It turns out PERF_IOC_FLAG_GROUP was broken from 75f937f24bd9 (in Linux 2.6.31, the initial perf_event release) until 724b6daa1 (Linux 3.4). I've done some extensive kernel source code digging plus running tests of various kernels and I hope the info presented is accurate now. Signed-off-by: Vince Weaver <vincent.weaver@xxxxxxxxx> diff --git a/man2/perf_event_open.2 b/man2/perf_event_open.2 index 504e801..ff7e4af 100644 --- a/man2/perf_event_open.2 +++ b/man2/perf_event_open.2 @@ -1803,7 +1803,8 @@ file descriptor argument. If the .B PERF_IOC_FLAG_GROUP bit is set in the ioctl argument, then all events in a group are -enabled, even if the event specified is not the group leader. +enabled, even if the event specified is not the group leader +(note this flag was broken until Linux 3.4). .TP .B PERF_EVENT_IOC_DISABLE Disables the individual counter or event group specified by the @@ -1819,7 +1820,8 @@ stops that counter from counting but doesn't affect any other counter. If the .B PERF_IOC_FLAG_GROUP bit is set in the ioctl argument, then all events in a group are -disabled, even if the event specified is not the group leader. +disabled, even if the event specified is not the group leader +(note this flag was broken until Linux 3.4). .TP .B PERF_EVENT_IOC_REFRESH Non-inherited overflow counters can use this @@ -1848,14 +1850,9 @@ values. If the .B PERF_IOC_FLAG_GROUP bit is set in the ioctl argument, then all events in a group are -reset, even if the event specified is not the group leader. +reset, even if the event specified is not the group leader +(note this flag was broken until Linux 3.4). -If the -.B PERF_IOC_FLAG_GROUP -bit is not set, then the behavior is somwhat unexpected: -when sent to a group leader only the leader is reset -(children are left alone); -when sent to a child all events in a group are reset. .TP .B PERF_EVENT_IOC_PERIOD IOC_PERIOD is the command to update the period; it @@ -2042,6 +2039,12 @@ Linux 2.6.36 and Linux 3.0 that ignores the was chosen if the union has a non-zero value in it. +From Linux 2.6.31 to Linux 3.4 the +.B PERF_IOC_FLAG_GROUP +ioctl argument was broken and would repeatedly operate +on the event specified rather than iterating across +all sibling events in a group. + Always double-check your results! Various generalized events have had wrong values. For example, retired branches measured -- To unsubscribe from this list: send the line "unsubscribe linux-man" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
