[GIT PULL] perf fixes

From: Ingo Molnar
Date: Thu May 22 2014 - 04:04:38 EST


Linus,

Please pull the latest perf-urgent-for-linus git tree from:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git perf-urgent-for-linus

# HEAD: b69cf53640da2b86439596118cfa95233154ee76 perf: Fix a race between ring_buffer_detach() and ring_buffer_attach()

The biggest changes are fixes for races that kept triggering Trinity
crashes, plus liblockdep build fixes and smaller misc fixes.

... liblockdep bits in perf/urgent are a pull mistake - they should
have been in locking/urgent - but by the time I noticed other commits
were added and testing was done :-/ Sorry about that.

Thanks,

Ingo

------------------>
Jiri Olsa (1):
perf: Prevent false warning in perf_swevent_add

Peter Zijlstra (4):
perf: Fix race in removing an event
perf: Fix perf_event_init_context()
perf: Limit perf_event_attr::sample_period to 63 bits
perf: Fix a race between ring_buffer_detach() and ring_buffer_attach()

S. Lockwood-Childs (1):
tools/liblockdep: Build liblockdep from tools/Makefile

Sasha Levin (1):
tools/liblockdep: Remove all build files when doing make clean

Yan, Zheng (1):
perf/x86/intel: Fix Silvermont's event constraints


arch/x86/kernel/cpu/perf_event_intel.c | 1 -
include/linux/perf_event.h | 2 +
kernel/events/core.c | 174 +++++++++++++++++----------------
tools/Makefile | 6 ++
tools/lib/lockdep/Makefile | 5 +-
5 files changed, 102 insertions(+), 86 deletions(-)

diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
index aa333d9..adb02aa 100644
--- a/arch/x86/kernel/cpu/perf_event_intel.c
+++ b/arch/x86/kernel/cpu/perf_event_intel.c
@@ -169,7 +169,6 @@ static struct event_constraint intel_slm_event_constraints[] __read_mostly =
{
FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */
- FIXED_EVENT_CONSTRAINT(0x013c, 2), /* CPU_CLK_UNHALTED.REF */
FIXED_EVENT_CONSTRAINT(0x0300, 2), /* pseudo CPU_CLK_UNHALTED.REF */
EVENT_CONSTRAINT_END
};
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 3356abc..3ef6ea1 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -402,6 +402,8 @@ struct perf_event {

struct ring_buffer *rb;
struct list_head rb_entry;
+ unsigned long rcu_batches;
+ int rcu_pending;

/* poll related */
wait_queue_head_t waitq;
diff --git a/kernel/events/core.c b/kernel/events/core.c
index f83a71a..440eefc 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -1443,6 +1443,11 @@ group_sched_out(struct perf_event *group_event,
cpuctx->exclusive = 0;
}

+struct remove_event {
+ struct perf_event *event;
+ bool detach_group;
+};
+
/*
* Cross CPU call to remove a performance event
*
@@ -1451,12 +1456,15 @@ group_sched_out(struct perf_event *group_event,
*/
static int __perf_remove_from_context(void *info)
{
- struct perf_event *event = info;
+ struct remove_event *re = info;
+ struct perf_event *event = re->event;
struct perf_event_context *ctx = event->ctx;
struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);

raw_spin_lock(&ctx->lock);
event_sched_out(event, cpuctx, ctx);
+ if (re->detach_group)
+ perf_group_detach(event);
list_del_event(event, ctx);
if (!ctx->nr_events && cpuctx->task_ctx == ctx) {
ctx->is_active = 0;
@@ -1481,10 +1489,14 @@ static int __perf_remove_from_context(void *info)
* When called from perf_event_exit_task, it's OK because the
* context has been detached from its task.
*/
-static void perf_remove_from_context(struct perf_event *event)
+static void perf_remove_from_context(struct perf_event *event, bool detach_group)
{
struct perf_event_context *ctx = event->ctx;
struct task_struct *task = ctx->task;
+ struct remove_event re = {
+ .event = event,
+ .detach_group = detach_group,
+ };

lockdep_assert_held(&ctx->mutex);

@@ -1493,12 +1505,12 @@ static void perf_remove_from_context(struct perf_event *event)
* Per cpu events are removed via an smp call and
* the removal is always successful.
*/
- cpu_function_call(event->cpu, __perf_remove_from_context, event);
+ cpu_function_call(event->cpu, __perf_remove_from_context, &re);
return;
}

retry:
- if (!task_function_call(task, __perf_remove_from_context, event))
+ if (!task_function_call(task, __perf_remove_from_context, &re))
return;

raw_spin_lock_irq(&ctx->lock);
@@ -1515,6 +1527,8 @@ retry:
* Since the task isn't running, its safe to remove the event, us
* holding the ctx->lock ensures the task won't get scheduled in.
*/
+ if (detach_group)
+ perf_group_detach(event);
list_del_event(event, ctx);
raw_spin_unlock_irq(&ctx->lock);
}
@@ -3178,7 +3192,8 @@ static void free_event_rcu(struct rcu_head *head)
}

static void ring_buffer_put(struct ring_buffer *rb);
-static void ring_buffer_detach(struct perf_event *event, struct ring_buffer *rb);
+static void ring_buffer_attach(struct perf_event *event,
+ struct ring_buffer *rb);

static void unaccount_event_cpu(struct perf_event *event, int cpu)
{
@@ -3238,8 +3253,6 @@ static void free_event(struct perf_event *event)
unaccount_event(event);

if (event->rb) {
- struct ring_buffer *rb;
-
/*
* Can happen when we close an event with re-directed output.
*
@@ -3247,12 +3260,7 @@ static void free_event(struct perf_event *event)
* over us; possibly making our ring_buffer_put() the last.
*/
mutex_lock(&event->mmap_mutex);
- rb = event->rb;
- if (rb) {
- rcu_assign_pointer(event->rb, NULL);
- ring_buffer_detach(event, rb);
- ring_buffer_put(rb); /* could be last */
- }
+ ring_buffer_attach(event, NULL);
mutex_unlock(&event->mmap_mutex);
}

@@ -3281,10 +3289,7 @@ int perf_event_release_kernel(struct perf_event *event)
* to trigger the AB-BA case.
*/
mutex_lock_nested(&ctx->mutex, SINGLE_DEPTH_NESTING);
- raw_spin_lock_irq(&ctx->lock);
- perf_group_detach(event);
- raw_spin_unlock_irq(&ctx->lock);
- perf_remove_from_context(event);
+ perf_remove_from_context(event, true);
mutex_unlock(&ctx->mutex);

free_event(event);
@@ -3839,28 +3844,47 @@ unlock:
static void ring_buffer_attach(struct perf_event *event,
struct ring_buffer *rb)
{
+ struct ring_buffer *old_rb = NULL;
unsigned long flags;

- if (!list_empty(&event->rb_entry))
- return;
+ if (event->rb) {
+ /*
+ * Should be impossible, we set this when removing
+ * event->rb_entry and wait/clear when adding event->rb_entry.
+ */
+ WARN_ON_ONCE(event->rcu_pending);

- spin_lock_irqsave(&rb->event_lock, flags);
- if (list_empty(&event->rb_entry))
- list_add(&event->rb_entry, &rb->event_list);
- spin_unlock_irqrestore(&rb->event_lock, flags);
-}
+ old_rb = event->rb;
+ event->rcu_batches = get_state_synchronize_rcu();
+ event->rcu_pending = 1;

-static void ring_buffer_detach(struct perf_event *event, struct ring_buffer *rb)
-{
- unsigned long flags;
+ spin_lock_irqsave(&old_rb->event_lock, flags);
+ list_del_rcu(&event->rb_entry);
+ spin_unlock_irqrestore(&old_rb->event_lock, flags);
+ }

- if (list_empty(&event->rb_entry))
- return;
+ if (event->rcu_pending && rb) {
+ cond_synchronize_rcu(event->rcu_batches);
+ event->rcu_pending = 0;
+ }
+
+ if (rb) {
+ spin_lock_irqsave(&rb->event_lock, flags);
+ list_add_rcu(&event->rb_entry, &rb->event_list);
+ spin_unlock_irqrestore(&rb->event_lock, flags);
+ }
+
+ rcu_assign_pointer(event->rb, rb);

- spin_lock_irqsave(&rb->event_lock, flags);
- list_del_init(&event->rb_entry);
- wake_up_all(&event->waitq);
- spin_unlock_irqrestore(&rb->event_lock, flags);
+ if (old_rb) {
+ ring_buffer_put(old_rb);
+ /*
+ * Since we detached before setting the new rb, so that we
+ * could attach the new rb, we could have missed a wakeup.
+ * Provide it now.
+ */
+ wake_up_all(&event->waitq);
+ }
}

static void ring_buffer_wakeup(struct perf_event *event)
@@ -3929,7 +3953,7 @@ static void perf_mmap_close(struct vm_area_struct *vma)
{
struct perf_event *event = vma->vm_file->private_data;

- struct ring_buffer *rb = event->rb;
+ struct ring_buffer *rb = ring_buffer_get(event);
struct user_struct *mmap_user = rb->mmap_user;
int mmap_locked = rb->mmap_locked;
unsigned long size = perf_data_size(rb);
@@ -3937,18 +3961,14 @@ static void perf_mmap_close(struct vm_area_struct *vma)
atomic_dec(&rb->mmap_count);

if (!atomic_dec_and_mutex_lock(&event->mmap_count, &event->mmap_mutex))
- return;
+ goto out_put;

- /* Detach current event from the buffer. */
- rcu_assign_pointer(event->rb, NULL);
- ring_buffer_detach(event, rb);
+ ring_buffer_attach(event, NULL);
mutex_unlock(&event->mmap_mutex);

/* If there's still other mmap()s of this buffer, we're done. */
- if (atomic_read(&rb->mmap_count)) {
- ring_buffer_put(rb); /* can't be last */
- return;
- }
+ if (atomic_read(&rb->mmap_count))
+ goto out_put;

/*
* No other mmap()s, detach from all other events that might redirect
@@ -3978,11 +3998,9 @@ again:
* still restart the iteration to make sure we're not now
* iterating the wrong list.
*/
- if (event->rb == rb) {
- rcu_assign_pointer(event->rb, NULL);
- ring_buffer_detach(event, rb);
- ring_buffer_put(rb); /* can't be last, we still have one */
- }
+ if (event->rb == rb)
+ ring_buffer_attach(event, NULL);
+
mutex_unlock(&event->mmap_mutex);
put_event(event);

@@ -4007,6 +4025,7 @@ again:
vma->vm_mm->pinned_vm -= mmap_locked;
free_uid(mmap_user);

+out_put:
ring_buffer_put(rb); /* could be last */
}

@@ -4124,7 +4143,6 @@ again:
vma->vm_mm->pinned_vm += extra;

ring_buffer_attach(event, rb);
- rcu_assign_pointer(event->rb, rb);

perf_event_init_userpage(event);
perf_event_update_userpage(event);
@@ -5408,6 +5426,9 @@ struct swevent_htable {

/* Recursion avoidance in each contexts */
int recursion[PERF_NR_CONTEXTS];
+
+ /* Keeps track of cpu being initialized/exited */
+ bool online;
};

static DEFINE_PER_CPU(struct swevent_htable, swevent_htable);
@@ -5654,8 +5675,14 @@ static int perf_swevent_add(struct perf_event *event, int flags)
hwc->state = !(flags & PERF_EF_START);

head = find_swevent_head(swhash, event);
- if (WARN_ON_ONCE(!head))
+ if (!head) {
+ /*
+ * We can race with cpu hotplug code. Do not
+ * WARN if the cpu just got unplugged.
+ */
+ WARN_ON_ONCE(swhash->online);
return -EINVAL;
+ }

hlist_add_head_rcu(&event->hlist_entry, head);

@@ -6914,7 +6941,7 @@ err_size:
static int
perf_event_set_output(struct perf_event *event, struct perf_event *output_event)
{
- struct ring_buffer *rb = NULL, *old_rb = NULL;
+ struct ring_buffer *rb = NULL;
int ret = -EINVAL;

if (!output_event)
@@ -6942,8 +6969,6 @@ set:
if (atomic_read(&event->mmap_count))
goto unlock;

- old_rb = event->rb;
-
if (output_event) {
/* get the rb we want to redirect to */
rb = ring_buffer_get(output_event);
@@ -6951,23 +6976,7 @@ set:
goto unlock;
}

- if (old_rb)
- ring_buffer_detach(event, old_rb);
-
- if (rb)
- ring_buffer_attach(event, rb);
-
- rcu_assign_pointer(event->rb, rb);
-
- if (old_rb) {
- ring_buffer_put(old_rb);
- /*
- * Since we detached before setting the new rb, so that we
- * could attach the new rb, we could have missed a wakeup.
- * Provide it now.
- */
- wake_up_all(&event->waitq);
- }
+ ring_buffer_attach(event, rb);

ret = 0;
unlock:
@@ -7018,6 +7027,9 @@ SYSCALL_DEFINE5(perf_event_open,
if (attr.freq) {
if (attr.sample_freq > sysctl_perf_event_sample_rate)
return -EINVAL;
+ } else {
+ if (attr.sample_period & (1ULL << 63))
+ return -EINVAL;
}

/*
@@ -7165,7 +7177,7 @@ SYSCALL_DEFINE5(perf_event_open,
struct perf_event_context *gctx = group_leader->ctx;

mutex_lock(&gctx->mutex);
- perf_remove_from_context(group_leader);
+ perf_remove_from_context(group_leader, false);

/*
* Removing from the context ends up with disabled
@@ -7175,7 +7187,7 @@ SYSCALL_DEFINE5(perf_event_open,
perf_event__state_init(group_leader);
list_for_each_entry(sibling, &group_leader->sibling_list,
group_entry) {
- perf_remove_from_context(sibling);
+ perf_remove_from_context(sibling, false);
perf_event__state_init(sibling);
put_ctx(gctx);
}
@@ -7305,7 +7317,7 @@ void perf_pmu_migrate_context(struct pmu *pmu, int src_cpu, int dst_cpu)
mutex_lock(&src_ctx->mutex);
list_for_each_entry_safe(event, tmp, &src_ctx->event_list,
event_entry) {
- perf_remove_from_context(event);
+ perf_remove_from_context(event, false);
unaccount_event_cpu(event, src_cpu);
put_ctx(src_ctx);
list_add(&event->migrate_entry, &events);
@@ -7367,13 +7379,7 @@ __perf_event_exit_task(struct perf_event *child_event,
struct perf_event_context *child_ctx,
struct task_struct *child)
{
- if (child_event->parent) {
- raw_spin_lock_irq(&child_ctx->lock);
- perf_group_detach(child_event);
- raw_spin_unlock_irq(&child_ctx->lock);
- }
-
- perf_remove_from_context(child_event);
+ perf_remove_from_context(child_event, !!child_event->parent);

/*
* It can happen that the parent exits first, and has events
@@ -7724,6 +7730,8 @@ int perf_event_init_context(struct task_struct *child, int ctxn)
* swapped under us.
*/
parent_ctx = perf_pin_task_context(parent, ctxn);
+ if (!parent_ctx)
+ return 0;

/*
* No need to check if parent_ctx != NULL here; since we saw
@@ -7835,6 +7843,7 @@ static void perf_event_init_cpu(int cpu)
struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);

mutex_lock(&swhash->hlist_mutex);
+ swhash->online = true;
if (swhash->hlist_refcount > 0) {
struct swevent_hlist *hlist;

@@ -7857,14 +7866,14 @@ static void perf_pmu_rotate_stop(struct pmu *pmu)

static void __perf_event_exit_context(void *__info)
{
+ struct remove_event re = { .detach_group = false };
struct perf_event_context *ctx = __info;
- struct perf_event *event;

perf_pmu_rotate_stop(ctx->pmu);

rcu_read_lock();
- list_for_each_entry_rcu(event, &ctx->event_list, event_entry)
- __perf_remove_from_context(event);
+ list_for_each_entry_rcu(re.event, &ctx->event_list, event_entry)
+ __perf_remove_from_context(&re);
rcu_read_unlock();
}

@@ -7892,6 +7901,7 @@ static void perf_event_exit_cpu(int cpu)
perf_event_exit_cpu_context(cpu);

mutex_lock(&swhash->hlist_mutex);
+ swhash->online = false;
swevent_hlist_release(swhash);
mutex_unlock(&swhash->hlist_mutex);
}
diff --git a/tools/Makefile b/tools/Makefile
index bcae806..9a617ad 100644
--- a/tools/Makefile
+++ b/tools/Makefile
@@ -44,6 +44,9 @@ cpupower: FORCE
cgroup firewire hv guest usb virtio vm net: FORCE
$(call descend,$@)

+liblockdep: FORCE
+ $(call descend,lib/lockdep)
+
libapikfs: FORCE
$(call descend,lib/api)

@@ -91,6 +94,9 @@ cpupower_clean:
cgroup_clean hv_clean firewire_clean lguest_clean usb_clean virtio_clean vm_clean net_clean:
$(call descend,$(@:_clean=),clean)

+liblockdep_clean:
+ $(call descend,lib/lockdep,clean)
+
libapikfs_clean:
$(call descend,lib/api,clean)

diff --git a/tools/lib/lockdep/Makefile b/tools/lib/lockdep/Makefile
index cb09d3f..bba2f52 100644
--- a/tools/lib/lockdep/Makefile
+++ b/tools/lib/lockdep/Makefile
@@ -1,8 +1,7 @@
# file format version
FILE_VERSION = 1

-MAKEFLAGS += --no-print-directory
-LIBLOCKDEP_VERSION=$(shell make -sC ../../.. kernelversion)
+LIBLOCKDEP_VERSION=$(shell make --no-print-directory -sC ../../.. kernelversion)

# Makefiles suck: This macro sets a default value of $(2) for the
# variable named by $(1), unless the variable has been set by
@@ -231,7 +230,7 @@ install_lib: all_cmd
install: install_lib

clean:
- $(RM) *.o *~ $(TARGETS) *.a *.so $(VERSION_FILES) .*.d
+ $(RM) *.o *~ $(TARGETS) *.a *liblockdep*.so* $(VERSION_FILES) .*.d
$(RM) tags TAGS

endif # skip-makefile
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/