Re: [PATCH V2 2/2] bpf: control a set of perf events by creating a new ioctl PERF_EVENT_IOC_SET_ENABLER

[xiakaixu <xiakaixu@huawei.com>]

于 2015/10/15 5:28, Alexei Starovoitov 写道:
> On 10/14/15 5:37 AM, Kaixu Xia wrote:
>> +    event->p_sample_disable = &enabler_event->sample_disable;
> 
> I don't like it as a concept and it's buggy implementation.
> What happens here when enabler is alive, but other event is destroyed?
> 
>> --- a/kernel/trace/bpf_trace.c
>> +++ b/kernel/trace/bpf_trace.c
>> @@ -221,9 +221,12 @@ static u64 bpf_perf_event_sample_control(u64 r1, u64 index, u64 flag, u64 r4, u6
>>       struct bpf_array *array = container_of(map, struct bpf_array, map);
>>       struct perf_event *event;
>>
>> -    if (unlikely(index >= array->map.max_entries))
>> +    if (unlikely(index > array->map.max_entries))
>>           return -E2BIG;
>>
>> +    if (index == array->map.max_entries)
>> +        index = 0;
> 
> what is this hack for ?
> 
> Either use notification and user space disable or
> call bpf_perf_event_sample_control() manually for each cpu.

I will discard current implemention that controlling a set of
perf events by the 'enabler' event. Call bpf_perf_event_sample_control()
manually for each cpu is fine. Maybe we can add a loop to control all the
events stored in maps by judging the index, OK?
> 
> 
> 
> .
>