Skip to contentSkip to conversation

bpf: Remove config check to enable bpf support for branch records

PATCH Superseded Kajol Jain 3 participants 6 messages Nov 18, 2021 – Nov 23, 2021

Subsystems (4)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Kajol Jain

author
#0 lore raw

PATCH v2 1/1 bpf: Remove config check to enable bpf support for branch records

Branch data available to bpf programs can be very useful to get stack traces out of userspace application.

Commit fff7b64355ea ("bpf: Add bpf_read_branch_records() helper") added bpf support to capture branch records in x86. Enable this feature for other architectures as well by removing check specific to x86. Incase any platform didn't support branch stack, it will return with -EINVAL.

Selftest 'perf_branches' result on power9 machine with branch stacks support.

Before this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:FAIL
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:FAIL
Summary: 0/1 PASSED, 0 SKIPPED, 1 FAILED

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:OK
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/2 PASSED, 0 SKIPPED, 0 FAILED

Selftest 'perf_branches' result on power9 machine which doesn't support branch stack

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:SKIP
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/1 PASSED, 1 SKIPPED, 0 FAILED

Fixes fff7b64355eac ("bpf: Add bpf_read_branch_records() helper")

---

Tested this patch changes on power9 machine using selftest 'perf branches' which is added in commit 67306f84ca78 ("selftests/bpf:
Add bpf_read_branch_records()")

Changelog:
v1 -> v2
- Inorder to add bpf support to capture branch record in powerpc, rather then adding config for powerpc, entirely remove config check from bpf_read_branch_records function as suggested by Peter Zijlstra

- Link to the v1 patch: https://lkml.org/lkml/2021/11/14/434

 kernel/trace/bpf_trace.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/trace/bpf_trace.c +0 -4
index 7396488793ff..5e445985c6b4 100644
@@ -1402,9 +1402,6 @@ static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
 	   void *, buf, u32, size, u64, flags)
 {
-#ifndef CONFIG_X86
-	return -ENOENT;
-#else
 	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
 	struct perf_branch_stack *br_stack = ctx->data->br_stack;
 	u32 to_copy;
@@ -1425,7 +1422,6 @@ BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
 	memcpy(buf, br_stack->entries, to_copy);
 
 	return to_copy;
-#endif
 }
 
 static const struct bpf_func_proto bpf_read_branch_records_proto = {
Signature · 1 line
2.27.0
#1 lore raw
Quoted 46 lines · Kajol Jain

Branch data available to bpf programs can be very useful to get stack traces out of userspace application.

Commit fff7b64355ea ("bpf: Add bpf_read_branch_records() helper") added bpf support to capture branch records in x86. Enable this feature for other architectures as well by removing check specific to x86. Incase any platform didn't support branch stack, it will return with -EINVAL.

Selftest 'perf_branches' result on power9 machine with branch stacks support.

Before this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:FAIL
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:FAIL
Summary: 0/1 PASSED, 0 SKIPPED, 1 FAILED

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:OK
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/2 PASSED, 0 SKIPPED, 0 FAILED

Selftest 'perf_branches' result on power9 machine which doesn't support branch stack

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:SKIP
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/1 PASSED, 1 SKIPPED, 0 FAILED

Fixes fff7b64355eac ("bpf: Add bpf_read_branch_records() helper")

---

Tested this patch changes on power9 machine using selftest 'perf branches' which is added in commit 67306f84ca78 ("selftests/bpf:
Add bpf_read_branch_records()")

Quoted 6 lines

Changelog:
v1 -> v2
- Inorder to add bpf support to capture branch record in powerpc, rather then adding config for powerpc, entirely remove config check from bpf_read_branch_records function as suggested by Peter Zijlstra

what will be returned for architectures that don't support branch records? Will it be zero instead of -ENOENT?

Quoted 31 lines

- Link to the v1 patch: https://lkml.org/lkml/2021/11/14/434

 kernel/trace/bpf_trace.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/trace/bpf_trace.c +0 -5
index 7396488793ff..5e445985c6b4 100644
@@ -1402,9 +1402,6 @@ static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
           void *, buf, u32, size, u64, flags)
 {
-#ifndef CONFIG_X86
-       return -ENOENT;
-#else
        static const u32 br_entry_size = sizeof(struct perf_branch_entry);
        struct perf_branch_stack *br_stack = ctx->data->br_stack;
        u32 to_copy;
@@ -1425,7 +1422,6 @@ BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
        memcpy(buf, br_stack->entries, to_copy);
 
        return to_copy;
-#endif
 }
 
 static const struct bpf_func_proto bpf_read_branch_records_proto = {
--

2.27.0

author
#2 lore raw
Quoted 54 lines · Andrii Nakryiko
Quoted 46 lines · Kajol Jain

Branch data available to bpf programs can be very useful to get stack traces out of userspace application.

Commit fff7b64355ea ("bpf: Add bpf_read_branch_records() helper") added bpf support to capture branch records in x86. Enable this feature for other architectures as well by removing check specific to x86. Incase any platform didn't support branch stack, it will return with -EINVAL.

Selftest 'perf_branches' result on power9 machine with branch stacks support.

Before this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:FAIL
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:FAIL
Summary: 0/1 PASSED, 0 SKIPPED, 1 FAILED

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:OK
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/2 PASSED, 0 SKIPPED, 0 FAILED

Selftest 'perf_branches' result on power9 machine which doesn't support branch stack

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:SKIP
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/1 PASSED, 1 SKIPPED, 0 FAILED

Fixes fff7b64355eac ("bpf: Add bpf_read_branch_records() helper")

---

Tested this patch changes on power9 machine using selftest 'perf branches' which is added in commit 67306f84ca78 ("selftests/bpf:
Add bpf_read_branch_records()")

Quoted 6 lines

Changelog:
v1 -> v2
- Inorder to add bpf support to capture branch record in powerpc, rather then adding config for powerpc, entirely remove config check from bpf_read_branch_records function as suggested by Peter Zijlstra

Quoted 3 lines

what will be returned for architectures that don't support branch records? Will it be zero instead of -ENOENT?

Hi Andrii,
Incase any architecture doesn't support branch records and if it
tries to do branch sampling with sample type as PERF_SAMPLE_BRANCH_STACK, perf_event_open itself will fail.

And even if, perf_event_open succeeds we have appropriate checks in bpf_read_branch_records function, which will return -EINVAL for those architectures.

Reference from linux/kernel/trace/bpf_trace.c

Here, br_stack will be empty, for unsupported architectures.

BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,

	   void *, buf, u32, size, u64, flags)
{
.....
	if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
		return -EINVAL;

	if (unlikely(!br_stack))
		return -EINVAL;

....
}

Thanks,
Kajol Jain

Quoted 31 lines
Quoted 31 lines

- Link to the v1 patch: https://lkml.org/lkml/2021/11/14/434

 kernel/trace/bpf_trace.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/trace/bpf_trace.c +0 -5
index 7396488793ff..5e445985c6b4 100644
@@ -1402,9 +1402,6 @@ static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
           void *, buf, u32, size, u64, flags)
 {
-#ifndef CONFIG_X86
-       return -ENOENT;
-#else
        static const u32 br_entry_size = sizeof(struct perf_branch_entry);
        struct perf_branch_stack *br_stack = ctx->data->br_stack;
        u32 to_copy;
@@ -1425,7 +1422,6 @@ BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
        memcpy(buf, br_stack->entries, to_copy);
 
        return to_copy;
-#endif
 }
 
 static const struct bpf_func_proto bpf_read_branch_records_proto = {
--

2.27.0

maintainer
#3 lore raw
Quoted 72 lines · kajoljain
Quoted 54 lines · Andrii Nakryiko
Quoted 46 lines · Kajol Jain

Branch data available to bpf programs can be very useful to get stack traces out of userspace application.

Commit fff7b64355ea ("bpf: Add bpf_read_branch_records() helper") added bpf support to capture branch records in x86. Enable this feature for other architectures as well by removing check specific to x86. Incase any platform didn't support branch stack, it will return with -EINVAL.

Selftest 'perf_branches' result on power9 machine with branch stacks support.

Before this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:FAIL
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:FAIL
Summary: 0/1 PASSED, 0 SKIPPED, 1 FAILED

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:OK
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/2 PASSED, 0 SKIPPED, 0 FAILED

Selftest 'perf_branches' result on power9 machine which doesn't support branch stack

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:SKIP
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/1 PASSED, 1 SKIPPED, 0 FAILED

Fixes fff7b64355eac ("bpf: Add bpf_read_branch_records() helper")

---

Tested this patch changes on power9 machine using selftest 'perf branches' which is added in commit 67306f84ca78 ("selftests/bpf:
Add bpf_read_branch_records()")

Quoted 6 lines

Changelog:
v1 -> v2
- Inorder to add bpf support to capture branch record in powerpc, rather then adding config for powerpc, entirely remove config check from bpf_read_branch_records function as suggested by Peter Zijlstra

Quoted 2 lines

what will be returned for architectures that don't support branch records? Will it be zero instead of -ENOENT?

Hi Andrii,
Incase any architecture doesn't support branch records and if it
tries to do branch sampling with sample type as PERF_SAMPLE_BRANCH_STACK, perf_event_open itself will fail.

And even if, perf_event_open succeeds we have appropriate checks in bpf_read_branch_records function, which will return -EINVAL for those architectures.

Reference from linux/kernel/trace/bpf_trace.c

Here, br_stack will be empty, for unsupported architectures.

BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,

Quoted 8 lines
	   void *, buf, u32, size, u64, flags)
{
.....
	if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
		return -EINVAL;

	if (unlikely(!br_stack))
		return -EINVAL;

In that case for unsupported archs we should probably bail out with -ENOENT here as helper doc says '**-ENOENT** if architecture does not support branch records' (see bpf_read_branch_records() doc in include/uapi/linux/bpf.h).

Quoted 5 lines

....
}

Thanks,
Kajol Jain

#4 lore raw
Quoted 80 lines · Daniel Borkmann
Quoted 78 lines · kajoljain
Quoted 54 lines · Andrii Nakryiko
Quoted 46 lines · Kajol Jain

Branch data available to bpf programs can be very useful to get stack traces out of userspace application.

Commit fff7b64355ea ("bpf: Add bpf_read_branch_records() helper") added bpf support to capture branch records in x86. Enable this feature for other architectures as well by removing check specific to x86. Incase any platform didn't support branch stack, it will return with -EINVAL.

Selftest 'perf_branches' result on power9 machine with branch stacks support.

Before this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:FAIL
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:FAIL
Summary: 0/1 PASSED, 0 SKIPPED, 1 FAILED

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:OK
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/2 PASSED, 0 SKIPPED, 0 FAILED

Selftest 'perf_branches' result on power9 machine which doesn't support branch stack

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:SKIP
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/1 PASSED, 1 SKIPPED, 0 FAILED

Fixes fff7b64355eac ("bpf: Add bpf_read_branch_records() helper")

---

Tested this patch changes on power9 machine using selftest 'perf branches' which is added in commit 67306f84ca78 ("selftests/bpf:
Add bpf_read_branch_records()")

Quoted 6 lines

Changelog:
v1 -> v2
- Inorder to add bpf support to capture branch record in powerpc, rather then adding config for powerpc, entirely remove config check from bpf_read_branch_records function as suggested by Peter Zijlstra

Quoted 2 lines

what will be returned for architectures that don't support branch records? Will it be zero instead of -ENOENT?

Hi Andrii,
Incase any architecture doesn't support branch records and if it
tries to do branch sampling with sample type as PERF_SAMPLE_BRANCH_STACK, perf_event_open itself will fail.

And even if, perf_event_open succeeds we have appropriate checks in bpf_read_branch_records function, which will return -EINVAL for those architectures.

Reference from linux/kernel/trace/bpf_trace.c

Here, br_stack will be empty, for unsupported architectures.

BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,

         void *, buf, u32, size, u64, flags)

{
.....
if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
return -EINVAL;

Quoted 6 lines
Quoted 2 lines
      if (unlikely(!br_stack))
              return -EINVAL;

In that case for unsupported archs we should probably bail out with -ENOENT here as helper doc says '**-ENOENT** if architecture does not support branch records' (see bpf_read_branch_records() doc in include/uapi/linux/bpf.h).

Yep, I think so too.

Quoted 6 lines
Quoted 6 lines

....
}

Thanks,
Kajol Jain

author
#5 lore raw
Quoted 81 lines · Andrii Nakryiko
Quoted 80 lines · Daniel Borkmann
Quoted 78 lines · kajoljain
Quoted 54 lines · Andrii Nakryiko
Quoted 46 lines · Kajol Jain

Branch data available to bpf programs can be very useful to get stack traces out of userspace application.

Commit fff7b64355ea ("bpf: Add bpf_read_branch_records() helper") added bpf support to capture branch records in x86. Enable this feature for other architectures as well by removing check specific to x86. Incase any platform didn't support branch stack, it will return with -EINVAL.

Selftest 'perf_branches' result on power9 machine with branch stacks support.

Before this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:FAIL
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:FAIL
Summary: 0/1 PASSED, 0 SKIPPED, 1 FAILED

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:OK
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/2 PASSED, 0 SKIPPED, 0 FAILED

Selftest 'perf_branches' result on power9 machine which doesn't support branch stack

After this patch changes:
[command]# ./test_progs -t perf_branches
#88/1 perf_branches/perf_branches_hw:SKIP
#88/2 perf_branches/perf_branches_no_hw:OK
#88 perf_branches:OK
Summary: 1/1 PASSED, 1 SKIPPED, 0 FAILED

Fixes fff7b64355eac ("bpf: Add bpf_read_branch_records() helper")

---

Tested this patch changes on power9 machine using selftest 'perf branches' which is added in commit 67306f84ca78 ("selftests/bpf:
Add bpf_read_branch_records()")

Quoted 6 lines

Changelog:
v1 -> v2
- Inorder to add bpf support to capture branch record in powerpc, rather then adding config for powerpc, entirely remove config check from bpf_read_branch_records function as suggested by Peter Zijlstra

Quoted 2 lines

what will be returned for architectures that don't support branch records? Will it be zero instead of -ENOENT?

Hi Andrii,
Incase any architecture doesn't support branch records and if it
tries to do branch sampling with sample type as PERF_SAMPLE_BRANCH_STACK, perf_event_open itself will fail.

And even if, perf_event_open succeeds we have appropriate checks in bpf_read_branch_records function, which will return -EINVAL for those architectures.

Reference from linux/kernel/trace/bpf_trace.c

Here, br_stack will be empty, for unsupported architectures.

BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,

         void *, buf, u32, size, u64, flags)

{
.....
if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
return -EINVAL;

Quoted 9 lines
Quoted 6 lines
Quoted 2 lines
      if (unlikely(!br_stack))
              return -EINVAL;

In that case for unsupported archs we should probably bail out with -ENOENT here as helper doc says '**-ENOENT** if architecture does not support branch records' (see bpf_read_branch_records() doc in include/uapi/linux/bpf.h).

Yep, I think so too.

Hi Andrii/Daniel,
I agree, changing return type to -ENOENT make sense, I will update
in next version of this patch.

Thanks,
Kajol Jain

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

....
}

Thanks,
Kajol Jain