Re: [PATCH v2] bpf: Remove config check to enable bpf support for branch records

[Daniel Borkmann <daniel@iogearbox.net>]

On 11/19/21 10:35 AM, kajoljain wrote:
> On 11/19/21 4:18 AM, Andrii Nakryiko wrote:
>> On Thu, Nov 18, 2021 at 5:10 AM Kajol Jain <kjain@linux.ibm.com> wrote:
>>>
>>> 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")
>>> Suggested-by: Peter Zijlstra <peterz@infradead.org>
>>> Signed-off-by: Kajol Jain <kjain@linux.ibm.com>
>>> ---
>>>
>>> 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
>>
>> 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;

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).

> ....
> }
> 
> Thanks,
> Kajol Jain