Re: [PATCH] mm: don't invoke __alloc_pages_direct_compact when order 0

[JoonSoo Kim]

2012/7/9 David Rientjes <rientjes@xxxxxxxxxx>:
> On Sun, 8 Jul 2012, JoonSoo Kim wrote:
>
>> >> __alloc_pages_direct_compact has many arguments so invoking it is very costly.
>> >> And in almost invoking case, order is 0, so return immediately.
>> >>
>> >
>> > If "zero cost" is "very costly", then this might make sense.
>> >
>> > __alloc_pages_direct_compact() is inlined by gcc.
>>
>> In my kernel image, __alloc_pages_direct_compact() is not inlined by gcc.
>
> Adding Andrew and Mel to the thread since this would require that we
> revert 11e33f6a55ed ("page allocator: break up the allocator entry point
> into fast and slow paths") which would obviously not be a clean revert
> since there have been several changes to these functions over the past
> three years.

Only "__alloc_pages_direct_compact()" is not inlined.
All others (__alloc_pages_high_priority, __alloc_pages_direct_reclaim,
...) are inlined correctly.
So revert is not needed.

I think __alloc_pages_direct_compact() can't be inlined by gcc,
because it is so big and is invoked two times in __alloc_pages_nodemask().

> I'm stunned (and skeptical) that __alloc_pages_direct_compact() is not
> inlined by your gcc, especially since the kernel must be compiled with
> optimization (either -O1 or -O2 which causes these functions to be
> inlined).  What version of gcc are you using and on what architecture?
> Please do "make mm/page_alloc.s" and send it to me privately, I'll file
> this and fix it up on gcc-bugs.

I will send result of "make mm/page_alloc.s" to you privately.
My environments is "x86_64, GNU C version 4.6.3"

> I'll definitely be following up on this.

Thanks for comments.
--
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/