[PATCH 0/9] General updates and two new drivers for HiSilicon Uncore PMU

[PATCH 0/9] General updates and two new drivers for HiSilicon Uncore PMU

[Yicong Yang]

From: Yicong Yang <yangyicong@hisilicon.com>

Support new version of DDRC/SLLC PMU identified with updated ACPI HID and
register definition. In order to support this, we do a preliminary refactor
to initialize device of each version by using driver data of each HID
rather than checking the version. This will also make the driver easier to
maintain and extend, since only the HID specific information along
with the new HID will be added to support the new version without touching
the common logic.

Two new Uncore PMU drivers is also added to support the monitoring the
events of the system bus (by NoC PMU) and the DVM opertaions (by MN PMU).

Junhao He (7):
  drivers/perf: hisi: Extend struct hisi_pmu_dev_info
  drivers/perf: hisi: Simplify the probe process for each DDRC version
  drivers/perf: hisi: Add support for HiSilicon DDRC v3 PMU driver
  drivers/perf: hisi: Use ACPI driver_data to retrieve SLLC PMU
    information
  drivers/perf: hisi: Add support for HiSilicon SLLC v3 PMU driver
  drivers/perf: hisi: Relax the event number check of v2 PMUs
  drivers/perf: hisi: Add support for HiSilicon MN PMU driver

Yicong Yang (2):
  drivers/perf: hisi: Support PMUs with no interrupt
  drivers/perf: hisi: Add support for HiSilicon NoC PMU

 Documentation/admin-guide/perf/hisi-pmu.rst   |  11 +
 drivers/perf/hisilicon/Makefile               |   3 +-
 drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c | 354 ++++++++--------
 drivers/perf/hisilicon/hisi_uncore_hha_pmu.c  |   7 +-
 drivers/perf/hisilicon/hisi_uncore_mn_pmu.c   | 355 ++++++++++++++++
 drivers/perf/hisilicon/hisi_uncore_noc_pmu.c  | 391 ++++++++++++++++++
 drivers/perf/hisilicon/hisi_uncore_pa_pmu.c   |   2 +-
 drivers/perf/hisilicon/hisi_uncore_pmu.c      |  11 +-
 drivers/perf/hisilicon/hisi_uncore_pmu.h      |   2 +
 drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 234 ++++++++---
 10 files changed, 1104 insertions(+), 266 deletions(-)
 create mode 100644 drivers/perf/hisilicon/hisi_uncore_mn_pmu.c
 create mode 100644 drivers/perf/hisilicon/hisi_uncore_noc_pmu.c

-- 
2.24.0

[PATCH 1/9] drivers/perf: hisi: Extend struct hisi_pmu_dev_info

[Yicong Yang]

From: Junhao He <hejunhao3@huawei.com>

The counter bits and event range may differ from versions of a
certain uncore PMU. Make this device specific information into
struct hisi_pmu_dev_info. This will help to simplify the
initialization process by using a list of struct hisi_pmu_dev_info
rather than checking the version.

Signed-off-by: Junhao He <hejunhao3@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 drivers/perf/hisilicon/hisi_uncore_pmu.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/perf/hisilicon/hisi_uncore_pmu.h b/drivers/perf/hisilicon/hisi_uncore_pmu.h
index f4fed2544877..777675838b80 100644
--- a/drivers/perf/hisilicon/hisi_uncore_pmu.h
+++ b/drivers/perf/hisilicon/hisi_uncore_pmu.h
@@ -72,6 +72,8 @@ struct hisi_uncore_ops {
 struct hisi_pmu_dev_info {
 	const char *name;
 	const struct attribute_group **attr_groups;
+	u32 counter_bits;
+	u32 check_event;
 	void *private;
 };
 
-- 
2.24.0

Re: [PATCH 1/9] drivers/perf: hisi: Extend struct hisi_pmu_dev_info

[Will Deacon]

On Tue, Feb 18, 2025 at 05:19:52PM +0800, Yicong Yang wrote:
> From: Junhao He <hejunhao3@huawei.com>
> 
> The counter bits and event range may differ from versions of a
> certain uncore PMU. Make this device specific information into
> struct hisi_pmu_dev_info. This will help to simplify the
> initialization process by using a list of struct hisi_pmu_dev_info
> rather than checking the version.
> 
> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
> ---
>  drivers/perf/hisilicon/hisi_uncore_pmu.h | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/perf/hisilicon/hisi_uncore_pmu.h b/drivers/perf/hisilicon/hisi_uncore_pmu.h
> index f4fed2544877..777675838b80 100644
> --- a/drivers/perf/hisilicon/hisi_uncore_pmu.h
> +++ b/drivers/perf/hisilicon/hisi_uncore_pmu.h
> @@ -72,6 +72,8 @@ struct hisi_uncore_ops {
>  struct hisi_pmu_dev_info {
>  	const char *name;
>  	const struct attribute_group **attr_groups;
> +	u32 counter_bits;
> +	u32 check_event;
>  	void *private;
>  };

This doesn't really do anything on its own; please just fold it into the
patch where the new fields are used.

Will

Re: [PATCH 1/9] drivers/perf: hisi: Extend struct hisi_pmu_dev_info

[Yicong Yang]

On 2025/3/1 14:23, Will Deacon wrote:
> On Tue, Feb 18, 2025 at 05:19:52PM +0800, Yicong Yang wrote:
>> From: Junhao He <hejunhao3@huawei.com>
>>
>> The counter bits and event range may differ from versions of a
>> certain uncore PMU. Make this device specific information into
>> struct hisi_pmu_dev_info. This will help to simplify the
>> initialization process by using a list of struct hisi_pmu_dev_info
>> rather than checking the version.
>>
>> Signed-off-by: Junhao He <hejunhao3@huawei.com>
>> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
>> ---
>>  drivers/perf/hisilicon/hisi_uncore_pmu.h | 2 ++
>>  1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/perf/hisilicon/hisi_uncore_pmu.h b/drivers/perf/hisilicon/hisi_uncore_pmu.h
>> index f4fed2544877..777675838b80 100644
>> --- a/drivers/perf/hisilicon/hisi_uncore_pmu.h
>> +++ b/drivers/perf/hisilicon/hisi_uncore_pmu.h
>> @@ -72,6 +72,8 @@ struct hisi_uncore_ops {
>>  struct hisi_pmu_dev_info {
>>  	const char *name;
>>  	const struct attribute_group **attr_groups;
>> +	u32 counter_bits;
>> +	u32 check_event;
>>  	void *private;
>>  };
> 
> This doesn't really do anything on its own; please just fold it into the
> patch where the new fields are used.
> 

sure, will make this into the following patch.

Thanks.

Re: [PATCH 1/9] drivers/perf: hisi: Extend struct hisi_pmu_dev_info

[Jonathan Cameron]

On Mon, 3 Mar 2025 22:43:35 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> On 2025/3/1 14:23, Will Deacon wrote:
> > On Tue, Feb 18, 2025 at 05:19:52PM +0800, Yicong Yang wrote:  
> >> From: Junhao He <hejunhao3@huawei.com>
> >>
> >> The counter bits and event range may differ from versions of a
> >> certain uncore PMU. Make this device specific information into
> >> struct hisi_pmu_dev_info. This will help to simplify the
> >> initialization process by using a list of struct hisi_pmu_dev_info
> >> rather than checking the version.
> >>
> >> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> >> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
> >> ---
> >>  drivers/perf/hisilicon/hisi_uncore_pmu.h | 2 ++
> >>  1 file changed, 2 insertions(+)
> >>
> >> diff --git a/drivers/perf/hisilicon/hisi_uncore_pmu.h b/drivers/perf/hisilicon/hisi_uncore_pmu.h
> >> index f4fed2544877..777675838b80 100644
> >> --- a/drivers/perf/hisilicon/hisi_uncore_pmu.h
> >> +++ b/drivers/perf/hisilicon/hisi_uncore_pmu.h
> >> @@ -72,6 +72,8 @@ struct hisi_uncore_ops {
> >>  struct hisi_pmu_dev_info {
> >>  	const char *name;
> >>  	const struct attribute_group **attr_groups;
> >> +	u32 counter_bits;
> >> +	u32 check_event;
This name is a little confusing.  Maybe nr_events or similar?


> >>  	void *private;
> >>  };  
> > 
> > This doesn't really do anything on its own; please just fold it into the
> > patch where the new fields are used.
> >   
> 
> sure, will make this into the following patch.
> 
> Thanks.
> 

[PATCH 2/9] drivers/perf: hisi: Simplify the probe process for each DDRC version

[Yicong Yang]

From: Junhao He <hejunhao3@huawei.com>

Version 1 and 2 of DDRC PMU also use different HID. Make use of
struct acpi_device_id::driver_data for version specific information
rather than judge the version register. This will help to
simplify the probe process and also a bit easier for extension.

Signed-off-by: Junhao He <hejunhao3@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c | 328 ++++++++----------
 1 file changed, 140 insertions(+), 188 deletions(-)

diff --git a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
index 7e490f8868f2..7e3c2436e96b 100644
--- a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
@@ -50,6 +50,10 @@
 #define DDRC_V1_NR_EVENTS	0x7
 #define DDRC_V2_NR_EVENTS	0x90
 
+#define DDRC_EVENT_CNTn(base, n)	((base) + (n) * 8)
+#define DDRC_EVENT_TYPEn(base, n)	((base) + (n) * 4)
+#define DDRC_UNIMPLEMENTED_REG		GENMASK(31, 0)
+
 /*
  * For PMU v1, there are eight-events and every event has been mapped
  * to fixed-purpose counters which register offset is not consistent.
@@ -63,47 +67,37 @@ static const u32 ddrc_reg_off[] = {
 	DDRC_PRE_CMD, DDRC_ACT_CMD, DDRC_RNK_CHG, DDRC_RW_CHG
 };
 
-/*
- * Select the counter register offset using the counter index.
- * In PMU v1, there are no programmable counter, the count
- * is read form the statistics counter register itself.
- */
-static u32 hisi_ddrc_pmu_v1_get_counter_offset(int cntr_idx)
-{
-	return ddrc_reg_off[cntr_idx];
-}
-
-static u32 hisi_ddrc_pmu_v2_get_counter_offset(int cntr_idx)
-{
-	return DDRC_V2_EVENT_CNT + cntr_idx * 8;
-}
+struct hisi_ddrc_pmu_regs {
+	u32 event_cnt;
+	u32 event_ctrl;
+	u32 event_type;
+	u32 perf_ctrl;
+	u32 perf_ctrl_en;
+	u32 int_mask;
+	u32 int_clear;
+	u32 int_status;
+};
 
-static u64 hisi_ddrc_pmu_v1_read_counter(struct hisi_pmu *ddrc_pmu,
+static u64 hisi_ddrc_pmu_read_counter(struct hisi_pmu *ddrc_pmu,
 				      struct hw_perf_event *hwc)
 {
-	return readl(ddrc_pmu->base +
-		     hisi_ddrc_pmu_v1_get_counter_offset(hwc->idx));
-}
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 
-static void hisi_ddrc_pmu_v1_write_counter(struct hisi_pmu *ddrc_pmu,
-					struct hw_perf_event *hwc, u64 val)
-{
-	writel((u32)val,
-	       ddrc_pmu->base + hisi_ddrc_pmu_v1_get_counter_offset(hwc->idx));
-}
+	if (regs->event_cnt == DDRC_UNIMPLEMENTED_REG)
+		return readl(ddrc_pmu->base + ddrc_reg_off[hwc->idx]);
 
-static u64 hisi_ddrc_pmu_v2_read_counter(struct hisi_pmu *ddrc_pmu,
-					 struct hw_perf_event *hwc)
-{
-	return readq(ddrc_pmu->base +
-		     hisi_ddrc_pmu_v2_get_counter_offset(hwc->idx));
+	return readq(ddrc_pmu->base + DDRC_EVENT_CNTn(regs->event_cnt, hwc->idx));
 }
 
-static void hisi_ddrc_pmu_v2_write_counter(struct hisi_pmu *ddrc_pmu,
-					   struct hw_perf_event *hwc, u64 val)
+static void hisi_ddrc_pmu_write_counter(struct hisi_pmu *ddrc_pmu,
+					struct hw_perf_event *hwc, u64 val)
 {
-	writeq(val,
-	       ddrc_pmu->base + hisi_ddrc_pmu_v2_get_counter_offset(hwc->idx));
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
+
+	if (regs->event_cnt == DDRC_UNIMPLEMENTED_REG)
+		writel((u32)val, ddrc_pmu->base + ddrc_reg_off[hwc->idx]);
+	else
+		writeq(val, ddrc_pmu->base + DDRC_EVENT_CNTn(regs->event_cnt, hwc->idx));
 }
 
 /*
@@ -114,54 +108,12 @@ static void hisi_ddrc_pmu_v2_write_counter(struct hisi_pmu *ddrc_pmu,
 static void hisi_ddrc_pmu_write_evtype(struct hisi_pmu *ddrc_pmu, int idx,
 				       u32 type)
 {
-	u32 offset;
-
-	if (ddrc_pmu->identifier >= HISI_PMU_V2) {
-		offset = DDRC_V2_EVENT_TYPE + 4 * idx;
-		writel(type, ddrc_pmu->base + offset);
-	}
-}
-
-static void hisi_ddrc_pmu_v1_start_counters(struct hisi_pmu *ddrc_pmu)
-{
-	u32 val;
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 
-	/* Set perf_enable in DDRC_PERF_CTRL to start event counting */
-	val = readl(ddrc_pmu->base + DDRC_PERF_CTRL);
-	val |= DDRC_V1_PERF_CTRL_EN;
-	writel(val, ddrc_pmu->base + DDRC_PERF_CTRL);
-}
+	if (regs->event_type == DDRC_UNIMPLEMENTED_REG)
+		return;
 
-static void hisi_ddrc_pmu_v1_stop_counters(struct hisi_pmu *ddrc_pmu)
-{
-	u32 val;
-
-	/* Clear perf_enable in DDRC_PERF_CTRL to stop event counting */
-	val = readl(ddrc_pmu->base + DDRC_PERF_CTRL);
-	val &= ~DDRC_V1_PERF_CTRL_EN;
-	writel(val, ddrc_pmu->base + DDRC_PERF_CTRL);
-}
-
-static void hisi_ddrc_pmu_v1_enable_counter(struct hisi_pmu *ddrc_pmu,
-					    struct hw_perf_event *hwc)
-{
-	u32 val;
-
-	/* Set counter index(event code) in DDRC_EVENT_CTRL register */
-	val = readl(ddrc_pmu->base + DDRC_EVENT_CTRL);
-	val |= (1 << GET_DDRC_EVENTID(hwc));
-	writel(val, ddrc_pmu->base + DDRC_EVENT_CTRL);
-}
-
-static void hisi_ddrc_pmu_v1_disable_counter(struct hisi_pmu *ddrc_pmu,
-					     struct hw_perf_event *hwc)
-{
-	u32 val;
-
-	/* Clear counter index(event code) in DDRC_EVENT_CTRL register */
-	val = readl(ddrc_pmu->base + DDRC_EVENT_CTRL);
-	val &= ~(1 << GET_DDRC_EVENTID(hwc));
-	writel(val, ddrc_pmu->base + DDRC_EVENT_CTRL);
+	writel(type, ddrc_pmu->base + DDRC_EVENT_TYPEn(regs->event_type, idx));
 }
 
 static int hisi_ddrc_pmu_v1_get_event_idx(struct perf_event *event)
@@ -180,120 +132,96 @@ static int hisi_ddrc_pmu_v1_get_event_idx(struct perf_event *event)
 	return idx;
 }
 
-static int hisi_ddrc_pmu_v2_get_event_idx(struct perf_event *event)
+static int hisi_ddrc_pmu_get_event_idx(struct perf_event *event)
 {
+	struct hisi_pmu *ddrc_pmu = to_hisi_pmu(event->pmu);
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
+
+	if (regs->event_type == DDRC_UNIMPLEMENTED_REG)
+		return hisi_ddrc_pmu_v1_get_event_idx(event);
+
 	return hisi_uncore_pmu_get_event_idx(event);
 }
 
-static void hisi_ddrc_pmu_v2_start_counters(struct hisi_pmu *ddrc_pmu)
+static void hisi_ddrc_pmu_start_counters(struct hisi_pmu *ddrc_pmu)
 {
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(ddrc_pmu->base + DDRC_V2_PERF_CTRL);
-	val |= DDRC_V2_PERF_CTRL_EN;
-	writel(val, ddrc_pmu->base + DDRC_V2_PERF_CTRL);
+	val = readl(ddrc_pmu->base + regs->perf_ctrl);
+	val |= regs->perf_ctrl_en;
+	writel(val, ddrc_pmu->base + regs->perf_ctrl);
 }
 
-static void hisi_ddrc_pmu_v2_stop_counters(struct hisi_pmu *ddrc_pmu)
+static void hisi_ddrc_pmu_stop_counters(struct hisi_pmu *ddrc_pmu)
 {
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(ddrc_pmu->base + DDRC_V2_PERF_CTRL);
-	val &= ~DDRC_V2_PERF_CTRL_EN;
-	writel(val, ddrc_pmu->base + DDRC_V2_PERF_CTRL);
+	val = readl(ddrc_pmu->base + regs->perf_ctrl);
+	val &= ~regs->perf_ctrl_en;
+	writel(val, ddrc_pmu->base + regs->perf_ctrl);
 }
 
-static void hisi_ddrc_pmu_v2_enable_counter(struct hisi_pmu *ddrc_pmu,
+static void hisi_ddrc_pmu_enable_counter(struct hisi_pmu *ddrc_pmu,
 					    struct hw_perf_event *hwc)
 {
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(ddrc_pmu->base + DDRC_V2_EVENT_CTRL);
-	val |= 1 << hwc->idx;
-	writel(val, ddrc_pmu->base + DDRC_V2_EVENT_CTRL);
+	val = readl(ddrc_pmu->base + regs->event_ctrl);
+	val |= BIT_ULL(hwc->idx);
+	writel(val, ddrc_pmu->base + regs->event_ctrl);
 }
 
-static void hisi_ddrc_pmu_v2_disable_counter(struct hisi_pmu *ddrc_pmu,
+static void hisi_ddrc_pmu_disable_counter(struct hisi_pmu *ddrc_pmu,
 					     struct hw_perf_event *hwc)
 {
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(ddrc_pmu->base + DDRC_V2_EVENT_CTRL);
-	val &= ~(1 << hwc->idx);
-	writel(val, ddrc_pmu->base + DDRC_V2_EVENT_CTRL);
-}
-
-static void hisi_ddrc_pmu_v1_enable_counter_int(struct hisi_pmu *ddrc_pmu,
-						struct hw_perf_event *hwc)
-{
-	u32 val;
-
-	/* Write 0 to enable interrupt */
-	val = readl(ddrc_pmu->base + DDRC_INT_MASK);
-	val &= ~(1 << hwc->idx);
-	writel(val, ddrc_pmu->base + DDRC_INT_MASK);
-}
-
-static void hisi_ddrc_pmu_v1_disable_counter_int(struct hisi_pmu *ddrc_pmu,
-						 struct hw_perf_event *hwc)
-{
-	u32 val;
-
-	/* Write 1 to mask interrupt */
-	val = readl(ddrc_pmu->base + DDRC_INT_MASK);
-	val |= 1 << hwc->idx;
-	writel(val, ddrc_pmu->base + DDRC_INT_MASK);
+	val = readl(ddrc_pmu->base + regs->event_ctrl);
+	val &= ~BIT_ULL(hwc->idx);
+	writel(val, ddrc_pmu->base + regs->event_ctrl);
 }
 
-static void hisi_ddrc_pmu_v2_enable_counter_int(struct hisi_pmu *ddrc_pmu,
-						struct hw_perf_event *hwc)
+static void hisi_ddrc_pmu_enable_counter_int(struct hisi_pmu *ddrc_pmu,
+					     struct hw_perf_event *hwc)
 {
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(ddrc_pmu->base + DDRC_V2_INT_MASK);
-	val &= ~(1 << hwc->idx);
-	writel(val, ddrc_pmu->base + DDRC_V2_INT_MASK);
+	val = readl(ddrc_pmu->base + regs->int_mask);
+	val &= ~BIT_ULL(hwc->idx);
+	writel(val, ddrc_pmu->base + regs->int_mask);
 }
 
-static void hisi_ddrc_pmu_v2_disable_counter_int(struct hisi_pmu *ddrc_pmu,
-						struct hw_perf_event *hwc)
+static void hisi_ddrc_pmu_disable_counter_int(struct hisi_pmu *ddrc_pmu,
+					      struct hw_perf_event *hwc)
 {
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(ddrc_pmu->base + DDRC_V2_INT_MASK);
-	val |= 1 << hwc->idx;
-	writel(val, ddrc_pmu->base + DDRC_V2_INT_MASK);
+	val = readl(ddrc_pmu->base + regs->int_mask);
+	val |= BIT_ULL(hwc->idx);
+	writel(val, ddrc_pmu->base + regs->int_mask);
 }
 
-static u32 hisi_ddrc_pmu_v1_get_int_status(struct hisi_pmu *ddrc_pmu)
+static u32 hisi_ddrc_pmu_get_int_status(struct hisi_pmu *ddrc_pmu)
 {
-	return readl(ddrc_pmu->base + DDRC_INT_STATUS);
-}
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 
-static void hisi_ddrc_pmu_v1_clear_int_status(struct hisi_pmu *ddrc_pmu,
-					      int idx)
-{
-	writel(1 << idx, ddrc_pmu->base + DDRC_INT_CLEAR);
+	return readl(ddrc_pmu->base + regs->int_status);
 }
 
-static u32 hisi_ddrc_pmu_v2_get_int_status(struct hisi_pmu *ddrc_pmu)
+static void hisi_ddrc_pmu_clear_int_status(struct hisi_pmu *ddrc_pmu,
+					   int idx)
 {
-	return readl(ddrc_pmu->base + DDRC_V2_INT_STATUS);
-}
+	struct hisi_ddrc_pmu_regs *regs = ddrc_pmu->dev_info->private;
 
-static void hisi_ddrc_pmu_v2_clear_int_status(struct hisi_pmu *ddrc_pmu,
-					      int idx)
-{
-	writel(1 << idx, ddrc_pmu->base + DDRC_V2_INT_CLEAR);
+	writel(1 << idx, ddrc_pmu->base + regs->int_clear);
 }
 
-static const struct acpi_device_id hisi_ddrc_pmu_acpi_match[] = {
-	{ "HISI0233", },
-	{ "HISI0234", },
-	{}
-};
-MODULE_DEVICE_TABLE(acpi, hisi_ddrc_pmu_acpi_match);
-
 static int hisi_ddrc_pmu_init_data(struct platform_device *pdev,
 				   struct hisi_pmu *ddrc_pmu)
 {
@@ -314,6 +242,10 @@ static int hisi_ddrc_pmu_init_data(struct platform_device *pdev,
 		return -EINVAL;
 	}
 
+	ddrc_pmu->dev_info = device_get_match_data(&pdev->dev);
+	if (!ddrc_pmu->dev_info)
+		return -ENODEV;
+
 	ddrc_pmu->base = devm_platform_ioremap_resource(pdev, 0);
 	if (IS_ERR(ddrc_pmu->base)) {
 		dev_err(&pdev->dev, "ioremap failed for ddrc_pmu resource\n");
@@ -396,34 +328,19 @@ static const struct attribute_group *hisi_ddrc_pmu_v2_attr_groups[] = {
 	NULL
 };
 
-static const struct hisi_uncore_ops hisi_uncore_ddrc_v1_ops = {
+static const struct hisi_uncore_ops hisi_uncore_ddrc_ops = {
 	.write_evtype           = hisi_ddrc_pmu_write_evtype,
-	.get_event_idx		= hisi_ddrc_pmu_v1_get_event_idx,
-	.start_counters		= hisi_ddrc_pmu_v1_start_counters,
-	.stop_counters		= hisi_ddrc_pmu_v1_stop_counters,
-	.enable_counter		= hisi_ddrc_pmu_v1_enable_counter,
-	.disable_counter	= hisi_ddrc_pmu_v1_disable_counter,
-	.enable_counter_int	= hisi_ddrc_pmu_v1_enable_counter_int,
-	.disable_counter_int	= hisi_ddrc_pmu_v1_disable_counter_int,
-	.write_counter		= hisi_ddrc_pmu_v1_write_counter,
-	.read_counter		= hisi_ddrc_pmu_v1_read_counter,
-	.get_int_status		= hisi_ddrc_pmu_v1_get_int_status,
-	.clear_int_status	= hisi_ddrc_pmu_v1_clear_int_status,
-};
-
-static const struct hisi_uncore_ops hisi_uncore_ddrc_v2_ops = {
-	.write_evtype           = hisi_ddrc_pmu_write_evtype,
-	.get_event_idx		= hisi_ddrc_pmu_v2_get_event_idx,
-	.start_counters		= hisi_ddrc_pmu_v2_start_counters,
-	.stop_counters		= hisi_ddrc_pmu_v2_stop_counters,
-	.enable_counter		= hisi_ddrc_pmu_v2_enable_counter,
-	.disable_counter	= hisi_ddrc_pmu_v2_disable_counter,
-	.enable_counter_int	= hisi_ddrc_pmu_v2_enable_counter_int,
-	.disable_counter_int	= hisi_ddrc_pmu_v2_disable_counter_int,
-	.write_counter		= hisi_ddrc_pmu_v2_write_counter,
-	.read_counter		= hisi_ddrc_pmu_v2_read_counter,
-	.get_int_status		= hisi_ddrc_pmu_v2_get_int_status,
-	.clear_int_status	= hisi_ddrc_pmu_v2_clear_int_status,
+	.get_event_idx		= hisi_ddrc_pmu_get_event_idx,
+	.start_counters		= hisi_ddrc_pmu_start_counters,
+	.stop_counters		= hisi_ddrc_pmu_stop_counters,
+	.enable_counter		= hisi_ddrc_pmu_enable_counter,
+	.disable_counter	= hisi_ddrc_pmu_disable_counter,
+	.enable_counter_int	= hisi_ddrc_pmu_enable_counter_int,
+	.disable_counter_int	= hisi_ddrc_pmu_disable_counter_int,
+	.write_counter		= hisi_ddrc_pmu_write_counter,
+	.read_counter		= hisi_ddrc_pmu_read_counter,
+	.get_int_status		= hisi_ddrc_pmu_get_int_status,
+	.clear_int_status	= hisi_ddrc_pmu_clear_int_status,
 };
 
 static int hisi_ddrc_pmu_dev_probe(struct platform_device *pdev,
@@ -439,18 +356,10 @@ static int hisi_ddrc_pmu_dev_probe(struct platform_device *pdev,
 	if (ret)
 		return ret;
 
-	if (ddrc_pmu->identifier >= HISI_PMU_V2) {
-		ddrc_pmu->counter_bits = 48;
-		ddrc_pmu->check_event = DDRC_V2_NR_EVENTS;
-		ddrc_pmu->pmu_events.attr_groups = hisi_ddrc_pmu_v2_attr_groups;
-		ddrc_pmu->ops = &hisi_uncore_ddrc_v2_ops;
-	} else {
-		ddrc_pmu->counter_bits = 32;
-		ddrc_pmu->check_event = DDRC_V1_NR_EVENTS;
-		ddrc_pmu->pmu_events.attr_groups = hisi_ddrc_pmu_v1_attr_groups;
-		ddrc_pmu->ops = &hisi_uncore_ddrc_v1_ops;
-	}
-
+	ddrc_pmu->pmu_events.attr_groups = ddrc_pmu->dev_info->attr_groups;
+	ddrc_pmu->counter_bits = ddrc_pmu->dev_info->counter_bits;
+	ddrc_pmu->check_event = ddrc_pmu->dev_info->check_event;
+	ddrc_pmu->ops = &hisi_uncore_ddrc_ops;
 	ddrc_pmu->num_counters = DDRC_NR_COUNTERS;
 	ddrc_pmu->dev = &pdev->dev;
 	ddrc_pmu->on_cpu = -1;
@@ -515,6 +424,49 @@ static void hisi_ddrc_pmu_remove(struct platform_device *pdev)
 					    &ddrc_pmu->node);
 }
 
+static struct hisi_ddrc_pmu_regs hisi_ddrc_v1_pmu_regs = {
+	.event_cnt = DDRC_UNIMPLEMENTED_REG,
+	.event_ctrl = DDRC_EVENT_CTRL,
+	.event_type = DDRC_UNIMPLEMENTED_REG,
+	.perf_ctrl = DDRC_PERF_CTRL,
+	.perf_ctrl_en = DDRC_V1_PERF_CTRL_EN,
+	.int_mask = DDRC_INT_MASK,
+	.int_clear = DDRC_INT_CLEAR,
+	.int_status = DDRC_INT_STATUS,
+};
+
+static const struct hisi_pmu_dev_info hisi_ddrc_v1 = {
+	.counter_bits = 32,
+	.check_event = DDRC_V1_NR_EVENTS,
+	.attr_groups = hisi_ddrc_pmu_v1_attr_groups,
+	.private = &hisi_ddrc_v1_pmu_regs,
+};
+
+static struct hisi_ddrc_pmu_regs hisi_ddrc_v2_pmu_regs = {
+	.event_cnt = DDRC_V2_EVENT_CNT,
+	.event_ctrl = DDRC_V2_EVENT_CTRL,
+	.event_type = DDRC_V2_EVENT_TYPE,
+	.perf_ctrl = DDRC_V2_PERF_CTRL,
+	.perf_ctrl_en = DDRC_V2_PERF_CTRL_EN,
+	.int_mask = DDRC_V2_INT_MASK,
+	.int_clear = DDRC_V2_INT_CLEAR,
+	.int_status = DDRC_V2_INT_STATUS,
+};
+
+static const struct hisi_pmu_dev_info hisi_ddrc_v2 = {
+	.counter_bits = 48,
+	.check_event = DDRC_V2_NR_EVENTS,
+	.attr_groups = hisi_ddrc_pmu_v2_attr_groups,
+	.private = &hisi_ddrc_v2_pmu_regs,
+};
+
+static const struct acpi_device_id hisi_ddrc_pmu_acpi_match[] = {
+	{ "HISI0233", (kernel_ulong_t)&hisi_ddrc_v1 },
+	{ "HISI0234", (kernel_ulong_t)&hisi_ddrc_v2 },
+	{}
+};
+MODULE_DEVICE_TABLE(acpi, hisi_ddrc_pmu_acpi_match);
+
 static struct platform_driver hisi_ddrc_pmu_driver = {
 	.driver = {
 		.name = "hisi_ddrc_pmu",
-- 
2.24.0

Re: [PATCH 2/9] drivers/perf: hisi: Simplify the probe process for each DDRC version

[Jonathan Cameron]

On Tue, 18 Feb 2025 17:19:53 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> From: Junhao He <hejunhao3@huawei.com>
> 
> Version 1 and 2 of DDRC PMU also use different HID. Make use of
> struct acpi_device_id::driver_data for version specific information
> rather than judge the version register. This will help to
> simplify the probe process and also a bit easier for extension.
> 
> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Looks good to me. Looking at this I can see the check_reg naming
is an existing thing so maybe leave that alone for now and consider
renaming as a future improvement in readability.

Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>

[PATCH 3/9] drivers/perf: hisi: Add support for HiSilicon DDRC v3 PMU driver

[Yicong Yang]

From: Junhao He <hejunhao3@huawei.com>

HiSilicon DDRC v3 PMU has the different interrupt register offset
compared to the v2. Add device information of v3 PMU with ACPI
HID HISI0235.

Signed-off-by: Junhao He <hejunhao3@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c | 24 +++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
index 7e3c2436e96b..26eaa6d20c00 100644
--- a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
@@ -43,6 +43,11 @@
 #define DDRC_V2_EVENT_TYPE	0xe74
 #define DDRC_V2_PERF_CTRL	0xeA0
 
+/* DDRC interrupt registers definition in v3 */
+#define DDRC_V3_INT_MASK	0x534
+#define DDRC_V3_INT_STATUS	0x538
+#define DDRC_V3_INT_CLEAR	0x53C
+
 /* DDRC has 8-counters */
 #define DDRC_NR_COUNTERS	0x8
 #define DDRC_V1_PERF_CTRL_EN	0x2
@@ -460,9 +465,28 @@ static const struct hisi_pmu_dev_info hisi_ddrc_v2 = {
 	.private = &hisi_ddrc_v2_pmu_regs,
 };
 
+static struct hisi_ddrc_pmu_regs hisi_ddrc_v3_pmu_regs = {
+	.event_cnt = DDRC_V2_EVENT_CNT,
+	.event_ctrl = DDRC_V2_EVENT_CTRL,
+	.event_type = DDRC_V2_EVENT_TYPE,
+	.perf_ctrl = DDRC_V2_PERF_CTRL,
+	.perf_ctrl_en = DDRC_V2_PERF_CTRL_EN,
+	.int_mask = DDRC_V3_INT_MASK,
+	.int_clear = DDRC_V3_INT_CLEAR,
+	.int_status = DDRC_V3_INT_STATUS,
+};
+
+static const struct hisi_pmu_dev_info hisi_ddrc_v3 = {
+	.counter_bits = 48,
+	.check_event = DDRC_V2_NR_EVENTS,
+	.attr_groups = hisi_ddrc_pmu_v2_attr_groups,
+	.private = &hisi_ddrc_v3_pmu_regs,
+};
+
 static const struct acpi_device_id hisi_ddrc_pmu_acpi_match[] = {
 	{ "HISI0233", (kernel_ulong_t)&hisi_ddrc_v1 },
 	{ "HISI0234", (kernel_ulong_t)&hisi_ddrc_v2 },
+	{ "HISI0235", (kernel_ulong_t)&hisi_ddrc_v3 },
 	{}
 };
 MODULE_DEVICE_TABLE(acpi, hisi_ddrc_pmu_acpi_match);
-- 
2.24.0

Re: [PATCH 3/9] drivers/perf: hisi: Add support for HiSilicon DDRC v3 PMU driver

[Jonathan Cameron]

On Tue, 18 Feb 2025 17:19:54 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> From: Junhao He <hejunhao3@huawei.com>
> 
> HiSilicon DDRC v3 PMU has the different interrupt register offset
> compared to the v2. Add device information of v3 PMU with ACPI
> HID HISI0235.
> 
> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>

[PATCH 4/9] drivers/perf: hisi: Use ACPI driver_data to retrieve SLLC PMU information

[Yicong Yang]

From: Junhao He <hejunhao3@huawei.com>

Make use of struct acpi_device_id::driver_data for version specific
information rather than judge the version register. This will help
to simplify the probe process and also a bit easier for extension.

Factor out SLLC register definition to struct hisi_sllc_pmu_regs.
No functional changes intended.

Signed-off-by: Junhao He <hejunhao3@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 191 ++++++++++++------
 1 file changed, 125 insertions(+), 66 deletions(-)

diff --git a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
index dbd079016fc4..c1fd60d397c3 100644
--- a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
@@ -36,11 +36,14 @@
 #define SLLC_SRCID_NONE			0x0
 #define SLLC_TGTID_EN			BIT(5)
 #define SLLC_TGTID_NONE			0x0
-#define SLLC_TGTID_MIN_SHIFT		1
-#define SLLC_TGTID_MAX_SHIFT		12
-#define SLLC_SRCID_CMD_SHIFT		1
-#define SLLC_SRCID_MSK_SHIFT		12
+#define SLLC_TGTID_MIN_MSK		GENMASK(11, 1)
+#define SLLC_TGTID_MAX_MSK		GENMASK(22, 12)
+#define SLLC_SRCID_CMD_MSK		GENMASK(11, 1)
+#define SLLC_SRCID_MSK_MSK		GENMASK(22, 12)
 #define SLLC_NR_EVENTS			0x80
+#define SLLC_EVENT_CNTn(cnt0, n)	((cnt0) + (n) * 8)
+#define SLLC_FIRST_BIT(_mask)		(find_first_bit((const unsigned long *)&(_mask), 32))
+#define SLLC_FIELD_PREP(_mask, _val)	(_mask & (_val << SLLC_FIRST_BIT(_mask)))
 
 HISI_PMU_EVENT_ATTR_EXTRACTOR(tgtid_min, config1, 10, 0);
 HISI_PMU_EVENT_ATTR_EXTRACTOR(tgtid_max, config1, 21, 11);
@@ -48,6 +51,23 @@ HISI_PMU_EVENT_ATTR_EXTRACTOR(srcid_cmd, config1, 32, 22);
 HISI_PMU_EVENT_ATTR_EXTRACTOR(srcid_msk, config1, 43, 33);
 HISI_PMU_EVENT_ATTR_EXTRACTOR(tracetag_en, config1, 44, 44);
 
+struct hisi_sllc_pmu_regs {
+	u32 int_mask;
+	u32 int_clear;
+	u32 int_status;
+	u32 perf_ctrl;
+	u32 srcid_ctrl;
+	u32 srcid_mask_msk;
+	u32 srcid_cmd_msk;
+	u32 tgtid_ctrl;
+	u32 tgtid_max_msk;
+	u32 tgtid_min_msk;
+	u32 event_ctrl;
+	u32 event_type0;
+	u32 version;
+	u32 event_cnt0;
+};
+
 static bool tgtid_is_valid(u32 max, u32 min)
 {
 	return max > 0 && max >= min;
@@ -56,96 +76,103 @@ static bool tgtid_is_valid(u32 max, u32 min)
 static void hisi_sllc_pmu_enable_tracetag(struct perf_event *event)
 {
 	struct hisi_pmu *sllc_pmu = to_hisi_pmu(event->pmu);
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 tt_en = hisi_get_tracetag_en(event);
 
 	if (tt_en) {
 		u32 val;
 
-		val = readl(sllc_pmu->base + SLLC_PERF_CTRL);
+		val = readl(sllc_pmu->base + regs->perf_ctrl);
 		val |= SLLC_TRACETAG_EN | SLLC_FILT_EN;
-		writel(val, sllc_pmu->base + SLLC_PERF_CTRL);
+		writel(val, sllc_pmu->base + regs->perf_ctrl);
 	}
 }
 
 static void hisi_sllc_pmu_disable_tracetag(struct perf_event *event)
 {
 	struct hisi_pmu *sllc_pmu = to_hisi_pmu(event->pmu);
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 tt_en = hisi_get_tracetag_en(event);
 
 	if (tt_en) {
 		u32 val;
 
-		val = readl(sllc_pmu->base + SLLC_PERF_CTRL);
+		val = readl(sllc_pmu->base + regs->perf_ctrl);
 		val &= ~(SLLC_TRACETAG_EN | SLLC_FILT_EN);
-		writel(val, sllc_pmu->base + SLLC_PERF_CTRL);
+		writel(val, sllc_pmu->base + regs->perf_ctrl);
 	}
 }
 
 static void hisi_sllc_pmu_config_tgtid(struct perf_event *event)
 {
 	struct hisi_pmu *sllc_pmu = to_hisi_pmu(event->pmu);
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 min = hisi_get_tgtid_min(event);
 	u32 max = hisi_get_tgtid_max(event);
 
 	if (tgtid_is_valid(max, min)) {
-		u32 val = (max << SLLC_TGTID_MAX_SHIFT) | (min << SLLC_TGTID_MIN_SHIFT);
+		u32 val = SLLC_FIELD_PREP(regs->tgtid_max_msk, max) |
+			  SLLC_FIELD_PREP(regs->tgtid_min_msk, min);
 
-		writel(val, sllc_pmu->base + SLLC_TGTID_CTRL);
+		writel(val, sllc_pmu->base + regs->tgtid_ctrl);
 		/* Enable the tgtid */
-		val = readl(sllc_pmu->base + SLLC_PERF_CTRL);
+		val = readl(sllc_pmu->base + regs->perf_ctrl);
 		val |= SLLC_TGTID_EN | SLLC_FILT_EN;
-		writel(val, sllc_pmu->base + SLLC_PERF_CTRL);
+		writel(val, sllc_pmu->base + regs->perf_ctrl);
 	}
 }
 
 static void hisi_sllc_pmu_clear_tgtid(struct perf_event *event)
 {
 	struct hisi_pmu *sllc_pmu = to_hisi_pmu(event->pmu);
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 min = hisi_get_tgtid_min(event);
 	u32 max = hisi_get_tgtid_max(event);
 
 	if (tgtid_is_valid(max, min)) {
 		u32 val;
 
-		writel(SLLC_TGTID_NONE, sllc_pmu->base + SLLC_TGTID_CTRL);
+		writel(SLLC_TGTID_NONE, sllc_pmu->base + regs->tgtid_ctrl);
 		/* Disable the tgtid */
-		val = readl(sllc_pmu->base + SLLC_PERF_CTRL);
+		val = readl(sllc_pmu->base + regs->perf_ctrl);
 		val &= ~(SLLC_TGTID_EN | SLLC_FILT_EN);
-		writel(val, sllc_pmu->base + SLLC_PERF_CTRL);
+		writel(val, sllc_pmu->base + regs->perf_ctrl);
 	}
 }
 
 static void hisi_sllc_pmu_config_srcid(struct perf_event *event)
 {
 	struct hisi_pmu *sllc_pmu = to_hisi_pmu(event->pmu);
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 cmd = hisi_get_srcid_cmd(event);
 
 	if (cmd) {
-		u32 val, msk;
+		u32 msk = hisi_get_srcid_msk(event);
+		u32 val = SLLC_FIELD_PREP(regs->srcid_mask_msk, msk) |
+			  SLLC_FIELD_PREP(regs->srcid_cmd_msk, cmd);
 
-		msk = hisi_get_srcid_msk(event);
-		val = (cmd << SLLC_SRCID_CMD_SHIFT) | (msk << SLLC_SRCID_MSK_SHIFT);
-		writel(val, sllc_pmu->base + SLLC_SRCID_CTRL);
+		writel(val, sllc_pmu->base + regs->srcid_ctrl);
 		/* Enable the srcid */
-		val = readl(sllc_pmu->base + SLLC_PERF_CTRL);
+		val = readl(sllc_pmu->base + regs->perf_ctrl);
 		val |= SLLC_SRCID_EN | SLLC_FILT_EN;
-		writel(val, sllc_pmu->base + SLLC_PERF_CTRL);
+		writel(val, sllc_pmu->base + regs->perf_ctrl);
 	}
 }
 
 static void hisi_sllc_pmu_clear_srcid(struct perf_event *event)
 {
 	struct hisi_pmu *sllc_pmu = to_hisi_pmu(event->pmu);
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 cmd = hisi_get_srcid_cmd(event);
 
 	if (cmd) {
 		u32 val;
 
-		writel(SLLC_SRCID_NONE, sllc_pmu->base + SLLC_SRCID_CTRL);
+		writel(SLLC_SRCID_NONE, sllc_pmu->base + regs->srcid_ctrl);
 		/* Disable the srcid */
-		val = readl(sllc_pmu->base + SLLC_PERF_CTRL);
+		val = readl(sllc_pmu->base + regs->perf_ctrl);
 		val &= ~(SLLC_SRCID_EN | SLLC_FILT_EN);
-		writel(val, sllc_pmu->base + SLLC_PERF_CTRL);
+		writel(val, sllc_pmu->base + regs->perf_ctrl);
 	}
 }
 
@@ -167,29 +194,27 @@ static void hisi_sllc_pmu_clear_filter(struct perf_event *event)
 	}
 }
 
-static u32 hisi_sllc_pmu_get_counter_offset(int idx)
-{
-	return (SLLC_EVENT_CNT0_L + idx * 8);
-}
-
 static u64 hisi_sllc_pmu_read_counter(struct hisi_pmu *sllc_pmu,
 				      struct hw_perf_event *hwc)
 {
-	return readq(sllc_pmu->base +
-		     hisi_sllc_pmu_get_counter_offset(hwc->idx));
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
+
+	return readq(sllc_pmu->base + SLLC_EVENT_CNTn(regs->event_cnt0, hwc->idx));
 }
 
 static void hisi_sllc_pmu_write_counter(struct hisi_pmu *sllc_pmu,
 					struct hw_perf_event *hwc, u64 val)
 {
-	writeq(val, sllc_pmu->base +
-	       hisi_sllc_pmu_get_counter_offset(hwc->idx));
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
+
+	writeq(val, sllc_pmu->base + SLLC_EVENT_CNTn(regs->event_cnt0, hwc->idx));
 }
 
 static void hisi_sllc_pmu_write_evtype(struct hisi_pmu *sllc_pmu, int idx,
 				       u32 type)
 {
-	u32 reg, reg_idx, shift, val;
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
+	u32 reg, val;
 
 	/*
 	 * Select the appropriate event select register(SLLC_EVENT_TYPE0/1).
@@ -198,96 +223,98 @@ static void hisi_sllc_pmu_write_evtype(struct hisi_pmu *sllc_pmu, int idx,
 	 * SLLC_EVENT_TYPE0 is chosen. For the latter 4 hardware counters,
 	 * SLLC_EVENT_TYPE1 is chosen.
 	 */
-	reg = SLLC_EVENT_TYPE0 + (idx / 4) * 4;
-	reg_idx = idx % 4;
-	shift = 8 * reg_idx;
+	reg = regs->event_type0 + (idx / 4) * 4;
 
 	/* Write event code to SLLC_EVENT_TYPEx Register */
 	val = readl(sllc_pmu->base + reg);
-	val &= ~(SLLC_EVTYPE_MASK << shift);
-	val |= (type << shift);
+	val &= ~(SLLC_EVTYPE_MASK << HISI_PMU_EVTYPE_SHIFT(idx));
+	val |= (type << HISI_PMU_EVTYPE_SHIFT(idx));
 	writel(val, sllc_pmu->base + reg);
 }
 
 static void hisi_sllc_pmu_start_counters(struct hisi_pmu *sllc_pmu)
 {
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(sllc_pmu->base + SLLC_PERF_CTRL);
+	val = readl(sllc_pmu->base + regs->perf_ctrl);
 	val |= SLLC_PERF_CTRL_EN;
-	writel(val, sllc_pmu->base + SLLC_PERF_CTRL);
+	writel(val, sllc_pmu->base + regs->perf_ctrl);
 }
 
 static void hisi_sllc_pmu_stop_counters(struct hisi_pmu *sllc_pmu)
 {
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(sllc_pmu->base + SLLC_PERF_CTRL);
+	val = readl(sllc_pmu->base + regs->perf_ctrl);
 	val &= ~(SLLC_PERF_CTRL_EN);
-	writel(val, sllc_pmu->base + SLLC_PERF_CTRL);
+	writel(val, sllc_pmu->base + regs->perf_ctrl);
 }
 
 static void hisi_sllc_pmu_enable_counter(struct hisi_pmu *sllc_pmu,
 					 struct hw_perf_event *hwc)
 {
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(sllc_pmu->base + SLLC_EVENT_CTRL);
-	val |= 1 << hwc->idx;
-	writel(val, sllc_pmu->base + SLLC_EVENT_CTRL);
+	val = readl(sllc_pmu->base + regs->event_ctrl);
+	val |= BIT_ULL(hwc->idx);
+	writel(val, sllc_pmu->base + regs->event_ctrl);
 }
 
 static void hisi_sllc_pmu_disable_counter(struct hisi_pmu *sllc_pmu,
 					  struct hw_perf_event *hwc)
 {
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(sllc_pmu->base + SLLC_EVENT_CTRL);
-	val &= ~(1 << hwc->idx);
-	writel(val, sllc_pmu->base + SLLC_EVENT_CTRL);
+	val = readl(sllc_pmu->base + regs->event_ctrl);
+	val &= ~BIT_ULL(hwc->idx);
+	writel(val, sllc_pmu->base + regs->event_ctrl);
 }
 
 static void hisi_sllc_pmu_enable_counter_int(struct hisi_pmu *sllc_pmu,
 					     struct hw_perf_event *hwc)
 {
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(sllc_pmu->base + SLLC_INT_MASK);
-	/* Write 0 to enable interrupt */
-	val &= ~(1 << hwc->idx);
-	writel(val, sllc_pmu->base + SLLC_INT_MASK);
+	val = readl(sllc_pmu->base + regs->int_mask);
+	val &= ~BIT_ULL(hwc->idx);
+	writel(val, sllc_pmu->base + regs->int_mask);
 }
 
 static void hisi_sllc_pmu_disable_counter_int(struct hisi_pmu *sllc_pmu,
 					      struct hw_perf_event *hwc)
 {
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 	u32 val;
 
-	val = readl(sllc_pmu->base + SLLC_INT_MASK);
-	/* Write 1 to mask interrupt */
-	val |= 1 << hwc->idx;
-	writel(val, sllc_pmu->base + SLLC_INT_MASK);
+	val = readl(sllc_pmu->base + regs->int_mask);
+	val |= BIT_ULL(hwc->idx);
+	writel(val, sllc_pmu->base + regs->int_mask);
 }
 
 static u32 hisi_sllc_pmu_get_int_status(struct hisi_pmu *sllc_pmu)
 {
-	return readl(sllc_pmu->base + SLLC_INT_STATUS);
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
+
+	return readl(sllc_pmu->base + regs->int_status);
 }
 
 static void hisi_sllc_pmu_clear_int_status(struct hisi_pmu *sllc_pmu, int idx)
 {
-	writel(1 << idx, sllc_pmu->base + SLLC_INT_CLEAR);
-}
+	struct hisi_sllc_pmu_regs *regs = sllc_pmu->dev_info->private;
 
-static const struct acpi_device_id hisi_sllc_pmu_acpi_match[] = {
-	{ "HISI0263", },
-	{}
-};
-MODULE_DEVICE_TABLE(acpi, hisi_sllc_pmu_acpi_match);
+	writel(BIT_ULL(idx), sllc_pmu->base + regs->int_clear);
+}
 
 static int hisi_sllc_pmu_init_data(struct platform_device *pdev,
 				   struct hisi_pmu *sllc_pmu)
 {
+	struct hisi_sllc_pmu_regs *regs;
+
 	hisi_uncore_pmu_init_topology(sllc_pmu, &pdev->dev);
 
 	/*
@@ -304,13 +331,18 @@ static int hisi_sllc_pmu_init_data(struct platform_device *pdev,
 		return -EINVAL;
 	}
 
+	sllc_pmu->dev_info = device_get_match_data(&pdev->dev);
+	if (!sllc_pmu->dev_info)
+		return -ENODEV;
+
 	sllc_pmu->base = devm_platform_ioremap_resource(pdev, 0);
 	if (IS_ERR(sllc_pmu->base)) {
 		dev_err(&pdev->dev, "ioremap failed for sllc_pmu resource.\n");
 		return PTR_ERR(sllc_pmu->base);
 	}
 
-	sllc_pmu->identifier = readl(sllc_pmu->base + SLLC_VERSION);
+	regs = sllc_pmu->dev_info->private;
+	sllc_pmu->identifier = readl(sllc_pmu->base + regs->version);
 
 	return 0;
 }
@@ -352,6 +384,27 @@ static const struct attribute_group *hisi_sllc_pmu_v2_attr_groups[] = {
 	NULL
 };
 
+static struct hisi_sllc_pmu_regs hisi_sllc_v2_pmu_regs = {
+	.int_mask = SLLC_INT_MASK,
+	.int_clear = SLLC_INT_CLEAR,
+	.int_status = SLLC_INT_STATUS,
+	.perf_ctrl = SLLC_PERF_CTRL,
+	.srcid_ctrl = SLLC_SRCID_CTRL,
+	.srcid_mask_msk = SLLC_SRCID_MSK_MSK,
+	.srcid_cmd_msk = SLLC_SRCID_CMD_MSK,
+	.tgtid_ctrl = SLLC_TGTID_CTRL,
+	.tgtid_max_msk = SLLC_TGTID_MAX_MSK,
+	.tgtid_min_msk = SLLC_TGTID_MIN_MSK,
+	.event_ctrl = SLLC_EVENT_CTRL,
+	.event_type0 = SLLC_EVENT_TYPE0,
+	.version = SLLC_VERSION,
+	.event_cnt0 = SLLC_EVENT_CNT0_L,
+};
+
+static const struct hisi_pmu_dev_info hisi_sllc_v2 = {
+	.private = &hisi_sllc_v2_pmu_regs,
+};
+
 static const struct hisi_uncore_ops hisi_uncore_sllc_ops = {
 	.write_evtype		= hisi_sllc_pmu_write_evtype,
 	.get_event_idx		= hisi_uncore_pmu_get_event_idx,
@@ -443,6 +496,12 @@ static void hisi_sllc_pmu_remove(struct platform_device *pdev)
 					    &sllc_pmu->node);
 }
 
+static const struct acpi_device_id hisi_sllc_pmu_acpi_match[] = {
+	{ "HISI0263", (kernel_ulong_t)&hisi_sllc_v2 },
+	{}
+};
+MODULE_DEVICE_TABLE(acpi, hisi_sllc_pmu_acpi_match);
+
 static struct platform_driver hisi_sllc_pmu_driver = {
 	.driver = {
 		.name = "hisi_sllc_pmu",
-- 
2.24.0

Re: [PATCH 4/9] drivers/perf: hisi: Use ACPI driver_data to retrieve SLLC PMU information

[Will Deacon]

On Tue, Feb 18, 2025 at 05:19:55PM +0800, Yicong Yang wrote:
> From: Junhao He <hejunhao3@huawei.com>
> 
> Make use of struct acpi_device_id::driver_data for version specific
> information rather than judge the version register. This will help
> to simplify the probe process and also a bit easier for extension.
> 
> Factor out SLLC register definition to struct hisi_sllc_pmu_regs.
> No functional changes intended.
> 
> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
> ---
>  drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 191 ++++++++++++------
>  1 file changed, 125 insertions(+), 66 deletions(-)
> 
> diff --git a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> index dbd079016fc4..c1fd60d397c3 100644
> --- a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> +++ b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> @@ -36,11 +36,14 @@
>  #define SLLC_SRCID_NONE			0x0
>  #define SLLC_TGTID_EN			BIT(5)
>  #define SLLC_TGTID_NONE			0x0
> -#define SLLC_TGTID_MIN_SHIFT		1
> -#define SLLC_TGTID_MAX_SHIFT		12
> -#define SLLC_SRCID_CMD_SHIFT		1
> -#define SLLC_SRCID_MSK_SHIFT		12
> +#define SLLC_TGTID_MIN_MSK		GENMASK(11, 1)
> +#define SLLC_TGTID_MAX_MSK		GENMASK(22, 12)
> +#define SLLC_SRCID_CMD_MSK		GENMASK(11, 1)
> +#define SLLC_SRCID_MSK_MSK		GENMASK(22, 12)
>  #define SLLC_NR_EVENTS			0x80
> +#define SLLC_EVENT_CNTn(cnt0, n)	((cnt0) + (n) * 8)
> +#define SLLC_FIRST_BIT(_mask)		(find_first_bit((const unsigned long *)&(_mask), 32))
> +#define SLLC_FIELD_PREP(_mask, _val)	(_mask & (_val << SLLC_FIRST_BIT(_mask)))

It's a bit of a shame to have to compute this dynamically given that the
input mask is constant for a given device. Is it not possible to use the
generic FIELD_PREP macro in per-device code and then just dispatch to
that, instead of funneling everything through hisi_sllc_pmu_regs?

Will

Re: [PATCH 4/9] drivers/perf: hisi: Use ACPI driver_data to retrieve SLLC PMU information

[Jonathan Cameron]

On Sat, 1 Mar 2025 06:43:49 +0000
Will Deacon <will@kernel.org> wrote:

> On Tue, Feb 18, 2025 at 05:19:55PM +0800, Yicong Yang wrote:
> > From: Junhao He <hejunhao3@huawei.com>
> > 
> > Make use of struct acpi_device_id::driver_data for version specific
> > information rather than judge the version register. This will help
> > to simplify the probe process and also a bit easier for extension.
> > 
> > Factor out SLLC register definition to struct hisi_sllc_pmu_regs.
> > No functional changes intended.
> > 
> > Signed-off-by: Junhao He <hejunhao3@huawei.com>
> > Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
> > ---
> >  drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 191 ++++++++++++------
> >  1 file changed, 125 insertions(+), 66 deletions(-)
> > 
> > diff --git a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> > index dbd079016fc4..c1fd60d397c3 100644
> > --- a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> > +++ b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> > @@ -36,11 +36,14 @@
> >  #define SLLC_SRCID_NONE			0x0
> >  #define SLLC_TGTID_EN			BIT(5)
> >  #define SLLC_TGTID_NONE			0x0
> > -#define SLLC_TGTID_MIN_SHIFT		1
> > -#define SLLC_TGTID_MAX_SHIFT		12
> > -#define SLLC_SRCID_CMD_SHIFT		1
> > -#define SLLC_SRCID_MSK_SHIFT		12
> > +#define SLLC_TGTID_MIN_MSK		GENMASK(11, 1)
> > +#define SLLC_TGTID_MAX_MSK		GENMASK(22, 12)
> > +#define SLLC_SRCID_CMD_MSK		GENMASK(11, 1)
> > +#define SLLC_SRCID_MSK_MSK		GENMASK(22, 12)
> >  #define SLLC_NR_EVENTS			0x80
> > +#define SLLC_EVENT_CNTn(cnt0, n)	((cnt0) + (n) * 8)
> > +#define SLLC_FIRST_BIT(_mask)		(find_first_bit((const unsigned long *)&(_mask), 32))
> > +#define SLLC_FIELD_PREP(_mask, _val)	(_mask & (_val << SLLC_FIRST_BIT(_mask)))  
> 
> It's a bit of a shame to have to compute this dynamically given that th
> input mask is constant for a given device. Is it not possible to use the
> generic FIELD_PREP macro in per-device code and then just dispatch to
> that, instead of funneling everything through hisi_sllc_pmu_regs?
There is yet another ongoing discussion of handling non const
field cases:
https://lore.kernel.org/all/cover.1739540679.git.geert+renesas@glider.be/

Maybe that will resolve with a nicer solution but I doubt it.

Alternatively best plan may be a shift (and length) instead
of the neat solution of FIELD_PREP().

Jonathan

> 
> Will

Re: [PATCH 4/9] drivers/perf: hisi: Use ACPI driver_data to retrieve SLLC PMU information

[Jonathan Cameron]

On Tue, 18 Feb 2025 17:19:55 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> From: Junhao He <hejunhao3@huawei.com>
> 
> Make use of struct acpi_device_id::driver_data for version specific
> information rather than judge the version register. This will help
> to simplify the probe process and also a bit easier for extension.
> 
> Factor out SLLC register definition to struct hisi_sllc_pmu_regs.
> No functional changes intended.
> 
> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
> ---
>  drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 191 ++++++++++++------
>  1 file changed, 125 insertions(+), 66 deletions(-)
> 
> diff --git a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> index dbd079016fc4..c1fd60d397c3 100644
> --- a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> +++ b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> @@ -36,11 +36,14 @@
>  #define SLLC_SRCID_NONE			0x0
>  #define SLLC_TGTID_EN			BIT(5)
>  #define SLLC_TGTID_NONE			0x0
> -#define SLLC_TGTID_MIN_SHIFT		1
> -#define SLLC_TGTID_MAX_SHIFT		12
> -#define SLLC_SRCID_CMD_SHIFT		1
> -#define SLLC_SRCID_MSK_SHIFT		12
> +#define SLLC_TGTID_MIN_MSK		GENMASK(11, 1)
> +#define SLLC_TGTID_MAX_MSK		GENMASK(22, 12)
> +#define SLLC_SRCID_CMD_MSK		GENMASK(11, 1)
> +#define SLLC_SRCID_MSK_MSK		GENMASK(22, 12)
>  #define SLLC_NR_EVENTS			0x80
> +#define SLLC_EVENT_CNTn(cnt0, n)	((cnt0) + (n) * 8)
> +#define SLLC_FIRST_BIT(_mask)		(find_first_bit((const unsigned long *)&(_mask), 32))

If you do end up keeping this, I think there should be no need for outer brackets here.

> +#define SLLC_FIELD_PREP(_mask, _val)	(_mask & (_val << SLLC_FIRST_BIT(_mask)))

[PATCH 5/9] drivers/perf: hisi: Add support for HiSilicon SLLC v3 PMU driver

[Yicong Yang]

From: Junhao He <hejunhao3@huawei.com>

SLLC v3 PMU has the following changes compared to previous version:
a) update the register layout
b) update the definition of SRCID_CTRL and TGTID_CTRL registers.
   To be compatible with v2, we use maximum width (11 bits)
   and mask the extra length for themselves.
c) remove latency events (driver does not need to be adapted).

SLLC v3 PMU is identified with HID HISI0264.

Signed-off-by: Junhao He <hejunhao3@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 40 +++++++++++++++++++
 1 file changed, 40 insertions(+)

diff --git a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
index c1fd60d397c3..dce7195320f2 100644
--- a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
@@ -28,6 +28,18 @@
 #define SLLC_VERSION			0x1cf0
 #define SLLC_EVENT_CNT0_L		0x1d00
 
+/* SLLC registers definition in v3 */
+#define SLLC_V3_INT_MASK		0x6834
+#define SLLC_V3_INT_STATUS		0x6838
+#define SLLC_V3_INT_CLEAR		0x683c
+#define SLLC_V3_VERSION			0x6c00
+#define SLLC_V3_PERF_CTRL		0x6d00
+#define SLLC_V3_SRCID_CTRL		0x6d04
+#define SLLC_V3_TGTID_CTRL		0x6d08
+#define SLLC_V3_EVENT_CTRL		0x6d14
+#define SLLC_V3_EVENT_TYPE0		0x6d18
+#define SLLC_V3_EVENT_CNT0_L		0x6e00
+
 #define SLLC_EVTYPE_MASK		0xff
 #define SLLC_PERF_CTRL_EN		BIT(0)
 #define SLLC_FILT_EN			BIT(1)
@@ -40,6 +52,12 @@
 #define SLLC_TGTID_MAX_MSK		GENMASK(22, 12)
 #define SLLC_SRCID_CMD_MSK		GENMASK(11, 1)
 #define SLLC_SRCID_MSK_MSK		GENMASK(22, 12)
+
+#define SLLC_V3_TGTID_MIN_MSK		GENMASK(9, 1)
+#define SLLC_V3_TGTID_MAX_MSK		GENMASK(18, 10)
+#define SLLC_V3_SRCID_CMD_MSK		GENMASK(9, 1)
+#define SLLC_V3_SRCID_MSK_MSK		GENMASK(18, 10)
+
 #define SLLC_NR_EVENTS			0x80
 #define SLLC_EVENT_CNTn(cnt0, n)	((cnt0) + (n) * 8)
 #define SLLC_FIRST_BIT(_mask)		(find_first_bit((const unsigned long *)&(_mask), 32))
@@ -405,6 +423,27 @@ static const struct hisi_pmu_dev_info hisi_sllc_v2 = {
 	.private = &hisi_sllc_v2_pmu_regs,
 };
 
+static struct hisi_sllc_pmu_regs hisi_sllc_v3_pmu_regs = {
+	.int_mask = SLLC_V3_INT_MASK,
+	.int_clear = SLLC_V3_INT_CLEAR,
+	.int_status = SLLC_V3_INT_STATUS,
+	.perf_ctrl = SLLC_V3_PERF_CTRL,
+	.srcid_ctrl = SLLC_V3_SRCID_CTRL,
+	.srcid_mask_msk = SLLC_V3_SRCID_MSK_MSK,
+	.srcid_cmd_msk = SLLC_V3_SRCID_CMD_MSK,
+	.tgtid_ctrl = SLLC_V3_TGTID_CTRL,
+	.tgtid_max_msk = SLLC_V3_TGTID_MAX_MSK,
+	.tgtid_min_msk = SLLC_V3_TGTID_MIN_MSK,
+	.event_ctrl = SLLC_V3_EVENT_CTRL,
+	.event_type0 = SLLC_V3_EVENT_TYPE0,
+	.version = SLLC_V3_VERSION,
+	.event_cnt0 = SLLC_V3_EVENT_CNT0_L,
+};
+
+static const struct hisi_pmu_dev_info hisi_sllc_v3 = {
+	.private = &hisi_sllc_v3_pmu_regs,
+};
+
 static const struct hisi_uncore_ops hisi_uncore_sllc_ops = {
 	.write_evtype		= hisi_sllc_pmu_write_evtype,
 	.get_event_idx		= hisi_uncore_pmu_get_event_idx,
@@ -498,6 +537,7 @@ static void hisi_sllc_pmu_remove(struct platform_device *pdev)
 
 static const struct acpi_device_id hisi_sllc_pmu_acpi_match[] = {
 	{ "HISI0263", (kernel_ulong_t)&hisi_sllc_v2 },
+	{ "HISI0264", (kernel_ulong_t)&hisi_sllc_v3 },
 	{}
 };
 MODULE_DEVICE_TABLE(acpi, hisi_sllc_pmu_acpi_match);
-- 
2.24.0

Re: [PATCH 5/9] drivers/perf: hisi: Add support for HiSilicon SLLC v3 PMU driver

[Jonathan Cameron]

On Tue, 18 Feb 2025 17:19:56 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> From: Junhao He <hejunhao3@huawei.com>
> 
> SLLC v3 PMU has the following changes compared to previous version:
> a) update the register layout
> b) update the definition of SRCID_CTRL and TGTID_CTRL registers.
>    To be compatible with v2, we use maximum width (11 bits)
>    and mask the extra length for themselves.
> c) remove latency events (driver does not need to be adapted).
> 
> SLLC v3 PMU is identified with HID HISI0264.
> 
> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>

Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>

[PATCH 6/9] drivers/perf: hisi: Relax the event number check of v2 PMUs

[Yicong Yang]

From: Junhao He <hejunhao3@huawei.com>

The supported event number range of each Uncore PMUs is provided by
each driver in hisi_pmu::check_event and out of range events
will be rejected. A later version with expanded event number range
needs to register the PMU with updated hisi_pmu::check_event
even if it's the only update, which means the expanded events
cannot be used unless the driver's updated. However the unsupported
events won't be counted by the hardware so we can relax the event
number check to allow the use the expanded events.

Signed-off-by: Junhao He <hejunhao3@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c | 2 +-
 drivers/perf/hisilicon/hisi_uncore_hha_pmu.c  | 7 +++----
 drivers/perf/hisilicon/hisi_uncore_pa_pmu.c   | 2 +-
 drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 3 +--
 4 files changed, 6 insertions(+), 8 deletions(-)

diff --git a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
index 26eaa6d20c00..21c494881ca0 100644
--- a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
@@ -53,7 +53,7 @@
 #define DDRC_V1_PERF_CTRL_EN	0x2
 #define DDRC_V2_PERF_CTRL_EN	0x1
 #define DDRC_V1_NR_EVENTS	0x7
-#define DDRC_V2_NR_EVENTS	0x90
+#define DDRC_V2_NR_EVENTS	0xFF
 
 #define DDRC_EVENT_CNTn(base, n)	((base) + (n) * 8)
 #define DDRC_EVENT_TYPEn(base, n)	((base) + (n) * 4)
diff --git a/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c b/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c
index ca609db86046..78cd6d67f209 100644
--- a/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c
@@ -47,9 +47,8 @@
 #define HHA_SRCID_CMD		GENMASK(16, 6)
 #define HHA_SRCID_MSK		GENMASK(30, 20)
 #define HHA_DATSRC_SKT_EN	BIT(23)
-#define HHA_EVTYPE_NONE		0xff
+#define HHA_EVTYPE_MASK		GENMASK(7, 0)
 #define HHA_V1_NR_EVENT		0x65
-#define HHA_V2_NR_EVENT		0xCE
 
 HISI_PMU_EVENT_ATTR_EXTRACTOR(srcid_cmd, config1, 10, 0);
 HISI_PMU_EVENT_ATTR_EXTRACTOR(srcid_msk, config1, 21, 11);
@@ -197,7 +196,7 @@ static void hisi_hha_pmu_write_evtype(struct hisi_pmu *hha_pmu, int idx,
 
 	/* Write event code to HHA_EVENT_TYPEx register */
 	val = readl(hha_pmu->base + reg);
-	val &= ~(HHA_EVTYPE_NONE << shift);
+	val &= ~(HHA_EVTYPE_MASK << shift);
 	val |= (type << shift);
 	writel(val, hha_pmu->base + reg);
 }
@@ -453,7 +452,7 @@ static int hisi_hha_pmu_dev_probe(struct platform_device *pdev,
 
 	if (hha_pmu->identifier >= HISI_PMU_V2) {
 		hha_pmu->counter_bits = 64;
-		hha_pmu->check_event = HHA_V2_NR_EVENT;
+		hha_pmu->check_event = HHA_EVTYPE_MASK;
 		hha_pmu->pmu_events.attr_groups = hisi_hha_pmu_v2_attr_groups;
 		hha_pmu->num_counters = HHA_V2_NR_COUNTERS;
 	} else {
diff --git a/drivers/perf/hisilicon/hisi_uncore_pa_pmu.c b/drivers/perf/hisilicon/hisi_uncore_pa_pmu.c
index a0142684e379..80108c63cb60 100644
--- a/drivers/perf/hisilicon/hisi_uncore_pa_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_pa_pmu.c
@@ -440,7 +440,7 @@ static int hisi_pa_pmu_dev_probe(struct platform_device *pdev,
 	pa_pmu->pmu_events.attr_groups = pa_pmu->dev_info->attr_groups;
 	pa_pmu->num_counters = PA_NR_COUNTERS;
 	pa_pmu->ops = &hisi_uncore_pa_ops;
-	pa_pmu->check_event = 0xB0;
+	pa_pmu->check_event = PA_EVTYPE_MASK;
 	pa_pmu->counter_bits = 64;
 	pa_pmu->dev = &pdev->dev;
 	pa_pmu->on_cpu = -1;
diff --git a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
index dce7195320f2..ea31b64e05e2 100644
--- a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
@@ -58,7 +58,6 @@
 #define SLLC_V3_SRCID_CMD_MSK		GENMASK(9, 1)
 #define SLLC_V3_SRCID_MSK_MSK		GENMASK(18, 10)
 
-#define SLLC_NR_EVENTS			0x80
 #define SLLC_EVENT_CNTn(cnt0, n)	((cnt0) + (n) * 8)
 #define SLLC_FIRST_BIT(_mask)		(find_first_bit((const unsigned long *)&(_mask), 32))
 #define SLLC_FIELD_PREP(_mask, _val)	(_mask & (_val << SLLC_FIRST_BIT(_mask)))
@@ -476,7 +475,7 @@ static int hisi_sllc_pmu_dev_probe(struct platform_device *pdev,
 
 	sllc_pmu->pmu_events.attr_groups = hisi_sllc_pmu_v2_attr_groups;
 	sllc_pmu->ops = &hisi_uncore_sllc_ops;
-	sllc_pmu->check_event = SLLC_NR_EVENTS;
+	sllc_pmu->check_event = SLLC_EVTYPE_MASK;
 	sllc_pmu->counter_bits = 64;
 	sllc_pmu->num_counters = 8;
 	sllc_pmu->dev = &pdev->dev;
-- 
2.24.0

Re: [PATCH 6/9] drivers/perf: hisi: Relax the event number check of v2 PMUs

[Jonathan Cameron]

On Tue, 18 Feb 2025 17:19:57 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> From: Junhao He <hejunhao3@huawei.com>
> 
> The supported event number range of each Uncore PMUs is provided by
> each driver in hisi_pmu::check_event and out of range events
> will be rejected. A later version with expanded event number range
> needs to register the PMU with updated hisi_pmu::check_event
> even if it's the only update, which means the expanded events
> cannot be used unless the driver's updated. However the unsupported
> events won't be counted by the hardware so we can relax the event
> number check to allow the use the expanded events.
> 
> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
> ---
>  drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c | 2 +-
>  drivers/perf/hisilicon/hisi_uncore_hha_pmu.c  | 7 +++----
>  drivers/perf/hisilicon/hisi_uncore_pa_pmu.c   | 2 +-
>  drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 3 +--
>  4 files changed, 6 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
> index 26eaa6d20c00..21c494881ca0 100644
> --- a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
> +++ b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
> @@ -53,7 +53,7 @@
>  #define DDRC_V1_PERF_CTRL_EN	0x2
>  #define DDRC_V2_PERF_CTRL_EN	0x1
>  #define DDRC_V1_NR_EVENTS	0x7
> -#define DDRC_V2_NR_EVENTS	0x90
> +#define DDRC_V2_NR_EVENTS	0xFF
>  
>  #define DDRC_EVENT_CNTn(base, n)	((base) + (n) * 8)
>  #define DDRC_EVENT_TYPEn(base, n)	((base) + (n) * 4)
> diff --git a/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c b/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c
> index ca609db86046..78cd6d67f209 100644
> --- a/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c
> +++ b/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c
> @@ -47,9 +47,8 @@
>  #define HHA_SRCID_CMD		GENMASK(16, 6)
>  #define HHA_SRCID_MSK		GENMASK(30, 20)
>  #define HHA_DATSRC_SKT_EN	BIT(23)
> -#define HHA_EVTYPE_NONE		0xff
> +#define HHA_EVTYPE_MASK		GENMASK(7, 0)
Using something called mask in places where we previously
had something called nr_events seems a little odd.

renaming EVTYPE_NONE to EVTYPE_MASK seems valid given the
useage but I'd have a different define for the number
of events and not make both changes in one patch.

>  #define HHA_V1_NR_EVENT		0x65
> -#define HHA_V2_NR_EVENT		0xCE
>  
>  HISI_PMU_EVENT_ATTR_EXTRACTOR(srcid_cmd, config1, 10, 0);
>  HISI_PMU_EVENT_ATTR_EXTRACTOR(srcid_msk, config1, 21, 11);
> @@ -197,7 +196,7 @@ static void hisi_hha_pmu_write_evtype(struct hisi_pmu *hha_pmu, int idx,
>  
>  	/* Write event code to HHA_EVENT_TYPEx register */
>  	val = readl(hha_pmu->base + reg);
> -	val &= ~(HHA_EVTYPE_NONE << shift);
> +	val &= ~(HHA_EVTYPE_MASK << shift);
>  	val |= (type << shift);
>  	writel(val, hha_pmu->base + reg);
>  }
> @@ -453,7 +452,7 @@ static int hisi_hha_pmu_dev_probe(struct platform_device *pdev,
>  
>  	if (hha_pmu->identifier >= HISI_PMU_V2) {
>  		hha_pmu->counter_bits = 64;
> -		hha_pmu->check_event = HHA_V2_NR_EVENT;
> +		hha_pmu->check_event = HHA_EVTYPE_MASK;
To me this makes little sense.  Should be HHA_MAX_NR_EVENT
or something like that.

>  		hha_pmu->pmu_events.attr_groups = hisi_hha_pmu_v2_attr_groups;
>  		hha_pmu->num_counters = HHA_V2_NR_COUNTERS;
>  	} else {

Jonathan

[PATCH 7/9] drivers/perf: hisi: Support PMUs with no interrupt

[Yicong Yang]

From: Yicong Yang <yangyicong@hisilicon.com>

We'll have PMUs don't have an interrupt to indicate the counter
overflow, but the Uncore PMU core assume all the PMUs have
interrupt. So handle this case in the core. The existing PMUs
won't be affected.

Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 drivers/perf/hisilicon/hisi_uncore_pmu.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/drivers/perf/hisilicon/hisi_uncore_pmu.c b/drivers/perf/hisilicon/hisi_uncore_pmu.c
index ef058b1dd509..a449651f79c9 100644
--- a/drivers/perf/hisilicon/hisi_uncore_pmu.c
+++ b/drivers/perf/hisilicon/hisi_uncore_pmu.c
@@ -510,7 +510,9 @@ int hisi_uncore_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
 			return 0;
 
 		hisi_pmu->on_cpu = cpumask_local_spread(0, dev_to_node(hisi_pmu->dev));
-		WARN_ON(irq_set_affinity(hisi_pmu->irq, cpumask_of(hisi_pmu->on_cpu)));
+		if (hisi_pmu->irq > 0)
+			WARN_ON(irq_set_affinity(hisi_pmu->irq,
+						 cpumask_of(hisi_pmu->on_cpu)));
 		return 0;
 	}
 
@@ -525,7 +527,8 @@ int hisi_uncore_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
 	hisi_pmu->on_cpu = cpu;
 
 	/* Overflow interrupt also should use the same CPU */
-	WARN_ON(irq_set_affinity(hisi_pmu->irq, cpumask_of(cpu)));
+	if (hisi_pmu->irq > 0)
+		WARN_ON(irq_set_affinity(hisi_pmu->irq, cpumask_of(cpu)));
 
 	return 0;
 }
@@ -560,7 +563,9 @@ int hisi_uncore_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
 	perf_pmu_migrate_context(&hisi_pmu->pmu, cpu, target);
 	/* Use this CPU for event counting */
 	hisi_pmu->on_cpu = target;
-	WARN_ON(irq_set_affinity(hisi_pmu->irq, cpumask_of(target)));
+
+	if (hisi_pmu->irq > 0)
+		WARN_ON(irq_set_affinity(hisi_pmu->irq, cpumask_of(target)));
 
 	return 0;
 }
-- 
2.24.0

Re: [PATCH 7/9] drivers/perf: hisi: Support PMUs with no interrupt

[Jonathan Cameron]

On Tue, 18 Feb 2025 17:19:58 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> From: Yicong Yang <yangyicong@hisilicon.com>
> 
> We'll have PMUs don't have an interrupt to indicate the counter
> overflow, but the Uncore PMU core assume all the PMUs have
> interrupt. So handle this case in the core. The existing PMUs
> won't be affected.
> 
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>

[PATCH 8/9] drivers/perf: hisi: Add support for HiSilicon NoC PMU

[Yicong Yang]

From: Yicong Yang <yangyicong@hisilicon.com>

Adds the support for HiSilicon NoC (Network on Chip) PMU which
will be used to monitor the events on the system bus. The PMU
device will be named after the SCL ID (either Super CPU cluster
or Super IO cluster) and the index ID, just similar to other
HiSilicon Uncore PMUs. Below PMU formats are provided besides
the event:

- ch: the transaction channel (data, request, response, etc) which
  can be used to filter the counting.
- tt_en: tracetag filtering enable. Just as other HiSilicon Uncore
  PMUs the NoC PMU supports only counting the transactions with
  tracetag.

The NoC PMU doesn't have an interrupt to indicate the overflow.
However we have a 64 bit counter which is large enough and it's
nearly impossible to overflow.

Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 Documentation/admin-guide/perf/hisi-pmu.rst  |  11 +
 drivers/perf/hisilicon/Makefile              |   3 +-
 drivers/perf/hisilicon/hisi_uncore_noc_pmu.c | 391 +++++++++++++++++++
 3 files changed, 404 insertions(+), 1 deletion(-)
 create mode 100644 drivers/perf/hisilicon/hisi_uncore_noc_pmu.c

diff --git a/Documentation/admin-guide/perf/hisi-pmu.rst b/Documentation/admin-guide/perf/hisi-pmu.rst
index 48992a0b8e94..d858bd3a8802 100644
--- a/Documentation/admin-guide/perf/hisi-pmu.rst
+++ b/Documentation/admin-guide/perf/hisi-pmu.rst
@@ -112,6 +112,17 @@ uring channel. It is 2 bits. Some important codes are as follows:
 - 2'b00: default value, count the events which sent to the both uring and
   uring_ext channel;
 
+6. ch: NoC PMU supports to filter the event counts of certain transaction
+channel with this option. The current supported channels are as follows:
+
+- 3'b010: Request channel
+- 3'b100: Snoop channel
+- 3'b110: Response channel
+- 3'b111: Data channel
+
+7. tt_en: NoC PMU supports to only count the transactions with tracetag by
+this option. See the 2nd list for more information about tracetag.
+
 Users could configure IDs to count data come from specific CCL/ICL, by setting
 srcid_cmd & srcid_msk, and data desitined for specific CCL/ICL by setting
 tgtid_cmd & tgtid_msk. A set bit in srcid_msk/tgtid_msk means the PMU will not
diff --git a/drivers/perf/hisilicon/Makefile b/drivers/perf/hisilicon/Makefile
index 48dcc8381ea7..dcec8f39719d 100644
--- a/drivers/perf/hisilicon/Makefile
+++ b/drivers/perf/hisilicon/Makefile
@@ -1,7 +1,8 @@
 # SPDX-License-Identifier: GPL-2.0-only
 obj-$(CONFIG_HISI_PMU) += hisi_uncore_pmu.o hisi_uncore_l3c_pmu.o \
 			  hisi_uncore_hha_pmu.o hisi_uncore_ddrc_pmu.o hisi_uncore_sllc_pmu.o \
-			  hisi_uncore_pa_pmu.o hisi_uncore_cpa_pmu.o hisi_uncore_uc_pmu.o
+			  hisi_uncore_pa_pmu.o hisi_uncore_cpa_pmu.o hisi_uncore_uc_pmu.o \
+			  hisi_uncore_noc_pmu.o
 
 obj-$(CONFIG_HISI_PCIE_PMU) += hisi_pcie_pmu.o
 obj-$(CONFIG_HNS3_PMU) += hns3_pmu.o
diff --git a/drivers/perf/hisilicon/hisi_uncore_noc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_noc_pmu.c
new file mode 100644
index 000000000000..e755f6d584e4
--- /dev/null
+++ b/drivers/perf/hisilicon/hisi_uncore_noc_pmu.c
@@ -0,0 +1,391 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Driver for HiSilicon Uncore NoC (Network on Chip) PMU device
+ *
+ * Copyright (c) 2025 HiSilicon Technologies Co., Ltd.
+ * Author: Yicong Yang <yangyicong@hisilicon.com>
+ */
+#include <linux/bitops.h>
+#include <linux/cpuhotplug.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/mod_devicetable.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/property.h>
+#include <linux/sysfs.h>
+
+#include "hisi_uncore_pmu.h"
+
+#define NOC_PMU_VERSION			0x1e00
+#define NOC_PMU_GLOBAL_CTRL		0x1e04
+#define   NOC_PMU_GLOBAL_CTRL_PMU_EN	BIT(0)
+#define   NOC_PMU_GLOBAL_CTRL_TT_EN	BIT(1)
+#define NOC_PMU_CNT_INFO		0x1e08
+#define   NOC_PMU_CNT_INFO_OVERFLOW(n)	BIT(n)
+#define NOC_PMU_EVENT_CTRL(n)		(0x1e20 + 4 * (n))
+#define   NOC_PMU_EVENT_CTRL_TYPE	GENMASK(4, 0)
+/*
+ * Note channel of 0x0 will reset the counter value, so don't do it before
+ * we read out the counter.
+ */
+#define   NOC_PMU_EVENT_CTRL_CHANNEL	GENMASK(10, 8)
+#define   NOC_PMU_EVENT_CTRL_EN		BIT(11)
+#define NOC_PMU_EVENT_COUNTER(n)	(0x1e80 + 8 * (n))
+
+#define NOC_PMU_NR_COUNTERS		4
+#define NOC_PMU_COUNTER_BITS		64
+
+#define NOC_PMU_CH_DEFAULT	0x7
+
+HISI_PMU_EVENT_ATTR_EXTRACTOR(ch, config1, 2, 0);
+HISI_PMU_EVENT_ATTR_EXTRACTOR(tt_en, config1, 3, 3);
+
+/* Dynamic CPU hotplug state used by this PMU driver */
+static enum cpuhp_state hisi_noc_pmu_cpuhp_state;
+
+/*
+ * Tracetag filtering is not per event and all the events should keep
+ * the consistence. Return true if the new comer doesn't match the
+ * tracetag filtering configuration of the current scheduled events.
+ */
+static bool hisi_noc_pmu_check_global_filter(struct perf_event *curr,
+					     struct perf_event *new)
+{
+	return hisi_get_tt_en(curr) == hisi_get_tt_en(new);
+}
+
+static void hisi_noc_pmu_write_evtype(struct hisi_pmu *noc_pmu, int idx, u32 type)
+{
+	u32 reg;
+
+	reg = readl(noc_pmu->base + NOC_PMU_EVENT_CTRL(idx));
+	reg &= ~NOC_PMU_EVENT_CTRL_TYPE;
+	reg |= FIELD_PREP(NOC_PMU_EVENT_CTRL_TYPE, type);
+	writel(reg, noc_pmu->base + NOC_PMU_EVENT_CTRL(idx));
+}
+
+static int hisi_noc_pmu_get_event_idx(struct perf_event *event)
+{
+	struct hisi_pmu *noc_pmu = to_hisi_pmu(event->pmu);
+	struct hisi_pmu_hwevents *pmu_events = &noc_pmu->pmu_events;
+	int cur_idx;
+
+	cur_idx = find_first_bit(pmu_events->used_mask, noc_pmu->num_counters);
+	if (cur_idx != noc_pmu->num_counters &&
+	    !hisi_noc_pmu_check_global_filter(pmu_events->hw_events[cur_idx], event))
+		return -EAGAIN;
+
+	return hisi_uncore_pmu_get_event_idx(event);
+}
+
+static u64 hisi_noc_pmu_read_counter(struct hisi_pmu *noc_pmu,
+				     struct hw_perf_event *hwc)
+{
+	return readq(noc_pmu->base + NOC_PMU_EVENT_COUNTER(hwc->idx));
+}
+
+static void hisi_noc_pmu_write_counter(struct hisi_pmu *noc_pmu,
+				       struct hw_perf_event *hwc, u64 val)
+{
+	writeq(val, noc_pmu->base + NOC_PMU_EVENT_COUNTER(hwc->idx));
+}
+
+static void hisi_noc_pmu_enable_counter(struct hisi_pmu *noc_pmu,
+					struct hw_perf_event *hwc)
+{
+	u32 reg;
+
+	reg = readl(noc_pmu->base + NOC_PMU_EVENT_CTRL(hwc->idx));
+	reg |= NOC_PMU_EVENT_CTRL_EN;
+	writel(reg, noc_pmu->base + NOC_PMU_EVENT_CTRL(hwc->idx));
+}
+
+static void hisi_noc_pmu_disable_counter(struct hisi_pmu *noc_pmu,
+					 struct hw_perf_event *hwc)
+{
+	u32 reg;
+
+	reg = readl(noc_pmu->base + NOC_PMU_EVENT_CTRL(hwc->idx));
+	reg &= ~NOC_PMU_EVENT_CTRL_EN;
+	writel(reg, noc_pmu->base + NOC_PMU_EVENT_CTRL(hwc->idx));
+}
+
+static void hisi_noc_pmu_enable_counter_int(struct hisi_pmu *noc_pmu,
+					    struct hw_perf_event *hwc)
+{
+	/* We don't support interrupt, so a stub here. */
+}
+
+static void hisi_noc_pmu_disable_counter_int(struct hisi_pmu *noc_pmu,
+					     struct hw_perf_event *hwc)
+{
+}
+
+static void hisi_noc_pmu_start_counters(struct hisi_pmu *noc_pmu)
+{
+	u32 reg;
+
+	reg = readl(noc_pmu->base + NOC_PMU_GLOBAL_CTRL);
+	reg |= NOC_PMU_GLOBAL_CTRL_PMU_EN;
+	writel(reg, noc_pmu->base + NOC_PMU_GLOBAL_CTRL);
+}
+
+static void hisi_noc_pmu_stop_counters(struct hisi_pmu *noc_pmu)
+{
+	u32 reg;
+
+	reg = readl(noc_pmu->base + NOC_PMU_GLOBAL_CTRL);
+	reg &= ~NOC_PMU_GLOBAL_CTRL_PMU_EN;
+	writel(reg, noc_pmu->base + NOC_PMU_GLOBAL_CTRL);
+}
+
+static u32 hisi_noc_pmu_get_int_status(struct hisi_pmu *noc_pmu)
+{
+	return readl(noc_pmu->base + NOC_PMU_CNT_INFO);
+}
+
+static void hisi_noc_pmu_clear_int_status(struct hisi_pmu *noc_pmu, int idx)
+{
+	writel(~NOC_PMU_CNT_INFO_OVERFLOW(idx), noc_pmu->base + NOC_PMU_CNT_INFO);
+}
+
+static void hisi_noc_pmu_enable_filter(struct perf_event *event)
+{
+	struct hisi_pmu *noc_pmu = to_hisi_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	u32 tt_en = hisi_get_tt_en(event);
+	u32 ch = hisi_get_ch(event);
+	u32 reg;
+
+	if (!ch)
+		ch = NOC_PMU_CH_DEFAULT;
+
+	reg = readl(noc_pmu->base + NOC_PMU_EVENT_CTRL(hwc->idx));
+	reg &= ~NOC_PMU_EVENT_CTRL_CHANNEL;
+	reg |= FIELD_PREP(NOC_PMU_EVENT_CTRL_CHANNEL, ch);
+	writel(reg, noc_pmu->base + NOC_PMU_EVENT_CTRL(hwc->idx));
+
+	/*
+	 * Since tracetag filter applies to all the counters, don't touch it
+	 * if user doesn't specify it explicitly.
+	 */
+	if (tt_en) {
+		reg = readl(noc_pmu->base + NOC_PMU_GLOBAL_CTRL);
+		reg |= NOC_PMU_GLOBAL_CTRL_TT_EN;
+		writel(reg, noc_pmu->base + NOC_PMU_GLOBAL_CTRL);
+	}
+}
+
+static void hisi_noc_pmu_disable_filter(struct perf_event *event)
+{
+	struct hisi_pmu *noc_pmu = to_hisi_pmu(event->pmu);
+	u32 tt_en = hisi_get_tt_en(event);
+	u32 reg;
+
+	/*
+	 * If we're not the last counter, don't touch the global tracetag
+	 * configuration.
+	 */
+	if (bitmap_weight(noc_pmu->pmu_events.used_mask, noc_pmu->num_counters) > 1)
+		return;
+
+	if (tt_en) {
+		reg = readl(noc_pmu->base + NOC_PMU_GLOBAL_CTRL);
+		reg &= ~NOC_PMU_GLOBAL_CTRL_TT_EN;
+		writel(reg, noc_pmu->base + NOC_PMU_GLOBAL_CTRL);
+	}
+}
+
+static const struct hisi_uncore_ops hisi_uncore_noc_ops = {
+	.write_evtype		= hisi_noc_pmu_write_evtype,
+	.get_event_idx		= hisi_noc_pmu_get_event_idx,
+	.read_counter		= hisi_noc_pmu_read_counter,
+	.write_counter		= hisi_noc_pmu_write_counter,
+	.enable_counter		= hisi_noc_pmu_enable_counter,
+	.disable_counter	= hisi_noc_pmu_disable_counter,
+	.enable_counter_int	= hisi_noc_pmu_enable_counter_int,
+	.disable_counter_int	= hisi_noc_pmu_disable_counter_int,
+	.start_counters		= hisi_noc_pmu_start_counters,
+	.stop_counters		= hisi_noc_pmu_stop_counters,
+	.get_int_status		= hisi_noc_pmu_get_int_status,
+	.clear_int_status	= hisi_noc_pmu_clear_int_status,
+	.enable_filter		= hisi_noc_pmu_enable_filter,
+	.disable_filter		= hisi_noc_pmu_disable_filter,
+};
+
+static struct attribute *hisi_noc_pmu_format_attrs[] = {
+	HISI_PMU_FORMAT_ATTR(event, "config:0-7"),
+	HISI_PMU_FORMAT_ATTR(ch, "config1:0-2"),
+	HISI_PMU_FORMAT_ATTR(tt_en, "config1:3"),
+	NULL
+};
+
+static const struct attribute_group hisi_noc_pmu_format_group = {
+	.name = "format",
+	.attrs = hisi_noc_pmu_format_attrs,
+};
+
+static struct attribute *hisi_noc_pmu_events_attrs[] = {
+	HISI_PMU_EVENT_ATTR(cycles, 0x0e),
+	/* Flux on/off the ring */
+	HISI_PMU_EVENT_ATTR(ingress_flow_sum, 0x1a),
+	HISI_PMU_EVENT_ATTR(egress_flow_sum, 0x17),
+	/* Buffer full duration on/off the ring */
+	HISI_PMU_EVENT_ATTR(ingress_buf_full, 0x19),
+	HISI_PMU_EVENT_ATTR(egress_buf_full, 0x12),
+	/* Failure packets count on/off the ring */
+	HISI_PMU_EVENT_ATTR(cw_ingress_fail, 0x01),
+	HISI_PMU_EVENT_ATTR(cc_ingress_fail, 0x09),
+	HISI_PMU_EVENT_ATTR(cw_egress_fail, 0x03),
+	HISI_PMU_EVENT_ATTR(cc_egress_fail, 0x0b),
+	/* Flux of the ring */
+	HISI_PMU_EVENT_ATTR(cw_main_flow_sum, 0x05),
+	HISI_PMU_EVENT_ATTR(cc_main_flow_sum, 0x0d),
+	NULL
+};
+
+static const struct attribute_group hisi_noc_pmu_events_group = {
+	.name = "events",
+	.attrs = hisi_noc_pmu_events_attrs,
+};
+
+static const struct attribute_group *hisi_noc_pmu_attr_groups[] = {
+	&hisi_noc_pmu_format_group,
+	&hisi_noc_pmu_events_group,
+	&hisi_pmu_cpumask_attr_group,
+	&hisi_pmu_identifier_group,
+	NULL
+};
+
+static int hisi_noc_pmu_dev_init(struct platform_device *pdev, struct hisi_pmu *noc_pmu)
+{
+	hisi_uncore_pmu_init_topology(noc_pmu, &pdev->dev);
+
+	if (noc_pmu->topo.scl_id < 0)
+		return dev_err_probe(&pdev->dev, -EINVAL, "failed to get scl-id\n");
+
+	if (noc_pmu->topo.index_id < 0)
+		return dev_err_probe(&pdev->dev, -EINVAL, "failed to get idx-id\n");
+
+	if (noc_pmu->topo.sub_id < 0)
+		return dev_err_probe(&pdev->dev, -EINVAL, "failed to get sub-id\n");
+
+	noc_pmu->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(noc_pmu->base))
+		return dev_err_probe(&pdev->dev, PTR_ERR(noc_pmu->base),
+				     "fail to remap io memory\n");
+
+	noc_pmu->on_cpu = -1;
+	noc_pmu->dev = &pdev->dev;
+	noc_pmu->ops = &hisi_uncore_noc_ops;
+	noc_pmu->pmu_events.attr_groups = hisi_noc_pmu_attr_groups;
+	noc_pmu->num_counters = NOC_PMU_NR_COUNTERS;
+	noc_pmu->counter_bits = NOC_PMU_COUNTER_BITS;
+	noc_pmu->check_event = NOC_PMU_EVENT_CTRL_TYPE;
+	noc_pmu->identifier = readl(noc_pmu->base + NOC_PMU_VERSION);
+	return 0;
+}
+
+static void hisi_noc_pmu_remove_cpuhp_instance(void *hotplug_node)
+{
+	cpuhp_state_remove_instance_nocalls(hisi_noc_pmu_cpuhp_state, hotplug_node);
+}
+
+static void hisi_noc_pmu_unregister_pmu(void *pmu)
+{
+	perf_pmu_unregister(pmu);
+}
+
+static int hisi_noc_pmu_probe(struct platform_device *pdev)
+{
+	struct hisi_pmu *noc_pmu;
+	char *name;
+	int ret;
+
+	noc_pmu = devm_kzalloc(&pdev->dev, sizeof(*noc_pmu), GFP_KERNEL);
+	if (!noc_pmu)
+		return -ENOMEM;
+
+	/*
+	 * HiSilicon Uncore PMU framework needs to get common hisi_pmu device
+	 * by device's drvdata.
+	 */
+	platform_set_drvdata(pdev, noc_pmu);
+
+	ret = hisi_noc_pmu_dev_init(pdev, noc_pmu);
+	if (ret)
+		return ret;
+
+	ret = cpuhp_state_add_instance(hisi_noc_pmu_cpuhp_state, &noc_pmu->node);
+	if (ret)
+		return dev_err_probe(&pdev->dev, ret, "Fail to register cpuhp instance\n");
+
+	ret = devm_add_action_or_reset(&pdev->dev, hisi_noc_pmu_remove_cpuhp_instance,
+				       &noc_pmu->node);
+	if (ret)
+		return ret;
+
+	hisi_pmu_init(noc_pmu, THIS_MODULE);
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "hisi_scl%d_noc%d_%d",
+			      noc_pmu->topo.scl_id, noc_pmu->topo.index_id,
+			      noc_pmu->topo.sub_id);
+	if (!name)
+		return -ENOMEM;
+
+	ret = perf_pmu_register(&noc_pmu->pmu, name, -1);
+	if (ret)
+		return dev_err_probe(&pdev->dev, ret, "Fail to register PMU\n");
+
+	return devm_add_action_or_reset(&pdev->dev, hisi_noc_pmu_unregister_pmu,
+					&noc_pmu->pmu);
+}
+
+const struct acpi_device_id hisi_noc_pmu_ids[] = {
+	{ "HISI04E0", },
+	{ }
+};
+MODULE_DEVICE_TABLE(acpi, hisi_noc_pmu_ids);
+
+static struct platform_driver hisi_noc_pmu_driver = {
+	.driver = {
+		.name = "hisi_noc_pmu",
+		.acpi_match_table = hisi_noc_pmu_ids,
+		.suppress_bind_attrs = true,
+	},
+	.probe = hisi_noc_pmu_probe,
+};
+
+static int __init hisi_noc_pmu_module_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "perf/hisi/noc:online",
+				      hisi_uncore_pmu_online_cpu,
+				      hisi_uncore_pmu_offline_cpu);
+	if (ret < 0) {
+		pr_err("hisi_noc_pmu: Fail to setup cpuhp callbacks, ret = %d\n", ret);
+		return ret;
+	}
+	hisi_noc_pmu_cpuhp_state = ret;
+
+	ret = platform_driver_register(&hisi_noc_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(hisi_noc_pmu_cpuhp_state);
+
+	return ret;
+}
+module_init(hisi_noc_pmu_module_init);
+
+static void __exit hisi_noc_pmu_module_exit(void)
+{
+	platform_driver_unregister(&hisi_noc_pmu_driver);
+	cpuhp_remove_multi_state(hisi_noc_pmu_cpuhp_state);
+}
+module_exit(hisi_noc_pmu_module_exit);
+
+MODULE_IMPORT_NS("HISI_PMU");
+MODULE_DESCRIPTION("HiSilicon SoC Uncore NoC PMU driver");
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Yicong Yang <yangyicong@hisilicon.com>");
-- 
2.24.0

Re: [PATCH 8/9] drivers/perf: hisi: Add support for HiSilicon NoC PMU

[Jonathan Cameron]

On Tue, 18 Feb 2025 17:19:59 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> From: Yicong Yang <yangyicong@hisilicon.com>
> 
> Adds the support for HiSilicon NoC (Network on Chip) PMU which
> will be used to monitor the events on the system bus. The PMU
> device will be named after the SCL ID (either Super CPU cluster
> or Super IO cluster) and the index ID, just similar to other
> HiSilicon Uncore PMUs. Below PMU formats are provided besides
> the event:
> 
> - ch: the transaction channel (data, request, response, etc) which
>   can be used to filter the counting.
> - tt_en: tracetag filtering enable. Just as other HiSilicon Uncore
>   PMUs the NoC PMU supports only counting the transactions with
>   tracetag.
> 
> The NoC PMU doesn't have an interrupt to indicate the overflow.
> However we have a 64 bit counter which is large enough and it's
> nearly impossible to overflow.
> 
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
A few minor comments inline.
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
> ---
>  Documentation/admin-guide/perf/hisi-pmu.rst  |  11 +
>  drivers/perf/hisilicon/Makefile              |   3 +-
>  drivers/perf/hisilicon/hisi_uncore_noc_pmu.c | 391 +++++++++++++++++++
>  3 files changed, 404 insertions(+), 1 deletion(-)
>  create mode 100644 drivers/perf/hisilicon/hisi_uncore_noc_pmu.c
> 
> diff --git a/Documentation/admin-guide/perf/hisi-pmu.rst b/Documentation/admin-guide/perf/hisi-pmu.rst
> index 48992a0b8e94..d858bd3a8802 100644
> --- a/Documentation/admin-guide/perf/hisi-pmu.rst
> +++ b/Documentation/admin-guide/perf/hisi-pmu.rst
> @@ -112,6 +112,17 @@ uring channel. It is 2 bits. Some important codes are as follows:
>  - 2'b00: default value, count the events which sent to the both uring and
>    uring_ext channel;
>  
> +6. ch: NoC PMU supports to filter the event counts of certain transaction
supports filtering the event counts...
> +channel with this option. The current supported channels are as follows:
> +
> +- 3'b010: Request channel
> +- 3'b100: Snoop channel
> +- 3'b110: Response channel
> +- 3'b111: Data channel
> +
> +7. tt_en: NoC PMU supports to only count the transactions with tracetag by

supports counting only transactions that have tracetag set if this option is
set.

> +this option. See the 2nd list for more information about tracetag.
> +


> +static int hisi_noc_pmu_probe(struct platform_device *pdev)
> +{
> +	struct hisi_pmu *noc_pmu;
> +	char *name;
> +	int ret;
> +
> +	noc_pmu = devm_kzalloc(&pdev->dev, sizeof(*noc_pmu), GFP_KERNEL);
Quite a bit of use of &pdev->dev, maybe worth
struct device *dev = &pdev->dev; to shorten all these lines.

> +	if (!noc_pmu)
> +		return -ENOMEM;
> +
> +	/*
> +	 * HiSilicon Uncore PMU framework needs to get common hisi_pmu device
> +	 * by device's drvdata.

from device's drvdata

[PATCH 9/9] drivers/perf: hisi: Add support for HiSilicon MN PMU driver

[Yicong Yang]

From: Junhao He <hejunhao3@huawei.com>

MN (Miscellaneous Node) is a hybrid node in ARM CHI. It broadcasts the
following two types of requests: DVM operations and PCIe configuration.
MN PMU devices exist on both SCCL and SICL, so we named the MN pmu
driver after SCL (Super cluster) ID.
The MN PMU driver using the HiSilicon uncore PMU framework. And only
the event parameter is supported.

Signed-off-by: Junhao He <hejunhao3@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
---
 drivers/perf/hisilicon/Makefile             |   2 +-
 drivers/perf/hisilicon/hisi_uncore_mn_pmu.c | 355 ++++++++++++++++++++
 2 files changed, 356 insertions(+), 1 deletion(-)
 create mode 100644 drivers/perf/hisilicon/hisi_uncore_mn_pmu.c

diff --git a/drivers/perf/hisilicon/Makefile b/drivers/perf/hisilicon/Makefile
index dcec8f39719d..186be3d02238 100644
--- a/drivers/perf/hisilicon/Makefile
+++ b/drivers/perf/hisilicon/Makefile
@@ -2,7 +2,7 @@
 obj-$(CONFIG_HISI_PMU) += hisi_uncore_pmu.o hisi_uncore_l3c_pmu.o \
 			  hisi_uncore_hha_pmu.o hisi_uncore_ddrc_pmu.o hisi_uncore_sllc_pmu.o \
 			  hisi_uncore_pa_pmu.o hisi_uncore_cpa_pmu.o hisi_uncore_uc_pmu.o \
-			  hisi_uncore_noc_pmu.o
+			  hisi_uncore_noc_pmu.o hisi_uncore_mn_pmu.o
 
 obj-$(CONFIG_HISI_PCIE_PMU) += hisi_pcie_pmu.o
 obj-$(CONFIG_HNS3_PMU) += hns3_pmu.o
diff --git a/drivers/perf/hisilicon/hisi_uncore_mn_pmu.c b/drivers/perf/hisilicon/hisi_uncore_mn_pmu.c
new file mode 100644
index 000000000000..bf9b326e40cf
--- /dev/null
+++ b/drivers/perf/hisilicon/hisi_uncore_mn_pmu.c
@@ -0,0 +1,355 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * HiSilicon SoC MN uncore Hardware event counters support
+ *
+ * Copyright (c) 2025 HiSilicon Technologies Co., Ltd.
+ */
+#include <linux/cpuhotplug.h>
+#include <linux/interrupt.h>
+#include <linux/iopoll.h>
+#include <linux/irq.h>
+#include <linux/list.h>
+#include <linux/mod_devicetable.h>
+#include <linux/property.h>
+
+#include "hisi_uncore_pmu.h"
+
+/* Dynamic CPU hotplug state used by MN PMU */
+static enum cpuhp_state hisi_mn_pmu_online;
+
+/* MN register definition */
+#define HISI_MN_DYNAMIC_CTRL		0x400
+#define   HISI_MN_DYNAMIC_CTRL_EN	BIT(0)
+#define HISI_MN_PERF_CTRL_REG		0x408
+#define   HISI_MN_PERF_CTRL_EN		BIT(6)
+#define HISI_MN_INT_MASK_REG		0x800
+#define HISI_MN_INT_STATUS_REG		0x808
+#define HISI_MN_INT_CLEAR_REG		0x80C
+#define HISI_MN_EVENT_CTRL_REG		0x1C00
+#define HISI_MN_VERSION_REG		0x1C04
+#define HISI_MN_EVTYPE_REGn(n)		(0x1d00 + (n) * 4)
+#define   HISI_MN_EVTYPE_MASK		GENMASK(7, 0)
+#define HISI_MN_CNTR_REGn(n)		(0x1e00 + (n) * 8)
+
+#define HISI_MN_NR_COUNTERS		4
+#define HISI_MN_COUNTER_BITS		48
+#define HISI_MN_TIMEOUT_US		500U
+
+/*
+ * Each event request takes a certain amount of time to complete. If
+ * we counting the latency related event, we need to wait for the all
+ * requests complete. Otherwise, the value of counter is slightly larger.
+ */
+static void hisi_mn_pmu_counter_flush(struct hisi_pmu *mn_pmu)
+{
+	int ret;
+	u32 val;
+
+	val = readl(mn_pmu->base + HISI_MN_DYNAMIC_CTRL);
+	val |= HISI_MN_DYNAMIC_CTRL_EN;
+	writel(val, mn_pmu->base + HISI_MN_DYNAMIC_CTRL);
+
+	ret = readl_poll_timeout_atomic(mn_pmu->base + HISI_MN_DYNAMIC_CTRL,
+					val, !(val & HISI_MN_DYNAMIC_CTRL_EN),
+					1, HISI_MN_TIMEOUT_US);
+	if (ret)
+		dev_warn(mn_pmu->dev, "Counter flush timeout\n");
+}
+
+static u64 hisi_mn_pmu_read_counter(struct hisi_pmu *mn_pmu,
+				    struct hw_perf_event *hwc)
+{
+	return readq(mn_pmu->base + HISI_MN_CNTR_REGn(hwc->idx));
+}
+
+static void hisi_mn_pmu_write_counter(struct hisi_pmu *mn_pmu,
+				      struct hw_perf_event *hwc, u64 val)
+{
+	writeq(val, mn_pmu->base + HISI_MN_CNTR_REGn(hwc->idx));
+}
+
+static void hisi_mn_pmu_write_evtype(struct hisi_pmu *mn_pmu, int idx, u32 type)
+{
+	u32 val;
+
+	/*
+	 * Select the appropriate event select register.
+	 * There are 2 32-bit event select registers for the
+	 * 8 hardware counters, each event code is 8-bit wide.
+	 */
+	val = readl(mn_pmu->base + HISI_MN_EVTYPE_REGn(idx / 4));
+	val &= ~(HISI_MN_EVTYPE_MASK << HISI_PMU_EVTYPE_SHIFT(idx));
+	val |= (type << HISI_PMU_EVTYPE_SHIFT(idx));
+	writel(val, mn_pmu->base + HISI_MN_EVTYPE_REGn(idx / 4));
+}
+
+static void hisi_mn_pmu_start_counters(struct hisi_pmu *mn_pmu)
+{
+	u32 val;
+
+	val = readl(mn_pmu->base + HISI_MN_PERF_CTRL_REG);
+	val |= HISI_MN_PERF_CTRL_EN;
+	writel(val, mn_pmu->base + HISI_MN_PERF_CTRL_REG);
+}
+
+static void hisi_mn_pmu_stop_counters(struct hisi_pmu *mn_pmu)
+{
+	u32 val;
+
+	val = readl(mn_pmu->base + HISI_MN_PERF_CTRL_REG);
+	val &= ~HISI_MN_PERF_CTRL_EN;
+	writel(val, mn_pmu->base + HISI_MN_PERF_CTRL_REG);
+
+	hisi_mn_pmu_counter_flush(mn_pmu);
+}
+
+static void hisi_mn_pmu_enable_counter(struct hisi_pmu *mn_pmu,
+				       struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	val = readl(mn_pmu->base + HISI_MN_EVENT_CTRL_REG);
+	val |= BIT(hwc->idx);
+	writel(val, mn_pmu->base + HISI_MN_EVENT_CTRL_REG);
+}
+
+static void hisi_mn_pmu_disable_counter(struct hisi_pmu *mn_pmu,
+					struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	val = readl(mn_pmu->base + HISI_MN_EVENT_CTRL_REG);
+	val &= ~BIT(hwc->idx);
+	writel(val, mn_pmu->base + HISI_MN_EVENT_CTRL_REG);
+}
+
+static void hisi_mn_pmu_enable_counter_int(struct hisi_pmu *mn_pmu,
+					   struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	val = readl(mn_pmu->base + HISI_MN_INT_MASK_REG);
+	val &= ~BIT(hwc->idx);
+	writel(val, mn_pmu->base + HISI_MN_INT_MASK_REG);
+}
+
+static void hisi_mn_pmu_disable_counter_int(struct hisi_pmu *mn_pmu,
+					    struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	val = readl(mn_pmu->base + HISI_MN_INT_MASK_REG);
+	val |= BIT(hwc->idx);
+	writel(val, mn_pmu->base + HISI_MN_INT_MASK_REG);
+}
+
+static u32 hisi_mn_pmu_get_int_status(struct hisi_pmu *mn_pmu)
+{
+	return readl(mn_pmu->base + HISI_MN_INT_STATUS_REG);
+}
+
+static void hisi_mn_pmu_clear_int_status(struct hisi_pmu *mn_pmu, int idx)
+{
+	writel(BIT(idx), mn_pmu->base + HISI_MN_INT_CLEAR_REG);
+}
+
+static struct attribute *hisi_mn_pmu_format_attr[] = {
+	HISI_PMU_FORMAT_ATTR(event, "config:0-7"),
+	NULL
+};
+
+static const struct attribute_group hisi_mn_pmu_format_group = {
+	.name = "format",
+	.attrs = hisi_mn_pmu_format_attr,
+};
+
+static struct attribute *hisi_mn_pmu_events_attr[] = {
+	HISI_PMU_EVENT_ATTR(req_eobarrier_num,		0x00),
+	HISI_PMU_EVENT_ATTR(req_ecbarrier_num,		0x01),
+	HISI_PMU_EVENT_ATTR(req_dvmop_num,		0x02),
+	HISI_PMU_EVENT_ATTR(req_dvmsync_num,		0x03),
+	HISI_PMU_EVENT_ATTR(req_retry_num,		0x04),
+	HISI_PMU_EVENT_ATTR(req_writenosnp_num,		0x05),
+	HISI_PMU_EVENT_ATTR(req_readnosnp_num,		0x06),
+	HISI_PMU_EVENT_ATTR(snp_dvm_num,		0x07),
+	HISI_PMU_EVENT_ATTR(snp_dvmsync_num,		0x08),
+	HISI_PMU_EVENT_ATTR(l3t_req_dvm_num,		0x09),
+	HISI_PMU_EVENT_ATTR(l3t_req_dvmsync_num,	0x0A),
+	HISI_PMU_EVENT_ATTR(mn_req_dvm_num,		0x0B),
+	HISI_PMU_EVENT_ATTR(mn_req_dvmsync_num,		0x0C),
+	HISI_PMU_EVENT_ATTR(pa_req_dvm_num,		0x0D),
+	HISI_PMU_EVENT_ATTR(pa_req_dvmsync_num,		0x0E),
+	HISI_PMU_EVENT_ATTR(snp_dvm_latency,		0x80),
+	HISI_PMU_EVENT_ATTR(snp_dvmsync_latency,	0x81),
+	HISI_PMU_EVENT_ATTR(l3t_req_dvm_latency,	0x82),
+	HISI_PMU_EVENT_ATTR(l3t_req_dvmsync_latency,	0x83),
+	HISI_PMU_EVENT_ATTR(mn_req_dvm_latency,		0x84),
+	HISI_PMU_EVENT_ATTR(mn_req_dvmsync_latency,	0x85),
+	HISI_PMU_EVENT_ATTR(pa_req_dvm_latency,		0x86),
+	HISI_PMU_EVENT_ATTR(pa_req_dvmsync_latency,	0x87),
+	NULL
+};
+
+static const struct attribute_group hisi_mn_pmu_events_group = {
+	.name = "events",
+	.attrs = hisi_mn_pmu_events_attr,
+};
+
+static const struct attribute_group *hisi_mn_pmu_attr_groups[] = {
+	&hisi_mn_pmu_format_group,
+	&hisi_mn_pmu_events_group,
+	&hisi_pmu_cpumask_attr_group,
+	&hisi_pmu_identifier_group,
+	NULL
+};
+
+static const struct hisi_uncore_ops hisi_uncore_mn_ops = {
+	.write_evtype		= hisi_mn_pmu_write_evtype,
+	.get_event_idx		= hisi_uncore_pmu_get_event_idx,
+	.start_counters		= hisi_mn_pmu_start_counters,
+	.stop_counters		= hisi_mn_pmu_stop_counters,
+	.enable_counter		= hisi_mn_pmu_enable_counter,
+	.disable_counter	= hisi_mn_pmu_disable_counter,
+	.enable_counter_int	= hisi_mn_pmu_enable_counter_int,
+	.disable_counter_int	= hisi_mn_pmu_disable_counter_int,
+	.write_counter		= hisi_mn_pmu_write_counter,
+	.read_counter		= hisi_mn_pmu_read_counter,
+	.get_int_status		= hisi_mn_pmu_get_int_status,
+	.clear_int_status	= hisi_mn_pmu_clear_int_status,
+};
+
+static int hisi_mn_pmu_dev_init(struct platform_device *pdev,
+				struct hisi_pmu *mn_pmu)
+{
+	int ret;
+
+	hisi_uncore_pmu_init_topology(mn_pmu, &pdev->dev);
+
+	if (mn_pmu->topo.scl_id < 0)
+		return dev_err_probe(&pdev->dev, -EINVAL,
+				     "Failed to read MN scl id\n");
+
+	if (mn_pmu->topo.index_id < 0)
+		return dev_err_probe(&pdev->dev, -EINVAL,
+				     "Failed to read MN index id\n");
+
+	mn_pmu->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(mn_pmu->base))
+		return dev_err_probe(&pdev->dev, PTR_ERR(mn_pmu->base),
+				     "Failed to ioremap resource\n");
+
+	ret = hisi_uncore_pmu_init_irq(mn_pmu, pdev);
+	if (ret)
+		return ret;
+
+	mn_pmu->on_cpu = -1;
+	mn_pmu->dev = &pdev->dev;
+	mn_pmu->ops = &hisi_uncore_mn_ops;
+	mn_pmu->pmu_events.attr_groups = hisi_mn_pmu_attr_groups;
+	mn_pmu->check_event = HISI_MN_EVTYPE_MASK;
+	mn_pmu->num_counters = HISI_MN_NR_COUNTERS;
+	mn_pmu->counter_bits = HISI_MN_COUNTER_BITS;
+	mn_pmu->identifier = readl(mn_pmu->base + HISI_MN_VERSION_REG);
+	return 0;
+}
+
+static void hisi_mn_pmu_remove_cpuhp(void *hotplug_node)
+{
+	cpuhp_state_remove_instance_nocalls(hisi_mn_pmu_online, hotplug_node);
+}
+
+static void hisi_mn_pmu_unregister(void *pmu)
+{
+	perf_pmu_unregister(pmu);
+}
+
+static int hisi_mn_pmu_probe(struct platform_device *pdev)
+{
+	struct hisi_pmu *mn_pmu;
+	char *name;
+	int ret;
+
+	mn_pmu = devm_kzalloc(&pdev->dev, sizeof(*mn_pmu), GFP_KERNEL);
+	if (!mn_pmu)
+		return -ENOMEM;
+
+	platform_set_drvdata(pdev, mn_pmu);
+
+	ret = hisi_mn_pmu_dev_init(pdev, mn_pmu);
+	if (ret)
+		return ret;
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "hisi_scl%d_mn%d",
+				mn_pmu->topo.scl_id, mn_pmu->topo.index_id);
+	if (!name)
+		return -ENOMEM;
+
+	ret = cpuhp_state_add_instance(hisi_mn_pmu_online, &mn_pmu->node);
+	if (ret)
+		return dev_err_probe(&pdev->dev, ret, "Failed to register cpu hotplug\n");
+
+	ret = devm_add_action_or_reset(&pdev->dev, hisi_mn_pmu_remove_cpuhp, &mn_pmu->node);
+	if (ret)
+		return ret;
+
+	hisi_pmu_init(mn_pmu, THIS_MODULE);
+
+	ret = perf_pmu_register(&mn_pmu->pmu, name, -1);
+	if (ret)
+		return dev_err_probe(mn_pmu->dev, ret, "Failed to register MN PMU\n");
+
+	return devm_add_action_or_reset(&pdev->dev, hisi_mn_pmu_unregister, &mn_pmu->pmu);
+}
+
+static const struct acpi_device_id hisi_mn_pmu_acpi_match[] = {
+	{ "HISI0222", },
+	{ }
+};
+MODULE_DEVICE_TABLE(acpi, hisi_mn_pmu_acpi_match);
+
+static struct platform_driver hisi_mn_pmu_driver = {
+	.driver = {
+		.name = "hisi_mn_pmu",
+		.acpi_match_table = hisi_mn_pmu_acpi_match,
+		/*
+		 * We have not worked out a safe bind/unbind process,
+		 * Forcefully unbinding during sampling will lead to a
+		 * kernel panic, so this is not supported yet.
+		 */
+		.suppress_bind_attrs = true,
+	},
+	.probe = hisi_mn_pmu_probe,
+};
+
+static int __init hisi_mn_pmu_module_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "perf/hisi/mn:online",
+				      hisi_uncore_pmu_online_cpu,
+				      hisi_uncore_pmu_offline_cpu);
+	if (ret < 0) {
+		pr_err("hisi_mn_pmu: Failed to setup MN PMU hotplug: %d\n", ret);
+		return ret;
+	}
+	hisi_mn_pmu_online = ret;
+
+	ret = platform_driver_register(&hisi_mn_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(hisi_mn_pmu_online);
+
+	return ret;
+}
+module_init(hisi_mn_pmu_module_init);
+
+static void __exit hisi_mn_pmu_module_exit(void)
+{
+	platform_driver_unregister(&hisi_mn_pmu_driver);
+	cpuhp_remove_multi_state(hisi_mn_pmu_online);
+}
+module_exit(hisi_mn_pmu_module_exit);
+
+MODULE_IMPORT_NS("HISI_PMU");
+MODULE_DESCRIPTION("HiSilicon SoC MN uncore PMU driver");
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Junhao He <hejunhao3@huawei.com>");
-- 
2.24.0

Re: [PATCH 9/9] drivers/perf: hisi: Add support for HiSilicon MN PMU driver

[Jonathan Cameron]

On Tue, 18 Feb 2025 17:20:00 +0800
Yicong Yang <yangyicong@huawei.com> wrote:

> From: Junhao He <hejunhao3@huawei.com>
> 
> MN (Miscellaneous Node) is a hybrid node in ARM CHI. It broadcasts the
> following two types of requests: DVM operations and PCIe configuration.
> MN PMU devices exist on both SCCL and SICL, so we named the MN pmu
> driver after SCL (Super cluster) ID.
> The MN PMU driver using the HiSilicon uncore PMU framework. And only
> the event parameter is supported.
> 
> Signed-off-by: Junhao He <hejunhao3@huawei.com>
> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Short and sweet

Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>