[PATCH] [RFC] crypto: ccree - fix retry handling in cc_send_sync_request()

[PATCH] [RFC] crypto: ccree - fix retry handling in cc_send_sync_request()

[Geert Uytterhoeven]

If cc_queues_status() indicates that the queue is full,
cc_send_sync_request() should loop and retry.

However, cc_queues_status() returns either 0 (for success), or -ENOSPC
(for queue full), while cc_send_sync_request() checks for real errors by
comparing with -EAGAIN.  Hence -ENOSPC is always considered a real
error, and the code never retries the operation.

Fix this by just removing the check, as cc_queues_status() never returns
any other error value than -ENOSPC.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
I am not 100% sure what was intended originally.
Perhaps the second -ENOSPC error in cc_queues_status() should have been
-EAGAIN?
---
 drivers/crypto/ccree/cc_request_mgr.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/drivers/crypto/ccree/cc_request_mgr.c b/drivers/crypto/ccree/cc_request_mgr.c
index 9d61e6f1247819e2..b2a18122f320b7b2 100644
--- a/drivers/crypto/ccree/cc_request_mgr.c
+++ b/drivers/crypto/ccree/cc_request_mgr.c
@@ -476,10 +476,6 @@ int cc_send_sync_request(struct cc_drvdata *drvdata,
 			break;
 
 		spin_unlock_bh(&mgr->hw_lock);
-		if (rc != -EAGAIN) {
-			cc_pm_put_suspend(dev);
-			return rc;
-		}
 		wait_for_completion_interruptible(&drvdata->hw_queue_avail);
 		reinit_completion(&drvdata->hw_queue_avail);
 	}
-- 
2.17.1

Re: [PATCH] [RFC] crypto: ccree - fix retry handling in cc_send_sync_request()

[Gilad Ben-Yossef]

On Tue, Jan 28, 2020 at 9:09 PM Geert Uytterhoeven
<geert+renesas@glider.be> wrote:
>
> If cc_queues_status() indicates that the queue is full,
> cc_send_sync_request() should loop and retry.
>
> However, cc_queues_status() returns either 0 (for success), or -ENOSPC
> (for queue full), while cc_send_sync_request() checks for real errors by
> comparing with -EAGAIN.  Hence -ENOSPC is always considered a real
> error, and the code never retries the operation.
>
> Fix this by just removing the check, as cc_queues_status() never returns
> any other error value than -ENOSPC.

Thank you for spotting this!

The error is simply checking for the wrong error value.
We should be checking for -ENOSPC!

What this does aims to do is wait for the hardware queue to free up if
we were asked to queue a synchronous request and there was no room in
the hardware queue.
The cc_queue_status() function used to return -EAGAIN in this scenario
and this was missed in the change.

I'm curious as to how you found this - did you run into some problem
and traced it to this?
This can lead to a setkey() failing in very high load situations but I
expect this occurrence to be very rare indeed since cc_queue_status()
already loops several times waiting for the room to be freeed.


Gilad



-- 
Gilad Ben-Yossef
Chief Coffee Drinker

values of β will give rise to dom!

Re: [PATCH] [RFC] crypto: ccree - fix retry handling in cc_send_sync_request()

[Gilad Ben-Yossef]

On Wed, Jan 29, 2020 at 11:11 AM Gilad Ben-Yossef <gilad@benyossef.com> wrote:
>
> On Tue, Jan 28, 2020 at 9:09 PM Geert Uytterhoeven
> <geert+renesas@glider.be> wrote:
> >
> > If cc_queues_status() indicates that the queue is full,
> > cc_send_sync_request() should loop and retry.
> >
> > However, cc_queues_status() returns either 0 (for success), or -ENOSPC
> > (for queue full), while cc_send_sync_request() checks for real errors by
> > comparing with -EAGAIN.  Hence -ENOSPC is always considered a real
> > error, and the code never retries the operation.
> >
> > Fix this by just removing the check, as cc_queues_status() never returns
> > any other error value than -ENOSPC.
>

OK, it took me a minute but I now see what you mean... your patch does
the right thing.

Acked-by: Gilad Ben-Yossef <gilad@benyossef.com>

Gilad

Re: [PATCH] [RFC] crypto: ccree - fix retry handling in cc_send_sync_request()

[Geert Uytterhoeven]

Hi Gilad,

On Wed, Jan 29, 2020 at 10:11 AM Gilad Ben-Yossef <gilad@benyossef.com> wrote:
> On Tue, Jan 28, 2020 at 9:09 PM Geert Uytterhoeven
> <geert+renesas@glider.be> wrote:
> > If cc_queues_status() indicates that the queue is full,
> > cc_send_sync_request() should loop and retry.
> >
> > However, cc_queues_status() returns either 0 (for success), or -ENOSPC
> > (for queue full), while cc_send_sync_request() checks for real errors by
> > comparing with -EAGAIN.  Hence -ENOSPC is always considered a real
> > error, and the code never retries the operation.
> >
> > Fix this by just removing the check, as cc_queues_status() never returns
> > any other error value than -ENOSPC.
>
> Thank you for spotting this!
>
> The error is simply checking for the wrong error value.
> We should be checking for -ENOSPC!
>
> What this does aims to do is wait for the hardware queue to free up if
> we were asked to queue a synchronous request and there was no room in
> the hardware queue.
> The cc_queue_status() function used to return -EAGAIN in this scenario
> and this was missed in the change.
>
> I'm curious as to how you found this - did you run into some problem
> and traced it to this?

I didn't run into a specific problem, but I'm working on cleaning up the driver
a bit.

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds