Date:	Fri, 22 Jan 2010 12:43:01 +0200
From:	Stephane Eranian <eranian@...gle.com>
To:	linux-kernel@...r.kernel.org
Cc:	peterz@...radead.org, mingo@...e.hu, paulus@...ba.org,
	davem@...emloft.net, fweisbec@...il.com, robert.richter@....com,
	perfmon2-devel@...ts.sf.net, eranian@...il.com, eranian@...gle.com
Subject: [PATCH]  perf_events: AMD event scheduling (v1)

	
	This patch adds correct AMD Northbridge event scheduling.
	It must be applied on top of my v5 + v6 incremental event
	scheduling patch.

	AMD Northbridge (NB) events measure L3 and Hypertransport
	activities.  There is a documented restriction on how NB
	events can be programmed (refer to BKDG section 3.12).

	No two cores can use the same counter to measure NB events.
	This patch implements this restriction by maintaining a per
	Northbridge counter allocation table. All cores attached to
	the same NB compete to allocate NB events. Given that you have
	4 counters, this means that at most 1 NB event can be measured by
	all cores. The better alternative is to measure all NB events
	from a single core. Both approaches are possible using this patch.
	If there is more NB events than there are counters, some NB events
	will not be scheduled, e.g., 2 NB events on each core on a 4-core
	package.

	The patch also takes care of hotplug CPU.

	Signed-off-by: Stephane Eranian <eranian@...gle.com>

--
 arch/x86/kernel/cpu/perf_event.c |  252 ++++++++++++++++++++++++++++++++++++++-
 kernel/perf_event.c              |    5 
 2 files changed, 254 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
index a961b1f..a97a744 100644
--- a/arch/x86/kernel/cpu/perf_event.c
+++ b/arch/x86/kernel/cpu/perf_event.c
@@ -69,6 +69,12 @@ struct debug_store {
 	u64	pebs_event_reset[MAX_PEBS_EVENTS];
 };
 
+struct amd_nb {
+        int nb_id; /* Northbridge id */
+	int refcnt; /* refernce count */
+	struct perf_event *owners[X86_PMC_IDX_MAX];
+};
+
 #define BITS_TO_U64(nr) DIV_ROUND_UP(nr, BITS_PER_BYTE * sizeof(u64))
 
 struct event_constraint {
@@ -89,6 +95,7 @@ struct cpu_hw_events {
 	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 amd_nb		*amd_nb;
 };
 
 #define EVENT_CONSTRAINT(c, n, m) { \
@@ -134,6 +141,8 @@ struct x86_pmu {
 
 static struct x86_pmu x86_pmu __read_mostly;
 
+static raw_spinlock_t amd_nb_lock;
+
 static DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events) = {
 	.enabled = 1,
 };
@@ -2199,12 +2208,144 @@ static void intel_get_event_constraints(struct cpu_hw_events *cpuc,
 	bitmap_fill((unsigned long *)idxmsk, x86_pmu.num_events);
 }
 
+/*
+ * AMD64 events are detected based on their event codes.
+ */
+static inline int amd_is_nb_event(struct hw_perf_event *hwc)
+{
+	u64 val = hwc->config;
+        /* event code : bits [35-32] | [7-0] */
+        val = (val >> 24) | ( val & 0xff);
+        return val >= 0x0e0;
+}
+
+static void amd_put_event_constraints(struct cpu_hw_events *cpuc,
+			       struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct perf_event *old;
+	struct amd_nb *nb;
+	int i;
+
+	/*
+	 * only care about NB events
+	 */
+	if(!amd_is_nb_event(hwc))
+		return;
+
+	/*
+	 * NB not initialized
+	 */
+	nb = cpuc->amd_nb;
+	if (!nb)
+		return;
+
+	if (hwc->idx == -1)
+		return;
+
+	/*
+	 * need to scan whole list because event may not have
+	 * been assigned during scheduling
+	 */
+	for(i=0; i < x86_pmu.num_events; i++) {
+		if (nb->owners[i] == event) {
+			old = cmpxchg(nb->owners+i, event, NULL);
+			WARN_ON(old != event);
+			return;
+		}
+	}
+}
+
+/*
+ * AMD64 Northbridge events need special treatment because
+ * counter access needs to be synchronized across all cores
+ * of a package. Refer to BKDG section 3.12
+ *
+ * NB events are events measuring L3 cache, Hypertransport
+ * traffic. They are identified by an event code  >= 0xe0.
+ *
+ * No two cores can be measuring NB events using the same
+ * counter. In other words, for NB events, it is as if there
+ * was only one set of counters per package (or cores sharing
+ * the same NB). Thus, we need to maintain a per-NB * allocation
+ * table. The available slot is propagated using the bitmask.
+ * We provide only one choice for each NB events based on
+ * the fact that only NB events have restrictions. Consequently,
+ * if a counter is available, there is a guarantee the NB event
+ * will be assigned to it. If no slot is available, an empty
+ * bitmask is returned and scheduling fails.
+ *
+ * Note that all cores attached the same NB compete for the same
+ * counters to host NB events, this is why we use atomic ops.
+ *
+ * Given that resources are allocated (cmpxchg), they must be
+ * eventually freed for others to use. This is accomplished by
+ * calling amd_put_event_constraints().
+ *
+ * Non NB events are not impacted by this restriction.
+ */
 static void amd_get_event_constraints(struct cpu_hw_events *cpuc,
 				      struct perf_event *event,
 				      u64 *idxmsk)
 {
-	/* no constraints, means supports all generic counters */
-	bitmap_fill((unsigned long *)idxmsk, x86_pmu.num_events);
+	struct hw_perf_event *hwc = &event->hw;
+	struct amd_nb *nb = cpuc->amd_nb;
+	struct perf_event *old = NULL;
+	int max = x86_pmu.num_events;
+	int i, j, k = -1;
+
+	/*
+	 * clean up vector
+	 */
+	bitmap_zero((unsigned long *)idxmsk, X86_PMC_IDX_MAX);
+
+	/*
+	 * if not NB event or no NB, then no constraints
+	 */
+	if (!amd_is_nb_event(hwc) || !nb) {
+		bitmap_fill((unsigned long *)idxmsk, x86_pmu.num_events);
+		return;
+	}
+	/*
+	 * detect if already present, if so reuse
+	 *
+	 * cannot merge with actual allocation
+	 * because of possible holes
+	 *
+	 * event can already be present yet not assigned (in hwc->idx)
+	 * because of successive calls to x86_schedule_events() from
+	 * hw_perf_group_sched_in() without hw_perf_enable()
+	 */
+	for(i=0; i < max; i++) {
+		/*
+		 * keep track of first free slot
+		 */
+		if (k == -1 && !nb->owners[i])
+			k = i;
+
+		/* already present, reuse */
+		if (nb->owners[i] == event)
+			goto skip;
+	}
+	/*
+	 * not present, so grab a new slot
+	 *
+	 * try to alllcate same counter as before if
+	 * event has already been assigned once. Otherwise,
+	 * try to use free counter k obtained during the 1st
+	 * pass above.
+	 */
+	i = j = hwc->idx != -1 ? hwc->idx : (k == -1 ? 0 : k);
+	do {
+		old = cmpxchg(nb->owners+i, NULL, event);
+		if (!old)
+			break;
+		if (++i == x86_pmu.num_events)
+			i = 0;
+	} while (i != j);
+skip:
+	if (!old)
+		set_bit(i, (unsigned long *)idxmsk);
 }
 
 static int x86_event_sched_in(struct perf_event *event,
@@ -2394,7 +2535,8 @@ static __initconst struct x86_pmu amd_pmu = {
 	.apic			= 1,
 	/* use highest bit to detect overflow */
 	.max_period		= (1ULL << 47) - 1,
-	.get_event_constraints	= amd_get_event_constraints
+	.get_event_constraints	= amd_get_event_constraints,
+	.put_event_constraints	= amd_put_event_constraints
 };
 
 static __init int p6_pmu_init(void)
@@ -2501,6 +2643,87 @@ static __init int intel_pmu_init(void)
 	return 0;
 }
 
+static struct amd_nb *amd_alloc_nb(int cpu, int nb_id)
+{
+        struct amd_nb *nb;
+
+        nb= vmalloc_node(sizeof(struct amd_nb), cpu_to_node(cpu));
+        if (!nb)
+                return NULL;
+
+        memset(nb, 0, sizeof(*nb));
+        nb->nb_id = nb_id;
+        return nb;
+}
+
+static void amd_pmu_cpu_online(int cpu)
+{
+	struct cpu_hw_events *cpu1, *cpu2;
+	struct amd_nb *nb = NULL;
+	int i, nb_id;
+
+	if (boot_cpu_data.x86_max_cores < 2)
+		return;
+
+	/*
+	 * function may be called too early in the
+	 * boot process, in which case nb_id is bogus
+	 *
+	 * for BSP, there is an explicit call from
+	 * amd_pmu_init()
+	 */
+	nb_id = amd_get_nb_id(cpu);
+	if (nb_id == BAD_APICID)
+		return;
+
+	cpu1 = &per_cpu(cpu_hw_events, cpu);
+	cpu1->amd_nb = NULL;
+
+	raw_spin_lock(&amd_nb_lock);
+
+	for_each_online_cpu(i) {
+		cpu2 = &per_cpu(cpu_hw_events, i);
+		nb = cpu2->amd_nb;
+		if (!nb)
+			continue;
+		if (nb->nb_id == nb_id)
+			goto found;
+	}
+
+	nb = amd_alloc_nb(cpu, nb_id);
+	if (!nb) {
+		pr_err("perf_events: failed to allocate NB storage for CPU%d\n", cpu);
+		raw_spin_unlock(&amd_nb_lock);
+		return;
+	}
+found:
+	nb->refcnt++;
+	cpu1->amd_nb = nb;
+
+	raw_spin_unlock(&amd_nb_lock);
+
+	pr_info("CPU%d NB%d ref=%d\n", cpu, nb_id, nb->refcnt);
+}
+
+static void amd_pmu_cpu_offline(int cpu)
+{
+	struct cpu_hw_events *cpuhw;
+
+	if (boot_cpu_data.x86_max_cores < 2)
+		return;
+
+	cpuhw = &per_cpu(cpu_hw_events, cpu);
+
+	raw_spin_lock(&amd_nb_lock);
+
+	if (--cpuhw->amd_nb->refcnt == 0)
+		vfree(cpuhw->amd_nb);
+
+	cpuhw->amd_nb = NULL;
+
+	raw_spin_unlock(&amd_nb_lock);
+}
+
 static __init int amd_pmu_init(void)
 {
 	/* Performance-monitoring supported from K7 and later: */
@@ -2513,6 +2736,8 @@ static __init int amd_pmu_init(void)
 	memcpy(hw_cache_event_ids, amd_hw_cache_event_ids,
 	       sizeof(hw_cache_event_ids));
 
+	/* initialize BSP */
+	amd_pmu_cpu_online(smp_processor_id());
 	return 0;
 }
 
@@ -2842,4 +3067,25 @@ struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 void hw_perf_event_setup_online(int cpu)
 {
 	init_debug_store_on_cpu(cpu);
+
+	switch (boot_cpu_data.x86_vendor) {
+	case X86_VENDOR_AMD:
+		amd_pmu_cpu_online(cpu);
+		break;
+	default:
+		return;
+	}
+}
+
+void hw_perf_event_setup_offline(int cpu)
+{
+	init_debug_store_on_cpu(cpu);
+
+	switch (boot_cpu_data.x86_vendor) {
+	case X86_VENDOR_AMD:
+		amd_pmu_cpu_offline(cpu);
+		break;
+	default:
+		return;
+	}
 }
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 27f69a0..20f212e 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -98,6 +98,7 @@ void __weak hw_perf_enable(void)		{ barrier(); }
 
 void __weak hw_perf_event_setup(int cpu)	{ barrier(); }
 void __weak hw_perf_event_setup_online(int cpu)	{ barrier(); }
+void __weak hw_perf_event_setup_offline(int cpu){ barrier(); }
 
 int __weak
 hw_perf_group_sched_in(struct perf_event *group_leader,
@@ -5251,6 +5252,10 @@ perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu)
 		perf_event_exit_cpu(cpu);
 		break;
 
+        case CPU_DEAD:
+                hw_perf_event_setup_offline(cpu);
+                break;
+
 	default:
 		break;
 	}
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists