On 2021/1/21 21:36, Jason Gunthorpe wrote:
On Thu, Jan 21, 2021 at 01:33:42PM +0000, liweihang wrote:
We need to allocate memory while spin_lock is hold, how about using GFP_KERNEL or GFP_NOWAIT?
You should try hard not to do that. Convert he spinlock to a mutex, for instance.
Jason
But what if some kernel users call ib_post_send() when holding a spinlock?
Thanks Weihang