static DEFINE_PER_CPU(atomic_t, perf_cgroup_events);
static DEFINE_PER_CPU(int, perf_sched_cb_usages);
+static DEFINE_PER_CPU(struct pmu_event_list, pmu_sb_events);
static atomic_t nr_mmap_events __read_mostly;
static atomic_t nr_comm_events __read_mostly;
if (ret || !write)
return ret;
+ /*
+ * If throttling is disabled don't allow the write:
+ */
+ if (sysctl_perf_cpu_time_max_percent == 100 ||
+ sysctl_perf_cpu_time_max_percent == 0)
+ return -EINVAL;
+
max_samples_per_tick = DIV_ROUND_UP(sysctl_perf_event_sample_rate, HZ);
perf_sample_period_ns = NSEC_PER_SEC / sysctl_perf_event_sample_rate;
update_perf_cpu_limits();
static void perf_duration_warn(struct irq_work *w)
{
- printk_ratelimited(KERN_WARNING
+ printk_ratelimited(KERN_INFO
"perf: interrupt took too long (%lld > %lld), lowering "
"kernel.perf_event_max_sample_rate to %d\n",
__report_avg, __report_allowed,
static inline int
event_filter_match(struct perf_event *event)
{
- return (event->cpu == -1 || event->cpu == smp_processor_id())
- && perf_cgroup_match(event) && pmu_filter_match(event);
+ return (event->cpu == -1 || event->cpu == smp_processor_id()) &&
+ perf_cgroup_match(event) && pmu_filter_match(event);
}
static void
* maintained, otherwise bogus information is return
* via read() for time_enabled, time_running:
*/
- if (event->state == PERF_EVENT_STATE_INACTIVE
- && !event_filter_match(event)) {
+ if (event->state == PERF_EVENT_STATE_INACTIVE &&
+ !event_filter_match(event)) {
delta = tstamp - event->tstamp_stopped;
event->tstamp_running += delta;
event->tstamp_stopped = tstamp;
lockdep_assert_held(&ctx->mutex);
- event->ctx = ctx;
if (event->cpu != -1)
event->cpu = cpu;
+ /*
+ * Ensures that if we can observe event->ctx, both the event and ctx
+ * will be 'complete'. See perf_iterate_sb_cpu().
+ */
+ smp_store_release(&event->ctx, ctx);
+
if (!task) {
cpu_function_call(cpu, __perf_install_in_context, event);
return;
static void ring_buffer_attach(struct perf_event *event,
struct ring_buffer *rb);
+static void detach_sb_event(struct perf_event *event)
+{
+ struct pmu_event_list *pel = per_cpu_ptr(&pmu_sb_events, event->cpu);
+
+ raw_spin_lock(&pel->lock);
+ list_del_rcu(&event->sb_list);
+ raw_spin_unlock(&pel->lock);
+}
+
+static bool is_sb_event(struct perf_event *event)
+{
+ struct perf_event_attr *attr = &event->attr;
+
+ if (event->parent)
+ return false;
+
+ if (event->attach_state & PERF_ATTACH_TASK)
+ return false;
+
+ if (attr->mmap || attr->mmap_data || attr->mmap2 ||
+ attr->comm || attr->comm_exec ||
+ attr->task ||
+ attr->context_switch)
+ return true;
+ return false;
+}
+
+static void unaccount_pmu_sb_event(struct perf_event *event)
+{
+ if (is_sb_event(event))
+ detach_sb_event(event);
+}
+
static void unaccount_event_cpu(struct perf_event *event, int cpu)
{
if (event->parent)
}
unaccount_event_cpu(event, event->cpu);
+
+ unaccount_pmu_sb_event(event);
}
static void perf_sched_delayed(struct work_struct *work)
}
if (sample_type & PERF_SAMPLE_RAW) {
- if (data->raw) {
- u32 raw_size = data->raw->size;
- u32 real_size = round_up(raw_size + sizeof(u32),
- sizeof(u64)) - sizeof(u32);
- u64 zero = 0;
-
- perf_output_put(handle, real_size);
- __output_copy(handle, data->raw->data, raw_size);
- if (real_size - raw_size)
- __output_copy(handle, &zero, real_size - raw_size);
+ struct perf_raw_record *raw = data->raw;
+
+ if (raw) {
+ struct perf_raw_frag *frag = &raw->frag;
+
+ perf_output_put(handle, raw->size);
+ do {
+ if (frag->copy) {
+ __output_custom(handle, frag->copy,
+ frag->data, frag->size);
+ } else {
+ __output_copy(handle, frag->data,
+ frag->size);
+ }
+ if (perf_raw_frag_last(frag))
+ break;
+ frag = frag->next;
+ } while (1);
+ if (frag->pad)
+ __output_skip(handle, NULL, frag->pad);
} else {
struct {
u32 size;
}
if (sample_type & PERF_SAMPLE_RAW) {
- int size = sizeof(u32);
-
- if (data->raw)
- size += data->raw->size;
- else
- size += sizeof(u32);
+ struct perf_raw_record *raw = data->raw;
+ int size;
+
+ if (raw) {
+ struct perf_raw_frag *frag = &raw->frag;
+ u32 sum = 0;
+
+ do {
+ sum += frag->size;
+ if (perf_raw_frag_last(frag))
+ break;
+ frag = frag->next;
+ } while (1);
+
+ size = round_up(sum + sizeof(u32), sizeof(u64));
+ raw->size = size - sizeof(u32);
+ frag->pad = raw->size - sum;
+ } else {
+ size = sizeof(u64);
+ }
- header->size += round_up(size, sizeof(u64));
+ header->size += size;
}
if (sample_type & PERF_SAMPLE_BRANCH_STACK) {
perf_output_end(&handle);
}
-typedef void (perf_event_aux_output_cb)(struct perf_event *event, void *data);
+typedef void (perf_iterate_f)(struct perf_event *event, void *data);
static void
-perf_event_aux_ctx(struct perf_event_context *ctx,
- perf_event_aux_output_cb output,
+perf_iterate_ctx(struct perf_event_context *ctx,
+ perf_iterate_f output,
void *data, bool all)
{
struct perf_event *event;
}
}
-static void
-perf_event_aux_task_ctx(perf_event_aux_output_cb output, void *data,
- struct perf_event_context *task_ctx)
+static void perf_iterate_sb_cpu(perf_iterate_f output, void *data)
{
- rcu_read_lock();
- preempt_disable();
- perf_event_aux_ctx(task_ctx, output, data, false);
- preempt_enable();
- rcu_read_unlock();
+ struct pmu_event_list *pel = this_cpu_ptr(&pmu_sb_events);
+ struct perf_event *event;
+
+ list_for_each_entry_rcu(event, &pel->list, sb_list) {
+ /*
+ * Skip events that are not fully formed yet; ensure that
+ * if we observe event->ctx, both event and ctx will be
+ * complete enough. See perf_install_in_context().
+ */
+ if (!smp_load_acquire(&event->ctx))
+ continue;
+
+ if (event->state < PERF_EVENT_STATE_INACTIVE)
+ continue;
+ if (!event_filter_match(event))
+ continue;
+ output(event, data);
+ }
}
+/*
+ * Iterate all events that need to receive side-band events.
+ *
+ * For new callers; ensure that account_pmu_sb_event() includes
+ * your event, otherwise it might not get delivered.
+ */
static void
-perf_event_aux(perf_event_aux_output_cb output, void *data,
+perf_iterate_sb(perf_iterate_f output, void *data,
struct perf_event_context *task_ctx)
{
- struct perf_cpu_context *cpuctx;
struct perf_event_context *ctx;
- struct pmu *pmu;
int ctxn;
+ rcu_read_lock();
+ preempt_disable();
+
/*
- * If we have task_ctx != NULL we only notify
- * the task context itself. The task_ctx is set
- * only for EXIT events before releasing task
+ * If we have task_ctx != NULL we only notify the task context itself.
+ * The task_ctx is set only for EXIT events before releasing task
* context.
*/
if (task_ctx) {
- perf_event_aux_task_ctx(output, data, task_ctx);
- return;
+ perf_iterate_ctx(task_ctx, output, data, false);
+ goto done;
}
- rcu_read_lock();
- list_for_each_entry_rcu(pmu, &pmus, entry) {
- cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
- if (cpuctx->unique_pmu != pmu)
- goto next;
- perf_event_aux_ctx(&cpuctx->ctx, output, data, false);
- ctxn = pmu->task_ctx_nr;
- if (ctxn < 0)
- goto next;
+ perf_iterate_sb_cpu(output, data);
+
+ for_each_task_context_nr(ctxn) {
ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
if (ctx)
- perf_event_aux_ctx(ctx, output, data, false);
-next:
- put_cpu_ptr(pmu->pmu_cpu_context);
+ perf_iterate_ctx(ctx, output, data, false);
}
+done:
+ preempt_enable();
rcu_read_unlock();
}
perf_event_enable_on_exec(ctxn);
- perf_event_aux_ctx(ctx, perf_event_addr_filters_exec, NULL,
+ perf_iterate_ctx(ctx, perf_event_addr_filters_exec, NULL,
true);
}
rcu_read_unlock();
};
rcu_read_lock();
- perf_event_aux_ctx(&cpuctx->ctx, __perf_event_output_stop, &ro, false);
+ perf_iterate_ctx(&cpuctx->ctx, __perf_event_output_stop, &ro, false);
if (cpuctx->task_ctx)
- perf_event_aux_ctx(cpuctx->task_ctx, __perf_event_output_stop,
+ perf_iterate_ctx(cpuctx->task_ctx, __perf_event_output_stop,
&ro, false);
rcu_read_unlock();
},
};
- perf_event_aux(perf_event_task_output,
+ perf_iterate_sb(perf_event_task_output,
&task_event,
task_ctx);
}
comm_event->event_id.header.size = sizeof(comm_event->event_id) + size;
- perf_event_aux(perf_event_comm_output,
+ perf_iterate_sb(perf_event_comm_output,
comm_event,
NULL);
}
mmap_event->event_id.header.size = sizeof(mmap_event->event_id) + size;
- perf_event_aux(perf_event_mmap_output,
+ perf_iterate_sb(perf_event_mmap_output,
mmap_event,
NULL);
if (!ctx)
continue;
- perf_event_aux_ctx(ctx, __perf_addr_filters_adjust, vma, true);
+ perf_iterate_ctx(ctx, __perf_addr_filters_adjust, vma, true);
}
rcu_read_unlock();
}
},
};
- perf_event_aux(perf_event_switch_output,
+ perf_iterate_sb(perf_event_switch_output,
&switch_event,
NULL);
}
static int perf_tp_filter_match(struct perf_event *event,
struct perf_sample_data *data)
{
- void *record = data->raw->data;
+ void *record = data->raw->frag.data;
/* only top level events have filters set */
if (event->parent)
struct perf_event *event;
struct perf_raw_record raw = {
- .size = entry_size,
- .data = record,
+ .frag = {
+ .size = entry_size,
+ .data = record,
+ },
};
perf_sample_data_init(&data, 0, 0);
prog = event->tp_event->prog;
if (prog) {
event->tp_event->prog = NULL;
- bpf_prog_put_rcu(prog);
+ bpf_prog_put(prog);
}
}
return pmu;
}
+static void attach_sb_event(struct perf_event *event)
+{
+ struct pmu_event_list *pel = per_cpu_ptr(&pmu_sb_events, event->cpu);
+
+ raw_spin_lock(&pel->lock);
+ list_add_rcu(&event->sb_list, &pel->list);
+ raw_spin_unlock(&pel->lock);
+}
+
+/*
+ * We keep a list of all !task (and therefore per-cpu) events
+ * that need to receive side-band records.
+ *
+ * This avoids having to scan all the various PMU per-cpu contexts
+ * looking for them.
+ */
+static void account_pmu_sb_event(struct perf_event *event)
+{
+ if (is_sb_event(event))
+ attach_sb_event(event);
+}
+
static void account_event_cpu(struct perf_event *event, int cpu)
{
if (event->parent)
enabled:
account_event_cpu(event, event->cpu);
+
+ account_pmu_sb_event(event);
}
/*
if (!event->parent) {
if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) {
- err = get_callchain_buffers();
+ err = get_callchain_buffers(attr->sample_max_stack);
if (err)
goto err_addr_filters;
}
return -EINVAL;
}
+ if (!attr.sample_max_stack)
+ attr.sample_max_stack = sysctl_perf_event_max_stack;
+
/*
* In cgroup mode, the pid argument is used to pass the fd
* opened to the cgroup directory in cgroupfs. The cpu argument
if (is_sampling_event(event)) {
if (event->pmu->capabilities & PERF_PMU_CAP_NO_INTERRUPT) {
- err = -ENOTSUPP;
+ err = -EOPNOTSUPP;
goto err_alloc;
}
}
swhash = &per_cpu(swevent_htable, cpu);
mutex_init(&swhash->hlist_mutex);
INIT_LIST_HEAD(&per_cpu(active_ctx_list, cpu));
+
+ INIT_LIST_HEAD(&per_cpu(pmu_sb_events.list, cpu));
+ raw_spin_lock_init(&per_cpu(pmu_sb_events.lock, cpu));
}
}
-static void perf_event_init_cpu(int cpu)
+int perf_event_init_cpu(unsigned int cpu)
{
struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
rcu_assign_pointer(swhash->swevent_hlist, hlist);
}
mutex_unlock(&swhash->hlist_mutex);
+ return 0;
}
#if defined CONFIG_HOTPLUG_CPU || defined CONFIG_KEXEC_CORE
}
srcu_read_unlock(&pmus_srcu, idx);
}
+#else
-static void perf_event_exit_cpu(int cpu)
+static void perf_event_exit_cpu_context(int cpu) { }
+
+#endif
+
+int perf_event_exit_cpu(unsigned int cpu)
{
perf_event_exit_cpu_context(cpu);
+ return 0;
}
-#else
-static inline void perf_event_exit_cpu(int cpu) { }
-#endif
static int
perf_reboot(struct notifier_block *notifier, unsigned long val, void *v)
.priority = INT_MIN,
};
-static int
-perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu)
-{
- unsigned int cpu = (long)hcpu;
-
- switch (action & ~CPU_TASKS_FROZEN) {
-
- case CPU_UP_PREPARE:
- /*
- * This must be done before the CPU comes alive, because the
- * moment we can run tasks we can encounter (software) events.
- *
- * Specifically, someone can have inherited events on kthreadd
- * or a pre-existing worker thread that gets re-bound.
- */
- perf_event_init_cpu(cpu);
- break;
-
- case CPU_DOWN_PREPARE:
- /*
- * This must be done before the CPU dies because after that an
- * active event might want to IPI the CPU and that'll not work
- * so great for dead CPUs.
- *
- * XXX smp_call_function_single() return -ENXIO without a warn
- * so we could possibly deal with this.
- *
- * This is safe against new events arriving because
- * sys_perf_event_open() serializes against hotplug using
- * get_online_cpus().
- */
- perf_event_exit_cpu(cpu);
- break;
- default:
- break;
- }
-
- return NOTIFY_OK;
-}
-
void __init perf_event_init(void)
{
int ret;
perf_pmu_register(&perf_cpu_clock, NULL, -1);
perf_pmu_register(&perf_task_clock, NULL, -1);
perf_tp_register();
- perf_cpu_notifier(perf_cpu_notify);
+ perf_event_init_cpu(smp_processor_id());
register_reboot_notifier(&perf_reboot_notifier);
ret = init_hw_breakpoint();