linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH v6 10/11] drivers: perf: hisi: use poll method when no IRQ for MN counter overflow
@ 2017-03-10  6:29 Anurup M
  2017-03-21 17:17 ` Mark Rutland
  0 siblings, 1 reply; 3+ messages in thread
From: Anurup M @ 2017-03-10  6:29 UTC (permalink / raw)
  To: mark.rutland, will.deacon
  Cc: linux-kernel, linux-arm-kernel, anurup.m, zhangshaokun,
	tanxiaojun, xuwei5, sanil.kumar, john.garry, gabriele.paoloni,
	shiju.jose, huangdaode, linuxarm, dikshit.n, shyju.pv,
	anurupvasu

When no IRQ is supported in hardware, use hrtimer to poll and
update event counter and avoid overflow condition for MN PMU.
An interval of 8 seconds is used for the hrtimer.

Signed-off-by: Anurup M <anurup.m@huawei.com>
Signed-off-by: Shaokun Zhang <zhangshaokun@hisilicon.com>
Signed-off-by: Dikshit N <dikshit.n@huawei.com>
---
 drivers/perf/hisilicon/hisi_uncore_mn.c | 39 +++++++++++++++++++++++++++++++++
 1 file changed, 39 insertions(+)

diff --git a/drivers/perf/hisilicon/hisi_uncore_mn.c b/drivers/perf/hisilicon/hisi_uncore_mn.c
index 18316dc..55b6165 100644
--- a/drivers/perf/hisilicon/hisi_uncore_mn.c
+++ b/drivers/perf/hisilicon/hisi_uncore_mn.c
@@ -20,6 +20,8 @@
  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 #include <linux/bitmap.h>
+#include <linux/hrtimer.h>
+#include <linux/ktime.h>
 #include <linux/module.h>
 #include <linux/of.h>
 #include <linux/of_device.h>
@@ -56,6 +58,15 @@ enum armv8_hisi_mn_counters {
 
 #define GET_MODULE_ID(hwmod_data) hwmod_data->mn_hwcfg.module_id
 
+/*
+ * Default timer frequency to poll and avoid counter overflow.
+ * CPU speed = 2.4Ghz and number of CPU cores in a SCCL is 16.
+ * For a single MN event on a CPU core consumes 200 cycles.
+ * So overflow time = (2^31 * 200) / (16 * 2.4G) which is about 21 seconds
+ * So on a safe side we use a timer interval of 8sec
+ */
+#define MN1_HRTIMER_INTERVAL (8LL * MSEC_PER_SEC)
+
 struct hisi_mn_hwcfg {
 	u32 module_id;
 };
@@ -272,6 +283,15 @@ static const struct of_device_id mn_of_match[] = {
 };
 MODULE_DEVICE_TABLE(of, mn_of_match);
 
+/* Use hrtimer when no IRQ, to poll for avoiding counter overflow */
+static void hisi_mn_hrtimer_init(struct hisi_pmu *mn_pmu)
+{
+	INIT_LIST_HEAD(&mn_pmu->active_list);
+	mn_pmu->ops->start_hrtimer = hisi_hrtimer_start;
+	mn_pmu->ops->stop_hrtimer = hisi_hrtimer_stop;
+	hisi_hrtimer_init(mn_pmu, MN1_HRTIMER_INTERVAL);
+}
+
 static int hisi_mn_init_data(struct hisi_pmu *mn_pmu,
 			     struct hisi_djtag_client *client)
 {
@@ -314,6 +334,12 @@ static int hisi_mn_init_data(struct hisi_pmu *mn_pmu,
 	} else
 		return -EINVAL;
 
+	/*
+	 * Use poll method to avoid counter overflow as
+	 * overflow IRQ is not supported in hardware.
+	 */
+	hisi_mn_hrtimer_init(mn_pmu);
+
 	return 0;
 }
 
@@ -361,11 +387,24 @@ static const struct attribute_group hisi_mn_cpumask_attr_group = {
 	.attrs = hisi_mn_cpumask_attrs,
 };
 
+static DEVICE_ATTR(hrtimer_interval, 0444, hisi_hrtimer_interval_sysfs_show,
+		   NULL);
+
+static struct attribute *hisi_mn_hrtimer_interval_attrs[] = {
+	&dev_attr_hrtimer_interval.attr,
+	NULL,
+};
+
+static const struct attribute_group hisi_mn_hrtimer_interval_attr_group = {
+	.attrs = hisi_mn_hrtimer_interval_attrs,
+};
+
 static const struct attribute_group *hisi_mn_pmu_attr_groups[] = {
 	&hisi_mn_attr_group,
 	&hisi_mn_format_group,
 	&hisi_mn_events_group,
 	&hisi_mn_cpumask_attr_group,
+	&hisi_mn_hrtimer_interval_attr_group,
 	NULL,
 };
 
-- 
2.1.4

^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH v6 10/11] drivers: perf: hisi: use poll method when no IRQ for MN counter overflow
  2017-03-10  6:29 [PATCH v6 10/11] drivers: perf: hisi: use poll method when no IRQ for MN counter overflow Anurup M
@ 2017-03-21 17:17 ` Mark Rutland
  2017-03-24  6:43   ` Anurup M
  0 siblings, 1 reply; 3+ messages in thread
From: Mark Rutland @ 2017-03-21 17:17 UTC (permalink / raw)
  To: Anurup M
  Cc: will.deacon, linux-kernel, linux-arm-kernel, anurup.m,
	zhangshaokun, tanxiaojun, xuwei5, sanil.kumar, john.garry,
	gabriele.paoloni, shiju.jose, huangdaode, linuxarm, dikshit.n,
	shyju.pv

On Fri, Mar 10, 2017 at 01:29:01AM -0500, Anurup M wrote:
> When no IRQ is supported in hardware, use hrtimer to poll and
> update event counter and avoid overflow condition for MN PMU.
> An interval of 8 seconds is used for the hrtimer.

Please fold this into the main MN patch.

The comments on the L3 hrtimer patch also apply here.

Thanks,
Mark.

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH v6 10/11] drivers: perf: hisi: use poll method when no IRQ for MN counter overflow
  2017-03-21 17:17 ` Mark Rutland
@ 2017-03-24  6:43   ` Anurup M
  0 siblings, 0 replies; 3+ messages in thread
From: Anurup M @ 2017-03-24  6:43 UTC (permalink / raw)
  To: Mark Rutland
  Cc: will.deacon, linux-kernel, linux-arm-kernel, anurup.m,
	zhangshaokun, tanxiaojun, xuwei5, sanil.kumar, john.garry,
	gabriele.paoloni, shiju.jose, huangdaode, linuxarm, dikshit.n,
	shyju.pv



On Tuesday 21 March 2017 10:47 PM, Mark Rutland wrote:
> On Fri, Mar 10, 2017 at 01:29:01AM -0500, Anurup M wrote:
>> When no IRQ is supported in hardware, use hrtimer to poll and
>> update event counter and avoid overflow condition for MN PMU.
>> An interval of 8 seconds is used for the hrtimer.
> Please fold this into the main MN patch.
>
> The comments on the L3 hrtimer patch also apply here.

Sure. shall fix these comments. and fold to main MN patch.

Thanks,
Anurup

> Thanks,
> Mark.

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2017-03-24  6:43 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-03-10  6:29 [PATCH v6 10/11] drivers: perf: hisi: use poll method when no IRQ for MN counter overflow Anurup M
2017-03-21 17:17 ` Mark Rutland
2017-03-24  6:43   ` Anurup M

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).