Skip to contentSkip to conversation

cpufreq: scmi: Fix OPP addition failure with a dummy clock provider

PATCH Applied Sudeep Holla 2 participants 4 messages Nov 10, 2020 – Nov 17, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 cpufreq: scmi: Fix OPP addition failure with a dummy clock provider

Commit dd461cd9183f ("opp: Allow dev_pm_opp_get_opp_table() to return -EPROBE_DEFER") handles -EPROBE_DEFER for the clock/interconnects within _allocate_opp_table() which is called from dev_pm_opp_add and it now propagates the error back to the caller.

SCMI performance domain re-used clock bindings to keep it simple. However with the above mentioned change, if clock property is present in a device node, opps fails to get added with below errors until clk_get succeeds.

cpu0: failed to add opp 450000000Hz
cpu0: failed to add opps to the device
....(errors on cpu1-cpu4)
cpu5: failed to add opp 450000000Hz
cpu5: failed to add opps to the device

So, in order to fix the issue, we need to register dummy clock provider. With the dummy clock provider, clk_get returns NULL(no errors!), then opp core proceeds to add OPPs for the CPUs.

Cc Rafael J. Wysocki <[email protected]>
Cc Viresh Kumar <[email protected]>
Fixes dd461cd9183f ("opp: Allow dev_pm_opp_get_opp_table() to return -EPROBE_DEFER")
---
 drivers/cpufreq/scmi-cpufreq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/cpufreq/scmi-cpufreq.c +6 -0
index e855e8612a67..78318508a6d6 100644
@@ -8,6 +8,7 @@
 
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 
+#include <linux/clk-provider.h>
 #include <linux/cpu.h>
 #include <linux/cpufreq.h>
 #include <linux/cpumask.h>
@@ -228,12 +229,17 @@ static struct cpufreq_driver scmi_cpufreq_driver = {
 static int scmi_cpufreq_probe(struct scmi_device *sdev)
 {
 	int ret;
+	struct device *dev = &sdev->dev;
 
 	handle = sdev->handle;
 
 	if (!handle || !handle->perf_ops)
 		return -ENODEV;
 
+	/* dummy clock provider as needed by OPP if clocks property is used */
+	if (of_find_property(dev->of_node, "#clock-cells", NULL))
+		devm_of_clk_add_hw_provider(dev, of_clk_hw_simple_get, NULL);
+
 	ret = cpufreq_register_driver(&scmi_cpufreq_driver);
 	if (ret) {
 		dev_err(&sdev->dev, "%s: registering cpufreq failed, err: %d\n",
Signature · 1 line
2.25.1
maintainer Acked-by on v1 1/1
#1 lore raw
Quoted 47 lines · Sudeep Holla

Commit dd461cd9183f ("opp: Allow dev_pm_opp_get_opp_table() to return -EPROBE_DEFER") handles -EPROBE_DEFER for the clock/interconnects within _allocate_opp_table() which is called from dev_pm_opp_add and it now propagates the error back to the caller.

SCMI performance domain re-used clock bindings to keep it simple. However with the above mentioned change, if clock property is present in a device node, opps fails to get added with below errors until clk_get succeeds.

cpu0: failed to add opp 450000000Hz
cpu0: failed to add opps to the device
....(errors on cpu1-cpu4)
cpu5: failed to add opp 450000000Hz
cpu5: failed to add opps to the device

So, in order to fix the issue, we need to register dummy clock provider. With the dummy clock provider, clk_get returns NULL(no errors!), then opp core proceeds to add OPPs for the CPUs.

Cc Rafael J. Wysocki <[email protected]>
Cc Viresh Kumar <[email protected]>
Fixes dd461cd9183f ("opp: Allow dev_pm_opp_get_opp_table() to return -EPROBE_DEFER")
---
 drivers/cpufreq/scmi-cpufreq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/cpufreq/scmi-cpufreq.c +2 -0
index e855e8612a67..78318508a6d6 100644
@@ -8,6 +8,7 @@
 
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 
+#include <linux/clk-provider.h>
 #include <linux/cpu.h>
 #include <linux/cpufreq.h>
 #include <linux/cpumask.h>
@@ -228,12 +229,17 @@ static struct cpufreq_driver scmi_cpufreq_driver = {
 static int scmi_cpufreq_probe(struct scmi_device *sdev)
 {
 	int ret;
+	struct device *dev = &sdev->dev;
 
 	handle = sdev->handle;
 

drivers/cpufreq/scmi-cpufreq.c static struct cpufreq_driver scmi_cpufreq_driver = { · line 245

Quoted 10 lines
drivers/cpufreq/scmi-cpufreq.c +4 -0
 	if (!handle || !handle->perf_ops)
 		return -ENODEV;
 
+	/* dummy clock provider as needed by OPP if clocks property is used */
+	if (of_find_property(dev->of_node, "#clock-cells", NULL))
+		devm_of_clk_add_hw_provider(dev, of_clk_hw_simple_get, NULL);
+
 	ret = cpufreq_register_driver(&scmi_cpufreq_driver);
 	if (ret) {
 		dev_err(&sdev->dev, "%s: registering cpufreq failed, err: %d\n",

For 5.10-rc.

Signature · 1 line
viresh
#2 lore raw
Quoted 61 lines · Viresh Kumar
Quoted 47 lines · Sudeep Holla

Commit dd461cd9183f ("opp: Allow dev_pm_opp_get_opp_table() to return -EPROBE_DEFER") handles -EPROBE_DEFER for the clock/interconnects within _allocate_opp_table() which is called from dev_pm_opp_add and it now propagates the error back to the caller.

SCMI performance domain re-used clock bindings to keep it simple. However with the above mentioned change, if clock property is present in a device node, opps fails to get added with below errors until clk_get succeeds.

cpu0: failed to add opp 450000000Hz
cpu0: failed to add opps to the device
....(errors on cpu1-cpu4)
cpu5: failed to add opp 450000000Hz
cpu5: failed to add opps to the device

So, in order to fix the issue, we need to register dummy clock provider. With the dummy clock provider, clk_get returns NULL(no errors!), then opp core proceeds to add OPPs for the CPUs.

Cc Rafael J. Wysocki <[email protected]>
Cc Viresh Kumar <[email protected]>
Fixes dd461cd9183f ("opp: Allow dev_pm_opp_get_opp_table() to return -EPROBE_DEFER")
---
 drivers/cpufreq/scmi-cpufreq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/cpufreq/scmi-cpufreq.c +2 -0
index e855e8612a67..78318508a6d6 100644
@@ -8,6 +8,7 @@
 
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 
+#include <linux/clk-provider.h>
 #include <linux/cpu.h>
 #include <linux/cpufreq.h>
 #include <linux/cpumask.h>
@@ -228,12 +229,17 @@ static struct cpufreq_driver scmi_cpufreq_driver = {
 static int scmi_cpufreq_probe(struct scmi_device *sdev)
 {
 	int ret;
+	struct device *dev = &sdev->dev;
 
 	handle = sdev->handle;
 

drivers/cpufreq/scmi-cpufreq.c static struct cpufreq_driver scmi_cpufreq_driver = { · line 245

Quoted 10 lines
drivers/cpufreq/scmi-cpufreq.c +4 -0
 	if (!handle || !handle->perf_ops)
 		return -ENODEV;
 
+	/* dummy clock provider as needed by OPP if clocks property is used */
+	if (of_find_property(dev->of_node, "#clock-cells", NULL))
+		devm_of_clk_add_hw_provider(dev, of_clk_hw_simple_get, NULL);
+
 	ret = cpufreq_register_driver(&scmi_cpufreq_driver);
 	if (ret) {
 		dev_err(&sdev->dev, "%s: registering cpufreq failed, err: %d\n",

For 5.10-rc.

Quoted 2 lines

Thanks Viresh, I assume Rafael will take this directly as fix for v5.10

Signature · 2 lines
Regards,
Sudeep
maintainer Applied
#3 lore raw
In reply to Sudeep Holla
Quoted 19 lines · Sudeep Holla

Commit dd461cd9183f ("opp: Allow dev_pm_opp_get_opp_table() to return -EPROBE_DEFER") handles -EPROBE_DEFER for the clock/interconnects within _allocate_opp_table() which is called from dev_pm_opp_add and it now propagates the error back to the caller.

SCMI performance domain re-used clock bindings to keep it simple. However with the above mentioned change, if clock property is present in a device node, opps fails to get added with below errors until clk_get succeeds.

cpu0: failed to add opp 450000000Hz
cpu0: failed to add opps to the device
....(errors on cpu1-cpu4)
cpu5: failed to add opp 450000000Hz
cpu5: failed to add opps to the device

So, in order to fix the issue, we need to register dummy clock provider. With the dummy clock provider, clk_get returns NULL(no errors!), then opp core proceeds to add OPPs for the CPUs.

Quoted 7 lines
Cc Rafael J. Wysocki <[email protected]>
Cc Viresh Kumar <[email protected]>
Fixes dd461cd9183f ("opp: Allow dev_pm_opp_get_opp_table() to return -EPROBE_DEFER")
---
 drivers/cpufreq/scmi-cpufreq.c | 6 ++++++
 1 file changed, 6 insertions(+)

Applied. Thanks.

Signature · 1 line
viresh