From 71a851b4d2a815adcfac09c1adda7ef6811fde66 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 10 Jul 2009 09:06:56 +0200 Subject: perf_counter: Stop open coding unclone_ctx Instead of open coding the unclone context thingy, put it in a common function. Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 36 +++++++++++++++++------------------- 1 file changed, 17 insertions(+), 19 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d55a50da234..8bf997d86bf 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -146,6 +146,14 @@ 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; + } +} + /* * Get the perf_counter_context for a task and lock it. * This has to cope with with the fact that until it is locked, @@ -1463,10 +1471,8 @@ static void perf_counter_enable_on_exec(struct task_struct *task) /* * Unclone this context if we enabled any counter. */ - if (enabled && ctx->parent_ctx) { - put_ctx(ctx->parent_ctx); - ctx->parent_ctx = NULL; - } + if (enabled) + unclone_ctx(ctx); spin_unlock(&ctx->lock); @@ -1526,7 +1532,6 @@ __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; @@ -1586,11 +1591,7 @@ 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 */ - } + unclone_ctx(ctx); spin_unlock_irqrestore(&ctx->lock, flags); } @@ -4255,15 +4256,12 @@ 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; - } + /* + * 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(&child_ctx->lock); local_irq_restore(flags); -- cgit v1.2.3 From d4d7d0b9545721d3cabb19d15163bbc66b797707 Mon Sep 17 00:00:00 2001 From: Chris Wilson Date: Mon, 6 Jul 2009 09:31:33 +0100 Subject: perf_counter: Fix the tracepoint channel to perfcounters Fix a missed rename in EVENT_PROFILE support so that it gets built and allows tracepoint tracing from the 'perf' tool. Fix a typo in the (never before built & enabled) portion in perf_counter.c as well, and update that code to the attr.config changes as well. Signed-off-by: Chris Wilson Cc: Ben Gamari Cc: Jason Baron Cc: Frederic Weisbecker Cc: Mike Galbraith Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt LKML-Reference: <1246869094-21237-1-git-send-email-chris@chris-wilson.co.uk> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d55a50da234..c6c38fb7766 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3671,7 +3671,7 @@ static const struct pmu perf_ops_task_clock = { void perf_tpcounter_event(int event_id) { struct perf_sample_data data = { - .regs = get_irq_regs(); + .regs = get_irq_regs(), .addr = 0, }; @@ -3687,16 +3687,12 @@ 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; - - ret = ftrace_profile_enable(event_id); - if (ret) + if (ftrace_profile_enable(counter->attr.config)) return NULL; counter->destroy = tp_perf_counter_destroy; -- cgit v1.2.3 From 413ee3b48ab582ffea33e7e140c7a2c5ea657e9a Mon Sep 17 00:00:00 2001 From: Anton Blanchard Date: Thu, 16 Jul 2009 15:15:52 +0200 Subject: perf_counter: Make sure we dont leak kernel memory to userspace There are a few places we are leaking tiny amounts of kernel memory to userspace. This happens when writing out strings because we always align the end to 64 bits. To avoid this we should always use an appropriately sized temporary buffer and ensure it is zeroed. Since d_path assembles the string from the end of the buffer backwards, we need to add 64 bits after the buffer to allow for alignment. We also need to copy arch_vma_name to the temporary buffer, because if we use it directly we may end up copying to userspace a number of bytes after the end of the string constant. Signed-off-by: Anton Blanchard Signed-off-by: Peter Zijlstra LKML-Reference: <20090716104817.273972048@samba.org> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c6c38fb7766..f7a8ab9576e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2968,8 +2968,10 @@ 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; @@ -3088,8 +3090,15 @@ 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; @@ -3100,9 +3109,11 @@ 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)); -- cgit v1.2.3 From ed900c054b541254f0ce5cedaf75206e29bd614e Mon Sep 17 00:00:00 2001 From: Anton Blanchard Date: Thu, 16 Jul 2009 15:44:29 +0200 Subject: perf_counter: Log vfork as a fork event Right now we don't output vfork events. Even though we should always see an exec after a vfork, we may get perfcounter samples between the vfork and exec. These samples can lead to some confusion when parsing perfcounter data. To keep things consistent we should always log a fork event. It will result in a little more log data, but is less confusing to trace parsing tools. Signed-off-by: Anton Blanchard Signed-off-by: Peter Zijlstra LKML-Reference: <20090716104817.589309391@samba.org> Signed-off-by: Ingo Molnar --- kernel/fork.c | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index 467746b3f0a..4812d60b29f 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1408,14 +1408,11 @@ long do_fork(unsigned long clone_flags, if (clone_flags & CLONE_VFORK) { p->vfork_done = &vfork; init_completion(&vfork); - } else if (!(clone_flags & CLONE_VM)) { - /* - * vfork will do an exec which will call - * set_task_comm() - */ - perf_counter_fork(p); } + if (!(clone_flags & CLONE_THREAD)) + perf_counter_fork(p); + audit_finish_fork(p); tracehook_report_clone(regs, clone_flags, nr, p); -- cgit v1.2.3 From c9f73a3dd27e03411f18a58c0814d51392d2b17a Mon Sep 17 00:00:00 2001 From: Arjan van de Ven Date: Tue, 21 Jul 2009 00:55:05 -0700 Subject: perf: Fix stack data leak the "reserved" field was not initialized to zero, resulting in 4 bytes of stack data leaking to userspace.... Signed-off-by: Arjan van de Ven Signed-off-by: Peter Zijlstra --- kernel/perf_counter.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 5c6fae4f43d..ff854fd89a8 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2666,6 +2666,7 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, header.size += sizeof(cpu_entry); cpu_entry.cpu = raw_smp_processor_id(); + cpu_entry.reserved = 0; } if (sample_type & PERF_SAMPLE_PERIOD) -- cgit v1.2.3 From 573402db02746179b3f95f83a11a787501f52d0a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 22 Jul 2009 11:13:50 +0200 Subject: perf_counter: Plug more stack leaks Per example of Arjan's patch, I went through and found a few more. Signed-off-by: Peter Zijlstra --- kernel/perf_counter.c | 23 +++++++++++++++++++++-- 1 file changed, 21 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ff854fd89a8..e1d6a3aa133 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2897,8 +2897,11 @@ void perf_counter_fork(struct task_struct *task) .event = { .header = { .type = PERF_EVENT_FORK, + .misc = 0, .size = sizeof(fork_event.event), }, + /* .pid */ + /* .ppid */ }, }; @@ -3008,8 +3011,16 @@ void perf_counter_comm(struct task_struct *task) 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 */ }, }; @@ -3160,8 +3171,16 @@ 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, -- cgit v1.2.3 From 7f453c24b95a085fc7bd35d53b33abc4dc5a048b Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 21 Jul 2009 13:19:40 +0200 Subject: perf_counter: PERF_SAMPLE_ID and inherited counters Anton noted that for inherited counters the counter-id as provided by PERF_SAMPLE_ID isn't mappable to the id found through PERF_RECORD_ID because each inherited counter gets its own id. His suggestion was to always return the parent counter id, since that is the primary counter id as exposed. However, these inherited counters have a unique identifier so that events like PERF_EVENT_PERIOD and PERF_EVENT_THROTTLE can be specific about which counter gets modified, which is important when trying to normalize the sample streams. This patch removes PERF_EVENT_PERIOD in favour of PERF_SAMPLE_PERIOD, which is more useful anyway, since changing periods became a lot more common than initially thought -- rendering PERF_EVENT_PERIOD the less useful solution (also, PERF_SAMPLE_PERIOD reports the more accurate value, since it reports the value used to trigger the overflow, whereas PERF_EVENT_PERIOD simply reports the requested period changed, which might only take effect on the next cycle). This still leaves us PERF_EVENT_THROTTLE to consider, but since that _should_ be a rare occurrence, and linking it to a primary id is the most useful bit to diagnose the problem, we introduce a PERF_SAMPLE_STREAM_ID, for those few cases where the full reconstruction is important. [Does change the ABI a little, but I see no other way out] Suggested-by: Anton Blanchard Signed-off-by: Peter Zijlstra LKML-Reference: <1248095846.15751.8781.camel@twins> --- kernel/perf_counter.c | 92 +++++++++++++++++---------------------------------- 1 file changed, 31 insertions(+), 61 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index e1d6a3aa133..7530588fa5c 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -154,6 +154,20 @@ static void unclone_ctx(struct perf_counter_context *ctx) } } +/* + * 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, @@ -1296,7 +1310,6 @@ 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) { @@ -1315,8 +1328,6 @@ 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; } @@ -1705,7 +1716,7 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) 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; + values[n++] = primary_counter_id(counter); mutex_unlock(&counter->child_mutex); if (count < n * sizeof(u64)) @@ -1812,8 +1823,6 @@ 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; } @@ -2662,6 +2671,9 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, 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.size += sizeof(cpu_entry); @@ -2705,7 +2717,13 @@ 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) @@ -2728,7 +2746,7 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, if (sub != counter) sub->pmu->read(sub); - group_entry.id = sub->id; + group_entry.id = primary_counter_id(sub); group_entry.counter = atomic64_read(&sub->count); perf_output_put(&handle, group_entry); @@ -2788,15 +2806,8 @@ perf_counter_read_event(struct perf_counter *counter, } if (counter->attr.read_format & PERF_FORMAT_ID) { - u64 id; - event.header.size += sizeof(u64); - if (counter->parent) - id = counter->parent->id; - else - id = counter->id; - - event.format[i++] = id; + event.format[i++] = primary_counter_id(counter); } ret = perf_output_begin(&handle, counter, event.header.size, 0, 0); @@ -3190,49 +3201,6 @@ void __perf_counter_mmap(struct vm_area_struct *vma) perf_counter_mmap_event(&mmap_event); } -/* - * 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 */ @@ -3246,14 +3214,16 @@ 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, .misc = 0, .size = sizeof(throttle_event), }, - .time = sched_clock(), - .id = counter->id, + .time = sched_clock(), + .id = primary_counter_id(counter), + .stream_id = counter->id, }; ret = perf_output_begin(&handle, counter, sizeof(throttle_event), 1, 0); -- cgit v1.2.3 From 966ee4d6b887c14159043ac80b8c3661d2bbe5e2 Mon Sep 17 00:00:00 2001 From: Anton Blanchard Date: Wed, 22 Jul 2009 23:05:46 +1000 Subject: perf_counter: Fix throttle/unthrottle event logging Right now we only print PERF_EVENT_THROTTLE + 1 (ie PERF_EVENT_UNTHROTTLE). Fix this to print both a throttle and unthrottle event. Signed-off-by: Anton Blanchard Signed-off-by: Peter Zijlstra LKML-Reference: <20090722130546.GE9029@kryten> --- kernel/perf_counter.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 7530588fa5c..787d4daef18 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3217,7 +3217,7 @@ static void perf_log_throttle(struct perf_counter *counter, int enable) u64 stream_id; } throttle_event = { .header = { - .type = PERF_EVENT_THROTTLE + 1, + .type = PERF_EVENT_THROTTLE, .misc = 0, .size = sizeof(throttle_event), }, @@ -3226,6 +3226,9 @@ static void perf_log_throttle(struct perf_counter *counter, int enable) .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; -- cgit v1.2.3