Re: [patch -mm] i386: use pte_update_defer in ptep_test_and_clear_{dirty,young}

[Zachary Amsden <zach@vmware.com>]

David Rientjes wrote:
> Sure, but what I really like about the patch is that we're only flushing 
> something if !flush_end in the first place.  So we can eliminate any TLB 
> flushing if that VMA didn't need it; that's a change from the current 
> behavior.  And since the most obvious use-case for /proc/pid/clear_refs is 
> in conjunction with /proc/pid/smaps for approximating memory footprint, 
> we'll end up saving TLB flushes because the granularity with which that 
> measurement is taken is usually very fine.
>
> Acked-by: David Rientjes <rientjes@google.com>
>   

I like the patch even better if you still batch the flushes, but keep 
the !flush_end machinery.  If I read it correctly, flush_start stays at 
the lower bound for the whole function, so it is still accurate later.  
And with the flush outside the spinlock, contention time is lower.

Thanks,

Zach