On 3/23/21 7:06 PM, Arnaldo Carvalho de Melo wrote:

Em Mon, Mar 22, 2021 at 01:53:39PM +0100, Thomas Richter escreveu:

quoted

For some time now the perf test 42: BPF filter returns an error
on bpf relocation subtest, at least on x86 and s390. This is caused by

commit d859900c4c56 ("bpf, libbpf: support global data/bss/rodata sections")

which introduces support for global variables in eBPF programs. At least
for global variables defined static.

Perf test 42 checks that the eBPF relocation fails when the eBPF program
contains a global variable. It returns OK when the eBPF program
could not be loaded and FAILED otherwise.

With above commit the test logic for the eBPF relocation need to change:
1. The function prepare_bpf() now always succeeds, the eBPF program
   compiled without errors and returns a valid object pointer instead of
   NULL.
2. There is no kprobe named sys_write, it now named ksys_write.
3. The function do_test() now returns TEST_FAIL because function
   parse_events_load_bpf_obj() can not execute the eBPF program. The
   eBPF verifier complains on an invalid map pointer:
      libbpf: load bpf program failed: Permission denied
      libbpf: -- BEGIN DUMP LOG ---
      libbpf:
      0: (b7) r1 = 0
      1: (63) *(u32 *)(r10 -4) = r1
      last_idx 1 first_idx 0
      regs=2 stack=0 before 0: (b7) r1 = 0
      2: (63) *(u32 *)(r10 -8) = r1
      3: (bf) r2 = r10
      4: (07) r2 += -4
      5: (bf) r3 = r10
      6: (07) r3 += -8
      7: (18) r1 = 0x380006ce000
      9: (b7) r4 = 0
      10: (85) call bpf_map_update_elem#2
      R1 type=map_value expected=map_ptr

Fix this by added logic to handle the kernel verifier return code:
1. Add function myksys_write() to cope with successful compile.
2. Use kprobe ksys_write
3. Handle eBPF verifier error.

Output after:
 42: BPF filter                          :
 42.1: Basic BPF filtering               : Ok
 42.2: BPF pinning                       : Ok
 42.3: BPF prologue generation           : Ok
 42.4: BPF relocation checker            : Failed
 #

Output after:
 # ./perf test -F 42
 42: BPF filter                          :
 42.1: Basic BPF filtering               : Ok
 42.2: BPF pinning                       : Ok
 42.3: BPF prologue generation           : Ok
 42.4: BPF relocation checker            : Ok
 #

Signed-off-by: Thomas Richter <redacted>
---
 tools/perf/tests/bpf-script-test-relocation.c |  4 ++--
 tools/perf/tests/bpf.c                        | 11 +++++++++++
 2 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/tools/perf/tests/bpf-script-test-relocation.c b/tools/perf/tests/bpf-script-test-relocation.c
index 74006e4b2d24..f8f8176ad4d1 100644
--- a/tools/perf/tests/bpf-script-test-relocation.c
+++ b/tools/perf/tests/bpf-script-test-relocation.c

@@ -34,8 +34,8 @@ struct bpf_map_def SEC("maps") my_table = {
 
 int this_is_a_global_val;
 
-SEC("func=sys_write")
-int bpf_func__sys_write(void *ctx)
+SEC("func=ksys_write")
+int bpf_func__ksys_write(void *ctx)
 {
 	int key = 0;
 	int value = 0;

diff --git a/tools/perf/tests/bpf.c b/tools/perf/tests/bpf.c
index f57e075b0ed2..d60ef9472d3d 100644
--- a/tools/perf/tests/bpf.c
+++ b/tools/perf/tests/bpf.c

@@ -59,6 +59,11 @@ static int llseek_loop(void)
 
 #endif
 
+static int myksys_write(void)
+{
+	return 0;
+}
+
 static struct {
 	enum test_llvm__testcase prog_id;
 	const char *desc;

@@ -105,6 +110,7 @@ static struct {
 		.name		  = "[bpf_relocation_test]",
 		.msg_compile_fail = "fix 'perf test LLVM' first",
 		.msg_load_fail	  = "libbpf error when dealing with relocation",
+		.target_func	  = &myksys_write,
 	},
 };
 

@@ -258,6 +264,11 @@ static int __test__bpf(int idx)
 		ret = do_test(obj,
 			      bpf_testcase_table[idx].target_func,
 			      bpf_testcase_table[idx].expect_result);
+		if (bpf_testcase_table[idx].prog_id == LLVM_TESTCASE_BPF_RELOCATION
+		    && ret == TEST_FAIL) {
+			ret = TEST_OK;
+			goto out;
+		}

At this point, if it doesn't matter if it fails or succeeds, just drop
this test case?

- Arnaldo

Arnaldo,

dropping this test case is fine with me. Support for eBPF global variables is now
part of the linux kernel and there are test programs in the 
.../testing/selftests/bpf directory. File prog_tests/global_data.c for example.

If you agree, I will provide a patch to remove the test case 42.4.

-- 
Thomas Richter, Dept 3303, IBM s390 Linux Development, Boeblingen, Germany
--
Vorsitzender des Aufsichtsrats: Gregor Pillen
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294