[PATCH RESEND] crypto: sun8i-ss: do not use sg_dma_len before calling DMA functions

[PATCH RESEND] crypto: sun8i-ss: do not use sg_dma_len before calling DMA functions

[Corentin Labbe]

When testing sun8i-ss with multi_v7_defconfig, all CBC algorithm fail crypto
selftests.
This is strange since on sunxi_defconfig, everything was ok.
The problem was in the IV setup loop which never run because sg_dma_len
was 0.

Fixes: 359e893e8af4 ("crypto: sun8i-ss - rework handling of IV")
Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
---

If someone know why sunxi_defconfig have sg_dma_len() which always works
even with any DMA call not done.

 drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c b/drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c
index 4caf17310e90..ddec1b08d4f6 100644
--- a/drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c
+++ b/drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c
@@ -141,7 +141,7 @@ static int sun8i_ss_setup_ivs(struct skcipher_request *areq)
 
 	/* we need to copy all IVs from source in case DMA is bi-directionnal */
 	while (sg && len) {
-		if (sg_dma_len(sg) == 0) {
+		if (sg->length == 0) {
 			sg = sg_next(sg);
 			continue;
 		}
-- 
2.49.0

Re: [PATCH RESEND] crypto: sun8i-ss: do not use sg_dma_len before calling DMA functions

[Ovidiu Panait]

Hi,

On 4/27/25 2:12 PM, Corentin Labbe wrote:
> When testing sun8i-ss with multi_v7_defconfig, all CBC algorithm fail crypto
> selftests.
> This is strange since on sunxi_defconfig, everything was ok.
> The problem was in the IV setup loop which never run because sg_dma_len
> was 0.
> 
> Fixes: 359e893e8af4 ("crypto: sun8i-ss - rework handling of IV")
> Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
> ---
> 
> If someone know why sunxi_defconfig have sg_dma_len() which always works
> even with any DMA call not done.
> 

It seems that sunxi_defconfig has CONFIG_NEED_SG_DMA_LENGTH disabled, so
sg_dma_len() defaults to sg->length.

From include/linux/scatterlist.h:

/*
 * These macros should be used after a dma_map_sg call has been done
 * to get bus addresses of each of the SG entries and their lengths.
 * You should only work with the number of sg entries dma_map_sg
 * returns, or alternatively stop on the first sg_dma_len(sg) which
 * is 0.
 */
...
#ifdef CONFIG_NEED_SG_DMA_LENGTH
#define sg_dma_len(sg)		((sg)->dma_length)
#else
#define sg_dma_len(sg)		((sg)->length)
#endif


On the other hand, multi_v7_defconfig has CONFIG_NEED_SG_DMA_LENGTH
enabled, so sg_dma_len(sg) defaults to sg->dma_length and it would need
to be used after calling dma_map_sg() (as indicated in the comment from
include/linux/scatterlist.h).


Ovidiu

>  drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c b/drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c
> index 4caf17310e90..ddec1b08d4f6 100644
> --- a/drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c
> +++ b/drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c
> @@ -141,7 +141,7 @@ static int sun8i_ss_setup_ivs(struct skcipher_request *areq)
>  
>  	/* we need to copy all IVs from source in case DMA is bi-directionnal */
>  	while (sg && len) {
> -		if (sg_dma_len(sg) == 0) {
> +		if (sg->length == 0) {
>  			sg = sg_next(sg);
>  			continue;
>  		}

Re: [PATCH RESEND] crypto: sun8i-ss: do not use sg_dma_len before calling DMA functions

[Herbert Xu]

On Sun, Apr 27, 2025 at 09:23:40PM +0300, Ovidiu Panait wrote:
> 
> On the other hand, multi_v7_defconfig has CONFIG_NEED_SG_DMA_LENGTH
> enabled, so sg_dma_len(sg) defaults to sg->dma_length and it would need
> to be used after calling dma_map_sg() (as indicated in the comment from
> include/linux/scatterlist.h).

In general, only use sg_dma_len if you are mapping an entire
SG list with dma_map_sg.

If you're using dma_map_single, do not use sg_dma_len.

This is because dma_map_sg will try to merge entries together,
thus creating an sg_dma_len that is bigger than the original
(and reducing the SG list length accordingly).

OTOH dma_map_single can't possibly do that.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH RESEND] crypto: sun8i-ss: do not use sg_dma_len before calling DMA functions

[Herbert Xu]

On Sun, Apr 27, 2025 at 01:12:36PM +0200, Corentin Labbe wrote:
> When testing sun8i-ss with multi_v7_defconfig, all CBC algorithm fail crypto
> selftests.
> This is strange since on sunxi_defconfig, everything was ok.
> The problem was in the IV setup loop which never run because sg_dma_len
> was 0.
> 
> Fixes: 359e893e8af4 ("crypto: sun8i-ss - rework handling of IV")
> Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
> ---
> 
> If someone know why sunxi_defconfig have sg_dma_len() which always works
> even with any DMA call not done.
> 
>  drivers/crypto/allwinner/sun8i-ss/sun8i-ss-cipher.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

Patch applied.  Thanks.
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt