Wrong %ebp after skipping a frame for which the instruction pointer is not in a known module.

When the calculated return address is not in a known module (which can happen
for dynamically generated code or for certain types of crashes), Breakpad scans
the stack for a "better" return address and if such an address is found, the
frame with the "invalid" instruction pointer is skipped.  This usually results
in skipped frames.  The problem is that when such frames are skipped register
%ebp is not properly redirected which frequently results in truncated callstacks
after the skipped frame due to wrong %ebp.

[Ivan Penkov]

.

[Mark Mentovai]

LGTM

[Ted Mielczarek]

LGTM. Would the x86-64 stackwalker benefit from a similar fix? (Probably not as
useful since you're less likely to have a frame pointer.)

http://breakpad.appspot.com/494002/diff/1/processor/stackwalker_x86.cc
File processor/stackwalker_x86.cc (right):

http://breakpad.appspot.com/494002/diff/1/processor/stackwalker_x86.cc#newcod...
processor/stackwalker_x86.cc:395: // the lowest, because we expect that the
function's prolog would have
Fix the "we" in the comment while you're here?

[Ivan Penkov]

On 2012/11/02 16:08:30, Ted Mielczarek wrote:
> LGTM. Would the x86-64 stackwalker benefit from a similar fix? (Probably not
as
> useful since you're less likely to have a frame pointer.)
> 
> http://breakpad.appspot.com/494002/diff/1/processor/stackwalker_x86.cc
> File processor/stackwalker_x86.cc (right):
> 
>
http://breakpad.appspot.com/494002/diff/1/processor/stackwalker_x86.cc#newcod...
> processor/stackwalker_x86.cc:395: // the lowest, because we expect that the
> function's prolog would have
> Fix the "we" in the comment while you're here?

The AMD64 stackwalking code is quite different.  It doesn't seem to be using
Windows Frame Information.  Scanning for return address is done in
StackwalkerAMD64::GetCallerByStackScan but I also see code that redirects %rbp
in that routine.  I don't see a direct/easy way (and need) to port the current
fix to AMD64.

Thanks,
-Ivan

[Ivan Penkov]

The comment was fixed.

Thanks,
-Ivan

http://breakpad.appspot.com/494002/diff/1/processor/stackwalker_x86.cc
File processor/stackwalker_x86.cc (right):

http://breakpad.appspot.com/494002/diff/1/processor/stackwalker_x86.cc#newcod...
processor/stackwalker_x86.cc:395: // the lowest, because we expect that the
function's prolog would have
On 2012/11/02 16:08:30, Ted Mielczarek wrote:
> Fix the "we" in the comment while you're here?

Done.