Re: Bug in PCI core

[Matthew Wilcox <matthew@wil.cx>]

On Fri, Oct 13, 2006 at 06:09:09PM +0100, Alan Cox wrote:
> Ar Gwe, 2006-10-13 am 12:34 -0400, ysgrifennodd Adam Belay:
> > I agree this needs to be fixed.  However, as I previously mentioned,
> > this isn't the right place to attack the problem.  Remember, this wasn't
> > originally a kernel regression.  Rather it's a workaround for a known
> 
> It's a kernel regression. It used to be reliable to read X resource
> addresses at any time.

No it didn't.  It's undefined behaviour to perform *any* PCI config
access to the device while it's doing a D-state transition.  It may have
happened to work with the chips you tried it with, but more likely you
never hit that window because X simply didn't try to do that.

> > Finally, it's worth noting that this issue is really a corner-case, and
> > in most systems it's extremely rare that even incorrect userspace apps
> > would have any issue.
> 
> Except just occasionally and randomly in the field, probably almost
> undebuggable and irreproducable - the very worst conceivable kind of
> bug.

Indeed.  Only now we have software producing it, rather than hardware
producing it.  That's actually an improvement I think, since it forces
awareness of the issue.