On Mon, 24 Feb 2014, Peter Zijlstra wrote: > On Fri, Feb 21, 2014 at 03:18:38PM -0500, Vince Weaver wrote: > > I've applied the patch and have been unable to trigger the warning with > > either my testcase or a few hours of fuzzing. > > Yay. > > > My only comment on the patch is it could always use some comments. > > > > The perf_event code is really hard to follow as is, without adding > > more uncommented special cases. > > Does the below help a bit? Or is there anywhere in particular you want > more comments?
yes, every little bit helps. While chasing these fuzzer-related bugs I end up deep in the perf_event code and many of the routines have no comments at all. Eventually I have to dig out the K+R book to figure out order precendece of ++ prefix operators, have at least 2-3 different files open in editors, plus a bunch of firefox tabs open to http://lxr.free-electrons.com, and even then I misunderstand the code a lot. Vince > > --- > Subject: perf, x86: Add a few more comments > From: Peter Zijlstra <pet...@infradead.org> > Date: Mon Feb 24 12:26:21 CET 2014 > > Add a few comments on the ->add(), ->del() and ->*_txn() > implementation. > > Requested-by: Vince Weaver <vincent.wea...@maine.edu> > Signed-off-by: Peter Zijlstra <pet...@infradead.org> > --- > arch/x86/kernel/cpu/perf_event.c | 49 > +++++++++++++++++++++++++++------------ > arch/x86/kernel/cpu/perf_event.h | 8 +++--- > 2 files changed, 39 insertions(+), 18 deletions(-) > > --- a/arch/x86/kernel/cpu/perf_event.c > +++ b/arch/x86/kernel/cpu/perf_event.c > @@ -892,7 +892,6 @@ static void x86_pmu_enable(struct pmu *p > * hw_perf_group_sched_in() or x86_pmu_enable() > * > * step1: save events moving to new counters > - * step2: reprogram moved events into new counters > */ > for (i = 0; i < n_running; i++) { > event = cpuc->event_list[i]; > @@ -918,6 +917,9 @@ static void x86_pmu_enable(struct pmu *p > x86_pmu_stop(event, PERF_EF_UPDATE); > } > > + /* > + * step2: reprogram moved events into new counters > + */ > for (i = 0; i < cpuc->n_events; i++) { > event = cpuc->event_list[i]; > hwc = &event->hw; > @@ -1043,7 +1045,7 @@ static int x86_pmu_add(struct perf_event > /* > * If group events scheduling transaction was started, > * skip the schedulability test here, it will be performed > - * at commit time (->commit_txn) as a whole > + * at commit time (->commit_txn) as a whole. > */ > if (cpuc->group_flag & PERF_EVENT_TXN) > goto done_collect; > @@ -1058,6 +1060,10 @@ static int x86_pmu_add(struct perf_event > memcpy(cpuc->assign, assign, n*sizeof(int)); > > done_collect: > + /* > + * Commit the collect_events() state. See x86_pmu_del() and > + * x86_pmu_*_txn(). > + */ > cpuc->n_events = n; > cpuc->n_added += n - n0; > cpuc->n_txn += n - n0; > @@ -1183,28 +1189,38 @@ static void x86_pmu_del(struct perf_even > * If we're called during a txn, we don't need to do anything. > * The events never got scheduled and ->cancel_txn will truncate > * the event_list. > + * > + * XXX assumes any ->del() called during a TXN will only be on > + * an event added during that same TXN. > */ > if (cpuc->group_flag & PERF_EVENT_TXN) > return; > > + /* > + * Not a TXN, therefore cleanup properly. > + */ > x86_pmu_stop(event, PERF_EF_UPDATE); > > for (i = 0; i < cpuc->n_events; i++) { > - if (event == cpuc->event_list[i]) { > - > - if (i >= cpuc->n_events - cpuc->n_added) > - --cpuc->n_added; > + if (event == cpuc->event_list[i]) > + break; > + } > > - if (x86_pmu.put_event_constraints) > - x86_pmu.put_event_constraints(cpuc, event); > + if (WARN_ON_ONCE(i == cpuc->n_events)) /* called ->del() without > ->add() ? */ > + return; > > - while (++i < cpuc->n_events) > - cpuc->event_list[i-1] = cpuc->event_list[i]; > + /* If we have a newly added event; make sure to decrease n_added. */ > + if (i >= cpuc->n_events - cpuc->n_added) > + --cpuc->n_added; > + > + if (x86_pmu.put_event_constraints) > + x86_pmu.put_event_constraints(cpuc, event); > + > + /* Delete the array entry. */ > + while (++i < cpuc->n_events) > + cpuc->event_list[i-1] = cpuc->event_list[i]; > + --cpuc->n_events; > > - --cpuc->n_events; > - break; > - } > - } > perf_event_update_userpage(event); > } > > @@ -1598,7 +1614,8 @@ static void x86_pmu_cancel_txn(struct pm > { > __this_cpu_and(cpu_hw_events.group_flag, ~PERF_EVENT_TXN); > /* > - * Truncate the collected events. > + * Truncate collected array by the number of events added in this > + * transaction. See x86_pmu_add() and x86_pmu_*_txn(). > */ > __this_cpu_sub(cpu_hw_events.n_added, > __this_cpu_read(cpu_hw_events.n_txn)); > __this_cpu_sub(cpu_hw_events.n_events, > __this_cpu_read(cpu_hw_events.n_txn)); > @@ -1609,6 +1626,8 @@ static void x86_pmu_cancel_txn(struct pm > * Commit group events scheduling transaction > * Perform the group schedulability test as a whole > * Return 0 if success > + * > + * Does not cancel the transaction on failure; expects the caller to do this. > */ > static int x86_pmu_commit_txn(struct pmu *pmu) > { > --- a/arch/x86/kernel/cpu/perf_event.h > +++ b/arch/x86/kernel/cpu/perf_event.h > @@ -130,9 +130,11 @@ struct cpu_hw_events { > unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; > int enabled; > > - int n_events; > - int n_added; > - int n_txn; > + int n_events; /* the # of events in the below > arrays */ > + int n_added; /* the # last events in the below > arrays; > + they've never been enabled yet */ > + int n_txn; /* the # last events in the below > arrays; > + added in the current transaction */ > int assign[X86_PMC_IDX_MAX]; /* event to counter > assignment */ > u64 tags[X86_PMC_IDX_MAX]; > struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled > order */ > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
