Re: sparsemem panic in 2.6.17-rc5-mm1 and -mm2

[Andy Whitcroft <apw@shadowen.org>]

Andrew Morton wrote:
> On Wed, 07 Jun 2006 10:26:03 +0100
> Andy Whitcroft <apw@shadowen.org> wrote:
> 
> 
>>>btw Andy, that UNALIGNED_ZONE_BOUNDARIES message is useless.  Only 0.1% of
>>>users even have the knowledge how to recompile their kernel, let alone the
>>>inclination.  Can we do something smarter here?
>>
>>Yes, valid point there.  The overall plan is that this should never come
>>out as the option should be on unless the architecture is ensuring
>>alignment.  Right now the only architecture which is so marked is x86.
>>I wonder if we should also be tainting the kernel at that point so its
>>obvious to 'us' that a kernel has this problem?
> 
> 
> Better to make things just work if we can.
> 
> 
>>The other option is to just turn the check on all the time.  It is two
>>shift and mask + a compare on two cache lines that we definatly are
>>examining anyhow to make the merge checks.
> 
> 
> Sounds OK to me.
> 
> Note that the code can be optimised:
> 
> 	if (page_zone_id(page) != page_zone_id(buddy))
> 
> ...
> 
> static inline int page_zone_id(struct page *page)
> {
> 	return (page->flags >> ZONETABLE_PGSHIFT) & ZONETABLE_MASK;
> }
> 
> We don't need to perform the shift to make that comparison.  If the
> compiler's sufficiently smart it will be able to optimise that for us.
> 
> <checks>
> 
>         shrl    $30, %edx       #, <variable>.flags
>         shrl    $30, %eax       #, <variable>.flags
>         cmpl    %eax, %edx      # <variable>.flags, <variable>.flags
> 
> Nope, not smart enough.

Piece of junk compiler ...   Ok.  I'll put together the minimum check
without the shift and test that.  See if its visible in the performance.

-apw