Re: [PATCH] x86_64: another fix for canonical RIPs during signal handling

[Willy Tarreau <w@1wt.eu>]

On Fri, Jun 23, 2006 at 03:46:21PM +0200, Andi Kleen wrote:
> 
> > If I understand it well, an application which maps address 0 has no way to
> > be notified that the kernel detected a corrupted stack pointer. 
> 
> It will just not crash again after the application tried to deliberately
> crash the kernel.

"deliberately" is a bit exagerated here. Failed stack overflows,
hardware memory corruption and various bugs that happen to most
application developpers at early coding stage are not what can be
called "deliberate".

Also, I don't know if memory leak detectors rely on getting a SEGV,
but this patch would make them useless on apps which map addr 0.

> > I agree 
> > that if the proposed patch avoids to make this undesired distinction between
> > apps that map addr 0 and those which don't, it would be better to merge it.
> > Andi, you said there was nothing wrong with it, do you accept that it gets
> > merged ?
> 
> As I said, it's not wrong, just not necessary.

I understand your point, but I think that covering most situations the
same way helps reducing exceptions, and helps troubleshooting.

> -Andi

Regards,
Willy