[PATCH bpf-next 0/2] Fix verifier test failures in verbose mode

[PATCH bpf-next 0/2] Fix verifier test failures in verbose mode

[Gregory Bell]

This patch series fixes two issues that cause false failures in the
BPF verifier test suite when run with verbose output (`-v`).

The following tests fail only when running the test_verifier in
verbose.

#458/p ld_dw: xor semi-random 64 bit imms, test 5 FAIL
#494/p precise: test 1 FAIL
#495/p precise: test 2 FAIL
#497/p precise: ST zero to stack insn is supported FAIL
#498/p precise: STX insn causing spi > allocated_stack FAIL
#501/p scale: scale test 1 FAIL
#502/p scale: scale test 2 FAIL

This leads to inconsistent results across verbose and
non-verbose runs.

Patch 1 addresses an issue where the verbose flag (`-v`) unintentionally
overrides the `opts.log_level`, leading to incorrect contents when checking
bpf_vlog in tests with `expected_ret == VERBOSE_ACCEPT`. This occurs when
running verbose with `-v` but not `-vv`

Patch 2 increases the size of the `bpf_vlog[]` buffer to prevent truncation
of large verifier logs, which was causing failures in several scale and
64-bit immediate tests.


Before patches:
./test_verifier | grep FAIL
Summary: 790 PASSED, 0 SKIPPED, 0 FAILED

./test_verifier -v | grep FAIL
#115/p BPF_ST_MEM stack imm sign FAIL
#458/p ld_dw: xor semi-random 64 bit imms, test 5 FAIL
#494/p precise: test 1 FAIL
#495/p precise: test 2 FAIL
#497/p precise: ST zero to stack insn is supported FAIL
#498/p precise: STX insn causing spi > allocated_stack FAIL
#501/p scale: scale test 1 FAIL
#502/p scale: scale test 2 FAIL
Summary: 782 PASSED, 0 SKIPPED, 8 FAILED

./test_verifier -vv | grep FAIL
#458/p ld_dw: xor semi-random 64 bit imms, test 5 FAIL
#501/p scale: scale test 1 FAIL
#502/p scale: scale test 2 FAIL
Summary: 787 PASSED, 0 SKIPPED, 3 FAILED

After patches:
./test_verifier -v | grep FAIL
Summary: 790 PASSED, 0 SKIPPED, 0 FAILED
./test_verifier -vv | grep FAIL
Summary: 790 PASSED, 0 SKIPPED, 0 FAILED

These fixes improve test reliability and ensure consistent behavior across
verbose and non-verbose runs.

Gregory Bell (2):
  selftests/bpf: test_verifier verbose causes erroneous failures
  selftests/bpf: test_verifier verbose log overflows

 tools/testing/selftests/bpf/test_verifier.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

-- 
2.49.0

[PATCH bpf-next 1/2] selftests/bpf: test_verifier verbose causes erroneous failures

[Gregory Bell]

When running test_verifier with the -v flag and a test with
`expected_ret==VERBOSE_ACCEPT`, the opts.log_level is unintentionally
overwritten because the verbose flag takes precedence. This leads to
a mismatch in the expected and actual contents of bpf_vlog, causing
tests to fail incorrectly.

Reorder the conditional logic that sets opts.log_level to preserve
the expected log level and prevent it from being overridden by -v.

Signed-off-by: Gregory Bell <grbell@redhat.com>
---
 tools/testing/selftests/bpf/test_verifier.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/tools/testing/selftests/bpf/test_verifier.c b/tools/testing/selftests/bpf/test_verifier.c
index 447b68509d76..2d13e862b078 100644
--- a/tools/testing/selftests/bpf/test_verifier.c
+++ b/tools/testing/selftests/bpf/test_verifier.c
@@ -1559,10 +1559,10 @@ static void do_test_single(struct bpf_test *test, bool unpriv,
 		       test->errstr_unpriv : test->errstr;
 
 	opts.expected_attach_type = test->expected_attach_type;
-	if (verbose)
-		opts.log_level = verif_log_level | 4; /* force stats */
-	else if (expected_ret == VERBOSE_ACCEPT)
+	if (expected_ret == VERBOSE_ACCEPT)
 		opts.log_level = 2;
+	else if (verbose)
+		opts.log_level = verif_log_level | 4; /* force stats */
 	else
 		opts.log_level = DEFAULT_LIBBPF_LOG_LEVEL;
 	opts.prog_flags = pflags;
-- 
2.49.0

[PATCH bpf-next 2/2] selftests/bpf: test_verifier verbose log overflows

[Gregory Bell]

Tests:
 - 458/p ld_dw: xor semi-random 64-bit imms, test 5
 - 501/p scale: scale test 1
 - 502/p scale: scale test 2

fail in verbose mode due to bpf_vlog[] overflowing. These tests
generate large verifier logs that exceed the current buffer size,
causing them to fail to load.

Increase the size of the bpf_vlog[] buffer to accommodate larger
logs and prevent false failures during test runs with verbose output.

Signed-off-by: Gregory Bell <grbell@redhat.com>
---
 tools/testing/selftests/bpf/test_verifier.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/testing/selftests/bpf/test_verifier.c b/tools/testing/selftests/bpf/test_verifier.c
index 2d13e862b078..27db34ecf3f5 100644
--- a/tools/testing/selftests/bpf/test_verifier.c
+++ b/tools/testing/selftests/bpf/test_verifier.c
@@ -734,7 +734,7 @@ static __u32 btf_raw_types[] = {
 	BTF_MEMBER_ENC(71, 13, 128), /* struct prog_test_member __kptr *ptr; */
 };
 
-static char bpf_vlog[UINT_MAX >> 8];
+static char bpf_vlog[UINT_MAX >> 5];
 
 static int load_btf_spec(__u32 *types, int types_len,
 			 const char *strings, int strings_len)
-- 
2.49.0

Re: [PATCH bpf-next 0/2] Fix verifier test failures in verbose mode

[Eduard Zingerman]

Gregory Bell <grbell@redhat.com> writes:

> This patch series fixes two issues that cause false failures in the
> BPF verifier test suite when run with verbose output (`-v`).
>
> The following tests fail only when running the test_verifier in
> verbose.
>
> #458/p ld_dw: xor semi-random 64 bit imms, test 5 FAIL
> #494/p precise: test 1 FAIL
> #495/p precise: test 2 FAIL
> #497/p precise: ST zero to stack insn is supported FAIL
> #498/p precise: STX insn causing spi > allocated_stack FAIL
> #501/p scale: scale test 1 FAIL
> #502/p scale: scale test 2 FAIL
>
> This leads to inconsistent results across verbose and
> non-verbose runs.
>
> Patch 1 addresses an issue where the verbose flag (`-v`) unintentionally
> overrides the `opts.log_level`, leading to incorrect contents when checking
> bpf_vlog in tests with `expected_ret == VERBOSE_ACCEPT`. This occurs when
> running verbose with `-v` but not `-vv`
>
> Patch 2 increases the size of the `bpf_vlog[]` buffer to prevent truncation
> of large verifier logs, which was causing failures in several scale and
> 64-bit immediate tests.
>
>
> Before patches:
> ./test_verifier | grep FAIL
> Summary: 790 PASSED, 0 SKIPPED, 0 FAILED

Can reproduce the issue with -v option, the series fixes failures I see.

Tested-by: Eduard Zingerman <eddyz87@gmail.com>

Re: [PATCH bpf-next 0/2] Fix verifier test failures in verbose mode

[patchwork-bot+netdevbpf]

Hello:

This series was applied to bpf/bpf-next.git (master)
by Alexei Starovoitov <ast@kernel.org>:

On Mon, 12 May 2025 10:04:11 -0400 you wrote:
> This patch series fixes two issues that cause false failures in the
> BPF verifier test suite when run with verbose output (`-v`).
> 
> The following tests fail only when running the test_verifier in
> verbose.
> 
> #458/p ld_dw: xor semi-random 64 bit imms, test 5 FAIL
> #494/p precise: test 1 FAIL
> #495/p precise: test 2 FAIL
> #497/p precise: ST zero to stack insn is supported FAIL
> #498/p precise: STX insn causing spi > allocated_stack FAIL
> #501/p scale: scale test 1 FAIL
> #502/p scale: scale test 2 FAIL
> 
> [...]

Here is the summary with links:
  - [bpf-next,1/2] selftests/bpf: test_verifier verbose causes erroneous failures
    https://git.kernel.org/bpf/bpf-next/c/c5bcc8c78127
  - [bpf-next,2/2] selftests/bpf: test_verifier verbose log overflows
    https://git.kernel.org/bpf/bpf-next/c/af8a5125a04c

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html