[PATCH v2 0/2] crypto: engine - support for parallel and batch requests

[PATCH v2 0/2] crypto: engine - support for parallel and batch requests

[Iuliana Prodan]

Added support for executing multiple, independent or not, requests
for crypto engine. This is based on a callback, can_enqueue_more, which
asks the driver if the hardware has free space, to enqueue a new request.
If hardware supports batch requests, crypto-engine can handle this use-case
through do_batch_requests callback.

Since, these new features, cannot be supported by all hardware,
the crypto-engine framework is backward compatible:
- by using the crypto_engine_alloc_init function, to initialize
crypto-engine, the new callbacks are NULL and the engine will work
as before these changes;
- to support only multiple requests, in parallel, the can_enqueue_more
callback must be implemented in driver. On crypto_pump_requests, if
can_enqueue_more callback returns true, a new request is send
to hardware, until there is no space and the callback returns false.
- to support batch requests, do_batch_requests callback must be
implemented in driver, to execute a batch of requests. The link
between the requests, is expected to be done in driver, in do_one_request(). 

Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>

---
Changes since V1:
- changed the name of can_enqueue_hardware callback to can_enqueue_more, and
the argument of this callback to crypto_engine structure (for cases when more
than ore crypto-engine is used).
- added a new patch with support for batch requests.

Changes since V0 (RFC):
- removed max_no_req and no_req, as the number of request that can be
processed in parallel;
- added a new callback, can_enqueue_more, to check whether the hardware
can process a new request.


Iuliana Prodan (2):
  crypto: engine - support for parallel requests
  crypto: engine - support for batch requests

 crypto/crypto_engine.c  | 121 ++++++++++++++++++++++++++++++++----------------
 include/crypto/engine.h |  13 ++++--
 2 files changed, 90 insertions(+), 44 deletions(-)

-- 
2.1.0

[PATCH v2 1/2] crypto: engine - support for parallel requests

[Iuliana Prodan]

Added support for executing multiple requests, in parallel,
for crypto engine.
A new callback is added, can_enqueue_more, which asks the
driver if the hardware has free space, to enqueue a new request.
The new crypto_engine_alloc_init_and_set function, initialize
crypto-engine, sets the maximum size for crypto-engine software
queue (not hardcoded anymore) and the can_enqueue_more callback.
On crypto_pump_requests, if can_enqueue_more callback returns true,
a new request is send to hardware, until there is no space and the
callback returns false.

Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
---
 crypto/crypto_engine.c  | 106 ++++++++++++++++++++++++++++++------------------
 include/crypto/engine.h |  10 +++--
 2 files changed, 72 insertions(+), 44 deletions(-)

diff --git a/crypto/crypto_engine.c b/crypto/crypto_engine.c
index eb029ff..aba934f 100644
--- a/crypto/crypto_engine.c
+++ b/crypto/crypto_engine.c
@@ -22,32 +22,18 @@
  * @err: error number
  */
 static void crypto_finalize_request(struct crypto_engine *engine,
-			     struct crypto_async_request *req, int err)
+				    struct crypto_async_request *req, int err)
 {
-	unsigned long flags;
-	bool finalize_cur_req = false;
 	int ret;
 	struct crypto_engine_ctx *enginectx;
 
-	spin_lock_irqsave(&engine->queue_lock, flags);
-	if (engine->cur_req == req)
-		finalize_cur_req = true;
-	spin_unlock_irqrestore(&engine->queue_lock, flags);
-
-	if (finalize_cur_req) {
-		enginectx = crypto_tfm_ctx(req->tfm);
-		if (engine->cur_req_prepared &&
-		    enginectx->op.unprepare_request) {
-			ret = enginectx->op.unprepare_request(engine, req);
-			if (ret)
-				dev_err(engine->dev, "failed to unprepare request\n");
-		}
-		spin_lock_irqsave(&engine->queue_lock, flags);
-		engine->cur_req = NULL;
-		engine->cur_req_prepared = false;
-		spin_unlock_irqrestore(&engine->queue_lock, flags);
+	enginectx = crypto_tfm_ctx(req->tfm);
+	if (enginectx->op.prepare_request &&
+	    enginectx->op.unprepare_request) {
+		ret = enginectx->op.unprepare_request(engine, req);
+		if (ret)
+			dev_err(engine->dev, "failed to unprepare request\n");
 	}
-
 	req->complete(req, err);
 
 	kthread_queue_work(engine->kworker, &engine->pump_requests);
@@ -73,10 +59,6 @@ static void crypto_pump_requests(struct crypto_engine *engine,
 
 	spin_lock_irqsave(&engine->queue_lock, flags);
 
-	/* Make sure we are not already running a request */
-	if (engine->cur_req)
-		goto out;
-
 	/* If another context is idling then defer */
 	if (engine->idling) {
 		kthread_queue_work(engine->kworker, &engine->pump_requests);
@@ -108,13 +90,18 @@ static void crypto_pump_requests(struct crypto_engine *engine,
 		goto out;
 	}
 
+start_request:
+	/* If hw is busy, do not send any request */
+	if (engine->can_enqueue_more &&
+	    !engine->can_enqueue_more(engine))
+		goto out;
+
 	/* Get the fist request from the engine queue to handle */
 	backlog = crypto_get_backlog(&engine->queue);
 	async_req = crypto_dequeue_request(&engine->queue);
 	if (!async_req)
 		goto out;
 
-	engine->cur_req = async_req;
 	if (backlog)
 		backlog->complete(backlog, -EINPROGRESS);
 
@@ -130,7 +117,7 @@ static void crypto_pump_requests(struct crypto_engine *engine,
 		ret = engine->prepare_crypt_hardware(engine);
 		if (ret) {
 			dev_err(engine->dev, "failed to prepare crypt hardware\n");
-			goto req_err;
+			goto req_err_2;
 		}
 	}
 
@@ -141,26 +128,38 @@ static void crypto_pump_requests(struct crypto_engine *engine,
 		if (ret) {
 			dev_err(engine->dev, "failed to prepare request: %d\n",
 				ret);
-			goto req_err;
+			goto req_err_2;
 		}
-		engine->cur_req_prepared = true;
 	}
 	if (!enginectx->op.do_one_request) {
 		dev_err(engine->dev, "failed to do request\n");
 		ret = -EINVAL;
-		goto req_err;
+		goto req_err_1;
 	}
+
 	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;
+		goto req_err_1;
 	}
-	return;
 
-req_err:
-	crypto_finalize_request(engine, async_req, ret);
-	return;
+	goto retry;
+
+req_err_1:
+	if (enginectx->op.unprepare_request) {
+		ret = enginectx->op.unprepare_request(engine, async_req);
+		if (ret)
+			dev_err(engine->dev, "failed to unprepare request\n");
+	}
+req_err_2:
+	async_req->complete(async_req, ret);
 
+retry:
+	if (engine->can_enqueue_more) {
+		spin_lock_irqsave(&engine->queue_lock, flags);
+		goto start_request;
+	}
+	return;
 out:
 	spin_unlock_irqrestore(&engine->queue_lock, flags);
 }
@@ -386,15 +385,26 @@ int crypto_engine_stop(struct crypto_engine *engine)
 EXPORT_SYMBOL_GPL(crypto_engine_stop);
 
 /**
- * crypto_engine_alloc_init - allocate crypto hardware engine structure and
- * initialize it.
+ * crypto_engine_alloc_init_and_set - allocate crypto hardware engine structure
+ * and initialize it by setting the maximum number of entries in the software
+ * crypto-engine queue.
  * @dev: the device attached with one hardware engine
+ * @cbk_can_enq: pointer to a callback function to be invoked when pumping
+ *               requests to check whether the hardware can process a new
+ *               request.
+ *               This has the form:
+ *               callback(struct crypto_engine *engine)
+ *               where:
+ *               @engine: the crypto engine structure.
  * @rt: whether this queue is set to run as a realtime task
+ * @qlen: maximum size of the crypto-engine queue
  *
  * This must be called from context that can sleep.
  * Return: the crypto engine structure on success, else NULL.
  */
-struct crypto_engine *crypto_engine_alloc_init(struct device *dev, bool rt)
+struct crypto_engine *crypto_engine_alloc_init_and_set(struct device *dev,
+						       bool (*cbk_can_enq)(struct crypto_engine *engine),
+						       bool rt, int qlen)
 {
 	struct sched_param param = { .sched_priority = MAX_RT_PRIO / 2 };
 	struct crypto_engine *engine;
@@ -411,12 +421,12 @@ struct crypto_engine *crypto_engine_alloc_init(struct device *dev, bool rt)
 	engine->running = false;
 	engine->busy = false;
 	engine->idling = false;
-	engine->cur_req_prepared = false;
 	engine->priv_data = dev;
+	engine->can_enqueue_more = cbk_can_enq;
 	snprintf(engine->name, sizeof(engine->name),
 		 "%s-engine", dev_name(dev));
 
-	crypto_init_queue(&engine->queue, CRYPTO_ENGINE_MAX_QLEN);
+	crypto_init_queue(&engine->queue, qlen);
 	spin_lock_init(&engine->queue_lock);
 
 	engine->kworker = kthread_create_worker(0, "%s", engine->name);
@@ -433,6 +443,22 @@ struct crypto_engine *crypto_engine_alloc_init(struct device *dev, bool rt)
 
 	return engine;
 }
+EXPORT_SYMBOL_GPL(crypto_engine_alloc_init_and_set);
+
+/**
+ * crypto_engine_alloc_init - allocate crypto hardware engine structure and
+ * initialize it.
+ * @dev: the device attached with one hardware engine
+ * @rt: whether this queue is set to run as a realtime task
+ *
+ * This must be called from context that can sleep.
+ * Return: the crypto engine structure on success, else NULL.
+ */
+struct crypto_engine *crypto_engine_alloc_init(struct device *dev, bool rt)
+{
+	return crypto_engine_alloc_init_and_set(dev, NULL, rt,
+						CRYPTO_ENGINE_MAX_QLEN);
+}
 EXPORT_SYMBOL_GPL(crypto_engine_alloc_init);
 
 /**
diff --git a/include/crypto/engine.h b/include/crypto/engine.h
index e29cd67..07c3f80 100644
--- a/include/crypto/engine.h
+++ b/include/crypto/engine.h
@@ -24,7 +24,6 @@
  * @idling: the engine is entering idle state
  * @busy: request pump is busy
  * @running: the engine is on working
- * @cur_req_prepared: current request is prepared
  * @list: link with the global crypto engine list
  * @queue_lock: spinlock to syncronise access to request queue
  * @queue: the crypto queue of the engine
@@ -35,17 +34,17 @@
  * @unprepare_crypt_hardware: there are currently no more requests on the
  * queue so the subsystem notifies the driver that it may relax the
  * hardware by issuing this call
+ * @can_enqueue_more: callback to check whether the hardware can process
+ * a new request
  * @kworker: kthread worker struct for request pump
  * @pump_requests: work struct for scheduling work to the request pump
  * @priv_data: the engine private data
- * @cur_req: the current request which is on processing
  */
 struct crypto_engine {
 	char			name[ENGINE_NAME_LEN];
 	bool			idling;
 	bool			busy;
 	bool			running;
-	bool			cur_req_prepared;
 
 	struct list_head	list;
 	spinlock_t		queue_lock;
@@ -56,12 +55,12 @@ struct crypto_engine {
 
 	int (*prepare_crypt_hardware)(struct crypto_engine *engine);
 	int (*unprepare_crypt_hardware)(struct crypto_engine *engine);
+	bool (*can_enqueue_more)(struct crypto_engine *engine);
 
 	struct kthread_worker           *kworker;
 	struct kthread_work             pump_requests;
 
 	void				*priv_data;
-	struct crypto_async_request	*cur_req;
 };
 
 /*
@@ -102,6 +101,9 @@ void crypto_finalize_skcipher_request(struct crypto_engine *engine,
 int crypto_engine_start(struct crypto_engine *engine);
 int crypto_engine_stop(struct crypto_engine *engine);
 struct crypto_engine *crypto_engine_alloc_init(struct device *dev, bool rt);
+struct crypto_engine *crypto_engine_alloc_init_and_set(struct device *dev,
+						       bool (*cbk_can_enq)(struct crypto_engine *engine),
+						       bool rt, int qlen);
 int crypto_engine_exit(struct crypto_engine *engine);
 
 #endif /* _CRYPTO_ENGINE_H */
-- 
2.1.0

[PATCH v2 2/2] crypto: engine - support for batch requests

[Iuliana Prodan]

Added support for batch requests, per crypto engine.
A new callback is added, do_batch_requests, which executes a
batch of requests. This has the crypto_engine structure as argument
(for cases when more than one crypto-engine is used).
The crypto_engine_alloc_init_and_set function, initializes
crypto-engine, and also, sets the do_batch_requests callback.
On crypto_pump_requests, if do_batch_requests callback is
implemented in a driver, this will be executed. The link between
the requests will be done in driver, in do_one_request().

Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
---
 crypto/crypto_engine.c  | 17 ++++++++++++++++-
 include/crypto/engine.h |  3 +++
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/crypto/crypto_engine.c b/crypto/crypto_engine.c
index aba934f..378772e 100644
--- a/crypto/crypto_engine.c
+++ b/crypto/crypto_engine.c
@@ -162,6 +162,12 @@ static void crypto_pump_requests(struct crypto_engine *engine,
 	return;
 out:
 	spin_unlock_irqrestore(&engine->queue_lock, flags);
+	if (engine->do_batch_requests) {
+		ret = engine->do_batch_requests(engine);
+		if (ret)
+			dev_err(engine->dev, "failed to do batch requests: %d\n",
+				ret);
+	}
 }
 
 static void crypto_pump_work(struct kthread_work *work)
@@ -396,6 +402,12 @@ EXPORT_SYMBOL_GPL(crypto_engine_stop);
  *               callback(struct crypto_engine *engine)
  *               where:
  *               @engine: the crypto engine structure.
+ * @cbk_do_batch: pointer to a callback function to be invoked when executing a
+ *                a batch of requests.
+ *                This has the form:
+ *                callback(struct crypto_engine *engine)
+ *                where:
+ *                @engine: the crypto engine structure.
  * @rt: whether this queue is set to run as a realtime task
  * @qlen: maximum size of the crypto-engine queue
  *
@@ -404,6 +416,7 @@ EXPORT_SYMBOL_GPL(crypto_engine_stop);
  */
 struct crypto_engine *crypto_engine_alloc_init_and_set(struct device *dev,
 						       bool (*cbk_can_enq)(struct crypto_engine *engine),
+						       int (*cbk_do_batch)(struct crypto_engine *engine),
 						       bool rt, int qlen)
 {
 	struct sched_param param = { .sched_priority = MAX_RT_PRIO / 2 };
@@ -423,6 +436,8 @@ struct crypto_engine *crypto_engine_alloc_init_and_set(struct device *dev,
 	engine->idling = false;
 	engine->priv_data = dev;
 	engine->can_enqueue_more = cbk_can_enq;
+	engine->do_batch_requests = cbk_do_batch;
+
 	snprintf(engine->name, sizeof(engine->name),
 		 "%s-engine", dev_name(dev));
 
@@ -456,7 +471,7 @@ EXPORT_SYMBOL_GPL(crypto_engine_alloc_init_and_set);
  */
 struct crypto_engine *crypto_engine_alloc_init(struct device *dev, bool rt)
 {
-	return crypto_engine_alloc_init_and_set(dev, NULL, rt,
+	return crypto_engine_alloc_init_and_set(dev, NULL, NULL, rt,
 						CRYPTO_ENGINE_MAX_QLEN);
 }
 EXPORT_SYMBOL_GPL(crypto_engine_alloc_init);
diff --git a/include/crypto/engine.h b/include/crypto/engine.h
index 07c3f80..27cddc4 100644
--- a/include/crypto/engine.h
+++ b/include/crypto/engine.h
@@ -34,6 +34,7 @@
  * @unprepare_crypt_hardware: there are currently no more requests on the
  * queue so the subsystem notifies the driver that it may relax the
  * hardware by issuing this call
+ * @do_batch_requests: execute a batch of requests
  * @can_enqueue_more: callback to check whether the hardware can process
  * a new request
  * @kworker: kthread worker struct for request pump
@@ -55,6 +56,7 @@ struct crypto_engine {
 
 	int (*prepare_crypt_hardware)(struct crypto_engine *engine);
 	int (*unprepare_crypt_hardware)(struct crypto_engine *engine);
+	int (*do_batch_requests)(struct crypto_engine *engine);
 	bool (*can_enqueue_more)(struct crypto_engine *engine);
 
 	struct kthread_worker           *kworker;
@@ -103,6 +105,7 @@ int crypto_engine_stop(struct crypto_engine *engine);
 struct crypto_engine *crypto_engine_alloc_init(struct device *dev, bool rt);
 struct crypto_engine *crypto_engine_alloc_init_and_set(struct device *dev,
 						       bool (*cbk_can_enq)(struct crypto_engine *engine),
+						       int (*cbk_do_batch)(struct crypto_engine *engine),
 						       bool rt, int qlen);
 int crypto_engine_exit(struct crypto_engine *engine);
 
-- 
2.1.0

Re: [PATCH v2 1/2] crypto: engine - support for parallel requests

[Corentin Labbe]

On Tue, Feb 04, 2020 at 02:34:19PM +0200, Iuliana Prodan wrote:
> Added support for executing multiple requests, in parallel,
> for crypto engine.
> A new callback is added, can_enqueue_more, which asks the
> driver if the hardware has free space, to enqueue a new request.
> The new crypto_engine_alloc_init_and_set function, initialize
> crypto-engine, sets the maximum size for crypto-engine software
> queue (not hardcoded anymore) and the can_enqueue_more callback.
> On crypto_pump_requests, if can_enqueue_more callback returns true,
> a new request is send to hardware, until there is no space and the
> callback returns false.
> 
> Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
> ---
>  crypto/crypto_engine.c  | 106 ++++++++++++++++++++++++++++++------------------
>  include/crypto/engine.h |  10 +++--
>  2 files changed, 72 insertions(+), 44 deletions(-)
> 
> diff --git a/crypto/crypto_engine.c b/crypto/crypto_engine.c
> index eb029ff..aba934f 100644
> --- a/crypto/crypto_engine.c
> +++ b/crypto/crypto_engine.c
> @@ -22,32 +22,18 @@
>   * @err: error number
>   */
>  static void crypto_finalize_request(struct crypto_engine *engine,
> -			     struct crypto_async_request *req, int err)
> +				    struct crypto_async_request *req, int err)
>  {
> -	unsigned long flags;
> -	bool finalize_cur_req = false;
>  	int ret;
>  	struct crypto_engine_ctx *enginectx;
>  
> -	spin_lock_irqsave(&engine->queue_lock, flags);
> -	if (engine->cur_req == req)
> -		finalize_cur_req = true;
> -	spin_unlock_irqrestore(&engine->queue_lock, flags);
> -
> -	if (finalize_cur_req) {
> -		enginectx = crypto_tfm_ctx(req->tfm);
> -		if (engine->cur_req_prepared &&
> -		    enginectx->op.unprepare_request) {
> -			ret = enginectx->op.unprepare_request(engine, req);
> -			if (ret)
> -				dev_err(engine->dev, "failed to unprepare request\n");
> -		}
> -		spin_lock_irqsave(&engine->queue_lock, flags);
> -		engine->cur_req = NULL;
> -		engine->cur_req_prepared = false;
> -		spin_unlock_irqrestore(&engine->queue_lock, flags);
> +	enginectx = crypto_tfm_ctx(req->tfm);
> +	if (enginectx->op.prepare_request &&
> +	    enginectx->op.unprepare_request) {
> +		ret = enginectx->op.unprepare_request(engine, req);
> +		if (ret)
> +			dev_err(engine->dev, "failed to unprepare request\n");
>  	}
> -
>  	req->complete(req, err);
>  
>  	kthread_queue_work(engine->kworker, &engine->pump_requests);
> @@ -73,10 +59,6 @@ static void crypto_pump_requests(struct crypto_engine *engine,
>  
>  	spin_lock_irqsave(&engine->queue_lock, flags);
>  
> -	/* Make sure we are not already running a request */
> -	if (engine->cur_req)
> -		goto out;
> -

Hello

Your patch has the same problem than mine reported by Horia.
If a queue has more than one request, a first crypto_pump_requests() will send a request and for drivers which do not block on do_one_request() crypto_pump_requests() will end.
Then another crypto_pump_requests() will fire sending a second request while the driver does not support that.

So we need to replace engine->cur_req by another locking mechanism.
Perhaps the cleaner is to add a "request count" (increased when do_one_request, decreased in crypto_finalize_request)
I know that the early version have that and it was removed, but I do not see any better way.

Regards

Re: [PATCH v2 1/2] crypto: engine - support for parallel requests

[Iuliana Prodan]

On 2/5/2020 9:11 PM, Corentin Labbe wrote:
> On Tue, Feb 04, 2020 at 02:34:19PM +0200, Iuliana Prodan wrote:
>> Added support for executing multiple requests, in parallel,
>> for crypto engine.
>> A new callback is added, can_enqueue_more, which asks the
>> driver if the hardware has free space, to enqueue a new request.
>> The new crypto_engine_alloc_init_and_set function, initialize
>> crypto-engine, sets the maximum size for crypto-engine software
>> queue (not hardcoded anymore) and the can_enqueue_more callback.
>> On crypto_pump_requests, if can_enqueue_more callback returns true,
>> a new request is send to hardware, until there is no space and the
>> callback returns false.
>>
>> Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
>> ---
>>   crypto/crypto_engine.c  | 106 ++++++++++++++++++++++++++++++------------------
>>   include/crypto/engine.h |  10 +++--
>>   2 files changed, 72 insertions(+), 44 deletions(-)
>>
>> diff --git a/crypto/crypto_engine.c b/crypto/crypto_engine.c
>> index eb029ff..aba934f 100644
>> --- a/crypto/crypto_engine.c
>> +++ b/crypto/crypto_engine.c
>> @@ -22,32 +22,18 @@
>>    * @err: error number
>>    */
>>   static void crypto_finalize_request(struct crypto_engine *engine,
>> -			     struct crypto_async_request *req, int err)
>> +				    struct crypto_async_request *req, int err)
>>   {
>> -	unsigned long flags;
>> -	bool finalize_cur_req = false;
>>   	int ret;
>>   	struct crypto_engine_ctx *enginectx;
>>   
>> -	spin_lock_irqsave(&engine->queue_lock, flags);
>> -	if (engine->cur_req == req)
>> -		finalize_cur_req = true;
>> -	spin_unlock_irqrestore(&engine->queue_lock, flags);
>> -
>> -	if (finalize_cur_req) {
>> -		enginectx = crypto_tfm_ctx(req->tfm);
>> -		if (engine->cur_req_prepared &&
>> -		    enginectx->op.unprepare_request) {
>> -			ret = enginectx->op.unprepare_request(engine, req);
>> -			if (ret)
>> -				dev_err(engine->dev, "failed to unprepare request\n");
>> -		}
>> -		spin_lock_irqsave(&engine->queue_lock, flags);
>> -		engine->cur_req = NULL;
>> -		engine->cur_req_prepared = false;
>> -		spin_unlock_irqrestore(&engine->queue_lock, flags);
>> +	enginectx = crypto_tfm_ctx(req->tfm);
>> +	if (enginectx->op.prepare_request &&
>> +	    enginectx->op.unprepare_request) {
>> +		ret = enginectx->op.unprepare_request(engine, req);
>> +		if (ret)
>> +			dev_err(engine->dev, "failed to unprepare request\n");
>>   	}
>> -
>>   	req->complete(req, err);
>>   
>>   	kthread_queue_work(engine->kworker, &engine->pump_requests);
>> @@ -73,10 +59,6 @@ static void crypto_pump_requests(struct crypto_engine *engine,
>>   
>>   	spin_lock_irqsave(&engine->queue_lock, flags);
>>   
>> -	/* Make sure we are not already running a request */
>> -	if (engine->cur_req)
>> -		goto out;
>> -
> 
> Hello
> 
> Your patch has the same problem than mine reported by Horia.
> If a queue has more than one request, a first crypto_pump_requests() will send a request and for drivers which do not block on do_one_request() crypto_pump_requests() will end.
> Then another crypto_pump_requests() will fire sending a second request while the driver does not support that.

> So we need to replace engine->cur_req by another locking mechanism.
> Perhaps the cleaner is to add a "request count" (increased when do_one_request, decreased in crypto_finalize_request)
> I know that the early version have that and it was removed, but I do not see any better way.
> 

The "request count" I've change it to can_enqueue_more, so the hw can 
"answer" if it can enqueue or not.

I'll (re)add the cur_req in crypto-engine.
If the new callback, can_enqueue_more, is not implemented the crypto- 
engine will work as before - will send requests to hardware, one-by-one, 
on crypto_pump_requests, and complete it, on crypto_finalize_request, 
and so on.

Thanks,
Iulia

Re: [PATCH v2 1/2] crypto: engine - support for parallel requests

[Corentin Labbe]

On Fri, Feb 07, 2020 at 11:26:38AM +0000, Iuliana Prodan wrote:
> On 2/5/2020 9:11 PM, Corentin Labbe wrote:
> > On Tue, Feb 04, 2020 at 02:34:19PM +0200, Iuliana Prodan wrote:
> >> Added support for executing multiple requests, in parallel,
> >> for crypto engine.
> >> A new callback is added, can_enqueue_more, which asks the
> >> driver if the hardware has free space, to enqueue a new request.
> >> The new crypto_engine_alloc_init_and_set function, initialize
> >> crypto-engine, sets the maximum size for crypto-engine software
> >> queue (not hardcoded anymore) and the can_enqueue_more callback.
> >> On crypto_pump_requests, if can_enqueue_more callback returns true,
> >> a new request is send to hardware, until there is no space and the
> >> callback returns false.
> >>
> >> Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
> >> ---
> >>   crypto/crypto_engine.c  | 106 ++++++++++++++++++++++++++++++------------------
> >>   include/crypto/engine.h |  10 +++--
> >>   2 files changed, 72 insertions(+), 44 deletions(-)
> >>
> >> diff --git a/crypto/crypto_engine.c b/crypto/crypto_engine.c
> >> index eb029ff..aba934f 100644
> >> --- a/crypto/crypto_engine.c
> >> +++ b/crypto/crypto_engine.c
> >> @@ -22,32 +22,18 @@
> >>    * @err: error number
> >>    */
> >>   static void crypto_finalize_request(struct crypto_engine *engine,
> >> -			     struct crypto_async_request *req, int err)
> >> +				    struct crypto_async_request *req, int err)
> >>   {
> >> -	unsigned long flags;
> >> -	bool finalize_cur_req = false;
> >>   	int ret;
> >>   	struct crypto_engine_ctx *enginectx;
> >>   
> >> -	spin_lock_irqsave(&engine->queue_lock, flags);
> >> -	if (engine->cur_req == req)
> >> -		finalize_cur_req = true;
> >> -	spin_unlock_irqrestore(&engine->queue_lock, flags);
> >> -
> >> -	if (finalize_cur_req) {
> >> -		enginectx = crypto_tfm_ctx(req->tfm);
> >> -		if (engine->cur_req_prepared &&
> >> -		    enginectx->op.unprepare_request) {
> >> -			ret = enginectx->op.unprepare_request(engine, req);
> >> -			if (ret)
> >> -				dev_err(engine->dev, "failed to unprepare request\n");
> >> -		}
> >> -		spin_lock_irqsave(&engine->queue_lock, flags);
> >> -		engine->cur_req = NULL;
> >> -		engine->cur_req_prepared = false;
> >> -		spin_unlock_irqrestore(&engine->queue_lock, flags);
> >> +	enginectx = crypto_tfm_ctx(req->tfm);
> >> +	if (enginectx->op.prepare_request &&
> >> +	    enginectx->op.unprepare_request) {
> >> +		ret = enginectx->op.unprepare_request(engine, req);
> >> +		if (ret)
> >> +			dev_err(engine->dev, "failed to unprepare request\n");
> >>   	}
> >> -
> >>   	req->complete(req, err);
> >>   
> >>   	kthread_queue_work(engine->kworker, &engine->pump_requests);
> >> @@ -73,10 +59,6 @@ static void crypto_pump_requests(struct crypto_engine *engine,
> >>   
> >>   	spin_lock_irqsave(&engine->queue_lock, flags);
> >>   
> >> -	/* Make sure we are not already running a request */
> >> -	if (engine->cur_req)
> >> -		goto out;
> >> -
> > 
> > Hello
> > 
> > Your patch has the same problem than mine reported by Horia.
> > If a queue has more than one request, a first crypto_pump_requests() will send a request and for drivers which do not block on do_one_request() crypto_pump_requests() will end.
> > Then another crypto_pump_requests() will fire sending a second request while the driver does not support that.
> 
> > So we need to replace engine->cur_req by another locking mechanism.
> > Perhaps the cleaner is to add a "request count" (increased when do_one_request, decreased in crypto_finalize_request)
> > I know that the early version have that and it was removed, but I do not see any better way.
> > 
> 
> The "request count" I've change it to can_enqueue_more, so the hw can 
> "answer" if it can enqueue or not.
> 
> I'll (re)add the cur_req in crypto-engine.
> If the new callback, can_enqueue_more, is not implemented the crypto- 
> engine will work as before - will send requests to hardware, one-by-one, 
> on crypto_pump_requests, and complete it, on crypto_finalize_request, 
> and so on.
> 

But if the crypto_engine use can_enqueue_more, cur_req is a lie, so the name should be changed (or this fact need to be heavy documented on each of its occurence).