[Libunwind-devel] The Lrs-race test case intermittently fails

[Paul Pluzhnikov]

Greetings,

I am observing intermittent hangs and crashes in Lrs-race test case on
Ubuntu Precise variant, using current trunk
(65ac86741606e1d87aef75755c699e4fa6884230).

for j in $(seq 1 100); do alarm 10 make check-TESTS TESTS=Lrs-race &>
/dev/null || echo $j; done
10
43
64
77
80
96


That is, 6% of executions either crash or hang.
Of the 6, 2 were hangs, and 4 were SIGSEGVs.

Here is a sample crash:

Program terminated with signal SIGSEGV, Segmentation fault.
#0  rs_new (c=0x2ae1a7227240, cache=0x2ae1a683f258
<local_addr_space+88>) at
../../libunwind-pristine/src/dwarf/Gparser.c:648
648               if (tmp->coll_chain >= DWARF_UNW_CACHE_SIZE)
(gdb) bt
#0  rs_new (c=0x2ae1a7227240, cache=0x2ae1a683f258
<local_addr_space+88>) at
../../libunwind-pristine/src/dwarf/Gparser.c:648
#1  _ULx86_64_dwarf_find_save_locs (c=0x2ae1a7227240) at
../../libunwind-pristine/src/dwarf/Gparser.c:893
#2  0x00002ae1a6637889 in _ULx86_64_dwarf_step (c=0x2ae1a7227240) at
../../libunwind-pristine/src/dwarf/Gstep.c:34
#3  0x00002ae1a6633a45 in _ULx86_64_step (cursor=0x2ae1a7227240) at
../../libunwind-pristine/src/x86_64/Gstep.c:71
#4  0x00002ae1a6634488 in trace_init_addr (rsp=47148660063968, rbp=0,
rip=4203294, cfa=47148660063968, cursor=0x2ae1a7227240,
f=0x2ae1a64638e0) at ../../libunwind-pristine/src/x86_64/Gtrace.c:248
#5  trace_lookup (rsp=47148660063968, rbp=0, rip=4203294,
cfa=47148660063968, cache=<optimized out>, cursor=0x2ae1a7227240) at
../../libunwind-pristine/src/x86_64/Gtrace.c:330
#6  _ULx86_64_tdep_trace (cursor=0x2ae1a7227240, buffer=<optimized
out>, size=0x2ae1a7227dfc) at
../../libunwind-pristine/src/x86_64/Gtrace.c:447
#7  0x00002ae1a663266b in unw_backtrace (buffer=0x2ae1a7227e30,
size=20) at ../../libunwind-pristine/src/mi/backtrace.c:69
#8  0x0000000000401b44 in foo_98 () at
../../libunwind-pristine/tests/Lrs-race.c:1032
#9  0x000000000040231f in bar (p=<optimized out>) at
../../libunwind-pristine/tests/Lrs-race.c:1439
#10 0x00002ae1a6851e9a in start_thread (arg=0x2ae1a7228700) at
pthread_create.c:308
#11 0x00002ae1a6b5b3fd in clone () at
../sysdeps/unix/sysv/linux/x86_64/clone.S:112

(gdb) p index
$1 = 44 ','
(gdb) p cache->hash[index]
$2 = 120
(gdb) p tmp
$3 = (dwarf_reg_state_t *) 0x2ae1a7d3f338
(gdb) p *tmp
Cannot access memory at address 0x2ae1a7d3f338


Please let me know if you can't repro, or need additional details.

Looks like this has already been reported here as well:
http://savannah.nongnu.org/bugs/?39294


Thanks,
-- 
Paul Pluzhnikov