Re: inappropriate use of in_atomic()

[Roland Dreier <roland@topspin.com>]

    > Consequently the use of in_atomic() in the below files is probably
    > deadlocky if CONFIG_PREEMPT=n:
    ...
    > 	drivers/infiniband/core/mad.c

Thanks for pointing this out.  I'll get you a patch in the next day or
two.  As you can probably tell, the code is just trying to decide
whether to use GFP_ATOMIC or GFP_KERNEL to allocate a couple of
things, depending on what context we're being called from.  So at
worst we can just change to GFP_ATOMIC unconditionally.

I'll check into whether we can do something cleverer, but just going
the GFP_ATOMIC route won't be horrible.

Thanks,
  Roland