Re: [PATCH] Revert "sh: Handle calling csum_partial with misaligned data"

On Sun, 2024-03-24 at 16:18 -0700, Guenter Roeck wrote:
> This reverts commit cadc4e1a2b4d20d0cc0e81f2c6ba0588775e54e5.
> 
> Commit cadc4e1a2b4d ("sh: Handle calling csum_partial with misaligned
> data") causes bad checksum calculations on unaligned data. Reverting
> it fixes the problem.
> 
>     # Subtest: checksum
>     # module: checksum_kunit
>     1..5
>     # test_csum_fixed_random_inputs: ASSERTION FAILED at lib/checksum_kunit.c:500
>     Expected ( u64)result == ( u64)expec, but
>         ( u64)result == 53378 (0xd082)
>         ( u64)expec == 33488 (0x82d0)
>     # test_csum_fixed_random_inputs: pass:0 fail:1 skip:0 total:1
>     not ok 1 test_csum_fixed_random_inputs
>     # test_csum_all_carry_inputs: ASSERTION FAILED at lib/checksum_kunit.c:525
>     Expected ( u64)result == ( u64)expec, but
>         ( u64)result == 65281 (0xff01)
>         ( u64)expec == 65280 (0xff00)
>     # test_csum_all_carry_inputs: pass:0 fail:1 skip:0 total:1
>     not ok 2 test_csum_all_carry_inputs
>     # test_csum_no_carry_inputs: ASSERTION FAILED at lib/checksum_kunit.c:573
>     Expected ( u64)result == ( u64)expec, but
>         ( u64)result == 65535 (0xffff)
>         ( u64)expec == 65534 (0xfffe)
>     # test_csum_no_carry_inputs: pass:0 fail:1 skip:0 total:1
>     not ok 3 test_csum_no_carry_inputs
>     # test_ip_fast_csum: pass:1 fail:0 skip:0 total:1
>     ok 4 test_ip_fast_csum
>     # test_csum_ipv6_magic: pass:1 fail:0 skip:0 total:1
>     ok 5 test_csum_ipv6_magic
>  # checksum: pass:2 fail:3 skip:0 total:5
>  # Totals: pass:2 fail:3 skip:0 total:5
> not ok 22 checksum
> 
> Fixes: cadc4e1a2b4d ("sh: Handle calling csum_partial with misaligned data")
> Signed-off-by: Guenter Roeck <[email protected]>
> ---
>  arch/sh/lib/checksum.S | 67 ++++++++++++------------------------------
>  1 file changed, 18 insertions(+), 49 deletions(-)
> 
> diff --git a/arch/sh/lib/checksum.S b/arch/sh/lib/checksum.S
> index 3e07074e0098..06fed5a21e8b 100644
> --- a/arch/sh/lib/checksum.S
> +++ b/arch/sh/lib/checksum.S
> @@ -33,7 +33,8 @@
>   */
>  
>  /*	
> - * asmlinkage __wsum csum_partial(const void *buf, int len, __wsum sum);
> + * unsigned int csum_partial(const unsigned char *buf, int len,
> + *                           unsigned int sum);
>   */
>  
>  .text
> @@ -45,31 +46,11 @@ ENTRY(csum_partial)
>  	   * Fortunately, it is easy to convert 2-byte alignment to 4-byte
>  	   * alignment for the unrolled loop.
>  	   */
> +	mov	r5, r1
>  	mov	r4, r0
> -	tst	#3, r0		! Check alignment.
> -	bt/s	2f		! Jump if alignment is ok.
> -	 mov	r4, r7		! Keep a copy to check for alignment
> +	tst	#2, r0		! Check alignment.
> +	bt	2f		! Jump if alignment is ok.
>  	!
> -	tst	#1, r0		! Check alignment.
> -	bt	21f		! Jump if alignment is boundary of 2bytes.
> -
> -	! buf is odd
> -	tst	r5, r5
> -	add	#-1, r5
> -	bt	9f
> -	mov.b	@r4+, r0
> -	extu.b	r0, r0
> -	addc	r0, r6		! t=0 from previous tst
> -	mov	r6, r0
> -	shll8	r6
> -	shlr16	r0
> -	shlr8	r0
> -	or	r0, r6
> -	mov	r4, r0
> -	tst	#2, r0
> -	bt	2f
> -21:
> -	! buf is 2 byte aligned (len could be 0)
>  	add	#-2, r5		! Alignment uses up two bytes.
>  	cmp/pz	r5		!
>  	bt/s	1f		! Jump if we had at least two bytes.
> @@ -77,17 +58,16 @@ ENTRY(csum_partial)
>  	bra	6f
>  	 add	#2, r5		! r5 was < 2.  Deal with it.
>  1:
> +	mov	r5, r1		! Save new len for later use.
>  	mov.w	@r4+, r0
>  	extu.w	r0, r0
>  	addc	r0, r6
>  	bf	2f
>  	add	#1, r6
>  2:
> -	! buf is 4 byte aligned (len could be 0)
> -	mov	r5, r1
>  	mov	#-5, r0
> -	shld	r0, r1
> -	tst	r1, r1
> +	shld	r0, r5
> +	tst	r5, r5
>  	bt/s	4f		! if it's =0, go to 4f
>  	 clrt
>  	.align	2
> @@ -109,31 +89,30 @@ ENTRY(csum_partial)
>  	addc	r0, r6
>  	addc	r2, r6
>  	movt	r0
> -	dt	r1
> +	dt	r5
>  	bf/s	3b
>  	 cmp/eq	#1, r0
> -	! here, we know r1==0
> -	addc	r1, r6			! add carry to r6
> +	! here, we know r5==0
> +	addc	r5, r6			! add carry to r6
>  4:
> -	mov	r5, r0
> +	mov	r1, r0
>  	and	#0x1c, r0
>  	tst	r0, r0
> -	bt	6f
> -	! 4 bytes or more remaining
> -	mov	r0, r1
> -	shlr2	r1
> +	bt/s	6f
> +	 mov	r0, r5
> +	shlr2	r5
>  	mov	#0, r2
>  5:
>  	addc	r2, r6
>  	mov.l	@r4+, r2
>  	movt	r0
> -	dt	r1
> +	dt	r5
>  	bf/s	5b
>  	 cmp/eq	#1, r0
>  	addc	r2, r6
> -	addc	r1, r6		! r1==0 here, so it means add carry-bit
> +	addc	r5, r6		! r5==0 here, so it means add carry-bit
>  6:
> -	! 3 bytes or less remaining
> +	mov	r1, r5
>  	mov	#3, r0
>  	and	r0, r5
>  	tst	r5, r5
> @@ -159,16 +138,6 @@ ENTRY(csum_partial)
>  	mov	#0, r0
>  	addc	r0, r6
>  9:
> -	! Check if the buffer was misaligned, if so realign sum
> -	mov	r7, r0
> -	tst	#1, r0
> -	bt	10f
> -	mov	r6, r0
> -	shll8	r6
> -	shlr16	r0
> -	shlr8	r0
> -	or	r0, r6
> -10:
>  	rts
>  	 mov	r6, r0

Applied to my sh-linux tree in the for-next branch.

Thanks,
Adrian

-- 
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913