Re: [PATCH] block: blk-mq-sched: Replace GFP_ATOMIC with GFP_KERNEL in blk_mq_sched_assign_ioc

[Jia-Ju Bai]

On 2018/1/25 10:58, Al Viro wrote:
> On Thu, Jan 25, 2018 at 10:46:26AM +0800, Jia-Ju Bai wrote:
> > The function ioc_create_icq here is not called in atomic context.
> > Thus GFP_ATOMIC is not necessary, and it can be replaced with GFP_KERNEL.
> >
> > This is found by a static analysis tool named DCNS written by myself.
> Umm...  Some human-readable analysis would be welcome.  FWIW, I've tried to
> put a proof together, but...
> 	struct blk_mq_ops->timeout = nvme_timeout
> 	nvme_timeout()
> 		nvme_alloc_request()
> 			blk_mq_alloc_request_hctx()
> 				blk_mq_get_request()
> 					blk_mq_sched_assign_ioc()
> ... and while I have not traced the call chain further, the look of that
> function (nvme_timeout()) strongly suggests that it *is* meant to be
> called from bloody atomic context.
>
> "My tool has found that place/put together a proof" is nice, but it
> doesn't replace the proof itself...

Thanks for reply :)

I have checked the given call chain, and find that nvme_dev_disable in 
nvme_timeout calls mutex_lock that can sleep.
Thus, I suppose this call chain is not in atomic context.

Besides, how do you find that "function (nvme_timeout()) strongly 
suggests that it *is* meant to be called from bloody atomic context"?
I check the comments in nvme_timeout, and do not find related description...

By the way, do you mean that I should add "My tool has proved that this 
function is never called in atomic context" in the description?


Thanks,
Jia-Ju Bai