Re: [PATCH 1/2] ipc/sem.c: Race in sem_lock()

[Manfred Spraul <manfred@colorfullife.com>]

Hi all,

On 09/15/2013 08:09 AM, Mike Galbraith wrote:
> On Sat, 2013-09-14 at 23:34 +0200, Manfred Spraul wrote:
>
>> The bug is probably also present in 3.10 and 3.11, but for these kernels
>> is is probably simpler just to move the test of sma->complex_count after
>> the spin_is_locked() test.
> IMHO, your 6 patch series should go to stable as well.  Scalability is
> still BAD without them.  Now, you've shown the lock split to be buggy.
>
> Logically, the whole thing should be reverted entirely in stable, or
> fixed up properly.
Davidlohr: Are you working on fixing the open issues?

IMHO Mike is right, especially for the 3.10 long-term kernel:
Either everything in ipc/*.c must be reverted or it should be fixed 
properly (i.e.: cherry-pick ipc/*)

I have created bugzilla entries for all issues I'm aware of:

https://bugzilla.kernel.org/show_bug.cgi?id=61351
     I sent a patch yesterday.

https://bugzilla.kernel.org/show_bug.cgi?id=61321
https://bugzilla.kernel.org/show_bug.cgi?id=61331
https://bugzilla.kernel.org/show_bug.cgi?id=61341
https://bugzilla.kernel.org/show_bug.cgi?id=61361
https://bugzilla.kernel.org/show_bug.cgi?id=61371
     No patches for theses 5 bugs.

And: Given these numbers from Mike, I would hate to revert anything:
On 09/15/2013 10:06 AM, Mike Galbraith wrote:
> On Sun, 2013-09-15 at 08:09 +0200, Mike Galbraith wrote:
>
>> Humongous improvements...
> (a couple sem-waitzero numbers)
>
> master:  Cpus 64, interleave 1 delay 0: 10039494796 in 30 secs
> 3.10.10: Cpus 64, interleave 1 delay 0:   129236313 in 30 secs
>
> (rapidly scrolling micro-font bench vs reality disclaimer)
One semop() completed every 3 ns, around 600 cpu ticks per operation.

--
     Manfred