On Thu, 26 Jul 2012 21:42:50 +0200
Peter Zijlstra [off-list ref] wrote:

On Thu, 2012-07-26 at 17:47 +0200, Martin Schwidefsky wrote:

quoted

A code review revealed another potential race in regard to TLB flushing
on s390. See patch #2 for the ugly details. To fix this I would like
to use the arch_enter_lazy_mmu_mode/arch_leave_lazy_mmu_mode but to do
that the pointer to the mm in question needs to be added to the functions.
To keep things symmetrical arch_flush_lazy_mmu_mode should grow an mm
argument as well.

powerpc 

I have a patch that makes sparc64 do the same thing.

That is good, I guess we are in agreement then to add the mm argument.
 

quoted

and x86 have a non-empty implementation for the lazy mmu flush
primitives and tile calls the generic definition in the architecture
files (which is a bit strange because the generic definition is empty).
Comments?

argh.. you're making my head hurt.

Fun, isn't it ?

I guess my first question is where is lazy_mmu_mode active crossing an
mm? I thought it was only ever held across operations on a single mm.

My take is never, it is only ever used in a single mm.

The second question would be if you could use that detach_mm thing I
proposed a while back ( http://marc.info/?l=linux-mm&m=134090072917840 )
or can we rework the active_mm magic in general to make all this easier?

No, that is not good enough. The issue I'm trying to fix is for a multi-
threaded application where the same mm is attached to multiple cpus. To
detach it on the local cpu won't help, it would have to be detached
everywhere.

Your 2/2 patch makes me shiver..

 
Ask me about it :-/

-- 
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>