X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=kernel%2Fperf_counter.c;h=8cb94a52d1bb3c21154e8dd028eecc93e1588866;hb=d1737fdbec7f90edc52dd0c5c3767457f28e78d8;hp=7e9108efd3053e6dafccf81d82e304a7cc776327;hpb=43a21ea81a2400992561146327c4785ce7f7be38;p=safe%2Fjmp%2Flinux-2.6 diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 7e9108e..8cb94a5 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -42,14 +42,21 @@ static int perf_overcommit __read_mostly = 1; static atomic_t nr_counters __read_mostly; static atomic_t nr_mmap_counters __read_mostly; static atomic_t nr_comm_counters __read_mostly; +static atomic_t nr_task_counters __read_mostly; /* * perf counter paranoia level: - * 0 - not paranoid - * 1 - disallow cpu counters to unpriv - * 2 - disallow kernel profiling to unpriv + * -1 - not paranoid at all + * 0 - disallow raw tracepoint access for unpriv + * 1 - disallow cpu counters for unpriv + * 2 - disallow kernel profiling for unpriv */ -int sysctl_perf_counter_paranoid __read_mostly; +int sysctl_perf_counter_paranoid __read_mostly = 1; + +static inline bool perf_paranoid_tracepoint_raw(void) +{ + return sysctl_perf_counter_paranoid > -1; +} static inline bool perf_paranoid_cpu(void) { @@ -87,6 +94,7 @@ void __weak hw_perf_disable(void) { barrier(); } void __weak hw_perf_enable(void) { barrier(); } void __weak hw_perf_counter_setup(int cpu) { barrier(); } +void __weak hw_perf_counter_setup_online(int cpu) { barrier(); } int __weak hw_perf_group_sched_in(struct perf_counter *group_leader, @@ -98,16 +106,16 @@ hw_perf_group_sched_in(struct perf_counter *group_leader, void __weak perf_counter_print_debug(void) { } -static DEFINE_PER_CPU(int, disable_count); +static DEFINE_PER_CPU(int, perf_disable_count); void __perf_disable(void) { - __get_cpu_var(disable_count)++; + __get_cpu_var(perf_disable_count)++; } bool __perf_enable(void) { - return !--__get_cpu_var(disable_count); + return !--__get_cpu_var(perf_disable_count); } void perf_disable(void) @@ -124,7 +132,7 @@ void perf_enable(void) static void get_ctx(struct perf_counter_context *ctx) { - atomic_inc(&ctx->refcount); + WARN_ON(!atomic_inc_not_zero(&ctx->refcount)); } static void free_ctx(struct rcu_head *head) @@ -146,6 +154,28 @@ static void put_ctx(struct perf_counter_context *ctx) } } +static void unclone_ctx(struct perf_counter_context *ctx) +{ + if (ctx->parent_ctx) { + put_ctx(ctx->parent_ctx); + ctx->parent_ctx = NULL; + } +} + +/* + * If we inherit counters we want to return the parent counter id + * to userspace. + */ +static u64 primary_counter_id(struct perf_counter *counter) +{ + u64 id = counter->id; + + if (counter->parent) + id = counter->parent->id; + + return id; +} + /* * Get the perf_counter_context for a task and lock it. * This has to cope with with the fact that until it is locked, @@ -175,6 +205,11 @@ perf_lock_task_context(struct task_struct *task, unsigned long *flags) spin_unlock_irqrestore(&ctx->lock, *flags); goto retry; } + + if (!atomic_inc_not_zero(&ctx->refcount)) { + spin_unlock_irqrestore(&ctx->lock, *flags); + ctx = NULL; + } } rcu_read_unlock(); return ctx; @@ -193,7 +228,6 @@ static struct perf_counter_context *perf_pin_task_context(struct task_struct *ta ctx = perf_lock_task_context(task, &flags); if (ctx) { ++ctx->pin_count; - get_ctx(ctx); spin_unlock_irqrestore(&ctx->lock, flags); } return ctx; @@ -232,6 +266,8 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) list_add_rcu(&counter->event_entry, &ctx->event_list); ctx->nr_counters++; + if (counter->attr.inherit_stat) + ctx->nr_stat++; } /* @@ -246,6 +282,8 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) if (list_empty(&counter->list_entry)) return; ctx->nr_counters--; + if (counter->attr.inherit_stat) + ctx->nr_stat--; list_del_init(&counter->list_entry); list_del_rcu(&counter->event_entry); @@ -275,6 +313,10 @@ counter_sched_out(struct perf_counter *counter, return; counter->state = PERF_COUNTER_STATE_INACTIVE; + if (counter->pending_disable) { + counter->pending_disable = 0; + counter->state = PERF_COUNTER_STATE_OFF; + } counter->tstamp_stopped = ctx->time; counter->pmu->disable(counter); counter->oncpu = -1; @@ -433,7 +475,8 @@ static void update_counter_times(struct perf_counter *counter) struct perf_counter_context *ctx = counter->ctx; u64 run_end; - if (counter->state < PERF_COUNTER_STATE_INACTIVE) + if (counter->state < PERF_COUNTER_STATE_INACTIVE || + counter->group_leader->state < PERF_COUNTER_STATE_INACTIVE) return; counter->total_time_enabled = ctx->time - counter->tstamp_enabled; @@ -482,7 +525,7 @@ static void __perf_counter_disable(void *info) */ if (counter->state >= PERF_COUNTER_STATE_INACTIVE) { update_context_time(ctx); - update_counter_times(counter); + update_group_times(counter); if (counter == counter->group_leader) group_sched_out(counter, cpuctx, ctx); else @@ -537,7 +580,7 @@ static void perf_counter_disable(struct perf_counter *counter) * in, so we can change the state safely. */ if (counter->state == PERF_COUNTER_STATE_INACTIVE) { - update_counter_times(counter); + update_group_times(counter); counter->state = PERF_COUNTER_STATE_OFF; } @@ -815,6 +858,27 @@ retry: } /* + * Put a counter into inactive state and update time fields. + * Enabling the leader of a group effectively enables all + * the group members that aren't explicitly disabled, so we + * have to update their ->tstamp_enabled also. + * Note: this works for group members as well as group leaders + * since the non-leader members' sibling_lists will be empty. + */ +static void __perf_counter_mark_enabled(struct perf_counter *counter, + struct perf_counter_context *ctx) +{ + struct perf_counter *sub; + + counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->tstamp_enabled = ctx->time - counter->total_time_enabled; + list_for_each_entry(sub, &counter->sibling_list, list_entry) + if (sub->state >= PERF_COUNTER_STATE_INACTIVE) + sub->tstamp_enabled = + ctx->time - sub->total_time_enabled; +} + +/* * Cross CPU call to enable a performance counter */ static void __perf_counter_enable(void *info) @@ -841,8 +905,7 @@ static void __perf_counter_enable(void *info) if (counter->state >= PERF_COUNTER_STATE_INACTIVE) goto unlock; - counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_enabled = ctx->time - counter->total_time_enabled; + __perf_counter_mark_enabled(counter, ctx); /* * If the counter is in a group and isn't the group leader, @@ -935,11 +998,9 @@ static void perf_counter_enable(struct perf_counter *counter) * Since we have the lock this context can't be scheduled * in, so we can change the state safely. */ - if (counter->state == PERF_COUNTER_STATE_OFF) { - counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_enabled = - ctx->time - counter->total_time_enabled; - } + if (counter->state == PERF_COUNTER_STATE_OFF) + __perf_counter_mark_enabled(counter, ctx); + out: spin_unlock_irq(&ctx->lock); } @@ -1002,6 +1063,81 @@ static int context_equiv(struct perf_counter_context *ctx1, && !ctx1->pin_count && !ctx2->pin_count; } +static void __perf_counter_read(void *counter); + +static void __perf_counter_sync_stat(struct perf_counter *counter, + struct perf_counter *next_counter) +{ + u64 value; + + if (!counter->attr.inherit_stat) + return; + + /* + * Update the counter value, we cannot use perf_counter_read() + * because we're in the middle of a context switch and have IRQs + * disabled, which upsets smp_call_function_single(), however + * we know the counter must be on the current CPU, therefore we + * don't need to use it. + */ + switch (counter->state) { + case PERF_COUNTER_STATE_ACTIVE: + __perf_counter_read(counter); + break; + + case PERF_COUNTER_STATE_INACTIVE: + update_counter_times(counter); + break; + + default: + break; + } + + /* + * In order to keep per-task stats reliable we need to flip the counter + * values when we flip the contexts. + */ + value = atomic64_read(&next_counter->count); + value = atomic64_xchg(&counter->count, value); + atomic64_set(&next_counter->count, value); + + swap(counter->total_time_enabled, next_counter->total_time_enabled); + swap(counter->total_time_running, next_counter->total_time_running); + + /* + * Since we swizzled the values, update the user visible data too. + */ + perf_counter_update_userpage(counter); + perf_counter_update_userpage(next_counter); +} + +#define list_next_entry(pos, member) \ + list_entry(pos->member.next, typeof(*pos), member) + +static void perf_counter_sync_stat(struct perf_counter_context *ctx, + struct perf_counter_context *next_ctx) +{ + struct perf_counter *counter, *next_counter; + + if (!ctx->nr_stat) + return; + + counter = list_first_entry(&ctx->event_list, + struct perf_counter, event_entry); + + next_counter = list_first_entry(&next_ctx->event_list, + struct perf_counter, event_entry); + + while (&counter->event_entry != &ctx->event_list && + &next_counter->event_entry != &next_ctx->event_list) { + + __perf_counter_sync_stat(counter, next_counter); + + counter = list_next_entry(counter, event_entry); + next_counter = list_next_entry(next_counter, event_entry); + } +} + /* * Called from scheduler to remove the counters of the current task, * with interrupts disabled. @@ -1057,6 +1193,8 @@ void perf_counter_task_sched_out(struct task_struct *task, ctx->task = next; next_ctx->task = task; do_switch = 0; + + perf_counter_sync_stat(ctx, next_ctx); } spin_unlock(&next_ctx->lock); spin_unlock(&ctx->lock); @@ -1203,7 +1341,6 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) #define MAX_INTERRUPTS (~0ULL) static void perf_log_throttle(struct perf_counter *counter, int enable); -static void perf_log_period(struct perf_counter *counter, u64 period); static void perf_adjust_period(struct perf_counter *counter, u64 events) { @@ -1222,8 +1359,6 @@ static void perf_adjust_period(struct perf_counter *counter, u64 events) if (!sample_period) sample_period = 1; - perf_log_period(counter, sample_period); - hwc->sample_period = sample_period; } @@ -1344,14 +1479,68 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) } /* + * Enable all of a task's counters that have been marked enable-on-exec. + * This expects task == current. + */ +static void perf_counter_enable_on_exec(struct task_struct *task) +{ + struct perf_counter_context *ctx; + struct perf_counter *counter; + unsigned long flags; + int enabled = 0; + + local_irq_save(flags); + ctx = task->perf_counter_ctxp; + if (!ctx || !ctx->nr_counters) + goto out; + + __perf_counter_task_sched_out(ctx); + + spin_lock(&ctx->lock); + + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + if (!counter->attr.enable_on_exec) + continue; + counter->attr.enable_on_exec = 0; + if (counter->state >= PERF_COUNTER_STATE_INACTIVE) + continue; + __perf_counter_mark_enabled(counter, ctx); + enabled = 1; + } + + /* + * Unclone this context if we enabled any counter. + */ + if (enabled) + unclone_ctx(ctx); + + spin_unlock(&ctx->lock); + + perf_counter_task_sched_in(task, smp_processor_id()); + out: + local_irq_restore(flags); +} + +/* * Cross CPU call to read the hardware counter */ -static void __read(void *info) +static void __perf_counter_read(void *info) { + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter *counter = info; struct perf_counter_context *ctx = counter->ctx; unsigned long flags; + /* + * If this is a task context, we need to check whether it is + * the current task context of this cpu. If not it has been + * scheduled out before the smp call arrived. In that case + * counter->count would have been updated to a recent sample + * when the counter was scheduled out. + */ + if (ctx->task && cpuctx->task_ctx != ctx) + return; + local_irq_save(flags); if (ctx->is_active) update_context_time(ctx); @@ -1368,7 +1557,7 @@ static u64 perf_counter_read(struct perf_counter *counter) */ if (counter->state == PERF_COUNTER_STATE_ACTIVE) { smp_call_function_single(counter->oncpu, - __read, counter, 1); + __perf_counter_read, counter, 1); } else if (counter->state == PERF_COUNTER_STATE_INACTIVE) { update_counter_times(counter); } @@ -1394,7 +1583,6 @@ __perf_counter_init_context(struct perf_counter_context *ctx, static struct perf_counter_context *find_get_context(pid_t pid, int cpu) { - struct perf_counter_context *parent_ctx; struct perf_counter_context *ctx; struct perf_cpu_context *cpuctx; struct task_struct *task; @@ -1454,16 +1642,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) retry: ctx = perf_lock_task_context(task, &flags); if (ctx) { - parent_ctx = ctx->parent_ctx; - if (parent_ctx) { - put_ctx(parent_ctx); - ctx->parent_ctx = NULL; /* no longer a clone */ - } - /* - * Get an extra reference before dropping the lock so that - * this context won't get freed if the task exits. - */ - get_ctx(ctx); + unclone_ctx(ctx); spin_unlock_irqrestore(&ctx->lock, flags); } @@ -1509,11 +1688,20 @@ static void free_counter(struct perf_counter *counter) { perf_pending_sync(counter); - atomic_dec(&nr_counters); - if (counter->attr.mmap) - atomic_dec(&nr_mmap_counters); - if (counter->attr.comm) - atomic_dec(&nr_comm_counters); + if (!counter->parent) { + atomic_dec(&nr_counters); + if (counter->attr.mmap) + atomic_dec(&nr_mmap_counters); + if (counter->attr.comm) + atomic_dec(&nr_comm_counters); + if (counter->attr.task) + atomic_dec(&nr_task_counters); + } + + if (counter->output) { + fput(counter->output->filp); + counter->output = NULL; + } if (counter->destroy) counter->destroy(counter); @@ -1547,14 +1735,133 @@ static int perf_release(struct inode *inode, struct file *file) return 0; } +static int perf_counter_read_size(struct perf_counter *counter) +{ + int entry = sizeof(u64); /* value */ + int size = 0; + int nr = 1; + + if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + size += sizeof(u64); + + if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + size += sizeof(u64); + + if (counter->attr.read_format & PERF_FORMAT_ID) + entry += sizeof(u64); + + if (counter->attr.read_format & PERF_FORMAT_GROUP) { + nr += counter->group_leader->nr_siblings; + size += sizeof(u64); + } + + size += entry * nr; + + return size; +} + +static u64 perf_counter_read_value(struct perf_counter *counter) +{ + struct perf_counter *child; + u64 total = 0; + + total += perf_counter_read(counter); + list_for_each_entry(child, &counter->child_list, child_list) + total += perf_counter_read(child); + + return total; +} + +static int perf_counter_read_entry(struct perf_counter *counter, + u64 read_format, char __user *buf) +{ + int n = 0, count = 0; + u64 values[2]; + + values[n++] = perf_counter_read_value(counter); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(counter); + + count = n * sizeof(u64); + + if (copy_to_user(buf, values, count)) + return -EFAULT; + + return count; +} + +static int perf_counter_read_group(struct perf_counter *counter, + u64 read_format, char __user *buf) +{ + struct perf_counter *leader = counter->group_leader, *sub; + int n = 0, size = 0, err = -EFAULT; + u64 values[3]; + + 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); + } + + size = n * sizeof(u64); + + if (copy_to_user(buf, values, size)) + return -EFAULT; + + err = perf_counter_read_entry(leader, read_format, buf + size); + if (err < 0) + return err; + + size += err; + + list_for_each_entry(sub, &leader->sibling_list, list_entry) { + err = perf_counter_read_entry(sub, read_format, + buf + size); + if (err < 0) + return err; + + size += err; + } + + return size; +} + +static int perf_counter_read_one(struct perf_counter *counter, + u64 read_format, char __user *buf) +{ + u64 values[4]; + int n = 0; + + values[n++] = perf_counter_read_value(counter); + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { + values[n++] = counter->total_time_enabled + + atomic64_read(&counter->child_total_time_enabled); + } + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { + values[n++] = counter->total_time_running + + atomic64_read(&counter->child_total_time_running); + } + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(counter); + + if (copy_to_user(buf, values, n * sizeof(u64))) + return -EFAULT; + + return n * sizeof(u64); +} + /* * Read the performance counter - simple non blocking version for now */ static ssize_t perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) { - u64 values[4]; - int n; + u64 read_format = counter->attr.read_format; + int ret; /* * Return end-of-file for a read on a counter that is in @@ -1564,28 +1871,18 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) if (counter->state == PERF_COUNTER_STATE_ERROR) return 0; + if (count < perf_counter_read_size(counter)) + return -ENOSPC; + WARN_ON_ONCE(counter->ctx->parent_ctx); mutex_lock(&counter->child_mutex); - values[0] = perf_counter_read(counter); - n = 1; - if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) - values[n++] = counter->total_time_enabled + - atomic64_read(&counter->child_total_time_enabled); - if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) - values[n++] = counter->total_time_running + - atomic64_read(&counter->child_total_time_running); - if (counter->attr.read_format & PERF_FORMAT_ID) - values[n++] = counter->id; + if (read_format & PERF_FORMAT_GROUP) + ret = perf_counter_read_group(counter, read_format, buf); + else + ret = perf_counter_read_one(counter, read_format, buf); mutex_unlock(&counter->child_mutex); - if (count < n * sizeof(u64)) - return -EINVAL; - count = n * sizeof(u64); - - if (copy_to_user(buf, values, count)) - return -EFAULT; - - return count; + return ret; } static ssize_t @@ -1682,8 +1979,6 @@ static int perf_counter_period(struct perf_counter *counter, u64 __user *arg) counter->attr.sample_freq = value; } else { - perf_log_period(counter, value); - counter->attr.sample_period = value; counter->hw.sample_period = value; } @@ -1693,6 +1988,8 @@ unlock: return ret; } +int perf_counter_set_output(struct perf_counter *counter, int output_fd); + static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) { struct perf_counter *counter = file->private_data; @@ -1716,6 +2013,9 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) case PERF_COUNTER_IOC_PERIOD: return perf_counter_period(counter, (u64 __user *)arg); + case PERF_COUNTER_IOC_SET_OUTPUT: + return perf_counter_set_output(counter, arg); + default: return -ENOTTY; } @@ -1752,6 +2052,18 @@ int perf_counter_task_disable(void) return 0; } +#ifndef PERF_COUNTER_INDEX_OFFSET +# define PERF_COUNTER_INDEX_OFFSET 0 +#endif + +static int perf_counter_index(struct perf_counter *counter) +{ + if (counter->state != PERF_COUNTER_STATE_ACTIVE) + return 0; + + return counter->hw.idx + 1 - PERF_COUNTER_INDEX_OFFSET; +} + /* * Callers need to ensure there can be no nesting of this function, otherwise * the seqlock logic goes bad. We can not serialize this because the arch @@ -1776,11 +2088,17 @@ void perf_counter_update_userpage(struct perf_counter *counter) preempt_disable(); ++userpg->lock; barrier(); - userpg->index = counter->hw.idx; + userpg->index = perf_counter_index(counter); userpg->offset = atomic64_read(&counter->count); if (counter->state == PERF_COUNTER_STATE_ACTIVE) userpg->offset -= atomic64_read(&counter->hw.prev_count); + userpg->time_enabled = counter->total_time_enabled + + atomic64_read(&counter->child_total_time_enabled); + + userpg->time_running = counter->total_time_running + + atomic64_read(&counter->child_total_time_running); + barrier(); ++userpg->lock; preempt_enable(); @@ -1877,7 +2195,7 @@ fail: static void perf_mmap_free_page(unsigned long addr) { - struct page *page = virt_to_page(addr); + struct page *page = virt_to_page((void *)addr); page->mapping = NULL; __free_page(page); @@ -1968,6 +2286,11 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) WARN_ON_ONCE(counter->ctx->parent_ctx); mutex_lock(&counter->mmap_mutex); + if (counter->output) { + ret = -EINVAL; + goto unlock; + } + if (atomic_inc_not_zero(&counter->mmap_count)) { if (nr_pages != counter->data->nr_pages) ret = -EINVAL; @@ -2077,7 +2400,7 @@ static void perf_pending_counter(struct perf_pending_entry *entry) if (counter->pending_disable) { counter->pending_disable = 0; - perf_counter_disable(counter); + __perf_counter_disable(counter); } if (counter->pending_wakeup) { @@ -2353,6 +2676,7 @@ static int perf_output_begin(struct perf_output_handle *handle, struct perf_counter *counter, unsigned int size, int nmi, int sample) { + struct perf_counter *output_counter; struct perf_mmap_data *data; unsigned int offset, head; int have_lost; @@ -2362,13 +2686,17 @@ static int perf_output_begin(struct perf_output_handle *handle, u64 lost; } lost_event; + rcu_read_lock(); /* * For inherited counters we send all the output towards the parent. */ if (counter->parent) counter = counter->parent; - rcu_read_lock(); + output_counter = rcu_dereference(counter->output); + if (output_counter) + counter = output_counter; + data = rcu_dereference(counter->data); if (!data) goto out; @@ -2462,46 +2790,113 @@ static u32 perf_counter_tid(struct perf_counter *counter, struct task_struct *p) return task_pid_nr_ns(p, counter->ns); } -static void perf_counter_output(struct perf_counter *counter, int nmi, - struct perf_sample_data *data) +static void perf_output_read_one(struct perf_output_handle *handle, + struct perf_counter *counter) { - int ret; - u64 sample_type = counter->attr.sample_type; - struct perf_output_handle handle; - struct perf_event_header header; - u64 ip; - struct { - u32 pid, tid; - } tid_entry; - struct { - u64 id; - u64 counter; - } group_entry; - struct perf_callchain_entry *callchain = NULL; - int callchain_size = 0; - u64 time; - struct { - u32 cpu, reserved; - } cpu_entry; - - header.type = 0; - header.size = sizeof(header); - - header.misc = PERF_EVENT_MISC_OVERFLOW; - header.misc |= perf_misc_flags(data->regs); + u64 read_format = counter->attr.read_format; + u64 values[4]; + int n = 0; - if (sample_type & PERF_SAMPLE_IP) { - ip = perf_instruction_pointer(data->regs); - header.type |= PERF_SAMPLE_IP; - header.size += sizeof(ip); + values[n++] = atomic64_read(&counter->count); + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { + values[n++] = counter->total_time_enabled + + atomic64_read(&counter->child_total_time_enabled); + } + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { + values[n++] = counter->total_time_running + + atomic64_read(&counter->child_total_time_running); } + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(counter); - if (sample_type & PERF_SAMPLE_TID) { - /* namespace issues */ + perf_output_copy(handle, values, n * sizeof(u64)); +} + +/* + * XXX PERF_FORMAT_GROUP vs inherited counters seems difficult. + */ +static void perf_output_read_group(struct perf_output_handle *handle, + struct perf_counter *counter) +{ + struct perf_counter *leader = counter->group_leader, *sub; + u64 read_format = counter->attr.read_format; + u64 values[5]; + int n = 0; + + values[n++] = 1 + leader->nr_siblings; + + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + values[n++] = leader->total_time_enabled; + + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + values[n++] = leader->total_time_running; + + if (leader != counter) + leader->pmu->read(leader); + + values[n++] = atomic64_read(&leader->count); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(leader); + + perf_output_copy(handle, values, n * sizeof(u64)); + + list_for_each_entry(sub, &leader->sibling_list, list_entry) { + n = 0; + + if (sub != counter) + sub->pmu->read(sub); + + values[n++] = atomic64_read(&sub->count); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_counter_id(sub); + + perf_output_copy(handle, values, n * sizeof(u64)); + } +} + +static void perf_output_read(struct perf_output_handle *handle, + struct perf_counter *counter) +{ + if (counter->attr.read_format & PERF_FORMAT_GROUP) + perf_output_read_group(handle, counter); + else + perf_output_read_one(handle, counter); +} + +void perf_counter_output(struct perf_counter *counter, int nmi, + struct perf_sample_data *data) +{ + int ret; + u64 sample_type = counter->attr.sample_type; + struct perf_output_handle handle; + struct perf_event_header header; + u64 ip; + struct { + u32 pid, tid; + } tid_entry; + struct perf_callchain_entry *callchain = NULL; + int callchain_size = 0; + u64 time; + struct { + u32 cpu, reserved; + } cpu_entry; + + header.type = PERF_EVENT_SAMPLE; + header.size = sizeof(header); + + header.misc = 0; + header.misc |= perf_misc_flags(data->regs); + + if (sample_type & PERF_SAMPLE_IP) { + ip = perf_instruction_pointer(data->regs); + header.size += sizeof(ip); + } + + if (sample_type & PERF_SAMPLE_TID) { + /* namespace issues */ tid_entry.pid = perf_counter_pid(counter, current); tid_entry.tid = perf_counter_tid(counter, current); - header.type |= PERF_SAMPLE_TID; header.size += sizeof(tid_entry); } @@ -2511,47 +2906,51 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, */ time = sched_clock(); - header.type |= PERF_SAMPLE_TIME; header.size += sizeof(u64); } - if (sample_type & PERF_SAMPLE_ADDR) { - header.type |= PERF_SAMPLE_ADDR; + if (sample_type & PERF_SAMPLE_ADDR) header.size += sizeof(u64); - } - if (sample_type & PERF_SAMPLE_ID) { - header.type |= PERF_SAMPLE_ID; + if (sample_type & PERF_SAMPLE_ID) + header.size += sizeof(u64); + + if (sample_type & PERF_SAMPLE_STREAM_ID) header.size += sizeof(u64); - } if (sample_type & PERF_SAMPLE_CPU) { - header.type |= PERF_SAMPLE_CPU; header.size += sizeof(cpu_entry); cpu_entry.cpu = raw_smp_processor_id(); + cpu_entry.reserved = 0; } - if (sample_type & PERF_SAMPLE_PERIOD) { - header.type |= PERF_SAMPLE_PERIOD; + if (sample_type & PERF_SAMPLE_PERIOD) header.size += sizeof(u64); - } - if (sample_type & PERF_SAMPLE_GROUP) { - header.type |= PERF_SAMPLE_GROUP; - header.size += sizeof(u64) + - counter->nr_siblings * sizeof(group_entry); - } + if (sample_type & PERF_SAMPLE_READ) + header.size += perf_counter_read_size(counter); if (sample_type & PERF_SAMPLE_CALLCHAIN) { callchain = perf_callchain(data->regs); if (callchain) { callchain_size = (1 + callchain->nr) * sizeof(u64); - - header.type |= PERF_SAMPLE_CALLCHAIN; header.size += callchain_size; - } + } else + header.size += sizeof(u64); + } + + if (sample_type & PERF_SAMPLE_RAW) { + int size = sizeof(u32); + + if (data->raw) + size += data->raw->size; + else + size += sizeof(u32); + + WARN_ON_ONCE(size & (sizeof(u64)-1)); + header.size += size; } ret = perf_output_begin(&handle, counter, header.size, nmi, 1); @@ -2572,7 +2971,13 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, if (sample_type & PERF_SAMPLE_ADDR) perf_output_put(&handle, data->addr); - if (sample_type & PERF_SAMPLE_ID) + if (sample_type & PERF_SAMPLE_ID) { + u64 id = primary_counter_id(counter); + + perf_output_put(&handle, id); + } + + if (sample_type & PERF_SAMPLE_STREAM_ID) perf_output_put(&handle, counter->id); if (sample_type & PERF_SAMPLE_CPU) @@ -2581,76 +2986,125 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, if (sample_type & PERF_SAMPLE_PERIOD) perf_output_put(&handle, data->period); - /* - * XXX PERF_SAMPLE_GROUP vs inherited counters seems difficult. - */ - if (sample_type & PERF_SAMPLE_GROUP) { - struct perf_counter *leader, *sub; - u64 nr = counter->nr_siblings; - - perf_output_put(&handle, nr); + if (sample_type & PERF_SAMPLE_READ) + perf_output_read(&handle, counter); - leader = counter->group_leader; - list_for_each_entry(sub, &leader->sibling_list, list_entry) { - if (sub != counter) - sub->pmu->read(sub); - - group_entry.id = sub->id; - group_entry.counter = atomic64_read(&sub->count); + if (sample_type & PERF_SAMPLE_CALLCHAIN) { + if (callchain) + perf_output_copy(&handle, callchain, callchain_size); + else { + u64 nr = 0; + perf_output_put(&handle, nr); + } + } - perf_output_put(&handle, group_entry); + if (sample_type & PERF_SAMPLE_RAW) { + if (data->raw) { + perf_output_put(&handle, data->raw->size); + perf_output_copy(&handle, data->raw->data, data->raw->size); + } else { + struct { + u32 size; + u32 data; + } raw = { + .size = sizeof(u32), + .data = 0, + }; + perf_output_put(&handle, raw); } } - if (callchain) - perf_output_copy(&handle, callchain, callchain_size); + perf_output_end(&handle); +} + +/* + * read event + */ + +struct perf_read_event { + struct perf_event_header header; + + u32 pid; + u32 tid; +}; + +static void +perf_counter_read_event(struct perf_counter *counter, + struct task_struct *task) +{ + struct perf_output_handle handle; + struct perf_read_event event = { + .header = { + .type = PERF_EVENT_READ, + .misc = 0, + .size = sizeof(event) + perf_counter_read_size(counter), + }, + .pid = perf_counter_pid(counter, task), + .tid = perf_counter_tid(counter, task), + }; + int ret; + + ret = perf_output_begin(&handle, counter, event.header.size, 0, 0); + if (ret) + return; + + perf_output_put(&handle, event); + perf_output_read(&handle, counter); perf_output_end(&handle); } /* - * fork tracking + * task tracking -- fork/exit + * + * enabled by: attr.comm | attr.mmap | attr.task */ -struct perf_fork_event { - struct task_struct *task; +struct perf_task_event { + struct task_struct *task; + struct perf_counter_context *task_ctx; struct { struct perf_event_header header; u32 pid; u32 ppid; + u32 tid; + u32 ptid; } event; }; -static void perf_counter_fork_output(struct perf_counter *counter, - struct perf_fork_event *fork_event) +static void perf_counter_task_output(struct perf_counter *counter, + struct perf_task_event *task_event) { struct perf_output_handle handle; - int size = fork_event->event.header.size; - struct task_struct *task = fork_event->task; + int size = task_event->event.header.size; + struct task_struct *task = task_event->task; int ret = perf_output_begin(&handle, counter, size, 0, 0); if (ret) return; - fork_event->event.pid = perf_counter_pid(counter, task); - fork_event->event.ppid = perf_counter_pid(counter, task->real_parent); + task_event->event.pid = perf_counter_pid(counter, task); + task_event->event.ppid = perf_counter_pid(counter, current); - perf_output_put(&handle, fork_event->event); + task_event->event.tid = perf_counter_tid(counter, task); + task_event->event.ptid = perf_counter_tid(counter, current); + + perf_output_put(&handle, task_event->event); perf_output_end(&handle); } -static int perf_counter_fork_match(struct perf_counter *counter) +static int perf_counter_task_match(struct perf_counter *counter) { - if (counter->attr.comm || counter->attr.mmap) + if (counter->attr.comm || counter->attr.mmap || counter->attr.task) return 1; return 0; } -static void perf_counter_fork_ctx(struct perf_counter_context *ctx, - struct perf_fork_event *fork_event) +static void perf_counter_task_ctx(struct perf_counter_context *ctx, + struct perf_task_event *task_event) { struct perf_counter *counter; @@ -2659,51 +3113,62 @@ static void perf_counter_fork_ctx(struct perf_counter_context *ctx, rcu_read_lock(); list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_counter_fork_match(counter)) - perf_counter_fork_output(counter, fork_event); + if (perf_counter_task_match(counter)) + perf_counter_task_output(counter, task_event); } rcu_read_unlock(); } -static void perf_counter_fork_event(struct perf_fork_event *fork_event) +static void perf_counter_task_event(struct perf_task_event *task_event) { struct perf_cpu_context *cpuctx; - struct perf_counter_context *ctx; + struct perf_counter_context *ctx = task_event->task_ctx; cpuctx = &get_cpu_var(perf_cpu_context); - perf_counter_fork_ctx(&cpuctx->ctx, fork_event); + perf_counter_task_ctx(&cpuctx->ctx, task_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_counter_ctxp); + if (!ctx) + ctx = rcu_dereference(task_event->task->perf_counter_ctxp); if (ctx) - perf_counter_fork_ctx(ctx, fork_event); + perf_counter_task_ctx(ctx, task_event); rcu_read_unlock(); } -void perf_counter_fork(struct task_struct *task) +static void perf_counter_task(struct task_struct *task, + struct perf_counter_context *task_ctx, + int new) { - struct perf_fork_event fork_event; + struct perf_task_event task_event; if (!atomic_read(&nr_comm_counters) && - !atomic_read(&nr_mmap_counters)) + !atomic_read(&nr_mmap_counters) && + !atomic_read(&nr_task_counters)) return; - fork_event = (struct perf_fork_event){ - .task = task, - .event = { + task_event = (struct perf_task_event){ + .task = task, + .task_ctx = task_ctx, + .event = { .header = { - .type = PERF_EVENT_FORK, - .size = sizeof(fork_event.event), + .type = new ? PERF_EVENT_FORK : PERF_EVENT_EXIT, + .misc = 0, + .size = sizeof(task_event.event), }, + /* .pid */ + /* .ppid */ + /* .tid */ + /* .ptid */ }, }; - perf_counter_fork_event(&fork_event); + perf_counter_task_event(&task_event); +} + +void perf_counter_fork(struct task_struct *task) +{ + perf_counter_task(task, NULL, 1); } /* @@ -2771,8 +3236,10 @@ static void perf_counter_comm_event(struct perf_comm_event *comm_event) struct perf_cpu_context *cpuctx; struct perf_counter_context *ctx; unsigned int size; - char *comm = comm_event->task->comm; + char comm[TASK_COMM_LEN]; + memset(comm, 0, sizeof(comm)); + strncpy(comm, comm_event->task->comm, sizeof(comm)); size = ALIGN(strlen(comm)+1, sizeof(u64)); comm_event->comm = comm; @@ -2799,13 +3266,24 @@ void perf_counter_comm(struct task_struct *task) { struct perf_comm_event comm_event; + if (task->perf_counter_ctxp) + perf_counter_enable_on_exec(task); + if (!atomic_read(&nr_comm_counters)) return; comm_event = (struct perf_comm_event){ .task = task, + /* .comm */ + /* .comm_size */ .event = { - .header = { .type = PERF_EVENT_COMM, }, + .header = { + .type = PERF_EVENT_COMM, + .misc = 0, + /* .size */ + }, + /* .pid */ + /* .tid */ }, }; @@ -2888,8 +3366,15 @@ static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) char *buf = NULL; const char *name; + memset(tmp, 0, sizeof(tmp)); + if (file) { - buf = kzalloc(PATH_MAX, GFP_KERNEL); + /* + * d_path works from the end of the buffer backwards, so we + * need to add enough zero bytes after the string to handle + * the 64bit alignment we do later. + */ + buf = kzalloc(PATH_MAX + sizeof(u64), GFP_KERNEL); if (!buf) { name = strncpy(tmp, "//enomem", sizeof(tmp)); goto got_name; @@ -2900,9 +3385,11 @@ static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) goto got_name; } } else { - name = arch_vma_name(mmap_event->vma); - if (name) + if (arch_vma_name(mmap_event->vma)) { + name = strncpy(tmp, arch_vma_name(mmap_event->vma), + sizeof(tmp)); goto got_name; + } if (!vma->vm_mm) { name = strncpy(tmp, "[vdso]", sizeof(tmp)); @@ -2947,8 +3434,16 @@ void __perf_counter_mmap(struct vm_area_struct *vma) mmap_event = (struct perf_mmap_event){ .vma = vma, + /* .file_name */ + /* .file_size */ .event = { - .header = { .type = PERF_EVENT_MMAP, }, + .header = { + .type = PERF_EVENT_MMAP, + .misc = 0, + /* .size */ + }, + /* .pid */ + /* .tid */ .start = vma->vm_start, .len = vma->vm_end - vma->vm_start, .pgoff = vma->vm_pgoff, @@ -2959,49 +3454,6 @@ void __perf_counter_mmap(struct vm_area_struct *vma) } /* - * Log sample_period changes so that analyzing tools can re-normalize the - * event flow. - */ - -struct freq_event { - struct perf_event_header header; - u64 time; - u64 id; - u64 period; -}; - -static void perf_log_period(struct perf_counter *counter, u64 period) -{ - struct perf_output_handle handle; - struct freq_event event; - int ret; - - if (counter->hw.sample_period == period) - return; - - if (counter->attr.sample_type & PERF_SAMPLE_PERIOD) - return; - - event = (struct freq_event) { - .header = { - .type = PERF_EVENT_PERIOD, - .misc = 0, - .size = sizeof(event), - }, - .time = sched_clock(), - .id = counter->id, - .period = period, - }; - - ret = perf_output_begin(&handle, counter, sizeof(event), 1, 0); - if (ret) - return; - - perf_output_put(&handle, event); - perf_output_end(&handle); -} - -/* * IRQ throttle logging */ @@ -3014,16 +3466,21 @@ static void perf_log_throttle(struct perf_counter *counter, int enable) struct perf_event_header header; u64 time; u64 id; + u64 stream_id; } throttle_event = { .header = { - .type = PERF_EVENT_THROTTLE + 1, + .type = PERF_EVENT_THROTTLE, .misc = 0, .size = sizeof(throttle_event), }, - .time = sched_clock(), - .id = counter->id, + .time = sched_clock(), + .id = primary_counter_id(counter), + .stream_id = counter->id, }; + if (enable) + throttle_event.header.type = PERF_EVENT_UNTHROTTLE; + ret = perf_output_begin(&handle, counter, sizeof(throttle_event), 1, 0); if (ret) return; @@ -3100,130 +3557,111 @@ int perf_counter_overflow(struct perf_counter *counter, int nmi, * Generic software counter infrastructure */ -static void perf_swcounter_update(struct perf_counter *counter) +/* + * We directly increment counter->count and keep a second value in + * counter->hw.period_left to count intervals. This period counter + * is kept in the range [-sample_period, 0] so that we can use the + * sign as trigger. + */ + +static u64 perf_swcounter_set_period(struct perf_counter *counter) { struct hw_perf_counter *hwc = &counter->hw; - u64 prev, now; - s64 delta; + u64 period = hwc->last_period; + u64 nr, offset; + s64 old, val; + + hwc->last_period = hwc->sample_period; again: - prev = atomic64_read(&hwc->prev_count); - now = atomic64_read(&hwc->count); - if (atomic64_cmpxchg(&hwc->prev_count, prev, now) != prev) - goto again; + old = val = atomic64_read(&hwc->period_left); + if (val < 0) + return 0; - delta = now - prev; + nr = div64_u64(period + val, period); + offset = nr * period; + val -= offset; + if (atomic64_cmpxchg(&hwc->period_left, old, val) != old) + goto again; - atomic64_add(delta, &counter->count); - atomic64_sub(delta, &hwc->period_left); + return nr; } -static void perf_swcounter_set_period(struct perf_counter *counter) +static void perf_swcounter_overflow(struct perf_counter *counter, + int nmi, struct perf_sample_data *data) { struct hw_perf_counter *hwc = &counter->hw; - s64 left = atomic64_read(&hwc->period_left); - s64 period = hwc->sample_period; + u64 overflow; - if (unlikely(left <= -period)) { - left = period; - atomic64_set(&hwc->period_left, left); - hwc->last_period = period; - } + data->period = counter->hw.last_period; + overflow = perf_swcounter_set_period(counter); - if (unlikely(left <= 0)) { - left += period; - atomic64_add(period, &hwc->period_left); - hwc->last_period = period; - } + if (hwc->interrupts == MAX_INTERRUPTS) + return; - atomic64_set(&hwc->prev_count, -left); - atomic64_set(&hwc->count, -left); + for (; overflow; overflow--) { + if (perf_counter_overflow(counter, nmi, data)) { + /* + * We inhibit the overflow from happening when + * hwc->interrupts == MAX_INTERRUPTS. + */ + break; + } + } } -static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) +static void perf_swcounter_unthrottle(struct perf_counter *counter) { - enum hrtimer_restart ret = HRTIMER_RESTART; - struct perf_sample_data data; - struct perf_counter *counter; - u64 period; - - counter = container_of(hrtimer, struct perf_counter, hw.hrtimer); - counter->pmu->read(counter); - - data.addr = 0; - data.regs = get_irq_regs(); /* - * In case we exclude kernel IPs or are somehow not in interrupt - * context, provide the next best thing, the user IP. + * Nothing to do, we already reset hwc->interrupts. */ - if ((counter->attr.exclude_kernel || !data.regs) && - !counter->attr.exclude_user) - data.regs = task_pt_regs(current); - - if (data.regs) { - if (perf_counter_overflow(counter, 0, &data)) - ret = HRTIMER_NORESTART; - } - - period = max_t(u64, 10000, counter->hw.sample_period); - hrtimer_forward_now(hrtimer, ns_to_ktime(period)); - - return ret; } -static void perf_swcounter_overflow(struct perf_counter *counter, - int nmi, struct pt_regs *regs, u64 addr) +static void perf_swcounter_add(struct perf_counter *counter, u64 nr, + int nmi, struct perf_sample_data *data) { - struct perf_sample_data data = { - .regs = regs, - .addr = addr, - .period = counter->hw.last_period, - }; + struct hw_perf_counter *hwc = &counter->hw; + + atomic64_add(nr, &counter->count); - perf_swcounter_update(counter); - perf_swcounter_set_period(counter); - if (perf_counter_overflow(counter, nmi, &data)) - /* soft-disable the counter */ - ; + if (!hwc->sample_period) + return; + + if (!data->regs) + return; + if (!atomic64_add_negative(nr, &hwc->period_left)) + perf_swcounter_overflow(counter, nmi, data); } static int perf_swcounter_is_counting(struct perf_counter *counter) { - struct perf_counter_context *ctx; - unsigned long flags; - int count; - + /* + * The counter is active, we're good! + */ if (counter->state == PERF_COUNTER_STATE_ACTIVE) return 1; + /* + * The counter is off/error, not counting. + */ if (counter->state != PERF_COUNTER_STATE_INACTIVE) return 0; /* - * If the counter is inactive, it could be just because - * its task is scheduled out, or because it's in a group - * which could not go on the PMU. We want to count in - * the first case but not the second. If the context is - * currently active then an inactive software counter must - * be the second case. If it's not currently active then - * we need to know whether the counter was active when the - * context was last active, which we can determine by - * comparing counter->tstamp_stopped with ctx->time. - * - * We are within an RCU read-side critical section, - * which protects the existence of *ctx. + * The counter is inactive, if the context is active + * we're part of a group that didn't make it on the 'pmu', + * not counting. */ - ctx = counter->ctx; - spin_lock_irqsave(&ctx->lock, flags); - count = 1; - /* Re-check state now we have the lock */ - if (counter->state < PERF_COUNTER_STATE_INACTIVE || - counter->ctx->is_active || - counter->tstamp_stopped < ctx->time) - count = 0; - spin_unlock_irqrestore(&ctx->lock, flags); - return count; + if (counter->ctx->is_active) + return 0; + + /* + * We're inactive and the context is too, this means the + * task is scheduled out, we're counting events that happen + * to us, like migration events. + */ + return 1; } static int perf_swcounter_match(struct perf_counter *counter, @@ -3249,19 +3687,10 @@ static int perf_swcounter_match(struct perf_counter *counter, return 1; } -static void perf_swcounter_add(struct perf_counter *counter, u64 nr, - int nmi, struct pt_regs *regs, u64 addr) -{ - int neg = atomic64_add_negative(nr, &counter->hw.count); - - if (counter->hw.sample_period && !neg && regs) - perf_swcounter_overflow(counter, nmi, regs, addr); -} - static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, - enum perf_type_id type, u32 event, - u64 nr, int nmi, struct pt_regs *regs, - u64 addr) + enum perf_type_id type, + u32 event, u64 nr, int nmi, + struct perf_sample_data *data) { struct perf_counter *counter; @@ -3270,8 +3699,8 @@ static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, rcu_read_lock(); list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_swcounter_match(counter, type, event, regs)) - perf_swcounter_add(counter, nr, nmi, regs, addr); + if (perf_swcounter_match(counter, type, event, data->regs)) + perf_swcounter_add(counter, nr, nmi, data); } rcu_read_unlock(); } @@ -3290,9 +3719,9 @@ static int *perf_swcounter_recursion_context(struct perf_cpu_context *cpuctx) return &cpuctx->recursion[0]; } -static void __perf_swcounter_event(enum perf_type_id type, u32 event, - u64 nr, int nmi, struct pt_regs *regs, - u64 addr) +static void do_perf_swcounter_event(enum perf_type_id type, u32 event, + u64 nr, int nmi, + struct perf_sample_data *data) { struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); int *recursion = perf_swcounter_recursion_context(cpuctx); @@ -3305,7 +3734,7 @@ static void __perf_swcounter_event(enum perf_type_id type, u32 event, barrier(); perf_swcounter_ctx_event(&cpuctx->ctx, type, event, - nr, nmi, regs, addr); + nr, nmi, data); rcu_read_lock(); /* * doesn't really matter which of the child contexts the @@ -3313,7 +3742,7 @@ static void __perf_swcounter_event(enum perf_type_id type, u32 event, */ ctx = rcu_dereference(current->perf_counter_ctxp); if (ctx) - perf_swcounter_ctx_event(ctx, type, event, nr, nmi, regs, addr); + perf_swcounter_ctx_event(ctx, type, event, nr, nmi, data); rcu_read_unlock(); barrier(); @@ -3323,35 +3752,79 @@ out: put_cpu_var(perf_cpu_context); } -void -perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr) +void __perf_swcounter_event(u32 event, u64 nr, int nmi, + struct pt_regs *regs, u64 addr) { - __perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, regs, addr); + struct perf_sample_data data = { + .regs = regs, + .addr = addr, + }; + + do_perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, &data); } static void perf_swcounter_read(struct perf_counter *counter) { - perf_swcounter_update(counter); } static int perf_swcounter_enable(struct perf_counter *counter) { - perf_swcounter_set_period(counter); + struct hw_perf_counter *hwc = &counter->hw; + + if (hwc->sample_period) { + hwc->last_period = hwc->sample_period; + perf_swcounter_set_period(counter); + } return 0; } static void perf_swcounter_disable(struct perf_counter *counter) { - perf_swcounter_update(counter); } static const struct pmu perf_ops_generic = { .enable = perf_swcounter_enable, .disable = perf_swcounter_disable, .read = perf_swcounter_read, + .unthrottle = perf_swcounter_unthrottle, }; /* + * hrtimer based swcounter callback + */ + +static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) +{ + enum hrtimer_restart ret = HRTIMER_RESTART; + struct perf_sample_data data; + struct perf_counter *counter; + u64 period; + + counter = container_of(hrtimer, struct perf_counter, hw.hrtimer); + counter->pmu->read(counter); + + data.addr = 0; + data.regs = get_irq_regs(); + /* + * In case we exclude kernel IPs or are somehow not in interrupt + * context, provide the next best thing, the user IP. + */ + if ((counter->attr.exclude_kernel || !data.regs) && + !counter->attr.exclude_user) + data.regs = task_pt_regs(current); + + if (data.regs) { + if (perf_counter_overflow(counter, 0, &data)) + ret = HRTIMER_NORESTART; + } + + period = max_t(u64, 10000, counter->hw.sample_period); + hrtimer_forward_now(hrtimer, ns_to_ktime(period)); + + return ret; +} + +/* * Software counter: cpu wall time clock */ @@ -3467,36 +3940,25 @@ static const struct pmu perf_ops_task_clock = { .read = task_clock_perf_counter_read, }; -/* - * Software counter: cpu migrations - */ -void perf_counter_task_migration(struct task_struct *task, int cpu) -{ - struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx; - - perf_swcounter_ctx_event(&cpuctx->ctx, PERF_TYPE_SOFTWARE, - PERF_COUNT_SW_CPU_MIGRATIONS, - 1, 1, NULL, 0); - - ctx = perf_pin_task_context(task); - if (ctx) { - perf_swcounter_ctx_event(ctx, PERF_TYPE_SOFTWARE, - PERF_COUNT_SW_CPU_MIGRATIONS, - 1, 1, NULL, 0); - perf_unpin_context(ctx); - } -} - #ifdef CONFIG_EVENT_PROFILE -void perf_tpcounter_event(int event_id) +void perf_tpcounter_event(int event_id, u64 addr, u64 count, void *record, + int entry_size) { - struct pt_regs *regs = get_irq_regs(); + struct perf_raw_record raw = { + .size = entry_size, + .data = record, + }; - if (!regs) - regs = task_pt_regs(current); + struct perf_sample_data data = { + .regs = get_irq_regs(), + .addr = addr, + .raw = &raw, + }; + + if (!data.regs) + data.regs = task_pt_regs(current); - __perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, 1, 1, regs, 0); + do_perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, &data); } EXPORT_SYMBOL_GPL(perf_tpcounter_event); @@ -3505,16 +3967,21 @@ extern void ftrace_profile_disable(int); static void tp_perf_counter_destroy(struct perf_counter *counter) { - ftrace_profile_disable(perf_event_id(&counter->attr)); + ftrace_profile_disable(counter->attr.config); } static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) { - int event_id = perf_event_id(&counter->attr); - int ret; + /* + * Raw tracepoint data is a severe data leak, only allow root to + * have these. + */ + if ((counter->attr.sample_type & PERF_SAMPLE_RAW) && + perf_paranoid_tracepoint_raw() && + !capable(CAP_SYS_ADMIN)) + return ERR_PTR(-EPERM); - ret = ftrace_profile_enable(event_id); - if (ret) + if (ftrace_profile_enable(counter->attr.config)) return NULL; counter->destroy = tp_perf_counter_destroy; @@ -3528,9 +3995,21 @@ static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) } #endif +atomic_t perf_swcounter_enabled[PERF_COUNT_SW_MAX]; + +static void sw_perf_counter_destroy(struct perf_counter *counter) +{ + u64 event = counter->attr.config; + + WARN_ON(counter->parent); + + atomic_dec(&perf_swcounter_enabled[event]); +} + static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) { const struct pmu *pmu = NULL; + u64 event = counter->attr.config; /* * Software counters (currently) can't in general distinguish @@ -3539,7 +4018,7 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) * to be kernel events, and page faults are never hypervisor * events. */ - switch (counter->attr.config) { + switch (event) { case PERF_COUNT_SW_CPU_CLOCK: pmu = &perf_ops_cpu_clock; @@ -3560,6 +4039,10 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_SW_PAGE_FAULTS_MAJ: case PERF_COUNT_SW_CONTEXT_SWITCHES: case PERF_COUNT_SW_CPU_MIGRATIONS: + if (!counter->parent) { + atomic_inc(&perf_swcounter_enabled[event]); + counter->destroy = sw_perf_counter_destroy; + } pmu = &perf_ops_generic; break; } @@ -3575,6 +4058,7 @@ perf_counter_alloc(struct perf_counter_attr *attr, int cpu, struct perf_counter_context *ctx, struct perf_counter *group_leader, + struct perf_counter *parent_counter, gfp_t gfpflags) { const struct pmu *pmu; @@ -3610,6 +4094,8 @@ perf_counter_alloc(struct perf_counter_attr *attr, counter->ctx = ctx; counter->oncpu = -1; + counter->parent = parent_counter; + counter->ns = get_pid_ns(current->nsproxy->pid_ns); counter->id = atomic64_inc_return(&perf_counter_id); @@ -3624,13 +4110,14 @@ perf_counter_alloc(struct perf_counter_attr *attr, hwc->sample_period = attr->sample_period; if (attr->freq && attr->sample_freq) hwc->sample_period = 1; + hwc->last_period = hwc->sample_period; atomic64_set(&hwc->period_left, hwc->sample_period); /* - * we currently do not support PERF_SAMPLE_GROUP on inherited counters + * we currently do not support PERF_FORMAT_GROUP on inherited counters */ - if (attr->inherit && (attr->sample_type & PERF_SAMPLE_GROUP)) + if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP)) goto done; switch (attr->type) { @@ -3667,11 +4154,15 @@ done: counter->pmu = pmu; - atomic_inc(&nr_counters); - if (counter->attr.mmap) - atomic_inc(&nr_mmap_counters); - if (counter->attr.comm) - atomic_inc(&nr_comm_counters); + if (!counter->parent) { + atomic_inc(&nr_counters); + if (counter->attr.mmap) + atomic_inc(&nr_mmap_counters); + if (counter->attr.comm) + atomic_inc(&nr_comm_counters); + if (counter->attr.task) + atomic_inc(&nr_task_counters); + } return counter; } @@ -3724,6 +4215,7 @@ static int perf_copy_attr(struct perf_counter_attr __user *uattr, if (val) goto err_size; } + size = sizeof(*attr); } ret = copy_from_user(attr, uattr, size); @@ -3755,6 +4247,57 @@ err_size: goto out; } +int perf_counter_set_output(struct perf_counter *counter, int output_fd) +{ + struct perf_counter *output_counter = NULL; + struct file *output_file = NULL; + struct perf_counter *old_output; + int fput_needed = 0; + int ret = -EINVAL; + + if (!output_fd) + goto set; + + output_file = fget_light(output_fd, &fput_needed); + if (!output_file) + return -EBADF; + + if (output_file->f_op != &perf_fops) + goto out; + + output_counter = output_file->private_data; + + /* Don't chain output fds */ + if (output_counter->output) + goto out; + + /* Don't set an output fd when we already have an output channel */ + if (counter->data) + goto out; + + atomic_long_inc(&output_file->f_count); + +set: + mutex_lock(&counter->mmap_mutex); + old_output = counter->output; + rcu_assign_pointer(counter->output, output_counter); + mutex_unlock(&counter->mmap_mutex); + + if (old_output) { + /* + * we need to make sure no existing perf_output_*() + * is still referencing this counter. + */ + synchronize_rcu(); + fput(old_output->filp); + } + + ret = 0; +out: + fput_light(output_file, fput_needed); + return ret; +} + /** * sys_perf_counter_open - open a performance counter, associate it to a task/cpu * @@ -3774,15 +4317,15 @@ SYSCALL_DEFINE5(perf_counter_open, struct file *group_file = NULL; int fput_needed = 0; int fput_needed2 = 0; - int ret; + int err; /* for future expandability... */ - if (flags) + if (flags & ~(PERF_FLAG_FD_NO_GROUP | PERF_FLAG_FD_OUTPUT)) return -EINVAL; - ret = perf_copy_attr(attr_uptr, &attr); - if (ret) - return ret; + err = perf_copy_attr(attr_uptr, &attr); + if (err) + return err; if (!attr.exclude_kernel) { if (perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN)) @@ -3805,8 +4348,8 @@ SYSCALL_DEFINE5(perf_counter_open, * Look up the group leader (we will attach this counter to it): */ group_leader = NULL; - if (group_fd != -1) { - ret = -EINVAL; + if (group_fd != -1 && !(flags & PERF_FLAG_FD_NO_GROUP)) { + err = -EINVAL; group_file = fget_light(group_fd, &fput_needed); if (!group_file) goto err_put_context; @@ -3834,19 +4377,25 @@ SYSCALL_DEFINE5(perf_counter_open, } counter = perf_counter_alloc(&attr, cpu, ctx, group_leader, - GFP_KERNEL); - ret = PTR_ERR(counter); + NULL, GFP_KERNEL); + err = PTR_ERR(counter); if (IS_ERR(counter)) goto err_put_context; - ret = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0); - if (ret < 0) + err = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0); + if (err < 0) goto err_free_put_context; - counter_file = fget_light(ret, &fput_needed2); + counter_file = fget_light(err, &fput_needed2); if (!counter_file) goto err_free_put_context; + if (flags & PERF_FLAG_FD_OUTPUT) { + err = perf_counter_set_output(counter, group_fd); + if (err) + goto err_fput_free_put_context; + } + counter->filp = counter_file; WARN_ON_ONCE(ctx->parent_ctx); mutex_lock(&ctx->mutex); @@ -3860,20 +4409,20 @@ SYSCALL_DEFINE5(perf_counter_open, list_add_tail(&counter->owner_entry, ¤t->perf_counter_list); mutex_unlock(¤t->perf_counter_mutex); +err_fput_free_put_context: fput_light(counter_file, fput_needed2); -out_fput: - fput_light(group_file, fput_needed); - - return ret; - err_free_put_context: - kfree(counter); + if (err < 0) + kfree(counter); err_put_context: - put_ctx(ctx); + if (err < 0) + put_ctx(ctx); + + fput_light(group_file, fput_needed); - goto out_fput; + return err; } /* @@ -3900,7 +4449,8 @@ inherit_counter(struct perf_counter *parent_counter, child_counter = perf_counter_alloc(&parent_counter->attr, parent_counter->cpu, child_ctx, - group_leader, GFP_KERNEL); + group_leader, parent_counter, + GFP_KERNEL); if (IS_ERR(child_counter)) return child_counter; get_ctx(child_ctx); @@ -3923,12 +4473,6 @@ inherit_counter(struct perf_counter *parent_counter, */ add_counter_to_ctx(child_counter, child_ctx); - child_counter->parent = parent_counter; - /* - * inherit into child's child as well: - */ - child_counter->attr.inherit = 1; - /* * Get a reference to the parent filp - we will fput it * when the child counter exits. This is safe to do because @@ -3972,10 +4516,14 @@ static int inherit_group(struct perf_counter *parent_counter, } static void sync_child_counter(struct perf_counter *child_counter, - struct perf_counter *parent_counter) + struct task_struct *child) { + struct perf_counter *parent_counter = child_counter->parent; u64 child_val; + if (child_counter->attr.inherit_stat) + perf_counter_read_event(child_counter, child); + child_val = atomic64_read(&child_counter->count); /* @@ -4004,7 +4552,8 @@ static void sync_child_counter(struct perf_counter *child_counter, static void __perf_counter_exit_task(struct perf_counter *child_counter, - struct perf_counter_context *child_ctx) + struct perf_counter_context *child_ctx, + struct task_struct *child) { struct perf_counter *parent_counter; @@ -4018,7 +4567,7 @@ __perf_counter_exit_task(struct perf_counter *child_counter, * counters need to be zapped - but otherwise linger. */ if (parent_counter) { - sync_child_counter(child_counter, parent_counter); + sync_child_counter(child_counter, child); free_counter(child_counter); } } @@ -4032,8 +4581,10 @@ void perf_counter_exit_task(struct task_struct *child) struct perf_counter_context *child_ctx; unsigned long flags; - if (likely(!child->perf_counter_ctxp)) + if (likely(!child->perf_counter_ctxp)) { + perf_counter_task(child, NULL, 0); return; + } local_irq_save(flags); /* @@ -4052,17 +4603,20 @@ void perf_counter_exit_task(struct task_struct *child) */ spin_lock(&child_ctx->lock); child->perf_counter_ctxp = NULL; - if (child_ctx->parent_ctx) { - /* - * This context is a clone; unclone it so it can't get - * swapped to another process while we're removing all - * the counters from it. - */ - put_ctx(child_ctx->parent_ctx); - child_ctx->parent_ctx = NULL; - } - spin_unlock(&child_ctx->lock); - local_irq_restore(flags); + /* + * If this context is a clone; unclone it so it can't get + * swapped to another process while we're removing all + * the counters from it. + */ + unclone_ctx(child_ctx); + spin_unlock_irqrestore(&child_ctx->lock, flags); + + /* + * Report the task dead after unscheduling the counters so that we + * won't get any samples after PERF_EVENT_EXIT. We can however still + * get a few PERF_EVENT_READ events. + */ + perf_counter_task(child, child_ctx, 0); /* * We can recurse on the same lock type through: @@ -4080,7 +4634,7 @@ void perf_counter_exit_task(struct task_struct *child) again: list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list, list_entry) - __perf_counter_exit_task(child_counter, child_ctx); + __perf_counter_exit_task(child_counter, child_ctx, child); /* * If the last counter was a group counter, it will have appended all @@ -4283,6 +4837,11 @@ perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu) perf_counter_init_cpu(cpu); break; + case CPU_ONLINE: + case CPU_ONLINE_FROZEN: + hw_perf_counter_setup_online(cpu); + break; + case CPU_DOWN_PREPARE: case CPU_DOWN_PREPARE_FROZEN: perf_counter_exit_cpu(cpu); @@ -4307,6 +4866,8 @@ void __init perf_counter_init(void) { perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, (void *)(long)smp_processor_id()); + perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE, + (void *)(long)smp_processor_id()); register_cpu_notifier(&perf_cpu_nb); }