Skip to contentSkip to conversation

: x86_64: Remove unnecessary cast in prefetch()

PATCH In review Serge Belyshev 3 participants 5 messages Sep 6, 2007 – Sep 15, 2007

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 : x86_64: Remove unnecessary cast in prefetch()

It is ok to call prefetch() function with NULL argument, as specifically commented in include/linux/prefetch.h. But in standard C, it is invalid to dereference NULL pointer (see C99 standard 6.5.3.2 paragraph 4 and note #84). Newer gcc versions (4.3 and above) will use that to conclude that "x" argument is non-null and thus wreaking havok everywhere prefetch() was inlined. Fixed by removing cast and changing asm constraint.

This can be fixed better by using gcc's __builtin_prefetch().

---
 include/asm-x86_64/processor.h |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/asm-x86_64/processor.h +1 -1
Index: linux/include/asm-x86_64/processor.h
===================================================================
@@ -373,7 +373,7 @@ static inline void sync_core(void)
 #define ARCH_HAS_PREFETCH
 static inline void prefetch(void *x) 
 { 
-	asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x));
+	asm volatile("prefetcht0 (%0)" :: "r" (x));
 } 
 
 #define ARCH_HAS_PREFETCHW 1
#1 lore raw
Quoted 8 lines · Serge Belyshev

It is ok to call prefetch() function with NULL argument, as specifically commented in include/linux/prefetch.h. But in standard C, it is invalid to dereference NULL pointer (see C99 standard 6.5.3.2 paragraph 4 and note #84). Newer gcc versions (4.3 and above) will use that to conclude that "x" argument is non-null and thus wreaking havok everywhere prefetch() was inlined. Fixed by removing cast and changing asm constraint.

This can be fixed better by using gcc's __builtin_prefetch().

I changed it to just use that. Thanks.

It seems like gcc 3.1/3.2 already supported it and that's the earliest gcc still supported so it can be used unconditionally.

-Andi

#2 lore raw
Quoted 7 lines · Andi Kleen
Quoted 1 line

This can be fixed better by using gcc's __builtin_prefetch().

I changed it to just use that. Thanks.

It seems like gcc 3.1/3.2 already supported it and that's the earliest gcc still supported so it can be used unconditionally.

Hi!

Will you submit this patch for inclusion into 2.6.23? It is important to make kernel work with GCC 4.3 and above. (Also note that gcc 4.2 already smart enough to break that code, but kernel is just lucky currently).

#3 lore raw
Quoted 6 lines · Serge Belyshev
Quoted 4 lines · Andi Kleen
Quoted 1 line

This can be fixed better by using gcc's __builtin_prefetch().

I changed it to just use that. Thanks.

Quoted 7 lines
Quoted 3 lines

It seems like gcc 3.1/3.2 already supported it and that's the earliest gcc still supported so it can be used unconditionally.

Hi!

Will you submit this patch for inclusion into 2.6.23? It is important

Didn't plan it currently.

Quoted 2 lines

to make kernel work with GCC 4.3 and above. (Also note that gcc 4.2 already smart enough to break that code, but kernel is just lucky currently).

How would it break the code exactly? It more sounded like an optimization to me. Would it generate incorrect code without it?

-Andi

#4 lore raw

Andi Kleen <[email protected]> writes:

...

Quoted 5 lines
Quoted 2 lines

to make kernel work with GCC 4.3 and above. (Also note that gcc 4.2 already smart enough to break that code, but kernel is just lucky currently).

How would it break the code exactly? It more sounded like an optimization to me. Would it generate incorrect code without it?

See http://gcc.gnu.org/PR33294 for testcase which fails if compiled with gcc 4.2 and above. Yes, incorrect code can be generated without the patch as demonstrated by testcase, but i didn't analyze why kernel appears to work with 4.2 currently.