Re: [patch 2/2] reduce tlb/cache flush times of agpgart memory allocation

["H. Peter Anvin" <hpa@zytor.com>]

Ingo Molnar wrote:
> * Li, Shaohua <shaohua.li@intel.com> wrote:
> 
>>> it would be great if you had time to update his patch and this to 
>>> it...
>> I'll do it soon.
> 
> great! Please do it as a delta patch against tip/master:
> 
>    http://people.redhat.com/mingo/tip.git/README
> 
> as your first two patches are being tested already:
> 
>  466ae83: reduce tlb/cache flush times of agpgart memory allocation
>  1ac2f7d: introduce two APIs for page attribute
> 
> [ but it can all only go upstream once the observations from Arjan have
>   been addressed. ]
> 
>>> and the logic probably should be "if there's more than X pags in the 
>>> the array, just use wbinvd". Although wbinvd is very painful if you 
>>> have 12Mb of cache and you wipe it for all cores in the system ;-(
>> This might not be that bad, changing attribute isn't frequently used.
> 
> dont some X/DRM drivers do it for a wide range of GX ops?
> 

I think so... at least it's likely to become more common. 
Realistically, it means WBINVD, being uninterruptible, is probably 
unsafe even for very large amounts.

	-hpa