Re: [PATCH] slub: reduce overhead of slub_debug

[Pekka Enberg]

On Thu, Jul 7, 2011 at 10:21 PM, David Miller <davem@xxxxxxxxxxxxx> wrote:
> From: Christoph Lameter <cl@xxxxxxxxx>
> Date: Thu, 7 Jul 2011 14:12:37 -0500 (CDT)
>
>> On Thu, 7 Jul 2011, Pekka Enberg wrote:
>>
>>> On Thu, 7 Jul 2011, Pekka Enberg wrote:
>>> > > Looks good to me. Christoph, David, ?
>>>
>>> On Thu, 2011-07-07 at 13:17 -0500, Christoph Lameter wrote:
>>> > The reason debug code is there is because it is useless overhead typically
>>> > not needed. There is no point in optimizing the code that is not run in
>>> > production environments unless there are gross performance issues that
>>> > make debugging difficult. A performance patch for debugging would have to
>>> > cause significant performance improvements. This patch does not do that
>>> > nor was there such an issue to be addressed in the first place.
>>>
>>> Is there something technically wrong with the patch? Quoting the patch
>>> email:
>>>
>>>   (Compiling some project with different options)
>>>                                  make -j12    make clean
>>>   slub_debug disabled:             1m 27s       1.2 s
>>>   slub_debug enabled:              1m 46s       7.6 s
>>>   slub_debug enabled + this patch: 1m 33s       3.2 s
>>>
>>>   check_bytes still shows up high, but not always at the top.
>>>
>>> That's significant enough speedup for me!
>>
>> Ok. I had a different set of numbers in mind from earlier posts.
>>
>> The benefit here comes from accessing memory in larger (word) chunks
>> instead of byte wise. This is a form of memscan() with inverse matching.
>>
>> Isnt there an asm optimized version that can do this much better (there is
>> one for memscan())? Optimizing this in core code by codeing something as
>> generic as that is not that good since the arch code can deliver better
>> performance and it seems that this is functionality that could be useful
>> elsewhere.
>
> You're being so unreasonable, just let the optimization in, refine it
> with follow-on patches.

I applied the patch. I think a follow up patch that moves the function
to lib/string.c with proper generic name would be in order. Thanks!
--
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/