h +5
-2@@ -227,9 +230,12 @@ struct perf_event_attr {__u32 bp_type;union {__u64 bp_addr; -__u64 config1;/* Extra for some events */ +__u64 config1;/* extension of config0 */ + }; +union { +__u64 bp_len; +__u64 config2;/* extension of config1 */ }; -__u64 bp_len; };/*
Quoted 2 lines
Giving us basically 192 bits of config{,1,2} space and the possibility of eventually deprecating the
bp_stuff.
OK,
Do you mean it's possible to encode bp_type/bp_addr/bp_len into config1/config2?
Lin,
So I tested this patch on my NHM system and it has a major event scheduling issue. It does not ensure mutual exclusion with two offcore response events have different extra_config values. This is demonstrated with libpfm4:
$ task -e offcore_response_0:DMND_DATA_RD:LOCAL_DRAM -e
offcore_response_0:DMND_RFO:LOCAL_DRAM noploop 1
PERF[type=4 val=0x5001b7 e_u=0 e_k=0 e_hv=1 period=0 freq=0 precise=0
event_xtra=4001] offcore_response_0:DMND_DATA_RD:LOCAL_DRAM
PERF[type=4 val=0x5001b7 e_u=0 e_k=0 e_hv=1 period=0 freq=0 precise=0
event_xtra=4003] offcore_response_0:DMND_RFO:LOCAL_DRAM
noploop for 1 seconds
283,537 offcore_response_0:DMND_DATA_RD:LOCAL_DRAM
(99.99% scaling, ena=995,676,390, run=119,395)
763,110 offcore_response_0:DMND_RFO:LOCAL_DRAM (99.90%
scaling, ena=996,675,054, run=991,306)Scaling should be 50% for 1st event, and 50% for the second. Both events are in separate event groups here.
I tracked this down to a mismatch between intel_percore_constraints() and the generic layer incremental scheduling approach.
The reason exclusion is not enforced is that in this case the era->ref = 2 when it should not. The reason is pretty tricky.
The scheduling algorithm schedules event groups. I illustrate this without the offcore response event, but just regular events:
$ task -e llc_misses -e uops_retired date noploop 1
// first group
[ 397.477498] CPU0 commit_txn [ 397.477600] CPU0 get cpuc=ffff88007fc0b600 used=0 per=ffff88007f80f780 cfg=13412e xtra=0 [ 397.477672] CPU0 commit_txn ret=0
// first group + second group
[ 397.477703] CPU0 commit_txn [ 397.477804] CPU0 get cpuc=ffff88007fc0b600 used=0 per=ffff88007f80f780 cfg=13412e xtra=0 [ 397.477877] CPU0 get cpuc=ffff88007fc0b600 used=0 per=ffff88007f80f780 cfg=1301c2 xtra=0 [ 397.477949] CPU0 commit_txn ret=0
Two groups = two calls to group_sched_in() = 2 calls to commit_txn() = 2 calls to x86_schedule_events()
= 2 calls to intel_percore_constraints().
In the second call, the first event is resubmitted because this is incremental. The scheduler needs to revisit
the entire assignment. Thus, you have an imbalance of get_constraints() vs. put_constraints() in the end
and it's game over.
One may wonder how come AMD NB events don't suffer from the same problem, after all there is some shared
state. I think the reason is that we track down the owner, i.e., the event which allocated the resource first. If we
come twice with the same event, we don't do the allocation again.
I think the problem here is slightly different, though, because you want to allow other events with the identicalextra_config value. Given this is shared between HT threads, you don't want to release the resource and reallocate
again because you may lose access. Maybe a solution would be to tag events as percore allocated, and skip
those the second time around. There cannot be a ref++ on an event that has already done ref++ once and without
callin put_constraint() in between.
Anyway, there is a scheduling problem and it needs some more thinking.....
Quoted 463 lines · Lin Ming
From: Andi Kleen <[email protected]>
Intel Nehalem/Westmere have a special
OFFCORE_RESPONSEevent 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 adds:
- Teachesperf_eventsthatOFFCORE_RESPONSEneeds extra parameters. The extra parameters are passed by user space in theperf_event_attr::event_extrafield.
- Add support to the Intelperf_eventcore 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 <[email protected]>Signed-off-by Lin Ming <[email protected]>---
arch/x86/kernel/cpu/perf_event.c | 64 ++++++++++arch/x86/kernel/cpu/perf_event_intel.c| 198 ++++++++++++++++++++++++++++++++include/linux/perf_event.h| 7 +-3 files changed, 268 insertions(+), 1 deletions(-)arch/x86/kernel/cpu/perf_event.c +32
-0@@ -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 {@@ -128,6 +130,13 @@ struct cpu_hw_events { 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 */ struct amd_nb *amd_nb;@@ -175,6 +184,28 @@ 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;@@ -219,6 +250,7 @@ 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;@@ -247,6 +279,11 @@ struct x86_pmu {*/ unsigned long
lbr_tos,lbr_from,lbr_to; /* MSR base regs */ intlbr_nr; /* hardware stack size */ +
+ /*
+ * Extra registers for events
+ */
+ structextra_reg*extra_regs;
};static struct
x86_pmux86_pmu__read_mostly;arch/x86/kernel/cpu/perf_event.c +27
-0@@ -339,6 +376,31 @@ 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.event_extra & ~er->valid_mask) + return -EINVAL; + event->hw.extra_reg = er->msr; + event->hw.extra_config = event->attr.event_extra; + break; + } + return 0; +} + static atomic_t active_events; static DEFINE_MUTEX(pmc_reserve_mutex);@@ -663,6 +725,8 @@ 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); }arch/x86/kernel/cpu/perf_event_intel.c +22
-0@@ -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
};arch/x86/kernel/cpu/perf_event_intel.c +12
-0+static struct extra_reg intel_nehalem_extra_regs[] = +{ + INTEL_EVENT_EXTRA_REG(0xb7, 0x1a6, 0xffff), /* OFFCORE_RESPONSE_0 */ + EVENT_EXTRA_END +}; + +static struct event_constraint intel_nehalem_percore_constraints[] = +{ + INTEL_EVENT_CONSTRAINT(0xb7, 0), + EVENT_CONSTRAINT_END +}; +static struct
event_constraintintel_westmere_event_constraints[] = {
FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */arch/x86/kernel/cpu/perf_event_intel.c +0
-0@@ -76,6 +110,20 @@ static struct event_constraint intel_westmere_event_constraints[] =EVENT_CONSTRAINT_END
};arch/x86/kernel/cpu/perf_event_intel.c +14
-0+static struct extra_reg intel_westmere_extra_regs[] = +{ + INTEL_EVENT_EXTRA_REG(0xb7, 0x1a6, 0xffff), /* OFFCORE_RESPONSE_0 */ + INTEL_EVENT_EXTRA_REG(0xbb, 0x1a7, 0xffff), /* OFFCORE_RESPONSE_1 */ + 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_constraintintel_gen_event_constraints[] = {
FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */arch/x86/kernel/cpu/perf_event_intel.c +0
-0@@ -794,6 +842,66 @@ intel_bts_constraints(struct perf_event *event) }static struct
event_constraint*arch/x86/kernel/cpu/perf_event_intel.c +60
-0+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) + return NULL; + + for (c = x86_pmu.percore_constraints; c->cmask; c++) { + if (e != c->code) + continue; + + /* + * Allocate resource per core. + */ + c = NULL; + 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; + 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; + c = NULL; + } + raw_spin_unlock(&pc->lock); + return c; + } + + return NULL; +} + +static struct event_constraint *intel_get_event_constraints(struct
cpu_hw_events*cpuc, structperf_event*event) {
structevent_constraint*c;arch/x86/kernel/cpu/perf_event_intel.c +90
-0@@ -806,9 +914,50 @@ 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--; + 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);@@ -880,11 +1029,43 @@ 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_online_cpu(i) { + 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. @@ -894,6 +1075,15 @@ 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); }@@ -918,7 +1108,9 @@ 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, };@@ -1036,7 +1228,10 @@ static __init int intel_pmu_init(void) intel_pmu_lbr_init_nhm(); x86_pmu.event_constraints = intel_nehalem_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;@@ -1058,7 +1253,10 @@ 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.extra_regs = intel_westmere_extra_regs; pr_cont("Westmere events, "); break;include/linux/perf_event.h +6
-1@@ -225,7 +225,10 @@ struct perf_event_attr { }; __u32 bp_type; - __u64 bp_addr; + union { + __u64 bp_addr; + __u64 event_extra; /* Extra for some events */ + }; __u64 bp_len; };@@ -541,6 +544,8 @@ struct hw_perf_event { unsigned long event_base; int idx; int last_cpu; + unsigned int extra_reg; + u64 extra_config; };struct { /* software */
struct hrtimer hrtimer;
--
1.7.3
Quoted 8 lines · Stephane Eranian
Lin,
So I tested this patch on my NHM system and it has a major event scheduling issue. It does not ensure mutual exclusion with two offcore response events have different
extra_configvalues. This is demonstrated with libpfm4:$ task -e offcore_response_0:DMND_DATA_RD:LOCAL_DRAM -e offcore_response_0:DMND_RFO:LOCAL_DRAM noploop 1
Hi, Stephane
$ cd libpfm4/perf_examples $ ./task -e offcore_response_0:DMND_DATA_RD:LOCAL_DRAM -e offcore_response_0:DMND_RFO:LOCAL_DRAM ./noploop 1 noploop for 1 seconds
0 offcore_response_0:DMND_DATA_RD:LOCAL_DRAM (0.00% scaling, ena=1,000,019,277, run=1,000,019,277)
0 offcore_response_0:DMND_RFO:LOCAL_DRAM (0.00% scaling, ena=1,000,019,277, run=1,000,019,277)The "task" example does not run well on my NHM system. Do I miss some other setting?
Lin Ming
Quoted 532 lines
PERF[type=4 val=0x5001b7 e_u=0 e_k=0 e_hv=1 period=0 freq=0 precise=0 event_xtra=4001] offcore_response_0:DMND_DATA_RD:LOCAL_DRAM PERF[type=4 val=0x5001b7 e_u=0 e_k=0 e_hv=1 period=0 freq=0 precise=0 event_xtra=4003] offcore_response_0:DMND_RFO:LOCAL_DRAM noploop for 1 seconds
283,537 offcore_response_0:DMND_DATA_RD:LOCAL_DRAM (99.99% scaling, ena=995,676,390, run=119,395) 763,110 offcore_response_0:DMND_RFO:LOCAL_DRAM (99.90%scaling, ena=996,675,054, run=991,306)
Scaling should be 50% for 1st event, and 50% for the second. Both events are in separate event groups here.
I tracked this down to a mismatch between
intel_percore_constraints()and the generic layer incremental scheduling approach.The reason exclusion is not enforced is that in this case the
era->ref= 2 when it should not. The reason is pretty tricky.The scheduling algorithm schedules event groups. I illustrate this without the offcore response event, but just regular events:
$ task -e llc_misses -e uops_retired date noploop 1// first group
[ 397.477498] CPU0 commit_txn [ 397.477600] CPU0 get cpuc=ffff88007fc0b600 used=0 per=ffff88007f80f780 cfg=13412e xtra=0 [ 397.477672] CPU0 commit_txn ret=0// first group + second group
[ 397.477703] CPU0 commit_txn [ 397.477804] CPU0 get cpuc=ffff88007fc0b600 used=0 per=ffff88007f80f780 cfg=13412e xtra=0 [ 397.477877] CPU0 get cpuc=ffff88007fc0b600 used=0 per=ffff88007f80f780 cfg=1301c2 xtra=0 [ 397.477949] CPU0 commit_txn ret=0Two groups = two calls to
group_sched_in()= 2 calls tocommit_txn()= 2 calls tox86_schedule_events()
= 2 calls tointel_percore_constraints().In the second call, the first event is resubmitted because this is incremental. The scheduler needs to revisit
the entire assignment. Thus, you have an imbalance ofget_constraints()vs.put_constraints()in the end
and it's game over.One may wonder how come AMD NB events don't suffer from the same problem, after all there is some shared
state. I think the reason is that we track down the owner, i.e., the event which allocated the resource first. If we
come twice with the same event, we don't do the allocation again.I think the problem here is slightly different, though, because you want to allow other events with the identical
extra_configvalue. Given this is shared between HT threads, you don't want to release the resource and reallocate
again because you may lose access. Maybe a solution would be to tag events as percore allocated, and skip
those the second time around. There cannot be a ref++ on an event that has already done ref++ once and without
callinput_constraint()in between.Anyway, there is a scheduling problem and it needs some more thinking.....
Quoted 463 lines · Lin Ming
From: Andi Kleen <[email protected]>
Intel Nehalem/Westmere have a special
OFFCORE_RESPONSEevent 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 adds:
- Teachesperf_eventsthatOFFCORE_RESPONSEneeds extra parameters. The extra parameters are passed by user space in theperf_event_attr::event_extrafield.
- Add support to the Intelperf_eventcore 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 <[email protected]>Signed-off-by Lin Ming <[email protected]>--- arch/x86/kernel/cpu/perf_event.c | 64 ++++++++++ arch/x86/kernel/cpu/perf_event_intel.c | 198 ++++++++++++++++++++++++++++++++ include/linux/perf_event.h | 7 +- 3 files changed, 268 insertions(+), 1 deletions(-)arch/x86/kernel/cpu/perf_event.c +64
-0@@ -93,6 +93,8 @@ struct amd_nb {struct event_constraint event_constraints[X86_PMC_IDX_MAX ]; }; +struct intel_percore ; +#define MAX_LBR_ENTRIES 16struct cpu_hw_events {@@ -128,6 +130,13 @@ struct cpu_hw_events {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 */ struct amd_nb *amd_nb;@@ -175,6 +184,28 @@ 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 ;@@ -219,6 +250,7 @@ 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;@@ -247,6 +279,11 @@ 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 ;@@ -339,6 +376,31 @@ 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.event_extra & ~er->valid_mask) +return -EINVAL ; + event->hw.extra_reg = er->msr; + event->hw.extra_config = event->attr.event_extra; +break ; + } +return 0 ; +} +static atomic_t active_events;static DEFINE_MUTEX (pmc_reserve_mutex);@@ -663,6 +725,8 @@ 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); }arch/x86/kernel/cpu/perf_event_intel.c +198
-0@@ -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 ,0x1a6 ,0xffff ),/* OFFCORE_RESPONSE_0 */ +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 */ @@ -76,6 +110,20 @@ static struct event_constraint intel_westmere_event_constraints[] =EVENT_CONSTRAINT_END }; +static struct extra_reg intel_westmere_extra_regs[] = +{ +INTEL_EVENT_EXTRA_REG (0xb7 ,0x1a6 ,0xffff ),/* OFFCORE_RESPONSE_0 */ +INTEL_EVENT_EXTRA_REG (0xbb ,0x1a7 ,0xffff ),/* OFFCORE_RESPONSE_1 */ +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 */ @@ -794,6 +842,66 @@ intel_bts_constraints(struct perf_event *event) }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) +return NULL ; + +for (c = x86_pmu.percore_constraints; c->cmask; c++) { +if (e != c->code) +continue ; + +/* +* Allocate resource per core. +*/ + c =NULL ; + 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 ; + 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 ; + 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) {struct event_constraint *c;@@ -806,9 +914,50 @@ intel_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *eventif (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--; +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);@@ -880,11 +1029,43 @@ 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_online_cpu (i) { +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. @@ -894,6 +1075,15 @@ 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); }@@ -918,7 +1108,9 @@ 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, };@@ -1036,7 +1228,10 @@ static __init int intel_pmu_init(void)intel_pmu_lbr_init_nhm (); x86_pmu.event_constraints = intel_nehalem_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 ;@@ -1058,7 +1253,10 @@ 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.extra_regs = intel_westmere_extra_regs;pr_cont ("Westmere events, " );break ;include/linux/perf_event.h +6
-2@@ -225,7 +225,10 @@ struct perf_event_attr { };__u32 bp_type; -__u64 bp_addr; +union { +__u64 bp_addr; +__u64 event_extra;/* Extra for some events */ + };__u64 bp_len; };@@ -541,6 +544,8 @@ struct hw_perf_event {unsigned long event_base;int idx;int last_cpu; +unsigned int extra_reg; +u64 extra_config; };struct {/* software */ struct hrtimer hrtimer; --1.7.3