Re: [PATCH] crypto: hisilicon/hpre - don't use GFP_KERNEL to alloc mem during softirq

[shaozhengchao <shaozhengchao@xxxxxxxxxx>]

On 2022/7/19 10:16, Herbert Xu wrote:
> On Tue, Jul 19, 2022 at 10:10:42AM +0800, Zhengchao Shao wrote:
> > @@ -252,7 +252,7 @@ static int hpre_prepare_dma_buf(struct hpre_asym_request *hpre_req,
> >   	if (unlikely(shift < 0))
> >   		return -EINVAL;
> >   
> > -	ptr = dma_alloc_coherent(dev, ctx->key_sz, tmp, GFP_KERNEL);
> > +	ptr = dma_alloc_coherent(dev, ctx->key_sz, tmp, flags);
>
> How about just using GFP_ATOMIC unconditinoally?
>
> Cheers,

Hi Herbert:
	Thank you for your reply. I think differentiate the application 
scenarios of GFP flags, which can be more suitable to allocate memory.

Zhengchao Shao