linux-arm-kernel.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
* [RFC PATCH 00/15] ARM: perf: support multiple PMUs
@ 2011-08-15 13:55 Mark Rutland
  2011-08-15 13:55 ` [RFC PATCH 01/15] perf: provide PMU when initing events Mark Rutland
                   ` (15 more replies)
  0 siblings, 16 replies; 18+ messages in thread
From: Mark Rutland @ 2011-08-15 13:55 UTC (permalink / raw)
  To: linux-arm-kernel

System (AKA nest or uncore) PMUs exist on devices which are not affine
to a single CPU. They usually cannot be directly associated with
individual tasks and are asynchronous with respect to the current
execution. Examples of devices which could have system PMUs include L2
cache controllers, GPUs and memory buses.

The following patch series refactors the ARM PMU backend, enabling
new PMUs to reuse the existing code. This should allow for system PMUs
to be supported in future. Further work will be required to get perf to
fully understand system PMUs, but this provides something usable.

The framework is intended to be used by system PMUs which hang off core
platform components (e.g. L2 cache, AXI bus). If a device is complex
enough or separate enough from core functionality to have its own
driver, it should implement its own PMU handling using the core perf
API directly.

The first patch ("perf: provide PMU when initing events") is currently
sitting in the tip tree, but as it's required for event initialization
to function (and hence for the PMU to be usable), it's provided here
for convenience.

The series is based on Will Deacon's perf-updates branch at:
	git://linux-arm.org/linux-2.6-wd.git perf-updates

An example driver using the framework (supporting the PMU present in
L220/PL310 level 2 cache controllers) can be found at:
	git://linux-arm.org/linux-2.6-wd.git perf-l2x0-wip

Any comments would be welcome.

Thanks,
Mark.

Mark Rutland (15):
  perf: provide PMU when initing events
  ARM: perf: only register a CPU PMU when present
  ARM: perf: clean up event group validation
  ARM: perf: remove active_mask
  ARM: perf: move active_events into struct arm_pmu
  ARM: perf: move platform device to struct arm_pmu
  ARM: perf: indirect access to cpu_hw_events
  ARM: perf: remove unnecessary armpmu->stop
  ARM: perf: lock PMU registers per-CPU
  ARM: perf: add type field to struct arm_pmu
  ARM: perf: refactor event mapping
  ARM: perf: add support for multiple PMUs
  ARM: perf: remove event limit from pmu_hw_events
  ARM: perf: remove cpu-related misnomers
  ARM: perf: move arm_pmu into <asm/pmu.h>

 arch/arm/include/asm/pmu.h          |   64 +++++++
 arch/arm/kernel/perf_event.c        |  318 +++++++++++++++++------------------
 arch/arm/kernel/perf_event_v6.c     |   73 ++++++---
 arch/arm/kernel/perf_event_v7.c     |   74 +++++---
 arch/arm/kernel/perf_event_xscale.c |   76 +++++----
 kernel/events/core.c                |    4 +-
 6 files changed, 359 insertions(+), 250 deletions(-)

^ permalink raw reply	[flat|nested] 18+ messages in thread
* [RFC PATCH 11/15] ARM: perf: refactor event mapping
@ 2011-04-28 14:47 Mark Rutland
  0 siblings, 0 replies; 18+ messages in thread
From: Mark Rutland @ 2011-04-28 14:47 UTC (permalink / raw)
  To: linux-arm-kernel

Currently mapping an event type to a hardware configuration value
depends on the data being pointed to from struct arm_pmu. These fields
(cache_map, event_map, raw_event_mask) are currently specific to CPU
PMUs, and do not serve the general case well.

This patch replaces the event map pointers on struct arm_pmu with a new
'map_event' function pointer. Small shim functions are used to reuse
the existing common code.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Will Deacon <will.deacon@arm.com>
---
 arch/arm/kernel/perf_event.c        |   67 +++++++++++++++++++----------------
 arch/arm/kernel/perf_event_v6.c     |   21 ++++++++---
 arch/arm/kernel/perf_event_v7.c     |   37 ++++++++++++++-----
 arch/arm/kernel/perf_event_xscale.c |   14 ++++---
 4 files changed, 87 insertions(+), 52 deletions(-)

diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c
index 1a2ebbf..b13bf23 100644
--- a/arch/arm/kernel/perf_event.c
+++ b/arch/arm/kernel/perf_event.c
@@ -75,11 +75,7 @@ struct arm_pmu {
 	void		(*start)(void);
 	void		(*stop)(void);
 	void		(*reset)(void *);
-	const unsigned	(*cache_map)[PERF_COUNT_HW_CACHE_MAX]
-				    [PERF_COUNT_HW_CACHE_OP_MAX]
-				    [PERF_COUNT_HW_CACHE_RESULT_MAX];
-	const unsigned	(*event_map)[PERF_COUNT_HW_MAX];
-	u32		raw_event_mask;
+	int		(*map_event)(struct perf_event *event);
 	int		num_events;
 	atomic_t	active_events;
 	struct mutex	reserve_mutex;
@@ -129,7 +125,11 @@ EXPORT_SYMBOL_GPL(perf_num_counters);
 #define CACHE_OP_UNSUPPORTED		0xFFFF
 
 static int
-armpmu_map_cache_event(u64 config)
+armpmu_map_cache_event(const unsigned (*cache_map)
+				      [PERF_COUNT_HW_CACHE_MAX]
+				      [PERF_COUNT_HW_CACHE_OP_MAX]
+				      [PERF_COUNT_HW_CACHE_RESULT_MAX],
+		       u64 config)
 {
 	unsigned int cache_type, cache_op, cache_result, ret;
 
@@ -145,7 +145,7 @@ armpmu_map_cache_event(u64 config)
 	if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX)
 		return -EINVAL;
 
-	ret = (int)(*armpmu->cache_map)[cache_type][cache_op][cache_result];
+	ret = (int)(*cache_map)[cache_type][cache_op][cache_result];
 
 	if (ret == CACHE_OP_UNSUPPORTED)
 		return -ENOENT;
@@ -154,16 +154,38 @@ armpmu_map_cache_event(u64 config)
 }
 
 static int
-armpmu_map_event(u64 config)
+armpmu_map_event(const unsigned (*event_map)[PERF_COUNT_HW_MAX], u64 config)
 {
-	int mapping = (*armpmu->event_map)[config];
-	return mapping == HW_OP_UNSUPPORTED ? -EOPNOTSUPP : mapping;
+	int mapping = (*event_map)[config];
+	return mapping == HW_OP_UNSUPPORTED ? -ENOENT : mapping;
 }
 
 static int
-armpmu_map_raw_event(u64 config)
+armpmu_map_raw_event(u32 raw_event_mask, u64 config)
 {
-	return (int)(config & armpmu->raw_event_mask);
+	return (int)(config & raw_event_mask);
+}
+
+static int map_cpu_event(struct perf_event *event,
+			 const unsigned (*event_map)[PERF_COUNT_HW_MAX],
+			 const unsigned (*cache_map)
+					[PERF_COUNT_HW_CACHE_MAX]
+					[PERF_COUNT_HW_CACHE_OP_MAX]
+					[PERF_COUNT_HW_CACHE_RESULT_MAX],
+			 u32 raw_event_mask)
+{
+	u64 config = event->attr.config;
+
+	switch (event->attr.type) {
+	case PERF_TYPE_HARDWARE:
+		return armpmu_map_event(event_map, config);
+	case PERF_TYPE_HW_CACHE:
+		return armpmu_map_cache_event(cache_map, config);
+	case PERF_TYPE_RAW:
+		return armpmu_map_raw_event(raw_event_mask, config);
+	}
+
+	return -ENOENT;
 }
 
 static int
@@ -484,17 +506,7 @@ __hw_perf_event_init(struct perf_event *event)
 	struct hw_perf_event *hwc = &event->hw;
 	int mapping, err;
 
-	/* Decode the generic type into an ARM event identifier. */
-	if (PERF_TYPE_HARDWARE == event->attr.type) {
-		mapping = armpmu_map_event(event->attr.config);
-	} else if (PERF_TYPE_HW_CACHE == event->attr.type) {
-		mapping = armpmu_map_cache_event(event->attr.config);
-	} else if (PERF_TYPE_RAW == event->attr.type) {
-		mapping = armpmu_map_raw_event(event->attr.config);
-	} else {
-		pr_debug("event type %x not supported\n", event->attr.type);
-		return -EOPNOTSUPP;
-	}
+	mapping = armpmu->map_event(event);
 
 	if (mapping < 0) {
 		pr_debug("event %x:%llx not supported\n", event->attr.type,
@@ -550,15 +562,8 @@ static int armpmu_event_init(struct perf_event *event)
 	int err = 0;
 	atomic_t *active_events = &armpmu->active_events;
 
-	switch (event->attr.type) {
-	case PERF_TYPE_RAW:
-	case PERF_TYPE_HARDWARE:
-	case PERF_TYPE_HW_CACHE:
-		break;
-
-	default:
+	if (armpmu->map_event(event) == -ENOENT)
 		return -ENOENT;
-	}
 
 	event->destroy = hw_perf_event_destroy;
 
diff --git a/arch/arm/kernel/perf_event_v6.c b/arch/arm/kernel/perf_event_v6.c
index 68cf704..a4c5aa9 100644
--- a/arch/arm/kernel/perf_event_v6.c
+++ b/arch/arm/kernel/perf_event_v6.c
@@ -657,6 +657,12 @@ armv6mpcore_pmu_disable_event(struct hw_perf_event *hwc,
 	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
 }
 
+static int armv6_map_event(struct perf_event *event)
+{
+	return map_cpu_event(event, &armv6_perf_map,
+				&armv6_perf_cache_map, 0xFF);
+}
+
 static struct arm_pmu armv6pmu = {
 	.id			= ARM_PERF_PMU_ID_V6,
 	.name			= "v6",
@@ -668,9 +674,7 @@ static struct arm_pmu armv6pmu = {
 	.get_event_idx		= armv6pmu_get_event_idx,
 	.start			= armv6pmu_start,
 	.stop			= armv6pmu_stop,
-	.cache_map		= &armv6_perf_cache_map,
-	.event_map		= &armv6_perf_map,
-	.raw_event_mask		= 0xFF,
+	.map_event		= armv6_map_event,
 	.num_events		= 3,
 	.max_period		= (1LLU << 32) - 1,
 };
@@ -687,6 +691,13 @@ static struct arm_pmu *__init armv6pmu_init(void)
  * disable the interrupt reporting and update the event. When unthrottling we
  * reset the period and enable the interrupt reporting.
  */
+
+static int armv6mpcore_map_event(struct perf_event *event)
+{
+	return map_cpu_event(event, &armv6mpcore_perf_map,
+				&armv6mpcore_perf_cache_map, 0xFF);
+}
+
 static struct arm_pmu armv6mpcore_pmu = {
 	.id			= ARM_PERF_PMU_ID_V6MP,
 	.name			= "v6mpcore",
@@ -698,9 +709,7 @@ static struct arm_pmu armv6mpcore_pmu = {
 	.get_event_idx		= armv6pmu_get_event_idx,
 	.start			= armv6pmu_start,
 	.stop			= armv6pmu_stop,
-	.cache_map		= &armv6mpcore_perf_cache_map,
-	.event_map		= &armv6mpcore_perf_map,
-	.raw_event_mask		= 0xFF,
+	.map_event		= armv6mpcore_map_event,
 	.num_events		= 3,
 	.max_period		= (1LLU << 32) - 1,
 };
diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c
index 68ac522..be7b58a 100644
--- a/arch/arm/kernel/perf_event_v7.c
+++ b/arch/arm/kernel/perf_event_v7.c
@@ -1140,6 +1140,30 @@ static void armv7pmu_reset(void *info)
 	armv7_pmnc_write(ARMV7_PMNC_P | ARMV7_PMNC_C);
 }
 
+static int armv7_a8_map_event(struct perf_event *event)
+{
+	return map_cpu_event(event, &armv7_a8_perf_map,
+				&armv7_a8_perf_cache_map, 0xFF);
+}
+
+static int armv7_a9_map_event(struct perf_event *event)
+{
+	return map_cpu_event(event, &armv7_a9_perf_map,
+				&armv7_a9_perf_cache_map, 0xFF);
+}
+
+static int armv7_a5_map_event(struct perf_event *event)
+{
+	return map_cpu_event(event, &armv7_a5_perf_map,
+				&armv7_a5_perf_cache_map, 0xFF);
+}
+
+static int armv7_a15_map_event(struct perf_event *event)
+{
+	return map_cpu_event(event, &armv7_a15_perf_map,
+				&armv7_a15_perf_cache_map, 0xFF);
+}
+
 static struct arm_pmu armv7pmu = {
 	.handle_irq		= armv7pmu_handle_irq,
 	.enable			= armv7pmu_enable_event,
@@ -1150,7 +1174,6 @@ static struct arm_pmu armv7pmu = {
 	.start			= armv7pmu_start,
 	.stop			= armv7pmu_stop,
 	.reset			= armv7pmu_reset,
-	.raw_event_mask		= 0xFF,
 	.max_period		= (1LLU << 32) - 1,
 };
 
@@ -1169,8 +1192,7 @@ static struct arm_pmu *__init armv7_a8_pmu_init(void)
 {
 	armv7pmu.id		= ARM_PERF_PMU_ID_CA8;
 	armv7pmu.name		= "ARMv7 Cortex-A8";
-	armv7pmu.cache_map	= &armv7_a8_perf_cache_map;
-	armv7pmu.event_map	= &armv7_a8_perf_map;
+	armv7pmu.map_event	= armv7_a8_map_event;
 	armv7pmu.num_events	= armv7_read_num_pmnc_events();
 	return &armv7pmu;
 }
@@ -1179,8 +1201,7 @@ static struct arm_pmu *__init armv7_a9_pmu_init(void)
 {
 	armv7pmu.id		= ARM_PERF_PMU_ID_CA9;
 	armv7pmu.name		= "ARMv7 Cortex-A9";
-	armv7pmu.cache_map	= &armv7_a9_perf_cache_map;
-	armv7pmu.event_map	= &armv7_a9_perf_map;
+	armv7pmu.map_event	= armv7_a9_map_event;
 	armv7pmu.num_events	= armv7_read_num_pmnc_events();
 	return &armv7pmu;
 }
@@ -1189,8 +1210,7 @@ static struct arm_pmu *__init armv7_a5_pmu_init(void)
 {
 	armv7pmu.id		= ARM_PERF_PMU_ID_CA5;
 	armv7pmu.name		= "ARMv7 Cortex-A5";
-	armv7pmu.cache_map	= &armv7_a5_perf_cache_map;
-	armv7pmu.event_map	= &armv7_a5_perf_map;
+	armv7pmu.map_event	= armv7_a5_map_event;
 	armv7pmu.num_events	= armv7_read_num_pmnc_events();
 	return &armv7pmu;
 }
@@ -1199,8 +1219,7 @@ static struct arm_pmu *__init armv7_a15_pmu_init(void)
 {
 	armv7pmu.id		= ARM_PERF_PMU_ID_CA15;
 	armv7pmu.name		= "ARMv7 Cortex-A15";
-	armv7pmu.cache_map	= &armv7_a15_perf_cache_map;
-	armv7pmu.event_map	= &armv7_a15_perf_map;
+	armv7pmu.map_event	= armv7_a15_map_event;
 	armv7pmu.num_events	= armv7_read_num_pmnc_events();
 	armv7pmu.set_event_filter = armv7pmu_set_event_filter;
 	return &armv7pmu;
diff --git a/arch/arm/kernel/perf_event_xscale.c b/arch/arm/kernel/perf_event_xscale.c
index 18e4823..d4c7610 100644
--- a/arch/arm/kernel/perf_event_xscale.c
+++ b/arch/arm/kernel/perf_event_xscale.c
@@ -425,6 +425,12 @@ xscale1pmu_write_counter(int counter, u32 val)
 	}
 }
 
+static int xscale_map_event(struct perf_event *event)
+{
+	return map_cpu_event(event, &xscale_perf_map,
+				&xscale_perf_cache_map, 0xFF);
+}
+
 static struct arm_pmu xscale1pmu = {
 	.id		= ARM_PERF_PMU_ID_XSCALE1,
 	.name		= "xscale1",
@@ -436,9 +442,7 @@ static struct arm_pmu xscale1pmu = {
 	.get_event_idx	= xscale1pmu_get_event_idx,
 	.start		= xscale1pmu_start,
 	.stop		= xscale1pmu_stop,
-	.cache_map	= &xscale_perf_cache_map,
-	.event_map	= &xscale_perf_map,
-	.raw_event_mask	= 0xFF,
+	.map_event	= xscale_map_event,
 	.num_events	= 3,
 	.max_period	= (1LLU << 32) - 1,
 };
@@ -799,9 +803,7 @@ static struct arm_pmu xscale2pmu = {
 	.get_event_idx	= xscale2pmu_get_event_idx,
 	.start		= xscale2pmu_start,
 	.stop		= xscale2pmu_stop,
-	.cache_map	= &xscale_perf_cache_map,
-	.event_map	= &xscale_perf_map,
-	.raw_event_mask	= 0xFF,
+	.map_event	= xscale_map_event,
 	.num_events	= 5,
 	.max_period	= (1LLU << 32) - 1,
 };
-- 
1.7.0.4

^ permalink raw reply related	[flat|nested] 18+ messages in thread
end of thread, other threads:[~2011-08-17 14:12 UTC | newest]

Thread overview: 18+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-08-15 13:55 [RFC PATCH 00/15] ARM: perf: support multiple PMUs Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 01/15] perf: provide PMU when initing events Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 02/15] ARM: perf: only register a CPU PMU when present Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 03/15] ARM: perf: clean up event group validation Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 04/15] ARM: perf: remove active_mask Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 05/15] ARM: perf: move active_events into struct arm_pmu Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 06/15] ARM: perf: move platform device to " Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 07/15] ARM: perf: indirect access to cpu_hw_events Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 08/15] ARM: perf: remove unnecessary armpmu->stop Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 09/15] ARM: perf: lock PMU registers per-CPU Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 10/15] ARM: perf: add type field to struct arm_pmu Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 11/15] ARM: perf: refactor event mapping Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 12/15] ARM: perf: add support for multiple PMUs Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 13/15] ARM: perf: remove event limit from pmu_hw_events Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 14/15] ARM: perf: remove cpu-related misnomers Mark Rutland
2011-08-15 13:55 ` [RFC PATCH 15/15] ARM: perf: move arm_pmu into <asm/pmu.h> Mark Rutland
2011-08-17 14:12 ` [RFC PATCH 00/15] ARM: perf: support multiple PMUs Jamie Iles
  -- strict thread matches above, loose matches on Subject: below --
2011-04-28 14:47 [RFC PATCH 11/15] ARM: perf: refactor event mapping Mark Rutland

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).