Messages in this thread | | | Date | Sun, 18 Jul 2010 12:26:59 -0700 | Subject | Re: [patch 1/2] x86_64 page fault NMI-safe | From | Linus Torvalds <> |
| |
On Sun, Jul 18, 2010 at 11:43 AM, Steven Rostedt <rostedt@goodmis.org> wrote: > > Are you sure you don't want to use Mathieu's 2/2 patch?
Yeah, I'm pretty sure. Unless somebody can show that it's faster, I really don't want to muck with regular iret's. Also, as shown during the discussion, even with Mathieu's 2/2 patch, we'd _still_ need NMI to also save cr2 etc.
So the sane thing to do is to put all the NMI crap where it belongs. NMI's need to know about the fact that them taking exceptions is special. That whole "vmalloc_sync_all()" is simply pure brokenness.
In other words, it is _not_ just about 'iret' fixup. It's a bigger thing. NMI's are special, and we don't want to spread that specialness around.
Linus
| |