Re: [PATCH bpf-next v1 1/2] selftests/bpf: Check alignment flag if expected result is REJECT

[Paul Chaignon <paul.chaignon@gmail.com>]

On Thu, Mar 12, 2026 at 02:59:57PM +0800, Tiezhu Yang wrote:
> Hi Emil and Paul,

Hi Tiezhu,

> 
> On 2026/3/11 下午11:56, Emil Tsalapatis wrote:
> > On Tue Mar 10, 2026 at 2:45 AM EDT, Tiezhu Yang wrote:
> > > CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS is set unconditionally for the
> > > most archs such as x86_64, aarch64, ppc64el and s390x, but this config
> > > may be not set by users for riscv64 and loongarch64.
> > > 
> > > If CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS is not set, the BPF verifier
> > > detects if a program has unaligned access and then rejects them. So it
> > > should also check the flag F_NEEDS_EFFICIENT_UNALIGNED_ACCESS if the
> > > expected result is REJECT and set alignment_prevented_execution as 1,
> > > then the message "(NOTE: not executed due to unknown alignment)" can
> > > be printed for some testcases of test_verifier to reflect the reality.
> > > 
> > > Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
> > > ---
> > >   tools/testing/selftests/bpf/test_verifier.c | 5 +++++
> > >   1 file changed, 5 insertions(+)
> > > 
> > > diff --git a/tools/testing/selftests/bpf/test_verifier.c b/tools/testing/selftests/bpf/test_verifier.c
> > > index a8ae03c57bba..a1ae2f044e96 100644
> > > --- a/tools/testing/selftests/bpf/test_verifier.c
> > > +++ b/tools/testing/selftests/bpf/test_verifier.c
> > > @@ -1640,6 +1640,11 @@ static void do_test_single(struct bpf_test *test, bool unpriv,
> > >   			printf("FAIL\nUnexpected success to load!\n");
> > >   			goto fail_log;
> > >   		}
> > > +#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
> > > +		if (fd_prog < 0 &&
> > > +		    (test->flags & F_NEEDS_EFFICIENT_UNALIGNED_ACCESS))
> > > +			alignment_prevented_execution = 1;
> > > +#endif
> > 
> > This doesn't look like it's breaking anything, but it will cause all
> > tests with F_NEEDS_EFFICIENT_UNALIGNED ACCESS to be reported as failing
> > due to unaligned accesses even if they actually failed due to expected
> > errors at load time.
> > 
> > Which test programs is this fix targeted towards? Can't we just skip
> > those tests for riscv/loongarch instead of adding this workaround?
> 
> Thanks for your reviews.
> 
> It does not fix any failed testcases, the testcases passed without
> and with this patch. It only affects the output for some testcases
> such as test_verifier 14, 153, 190, 191, 193, 197, 200, 201, 202,
> 204, 255, 256, 295, 297, 298, 300, 301, 302, 307, 308, 331, 341,
> 342, 475, 477, 500, 512, 514, 515, 517, 520, 522, 523, 525.
> 
> For example, for the test_verifier 14,
> verifier/atomic_cmpxchg.c:215
> "Dest pointer in r0 - succeed, check 5",
> if remove ".flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,",

I think I'm missing something. Why would you remove this flag?

Are you trying to illustrate that this is a program that could be
rejected because of misaligned accesses? I agree, but here the flag is
set and this program is therefore rejected because of an invalid memory
access; the misalgined access check is skipped in the verifier.

> here is the test result if HAVE_EFFICIENT_UNALIGNED_ACCESS is not set:
> 
> $ sudo ./test_verifier 14
> #14/u Dest pointer in r0 - succeed, check 5 OK
> #14/p Dest pointer in r0 - succeed, check 5 FAIL
> Unexpected verifier log!
> EXP: R0 invalid mem access
> RES:
> FAIL
> Unexpected error message!
> 	EXP: R0 invalid mem access
> 	RES: misaligned access off (0x0; 0xffffffff)+-8 size 4
> verification time 27 usec
> stack depth 8
> processed 5 insns (limit 1000000) max_states_per_insn 0 total_states 0
> peak_states 0 mark_read 0
> 
> misaligned access off (0x0; 0xffffffff)+-8 size 4
> verification time 27 usec
> stack depth 8
> processed 5 insns (limit 1000000) max_states_per_insn 0 total_states 0
> peak_states 0 mark_read 0
> Summary: 1 PASSED, 0 SKIPPED, 1 FAILED
> 
> It seems that this is because the program has unaligned accesses,
> but the kernel sets CONFIG_ARCH_STRICT_ALIGN by default to enable
> -mstrict-align to prevent unaligned accesses.
> 
> So I think add "(NOTE: not executed due to unknown alignment)" to the
> output is reasonable. Something I missed?
> 
> Without this patch:
> 
>   sudo ./test_verifier > log.before
> 
> With this patch:
> 
>   sudo ./test_verifier > log.after
> 
> Here is the diff:
> 
> $ diff -uNr log.before log.after
> --- log.before	2026-03-12 14:19:04.119957887 +0800
> +++ log.after	2026-03-12 14:23:03.468251328 +0800
> @@ -26,7 +26,7 @@
>  #12/p Dest pointer in r0 - succeed, check 3 OK
>  #13/u Dest pointer in r0 - succeed, check 4 OK
>  #13/p Dest pointer in r0 - succeed, check 4 OK
> -#14/p Dest pointer in r0 - succeed, check 5 OK
> +#14/p Dest pointer in r0 - succeed, check 5 OK (NOTE: not executed due to
> unknown alignment)

As stated in my previous answer, this note would be incorrect: the
program was executed in test case #14 because it was rejected, not
because of unknown alignment.

[...]