perf/x86: Fix event/group validation
authorPeter Zijlstra <peterz@infradead.org>
Thu, 21 May 2015 08:57:13 +0000 (10:57 +0200)
committerIngo Molnar <mingo@kernel.org>
Wed, 27 May 2015 06:46:44 +0000 (08:46 +0200)
Commit 43b4578071c0 ("perf/x86: Reduce stack usage of
x86_schedule_events()") violated the rule that 'fake' scheduling; as
used for event/group validation; should not change the event state.

This went mostly un-noticed because repeated calls of
x86_pmu::get_event_constraints() would give the same result. And
x86_pmu::put_event_constraints() would mostly not do anything.

Commit e979121b1b15 ("perf/x86/intel: Implement cross-HT corruption
bug workaround") made the situation much worse by actually setting the
event->hw.constraint value to NULL, so when validation and actual
scheduling interact we get NULL ptr derefs.

Fix it by removing the constraint pointer from the event and move it
back to an array, this time in cpuc instead of on the stack.

validate_group()
  x86_schedule_events()
    event->hw.constraint = c; # store

      <context switch>
        perf_task_event_sched_in()
          ...
            x86_schedule_events();
              event->hw.constraint = c2; # store

              ...

              put_event_constraints(event); # assume failure to schedule
                intel_put_event_constraints()
                  event->hw.constraint = NULL;

      <context switch end>

    c = event->hw.constraint; # read -> NULL

    if (!test_bit(hwc->idx, c->idxmsk)) # <- *BOOM* NULL deref

This in particular is possible when the event in question is a
cpu-wide event and group-leader, where the validate_group() tries to
add an event to the group.

Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Hunter <ahh@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Maria Dimakopoulou <maria.n.dimakopoulou@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 43b4578071c0 ("perf/x86: Reduce stack usage of x86_schedule_events()")
Fixes: e979121b1b15 ("perf/x86/intel: Implement cross-HT corruption bug workaround")
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/kernel/cpu/perf_event.c
arch/x86/kernel/cpu/perf_event.h
arch/x86/kernel/cpu/perf_event_intel.c
arch/x86/kernel/cpu/perf_event_intel_ds.c
arch/x86/kernel/cpu/perf_event_intel_uncore.c
arch/x86/kernel/cpu/perf_event_intel_uncore.h
include/linux/perf_event.h

index 87848eb..1664eee 100644 (file)
@@ -620,7 +620,7 @@ struct sched_state {
 struct perf_sched {
        int                     max_weight;
        int                     max_events;
-       struct perf_event       **events;
+       struct event_constraint **constraints;
        struct sched_state      state;
        int                     saved_states;
        struct sched_state      saved[SCHED_STATES_MAX];
@@ -629,7 +629,7 @@ struct perf_sched {
 /*
  * Initialize interator that runs through all events and counters.
  */
-static void perf_sched_init(struct perf_sched *sched, struct perf_event **events,
+static void perf_sched_init(struct perf_sched *sched, struct event_constraint **constraints,
                            int num, int wmin, int wmax)
 {
        int idx;
@@ -637,10 +637,10 @@ static void perf_sched_init(struct perf_sched *sched, struct perf_event **events
        memset(sched, 0, sizeof(*sched));
        sched->max_events       = num;
        sched->max_weight       = wmax;
-       sched->events           = events;
+       sched->constraints      = constraints;
 
        for (idx = 0; idx < num; idx++) {
-               if (events[idx]->hw.constraint->weight == wmin)
+               if (constraints[idx]->weight == wmin)
                        break;
        }
 
@@ -687,7 +687,7 @@ static bool __perf_sched_find_counter(struct perf_sched *sched)
        if (sched->state.event >= sched->max_events)
                return false;
 
-       c = sched->events[sched->state.event]->hw.constraint;
+       c = sched->constraints[sched->state.event];
        /* Prefer fixed purpose counters */
        if (c->idxmsk64 & (~0ULL << INTEL_PMC_IDX_FIXED)) {
                idx = INTEL_PMC_IDX_FIXED;
@@ -745,7 +745,7 @@ static bool perf_sched_next_event(struct perf_sched *sched)
                        if (sched->state.weight > sched->max_weight)
                                return false;
                }
-               c = sched->events[sched->state.event]->hw.constraint;
+               c = sched->constraints[sched->state.event];
        } while (c->weight != sched->state.weight);
 
        sched->state.counter = 0;       /* start with first counter */
@@ -756,12 +756,12 @@ static bool perf_sched_next_event(struct perf_sched *sched)
 /*
  * Assign a counter for each event.
  */
-int perf_assign_events(struct perf_event **events, int n,
+int perf_assign_events(struct event_constraint **constraints, int n,
                        int wmin, int wmax, int *assign)
 {
        struct perf_sched sched;
 
-       perf_sched_init(&sched, events, n, wmin, wmax);
+       perf_sched_init(&sched, constraints, n, wmin, wmax);
 
        do {
                if (!perf_sched_find_counter(&sched))
@@ -788,9 +788,9 @@ int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign)
                x86_pmu.start_scheduling(cpuc);
 
        for (i = 0, wmin = X86_PMC_IDX_MAX, wmax = 0; i < n; i++) {
-               hwc = &cpuc->event_list[i]->hw;
+               cpuc->event_constraint[i] = NULL;
                c = x86_pmu.get_event_constraints(cpuc, i, cpuc->event_list[i]);
-               hwc->constraint = c;
+               cpuc->event_constraint[i] = c;
 
                wmin = min(wmin, c->weight);
                wmax = max(wmax, c->weight);
@@ -801,7 +801,7 @@ int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign)
         */
        for (i = 0; i < n; i++) {
                hwc = &cpuc->event_list[i]->hw;
-               c = hwc->constraint;
+               c = cpuc->event_constraint[i];
 
                /* never assigned */
                if (hwc->idx == -1)
@@ -821,9 +821,10 @@ int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign)
        }
 
        /* slow path */
-       if (i != n)
-               unsched = perf_assign_events(cpuc->event_list, n, wmin,
+       if (i != n) {
+               unsched = perf_assign_events(cpuc->event_constraint, n, wmin,
                                             wmax, assign);
+       }
 
        /*
         * In case of success (unsched = 0), mark events as committed,
@@ -840,7 +841,7 @@ int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign)
                        e = cpuc->event_list[i];
                        e->hw.flags |= PERF_X86_EVENT_COMMITTED;
                        if (x86_pmu.commit_scheduling)
-                               x86_pmu.commit_scheduling(cpuc, e, assign[i]);
+                               x86_pmu.commit_scheduling(cpuc, i, assign[i]);
                }
        }
 
@@ -1292,8 +1293,10 @@ static void x86_pmu_del(struct perf_event *event, int flags)
                x86_pmu.put_event_constraints(cpuc, event);
 
        /* Delete the array entry. */
-       while (++i < cpuc->n_events)
+       while (++i < cpuc->n_events) {
                cpuc->event_list[i-1] = cpuc->event_list[i];
+               cpuc->event_constraint[i-1] = cpuc->event_constraint[i];
+       }
        --cpuc->n_events;
 
        perf_event_update_userpage(event);
index 6ac5cb7..fdfaab7 100644 (file)
@@ -172,7 +172,10 @@ struct cpu_hw_events {
                                             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 */
+       struct event_constraint *event_constraint[X86_PMC_IDX_MAX];
+
 
        unsigned int            group_flag;
        int                     is_fake;
@@ -519,9 +522,7 @@ struct x86_pmu {
        void            (*put_event_constraints)(struct cpu_hw_events *cpuc,
                                                 struct perf_event *event);
 
-       void            (*commit_scheduling)(struct cpu_hw_events *cpuc,
-                                            struct perf_event *event,
-                                            int cntr);
+       void            (*commit_scheduling)(struct cpu_hw_events *cpuc, int idx, int cntr);
 
        void            (*start_scheduling)(struct cpu_hw_events *cpuc);
 
@@ -717,7 +718,7 @@ static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
 
 void x86_pmu_enable_all(int added);
 
-int perf_assign_events(struct perf_event **events, int n,
+int perf_assign_events(struct event_constraint **constraints, int n,
                        int wmin, int wmax, int *assign);
 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
 
index 3998131..7a58fb5 100644 (file)
@@ -2106,7 +2106,7 @@ static struct event_constraint *
 intel_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
                            struct perf_event *event)
 {
-       struct event_constraint *c1 = event->hw.constraint;
+       struct event_constraint *c1 = cpuc->event_constraint[idx];
        struct event_constraint *c2;
 
        /*
@@ -2188,8 +2188,6 @@ intel_put_shared_regs_event_constraints(struct cpu_hw_events *cpuc,
 static void intel_put_event_constraints(struct cpu_hw_events *cpuc,
                                        struct perf_event *event)
 {
-       struct event_constraint *c = event->hw.constraint;
-
        intel_put_shared_regs_event_constraints(cpuc, event);
 
        /*
@@ -2197,19 +2195,14 @@ static void intel_put_event_constraints(struct cpu_hw_events *cpuc,
         * all events are subject to and must call the
         * put_excl_constraints() routine
         */
-       if (c && cpuc->excl_cntrs)
+       if (cpuc->excl_cntrs)
                intel_put_excl_constraints(cpuc, event);
-
-       /* cleanup dynamic constraint */
-       if (c && (c->flags & PERF_X86_EVENT_DYNAMIC))
-               event->hw.constraint = NULL;
 }
 
-static void intel_commit_scheduling(struct cpu_hw_events *cpuc,
-                                   struct perf_event *event, int cntr)
+static void intel_commit_scheduling(struct cpu_hw_events *cpuc, int idx, int cntr)
 {
        struct intel_excl_cntrs *excl_cntrs = cpuc->excl_cntrs;
-       struct event_constraint *c = event->hw.constraint;
+       struct event_constraint *c = cpuc->event_constraint[idx];
        struct intel_excl_states *xlo, *xl;
        int tid = cpuc->excl_thread_id;
        int o_tid = 1 - tid;
index 813f75d..7f73b35 100644 (file)
@@ -706,9 +706,9 @@ void intel_pmu_pebs_disable(struct perf_event *event)
 
        cpuc->pebs_enabled &= ~(1ULL << hwc->idx);
 
-       if (event->hw.constraint->flags & PERF_X86_EVENT_PEBS_LDLAT)
+       if (event->hw.flags & PERF_X86_EVENT_PEBS_LDLAT)
                cpuc->pebs_enabled &= ~(1ULL << (hwc->idx + 32));
-       else if (event->hw.constraint->flags & PERF_X86_EVENT_PEBS_ST)
+       else if (event->hw.flags & PERF_X86_EVENT_PEBS_ST)
                cpuc->pebs_enabled &= ~(1ULL << 63);
 
        if (cpuc->enabled)
index c635b8b..ec2ba57 100644 (file)
@@ -365,9 +365,8 @@ static int uncore_assign_events(struct intel_uncore_box *box, int assign[], int
        bitmap_zero(used_mask, UNCORE_PMC_IDX_MAX);
 
        for (i = 0, wmin = UNCORE_PMC_IDX_MAX, wmax = 0; i < n; i++) {
-               hwc = &box->event_list[i]->hw;
                c = uncore_get_event_constraint(box, box->event_list[i]);
-               hwc->constraint = c;
+               box->event_constraint[i] = c;
                wmin = min(wmin, c->weight);
                wmax = max(wmax, c->weight);
        }
@@ -375,7 +374,7 @@ static int uncore_assign_events(struct intel_uncore_box *box, int assign[], int
        /* fastpath, try to reuse previous register */
        for (i = 0; i < n; i++) {
                hwc = &box->event_list[i]->hw;
-               c = hwc->constraint;
+               c = box->event_constraint[i];
 
                /* never assigned */
                if (hwc->idx == -1)
@@ -395,7 +394,7 @@ static int uncore_assign_events(struct intel_uncore_box *box, int assign[], int
        }
        /* slow path */
        if (i != n)
-               ret = perf_assign_events(box->event_list, n,
+               ret = perf_assign_events(box->event_constraint, n,
                                         wmin, wmax, assign);
 
        if (!assign || ret) {
index 6c8c1e7..f789ec9 100644 (file)
@@ -97,6 +97,7 @@ struct intel_uncore_box {
        atomic_t refcnt;
        struct perf_event *events[UNCORE_PMC_IDX_MAX];
        struct perf_event *event_list[UNCORE_PMC_IDX_MAX];
+       struct event_constraint *event_constraint[UNCORE_PMC_IDX_MAX];
        unsigned long active_mask[BITS_TO_LONGS(UNCORE_PMC_IDX_MAX)];
        u64 tags[UNCORE_PMC_IDX_MAX];
        struct pci_dev *pci_dev;
index 61992cf..d8a82a8 100644 (file)
@@ -92,8 +92,6 @@ struct hw_perf_event_extra {
        int             idx;    /* index in shared_regs->regs[] */
 };
 
-struct event_constraint;
-
 /**
  * struct hw_perf_event - performance event hardware details:
  */
@@ -112,8 +110,6 @@ struct hw_perf_event {
 
                        struct hw_perf_event_extra extra_reg;
                        struct hw_perf_event_extra branch_reg;
-
-                       struct event_constraint *constraint;
                };
                struct { /* software */
                        struct hrtimer  hrtimer;