[PATCH] Proper fix for highmem kmap_atomic functions for VMI for 2.6.21
authorZachary Amsden <zach@vmware.com>
Sun, 8 Apr 2007 23:04:01 +0000 (16:04 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Mon, 9 Apr 2007 02:47:55 +0000 (19:47 -0700)
commit49f19710512c825aaea73b9207b3a848027cda1d
tree06da31bd9a84273e12aa43f536f90eb8146ff92e
parenta5bfffac645a7b2d8119f8bbae34df5c94832799
[PATCH] Proper fix for highmem kmap_atomic functions for VMI for 2.6.21

Since lazy MMU batching mode still allows interrupts to enter, it is
possible for interrupt handlers to try to use kmap_atomic, which fails when
lazy mode is active, since the PTE update to highmem will be delayed.  The
best workaround is to issue an explicit flush in kmap_atomic_functions
case; this is the only way nested PTE updates can happen in the interrupt
handler.

Thanks to Jeremy Fitzhardinge for noting the bug and suggestions on a fix.

This patch gets reverted again when we start 2.6.22 and the bug gets fixed
differently.

Signed-off-by: Zachary Amsden <zach@vmware.com>
Cc: Andi Kleen <ak@muc.de>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/i386/kernel/vmi.c
arch/i386/mm/highmem.c
include/asm-generic/pgtable.h
include/asm-i386/paravirt.h