[PATCH 0/5] introduce Kernel PMU events support

[PATCH 0/5] introduce Kernel PMU events support

[Eugeniy Paltsev]

Introduce Kernel PMU events support and refactor ARC-specific perf code. 

Eugeniy Paltsev (5):
  ARC: perf: trivial code cleanup
  ARC: perf: introduce Kernel PMU events support
  ARC: perf: fix of kernel data types using
  ARC: perf: fix description comment
  ARC: perf: remove useless ifdefs

 arch/arc/kernel/perf_event.c | 179 ++++++++++++++++++++++++++++++++++---------
 1 file changed, 141 insertions(+), 38 deletions(-)

-- 
2.14.5

[PATCH 1/5] ARC: perf: trivial code cleanup

[Eugeniy Paltsev]

Use BIT(), lower_32_bits(), upper_32_bits() macroses,
fix code style violations.

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
---
 arch/arc/kernel/perf_event.c | 31 ++++++++++++++++++-------------
 1 file changed, 18 insertions(+), 13 deletions(-)

diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
index 8aec462d90fb..811a07a2ca21 100644
--- a/arch/arc/kernel/perf_event.c
+++ b/arch/arc/kernel/perf_event.c
@@ -19,6 +19,9 @@
 #include <asm/arcregs.h>
 #include <asm/stacktrace.h>
 
+/* HW holds 8 symbols + one for null terminator */
+#define ARCPMU_EVENT_NAME_LEN	9
+
 struct arc_pmu {
 	struct pmu	pmu;
 	unsigned int	irq;
@@ -155,7 +158,7 @@ static int arc_pmu_event_init(struct perf_event *event)
 	int ret;
 
 	if (!is_sampling_event(event)) {
-		hwc->sample_period  = arc_pmu->max_period;
+		hwc->sample_period = arc_pmu->max_period;
 		hwc->last_period = hwc->sample_period;
 		local64_set(&hwc->period_left, hwc->sample_period);
 	}
@@ -192,6 +195,7 @@ static int arc_pmu_event_init(struct perf_event *event)
 		pr_debug("init cache event with h/w %08x \'%s\'\n",
 			 (int)hwc->config, arc_pmu_ev_hw_map[ret]);
 		return 0;
+
 	default:
 		return -ENOENT;
 	}
@@ -246,8 +250,8 @@ static int arc_pmu_event_set_period(struct perf_event *event)
 	write_aux_reg(ARC_REG_PCT_INDEX, idx);
 
 	/* Write value */
-	write_aux_reg(ARC_REG_PCT_COUNTL, (u32)value);
-	write_aux_reg(ARC_REG_PCT_COUNTH, (value >> 32));
+	write_aux_reg(ARC_REG_PCT_COUNTL, lower_32_bits(value));
+	write_aux_reg(ARC_REG_PCT_COUNTH, upper_32_bits(value));
 
 	perf_event_update_userpage(event);
 
@@ -277,7 +281,7 @@ static void arc_pmu_start(struct perf_event *event, int flags)
 	/* Enable interrupt for this counter */
 	if (is_sampling_event(event))
 		write_aux_reg(ARC_REG_PCT_INT_CTRL,
-			      read_aux_reg(ARC_REG_PCT_INT_CTRL) | (1 << idx));
+			      read_aux_reg(ARC_REG_PCT_INT_CTRL) | BIT(idx));
 
 	/* enable ARC pmu here */
 	write_aux_reg(ARC_REG_PCT_INDEX, idx);		/* counter # */
@@ -295,9 +299,9 @@ static void arc_pmu_stop(struct perf_event *event, int flags)
 		 * Reset interrupt flag by writing of 1. This is required
 		 * to make sure pending interrupt was not left.
 		 */
-		write_aux_reg(ARC_REG_PCT_INT_ACT, 1 << idx);
+		write_aux_reg(ARC_REG_PCT_INT_ACT, BIT(idx));
 		write_aux_reg(ARC_REG_PCT_INT_CTRL,
-			      read_aux_reg(ARC_REG_PCT_INT_CTRL) & ~(1 << idx));
+			      read_aux_reg(ARC_REG_PCT_INT_CTRL) & ~BIT(idx));
 	}
 
 	if (!(event->hw.state & PERF_HES_STOPPED)) {
@@ -349,9 +353,10 @@ static int arc_pmu_add(struct perf_event *event, int flags)
 
 	if (is_sampling_event(event)) {
 		/* Mimic full counter overflow as other arches do */
-		write_aux_reg(ARC_REG_PCT_INT_CNTL, (u32)arc_pmu->max_period);
+		write_aux_reg(ARC_REG_PCT_INT_CNTL,
+			      lower_32_bits(arc_pmu->max_period));
 		write_aux_reg(ARC_REG_PCT_INT_CNTH,
-			      (arc_pmu->max_period >> 32));
+			      upper_32_bits(arc_pmu->max_period));
 	}
 
 	write_aux_reg(ARC_REG_PCT_CONFIG, 0);
@@ -392,7 +397,7 @@ static irqreturn_t arc_pmu_intr(int irq, void *dev)
 		idx = __ffs(active_ints);
 
 		/* Reset interrupt flag by writing of 1 */
-		write_aux_reg(ARC_REG_PCT_INT_ACT, 1 << idx);
+		write_aux_reg(ARC_REG_PCT_INT_ACT, BIT(idx));
 
 		/*
 		 * On reset of "interrupt active" bit corresponding
@@ -400,7 +405,7 @@ static irqreturn_t arc_pmu_intr(int irq, void *dev)
 		 * Now we need to re-enable interrupt for the counter.
 		 */
 		write_aux_reg(ARC_REG_PCT_INT_CTRL,
-			read_aux_reg(ARC_REG_PCT_INT_CTRL) | (1 << idx));
+			read_aux_reg(ARC_REG_PCT_INT_CTRL) | BIT(idx));
 
 		event = pmu_cpu->act_counter[idx];
 		hwc = &event->hw;
@@ -414,7 +419,7 @@ static irqreturn_t arc_pmu_intr(int irq, void *dev)
 				arc_pmu_stop(event, 0);
 		}
 
-		active_ints &= ~(1U << idx);
+		active_ints &= ~BIT(idx);
 	} while (active_ints);
 
 done:
@@ -453,7 +458,7 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
 			uint32_t word0, word1;
 			char sentinel;
 		} indiv;
-		char str[9];
+		char str[ARCPMU_EVENT_NAME_LEN];
 	} cc_name;
 
 
@@ -483,7 +488,7 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
 		arc_pmu->n_counters, counter_size, cc_bcr.c,
 		has_interrupts ? ", [overflow IRQ support]":"");
 
-	cc_name.str[8] = 0;
+	cc_name.str[ARCPMU_EVENT_NAME_LEN - 1] = 0;
 	for (i = 0; i < PERF_COUNT_ARC_HW_MAX; i++)
 		arc_pmu->ev_hw_idx[i] = -1;
 
-- 
2.14.5

Re: [PATCH 1/5] ARC: perf: trivial code cleanup

[Vineet Gupta]

On 12/5/18 9:06 AM, Eugeniy Paltsev wrote:
> Use BIT(), lower_32_bits(), upper_32_bits() macroses,
> fix code style violations.
>
> Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>

Acked-by: Vineet Gupta <vgupta@synopsys.com>

Thx,
-Vineet

[PATCH 2/5] ARC: perf: introduce Kernel PMU events support

[Eugeniy Paltsev]

Export all available ARC architected hardware events as
kernel PMU events to make non-generic events accessible.

ARC PMU HW allow us to read the list of all available
events names. So we generate kernel PMU event list
dynamically in arc_pmu_device_probe() using
human-readable events names we got from HW instead of
using pre-defined events list.

-------------------------->8--------------------------
$ perf list
  [snip]
  arc_pmu/bdata64/                  [Kernel PMU event]
  arc_pmu/bdcstall/                 [Kernel PMU event]
  arc_pmu/bdslot/                   [Kernel PMU event]
  arc_pmu/bfbmp/                    [Kernel PMU event]
  arc_pmu/bfirqex/                  [Kernel PMU event]
  arc_pmu/bflgstal/                 [Kernel PMU event]
  arc_pmu/bflush/                   [Kernel PMU event]
-------------------------->8--------------------------

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
---
 arch/arc/kernel/perf_event.c | 107 ++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 106 insertions(+), 1 deletion(-)

diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
index 811a07a2ca21..97b88b00c418 100644
--- a/arch/arc/kernel/perf_event.c
+++ b/arch/arc/kernel/perf_event.c
@@ -22,12 +22,28 @@
 /* HW holds 8 symbols + one for null terminator */
 #define ARCPMU_EVENT_NAME_LEN	9
 
+enum arc_pmu_attr_groups {
+	ARCPMU_ATTR_GR_EVENTS,
+	ARCPMU_ATTR_GR_FORMATS,
+	ARCPMU_NR_ATTR_GR
+};
+
+struct arc_pmu_raw_event_entry {
+	char name[ARCPMU_EVENT_NAME_LEN];
+};
+
 struct arc_pmu {
 	struct pmu	pmu;
 	unsigned int	irq;
 	int		n_counters;
+	int		n_events;
 	u64		max_period;
 	int		ev_hw_idx[PERF_COUNT_ARC_HW_MAX];
+
+	struct arc_pmu_raw_event_entry	*raw_entry;
+	struct attribute		**attrs;
+	struct perf_pmu_events_attr	*attr;
+	const struct attribute_group	*attr_groups[ARCPMU_NR_ATTR_GR + 1];
 };
 
 struct arc_pmu_cpu {
@@ -196,6 +212,17 @@ static int arc_pmu_event_init(struct perf_event *event)
 			 (int)hwc->config, arc_pmu_ev_hw_map[ret]);
 		return 0;
 
+	case PERF_TYPE_RAW:
+		if (event->attr.config >= arc_pmu->n_events)
+			return -ENOENT;
+
+		hwc->config |= event->attr.config;
+		pr_debug("init raw event with idx %lld \'%s\'\n",
+			 event->attr.config,
+			 arc_pmu->raw_entry[event->attr.config].name);
+
+		return 0;
+
 	default:
 		return -ENOENT;
 	}
@@ -446,6 +473,68 @@ static void arc_cpu_pmu_irq_init(void *data)
 	write_aux_reg(ARC_REG_PCT_INT_ACT, 0xffffffff);
 }
 
+/* Event field occupies the bottom 15 bits of our config field */
+PMU_FORMAT_ATTR(event, "config:0-14");
+static struct attribute *arc_pmu_format_attrs[] = {
+	&format_attr_event.attr,
+	NULL,
+};
+
+static struct attribute_group arc_pmu_format_attr_gr = {
+	.name = "format",
+	.attrs = arc_pmu_format_attrs,
+};
+
+static ssize_t
+arc_pmu_events_sysfs_show(struct device *dev,
+			  struct device_attribute *attr, char *page)
+{
+	struct perf_pmu_events_attr *pmu_attr;
+
+	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
+	return sprintf(page, "event=0x%04llx\n", pmu_attr->id);
+}
+
+/*
+ * We don't add attrs here as we don't have pre-defined list of perf events.
+ * We will generate and add attrs dynamically in probe() after we read HW
+ * configuration.
+ */
+static struct attribute_group arc_pmu_events_attr_gr = {
+	.name = "events",
+};
+
+static void arc_pmu_add_raw_event_attr(int j, char *str)
+{
+	memmove(arc_pmu->raw_entry[j].name, str, ARCPMU_EVENT_NAME_LEN - 1);
+	arc_pmu->attr[j].attr.attr.name = arc_pmu->raw_entry[j].name;
+	arc_pmu->attr[j].attr.attr.mode = VERIFY_OCTAL_PERMISSIONS(0444);
+	arc_pmu->attr[j].attr.show = arc_pmu_events_sysfs_show;
+	arc_pmu->attr[j].id = j;
+	arc_pmu->attrs[j] = &(arc_pmu->attr[j].attr.attr);
+}
+
+static int arc_pmu_raw_alloc(struct device *dev)
+{
+	arc_pmu->attr = devm_kmalloc_array(dev, arc_pmu->n_events + 1,
+		sizeof(struct perf_pmu_events_attr), GFP_KERNEL | __GFP_ZERO);
+	if (!arc_pmu->attr)
+		return -ENOMEM;
+
+	arc_pmu->attrs = devm_kmalloc_array(dev, arc_pmu->n_events + 1,
+		sizeof(*arc_pmu->attrs), GFP_KERNEL | __GFP_ZERO);
+	if (!arc_pmu->attrs)
+		return -ENOMEM;
+
+	arc_pmu->raw_entry = devm_kmalloc_array(dev, arc_pmu->n_events,
+		sizeof(struct arc_pmu_raw_event_entry),
+		GFP_KERNEL | __GFP_ZERO);
+	if (!arc_pmu->raw_entry)
+		return -ENOMEM;
+
+	return 0;
+}
+
 static int arc_pmu_device_probe(struct platform_device *pdev)
 {
 	struct arc_reg_pct_build pct_bcr;
@@ -477,6 +566,11 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
 	if (!arc_pmu)
 		return -ENOMEM;
 
+	arc_pmu->n_events = cc_bcr.c;
+
+	if (arc_pmu_raw_alloc(&pdev->dev))
+		return -ENOMEM;
+
 	has_interrupts = is_isa_arcv2() ? pct_bcr.i : 0;
 
 	arc_pmu->n_counters = pct_bcr.c;
@@ -508,8 +602,14 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
 				arc_pmu->ev_hw_idx[i] = j;
 			}
 		}
+
+		arc_pmu_add_raw_event_attr(j, cc_name.str);
 	}
 
+	arc_pmu_events_attr_gr.attrs = arc_pmu->attrs;
+	arc_pmu->attr_groups[ARCPMU_ATTR_GR_EVENTS] = &arc_pmu_events_attr_gr;
+	arc_pmu->attr_groups[ARCPMU_ATTR_GR_FORMATS] = &arc_pmu_format_attr_gr;
+
 	arc_pmu->pmu = (struct pmu) {
 		.pmu_enable	= arc_pmu_enable,
 		.pmu_disable	= arc_pmu_disable,
@@ -519,6 +619,7 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
 		.start		= arc_pmu_start,
 		.stop		= arc_pmu_stop,
 		.read		= arc_pmu_read,
+		.attr_groups	= arc_pmu->attr_groups,
 	};
 
 	if (has_interrupts) {
@@ -540,7 +641,11 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
 	} else
 		arc_pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
 
-	return perf_pmu_register(&arc_pmu->pmu, pdev->name, PERF_TYPE_RAW);
+	/*
+	 * perf parser doesn't really like '-' symbol in events name, so let's
+	 * use '_' in arc pmu name as it goes to kernel PMU event prefix.
+	 */
+	return perf_pmu_register(&arc_pmu->pmu, "arc_pmu", PERF_TYPE_RAW);
 }
 
 #ifdef CONFIG_OF
-- 
2.14.5

Re: [PATCH 2/5] ARC: perf: introduce Kernel PMU events support

[Vineet Gupta]

On 12/5/18 9:06 AM, Eugeniy Paltsev wrote:
> Export all available ARC architected hardware events as
> kernel PMU events to make non-generic events accessible.
>
> ARC PMU HW allow us to read the list of all available
> events names. So we generate kernel PMU event list
> dynamically in arc_pmu_device_probe() using
> human-readable events names we got from HW instead of
> using pre-defined events list.
>
> -------------------------->8--------------------------
> $ perf list
>   [snip]
>   arc_pmu/bdata64/                  [Kernel PMU event]
>   arc_pmu/bdcstall/                 [Kernel PMU event]
>   arc_pmu/bdslot/                   [Kernel PMU event]
>   arc_pmu/bfbmp/                    [Kernel PMU event]
>   arc_pmu/bfirqex/                  [Kernel PMU event]
>   arc_pmu/bflgstal/                 [Kernel PMU event]
>   arc_pmu/bflush/                   [Kernel PMU event]

Lets call this pct iso pmu since pmu has more of power mgmt connotation. I know
the code has pmu littered all over but atleast the user interface could be more
intuitive.

BTW this approach seems more user friendly and is different from Alexey's earlier
stab at implementing raw events [1]. He didn't keep around any list (and relied on
use rlooking in the PRM to find his interesting event of the day) and pass as
*r*foobar.  PeterZ at the time had some reservations with that which I never fully
understood.

[1] https://lore.kernel.org/patchwork/patch/568769/

> -------------------------->8--------------------------
>
> Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
> ---
>  arch/arc/kernel/perf_event.c | 107 ++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 106 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
> index 811a07a2ca21..97b88b00c418 100644
> --- a/arch/arc/kernel/perf_event.c
> +++ b/arch/arc/kernel/perf_event.c
> @@ -22,12 +22,28 @@
>  /* HW holds 8 symbols + one for null terminator */
>  #define ARCPMU_EVENT_NAME_LEN	9
>  
> +enum arc_pmu_attr_groups {
> +	ARCPMU_ATTR_GR_EVENTS,
> +	ARCPMU_ATTR_GR_FORMATS,
> +	ARCPMU_NR_ATTR_GR
> +};
> +
> +struct arc_pmu_raw_event_entry {
> +	char name[ARCPMU_EVENT_NAME_LEN];
> +};
> +
>  struct arc_pmu {
>  	struct pmu	pmu;
>  	unsigned int	irq;
>  	int		n_counters;
> +	int		n_events;
>  	u64		max_period;
>  	int		ev_hw_idx[PERF_COUNT_ARC_HW_MAX];
> +
> +	struct arc_pmu_raw_event_entry	*raw_entry;
> +	struct attribute		**attrs;
> +	struct perf_pmu_events_attr	*attr;
> +	const struct attribute_group	*attr_groups[ARCPMU_NR_ATTR_GR + 1];
>  };
>  
>  struct arc_pmu_cpu {
> @@ -196,6 +212,17 @@ static int arc_pmu_event_init(struct perf_event *event)
>  			 (int)hwc->config, arc_pmu_ev_hw_map[ret]);
>  		return 0;
>  
> +	case PERF_TYPE_RAW:
> +		if (event->attr.config >= arc_pmu->n_events)
> +			return -ENOENT;
> +
> +		hwc->config |= event->attr.config;
> +		pr_debug("init raw event with idx %lld \'%s\'\n",
> +			 event->attr.config,
> +			 arc_pmu->raw_entry[event->attr.config].name);
> +
> +		return 0;
> +
>  	default:
>  		return -ENOENT;
>  	}
> @@ -446,6 +473,68 @@ static void arc_cpu_pmu_irq_init(void *data)
>  	write_aux_reg(ARC_REG_PCT_INT_ACT, 0xffffffff);
>  }
>  
> +/* Event field occupies the bottom 15 bits of our config field */
> +PMU_FORMAT_ATTR(event, "config:0-14");
> +static struct attribute *arc_pmu_format_attrs[] = {
> +	&format_attr_event.attr,
> +	NULL,
> +};
> +
> +static struct attribute_group arc_pmu_format_attr_gr = {
> +	.name = "format",
> +	.attrs = arc_pmu_format_attrs,
> +};
> +
> +static ssize_t
> +arc_pmu_events_sysfs_show(struct device *dev,
> +			  struct device_attribute *attr, char *page)
> +{
> +	struct perf_pmu_events_attr *pmu_attr;
> +
> +	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
> +	return sprintf(page, "event=0x%04llx\n", pmu_attr->id);
> +}
> +
> +/*
> + * We don't add attrs here as we don't have pre-defined list of perf events.
> + * We will generate and add attrs dynamically in probe() after we read HW
> + * configuration.
> + */
> +static struct attribute_group arc_pmu_events_attr_gr = {
> +	.name = "events",
> +};
> +
> +static void arc_pmu_add_raw_event_attr(int j, char *str)
> +{
> +	memmove(arc_pmu->raw_entry[j].name, str, ARCPMU_EVENT_NAME_LEN - 1);
> +	arc_pmu->attr[j].attr.attr.name = arc_pmu->raw_entry[j].name;
> +	arc_pmu->attr[j].attr.attr.mode = VERIFY_OCTAL_PERMISSIONS(0444);
> +	arc_pmu->attr[j].attr.show = arc_pmu_events_sysfs_show;
> +	arc_pmu->attr[j].id = j;
> +	arc_pmu->attrs[j] = &(arc_pmu->attr[j].attr.attr);
> +}
> +
> +static int arc_pmu_raw_alloc(struct device *dev)
> +{
> +	arc_pmu->attr = devm_kmalloc_array(dev, arc_pmu->n_events + 1,
> +		sizeof(struct perf_pmu_events_attr), GFP_KERNEL | __GFP_ZERO);
> +	if (!arc_pmu->attr)
> +		return -ENOMEM;
> +
> +	arc_pmu->attrs = devm_kmalloc_array(dev, arc_pmu->n_events + 1,
> +		sizeof(*arc_pmu->attrs), GFP_KERNEL | __GFP_ZERO);
> +	if (!arc_pmu->attrs)
> +		return -ENOMEM;
> +
> +	arc_pmu->raw_entry = devm_kmalloc_array(dev, arc_pmu->n_events,
> +		sizeof(struct arc_pmu_raw_event_entry),
> +		GFP_KERNEL | __GFP_ZERO);
> +	if (!arc_pmu->raw_entry)
> +		return -ENOMEM;
> +
> +	return 0;
> +}
> +
>  static int arc_pmu_device_probe(struct platform_device *pdev)
>  {
>  	struct arc_reg_pct_build pct_bcr;
> @@ -477,6 +566,11 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
>  	if (!arc_pmu)
>  		return -ENOMEM;
>  
> +	arc_pmu->n_events = cc_bcr.c;
> +
> +	if (arc_pmu_raw_alloc(&pdev->dev))
> +		return -ENOMEM;
> +
>  	has_interrupts = is_isa_arcv2() ? pct_bcr.i : 0;
>  
>  	arc_pmu->n_counters = pct_bcr.c;
> @@ -508,8 +602,14 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
>  				arc_pmu->ev_hw_idx[i] = j;
>  			}
>  		}
> +
> +		arc_pmu_add_raw_event_attr(j, cc_name.str);
>  	}
>  
> +	arc_pmu_events_attr_gr.attrs = arc_pmu->attrs;
> +	arc_pmu->attr_groups[ARCPMU_ATTR_GR_EVENTS] = &arc_pmu_events_attr_gr;
> +	arc_pmu->attr_groups[ARCPMU_ATTR_GR_FORMATS] = &arc_pmu_format_attr_gr;
> +
>  	arc_pmu->pmu = (struct pmu) {
>  		.pmu_enable	= arc_pmu_enable,
>  		.pmu_disable	= arc_pmu_disable,
> @@ -519,6 +619,7 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
>  		.start		= arc_pmu_start,
>  		.stop		= arc_pmu_stop,
>  		.read		= arc_pmu_read,
> +		.attr_groups	= arc_pmu->attr_groups,
>  	};
>  
>  	if (has_interrupts) {
> @@ -540,7 +641,11 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
>  	} else
>  		arc_pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
>  
> -	return perf_pmu_register(&arc_pmu->pmu, pdev->name, PERF_TYPE_RAW);
> +	/*
> +	 * perf parser doesn't really like '-' symbol in events name, so let's
> +	 * use '_' in arc pmu name as it goes to kernel PMU event prefix.
> +	 */
> +	return perf_pmu_register(&arc_pmu->pmu, "arc_pmu", PERF_TYPE_RAW);
>  }
>  
>  #ifdef CONFIG_OF

Re: [PATCH 2/5] ARC: perf: introduce Kernel PMU events support

[Vineet Gupta]

On 12/5/18 9:06 AM, Eugeniy Paltsev wrote:
> Export all available ARC architected hardware events as
> kernel PMU events to make non-generic events accessible.
> 
> ARC PMU HW allow us to read the list of all available
> events names. So we generate kernel PMU event list
> dynamically in arc_pmu_device_probe() using
> human-readable events names we got from HW instead of
> using pre-defined events list.
> 
> -------------------------->8--------------------------
> $ perf list
>   [snip]
>   arc_pmu/bdata64/                  [Kernel PMU event]
>   arc_pmu/bdcstall/                 [Kernel PMU event]
>   arc_pmu/bdslot/                   [Kernel PMU event]
>   arc_pmu/bfbmp/                    [Kernel PMU event]
>   arc_pmu/bfirqex/                  [Kernel PMU event]
>   arc_pmu/bflgstal/                 [Kernel PMU event]
>   arc_pmu/bflush/                   [Kernel PMU event]
> -------------------------->8--------------------------


@Peter do you have any comments on this patch. I'd really like to have this
upstream for next release, so any thoughts you have are more than welcome.

-Vineet

> Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>

> ---
>  arch/arc/kernel/perf_event.c | 107 ++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 106 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
> index 811a07a2ca21..97b88b00c418 100644
> --- a/arch/arc/kernel/perf_event.c
> +++ b/arch/arc/kernel/perf_event.c
> @@ -22,12 +22,28 @@
>  /* HW holds 8 symbols + one for null terminator */
>  #define ARCPMU_EVENT_NAME_LEN	9
>  
> +enum arc_pmu_attr_groups {
> +	ARCPMU_ATTR_GR_EVENTS,
> +	ARCPMU_ATTR_GR_FORMATS,
> +	ARCPMU_NR_ATTR_GR
> +};
> +
> +struct arc_pmu_raw_event_entry {
> +	char name[ARCPMU_EVENT_NAME_LEN];
> +};
> +
>  struct arc_pmu {
>  	struct pmu	pmu;
>  	unsigned int	irq;
>  	int		n_counters;
> +	int		n_events;
>  	u64		max_period;
>  	int		ev_hw_idx[PERF_COUNT_ARC_HW_MAX];
> +
> +	struct arc_pmu_raw_event_entry	*raw_entry;
> +	struct attribute		**attrs;
> +	struct perf_pmu_events_attr	*attr;
> +	const struct attribute_group	*attr_groups[ARCPMU_NR_ATTR_GR + 1];
>  };
>  
>  struct arc_pmu_cpu {
> @@ -196,6 +212,17 @@ static int arc_pmu_event_init(struct perf_event *event)
>  			 (int)hwc->config, arc_pmu_ev_hw_map[ret]);
>  		return 0;
>  
> +	case PERF_TYPE_RAW:
> +		if (event->attr.config >= arc_pmu->n_events)
> +			return -ENOENT;
> +
> +		hwc->config |= event->attr.config;
> +		pr_debug("init raw event with idx %lld \'%s\'\n",
> +			 event->attr.config,
> +			 arc_pmu->raw_entry[event->attr.config].name);
> +
> +		return 0;
> +
>  	default:
>  		return -ENOENT;
>  	}
> @@ -446,6 +473,68 @@ static void arc_cpu_pmu_irq_init(void *data)
>  	write_aux_reg(ARC_REG_PCT_INT_ACT, 0xffffffff);
>  }
>  
> +/* Event field occupies the bottom 15 bits of our config field */
> +PMU_FORMAT_ATTR(event, "config:0-14");
> +static struct attribute *arc_pmu_format_attrs[] = {
> +	&format_attr_event.attr,
> +	NULL,
> +};
> +
> +static struct attribute_group arc_pmu_format_attr_gr = {
> +	.name = "format",
> +	.attrs = arc_pmu_format_attrs,
> +};
> +
> +static ssize_t
> +arc_pmu_events_sysfs_show(struct device *dev,
> +			  struct device_attribute *attr, char *page)
> +{
> +	struct perf_pmu_events_attr *pmu_attr;
> +
> +	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
> +	return sprintf(page, "event=0x%04llx\n", pmu_attr->id);
> +}
> +
> +/*
> + * We don't add attrs here as we don't have pre-defined list of perf events.
> + * We will generate and add attrs dynamically in probe() after we read HW
> + * configuration.
> + */
> +static struct attribute_group arc_pmu_events_attr_gr = {
> +	.name = "events",
> +};
> +
> +static void arc_pmu_add_raw_event_attr(int j, char *str)
> +{
> +	memmove(arc_pmu->raw_entry[j].name, str, ARCPMU_EVENT_NAME_LEN - 1);
> +	arc_pmu->attr[j].attr.attr.name = arc_pmu->raw_entry[j].name;
> +	arc_pmu->attr[j].attr.attr.mode = VERIFY_OCTAL_PERMISSIONS(0444);
> +	arc_pmu->attr[j].attr.show = arc_pmu_events_sysfs_show;
> +	arc_pmu->attr[j].id = j;
> +	arc_pmu->attrs[j] = &(arc_pmu->attr[j].attr.attr);
> +}
> +
> +static int arc_pmu_raw_alloc(struct device *dev)
> +{
> +	arc_pmu->attr = devm_kmalloc_array(dev, arc_pmu->n_events + 1,
> +		sizeof(struct perf_pmu_events_attr), GFP_KERNEL | __GFP_ZERO);
> +	if (!arc_pmu->attr)
> +		return -ENOMEM;
> +
> +	arc_pmu->attrs = devm_kmalloc_array(dev, arc_pmu->n_events + 1,
> +		sizeof(*arc_pmu->attrs), GFP_KERNEL | __GFP_ZERO);
> +	if (!arc_pmu->attrs)
> +		return -ENOMEM;
> +
> +	arc_pmu->raw_entry = devm_kmalloc_array(dev, arc_pmu->n_events,
> +		sizeof(struct arc_pmu_raw_event_entry),
> +		GFP_KERNEL | __GFP_ZERO);
> +	if (!arc_pmu->raw_entry)
> +		return -ENOMEM;
> +
> +	return 0;
> +}
> +
>  static int arc_pmu_device_probe(struct platform_device *pdev)
>  {
>  	struct arc_reg_pct_build pct_bcr;
> @@ -477,6 +566,11 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
>  	if (!arc_pmu)
>  		return -ENOMEM;
>  
> +	arc_pmu->n_events = cc_bcr.c;
> +
> +	if (arc_pmu_raw_alloc(&pdev->dev))
> +		return -ENOMEM;
> +
>  	has_interrupts = is_isa_arcv2() ? pct_bcr.i : 0;
>  
>  	arc_pmu->n_counters = pct_bcr.c;
> @@ -508,8 +602,14 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
>  				arc_pmu->ev_hw_idx[i] = j;
>  			}
>  		}
> +
> +		arc_pmu_add_raw_event_attr(j, cc_name.str);
>  	}
>  
> +	arc_pmu_events_attr_gr.attrs = arc_pmu->attrs;
> +	arc_pmu->attr_groups[ARCPMU_ATTR_GR_EVENTS] = &arc_pmu_events_attr_gr;
> +	arc_pmu->attr_groups[ARCPMU_ATTR_GR_FORMATS] = &arc_pmu_format_attr_gr;
> +
>  	arc_pmu->pmu = (struct pmu) {
>  		.pmu_enable	= arc_pmu_enable,
>  		.pmu_disable	= arc_pmu_disable,
> @@ -519,6 +619,7 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
>  		.start		= arc_pmu_start,
>  		.stop		= arc_pmu_stop,
>  		.read		= arc_pmu_read,
> +		.attr_groups	= arc_pmu->attr_groups,
>  	};
>  
>  	if (has_interrupts) {
> @@ -540,7 +641,11 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
>  	} else
>  		arc_pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
>  
> -	return perf_pmu_register(&arc_pmu->pmu, pdev->name, PERF_TYPE_RAW);
> +	/*
> +	 * perf parser doesn't really like '-' symbol in events name, so let's
> +	 * use '_' in arc pmu name as it goes to kernel PMU event prefix.
> +	 */
> +	return perf_pmu_register(&arc_pmu->pmu, "arc_pmu", PERF_TYPE_RAW);
>  }
>  
>  #ifdef CONFIG_OF
> 

[PATCH 3/5] ARC: perf: fix of kernel data types using

[Eugeniy Paltsev]

Use u32, u64, s64 instead of uint32_t, uint64_t, int64_t

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
---
 arch/arc/kernel/perf_event.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
index 97b88b00c418..8c88837fcd83 100644
--- a/arch/arc/kernel/perf_event.c
+++ b/arch/arc/kernel/perf_event.c
@@ -101,10 +101,10 @@ static struct arc_pmu *arc_pmu;
 static DEFINE_PER_CPU(struct arc_pmu_cpu, arc_pmu_cpu);
 
 /* read counter #idx; note that counter# != event# on ARC! */
-static uint64_t arc_pmu_read_counter(int idx)
+static u64 arc_pmu_read_counter(int idx)
 {
-	uint32_t tmp;
-	uint64_t result;
+	u32 tmp;
+	u64 result;
 
 	/*
 	 * ARC supports making 'snapshots' of the counters, so we don't
@@ -113,7 +113,7 @@ static uint64_t arc_pmu_read_counter(int idx)
 	write_aux_reg(ARC_REG_PCT_INDEX, idx);
 	tmp = read_aux_reg(ARC_REG_PCT_CONTROL);
 	write_aux_reg(ARC_REG_PCT_CONTROL, tmp | ARC_REG_PCT_CONTROL_SN);
-	result = (uint64_t) (read_aux_reg(ARC_REG_PCT_SNAPH)) << 32;
+	result = (u64) (read_aux_reg(ARC_REG_PCT_SNAPH)) << 32;
 	result |= read_aux_reg(ARC_REG_PCT_SNAPL);
 
 	return result;
@@ -122,9 +122,9 @@ static uint64_t arc_pmu_read_counter(int idx)
 static void arc_perf_event_update(struct perf_event *event,
 				  struct hw_perf_event *hwc, int idx)
 {
-	uint64_t prev_raw_count = local64_read(&hwc->prev_count);
-	uint64_t new_raw_count = arc_pmu_read_counter(idx);
-	int64_t delta = new_raw_count - prev_raw_count;
+	u64 prev_raw_count = local64_read(&hwc->prev_count);
+	u64 new_raw_count = arc_pmu_read_counter(idx);
+	s64 delta = new_raw_count - prev_raw_count;
 
 	/*
 	 * We aren't afraid of hwc->prev_count changing beneath our feet
@@ -231,7 +231,7 @@ static int arc_pmu_event_init(struct perf_event *event)
 /* starts all counters */
 static void arc_pmu_enable(struct pmu *pmu)
 {
-	uint32_t tmp;
+	u32 tmp;
 	tmp = read_aux_reg(ARC_REG_PCT_CONTROL);
 	write_aux_reg(ARC_REG_PCT_CONTROL, (tmp & 0xffff0000) | 0x1);
 }
@@ -239,7 +239,7 @@ static void arc_pmu_enable(struct pmu *pmu)
 /* stops all counters */
 static void arc_pmu_disable(struct pmu *pmu)
 {
-	uint32_t tmp;
+	u32 tmp;
 	tmp = read_aux_reg(ARC_REG_PCT_CONTROL);
 	write_aux_reg(ARC_REG_PCT_CONTROL, (tmp & 0xffff0000) | 0x0);
 }
@@ -544,7 +544,7 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
 
 	union cc_name {
 		struct {
-			uint32_t word0, word1;
+			u32 word0, word1;
 			char sentinel;
 		} indiv;
 		char str[ARCPMU_EVENT_NAME_LEN];
-- 
2.14.5

Re: [PATCH 3/5] ARC: perf: fix of kernel data types using

[Vineet Gupta]

On 12/5/18 9:06 AM, Eugeniy Paltsev wrote:
> Use u32, u64, s64 instead of uint32_t, uint64_t, int64_t
>
> Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>


Maybe squash this into trivial cleanups 1/5 ?

Acked-by: Vineet Gupta <vgupta@synopsys.com>

Thx,
-Vineet

[PATCH 4/5] ARC: perf: fix description comment

[Eugeniy Paltsev]

Fix description comment as this code doesn't belong only to
ARC700 anymore.

Also while I'm at it, use SPDX License Identifier.

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
---
 arch/arc/kernel/perf_event.c | 19 +++++++------------
 1 file changed, 7 insertions(+), 12 deletions(-)

diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
index 8c88837fcd83..b6207090aaa7 100644
--- a/arch/arc/kernel/perf_event.c
+++ b/arch/arc/kernel/perf_event.c
@@ -1,15 +1,10 @@
-/*
- * Linux performance counter support for ARC700 series
- *
- * Copyright (C) 2013-2015 Synopsys, Inc. (www.synopsys.com)
- *
- * This code is inspired by the perf support of various other architectures.
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License version 2 as
- * published by the Free Software Foundation.
- *
- */
+// SPDX-License-Identifier: GPL-2.0+
+//
+// Linux performance counter support for ARC CPUs.
+// This code is inspired by the perf support of various other architectures.
+//
+// Copyright (C) 2013-2018 Synopsys, Inc. (www.synopsys.com)
+
 #include <linux/errno.h>
 #include <linux/interrupt.h>
 #include <linux/module.h>
-- 
2.14.5

Re: [PATCH 4/5] ARC: perf: fix description comment

[Vineet Gupta]

On 12/5/18 9:06 AM, Eugeniy Paltsev wrote:
> Fix description comment as this code doesn't belong only to
> ARC700 anymore.
>
> Also while I'm at it, use SPDX License Identifier.
>
> Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>

Maybe squash this into trivial cleanups 1/5 ?

Acked-by: Vineet Gupta <vgupta@synopsys.com>

Thx,
-Vineet

Re: [PATCH 4/5] ARC: perf: fix description comment

[Eugeniy Paltsev]

On Wed, 2018-12-05 at 17:13 +0000, Vineet Gupta wrote:
> On 12/5/18 9:06 AM, Eugeniy Paltsev wrote:
> > Fix description comment as this code doesn't belong only to
> > ARC700 anymore.
> > 
> > Also while I'm at it, use SPDX License Identifier.
> > 
> > Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
> 
> Maybe squash this into trivial cleanups 1/5 ?

I split cleanup for several patches to make review easier.
I can squash them into single patch in v2 if you want it.


> Acked-by: Vineet Gupta <vgupta@synopsys.com>
> 
> Thx,
> -Vineet
-- 
 Eugeniy Paltsev

Re: [PATCH 4/5] ARC: perf: fix description comment

[Vineet Gupta]

On 12/6/18 7:49 AM, Eugeniy Paltsev wrote:
>> Maybe squash this into trivial cleanups 1/5 ?
> I split cleanup for several patches to make review easier.
> I can squash them into single patch in v2 if you want it.
>
>

Understand, but they are all trivial anyways so yeah better to squash - I can do
it here as well. Lets wait for PeterZ to comment on 2/5 before v2.

-Vineet

[PATCH 5/5] ARC: perf: remove useless ifdefs

[Eugeniy Paltsev]

ifdef around 'arc_pmu_match' structure declaration is useless
as we refer to 'arc_pmu_match' in several places which aren't
guarded with ifdef.

Nevertheless 'ARC' option selects 'OF' unconditionally so we
can simply get rid of this ifdef.

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
---
 arch/arc/kernel/perf_event.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
index b6207090aaa7..7786c1f005fb 100644
--- a/arch/arc/kernel/perf_event.c
+++ b/arch/arc/kernel/perf_event.c
@@ -643,14 +643,12 @@ static int arc_pmu_device_probe(struct platform_device *pdev)
 	return perf_pmu_register(&arc_pmu->pmu, "arc_pmu", PERF_TYPE_RAW);
 }
 
-#ifdef CONFIG_OF
 static const struct of_device_id arc_pmu_match[] = {
 	{ .compatible = "snps,arc700-pct" },
 	{ .compatible = "snps,archs-pct" },
 	{},
 };
 MODULE_DEVICE_TABLE(of, arc_pmu_match);
-#endif
 
 static struct platform_driver arc_pmu_driver = {
 	.driver	= {
-- 
2.14.5

Re: [PATCH 5/5] ARC: perf: remove useless ifdefs

[Vineet Gupta]

On 12/5/18 9:06 AM, Eugeniy Paltsev wrote:
> ifdef around 'arc_pmu_match' structure declaration is useless
> as we refer to 'arc_pmu_match' in several places which aren't
> guarded with ifdef.
>
> Nevertheless 'ARC' option selects 'OF' unconditionally so we
> can simply get rid of this ifdef.
>
> Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>


Maybe squash this into trivial cleanups 1/5 ?

Acked-by: Vineet Gupta <vgupta@synopsys.com>

Thx,
-Vineet