Some compilers (for example clang-15) optimize bpf_testmod_loop_test and
remove the loop:
gcc version
(gdb) disassemble bpf_testmod_loop_test
Dump of assembler code for function bpf_testmod_loop_test:
0x0000000000000570 <+0>: callq 0x575 <bpf_testmod_loop_test+5>
0x0000000000000575 <+5>: xor %eax,%eax
0x0000000000000577 <+7>: test %edi,%edi
0x0000000000000579 <+9>: jle 0x587 <bpf_testmod_loop_test+23>
0x000000000000057b <+11>: xor %edx,%edx
0x000000000000057d <+13>: add %edx,%eax
0x000000000000057f <+15>: add $0x1,%edx
0x0000000000000582 <+18>: cmp %edx,%edi
0x0000000000000584 <+20>: jne 0x57d <bpf_testmod_loop_test+13>
0x0000000000000586 <+22>: retq
0x0000000000000587 <+23>: retq
clang-15 version
(gdb) disassemble bpf_testmod_loop_test
Dump of assembler code for function bpf_testmod_loop_test:
0x0000000000000450 <+0>: nopl 0x0(%rax,%rax,1)
0x0000000000000455 <+5>: test %edi,%edi
0x0000000000000457 <+7>: jle 0x46b <bpf_testmod_loop_test+27>
0x0000000000000459 <+9>: lea -0x1(%rdi),%eax
0x000000000000045c <+12>: lea -0x2(%rdi),%ecx
0x000000000000045f <+15>: imul %rax,%rcx
0x0000000000000463 <+19>: shr %rcx
0x0000000000000466 <+22>: lea -0x1(%rdi,%rcx,1),%eax
0x000000000000046a <+26>: retq
0x000000000000046b <+27>: xor %eax,%eax
0x000000000000046d <+29>: retq
Note: The jne instruction is removed in clang-15 version.
Force the compile to keep the loop by making sum volatile.
Signed-off-by: Song Liu <song@xxxxxxxxxx>
---
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
index 7999476b9446..c5ad39bbe9af 100644
--- a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
+++ b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
@@ -130,7 +130,11 @@ bpf_testmod_test_btf_type_tag_percpu_2(struct bpf_testmod_btf_type_tag_3 *arg) {
noinline int bpf_testmod_loop_test(int n)
{
- int i, sum = 0;
+ /* Make sum volatile, so smart compilers, such as clang, will not
+ * optimize the code by removing the loop.
+ */
+ volatile int sum = 0;
+ int i;
/* the primary goal of this test is to test LBR. Create a lot of
* branches in the function, so we can catch it easily.
--
2.34.1
