Re: [PATCH bpf] bpf/tests: Remove test for MOVSX32 with offset=32

[Yonghong Song]

On 11/12/23 11:14 AM, Stanislav Fomichev wrote:
> On 11/10, Puranjay Mohan wrote:
> > MOVSX32 only supports sign extending 8-bit and 16-bit operands into 32
> > bit operands. The "ALU_MOVSX | BPF_W" test tries to sign extend a 32 bit
> > operand into a 32 bit operand which is equivalent to a normal BPF_MOV.
> >
> > Remove this test as it tries to run an invalid instruction.
> >
> > Fixes: daabb2b098e0 ("bpf/tests: add tests for cpuv4 instructions")
> > Signed-off-by: Puranjay Mohan <puranjay12@xxxxxxxxx>
> > Reported-by: kernel test robot <oliver.sang@xxxxxxxxx>
> > Closes: https://lore.kernel.org/oe-lkp/202310111838.46ff5b6a-oliver.sang@xxxxxxxxx
> Acked-by: Stanislav Fomichev <sdf@xxxxxxxxxx>
>
> (based on the fact that emit_movsx_reg doesn't handle 32 bit case under !is64)

The test failure signature is:
  #83 ALU_MOVSX | BPF_W jited:1 ret 2 != 1 (0x2 != 0x1)FAIL (1 times)

If the asm code went though verifier, we should get a verification failure.
But in test_bpf.ko case, the asm code is directly jited (see jited:1 above),
so there is no verification failure.

The current jit does not do error handling and assumes asm code are all legal,
so for !is64 case, the size = 32 asm code is silently ignored for x86 side
and caused the error. But the test is introduced by arm64 and the test passed
for arm64.

Not sure how we could prevent such issues. To handle *all* illegal insns in jit
definitely not a good idea. In any case, the change for this patch looks good.

Acked-by: Yonghong Song <yonghong.song@xxxxxxxxx>