Messages in this thread | | | Subject | Re: [patch 1/2] x86_64 page fault NMI-safe | From | Steven Rostedt <> | Date | Fri, 16 Jul 2010 08:54:58 -0400 |
| |
On Fri, 2010-07-16 at 14:00 +0200, Frederic Weisbecker wrote: > On Thu, Jul 15, 2010 at 07:51:55AM -0700, Linus Torvalds wrote:
> > Such new page table created that might race is only about top level page > right? Otherwise it wouldn't race since the top level entries are shared > and then updates inside lower level pages are naturally propagated, if > I understood you well. > > So, if only top level pages that gets added can generate such lazily > mapping update, I wonder why I experienced this fault everytime with > my patches. > > I allocated 8192 bytes per cpu in a x86-32 system that has only 2 GB. > I doubt there is a top level page table update there at this time with > such a small amount of available memory. But still it faults once on > access. > > I have troubles to visualize the race and the problem here. >
A few trace_printks and a tracing_off() on fault would probably show exactly what was happening ;-)
-- Steve
| |