[RFC][PATCH 00/13] perf pmu interface changes -v3

[RFC][PATCH 00/13] perf pmu interface changes -v3

[Peter Zijlstra]

The first 5 patches are fixes to the existing code and should probably
end up in .35, Dave has his sparc64 fix already queued I think, Paulus
could you take the FSL fixes?

The rest of these patches prepare the perf code for multiple pmus (no
user interface yet, Lin Ming is working on that). These patches remove
all weak functions and rework the struct pmu interface.

The patches are boot tested on x86_64 and compile tested on: powerpc, ppc-fsl,
sparc and arm and SH (by courtesy of Matt Fleming).

X86 -- appears to be fully functional.
FSL -- should, after the initial few fixes, at least compile again.
ARM -- is known to have grown some issues, Will was looking into that.
SH  -- has some pre-existing bugs, but hopefully Paulus' last observation about
       the missing -ENOENT in the pmu::event_init method cures the newly 
       introduced troubles.
PPC -- will be looked at by Paulus
SPARC -- Dave, when you have a moment.. ;-)

Once those are sorted, and I'm back from vacation I'll start on the previously
mentioned per pmu context support.

Patches also available in git format for easy testing (tip/master + patches):

  git://git.kernel.org/pub/scm/linux/kernel/git/peterz/linux-2.6-perf.git perf-pmu

HEAD should be ac9fd39c1

[PATCH 01/13] perf, x86: Fix Nehalem PMU quirk

[Peter Zijlstra]

[-- Attachment #1: perf-x86-fix-nehalem-quirk.patch --]
[-- Type: text/plain, Size: 748 bytes --]

The idea was to run all three counters, which means we need to
program 7 into the enable mask, not 3 (which is two bits).

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
Index: linux-2.6/arch/x86/kernel/cpu/perf_event_intel.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_event_intel.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_event_intel.c
@@ -504,7 +504,7 @@ static void intel_pmu_nhm_enable_all(int
 		wrmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + 1, 0x4300B1);
 		wrmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + 2, 0x4300B5);
 
-		wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0x3);
+		wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0x7);
 		wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0x0);
 
 		for (i = 0; i < 3; i++) {

[PATCH 02/13] sparc64: Fix maybe_change_configuration() PCR setting.

[Peter Zijlstra]

[-- Attachment #1: davem-sparc-pmu-fixup.patch --]
[-- Type: text/plain, Size: 788 bytes --]

From: David S. Miller <davem@davemloft.net>

Need to mask out the existing event bits before OR'ing in
the new ones.

Reported-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 arch/sparc/kernel/perf_event.c |    1 +
 1 file changed, 1 insertion(+)

Index: linux-2.6/arch/sparc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sparc/kernel/perf_event.c
+++ linux-2.6/arch/sparc/kernel/perf_event.c
@@ -657,6 +657,7 @@ static u64 maybe_change_configuration(st
 		cpuc->current_idx[i] = idx;
 
 		enc = perf_event_get_enc(cpuc->events[i]);
+		pcr &= ~mask_for_index(idx);
 		pcr |= event_encoding(enc, idx);
 	}
 out:

[PATCH 03/13] perf: Fix CPU hotplug

[Peter Zijlstra]

[-- Attachment #1: perf-x86-fix-cpuhotplug.patch --]
[-- Type: text/plain, Size: 789 bytes --]

Since we have UP_PREPARE, we should also have UP_CANCELED.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 kernel/perf_event.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -5743,15 +5743,15 @@ perf_cpu_notify(struct notifier_block *s
 {
 	unsigned int cpu = (long)hcpu;
 
-	switch (action) {
+	switch (action & ~CPU_TASKS_FROZEN) {
 
 	case CPU_UP_PREPARE:
-	case CPU_UP_PREPARE_FROZEN:
+	case CPU_DOWN_FAILED:
 		perf_event_init_cpu(cpu);
 		break;
 
+	case CPU_UP_CANCELED:
 	case CPU_DOWN_PREPARE:
-	case CPU_DOWN_PREPARE_FROZEN:
 		perf_event_exit_cpu(cpu);
 		break;
 

[PATCH 04/13] perf, powerpc: Use perf_sample_data_init() for the FSL code

[Peter Zijlstra]

[-- Attachment #1: perf-fsl-fix-sample_data.patch --]
[-- Type: text/plain, Size: 822 bytes --]

We should use perf_sample_data_init() to initialize struct perf_sample_data.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 arch/powerpc/kernel/perf_event_fsl_emb.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

Index: linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event_fsl_emb.c
+++ linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
@@ -566,9 +566,9 @@ static void record_and_restart(struct pe
 	 * Finally record data if requested.
 	 */
 	if (record) {
-		struct perf_sample_data data = {
-			.period	= event->hw.last_period,
-		};
+		struct perf_sample_data data;
+
+		perf_sample_data_init(&data, 0);
 
 		if (perf_event_overflow(event, nmi, &data, regs)) {
 			/*

[PATCH 05/13] perf, powerpc: Convert the FSL driver to use local64_t

[Peter Zijlstra]

[-- Attachment #1: perf-fsl-fix-local64_t.patch --]
[-- Type: text/plain, Size: 3298 bytes --]

For some reason the FSL driver got left out when we converted perf
to use local64_t instead of atomic64_t.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 arch/powerpc/kernel/perf_event_fsl_emb.c |   28 ++++++++++++++--------------
 1 file changed, 14 insertions(+), 14 deletions(-)

Index: linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event_fsl_emb.c
+++ linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
@@ -162,15 +162,15 @@ static void fsl_emb_pmu_read(struct perf
 	 * Therefore we treat them like NMIs.
 	 */
 	do {
-		prev = atomic64_read(&event->hw.prev_count);
+		prev = local64_read(&event->hw.prev_count);
 		barrier();
 		val = read_pmc(event->hw.idx);
-	} while (atomic64_cmpxchg(&event->hw.prev_count, prev, val) != prev);
+	} while (local64_cmpxchg(&event->hw.prev_count, prev, val) != prev);
 
 	/* The counters are only 32 bits wide */
 	delta = (val - prev) & 0xfffffffful;
-	atomic64_add(delta, &event->count);
-	atomic64_sub(delta, &event->hw.period_left);
+	local64_add(delta, &event->count);
+	local64_sub(delta, &event->hw.period_left);
 }
 
 /*
@@ -296,11 +296,11 @@ static int fsl_emb_pmu_enable(struct per
 
 	val = 0;
 	if (event->hw.sample_period) {
-		s64 left = atomic64_read(&event->hw.period_left);
+		s64 left = local64_read(&event->hw.period_left);
 		if (left < 0x80000000L)
 			val = 0x80000000L - left;
 	}
-	atomic64_set(&event->hw.prev_count, val);
+	local64_set(&event->hw.prev_count, val);
 	write_pmc(i, val);
 	perf_event_update_userpage(event);
 
@@ -371,8 +371,8 @@ static void fsl_emb_pmu_unthrottle(struc
 	if (left < 0x80000000L)
 		val = 0x80000000L - left;
 	write_pmc(event->hw.idx, val);
-	atomic64_set(&event->hw.prev_count, val);
-	atomic64_set(&event->hw.period_left, left);
+	local64_set(&event->hw.prev_count, val);
+	local64_set(&event->hw.period_left, left);
 	perf_event_update_userpage(event);
 	perf_enable();
 	local_irq_restore(flags);
@@ -500,7 +500,7 @@ const struct pmu *hw_perf_event_init(str
 		return ERR_PTR(-ENOTSUPP);
 
 	event->hw.last_period = event->hw.sample_period;
-	atomic64_set(&event->hw.period_left, event->hw.last_period);
+	local64_set(&event->hw.period_left, event->hw.last_period);
 
 	/*
 	 * See if we need to reserve the PMU.
@@ -541,16 +541,16 @@ static void record_and_restart(struct pe
 	int record = 0;
 
 	/* we don't have to worry about interrupts here */
-	prev = atomic64_read(&event->hw.prev_count);
+	prev = local64_read(&event->hw.prev_count);
 	delta = (val - prev) & 0xfffffffful;
-	atomic64_add(delta, &event->count);
+	local64_add(delta, &event->count);
 
 	/*
 	 * See if the total period for this event has expired,
 	 * and update for the next period.
 	 */
 	val = 0;
-	left = atomic64_read(&event->hw.period_left) - delta;
+	left = local64_read(&event->hw.period_left) - delta;
 	if (period) {
 		if (left <= 0) {
 			left += period;
@@ -584,8 +584,8 @@ static void record_and_restart(struct pe
 	}
 
 	write_pmc(event->hw.idx, val);
-	atomic64_set(&event->hw.prev_count, val);
-	atomic64_set(&event->hw.period_left, left);
+	local64_set(&event->hw.prev_count, val);
+	local64_set(&event->hw.period_left, left);
 	perf_event_update_userpage(event);
 }
 

[RFC][PATCH 06/13] perf: deconstify struct pmu

[Peter Zijlstra]

[-- Attachment #1: perf-deconst-pmu.patch --]
[-- Type: text/plain, Size: 12375 bytes --]

sed -ie 's/const struct pmu\>/struct pmu/g' `git grep -l "const struct pmu\>"`

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 arch/arm/kernel/perf_event.c             |    2 +-
 arch/powerpc/kernel/perf_event.c         |    8 ++++----
 arch/powerpc/kernel/perf_event_fsl_emb.c |    2 +-
 arch/sh/kernel/perf_event.c              |    4 ++--
 arch/sparc/kernel/perf_event.c           |   10 +++++-----
 arch/x86/kernel/cpu/perf_event.c         |   14 +++++++-------
 include/linux/perf_event.h               |   10 +++++-----
 kernel/perf_event.c                      |   26 +++++++++++++-------------
 8 files changed, 38 insertions(+), 38 deletions(-)

Index: linux-2.6/arch/arm/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/arm/kernel/perf_event.c
+++ linux-2.6/arch/arm/kernel/perf_event.c
@@ -491,7 +491,7 @@ __hw_perf_event_init(struct perf_event *
 	return err;
 }
 
-const struct pmu *
+struct pmu *
 hw_perf_event_init(struct perf_event *event)
 {
 	int err = 0;
Index: linux-2.6/arch/powerpc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event.c
+++ linux-2.6/arch/powerpc/kernel/perf_event.c
@@ -854,7 +854,7 @@ static void power_pmu_unthrottle(struct 
  * Set the flag to make pmu::enable() not perform the
  * schedulability test, it will be performed at commit time
  */
-void power_pmu_start_txn(const struct pmu *pmu)
+void power_pmu_start_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
@@ -867,7 +867,7 @@ void power_pmu_start_txn(const struct pm
  * Clear the flag and pmu::enable() will perform the
  * schedulability test.
  */
-void power_pmu_cancel_txn(const struct pmu *pmu)
+void power_pmu_cancel_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
@@ -879,7 +879,7 @@ void power_pmu_cancel_txn(const struct p
  * Perform the group schedulability test as a whole
  * Return 0 if success
  */
-int power_pmu_commit_txn(const struct pmu *pmu)
+int power_pmu_commit_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw;
 	long i, n;
@@ -1011,7 +1011,7 @@ static int hw_perf_cache_event(u64 confi
 	return 0;
 }
 
-const struct pmu *hw_perf_event_init(struct perf_event *event)
+struct pmu *hw_perf_event_init(struct perf_event *event)
 {
 	u64 ev;
 	unsigned long flags;
Index: linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event_fsl_emb.c
+++ linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
@@ -428,7 +428,7 @@ static int hw_perf_cache_event(u64 confi
 	return 0;
 }
 
-const struct pmu *hw_perf_event_init(struct perf_event *event)
+struct pmu *hw_perf_event_init(struct perf_event *event)
 {
 	u64 ev;
 	struct perf_event *events[MAX_HWEVENTS];
Index: linux-2.6/arch/sh/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sh/kernel/perf_event.c
+++ linux-2.6/arch/sh/kernel/perf_event.c
@@ -257,13 +257,13 @@ static void sh_pmu_read(struct perf_even
 	sh_perf_event_update(event, &event->hw, event->hw.idx);
 }
 
-static const struct pmu pmu = {
+static struct pmu pmu = {
 	.enable		= sh_pmu_enable,
 	.disable	= sh_pmu_disable,
 	.read		= sh_pmu_read,
 };
 
-const struct pmu *hw_perf_event_init(struct perf_event *event)
+struct pmu *hw_perf_event_init(struct perf_event *event)
 {
 	int err = __hw_perf_event_init(event);
 	if (unlikely(err)) {
Index: linux-2.6/arch/sparc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sparc/kernel/perf_event.c
+++ linux-2.6/arch/sparc/kernel/perf_event.c
@@ -1098,7 +1098,7 @@ static int __hw_perf_event_init(struct p
  * Set the flag to make pmu::enable() not perform the
  * schedulability test, it will be performed at commit time
  */
-static void sparc_pmu_start_txn(const struct pmu *pmu)
+static void sparc_pmu_start_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
@@ -1110,7 +1110,7 @@ static void sparc_pmu_start_txn(const st
  * Clear the flag and pmu::enable() will perform the
  * schedulability test.
  */
-static void sparc_pmu_cancel_txn(const struct pmu *pmu)
+static void sparc_pmu_cancel_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
@@ -1122,7 +1122,7 @@ static void sparc_pmu_cancel_txn(const s
  * Perform the group schedulability test as a whole
  * Return 0 if success
  */
-static int sparc_pmu_commit_txn(const struct pmu *pmu)
+static int sparc_pmu_commit_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	int n;
@@ -1141,7 +1141,7 @@ static int sparc_pmu_commit_txn(const st
 	return 0;
 }
 
-static const struct pmu pmu = {
+static struct pmu pmu = {
 	.enable		= sparc_pmu_enable,
 	.disable	= sparc_pmu_disable,
 	.read		= sparc_pmu_read,
@@ -1151,7 +1151,7 @@ static const struct pmu pmu = {
 	.commit_txn	= sparc_pmu_commit_txn,
 };
 
-const struct pmu *hw_perf_event_init(struct perf_event *event)
+struct pmu *hw_perf_event_init(struct perf_event *event)
 {
 	int err = __hw_perf_event_init(event);
 
Index: linux-2.6/arch/x86/kernel/cpu/perf_event.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_event.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_event.c
@@ -618,7 +618,7 @@ static void x86_pmu_enable_all(int added
 	}
 }
 
-static const struct pmu pmu;
+static struct pmu pmu;
 
 static inline int is_x86_event(struct perf_event *event)
 {
@@ -1394,7 +1394,7 @@ static inline void x86_pmu_read(struct p
  * Set the flag to make pmu::enable() not perform the
  * schedulability test, it will be performed at commit time
  */
-static void x86_pmu_start_txn(const struct pmu *pmu)
+static void x86_pmu_start_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 
@@ -1407,7 +1407,7 @@ static void x86_pmu_start_txn(const stru
  * Clear the flag and pmu::enable() will perform the
  * schedulability test.
  */
-static void x86_pmu_cancel_txn(const struct pmu *pmu)
+static void x86_pmu_cancel_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 
@@ -1424,7 +1424,7 @@ static void x86_pmu_cancel_txn(const str
  * Perform the group schedulability test as a whole
  * Return 0 if success
  */
-static int x86_pmu_commit_txn(const struct pmu *pmu)
+static int x86_pmu_commit_txn(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	int assign[X86_PMC_IDX_MAX];
@@ -1450,7 +1450,7 @@ static int x86_pmu_commit_txn(const stru
 	return 0;
 }
 
-static const struct pmu pmu = {
+static struct pmu pmu = {
 	.enable		= x86_pmu_enable,
 	.disable	= x86_pmu_disable,
 	.start		= x86_pmu_start,
@@ -1536,9 +1536,9 @@ out:
 	return ret;
 }
 
-const struct pmu *hw_perf_event_init(struct perf_event *event)
+struct pmu *hw_perf_event_init(struct perf_event *event)
 {
-	const struct pmu *tmp;
+	struct pmu *tmp;
 	int err;
 
 	err = __hw_perf_event_init(event);
Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -576,19 +576,19 @@ struct pmu {
 	 * Start the transaction, after this ->enable() doesn't need
 	 * to do schedulability tests.
 	 */
-	void (*start_txn)	(const struct pmu *pmu);
+	void (*start_txn)	(struct pmu *pmu);
 	/*
 	 * If ->start_txn() disabled the ->enable() schedulability test
 	 * then ->commit_txn() is required to perform one. On success
 	 * the transaction is closed. On error the transaction is kept
 	 * open until ->cancel_txn() is called.
 	 */
-	int  (*commit_txn)	(const struct pmu *pmu);
+	int  (*commit_txn)	(struct pmu *pmu);
 	/*
 	 * Will cancel the transaction, assumes ->disable() is called for
 	 * each successfull ->enable() during the transaction.
 	 */
-	void (*cancel_txn)	(const struct pmu *pmu);
+	void (*cancel_txn)	(struct pmu *pmu);
 };
 
 /**
@@ -667,7 +667,7 @@ struct perf_event {
 	int				nr_siblings;
 	int				group_flags;
 	struct perf_event		*group_leader;
-	const struct pmu		*pmu;
+	struct pmu		*pmu;
 
 	enum perf_event_active_state	state;
 	unsigned int			attach_state;
@@ -845,7 +845,7 @@ struct perf_output_handle {
  */
 extern int perf_max_events;
 
-extern const struct pmu *hw_perf_event_init(struct perf_event *event);
+extern struct pmu *hw_perf_event_init(struct perf_event *event);
 
 extern void perf_event_task_sched_in(struct task_struct *task);
 extern void perf_event_task_sched_out(struct task_struct *task, struct task_struct *next);
Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -75,7 +75,7 @@ static DEFINE_SPINLOCK(perf_resource_loc
 /*
  * Architecture provided APIs - weak aliases:
  */
-extern __weak const struct pmu *hw_perf_event_init(struct perf_event *event)
+extern __weak struct pmu *hw_perf_event_init(struct perf_event *event)
 {
 	return NULL;
 }
@@ -673,7 +673,7 @@ group_sched_in(struct perf_event *group_
 	       struct perf_event_context *ctx)
 {
 	struct perf_event *event, *partial_group = NULL;
-	const struct pmu *pmu = group_event->pmu;
+	struct pmu *pmu = group_event->pmu;
 	bool txn = false;
 
 	if (group_event->state == PERF_EVENT_STATE_OFF)
@@ -4291,7 +4291,7 @@ static int perf_swevent_int(struct perf_
 	return 0;
 }
 
-static const struct pmu perf_ops_generic = {
+static struct pmu perf_ops_generic = {
 	.enable		= perf_swevent_enable,
 	.disable	= perf_swevent_disable,
 	.start		= perf_swevent_int,
@@ -4404,7 +4404,7 @@ static void cpu_clock_perf_event_read(st
 	cpu_clock_perf_event_update(event);
 }
 
-static const struct pmu perf_ops_cpu_clock = {
+static struct pmu perf_ops_cpu_clock = {
 	.enable		= cpu_clock_perf_event_enable,
 	.disable	= cpu_clock_perf_event_disable,
 	.read		= cpu_clock_perf_event_read,
@@ -4461,7 +4461,7 @@ static void task_clock_perf_event_read(s
 	task_clock_perf_event_update(event, time);
 }
 
-static const struct pmu perf_ops_task_clock = {
+static struct pmu perf_ops_task_clock = {
 	.enable		= task_clock_perf_event_enable,
 	.disable	= task_clock_perf_event_disable,
 	.read		= task_clock_perf_event_read,
@@ -4575,7 +4575,7 @@ static int swevent_hlist_get(struct perf
 
 #ifdef CONFIG_EVENT_TRACING
 
-static const struct pmu perf_ops_tracepoint = {
+static struct pmu perf_ops_tracepoint = {
 	.enable		= perf_trace_enable,
 	.disable	= perf_trace_disable,
 	.start		= perf_swevent_int,
@@ -4639,7 +4639,7 @@ static void tp_perf_event_destroy(struct
 	perf_trace_destroy(event);
 }
 
-static const struct pmu *tp_perf_event_init(struct perf_event *event)
+static struct pmu *tp_perf_event_init(struct perf_event *event)
 {
 	int err;
 
@@ -4686,7 +4686,7 @@ static void perf_event_free_filter(struc
 
 #else
 
-static const struct pmu *tp_perf_event_init(struct perf_event *event)
+static struct pmu *tp_perf_event_init(struct perf_event *event)
 {
 	return NULL;
 }
@@ -4708,7 +4708,7 @@ static void bp_perf_event_destroy(struct
 	release_bp_slot(event);
 }
 
-static const struct pmu *bp_perf_event_init(struct perf_event *bp)
+static struct pmu *bp_perf_event_init(struct perf_event *bp)
 {
 	int err;
 
@@ -4732,7 +4732,7 @@ void perf_bp_event(struct perf_event *bp
 		perf_swevent_add(bp, 1, 1, &sample, regs);
 }
 #else
-static const struct pmu *bp_perf_event_init(struct perf_event *bp)
+static struct pmu *bp_perf_event_init(struct perf_event *bp)
 {
 	return NULL;
 }
@@ -4754,9 +4754,9 @@ static void sw_perf_event_destroy(struct
 	swevent_hlist_put(event);
 }
 
-static const struct pmu *sw_perf_event_init(struct perf_event *event)
+static struct pmu *sw_perf_event_init(struct perf_event *event)
 {
-	const struct pmu *pmu = NULL;
+	struct pmu *pmu = NULL;
 	u64 event_id = event->attr.config;
 
 	/*
@@ -4818,7 +4818,7 @@ perf_event_alloc(struct perf_event_attr 
 		   perf_overflow_handler_t overflow_handler,
 		   gfp_t gfpflags)
 {
-	const struct pmu *pmu;
+	struct pmu *pmu;
 	struct perf_event *event;
 	struct hw_perf_event *hwc;
 	long err;

[RFC][PATCH 07/13] perf: register pmu implementations

[Peter Zijlstra]

[-- Attachment #1: perf-register-pmu.patch --]
[-- Type: text/plain, Size: 34355 bytes --]

Simple registration interface for struct pmu, this provides the
infrastructure for removing all the weak functions.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 arch/arm/kernel/perf_event.c             |   38 +-
 arch/powerpc/kernel/perf_event.c         |   46 +-
 arch/powerpc/kernel/perf_event_fsl_emb.c |   37 +
 arch/sh/kernel/perf_event.c              |   35 +
 arch/sparc/kernel/perf_event.c           |   29 -
 arch/x86/kernel/cpu/perf_event.c         |   45 +-
 include/linux/perf_event.h               |   10 
 kernel/hw_breakpoint.c                   |   35 +
 kernel/perf_event.c                      |  588 +++++++++++++++----------------
 9 files changed, 466 insertions(+), 397 deletions(-)

Index: linux-2.6/arch/arm/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/arm/kernel/perf_event.c
+++ linux-2.6/arch/arm/kernel/perf_event.c
@@ -306,12 +306,7 @@ out:
 	return err;
 }
 
-static struct pmu pmu = {
-	.enable	    = armpmu_enable,
-	.disable    = armpmu_disable,
-	.unthrottle = armpmu_unthrottle,
-	.read	    = armpmu_read,
-};
+static struct pmu pmu;
 
 static int
 validate_event(struct cpu_hw_events *cpuc,
@@ -491,20 +486,29 @@ __hw_perf_event_init(struct perf_event *
 	return err;
 }
 
-struct pmu *
-hw_perf_event_init(struct perf_event *event)
+static int armpmu_event_init(struct perf_event *event)
 {
 	int err = 0;
 
+	switch (event->attr.type) {
+	case PERF_TYPE_RAW:
+	case PERF_TYPE_HARDWARE:
+	case PERF_TYPE_HW_CACHE:
+		break;
+
+	default:
+		return -ENOENT;
+	}
+
 	if (!armpmu)
-		return ERR_PTR(-ENODEV);
+		return -ENODEV;
 
 	event->destroy = hw_perf_event_destroy;
 
 	if (!atomic_inc_not_zero(&active_events)) {
 		if (atomic_read(&active_events) > perf_max_events) {
 			atomic_dec(&active_events);
-			return ERR_PTR(-ENOSPC);
+			return -ENOSPC;
 		}
 
 		mutex_lock(&pmu_reserve_mutex);
@@ -518,15 +522,23 @@ hw_perf_event_init(struct perf_event *ev
 	}
 
 	if (err)
-		return ERR_PTR(err);
+		return err;
 
 	err = __hw_perf_event_init(event);
 	if (err)
 		hw_perf_event_destroy(event);
 
-	return err ? ERR_PTR(err) : &pmu;
+	return err;
 }
 
+static struct pmu pmu = {
+	.event_init = armpmu_event_init,
+	.enable	    = armpmu_enable,
+	.disable    = armpmu_disable,
+	.unthrottle = armpmu_unthrottle,
+	.read	    = armpmu_read,
+};
+
 void
 hw_perf_enable(void)
 {
@@ -2994,6 +3006,8 @@ init_hw_perf_events(void)
 		perf_max_events = -1;
 	}
 
+	perf_pmu_register(&pmu);
+
 	return 0;
 }
 arch_initcall(init_hw_perf_events);
Index: linux-2.6/arch/powerpc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event.c
+++ linux-2.6/arch/powerpc/kernel/perf_event.c
@@ -901,16 +901,6 @@ int power_pmu_commit_txn(struct pmu *pmu
 	return 0;
 }
 
-struct pmu power_pmu = {
-	.enable		= power_pmu_enable,
-	.disable	= power_pmu_disable,
-	.read		= power_pmu_read,
-	.unthrottle	= power_pmu_unthrottle,
-	.start_txn	= power_pmu_start_txn,
-	.cancel_txn	= power_pmu_cancel_txn,
-	.commit_txn	= power_pmu_commit_txn,
-};
-
 /*
  * Return 1 if we might be able to put event on a limited PMC,
  * or 0 if not.
@@ -1011,7 +1001,7 @@ static int hw_perf_cache_event(u64 confi
 	return 0;
 }
 
-struct pmu *hw_perf_event_init(struct perf_event *event)
+static int power_pmu_event_init(struct perf_event *event)
 {
 	u64 ev;
 	unsigned long flags;
@@ -1023,25 +1013,27 @@ struct pmu *hw_perf_event_init(struct pe
 	struct cpu_hw_events *cpuhw;
 
 	if (!ppmu)
-		return ERR_PTR(-ENXIO);
+		return -ENOENT;
+
 	switch (event->attr.type) {
 	case PERF_TYPE_HARDWARE:
 		ev = event->attr.config;
 		if (ev >= ppmu->n_generic || ppmu->generic_events[ev] == 0)
-			return ERR_PTR(-EOPNOTSUPP);
+			return -EOPNOTSUPP;
 		ev = ppmu->generic_events[ev];
 		break;
 	case PERF_TYPE_HW_CACHE:
 		err = hw_perf_cache_event(event->attr.config, &ev);
 		if (err)
-			return ERR_PTR(err);
+			return err;
 		break;
 	case PERF_TYPE_RAW:
 		ev = event->attr.config;
 		break;
 	default:
-		return ERR_PTR(-EINVAL);
+		return -ENOENT;
 	}
+
 	event->hw.config_base = ev;
 	event->hw.idx = 0;
 
@@ -1078,7 +1070,7 @@ struct pmu *hw_perf_event_init(struct pe
 			 */
 			ev = normal_pmc_alternative(ev, flags);
 			if (!ev)
-				return ERR_PTR(-EINVAL);
+				return -EINVAL;
 		}
 	}
 
@@ -1092,19 +1084,19 @@ struct pmu *hw_perf_event_init(struct pe
 		n = collect_events(event->group_leader, ppmu->n_counter - 1,
 				   ctrs, events, cflags);
 		if (n < 0)
-			return ERR_PTR(-EINVAL);
+			return -EINVAL;
 	}
 	events[n] = ev;
 	ctrs[n] = event;
 	cflags[n] = flags;
 	if (check_excludes(ctrs, cflags, n, 1))
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 
 	cpuhw = &get_cpu_var(cpu_hw_events);
 	err = power_check_constraints(cpuhw, events, cflags, n + 1);
 	put_cpu_var(cpu_hw_events);
 	if (err)
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 
 	event->hw.config = events[n];
 	event->hw.event_base = cflags[n];
@@ -1129,11 +1121,20 @@ struct pmu *hw_perf_event_init(struct pe
 	}
 	event->destroy = hw_perf_event_destroy;
 
-	if (err)
-		return ERR_PTR(err);
-	return &power_pmu;
+	return err;
 }
 
+struct pmu power_pmu = {
+	.event_init	= power_pmu_event_init,
+	.enable		= power_pmu_enable,
+	.disable	= power_pmu_disable,
+	.read		= power_pmu_read,
+	.unthrottle	= power_pmu_unthrottle,
+	.start_txn	= power_pmu_start_txn,
+	.cancel_txn	= power_pmu_cancel_txn,
+	.commit_txn	= power_pmu_commit_txn,
+};
+
 /*
  * A counter has overflowed; update its count and record
  * things if requested.  Note that interrupts are hard-disabled
@@ -1339,6 +1340,7 @@ int register_power_pmu(struct power_pmu 
 		freeze_events_kernel = MMCR0_FCHV;
 #endif /* CONFIG_PPC64 */
 
+	perf_pmu_register(&power_pmu);
 	perf_cpu_notifier(power_pmu_notifier);
 
 	return 0;
Index: linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event_fsl_emb.c
+++ linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
@@ -378,13 +378,6 @@ static void fsl_emb_pmu_unthrottle(struc
 	local_irq_restore(flags);
 }
 
-static struct pmu fsl_emb_pmu = {
-	.enable		= fsl_emb_pmu_enable,
-	.disable	= fsl_emb_pmu_disable,
-	.read		= fsl_emb_pmu_read,
-	.unthrottle	= fsl_emb_pmu_unthrottle,
-};
-
 /*
  * Release the PMU if this is the last perf_event.
  */
@@ -428,7 +421,7 @@ static int hw_perf_cache_event(u64 confi
 	return 0;
 }
 
-struct pmu *hw_perf_event_init(struct perf_event *event)
+static int fsl_emb_pmu_event_init(struct perf_event *event)
 {
 	u64 ev;
 	struct perf_event *events[MAX_HWEVENTS];
@@ -441,14 +434,14 @@ struct pmu *hw_perf_event_init(struct pe
 	case PERF_TYPE_HARDWARE:
 		ev = event->attr.config;
 		if (ev >= ppmu->n_generic || ppmu->generic_events[ev] == 0)
-			return ERR_PTR(-EOPNOTSUPP);
+			return -EOPNOTSUPP;
 		ev = ppmu->generic_events[ev];
 		break;
 
 	case PERF_TYPE_HW_CACHE:
 		err = hw_perf_cache_event(event->attr.config, &ev);
 		if (err)
-			return ERR_PTR(err);
+			return err;
 		break;
 
 	case PERF_TYPE_RAW:
@@ -456,12 +449,12 @@ struct pmu *hw_perf_event_init(struct pe
 		break;
 
 	default:
-		return ERR_PTR(-EINVAL);
+		return -ENOENT;
 	}
 
 	event->hw.config = ppmu->xlate_event(ev);
 	if (!(event->hw.config & FSL_EMB_EVENT_VALID))
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 
 	/*
 	 * If this is in a group, check if it can go on with all the
@@ -473,7 +466,7 @@ struct pmu *hw_perf_event_init(struct pe
 		n = collect_events(event->group_leader,
 		                   ppmu->n_counter - 1, events);
 		if (n < 0)
-			return ERR_PTR(-EINVAL);
+			return -EINVAL;
 	}
 
 	if (event->hw.config & FSL_EMB_EVENT_RESTRICTED) {
@@ -484,7 +477,7 @@ struct pmu *hw_perf_event_init(struct pe
 		}
 
 		if (num_restricted >= ppmu->n_restricted)
-			return ERR_PTR(-EINVAL);
+			return -EINVAL;
 	}
 
 	event->hw.idx = -1;
@@ -497,7 +490,7 @@ struct pmu *hw_perf_event_init(struct pe
 	if (event->attr.exclude_kernel)
 		event->hw.config_base |= PMLCA_FCS;
 	if (event->attr.exclude_idle)
-		return ERR_PTR(-ENOTSUPP);
+		return -ENOTSUPP;
 
 	event->hw.last_period = event->hw.sample_period;
 	local64_set(&event->hw.period_left, event->hw.last_period);
@@ -523,11 +516,17 @@ struct pmu *hw_perf_event_init(struct pe
 	}
 	event->destroy = hw_perf_event_destroy;
 
-	if (err)
-		return ERR_PTR(err);
-	return &fsl_emb_pmu;
+	return err;
 }
 
+static struct pmu fsl_emb_pmu = {
+	.event_init	= fsl_emb_pmu_event_init,
+	.enable		= fsl_emb_pmu_enable,
+	.disable	= fsl_emb_pmu_disable,
+	.read		= fsl_emb_pmu_read,
+	.unthrottle	= fsl_emb_pmu_unthrottle,
+};
+
 /*
  * A counter has overflowed; update its count and record
  * things if requested.  Note that interrupts are hard-disabled
@@ -650,5 +649,7 @@ int register_fsl_emb_pmu(struct fsl_emb_
 	pr_info("%s performance monitor hardware support registered\n",
 		pmu->name);
 
+	perf_pmu_register(&fsl_emb_pmu);
+
 	return 0;
 }
Index: linux-2.6/arch/sh/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sh/kernel/perf_event.c
+++ linux-2.6/arch/sh/kernel/perf_event.c
@@ -257,26 +257,38 @@ static void sh_pmu_read(struct perf_even
 	sh_perf_event_update(event, &event->hw, event->hw.idx);
 }
 
-static struct pmu pmu = {
-	.enable		= sh_pmu_enable,
-	.disable	= sh_pmu_disable,
-	.read		= sh_pmu_read,
-};
-
-struct pmu *hw_perf_event_init(struct perf_event *event)
+static int sh_pmu_event_init(struct perf_event *event)
 {
-	int err = __hw_perf_event_init(event);
+	int err;
+
+	switch (event->attr.type) {
+	case PERF_TYPE_RAW:
+	case PERF_TYPE_HW_CACHE:
+	case PERF_TYPE_HARDWARE:
+		err = __hw_perf_event_init(event);
+		break;
+
+	default:
+		return -ENOENT;
+	}
+
 	if (unlikely(err)) {
 		if (event->destroy)
 			event->destroy(event);
-		return ERR_PTR(err);
 	}
 
-	return &pmu;
+	return err;
 }
 
+static struct pmu pmu = {
+	.event_init	= sh_pmu_event_init,
+	.enable		= sh_pmu_enable,
+	.disable	= sh_pmu_disable,
+	.read		= sh_pmu_read,
+};
+
 static void sh_pmu_setup(int cpu)
-{
+
 	struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu);
 
 	memset(cpuhw, 0, sizeof(struct cpu_hw_events));
@@ -325,6 +337,7 @@ int __cpuinit register_sh_pmu(struct sh_
 
 	WARN_ON(pmu->num_events > MAX_HWEVENTS);
 
+	perf_pmu_register(&pmu);
 	perf_cpu_notifier(sh_pmu_notifier);
 	return 0;
 }
Index: linux-2.6/arch/sparc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sparc/kernel/perf_event.c
+++ linux-2.6/arch/sparc/kernel/perf_event.c
@@ -1025,7 +1025,7 @@ out:
 	return ret;
 }
 
-static int __hw_perf_event_init(struct perf_event *event)
+static int sparc_pmu_event_init(struct perf_event *event)
 {
 	struct perf_event_attr *attr = &event->attr;
 	struct perf_event *evts[MAX_HWEVENTS];
@@ -1038,17 +1038,27 @@ static int __hw_perf_event_init(struct p
 	if (atomic_read(&nmi_active) < 0)
 		return -ENODEV;
 
-	if (attr->type == PERF_TYPE_HARDWARE) {
+	switch (attr->type) {
+	case PERF_TYPE_HARDWARE:
 		if (attr->config >= sparc_pmu->max_events)
 			return -EINVAL;
 		pmap = sparc_pmu->event_map(attr->config);
-	} else if (attr->type == PERF_TYPE_HW_CACHE) {
+		break;
+
+	case PERF_TYPE_HW_CACHE:
 		pmap = sparc_map_cache_event(attr->config);
 		if (IS_ERR(pmap))
 			return PTR_ERR(pmap);
-	} else
+		break;
+
+	case PERF_TYPE_RAW:
 		return -EOPNOTSUPP;
 
+	default:
+		return -ENOENT;
+
+	}
+
 	/* We save the enable bits in the config_base.  */
 	hwc->config_base = sparc_pmu->irq_bit;
 	if (!attr->exclude_user)
@@ -1143,6 +1153,7 @@ static int sparc_pmu_commit_txn(struct p
 }
 
 static struct pmu pmu = {
+	.event_init	= sparc_pmu_event_init,
 	.enable		= sparc_pmu_enable,
 	.disable	= sparc_pmu_disable,
 	.read		= sparc_pmu_read,
@@ -1152,15 +1163,6 @@ static struct pmu pmu = {
 	.commit_txn	= sparc_pmu_commit_txn,
 };
 
-struct pmu *hw_perf_event_init(struct perf_event *event)
-{
-	int err = __hw_perf_event_init(event);
-
-	if (err)
-		return ERR_PTR(err);
-	return &pmu;
-}
-
 void perf_event_print_debug(void)
 {
 	unsigned long flags;
@@ -1280,6 +1282,7 @@ void __init init_hw_perf_events(void)
 	/* All sparc64 PMUs currently have 2 events.  */
 	perf_max_events = 2;
 
+	perf_pmu_register(&pmu);
 	register_die_notifier(&perf_event_nmi_notifier);
 }
 
Index: linux-2.6/arch/x86/kernel/cpu/perf_event.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_event.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_event.c
@@ -530,7 +530,7 @@ static int x86_pmu_hw_config(struct perf
 /*
  * Setup the hardware configuration for a given attr_type
  */
-static int __hw_perf_event_init(struct perf_event *event)
+static int __x86_pmu_event_init(struct perf_event *event)
 {
 	int err;
 
@@ -1381,6 +1381,7 @@ void __init init_hw_perf_events(void)
 	pr_info("... fixed-purpose events:   %d\n",     x86_pmu.num_counters_fixed);
 	pr_info("... event mask:             %016Lx\n", x86_pmu.intel_ctrl);
 
+	perf_pmu_register(&pmu);
 	perf_cpu_notifier(x86_pmu_notifier);
 }
 
@@ -1450,18 +1451,6 @@ static int x86_pmu_commit_txn(struct pmu
 	return 0;
 }
 
-static struct pmu pmu = {
-	.enable		= x86_pmu_enable,
-	.disable	= x86_pmu_disable,
-	.start		= x86_pmu_start,
-	.stop		= x86_pmu_stop,
-	.read		= x86_pmu_read,
-	.unthrottle	= x86_pmu_unthrottle,
-	.start_txn	= x86_pmu_start_txn,
-	.cancel_txn	= x86_pmu_cancel_txn,
-	.commit_txn	= x86_pmu_commit_txn,
-};
-
 /*
  * validate that we can schedule this event
  */
@@ -1536,12 +1525,22 @@ out:
 	return ret;
 }
 
-struct pmu *hw_perf_event_init(struct perf_event *event)
+int x86_pmu_event_init(struct perf_event *event)
 {
 	struct pmu *tmp;
 	int err;
 
-	err = __hw_perf_event_init(event);
+	switch (event->attr.type) {
+	case PERF_TYPE_RAW:
+	case PERF_TYPE_HARDWARE:
+	case PERF_TYPE_HW_CACHE:
+		break;
+
+	default:
+		return -ENOENT;
+	}
+
+	err = __x86_pmu_event_init(event);
 	if (!err) {
 		/*
 		 * we temporarily connect event to its pmu
@@ -1561,12 +1560,24 @@ struct pmu *hw_perf_event_init(struct pe
 	if (err) {
 		if (event->destroy)
 			event->destroy(event);
-		return ERR_PTR(err);
 	}
 
-	return &pmu;
+	return err;
 }
 
+static struct pmu pmu = {
+	.event_init	= x86_pmu_event_init,
+	.enable		= x86_pmu_enable,
+	.disable	= x86_pmu_disable,
+	.start		= x86_pmu_start,
+	.stop		= x86_pmu_stop,
+	.read		= x86_pmu_read,
+	.unthrottle	= x86_pmu_unthrottle,
+	.start_txn	= x86_pmu_start_txn,
+	.cancel_txn	= x86_pmu_cancel_txn,
+	.commit_txn	= x86_pmu_commit_txn,
+};
+
 /*
  * callchain support
  */
Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -561,6 +561,13 @@ struct perf_event;
  * struct pmu - generic performance monitoring unit
  */
 struct pmu {
+	struct list_head		entry;
+
+	/*
+	 * Should return -ENOENT when the @event doesn't match this pmu
+	 */
+	int (*event_init)		(struct perf_event *event);
+
 	int (*enable)			(struct perf_event *event);
 	void (*disable)			(struct perf_event *event);
 	int (*start)			(struct perf_event *event);
@@ -847,7 +854,8 @@ struct perf_output_handle {
  */
 extern int perf_max_events;
 
-extern struct pmu *hw_perf_event_init(struct perf_event *event);
+extern int perf_pmu_register(struct pmu *pmu);
+extern void perf_pmu_unregister(struct pmu *pmu);
 
 extern void perf_event_task_sched_in(struct task_struct *task);
 extern void perf_event_task_sched_out(struct task_struct *task, struct task_struct *next);
Index: linux-2.6/kernel/hw_breakpoint.c
===================================================================
--- linux-2.6.orig/kernel/hw_breakpoint.c
+++ linux-2.6/kernel/hw_breakpoint.c
@@ -553,6 +553,34 @@ static struct notifier_block hw_breakpoi
 	.priority = 0x7fffffff
 };
 
+static void bp_perf_event_destroy(struct perf_event *event)
+{
+	release_bp_slot(event);
+}
+
+static int hw_breakpoint_event_init(struct perf_event *bp)
+{
+	int err;
+
+	if (bp->attr.type != PERF_TYPE_BREAKPOINT)
+		return -ENOENT;
+
+	err = register_perf_hw_breakpoint(bp);
+	if (err)
+		return err;
+
+	bp->destroy = bp_perf_event_destroy;
+
+	return 0;
+}
+
+static struct pmu perf_breakpoint = {
+	.event_init	= hw_breakpoint_event_init,
+	.enable		= arch_install_hw_breakpoint,
+	.disable	= arch_uninstall_hw_breakpoint,
+	.read		= hw_breakpoint_pmu_read,
+};
+
 static int __init init_hw_breakpoint(void)
 {
 	unsigned int **task_bp_pinned;
@@ -574,6 +602,8 @@ static int __init init_hw_breakpoint(voi
 
 	constraints_initialized = 1;
 
+	perf_pmu_register(&perf_breakpoint);
+
 	return register_die_notifier(&hw_breakpoint_exceptions_nb);
 
  err_alloc:
@@ -589,8 +619,3 @@ static int __init init_hw_breakpoint(voi
 core_initcall(init_hw_breakpoint);
 
 
-struct pmu perf_ops_bp = {
-	.enable		= arch_install_hw_breakpoint,
-	.disable	= arch_uninstall_hw_breakpoint,
-	.read		= hw_breakpoint_pmu_read,
-};
Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -31,7 +31,6 @@
 #include <linux/kernel_stat.h>
 #include <linux/perf_event.h>
 #include <linux/ftrace_event.h>
-#include <linux/hw_breakpoint.h>
 
 #include <asm/irq_regs.h>
 
@@ -72,14 +71,6 @@ static atomic64_t perf_event_id;
  */
 static DEFINE_SPINLOCK(perf_resource_lock);
 
-/*
- * Architecture provided APIs - weak aliases:
- */
-extern __weak struct pmu *hw_perf_event_init(struct perf_event *event)
-{
-	return NULL;
-}
-
 void __weak hw_perf_disable(void)		{ barrier(); }
 void __weak hw_perf_enable(void)		{ barrier(); }
 
@@ -4291,182 +4282,6 @@ static int perf_swevent_int(struct perf_
 	return 0;
 }
 
-static struct pmu perf_ops_generic = {
-	.enable		= perf_swevent_enable,
-	.disable	= perf_swevent_disable,
-	.start		= perf_swevent_int,
-	.stop		= perf_swevent_void,
-	.read		= perf_swevent_read,
-	.unthrottle	= perf_swevent_void, /* hwc->interrupts already reset */
-};
-
-/*
- * hrtimer based swevent callback
- */
-
-static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer)
-{
-	enum hrtimer_restart ret = HRTIMER_RESTART;
-	struct perf_sample_data data;
-	struct pt_regs *regs;
-	struct perf_event *event;
-	u64 period;
-
-	event = container_of(hrtimer, struct perf_event, hw.hrtimer);
-	event->pmu->read(event);
-
-	perf_sample_data_init(&data, 0);
-	data.period = event->hw.last_period;
-	regs = get_irq_regs();
-
-	if (regs && !perf_exclude_event(event, regs)) {
-		if (!(event->attr.exclude_idle && current->pid == 0))
-			if (perf_event_overflow(event, 0, &data, regs))
-				ret = HRTIMER_NORESTART;
-	}
-
-	period = max_t(u64, 10000, event->hw.sample_period);
-	hrtimer_forward_now(hrtimer, ns_to_ktime(period));
-
-	return ret;
-}
-
-static void perf_swevent_start_hrtimer(struct perf_event *event)
-{
-	struct hw_perf_event *hwc = &event->hw;
-
-	hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
-	hwc->hrtimer.function = perf_swevent_hrtimer;
-	if (hwc->sample_period) {
-		u64 period;
-
-		if (hwc->remaining) {
-			if (hwc->remaining < 0)
-				period = 10000;
-			else
-				period = hwc->remaining;
-			hwc->remaining = 0;
-		} else {
-			period = max_t(u64, 10000, hwc->sample_period);
-		}
-		__hrtimer_start_range_ns(&hwc->hrtimer,
-				ns_to_ktime(period), 0,
-				HRTIMER_MODE_REL, 0);
-	}
-}
-
-static void perf_swevent_cancel_hrtimer(struct perf_event *event)
-{
-	struct hw_perf_event *hwc = &event->hw;
-
-	if (hwc->sample_period) {
-		ktime_t remaining = hrtimer_get_remaining(&hwc->hrtimer);
-		hwc->remaining = ktime_to_ns(remaining);
-
-		hrtimer_cancel(&hwc->hrtimer);
-	}
-}
-
-/*
- * Software event: cpu wall time clock
- */
-
-static void cpu_clock_perf_event_update(struct perf_event *event)
-{
-	int cpu = raw_smp_processor_id();
-	s64 prev;
-	u64 now;
-
-	now = cpu_clock(cpu);
-	prev = local64_xchg(&event->hw.prev_count, now);
-	local64_add(now - prev, &event->count);
-}
-
-static int cpu_clock_perf_event_enable(struct perf_event *event)
-{
-	struct hw_perf_event *hwc = &event->hw;
-	int cpu = raw_smp_processor_id();
-
-	local64_set(&hwc->prev_count, cpu_clock(cpu));
-	perf_swevent_start_hrtimer(event);
-
-	return 0;
-}
-
-static void cpu_clock_perf_event_disable(struct perf_event *event)
-{
-	perf_swevent_cancel_hrtimer(event);
-	cpu_clock_perf_event_update(event);
-}
-
-static void cpu_clock_perf_event_read(struct perf_event *event)
-{
-	cpu_clock_perf_event_update(event);
-}
-
-static struct pmu perf_ops_cpu_clock = {
-	.enable		= cpu_clock_perf_event_enable,
-	.disable	= cpu_clock_perf_event_disable,
-	.read		= cpu_clock_perf_event_read,
-};
-
-/*
- * Software event: task time clock
- */
-
-static void task_clock_perf_event_update(struct perf_event *event, u64 now)
-{
-	u64 prev;
-	s64 delta;
-
-	prev = local64_xchg(&event->hw.prev_count, now);
-	delta = now - prev;
-	local64_add(delta, &event->count);
-}
-
-static int task_clock_perf_event_enable(struct perf_event *event)
-{
-	struct hw_perf_event *hwc = &event->hw;
-	u64 now;
-
-	now = event->ctx->time;
-
-	local64_set(&hwc->prev_count, now);
-
-	perf_swevent_start_hrtimer(event);
-
-	return 0;
-}
-
-static void task_clock_perf_event_disable(struct perf_event *event)
-{
-	perf_swevent_cancel_hrtimer(event);
-	task_clock_perf_event_update(event, event->ctx->time);
-
-}
-
-static void task_clock_perf_event_read(struct perf_event *event)
-{
-	u64 time;
-
-	if (!in_nmi()) {
-		update_context_time(event->ctx);
-		time = event->ctx->time;
-	} else {
-		u64 now = perf_clock();
-		u64 delta = now - event->ctx->timestamp;
-		time = event->ctx->time + delta;
-	}
-
-	task_clock_perf_event_update(event, time);
-}
-
-static struct pmu perf_ops_task_clock = {
-	.enable		= task_clock_perf_event_enable,
-	.disable	= task_clock_perf_event_disable,
-	.read		= task_clock_perf_event_read,
-};
-
 /* Deref the hlist from the update side */
 static inline struct swevent_hlist *
 swevent_hlist_deref(struct perf_cpu_context *cpuctx)
@@ -4573,17 +4388,63 @@ static int swevent_hlist_get(struct perf
 	return err;
 }
 
-#ifdef CONFIG_EVENT_TRACING
+atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX];
 
-static struct pmu perf_ops_tracepoint = {
-	.enable		= perf_trace_enable,
-	.disable	= perf_trace_disable,
+static void sw_perf_event_destroy(struct perf_event *event)
+{
+	u64 event_id = event->attr.config;
+
+	WARN_ON(event->parent);
+
+	atomic_dec(&perf_swevent_enabled[event_id]);
+	swevent_hlist_put(event);
+}
+
+static int perf_swevent_init(struct perf_event *event)
+{
+	int event_id = event->attr.config;
+
+	if (event->attr.type != PERF_TYPE_SOFTWARE)
+		return -ENOENT;
+
+	switch (event_id) {
+	case PERF_COUNT_SW_CPU_CLOCK:
+	case PERF_COUNT_SW_TASK_CLOCK:
+		return -ENOENT;
+
+	default:
+		break;
+	}
+
+	if (event_id > PERF_COUNT_SW_MAX)
+		return -ENOENT;
+
+	if (!event->parent) {
+		int err;
+
+		err = swevent_hlist_get(event);
+		if (err)
+			return err;
+
+		atomic_inc(&perf_swevent_enabled[event_id]);
+		event->destroy = sw_perf_event_destroy;
+	}
+
+	return 0;
+}
+
+static struct pmu perf_swevent = {
+	.event_init	= perf_swevent_init,
+	.enable		= perf_swevent_enable,
+	.disable	= perf_swevent_disable,
 	.start		= perf_swevent_int,
 	.stop		= perf_swevent_void,
 	.read		= perf_swevent_read,
-	.unthrottle	= perf_swevent_void,
+	.unthrottle	= perf_swevent_void, /* hwc->interrupts already reset */
 };
 
+#ifdef CONFIG_EVENT_TRACING
+
 static int perf_tp_filter_match(struct perf_event *event,
 				struct perf_sample_data *data)
 {
@@ -4639,10 +4500,13 @@ static void tp_perf_event_destroy(struct
 	perf_trace_destroy(event);
 }
 
-static struct pmu *tp_perf_event_init(struct perf_event *event)
+static int perf_tp_event_init(struct perf_event *event)
 {
 	int err;
 
+	if (event->attr.type != PERF_TYPE_TRACEPOINT)
+		return -ENOENT;
+
 	/*
 	 * Raw tracepoint data is a severe data leak, only allow root to
 	 * have these.
@@ -4650,15 +4514,30 @@ static struct pmu *tp_perf_event_init(st
 	if ((event->attr.sample_type & PERF_SAMPLE_RAW) &&
 			perf_paranoid_tracepoint_raw() &&
 			!capable(CAP_SYS_ADMIN))
-		return ERR_PTR(-EPERM);
+		return -EPERM;
 
 	err = perf_trace_init(event);
 	if (err)
-		return NULL;
+		return err;
 
 	event->destroy = tp_perf_event_destroy;
 
-	return &perf_ops_tracepoint;
+	return 0;
+}
+
+static struct pmu perf_tracepoint = {
+	.event_init	= perf_tp_event_init,
+	.enable		= perf_trace_enable,
+	.disable	= perf_trace_disable,
+	.start		= perf_swevent_int,
+	.stop		= perf_swevent_void,
+	.read		= perf_swevent_read,
+	.unthrottle	= perf_swevent_void,
+};
+
+static inline void perf_tp_register(void)
+{
+	perf_pmu_register(&perf_tracepoint);
 }
 
 static int perf_event_set_filter(struct perf_event *event, void __user *arg)
@@ -4686,9 +4565,8 @@ static void perf_event_free_filter(struc
 
 #else
 
-static struct pmu *tp_perf_event_init(struct perf_event *event)
+static inline void perf_tp_register(void)
 {
-	return NULL;
 }
 
 static int perf_event_set_filter(struct perf_event *event, void __user *arg)
@@ -4703,105 +4581,247 @@ static void perf_event_free_filter(struc
 #endif /* CONFIG_EVENT_TRACING */
 
 #ifdef CONFIG_HAVE_HW_BREAKPOINT
-static void bp_perf_event_destroy(struct perf_event *event)
+void perf_bp_event(struct perf_event *bp, void *data)
 {
-	release_bp_slot(event);
+	struct perf_sample_data sample;
+	struct pt_regs *regs = data;
+
+	perf_sample_data_init(&sample, bp->attr.bp_addr);
+
+	if (!perf_exclude_event(bp, regs))
+		perf_swevent_add(bp, 1, 1, &sample, regs);
 }
+#endif
 
-static struct pmu *bp_perf_event_init(struct perf_event *bp)
+/*
+ * hrtimer based swevent callback
+ */
+
+static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer)
 {
-	int err;
+	enum hrtimer_restart ret = HRTIMER_RESTART;
+	struct perf_sample_data data;
+	struct pt_regs *regs;
+	struct perf_event *event;
+	u64 period;
 
-	err = register_perf_hw_breakpoint(bp);
-	if (err)
-		return ERR_PTR(err);
+	event = container_of(hrtimer, struct perf_event, hw.hrtimer);
+	event->pmu->read(event);
+
+	perf_sample_data_init(&data, 0);
+	data.period = event->hw.last_period;
+	regs = get_irq_regs();
+
+	if (regs && !perf_exclude_event(event, regs)) {
+		if (!(event->attr.exclude_idle && current->pid == 0))
+			if (perf_event_overflow(event, 0, &data, regs))
+				ret = HRTIMER_NORESTART;
+	}
 
-	bp->destroy = bp_perf_event_destroy;
+	period = max_t(u64, 10000, event->hw.sample_period);
+	hrtimer_forward_now(hrtimer, ns_to_ktime(period));
 
-	return &perf_ops_bp;
+	return ret;
 }
 
-void perf_bp_event(struct perf_event *bp, void *data)
+static void perf_swevent_start_hrtimer(struct perf_event *event)
 {
-	struct perf_sample_data sample;
-	struct pt_regs *regs = data;
+	struct hw_perf_event *hwc = &event->hw;
 
-	perf_sample_data_init(&sample, bp->attr.bp_addr);
+	hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+	hwc->hrtimer.function = perf_swevent_hrtimer;
+	if (hwc->sample_period) {
+		u64 period;
 
-	if (!perf_exclude_event(bp, regs))
-		perf_swevent_add(bp, 1, 1, &sample, regs);
+		if (hwc->remaining) {
+			if (hwc->remaining < 0)
+				period = 10000;
+			else
+				period = hwc->remaining;
+			hwc->remaining = 0;
+		} else {
+			period = max_t(u64, 10000, hwc->sample_period);
+		}
+		__hrtimer_start_range_ns(&hwc->hrtimer,
+				ns_to_ktime(period), 0,
+				HRTIMER_MODE_REL, 0);
+	}
 }
-#else
-static struct pmu *bp_perf_event_init(struct perf_event *bp)
+
+static void perf_swevent_cancel_hrtimer(struct perf_event *event)
 {
-	return NULL;
+	struct hw_perf_event *hwc = &event->hw;
+
+	if (hwc->sample_period) {
+		ktime_t remaining = hrtimer_get_remaining(&hwc->hrtimer);
+		hwc->remaining = ktime_to_ns(remaining);
+
+		hrtimer_cancel(&hwc->hrtimer);
+	}
 }
 
-void perf_bp_event(struct perf_event *bp, void *regs)
+/*
+ * Software event: cpu wall time clock
+ */
+
+static void cpu_clock_event_update(struct perf_event *event)
 {
+	int cpu = raw_smp_processor_id();
+	s64 prev;
+	u64 now;
+
+	now = cpu_clock(cpu);
+	prev = local64_xchg(&event->hw.prev_count, now);
+	local64_add(now - prev, &event->count);
 }
-#endif
 
-atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX];
+static int cpu_clock_event_enable(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	int cpu = raw_smp_processor_id();
 
-static void sw_perf_event_destroy(struct perf_event *event)
+	local64_set(&hwc->prev_count, cpu_clock(cpu));
+	perf_swevent_start_hrtimer(event);
+
+	return 0;
+}
+
+static void cpu_clock_event_disable(struct perf_event *event)
 {
-	u64 event_id = event->attr.config;
+	perf_swevent_cancel_hrtimer(event);
+	cpu_clock_event_update(event);
+}
 
-	WARN_ON(event->parent);
+static void cpu_clock_event_read(struct perf_event *event)
+{
+	cpu_clock_event_update(event);
+}
 
-	atomic_dec(&perf_swevent_enabled[event_id]);
-	swevent_hlist_put(event);
+static int cpu_clock_event_init(struct perf_event *event)
+{
+	if (event->attr.type != PERF_TYPE_SOFTWARE)
+		return -ENOENT;
+
+	if (event->attr.config != PERF_COUNT_SW_CPU_CLOCK)
+		return -ENOENT;
+
+	return 0;
 }
 
-static struct pmu *sw_perf_event_init(struct perf_event *event)
+static struct pmu perf_cpu_clock = {
+	.event_init	= cpu_clock_event_init,
+	.enable		= cpu_clock_event_enable,
+	.disable	= cpu_clock_event_disable,
+	.read		= cpu_clock_event_read,
+};
+
+/*
+ * Software event: task time clock
+ */
+
+static void task_clock_event_update(struct perf_event *event, u64 now)
 {
-	struct pmu *pmu = NULL;
-	u64 event_id = event->attr.config;
+	u64 prev;
+	s64 delta;
 
-	/*
-	 * Software events (currently) can't in general distinguish
-	 * between user, kernel and hypervisor events.
-	 * However, context switches and cpu migrations are considered
-	 * to be kernel events, and page faults are never hypervisor
-	 * events.
-	 */
-	switch (event_id) {
-	case PERF_COUNT_SW_CPU_CLOCK:
-		pmu = &perf_ops_cpu_clock;
+	prev = local64_xchg(&event->hw.prev_count, now);
+	delta = now - prev;
+	local64_add(delta, &event->count);
+}
 
-		break;
-	case PERF_COUNT_SW_TASK_CLOCK:
-		/*
-		 * If the user instantiates this as a per-cpu event,
-		 * use the cpu_clock event instead.
-		 */
-		if (event->ctx->task)
-			pmu = &perf_ops_task_clock;
-		else
-			pmu = &perf_ops_cpu_clock;
+static int task_clock_event_enable(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u64 now;
 
-		break;
-	case PERF_COUNT_SW_PAGE_FAULTS:
-	case PERF_COUNT_SW_PAGE_FAULTS_MIN:
-	case PERF_COUNT_SW_PAGE_FAULTS_MAJ:
-	case PERF_COUNT_SW_CONTEXT_SWITCHES:
-	case PERF_COUNT_SW_CPU_MIGRATIONS:
-	case PERF_COUNT_SW_ALIGNMENT_FAULTS:
-	case PERF_COUNT_SW_EMULATION_FAULTS:
-		if (!event->parent) {
-			int err;
-
-			err = swevent_hlist_get(event);
-			if (err)
-				return ERR_PTR(err);
+	now = event->ctx->time;
+
+	local64_set(&hwc->prev_count, now);
+
+	perf_swevent_start_hrtimer(event);
+
+	return 0;
+}
+
+static void task_clock_event_disable(struct perf_event *event)
+{
+	perf_swevent_cancel_hrtimer(event);
+	task_clock_event_update(event, event->ctx->time);
+
+}
+
+static void task_clock_event_read(struct perf_event *event)
+{
+	u64 time;
+
+	if (!in_nmi()) {
+		update_context_time(event->ctx);
+		time = event->ctx->time;
+	} else {
+		u64 now = perf_clock();
+		u64 delta = now - event->ctx->timestamp;
+		time = event->ctx->time + delta;
+	}
+
+	task_clock_event_update(event, time);
+}
+
+static int task_clock_event_init(struct perf_event *event)
+{
+	if (event->attr.type != PERF_TYPE_SOFTWARE)
+		return -ENOENT;
+
+	if (event->attr.config != PERF_COUNT_SW_TASK_CLOCK)
+		return -ENOENT;
+
+	return 0;
+}
+
+static struct pmu perf_task_clock = {
+	.event_init	= task_clock_event_init,
+	.enable		= task_clock_event_enable,
+	.disable	= task_clock_event_disable,
+	.read		= task_clock_event_read,
+};
+
+static LIST_HEAD(pmus);
+static DEFINE_MUTEX(pmus_lock);
+static struct srcu_struct pmus_srcu;
+
+int perf_pmu_register(struct pmu *pmu)
+{
+	mutex_lock(&pmus_lock);
+	list_add_rcu(&pmu->entry, &pmus);
+	mutex_unlock(&pmus_lock);
+
+	return 0;
+}
+
+void perf_pmu_unregister(struct pmu *pmu)
+{
+	mutex_lock(&pmus_lock);
+	list_del_rcu(&pmu->entry);
+	mutex_unlock(&pmus_lock);
 
-			atomic_inc(&perf_swevent_enabled[event_id]);
-			event->destroy = sw_perf_event_destroy;
+	synchronize_srcu(&pmus_srcu);
+}
+
+struct pmu *perf_init_event(struct perf_event *event)
+{
+	struct pmu *pmu = NULL;
+	int idx;
+
+	idx = srcu_read_lock(&pmus_srcu);
+	list_for_each_entry_rcu(pmu, &pmus, entry) {
+		int ret = pmu->event_init(event);
+		if (!ret)
+			break;
+		if (ret != -ENOENT) {
+			pmu = ERR_PTR(ret);
+			break;
 		}
-		pmu = &perf_ops_generic;
-		break;
 	}
+	srcu_read_unlock(&pmus_srcu, idx);
 
 	return pmu;
 }
@@ -4882,29 +4902,8 @@ perf_event_alloc(struct perf_event_attr 
 	if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP))
 		goto done;
 
-	switch (attr->type) {
-	case PERF_TYPE_RAW:
-	case PERF_TYPE_HARDWARE:
-	case PERF_TYPE_HW_CACHE:
-		pmu = hw_perf_event_init(event);
-		break;
-
-	case PERF_TYPE_SOFTWARE:
-		pmu = sw_perf_event_init(event);
-		break;
-
-	case PERF_TYPE_TRACEPOINT:
-		pmu = tp_perf_event_init(event);
-		break;
+	pmu = perf_init_event(event);
 
-	case PERF_TYPE_BREAKPOINT:
-		pmu = bp_perf_event_init(event);
-		break;
-
-
-	default:
-		break;
-	}
 done:
 	err = 0;
 	if (!pmu)
@@ -5762,22 +5761,15 @@ perf_cpu_notify(struct notifier_block *s
 	return NOTIFY_OK;
 }
 
-/*
- * This has to have a higher priority than migration_notifier in sched.c.
- */
-static struct notifier_block __cpuinitdata perf_cpu_nb = {
-	.notifier_call		= perf_cpu_notify,
-	.priority		= 20,
-};
-
 void __init perf_event_init(void)
 {
 	perf_event_init_all_cpus();
-	perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE,
-			(void *)(long)smp_processor_id());
-	perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE,
-			(void *)(long)smp_processor_id());
-	register_cpu_notifier(&perf_cpu_nb);
+	init_srcu_struct(&pmus_srcu);
+	perf_pmu_register(&perf_swevent);
+	perf_pmu_register(&perf_cpu_clock);
+	perf_pmu_register(&perf_task_clock);
+	perf_tp_register();
+	perf_cpu_notifier(perf_cpu_notify);
 }
 
 static ssize_t perf_show_reserve_percpu(struct sysdev_class *class,

[RFC][PATCH 08/13] perf: Unindent labels

[Peter Zijlstra]

[-- Attachment #1: perf-whitespace.patch --]
[-- Type: text/plain, Size: 4256 bytes --]

Fixup random annoying style bits

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 kernel/perf_event.c |   43 ++++++++++++++++++++++++-------------------
 1 file changed, 24 insertions(+), 19 deletions(-)

Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -147,7 +147,7 @@ perf_lock_task_context(struct task_struc
 	struct perf_event_context *ctx;
 
 	rcu_read_lock();
- retry:
+retry:
 	ctx = rcu_dereference(task->perf_event_ctxp);
 	if (ctx) {
 		/*
@@ -601,7 +601,7 @@ void perf_event_disable(struct perf_even
 		return;
 	}
 
- retry:
+retry:
 	task_oncpu_function_call(task, __perf_event_disable, event);
 
 	raw_spin_lock_irq(&ctx->lock);
@@ -831,7 +831,7 @@ static void __perf_install_in_context(vo
 	if (!err && !ctx->task && cpuctx->max_pertask)
 		cpuctx->max_pertask--;
 
- unlock:
+unlock:
 	perf_enable();
 
 	raw_spin_unlock(&ctx->lock);
@@ -904,10 +904,12 @@ static void __perf_event_mark_enabled(st
 
 	event->state = PERF_EVENT_STATE_INACTIVE;
 	event->tstamp_enabled = ctx->time - event->total_time_enabled;
-	list_for_each_entry(sub, &event->sibling_list, group_entry)
-		if (sub->state >= PERF_EVENT_STATE_INACTIVE)
+	list_for_each_entry(sub, &event->sibling_list, group_entry) {
+		if (sub->state >= PERF_EVENT_STATE_INACTIVE) {
 			sub->tstamp_enabled =
 				ctx->time - sub->total_time_enabled;
+		}
+	}
 }
 
 /*
@@ -973,7 +975,7 @@ static void __perf_event_enable(void *in
 		}
 	}
 
- unlock:
+unlock:
 	raw_spin_unlock(&ctx->lock);
 }
 
@@ -1014,7 +1016,7 @@ void perf_event_enable(struct perf_event
 	if (event->state == PERF_EVENT_STATE_ERROR)
 		event->state = PERF_EVENT_STATE_OFF;
 
- retry:
+retry:
 	raw_spin_unlock_irq(&ctx->lock);
 	task_oncpu_function_call(task, __perf_event_enable, event);
 
@@ -1034,7 +1036,7 @@ void perf_event_enable(struct perf_event
 	if (event->state == PERF_EVENT_STATE_OFF)
 		__perf_event_mark_enabled(event, ctx);
 
- out:
+out:
 	raw_spin_unlock_irq(&ctx->lock);
 }
 
@@ -1074,17 +1076,19 @@ static void ctx_sched_out(struct perf_ev
 	if (!ctx->nr_active)
 		goto out_enable;
 
-	if (event_type & EVENT_PINNED)
+	if (event_type & EVENT_PINNED) {
 		list_for_each_entry(event, &ctx->pinned_groups, group_entry)
 			group_sched_out(event, cpuctx, ctx);
+	}
 
-	if (event_type & EVENT_FLEXIBLE)
+	if (event_type & EVENT_FLEXIBLE) {
 		list_for_each_entry(event, &ctx->flexible_groups, group_entry)
 			group_sched_out(event, cpuctx, ctx);
+	}
 
  out_enable:
 	perf_enable();
- out:
+out:
 	raw_spin_unlock(&ctx->lock);
 }
 
@@ -1323,9 +1327,10 @@ ctx_flexible_sched_in(struct perf_event_
 		if (event->cpu != -1 && event->cpu != smp_processor_id())
 			continue;
 
-		if (group_can_go_on(event, cpuctx, can_add_hw))
+		if (group_can_go_on(event, cpuctx, can_add_hw)) {
 			if (group_sched_in(event, cpuctx, ctx))
 				can_add_hw = 0;
+		}
 	}
 }
 
@@ -1355,7 +1360,7 @@ ctx_sched_in(struct perf_event_context *
 		ctx_flexible_sched_in(ctx, cpuctx);
 
 	perf_enable();
- out:
+out:
 	raw_spin_unlock(&ctx->lock);
 }
 
@@ -1696,7 +1701,7 @@ static void perf_event_enable_on_exec(st
 	raw_spin_unlock(&ctx->lock);
 
 	perf_event_task_sched_in(task);
- out:
+out:
 	local_irq_restore(flags);
 }
 
@@ -1825,7 +1830,7 @@ static struct perf_event_context *find_g
 	if (!ptrace_may_access(task, PTRACE_MODE_READ))
 		goto errout;
 
- retry:
+retry:
 	ctx = perf_lock_task_context(task, &flags);
 	if (ctx) {
 		unclone_ctx(ctx);
@@ -1853,7 +1858,7 @@ static struct perf_event_context *find_g
 	put_task_struct(task);
 	return ctx;
 
- errout:
+errout:
 	put_task_struct(task);
 	return ERR_PTR(err);
 }
@@ -3044,7 +3049,7 @@ again:
 	if (handle->wakeup != local_read(&buffer->wakeup))
 		perf_output_wakeup(handle);
 
- out:
+out:
 	preempt_enable();
 }
 
@@ -4352,7 +4357,7 @@ static int swevent_hlist_get_cpu(struct 
 		rcu_assign_pointer(cpuctx->swevent_hlist, hlist);
 	}
 	cpuctx->hlist_refcount++;
- exit:
+exit:
 	mutex_unlock(&cpuctx->hlist_mutex);
 
 	return err;
@@ -4377,7 +4382,7 @@ static int swevent_hlist_get(struct perf
 	put_online_cpus();
 
 	return 0;
- fail:
+fail:
 	for_each_possible_cpu(cpu) {
 		if (cpu == failed_cpu)
 			break;

[RFC][PATCH 09/13] perf: Reduce perf_disable() usage

[Peter Zijlstra]

[-- Attachment #1: perf-less-disable-1.patch --]
[-- Type: text/plain, Size: 11232 bytes --]

Since the current perf_disable() usage is only an optimization, remove
it for now. This eases the removal of the weak hw_perf_enable
interface.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 arch/powerpc/kernel/perf_event.c         |    3 ++
 arch/powerpc/kernel/perf_event_fsl_emb.c |    8 +++++-
 arch/sparc/kernel/perf_event.c           |    3 ++
 arch/x86/kernel/cpu/perf_event.c         |   22 +++++++++++-------
 include/linux/perf_event.h               |   20 ++++++++--------
 kernel/perf_event.c                      |   37 -------------------------------
 6 files changed, 37 insertions(+), 56 deletions(-)

Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -562,26 +562,26 @@ struct pmu {
 	struct list_head		entry;
 
 	/*
-	 * Should return -ENOENT when the @event doesn't match this pmu
+	 * Should return -ENOENT when the @event doesn't match this PMU.
 	 */
 	int (*event_init)		(struct perf_event *event);
 
-	int (*enable)			(struct perf_event *event);
+	int  (*enable)			(struct perf_event *event);
 	void (*disable)			(struct perf_event *event);
-	int (*start)			(struct perf_event *event);
+	int  (*start)			(struct perf_event *event);
 	void (*stop)			(struct perf_event *event);
 	void (*read)			(struct perf_event *event);
 	void (*unthrottle)		(struct perf_event *event);
 
 	/*
-	 * Group events scheduling is treated as a transaction, add group
-	 * events as a whole and perform one schedulability test. If the test
-	 * fails, roll back the whole group
+	 * Group events scheduling is treated as a transaction, add
+	 * group events as a whole and perform one schedulability test.
+	 * If the test fails, roll back the whole group
 	 */
 
 	/*
-	 * Start the transaction, after this ->enable() doesn't need
-	 * to do schedulability tests.
+	 * Start the transaction, after this ->enable() doesn't need to
+	 * do schedulability tests.
 	 */
 	void (*start_txn)	(struct pmu *pmu);
 	/*
@@ -592,8 +592,8 @@ struct pmu {
 	 */
 	int  (*commit_txn)	(struct pmu *pmu);
 	/*
-	 * Will cancel the transaction, assumes ->disable() is called for
-	 * each successfull ->enable() during the transaction.
+	 * Will cancel the transaction, assumes ->disable() is called
+	 * for each successfull ->enable() during the transaction.
 	 */
 	void (*cancel_txn)	(struct pmu *pmu);
 };
Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -460,11 +460,6 @@ static void __perf_event_remove_from_con
 		return;
 
 	raw_spin_lock(&ctx->lock);
-	/*
-	 * Protect the list operation against NMI by disabling the
-	 * events on a global level.
-	 */
-	perf_disable();
 
 	event_sched_out(event, cpuctx, ctx);
 
@@ -480,7 +475,6 @@ static void __perf_event_remove_from_con
 			    perf_max_events - perf_reserved_percpu);
 	}
 
-	perf_enable();
 	raw_spin_unlock(&ctx->lock);
 }
 
@@ -785,12 +779,6 @@ static void __perf_install_in_context(vo
 	ctx->is_active = 1;
 	update_context_time(ctx);
 
-	/*
-	 * Protect the list operation against NMI by disabling the
-	 * events on a global level. NOP for non NMI based events.
-	 */
-	perf_disable();
-
 	add_event_to_ctx(event, ctx);
 
 	if (event->cpu != -1 && event->cpu != smp_processor_id())
@@ -832,8 +820,6 @@ static void __perf_install_in_context(vo
 		cpuctx->max_pertask--;
 
 unlock:
-	perf_enable();
-
 	raw_spin_unlock(&ctx->lock);
 }
 
@@ -954,12 +940,10 @@ static void __perf_event_enable(void *in
 	if (!group_can_go_on(event, cpuctx, 1)) {
 		err = -EEXIST;
 	} else {
-		perf_disable();
 		if (event == leader)
 			err = group_sched_in(event, cpuctx, ctx);
 		else
 			err = event_sched_in(event, cpuctx, ctx);
-		perf_enable();
 	}
 
 	if (err) {
@@ -1072,9 +1056,8 @@ static void ctx_sched_out(struct perf_ev
 		goto out;
 	update_context_time(ctx);
 
-	perf_disable();
 	if (!ctx->nr_active)
-		goto out_enable;
+		goto out;
 
 	if (event_type & EVENT_PINNED) {
 		list_for_each_entry(event, &ctx->pinned_groups, group_entry)
@@ -1085,9 +1068,6 @@ static void ctx_sched_out(struct perf_ev
 		list_for_each_entry(event, &ctx->flexible_groups, group_entry)
 			group_sched_out(event, cpuctx, ctx);
 	}
-
- out_enable:
-	perf_enable();
 out:
 	raw_spin_unlock(&ctx->lock);
 }
@@ -1346,8 +1326,6 @@ ctx_sched_in(struct perf_event_context *
 
 	ctx->timestamp = perf_clock();
 
-	perf_disable();
-
 	/*
 	 * First go through the list and put on any pinned groups
 	 * in order to give them the best chance of going on.
@@ -1359,7 +1337,6 @@ ctx_sched_in(struct perf_event_context *
 	if (event_type & EVENT_FLEXIBLE)
 		ctx_flexible_sched_in(ctx, cpuctx);
 
-	perf_enable();
 out:
 	raw_spin_unlock(&ctx->lock);
 }
@@ -1407,8 +1384,6 @@ void perf_event_task_sched_in(struct tas
 	if (cpuctx->task_ctx == ctx)
 		return;
 
-	perf_disable();
-
 	/*
 	 * We want to keep the following priority order:
 	 * cpu pinned (that don't need to move), task pinned,
@@ -1421,8 +1396,6 @@ void perf_event_task_sched_in(struct tas
 	ctx_sched_in(ctx, cpuctx, EVENT_FLEXIBLE);
 
 	cpuctx->task_ctx = ctx;
-
-	perf_enable();
 }
 
 #define MAX_INTERRUPTS (~0ULL)
@@ -1537,11 +1510,9 @@ static void perf_adjust_period(struct pe
 	hwc->sample_period = sample_period;
 
 	if (local64_read(&hwc->period_left) > 8*sample_period) {
-		perf_disable();
 		perf_event_stop(event);
 		local64_set(&hwc->period_left, 0);
 		perf_event_start(event);
-		perf_enable();
 	}
 }
 
@@ -1570,15 +1541,12 @@ static void perf_ctx_adjust_freq(struct 
 		 */
 		if (interrupts == MAX_INTERRUPTS) {
 			perf_log_throttle(event, 1);
-			perf_disable();
 			event->pmu->unthrottle(event);
-			perf_enable();
 		}
 
 		if (!event->attr.freq || !event->attr.sample_freq)
 			continue;
 
-		perf_disable();
 		event->pmu->read(event);
 		now = local64_read(&event->count);
 		delta = now - hwc->freq_count_stamp;
@@ -1586,7 +1554,6 @@ static void perf_ctx_adjust_freq(struct 
 
 		if (delta > 0)
 			perf_adjust_period(event, TICK_NSEC, delta);
-		perf_enable();
 	}
 	raw_spin_unlock(&ctx->lock);
 }
@@ -1629,7 +1596,6 @@ void perf_event_task_tick(struct task_st
 	if (!rotate)
 		return;
 
-	perf_disable();
 	cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
 	if (ctx)
 		task_ctx_sched_out(ctx, EVENT_FLEXIBLE);
@@ -1641,7 +1607,6 @@ void perf_event_task_tick(struct task_st
 	cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE);
 	if (ctx)
 		task_ctx_sched_in(curr, EVENT_FLEXIBLE);
-	perf_enable();
 }
 
 static int event_enable_on_exec(struct perf_event *event,
Index: linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event_fsl_emb.c
+++ linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
@@ -262,7 +262,7 @@ static int collect_events(struct perf_ev
 	return n;
 }
 
-/* perf must be disabled, context locked on entry */
+/* context locked on entry */
 static int fsl_emb_pmu_enable(struct perf_event *event)
 {
 	struct cpu_hw_events *cpuhw;
@@ -271,6 +271,7 @@ static int fsl_emb_pmu_enable(struct per
 	u64 val;
 	int i;
 
+	perf_disable();
 	cpuhw = &get_cpu_var(cpu_hw_events);
 
 	if (event->hw.config & FSL_EMB_EVENT_RESTRICTED)
@@ -310,15 +311,17 @@ static int fsl_emb_pmu_enable(struct per
 	ret = 0;
  out:
 	put_cpu_var(cpu_hw_events);
+	perf_enable();
 	return ret;
 }
 
-/* perf must be disabled, context locked on entry */
+/* context locked on entry */
 static void fsl_emb_pmu_disable(struct perf_event *event)
 {
 	struct cpu_hw_events *cpuhw;
 	int i = event->hw.idx;
 
+	perf_disable();
 	if (i < 0)
 		goto out;
 
@@ -346,6 +349,7 @@ static void fsl_emb_pmu_disable(struct p
 	cpuhw->n_events--;
 
  out:
+	perf_enable();
 	put_cpu_var(cpu_hw_events);
 }
 
Index: linux-2.6/arch/x86/kernel/cpu/perf_event.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_event.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_event.c
@@ -969,10 +969,11 @@ static int x86_pmu_enable(struct perf_ev
 
 	hwc = &event->hw;
 
+	perf_disable();
 	n0 = cpuc->n_events;
-	n = collect_events(cpuc, event, false);
-	if (n < 0)
-		return n;
+	ret = n = collect_events(cpuc, event, false);
+	if (ret < 0)
+		goto out;
 
 	/*
 	 * If group events scheduling transaction was started,
@@ -980,23 +981,26 @@ static int x86_pmu_enable(struct perf_ev
 	 * at commit time(->commit_txn) as a whole
 	 */
 	if (cpuc->group_flag & PERF_EVENT_TXN)
-		goto out;
+		goto done_collect;
 
 	ret = x86_pmu.schedule_events(cpuc, n, assign);
 	if (ret)
-		return ret;
+		goto out;
 	/*
 	 * copy new assignment, now we know it is possible
 	 * will be used by hw_perf_enable()
 	 */
 	memcpy(cpuc->assign, assign, n*sizeof(int));
 
-out:
+done_collect:
 	cpuc->n_events = n;
 	cpuc->n_added += n - n0;
 	cpuc->n_txn += n - n0;
 
-	return 0;
+	ret = 0;
+out:
+	perf_enable();
+	return ret;
 }
 
 static int x86_pmu_start(struct perf_event *event)
@@ -1399,6 +1403,7 @@ static void x86_pmu_start_txn(struct pmu
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 
+	perf_disable();
 	cpuc->group_flag |= PERF_EVENT_TXN;
 	cpuc->n_txn = 0;
 }
@@ -1418,6 +1423,7 @@ static void x86_pmu_cancel_txn(struct pm
 	 */
 	cpuc->n_added -= cpuc->n_txn;
 	cpuc->n_events -= cpuc->n_txn;
+	perf_enable();
 }
 
 /*
@@ -1447,7 +1453,7 @@ static int x86_pmu_commit_txn(struct pmu
 	memcpy(cpuc->assign, assign, n*sizeof(int));
 
 	cpuc->group_flag &= ~PERF_EVENT_TXN;
-
+	perf_enable();
 	return 0;
 }
 
Index: linux-2.6/arch/powerpc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event.c
+++ linux-2.6/arch/powerpc/kernel/perf_event.c
@@ -858,6 +858,7 @@ void power_pmu_start_txn(struct pmu *pmu
 {
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
+	perf_disable();
 	cpuhw->group_flag |= PERF_EVENT_TXN;
 	cpuhw->n_txn_start = cpuhw->n_events;
 }
@@ -872,6 +873,7 @@ void power_pmu_cancel_txn(struct pmu *pm
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
 	cpuhw->group_flag &= ~PERF_EVENT_TXN;
+	perf_enable();
 }
 
 /*
@@ -898,6 +900,7 @@ int power_pmu_commit_txn(struct pmu *pmu
 		cpuhw->event[i]->hw.config = cpuhw->events[i];
 
 	cpuhw->group_flag &= ~PERF_EVENT_TXN;
+	perf_enable();
 	return 0;
 }
 
Index: linux-2.6/arch/sparc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sparc/kernel/perf_event.c
+++ linux-2.6/arch/sparc/kernel/perf_event.c
@@ -1112,6 +1112,7 @@ static void sparc_pmu_start_txn(struct p
 {
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
+	perf_disable();
 	cpuhw->group_flag |= PERF_EVENT_TXN;
 }
 
@@ -1125,6 +1126,7 @@ static void sparc_pmu_cancel_txn(struct 
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
 	cpuhw->group_flag &= ~PERF_EVENT_TXN;
+	perf_enable();
 }
 
 /*
@@ -1148,6 +1150,7 @@ static int sparc_pmu_commit_txn(struct p
 		return -EAGAIN;
 
 	cpuc->group_flag &= ~PERF_EVENT_TXN;
+	perf_enable();
 	return 0;
 }
 

[RFC][PATCH 10/13] perf: Per PMU disable

[Peter Zijlstra]

[-- Attachment #1: perf-pmu-disable.patch --]
[-- Type: text/plain, Size: 15792 bytes --]

Changes perf_disable() into perf_pmu_disable().

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 arch/arm/kernel/perf_event.c             |   24 ++++++++++-----------
 arch/powerpc/kernel/perf_event.c         |   26 ++++++++++++-----------
 arch/powerpc/kernel/perf_event_fsl_emb.c |   18 +++++++++-------
 arch/sh/kernel/perf_event.c              |   34 ++++++++++++++++---------------
 arch/sparc/kernel/perf_event.c           |   20 ++++++++++--------
 arch/x86/kernel/cpu/perf_event.c         |   16 ++++++++------
 include/linux/perf_event.h               |   13 ++++++-----
 kernel/perf_event.c                      |   30 ++++++++++++++++-----------
 8 files changed, 99 insertions(+), 82 deletions(-)

Index: linux-2.6/arch/arm/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/arm/kernel/perf_event.c
+++ linux-2.6/arch/arm/kernel/perf_event.c
@@ -531,16 +531,7 @@ static int armpmu_event_init(struct perf
 	return err;
 }
 
-static struct pmu pmu = {
-	.event_init = armpmu_event_init,
-	.enable	    = armpmu_enable,
-	.disable    = armpmu_disable,
-	.unthrottle = armpmu_unthrottle,
-	.read	    = armpmu_read,
-};
-
-void
-hw_perf_enable(void)
+static void armpmu_pmu_enable(struct pmu *pmu)
 {
 	/* Enable all of the perf events on hardware. */
 	int idx;
@@ -561,13 +552,22 @@ hw_perf_enable(void)
 	armpmu->start();
 }
 
-void
-hw_perf_disable(void)
+static void armpmu_pmu_disable(struct pmu *pmu)
 {
 	if (armpmu)
 		armpmu->stop();
 }
 
+static struct pmu pmu = {
+	.pmu_enable = armpmu_pmu_enable,
+	.pmu_disable= armpmu_pmu_disable,
+	.event_init = armpmu_event_init,
+	.enable	    = armpmu_enable,
+	.disable    = armpmu_disable,
+	.unthrottle = armpmu_unthrottle,
+	.read	    = armpmu_read,
+};
+
 /*
  * ARMv6 Performance counter handling code.
  *
Index: linux-2.6/arch/powerpc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event.c
+++ linux-2.6/arch/powerpc/kernel/perf_event.c
@@ -517,7 +517,7 @@ static void write_mmcr0(struct cpu_hw_ev
  * Disable all events to prevent PMU interrupts and to allow
  * events to be added or removed.
  */
-void hw_perf_disable(void)
+static void powerpc_pmu_pmu_disable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw;
 	unsigned long flags;
@@ -565,7 +565,7 @@ void hw_perf_disable(void)
  * If we were previously disabled and events were added, then
  * put the new config on the PMU.
  */
-void hw_perf_enable(void)
+static void powerpc_pmu_pmu_enable(struct pmu *pmu)
 {
 	struct perf_event *event;
 	struct cpu_hw_events *cpuhw;
@@ -735,7 +735,7 @@ static int power_pmu_enable(struct perf_
 	int ret = -EAGAIN;
 
 	local_irq_save(flags);
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 
 	/*
 	 * Add the event to the list (if there is room)
@@ -769,7 +769,7 @@ nocheck:
 
 	ret = 0;
  out:
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
 	return ret;
 }
@@ -784,7 +784,7 @@ static void power_pmu_disable(struct per
 	unsigned long flags;
 
 	local_irq_save(flags);
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 
 	power_pmu_read(event);
 
@@ -818,7 +818,7 @@ static void power_pmu_disable(struct per
 		cpuhw->mmcr[0] &= ~(MMCR0_PMXE | MMCR0_FCECE);
 	}
 
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
 }
 
@@ -834,7 +834,7 @@ static void power_pmu_unthrottle(struct 
 	if (!event->hw.idx || !event->hw.sample_period)
 		return;
 	local_irq_save(flags);
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 	power_pmu_read(event);
 	left = event->hw.sample_period;
 	event->hw.last_period = left;
@@ -845,7 +845,7 @@ static void power_pmu_unthrottle(struct 
 	local64_set(&event->hw.prev_count, val);
 	local64_set(&event->hw.period_left, left);
 	perf_event_update_userpage(event);
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
 }
 
@@ -858,7 +858,7 @@ void power_pmu_start_txn(struct pmu *pmu
 {
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
-	perf_disable();
+	perf_pmu_disable(pmu);
 	cpuhw->group_flag |= PERF_EVENT_TXN;
 	cpuhw->n_txn_start = cpuhw->n_events;
 }
@@ -873,7 +873,7 @@ void power_pmu_cancel_txn(struct pmu *pm
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
 	cpuhw->group_flag &= ~PERF_EVENT_TXN;
-	perf_enable();
+	perf_pmu_enable(pmu);
 }
 
 /*
@@ -900,7 +900,7 @@ int power_pmu_commit_txn(struct pmu *pmu
 		cpuhw->event[i]->hw.config = cpuhw->events[i];
 
 	cpuhw->group_flag &= ~PERF_EVENT_TXN;
-	perf_enable();
+	perf_pmu_enable(pmu);
 	return 0;
 }
 
@@ -1128,7 +1128,9 @@ static int power_pmu_event_init(struct p
 }
 
 struct pmu power_pmu = {
-	.event_init	= power_pmu_event_init,
+	.pmu_enable	= power_pmu_pmu_enable,
+	.pmu_disable	= power_pmu_pmu_disable,
+	.event_init	= pmwer_pmu_event_init,
 	.enable		= power_pmu_enable,
 	.disable	= power_pmu_disable,
 	.read		= power_pmu_read,
Index: linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event_fsl_emb.c
+++ linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
@@ -177,7 +177,7 @@ static void fsl_emb_pmu_read(struct perf
  * Disable all events to prevent PMU interrupts and to allow
  * events to be added or removed.
  */
-void hw_perf_disable(void)
+static void fsl_emb_pmu_pmu_disable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw;
 	unsigned long flags;
@@ -216,7 +216,7 @@ void hw_perf_disable(void)
  * If we were previously disabled and events were added, then
  * put the new config on the PMU.
  */
-void hw_perf_enable(void)
+static void fsl_emb_pmu_pmu_enable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw;
 	unsigned long flags;
@@ -271,7 +271,7 @@ static int fsl_emb_pmu_enable(struct per
 	u64 val;
 	int i;
 
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 	cpuhw = &get_cpu_var(cpu_hw_events);
 
 	if (event->hw.config & FSL_EMB_EVENT_RESTRICTED)
@@ -311,7 +311,7 @@ static int fsl_emb_pmu_enable(struct per
 	ret = 0;
  out:
 	put_cpu_var(cpu_hw_events);
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	return ret;
 }
 
@@ -321,7 +321,7 @@ static void fsl_emb_pmu_disable(struct p
 	struct cpu_hw_events *cpuhw;
 	int i = event->hw.idx;
 
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 	if (i < 0)
 		goto out;
 
@@ -349,7 +349,7 @@ static void fsl_emb_pmu_disable(struct p
 	cpuhw->n_events--;
 
  out:
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	put_cpu_var(cpu_hw_events);
 }
 
@@ -367,7 +367,7 @@ static void fsl_emb_pmu_unthrottle(struc
 	if (event->hw.idx < 0 || !event->hw.sample_period)
 		return;
 	local_irq_save(flags);
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 	fsl_emb_pmu_read(event);
 	left = event->hw.sample_period;
 	event->hw.last_period = left;
@@ -378,7 +378,7 @@ static void fsl_emb_pmu_unthrottle(struc
 	local64_set(&event->hw.prev_count, val);
 	local64_set(&event->hw.period_left, left);
 	perf_event_update_userpage(event);
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
 }
 
@@ -524,6 +524,8 @@ static int fsl_emb_pmu_event_init(struct
 }
 
 static struct pmu fsl_emb_pmu = {
+	.pmu_enable	= fsl_emb_pmu_pmu_enable,
+	.pmu_disable	= fsl_emb_pmu_pmu_disable,
 	.event_init	= fsl_emb_pmu_event_init,
 	.enable		= fsl_emb_pmu_enable,
 	.disable	= fsl_emb_pmu_disable,
Index: linux-2.6/arch/sh/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sh/kernel/perf_event.c
+++ linux-2.6/arch/sh/kernel/perf_event.c
@@ -268,7 +268,25 @@ static in sh_pmu_event_init(struct perf_
 	return err;
 }
 
+static void sh_pmu_pmu_enable(struct pmu *pmu)
+{
+	if (!sh_pmu_initialized())
+		return;
+
+	sh_pmu->enable_all();
+}
+
+static void sh_pmu_pmu_disable(struct pmu *pmu)
+{
+	if (!sh_pmu_initialized())
+		return;
+
+	sh_pmu->disable_all();
+}
+
 static struct pmu pmu = {
+	.pmu_enable	= sh_pmu_pmu_enable,
+	.pmu_disable	= sh_pmu_pmu_disable,
 	.event_init	= sh_pmu_event_init,
 	.enable		= sh_pmu_enable,
 	.disable	= sh_pmu_disable,
@@ -299,22 +317,6 @@ sh_pmu_notifier(struct notifier_block *s
 	return NOTIFY_OK;
 }
 
-void hw_perf_enable(void)
-{
-	if (!sh_pmu_initialized())
-		return;
-
-	sh_pmu->enable_all();
-}
-
-void hw_perf_disable(void)
-{
-	if (!sh_pmu_initialized())
-		return;
-
-	sh_pmu->disable_all();
-}
-
 int __cpuinit register_sh_pmu(struct sh_pmu *pmu)
 {
 	if (sh_pmu)
Index: linux-2.6/arch/sparc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sparc/kernel/perf_event.c
+++ linux-2.6/arch/sparc/kernel/perf_event.c
@@ -664,7 +664,7 @@ out:
 	return pcr;
 }
 
-void hw_perf_enable(void)
+static void sparc_pmu_pmu_enable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	u64 pcr;
@@ -691,7 +691,7 @@ void hw_perf_enable(void)
 	pcr_ops->write(cpuc->pcr);
 }
 
-void hw_perf_disable(void)
+static void sparc_pmu_pmu_disable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	u64 val;
@@ -718,7 +718,7 @@ static void sparc_pmu_disable(struct per
 	int i;
 
 	local_irq_save(flags);
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 
 	for (i = 0; i < cpuc->n_events; i++) {
 		if (event == cpuc->event[i]) {
@@ -748,7 +748,7 @@ static void sparc_pmu_disable(struct per
 		}
 	}
 
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
 }
 
@@ -991,7 +991,7 @@ static int sparc_pmu_enable(struct perf_
 	unsigned long flags;
 
 	local_irq_save(flags);
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 
 	n0 = cpuc->n_events;
 	if (n0 >= perf_max_events)
@@ -1020,7 +1020,7 @@ nocheck:
 
 	ret = 0;
 out:
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
 	return ret;
 }
@@ -1113,7 +1113,7 @@ static void sparc_pmu_start_txn(struct p
 {
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
-	perf_disable();
+	perf_pmu_disable(pmu);
 	cpuhw->group_flag |= PERF_EVENT_TXN;
 }
 
@@ -1127,7 +1127,7 @@ static void sparc_pmu_cancel_txn(struct 
 	struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events);
 
 	cpuhw->group_flag &= ~PERF_EVENT_TXN;
-	perf_enable();
+	perf_pmu_enable(pmu);
 }
 
 /*
@@ -1151,11 +1151,13 @@ static int sparc_pmu_commit_txn(struct p
 		return -EAGAIN;
 
 	cpuc->group_flag &= ~PERF_EVENT_TXN;
-	perf_enable();
+	perf_pmu_enable(pmu);
 	return 0;
 }
 
 static struct pmu pmu = {
+	.pmu_enable	= sparc_pmu_pmu_enable,
+	.pmu_disable	= sparc_pmu_pmu_disable,
 	.event_init	= sparc_pmu_event_init,
 	.enable		= sparc_pmu_enable,
 	.disable	= sparc_pmu_disable,
Index: linux-2.6/arch/x86/kernel/cpu/perf_event.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_event.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_event.c
@@ -583,7 +583,7 @@ static void x86_pmu_disable_all(void)
 	}
 }
 
-void hw_perf_disable(void)
+static void x86_pmu_pmu_disable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 
@@ -803,7 +803,7 @@ static inline int match_prev_assignment(
 static int x86_pmu_start(struct perf_event *event);
 static void x86_pmu_stop(struct perf_event *event);
 
-void hw_perf_enable(void)
+static void x86_pmu_pmu_enable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct perf_event *event;
@@ -969,7 +969,7 @@ static int x86_pmu_enable(struct perf_ev
 
 	hwc = &event->hw;
 
-	perf_disable();
+	perf_pmu_disable(event->pmu);
 	n0 = cpuc->n_events;
 	ret = n = collect_events(cpuc, event, false);
 	if (ret < 0)
@@ -999,7 +999,7 @@ done_collect:
 
 	ret = 0;
 out:
-	perf_enable();
+	perf_pmu_enable(event->pmu);
 	return ret;
 }
 
@@ -1403,7 +1403,7 @@ static void x86_pmu_start_txn(struct pmu
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 
-	perf_disable();
+	perf_pmu_disable(pmu);
 	cpuc->group_flag |= PERF_EVENT_TXN;
 	cpuc->n_txn = 0;
 }
@@ -1423,7 +1423,7 @@ static void x86_pmu_cancel_txn(struct pm
 	 */
 	cpuc->n_added -= cpuc->n_txn;
 	cpuc->n_events -= cpuc->n_txn;
-	perf_enable();
+	perf_pmu_enable(pmu);
 }
 
 /*
@@ -1453,7 +1453,7 @@ static int x86_pmu_commit_txn(struct pmu
 	memcpy(cpuc->assign, assign, n*sizeof(int));
 
 	cpuc->group_flag &= ~PERF_EVENT_TXN;
-	perf_enable();
+	perf_pmu_enable(pmu);
 	return 0;
 }
 
@@ -1572,6 +1572,8 @@ int x86_pmu_event_init(struct perf_event
 }
 
 static struct pmu pmu = {
+	.pmu_enable	= x86_pmu_pmu_enable,
+	.pmu_disable	= x86_pmu_pmu_disable,
 	.event_init	= x86_pmu_event_init,
 	.enable		= x86_pmu_enable,
 	.disable	= x86_pmu_disable,
Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -563,6 +563,11 @@ struct perf_event;
 struct pmu {
 	struct list_head		entry;
 
+	int				*pmu_disable_count;
+
+	void (*pmu_enable)		(struct pmu *pmu);
+	void (*pmu_disable)		(struct pmu *pmu);
+
 	/*
 	 * Should return -ENOENT when the @event doesn't match this PMU.
 	 */
@@ -866,10 +871,8 @@ extern void perf_event_free_task(struct 
 extern void set_perf_event_pending(void);
 extern void perf_event_do_pending(void);
 extern void perf_event_print_debug(void);
-extern void __perf_disable(void);
-extern bool __perf_enable(void);
-extern void perf_disable(void);
-extern void perf_enable(void);
+extern void perf_pmu_disable(struct pmu *pmu);
+extern void perf_pmu_enable(struct pmu *pmu);
 extern int perf_event_task_disable(void);
 extern int perf_event_task_enable(void);
 extern void perf_event_update_userpage(struct perf_event *event);
@@ -1040,8 +1043,6 @@ static inline void perf_event_exit_task(
 static inline void perf_event_free_task(struct task_struct *task)	{ }
 static inline void perf_event_do_pending(void)				{ }
 static inline void perf_event_print_debug(void)				{ }
-static inline void perf_disable(void)					{ }
-static inline void perf_enable(void)					{ }
 static inline int perf_event_task_disable(void)				{ return -EINVAL; }
 static inline int perf_event_task_enable(void)				{ return -EINVAL; }
 
Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -71,23 +71,20 @@ static atomic64_t perf_event_id;
  */
 static DEFINE_SPINLOCK(perf_resource_lock);
 
-void __weak hw_perf_disable(void)		{ barrier(); }
-void __weak hw_perf_enable(void)		{ barrier(); }
-
 void __weak perf_event_print_debug(void)	{ }
 
-static DEFINE_PER_CPU(int, perf_disable_count);
-
-void perf_disable(void)
+void perf_pmu_disable(struct pmu *pmu)
 {
-	if (!__get_cpu_var(perf_disable_count)++)
-		hw_perf_disable();
+	int *count = this_cpu_ptr(pmu->pmu_disable_count);
+	if (!(*count)++)
+		pmu->pmu_disable(pmu);
 }
 
-void perf_enable(void)
+void perf_pmu_enable(struct pmu *pmu)
 {
-	if (!--__get_cpu_var(perf_disable_count))
-		hw_perf_enable();
+	int *count = this_cpu_ptr(pmu->pmu_disable_count);
+	if (!--(*count))
+		pmu->pmu_enable(pmu);
 }
 
 static void get_ctx(struct perf_event_context *ctx)
@@ -4760,16 +4757,25 @@ static struct srcu_struct pmus_srcu;
 
 int perf_pmu_register(struct pmu *pmu)
 {
+	int ret;
+
 	mutex_lock(&pmus_lock);
+	ret = -ENOMEM;
+	pmu->pmu_disable_count = alloc_percpu(int);
+	if (!pmu->pmu_disable_count)
+		goto unlock;
 	list_add_rcu(&pmu->entry, &pmus);
+	ret = 0;
+unlock:
 	mutex_unlock(&pmus_lock);
 
-	return 0;
+	return ret;
 }
 
 void perf_pmu_unregister(struct pmu *pmu)
 {
 	mutex_lock(&pmus_lock);
+	free_percpu(pmu->pmu_disable_count);
 	list_del_rcu(&pmu->entry);
 	mutex_unlock(&pmus_lock);
 

Re: [RFC][PATCH 10/13] perf: Per PMU disable

[Paul Mackerras]

On Fri, Jul 09, 2010 at 10:21:27AM +0200, Peter Zijlstra wrote:

> -void hw_perf_disable(void)
> +static void powerpc_pmu_pmu_disable(struct pmu *pmu)

> -void hw_perf_enable(void)
> +static void powerpc_pmu_pmu_enable(struct pmu *pmu)

>  struct pmu power_pmu = {
> -	.event_init	= power_pmu_event_init,
> +	.pmu_enable	= power_pmu_pmu_enable,
> +	.pmu_disable	= power_pmu_pmu_disable,
> +	.event_init	= pmwer_pmu_event_init,

Doh!  Think you posted the old version of the patch again...

:)

Paul.

Re: [RFC][PATCH 10/13] perf: Per PMU disable

[Peter Zijlstra]

On Fri, 2010-07-09 at 20:17 +1000, Paul Mackerras wrote:
> On Fri, Jul 09, 2010 at 10:21:27AM +0200, Peter Zijlstra wrote:
> 
> > -void hw_perf_disable(void)
> > +static void powerpc_pmu_pmu_disable(struct pmu *pmu)
> 
> > -void hw_perf_enable(void)
> > +static void powerpc_pmu_pmu_enable(struct pmu *pmu)
> 
> >  struct pmu power_pmu = {
> > -	.event_init	= power_pmu_event_init,
> > +	.pmu_enable	= power_pmu_pmu_enable,
> > +	.pmu_disable	= power_pmu_pmu_disable,
> > +	.event_init	= pmwer_pmu_event_init,
> 
> Doh!  Think you posted the old version of the patch again...

*sigh*

Re: [RFC][PATCH 10/13] perf: Per PMU disable

[Peter Zijlstra]

On Fri, 2010-07-09 at 12:30 +0200, Peter Zijlstra wrote:
> On Fri, 2010-07-09 at 20:17 +1000, Paul Mackerras wrote:
> > On Fri, Jul 09, 2010 at 10:21:27AM +0200, Peter Zijlstra wrote:
> > 
> > > -void hw_perf_disable(void)
> > > +static void powerpc_pmu_pmu_disable(struct pmu *pmu)
> > 
> > > -void hw_perf_enable(void)
> > > +static void powerpc_pmu_pmu_enable(struct pmu *pmu)
> > 
> > >  struct pmu power_pmu = {
> > > -	.event_init	= power_pmu_event_init,
> > > +	.pmu_enable	= power_pmu_pmu_enable,
> > > +	.pmu_disable	= power_pmu_pmu_disable,
> > > +	.event_init	= pmwer_pmu_event_init,
> > 
> > Doh!  Think you posted the old version of the patch again...
> 
> *sigh*

The hunks ended up in the last patch.. I've pushed out a new git tree.

[RFC][PATCH 11/13] perf: Default PMU ops

[Peter Zijlstra]

[-- Attachment #1: perf-default-ops.patch --]
[-- Type: text/plain, Size: 4086 bytes --]

Provide default implementations for the pmu txn methods, this allows
us to remove some conditional code.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 include/linux/perf_event.h |   10 +++----
 kernel/perf_event.c        |   64 ++++++++++++++++++++++++++++++++++++---------
 2 files changed, 57 insertions(+), 17 deletions(-)

Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -656,21 +656,14 @@ group_sched_in(struct perf_event *group_
 {
 	struct perf_event *event, *partial_group = NULL;
 	struct pmu *pmu = group_event->pmu;
-	bool txn = false;
 
 	if (group_event->state == PERF_EVENT_STATE_OFF)
 		return 0;
 
-	/* Check if group transaction availabe */
-	if (pmu->start_txn)
-		txn = true;
-
-	if (txn)
-		pmu->start_txn(pmu);
+	pmu->start_txn(pmu);
 
 	if (event_sched_in(group_event, cpuctx, ctx)) {
-		if (txn)
-			pmu->cancel_txn(pmu);
+		pmu->cancel_txn(pmu);
 		return -EAGAIN;
 	}
 
@@ -684,7 +677,7 @@ group_sched_in(struct perf_event *group_
 		}
 	}
 
-	if (!txn || !pmu->commit_txn(pmu))
+	if (!pmu->commit_txn(pmu))
 		return 0;
 
 group_error:
@@ -699,8 +692,7 @@ group_error:
 	}
 	event_sched_out(group_event, cpuctx, ctx);
 
-	if (txn)
-		pmu->cancel_txn(pmu);
+	pmu->cancel_txn(pmu);
 
 	return -EAGAIN;
 }
@@ -4755,6 +4747,31 @@ static LIST_HEAD(pmus);
 static DEFINE_MUTEX(pmus_lock);
 static struct srcu_struct pmus_srcu;
 
+static void perf_pmu_nop_void(struct pmu *pmu)
+{
+}
+
+static int perf_pmu_nop_int(struct pmu *pmu)
+{
+	return 0;
+}
+
+static void perf_pmu_start_txn(struct pmu *pmu)
+{
+	perf_pmu_disable(pmu);
+}
+
+static int perf_pmu_commit_txn(struct pmu *pmu)
+{
+	perf_pmu_enable(pmu);
+	return 0;
+}
+
+static void perf_pmu_cancel_txn(struct pmu *pmu)
+{
+	perf_pmu_enable(pmu);
+}
+
 int perf_pmu_register(struct pmu *pmu)
 {
 	int ret;
@@ -4764,6 +4781,29 @@ int perf_pmu_register(struct pmu *pmu)
 	pmu->pmu_disable_count = alloc_percpu(int);
 	if (!pmu->pmu_disable_count)
 		goto unlock;
+
+	if (!pmu->start_txn) {
+		if (pmu->pmu_enable) {
+			/*
+			 * If we have pmu_enable/pmu_disable calls, install
+			 * transaction stubs that use that to try and batch
+			 * hardware accesses.
+			 */
+			pmu->start_txn  = perf_pmu_start_txn;
+			pmu->commit_txn = perf_pmu_commit_txn;
+			pmu->cancel_txn = perf_pmu_cancel_txn;
+		} else {
+			pmu->start_txn  = perf_pmu_nop_void;
+			pmu->commit_txn = perf_pmu_nop_int;
+			pmu->cancel_txn = perf_pmu_nop_void;
+		}
+	}
+
+	if (!pmu->pmu_enable) {
+		pmu->pmu_enable  = perf_pmu_nop_void;
+		pmu->pmu_disable = perf_pmu_nop_void;
+	}
+
 	list_add_rcu(&pmu->entry, &pmus);
 	ret = 0;
 unlock:
Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -565,8 +565,8 @@ struct pmu {
 
 	int				*pmu_disable_count;
 
-	void (*pmu_enable)		(struct pmu *pmu);
-	void (*pmu_disable)		(struct pmu *pmu);
+	void (*pmu_enable)		(struct pmu *pmu); /* optional */
+	void (*pmu_disable)		(struct pmu *pmu); /* optional */
 
 	/*
 	 * Should return -ENOENT when the @event doesn't match this PMU.
@@ -590,19 +590,19 @@ struct pmu {
 	 * Start the transaction, after this ->enable() doesn't need to
 	 * do schedulability tests.
 	 */
-	void (*start_txn)	(struct pmu *pmu);
+	void (*start_txn)	(struct pmu *pmu); /* optional */
 	/*
 	 * If ->start_txn() disabled the ->enable() schedulability test
 	 * then ->commit_txn() is required to perform one. On success
 	 * the transaction is closed. On error the transaction is kept
 	 * open until ->cancel_txn() is called.
 	 */
-	int  (*commit_txn)	(struct pmu *pmu);
+	int  (*commit_txn)	(struct pmu *pmu); /* optional */
 	/*
 	 * Will cancel the transaction, assumes ->disable() is called
 	 * for each successfull ->enable() during the transaction.
 	 */
-	void (*cancel_txn)	(struct pmu *pmu);
+	void (*cancel_txn)	(struct pmu *pmu); /* optional */
 };
 
 /**

[RFC][PATCH 12/13] perf: Shrink hw_perf_event

[Peter Zijlstra]

[-- Attachment #1: perf-swtimer-period_left.patch --]
[-- Type: text/plain, Size: 1701 bytes --]

Use hw_perf_event::period_left instead of hw_perf_event::remaning and
win back 8 bytes.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 include/linux/perf_event.h |    1 -
 kernel/perf_event.c        |   13 ++++++-------
 2 files changed, 6 insertions(+), 8 deletions(-)

Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -529,7 +529,6 @@ struct hw_perf_event {
 			int		last_cpu;
 		};
 		struct { /* software */
-			s64		remaining;
 			struct hrtimer	hrtimer;
 		};
 #ifdef CONFIG_HAVE_HW_BREAKPOINT
Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -4590,14 +4590,13 @@ static void perf_swevent_start_hrtimer(s
 	hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	hwc->hrtimer.function = perf_swevent_hrtimer;
 	if (hwc->sample_period) {
-		u64 period;
+		s64 period = local64_read(&hwc->period_left);
 
-		if (hwc->remaining) {
-			if (hwc->remaining < 0)
+		if (period) {
+			if (period < 0)
 				period = 10000;
-			else
-				period = hwc->remaining;
-			hwc->remaining = 0;
+
+			local64_set(&hwc->period_left, 0);
 		} else {
 			period = max_t(u64, 10000, hwc->sample_period);
 		}
@@ -4613,7 +4612,7 @@ static void perf_swevent_cancel_hrtimer(
 
 	if (hwc->sample_period) {
 		ktime_t remaining = hrtimer_get_remaining(&hwc->hrtimer);
-		hwc->remaining = ktime_to_ns(remaining);
+		local64_set(&hwc->period_left, ktime_to_ns(remaining));
 
 		hrtimer_cancel(&hwc->hrtimer);
 	}

[RFC][PATCH 13/13] perf: Rework the PMU methods

[Peter Zijlstra]

[-- Attachment #1: perf-change-ops.patch --]
[-- Type: text/plain, Size: 50112 bytes --]

Replace pmu::{enable,disable,start,stop,unthrottle} with
pmu::{add,del,start,stop}, all of which take a flags argument.

The new interface extends the capability to stop a counter while
keeping it scheduled on the PMU. We replace the throttled state with
the generic stopped state.

This also allows us to efficiently stop/start counters over certain
code paths (like IRQ handlers).

It also allows scheduling a counter without it starting, allowing for
a generic frozen state (useful for rotating stopped counters).

The stopped state is implemented in two different ways, depending on
how the architecture implemented the throttled state:

 1) We disable the counter:
    a) the pmu has per-counter enable bits, we flip that
    b) we program a NOP event, preserving the counter state

 2) We store the counter state and ignore all read/overflow events

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 arch/arm/kernel/perf_event.c              |   96 +++++++++++++-------
 arch/powerpc/kernel/perf_event.c          |  107 ++++++++++++++--------
 arch/powerpc/kernel/perf_event_fsl_emb.c  |  108 ++++++++++++++---------
 arch/sh/kernel/perf_event.c               |   75 ++++++++++------
 arch/sparc/kernel/perf_event.c            |  109 ++++++++++++++---------
 arch/x86/kernel/cpu/perf_event.c          |  106 +++++++++++++---------
 arch/x86/kernel/cpu/perf_event_intel.c    |    2 
 arch/x86/kernel/cpu/perf_event_intel_ds.c |    2 
 include/linux/ftrace_event.h              |    4 
 include/linux/perf_event.h                |   54 ++++++++---
 kernel/hw_breakpoint.c                    |   29 +++++-
 kernel/perf_event.c                       |  140 +++++++++++++++---------------
 kernel/trace/trace_event_perf.c           |    7 +
 13 files changed, 525 insertions(+), 314 deletions(-)

Index: linux-2.6/arch/x86/kernel/cpu/perf_event.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_event.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_event.c
@@ -583,7 +583,7 @@ static void x86_pmu_disable_all(void)
 	}
 }
 
-static void x86_pmu_pmu_disable(struct pmu *pmu)
+static void x86_pmu_disable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 
@@ -800,10 +800,10 @@ static inline int match_prev_assignment(
 		hwc->last_tag == cpuc->tags[i];
 }
 
-static int x86_pmu_start(struct perf_event *event);
-static void x86_pmu_stop(struct perf_event *event);
+static void x86_pmu_start(struct perf_event *event, int flags);
+static void x86_pmu_stop(struct perf_event *event, int flags);
 
-static void x86_pmu_pmu_enable(struct pmu *pmu)
+static void x86_pmu_enable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct perf_event *event;
@@ -839,7 +839,14 @@ static void x86_pmu_pmu_enable(struct pm
 			    match_prev_assignment(hwc, cpuc, i))
 				continue;
 
-			x86_pmu_stop(event);
+			/*
+			 * Ensure we don't accidentally enable a stopped
+			 * counter simply because we rescheduled.
+			 */
+			if (hwc->state & PERF_HES_STOPPED)
+				hwc->state |= PERF_HES_ARCH;
+
+			x86_pmu_stop(event, PERF_EF_UPDATE);
 		}
 
 		for (i = 0; i < cpuc->n_events; i++) {
@@ -851,7 +858,10 @@ static void x86_pmu_pmu_enable(struct pm
 			else if (i < n_running)
 				continue;
 
-			x86_pmu_start(event);
+			if (hwc->state & PERF_HES_ARCH)
+				continue;
+
+			x86_pmu_start(event, PERF_EF_RELOAD);
 		}
 		cpuc->n_added = 0;
 		perf_events_lapic_init();
@@ -952,15 +962,12 @@ static void x86_pmu_enable_event(struct 
 }
 
 /*
- * activate a single event
+ * Add a single event to the PMU.
  *
  * The event is added to the group of enabled events
  * but only if it can be scehduled with existing events.
- *
- * Called with PMU disabled. If successful and return value 1,
- * then guaranteed to call perf_enable() and hw_perf_enable()
  */
-static int x86_pmu_enable(struct perf_event *event)
+static int x86_pmu_add(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct hw_perf_event *hwc;
@@ -975,10 +982,14 @@ static int x86_pmu_enable(struct perf_ev
 	if (ret < 0)
 		goto out;
 
+	hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+	if (!(flags & PERF_EF_START))
+		hwc->state |= PERF_HES_ARCH;
+
 	/*
 	 * If group events scheduling transaction was started,
 	 * skip the schedulability test here, it will be peformed
-	 * 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;
@@ -1003,27 +1014,28 @@ out:
 	return ret;
 }
 
-static int x86_pmu_start(struct perf_event *event)
+static void x86_pmu_start(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	int idx = event->hw.idx;
 
-	if (idx == -1)
-		return -EAGAIN;
+	if (WARN_ON_ONCE(!(event->hw.state & PERF_HES_STOPPED)))
+		return;
+
+	if (WARN_ON_ONCE(idx == -1))
+		return;
+
+	if (flags & PERF_EF_RELOAD) {
+		WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE));
+		x86_perf_event_set_period(event);
+	}
+
+	event->hw.state = 0;
 
-	x86_perf_event_set_period(event);
 	cpuc->events[idx] = event;
 	__set_bit(idx, cpuc->active_mask);
 	x86_pmu.enable(event);
 	perf_event_update_userpage(event);
-
-	return 0;
-}
-
-static void x86_pmu_unthrottle(struct perf_event *event)
-{
-	int ret = x86_pmu_start(event);
-	WARN_ON_ONCE(ret);
 }
 
 void perf_event_print_debug(void)
@@ -1080,27 +1092,29 @@ void perf_event_print_debug(void)
 	local_irq_restore(flags);
 }
 
-static void x86_pmu_stop(struct perf_event *event)
+static void x86_pmu_stop(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct hw_perf_event *hwc = &event->hw;
-	int idx = hwc->idx;
-
-	if (!__test_and_clear_bit(idx, cpuc->active_mask))
-		return;
-
-	x86_pmu.disable(event);
 
-	/*
-	 * Drain the remaining delta count out of a event
-	 * that we are disabling:
-	 */
-	x86_perf_event_update(event);
+	if (__test_and_clear_bit(hwc->idx, cpuc->active_mask)) {
+		x86_pmu.disable(event);
+		cpuc->events[hwc->idx] = NULL;
+		WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+		hwc->state |= PERF_HES_STOPPED;
+	}
 
-	cpuc->events[idx] = NULL;
+	if ((flags & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
+		/*
+		 * Drain the remaining delta count out of a event
+		 * that we are disabling:
+		 */
+		x86_perf_event_update(event);
+		hwc->state |= PERF_HES_UPTODATE;
+	}
 }
 
-static void x86_pmu_disable(struct perf_event *event)
+static void x86_pmu_del(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	int i;
@@ -1113,7 +1127,7 @@ static void x86_pmu_disable(struct perf_
 	if (cpuc->group_flag & PERF_EVENT_TXN)
 		return;
 
-	x86_pmu_stop(event);
+	x86_pmu_stop(event, PERF_EF_UPDATE);
 
 	for (i = 0; i < cpuc->n_events; i++) {
 		if (event == cpuc->event_list[i]) {
@@ -1165,7 +1179,7 @@ static int x86_pmu_handle_irq(struct pt_
 			continue;
 
 		if (perf_event_overflow(event, 1, &data, regs))
-			x86_pmu_stop(event);
+			x86_pmu_stop(event, 0);
 	}
 
 	if (handled)
@@ -1572,15 +1586,17 @@ int x86_pmu_event_init(struct perf_event
 }
 
 static struct pmu pmu = {
-	.pmu_enable	= x86_pmu_pmu_enable,
-	.pmu_disable	= x86_pmu_pmu_disable,
+	.pmu_enable	= x86_pmu_enable,
+	.pmu_disable	= x86_pmu_disable,
+
 	.event_init	= x86_pmu_event_init,
-	.enable		= x86_pmu_enable,
-	.disable	= x86_pmu_disable,
+
+	.add		= x86_pmu_add,
+	.del		= x86_pmu_del,
 	.start		= x86_pmu_start,
 	.stop		= x86_pmu_stop,
 	.read		= x86_pmu_read,
-	.unthrottle	= x86_pmu_unthrottle,
+
 	.start_txn	= x86_pmu_start_txn,
 	.cancel_txn	= x86_pmu_cancel_txn,
 	.commit_txn	= x86_pmu_commit_txn,
Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -404,7 +404,7 @@ event_sched_out(struct perf_event *event
 		event->state = PERF_EVENT_STATE_OFF;
 	}
 	event->tstamp_stopped = ctx->time;
-	event->pmu->disable(event);
+	event->pmu->del(event, 0);
 	event->oncpu = -1;
 
 	if (!is_software_event(event))
@@ -631,7 +631,7 @@ event_sched_in(struct perf_event *event,
 	 */
 	smp_wmb();
 
-	if (event->pmu->enable(event)) {
+	if (event->pmu->add(event, PERF_EF_START)) {
 		event->state = PERF_EVENT_STATE_INACTIVE;
 		event->oncpu = -1;
 		return -EAGAIN;
@@ -1464,22 +1464,6 @@ do {					\
 	return div64_u64(dividend, divisor);
 }
 
-static void perf_event_stop(struct perf_event *event)
-{
-	if (!event->pmu->stop)
-		return event->pmu->disable(event);
-
-	return event->pmu->stop(event);
-}
-
-static int perf_event_start(struct perf_event *event)
-{
-	if (!event->pmu->start)
-		return event->pmu->enable(event);
-
-	return event->pmu->start(event);
-}
-
 static void perf_adjust_period(struct perf_event *event, u64 nsec, u64 count)
 {
 	struct hw_perf_event *hwc = &event->hw;
@@ -1499,9 +1483,9 @@ static void perf_adjust_period(struct pe
 	hwc->sample_period = sample_period;
 
 	if (local64_read(&hwc->period_left) > 8*sample_period) {
-		perf_event_stop(event);
+		event->pmu->stop(event, PERF_EF_UPDATE);
 		local64_set(&hwc->period_left, 0);
-		perf_event_start(event);
+		event->pmu->start(event, PERF_EF_RELOAD);
 	}
 }
 
@@ -1530,7 +1514,7 @@ static void perf_ctx_adjust_freq(struct 
 		 */
 		if (interrupts == MAX_INTERRUPTS) {
 			perf_log_throttle(event, 1);
-			event->pmu->unthrottle(event);
+			event->pmu->start(event, 0);
 		}
 
 		if (!event->attr.freq || !event->attr.sample_freq)
@@ -2276,6 +2260,9 @@ int perf_event_task_disable(void)
 
 static int perf_event_index(struct perf_event *event)
 {
+	if (event->hw.state & PERF_HES_STOPPED)
+		return 0;
+
 	if (event->state != PERF_EVENT_STATE_ACTIVE)
 		return 0;
 
@@ -3894,8 +3881,6 @@ static int __perf_event_overflow(struct 
 	struct hw_perf_event *hwc = &event->hw;
 	int ret = 0;
 
-	throttle = (throttle && event->pmu->unthrottle != NULL);
-
 	if (!throttle) {
 		hwc->interrupts++;
 	} else {
@@ -4020,7 +4005,7 @@ static void perf_swevent_overflow(struct
 	}
 }
 
-static void perf_swevent_add(struct perf_event *event, u64 nr,
+static void perf_swevent_event(struct perf_event *event, u64 nr,
 			       int nmi, struct perf_sample_data *data,
 			       struct pt_regs *regs)
 {
@@ -4046,6 +4031,9 @@ static void perf_swevent_add(struct perf
 static int perf_exclude_event(struct perf_event *event,
 			      struct pt_regs *regs)
 {
+	if (event->hw.state & PERF_HES_STOPPED)
+		return 0;
+
 	if (regs) {
 		if (event->attr.exclude_user && user_mode(regs))
 			return 1;
@@ -4145,7 +4133,7 @@ static void do_perf_sw_event(enum perf_t
 
 	hlist_for_each_entry_rcu(event, node, head, hlist_entry) {
 		if (perf_swevent_match(event, type, event_id, data, regs))
-			perf_swevent_add(event, nr, nmi, data, regs);
+			perf_swevent_event(event, nr, nmi, data, regs);
 	}
 end:
 	rcu_read_unlock();
@@ -4205,7 +4193,7 @@ static void perf_swevent_read(struct per
 {
 }
 
-static int perf_swevent_enable(struct perf_event *event)
+static int perf_swevent_add(struct perf_event *event, int flags)
 {
 	struct hw_perf_event *hwc = &event->hw;
 	struct perf_cpu_context *cpuctx;
@@ -4218,6 +4206,8 @@ static int perf_swevent_enable(struct pe
 		perf_swevent_set_period(event);
 	}
 
+	hwc->state = !(flags & PERF_EF_START);
+
 	head = find_swevent_head(cpuctx, event);
 	if (WARN_ON_ONCE(!head))
 		return -EINVAL;
@@ -4227,18 +4217,19 @@ static int perf_swevent_enable(struct pe
 	return 0;
 }
 
-static void perf_swevent_disable(struct perf_event *event)
+static void perf_swevent_del(struct perf_event *event, int flags)
 {
 	hlist_del_rcu(&event->hlist_entry);
 }
 
-static void perf_swevent_void(struct perf_event *event)
+static void perf_swevent_start(struct perf_event *event, int flags)
 {
+	event->hw.state = 0;
 }
 
-static int perf_swevent_int(struct perf_event *event)
+static void perf_swevent_stop(struct perf_event *event, int flags)
 {
-	return 0;
+	event->hw.state = PERF_HES_STOPPED;
 }
 
 /* Deref the hlist from the update side */
@@ -4394,12 +4385,11 @@ static int perf_swevent_init(struct perf
 
 static struct pmu perf_swevent = {
 	.event_init	= perf_swevent_init,
-	.enable		= perf_swevent_enable,
-	.disable	= perf_swevent_disable,
-	.start		= perf_swevent_int,
-	.stop		= perf_swevent_void,
+	.add		= perf_swevent_add,
+	.del		= perf_swevent_del,
+	.start		= perf_swevent_start,
+	.stop		= perf_swevent_stop,
 	.read		= perf_swevent_read,
-	.unthrottle	= perf_swevent_void, /* hwc->interrupts already reset */
 };
 
 #ifdef CONFIG_EVENT_TRACING
@@ -4447,7 +4437,7 @@ void perf_tp_event(u64 addr, u64 count, 
 
 	hlist_for_each_entry_rcu(event, node, head, hlist_entry) {
 		if (perf_tp_event_match(event, &data, regs))
-			perf_swevent_add(event, count, 1, &data, regs);
+			perf_swevent_event(event, count, 1, &data, regs);
 	}
 
 	perf_swevent_put_recursion_context(rctx);
@@ -4486,12 +4476,11 @@ static int perf_tp_event_init(struct per
 
 static struct pmu perf_tracepoint = {
 	.event_init	= perf_tp_event_init,
-	.enable		= perf_trace_enable,
-	.disable	= perf_trace_disable,
-	.start		= perf_swevent_int,
-	.stop		= perf_swevent_void,
+	.add		= perf_trace_add,
+	.del		= perf_trace_del,
+	.start		= perf_swevent_start,
+	.stop		= perf_swevent_stop,
 	.read		= perf_swevent_read,
-	.unthrottle	= perf_swevent_void,
 };
 
 static inline void perf_tp_register(void)
@@ -4547,8 +4536,8 @@ void perf_bp_event(struct perf_event *bp
 
 	perf_sample_data_init(&sample, bp->attr.bp_addr);
 
-	if (!perf_exclude_event(bp, regs))
-		perf_swevent_add(bp, 1, 1, &sample, regs);
+	if (!bp->hw.state && !perf_exclude_event(bp, regs))
+		perf_swevent_event(bp, 1, 1, &sample, regs);
 }
 #endif
 
@@ -4624,32 +4613,39 @@ static void perf_swevent_cancel_hrtimer(
 
 static void cpu_clock_event_update(struct perf_event *event)
 {
-	int cpu = raw_smp_processor_id();
 	s64 prev;
 	u64 now;
 
-	now = cpu_clock(cpu);
+	now = local_clock();
 	prev = local64_xchg(&event->hw.prev_count, now);
 	local64_add(now - prev, &event->count);
 }
 
-static int cpu_clock_event_enable(struct perf_event *event)
+static void cpu_clock_event_start(struct perf_event *event, int flags)
 {
-	struct hw_perf_event *hwc = &event->hw;
-	int cpu = raw_smp_processor_id();
-
-	local64_set(&hwc->prev_count, cpu_clock(cpu));
+	local64_set(&event->hw.prev_count, local_clock());
 	perf_swevent_start_hrtimer(event);
-
-	return 0;
 }
 
-static void cpu_clock_event_disable(struct perf_event *event)
+static void cpu_clock_event_stop(struct perf_event *event, int flags)
 {
 	perf_swevent_cancel_hrtimer(event);
 	cpu_clock_event_update(event);
 }
 
+static int cpu_clock_event_add(struct perf_event *event, int flags)
+{
+	if (flags & PERF_EF_START)
+		cpu_clock_event_start(event, flags);
+
+	return 0;
+}
+
+static void cpu_clock_event_del(struct perf_event *event, int flags)
+{
+	cpu_clock_event_stop(event, flags);
+}
+
 static void cpu_clock_event_read(struct perf_event *event)
 {
 	cpu_clock_event_update(event);
@@ -4668,8 +4664,10 @@ static int cpu_clock_event_init(struct p
 
 static struct pmu perf_cpu_clock = {
 	.event_init	= cpu_clock_event_init,
-	.enable		= cpu_clock_event_enable,
-	.disable	= cpu_clock_event_disable,
+	.add		= cpu_clock_event_add,
+	.del		= cpu_clock_event_del,
+	.start		= cpu_clock_event_start,
+	.stop		= cpu_clock_event_stop,
 	.read		= cpu_clock_event_read,
 };
 
@@ -4687,25 +4685,29 @@ static void task_clock_event_update(stru
 	local64_add(delta, &event->count);
 }
 
-static int task_clock_event_enable(struct perf_event *event)
+static void task_clock_event_start(struct perf_event *event, int flags)
 {
-	struct hw_perf_event *hwc = &event->hw;
-	u64 now;
-
-	now = event->ctx->time;
-
-	local64_set(&hwc->prev_count, now);
-
+	local64_set(&event->hw.prev_count, event->ctx->time);
 	perf_swevent_start_hrtimer(event);
-
-	return 0;
 }
 
-static void task_clock_event_disable(struct perf_event *event)
+static void task_clock_event_stop(struct perf_event *event, int flags)
 {
 	perf_swevent_cancel_hrtimer(event);
 	task_clock_event_update(event, event->ctx->time);
+}
+
+static int task_clock_event_add(struct perf_event *event, int flags)
+{
+	if (flags & PERF_EF_START)
+		task_clock_event_start(event, flags);
 
+	return 0;
+}
+
+static void task_clock_event_del(struct perf_event *event, int flags)
+{
+	task_clock_event_stop(event, PERF_EF_UPDATE);
 }
 
 static void task_clock_event_read(struct perf_event *event)
@@ -4737,8 +4739,10 @@ static int task_clock_event_init(struct 
 
 static struct pmu perf_task_clock = {
 	.event_init	= task_clock_event_init,
-	.enable		= task_clock_event_enable,
-	.disable	= task_clock_event_disable,
+	.add		= task_clock_event_add,
+	.del		= task_clock_event_del,
+	.start		= task_clock_event_start,
+	.stop		= task_clock_event_stop,
 	.read		= task_clock_event_read,
 };
 
Index: linux-2.6/kernel/trace/trace_event_perf.c
===================================================================
--- linux-2.6.orig/kernel/trace/trace_event_perf.c
+++ linux-2.6/kernel/trace/trace_event_perf.c
@@ -99,7 +99,7 @@ int perf_trace_init(struct perf_event *p
 	return ret;
 }
 
-int perf_trace_enable(struct perf_event *p_event)
+int perf_trace_add(struct perf_event *p_event, int flags)
 {
 	struct ftrace_event_call *tp_event = p_event->tp_event;
 	struct hlist_head *list;
@@ -108,13 +108,16 @@ int perf_trace_enable(struct perf_event 
 	if (WARN_ON_ONCE(!list))
 		return -EINVAL;
 
+	if (!(flags & PERF_EF_START))
+		p_event->hw.state = PERF_HES_STOPPED;
+
 	list = this_cpu_ptr(list);
 	hlist_add_head_rcu(&p_event->hlist_entry, list);
 
 	return 0;
 }
 
-void perf_trace_disable(struct perf_event *p_event)
+void perf_trace_del(struct perf_event *p_event, int flags)
 {
 	hlist_del_rcu(&p_event->hlist_entry);
 }
Index: linux-2.6/arch/sh/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sh/kernel/perf_event.c
+++ linux-2.6/arch/sh/kernel/perf_event.c
@@ -206,46 +206,71 @@ again:
 	local64_add(delta, &event->count);
 }
 
-static void sh_pmu_disable(struct perf_event *event)
+static void sh_pmu_stop(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct hw_perf_event *hwc = &event->hw;
 	int idx = hwc->idx;
 
-	clear_bit(idx, cpuc->active_mask);
-	sh_pmu->disable(hwc, idx);
+	if (!(event->hw.state & PERF_HES_STOPPED)) {
+		sh_pmu->disable(hwc, idx);
+		cpuc->events[idx] = NULL;
+		event->hw.state |= PERF_HES_STOPPED;
+	}
 
-	barrier();
+	if ((flags & PERF_EF_UPDATE) && !(event->hw.state & PERF_HES_UPTODATE)) {
+		sh_perf_event_update(event, &event->hw, idx);
+		event->hw.state |= PERF_HES_UPTODATE;
+	}
+}
 
-	sh_perf_event_update(event, &event->hw, idx);
+static void sh_pmu_start(struct perf_event *event, int flags)
+{
+	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
 
-	cpuc->events[idx] = NULL;
-	clear_bit(idx, cpuc->used_mask);
+	if (WARN_ON_ONCE(idx == -1))
+		return;
+
+	if (flags & PERF_EF_RELOAD)
+		WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE));
+
+	cpuc->events[idx] = event;
+	event->hw.state = 0;
+	sh_pmu->enable(hwc, idx);
+}
+
+static void sh_pmu_del(struct perf_event *event, int flags)
+{
+	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+
+	sh_pmu_stop(event, PERF_EF_UPDATE);
+	__clear_bit(event->hw.idx, cpuc->used_mask);
 
 	perf_event_update_userpage(event);
 }
 
-static int sh_pmu_enable(struct perf_event *event)
+static int sh_pmu_add(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct hw_perf_event *hwc = &event->hw;
 	int idx = hwc->idx;
 
-	if (test_and_set_bit(idx, cpuc->used_mask)) {
+	if (__test_and_set_bit(idx, cpuc->used_mask)) {
 		idx = find_first_zero_bit(cpuc->used_mask, sh_pmu->num_events);
 		if (idx == sh_pmu->num_events)
 			return -EAGAIN;
 
-		set_bit(idx, cpuc->used_mask);
+		__set_bit(idx, cpuc->used_mask);
 		hwc->idx = idx;
 	}
 
 	sh_pmu->disable(hwc, idx);
 
-	cpuc->events[idx] = event;
-	set_bit(idx, cpuc->active_mask);
-
-	sh_pmu->enable(hwc, idx);
+	event->hw.state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+	if (flags & PERF_EF_START)
+		sh_pmu_start(event, PERF_EF_RELOAD);
 
 	perf_event_update_userpage(event);
 
@@ -280,7 +305,7 @@ static int sh_pmu_event_init(struct perf
 	return err;
 }
 
-static void sh_pmu_pmu_enable(struct pmu *pmu)
+static void sh_pmu_enable(struct pmu *pmu)
 {
 	if (!sh_pmu_initialized())
 		return;
@@ -288,7 +313,7 @@ static void sh_pmu_pmu_enable(struct pmu
 	sh_pmu->enable_all();
 }
 
-static void sh_pmu_pmu_disable(struct pmu *pmu)
+static void sh_pmu_disable(struct pmu *pmu)
 {
 	if (!sh_pmu_initialized())
 		return;
@@ -297,11 +322,13 @@ static void sh_pmu_pmu_disable(struct pm
 }
 
 static struct pmu pmu = {
-	.pmu_enable	= sh_pmu_pmu_enable,
-	.pmu_disable	= sh_pmu_pmu_disable,
+	.pmu_enable	= sh_pmu_enable,
+	.pmu_disable	= sh_pmu_disable,
 	.event_init	= sh_pmu_event_init,
-	.enable		= sh_pmu_enable,
-	.disable	= sh_pmu_disable,
+	.add		= sh_pmu_add,
+	.del		= sh_pmu_del,
+	.start		= sh_pmu_start,
+	.stop		= sh_pmu_stop,
 	.read		= sh_pmu_read,
 };
 
@@ -329,15 +356,15 @@ sh_pmu_notifier(struct notifier_block *s
 	return NOTIFY_OK;
 }
 
-int __cpuinit register_sh_pmu(struct sh_pmu *pmu)
+int __cpuinit register_sh_pmu(struct sh_pmu *_pmu)
 {
 	if (sh_pmu)
 		return -EBUSY;
-	sh_pmu = pmu;
+	sh_pmu = _pmu;
 
-	pr_info("Performance Events: %s support registered\n", pmu->name);
+	pr_info("Performance Events: %s support registered\n", _pmu->name);
 
-	WARN_ON(pmu->num_events > MAX_HWEVENTS);
+	WARN_ON(_pmu->num_events > MAX_HWEVENTS);
 
 	perf_pmu_register(&pmu);
 	perf_cpu_notifier(sh_pmu_notifier);
Index: linux-2.6/arch/arm/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/arm/kernel/perf_event.c
+++ linux-2.6/arch/arm/kernel/perf_event.c
@@ -221,46 +221,56 @@ again:
 }
 
 static void
-armpmu_disable(struct perf_event *event)
+armpmu_read(struct perf_event *event)
 {
-	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct hw_perf_event *hwc = &event->hw;
-	int idx = hwc->idx;
-
-	WARN_ON(idx < 0);
-
-	clear_bit(idx, cpuc->active_mask);
-	armpmu->disable(hwc, idx);
-
-	barrier();
 
-	armpmu_event_update(event, hwc, idx);
-	cpuc->events[idx] = NULL;
-	clear_bit(idx, cpuc->used_mask);
+	/* Don't read disabled counters! */
+	if (hwc->idx < 0)
+		return;
 
-	perf_event_update_userpage(event);
+	armpmu_event_update(event, hwc, hwc->idx);
 }
 
 static void
-armpmu_read(struct perf_event *event)
+armpmu_stop(struct perf_event *event, int flags)
 {
 	struct hw_perf_event *hwc = &event->hw;
 
-	/* Don't read disabled counters! */
-	if (hwc->idx < 0)
+	if (!armpmu)
 		return;
 
-	armpmu_event_update(event, hwc, hwc->idx);
+	/*
+	 * ARM pmu always has to update the counter, so ignore
+	 * PERF_EF_UPDATE, see comments in armpmu_start().
+	 */
+	if (!(hwc->state & PERF_HES_STOPPED)) {
+		armpmu->disable(hwc, hwc->idx);
+		barrier(); /* why? */
+		armpmu_event_update(event, hwc, hwc->idx);
+		hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	}
 }
 
 static void
-armpmu_unthrottle(struct perf_event *event)
+armpmu_start(struct perf_event *event, int flags)
 {
 	struct hw_perf_event *hwc = &event->hw;
 
+	if (!armpmu)
+		return;
+
+	/*
+	 * ARM pmu always has to reprogram the period, so ignore
+	 * PERF_EF_RELOAD, see the comment below.
+	 */
+	if (flags & PERF_EF_RELOAD)
+		WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+
+	hwc->state = 0;
 	/*
 	 * Set the period again. Some counters can't be stopped, so when we
-	 * were throttled we simply disabled the IRQ source and the counter
+	 * were stopped we simply disabled the IRQ source and the counter
 	 * may have been left counting. If we don't do this step then we may
 	 * get an interrupt too soon or *way* too late if the overflow has
 	 * happened since disabling.
@@ -269,8 +279,25 @@ armpmu_unthrottle(struct perf_event *eve
 	armpmu->enable(hwc, hwc->idx);
 }
 
+static void
+armpmu_del(struct perf_event *event, int flags)
+{
+	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	WARN_ON(idx < 0);
+
+	clear_bit(idx, cpuc->active_mask);
+	armpmu_stop(event, PERF_EF_UPDATE);
+	cpuc->events[idx] = NULL;
+	clear_bit(idx, cpuc->used_mask);
+
+	perf_event_update_userpage(event);
+}
+
 static int
-armpmu_enable(struct perf_event *event)
+armpmu_add(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct hw_perf_event *hwc = &event->hw;
@@ -293,11 +320,9 @@ armpmu_enable(struct perf_event *event)
 	cpuc->events[idx] = event;
 	set_bit(idx, cpuc->active_mask);
 
-	/* Set the period for the event. */
-	armpmu_event_set_period(event, hwc, idx);
-
-	/* Enable the event. */
-	armpmu->enable(hwc, idx);
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	if (flags & PERF_EF_START)
+		armpmu_start(event, PERF_EF_RELOAD);
 
 	/* Propagate our changes to the userspace mapping. */
 	perf_event_update_userpage(event);
@@ -531,7 +556,7 @@ static int armpmu_event_init(struct perf
 	return err;
 }
 
-static void armpmu_pmu_enable(struct pmu *pmu)
+static void armpmu_enable(struct pmu *pmu)
 {
 	/* Enable all of the perf events on hardware. */
 	int idx;
@@ -552,20 +577,21 @@ static void armpmu_pmu_enable(struct pmu
 	armpmu->start();
 }
 
-static void armpmu_pmu_disable(struct pmu *pmu)
+static void armpmu_disable(struct pmu *pmu)
 {
 	if (armpmu)
 		armpmu->stop();
 }
 
 static struct pmu pmu = {
-	.pmu_enable = armpmu_pmu_enable,
-	.pmu_disable= armpmu_pmu_disable,
-	.event_init = armpmu_event_init,
-	.enable	    = armpmu_enable,
-	.disable    = armpmu_disable,
-	.unthrottle = armpmu_unthrottle,
-	.read	    = armpmu_read,
+	.pmu_enable	= armpmu_enable,
+	.pmu_disable	= armpmu_disable,
+	.event_init	= armpmu_event_init,
+	.add		= armpmu_add,
+	.del		= armpmu_del,
+	.start		= armpmu_start,
+	.stop		= armpmu_stop,
+	.read		= armpmu_read,
 };
 
 /*
Index: linux-2.6/arch/powerpc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event.c
+++ linux-2.6/arch/powerpc/kernel/perf_event.c
@@ -402,6 +402,9 @@ static void power_pmu_read(struct perf_e
 {
 	s64 val, delta, prev;
 
+	if (event->hw.state & PERF_HES_STOPPED)
+		return;
+
 	if (!event->hw.idx)
 		return;
 	/*
@@ -517,7 +520,7 @@ static void write_mmcr0(struct cpu_hw_ev
  * Disable all events to prevent PMU interrupts and to allow
  * events to be added or removed.
  */
-static void powerpc_pmu_pmu_disable(struct pmu *pmu)
+static void power_pmu_disable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw;
 	unsigned long flags;
@@ -565,7 +568,7 @@ static void powerpc_pmu_pmu_disable(stru
  * If we were previously disabled and events were added, then
  * put the new config on the PMU.
  */
-static void powerpc_pmu_pmu_enable(struct pmu *pmu)
+static void power_pmu_enable(struct pmu *pmu)
 {
 	struct perf_event *event;
 	struct cpu_hw_events *cpuhw;
@@ -672,6 +675,8 @@ static void powerpc_pmu_pmu_enable(struc
 		}
 		local64_set(&event->hw.prev_count, val);
 		event->hw.idx = idx;
+		if (event->hw.state & PERF_HES_STOPPED)
+			val = 0;
 		write_pmc(idx, val);
 		perf_event_update_userpage(event);
 	}
@@ -727,7 +732,7 @@ static int collect_events(struct perf_ev
  * re-enable the PMU in order to get hw_perf_enable to do the
  * actual work of reconfiguring the PMU.
  */
-static int power_pmu_enable(struct perf_event *event)
+static int power_pmu_add(struct perf_event *event, int ef_flags)
 {
 	struct cpu_hw_events *cpuhw;
 	unsigned long flags;
@@ -749,6 +754,9 @@ static int power_pmu_enable(struct perf_
 	cpuhw->events[n0] = event->hw.config;
 	cpuhw->flags[n0] = event->hw.event_base;
 
+	if (!(ef_flags & PERF_EF_START))
+		event->hw.state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+
 	/*
 	 * If group events scheduling transaction was started,
 	 * skip the schedulability test here, it will be peformed
@@ -777,7 +785,7 @@ nocheck:
 /*
  * Remove a event from the PMU.
  */
-static void power_pmu_disable(struct perf_event *event)
+static void power_pmu_del(struct perf_event *event, int ef_flags)
 {
 	struct cpu_hw_events *cpuhw;
 	long i;
@@ -823,27 +831,53 @@ static void power_pmu_disable(struct per
 }
 
 /*
- * Re-enable interrupts on a event after they were throttled
- * because they were coming too fast.
+ * POWER-PMU does not support disabling individual counters, hence
+ * program their cycle counter to their max value and ignore the interrupts.
  */
-static void power_pmu_unthrottle(struct perf_event *event)
+
+static void power_pmu_start(struct perf_event *event, int ef_flags)
 {
-	s64 val, left;
 	unsigned long flags;
+	s64 left;
 
 	if (!event->hw.idx || !event->hw.sample_period)
 		return;
+
+	if (!(event->hw.state & PERF_HES_STOPPED))
+		return;
+
+	if (ef_flags & PERF_EF_RELOAD)
+		WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE));
+
+	local_irq_save(flags);
+	perf_pmu_disable(event->pmu);
+
+	event->hw.state = 0;
+	left = local64_read(&event->hw.period_left);
+	write_pmc(event->hw.idx, left);
+
+	perf_event_update_userpage(event);
+	perf_pmu_enable(event->pmu);
+	local_irq_restore(flags);
+}
+
+static void power_pmu_stop(struct perf_event *event, int ef_flags)
+{
+	unsigned long flags;
+
+	if (!event->hw.idx || !event->hw.sample_period)
+		return;
+
+	if (event->hw.state & PERF_HES_STOPPED)
+		return;
+
 	local_irq_save(flags);
 	perf_pmu_disable(event->pmu);
+
 	power_pmu_read(event);
-	left = event->hw.sample_period;
-	event->hw.last_period = left;
-	val = 0;
-	if (left < 0x80000000L)
-		val = 0x80000000L - left;
-	write_pmc(event->hw.idx, val);
-	local64_set(&event->hw.prev_count, val);
-	local64_set(&event->hw.period_left, left);
+	event->hw.state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	write_pmc(event->hw.idx, 0);
+
 	perf_event_update_userpage(event);
 	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
@@ -1128,13 +1162,14 @@ static int power_pmu_event_init(struct p
 }
 
 struct pmu power_pmu = {
-	.pmu_enable	= power_pmu_pmu_enable,
-	.pmu_disable	= power_pmu_pmu_disable,
-	.event_init	= pmwer_pmu_event_init,
-	.enable		= power_pmu_enable,
-	.disable	= power_pmu_disable,
+	.pmu_enable	= power_pmu_enable,
+	.pmu_disable	= power_pmu_disable,
+	.event_init	= power_pmu_event_init,
+	.add		= power_pmu_add,
+	.del		= power_pmu_del,
+	.start		= power_pmu_start,
+	.stop		= power_pmu_stop,
 	.read		= power_pmu_read,
-	.unthrottle	= power_pmu_unthrottle,
 	.start_txn	= power_pmu_start_txn,
 	.cancel_txn	= power_pmu_cancel_txn,
 	.commit_txn	= power_pmu_commit_txn,
@@ -1152,6 +1187,11 @@ static void record_and_restart(struct pe
 	s64 prev, delta, left;
 	int record = 0;
 
+	if (event->hw.state & PERF_HES_STOPPED) {
+		write_pmc(event->hw.idx, 0);
+		return;
+	}
+
 	/* we don't have to worry about interrupts here */
 	prev = local64_read(&event->hw.prev_count);
 	delta = (val - prev) & 0xfffffffful;
@@ -1174,6 +1214,11 @@ static void record_and_restart(struct pe
 			val = 0x80000000LL - left;
 	}
 
+	write_pmc(event->hw.idx, val);
+	local64_set(&event->hw.prev_count, val);
+	local64_set(&event->hw.period_left, left);
+	perf_event_update_userpage(event);
+
 	/*
 	 * Finally record data if requested.
 	 */
@@ -1186,23 +1231,9 @@ static void record_and_restart(struct pe
 		if (event->attr.sample_type & PERF_SAMPLE_ADDR)
 			perf_get_data_addr(regs, &data.addr);
 
-		if (perf_event_overflow(event, nmi, &data, regs)) {
-			/*
-			 * Interrupts are coming too fast - throttle them
-			 * by setting the event to 0, so it will be
-			 * at least 2^30 cycles until the next interrupt
-			 * (assuming each event counts at most 2 counts
-			 * per cycle).
-			 */
-			val = 0;
-			left = ~0ULL >> 1;
-		}
+		if (perf_event_overflow(event, nmi, &data, regs))
+			power_pmu_stop(event, 0);
 	}
-
-	write_pmc(event->hw.idx, val);
-	local64_set(&event->hw.prev_count, val);
-	local64_set(&event->hw.period_left, left);
-	perf_event_update_userpage(event);
 }
 
 /*
Index: linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/perf_event_fsl_emb.c
+++ linux-2.6/arch/powerpc/kernel/perf_event_fsl_emb.c
@@ -156,6 +156,9 @@ static void fsl_emb_pmu_read(struct perf
 {
 	s64 val, delta, prev;
 
+	if (event->hw.state & PERF_HES_STOPPED)
+		return;
+
 	/*
 	 * Performance monitor interrupts come even when interrupts
 	 * are soft-disabled, as long as interrupts are hard-enabled.
@@ -177,7 +180,7 @@ static void fsl_emb_pmu_read(struct perf
  * Disable all events to prevent PMU interrupts and to allow
  * events to be added or removed.
  */
-static void fsl_emb_pmu_pmu_disable(struct pmu *pmu)
+static void fsl_emb_pmu_disable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw;
 	unsigned long flags;
@@ -216,7 +219,7 @@ static void fsl_emb_pmu_pmu_disable(stru
  * If we were previously disabled and events were added, then
  * put the new config on the PMU.
  */
-static void fsl_emb_pmu_pmu_enable(struct pmu *pmu)
+static void fsl_emb_pmu_enable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuhw;
 	unsigned long flags;
@@ -263,7 +266,7 @@ static int collect_events(struct perf_ev
 }
 
 /* context locked on entry */
-static int fsl_emb_pmu_enable(struct perf_event *event)
+static int fsl_emb_pmu_add(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuhw;
 	int ret = -EAGAIN;
@@ -302,6 +305,12 @@ static int fsl_emb_pmu_enable(struct per
 			val = 0x80000000L - left;
 	}
 	local64_set(&event->hw.prev_count, val);
+
+	if (!(flags & PERF_EF_START)) {
+		event->hw.state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+		val = 0;
+	}
+
 	write_pmc(i, val);
 	perf_event_update_userpage(event);
 
@@ -316,7 +325,7 @@ static int fsl_emb_pmu_enable(struct per
 }
 
 /* context locked on entry */
-static void fsl_emb_pmu_disable(struct perf_event *event)
+static void fsl_emb_pmu_del(struct perf_event *event, int flags)
 {
 	struct cpu_hw_events *cpuhw;
 	int i = event->hw.idx;
@@ -353,30 +362,49 @@ static void fsl_emb_pmu_disable(struct p
 	put_cpu_var(cpu_hw_events);
 }
 
-/*
- * Re-enable interrupts on a event after they were throttled
- * because they were coming too fast.
- *
- * Context is locked on entry, but perf is not disabled.
- */
-static void fsl_emb_pmu_unthrottle(struct perf_event *event)
+static void fsl_emb_pmu_start(struct perf_event *event, int ef_flags)
+{
+	unsigned long flags;
+	s64 left;
+
+	if (event->hw.idx < 0 || !event->hw.sample_period)
+		return;
+
+	if (!(event->hw.state & PERF_HES_STOPPED))
+		return;
+
+	if (ef_flags & PERF_EF_RELOAD)
+		WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE));
+
+	local_irq_save(flags);
+	perf_pmu_disable(event->pmu);
+
+	event->hw.state = 0;
+	left = local64_read(&event->hw.period_left);
+	write_pmc(event->hw.idx, left);
+
+	perf_event_update_userpage(event);
+	perf_pmu_enable(event->pmu);
+	local_irq_restore(flags);
+}
+
+static void fsl_emb_pmu_stop(struct perf_event *event, int ef_flags)
 {
-	s64 val, left;
 	unsigned long flags;
 
 	if (event->hw.idx < 0 || !event->hw.sample_period)
 		return;
+
+	if (event->hw.state & PERF_HES_STOPPED)
+		return;
+
 	local_irq_save(flags);
 	perf_pmu_disable(event->pmu);
+
 	fsl_emb_pmu_read(event);
-	left = event->hw.sample_period;
-	event->hw.last_period = left;
-	val = 0;
-	if (left < 0x80000000L)
-		val = 0x80000000L - left;
-	write_pmc(event->hw.idx, val);
-	local64_set(&event->hw.prev_count, val);
-	local64_set(&event->hw.period_left, left);
+	event->hw.state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	write_pmc(event->hw.idx, 0);
+
 	perf_event_update_userpage(event);
 	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
@@ -524,13 +552,14 @@ static int fsl_emb_pmu_event_init(struct
 }
 
 static struct pmu fsl_emb_pmu = {
-	.pmu_enable	= fsl_emb_pmu_pmu_enable,
-	.pmu_disable	= fsl_emb_pmu_pmu_disable,
+	.pmu_enable	= fsl_emb_pmu_enable,
+	.pmu_disable	= fsl_emb_pmu_disable,
 	.event_init	= fsl_emb_pmu_event_init,
-	.enable		= fsl_emb_pmu_enable,
-	.disable	= fsl_emb_pmu_disable,
+	.add		= fsl_emb_pmu_add,
+	.del		= fsl_emb_pmu_del,
+	.start		= fsl_emb_pmu_start,
+	.stop		= fsl_emb_pmu_stop,
 	.read		= fsl_emb_pmu_read,
-	.unthrottle	= fsl_emb_pmu_unthrottle,
 };
 
 /*
@@ -545,6 +574,11 @@ static void record_and_restart(struct pe
 	s64 prev, delta, left;
 	int record = 0;
 
+	if (event->hw.state & PERF_HES_STOPPED) {
+		write_pmc(event->hw.idx, 0);
+		return;
+	}
+
 	/* we don't have to worry about interrupts here */
 	prev = local64_read(&event->hw.prev_count);
 	delta = (val - prev) & 0xfffffffful;
@@ -567,6 +601,11 @@ static void record_and_restart(struct pe
 			val = 0x80000000LL - left;
 	}
 
+	write_pmc(event->hw.idx, val);
+	local64_set(&event->hw.prev_count, val);
+	local64_set(&event->hw.period_left, left);
+	perf_event_update_userpage(event);
+
 	/*
 	 * Finally record data if requested.
 	 */
@@ -574,24 +613,11 @@ static void record_and_restart(struct pe
 		struct perf_sample_data data;
 
 		perf_sample_data_init(&data, 0);
+		data.period = event->hw.last_period;
 
-		if (perf_event_overflow(event, nmi, &data, regs)) {
-			/*
-			 * Interrupts are coming too fast - throttle them
-			 * by setting the event to 0, so it will be
-			 * at least 2^30 cycles until the next interrupt
-			 * (assuming each event counts at most 2 counts
-			 * per cycle).
-			 */
-			val = 0;
-			left = ~0ULL >> 1;
-		}
+		if (perf_event_overflow(event, nmi, &data, regs))
+			fsl_emb_pmu_stop(event, 0);
 	}
-
-	write_pmc(event->hw.idx, val);
-	local64_set(&event->hw.prev_count, val);
-	local64_set(&event->hw.period_left, left);
-	perf_event_update_userpage(event);
 }
 
 static void perf_event_interrupt(struct pt_regs *regs)
Index: linux-2.6/arch/sparc/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sparc/kernel/perf_event.c
+++ linux-2.6/arch/sparc/kernel/perf_event.c
@@ -658,13 +658,16 @@ static u64 maybe_change_configuration(st
 
 		enc = perf_event_get_enc(cpuc->events[i]);
 		pcr &= ~mask_for_index(idx);
-		pcr |= event_encoding(enc, idx);
+		if (hwc->state & PERF_HES_STOPPED)
+			pcr |= nop_for_index(idx);
+		else
+			pcr |= event_encoding(enc, idx);
 	}
 out:
 	return pcr;
 }
 
-static void sparc_pmu_pmu_enable(struct pmu *pmu)
+static void sparc_pmu_enable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	u64 pcr;
@@ -691,7 +694,7 @@ static void sparc_pmu_pmu_enable(struct 
 	pcr_ops->write(cpuc->pcr);
 }
 
-static void sparc_pmu_pmu_disable(struct pmu *pmu)
+static void sparc_pmu_disable(struct pmu *pmu)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	u64 val;
@@ -710,10 +713,53 @@ static void sparc_pmu_pmu_disable(struct
 	pcr_ops->write(cpuc->pcr);
 }
 
-static void sparc_pmu_disable(struct perf_event *event)
+static int active_event_index(struct cpu_hw_events *cpuc,
+			      struct perf_event *event)
+{
+	int i;
+
+	for (i = 0; i < cpuc->n_events; i++) {
+		if (cpuc->event[i] == event)
+			break;
+	}
+	BUG_ON(i == cpuc->n_events);
+	return cpuc->current_idx[i];
+}
+
+static void sparc_pmu_start(struct perf_event *event, int flags)
+{
+	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+	int idx = active_event_index(cpuc, event);
+
+	if (flags & PERF_EF_RELOAD) {
+		WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE));
+		sparc_perf_event_set_period(event, &event->hw, idx);
+	}
+
+	event->hw.state = 0;
+
+	sparc_pmu_enable_event(cpuc, &event->hw, idx);
+}
+
+static void sparc_pmu_stop(struct perf_event *event, int flags)
+{
+	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+	int idx = active_event_index(cpuc, event);
+
+	if (!(event->hw.state & PERF_HES_STOPPED)) {
+		sparc_pmu_disable_event(cpuc, &event->hw, idx);
+		event->hw.state |= PERF_HES_STOPPED;
+	}
+
+	if (!(event->hw.state & PERF_HES_UPTODATE) && (flags & PERF_EF_UPDATE)) {
+		sparc_perf_event_update(event, &event->hw, idx);
+		event->hw.state |= PERF_HES_UPTODATE;
+	}
+}
+
+static void sparc_pmu_del(struct perf_event *event, int _flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
-	struct hw_perf_event *hwc = &event->hw;
 	unsigned long flags;
 	int i;
 
@@ -722,7 +768,10 @@ static void sparc_pmu_disable(struct per
 
 	for (i = 0; i < cpuc->n_events; i++) {
 		if (event == cpuc->event[i]) {
-			int idx = cpuc->current_idx[i];
+			/* Absorb the final count and turn off the
+			 * event.
+			 */
+			sparc_pmu_stop(event, PERF_EF_UPDATE);
 
 			/* Shift remaining entries down into
 			 * the existing slot.
@@ -734,13 +783,6 @@ static void sparc_pmu_disable(struct per
 					cpuc->current_idx[i];
 			}
 
-			/* Absorb the final count and turn off the
-			 * event.
-			 */
-			sparc_pmu_disable_event(cpuc, hwc, idx);
-			barrier();
-			sparc_perf_event_update(event, hwc, idx);
-
 			perf_event_update_userpage(event);
 
 			cpuc->n_events--;
@@ -752,19 +794,6 @@ static void sparc_pmu_disable(struct per
 	local_irq_restore(flags);
 }
 
-static int active_event_index(struct cpu_hw_events *cpuc,
-			      struct perf_event *event)
-{
-	int i;
-
-	for (i = 0; i < cpuc->n_events; i++) {
-		if (cpuc->event[i] == event)
-			break;
-	}
-	BUG_ON(i == cpuc->n_events);
-	return cpuc->current_idx[i];
-}
-
 static void sparc_pmu_read(struct perf_event *event)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
@@ -774,15 +803,6 @@ static void sparc_pmu_read(struct perf_e
 	sparc_perf_event_update(event, hwc, idx);
 }
 
-static void sparc_pmu_unthrottle(struct perf_event *event)
-{
-	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
-	int idx = active_event_index(cpuc, event);
-	struct hw_perf_event *hwc = &event->hw;
-
-	sparc_pmu_enable_event(cpuc, hwc, idx);
-}
-
 static atomic_t active_events = ATOMIC_INIT(0);
 static DEFINE_MUTEX(pmc_grab_mutex);
 
@@ -984,7 +1004,7 @@ static int collect_events(struct perf_ev
 	return n;
 }
 
-static int sparc_pmu_enable(struct perf_event *event)
+static int sparc_pmu_add(struct perf_event *event, int ef_flags)
 {
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	int n0, ret = -EAGAIN;
@@ -1001,6 +1021,10 @@ static int sparc_pmu_enable(struct perf_
 	cpuc->events[n0] = event->hw.event_base;
 	cpuc->current_idx[n0] = PIC_NO_INDEX;
 
+	event->hw.state = PERF_HES_UPTODATE;
+	if (!(ef_flags & PERF_EF_START))
+		event->hw.state |= PERF_HES_STOPPED;
+
 	/*
 	 * If group events scheduling transaction was started,
 	 * skip the schedulability test here, it will be peformed
@@ -1156,13 +1180,14 @@ static int sparc_pmu_commit_txn(struct p
 }
 
 static struct pmu pmu = {
-	.pmu_enable	= sparc_pmu_pmu_enable,
-	.pmu_disable	= sparc_pmu_pmu_disable,
+	.pmu_enable	= sparc_pmu_enable,
+	.pmu_disable	= sparc_pmu_disable,
 	.event_init	= sparc_pmu_event_init,
-	.enable		= sparc_pmu_enable,
-	.disable	= sparc_pmu_disable,
+	.add		= sparc_pmu_add,
+	.del		= sparc_pmu_del,
+	.start		= sparc_pmu_start,
+	.stop		= sparc_pmu_stop,
 	.read		= sparc_pmu_read,
-	.unthrottle	= sparc_pmu_unthrottle,
 	.start_txn	= sparc_pmu_start_txn,
 	.cancel_txn	= sparc_pmu_cancel_txn,
 	.commit_txn	= sparc_pmu_commit_txn,
@@ -1243,7 +1268,7 @@ static int __kprobes perf_event_nmi_hand
 			continue;
 
 		if (perf_event_overflow(event, 1, &data, regs))
-			sparc_pmu_disable_event(cpuc, hwc, idx);
+			sparc_pmu_stop(event, 0);
 	}
 
 	return NOTIFY_STOP;
Index: linux-2.6/include/linux/ftrace_event.h
===================================================================
--- linux-2.6.orig/include/linux/ftrace_event.h
+++ linux-2.6/include/linux/ftrace_event.h
@@ -248,8 +248,8 @@ DECLARE_PER_CPU(struct pt_regs, perf_tra
 
 extern int  perf_trace_init(struct perf_event *event);
 extern void perf_trace_destroy(struct perf_event *event);
-extern int  perf_trace_enable(struct perf_event *event);
-extern void perf_trace_disable(struct perf_event *event);
+extern int  perf_trace_add(struct perf_event *event, int flags);
+extern void perf_trace_del(struct perf_event *event, int flags);
 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
 				     char *filter_str);
 extern void ftrace_profile_free_filter(struct perf_event *event);
Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -538,6 +538,7 @@ struct hw_perf_event {
 		};
 #endif
 	};
+	int				state;
 	local64_t			prev_count;
 	u64				sample_period;
 	u64				last_period;
@@ -549,6 +550,13 @@ struct hw_perf_event {
 #endif
 };
 
+/*
+ * hw_perf_event::state flags
+ */
+#define PERF_HES_STOPPED	0x01 /* the counter is stopped */
+#define PERF_HES_UPTODATE	0x02 /* event->count up-to-date */
+#define PERF_HES_ARCH		0x04
+
 struct perf_event;
 
 /*
@@ -564,42 +572,62 @@ struct pmu {
 
 	int				*pmu_disable_count;
 
+	/*
+	 * Fully disable/enable this PMU, can be used to protect from the PMI
+	 * as well as for lazy/batch writing of the MSRs.
+	 */
 	void (*pmu_enable)		(struct pmu *pmu); /* optional */
 	void (*pmu_disable)		(struct pmu *pmu); /* optional */
 
 	/*
+	 * Try and initialize the event for this PMU.
 	 * Should return -ENOENT when the @event doesn't match this PMU.
 	 */
 	int (*event_init)		(struct perf_event *event);
 
-	int  (*enable)			(struct perf_event *event);
-	void (*disable)			(struct perf_event *event);
-	int  (*start)			(struct perf_event *event);
-	void (*stop)			(struct perf_event *event);
+#define PERF_EF_START	0x01		/* start the counter when adding    */
+#define PERF_EF_RELOAD	0x02		/* reload the counter when starting */
+#define PERF_EF_UPDATE	0x04		/* update the counter when stopping */
+
+	/*
+	 * Adds/Removes a counter to/from the PMU, can be done inside
+	 * a transaction, see the ->*_txn() methods.
+	 */
+	int  (*add)			(struct perf_event *event, int flags);
+	void (*del)			(struct perf_event *event, int flags);
+
+	/*
+	 * Starts/Stops a counter present on the PMU. The PMI handler
+	 * should stop the counter when perf_event_overflow() returns
+	 * !0. ->start() will be used to continue.
+	 */
+	void (*start)			(struct perf_event *event, int flags);
+	void (*stop)			(struct perf_event *event, int flags);
+
+	/*
+	 * Updates the counter value of the event.
+	 */
 	void (*read)			(struct perf_event *event);
-	void (*unthrottle)		(struct perf_event *event);
 
 	/*
 	 * Group events scheduling is treated as a transaction, add
 	 * group events as a whole and perform one schedulability test.
 	 * If the test fails, roll back the whole group
-	 */
-
-	/*
-	 * Start the transaction, after this ->enable() doesn't need to
+	 *
+	 * Start the transaction, after this ->add() doesn't need to
 	 * do schedulability tests.
 	 */
 	void (*start_txn)	(struct pmu *pmu); /* optional */
 	/*
-	 * If ->start_txn() disabled the ->enable() schedulability test
+	 * If ->start_txn() disabled the ->add() schedulability test
 	 * then ->commit_txn() is required to perform one. On success
 	 * the transaction is closed. On error the transaction is kept
 	 * open until ->cancel_txn() is called.
 	 */
 	int  (*commit_txn)	(struct pmu *pmu); /* optional */
 	/*
-	 * Will cancel the transaction, assumes ->disable() is called
-	 * for each successfull ->enable() during the transaction.
+	 * Will cancel the transaction, assumes ->del() is called
+	 * for each successfull ->add() during the transaction.
 	 */
 	void (*cancel_txn)	(struct pmu *pmu); /* optional */
 };
@@ -680,7 +708,7 @@ struct perf_event {
 	int				nr_siblings;
 	int				group_flags;
 	struct perf_event		*group_leader;
-	struct pmu		*pmu;
+	struct pmu			*pmu;
 
 	enum perf_event_active_state	state;
 	unsigned int			attach_state;
Index: linux-2.6/arch/x86/kernel/cpu/perf_event_intel.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_event_intel.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_event_intel.c
@@ -713,7 +713,7 @@ again:
 		data.period = event->hw.last_period;
 
 		if (perf_event_overflow(event, 1, &data, regs))
-			x86_pmu_stop(event);
+			x86_pmu_stop(event, 0);
 	}
 
 	intel_pmu_ack_status(ack);
Index: linux-2.6/arch/x86/kernel/cpu/perf_event_intel_ds.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/cpu/perf_event_intel_ds.c
+++ linux-2.6/arch/x86/kernel/cpu/perf_event_intel_ds.c
@@ -491,7 +491,7 @@ static void __intel_pmu_pebs_event(struc
 		regs.flags &= ~PERF_EFLAGS_EXACT;
 
 	if (perf_event_overflow(event, 1, &data, &regs))
-		x86_pmu_stop(event);
+		x86_pmu_stop(event, 0);
 }
 
 static void intel_pmu_drain_pebs_core(struct pt_regs *iregs)
Index: linux-2.6/kernel/hw_breakpoint.c
===================================================================
--- linux-2.6.orig/kernel/hw_breakpoint.c
+++ linux-2.6/kernel/hw_breakpoint.c
@@ -574,10 +574,35 @@ static int hw_breakpoint_event_init(stru
 	return 0;
 }
 
+static int hw_breakpoint_add(struct perf_event *bp, int flags)
+{
+	if (!(flags & PERF_EF_START))
+		bp->hw.state = PERF_HES_STOPPED;
+
+	return arch_install_hw_breakpoint(bp);
+}
+
+static void hw_breakpoint_del(struct perf_event *bp, int flags)
+{
+	arch_uninstall_hw_breakpoint(bp);
+}
+
+static void hw_breakpoint_start(struct perf_event *bp, int flags)
+{
+	bp->hw.state = 0;
+}
+
+static void hw_breakpoint_stop(struct perf_event *bp, int flags)
+{
+	bp->hw.state = PERF_HES_STOPPED;
+}
+
 static struct pmu perf_breakpoint = {
 	.event_init	= hw_breakpoint_event_init,
-	.enable		= arch_install_hw_breakpoint,
-	.disable	= arch_uninstall_hw_breakpoint,
+	.add		= hw_breakpoint_add,
+	.del		= hw_breakpoint_del,
+	.start		= hw_breakpoint_start,
+	.stop		= hw_breakpoint_stop,
 	.read		= hw_breakpoint_pmu_read,
 };
 

Re: [RFC][PATCH 13/13] perf: Rework the PMU methods

[Frederic Weisbecker]

On Fri, Jul 09, 2010 at 10:21:30AM +0200, Peter Zijlstra wrote:
> -static void x86_pmu_pmu_enable(struct pmu *pmu)
> +static void x86_pmu_enable(struct pmu *pmu)
>  {
>  	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
>  	struct perf_event *event;
> @@ -839,7 +839,14 @@ static void x86_pmu_pmu_enable(struct pm
>  			    match_prev_assignment(hwc, cpuc, i))
>  				continue;
>  
> -			x86_pmu_stop(event);
> +			/*
> +			 * Ensure we don't accidentally enable a stopped
> +			 * counter simply because we rescheduled.
> +			 */
> +			if (hwc->state & PERF_HES_STOPPED)
> +				hwc->state |= PERF_HES_ARCH;



So, if I understood well, this is to avoid enabling an event from perf_pmu_enable()
while the event was already stopped.

Looks good but why this ARCH name then? This should be a kind of PERF_HES_ONCPU
or so.

Re: [RFC][PATCH 13/13] perf: Rework the PMU methods

[Peter Zijlstra]

On Sat, 2010-07-10 at 15:36 +0200, Frederic Weisbecker wrote:

> So, if I understood well, this is to avoid enabling an event from perf_pmu_enable()
> while the event was already stopped.

Right.

> Looks good but why this ARCH name then? This should be a kind of PERF_HES_ONCPU
> or so.

Because I failed to come up with a sensible name, ONCPU doesn't really
make sense either, its got nothing to do with CPUs.

_REALLY_STOPPED maybe, but then, that's a bad name too ;-)

Re: [RFC][PATCH 00/13] perf pmu interface changes -v3

[Peter Zijlstra]

On Fri, 2010-07-09 at 10:21 +0200, Peter Zijlstra wrote:

Because I'm full of fail today and again had wandering hunks:

> Patches also available in git format for easy testing (tip/master + patches):
> 
>   git://git.kernel.org/pub/scm/linux/kernel/git/peterz/linux-2.6-perf.git perf-pmu

The new HEAD should be 6aa2867187

Re: [RFC][PATCH 00/13] perf pmu interface changes -v3

[Will Deacon]

Hi Peter,

On Fri, 2010-07-09 at 09:21 +0100, Peter Zijlstra wrote:
> The patches are boot tested on x86_64 and compile tested on: powerpc,
> ppc-fsl,
> sparc and arm and SH (by courtesy of Matt Fleming).
> 
> X86 -- appears to be fully functional.
> FSL -- should, after the initial few fixes, at least compile again.
> ARM -- is known to have grown some issues, Will was looking into that.

I've been looking at this and after fixing atomic64 operations and a
sign-extension bug, I'm getting closer to finding the *real* cause of
the problem! It appears that the PMU isn't being enabled for pinned
events. The following patch fixes that, but I'm not sure that it's the
correct thing to do:

diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c
index c440ae1..a946a77 100644
--- a/arch/arm/kernel/perf_event.c
+++ b/arch/arm/kernel/perf_event.c
@@ -304,6 +304,8 @@ armpmu_add(struct perf_event *event, int flags)
        int idx;
        int err = 0;
 
+       perf_pmu_disable(event->pmu);
+
        /* If we don't have a space for the counter then finish early.
*/
        idx = armpmu->get_event_idx(cpuc, hwc);
        if (idx < 0) {
@@ -328,6 +330,7 @@ armpmu_add(struct perf_event *event, int flags)
        perf_event_update_userpage(event);
 
 out:
+       perf_pmu_enable(event->pmu);
        return err;
 }

Because only pinned events seem to be broken, I'm worried that this is
just hiding the problem... or is pmu->add(...) supposed to enable the
PMU as well as installing the event?

The function names aren't especially helpful here either:
armpmu_{start,stop} call armpmu->{enable,disable} and
armpmu_{enable,disable} call armpmu->{start,stop}!

Any thoughts?

Will

Re: [RFC][PATCH 00/13] perf pmu interface changes -v3

[Peter Zijlstra]

On Fri, 2010-07-09 at 16:11 +0100, Will Deacon wrote:
> 
> I've been looking at this and after fixing atomic64 operations and a
> sign-extension bug, I'm getting closer to finding the *real* cause of
> the problem! It appears that the PMU isn't being enabled for pinned
> events. The following patch fixes that, but I'm not sure that it's the
> correct thing to do:

> Because only pinned events seem to be broken, I'm worried that this is
> just hiding the problem... or is pmu->add(...) supposed to enable the
> PMU as well as installing the event?
> 
> The function names aren't especially helpful here either:
> armpmu_{start,stop} call armpmu->{enable,disable} and
> armpmu_{enable,disable} call armpmu->{start,stop}!
> 
> Any thoughts? 

Ah yes.. Silly me..

We used to only call ->enable() [ now called ->add() ] inside
perf_disable()/perf_enable() regions.

But due to patch 9/13 that is no longer the case. I did move it inside
several ->enable methods, but not for ARM (and SH) as it wasn't evident
they needed it (that'll teach me to assume things).

Once we get context per pmu we can move it outside again.

So yes, your patch looks good and I added the SH bit as well. I'll push
it out to the git tree as and worry about rebasing the whole series once
I get back.

Thanks!

(new HEAD 0a20dff8474e5738f086e9acc38649f56938c0ad)

---
 arch/arm/kernel/perf_event.c |    3 +++
 arch/sh/kernel/perf_event.c  |   10 ++++++++--
 2 files changed, 11 insertions(+), 2 deletions(-)

Index: linux-2.6/arch/arm/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/arm/kernel/perf_event.c
+++ linux-2.6/arch/arm/kernel/perf_event.c
@@ -304,6 +304,8 @@ armpmu_add(struct perf_event *event, int
 	int idx;
 	int err = 0;
 
+	perf_pmu_disable(event->pmu);
+
 	/* If we don't have a space for the counter then finish early. */
 	idx = armpmu->get_event_idx(cpuc, hwc);
 	if (idx < 0) {
@@ -328,6 +330,7 @@ armpmu_add(struct perf_event *event, int
 	perf_event_update_userpage(event);
 
 out:
+	perf_pmu_enable(event->pmu);
 	return err;
 }
 
Index: linux-2.6/arch/sh/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/arch/sh/kernel/perf_event.c
+++ linux-2.6/arch/sh/kernel/perf_event.c
@@ -256,11 +256,14 @@ static int sh_pmu_add(struct perf_event 
 	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
 	struct hw_perf_event *hwc = &event->hw;
 	int idx = hwc->idx;
+	int ret = -EAGAIN;
+
+	perf_pmu_disable(event->pmu);
 
 	if (__test_and_set_bit(idx, cpuc->used_mask)) {
 		idx = find_first_zero_bit(cpuc->used_mask, sh_pmu->num_events);
 		if (idx == sh_pmu->num_events)
-			return -EAGAIN;
+			goto ret;
 
 		__set_bit(idx, cpuc->used_mask);
 		hwc->idx = idx;
@@ -273,8 +276,11 @@ static int sh_pmu_add(struct perf_event 
 		sh_pmu_start(event, PERF_EF_RELOAD);
 
 	perf_event_update_userpage(event);
+	ret = 0;
 
-	return 0;
+out:
+	perf_pmu_enable(event->pmu);
+	return ret;
 }
 
 static void sh_pmu_read(struct perf_event *event)

Re: [RFC][PATCH 00/13] perf pmu interface changes -v3

[Matt Fleming]

On Fri, 09 Jul 2010 17:52:14 +0200, Peter Zijlstra <peterz@infradead.org> wrote:
> Index: linux-2.6/arch/sh/kernel/perf_event.c
> ===================================================================
> --- linux-2.6.orig/arch/sh/kernel/perf_event.c
> +++ linux-2.6/arch/sh/kernel/perf_event.c
> @@ -256,11 +256,14 @@ static int sh_pmu_add(struct perf_event 
>  	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
>  	struct hw_perf_event *hwc = &event->hw;
>  	int idx = hwc->idx;
> +	int ret = -EAGAIN;
> +
> +	perf_pmu_disable(event->pmu);
>  
>  	if (__test_and_set_bit(idx, cpuc->used_mask)) {
>  		idx = find_first_zero_bit(cpuc->used_mask, sh_pmu->num_events);
>  		if (idx == sh_pmu->num_events)
> -			return -EAGAIN;
> +			goto ret;
>  
>  		__set_bit(idx, cpuc->used_mask);
>  		hwc->idx = idx;
> @@ -273,8 +276,11 @@ static int sh_pmu_add(struct perf_event 
>  		sh_pmu_start(event, PERF_EF_RELOAD);
>  
>  	perf_event_update_userpage(event);
> +	ret = 0;
>  
> -	return 0;
> +out:
> +	perf_pmu_enable(event->pmu);
> +	return ret;
>  }
>  
>  static void sh_pmu_read(struct perf_event *event)

This patch results in the following compilation error,

  CC      arch/sh/kernel/perf_event.o
cc1: warnings being treated as errors
arch/sh/kernel/perf_event.c: In function 'sh_pmu_add':
arch/sh/kernel/perf_event.c:281: error: label 'out' defined but not used
arch/sh/kernel/perf_event.c:266: error: label 'ret' used but not defined
arch/sh/kernel/perf_event.c: In function 'sh_pmu_setup':
arch/sh/kernel/perf_event.c:343: error: parameter 'cpuhw' is initialized
arch/sh/kernel/perf_event.c:343: error: 'cpu' undeclared (first use in this function)
arch/sh/kernel/perf_event.c:343: error: (Each undeclared identifier is reported only once
arch/sh/kernel/perf_event.c:343: error: for each function it appears in.)
arch/sh/kernel/perf_event.c:343: error: left-hand operand of comma expression has no effect
arch/sh/kernel/perf_event.c:345: error: expected declaration specifiers before 'memset'
arch/sh/kernel/perf_event.c:346: error: expected declaration specifiers before '}' token
arch/sh/kernel/perf_event.c:350: error: expected '=', ',', ';', 'asm' or '__attribute__' before '{' token
arch/sh/kernel/perf_event.c:366: error: expected '=', ',', ';', 'asm' or '__attribute__' before '{' token
arch/sh/kernel/perf_event.c:378: error: old-style parameter declarations in prototyped function definition
arch/sh/kernel/perf_event.c:378: error: expected '{' at end of input
make[1]: *** [arch/sh/kernel/perf_event.o] Error 1
make: *** [arch/sh/kernel/perf_event.o] Error 2

You can pick up a CodeSourcery SH toolchain from
http://www.codesourcery.com/sgpp/lite/superh/portal/release1298 in case
you wanted to build any further changes.

Does this patch look OK? It fixes the above compilation error for me.

diff --git a/arch/sh/kernel/perf_event.c b/arch/sh/kernel/perf_event.c
index 3bfd70b..bcfb325 100644
--- a/arch/sh/kernel/perf_event.c
+++ b/arch/sh/kernel/perf_event.c
@@ -263,7 +263,7 @@ static int sh_pmu_add(struct perf_event *event, int flags)
 	if (__test_and_set_bit(idx, cpuc->used_mask)) {
 		idx = find_first_zero_bit(cpuc->used_mask, sh_pmu->num_events);
 		if (idx == sh_pmu->num_events)
-			goto ret;
+			goto out;
 
 		__set_bit(idx, cpuc->used_mask);
 		hwc->idx = idx;
@@ -339,7 +339,7 @@ static struct pmu pmu = {
 };
 
 static void sh_pmu_setup(int cpu)
-
+{
 	struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu);
 
 	memset(cpuhw, 0, sizeof(struct cpu_hw_events));

Re: [RFC][PATCH 00/13] perf pmu interface changes -v3

[Peter Zijlstra]

On Fri, 2010-07-09 at 16:11 +0100, Will Deacon wrote:
> 
> The function names aren't especially helpful here either:
> armpmu_{start,stop} call armpmu->{enable,disable} and
> armpmu_{enable,disable} call armpmu->{start,stop}! 

Yes, that is rather unfortunate. Maybe we can clean that up once the
dust settles a bit.