Messages in this thread | | | Date | Thu, 15 Jul 2010 08:38:51 -0700 | Subject | Re: [patch 1/2] x86_64 page fault NMI-safe | From | Linus Torvalds <> |
| |
On Thu, Jul 15, 2010 at 7:51 AM, Linus Torvalds <torvalds@linux-foundation.org> wrote: > > So it's not the lazy page table fill that is the problem. Never has > been. We've been doing the lazy fill for a long time, and it was > simple and useful way back when.
Btw, this is true to the degree that I would _much_ rather just get rid of the crazy "vmalloc_sync_all()" crap entirely, and make it clear that non-lazy vmalloc page table fill is a bug.
Because quite frankly, it _is_ a bug to depend on non-lazy vmalloc. The whole function is only implemented on 32-bit x86, so any code that thinks it needs it is either just wrong, or will only work on 32-bit x86 anyway (and on other architectures by pure chance, likely because their VM fill granularity is so big that they never saw the problem).
So getting rid of vmalloc_sync_all() entirely would be a good thing. Then we wouldn't have that silly and pointless interface, and we wouldn't have that crazy "this only does something on x86-32, everywhere else it's a placebo".
Linus
| |