Hi,
We are debugging a segfault of perf in ordered_events__free().
Disassemble shows the segfault was caused by oe->buff == NULL
in the following line:
/*
* Current buffer might not have all the events allocated
* yet, we need to free only allocated ones ...
*/
list_del(&oe->buffer->list);
After poking around the code, I suspect it is caused by the following
condition in alloc_event():
} else if (oe->buffer) {
new = &oe->buffer->event[oe->buffer_idx];
if (++oe->buffer_idx == MAX_SAMPLE_BUFFER)
oe->buffer = NULL;
Does this theory make sense? If so, what would be the best fix?
Thanks in advance!
Song
On Wed, Jan 16, 2019 at 10:57:49AM -0800, Song Liu wrote:
> Hi,
>
> We are debugging a segfault of perf in ordered_events__free().
hi,
any backtrace or info on how to reproduce it?
> Disassemble shows the segfault was caused by oe->buff == NULL
> in the following line:
>
> /*
> * Current buffer might not have all the events allocated
> * yet, we need to free only allocated ones ...
> */
> list_del(&oe->buffer->list);
>
> After poking around the code, I suspect it is caused by the following
> condition in alloc_event():
>
> } else if (oe->buffer) {
> new = &oe->buffer->event[oe->buffer_idx];
> if (++oe->buffer_idx == MAX_SAMPLE_BUFFER)
> oe->buffer = NULL;
argh.. yea, we need to check oe->buffer in ordered_events__free
would attached change fix it for you?
thanks,
jirka
---
diff --git a/tools/perf/util/ordered-events.c b/tools/perf/util/ordered-events.c
index 897589507d97..ea523d3b248f 100644
--- a/tools/perf/util/ordered-events.c
+++ b/tools/perf/util/ordered-events.c
@@ -391,8 +391,10 @@ void ordered_events__free(struct ordered_events *oe)
* Current buffer might not have all the events allocated
* yet, we need to free only allocated ones ...
*/
- list_del(&oe->buffer->list);
- ordered_events_buffer__free(oe->buffer, oe->buffer_idx, oe);
+ if (oe->buffer) {
+ list_del(&oe->buffer->list);
+ ordered_events_buffer__free(oe->buffer, oe->buffer_idx, oe);
+ }
/* ... and continue with the rest */
list_for_each_entry_safe(buffer, tmp, &oe->to_free, list) {
Thanks Jiri!
On Wed, Jan 16, 2019 at 1:34 PM Jiri Olsa <[email protected]> wrote:
>
> On Wed, Jan 16, 2019 at 10:57:49AM -0800, Song Liu wrote:
> > Hi,
> >
> > We are debugging a segfault of perf in ordered_events__free().
>
> hi,
> any backtrace or info on how to reproduce it?
Here is the backtrace:
#0 0x0000000000500055 in ordered_events(float, long double,...)(...) ()
#1 0x0000000000500196 in ordered_events.reinit ()
#2 0x00000000004fe413 in perf_session.process_events ()
#3 0x0000000000440431 in cmd_record ()
#4 0x00000000004a439f in run_builtin ()
#5 0x000000000042b3e5 in main ()"
>
> > Disassemble shows the segfault was caused by oe->buff == NULL
> > in the following line:
> >
> > /*
> > * Current buffer might not have all the events allocated
> > * yet, we need to free only allocated ones ...
> > */
> > list_del(&oe->buffer->list);
> >
> > After poking around the code, I suspect it is caused by the following
> > condition in alloc_event():
> >
> > } else if (oe->buffer) {
> > new = &oe->buffer->event[oe->buffer_idx];
> > if (++oe->buffer_idx == MAX_SAMPLE_BUFFER)
> > oe->buffer = NULL;
>
>
> argh.. yea, we need to check oe->buffer in ordered_events__free
>
> would attached change fix it for you?
Let me try roll a fixed version to confirm.
Thanks again!
Song
>
> thanks,
> jirka
>
>
> ---
> diff --git a/tools/perf/util/ordered-events.c b/tools/perf/util/ordered-events.c
> index 897589507d97..ea523d3b248f 100644
> --- a/tools/perf/util/ordered-events.c
> +++ b/tools/perf/util/ordered-events.c
> @@ -391,8 +391,10 @@ void ordered_events__free(struct ordered_events *oe)
> * Current buffer might not have all the events allocated
> * yet, we need to free only allocated ones ...
> */
> - list_del(&oe->buffer->list);
> - ordered_events_buffer__free(oe->buffer, oe->buffer_idx, oe);
> + if (oe->buffer) {
> + list_del(&oe->buffer->list);
> + ordered_events_buffer__free(oe->buffer, oe->buffer_idx, oe);
> + }
>
> /* ... and continue with the rest */
> list_for_each_entry_safe(buffer, tmp, &oe->to_free, list) {
On Wed, Jan 16, 2019 at 11:33:55PM -0800, Song Liu wrote:
> On Wed, Jan 16, 2019 at 1:53 PM Song Liu <[email protected]> wrote:
> >
> > Thanks Jiri!
> >
> > On Wed, Jan 16, 2019 at 1:34 PM Jiri Olsa <[email protected]> wrote:
> > >
> > > On Wed, Jan 16, 2019 at 10:57:49AM -0800, Song Liu wrote:
> > > > Hi,
> > > >
> > > > We are debugging a segfault of perf in ordered_events__free().
> > >
> > > hi,
> > > any backtrace or info on how to reproduce it?
> >
> > Here is the backtrace:
> >
> > #0 0x0000000000500055 in ordered_events(float, long double,...)(...) ()
> > #1 0x0000000000500196 in ordered_events.reinit ()
> > #2 0x00000000004fe413 in perf_session.process_events ()
> > #3 0x0000000000440431 in cmd_record ()
> > #4 0x00000000004a439f in run_builtin ()
> > #5 0x000000000042b3e5 in main ()"
> >
> > >
> > > > Disassemble shows the segfault was caused by oe->buff == NULL
> > > > in the following line:
> > > >
> > > > /*
> > > > * Current buffer might not have all the events allocated
> > > > * yet, we need to free only allocated ones ...
> > > > */
> > > > list_del(&oe->buffer->list);
> > > >
> > > > After poking around the code, I suspect it is caused by the following
> > > > condition in alloc_event():
> > > >
> > > > } else if (oe->buffer) {
> > > > new = &oe->buffer->event[oe->buffer_idx];
> > > > if (++oe->buffer_idx == MAX_SAMPLE_BUFFER)
> > > > oe->buffer = NULL;
> > >
> > >
> > > argh.. yea, we need to check oe->buffer in ordered_events__free
> > >
> > > would attached change fix it for you?
> >
> > Let me try roll a fixed version to confirm.
>
> Yes, the patch fixes this segfault. Please CC me on the official patch to
> back port the official version.
thanks for testing, will post it today
jirka
On Wed, Jan 16, 2019 at 1:53 PM Song Liu <[email protected]> wrote:
>
> Thanks Jiri!
>
> On Wed, Jan 16, 2019 at 1:34 PM Jiri Olsa <[email protected]> wrote:
> >
> > On Wed, Jan 16, 2019 at 10:57:49AM -0800, Song Liu wrote:
> > > Hi,
> > >
> > > We are debugging a segfault of perf in ordered_events__free().
> >
> > hi,
> > any backtrace or info on how to reproduce it?
>
> Here is the backtrace:
>
> #0 0x0000000000500055 in ordered_events(float, long double,...)(...) ()
> #1 0x0000000000500196 in ordered_events.reinit ()
> #2 0x00000000004fe413 in perf_session.process_events ()
> #3 0x0000000000440431 in cmd_record ()
> #4 0x00000000004a439f in run_builtin ()
> #5 0x000000000042b3e5 in main ()"
>
> >
> > > Disassemble shows the segfault was caused by oe->buff == NULL
> > > in the following line:
> > >
> > > /*
> > > * Current buffer might not have all the events allocated
> > > * yet, we need to free only allocated ones ...
> > > */
> > > list_del(&oe->buffer->list);
> > >
> > > After poking around the code, I suspect it is caused by the following
> > > condition in alloc_event():
> > >
> > > } else if (oe->buffer) {
> > > new = &oe->buffer->event[oe->buffer_idx];
> > > if (++oe->buffer_idx == MAX_SAMPLE_BUFFER)
> > > oe->buffer = NULL;
> >
> >
> > argh.. yea, we need to check oe->buffer in ordered_events__free
> >
> > would attached change fix it for you?
>
> Let me try roll a fixed version to confirm.
Yes, the patch fixes this segfault. Please CC me on the official patch to
back port the official version.
Thanks,
Song