[PATCH] selftests/bpf: Fix fill_link_info selftest on powerpc

[PATCH] selftests/bpf: Fix fill_link_info selftest on powerpc

[Saket Kumar Bhaskar]

With CONFIG_KPROBES_ON_FTRACE enabled on powerpc, ftrace_location_range
returns ftrace location for bpf_fentry_test1 at offset of 4 bytes from
function entry. This is because branch to _mcount function is at offset
of 4 bytes in function profile sequence.

To fix this, add entry_offset of 4 bytes while verifying the address for
kprobe entry address of bpf_fentry_test1 in verify_perf_link_info in
selftest, when CONFIG_KPROBES_ON_FTRACE is enabled.

Disassemble of bpf_fentry_test1:

c000000000e4b080 <bpf_fentry_test1>:
c000000000e4b080:       a6 02 08 7c     mflr    r0
c000000000e4b084:       b9 e2 22 4b     bl      c00000000007933c <_mcount>
c000000000e4b088:       01 00 63 38     addi    r3,r3,1
c000000000e4b08c:       b4 07 63 7c     extsw   r3,r3
c000000000e4b090:       20 00 80 4e     blr

When CONFIG_PPC_FTRACE_OUT_OF_LINE [1] is enabled, these function profile
sequence is moved out of line with an unconditional branch at offset 0.
So, the test works without altering the offset for
'CONFIG_KPROBES_ON_FTRACE && CONFIG_PPC_FTRACE_OUT_OF_LINE' case.

Disassemble of bpf_fentry_test1:

c000000000f95190 <bpf_fentry_test1>:
c000000000f95190:       00 00 00 60     nop
c000000000f95194:       01 00 63 38     addi    r3,r3,1
c000000000f95198:       b4 07 63 7c     extsw   r3,r3
c000000000f9519c:       20 00 80 4e     blr

[1] https://lore.kernel.org/all/20241030070850.1361304-13-hbathini@linux.ibm.com/

Fixes: 23cf7aa539dc ("selftests/bpf: Add selftest for fill_link_info")
Signed-off-by: Saket Kumar Bhaskar <skb99@linux.ibm.com>
---
 .../selftests/bpf/prog_tests/fill_link_info.c       |  4 ++++
 .../selftests/bpf/progs/test_fill_link_info.c       | 13 ++++++++++---
 2 files changed, 14 insertions(+), 3 deletions(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/fill_link_info.c b/tools/testing/selftests/bpf/prog_tests/fill_link_info.c
index d50cbd804..e59af2aa6 100644
--- a/tools/testing/selftests/bpf/prog_tests/fill_link_info.c
+++ b/tools/testing/selftests/bpf/prog_tests/fill_link_info.c
@@ -171,6 +171,10 @@ static void test_kprobe_fill_link_info(struct test_fill_link_info *skel,
 		/* See also arch_adjust_kprobe_addr(). */
 		if (skel->kconfig->CONFIG_X86_KERNEL_IBT)
 			entry_offset = 4;
+		if (skel->kconfig->CONFIG_PPC64 &&
+		    skel->kconfig->CONFIG_KPROBES_ON_FTRACE &&
+		    !skel->kconfig->CONFIG_PPC_FTRACE_OUT_OF_LINE)
+			entry_offset = 4;
 		err = verify_perf_link_info(link_fd, type, kprobe_addr, 0, entry_offset);
 		ASSERT_OK(err, "verify_perf_link_info");
 	} else {
diff --git a/tools/testing/selftests/bpf/progs/test_fill_link_info.c b/tools/testing/selftests/bpf/progs/test_fill_link_info.c
index 6afa83475..fac33a14f 100644
--- a/tools/testing/selftests/bpf/progs/test_fill_link_info.c
+++ b/tools/testing/selftests/bpf/progs/test_fill_link_info.c
@@ -6,13 +6,20 @@
 #include <stdbool.h>
 
 extern bool CONFIG_X86_KERNEL_IBT __kconfig __weak;
+extern bool CONFIG_PPC_FTRACE_OUT_OF_LINE __kconfig __weak;
+extern bool CONFIG_KPROBES_ON_FTRACE __kconfig __weak;
+extern bool CONFIG_PPC64 __kconfig __weak;
 
-/* This function is here to have CONFIG_X86_KERNEL_IBT
- * used and added to object BTF.
+/* This function is here to have CONFIG_X86_KERNEL_IBT,
+ * CONFIG_PPC_FTRACE_OUT_OF_LINE, CONFIG_KPROBES_ON_FTRACE,
+ * CONFIG_PPC6 used and added to object BTF.
  */
 int unused(void)
 {
-	return CONFIG_X86_KERNEL_IBT ? 0 : 1;
+	return CONFIG_X86_KERNEL_IBT ||
+			CONFIG_PPC_FTRACE_OUT_OF_LINE ||
+			CONFIG_KPROBES_ON_FTRACE ||
+			CONFIG_PPC64 ? 0 : 1;
 }
 
 SEC("kprobe")
-- 
2.45.2

Re: [PATCH] selftests/bpf: Fix fill_link_info selftest on powerpc

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to bpf/bpf-next.git (master)
by Andrii Nakryiko <andrii@kernel.org>:

On Mon,  9 Dec 2024 12:27:20 +0530 you wrote:
> With CONFIG_KPROBES_ON_FTRACE enabled on powerpc, ftrace_location_range
> returns ftrace location for bpf_fentry_test1 at offset of 4 bytes from
> function entry. This is because branch to _mcount function is at offset
> of 4 bytes in function profile sequence.
> 
> To fix this, add entry_offset of 4 bytes while verifying the address for
> kprobe entry address of bpf_fentry_test1 in verify_perf_link_info in
> selftest, when CONFIG_KPROBES_ON_FTRACE is enabled.
> 
> [...]

Here is the summary with links:
  - selftests/bpf: Fix fill_link_info selftest on powerpc
    https://git.kernel.org/bpf/bpf-next/c/4d33dc1bc31d

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