On Wed, 2008-09-03 at 15:41 +0200, Sebastien Dugue wrote:

On Wed, 20 Aug 2008 15:23:01 +1000 Benjamin Herrenschmidt [off-list ref] wrote:

quoted

BTW. It would be good to try to turn the GFP_ATOMIC into GFP_KERNEL,

  That would be nice indeed

quoted

maybe using a semaphore instead of a lock to protect insertion vs.
initialisation.

  a semaphore? are you meaning a mutex? If not, I fail to understand what you're
implying.

Right, a mutex, bad habit calling those semaphores from the old days :-)

  Right, that's the problem with this new scheme and I'm still trying
to find a way to handle memory allocation failures be it for GFP_ATOMIC or
GFP_KERNEL.

  I could not think of anything simple so far and I'm open for suggestions.

GFP_KERNEL should not fail, it will just block no ? If it fails, it's
probably catastrophic enough not to care. You can always fallback to
linear lookup. I don't know if it's worth trying to fire off a new
allocation attempt later, probably not.

Ben.