On 3/12/2020 2:45 PM, Iuliana Prodan wrote:
> On 3/12/2020 5:26 AM, Herbert Xu wrote:
>> On Mon, Mar 09, 2020 at 12:51:32AM +0200, Iuliana Prodan wrote:
>>>
>>> ret = enginectx->op.do_one_request(engine, async_req);
>>> - if (ret) {
>>> - dev_err(engine->dev, "Failed to do one request from queue: %d\n", ret);
>>> - goto req_err;
>>> + can_enq_more = ret;
>>> + if (can_enq_more < 0) {
>>> + dev_err(engine->dev, "Failed to do one request from queue: %d\n",
>>> + ret);
>>> + goto req_err_1;
>>> + }
>>
>> So this now includes the case of the hardware queue being full
>> and the request needs to be queued until space opens up again.
>> In this case, we should not do dev_err. So you need to be able
>> to distinguish between the hardware queue being full vs. a real
>> fatal error on the request (e.g., out-of-memory or some hardware
>> failure).
>>
> There are two aspects here:
> - if all requests go through crypto-engine, and, in this case, if there
> is no space in hw queue, do_one_req returns 0, and actually there will
> be no case of do_one_request() < 0;
> - if there are other requests (non crypto API) that go to hw for
> execution (in CAAM we have split key or RNG) those might occupy the hw
> queue, after crypto-engine returns that it still has space. This case
> wasn't supported before my modifications and neither with these patches.
> This use-case can be solved by retrying to send the request to hw -
> enqueue it back to crypto-engine, in the head of the queue (to keep the
> order, and send it again to hw).
> I've tried this, but it implies modifications in all drivers. For
> example, a driver, in case of error, it frees the resources of the
> request. So, will need to map again a request.
>
I believe this recovery mechanism should be handled in other patch (series).
Herbert, what is your proposal to move forward?
Thanks,
Iulia
