[openib-general] [RFC] patch for new send MAD allocation routines

[Roland Dreier]

Instead of hard-coding GFP_KERNEL:

    > +	buf = kmalloc(sizeof *send_buf + buf_size, GFP_KERNEL);

would it make sense to put a gfp_mask into the API and avoid some of
the heartburn about interrupt context that we've had lately?

 - R.