Re: [PATCH] Scalable page cache

[Andrew Morton]

"David S. Miller" wrote:
> 
>    From: Ingo Molnar <[e-mail]>
>    Date: Mon, 26 Nov 2001 22:09:02 +0100 (CET)
> 
>    something like this:
> 
>    - #define SetPageReferenced(page)    set_bit(PG_referenced, &(page)->flags)
>    + #define SetPageReferenced(page) \
>    +    if (!test_bit(PG_referenced), &(page)->flags) \
>    +            set_bit(PG_referenced, &(page)->flags)
> 
> On some platforms all the {test_,}{clear,change,set}_bit() ops give
> you this cache behavior.  How hard would it be to make ix86 act
> similarly?

It would be nice to have a full complement of non-atomic
bitops.  At present we have __set_bit, but no __clear_bit, etc.

So we often do buslocked RMW's in places where it's not needed.

-