On Thu, Sep 29, 2016 at 04:42:31PM +1000, Nicholas Piggin wrote:
> Take Alpha instead. It's using 32-bit ops.
Hmm, my Alpha docs are on the other machine, but I suppose the problem
is 64bit immediates (which would be a common problem I suppose, those
don't really work well on x86 either).
Yes, that does make it all more tricky than desired.
OTOH maybe this is a good excuse to (finally) sanitize the bitmap API to
use a fixed width. It using 'unsigned long' has been something of a pain
at times.
With that it would be 'obvious' bits need to be part of the same 32bit
word and we can use the normal smp_mb__{before,after}_atomic() again.
Given our bitops really are only about single bits, I can't see the
change making a real performance difference.
OTOH, the non atomic things like weight and ff[sz] do benefit from using
the longer words. Bother..
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>