Change ia64 prefetch to use lfetch.fault

Change ia64 prefetch to use lfetch.fault

[Chen, Kenneth W]

I don't remember whether this has been discussed earlier, won't
it make sense to use the fault variant of lfetch for prefetch()
and prefetchw()?  After all, the caller has high confidence that
the cache line will be used in the future.


Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>

--- linux-2.6.12/include/asm-ia64/processor.h.orig	2005-07-28 14:46:26.629519713 -0700
+++ linux-2.6.12/include/asm-ia64/processor.h	2005-07-28 14:48:17.037721485 -0700
@@ -682,13 +682,13 @@ ia64_imva (void *addr)
 static inline void
 prefetch (const void *x)
 {
-	 ia64_lfetch(ia64_lfhint_none, x);
+	ia64_lfetch_fault(ia64_lfhint_none, x);
 }
 
 static inline void
 prefetchw (const void *x)
 {
-	ia64_lfetch_excl(ia64_lfhint_none, x);
+	ia64_lfetch_fault_excl(ia64_lfhint_none, x);
 }
 
 #define spin_lock_prefetch(x)	prefetchw(x)

Re: Change ia64 prefetch to use lfetch.fault

[david mosberger]

No, it doesn't --- it actually hurt performance last time I measured
it.  Basically, Linux uses prefetch() too often with bad/NULL
pointers, IIRC.  Check my earlier mails where I fixed the IVT code to
properly handle lfetch.fault (from a couple of months ago).

  --david

On 7/28/05, Chen, Kenneth W <kenneth.w.chen@intel.com> wrote:
> I don't remember whether this has been discussed earlier, won't
> it make sense to use the fault variant of lfetch for prefetch()
> and prefetchw()?  After all, the caller has high confidence that
> the cache line will be used in the future.
> 
> 
> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
> 
> --- linux-2.6.12/include/asm-ia64/processor.h.orig      2005-07-28 14:46:26.629519713 -0700
> +++ linux-2.6.12/include/asm-ia64/processor.h   2005-07-28 14:48:17.037721485 -0700
> @@ -682,13 +682,13 @@ ia64_imva (void *addr)
>  static inline void
>  prefetch (const void *x)
>  {
> -        ia64_lfetch(ia64_lfhint_none, x);
> +       ia64_lfetch_fault(ia64_lfhint_none, x);
>  }
> 
>  static inline void
>  prefetchw (const void *x)
>  {
> -       ia64_lfetch_excl(ia64_lfhint_none, x);
> +       ia64_lfetch_fault_excl(ia64_lfhint_none, x);
>  }
> 
>  #define spin_lock_prefetch(x)  prefetchw(x)
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-ia64" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 


-- 
Mosberger Consulting LLC, voice/fax: 510-744-9372,
http://www.mosberger-consulting.com/
35706 Runckel Lane, Fremont, CA 94536