[SEGFAULT: perf]

[SEGFAULT: perf]

[Andrew Watts]

[-- Attachment #1: Type: text/plain, Size: 3743 bytes --]

perf report on 3.4.6 segfaults when certain pointers are null. Attached is
a patch that applies cleanly @3.4.6 and addresses (A). I am not comfortable
suggesting fixes for (B) and (C).

~ Andy

===

(A) perf report with sort on comm/pid/parent/dso:

(gdb) run report --sort=comm
Program received signal SIGSEGV, Segmentation fault.
0x0805c00b in perf_evsel__add_hist_entry (evsel=0x81c56a8, al=0xbfffe9b0,
sample=0xbfffea84, machine=0x81c3fa8) at builtin-report.c:171
171             if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
#0  0x0805c00b in perf_evsel__add_hist_entry (evsel=0x81c56a8, al=0xbfffe9b0,
    sample=0xbfffea84, machine=0x81c3fa8) at builtin-report.c:171
#1  0x0805c234 in process_sample_event (tool=0xbffff27c, event=0xb6db1f38,
    sample=0xbfffea84, evsel=0x81c56a8, machine=0x81c3fa8)
    at builtin-report.c:216
#2  0x080a3097 in perf_session_deliver_event (session=0x81c3f50, event=
    0xb6db1f38, sample=0xbfffea84, tool=0xbffff27c, file_offset=339768)
    at util/session.c:885
#3  0x080a24d1 in flush_sample_queue (s=0x81c3f50, tool=0xbffff27c)
    at util/session.c:587
#4  0x080a40e3 in __perf_session__process_events (session=0x81c3f50,
    data_offset=280, data_size=419640, file_size=419920, tool=0xbffff27c)
    at util/session.c:1257
#5  0x080a41d3 in perf_session__process_events (self=0x81c3f50, tool=
    0xbffff27c) at util/session.c:1273
(gdb) print he->ms.sym
$1 = (struct symbol *) 0x0

--

(B) perf report segfaults on sorts of symbol_from/symbol_to:

(gdb) run report --sort=symbol_from
Program received signal SIGSEGV, Segmentation fault.
0x080b8777 in sort__sym_from_cmp (left=0xbfffe878, right=0x84dfde0)
at util/sort.c:334
334             if (!from_l->sym && !from_r->sym
(gdb) bt
#0  0x080b8777 in sort__sym_from_cmp (left=0xbfffe878, right=0x84dfde0)
    at util/sort.c:334
#1  0x080ba0b8 in hist_entry__cmp (left=0xbfffe878, right=0x84dfde0)
    at util/hist.c:345
#2  0x080b9c31 in add_hist_entry (hists=0x81c571c, entry=0xbfffe878, al=
    0xbfffe9b0, period=333940) at util/hist.c:254
#3  0x080ba04c in __hists__add_entry (self=0x81c571c, al=0xbfffe9b0,
    sym_parent=0x0, period=333940) at util/hist.c:335
#4  0x0805bf50 in perf_evsel__add_hist_entry (evsel=0x81c56a8, al=0xbfffe9b0,
    sample=0xbfffea84, machine=0x81c3fa8) at builtin-report.c:149
#5  0x0805c234 in process_sample_event (tool=0xbffff27c, event=0xb6db1840,
    sample=0xbfffea84, evsel=0x81c56a8, machine=0x81c3fa8)
    at builtin-report.c:216
(gdb) print left->branch_info
$2 = (struct branch_info *) 0x0
(gdb) print right->branch_info
$3 = (struct branch_info *) 0x0

--

(C) perf report segfaults with dso_from/dso_to:

(gdb) run report --sort=dso_to
Program received signal SIGSEGV, Segmentation fault.
sort__dso_to_cmp (left=0xbfffe878, right=0x82346f0) at util/sort.c:317
317             return _sort__dso_cmp(left->branch_info->to.map,
(gdb) bt
#0  sort__dso_to_cmp (left=0xbfffe878, right=0x82346f0) at util/sort.c:317
#1  0x080ba0c8 in hist_entry__cmp (left=0xbfffe878, right=0x82346f0)
    at util/hist.c:345
#2  0x080b9c41 in add_hist_entry (hists=0x81c571c, entry=0xbfffe878, al=
    0xbfffe9b0, period=333331) at util/hist.c:254
#3  0x080ba05c in __hists__add_entry (self=0x81c571c, al=0xbfffe9b0,
    sym_parent=0x0, period=333331) at util/hist.c:335
#4  0x0805bf50 in perf_evsel__add_hist_entry (evsel=0x81c56a8, al=0xbfffe9b0,
    sample=0xbfffea84, machine=0x81c3fa8) at builtin-report.c:149
#5  0x0805c242 in process_sample_event (tool=0xbffff27c, event=0xb6dbf800,
    sample=0xbfffea84, evsel=0x81c56a8, machine=0x81c3fa8)
    at builtin-report.c:216
(gdb) print left->branch_info
$4 = (struct branch_info *) 0x0
(gdb) print right->branch_info
$5 = (struct branch_info *) 0x0

===


[-- Attachment #2: perf1.diff --]
[-- Type: text/plain, Size: 434 bytes --]

--- builtin-report.c.orig		2012-07-22
+++ builtin-report.c			2012-07-22
@@ -162,7 +162,7 @@ static int perf_evsel__add_hist_entry(st
 	 * so we don't allocated the extra space needed because the stdio
 	 * code will not use it.
 	 */
-	if (al->sym != NULL && use_browser > 0) {
+	if (al->sym != NULL && he->ms.sym != NULL && use_browser > 0) {
 		struct annotation *notes = symbol__annotation(he->ms.sym);
 
 		assert(evsel != NULL);

Re: [SEGFAULT: perf]

[Namhyung Kim]

Hi, Andrew

On Mon, 23 Jul 2012 08:52:17 -0500, Andrew Watts wrote:
> perf report on 3.4.6 segfaults when certain pointers are null. Attached is
> a patch that applies cleanly @3.4.6 and addresses (A). I am not comfortable
> suggesting fixes for (B) and (C).
>

(A) has a fix already in the tip tree [1], and maybe in mainline too.

And yes, I think this sort thing needs some love. It's on my TODO list
but not have time to do it yet.

[1] http://www.spinics.net/lists/linux-tip-commits/msg15695.html

Thanks,
Namhyung


> ~ Andy
>
> ===
>
> (A) perf report with sort on comm/pid/parent/dso:
>
> (gdb) run report --sort=comm
> Program received signal SIGSEGV, Segmentation fault.
> 0x0805c00b in perf_evsel__add_hist_entry (evsel=0x81c56a8, al=0xbfffe9b0,
> sample=0xbfffea84, machine=0x81c3fa8) at builtin-report.c:171
> 171             if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
> #0  0x0805c00b in perf_evsel__add_hist_entry (evsel=0x81c56a8, al=0xbfffe9b0,
>     sample=0xbfffea84, machine=0x81c3fa8) at builtin-report.c:171
> #1  0x0805c234 in process_sample_event (tool=0xbffff27c, event=0xb6db1f38,
>     sample=0xbfffea84, evsel=0x81c56a8, machine=0x81c3fa8)
>     at builtin-report.c:216
> #2  0x080a3097 in perf_session_deliver_event (session=0x81c3f50, event=
>     0xb6db1f38, sample=0xbfffea84, tool=0xbffff27c, file_offset=339768)
>     at util/session.c:885
> #3  0x080a24d1 in flush_sample_queue (s=0x81c3f50, tool=0xbffff27c)
>     at util/session.c:587
> #4  0x080a40e3 in __perf_session__process_events (session=0x81c3f50,
>     data_offset=280, data_size=419640, file_size=419920, tool=0xbffff27c)
>     at util/session.c:1257
> #5  0x080a41d3 in perf_session__process_events (self=0x81c3f50, tool=
>     0xbffff27c) at util/session.c:1273
> (gdb) print he->ms.sym
> $1 = (struct symbol *) 0x0
>
> --
>
> (B) perf report segfaults on sorts of symbol_from/symbol_to:
>
> (gdb) run report --sort=symbol_from
> Program received signal SIGSEGV, Segmentation fault.
> 0x080b8777 in sort__sym_from_cmp (left=0xbfffe878, right=0x84dfde0)
> at util/sort.c:334
> 334             if (!from_l->sym && !from_r->sym
> (gdb) bt
> #0  0x080b8777 in sort__sym_from_cmp (left=0xbfffe878, right=0x84dfde0)
>     at util/sort.c:334
> #1  0x080ba0b8 in hist_entry__cmp (left=0xbfffe878, right=0x84dfde0)
>     at util/hist.c:345
> #2  0x080b9c31 in add_hist_entry (hists=0x81c571c, entry=0xbfffe878, al=
>     0xbfffe9b0, period=333940) at util/hist.c:254
> #3  0x080ba04c in __hists__add_entry (self=0x81c571c, al=0xbfffe9b0,
>     sym_parent=0x0, period=333940) at util/hist.c:335
> #4  0x0805bf50 in perf_evsel__add_hist_entry (evsel=0x81c56a8, al=0xbfffe9b0,
>     sample=0xbfffea84, machine=0x81c3fa8) at builtin-report.c:149
> #5  0x0805c234 in process_sample_event (tool=0xbffff27c, event=0xb6db1840,
>     sample=0xbfffea84, evsel=0x81c56a8, machine=0x81c3fa8)
>     at builtin-report.c:216
> (gdb) print left->branch_info
> $2 = (struct branch_info *) 0x0
> (gdb) print right->branch_info
> $3 = (struct branch_info *) 0x0
>
> --
>
> (C) perf report segfaults with dso_from/dso_to:
>
> (gdb) run report --sort=dso_to
> Program received signal SIGSEGV, Segmentation fault.
> sort__dso_to_cmp (left=0xbfffe878, right=0x82346f0) at util/sort.c:317
> 317             return _sort__dso_cmp(left->branch_info->to.map,
> (gdb) bt
> #0  sort__dso_to_cmp (left=0xbfffe878, right=0x82346f0) at util/sort.c:317
> #1  0x080ba0c8 in hist_entry__cmp (left=0xbfffe878, right=0x82346f0)
>     at util/hist.c:345
> #2  0x080b9c41 in add_hist_entry (hists=0x81c571c, entry=0xbfffe878, al=
>     0xbfffe9b0, period=333331) at util/hist.c:254
> #3  0x080ba05c in __hists__add_entry (self=0x81c571c, al=0xbfffe9b0,
>     sym_parent=0x0, period=333331) at util/hist.c:335
> #4  0x0805bf50 in perf_evsel__add_hist_entry (evsel=0x81c56a8, al=0xbfffe9b0,
>     sample=0xbfffea84, machine=0x81c3fa8) at builtin-report.c:149
> #5  0x0805c242 in process_sample_event (tool=0xbffff27c, event=0xb6dbf800,
>     sample=0xbfffea84, evsel=0x81c56a8, machine=0x81c3fa8)
>     at builtin-report.c:216
> (gdb) print left->branch_info
> $4 = (struct branch_info *) 0x0
> (gdb) print right->branch_info
> $5 = (struct branch_info *) 0x0
>
> ===
>
>
> --- builtin-report.c.orig		2012-07-22
> +++ builtin-report.c			2012-07-22
> @@ -162,7 +162,7 @@ static int perf_evsel__add_hist_entry(st
>  	 * so we don't allocated the extra space needed because the stdio
>  	 * code will not use it.
>  	 */
> -	if (al->sym != NULL && use_browser > 0) {
> +	if (al->sym != NULL && he->ms.sym != NULL && use_browser > 0) {
>  		struct annotation *notes = symbol__annotation(he->ms.sym);
>  
>  		assert(evsel != NULL);

Re: [SEGFAULT: perf]

[Andrew Watts]

Hi Namhyung.

On Tue, Jul 24, 2012 at 09:18:24AM +0900, Namhyung Kim wrote:
> 
> (A) has a fix already in the tip tree [1], and maybe in mainline too.

Ah yes, I see the fix in 3.5 as well. Though it is slightly different
from the patch I submitted.

> And yes, I think this sort thing needs some love. It's on my TODO list
> but not have time to do it yet.

I understand all about time constraints. I am glad it is on your TODO 
though; will be on the lookout for commits on that front.

Cheers.

~ Andy