Point the child stack at the top-most address rather than the bottom-most add...

In Linux Breakpad, point the child stack at the top-most address rather than the bottom-most address before calling clone().

BUG=none
TEST=Official builds on Linux should be able to dump/upload reliably when running with --crash-test.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=17354

[agl, 2009-05-29 22:25:38]

http://codereview.chromium.org/115955/diff/1/2
File breakpad/linux/exception_handler.cc (right):

http://codereview.chromium.org/115955/diff/1/2#newcode273
Line 273: return 0;
return false;

http://codereview.chromium.org/115955/diff/1/2#newcode274
Line 274: stack += kChildStackSize;  // clone() needs the top-most address.
make it (kChildStackSize - 16) to stay safe.

Oh, and memset(stack + kChildStackSize - 16, 0, 16);

(Typically, on Intel, I think the stack pointer points *at* the next word to be
written to, so "- sizeof(int)" would probably be more technically correct.
However, tools will often assume that the stack is 'terminated' with a null
frame pointer and 16 bytes should be enough for that gubbins.

This is, of course, all stuff that I should have done in the first place, so
thanks :)

[Lei Zhang, 2009-05-29 22:51:10]

On 2009/05/29 22:25:38, agl wrote:
> http://codereview.chromium.org/115955/diff/1/2
> File breakpad/linux/exception_handler.cc (right):
> 
> http://codereview.chromium.org/115955/diff/1/2#newcode273
> Line 273: return 0;
> return false;
> 
> http://codereview.chromium.org/115955/diff/1/2#newcode274
> Line 274: stack += kChildStackSize;  // clone() needs the top-most address.
> make it (kChildStackSize - 16) to stay safe.
> 
> Oh, and memset(stack + kChildStackSize - 16, 0, 16);
> 
> (Typically, on Intel, I think the stack pointer points *at* the next word to
be
> written to, so "- sizeof(int)" would probably be more technically correct.
> However, tools will often assume that the stack is 'terminated' with a null
> frame pointer and 16 bytes should be enough for that gubbins.
> 
> This is, of course, all stuff that I should have done in the first place, so
> thanks :)

In linux_syscall_support.h we have the asm equivalent of:
child_stack = (child_stack & ~0xF) - 20;

Isn't this redundent?

[Lei Zhang, 2009-06-01 21:11:27]

Please take a look at patchset 2.

[agl, 2009-06-01 21:13:58]

LGTM