Re: [PATCH] crypto: sun4i-ss - Fix SHA1 hash on A33-variant with BE CPU

From: Corentin Labbe
Date: Mon Sep 07 2020 - 12:01:55 EST


On Mon, Sep 07, 2020 at 04:24:00PM +1000, Herbert Xu wrote:
> On Sun, Sep 06, 2020 at 04:52:24PM +0800, kernel test robot wrote:
> >
> > >> drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:483:35: sparse: sparse: incorrect type in assignment (different base types) @@ expected unsigned int [assigned] [usertype] v @@ got restricted __le32 [usertype] @@
> > drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:483:35: sparse: expected unsigned int [assigned] [usertype] v
> > drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:483:35: sparse: got restricted __le32 [usertype]
> > drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:485:35: sparse: sparse: incorrect type in assignment (different base types) @@ expected unsigned int [assigned] [usertype] v @@ got restricted __be32 [usertype] @@
> > drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:485:35: sparse: expected unsigned int [assigned] [usertype] v
> > drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:485:35: sparse: got restricted __be32 [usertype]
> > drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:490:27: sparse: sparse: incorrect type in assignment (different base types) @@ expected unsigned int [addressable] [assigned] [usertype] v @@ got restricted __le32 [usertype] @@
> > drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:490:27: sparse: expected unsigned int [addressable] [assigned] [usertype] v
> > drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c:490:27: sparse: got restricted __le32 [usertype]
>
> This appears to be a genuine bug, on big-endian CPUs at least.
>
> ---8<---
> When the hash is written out on the A33 variant, it is incorrectly
> swabbed on big-endian CPUs, when it should simply be written out as
> is because it's already in the right format. This was caught by
> sparse warnings.
>
> Instead of using cpu_to_Xe32 followed by a memcpy, this patch
> converts the final hash write to use put_unaligned instead. This
> simplifies the code and makes the A33 variant handling a lot clearer.
>
> This patch also fixes the incorrect endianness marking on wb,
> although this should have no effect in the genereated code.
>
> Fixes: 1e02e6fbdadb ("crypto: sun4i-ss - add the A33 variant of SS")
> Reported-by: kernel test robot <lkp@xxxxxxxxx>
> Signed-off-by: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
>
> diff --git a/drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c b/drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c
> index dc35edd90034..84f7921de577 100644
> --- a/drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c
> +++ b/drivers/crypto/allwinner/sun4i-ss/sun4i-ss-hash.c
> @@ -9,6 +9,7 @@
> * You could find the datasheet in Documentation/arm/sunxi.rst
> */
> #include "sun4i-ss.h"
> +#include <asm/unaligned.h>
> #include <linux/scatterlist.h>
>
> /* This is a totally arbitrary value */
> @@ -196,7 +197,7 @@ static int sun4i_hash(struct ahash_request *areq)
> struct sg_mapping_iter mi;
> int in_r, err = 0;
> size_t copied = 0;
> - __le32 wb = 0;
> + u32 wb = 0;
>
> dev_dbg(ss->dev, "%s %s bc=%llu len=%u mode=%x wl=%u h0=%0x",
> __func__, crypto_tfm_alg_name(areq->base.tfm),
> @@ -408,7 +409,7 @@ static int sun4i_hash(struct ahash_request *areq)
>
> nbw = op->len - 4 * nwait;
> if (nbw) {
> - wb = cpu_to_le32(*(u32 *)(op->buf + nwait * 4));
> + wb = le32_to_cpup((__le32 *)(op->buf + nwait * 4));
> wb &= GENMASK((nbw * 8) - 1, 0);
>
> op->byte_count += nbw;
> @@ -417,7 +418,7 @@ static int sun4i_hash(struct ahash_request *areq)
>
> /* write the remaining bytes of the nbw buffer */
> wb |= ((1 << 7) << (nbw * 8));
> - bf[j++] = le32_to_cpu(wb);
> + ((__le32 *)bf)[j++] = cpu_to_le32(wb);
>
> /*
> * number of space to pad to obtain 64o minus 8(size) minus 4 (final 1)
> @@ -479,16 +480,16 @@ static int sun4i_hash(struct ahash_request *areq)
> /* Get the hash from the device */
> if (op->mode == SS_OP_SHA1) {
> for (i = 0; i < 5; i++) {
> + v = readl(ss->base + SS_MD0 + i * 4);
> if (ss->variant->sha1_in_be)
> - v = cpu_to_le32(readl(ss->base + SS_MD0 + i * 4));
> + put_unaligned(v, areq->result + i * 4);

The put_unaligned should be _le32.

This fix the modprobe tcrypt fail.