On 2018/1/25 11:34, Jens Axboe wrote:

On 1/24/18 7:46 PM, Jia-Ju Bai wrote:

quoted

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.

But it's running off the IO submission path, so by definition the GFP
mask cannot include anything that will do IO. GFP_KERNEL will make
it deadlock prone.

It could be GFP_NOIO, but that's also overlooking the fact that we can
have preemption disabled here.

On top of all that, we want something quick here, and it's OK that
it fails. That's preferable to blocking. So we want an atomic alloc,
even if we could tolerate a blocking one.

So I think you need to fix your static analysis tool, it's missing
a few key things.

Okay, thanks for your advice.


Thanks,
Jia-Ju Bai