Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1630289 > unrolled thread

[PATCH] crypto: tcrypt: don't disable irqs and wait

Back to article view | Back to linux.kernel

Contents

  [PATCH] crypto: tcrypt: don't disable irqs and wait Gilad Ben-Yossef <gilad@benyossef.com> - 2017-04-25 10:30 +0200
    Re: [PATCH] crypto: tcrypt: don't disable irqs and wait Horia Geantă <horia.geanta@nxp.com> - 2017-04-26 14:10 +0200

#1630289 — [PATCH] crypto: tcrypt: don't disable irqs and wait

The tcrypt AEAD cycles speed tests disables irqs during the test, which is
broken at the very least since commit
'1425d2d17f7309c6 ("crypto: tcrypt - Fix AEAD speed tests")'
adds a wait for completion as part of the test and probably since
switching to the new AEAD API.

While the result of taking a cycle count diff may not mean much on SMP
systems if the task migrates, it's good enough for tcrypt being the quick
& dirty dev tool it is. It's also what all the other (i.e. hash) cycle
speed tests do.

Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Reported-by: Ofir Drang <ofir.drang@arm.com>
---
 crypto/tcrypt.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/crypto/tcrypt.c b/crypto/tcrypt.c
index 9a11f3c..0dd6a43 100644
--- a/crypto/tcrypt.c
+++ b/crypto/tcrypt.c
@@ -138,8 +138,6 @@ static int test_aead_cycles(struct aead_request *req, int enc, int blen)
 	int ret = 0;
 	int i;
 
-	local_irq_disable();
-
 	/* Warm-up run. */
 	for (i = 0; i < 4; i++) {
 		if (enc)
@@ -169,8 +167,6 @@ static int test_aead_cycles(struct aead_request *req, int enc, int blen)
 	}
 
 out:
-	local_irq_enable();
-
 	if (ret == 0)
 		printk("1 operation in %lu cycles (%d bytes)\n",
 		       (cycles + 4) / 8, blen);
-- 
2.1.4

[toc] | [next] | [standalone]

#1631419

On 4/25/2017 11:25 AM, Gilad Ben-Yossef wrote:
> The tcrypt AEAD cycles speed tests disables irqs during the test, which is
> broken at the very least since commit
> '1425d2d17f7309c6 ("crypto: tcrypt - Fix AEAD speed tests")'
> adds a wait for completion as part of the test and probably since
> switching to the new AEAD API.
> 
> While the result of taking a cycle count diff may not mean much on SMP
> systems if the task migrates, it's good enough for tcrypt being the quick
> & dirty dev tool it is. It's also what all the other (i.e. hash) cycle
> speed tests do.
> 
> Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
> Reported-by: Ofir Drang <ofir.drang@arm.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>

I've also noticed this in CAAM accelerator - spin_unlock_bh() called
from caam_jr_enqueue() complains due to disabled interrupts:
WARNING: CPU: 4 PID: 2585 at kernel/softirq.c:161
.__local_bh_enable_ip+0x70/0xcc

Thanks,
Horia

> ---
>  crypto/tcrypt.c | 4 ----
>  1 file changed, 4 deletions(-)
> 
> diff --git a/crypto/tcrypt.c b/crypto/tcrypt.c
> index 9a11f3c..0dd6a43 100644
> --- a/crypto/tcrypt.c
> +++ b/crypto/tcrypt.c
> @@ -138,8 +138,6 @@ static int test_aead_cycles(struct aead_request *req, int enc, int blen)
>  	int ret = 0;
>  	int i;
>  
> -	local_irq_disable();
> -
>  	/* Warm-up run. */
>  	for (i = 0; i < 4; i++) {
>  		if (enc)
> @@ -169,8 +167,6 @@ static int test_aead_cycles(struct aead_request *req, int enc, int blen)
>  	}
>  
>  out:
> -	local_irq_enable();
> -
>  	if (ret == 0)
>  		printk("1 operation in %lu cycles (%d bytes)\n",
>  		       (cycles + 4) / 8, blen);
> 

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web