Re: crypto API and GFP_ATOMIC

[Mikulas Patocka]

On Wed, 10 Jun 2020, Herbert Xu wrote:

> On Wed, Jun 10, 2020 at 08:02:23AM -0400, Mikulas Patocka wrote:
> >
> > Yes, fixing the drivers would be the best - but you can hardly find any 
> > person who has all the crypto hardware and who is willing to rewrite all 
> > the drivers for it.
> 
> We don't have to rewrite them straight away.  We could mark the
> known broken ones (or the known working ones) and then dm-crypt
> can allocate only those using the types/mask to crypto_alloc.
> 
> Cheers,

I triaged the drivers in drivers/crypto and unfortunatelly, most of them 
do memory allocation in the encryption routine. Some of the do GFP_KERNEL 
allocation even in the absence of CRYPTO_TFM_REQ_MAY_SLEEP.

I'm sending the patches:

The first patch adds a new flag CRYPTO_ALG_ALLOCATES_MEMORY.

The second patch passes CRYPTO_ALG_ALLOCATES_MEMORY through the crypto API 
stack (please check it - I am not an expert in this area).

The third patch sets CRYPTO_ALG_ALLOCATES_MEMORY on drivers that allocate 
memory in the encrypt/decrypt routine.

The fourth patch fixes the drivers that use GFP_KERNEL in non-blocking 
context.

Mikulas