2010-12-09 07:46:22

by Li Zefan

Subject: [PATCH] tracing: Fix preempt count leak

While running my ftrace stress test, this showed up:

BUG: sleeping function called from invalid context at mm/mmap.c:233
...
note: cat[3293] exited with preempt_count 1

The bug was introduced by commit 91e86e560d0b3ce4c5fc64fd2bbb99f856a30a4e
("tracing: Fix recursive user stack trace")

Signed-off-by: Li Zefan <[email protected]>
---
kernel/trace/trace.c | 6 ++----
1 files changed, 2 insertions(+), 4 deletions(-)

diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index c380612..5767457 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -1313,12 +1313,10 @@ ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)

__this_cpu_inc(user_stack_count);

-
-
event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
sizeof(*entry), flags, pc);
if (!event)
- return;
+ goto out_drop_count;
entry = ring_buffer_event_data(event);

entry->tgid = current->tgid;
@@ -1333,8 +1331,8 @@ ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
if (!filter_check_discard(call, entry, buffer, event))
ring_buffer_unlock_commit(buffer, event);

+ out_drop_count:
__this_cpu_dec(user_stack_count);
-
out:
preempt_enable();
}
--
1.6.3

2010-12-21 00:56:03

by Li Zefan

Subject: Re: [PATCH] tracing: Fix preempt count leak

Any comment?

15:47, Li Zefan wrote:
> While running my ftrace stress test, this showed up:
>
> BUG: sleeping function called from invalid context at mm/mmap.c:233
> ...
> note: cat[3293] exited with preempt_count 1
>
> The bug was introduced by commit 91e86e560d0b3ce4c5fc64fd2bbb99f856a30a4e
> ("tracing: Fix recursive user stack trace")
>
> Signed-off-by: Li Zefan <[email protected]>
> ---
> kernel/trace/trace.c | 6 ++----
> 1 files changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
> index c380612..5767457 100644
> --- a/kernel/trace/trace.c
> +++ b/kernel/trace/trace.c
> @@ -1313,12 +1313,10 @@ ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
>
> __this_cpu_inc(user_stack_count);
>
> -
> -
> event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
> sizeof(*entry), flags, pc);
> if (!event)
> - return;
> + goto out_drop_count;
> entry = ring_buffer_event_data(event);
>
> entry->tgid = current->tgid;
> @@ -1333,8 +1331,8 @@ ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
> if (!filter_check_discard(call, entry, buffer, event))
> ring_buffer_unlock_commit(buffer, event);
>
> + out_drop_count:
> __this_cpu_dec(user_stack_count);
> -
> out:
> preempt_enable();
> }

2010-12-21 01:34:03

by Steven Rostedt

Subject: Re: [PATCH] tracing: Fix preempt count leak

On Tue, 2010-12-21 at 08:55 +0800, Li Zefan wrote:
> Any comment?

oops, this was buried in my email. I'll look at it tomorrow.

-- Steve

>
> 15:47, Li Zefan wrote:
> > While running my ftrace stress test, this showed up:
> >
> > BUG: sleeping function called from invalid context at mm/mmap.c:233
> > ...
> > note: cat[3293] exited with preempt_count 1
> >
> > The bug was introduced by commit 91e86e560d0b3ce4c5fc64fd2bbb99f856a30a4e
> > ("tracing: Fix recursive user stack trace")
> >
> > Signed-off-by: Li Zefan <[email protected]>
> > ---
> > kernel/trace/trace.c | 6 ++----
> > 1 files changed, 2 insertions(+), 4 deletions(-)
> >
> > diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
> > index c380612..5767457 100644
> > --- a/kernel/trace/trace.c
> > +++ b/kernel/trace/trace.c
> > @@ -1313,12 +1313,10 @@ ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
> >
> > __this_cpu_inc(user_stack_count);
> >
> > -
> > -
> > event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
> > sizeof(*entry), flags, pc);
> > if (!event)
> > - return;
> > + goto out_drop_count;
> > entry = ring_buffer_event_data(event);
> >
> > entry->tgid = current->tgid;
> > @@ -1333,8 +1331,8 @@ ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
> > if (!filter_check_discard(call, entry, buffer, event))
> > ring_buffer_unlock_commit(buffer, event);
> >
> > + out_drop_count:
> > __this_cpu_dec(user_stack_count);
> > -
> > out:
> > preempt_enable();
> > }