linux.perf config question - followers

[sifangli-labs]

Hi, I can collect callstack samples using the following cfg at 97Hz. My further goal (not sure if supported) is to collect callstack sample when we have page faults - so I added the "followers" sections - but surprisingly, all the prev callback are gone in the trace when uncommented, and I do not see any additional effects of the two "followers" sections. BTW, the "page fault user" events are visible in the ftrace tracks in the collected pftrace.

My question is if my goal is doable or if the cfg has problems, thx.

data_sources {
  config {
    name: "linux.perf"
    perf_event_config {
      timebase {
        counter: SW_CPU_CLOCK
        frequency: 97
        timestamp_clock: PERF_CLOCK_BOOTTIME
      }
      # cannot enable the folowers below because the callstack is gone then
      # followers {
      #   counter: SW_PAGE_FAULTS
      #   tracepoint {
      #     name: "exceptions/page_fault_user"
      #   }
      # }
      # followers {
      #   counter: SW_PAGE_FAULTS
      #   tracepoint {
      #     name: "exceptions/page_fault_kernel"
      #   }
      # }
      callstack_sampling {

[LalitMaganti]

The correct way to "collect stack samples based on multiple different tracepoints" should be done by creating multiple data sources, each one targeting the event you want to collect. Here's an example from a config we use internally [1]

The use of followers for this is a misunderstanding of the API: followers are a very advanced feature and are rarely useful: they require deep knowledge of both perf and the hardware you're running on to figure out whether what you're doing makes sense.

[1]

      data_sources {
        config {
          name: "linux.perf"
          target_buffer: 0
          perf_event_config {
            unwind_state_clear_period_ms: 120000
            max_daemon_memory_kb: 524288
            timebase {
              frequency: 50
              tracepoint {
                name: "binder:binder_return"
                filter: "(cmd == 0x7212) || (cmd == 0x7214)"
              }
              name: "binder_to_frozen"
            }
            callstack_sampling {
              scope {
                exclude_pid: 0
              }
              kernel_frames: true
            }
          }
        }
      }
      data_sources {
        config {
          name: "linux.perf"
          target_buffer: 0
          perf_event_config {
            unwind_state_clear_period_ms: 120000
            max_daemon_memory_kb: 524288
            timebase {
              frequency: 50
              tracepoint {
                name: "f2fs:f2fs_set_page_dirty"
                filter: "dirty == 0"
              }
            }
            callstack_sampling {
              scope {
                exclude_pid: 0
              }
              kernel_frames: true
            }
          }
        }
      }
      data_sources {
        config {
          name: "linux.perf"
          target_buffer: 0
          perf_event_config {
            unwind_state_clear_period_ms: 120000
            max_daemon_memory_kb: 524288
            timebase {
              frequency: 50
              counter: SW_PAGE_FAULTS_MIN
              timestamp_clock: PERF_CLOCK_BOOTTIME
            }
            callstack_sampling {
              scope {
                exclude_pid: 0
              }
              kernel_frames: true
            }
          }
        }
      }