cpuidle: ladder: Add per CPU PM QoS resume latency support

Individual CPUs may have special requirements to not enter deep idle states. For example, a CPU running real time applications would not want to enter deep idle states to avoid latency impacts. At the same time other CPUs that do not have such a requirement could allow deep idle states to save power.

This was already implemented in the menu governor. Implementing similar changes in the ladder governor which gets selected when CONFIG_NO_HZ and CONFIG_NO_HZ_IDLE are not set. Refer following commits for the menu governor changes.

commit 9908859acaa9 ("cpuidle/menu: add per CPU PM QoS resume latency consideration")
commit 6dbf5cea05a7 ("cpuidle: menu: Avoid taking spinlock for accessing QoS values")

Signed-off-by Ramesh Thomas <[email protected]>

---

v2:
- use PM_QOS_RESUME_LATENCY_NO_CONSTRAINT for "no constraint" value Should be applied over https://patchwork.kernel.org/patch/10024157/

 drivers/cpuidle/governors/ladder.c | 7 +++++++
 1 file changed, 7 insertions(+)

drivers/cpuidle/governors/ladder.c +7 -1

index ce1a2ff..1ad8745 100644
@@ -17,6 +17,7 @@
 #include <linux/pm_qos.h>
 #include <linux/jiffies.h>
 #include <linux/tick.h>
+#include <linux/cpu.h>
 
 #include <asm/io.h>
 #include <linux/uaccess.h>

@@ -67,10 +68,16 @@ static int ladder_select_state(struct cpuidle_driver *drv,
                                struct cpuidle_device *dev)
 {
        struct ladder_device *ldev = this_cpu_ptr(&ladder_devices);
+       struct device *device = get_cpu_device(dev->cpu);
        struct ladder_device_state *last_state;
        int last_residency, last_idx = ldev->last_state_idx;
        int first_idx = drv->states[0].flags & CPUIDLE_FLAG_POLLING ? 1 : 0;
        int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY);
+       int resume_latency = dev_pm_qos_raw_read_value(device);
+
+       if (resume_latency < latency_req &&
+           resume_latency != PM_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+               latency_req = resume_latency;
 
        /* Special case when user has set very strict latency requirement */
        if (unlikely(latency_req == 0)) {
--

2.7.4

Quoted 57 lines · Ramesh Thomas

Individual CPUs may have special requirements to not enter deep idle states. For example, a CPU running real time applications would not want to enter deep idle states to avoid latency impacts. At the same time other CPUs that do not have such a requirement could allow deep idle states to save power.

This was already implemented in the menu governor. Implementing similar changes in the ladder governor which gets selected when CONFIG_NO_HZ and CONFIG_NO_HZ_IDLE are not set. Refer following commits for the menu governor changes.

commit 9908859acaa9 ("cpuidle/menu: add per CPU PM QoS resume latency consideration")
commit 6dbf5cea05a7 ("cpuidle: menu: Avoid taking spinlock for accessing QoS values")

Signed-off-by Ramesh Thomas <[email protected]>

---

v2:
- use PM_QOS_RESUME_LATENCY_NO_CONSTRAINT for "no constraint" value Should be applied over https://patchwork.kernel.org/patch/10024157/

 drivers/cpuidle/governors/ladder.c | 7 +++++++
 1 file changed, 7 insertions(+)

drivers/cpuidle/governors/ladder.c +7 -1

index ce1a2ff..1ad8745 100644
@@ -17,6 +17,7 @@
 #include <linux/pm_qos.h>
 #include <linux/jiffies.h>
 #include <linux/tick.h>
+#include <linux/cpu.h>
 
 #include <asm/io.h>
 #include <linux/uaccess.h>

@@ -67,10 +68,16 @@ static int ladder_select_state(struct cpuidle_driver *drv,
                                struct cpuidle_device *dev)
 {
        struct ladder_device *ldev = this_cpu_ptr(&ladder_devices);
+       struct device *device = get_cpu_device(dev->cpu);
        struct ladder_device_state *last_state;
        int last_residency, last_idx = ldev->last_state_idx;
        int first_idx = drv->states[0].flags & CPUIDLE_FLAG_POLLING ? 1 : 0;
        int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY);
+       int resume_latency = dev_pm_qos_raw_read_value(device);
+
+       if (resume_latency < latency_req &&
+           resume_latency != PM_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+               latency_req = resume_latency;
 
        /* Special case when user has set very strict latency requirement */
        if (unlikely(latency_req == 0)) {
--

Quoted 2 lines

2.7.4

Looks good to me.

I'll queue it up if nobody objects.