Re: [PATCH bpf-next] arm64: bpf: zero upper bits after rev32

[Puranjay Mohan <puranjay12@xxxxxxxxx>]

Artem Savkov <asavkov@xxxxxxxxxx> writes:

> On Wed, Mar 20, 2024 at 07:34:46PM +0800, Xu Kuohai wrote:
>> On 3/13/2024 10:02 PM, Artem Savkov wrote:
>> > Commit d63903bbc30c7 ("arm64: bpf: fix endianness conversion bugs")
>> > added upper bits zeroing to byteswap operations, but it assumes they
>> > will be already zeroed after rev32, which is not the case on some
>> > systems at least:
>> > 
>> > [ 9757.262607] test_bpf: #312 BSWAP 16: 0x0123456789abcdef -> 0xefcd jited:1 8 PASS
>> > [ 9757.264435] test_bpf: #313 BSWAP 32: 0x0123456789abcdef -> 0xefcdab89 jited:1 ret 1460850314 != -271733879 (0x5712ce8a != 0xefcdab89)FAIL (1 times)
>> > [ 9757.266260] test_bpf: #314 BSWAP 64: 0x0123456789abcdef -> 0x67452301 jited:1 8 PASS
>> > [ 9757.268000] test_bpf: #315 BSWAP 64: 0x0123456789abcdef >> 32 -> 0xefcdab89 jited:1 8 PASS
>> > [ 9757.269686] test_bpf: #316 BSWAP 16: 0xfedcba9876543210 -> 0x1032 jited:1 8 PASS
>> > [ 9757.271380] test_bpf: #317 BSWAP 32: 0xfedcba9876543210 -> 0x10325476 jited:1 ret -1460850316 != 271733878 (0xa8ed3174 != 0x10325476)FAIL (1 times)
>> > [ 9757.273022] test_bpf: #318 BSWAP 64: 0xfedcba9876543210 -> 0x98badcfe jited:1 7 PASS
>> > [ 9757.274721] test_bpf: #319 BSWAP 64: 0xfedcba9876543210 >> 32 -> 0x10325476 jited:1 9 PASS
>> > 
>> > Fixes: d63903bbc30c7 ("arm64: bpf: fix endianness conversion bugs")
>> > Signed-off-by: Artem Savkov <asavkov@xxxxxxxxxx>
>> > ---
>> >   arch/arm64/net/bpf_jit_comp.c | 3 ++-
>> >   1 file changed, 2 insertions(+), 1 deletion(-)
>> > 
>> > diff --git a/arch/arm64/net/bpf_jit_comp.c b/arch/arm64/net/bpf_jit_comp.c
>> > index c5b461dda4385..e86e5ba74dca2 100644
>> > --- a/arch/arm64/net/bpf_jit_comp.c
>> > +++ b/arch/arm64/net/bpf_jit_comp.c
>> > @@ -944,7 +944,8 @@ static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
>> >   			break;
>> >   		case 32:
>> >   			emit(A64_REV32(is64, dst, dst), ctx);
>> > -			/* upper 32 bits already cleared */
>> > +			/* zero-extend 32 bits into 64 bits */
>> > +			emit(A64_UXTW(is64, dst, dst), ctx);
>> 
>> I think the problem only occurs when is64 == 1. In this case, the generated rev32
>> insn reverses byte order in both high and low 32-bit word. To fix it, we could just
>> set the first arg to 0 for A64_REV32:
>> 
>> emit(A64_REV32(0, dst, dst), ctx);
>> 
>> No need to add an extra uxtw isnn.
>
> I can confirm this approach fixes the test issue as well.

Yes, the following diff fixes the issue:

diff --git a/arch/arm64/net/bpf_jit_comp.c b/arch/arm64/net/bpf_jit_comp.c
index bc16eb694..64deff221 100644
--- a/arch/arm64/net/bpf_jit_comp.c
+++ b/arch/arm64/net/bpf_jit_comp.c
@@ -943,7 +943,7 @@ static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
                        emit(A64_UXTH(is64, dst, dst), ctx);
                        break;
                case 32:
-                       emit(A64_REV32(is64, dst, dst), ctx);
+                       emit(A64_REV32(0, dst, dst), ctx);
                        /* upper 32 bits already cleared */
                        break;
                case 64:

All tests pass with this change:

test_bpf: Summary: 1049 PASSED, 0 FAILED, [1037/1037 JIT'ed]
test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

When you send a patch please add:

Tested-by: Puranjay Mohan <puranjay12@xxxxxxxxx>
Acked-by: Puranjay Mohan <puranjay12@xxxxxxxxx>


Thanks,
Puranjay