Re: [PATCH V6 2/2 RESEND] ksm: replace jhash2 with faster hash

[Mike Rapoport <rppt@linux.vnet.ibm.com>]

Hi,

On Thu, May 24, 2018 at 11:01:20AM +0300, Timofey Titovets wrote:
> N?N?, 23 D 1/4 D?N? 2018 D3. D2 17:24, Pavel Tatashin <pasha.tatashin@oracle.com>:
> 
> > Hi Timofey,

[ ... ]
 
> > It really feels wrong to keep  choice_fastest_hash() in fasthash(), it is
> > done only once and really belongs to the init function, like ksm_init().
> As
> 
> That possible to move decision from lazy load, to ksm_thread,
> that will allow us to start bench and not slowdown boot.
> 
> But for that to works, ksm must start later, after init of crypto.
 
What about moving choice_fastest_hash() to run_store()?

KSM anyway starts with ksm_run = KSM_RUN_STOP and does not scan until
userspace writes !0 to /sys/kernel/mm/ksm/run.

Selection of the hash function when KSM is actually enabled seems quite
appropriate...

> > I understand, you think it is a bad idea to keep it in ksm_init() because
> > it slows down boot by 0.25s, which I agree with your is substantial. But,
> I
> > really do not think that we should spend those 0.25s at all deciding what
> > hash function is optimal, and instead default to one or another during
> boot
> > based on hardware we are booting on. If crc32c without hw acceleration is
> > no worse than jhash2, maybe we should simply switch to  crc32c?
> 
> crc32c with no hw, are slower in compare to jhash2 on x86, so i think on
> other arches result will be same.
> 
> > Thank you,
> > Pavel
> 
> Thanks.
> 
> --
> Have a nice day,
> Timofey.
> 

-- 
Sincerely yours,
Mike.