[1/9] crypto: engine: workqueue can only be processed one by one
diff mbox series

Message ID 20200122104528.30084-2-clabbe.montjoie@gmail.com
State New
Headers show
Series
  • crypto: engine: permit to handle multiple requests
Related show

Commit Message

Corentin Labbe Jan. 22, 2020, 10:45 a.m. UTC
Some bykeshedding are unnecessary since a workqueue can only be executed
one by one.
This behaviour is documented in:
- kernel/kthread.c: comment of kthread_worker_fn()
- Documentation/core-api/workqueue.rst: the functions associated with the work items one after the other

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
---
 crypto/crypto_engine.c  | 13 -------------
 include/crypto/engine.h |  2 --
 2 files changed, 15 deletions(-)

Comments

Horia Geantă Jan. 28, 2020, 3:50 p.m. UTC | #1
On 1/22/2020 12:46 PM, Corentin Labbe wrote:
> Some bykeshedding are unnecessary since a workqueue can only be executed
> one by one.
> This behaviour is documented in:
> - kernel/kthread.c: comment of kthread_worker_fn()
> - Documentation/core-api/workqueue.rst: the functions associated with the work items one after the other
[...]
> @@ -73,16 +73,6 @@ 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;
> -
This check is here for a good reason, namely because crypto engine
cannot currently handle multiple crypto requests being in "flight"
in parallel.

More exactly, if this check is removed the following sequence could occur:
crypto_pump_work() -> crypto_pump_requests() -> .do_one_request(areq1)
crypto_pump_work() -> crypto_pump_requests() -> .do_one_request(areq2)
crypto_finalize_request(areq1)
crypto_finalize_request(areq2)

This isn't correctly handled in crypto_finalize_request(),
since .unprepare_request will be called only for areq2.

/**
 * crypto_finalize_request - finalize one request if the request is done
 * @engine: the hardware engine
 * @req: the request need to be finalized
 * @err: error number
 */
static void crypto_finalize_request(struct crypto_engine *engine,
			     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);
	}

	req->complete(req, err);

	kthread_queue_work(engine->kworker, &engine->pump_requests);
}

Horia
Corentin Labbe Jan. 28, 2020, 3:58 p.m. UTC | #2
On Tue, Jan 28, 2020 at 03:50:14PM +0000, Horia Geanta wrote:
> On 1/22/2020 12:46 PM, Corentin Labbe wrote:
> > Some bykeshedding are unnecessary since a workqueue can only be executed
> > one by one.
> > This behaviour is documented in:
> > - kernel/kthread.c: comment of kthread_worker_fn()
> > - Documentation/core-api/workqueue.rst: the functions associated with the work items one after the other
> [...]
> > @@ -73,16 +73,6 @@ 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;
> > -
> This check is here for a good reason, namely because crypto engine
> cannot currently handle multiple crypto requests being in "flight"
> in parallel.
> 
> More exactly, if this check is removed the following sequence could occur:
> crypto_pump_work() -> crypto_pump_requests() -> .do_one_request(areq1)
> crypto_pump_work() -> crypto_pump_requests() -> .do_one_request(areq2)
> crypto_finalize_request(areq1)
> crypto_finalize_request(areq2)
> 

As explained in the commitlog, crypto_pump_work() cannot be ran twice.

Regards
Corentin Labbe Jan. 28, 2020, 4:55 p.m. UTC | #3
On Tue, Jan 28, 2020 at 04:58:00PM +0100, Corentin Labbe wrote:
> On Tue, Jan 28, 2020 at 03:50:14PM +0000, Horia Geanta wrote:
> > On 1/22/2020 12:46 PM, Corentin Labbe wrote:
> > > Some bykeshedding are unnecessary since a workqueue can only be executed
> > > one by one.
> > > This behaviour is documented in:
> > > - kernel/kthread.c: comment of kthread_worker_fn()
> > > - Documentation/core-api/workqueue.rst: the functions associated with the work items one after the other
> > [...]
> > > @@ -73,16 +73,6 @@ 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;
> > > -
> > This check is here for a good reason, namely because crypto engine
> > cannot currently handle multiple crypto requests being in "flight"
> > in parallel.
> > 
> > More exactly, if this check is removed the following sequence could occur:
> > crypto_pump_work() -> crypto_pump_requests() -> .do_one_request(areq1)
> > crypto_pump_work() -> crypto_pump_requests() -> .do_one_request(areq2)
> > crypto_finalize_request(areq1)
> > crypto_finalize_request(areq2)
> > 
> 
> As explained in the commitlog, crypto_pump_work() cannot be ran twice.
> 

Sorry, I have misunderstood and wrongly answered.

Right since some driver does not block on do_one_request(), crypto_pump_work() can be ran one after one and so launch two request.

So this patch is bad.

Regards

Patch
diff mbox series

diff --git a/crypto/crypto_engine.c b/crypto/crypto_engine.c
index eb029ff1e05a..feb0d82dd454 100644
--- a/crypto/crypto_engine.c
+++ b/crypto/crypto_engine.c
@@ -73,16 +73,6 @@  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);
-		goto out;
-	}
-
 	/* Check if the engine queue is idle */
 	if (!crypto_queue_len(&engine->queue) || !engine->running) {
 		if (!engine->busy)
@@ -96,7 +86,6 @@  static void crypto_pump_requests(struct crypto_engine *engine,
 		}
 
 		engine->busy = false;
-		engine->idling = true;
 		spin_unlock_irqrestore(&engine->queue_lock, flags);
 
 		if (engine->unprepare_crypt_hardware &&
@@ -104,7 +93,6 @@  static void crypto_pump_requests(struct crypto_engine *engine,
 			dev_err(engine->dev, "failed to unprepare crypt hardware\n");
 
 		spin_lock_irqsave(&engine->queue_lock, flags);
-		engine->idling = false;
 		goto out;
 	}
 
@@ -410,7 +398,6 @@  struct crypto_engine *crypto_engine_alloc_init(struct device *dev, bool rt)
 	engine->rt = rt;
 	engine->running = false;
 	engine->busy = false;
-	engine->idling = false;
 	engine->cur_req_prepared = false;
 	engine->priv_data = dev;
 	snprintf(engine->name, sizeof(engine->name),
diff --git a/include/crypto/engine.h b/include/crypto/engine.h
index e29cd67f93c7..7e7cbd9ca3b5 100644
--- a/include/crypto/engine.h
+++ b/include/crypto/engine.h
@@ -21,7 +21,6 @@ 
 /*
  * struct crypto_engine - crypto hardware engine
  * @name: the engine name
- * @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
@@ -42,7 +41,6 @@ 
  */
 struct crypto_engine {
 	char			name[ENGINE_NAME_LEN];
-	bool			idling;
 	bool			busy;
 	bool			running;
 	bool			cur_req_prepared;