Re: REGRESSION: Performance regressions from switchinganon_vma->lock to mutex

[Peter Zijlstra]

On Wed, 2011-06-15 at 09:26 +0800, Shaohua Li wrote:

> On Wed, 2011-06-15 at 08:29 +0800, Tim Chen wrote:
> >          + 7.30% anon_vma_clone_batch

> what are you testing? I didn't see Andi's batch anon->lock for fork
> patches are merged in 2.6.39. 

Good spot that certainly isn't plain .39.

It looks like those (http://marc.info/?l=linux-mm&m=130533041726258) are
similar to Linus' patch, except Linus takes the hard line that the root
lock should stay the same. Let me try Linus' patch first to see if this
workload can trigger his WARN.

/me mutters something about patches in attachments and rebuilds.

OK, the WARN doesn't trigger, but it also doesn't improve things (quite
the opposite in fact):

-tip            260.092 messages/sec/core
    +sirq-rcu   271.078 messages/sec/core
    +linus      262.435 messages/sec/core

So Linus' patch makes the throughput drop from 271 to 262, weird.

/me goes re-test without the sirq-rcu bits mixed in just to make sure.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/