X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=kernel%2Fperf_event.c;h=603c0d8b5df17a93c8b9679be42dd8905b5eaf97;hb=285caad415f459f336247932b4db95a571357a02;hp=8e55b440e28a62b91b119a43b1fae266512299bb;hpb=0cff784ae41cc125368ae77f1c01328ae2fdc6b3;p=safe%2Fjmp%2Flinux-2.6 diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 8e55b44..603c0d8b 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -36,7 +36,7 @@ /* * Each CPU has a list of per CPU events: */ -DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context); +static DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context); int perf_max_events __read_mostly = 1; static int perf_reserved_percpu __read_mostly; @@ -203,14 +203,14 @@ perf_lock_task_context(struct task_struct *task, unsigned long *flags) * if so. If we locked the right context, then it * can't get swapped on us any more. */ - spin_lock_irqsave(&ctx->lock, *flags); + raw_spin_lock_irqsave(&ctx->lock, *flags); if (ctx != rcu_dereference(task->perf_event_ctxp)) { - spin_unlock_irqrestore(&ctx->lock, *flags); + raw_spin_unlock_irqrestore(&ctx->lock, *flags); goto retry; } if (!atomic_inc_not_zero(&ctx->refcount)) { - spin_unlock_irqrestore(&ctx->lock, *flags); + raw_spin_unlock_irqrestore(&ctx->lock, *flags); ctx = NULL; } } @@ -231,7 +231,7 @@ static struct perf_event_context *perf_pin_task_context(struct task_struct *task ctx = perf_lock_task_context(task, &flags); if (ctx) { ++ctx->pin_count; - spin_unlock_irqrestore(&ctx->lock, flags); + raw_spin_unlock_irqrestore(&ctx->lock, flags); } return ctx; } @@ -240,12 +240,55 @@ static void perf_unpin_context(struct perf_event_context *ctx) { unsigned long flags; - spin_lock_irqsave(&ctx->lock, flags); + raw_spin_lock_irqsave(&ctx->lock, flags); --ctx->pin_count; - spin_unlock_irqrestore(&ctx->lock, flags); + raw_spin_unlock_irqrestore(&ctx->lock, flags); put_ctx(ctx); } +static inline u64 perf_clock(void) +{ + return cpu_clock(smp_processor_id()); +} + +/* + * Update the record of the current time in a context. + */ +static void update_context_time(struct perf_event_context *ctx) +{ + u64 now = perf_clock(); + + ctx->time += now - ctx->timestamp; + ctx->timestamp = now; +} + +/* + * Update the total_time_enabled and total_time_running fields for a event. + */ +static void update_event_times(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + u64 run_end; + + if (event->state < PERF_EVENT_STATE_INACTIVE || + event->group_leader->state < PERF_EVENT_STATE_INACTIVE) + return; + + if (ctx->is_active) + run_end = ctx->time; + else + run_end = event->tstamp_stopped; + + event->total_time_enabled = run_end - event->tstamp_enabled; + + if (event->state == PERF_EVENT_STATE_INACTIVE) + run_end = event->tstamp_stopped; + else + run_end = ctx->time; + + event->total_time_running = run_end - event->tstamp_running; +} + /* * Add a event from the lists for its context. * Must be called with ctx->mutex and ctx->lock held. @@ -294,6 +337,18 @@ list_del_event(struct perf_event *event, struct perf_event_context *ctx) if (event->group_leader != event) event->group_leader->nr_siblings--; + update_event_times(event); + + /* + * If event was in error state, then keep it + * that way, otherwise bogus counts will be + * returned on read(). The only way to get out + * of error state is by explicit re-enabling + * of the event + */ + if (event->state > PERF_EVENT_STATE_OFF) + event->state = PERF_EVENT_STATE_OFF; + /* * If this was a group event with sibling events then * upgrade the siblings to singleton events by adding them @@ -372,7 +427,7 @@ static void __perf_event_remove_from_context(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); /* * Protect the list operation against NMI by disabling the * events on a global level. @@ -394,7 +449,7 @@ static void __perf_event_remove_from_context(void *info) } perf_enable(); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } @@ -421,7 +476,7 @@ static void perf_event_remove_from_context(struct perf_event *event) if (!task) { /* * Per cpu events are removed via an smp call and - * the removal is always sucessful. + * the removal is always successful. */ smp_call_function_single(event->cpu, __perf_event_remove_from_context, @@ -433,12 +488,12 @@ retry: task_oncpu_function_call(task, __perf_event_remove_from_context, event); - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); /* * If the context is active we need to retry the smp call. */ if (ctx->nr_active && !list_empty(&event->group_entry)) { - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); goto retry; } @@ -447,48 +502,9 @@ retry: * can remove the event safely, if the call above did not * succeed. */ - if (!list_empty(&event->group_entry)) { + if (!list_empty(&event->group_entry)) list_del_event(event, ctx); - } - spin_unlock_irq(&ctx->lock); -} - -static inline u64 perf_clock(void) -{ - return cpu_clock(smp_processor_id()); -} - -/* - * Update the record of the current time in a context. - */ -static void update_context_time(struct perf_event_context *ctx) -{ - u64 now = perf_clock(); - - ctx->time += now - ctx->timestamp; - ctx->timestamp = now; -} - -/* - * Update the total_time_enabled and total_time_running fields for a event. - */ -static void update_event_times(struct perf_event *event) -{ - struct perf_event_context *ctx = event->ctx; - u64 run_end; - - if (event->state < PERF_EVENT_STATE_INACTIVE || - event->group_leader->state < PERF_EVENT_STATE_INACTIVE) - return; - - event->total_time_enabled = ctx->time - event->tstamp_enabled; - - if (event->state == PERF_EVENT_STATE_INACTIVE) - run_end = event->tstamp_stopped; - else - run_end = ctx->time; - - event->total_time_running = run_end - event->tstamp_running; + raw_spin_unlock_irq(&ctx->lock); } /* @@ -519,7 +535,7 @@ static void __perf_event_disable(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); /* * If the event is on, turn it off. @@ -535,7 +551,7 @@ static void __perf_event_disable(void *info) event->state = PERF_EVENT_STATE_OFF; } - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -551,7 +567,7 @@ static void __perf_event_disable(void *info) * is the current context on this CPU and preemption is disabled, * hence we can't get into perf_event_task_sched_out for this context. */ -static void perf_event_disable(struct perf_event *event) +void perf_event_disable(struct perf_event *event) { struct perf_event_context *ctx = event->ctx; struct task_struct *task = ctx->task; @@ -568,12 +584,12 @@ static void perf_event_disable(struct perf_event *event) retry: task_oncpu_function_call(task, __perf_event_disable, event); - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); /* * If the event is still active, we need to retry the cross-call. */ if (event->state == PERF_EVENT_STATE_ACTIVE) { - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); goto retry; } @@ -586,7 +602,7 @@ static void perf_event_disable(struct perf_event *event) event->state = PERF_EVENT_STATE_OFF; } - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); } static int @@ -754,7 +770,7 @@ static void __perf_install_in_context(void *info) cpuctx->task_ctx = ctx; } - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); ctx->is_active = 1; update_context_time(ctx); @@ -766,6 +782,9 @@ static void __perf_install_in_context(void *info) add_event_to_ctx(event, ctx); + if (event->cpu != -1 && event->cpu != smp_processor_id()) + goto unlock; + /* * Don't put the event on if it is disabled or if * it is in a group and the group isn't on. @@ -804,7 +823,7 @@ static void __perf_install_in_context(void *info) unlock: perf_enable(); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -829,7 +848,7 @@ perf_install_in_context(struct perf_event_context *ctx, if (!task) { /* * Per cpu events are installed via an smp call and - * the install is always sucessful. + * the install is always successful. */ smp_call_function_single(cpu, __perf_install_in_context, event, 1); @@ -840,12 +859,12 @@ retry: task_oncpu_function_call(task, __perf_install_in_context, event); - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); /* * we need to retry the smp call. */ if (ctx->is_active && list_empty(&event->group_entry)) { - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); goto retry; } @@ -856,7 +875,7 @@ retry: */ if (list_empty(&event->group_entry)) add_event_to_ctx(event, ctx); - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); } /* @@ -901,7 +920,7 @@ static void __perf_event_enable(void *info) cpuctx->task_ctx = ctx; } - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); ctx->is_active = 1; update_context_time(ctx); @@ -909,6 +928,9 @@ static void __perf_event_enable(void *info) goto unlock; __perf_event_mark_enabled(event, ctx); + if (event->cpu != -1 && event->cpu != smp_processor_id()) + goto unlock; + /* * If the event is in a group and isn't the group leader, * then don't put it on unless the group is on. @@ -943,7 +965,7 @@ static void __perf_event_enable(void *info) } unlock: - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -955,7 +977,7 @@ static void __perf_event_enable(void *info) * perf_event_for_each_child or perf_event_for_each as described * for perf_event_disable. */ -static void perf_event_enable(struct perf_event *event) +void perf_event_enable(struct perf_event *event) { struct perf_event_context *ctx = event->ctx; struct task_struct *task = ctx->task; @@ -969,7 +991,7 @@ static void perf_event_enable(struct perf_event *event) return; } - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); if (event->state >= PERF_EVENT_STATE_INACTIVE) goto out; @@ -984,10 +1006,10 @@ static void perf_event_enable(struct perf_event *event) event->state = PERF_EVENT_STATE_OFF; retry: - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); task_oncpu_function_call(task, __perf_event_enable, event); - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); /* * If the context is active and the event is still off, @@ -1004,7 +1026,7 @@ static void perf_event_enable(struct perf_event *event) __perf_event_mark_enabled(event, ctx); out: - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); } static int perf_event_refresh(struct perf_event *event, int refresh) @@ -1026,20 +1048,20 @@ void __perf_event_sched_out(struct perf_event_context *ctx, { struct perf_event *event; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); ctx->is_active = 0; if (likely(!ctx->nr_events)) goto out; update_context_time(ctx); perf_disable(); - if (ctx->nr_active) + if (ctx->nr_active) { list_for_each_entry(event, &ctx->group_list, group_entry) group_sched_out(event, cpuctx, ctx); - + } perf_enable(); out: - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -1061,8 +1083,6 @@ static int context_equiv(struct perf_event_context *ctx1, && !ctx1->pin_count && !ctx2->pin_count; } -static void __perf_event_read(void *event); - static void __perf_event_sync_stat(struct perf_event *event, struct perf_event *next_event) { @@ -1080,8 +1100,8 @@ static void __perf_event_sync_stat(struct perf_event *event, */ switch (event->state) { case PERF_EVENT_STATE_ACTIVE: - __perf_event_read(event); - break; + event->pmu->read(event); + /* fall-through */ case PERF_EVENT_STATE_INACTIVE: update_event_times(event); @@ -1120,6 +1140,8 @@ static void perf_event_sync_stat(struct perf_event_context *ctx, if (!ctx->nr_stat) return; + update_context_time(ctx); + event = list_first_entry(&ctx->event_list, struct perf_event, event_entry); @@ -1163,8 +1185,6 @@ void perf_event_task_sched_out(struct task_struct *task, if (likely(!ctx || !cpuctx->task_ctx)) return; - update_context_time(ctx); - rcu_read_lock(); parent = rcu_dereference(ctx->parent_ctx); next_ctx = next->perf_event_ctxp; @@ -1179,8 +1199,8 @@ void perf_event_task_sched_out(struct task_struct *task, * order we take the locks because no other cpu could * be trying to lock both of these tasks. */ - spin_lock(&ctx->lock); - spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING); + raw_spin_lock(&ctx->lock); + raw_spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING); if (context_equiv(ctx, next_ctx)) { /* * XXX do we need a memory barrier of sorts @@ -1194,8 +1214,8 @@ void perf_event_task_sched_out(struct task_struct *task, perf_event_sync_stat(ctx, next_ctx); } - spin_unlock(&next_ctx->lock); - spin_unlock(&ctx->lock); + raw_spin_unlock(&next_ctx->lock); + raw_spin_unlock(&ctx->lock); } rcu_read_unlock(); @@ -1237,7 +1257,7 @@ __perf_event_sched_in(struct perf_event_context *ctx, struct perf_event *event; int can_add_hw = 1; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); ctx->is_active = 1; if (likely(!ctx->nr_events)) goto out; @@ -1292,7 +1312,7 @@ __perf_event_sched_in(struct perf_event_context *ctx, } perf_enable(); out: - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -1356,11 +1376,14 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) struct hw_perf_event *hwc; u64 interrupts, freq; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (event->state != PERF_EVENT_STATE_ACTIVE) continue; + if (event->cpu != -1 && event->cpu != smp_processor_id()) + continue; + hwc = &event->hw; interrupts = hwc->interrupts; @@ -1411,7 +1434,7 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) perf_enable(); } } - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } /* @@ -1424,7 +1447,7 @@ static void rotate_ctx(struct perf_event_context *ctx) if (!ctx->nr_events) return; - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); /* * Rotate the first entry last (works just fine for group events too): */ @@ -1435,7 +1458,7 @@ static void rotate_ctx(struct perf_event_context *ctx) } perf_enable(); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); } void perf_event_task_tick(struct task_struct *curr, int cpu) @@ -1484,7 +1507,7 @@ static void perf_event_enable_on_exec(struct task_struct *task) __perf_event_task_sched_out(ctx); - spin_lock(&ctx->lock); + raw_spin_lock(&ctx->lock); list_for_each_entry(event, &ctx->group_list, group_entry) { if (!event->attr.enable_on_exec) @@ -1502,7 +1525,7 @@ static void perf_event_enable_on_exec(struct task_struct *task) if (enabled) unclone_ctx(ctx); - spin_unlock(&ctx->lock); + raw_spin_unlock(&ctx->lock); perf_event_task_sched_in(task, smp_processor_id()); out: @@ -1517,7 +1540,6 @@ static void __perf_event_read(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_event *event = info; struct perf_event_context *ctx = event->ctx; - unsigned long flags; /* * If this is a task context, we need to check whether it is @@ -1529,12 +1551,12 @@ static void __perf_event_read(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - local_irq_save(flags); - if (ctx->is_active) - update_context_time(ctx); - event->pmu->read(event); + raw_spin_lock(&ctx->lock); + update_context_time(ctx); update_event_times(event); - local_irq_restore(flags); + raw_spin_unlock(&ctx->lock); + + event->pmu->read(event); } static u64 perf_event_read(struct perf_event *event) @@ -1547,7 +1569,13 @@ static u64 perf_event_read(struct perf_event *event) smp_call_function_single(event->oncpu, __perf_event_read, event, 1); } else if (event->state == PERF_EVENT_STATE_INACTIVE) { + struct perf_event_context *ctx = event->ctx; + unsigned long flags; + + raw_spin_lock_irqsave(&ctx->lock, flags); + update_context_time(ctx); update_event_times(event); + raw_spin_unlock_irqrestore(&ctx->lock, flags); } return atomic64_read(&event->count); @@ -1560,8 +1588,7 @@ static void __perf_event_init_context(struct perf_event_context *ctx, struct task_struct *task) { - memset(ctx, 0, sizeof(*ctx)); - spin_lock_init(&ctx->lock); + raw_spin_lock_init(&ctx->lock); mutex_init(&ctx->mutex); INIT_LIST_HEAD(&ctx->group_list); INIT_LIST_HEAD(&ctx->event_list); @@ -1577,15 +1604,12 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) unsigned long flags; int err; - /* - * If cpu is not a wildcard then this is a percpu event: - */ - if (cpu != -1) { + if (pid == -1 && cpu != -1) { /* Must be root to operate on a CPU event: */ if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN)) return ERR_PTR(-EACCES); - if (cpu < 0 || cpu > num_possible_cpus()) + if (cpu < 0 || cpu >= nr_cpumask_bits) return ERR_PTR(-EINVAL); /* @@ -1593,7 +1617,7 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) * offline CPU and activate it when the CPU comes up, but * that's for later. */ - if (!cpu_isset(cpu, cpu_online_map)) + if (!cpu_online(cpu)) return ERR_PTR(-ENODEV); cpuctx = &per_cpu(perf_cpu_context, cpu); @@ -1631,11 +1655,11 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu) ctx = perf_lock_task_context(task, &flags); if (ctx) { unclone_ctx(ctx); - spin_unlock_irqrestore(&ctx->lock, flags); + raw_spin_unlock_irqrestore(&ctx->lock, flags); } if (!ctx) { - ctx = kmalloc(sizeof(struct perf_event_context), GFP_KERNEL); + ctx = kzalloc(sizeof(struct perf_event_context), GFP_KERNEL); err = -ENOMEM; if (!ctx) goto errout; @@ -1701,16 +1725,10 @@ static void free_event(struct perf_event *event) call_rcu(&event->rcu_head, free_event_rcu); } -/* - * Called when the last reference to the file is gone. - */ -static int perf_release(struct inode *inode, struct file *file) +int perf_event_release_kernel(struct perf_event *event) { - struct perf_event *event = file->private_data; struct perf_event_context *ctx = event->ctx; - file->private_data = NULL; - WARN_ON_ONCE(ctx->parent_ctx); mutex_lock(&ctx->mutex); perf_event_remove_from_context(event); @@ -1725,26 +1743,19 @@ static int perf_release(struct inode *inode, struct file *file) return 0; } +EXPORT_SYMBOL_GPL(perf_event_release_kernel); -int perf_event_release_kernel(struct perf_event *event) +/* + * Called when the last reference to the file is gone. + */ +static int perf_release(struct inode *inode, struct file *file) { - struct perf_event_context *ctx = event->ctx; - - WARN_ON_ONCE(ctx->parent_ctx); - mutex_lock(&ctx->mutex); - perf_event_remove_from_context(event); - mutex_unlock(&ctx->mutex); - - mutex_lock(&event->owner->perf_event_mutex); - list_del_init(&event->owner_entry); - mutex_unlock(&event->owner->perf_event_mutex); - put_task_struct(event->owner); + struct perf_event *event = file->private_data; - free_event(event); + file->private_data = NULL; - return 0; + return perf_event_release_kernel(event); } -EXPORT_SYMBOL_GPL(perf_event_release_kernel); static int perf_event_read_size(struct perf_event *event) { @@ -1771,92 +1782,94 @@ static int perf_event_read_size(struct perf_event *event) return size; } -u64 perf_event_read_value(struct perf_event *event) +u64 perf_event_read_value(struct perf_event *event, u64 *enabled, u64 *running) { struct perf_event *child; u64 total = 0; + *enabled = 0; + *running = 0; + + mutex_lock(&event->child_mutex); total += perf_event_read(event); - list_for_each_entry(child, &event->child_list, child_list) + *enabled += event->total_time_enabled + + atomic64_read(&event->child_total_time_enabled); + *running += event->total_time_running + + atomic64_read(&event->child_total_time_running); + + list_for_each_entry(child, &event->child_list, child_list) { total += perf_event_read(child); + *enabled += child->total_time_enabled; + *running += child->total_time_running; + } + mutex_unlock(&event->child_mutex); return total; } EXPORT_SYMBOL_GPL(perf_event_read_value); -static int perf_event_read_entry(struct perf_event *event, - u64 read_format, char __user *buf) -{ - int n = 0, count = 0; - u64 values[2]; - - values[n++] = perf_event_read_value(event); - if (read_format & PERF_FORMAT_ID) - values[n++] = primary_event_id(event); - - count = n * sizeof(u64); - - if (copy_to_user(buf, values, count)) - return -EFAULT; - - return count; -} - static int perf_event_read_group(struct perf_event *event, u64 read_format, char __user *buf) { struct perf_event *leader = event->group_leader, *sub; - int n = 0, size = 0, err = -EFAULT; - u64 values[3]; + int n = 0, size = 0, ret = -EFAULT; + struct perf_event_context *ctx = leader->ctx; + u64 values[5]; + u64 count, enabled, running; + + mutex_lock(&ctx->mutex); + count = perf_event_read_value(leader, &enabled, &running); values[n++] = 1 + leader->nr_siblings; - if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { - values[n++] = leader->total_time_enabled + - atomic64_read(&leader->child_total_time_enabled); - } - if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { - values[n++] = leader->total_time_running + - atomic64_read(&leader->child_total_time_running); - } + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + values[n++] = enabled; + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + values[n++] = running; + values[n++] = count; + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(leader); size = n * sizeof(u64); if (copy_to_user(buf, values, size)) - return -EFAULT; - - err = perf_event_read_entry(leader, read_format, buf + size); - if (err < 0) - return err; + goto unlock; - size += err; + ret = size; list_for_each_entry(sub, &leader->sibling_list, group_entry) { - err = perf_event_read_entry(sub, read_format, - buf + size); - if (err < 0) - return err; + n = 0; + + values[n++] = perf_event_read_value(sub, &enabled, &running); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(sub); + + size = n * sizeof(u64); + + if (copy_to_user(buf + ret, values, size)) { + ret = -EFAULT; + goto unlock; + } - size += err; + ret += size; } +unlock: + mutex_unlock(&ctx->mutex); - return size; + return ret; } static int perf_event_read_one(struct perf_event *event, u64 read_format, char __user *buf) { + u64 enabled, running; u64 values[4]; int n = 0; - values[n++] = perf_event_read_value(event); - if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { - values[n++] = event->total_time_enabled + - atomic64_read(&event->child_total_time_enabled); - } - if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { - values[n++] = event->total_time_running + - atomic64_read(&event->child_total_time_running); - } + values[n++] = perf_event_read_value(event, &enabled, &running); + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + values[n++] = enabled; + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + values[n++] = running; if (read_format & PERF_FORMAT_ID) values[n++] = primary_event_id(event); @@ -1887,12 +1900,10 @@ perf_read_hw(struct perf_event *event, char __user *buf, size_t count) return -ENOSPC; WARN_ON_ONCE(event->ctx->parent_ctx); - mutex_lock(&event->child_mutex); if (read_format & PERF_FORMAT_GROUP) ret = perf_event_read_group(event, read_format, buf); else ret = perf_event_read_one(event, read_format, buf); - mutex_unlock(&event->child_mutex); return ret; } @@ -1982,7 +1993,7 @@ static int perf_event_period(struct perf_event *event, u64 __user *arg) if (!value) return -EINVAL; - spin_lock_irq(&ctx->lock); + raw_spin_lock_irq(&ctx->lock); if (event->attr.freq) { if (value > sysctl_perf_event_sample_rate) { ret = -EINVAL; @@ -1995,7 +2006,7 @@ static int perf_event_period(struct perf_event *event, u64 __user *arg) event->hw.sample_period = value; } unlock: - spin_unlock_irq(&ctx->lock); + raw_spin_unlock_irq(&ctx->lock); return ret; } @@ -2204,6 +2215,7 @@ static void perf_mmap_data_free(struct perf_mmap_data *data) perf_mmap_free_page((unsigned long)data->user_page); for (i = 0; i < data->nr_pages; i++) perf_mmap_free_page((unsigned long)data->data_pages[i]); + kfree(data); } #else @@ -2244,6 +2256,7 @@ static void perf_mmap_data_free_work(struct work_struct *work) perf_mmap_unmark_page(base + (i * PAGE_SIZE)); vfree(base); + kfree(data); } static void perf_mmap_data_free(struct perf_mmap_data *data) @@ -2337,7 +2350,7 @@ perf_mmap_data_init(struct perf_event *event, struct perf_mmap_data *data) } if (!data->watermark) - data->watermark = max_t(long, PAGE_SIZE, max_size / 2); + data->watermark = max_size / 2; rcu_assign_pointer(event->data, data); @@ -2349,7 +2362,6 @@ static void perf_mmap_data_free_rcu(struct rcu_head *rcu_head) data = container_of(rcu_head, struct perf_mmap_data, rcu_head); perf_mmap_data_free(data); - kfree(data); } static void perf_mmap_data_release(struct perf_event *event) @@ -3256,6 +3268,9 @@ static void perf_event_task_output(struct perf_event *event, static int perf_event_task_match(struct perf_event *event) { + if (event->cpu != -1 && event->cpu != smp_processor_id()) + return 0; + if (event->attr.comm || event->attr.mmap || event->attr.task) return 1; @@ -3267,15 +3282,10 @@ static void perf_event_task_ctx(struct perf_event_context *ctx, { struct perf_event *event; - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (perf_event_task_match(event)) perf_event_task_output(event, task_event); } - rcu_read_unlock(); } static void perf_event_task_event(struct perf_task_event *task_event) @@ -3283,15 +3293,14 @@ static void perf_event_task_event(struct perf_task_event *task_event) struct perf_cpu_context *cpuctx; struct perf_event_context *ctx = task_event->task_ctx; + rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_task_ctx(&cpuctx->ctx, task_event); - put_cpu_var(perf_cpu_context); - - rcu_read_lock(); if (!ctx) ctx = rcu_dereference(task_event->task->perf_event_ctxp); if (ctx) perf_event_task_ctx(ctx, task_event); + put_cpu_var(perf_cpu_context); rcu_read_unlock(); } @@ -3368,6 +3377,9 @@ static void perf_event_comm_output(struct perf_event *event, static int perf_event_comm_match(struct perf_event *event) { + if (event->cpu != -1 && event->cpu != smp_processor_id()) + return 0; + if (event->attr.comm) return 1; @@ -3379,15 +3391,10 @@ static void perf_event_comm_ctx(struct perf_event_context *ctx, { struct perf_event *event; - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (perf_event_comm_match(event)) perf_event_comm_output(event, comm_event); } - rcu_read_unlock(); } static void perf_event_comm_event(struct perf_comm_event *comm_event) @@ -3398,7 +3405,7 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event) char comm[TASK_COMM_LEN]; memset(comm, 0, sizeof(comm)); - strncpy(comm, comm_event->task->comm, sizeof(comm)); + strlcpy(comm, comm_event->task->comm, sizeof(comm)); size = ALIGN(strlen(comm)+1, sizeof(u64)); comm_event->comm = comm; @@ -3406,18 +3413,13 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event) comm_event->event_id.header.size = sizeof(comm_event->event_id) + size; + rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_comm_ctx(&cpuctx->ctx, comm_event); - put_cpu_var(perf_cpu_context); - - rcu_read_lock(); - /* - * doesn't really matter which of the child contexts the - * events ends up in. - */ ctx = rcu_dereference(current->perf_event_ctxp); if (ctx) perf_event_comm_ctx(ctx, comm_event); + put_cpu_var(perf_cpu_context); rcu_read_unlock(); } @@ -3492,6 +3494,9 @@ static void perf_event_mmap_output(struct perf_event *event, static int perf_event_mmap_match(struct perf_event *event, struct perf_mmap_event *mmap_event) { + if (event->cpu != -1 && event->cpu != smp_processor_id()) + return 0; + if (event->attr.mmap) return 1; @@ -3503,15 +3508,10 @@ static void perf_event_mmap_ctx(struct perf_event_context *ctx, { struct perf_event *event; - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (perf_event_mmap_match(event, mmap_event)) perf_event_mmap_output(event, mmap_event); } - rcu_read_unlock(); } static void perf_event_mmap_event(struct perf_mmap_event *mmap_event) @@ -3567,18 +3567,13 @@ got_name: mmap_event->event_id.header.size = sizeof(mmap_event->event_id) + size; + rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); perf_event_mmap_ctx(&cpuctx->ctx, mmap_event); - put_cpu_var(perf_cpu_context); - - rcu_read_lock(); - /* - * doesn't really matter which of the child contexts the - * events ends up in. - */ ctx = rcu_dereference(current->perf_event_ctxp); if (ctx) perf_event_mmap_ctx(ctx, mmap_event); + put_cpu_var(perf_cpu_context); rcu_read_unlock(); kfree(buf); @@ -3849,27 +3844,40 @@ static int perf_swevent_is_counting(struct perf_event *event) static int perf_tp_event_match(struct perf_event *event, struct perf_sample_data *data); +static int perf_exclude_event(struct perf_event *event, + struct pt_regs *regs) +{ + if (regs) { + if (event->attr.exclude_user && user_mode(regs)) + return 1; + + if (event->attr.exclude_kernel && !user_mode(regs)) + return 1; + } + + return 0; +} + static int perf_swevent_match(struct perf_event *event, enum perf_type_id type, u32 event_id, struct perf_sample_data *data, struct pt_regs *regs) { + if (event->cpu != -1 && event->cpu != smp_processor_id()) + return 0; + if (!perf_swevent_is_counting(event)) return 0; if (event->attr.type != type) return 0; + if (event->attr.config != event_id) return 0; - if (regs) { - if (event->attr.exclude_user && user_mode(regs)) - return 0; - - if (event->attr.exclude_kernel && !user_mode(regs)) - return 0; - } + if (perf_exclude_event(event, regs)) + return 0; if (event->attr.type == PERF_TYPE_TRACEPOINT && !perf_tp_event_match(event, data)) @@ -3886,49 +3894,59 @@ static void perf_swevent_ctx_event(struct perf_event_context *ctx, { struct perf_event *event; - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { if (perf_swevent_match(event, type, event_id, data, regs)) perf_swevent_add(event, nr, nmi, data, regs); } - rcu_read_unlock(); } -static int *perf_swevent_recursion_context(struct perf_cpu_context *cpuctx) +int perf_swevent_get_recursion_context(void) { + struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); + int rctx; + if (in_nmi()) - return &cpuctx->recursion[3]; + rctx = 3; + else if (in_irq()) + rctx = 2; + else if (in_softirq()) + rctx = 1; + else + rctx = 0; - if (in_irq()) - return &cpuctx->recursion[2]; + if (cpuctx->recursion[rctx]) { + put_cpu_var(perf_cpu_context); + return -1; + } - if (in_softirq()) - return &cpuctx->recursion[1]; + cpuctx->recursion[rctx]++; + barrier(); - return &cpuctx->recursion[0]; + return rctx; } +EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context); + +void perf_swevent_put_recursion_context(int rctx) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + barrier(); + cpuctx->recursion[rctx]--; + put_cpu_var(perf_cpu_context); +} +EXPORT_SYMBOL_GPL(perf_swevent_put_recursion_context); static void do_perf_sw_event(enum perf_type_id type, u32 event_id, u64 nr, int nmi, struct perf_sample_data *data, struct pt_regs *regs) { - struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); - int *recursion = perf_swevent_recursion_context(cpuctx); + struct perf_cpu_context *cpuctx; struct perf_event_context *ctx; - if (*recursion) - goto out; - - (*recursion)++; - barrier(); - + cpuctx = &__get_cpu_var(perf_cpu_context); + rcu_read_lock(); perf_swevent_ctx_event(&cpuctx->ctx, type, event_id, nr, nmi, data, regs); - rcu_read_lock(); /* * doesn't really matter which of the child contexts the * events ends up in. @@ -3937,23 +3955,24 @@ static void do_perf_sw_event(enum perf_type_id type, u32 event_id, if (ctx) perf_swevent_ctx_event(ctx, type, event_id, nr, nmi, data, regs); rcu_read_unlock(); - - barrier(); - (*recursion)--; - -out: - put_cpu_var(perf_cpu_context); } void __perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) { - struct perf_sample_data data = { - .addr = addr, - }; + struct perf_sample_data data; + int rctx; - do_perf_sw_event(PERF_TYPE_SOFTWARE, event_id, nr, nmi, - &data, regs); + rctx = perf_swevent_get_recursion_context(); + if (rctx < 0) + return; + + data.addr = addr; + data.raw = NULL; + + do_perf_sw_event(PERF_TYPE_SOFTWARE, event_id, nr, nmi, &data, regs); + + perf_swevent_put_recursion_context(rctx); } static void perf_swevent_read(struct perf_event *event) @@ -3998,6 +4017,8 @@ static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer) event->pmu->read(event); data.addr = 0; + data.raw = NULL; + data.period = event->hw.last_period; regs = get_irq_regs(); /* * In case we exclude kernel IPs or are somehow not in interrupt @@ -4066,8 +4087,7 @@ static void cpu_clock_perf_event_update(struct perf_event *event) u64 now; now = cpu_clock(cpu); - prev = atomic64_read(&event->hw.prev_count); - atomic64_set(&event->hw.prev_count, now); + prev = atomic64_xchg(&event->hw.prev_count, now); atomic64_add(now - prev, &event->count); } @@ -4176,6 +4196,7 @@ void perf_tp_event(int event_id, u64 addr, u64 count, void *record, if (!regs) regs = task_pt_regs(current); + /* Trace events already protected against recursion */ do_perf_sw_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, &data, regs); } @@ -4271,15 +4292,8 @@ static void bp_perf_event_destroy(struct perf_event *event) static const struct pmu *bp_perf_event_init(struct perf_event *bp) { int err; - /* - * The breakpoint is already filled if we haven't created the counter - * through perf syscall - * FIXME: manage to get trigerred to NULL if it comes from syscalls - */ - if (!bp->callback) - err = register_perf_hw_breakpoint(bp); - else - err = __register_perf_hw_breakpoint(bp); + + err = register_perf_hw_breakpoint(bp); if (err) return ERR_PTR(err); @@ -4288,15 +4302,18 @@ static const struct pmu *bp_perf_event_init(struct perf_event *bp) return &perf_ops_bp; } -void perf_bp_event(struct perf_event *bp, void *regs) +void perf_bp_event(struct perf_event *bp, void *data) { - /* TODO */ + struct perf_sample_data sample; + struct pt_regs *regs = data; + + sample.raw = NULL; + sample.addr = bp->attr.bp_addr; + + if (!perf_exclude_event(bp, regs)) + perf_swevent_add(bp, 1, 1, &sample, regs); } #else -static void bp_perf_event_destroy(struct perf_event *event) -{ -} - static const struct pmu *bp_perf_event_init(struct perf_event *bp) { return NULL; @@ -4373,7 +4390,7 @@ perf_event_alloc(struct perf_event_attr *attr, struct perf_event_context *ctx, struct perf_event *group_leader, struct perf_event *parent_event, - perf_callback_t callback, + perf_overflow_handler_t overflow_handler, gfp_t gfpflags) { const struct pmu *pmu; @@ -4416,10 +4433,10 @@ perf_event_alloc(struct perf_event_attr *attr, event->state = PERF_EVENT_STATE_INACTIVE; - if (!callback && parent_event) - callback = parent_event->callback; + if (!overflow_handler && parent_event) + overflow_handler = parent_event->overflow_handler; - event->callback = callback; + event->overflow_handler = overflow_handler; if (attr->disabled) event->state = PERF_EVENT_STATE_OFF; @@ -4554,7 +4571,7 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr, if (attr->type >= PERF_TYPE_MAX) return -EINVAL; - if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) + if (attr->__reserved_1 || attr->__reserved_2) return -EINVAL; if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) @@ -4707,7 +4724,7 @@ SYSCALL_DEFINE5(perf_event_open, if (IS_ERR(event)) goto err_put_context; - err = anon_inode_getfd("[perf_event]", &perf_fops, event, 0); + err = anon_inode_getfd("[perf_event]", &perf_fops, event, O_RDWR); if (err < 0) goto err_free_put_context; @@ -4759,7 +4776,8 @@ err_put_context: */ struct perf_event * perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, - pid_t pid, perf_callback_t callback) + pid_t pid, + perf_overflow_handler_t overflow_handler) { struct perf_event *event; struct perf_event_context *ctx; @@ -4770,14 +4788,17 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, */ ctx = find_get_context(pid, cpu); - if (IS_ERR(ctx)) - return NULL; + if (IS_ERR(ctx)) { + err = PTR_ERR(ctx); + goto err_exit; + } event = perf_event_alloc(attr, cpu, ctx, NULL, - NULL, callback, GFP_KERNEL); - err = PTR_ERR(event); - if (IS_ERR(event)) + NULL, overflow_handler, GFP_KERNEL); + if (IS_ERR(event)) { + err = PTR_ERR(event); goto err_put_context; + } event->filp = NULL; WARN_ON_ONCE(ctx->parent_ctx); @@ -4794,11 +4815,10 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, return event; -err_put_context: - if (err < 0) - put_ctx(ctx); - - return NULL; + err_put_context: + put_ctx(ctx); + err_exit: + return ERR_PTR(err); } EXPORT_SYMBOL_GPL(perf_event_create_kernel_counter); @@ -4936,7 +4956,6 @@ __perf_event_exit_task(struct perf_event *child_event, { struct perf_event *parent_event; - update_event_times(child_event); perf_event_remove_from_context(child_event); parent_event = child_event->parent; @@ -4980,7 +4999,7 @@ void perf_event_exit_task(struct task_struct *child) * reading child->perf_event_ctxp, we wait until it has * incremented the context's refcount before we do put_ctx below. */ - spin_lock(&child_ctx->lock); + raw_spin_lock(&child_ctx->lock); child->perf_event_ctxp = NULL; /* * If this context is a clone; unclone it so it can't get @@ -4988,7 +5007,8 @@ void perf_event_exit_task(struct task_struct *child) * the events from it. */ unclone_ctx(child_ctx); - spin_unlock_irqrestore(&child_ctx->lock, flags); + update_context_time(child_ctx); + raw_spin_unlock_irqrestore(&child_ctx->lock, flags); /* * Report the task dead after unscheduling the events so that we @@ -5071,7 +5091,7 @@ again: */ int perf_event_init_task(struct task_struct *child) { - struct perf_event_context *child_ctx, *parent_ctx; + struct perf_event_context *child_ctx = NULL, *parent_ctx; struct perf_event_context *cloned_ctx; struct perf_event *event; struct task_struct *parent = current; @@ -5087,20 +5107,6 @@ int perf_event_init_task(struct task_struct *child) return 0; /* - * This is executed from the parent task context, so inherit - * events that have been marked for cloning. - * First allocate and initialize a context for the child. - */ - - child_ctx = kmalloc(sizeof(struct perf_event_context), GFP_KERNEL); - if (!child_ctx) - return -ENOMEM; - - __perf_event_init_context(child_ctx, child); - child->perf_event_ctxp = child_ctx; - get_task_struct(child); - - /* * If the parent's context is a clone, pin it so it won't get * swapped under us. */ @@ -5130,6 +5136,26 @@ int perf_event_init_task(struct task_struct *child) continue; } + if (!child->perf_event_ctxp) { + /* + * This is executed from the parent task context, so + * inherit events that have been marked for cloning. + * First allocate and initialize a context for the + * child. + */ + + child_ctx = kzalloc(sizeof(struct perf_event_context), + GFP_KERNEL); + if (!child_ctx) { + ret = -ENOMEM; + break; + } + + __perf_event_init_context(child_ctx, child); + child->perf_event_ctxp = child_ctx; + get_task_struct(child); + } + ret = inherit_group(event, parent, parent_ctx, child, child_ctx); if (ret) { @@ -5138,7 +5164,7 @@ int perf_event_init_task(struct task_struct *child) } } - if (inherited_all) { + if (child_ctx && inherited_all) { /* * Mark the child context as a clone of the parent * context, or of whatever the parent is a clone of. @@ -5272,11 +5298,11 @@ perf_set_reserve_percpu(struct sysdev_class *class, perf_reserved_percpu = val; for_each_online_cpu(cpu) { cpuctx = &per_cpu(perf_cpu_context, cpu); - spin_lock_irq(&cpuctx->ctx.lock); + raw_spin_lock_irq(&cpuctx->ctx.lock); mpt = min(perf_max_events - cpuctx->ctx.nr_events, perf_max_events - perf_reserved_percpu); cpuctx->max_pertask = mpt; - spin_unlock_irq(&cpuctx->ctx.lock); + raw_spin_unlock_irq(&cpuctx->ctx.lock); } spin_unlock(&perf_resource_lock);