On Wed, 2024-07-10 at 12:25 +0100, Alan Maguire wrote:

On 09/07/2024 11:45, Geliang Tang wrote:

...

From: Geliang Tang tanggeliang@kylinos.cn

Run bpf_tcp_ca selftests (./test_progs -t bpf_tcp_ca) on a Loongarch platform, some "Segmentation fault" errors occur:

'''  test_dctcp:PASS:bpf_dctcp__open_and_load 0 nsec  test_dctcp:FAIL:bpf_map__attach_struct_ops unexpected error: -524  #29/1    bpf_tcp_ca/dctcp:FAIL  test_cubic:PASS:bpf_cubic__open_and_load 0 nsec  test_cubic:FAIL:bpf_map__attach_struct_ops unexpected error: -524  #29/2    bpf_tcp_ca/cubic:FAIL  test_dctcp_fallback:PASS:dctcp_skel 0 nsec  test_dctcp_fallback:PASS:bpf_dctcp__load 0 nsec  test_dctcp_fallback:FAIL:dctcp link unexpected error: -524  #29/4    bpf_tcp_ca/dctcp_fallback:FAIL  test_write_sk_pacing:PASS:open_and_load 0 nsec  test_write_sk_pacing:FAIL:attach_struct_ops unexpected error: -524  #29/6    bpf_tcp_ca/write_sk_pacing:FAIL  test_update_ca:PASS:open 0 nsec  test_update_ca:FAIL:attach_struct_ops unexpected error: -524  settcpca:FAIL:setsockopt unexpected setsockopt: \ actual -1 == expected -1  (network_helpers.c:99: errno: No such file or directory) \ Failed to call post_socket_cb  start_test:FAIL:start_server_str unexpected start_server_str: \ actual -1 == expected -1  test_update_ca:FAIL:ca1_ca1_cnt unexpected ca1_ca1_cnt: \ actual 0 <= expected 0  #29/9    bpf_tcp_ca/update_ca:FAIL  #29      bpf_tcp_ca:FAIL  Caught signal #11!  Stack trace:  ./test_progs(crash_handler+0x28)[0x5555567ed91c]  linux-vdso.so.1(__vdso_rt_sigreturn+0x0)[0x7ffffee408b0]  ./test_progs(bpf_link__update_map+0x80)[0x555556824a78]  ./test_progs(+0x94d68)[0x5555564c4d68]  ./test_progs(test_bpf_tcp_ca+0xe8)[0x5555564c6a88]  ./test_progs(+0x3bde54)[0x5555567ede54]  ./test_progs(main+0x61c)[0x5555567efd54]  /usr/lib64/libc.so.6(+0x22208)[0x7ffff2aaa208]  /usr/lib64/libc.so.6(__libc_start_main+0xac)[0x7ffff2aaa30c]  ./test_progs(_start+0x48)[0x55555646bca8]  Segmentation fault '''

This is because BPF trampoline is not implemented on Loongarch yet, "link" returned by bpf_map__attach_struct_ops() is NULL. test_progs crashs when this NULL link passes to bpf_link__update_map(). This patch adds NULL checks for all links in bpf_tcp_ca to fix these errors. If "link" is NULL, goto the newly added label "out" to destroy the skel.

v2:  - use "goto out" instead of "return" as Eduard suggested.

Signed-off-by: Geliang Tang tanggeliang@kylinos.cn

Reviewed-by: Alan Maguire alan.maguire@oracle.com

Maybe I'm missing it, but I'm not seeing this series on patchwork.kernel.org/project/netdevbpf, so we don't have an associated CI run (the series is in the kselftest patchwork however). Is there some patchwork bot magic that will do this?

Thanks for your review. I want to do a small update for this patch, so I changed this set as "Changes Requested". So they are not showed in patchwork. v2 will be sent soon. I will add your "reviewed-by" tag in it.

Thanks, -Geliang

...

---

.../selftests/bpf/prog_tests/bpf_tcp_ca.c     | 21 +++++++++++++--

1 file changed, 15 insertions(+), 6 deletions(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/bpf_tcp_ca.c b/tools/testing/selftests/bpf/prog_tests/bpf_tcp_ca.c index d842ff64bc2a..efc1bf2ff7de 100644 --- a/tools/testing/selftests/bpf/prog_tests/bpf_tcp_ca.c +++ b/tools/testing/selftests/bpf/prog_tests/bpf_tcp_ca.c @@ -411,7 +411,8 @@ static void test_update_ca(void)   return;     link = bpf_map__attach_struct_ops(skel->maps.ca_update_1);

• ASSERT_OK_PTR(link, "attach_struct_ops");

• if (!ASSERT_OK_PTR(link, "attach_struct_ops"))

• goto out;
  
  

do_test(&opts);   saved_ca1_cnt = skel->bss->ca1_cnt; @@ -425,6 +426,7 @@ static void test_update_ca(void)   ASSERT_GT(skel->bss->ca2_cnt, 0, "ca2_ca2_cnt");     bpf_link__destroy(link); +out:   tcp_ca_update__destroy(skel);  }   @@ -447,7 +449,8 @@ static void test_update_wrong(void)   return;     link = bpf_map__attach_struct_ops(skel->maps.ca_update_1);

• ASSERT_OK_PTR(link, "attach_struct_ops");

• if (!ASSERT_OK_PTR(link, "attach_struct_ops"))

• goto out;
  

do_test(&opts);   saved_ca1_cnt = skel->bss->ca1_cnt; @@ -460,11 +463,13 @@ static void test_update_wrong(void)   ASSERT_GT(skel->bss->ca1_cnt, saved_ca1_cnt, "ca2_ca1_cnt");     bpf_link__destroy(link); +out:   tcp_ca_update__destroy(skel);  }    static void test_mixed_links(void)  {

• struct bpf_link *link = NULL, *link_nl = NULL;

struct cb_opts cb_opts = {   .cc = "tcp_ca_update",   }; @@ -473,7 +478,6 @@ static void test_mixed_links(void)   .cb_opts = &cb_opts,   };   struct tcp_ca_update *skel;

• struct bpf_link *link, *link_nl;

int err;     skel = tcp_ca_update__open_and_load(); @@ -481,10 +485,12 @@ static void test_mixed_links(void)   return;     link_nl = bpf_map__attach_struct_ops(skel-

...

maps.ca_no_link);

• ASSERT_OK_PTR(link_nl, "attach_struct_ops_nl");

• if (!ASSERT_OK_PTR(link_nl, "attach_struct_ops_nl"))

• goto out;
  

link = bpf_map__attach_struct_ops(skel->maps.ca_update_1);

• ASSERT_OK_PTR(link, "attach_struct_ops");

• if (!ASSERT_OK_PTR(link, "attach_struct_ops"))

• goto out;
  

do_test(&opts);   ASSERT_GT(skel->bss->ca1_cnt, 0, "ca1_ca1_cnt"); @@ -492,6 +498,7 @@ static void test_mixed_links(void)   err = bpf_link__update_map(link, skel->maps.ca_no_link);   ASSERT_ERR(err, "update_map");   +out:   bpf_link__destroy(link);   bpf_link__destroy(link_nl);   tcp_ca_update__destroy(skel); @@ -536,7 +543,8 @@ static void test_link_replace(void)   bpf_link__destroy(link);     link = bpf_map__attach_struct_ops(skel->maps.ca_update_2);

• ASSERT_OK_PTR(link, "attach_struct_ops_2nd");

• if (!ASSERT_OK_PTR(link, "attach_struct_ops_2nd"))

• goto out;
  

/* BPF_F_REPLACE with a wrong old map Fd. It should fail!   * @@ -559,6 +567,7 @@ static void test_link_replace(void)     bpf_link__destroy(link);   +out:   tcp_ca_update__destroy(skel);  }

On 09/07/2024 11:45, Geliang Tang wrote:

From: Geliang Tang tanggeliang@kylinos.cn

Run dummy_st_ops selftests (./test_progs -t dummy_st_ops) on a Loongarch platform, some "unexpected arg" errors occur:

''' #78/1 dummy_st_ops/dummy_st_ops_attach:OK test_dummy_init_ret_value:FAIL:test_ret unexpected test_ret: \ actual 0 != expected 4076074229 #78/2 dummy_st_ops/dummy_init_ret_value:FAIL #78/3 dummy_st_ops/dummy_init_ptr_arg:SKIP test_dummy_multiple_args:FAIL:arg 0 unexpected arg 0: \ actual 0 != expected 7 test_dummy_multiple_args:FAIL:arg 1 unexpected arg 1: \ actual 0 != expected -100 test_dummy_multiple_args:FAIL:arg 2 unexpected arg 2: \ actual 0 != expected 35423 test_dummy_multiple_args:FAIL:arg 3 unexpected arg 3: \ actual 0 != expected 99 test_dummy_multiple_args:FAIL:arg 4 unexpected arg 4: \ actual 0 != expected 1311768467139281697 #78/4 dummy_st_ops/dummy_multiple_args:FAIL #78/5 dummy_st_ops/dummy_sleepable:SKIP #78/6 dummy_st_ops/dummy_sleepable_reject_null:OK #78/7 dummy_st_ops/test_unsupported_field_sleepable:OK #78 dummy_st_ops:FAIL '''

This is because BPF trampoline is not implemented on Loongarch yet, bpf_prog_test_run_opts() returns ENOTSUPP.

This patch checks the return values of bpf_prog_test_run_opts() in dummy_st_ops to fix these errors. If error returned, goto the newly added label "out" to destroy the skel.

Signed-off-by: Geliang Tang tanggeliang@kylinos.cn

Reviewed-by: Alan Maguire alan.maguire@oracle.com

Verified the test still passes on x86_64 too, so feel free to add

Tested-by: Alan Maguire alan.maguire@oracle.com

---

tools/testing/selftests/bpf/prog_tests/dummy_st_ops.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/dummy_st_ops.c b/tools/testing/selftests/bpf/prog_tests/dummy_st_ops.c index d3d94596ab79..a208801f524f 100644 --- a/tools/testing/selftests/bpf/prog_tests/dummy_st_ops.c +++ b/tools/testing/selftests/bpf/prog_tests/dummy_st_ops.c @@ -41,9 +41,11 @@ static void test_dummy_init_ret_value(void) fd = bpf_program__fd(skel->progs.test_1); err = bpf_prog_test_run_opts(fd, &attr);

• ASSERT_OK(err, "test_run");

• if (!ASSERT_OK(err, "test_run"))

• goto out;
  

  ASSERT_EQ(attr.retval, 0xf2f3f4f5, "test_ret");

+out: dummy_st_ops_success__destroy(skel); } @@ -115,13 +117,15 @@ static void test_dummy_multiple_args(void) fd = bpf_program__fd(skel->progs.test_2); err = bpf_prog_test_run_opts(fd, &attr);

• ASSERT_OK(err, "test_run");

• if (!ASSERT_OK(err, "test_run"))

• goto out;
  

  args[0] = 7; for (i = 0; i < ARRAY_SIZE(args); i++) { snprintf(name, sizeof(name), "arg %zu", i); ASSERT_EQ(skel->bss->test_2_args[i], args[i], name); }

+out: dummy_st_ops_success__destroy(skel); }

On 09/07/2024 11:45, Geliang Tang wrote:

From: Geliang Tang tanggeliang@kylinos.cn

If bpf_object__load() fails in test_xdp_adjust_frags_tail_grow(), "obj" opened before this should be closed. So use "goto out" to close it instead of using "return" here.

Signed-off-by: Geliang Tang tanggeliang@kylinos.cn

Reviewed-by: Alan Maguire alan.maguire@oracle.com

...with one suggestion below...

---

tools/testing/selftests/bpf/prog_tests/xdp_adjust_tail.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_tail.c b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_tail.c index f09505f8b038..53d6ad8c2257 100644 --- a/tools/testing/selftests/bpf/prog_tests/xdp_adjust_tail.c +++ b/tools/testing/selftests/bpf/prog_tests/xdp_adjust_tail.c @@ -222,7 +222,7 @@ static void test_xdp_adjust_frags_tail_grow(void) prog = bpf_object__next_program(obj, NULL); if (bpf_object__load(obj))

• return;
  

• goto out;
  
  

Nit: perhaps we should change this to

if (!ASSERT_OK(bpf_object__load(obj), "obj_load")) goto out;

?