Re: [PATCH RFC 06/10] crypto: engine: introduce ct

On Thu, Jan 16, 2020 at 11:34:19AM +0000, Iuliana Prodan wrote:

On 1/14/2020 4:00 PM, Corentin Labbe wrote:

quoted

We will store the number of request in a batch in engine->ct.
This patch adds all loop to unprepare all requests of a batch.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
---
  crypto/crypto_engine.c  | 30 ++++++++++++++++++------------
  include/crypto/engine.h |  2 ++
  2 files changed, 20 insertions(+), 12 deletions(-)

diff --git a/crypto/crypto_engine.c b/crypto/crypto_engine.c
index b72873550587..591dea5ddeec 100644
--- a/crypto/crypto_engine.c
+++ b/crypto/crypto_engine.c

@@ -28,6 +28,7 @@ static void crypto_finalize_request(struct crypto_engine *engine,
  	bool finalize_cur_req = false;
  	int ret;
  	struct crypto_engine_ctx *enginectx;
+	int i = 0;
  
  	spin_lock_irqsave(&engine->queue_lock, flags);
  	if (engine->cur_reqs[0].req == req)

You're checking here just the first request, but do the completion for 
all? Why? Shouldn't we check for each request if it was done by hw or not?

The first request is a sort of key for the whole batch.

I've also seen that the do_one_request is called only on the first 
request, from the batch.

Since the request are linked, this is not a problem.
But I miss this explanaition in the code.

In your driver you do the prepare/unprepare for the whole batch at once, 
but not all drivers, who uses crypto-engine, are doing this (see virtio, 
amlogic, stm32). And I don't know if they can...

prepare is optionnal, and unprepare is optional even if prepare is done.
Furthermore, doing prepare/unprepare is optional per request.
I have tested this serie on sun8i-ss and amlogic which dont use prepare/unprepare.


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel