diff mbox series

[RFC,2/4] powerpc/papr_scm: Add perf interface support

Message ID 20210512163824.255370-3-kjain@linux.ibm.com
State Changes Requested
Headers show
Series Add perf interface to expose nvdimm performance stats | expand

Checks

Context Check Description
snowpatch_ozlabs/apply_patch success Successfully applied on branch powerpc/merge (1fab3666d738e4af3a7450c44441310e4d7a7e53)
snowpatch_ozlabs/checkpatch warning total: 0 errors, 0 warnings, 1 checks, 332 lines checked
snowpatch_ozlabs/needsstable success Patch has no Fixes tags

Commit Message

kajoljain May 12, 2021, 4:38 p.m. UTC
This patch adds support for performance monitoring of papr
nvdimm devices via perf interface. It adds callbacks functions
like add/del/read/event_init for nvdimm_pmu structure.

Patch adds a new parameter 'priv' in pdev_archdata structure to save
nvdimm_pmu device pointer, to handle the unregistering of pmu device.

papr_scm_pmu_register function populates the nvdimm_pmu structure
with events, attribute groups along with event handling functions.
Event handling functions internally uses hcall to get events and
counter data. Finally the populated nvdimm_pmu structure is passed
to register the pmu device.

Result in power9 machine with 2 nvdimm device:

Ex: List all event by perf list

command:# perf list nmem

  nmem0/cchrhcnt/                                    [Kernel PMU event]
  nmem0/cchwhcnt/                                    [Kernel PMU event]
  nmem0/critrscu/                                    [Kernel PMU event]
  nmem0/ctlresct/                                    [Kernel PMU event]
  nmem0/ctlrestm/                                    [Kernel PMU event]
  nmem0/fastwcnt/                                    [Kernel PMU event]
  nmem0/hostlcnt/                                    [Kernel PMU event]
  nmem0/hostldur/                                    [Kernel PMU event]
  nmem0/hostscnt/                                    [Kernel PMU event]
  nmem0/hostsdur/                                    [Kernel PMU event]
  nmem0/medrcnt/                                     [Kernel PMU event]
  nmem0/medrdur/                                     [Kernel PMU event]
  nmem0/medwcnt/                                     [Kernel PMU event]
  nmem0/medwdur/                                     [Kernel PMU event]
  nmem0/memlife/                                     [Kernel PMU event]
  nmem0/noopstat/                                    [Kernel PMU event]
  nmem0/ponsecs/                                     [Kernel PMU event]
  nmem1/cchrhcnt/                                    [Kernel PMU event]
  nmem1/cchwhcnt/                                    [Kernel PMU event]
  nmem1/critrscu/                                    [Kernel PMU event]
  ...
  nmem1/noopstat/                                    [Kernel PMU event]
  nmem1/ponsecs/                                     [Kernel PMU event]

Signed-off-by: Kajol Jain <kjain@linux.ibm.com>
---
 arch/powerpc/include/asm/device.h         |   5 +
 arch/powerpc/platforms/pseries/papr_scm.c | 284 +++++++++++++++++++++-
 2 files changed, 288 insertions(+), 1 deletion(-)
diff mbox series

Patch

diff --git a/arch/powerpc/include/asm/device.h b/arch/powerpc/include/asm/device.h
index 219559d65864..47ed639f3b8f 100644
--- a/arch/powerpc/include/asm/device.h
+++ b/arch/powerpc/include/asm/device.h
@@ -48,6 +48,11 @@  struct dev_archdata {
 
 struct pdev_archdata {
 	u64 dma_mask;
+	/*
+	 * Pointer to nvdimm_pmu structure, to handle the unregistering
+	 * of pmu device
+	 */
+	void *priv;
 };
 
 #endif /* _ASM_POWERPC_DEVICE_H */
diff --git a/arch/powerpc/platforms/pseries/papr_scm.c b/arch/powerpc/platforms/pseries/papr_scm.c
index ef26fe40efb0..997d379094d0 100644
--- a/arch/powerpc/platforms/pseries/papr_scm.c
+++ b/arch/powerpc/platforms/pseries/papr_scm.c
@@ -18,6 +18,8 @@ 
 #include <asm/plpar_wrappers.h>
 #include <asm/papr_pdsm.h>
 #include <asm/mce.h>
+#include <linux/perf_event.h>
+#include <linux/ctype.h>
 
 #define BIND_ANY_ADDR (~0ul)
 
@@ -116,6 +118,9 @@  struct papr_scm_priv {
 
 	/* length of the stat buffer as expected by phyp */
 	size_t stat_buffer_len;
+
+	 /* array to have event_code and stat_id mappings */
+	char **nvdimm_events_map;
 };
 
 static int papr_scm_pmem_flush(struct nd_region *nd_region,
@@ -329,6 +334,271 @@  static ssize_t drc_pmem_query_stats(struct papr_scm_priv *p,
 	return 0;
 }
 
+static struct attribute_group nvdimm_pmu_events_group = {
+	.name = "events",
+	/* .attrs is set in papr_scm_pmu_check_events function */
+};
+
+PMU_FORMAT_ATTR(event, "config:0-37");
+
+static struct attribute *nvdimm_pmu_format_attr[] = {
+	&format_attr_event.attr,
+	NULL,
+};
+
+static struct attribute_group nvdimm_pmu_format_group = {
+	.name = "format",
+	.attrs = nvdimm_pmu_format_attr,
+};
+
+static const struct attribute_group *nvdimm_pmu_attr_groups[] = {
+	&nvdimm_pmu_format_group,
+	&nvdimm_pmu_events_group,
+	NULL,
+};
+
+static void papr_scm_pmu_get_value(struct perf_event *event, struct device *dev, u64 *count)
+{
+	struct papr_scm_perf_stat *stat;
+	struct papr_scm_perf_stats *stats;
+	struct papr_scm_priv *p = (struct papr_scm_priv *)dev->driver_data;
+	int rc, size;
+	u64 statval;
+
+	/* Allocate buffer to hold single performance stat */
+	size = sizeof(struct papr_scm_perf_stats) +
+		sizeof(struct papr_scm_perf_stat);
+
+	if (!p->nvdimm_events_map)
+		return;
+
+	stats = kzalloc(size, GFP_KERNEL);
+	if (!stats)
+		return;
+
+	stat = &stats->scm_statistic[0];
+	memcpy(&stat->stat_id,
+	       p->nvdimm_events_map[event->attr.config - 1],
+		sizeof(stat->stat_id));
+	stat->stat_val = 0;
+
+	rc = drc_pmem_query_stats(p, stats, 1);
+	if (rc < 0) {
+		kfree(stats);
+		return;
+	}
+
+	statval = be64_to_cpu(stat->stat_val);
+	*count = statval;
+	kfree(stats);
+}
+
+static int papr_scm_pmu_add(struct perf_event *event, int flags,  struct device *dev)
+{
+	u64 count = 0;
+
+	papr_scm_pmu_get_value(event, dev, &count);
+	local64_set(&event->hw.prev_count, count);
+	return 0;
+}
+
+static void papr_scm_pmu_read(struct perf_event *event, struct device *dev)
+{
+	u64 prev, now = 0;
+
+	papr_scm_pmu_get_value(event, dev, &now);
+	prev = local64_xchg(&event->hw.prev_count, now);
+
+	if (now - prev >= 0)
+		local64_add(now - prev, &event->count);
+}
+
+static void papr_scm_pmu_del(struct perf_event *event, int flags,  struct device *dev)
+{
+	papr_scm_pmu_read(event, dev);
+}
+
+static void nvdimm_pmu_uinit(struct nvdimm_pmu *nd_pmu)
+{
+	unregister_nvdimm_pmu(&nd_pmu->pmu);
+	kfree(nd_pmu);
+}
+
+static int papr_scm_pmu_register(struct papr_scm_priv *p)
+{
+	struct nvdimm_pmu *papr_scm_pmu;
+	int rc;
+
+	papr_scm_pmu = devm_kzalloc(&p->pdev->dev, sizeof(*papr_scm_pmu), GFP_KERNEL);
+	if (!papr_scm_pmu)
+		return -ENOMEM;
+
+	papr_scm_pmu->name = nvdimm_name(p->nvdimm);
+	papr_scm_pmu->read = papr_scm_pmu_read;
+	papr_scm_pmu->add = papr_scm_pmu_add;
+	papr_scm_pmu->del = papr_scm_pmu_del;
+	papr_scm_pmu->attr_groups = nvdimm_pmu_attr_groups;
+
+	rc = register_nvdimm_pmu(papr_scm_pmu, p->pdev);
+	if (rc)
+		goto pmu_register_err;
+
+	/*
+	 * Set archdata.priv value to nvdimm_pmu structure, to handle the
+	 * unregistering of pmu device.
+	 */
+	p->pdev->archdata.priv = papr_scm_pmu;
+	return 0;
+
+pmu_register_err:
+	kfree(papr_scm_pmu);
+	return rc;
+}
+
+static ssize_t device_show_string(struct device *dev, struct device_attribute *attr,
+				  char *buf)
+{
+	struct perf_pmu_events_attr *d;
+
+	d = container_of(attr, struct perf_pmu_events_attr, attr);
+
+	return sysfs_emit(buf, "%s\n", (char *)d->event_str);
+}
+
+static char *strtolower(char *updated_name)
+{
+	int i = 0;
+
+	while (updated_name[i]) {
+		if (isupper(updated_name[i]))
+			updated_name[i] = tolower(updated_name[i]);
+		i++;
+	}
+	updated_name[i] = '\0';
+	return strim(updated_name);
+}
+
+/* device_str_attr_create : Populate event "name" and string "str" in attribute */
+static struct attribute *device_str_attr_create_(char *name, char *str)
+{
+	struct perf_pmu_events_attr *attr;
+
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
+
+	if (!attr)
+		return NULL;
+
+	sysfs_attr_init(&attr->attr.attr);
+	attr->event_str = str;
+	attr->attr.attr.name = strtolower(name);
+	attr->attr.attr.mode = 0444;
+	attr->attr.show = device_show_string;
+
+	return &attr->attr.attr;
+}
+
+static int papr_scm_pmu_check_events(struct papr_scm_priv *p)
+{
+	struct papr_scm_perf_stat *stat;
+	struct papr_scm_perf_stats *stats, *single_stats;
+	int index, size, rc, attrs;
+	u32 total_events;
+	struct attribute **events;
+	char *eventcode, *eventname, *statid;
+
+	if (!p->stat_buffer_len)
+		return -ENOENT;
+
+	total_events = (p->stat_buffer_len  - sizeof(struct papr_scm_perf_stats))
+			/ sizeof(struct papr_scm_perf_stat);
+
+	/* Allocate the buffer for phyp where stats are written */
+	stats = kzalloc(p->stat_buffer_len, GFP_KERNEL);
+	if (!stats)
+		return -ENOMEM;
+
+	/* Allocate memory to nvdimm_event_map */
+	p->nvdimm_events_map = kcalloc(total_events, sizeof(char *), GFP_KERNEL);
+	if (!p->nvdimm_events_map) {
+		rc = -ENOMEM;
+		goto out_stats;
+	}
+
+	/* Called to get list of events supported */
+	rc = drc_pmem_query_stats(p, stats, 0);
+	if (rc)
+		goto out_nvdimm_events_map;
+
+	/* Allocate buffer to hold single performance stat */
+	size = sizeof(struct papr_scm_perf_stats) + sizeof(struct papr_scm_perf_stat);
+
+	single_stats = kzalloc(size, GFP_KERNEL);
+	if (!single_stats) {
+		rc = -ENOMEM;
+		goto out_nvdimm_events_map;
+	}
+
+	events = kzalloc(total_events * sizeof(struct attribute *), GFP_KERNEL);
+	if (!events) {
+		rc = -ENOMEM;
+		goto out_single_stats;
+	}
+
+	for (index = 0, stat = stats->scm_statistic, attrs = 0;
+		     index < total_events; index++, ++stat) {
+
+		single_stats->scm_statistic[0] = *stat;
+		rc = drc_pmem_query_stats(p, single_stats, 1);
+
+		if (rc < 0) {
+			pr_info("Event not supported %s for device %s\n",
+				stat->stat_id, nvdimm_name(p->nvdimm));
+		} else {
+			eventcode = kasprintf(GFP_KERNEL, "event=0x%x", attrs + 1);
+			eventname = kzalloc(strlen(stat->stat_id) + 1, GFP_KERNEL);
+			statid = kzalloc(strlen(stat->stat_id) + 1, GFP_KERNEL);
+
+			if (!eventname || !statid || !eventcode)
+				goto out;
+
+			strcpy(eventname, stat->stat_id);
+			events[attrs] = device_str_attr_create_(eventname,
+								eventcode);
+			if (!events[attrs])
+				goto out;
+
+			strcpy(statid, stat->stat_id);
+			p->nvdimm_events_map[attrs] = statid;
+			attrs++;
+			continue;
+out:
+			kfree(eventcode);
+			kfree(eventname);
+			kfree(statid);
+		}
+	}
+	events[attrs] = NULL;
+	p->nvdimm_events_map[attrs] = NULL;
+
+	if (!attrs)
+		goto out_events;
+
+	nvdimm_pmu_events_group.attrs = events;
+	kfree(single_stats);
+	kfree(stats);
+	return 0;
+
+out_events:
+	kfree(events);
+out_single_stats:
+	kfree(single_stats);
+out_nvdimm_events_map:
+	kfree(p->nvdimm_events_map);
+out_stats:
+	kfree(stats);
+	return rc;
+}
+
 /*
  * Issue hcall to retrieve dimm health info and populate papr_scm_priv with the
  * health information.
@@ -923,7 +1193,7 @@  static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
 	struct nd_mapping_desc mapping;
 	struct nd_region_desc ndr_desc;
 	unsigned long dimm_flags;
-	int target_nid, online_nid;
+	int target_nid, online_nid, rc;
 	ssize_t stat_size;
 
 	p->bus_desc.ndctl = papr_scm_ndctl;
@@ -1015,6 +1285,15 @@  static int papr_scm_nvdimm_init(struct papr_scm_priv *p)
 		p->stat_buffer_len = stat_size;
 		dev_dbg(&p->pdev->dev, "Max perf-stat size %lu-bytes\n",
 			p->stat_buffer_len);
+
+		rc = papr_scm_pmu_check_events(p);
+		if (rc) {
+			dev_info(&p->pdev->dev, "nvdimm pmu check events failed, rc=%d\n", rc);
+		} else {
+			rc = papr_scm_pmu_register(p);
+			if (rc)
+				dev_info(&p->pdev->dev, "nvdimm pmu didn't register rc=%d\n", rc);
+		}
 	} else {
 		dev_info(&p->pdev->dev, "Dimm performance stats unavailable\n");
 	}
@@ -1195,7 +1474,10 @@  static int papr_scm_remove(struct platform_device *pdev)
 
 	nvdimm_bus_unregister(p->bus);
 	drc_pmem_unbind(p);
+	nvdimm_pmu_uinit(pdev->archdata.priv);
+	pdev->archdata.priv = NULL;
 	kfree(p->bus_desc.provider_name);
+	kfree(p->nvdimm_events_map);
 	kfree(p);
 
 	return 0;