[PATCH] uml: tlb flushing fix

This patch fixes a fairly serious tlb flushing bug that makes aio use under
uml very unreliable -- SEGVs, Oops and panic()s occur as a result of stale
tlb entires being used by uml when aio switches mms due to the fact that
uml does not implement the activate_mm() hook.  This patch introduces a
simple but correct approach (read: hammer) for implementing activate_mm()
in uml by doing a force_flush_all() if the new mm is different from old.
With this patch in place, uml is able to succeed at the aio test case that
was randomly faulting for me before.

Cc: Jeff Dike <jdike@addtoit.com>
Cc: <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

diff --git a/include/asm-um/mmu_context.h b/include/asm-um/mmu_context.h
index 7529c9c853dda226c1e42749121ed74b4b3e958e..095bb627b96a1d52be4cc1291f48c308449f65a1 100644 (file)
--- a/include/asm-um/mmu_context.h
+++ b/include/asm-um/mmu_context.h
@@ -16,8 +16,12 @@
 
 #define deactivate_mm(tsk,mm)  do { } while (0)
 
+extern void force_flush_all(void);
+
 static inline void activate_mm(struct mm_struct *old, struct mm_struct *new)
 {
+       if (old != new)
+               force_flush_all();
 }
 
 static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,