sparc64 verifier failures..

I'm getting about 100 verifier failures on sparc64.

The vast majority of them seem to be due to misaligned packet
accesses.  Here is a sample of some of the failures.

#32/p invalid fp arithmetic FAIL
Unexpected error message!
0: (b7) r0 = 0
1: (bf) r1 = r10
2: (17) r1 -= 8
3: (7b) *(u64 *)(r1 +0) = r0
misaligned access off (0x0; 0xffffffffffffffff)+0+0 size 8
 ...
#160/p raw_stack: skb_load_bytes, spilled regs corruption FAIL
Unexpected error message!
0: (b7) r2 = 4
1: (bf) r6 = r10
2: (07) r6 += -8
3: (7b) *(u64 *)(r6 +0) = r1
4: (bf) r3 = r6
5: (b7) r4 = 8
6: (85) call bpf_skb_load_bytes#26
7: (79) r0 = *(u64 *)(r6 +0)
8: (61) r0 = *(u32 *)(r0 +8)
misaligned access off (0x0; 0xffffffffffffffff)+0+8 size 4
#161/p raw_stack: skb_load_bytes, spilled regs corruption 2 FAIL
Unexpected error message!
0: (b7) r2 = 4
1: (bf) r6 = r10
2: (07) r6 += -16
3: (7b) *(u64 *)(r6 -8) = r1
4: (7b) *(u64 *)(r6 +0) = r1
5: (7b) *(u64 *)(r6 +8) = r1
6: (bf) r3 = r6
7: (b7) r4 = 8
8: (85) call bpf_skb_load_bytes#26
9: (79) r0 = *(u64 *)(r6 -8)
10: (79) r2 = *(u64 *)(r6 +8)
11: (79) r3 = *(u64 *)(r6 +0)

194 more lines

On 12/19/2017 09:36 PM, David Miller wrote:

I'm getting about 100 verifier failures on sparc64.

The vast majority of them seem to be due to misaligned packet
accesses.  Here is a sample of some of the failures.

Thanks, I'll check it next days. It would probably make sense to have
a --strict-align mode for test_verifier that enables it on all tests
unconditionally so selftests suite would always run in both modes (at
least on archs that don't have this restriction).

From: Daniel Borkmann <daniel@iogearbox.net>
Date: Wed, 20 Dec 2017 14:05:36 +0100

On 12/19/2017 09:36 PM, David Miller wrote:

quoted

I'm getting about 100 verifier failures on sparc64.

The vast majority of them seem to be due to misaligned packet
accesses.  Here is a sample of some of the failures.

Thanks, I'll check it next days. It would probably make sense to have
a --strict-align mode for test_verifier that enables it on all tests
unconditionally so selftests suite would always run in both modes (at
least on archs that don't have this restriction).

Indeed, and for some reason I thought we were already doing this.