On Tue, Dec 27, 2016 at 02:49:07PM +0800, Gonglei wrote:
> crypto engine was introduced since 'commit 735d37b5424b ("crypto: engine
> - Introduce the block request crypto engine framework")' which uses work
> queue to realize the asynchronous processing for ablk_cipher and ahash.
>
> For virtio-crypto device, I register an engine for each
> data virtqueue so that we can use the capability of
> multiple data queues in future.
The above got applied as d79b5d0bbf2e.
What's the benefit of this change?
virtio has its own queue for requests. Adding a crypto_engine puts
a queue in front of that. So now there's a queue feeding a queue.
That seems to be a roundabout way of doing things, so I'm wondering
why this change was made? It seems to introduce complexity and
overhead with no apparent benefit.
The reason I'm asking is that I'm splitting sign/verify out of
virtio_crypto_akcipher_algs.c:
https://lore.kernel.org/all/ZscuLueUKl9rcCGr@xxxxxxxxx/
Nowadays sign/verify is no longer asynchronous. However the
crypto_engine indirection forces me to introduce a sig_request
struct which stores the input/output parameters for a sign/verify
operation, so that the crypto_engine can consume it asynchronously.
I'm tempted to instead remove crypto_engine support from
virtio_crypto_core.c to ease migration to synchronous sign/verify.
Thanks,
Lukas
