Test failures on PPC (Fedora 11, Rawhide)
GoogleCodeExporter opened this issue · comments
From the ppc failure log:
Fedora 11:
http://koji.fedoraproject.org/koji/getfile?taskID=1449784&name=build.log
Fedora 12 (Rawhide):
http://koji.fedoraproject.org/koji/getfile?taskID=1450299&name=build.log
While running:
LD_LIBRARY_PATH=./.libs make check
On ppc, it fails on this test:
...
Obtained 10 stack frames.
Obtained 10 stack frames.
./stacktrace_unittest [0x10000754] 0x10000754
./stacktrace_unittest [0x10000954] 0x10000954
./stacktrace_unittest [0x10000998] 0x10000998
./stacktrace_unittest [0x100009e8] 0x100009e8
./stacktrace_unittest [0x10000a38] 0x10000a38
./stacktrace_unittest [0x10000a88] 0x10000a88
./stacktrace_unittest [0x10000ac4] 0x10000ac4
/lib/libc.so.6 [0xfbbf51c] 0xfbbf51c
/lib/libc.so.6 [0xfbbf6c0] 0xfbbf6c0
[(nil)] (nil)
CheckStackTrace() addr: 0x10000a60
Backtrace 0: expected: 0x10000710..0x10000750 actual: 0x10000754 ... Check
failed: ret_addr <= (void *) ((char *) function_start_addr + kMaxFnLen)
FAIL: stacktrace_unittest
...
Original issue reported on code.google.com by spo...@gmail.com
on 2 Jul 2009 at 5:37
Interesting. Are the other tests passing? This test has to make a guess how
big
each function will be, and may get it wrong on some architectures.
Can you run the test in gdb:
LD_LIBRARY_PATH=.libs gdb .libs/lt-stacktrace_unittest
or something similar.
You can just put a breakpoint in main, and do something like
x 0x10000754
what function is it in? Is it the same function as
x 0x10000710
?
Likewise for some of the other addresses it prints out (10000954, etc).
Original comment by csilv...@gmail.com
on 2 Jul 2009 at 6:34
- Added labels: Priority-Medium, Type-Defect
(gdb) x 0x10000754
0x10000754 <printf>: 0x7fe4fb78
(gdb) x 0x10000710
0x10000710 <_Z19CheckStackTraceLeafv>: 0x9421ffa0
(gdb) x 0x10000954
0x10000954 <_Z16CheckStackTrace4i+36>: 0x37ffffff
(gdb) x 0x10000998
0x10000998 <_Z16CheckStackTrace3i+40>: 0x37ffffff
(gdb) x 0x100009e8
0x100009e8 <_Z16CheckStackTrace2i+40>: 0x37ffffff
(gdb) x 0x10000a38
0x10000a38 <_Z16CheckStackTrace1i+40>: 0x37ffffff
(gdb) x 0x10000a88
0x10000a88 <_Z15CheckStackTracei+40>: 0x37ffffff
(gdb) x 0x10000ac4
0x10000ac4 <printf>: 0x3c801000
And yes, all the other tests pass.
Original comment by spo...@gmail.com
on 2 Jul 2009 at 8:20
Weird. That doesn't seem quite plausible to me that printf is at 0x10000754. I
wonder what's going on here.
One more debugging question: can you do
(gdb) disassemble 0x10000710
and also
(gdb) disassemble 0x10000754
?
That will tell us how long these functions are, so we can see what's going on
better.
Original comment by csilv...@gmail.com
on 2 Jul 2009 at 8:26
Any more news on this? We changed this test a bit in perftools 1.4; if the bug
report was on an earlier version, I wonder if you could try the latest and see
if you
still get a failure.
Original comment by csilv...@gmail.com
on 14 Oct 2009 at 11:23
Any more news on this? If not, I'll have to close it CannotReproduce (I don't
have a
ppc machine for testing on, unfortunately).
Original comment by csilv...@gmail.com
on 10 Mar 2010 at 6:14
Go ahead and close it. The PPC folks can re-open if they're interested.
Original comment by spo...@gmail.com
on 10 Mar 2010 at 6:23
OK, done.
Original comment by csilv...@gmail.com
on 10 Mar 2010 at 6:51
- Changed state: CannotReproduce