diff options
author | Avi Kivity <avi@redhat.com> | 2011-12-25 15:44:43 +0200 |
---|---|---|
committer | Avi Kivity <avi@redhat.com> | 2011-12-27 11:22:24 +0200 |
commit | 9e31905f293ae84e4f120ed9e414031eaefa0bdf (patch) | |
tree | 153204ff0dca820e760007bc24075ec7fb46a276 /arch/x86/kernel/cpu/perf_event.c | |
parent | ff5c2c0316ff0e3e2dba3ca14167d994453df093 (diff) | |
parent | b3d9468a8bd218a695e3a0ff112cd4efd27b670a (diff) | |
download | linux-9e31905f293ae84e4f120ed9e414031eaefa0bdf.tar.bz2 |
Merge remote-tracking branch 'tip/perf/core' into kvm-updates/3.3
* tip/perf/core: (66 commits)
perf, x86: Expose perf capability to other modules
perf, x86: Implement arch event mask as quirk
x86, perf: Disable non available architectural events
jump_label: Provide jump_label_key initializers
jump_label, x86: Fix section mismatch
perf, core: Rate limit perf_sched_events jump_label patching
perf: Fix enable_on_exec for sibling events
perf: Remove superfluous arguments
perf, x86: Prefer fixed-purpose counters when scheduling
perf, x86: Fix event scheduler for constraints with overlapping counters
perf, x86: Implement event scheduler helper functions
perf: Avoid a useless pmu_disable() in the perf-tick
x86/tools: Add decoded instruction dump mode
x86: Update instruction decoder to support new AVX formats
x86/tools: Fix insn_sanity message outputs
x86/tools: Fix instruction decoder message output
x86: Fix instruction decoder to handle grouped AVX instructions
x86/tools: Fix Makefile to build all test tools
perf test: Soft errors shouldn't stop the "Validate PERF_RECORD_" test
perf test: Validate PERF_RECORD_ events and perf_sample fields
...
Signed-off-by: Avi Kivity <avi@redhat.com>
* commit 'b3d9468a8bd218a695e3a0ff112cd4efd27b670a': (66 commits)
perf, x86: Expose perf capability to other modules
perf, x86: Implement arch event mask as quirk
x86, perf: Disable non available architectural events
jump_label: Provide jump_label_key initializers
jump_label, x86: Fix section mismatch
perf, core: Rate limit perf_sched_events jump_label patching
perf: Fix enable_on_exec for sibling events
perf: Remove superfluous arguments
perf, x86: Prefer fixed-purpose counters when scheduling
perf, x86: Fix event scheduler for constraints with overlapping counters
perf, x86: Implement event scheduler helper functions
perf: Avoid a useless pmu_disable() in the perf-tick
x86/tools: Add decoded instruction dump mode
x86: Update instruction decoder to support new AVX formats
x86/tools: Fix insn_sanity message outputs
x86/tools: Fix instruction decoder message output
x86: Fix instruction decoder to handle grouped AVX instructions
x86/tools: Fix Makefile to build all test tools
perf test: Soft errors shouldn't stop the "Validate PERF_RECORD_" test
perf test: Validate PERF_RECORD_ events and perf_sample fields
...
Diffstat (limited to 'arch/x86/kernel/cpu/perf_event.c')
-rw-r--r-- | arch/x86/kernel/cpu/perf_event.c | 254 |
1 files changed, 198 insertions, 56 deletions
diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 2bda212a0010..930fe4879542 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -484,18 +484,195 @@ static inline int is_x86_event(struct perf_event *event) return event->pmu == &pmu; } +/* + * Event scheduler state: + * + * Assign events iterating over all events and counters, beginning + * with events with least weights first. Keep the current iterator + * state in struct sched_state. + */ +struct sched_state { + int weight; + int event; /* event index */ + int counter; /* counter index */ + int unassigned; /* number of events to be assigned left */ + unsigned long used[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; +}; + +/* Total max is X86_PMC_IDX_MAX, but we are O(n!) limited */ +#define SCHED_STATES_MAX 2 + +struct perf_sched { + int max_weight; + int max_events; + struct event_constraint **constraints; + struct sched_state state; + int saved_states; + struct sched_state saved[SCHED_STATES_MAX]; +}; + +/* + * Initialize interator that runs through all events and counters. + */ +static void perf_sched_init(struct perf_sched *sched, struct event_constraint **c, + int num, int wmin, int wmax) +{ + int idx; + + memset(sched, 0, sizeof(*sched)); + sched->max_events = num; + sched->max_weight = wmax; + sched->constraints = c; + + for (idx = 0; idx < num; idx++) { + if (c[idx]->weight == wmin) + break; + } + + sched->state.event = idx; /* start with min weight */ + sched->state.weight = wmin; + sched->state.unassigned = num; +} + +static void perf_sched_save_state(struct perf_sched *sched) +{ + if (WARN_ON_ONCE(sched->saved_states >= SCHED_STATES_MAX)) + return; + + sched->saved[sched->saved_states] = sched->state; + sched->saved_states++; +} + +static bool perf_sched_restore_state(struct perf_sched *sched) +{ + if (!sched->saved_states) + return false; + + sched->saved_states--; + sched->state = sched->saved[sched->saved_states]; + + /* continue with next counter: */ + clear_bit(sched->state.counter++, sched->state.used); + + return true; +} + +/* + * Select a counter for the current event to schedule. Return true on + * success. + */ +static bool __perf_sched_find_counter(struct perf_sched *sched) +{ + struct event_constraint *c; + int idx; + + if (!sched->state.unassigned) + return false; + + if (sched->state.event >= sched->max_events) + return false; + + c = sched->constraints[sched->state.event]; + + /* Prefer fixed purpose counters */ + if (x86_pmu.num_counters_fixed) { + idx = X86_PMC_IDX_FIXED; + for_each_set_bit_cont(idx, c->idxmsk, X86_PMC_IDX_MAX) { + if (!__test_and_set_bit(idx, sched->state.used)) + goto done; + } + } + /* Grab the first unused counter starting with idx */ + idx = sched->state.counter; + for_each_set_bit_cont(idx, c->idxmsk, X86_PMC_IDX_FIXED) { + if (!__test_and_set_bit(idx, sched->state.used)) + goto done; + } + + return false; + +done: + sched->state.counter = idx; + + if (c->overlap) + perf_sched_save_state(sched); + + return true; +} + +static bool perf_sched_find_counter(struct perf_sched *sched) +{ + while (!__perf_sched_find_counter(sched)) { + if (!perf_sched_restore_state(sched)) + return false; + } + + return true; +} + +/* + * Go through all unassigned events and find the next one to schedule. + * Take events with the least weight first. Return true on success. + */ +static bool perf_sched_next_event(struct perf_sched *sched) +{ + struct event_constraint *c; + + if (!sched->state.unassigned || !--sched->state.unassigned) + return false; + + do { + /* next event */ + sched->state.event++; + if (sched->state.event >= sched->max_events) { + /* next weight */ + sched->state.event = 0; + sched->state.weight++; + if (sched->state.weight > sched->max_weight) + return false; + } + c = sched->constraints[sched->state.event]; + } while (c->weight != sched->state.weight); + + sched->state.counter = 0; /* start with first counter */ + + return true; +} + +/* + * Assign a counter for each event. + */ +static int perf_assign_events(struct event_constraint **constraints, int n, + int wmin, int wmax, int *assign) +{ + struct perf_sched sched; + + perf_sched_init(&sched, constraints, n, wmin, wmax); + + do { + if (!perf_sched_find_counter(&sched)) + break; /* failed */ + if (assign) + assign[sched.state.event] = sched.state.counter; + } while (perf_sched_next_event(&sched)); + + return sched.state.unassigned; +} + int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign) { struct event_constraint *c, *constraints[X86_PMC_IDX_MAX]; unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; - int i, j, w, wmax, num = 0; + int i, wmin, wmax, num = 0; struct hw_perf_event *hwc; bitmap_zero(used_mask, X86_PMC_IDX_MAX); - for (i = 0; i < n; i++) { + for (i = 0, wmin = X86_PMC_IDX_MAX, wmax = 0; i < n; i++) { c = x86_pmu.get_event_constraints(cpuc, cpuc->event_list[i]); constraints[i] = c; + wmin = min(wmin, c->weight); + wmax = max(wmax, c->weight); } /* @@ -521,59 +698,11 @@ int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign) if (assign) assign[i] = hwc->idx; } - if (i == n) - goto done; - - /* - * begin slow path - */ - - bitmap_zero(used_mask, X86_PMC_IDX_MAX); - - /* - * weight = number of possible counters - * - * 1 = most constrained, only works on one counter - * wmax = least constrained, works on any counter - * - * assign events to counters starting with most - * constrained events. - */ - wmax = x86_pmu.num_counters; - - /* - * when fixed event counters are present, - * wmax is incremented by 1 to account - * for one more choice - */ - if (x86_pmu.num_counters_fixed) - wmax++; - - for (w = 1, num = n; num && w <= wmax; w++) { - /* for each event */ - for (i = 0; num && i < n; i++) { - c = constraints[i]; - hwc = &cpuc->event_list[i]->hw; - - if (c->weight != w) - continue; - - for_each_set_bit(j, c->idxmsk, X86_PMC_IDX_MAX) { - if (!test_bit(j, used_mask)) - break; - } - - if (j == X86_PMC_IDX_MAX) - break; - __set_bit(j, used_mask); + /* slow path */ + if (i != n) + num = perf_assign_events(constraints, n, wmin, wmax, assign); - if (assign) - assign[i] = j; - num--; - } - } -done: /* * scheduling failed or is just a simulation, * free resources if necessary @@ -1119,6 +1248,7 @@ static void __init pmu_check_apic(void) static int __init init_hw_perf_events(void) { + struct x86_pmu_quirk *quirk; struct event_constraint *c; int err; @@ -1147,8 +1277,8 @@ static int __init init_hw_perf_events(void) pr_cont("%s PMU driver.\n", x86_pmu.name); - if (x86_pmu.quirks) - x86_pmu.quirks(); + for (quirk = x86_pmu.quirks; quirk; quirk = quirk->next) + quirk->func(); if (x86_pmu.num_counters > X86_PMC_MAX_GENERIC) { WARN(1, KERN_ERR "hw perf events %d > max(%d), clipping!", @@ -1171,7 +1301,7 @@ static int __init init_hw_perf_events(void) unconstrained = (struct event_constraint) __EVENT_CONSTRAINT(0, (1ULL << x86_pmu.num_counters) - 1, - 0, x86_pmu.num_counters); + 0, x86_pmu.num_counters, 0); if (x86_pmu.event_constraints) { for_each_event_constraint(c, x86_pmu.event_constraints) { @@ -1566,3 +1696,15 @@ unsigned long perf_misc_flags(struct pt_regs *regs) return misc; } + +void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap) +{ + cap->version = x86_pmu.version; + cap->num_counters_gp = x86_pmu.num_counters; + cap->num_counters_fixed = x86_pmu.num_counters_fixed; + cap->bit_width_gp = x86_pmu.cntval_bits; + cap->bit_width_fixed = x86_pmu.cntval_bits; + cap->events_mask = (unsigned int)x86_pmu.events_maskl; + cap->events_mask_len = x86_pmu.events_mask_len; +} +EXPORT_SYMBOL_GPL(perf_get_x86_pmu_capability); |