increase exception handler stack size for dumping thread

increase exception handler stack size for dumping thread

When running unittests under ASAN, we see that these code paths can
slightly smash the stack.  Double it to avoid that.

[ RUN      ] ExceptionHandlerTest.InstructionPointerMemoryMinBound
=================================================================
==12775== ERROR: AddressSanitizer: stack-buffer-overflow on address 0xf6787614
at pc 0xf7516b29 bp 0xf6786d38 sp 0xf6786d30
READ of size 4 at 0xf6787614 thread T0
    #0 0xf7516b28
(/build/x86-generic/tmp/portage/chromeos-base/google-breakpad-1181-r66/work/google-breakpad-1181/build/src/client/linux/linux_client_unittest_shlib+0x69eb28)
Shadow byte and word:
  0x3ecf0ec2: f2
  0x3ecf0ec0: f2 f2 f2 f2
More shadow bytes:
  0x3ecf0eb0: f2 f2 f2 f2
  0x3ecf0eb4: 04 f4 f4 f4
  0x3ecf0eb8: f2 f2 f2 f2
  0x3ecf0ebc: 04 f4 f4 f4
=>0x3ecf0ec0: f2 f2 f2 f2
  0x3ecf0ec4: 04 f4 f4 f4
  0x3ecf0ec8: f2 f2 f2 f2
  0x3ecf0ecc: 04 f4 f4 f4
  0x3ecf0ed0: f2 f2 f2 f2
Stats: 0M malloced (0M for red zones) by 2757 calls
Stats: 0M realloced by 0 calls
Stats: 0M freed by 2229 calls
Stats: 0M really freed by 0 calls
Stats: 3M (899 full pages) mmaped in 7 calls
  mmaps   by size class: 7:4095; 8:2047; 9:1023; 10:511; 14:32; 16:16;
  mallocs by size class: 7:1831; 8:590; 9:85; 10:233; 14:3; 16:15;
  frees   by size class: 7:1459; 8:437; 9:84; 10:232; 14:2; 16:15;
  rfrees  by size class:
Stats: malloc large: 15 small slow: 25
==12775== ABORTING

BUG=chromium:293519
TEST=ran unittests under ASAN and they now pass
R=benchan@chromium.org

Committed: https://code.google.com/p/google-breakpad/source/detail?r=1224

[Ben Chan]

On 2013/10/06 10:26:27, vapier wrote:

lgtm.

btw, is the issue ASAN specific?

[vapier]

i think the issue is normally not hit as the normal stack usage in a func is
<100 bytes.  i think asan makes it worse due to the overhead of runtime checking
and stack padding -- it seems to add like 300 bytes per function call to the
stack.

i could make this size dependent upon ADDRESS_SANITIZE define if people prefer. 
it just didn't seem like it really mattered here.

[vapier]

Committed patchset #1 manually as r1224 (presubmit successful).