Re: [PATCH v3 1/2] crypto: engine - support for parallel requests

From: Iuliana Prodan
Date: Thu Feb 13 2020 - 20:25:57 EST


On 2/13/2020 8:18 AM, Herbert Xu wrote:
> On Fri, Feb 07, 2020 at 02:36:13PM +0200, Iuliana Prodan wrote:
>>
>> +start_request:
>> + /* If hardware is busy, do not send any request */
>> + if (engine->can_enqueue_more) {
>> + if (!engine->can_enqueue_more(engine))
>> + goto out;
>
> Instead of a driver callback I'd rather the driver called into
> the engine telling it to stop/start, similar to how net drivers
> work.
>

Given your suggestion, I’m thinking of implementing do_one_request, in
the driver, to return -IN_PROGRESS if the hw can enqueue more and -EBUSY
if otherwise (solution 1). But, this implies to update all the drivers
that use crypto-engine (something I wouldn’t mind doing, but I don’t
have the hw to test it).

My current proposal keeps the backward compatibility of crypto-engine,
so no need to change the other drivers. If they want to use multiple
requests or batch requests, they can implement can_enqueue_more,
respective do_batch_requests (solution 2).

Please, let me know how do you want me to proceed? Solution 1 or
solution 2, or… maybe I’ve missed something?

Thanks,
Iulia