[OpenAFS-devel] LWP and clang

[Benjamin Kaduk]

On Wed, 16 Oct 2013, Benjamin Kaduk wrote:

> On Wed, 16 Oct 2013, Marc Dionne wrote:
>
>>> 
>>> Hmm, the example I have handy (aklog.core from ~tip-of-1.6.x) shows the
>>> faulting line of source as:
>>> rx_lwp.c:174:
>>> 174     char name[MAXTHREADNAMELENGTH] = "srv_0";
>> 
>> Crashing here which is probably the first write into stack memory
>> after starting the rx_ListenerProc LWP, so that would seem consistent
>> with having a bogus stack pointer after switching lwps.
>
> I can access memory at $rsp at the faulting instruction, but it seems to only 
> be 8-byte aligned, which may be the issue.  (This is on amd64.)
>
> I will look around a bit more.

Yeah, if I tweak rsp before stepping into the listener proc from 
Create_Process_Part2, there's no SIGBUS.

Switching FBSD to use the same block as sys_x86_darwin_80 at line 396 of 
lwp.c lets aklog run to completion.  I'm not sure what it will take to 
convince me that this is right for the right reasons, though.

-Ben