[PATCH V5 0/6] csky: Add pmu hardware sampling support

[PATCH V5 0/6] csky: Add pmu hardware sampling support

[Mao Han]

This patch set add hardware sampling support for csky-pmu, and
also add some properties to pmu node definition. perf can record
on hardware event with this patch applied.

Cc: Guo Ren <guoren@kernel.org>

Changes since v4:
  - remove some function hook registration

Changes since v3:
  - change reg-io-width to count-width
  - use macro sign_extend64
  - update commit log

Changes since v2:
  - update dt-binding(csky pmu use rising edge interrupt)
  - use cpuhp_setup_state to enable irq(fix irq enable on smp)

Changes since v1:
  - do not update hpcr when event type is invalid(fix option
    --all-kernel/--all-user)

Guo Ren (1):
  csky: Fixup some error count in 810 & 860.

Mao Han (5):
  csky: Init pmu as a device
  csky: Add count-width property for csky pmu
  csky: Add pmu interrupt support
  dt-bindings: csky: Add csky PMU bindings
  csky: Fix perf record in kernel/user space

 Documentation/devicetree/bindings/csky/pmu.txt |  38 +++
 arch/csky/kernel/perf_event.c                  | 422 +++++++++++++++++++++++--
 2 files changed, 425 insertions(+), 35 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/csky/pmu.txt

-- 
2.7.4

[PATCH V5 1/6] csky: Init pmu as a device

[Mao Han]

This patch change the csky pmu initialization from arch init to
device init. The pmu can be configued with information from
device tree(pmu device name, irq number and etc.).

Signed-off-by: Mao Han <han_mao@c-sky.com>
Cc: Guo Ren <guoren@kernel.org>
---
 arch/csky/kernel/perf_event.c | 50 +++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 48 insertions(+), 2 deletions(-)

diff --git a/arch/csky/kernel/perf_event.c b/arch/csky/kernel/perf_event.c
index 376c972..2282554 100644
--- a/arch/csky/kernel/perf_event.c
+++ b/arch/csky/kernel/perf_event.c
@@ -949,7 +949,7 @@ static int csky_pmu_add(struct perf_event *event, int flags)
 	return 0;
 }
 
-int __init init_hw_perf_events(void)
+int init_hw_perf_events(void)
 {
 	csky_pmu.pmu = (struct pmu) {
 		.pmu_enable	= csky_pmu_enable,
@@ -1028,4 +1028,50 @@ int __init init_hw_perf_events(void)
 
 	return perf_pmu_register(&csky_pmu.pmu, "cpu", PERF_TYPE_RAW);
 }
-arch_initcall(init_hw_perf_events);
+
+int csky_pmu_device_probe(struct platform_device *pdev,
+			  const struct of_device_id *of_table)
+{
+	int ret;
+
+	ret = init_hw_perf_events();
+	if (ret) {
+		pr_notice("[perf] failed to probe PMU!\n");
+		return ret;
+	}
+
+	return ret;
+}
+
+const static struct of_device_id csky_pmu_of_device_ids[] = {
+	{.compatible = "csky,csky-pmu"},
+	{},
+};
+
+static int csky_pmu_dev_probe(struct platform_device *pdev)
+{
+	return csky_pmu_device_probe(pdev, csky_pmu_of_device_ids);
+}
+
+static struct platform_driver csky_pmu_driver = {
+	.driver = {
+		   .name = "csky-pmu",
+		   .of_match_table = csky_pmu_of_device_ids,
+		   },
+	.probe = csky_pmu_dev_probe,
+};
+
+static int __init csky_pmu_probe(void)
+{
+	int ret;
+
+	ret = platform_driver_register(&csky_pmu_driver);
+	if (ret)
+		pr_notice("[perf] PMU initialization failed\n");
+	else
+		pr_notice("[perf] PMU initialization done\n");
+
+	return ret;
+}
+
+device_initcall(csky_pmu_probe);
-- 
2.7.4

[PATCH V5 2/6] csky: Add count-width property for csky pmu

[Mao Han]

The csky pmu counter may have different io width. When the counter is
smaller then 64 bits and counter value is smaller than the old value, it
will result to a extremely large delta value. So the sampled value should
be extend to 64 bits to avoid this, the extension bits base on the
count-width property from dts.

Signed-off-by: Mao Han <han_mao@c-sky.com>
Cc: Guo Ren <guoren@kernel.org>
---
 arch/csky/kernel/perf_event.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/arch/csky/kernel/perf_event.c b/arch/csky/kernel/perf_event.c
index 2282554..a15b397 100644
--- a/arch/csky/kernel/perf_event.c
+++ b/arch/csky/kernel/perf_event.c
@@ -9,6 +9,7 @@
 #include <linux/platform_device.h>
 
 #define CSKY_PMU_MAX_EVENTS 32
+#define DEFAULT_COUNT_WIDTH 48
 
 #define HPCR		"<0, 0x0>"	/* PMU Control reg */
 #define HPCNTENR	"<0, 0x4>"	/* Count Enable reg */
@@ -18,6 +19,7 @@ static void (*hw_raw_write_mapping[CSKY_PMU_MAX_EVENTS])(uint64_t val);
 
 struct csky_pmu_t {
 	struct pmu	pmu;
+	uint32_t	count_width;
 	uint32_t	hpcr;
 } csky_pmu;
 
@@ -804,7 +806,12 @@ static void csky_perf_event_update(struct perf_event *event,
 				   struct hw_perf_event *hwc)
 {
 	uint64_t prev_raw_count = local64_read(&hwc->prev_count);
-	uint64_t new_raw_count = hw_raw_read_mapping[hwc->idx]();
+	/*
+	 * Sign extend count value to 64bit, otherwise delta calculation
+	 * would be incorrect when overflow occurs.
+	 */
+	uint64_t new_raw_count = sign_extend64(
+		hw_raw_read_mapping[hwc->idx](), csky_pmu.count_width - 1);
 	int64_t delta = new_raw_count - prev_raw_count;
 
 	/*
@@ -1032,6 +1039,7 @@ int init_hw_perf_events(void)
 int csky_pmu_device_probe(struct platform_device *pdev,
 			  const struct of_device_id *of_table)
 {
+	struct device_node *node = pdev->dev.of_node;
 	int ret;
 
 	ret = init_hw_perf_events();
@@ -1040,6 +1048,11 @@ int csky_pmu_device_probe(struct platform_device *pdev,
 		return ret;
 	}
 
+	if (of_property_read_u32(node, "count-width",
+				 &csky_pmu.count_width)) {
+		csky_pmu.count_width = DEFAULT_COUNT_WIDTH;
+	}
+
 	return ret;
 }
 
-- 
2.7.4

[PATCH V5 3/6] csky: Add pmu interrupt support

[Mao Han]

This patch add interrupt request and handler for csky pmu.
perf can record on hardware event with this patch applied.

Signed-off-by: Mao Han <han_mao@c-sky.com>
Cc: Guo Ren <guoren@kernel.org>
---
 arch/csky/kernel/perf_event.c | 273 +++++++++++++++++++++++++++++++++++++++---
 1 file changed, 258 insertions(+), 15 deletions(-)

diff --git a/arch/csky/kernel/perf_event.c b/arch/csky/kernel/perf_event.c
index a15b397..8d4547e 100644
--- a/arch/csky/kernel/perf_event.c
+++ b/arch/csky/kernel/perf_event.c
@@ -11,17 +11,48 @@
 #define CSKY_PMU_MAX_EVENTS 32
 #define DEFAULT_COUNT_WIDTH 48
 
-#define HPCR		"<0, 0x0>"	/* PMU Control reg */
-#define HPCNTENR	"<0, 0x4>"	/* Count Enable reg */
+#define HPCR		"<0, 0x0>"      /* PMU Control reg */
+#define HPSPR		"<0, 0x1>"      /* Start PC reg */
+#define HPEPR		"<0, 0x2>"      /* End PC reg */
+#define HPSIR		"<0, 0x3>"      /* Soft Counter reg */
+#define HPCNTENR	"<0, 0x4>"      /* Count Enable reg */
+#define HPINTENR	"<0, 0x5>"      /* Interrupt Enable reg */
+#define HPOFSR		"<0, 0x6>"      /* Interrupt Status reg */
+
+/* The events for a given PMU register set. */
+struct pmu_hw_events {
+	/*
+	 * The events that are active on the PMU for the given index.
+	 */
+	struct perf_event *events[CSKY_PMU_MAX_EVENTS];
+
+	/*
+	 * A 1 bit for an index indicates that the counter is being used for
+	 * an event. A 0 means that the counter can be used.
+	 */
+	unsigned long used_mask[BITS_TO_LONGS(CSKY_PMU_MAX_EVENTS)];
+
+	/*
+	 * Hardware lock to serialize accesses to PMU registers. Needed for the
+	 * read/modify/write sequences.
+	 */
+	raw_spinlock_t pmu_lock;
+};
 
 static uint64_t (*hw_raw_read_mapping[CSKY_PMU_MAX_EVENTS])(void);
 static void (*hw_raw_write_mapping[CSKY_PMU_MAX_EVENTS])(uint64_t val);
 
-struct csky_pmu_t {
-	struct pmu	pmu;
-	uint32_t	count_width;
-	uint32_t	hpcr;
+static struct csky_pmu_t {
+	struct pmu			pmu;
+	struct pmu_hw_events __percpu	*hw_events;
+	struct platform_device		*plat_device;
+	uint32_t			count_width;
+	uint32_t			hpcr;
+	u64				max_period;
 } csky_pmu;
+static int csky_pmu_irq;
+
+#define to_csky_pmu(p)  (container_of(p, struct csky_pmu, pmu))
 
 #define cprgr(reg)				\
 ({						\
@@ -802,6 +833,47 @@ static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
 	},
 };
 
+int  csky_pmu_event_set_period(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	s64 left = local64_read(&hwc->period_left);
+	s64 period = hwc->sample_period;
+	int ret = 0;
+
+	if (unlikely(left <= -period)) {
+		left = period;
+		local64_set(&hwc->period_left, left);
+		hwc->last_period = period;
+		ret = 1;
+	}
+
+	if (unlikely(left <= 0)) {
+		left += period;
+		local64_set(&hwc->period_left, left);
+		hwc->last_period = period;
+		ret = 1;
+	}
+
+	if (left > (s64)csky_pmu.max_period)
+		left = csky_pmu.max_period;
+
+	/*
+	 * The hw event starts counting from this event offset,
+	 * mark it to be able to extract future "deltas":
+	 */
+	local64_set(&hwc->prev_count, (u64)(-left));
+
+	if (hw_raw_write_mapping[hwc->idx] != NULL)
+		hw_raw_write_mapping[hwc->idx]((u64)(-left) &
+						csky_pmu.max_period);
+
+	cpwcr(HPOFSR, ~BIT(hwc->idx) & cprcr(HPOFSR));
+
+	perf_event_update_userpage(event);
+
+	return ret;
+}
+
 static void csky_perf_event_update(struct perf_event *event,
 				   struct hw_perf_event *hwc)
 {
@@ -823,6 +895,11 @@ static void csky_perf_event_update(struct perf_event *event,
 	local64_sub(delta, &hwc->period_left);
 }
 
+static void csky_pmu_reset(void *info)
+{
+	cpwcr(HPCR, BIT(31) | BIT(30) | BIT(1));
+}
+
 static void csky_pmu_read(struct perf_event *event)
 {
 	csky_perf_event_update(event, &event->hw);
@@ -899,7 +976,9 @@ static void csky_pmu_disable(struct pmu *pmu)
 
 static void csky_pmu_start(struct perf_event *event, int flags)
 {
+	unsigned long irq_flags;
 	struct hw_perf_event *hwc = &event->hw;
+	struct pmu_hw_events *events = this_cpu_ptr(csky_pmu.hw_events);
 	int idx = hwc->idx;
 
 	if (WARN_ON_ONCE(idx == -1))
@@ -910,16 +989,35 @@ static void csky_pmu_start(struct perf_event *event, int flags)
 
 	hwc->state = 0;
 
+	csky_pmu_event_set_period(event);
+
+	raw_spin_lock_irqsave(&events->pmu_lock, irq_flags);
+
+	cpwcr(HPINTENR, BIT(idx) | cprcr(HPINTENR));
 	cpwcr(HPCNTENR, BIT(idx) | cprcr(HPCNTENR));
+
+	raw_spin_unlock_irqrestore(&events->pmu_lock, irq_flags);
 }
 
-static void csky_pmu_stop(struct perf_event *event, int flags)
+static void csky_pmu_stop_event(struct perf_event *event)
 {
+	unsigned long irq_flags;
 	struct hw_perf_event *hwc = &event->hw;
+	struct pmu_hw_events *events = this_cpu_ptr(csky_pmu.hw_events);
 	int idx = hwc->idx;
 
+	raw_spin_lock_irqsave(&events->pmu_lock, irq_flags);
+
+	cpwcr(HPINTENR, ~BIT(idx) & cprcr(HPINTENR));
+	cpwcr(HPCNTENR, ~BIT(idx) & cprcr(HPCNTENR));
+
+	raw_spin_unlock_irqrestore(&events->pmu_lock, irq_flags);
+}
+
+static void csky_pmu_stop(struct perf_event *event, int flags)
+{
 	if (!(event->hw.state & PERF_HES_STOPPED)) {
-		cpwcr(HPCNTENR, ~BIT(idx) & cprcr(HPCNTENR));
+		csky_pmu_stop_event(event);
 		event->hw.state |= PERF_HES_STOPPED;
 	}
 
@@ -932,7 +1030,11 @@ static void csky_pmu_stop(struct perf_event *event, int flags)
 
 static void csky_pmu_del(struct perf_event *event, int flags)
 {
+	struct pmu_hw_events *hw_events = this_cpu_ptr(csky_pmu.hw_events);
+	struct hw_perf_event *hwc = &event->hw;
+
 	csky_pmu_stop(event, PERF_EF_UPDATE);
+	hw_events->events[hwc->idx] = NULL;
 
 	perf_event_update_userpage(event);
 }
@@ -940,12 +1042,10 @@ static void csky_pmu_del(struct perf_event *event, int flags)
 /* allocate hardware counter and optionally start counting */
 static int csky_pmu_add(struct perf_event *event, int flags)
 {
+	struct pmu_hw_events *hw_events = this_cpu_ptr(csky_pmu.hw_events);
 	struct hw_perf_event *hwc = &event->hw;
 
-	local64_set(&hwc->prev_count, 0);
-
-	if (hw_raw_write_mapping[hwc->idx] != NULL)
-		hw_raw_write_mapping[hwc->idx](0);
+	hw_events->events[hwc->idx] = event;
 
 	hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
 	if (flags & PERF_EF_START)
@@ -956,8 +1056,109 @@ static int csky_pmu_add(struct perf_event *event, int flags)
 	return 0;
 }
 
+static irqreturn_t csky_pmu_handle_irq(int irq_num, void *dev)
+{
+	struct perf_sample_data data;
+	struct pmu_hw_events *cpuc = this_cpu_ptr(csky_pmu.hw_events);
+	struct pt_regs *regs;
+	int idx;
+
+	/*
+	 * Did an overflow occur?
+	 */
+	if (!cprcr(HPOFSR))
+		return IRQ_NONE;
+
+	/*
+	 * Handle the counter(s) overflow(s)
+	 */
+	regs = get_irq_regs();
+
+	csky_pmu_disable(&csky_pmu.pmu);
+	for (idx = 0; idx < CSKY_PMU_MAX_EVENTS; ++idx) {
+		struct perf_event *event = cpuc->events[idx];
+		struct hw_perf_event *hwc;
+
+		/* Ignore if we don't have an event. */
+		if (!event)
+			continue;
+		/*
+		 * We have a single interrupt for all counters. Check that
+		 * each counter has overflowed before we process it.
+		 */
+		if (!(cprcr(HPOFSR) & 1 << idx))
+			continue;
+
+		hwc = &event->hw;
+		csky_perf_event_update(event, &event->hw);
+		perf_sample_data_init(&data, 0, hwc->last_period);
+		csky_pmu_event_set_period(event);
+
+		if (perf_event_overflow(event, &data, regs))
+			csky_pmu_stop_event(event);
+	}
+	csky_pmu_enable(&csky_pmu.pmu);
+	/*
+	 * Handle the pending perf events.
+	 *
+	 * Note: this call *must* be run with interrupts disabled. For
+	 * platforms that can have the PMU interrupts raised as an NMI, this
+	 * will not work.
+	 */
+	irq_work_run();
+
+	return IRQ_HANDLED;
+}
+
+static int csky_pmu_request_irq(irq_handler_t handler)
+{
+	int err, irqs;
+	struct platform_device *pmu_device = csky_pmu.plat_device;
+
+	if (!pmu_device)
+		return -ENODEV;
+
+	irqs = min(pmu_device->num_resources, num_possible_cpus());
+	if (irqs < 1) {
+		pr_err("no irqs for PMUs defined\n");
+		return -ENODEV;
+	}
+
+	csky_pmu_irq = platform_get_irq(pmu_device, 0);
+	if (csky_pmu_irq < 0)
+		return -ENODEV;
+	err = request_percpu_irq(csky_pmu_irq, handler, "csky-pmu",
+				 this_cpu_ptr(csky_pmu.hw_events));
+	if (err) {
+		pr_err("unable to request IRQ%d for CSKY PMU counters\n",
+		       csky_pmu_irq);
+		return err;
+	}
+
+	return 0;
+}
+
+static void csky_pmu_free_irq(void)
+{
+	int irq;
+	struct platform_device *pmu_device = csky_pmu.plat_device;
+
+	irq = platform_get_irq(pmu_device, 0);
+	if (irq >= 0)
+		free_percpu_irq(irq, this_cpu_ptr(csky_pmu.hw_events));
+}
+
 int init_hw_perf_events(void)
 {
+	int cpu;
+
+	csky_pmu.hw_events = alloc_percpu_gfp(struct pmu_hw_events,
+					      GFP_KERNEL);
+	if (!csky_pmu.hw_events) {
+		pr_info("failed to allocate per-cpu PMU data.\n");
+		return -ENOMEM;
+	}
+
 	csky_pmu.pmu = (struct pmu) {
 		.pmu_enable	= csky_pmu_enable,
 		.pmu_disable	= csky_pmu_disable,
@@ -969,6 +1170,13 @@ int init_hw_perf_events(void)
 		.read		= csky_pmu_read,
 	};
 
+	for_each_possible_cpu(cpu) {
+		struct pmu_hw_events *events;
+
+		events = per_cpu_ptr(csky_pmu.hw_events, cpu);
+		raw_spin_lock_init(&events->pmu_lock);
+	}
+
 	memset((void *)hw_raw_read_mapping, 0,
 		sizeof(hw_raw_read_mapping[CSKY_PMU_MAX_EVENTS]));
 
@@ -1029,11 +1237,19 @@ int init_hw_perf_events(void)
 	hw_raw_write_mapping[0x1a] = csky_pmu_write_l2wac;
 	hw_raw_write_mapping[0x1b] = csky_pmu_write_l2wmc;
 
-	csky_pmu.pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
+	return 0;
+}
 
-	cpwcr(HPCR, BIT(31) | BIT(30) | BIT(1));
+static int csky_pmu_starting_cpu(unsigned int cpu)
+{
+	enable_percpu_irq(csky_pmu_irq, 0);
+	return 0;
+}
 
-	return perf_pmu_register(&csky_pmu.pmu, "cpu", PERF_TYPE_RAW);
+static int csky_pmu_dying_cpu(unsigned int cpu)
+{
+	disable_percpu_irq(csky_pmu_irq);
+	return 0;
 }
 
 int csky_pmu_device_probe(struct platform_device *pdev,
@@ -1052,6 +1268,33 @@ int csky_pmu_device_probe(struct platform_device *pdev,
 				 &csky_pmu.count_width)) {
 		csky_pmu.count_width = DEFAULT_COUNT_WIDTH;
 	}
+	csky_pmu.max_period = ((u64)1 << csky_pmu.count_width) - 1;
+
+	csky_pmu.plat_device = pdev;
+
+	/* Ensure the PMU has sane values out of reset. */
+	on_each_cpu(csky_pmu_reset, &csky_pmu, 1);
+
+	ret = csky_pmu_request_irq(csky_pmu_handle_irq);
+	if (ret) {
+		csky_pmu.pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
+		pr_notice("[perf] PMU request irq fail!\n");
+	}
+
+	ret = cpuhp_setup_state(CPUHP_AP_PERF_ONLINE, "AP_PERF_ONLINE",
+				csky_pmu_starting_cpu,
+				csky_pmu_dying_cpu);
+	if (ret) {
+		csky_pmu_free_irq();
+		free_percpu(csky_pmu.hw_events);
+		return ret;
+	}
+
+	ret = perf_pmu_register(&csky_pmu.pmu, "cpu", PERF_TYPE_RAW);
+	if (ret) {
+		csky_pmu_free_irq();
+		free_percpu(csky_pmu.hw_events);
+	}
 
 	return ret;
 }
-- 
2.7.4

[PATCH V5 4/6] dt-bindings: csky: Add csky PMU bindings

[Mao Han]

This patch adds the documentation to describe that how to add pmu node in
dts.

Signed-off-by: Mao Han <han_mao@c-sky.com>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Guo Ren <guoren@kernel.org>
---
 Documentation/devicetree/bindings/csky/pmu.txt | 38 ++++++++++++++++++++++++++
 1 file changed, 38 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/csky/pmu.txt

diff --git a/Documentation/devicetree/bindings/csky/pmu.txt b/Documentation/devicetree/bindings/csky/pmu.txt
new file mode 100644
index 0000000..53c3b0a
--- /dev/null
+++ b/Documentation/devicetree/bindings/csky/pmu.txt
@@ -0,0 +1,38 @@
+============================
+C-SKY Performance Monitor Units
+============================
+
+C-SKY 8xx series cores often have a PMU for counting cpu and cache events.
+The C-SKY PMU representation in the device tree should be done as under:
+
+==============================
+PMU node bindings definition
+==============================
+
+	Description: Describes PMU
+
+	PROPERTIES
+
+	- compatible
+		Usage: required
+		Value type: <string>
+		Definition: must be "csky,csky-pmu"
+	- interrupts
+		Usage: required
+		Value type: <u32>
+		Definition: must be pmu irq num defined by soc
+	- count-width
+		Usage: optional
+		Value type: <u32>
+		Definition: the width of pmu counter
+
+Examples:
+---------
+
+        pmu {
+                compatible = "csky,csky-pmu";
+                interrupts = <0x17 IRQ_TYPE_EDGE_RISING>;
+                interrupt-parent = <&intc>;
+		count-width = <0x30>;
+        };
+
-- 
2.7.4

[PATCH V5 5/6] csky: Fixup some error count in 810 & 860.

[Mao Han]

From: Guo Ren <ren_guo@c-sky.com>

CK810 pmu only support event with index 0-8 and 0xd; CK860 only
support event 1~4, 0xa~0x1b. So do not register unsupport event
to hardware cache event, which may leader to unknown behavior.

Signed-off-by: Guo Ren <ren_guo@c-sky.com>
Signed-off-by: Mao Han <han_mao@c-sky.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: linux-csky@vger.kernel.org
---
 arch/csky/kernel/perf_event.c | 60 ++++++++++++++++++++++++++++++++++++++-----
 1 file changed, 54 insertions(+), 6 deletions(-)

diff --git a/arch/csky/kernel/perf_event.c b/arch/csky/kernel/perf_event.c
index 8d4547e..499427e 100644
--- a/arch/csky/kernel/perf_event.c
+++ b/arch/csky/kernel/perf_event.c
@@ -734,6 +734,20 @@ static const int csky_pmu_hw_map[PERF_COUNT_HW_MAX] = {
 #define CACHE_OP_UNSUPPORTED	0xffff
 static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
 	[C(L1D)] = {
+#ifdef CONFIG_CPU_CK810
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= 0x5,
+			[C(RESULT_MISS)]	= 0x6,
+		},
+#else
 		[C(OP_READ)] = {
 			[C(RESULT_ACCESS)]	= 0x14,
 			[C(RESULT_MISS)]	= 0x15,
@@ -743,9 +757,10 @@ static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
 			[C(RESULT_MISS)]	= 0x17,
 		},
 		[C(OP_PREFETCH)] = {
-			[C(RESULT_ACCESS)]	= 0x5,
-			[C(RESULT_MISS)]	= 0x6,
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
 		},
+#endif
 	},
 	[C(L1I)] = {
 		[C(OP_READ)] = {
@@ -762,6 +777,20 @@ static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
 		},
 	},
 	[C(LL)] = {
+#ifdef CONFIG_CPU_CK810
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= 0x7,
+			[C(RESULT_MISS)]	= 0x8,
+		},
+#else
 		[C(OP_READ)] = {
 			[C(RESULT_ACCESS)]	= 0x18,
 			[C(RESULT_MISS)]	= 0x19,
@@ -771,29 +800,48 @@ static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
 			[C(RESULT_MISS)]	= 0x1b,
 		},
 		[C(OP_PREFETCH)] = {
-			[C(RESULT_ACCESS)]	= 0x7,
-			[C(RESULT_MISS)]	= 0x8,
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
 		},
+#endif
 	},
 	[C(DTLB)] = {
+#ifdef CONFIG_CPU_CK810
 		[C(OP_READ)] = {
-			[C(RESULT_ACCESS)]	= 0x5,
-			[C(RESULT_MISS)]	= 0xb,
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
 		},
 		[C(OP_WRITE)] = {
 			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
 			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
 		},
+#else
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= 0x14,
+			[C(RESULT_MISS)]	= 0xb,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= 0x16,
+			[C(RESULT_MISS)]	= 0xb,
+		},
+#endif
 		[C(OP_PREFETCH)] = {
 			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
 			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
 		},
 	},
 	[C(ITLB)] = {
+#ifdef CONFIG_CPU_CK810
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+#else
 		[C(OP_READ)] = {
 			[C(RESULT_ACCESS)]	= 0x3,
 			[C(RESULT_MISS)]	= 0xa,
 		},
+#endif
 		[C(OP_WRITE)] = {
 			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
 			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
-- 
2.7.4

[PATCH V5 6/6] csky: Fix perf record in kernel/user space

[Mao Han]

csky_pmu_event_init is called several times during the perf record
initialzation. After configure the event counter in either kernel
space or user space, csky_pmu_event_init is called twice with no
attr specified. Configuration will be overwritten with sampling in
both kernel space and user space. --all-kernel/--all-user is
useless without this patch applied.

Signed-off-by: Mao Han <han_mao@c-sky.com>
Cc: Guo Ren <guoren@kernel.org>
Cc: linux-csky@vger.kernel.org
---
 arch/csky/kernel/perf_event.c | 26 ++++++++++++++------------
 1 file changed, 14 insertions(+), 12 deletions(-)

diff --git a/arch/csky/kernel/perf_event.c b/arch/csky/kernel/perf_event.c
index 499427e..13a254e 100644
--- a/arch/csky/kernel/perf_event.c
+++ b/arch/csky/kernel/perf_event.c
@@ -976,15 +976,6 @@ static int csky_pmu_event_init(struct perf_event *event)
 	struct hw_perf_event *hwc = &event->hw;
 	int ret;
 
-	if (event->attr.exclude_user)
-		csky_pmu.hpcr = BIT(2);
-	else if (event->attr.exclude_kernel)
-		csky_pmu.hpcr = BIT(3);
-	else
-		csky_pmu.hpcr = BIT(2) | BIT(3);
-
-	csky_pmu.hpcr |= BIT(1) | BIT(0);
-
 	switch (event->attr.type) {
 	case PERF_TYPE_HARDWARE:
 		if (event->attr.config >= PERF_COUNT_HW_MAX)
@@ -993,21 +984,32 @@ static int csky_pmu_event_init(struct perf_event *event)
 		if (ret == HW_OP_UNSUPPORTED)
 			return -ENOENT;
 		hwc->idx = ret;
-		return 0;
+		break;
 	case PERF_TYPE_HW_CACHE:
 		ret = csky_pmu_cache_event(event->attr.config);
 		if (ret == CACHE_OP_UNSUPPORTED)
 			return -ENOENT;
 		hwc->idx = ret;
-		return 0;
+		break;
 	case PERF_TYPE_RAW:
 		if (hw_raw_read_mapping[event->attr.config] == NULL)
 			return -ENOENT;
 		hwc->idx = event->attr.config;
-		return 0;
+		break;
 	default:
 		return -ENOENT;
 	}
+
+	if (event->attr.exclude_user)
+		csky_pmu.hpcr = BIT(2);
+	else if (event->attr.exclude_kernel)
+		csky_pmu.hpcr = BIT(3);
+	else
+		csky_pmu.hpcr = BIT(2) | BIT(3);
+
+	csky_pmu.hpcr |= BIT(1) | BIT(0);
+
+	return 0;
 }
 
 /* starts all counters */
-- 
2.7.4