lkml.org 
[lkml]   [2008]   [Jul]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [git pull] core, x86: make LIST_POISON less deadly

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> On Mon, 14 Jul 2008, Avi Kivity wrote:
> >
> > We could add a printk to the #GP handler that alerts the reader that a
> > poisoned list is suspected, if we find the address pattern in one of the
> > registers.
>
> I wouldn't worry too much - it's going to be pretty visible anyway.
>
> The only thing I _would_ worry about is the AMD prefetch bug - some AMD
> cores raise a spurious page fault for prefetch instructions, and we ignore
> it.
>
> I _think_ that bug is a pure TLB issue and would never trigger for an
> address that doesn't do page fault handling at all because it gets caught
> in the "is the address valid" code, but it does make me go "Hmm".
>
> See
>
> http://lkml.org/lkml/2003/9/10/397
>
> from Rich Brunner. The AMD errata listing does say just page fault,
> and does talk about speculative TLB reloads, so I think we're all
> good.

one of my testboxes has this erratum and i saw related bugs in the past
during stress-tests so it will show up if the non-canonical address
causes any problems. So i think 0xdead000000000000 will be fine, and
even if it isnt, we'll catch any problems fast enough.

Ingo


\
 
 \ /
  Last update: 2008-07-14 18:37    [W:0.034 / U:0.184 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site