Re: [PATCH] perf tool: Ensure forward progress unwinding frames

[David Ahern]

On 12/9/13, 3:02 AM, Jiri Olsa wrote:
> > On 12/1/13, 5:01 PM, David Ahern wrote:
> > > The dwarf option for callstacks is getting stuck on the unwind and showing
> > > repeated frames. For example,
> > >
> > > qemu-system-x86 13759 [001] 287663.084957: syscalls:sys_enter_futex: uaddr: 0x7f770a958700, op: 0x00000081, val: 0x00000001, utime: 0x7f75ea
> > >              7f7705fd23ea __lll_unlock_wake (/lib64/libpthread-2.14.90.so)
> > >              7f7705fcf045 _L_unlock_649 (/lib64/libpthread-2.14.90.so)
> > >              7f7705fcf045 _L_unlock_649 (/lib64/libpthread-2.14.90.so)
> > >              7f7705fcf045 _L_unlock_649 (/lib64/libpthread-2.14.90.so)
> > >              7f7705fcf045 _L_unlock_649 (/lib64/libpthread-2.14.90.so)
> > >              7f7705fcf045 _L_unlock_649 (/lib64/libpthread-2.14.90.so)
> > >              ...
>
> I wonder this is us or libunwind bug (or cfi code) ;-)
>
> looks like you can reproduce..? are you using the latest libunwind?

$ rpm -q libunwind
libunwind-0.99-2.20110424git1e10c293.fc16.x86_64

> > >              (Last frame repeats 126 times)
> > >
> > > Catch that the same frame is hit multiple times and break out of the loop.
> > >
> > > With this patch:
> > >
> > > qemu-system-x86 13759 [001] 287663.084967: syscalls:sys_exit_futex: 0x1
> > >              7f7705fd23ea __lll_unlock_wake (/lib64/libpthread-2.14.90.so)
> > >              7f7705fcf045 _L_unlock_649 (/lib64/libpthread-2.14.90.so)
> > >
> > > Signed-off-by: David Ahern <dsahern@xxxxxxxxx>
> > > Cc: Jiri Olsa <jolsa@xxxxxxxxxx>
> > > ---
> > >   tools/perf/util/unwind.c | 5 +++++
> > >   1 file changed, 5 insertions(+)
> > >
> > > diff --git a/tools/perf/util/unwind.c b/tools/perf/util/unwind.c
> > > index 0efd5393de85..817ffc1f3a00 100644
> > > --- a/tools/perf/util/unwind.c
> > > +++ b/tools/perf/util/unwind.c
> > > @@ -563,6 +563,7 @@ static int get_entries(struct unwind_info *ui, unwind_entry_cb_t cb,
> > >   {
> > >   	unw_addr_space_t addr_space;
> > >   	unw_cursor_t c;
> > > +	unw_word_t prev_ip = 0;
> > >   	int ret;
> > >
> > >   	addr_space = unw_create_addr_space(&accessors, 0);
> > > @@ -579,6 +580,10 @@ static int get_entries(struct unwind_info *ui, unwind_entry_cb_t cb,
> > >   		unw_word_t ip;
> > >
> > >   		unw_get_reg(&c, UNW_REG_IP, &ip);
> > > +		if (ip == prev_ip)
> > > +			break;
> > > +		prev_ip = ip;
>
> hum, how about recursion?
>
> how about some counter and break after some insane value

I chose > 1 as the insane value. Do you have a suggestion on something 
else? 5? 10?

> and maybe display some warning..
>
>    qemu-system-x86 13759 [001] 287663.084967: syscalls:sys_exit_futex: 0x1
>               7f7705fd23ea __lll_unlock_wake (/lib64/libpthread-2.14.90.so)
>               7f7705fcf045 _L_unlock_649 (/lib64/libpthread-2.14.90.so)
>               warning: unwind broken

sure.

David
--
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/