Hi Suzuki,
On Thu, 4 Mar 2021 at 12:13, Suzuki K Poulose suzuki.poulose@arm.com wrote:
On 1/28/21 5:09 PM, Mike Leach wrote:
Loaded coresight configurations are registered in the cs_etm\cs_config sub directory. This extends the etm-perf code to handle these registrations, and the cs_syscfg driver to perform the registration on load.
Signed-off-by: Mike Leach mike.leach@linaro.org
.../hwtracing/coresight/coresight-config.h | 5 +- .../hwtracing/coresight/coresight-etm-perf.c | 164 +++++++++++++++--- .../hwtracing/coresight/coresight-etm-perf.h | 8 + .../hwtracing/coresight/coresight-syscfg.c | 13 +- 4 files changed, 166 insertions(+), 24 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-config.h b/drivers/hwtracing/coresight/coresight-config.h index 9d66e0071f38..98380b496046 100644 --- a/drivers/hwtracing/coresight/coresight-config.h +++ b/drivers/hwtracing/coresight/coresight-config.h @@ -154,7 +154,8 @@ struct cscfg_config_feat_ref {
- @nr_presets: Number of sets of presets supplied by this configuration.
- @nr_total_params: Sum of all parameters declared by used features
- @presets: Array of preset values.
- @id_ea: Extended attribute for perf configid value
*/ struct cscfg_config_desc { const char *name;
- @event_ea: Extended attribute for perf event value
@@ -165,6 +166,8 @@ struct cscfg_config_desc { int nr_presets; int nr_total_params; const u64 *presets; /* nr_presets * nr_total_params */
struct dev_ext_attribute *id_ea;
struct dev_ext_attribute *event_ea;
};
/**
diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c index a608081bd446..e270bb1e0f7d 100644 --- a/drivers/hwtracing/coresight/coresight-etm-perf.c +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c @@ -18,8 +18,10 @@ #include <linux/types.h> #include <linux/workqueue.h>
+#include "coresight-config.h" #include "coresight-etm-perf.h" #include "coresight-priv.h" +#include "coresight-syscfg.h"
static struct pmu etm_pmu; static bool etm_perf_up; @@ -32,8 +34,13 @@ PMU_FORMAT_ATTR(cycacc, "config:" __stringify(ETM_OPT_CYCACC)); PMU_FORMAT_ATTR(contextid, "config:" __stringify(ETM_OPT_CTXTID)); PMU_FORMAT_ATTR(timestamp, "config:" __stringify(ETM_OPT_TS)); PMU_FORMAT_ATTR(retstack, "config:" __stringify(ETM_OPT_RETSTK)); +/* preset - if sink ID is used as a configuration selector */ +PMU_FORMAT_ATTR(preset, "config:0-3"); /* Sink ID - same for all ETMs */ PMU_FORMAT_ATTR(sinkid, "config2:0-31"); +/* config ID - set if a system configuration is selected */ +PMU_FORMAT_ATTR(configid, "config2:32-63");
static struct attribute *etm_config_formats_attr[] = { &format_attr_cycacc.attr,
@@ -41,6 +48,8 @@ static struct attribute *etm_config_formats_attr[] = { &format_attr_timestamp.attr, &format_attr_retstack.attr, &format_attr_sinkid.attr,
&format_attr_preset.attr,
};&format_attr_configid.attr, NULL,
@@ -58,9 +67,29 @@ static const struct attribute_group etm_pmu_sinks_group = { .attrs = etm_config_sinks_attr, };
+static struct attribute *etm_config_cscfg_attr[] = {
NULL,
+};
+static const struct attribute_group etm_pmu_cscfg_group = {
.name = "configurations",
.attrs = etm_config_cscfg_attr,
+};
+static struct attribute *etm_config_events_attr[] = {
NULL,
+};
+static const struct attribute_group etm_pmu_events_group = {
.name = "events",
.attrs = etm_config_events_attr,
+};
- static const struct attribute_group *etm_pmu_attr_groups[] = { &etm_pmu_format_group, &etm_pmu_sinks_group,
&etm_pmu_cscfg_group,
};&etm_pmu_events_group, NULL,
@@ -219,7 +248,7 @@ static void *etm_setup_aux(struct perf_event *event, void **pages, INIT_WORK(&event_data->work, free_event_data);
/* First get the selected sink from user space. */
if (event->attr.config2) {
if (event->attr.config2 & GENMASK_ULL(31, 0)) { id = (u32)event->attr.config2; sink = coresight_get_sink_by_id(id); }
@@ -537,21 +566,17 @@ static ssize_t etm_perf_sink_name_show(struct device *dev, return scnprintf(buf, PAGE_SIZE, "0x%lx\n", (unsigned long)(ea->var)); }
-int etm_perf_add_symlink_sink(struct coresight_device *csdev) +int etm_perf_add_symlink_group(struct device *dev,
struct dev_ext_attribute **ext_attr,
const char *name,
const char *group_name)
Couple of nits:
This could be:
struct dev_ext_attribute * etm_perf_add_to_group_hashed(struct device *dev, const char *group, const char *name) { }
This could return the ea on success avoiding the **ext_attr as argument. Also the name of the function indicates what it does.
OK - I'll look at this.
{
int ret;
struct dev_ext_attribute *ea; unsigned long hash;
const char *name;
int ret; struct device *pmu_dev = etm_pmu.dev;
struct device *dev = &csdev->dev;
struct dev_ext_attribute *ea;
if (csdev->type != CORESIGHT_DEV_TYPE_SINK &&
csdev->type != CORESIGHT_DEV_TYPE_LINKSINK)
return -EINVAL;
if (csdev->ea != NULL)
return -EINVAL;
*ext_attr = NULL; if (!etm_perf_up) return -EPROBE_DEFER;
@@ -560,7 +585,6 @@ int etm_perf_add_symlink_sink(struct coresight_device *csdev) if (!ea) return -ENOMEM;
name = dev_name(dev); /* See function coresight_get_sink_by_id() to know where this is used */ hash = hashlen_hash(hashlen_string(NULL, name));
@@ -574,31 +598,127 @@ int etm_perf_add_symlink_sink(struct coresight_device *csdev) ea->var = (unsigned long *)hash;
ret = sysfs_add_file_to_group(&pmu_dev->kobj,
&ea->attr.attr, "sinks");
&ea->attr.attr, group_name); if (!ret)
csdev->ea = ea;
*ext_attr = ea; return ret;
}
-void etm_perf_del_symlink_sink(struct coresight_device *csdev) +int etm_perf_add_symlink_sink(struct coresight_device *csdev) +{
const char *name;
struct device *dev = &csdev->dev;
if (csdev->type != CORESIGHT_DEV_TYPE_SINK &&
csdev->type != CORESIGHT_DEV_TYPE_LINKSINK)
return -EINVAL;
if (csdev->ea != NULL)
return -EINVAL;
name = dev_name(dev);
return etm_perf_add_symlink_group(dev, &csdev->ea, name, "sinks");
+}
+void etm_perf_del_symlink_group(struct dev_ext_attribute *ea, const char *group_name)
void etm_perf_del_from_group(group, ea)
+/* string to contain the attribute value */ +#define CSCFG_EVENT_STR_SIZE 32
+static ssize_t etm_perf_cscfg_event_show(struct device *dev,
struct device_attribute *dattr,
char *buf)
+{
struct dev_ext_attribute *ea;
ea = container_of(dattr, struct dev_ext_attribute, attr);
return scnprintf(buf, PAGE_SIZE, "%s\n", (const char *)(ea->var));
+}
"configid=0x%lx", (unsigned long)ea->var ?
ea->var _is_ "configid=0x%lx" due to the way perf handles the events sub-dir entries.
+static int etm_perf_add_cscfg_event(struct device *dev, struct cscfg_config_desc *cs_cfg) +{
struct dev_ext_attribute *ea;
unsigned long hash;
int ret;
struct device *pmu_dev = etm_pmu.dev;
ea = devm_kzalloc(dev, sizeof(*ea), GFP_KERNEL);
if (!ea)
return -ENOMEM;
hash = (unsigned long)cs_cfg->id_ea->var;
sysfs_attr_init(&ea->attr.attr);
ea->attr.attr.name = devm_kstrdup(dev, cs_cfg->name, GFP_KERNEL);
if (!ea->attr.attr.name)
return -ENOMEM;
/*
* attribute value is "configid=<hash>".
* this will be what perf evaluates when the config name is used
* on the command line.
*/
ea->var = devm_kzalloc(dev, CSCFG_EVENT_STR_SIZE, GFP_KERNEL);
if (!ea->var)
return -ENOMEM;
Could we drop this string and use the "hash" instead ?
No. My understanding is that we have added an events directory to cs_etm, and add the configurations in there:-
cs_etm/events/autofdo
Now the contents of autofdo are "configid=0x<hash-value>" - where hash-value is the hash of "autofdo".
On the perf command line:-
perf record -e cs_etm/autofdo/ .....
will result in perf parsing autofdo, looking in the events dir for cs_etm, seeing the configid=-string, and parsing that to assign to configid attribute - which we have allocated to config2:63:32 - this will then appear as a value in the perf_event and we can load the configuration when starting up the event on the ETM etc.
scnprintf(ea->var, CSCFG_EVENT_STR_SIZE, "configid=0x%lx", hash);
ea->attr.attr.mode = 0444;
ea->attr.show = etm_perf_cscfg_event_show;
ret = sysfs_add_file_to_group(&pmu_dev->kobj,
&ea->attr.attr, "events");
if (!ret)
cs_cfg->event_ea = ea;
return ret;
+}
+int etm_perf_add_symlink_cscfg(struct device *dev, struct cscfg_config_desc *cs_cfg)
etm_perf_add_cscfg() ?
+{
int err;
if (cs_cfg->id_ea != NULL)
return 0;
err = etm_perf_add_symlink_group(dev, &cs_cfg->id_ea,
cs_cfg->name, "configurations");
if (!err)
err = etm_perf_add_cscfg_event(dev, cs_cfg);
return err;
+}
+void etm_perf_del_symlink_cscfg(struct cscfg_config_desc *cs_cfg)
etm_perf_del_cscfg() ?
Both these seem reasonable.
Regards
Mike
Suzuki