Re: [PATCH] x86/tlb: replace INVALIDATE_TLB_VECTOR by CALL_FUNCTION_VECTOR

[Alex Shi]

On 05/18/2012 11:12 PM, Alex Shi wrote:

> There are only 32 INVALIDATE_TLB_VECTOR now in kernel.
> but modern x86 sever has more cpu number. That causes lock
> contention in TLB flushing.
> 
> Now, useing generic smp call function to replace it.
> In the NHM EX machine 4P * 8cores * HT = 64 CPUs, hackbench pthread
> has 3% performance increase.
> And no clear performance changes on NHM EP(16CPUs), WSM EP(23CPU)
> and SNB EP(32CPU) machines.
> 
> This patch is base on my tlb flush range support patchset.
> 
> Signed-off-by: Alex Shi <alex.shi@xxxxxxxxx>



refresh this patch will more clean up, and cc to PeterZ.
Actually, I tried to keep flush_tlb_info into dedicated cache line, but did find clear help.

Comments are appreciated!

---