On 04/04/2015 09:40, Corinna Vinschen wrote:
> On Apr  3 23:09, Jon TURNEY wrote:
>> On 01/04/2015 15:22, Corinna Vinschen wrote:
>>> On Apr  1 14:19, Jon TURNEY wrote:
>>>> Add ucontext.h header, defining ucontext_t and mcontext_t types.
>>>>
>>>> Provide sigaction sighandlers with a ucontext_t parameter, containing stack and
>>>> context information.
>>>>
>>>> 	* include/sys/ucontext.h : New header.
>>>> 	* include/ucontext.h : Ditto.
>>>> 	* exceptions.cc (call_signal_handler): Provide ucontext_t
>>>> 	parameter to signal handler function.
>>>
>>> Patch is ok with a single change:  Please add a "FIXME?" comment to:
>>>
>>>    else
>>>      RtlCaptureContext();
>>>
>>> On second thought, calling RtlCaptureContext here is probably wrong.
>>
>> Wrong and also dangerous.
>>
>> This causes random crashes on x86.
>>
>> It seems that RtlCaptureContext requires the framepointer of the calling
>> function in ebp, which it uses to report the rip and rsp of it's caller.
>>
>> It also seems that gcc can decide to optimize the setting of the
>> framepointer away, irrespective of the fact that -fomit-frame-pointer is not
>> used when building exceptions.cc
>>
>> If _cygtls::call_signal_handler() happens to get called with ebp pointing to
>> an invalid memory address, as seems to happen occasionally, we will fault in
>> RtlCaptureContext.  (in all cases, the eip and ebp in the returned context
>> are incorrect)
>>
>> I wrote the attached patch, which fakes a callframe for RtlCaptureContext to
>> avoid these possible crashes, but this needs more work to correctly report
>> eip and ebp
>
> Maybe it's simpler than that?  Looking into the GCC info pages, I found
> this:
>
>       Starting with GCC version 4.6, the default setting (when not
>       optimizing for size) for 32-bit GNU/Linux x86 and 32-bit Darwin x86
>       targets has been changed to '-fomit-frame-pointer'.  The default
>       can be reverted to '-fno-omit-frame-pointer' by configuring GCC
>       with the '--enable-frame-pointer' configure option.
>
>       Enabled at levels '-O', '-O2', '-O3', '-Os'.
>       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Very good.  I hadn't noticed that sentence and was just looking at the 
"-O also turns on -fomit-frame-pointer on machines where doing so does 
not interfere with debugging. "

> So it seems adding -fomit-frame-pointer file by file in Makefile.in
> (when building with -O2) is moot and only has an effect when building
> unoptimized, otherwise all files are built with -fomit-frame-pointer
> anyway.
>
> So, what if we drop all the -fomit-frame-pointer from Makefile.in and
> add an
>
>    exceptions_CFLAGS:=-fno-omit-frame-pointer
>
> Does that help?

Yes, that seems to do the trick.  Patch attached.