Branch data available to bpf programs can be very useful to get
stack traces out of userspace applications.
Commit fff7b64355ea ("bpf: Add bpf_read_branch_records() helper")
added bpf support to capture branch records in x86. Enable this feature
for powerpc as well.
Commit 67306f84ca78 ("selftests/bpf: Add bpf_read_branch_records()
selftest") adds selftest corresponding to bpf branch read
function bpf_read_branch_records(). Used this selftest to
test bpf support, for reading branch records in powerpc.
Selftest result in power9 box before this patch changes:
[command]# ./test_progs -t perf_branches
Failed to load bpf_testmod.ko into the kernel: -8
WARNING! Selftests relying on bpf_testmod.ko will be skipped.
test_perf_branches_common:PASS:test_perf_branches_load 0 nsec
test_perf_branches_common:PASS:attach_perf_event 0 nsec
test_perf_branches_common:PASS:set_affinity 0 nsec
check_good_sample:PASS:output not valid 0 nsec
check_good_sample:FAIL:read_branches_size err -2
check_good_sample:FAIL:read_branches_stack err -2
check_good_sample:FAIL:read_branches_stack stack bytes written=-2
not multiple of struct size=24
check_good_sample:FAIL:read_branches_global err -2
check_good_sample:FAIL:read_branches_global global bytes written=-2
not multiple of struct size=24
check_good_sample:PASS:read_branches_size 0 nsec
#75/1 perf_branches_hw:FAIL
#75/2 perf_branches_no_hw:OK
#75 perf_branches:FAIL
Summary: 0/1 PASSED, 0 SKIPPED, 2 FAILED
Selftest result in power9 box after this patch changes:
[command]#: ./test_progs -t perf_branches
#75/1 perf_branches_hw:OK
#75/2 perf_branches_no_hw:OK
#75 perf_branches:OK
Summary: 1/2 PASSED, 0 SKIPPED, 0 FAILED
Signed-off-by: Kajol Jain<[email protected]>
---
kernel/trace/bpf_trace.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index fdd14072fc3b..2b7343b64bb7 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -1245,7 +1245,7 @@ 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
+#if !(defined(CONFIG_X86) || defined(CONFIG_PPC64))
return -ENOENT;
#else
static const u32 br_entry_size = sizeof(struct perf_branch_entry);
--
2.27.0
On 11/15/21 5:44 AM, Kajol Jain wrote:
> Branch data available to bpf programs can be very useful to get
> stack traces out of userspace applications.
>
> Commit fff7b64355ea ("bpf: Add bpf_read_branch_records() helper")
> added bpf support to capture branch records in x86. Enable this feature
> for powerpc as well.
>
> Commit 67306f84ca78 ("selftests/bpf: Add bpf_read_branch_records()
> selftest") adds selftest corresponding to bpf branch read
> function bpf_read_branch_records(). Used this selftest to
> test bpf support, for reading branch records in powerpc.
>
> Selftest result in power9 box before this patch changes:
>
> [command]# ./test_progs -t perf_branches
> Failed to load bpf_testmod.ko into the kernel: -8
> WARNING! Selftests relying on bpf_testmod.ko will be skipped.
> test_perf_branches_common:PASS:test_perf_branches_load 0 nsec
> test_perf_branches_common:PASS:attach_perf_event 0 nsec
> test_perf_branches_common:PASS:set_affinity 0 nsec
> check_good_sample:PASS:output not valid 0 nsec
> check_good_sample:FAIL:read_branches_size err -2
> check_good_sample:FAIL:read_branches_stack err -2
> check_good_sample:FAIL:read_branches_stack stack bytes written=-2
> not multiple of struct size=24
> check_good_sample:FAIL:read_branches_global err -2
> check_good_sample:FAIL:read_branches_global global bytes written=-2
> not multiple of struct size=24
> check_good_sample:PASS:read_branches_size 0 nsec
> #75/1 perf_branches_hw:FAIL
> #75/2 perf_branches_no_hw:OK
> #75 perf_branches:FAIL
> Summary: 0/1 PASSED, 0 SKIPPED, 2 FAILED
>
> Selftest result in power9 box after this patch changes:
>
> [command]#: ./test_progs -t perf_branches
> #75/1 perf_branches_hw:OK
> #75/2 perf_branches_no_hw:OK
> #75 perf_branches:OK
> Summary: 1/2 PASSED, 0 SKIPPED, 0 FAILED
>
> Signed-off-by: Kajol Jain<[email protected]>
> ---
> kernel/trace/bpf_trace.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> index fdd14072fc3b..2b7343b64bb7 100644
> --- a/kernel/trace/bpf_trace.c
> +++ b/kernel/trace/bpf_trace.c
> @@ -1245,7 +1245,7 @@ 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
> +#if !(defined(CONFIG_X86) || defined(CONFIG_PPC64))
Can this really be enabled generically? Looking at 3925f46bb590 ("powerpc/perf: Enable
branch stack sampling framework") it says POWER8 [and beyond]. Should there be a generic
Kconfig symbol like ARCH_HAS_BRANCH_RECORDS that can be selected by archs instead?
> return -ENOENT;
> #else
> static const u32 br_entry_size = sizeof(struct perf_branch_entry);
>
On Tue, Nov 16, 2021 at 12:30:07AM +0100, Daniel Borkmann wrote:
> > diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> > index fdd14072fc3b..2b7343b64bb7 100644
> > --- a/kernel/trace/bpf_trace.c
> > +++ b/kernel/trace/bpf_trace.c
> > @@ -1245,7 +1245,7 @@ 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
> > +#if !(defined(CONFIG_X86) || defined(CONFIG_PPC64))
>
> Can this really be enabled generically? Looking at 3925f46bb590 ("powerpc/perf: Enable
> branch stack sampling framework") it says POWER8 [and beyond]. Should there be a generic
> Kconfig symbol like ARCH_HAS_BRANCH_RECORDS that can be selected by archs instead?
I conplained about it before as well. I'd just take it out entirely.
If perf_snapshot_branch_stack isn't implemnted it'll return 0 and then
we'll -Esomething anyway.
On 11/16/21 2:04 PM, Peter Zijlstra wrote:
> On Tue, Nov 16, 2021 at 12:30:07AM +0100, Daniel Borkmann wrote:
>
>>> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
>>> index fdd14072fc3b..2b7343b64bb7 100644
>>> --- a/kernel/trace/bpf_trace.c
>>> +++ b/kernel/trace/bpf_trace.c
>>> @@ -1245,7 +1245,7 @@ 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
>>> +#if !(defined(CONFIG_X86) || defined(CONFIG_PPC64))
>>
>> Can this really be enabled generically? Looking at 3925f46bb590 ("powerpc/perf: Enable
>> branch stack sampling framework") it says POWER8 [and beyond]. Should there be a generic
>> Kconfig symbol like ARCH_HAS_BRANCH_RECORDS that can be selected by archs instead?
>
Hi Peterz/Daniel,
Thanks for reviewing the patch
> I conplained about it before as well. I'd just take it out entirely.
I agree, it make more sense to entirely remove this arch check from
here. Because anyway, incase any arch doesn't support this
functionality, bpf_read_branch_records will return -EINVAL.
>
> If perf_snapshot_branch_stack isn't implemnted it'll return 0 and then
> we'll -Esomething anyway.
In this patch, we are basically adding powerpc support to capture
branch records via bpf_read_branch_records function. We are still
looking into adding support for perf_snapshot_branch_stack for powerpc.
I will send a follow up to remove arch check in bpf_read_branch_records
function.
Thanks,
Kajol Jain
>
>