[PATCH] powerpc: Avoid taking a data miss on every userspace instruction miss

[LEROY Christophe]

Anton Blanchard <anton at ozlabs.org> a écrit :

> From: Anton Blanchard <anton at samba.org>
>
> Early on in do_page_fault() we call store_updates_sp(), regardless of
> the type of exception. For an instruction miss this doesn't make
> sense, because we only use this information to detect if a data miss
> is the result of a stack expansion instruction or not.
>
> Worse still, it results in a data miss within every userspace
> instruction miss handler, because we try and load the very instruction
> we are about to install a pte for!
>
> A simple exec microbenchmark runs 6% faster on POWER8 with this fix:
>
>  #include <stdlib.h>
>  #include <stdio.h>
>  #include <unistd.h>
>
> int main(int argc, char *argv[])
> {
> 	unsigned long left = atol(argv[1]);
> 	char leftstr[16];
>
> 	if (left-- == 0)
> 		return 0;
>
> 	sprintf(leftstr, "%ld", left);
> 	execlp(argv[0], argv[0], leftstr, NULL);
> 	perror("exec failed\n");
>
> 	return 0;
> }
>
> Pass the number of iterations on the command line (eg 10000) and time
> how long it takes to execute.
>
> Signed-off-by: Anton Blanchard <anton at samba.org>
> ---
>  arch/powerpc/mm/fault.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
> index fd6484fc2fa9..3a7d580fdc59 100644
> --- a/arch/powerpc/mm/fault.c
> +++ b/arch/powerpc/mm/fault.c
> @@ -287,7 +287,7 @@ int do_page_fault(struct pt_regs *regs, unsigned  
> long address,
>  	 * can result in fault, which will cause a deadlock when called with
>  	 * mmap_sem held
>  	 */
> -	if (user_mode(regs))
> +	if (!is_exec && user_mode(regs))

Shouldn't it also check 'is_write' ?
If it is a store, is_write should be set, shouldn't it ?

Christophe

>  		store_update_sp = store_updates_sp(regs);
>
>  	if (user_mode(regs))
> --
> 2.11.0