LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [PATCH v8 0/2] drivers/perf: hisi: Add support for PCIe PMU
@ 2021-07-28  8:09 Qi Liu
  2021-07-28  8:09 ` [PATCH v8 1/2] docs: perf: Add description for HiSilicon PCIe PMU driver Qi Liu
  2021-07-28  8:09 ` [PATCH v8 2/2] drivers/perf: hisi: Add driver for HiSilicon PCIe PMU Qi Liu
  0 siblings, 2 replies; 6+ messages in thread
From: Qi Liu @ 2021-07-28  8:09 UTC (permalink / raw)
  To: will, mark.rutland, bhelgaas
  Cc: linux-pci, linux-arm-kernel, linux-kernel, linuxarm, zhangshaokun

This patchset adds support for HiSilicon PCIe Performance Monitoring
Unit(PMU). It is a PCIe Root Complex integrated End Point(RCiEP) device
added on Hip09. Each PCIe Core has a PMU RCiEP to monitor multi root
ports and all Endpoints downstream these root ports.

HiSilicon PCIe PMU is supported to collect performance data of PCIe bus,
such as: bandwidth, latency etc.

Changes since v7:
- Drop headerfile cpumask.h and cpuhotplug.h.
- Rename events in perf list: bw->flux, lat->delay, as driver doesn't
  process bandwidth and average latency data.
- Link: https://lore.kernel.org/linux-arm-kernel/1624532384-43002-1-git-send-email-liuqi115@huawei.com/

Changes since v6:
- Move the driver to drivers/perf/hisilicon.
- Treat content in PMU counter and ext_counter as different PMU events, and
  export them separately.
- Address the comments from Will and Krzysztof.
- Link: https://lore.kernel.org/linux-arm-kernel/1622467951-32114-1-git-send-email-liuqi115@huawei.com/

Changes since v5:
- Fix some errors when build under ARCH=xtensa.
- Link: https://lore.kernel.org/linux-arm-kernel/1621946795-14046-1-git-send-email-liuqi115@huawei.com/

Changes since v4:
- Replace irq_set_affinity_hint() with irq_set_affinity().
- Link: https://lore.kernel.org/linux-arm-kernel/1621417741-5229-1-git-send-email-liuqi115@huawei.com/

Changes since v3:
- Fix some warnings when build under 32bits architecture.
- Address the comments from John.
- Link: https://lore.kernel.org/linux-arm-kernel/1618490885-44612-1-git-send-email-liuqi115@huawei.com/

Changes since v2:
- Address the comments from John.
- Link: https://lore.kernel.org/linux-arm-kernel/1617959157-22956-1-git-send-email-liuqi115@huawei.com/

Changes since v1:
- Drop the internal Reviewed-by tag.
- Fix some build warnings when W=1.
- Link: https://lore.kernel.org/linux-arm-kernel/1617788943-52722-1-git-send-email-liuqi115@huawei.com/

Qi Liu (2):
  docs: perf: Add description for HiSilicon PCIe PMU driver
  drivers/perf: hisi: Add driver for HiSilicon PCIe PMU

 Documentation/admin-guide/perf/hisi-pcie-pmu.rst | 104 +++
 MAINTAINERS                                      |   2 +
 drivers/perf/hisilicon/Kconfig                   |   9 +
 drivers/perf/hisilicon/Makefile                  |   2 +
 drivers/perf/hisilicon/hisi_pcie_pmu.c           | 967 +++++++++++++++++++++++
 include/linux/cpuhotplug.h                       |   1 +
 6 files changed, 1085 insertions(+)
 create mode 100644 Documentation/admin-guide/perf/hisi-pcie-pmu.rst
 create mode 100644 drivers/perf/hisilicon/hisi_pcie_pmu.c

-- 
2.7.4


^ permalink raw reply	[flat|nested] 6+ messages in thread

* [PATCH v8 1/2] docs: perf: Add description for HiSilicon PCIe PMU driver
  2021-07-28  8:09 [PATCH v8 0/2] drivers/perf: hisi: Add support for PCIe PMU Qi Liu
@ 2021-07-28  8:09 ` Qi Liu
  2021-07-28  8:09 ` [PATCH v8 2/2] drivers/perf: hisi: Add driver for HiSilicon PCIe PMU Qi Liu
  1 sibling, 0 replies; 6+ messages in thread
From: Qi Liu @ 2021-07-28  8:09 UTC (permalink / raw)
  To: will, mark.rutland, bhelgaas
  Cc: linux-pci, linux-arm-kernel, linux-kernel, linuxarm, zhangshaokun

PCIe PMU Root Complex Integrated End Point(RCiEP) device is supported on
HiSilicon HIP09 platform. Document it to provide guidance on how to
use it.

Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Qi Liu <liuqi115@huawei.com>
---
 Documentation/admin-guide/perf/hisi-pcie-pmu.rst | 104 +++++++++++++++++++++++
 1 file changed, 104 insertions(+)
 create mode 100644 Documentation/admin-guide/perf/hisi-pcie-pmu.rst

diff --git a/Documentation/admin-guide/perf/hisi-pcie-pmu.rst b/Documentation/admin-guide/perf/hisi-pcie-pmu.rst
new file mode 100644
index 0000000..2084b53
--- /dev/null
+++ b/Documentation/admin-guide/perf/hisi-pcie-pmu.rst
@@ -0,0 +1,104 @@
+================================================
+HiSilicon PCIe Performance Monitoring Unit (PMU)
+================================================
+
+HiSilicon PCIe Performance Monitoring Unit (PMU) is a PCIe Root Complex
+integrated End Point (RCiEP) device. On Hip09, each PCIe Core has a PMU RCiEP
+to monitor multi Root Ports and all Endpoints downstream these Root Ports.
+
+HiSilicon PCIe PMU is supported to collect performance data of PCIe bus, such
+as: bandwidth, latency etc.
+
+
+HiSilicon PCIe PMU driver
+=========================
+
+The PCIe PMU driver registers a perf PMU with the name of its sicl-id and PCIe
+Core id.::
+
+  /sys/bus/event_source/hisi_pcie<sicl>_<core>
+
+PMU driver provides description of available events and filter options in sysfs,
+see /sys/bus/event_source/devices/hisi_pcie<sicl>_<core>.
+
+The "format" directory describes all formats of the config (events) and config1
+(filter options) fields of the perf_event_attr structure. The "events" directory
+describes all documented events shown in perf list.
+
+The "identifier" sysfs file allows users to identify the version of the
+PMU hardware device.
+
+The "bus" sysfs file allows users to get the bus number of Root Ports
+monitored by PMU.
+
+Example usage of perf::
+
+  $# perf list
+  hisi_pcie0_0/bw_rx/ [kernel PMU event]
+  ------------------------------------------
+
+  $# perf stat -e hisi_pcie0_0/bw_rx/ sleep 5
+
+The current driver does not support sampling. So "perf record" is unsupported.
+Also attach to a task is unsupported for PCIe PMU.
+
+Filter options
+--------------
+
+1. Target filter
+PMU could only monitor the performance of traffic downstream target Root Ports
+or downstream target Endpoint. PCIe PMU driver support "port" and "bdf"
+interfaces for users, and these two interfaces aren't supported at the same
+time.
+
+-port
+"port" filter can be used in all PCIe PMU events, target Root Port can be
+selected by configuring the 16-bits-bitmap "port". Multi ports can be selected
+for AP-layer-events, and only one port can be selected for TL/DL-layer-events.
+
+For example, if target Root Port is 0000:00:00.0 (x8 lanes), bit0 of bitmap
+should be set, port=0x1; if target Root Port is 0000:00:04.0 (x4 lanes),
+bit8 is set, port=0x100; if these two Root Ports are both monitored, port=0x101.
+
+Example usage of perf::
+
+  $# perf stat -e hisi_pcie0_0/bw_rx,port=0x1/ sleep 5
+
+-bdf
+
+"bdf" filter can only be used in bandwidth events, target Endpoint is selected
+by configuring BDF to "bdf". Counter only counts the bandwidth of message
+requested by target Endpoint.
+
+For example, "bdf=0x3900" means BDF of target Endpoint is 0000:39:00.0.
+
+Example usage of perf::
+
+  $# perf stat -e hisi_pcie0_0/bw_rx,bdf=0x3900/ sleep 5
+
+2. Trigger filter
+Event statistics start when the first time TLP length is greater/smaller
+than trigger condition. You can set the trigger condition by writing "trig_len",
+and set the trigger mode by writing "trig_mode". This filter can only be used
+in bandwidth events.
+
+For example, "trig_len=4" means trigger condition is 2^4 DW, "trig_mode=0"
+means statistics start when TLP length > trigger condition, "trig_mode=1"
+means start when TLP length < condition.
+
+Example usage of perf::
+
+  $# perf stat -e hisi_pcie0_0/bw_rx,trig_len=0x4,trig_mode=1/ sleep 5
+
+3. Threshold filter
+Counter counts when TLP length within the specified range. You can set the
+threshold by writing "thr_len", and set the threshold mode by writing
+"thr_mode". This filter can only be used in bandwidth events.
+
+For example, "thr_len=4" means threshold is 2^4 DW, "thr_mode=0" means
+counter counts when TLP length >= threshold, and "thr_mode=1" means counts
+when TLP length < threshold.
+
+Example usage of perf::
+
+  $# perf stat -e hisi_pcie0_0/bw_rx,thr_len=0x4,thr_mode=1/ sleep 5
-- 
2.7.4


^ permalink raw reply	[flat|nested] 6+ messages in thread

* [PATCH v8 2/2] drivers/perf: hisi: Add driver for HiSilicon PCIe PMU
  2021-07-28  8:09 [PATCH v8 0/2] drivers/perf: hisi: Add support for PCIe PMU Qi Liu
  2021-07-28  8:09 ` [PATCH v8 1/2] docs: perf: Add description for HiSilicon PCIe PMU driver Qi Liu
@ 2021-07-28  8:09 ` Qi Liu
  2021-08-02 10:03   ` Will Deacon
  1 sibling, 1 reply; 6+ messages in thread
From: Qi Liu @ 2021-07-28  8:09 UTC (permalink / raw)
  To: will, mark.rutland, bhelgaas
  Cc: linux-pci, linux-arm-kernel, linux-kernel, linuxarm, zhangshaokun

PCIe PMU Root Complex Integrated End Point(RCiEP) device is supported
to sample bandwidth, latency, buffer occupation etc.

Each PMU RCiEP device monitors multiple Root Ports, and each RCiEP is
registered as a PMU in /sys/bus/event_source/devices, so users can
select target PMU, and use filter to do further sets.

Filtering options contains:
event        - select the event.
subevent     - select the subevent.
port         - select target Root Ports. Information of Root Ports
               are shown under sysfs.
bdf          - select requester_id of target EP device.
trig_len     - set trigger condition for starting event statistics.
trigger_mode - set trigger mode. 0 means starting to statistic when
               bigger than trigger condition, and 1 means smaller.
thr_len      - set threshold for statistics.
thr_mode     - set threshold mode. 0 means count when bigger than
               threshold, and 1 means smaller.

Reviewed-by: John Garry <john.garry@huawei.com>
Signed-off-by: Qi Liu <liuqi115@huawei.com>
---
 MAINTAINERS                            |   2 +
 drivers/perf/hisilicon/Kconfig         |   9 +
 drivers/perf/hisilicon/Makefile        |   2 +
 drivers/perf/hisilicon/hisi_pcie_pmu.c | 966 +++++++++++++++++++++++++
 include/linux/cpuhotplug.h             |   1 +
 5 files changed, 980 insertions(+)
 create mode 100644 drivers/perf/hisilicon/hisi_pcie_pmu.c

diff --git a/MAINTAINERS b/MAINTAINERS
index 81e1edeceae4..8639c1c469d9 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -8228,8 +8228,10 @@ F:	Documentation/devicetree/bindings/misc/hisilicon-hikey-usb.yaml
 
 HISILICON PMU DRIVER
 M:	Shaokun Zhang <zhangshaokun@hisilicon.com>
+M:	Qi Liu <liuqi115@huawei.com>
 S:	Supported
 W:	http://www.hisilicon.com
+F:	Documentation/admin-guide/perf/hisi-pcie-pmu.rst
 F:	Documentation/admin-guide/perf/hisi-pmu.rst
 F:	drivers/perf/hisilicon
 
diff --git a/drivers/perf/hisilicon/Kconfig b/drivers/perf/hisilicon/Kconfig
index c5d1b7019fff..5546218b5598 100644
--- a/drivers/perf/hisilicon/Kconfig
+++ b/drivers/perf/hisilicon/Kconfig
@@ -5,3 +5,12 @@ config HISI_PMU
 	  help
 	  Support for HiSilicon SoC L3 Cache performance monitor, Hydra Home
 	  Agent performance monitor and DDR Controller performance monitor.
+
+config HISI_PCIE_PMU
+	tristate "HiSilicon PCIE PERF PMU"
+	depends on PCI && ARM64
+	help
+	  Provide support for HiSilicon PCIe performance monitoring unit (PMU)
+	  RCiEP devices.
+	  Adds the PCIe PMU into perf events system for monitoring latency,
+	  bandwidth etc.
diff --git a/drivers/perf/hisilicon/Makefile b/drivers/perf/hisilicon/Makefile
index 7643c9f93e36..506ed39e3266 100644
--- a/drivers/perf/hisilicon/Makefile
+++ b/drivers/perf/hisilicon/Makefile
@@ -2,3 +2,5 @@
 obj-$(CONFIG_HISI_PMU) += hisi_uncore_pmu.o hisi_uncore_l3c_pmu.o \
 			  hisi_uncore_hha_pmu.o hisi_uncore_ddrc_pmu.o hisi_uncore_sllc_pmu.o \
 			  hisi_uncore_pa_pmu.o
+
+obj-$(CONFIG_HISI_PCIE_PMU) += hisi_pcie_pmu.o
diff --git a/drivers/perf/hisilicon/hisi_pcie_pmu.c b/drivers/perf/hisilicon/hisi_pcie_pmu.c
new file mode 100644
index 000000000000..b7b566abbcd8
--- /dev/null
+++ b/drivers/perf/hisilicon/hisi_pcie_pmu.c
@@ -0,0 +1,966 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * This driver adds support for PCIe PMU RCiEP device. Related
+ * perf events are bandwidth, bandwidth utilization, latency
+ * etc.
+ *
+ * Copyright (C) 2021 HiSilicon Limited
+ * Author: Qi Liu <liuqi115@huawei.com>
+ */
+#include <linux/bitfield.h>
+#include <linux/bitmap.h>
+#include <linux/bug.h>
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/perf_event.h>
+
+/* Define registers */
+#define HISI_PCIE_GLOBAL_CTRL		0x00
+#define HISI_PCIE_EVENT_CTRL		0x010
+#define HISI_PCIE_CNT			0x090
+#define HISI_PCIE_EXT_CNT		0x110
+#define HISI_PCIE_INT_STAT		0x150
+#define HISI_PCIE_INT_MASK		0x154
+#define HISI_PCIE_REG_BDF		0xfe0
+#define HISI_PCIE_REG_VERSION		0xfe4
+#define HISI_PCIE_REG_INFO		0xfe8
+
+/* Define PCIE CTRL CMD */
+#define HISI_PCIE_GLOBAL_EN		0x01
+#define HISI_PCIE_GLOBAL_NONE		0
+#define HISI_PCIE_EVENT_EN		BIT_ULL(20)
+#define HISI_PCIE_RESET_CNT		BIT_ULL(22)
+#define HISI_PCIE_DEFAULT_SET		BIT_ULL(34)
+#define HISI_PCIE_THR_EN		BIT_ULL(26)
+#define HISI_PCIE_TARGET_EN		BIT_ULL(32)
+#define HISI_PCIE_TRIG_EN		BIT_ULL(52)
+
+/* Define offsets in event ctrl regesiter */
+#define HISI_PCIE_EVENT_M		GENMASK_ULL(7, 0)
+#define HISI_PCIE_SUBEVENT_M		GENMASK_ULL(15, 8)
+#define HISI_PCIE_THR_MODE_M		GENMASK_ULL(27, 27)
+#define HISI_PCIE_THR_M			GENMASK_ULL(31, 28)
+#define HISI_PCIE_TARGET_M		GENMASK_ULL(52, 36)
+#define HISI_PCIE_TRIG_MODE_M		GENMASK_ULL(53, 53)
+#define HISI_PCIE_TRIG_M		GENMASK_ULL(59, 56)
+
+#define HISI_PCIE_MAX_COUNTERS		8
+#define HISI_PCIE_REG_STEP		8
+#define HISI_PCIE_EVENT_MAX		0xa2
+#define HISI_PCIE_SUBEVENT_MAX		0x20
+#define HISI_PCIE_THR_MAX_VAL		10
+#define HISI_PCIE_TRIG_MAX_VAL		10
+#define HISI_PCIE_MAX_PERIOD		(GENMASK_ULL(63, 0))
+#define HISI_PCIE_DEFAULT_VALUE		BIT_ULL(63)
+
+struct hisi_pcie_pmu {
+	struct perf_event *hw_events[HISI_PCIE_MAX_COUNTERS];
+	struct hlist_node node;
+	struct pci_dev *pdev;
+	struct pmu pmu;
+	void __iomem *base;
+	int irq;
+	u32 identifier;
+	/* Minimum and maximum bdf of root ports monitored by PMU */
+	u16 bdf_min;
+	u16 bdf_max;
+	int on_cpu;
+};
+
+struct hisi_pcie_reg_pair {
+	u16 lo;
+	u16 hi;
+};
+
+#define to_pcie_pmu(p)  (container_of((p), struct hisi_pcie_pmu, pmu))
+#define GET_PCI_DEVFN(bdf)  ((bdf) & 0xff)
+#define EXT_COUNTER_IS_USED(idx)		((idx) & GENMASK(15, 8))
+
+#define HISI_PCIE_PMU_FILTER_ATTR(_name, _config, _hi, _lo)		  \
+	static u64 hisi_pcie_get_##_name(struct perf_event *event)	  \
+	{								  \
+		return FIELD_GET(GENMASK(_hi, _lo), event->attr._config); \
+	}								  \
+
+HISI_PCIE_PMU_FILTER_ATTR(event, config, 15, 0);
+HISI_PCIE_PMU_FILTER_ATTR(subevent, config, 23, 16);
+HISI_PCIE_PMU_FILTER_ATTR(thr_len, config1, 3, 0);
+HISI_PCIE_PMU_FILTER_ATTR(thr_mode, config1, 4, 4);
+HISI_PCIE_PMU_FILTER_ATTR(trig_len, config1, 8, 5);
+HISI_PCIE_PMU_FILTER_ATTR(trig_mode, config1, 9, 9);
+HISI_PCIE_PMU_FILTER_ATTR(port, config2, 15, 0);
+HISI_PCIE_PMU_FILTER_ATTR(bdf, config2, 31, 16);
+
+static ssize_t hisi_pcie_format_sysfs_show(struct device *dev,
+				    struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr;
+
+	eattr = container_of(attr, struct dev_ext_attribute, attr);
+
+	return sysfs_emit(buf, "%s\n", (char *)eattr->var);
+}
+
+static ssize_t hisi_pcie_event_sysfs_show(struct device *dev,
+				   struct device_attribute *attr, char *buf)
+{
+	struct perf_pmu_events_attr *pmu_attr =
+		container_of(attr, struct perf_pmu_events_attr, attr);
+
+	return sysfs_emit(buf, "config=0x%llx\n", pmu_attr->id);
+}
+
+#define HISI_PCIE_PMU_FORMAT_ATTR(_name, _format)                              \
+	(&((struct dev_ext_attribute[]){                                       \
+		{ .attr = __ATTR(_name, 0444, hisi_pcie_format_sysfs_show,     \
+				 NULL),                                        \
+		  .var = (void *)_format }                                     \
+	})[0].attr.attr)
+
+#define HISI_PCIE_PMU_EVENT_ATTR(_name, _id)			\
+	PMU_EVENT_ATTR_ID(_name, hisi_pcie_event_sysfs_show, _id)
+
+static ssize_t cpumask_show(struct device *dev, struct device_attribute *attr,
+			    char *buf)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(dev_get_drvdata(dev));
+
+	return cpumap_print_to_pagebuf(true, buf, cpumask_of(pcie_pmu->on_cpu));
+}
+static DEVICE_ATTR_RO(cpumask);
+
+static ssize_t identifier_show(struct device *dev,
+			       struct device_attribute *attr, char *buf)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(dev_get_drvdata(dev));
+
+	return sysfs_emit(buf, "%#x\n", pcie_pmu->identifier);
+}
+static DEVICE_ATTR_RO(identifier);
+
+static ssize_t bus_show(struct device *dev, struct device_attribute *attr,
+			char *buf)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(dev_get_drvdata(dev));
+
+	return sysfs_emit(buf, "%#04x\n", PCI_BUS_NUM(pcie_pmu->bdf_min));
+}
+static DEVICE_ATTR_RO(bus);
+
+static struct hisi_pcie_reg_pair
+hisi_pcie_parse_reg_value(struct hisi_pcie_pmu *pcie_pmu, u32 reg_off)
+{
+	u32 val = readl_relaxed(pcie_pmu->base + reg_off);
+	struct hisi_pcie_reg_pair regs = {
+		.lo = val,
+		.hi = val >> 16,
+	};
+
+	return regs;
+}
+
+/*
+ * Hardware counter and ext_counter work together for bandwidth, latency,
+ * bus utilization and buffer occupancy events. For example, for latency
+ * events(idx = 0x10), counter counts total delay cycles and ext_counter
+ * counts PCIe packets number.
+ *
+ * As we don't want PMU driver to process these two data, "delay cycles"
+ * can be treated as an event(id = 0x10), "packets number" as another event
+ * (id = 0x10 << 8), and driver could export these data separately. So in
+ * this function, the "real" event idx is get to set HISI_PCIE_EVENT_CTRL.
+ */
+static u32 hisi_pcie_pmu_get_real_idx(struct perf_event *event)
+{
+	u32 event_idx = hisi_pcie_get_event(event);
+
+	if (!EXT_COUNTER_IS_USED(event_idx))
+		return event_idx;
+
+	return FIELD_GET(GENMASK(15, 8), event_idx);
+}
+
+static u32 hisi_pcie_pmu_get_offset(u32 offset, u32 idx)
+{
+	return offset + HISI_PCIE_REG_STEP * idx;
+}
+
+static u32 hisi_pcie_pmu_readl(struct hisi_pcie_pmu *pcie_pmu, u32 reg_offset,
+			       u32 idx)
+{
+	u32 offset = hisi_pcie_pmu_get_offset(reg_offset, idx);
+
+	return readl_relaxed(pcie_pmu->base + offset);
+}
+
+static void hisi_pcie_pmu_writel(struct hisi_pcie_pmu *pcie_pmu, u32 reg_offset,
+				 u32 idx, u32 val)
+{
+	u32 offset = hisi_pcie_pmu_get_offset(reg_offset, idx);
+
+	writel_relaxed(val, pcie_pmu->base + offset);
+}
+
+static u64 hisi_pcie_pmu_readq(struct hisi_pcie_pmu *pcie_pmu, u32 reg_offset,
+			       u32 idx)
+{
+	u32 offset = hisi_pcie_pmu_get_offset(reg_offset, idx);
+
+	return readq_relaxed(pcie_pmu->base + offset);
+}
+
+static void hisi_pcie_pmu_writeq(struct hisi_pcie_pmu *pcie_pmu, u32 reg_offset,
+				 u32 idx, u64 val)
+{
+	u32 offset = hisi_pcie_pmu_get_offset(reg_offset, idx);
+
+	writeq_relaxed(val, pcie_pmu->base + offset);
+}
+
+static void hisi_pcie_pmu_config_filter(struct perf_event *event)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+	u32 event_idx = hisi_pcie_pmu_get_real_idx(event);
+	struct hw_perf_event *hwc = &event->hw;
+	u64 reg = HISI_PCIE_DEFAULT_SET;
+	u64 port, trig_len, thr_len;
+	u32 idx = hwc->idx;
+
+	/* Config HISI_PCIE_EVENT_CTRL according to event and subevent. */
+	reg |= FIELD_PREP(HISI_PCIE_EVENT_M, event_idx) |
+	       FIELD_PREP(HISI_PCIE_SUBEVENT_M, hisi_pcie_get_subevent(event));
+
+	/* Config HISI_PCIE_EVENT_CTRL according to ROOT PORT or EP device. */
+	port = hisi_pcie_get_port(event);
+	if (port)
+		reg |= FIELD_PREP(HISI_PCIE_TARGET_M, port);
+	else
+		reg |= HISI_PCIE_TARGET_EN |
+		       FIELD_PREP(HISI_PCIE_TARGET_M, hisi_pcie_get_bdf(event));
+
+	/* Config HISI_PCIE_EVENT_CTRL according to trigger condition. */
+	trig_len = hisi_pcie_get_trig_len(event);
+	if (trig_len) {
+		reg |= FIELD_PREP(HISI_PCIE_TRIG_M, trig_len);
+		reg |= FIELD_PREP(HISI_PCIE_TRIG_MODE_M,
+				  hisi_pcie_get_trig_mode(event));
+		reg |= HISI_PCIE_TRIG_EN;
+	}
+
+	/* Config HISI_PCIE_EVENT_CTRL according to threshold condition. */
+	thr_len = hisi_pcie_get_thr_len(event);
+	if (thr_len) {
+		reg |= FIELD_PREP(HISI_PCIE_THR_M, thr_len);
+		reg |= FIELD_PREP(HISI_PCIE_THR_MODE_M,
+		       hisi_pcie_get_thr_mode(event));
+		reg |= HISI_PCIE_THR_EN;
+	}
+
+	hisi_pcie_pmu_writeq(pcie_pmu, HISI_PCIE_EVENT_CTRL, idx, reg);
+}
+
+static void hisi_pcie_pmu_clear_filter(struct perf_event *event)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	hisi_pcie_pmu_writeq(pcie_pmu, HISI_PCIE_EVENT_CTRL, hwc->idx,
+			     HISI_PCIE_DEFAULT_SET);
+}
+
+static bool hisi_pcie_pmu_valid_port(struct hisi_pcie_pmu *pcie_pmu, u16 rp_bdf)
+{
+	return rp_bdf >= pcie_pmu->bdf_min && rp_bdf <= pcie_pmu->bdf_max;
+}
+
+static bool hisi_pcie_pmu_valid_requester_id(struct hisi_pcie_pmu *pcie_pmu,
+					     u32 bdf)
+{
+	struct pci_dev *root_port, *pdev;
+
+	pdev = pci_get_domain_bus_and_slot(pci_domain_nr(pcie_pmu->pdev->bus),
+					   PCI_BUS_NUM(bdf),
+					   GET_PCI_DEVFN(bdf));
+	if (!pdev)
+		return false;
+
+	root_port = pcie_find_root_port(pdev);
+	if (!root_port)
+		return false;
+
+	pci_dev_put(pdev);
+	return hisi_pcie_pmu_valid_port(pcie_pmu, pci_dev_id(root_port));
+}
+
+static bool hisi_pcie_pmu_valid_filter(struct perf_event *event,
+				       struct hisi_pcie_pmu *pcie_pmu)
+{
+	u32 event_idx = hisi_pcie_pmu_get_real_idx(event);
+	u32 subev_idx = hisi_pcie_get_subevent(event);
+	u32 requester_id = hisi_pcie_get_bdf(event);
+
+	if (subev_idx > HISI_PCIE_SUBEVENT_MAX ||
+	    event_idx > HISI_PCIE_EVENT_MAX)
+		return false;
+
+	if (hisi_pcie_get_thr_len(event) > HISI_PCIE_THR_MAX_VAL)
+		return false;
+
+	if (hisi_pcie_get_trig_len(event) > HISI_PCIE_TRIG_MAX_VAL)
+		return false;
+
+	if (requester_id) {
+		if (!hisi_pcie_pmu_valid_requester_id(pcie_pmu, requester_id))
+			return false;
+	}
+
+	return true;
+}
+
+static bool hisi_pcie_pmu_validate_event_group(struct perf_event *event)
+{
+	struct perf_event *sibling, *leader = event->group_leader;
+	int counters = 1;
+
+	if (!is_software_event(leader)) {
+		if (leader->pmu != event->pmu)
+			return false;
+
+		if (leader != event)
+			counters++;
+	}
+
+	for_each_sibling_event(sibling, event->group_leader) {
+		if (is_software_event(sibling))
+			continue;
+
+		if (sibling->pmu != event->pmu)
+			return false;
+
+		counters++;
+	}
+
+	return counters <= HISI_PCIE_MAX_COUNTERS;
+}
+
+static int hisi_pcie_pmu_event_init(struct perf_event *event)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+
+	event->cpu = pcie_pmu->on_cpu;
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/* Sampling is not supported. */
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EOPNOTSUPP;
+
+	if (!hisi_pcie_pmu_valid_filter(event, pcie_pmu)) {
+		pci_err(pcie_pmu->pdev, "Invalid filter!\n");
+		return -EINVAL;
+	}
+
+	if (!hisi_pcie_pmu_validate_event_group(event))
+		return -EINVAL;
+
+	return 0;
+}
+
+static u64 hisi_pcie_pmu_read_counter(struct perf_event *event)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+	u32 event_idx = hisi_pcie_get_event(event);
+	u32 idx = event->hw.idx;
+
+	if (EXT_COUNTER_IS_USED(event_idx))
+		return hisi_pcie_pmu_readq(pcie_pmu, HISI_PCIE_EXT_CNT, idx);
+
+	return hisi_pcie_pmu_readq(pcie_pmu, HISI_PCIE_CNT, idx);
+}
+
+static void hisi_pcie_pmu_write_counter(struct perf_event *event, u64 val)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+	u32 event_idx = hisi_pcie_get_event(event);
+	u32 idx = event->hw.idx;
+
+	if (EXT_COUNTER_IS_USED(event_idx))
+		hisi_pcie_pmu_writeq(pcie_pmu, HISI_PCIE_EXT_CNT, idx, val);
+	else
+		hisi_pcie_pmu_writeq(pcie_pmu, HISI_PCIE_CNT, idx, val);
+}
+
+static int hisi_pcie_pmu_get_event_idx(struct hisi_pcie_pmu *pcie_pmu)
+{
+	int idx;
+
+	for (idx = 0; idx < HISI_PCIE_MAX_COUNTERS; idx++) {
+		if (!pcie_pmu->hw_events[idx])
+			return idx;
+	}
+
+	return -EINVAL;
+}
+
+static void hisi_pcie_pmu_event_update(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u64 new_cnt, prev_cnt, delta;
+
+	do {
+		prev_cnt = local64_read(&hwc->prev_count);
+		new_cnt = hisi_pcie_pmu_read_counter(event);
+	} while (local64_cmpxchg(&hwc->prev_count, prev_cnt,
+				 new_cnt) != prev_cnt);
+
+	delta = (new_cnt - prev_cnt) & HISI_PCIE_MAX_PERIOD;
+	local64_add(delta, &event->count);
+}
+
+static void hisi_pcie_pmu_read(struct perf_event *event)
+{
+	hisi_pcie_pmu_event_update(event);
+}
+
+static void hisi_pcie_pmu_set_period(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	local64_set(&hwc->prev_count, HISI_PCIE_DEFAULT_VALUE);
+	hisi_pcie_pmu_write_counter(event, HISI_PCIE_DEFAULT_VALUE);
+}
+
+static void hisi_pcie_pmu_enable_counter(struct hisi_pcie_pmu *pcie_pmu,
+					 struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+	u64 val;
+
+	val = hisi_pcie_pmu_readq(pcie_pmu, HISI_PCIE_EVENT_CTRL, idx);
+	val |= HISI_PCIE_EVENT_EN;
+	hisi_pcie_pmu_writeq(pcie_pmu, HISI_PCIE_EVENT_CTRL, idx, val);
+}
+
+static void hisi_pcie_pmu_disable_counter(struct hisi_pcie_pmu *pcie_pmu,
+					  struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+	u64 val;
+
+	val = hisi_pcie_pmu_readq(pcie_pmu, HISI_PCIE_EVENT_CTRL, idx);
+	val &= ~HISI_PCIE_EVENT_EN;
+	hisi_pcie_pmu_writeq(pcie_pmu, HISI_PCIE_EVENT_CTRL, idx, val);
+}
+
+static void hisi_pcie_pmu_enable_int(struct hisi_pcie_pmu *pcie_pmu,
+				     struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+
+	hisi_pcie_pmu_writel(pcie_pmu, HISI_PCIE_INT_MASK, idx, 0);
+}
+
+static void hisi_pcie_pmu_disable_int(struct hisi_pcie_pmu *pcie_pmu,
+				      struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+
+	hisi_pcie_pmu_writel(pcie_pmu, HISI_PCIE_INT_MASK, idx, 1);
+}
+
+static void hisi_pcie_pmu_reset_counter(struct hisi_pcie_pmu *pcie_pmu,
+					struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+
+	hisi_pcie_pmu_writeq(pcie_pmu, HISI_PCIE_EVENT_CTRL, idx,
+			     HISI_PCIE_RESET_CNT);
+	hisi_pcie_pmu_writeq(pcie_pmu, HISI_PCIE_EVENT_CTRL, idx,
+			     HISI_PCIE_DEFAULT_SET);
+}
+
+static void hisi_pcie_pmu_start(struct perf_event *event, int flags)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	u64 prev_cnt;
+
+	if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
+		return;
+
+	WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+	hwc->state = 0;
+
+	hisi_pcie_pmu_config_filter(event);
+	hisi_pcie_pmu_enable_counter(pcie_pmu, hwc);
+	hisi_pcie_pmu_enable_int(pcie_pmu, hwc);
+	hisi_pcie_pmu_set_period(event);
+
+	if (flags & PERF_EF_RELOAD) {
+		prev_cnt = local64_read(&hwc->prev_count);
+		hisi_pcie_pmu_write_counter(event, prev_cnt);
+	}
+
+	perf_event_update_userpage(event);
+}
+
+static void hisi_pcie_pmu_stop(struct perf_event *event, int flags)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	hisi_pcie_pmu_event_update(event);
+	hisi_pcie_pmu_disable_int(pcie_pmu, hwc);
+	hisi_pcie_pmu_disable_counter(pcie_pmu, hwc);
+	hisi_pcie_pmu_clear_filter(event);
+	WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+	hwc->state |= PERF_HES_STOPPED;
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	hwc->state |= PERF_HES_UPTODATE;
+}
+
+static int hisi_pcie_pmu_add(struct perf_event *event, int flags)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx;
+
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	idx = hisi_pcie_pmu_get_event_idx(pcie_pmu);
+	if (idx < 0)
+		return idx;
+
+	hwc->idx = idx;
+	pcie_pmu->hw_events[idx] = event;
+
+	/* Reset Counter to avoid interference caused by previous statistic. */
+	hisi_pcie_pmu_reset_counter(pcie_pmu, hwc);
+
+	if (flags & PERF_EF_START)
+		hisi_pcie_pmu_start(event, PERF_EF_RELOAD);
+
+	return 0;
+}
+
+static void hisi_pcie_pmu_del(struct perf_event *event, int flags)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	hisi_pcie_pmu_stop(event, PERF_EF_UPDATE);
+	pcie_pmu->hw_events[hwc->idx] = NULL;
+	perf_event_update_userpage(event);
+}
+
+static void hisi_pcie_pmu_enable(struct pmu *pmu)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(pmu);
+	int num;
+
+	for (num = 0; num < HISI_PCIE_MAX_COUNTERS; num++) {
+		if (pcie_pmu->hw_events[num])
+			break;
+	}
+
+	if (num == HISI_PCIE_MAX_COUNTERS)
+		return;
+
+	writel(HISI_PCIE_GLOBAL_EN, pcie_pmu->base + HISI_PCIE_GLOBAL_CTRL);
+}
+
+static void hisi_pcie_pmu_disable(struct pmu *pmu)
+{
+	struct hisi_pcie_pmu *pcie_pmu = to_pcie_pmu(pmu);
+
+	writel(HISI_PCIE_GLOBAL_NONE, pcie_pmu->base + HISI_PCIE_GLOBAL_CTRL);
+}
+
+static irqreturn_t hisi_pcie_pmu_irq(int irq, void *data)
+{
+	struct hisi_pcie_pmu *pcie_pmu = data;
+	irqreturn_t ret = IRQ_NONE;
+	struct perf_event *event;
+	u32 overflown;
+	int idx;
+
+	for (idx = 0; idx < HISI_PCIE_MAX_COUNTERS; idx++) {
+		overflown = hisi_pcie_pmu_readl(pcie_pmu, HISI_PCIE_INT_STAT,
+						idx);
+		if (!overflown)
+			continue;
+
+		/* Clear status of interrupt. */
+		hisi_pcie_pmu_writel(pcie_pmu, HISI_PCIE_INT_STAT, idx, 1);
+		event = pcie_pmu->hw_events[idx];
+		if (!event)
+			continue;
+
+		hisi_pcie_pmu_event_update(event);
+		hisi_pcie_pmu_set_period(event);
+		ret = IRQ_HANDLED;
+	}
+
+	return ret;
+}
+
+static int hisi_pcie_pmu_irq_register(struct pci_dev *pdev,
+				      struct hisi_pcie_pmu *pcie_pmu)
+{
+	int irq, ret;
+
+	ret = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI);
+	if (ret < 0) {
+		pci_err(pdev, "Failed to enable MSI vectors: %d!\n", ret);
+		return ret;
+	}
+
+	irq = pci_irq_vector(pdev, 0);
+	ret = request_irq(irq, hisi_pcie_pmu_irq,
+			  IRQF_NOBALANCING | IRQF_NO_THREAD, "hisi_pcie_pmu",
+			  pcie_pmu);
+	if (ret) {
+		pci_err(pdev, "Failed to register IRQ: %d!\n", ret);
+		pci_free_irq_vectors(pdev);
+		return ret;
+	}
+
+	pcie_pmu->irq = irq;
+
+	return 0;
+}
+
+static void hisi_pcie_pmu_irq_unregister(struct pci_dev *pdev,
+					 struct hisi_pcie_pmu *pcie_pmu)
+{
+	free_irq(pcie_pmu->irq, pcie_pmu);
+	pci_free_irq_vectors(pdev);
+}
+
+static int hisi_pcie_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct hisi_pcie_pmu *pcie_pmu = hlist_entry_safe(node,
+					 struct hisi_pcie_pmu, node);
+
+	if (pcie_pmu->on_cpu == -1) {
+		pcie_pmu->on_cpu = cpu;
+		WARN_ON(irq_set_affinity(pcie_pmu->irq, cpumask_of(cpu)));
+	}
+
+	return 0;
+}
+
+static int hisi_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct hisi_pcie_pmu *pcie_pmu = hlist_entry_safe(node,
+					 struct hisi_pcie_pmu, node);
+	unsigned int target;
+
+	/* Nothing to do if this CPU doesn't own the PMU */
+	if (pcie_pmu->on_cpu != cpu)
+		return 0;
+
+	pcie_pmu->on_cpu = -1;
+	/* Choose a new CPU from all online cpus. */
+	target = cpumask_first(cpu_online_mask);
+	if (target >= nr_cpu_ids) {
+		pci_err(pcie_pmu->pdev, "There is no CPU to set!\n");
+		return 0;
+	}
+
+	perf_pmu_migrate_context(&pcie_pmu->pmu, cpu, target);
+	/* Use this CPU for event counting */
+	pcie_pmu->on_cpu = target;
+	WARN_ON(irq_set_affinity(pcie_pmu->irq, cpumask_of(target)));
+
+	return 0;
+}
+
+/*
+ * Events with the "dl" suffix in their names count performance in DL layer,
+ * otherswise, events count performance in TL layer.
+ */
+static struct attribute *hisi_pcie_pmu_events_attr[] = {
+	HISI_PCIE_PMU_EVENT_ATTR(flux_rx_mwr, 0x010004),
+	HISI_PCIE_PMU_EVENT_ATTR(flux_rx_mrd, 0x100005),
+	HISI_PCIE_PMU_EVENT_ATTR(flux_tx_mwr, 0x010005),
+	HISI_PCIE_PMU_EVENT_ATTR(flux_tx_mrd, 0x200004),
+	HISI_PCIE_PMU_EVENT_ATTR(delay_rx_mwr, 0x000010),
+	HISI_PCIE_PMU_EVENT_ATTR(delay_rx_mrd, 0x020010),
+	HISI_PCIE_PMU_EVENT_ATTR(delay_tx_mrd, 0x000011),
+	HISI_PCIE_PMU_EVENT_ATTR(flux_rx_dl, 0x010084),
+	HISI_PCIE_PMU_EVENT_ATTR(flux_tx_dl, 0x030084),
+	NULL
+};
+
+static struct attribute_group hisi_pcie_pmu_events_group = {
+	.name = "events",
+	.attrs = hisi_pcie_pmu_events_attr,
+};
+
+static struct attribute *hisi_pcie_pmu_format_attr[] = {
+	HISI_PCIE_PMU_FORMAT_ATTR(event, "config:0-15"),
+	HISI_PCIE_PMU_FORMAT_ATTR(subevent, "config:16-23"),
+	HISI_PCIE_PMU_FORMAT_ATTR(thr_len, "config1:0-3"),
+	HISI_PCIE_PMU_FORMAT_ATTR(thr_mode, "config1:4"),
+	HISI_PCIE_PMU_FORMAT_ATTR(trig_len, "config1:5-8"),
+	HISI_PCIE_PMU_FORMAT_ATTR(trig_mode, "config1:9"),
+	HISI_PCIE_PMU_FORMAT_ATTR(port, "config2:0-15"),
+	HISI_PCIE_PMU_FORMAT_ATTR(bdf, "config2:16-31"),
+	NULL
+};
+
+static const struct attribute_group hisi_pcie_pmu_format_group = {
+	.name = "format",
+	.attrs = hisi_pcie_pmu_format_attr,
+};
+
+static struct attribute *hisi_pcie_pmu_bus_attrs[] = {
+	&dev_attr_bus.attr,
+	NULL
+};
+
+static const struct attribute_group hisi_pcie_pmu_bus_attr_group = {
+	.attrs = hisi_pcie_pmu_bus_attrs,
+};
+
+static struct attribute *hisi_pcie_pmu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL
+};
+
+static const struct attribute_group hisi_pcie_pmu_cpumask_attr_group = {
+	.attrs = hisi_pcie_pmu_cpumask_attrs,
+};
+
+static struct attribute *hisi_pcie_pmu_identifier_attrs[] = {
+	&dev_attr_identifier.attr,
+	NULL
+};
+
+static const struct attribute_group hisi_pcie_pmu_identifier_attr_group = {
+	.attrs = hisi_pcie_pmu_identifier_attrs,
+};
+
+static const struct attribute_group *hisi_pcie_pmu_attr_groups[] = {
+	&hisi_pcie_pmu_events_group,
+	&hisi_pcie_pmu_format_group,
+	&hisi_pcie_pmu_bus_attr_group,
+	&hisi_pcie_pmu_cpumask_attr_group,
+	&hisi_pcie_pmu_identifier_attr_group,
+	NULL
+};
+
+static int hisi_pcie_alloc_pmu(struct pci_dev *pdev,
+			       struct hisi_pcie_pmu *pcie_pmu)
+{
+	struct hisi_pcie_reg_pair regs;
+	u16 sicl_id, device_id;
+	char *name;
+
+	regs = hisi_pcie_parse_reg_value(pcie_pmu, HISI_PCIE_REG_BDF);
+	pcie_pmu->bdf_min = regs.lo;
+	pcie_pmu->bdf_max = regs.hi;
+
+	regs = hisi_pcie_parse_reg_value(pcie_pmu, HISI_PCIE_REG_INFO);
+	sicl_id = regs.hi;
+	device_id = regs.lo;
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL,
+			      "hisi_pcie%u_%u", sicl_id, device_id);
+	if (!name)
+		return -ENOMEM;
+
+	pcie_pmu->pdev = pdev;
+	pcie_pmu->on_cpu = -1;
+	pcie_pmu->identifier = readl(pcie_pmu->base + HISI_PCIE_REG_VERSION);
+	pcie_pmu->pmu = (struct pmu) {
+		.name		= name,
+		.module		= THIS_MODULE,
+		.event_init	= hisi_pcie_pmu_event_init,
+		.pmu_enable	= hisi_pcie_pmu_enable,
+		.pmu_disable	= hisi_pcie_pmu_disable,
+		.add		= hisi_pcie_pmu_add,
+		.del		= hisi_pcie_pmu_del,
+		.start		= hisi_pcie_pmu_start,
+		.stop		= hisi_pcie_pmu_stop,
+		.read		= hisi_pcie_pmu_read,
+		.task_ctx_nr	= perf_invalid_context,
+		.attr_groups	= hisi_pcie_pmu_attr_groups,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	return 0;
+}
+
+static int hisi_pcie_init_pmu(struct pci_dev *pdev,
+			      struct hisi_pcie_pmu *pcie_pmu)
+{
+	int ret;
+
+	pcie_pmu->base = pci_ioremap_bar(pdev, 2);
+	if (!pcie_pmu->base) {
+		pci_err(pdev, "Ioremap failed for pcie_pmu resource.\n");
+		return -ENOMEM;
+	}
+
+	ret = hisi_pcie_alloc_pmu(pdev, pcie_pmu);
+	if (ret)
+		return ret;
+
+	ret = hisi_pcie_pmu_irq_register(pdev, pcie_pmu);
+	if (ret)
+		goto err_set_pmu_fail;
+
+	ret = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_HISI_PCIE_PMU_ONLINE,
+				       &pcie_pmu->node);
+	if (ret) {
+		pci_err(pdev, "Failed to register hotplug: %d.\n", ret);
+		goto err_irq_unregister;
+	}
+
+	ret = perf_pmu_register(&pcie_pmu->pmu, pcie_pmu->pmu.name, -1);
+	if (ret) {
+		pci_err(pdev, "Failed to register PCIe PMU: %d.\n", ret);
+		goto err_hotplug_unregister;
+	}
+
+	return ret;
+
+err_hotplug_unregister:
+	cpuhp_state_remove_instance_nocalls(
+		CPUHP_AP_PERF_ARM_HISI_PCIE_PMU_ONLINE, &pcie_pmu->node);
+
+err_irq_unregister:
+	hisi_pcie_pmu_irq_unregister(pdev, pcie_pmu);
+
+err_set_pmu_fail:
+	iounmap(pcie_pmu->base);
+
+	return ret;
+}
+
+static void hisi_pcie_uninit_pmu(struct pci_dev *pdev)
+{
+	struct hisi_pcie_pmu *pcie_pmu = pci_get_drvdata(pdev);
+
+	perf_pmu_unregister(&pcie_pmu->pmu);
+	cpuhp_state_remove_instance_nocalls(
+		CPUHP_AP_PERF_ARM_HISI_PCIE_PMU_ONLINE, &pcie_pmu->node);
+	hisi_pcie_pmu_irq_unregister(pdev, pcie_pmu);
+	iounmap(pcie_pmu->base);
+}
+
+static int hisi_pcie_init_dev(struct pci_dev *pdev)
+{
+	int ret;
+
+	ret = pci_enable_device(pdev);
+	if (ret) {
+		pci_err(pdev, "Failed to enable PCI device: %d.\n", ret);
+		return ret;
+	}
+
+	ret = pci_request_mem_regions(pdev, "hisi_pcie_pmu");
+	if (ret < 0) {
+		pci_err(pdev, "Failed to request PCI mem regions: %d.\n", ret);
+		pci_disable_device(pdev);
+		return ret;
+	}
+
+	pci_set_master(pdev);
+
+	return 0;
+}
+
+static void hisi_pcie_uninit_dev(struct pci_dev *pdev)
+{
+	pci_clear_master(pdev);
+	pci_release_mem_regions(pdev);
+	pci_disable_device(pdev);
+}
+
+static int hisi_pcie_pmu_probe(struct pci_dev *pdev,
+			       const struct pci_device_id *id)
+{
+	struct hisi_pcie_pmu *pcie_pmu;
+	int ret;
+
+	pcie_pmu = devm_kzalloc(&pdev->dev, sizeof(*pcie_pmu), GFP_KERNEL);
+	if (!pcie_pmu)
+		return -ENOMEM;
+
+	ret = hisi_pcie_init_dev(pdev);
+	if (ret)
+		return ret;
+
+	ret = hisi_pcie_init_pmu(pdev, pcie_pmu);
+	if (ret)
+		hisi_pcie_uninit_dev(pdev);
+
+	pci_set_drvdata(pdev, pcie_pmu);
+
+	return ret;
+}
+
+static void hisi_pcie_pmu_remove(struct pci_dev *pdev)
+{
+	hisi_pcie_uninit_pmu(pdev);
+	hisi_pcie_uninit_dev(pdev);
+	pci_set_drvdata(pdev, NULL);
+}
+
+static const struct pci_device_id hisi_pcie_pmu_ids[] = {
+	{ PCI_DEVICE(PCI_VENDOR_ID_HUAWEI, 0xa12d) },
+	{ 0, }
+};
+MODULE_DEVICE_TABLE(pci, hisi_pcie_pmu_ids);
+
+static struct pci_driver hisi_pcie_pmu_driver = {
+	.name = "hisi_pcie_pmu",
+	.id_table = hisi_pcie_pmu_ids,
+	.probe = hisi_pcie_pmu_probe,
+	.remove = hisi_pcie_pmu_remove,
+};
+
+static int __init hisi_pcie_module_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_HISI_PCIE_PMU_ONLINE,
+				      "AP_PERF_ARM_HISI_PCIE_PMU_ONLINE",
+				      hisi_pcie_pmu_online_cpu,
+				      hisi_pcie_pmu_offline_cpu);
+	if (ret) {
+		pr_err("Failed to setup PCIe PMU hotplug, ret = %d.\n", ret);
+		return ret;
+	}
+
+	ret = pci_register_driver(&hisi_pcie_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(
+				CPUHP_AP_PERF_ARM_HISI_PCIE_PMU_ONLINE);
+
+	return ret;
+}
+module_init(hisi_pcie_module_init);
+
+static void __exit hisi_pcie_module_exit(void)
+{
+	pci_unregister_driver(&hisi_pcie_pmu_driver);
+	cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_HISI_PCIE_PMU_ONLINE);
+}
+module_exit(hisi_pcie_module_exit);
+
+MODULE_DESCRIPTION("HiSilicon PCIe PMU driver");
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Qi Liu <liuqi115@huawei.com>");
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index 4a62b3980642..a9776a2f955b 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -180,6 +180,7 @@ enum cpuhp_state {
 	CPUHP_AP_PERF_ARM_HISI_L3_ONLINE,
 	CPUHP_AP_PERF_ARM_HISI_PA_ONLINE,
 	CPUHP_AP_PERF_ARM_HISI_SLLC_ONLINE,
+	CPUHP_AP_PERF_ARM_HISI_PCIE_PMU_ONLINE,
 	CPUHP_AP_PERF_ARM_L2X0_ONLINE,
 	CPUHP_AP_PERF_ARM_QCOM_L2_ONLINE,
 	CPUHP_AP_PERF_ARM_QCOM_L3_ONLINE,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v8 2/2] drivers/perf: hisi: Add driver for HiSilicon PCIe PMU
  2021-07-28  8:09 ` [PATCH v8 2/2] drivers/perf: hisi: Add driver for HiSilicon PCIe PMU Qi Liu
@ 2021-08-02 10:03   ` Will Deacon
  2021-08-04  7:29     ` liuqi (BA)
  0 siblings, 1 reply; 6+ messages in thread
From: Will Deacon @ 2021-08-02 10:03 UTC (permalink / raw)
  To: Qi Liu
  Cc: mark.rutland, bhelgaas, linux-pci, linux-arm-kernel,
	linux-kernel, linuxarm, zhangshaokun

On Wed, Jul 28, 2021 at 04:09:32PM +0800, Qi Liu wrote:
> PCIe PMU Root Complex Integrated End Point(RCiEP) device is supported
> to sample bandwidth, latency, buffer occupation etc.
> 
> Each PMU RCiEP device monitors multiple Root Ports, and each RCiEP is
> registered as a PMU in /sys/bus/event_source/devices, so users can
> select target PMU, and use filter to do further sets.
> 
> Filtering options contains:
> event        - select the event.
> subevent     - select the subevent.

Hmm, I was hoping that you would expose all the events as proper perf_events
and get rid of the subevents entirely.

Then userspace could do things like:

  // Count number of RX memory reads
  $ perf stat -e hisi_pcie0_0/rx_memory_read/

  // Count delay cycles
  $ perf stat -e hisi_pcie0_0/latency/

  // Count both of the above (events must be in the same group)
  $ perf stat -g -e hisi_pcie0_0/latency/ -e hisi_pcie0_0/rx_memory_read/

Note that in all three of these cases the hardware will be programmed in
the same way and both HISI_PCIE_CNT and HISI_PCIE_EXT_CNT are allocated!

So for example, doing this (i.e. without the '-g'):

  $ perf stat -e hisi_pcie0_0/latency/ -e hisi_pcie0_0/rx_memory_read/

would fail because the first event would allocate both of the counters.

All you need to do is check the counter scheduling constraints when
accepting an event group in the driver. No need for subevents at all.

Does that make sense?

Will

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v8 2/2] drivers/perf: hisi: Add driver for HiSilicon PCIe PMU
  2021-08-02 10:03   ` Will Deacon
@ 2021-08-04  7:29     ` liuqi (BA)
  2021-08-13 14:40       ` Will Deacon
  0 siblings, 1 reply; 6+ messages in thread
From: liuqi (BA) @ 2021-08-04  7:29 UTC (permalink / raw)
  To: Will Deacon, Linuxarm
  Cc: mark.rutland, bhelgaas, linux-pci, linux-arm-kernel,
	linux-kernel, zhangshaokun


Hi Will,
> Hmm, I was hoping that you would expose all the events as proper perf_events
> and get rid of the subevents entirely.
> 
> Then userspace could do things like:
> 
>    // Count number of RX memory reads
>    $ perf stat -e hisi_pcie0_0/rx_memory_read/
> 
>    // Count delay cycles
>    $ perf stat -e hisi_pcie0_0/latency/
> 
>    // Count both of the above (events must be in the same group)
>    $ perf stat -g -e hisi_pcie0_0/latency/ -e hisi_pcie0_0/rx_memory_read/
> 
> Note that in all three of these cases the hardware will be programmed in
> the same way and both HISI_PCIE_CNT and HISI_PCIE_EXT_CNT are allocated!
> 
> So for example, doing this (i.e. without the '-g'):
> 
>    $ perf stat -e hisi_pcie0_0/latency/ -e hisi_pcie0_0/rx_memory_read/
> 
> would fail because the first event would allocate both of the counters.

I'm confused with this situation when getting rid of subevent:

$ perf stat -e hisi_pcie0_0/latency/ -e hisi_pcie0_0/rx_memory_read/

In this case, driver checks the relationship of "latency" and 
"rx_memory_read" in pmu->add() function and return a -EINVAL, but this 
seems lead to time division multiplexing.

	if (event->pmu->add(event, PERF_EF_START)) {
		perf_event_set_state(event, PERF_EVENT_STATE_INACTIVE);
		event->oncpu = -1;
		ret = -EAGAIN;
		goto out;
	}
	...
out:
	perf_pmu_enable(event->pmu);

This result doesn't meet our expection, do I miss something here?

How about add an array to record events and check the relationship in 
event_init() function? It seems that perf stat could only failed when 
driver return invalid value in pmu->event_init() function.

Thanks,
Qi
> 
> All you need to do is check the counter scheduling constraints when
> accepting an event group in the driver. No need for subevents at all.
> 
> Does that make sense?
> 
> Will
> .
> 


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v8 2/2] drivers/perf: hisi: Add driver for HiSilicon PCIe PMU
  2021-08-04  7:29     ` liuqi (BA)
@ 2021-08-13 14:40       ` Will Deacon
  0 siblings, 0 replies; 6+ messages in thread
From: Will Deacon @ 2021-08-13 14:40 UTC (permalink / raw)
  To: liuqi (BA)
  Cc: Linuxarm, mark.rutland, bhelgaas, linux-pci, linux-arm-kernel,
	linux-kernel, zhangshaokun

On Wed, Aug 04, 2021 at 03:29:54PM +0800, liuqi (BA) wrote:
> 
> Hi Will,
> > Hmm, I was hoping that you would expose all the events as proper perf_events
> > and get rid of the subevents entirely.
> > 
> > Then userspace could do things like:
> > 
> >    // Count number of RX memory reads
> >    $ perf stat -e hisi_pcie0_0/rx_memory_read/
> > 
> >    // Count delay cycles
> >    $ perf stat -e hisi_pcie0_0/latency/
> > 
> >    // Count both of the above (events must be in the same group)
> >    $ perf stat -g -e hisi_pcie0_0/latency/ -e hisi_pcie0_0/rx_memory_read/
> > 
> > Note that in all three of these cases the hardware will be programmed in
> > the same way and both HISI_PCIE_CNT and HISI_PCIE_EXT_CNT are allocated!
> > 
> > So for example, doing this (i.e. without the '-g'):
> > 
> >    $ perf stat -e hisi_pcie0_0/latency/ -e hisi_pcie0_0/rx_memory_read/
> > 
> > would fail because the first event would allocate both of the counters.
> 
> I'm confused with this situation when getting rid of subevent:
> 
> $ perf stat -e hisi_pcie0_0/latency/ -e hisi_pcie0_0/rx_memory_read/
> 
> In this case, driver checks the relationship of "latency" and
> "rx_memory_read" in pmu->add() function and return a -EINVAL, but this seems
> lead to time division multiplexing.
> 
> 	if (event->pmu->add(event, PERF_EF_START)) {
> 		perf_event_set_state(event, PERF_EVENT_STATE_INACTIVE);
> 		event->oncpu = -1;
> 		ret = -EAGAIN;
> 		goto out;
> 	}
> 	...
> out:
> 	perf_pmu_enable(event->pmu);
> 
> This result doesn't meet our expection, do I miss something here?

This is how perf works. If you don't want multiplexing, put the events in a
group. What's the problem with that?

Will

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2021-08-13 14:40 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-07-28  8:09 [PATCH v8 0/2] drivers/perf: hisi: Add support for PCIe PMU Qi Liu
2021-07-28  8:09 ` [PATCH v8 1/2] docs: perf: Add description for HiSilicon PCIe PMU driver Qi Liu
2021-07-28  8:09 ` [PATCH v8 2/2] drivers/perf: hisi: Add driver for HiSilicon PCIe PMU Qi Liu
2021-08-02 10:03   ` Will Deacon
2021-08-04  7:29     ` liuqi (BA)
2021-08-13 14:40       ` Will Deacon

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