Skip to contentSkip to conversation

allow idle state to be found as deepest state for s2idle only

RFC PATCH In review [email protected] 2 participants 7 messages Apr 13, 2020 – Apr 21, 2020

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

cover allow idle state to be found as deepest state for s2idle only

From: Chunyan Zhang <[email protected]>

We can add a new c-state to let system suspend enter into the same deep state as s2ram when using s2idle, but we don't want other cases to use that c-state, such as play_idle() which would find the deepest state by invoking find_deepest_state() as well. This patch-set provides users an option which can restrict c-state to be found as deepest state for s2idle only.

Chunyan Zhang (2):
cpuidle: allow idle state to be found as deepest state for s2idle only
dt-bindings: arm: Add description to the new property for-s2idle-only

 Documentation/devicetree/bindings/arm/idle-states.yaml | 10 +++++++++-
 drivers/cpuidle/cpuidle.c                              |  3 ++-
 drivers/cpuidle/dt_idle_states.c                       |  3 +++
 include/linux/cpuidle.h                                |  1 +
 4 files changed, 15 insertions(+), 2 deletions(-)
Signature · 1 line
2.20.1
#1 lore raw

RFC PATCH 1/2 cpuidle: allow idle state to be found as deepest state for s2idle only

From: Chunyan Zhang <[email protected]>

Add a new flag CPUIDLE_FLAG_S2IDLE to allow c-state to be found as deepest state for s2idle only, so that users can add a new c-state for using s2idle and don't worry disturbing other use cases such as play_idle() which probably don't want to enter into so much deep idle state since devices are not suspended for that kind of cases.

---
 drivers/cpuidle/cpuidle.c        | 3 ++-
 drivers/cpuidle/dt_idle_states.c | 3 +++
 include/linux/cpuidle.h          | 1 +
 3 files changed, 6 insertions(+), 1 deletion(-)
drivers/cpuidle/cpuidle.c +2 -1
index de81298051b3..bb61f0c271d2 100644
@@ -89,7 +89,8 @@ static int find_deepest_state(struct cpuidle_driver *drv,
 		    s->exit_latency_ns <= latency_req ||
 		    s->exit_latency_ns > max_latency_ns ||
 		    (s->flags & forbidden_flags) ||
-		    (s2idle && !s->enter_s2idle))
+		    (s2idle && !s->enter_s2idle) ||
+		    (!s2idle && (s->flags & CPUIDLE_FLAG_S2ILDE)))
 			continue;
 
 		latency_req = s->exit_latency_ns;
drivers/cpuidle/dt_idle_states.c +3 -0
index 252f2a9686a6..530db2726c05 100644
@@ -80,6 +80,9 @@ static int init_state_node(struct cpuidle_state *idle_state,
 	idle_state->flags = 0;
 	if (of_property_read_bool(state_node, "local-timer-stop"))
 		idle_state->flags |= CPUIDLE_FLAG_TIMER_STOP;
+
+	if (of_property_read_bool(state_node, "for-s2idle-only"))
+		idle_state->flags |= CPUIDLE_FLAG_S2ILDE;
 	/*
 	 * TODO:
 	 *	replace with kstrdup and pointer assignment when name
include/linux/cpuidle.h +1 -0
index ec2ef63771f0..08da701f74cd 100644
@@ -78,6 +78,7 @@ struct cpuidle_state {
 #define CPUIDLE_FLAG_TIMER_STOP BIT(2) /* timer is stopped on this state */
 #define CPUIDLE_FLAG_UNUSABLE	BIT(3) /* avoid using this state */
 #define CPUIDLE_FLAG_OFF	BIT(4) /* disable this state by default */
+#define CPUIDLE_FLAG_S2ILDE	BIT(5) /* state is used for s2idle only */
 
 struct cpuidle_device_kobj;
 struct cpuidle_state_kobj;
Signature · 1 line
2.20.1
#2 lore raw

Re: [RFC PATCH v1 1/2] cpuidle: allow idle state to be found as deepest state for s2idle only

Hello,

Any comments or suggests on this? That would be very appreciated.

Thanks,
Chunyan

Quoted 59 lines · On Mon, Apr 13, 2020 at 5:09 PM <[email protected]> wrote:

From: Chunyan Zhang <[email protected]>

Add a new flag CPUIDLE_FLAG_S2IDLE to allow c-state to be found as deepest state for s2idle only, so that users can add a new c-state for using s2idle and don't worry disturbing other use cases such as play_idle() which probably don't want to enter into so much deep idle state since devices are not suspended for that kind of cases.

---
 drivers/cpuidle/cpuidle.c        | 3 ++-
 drivers/cpuidle/dt_idle_states.c | 3 +++
 include/linux/cpuidle.h          | 1 +
 3 files changed, 6 insertions(+), 1 deletion(-)
drivers/cpuidle/cpuidle.c +2 -1
index de81298051b3..bb61f0c271d2 100644
@@ -89,7 +89,8 @@ static int find_deepest_state(struct cpuidle_driver *drv,
                    s->exit_latency_ns <= latency_req ||
                    s->exit_latency_ns > max_latency_ns ||
                    (s->flags & forbidden_flags) ||
-                   (s2idle && !s->enter_s2idle))
+                   (s2idle && !s->enter_s2idle) ||
+                   (!s2idle && (s->flags & CPUIDLE_FLAG_S2ILDE)))
                        continue;
 
                latency_req = s->exit_latency_ns;
drivers/cpuidle/dt_idle_states.c +3 -0
index 252f2a9686a6..530db2726c05 100644
@@ -80,6 +80,9 @@ static int init_state_node(struct cpuidle_state *idle_state,
        idle_state->flags = 0;
        if (of_property_read_bool(state_node, "local-timer-stop"))
                idle_state->flags |= CPUIDLE_FLAG_TIMER_STOP;
+
+       if (of_property_read_bool(state_node, "for-s2idle-only"))
+               idle_state->flags |= CPUIDLE_FLAG_S2ILDE;
        /*
         * TODO:
         *      replace with kstrdup and pointer assignment when name
include/linux/cpuidle.h +1 -1
index ec2ef63771f0..08da701f74cd 100644
@@ -78,6 +78,7 @@ struct cpuidle_state {
 #define CPUIDLE_FLAG_TIMER_STOP BIT(2) /* timer is stopped on this state */
 #define CPUIDLE_FLAG_UNUSABLE  BIT(3) /* avoid using this state */
 #define CPUIDLE_FLAG_OFF       BIT(4) /* disable this state by default */
+#define CPUIDLE_FLAG_S2ILDE    BIT(5) /* state is used for s2idle only */
 
 struct cpuidle_device_kobj;
 struct cpuidle_state_kobj;
--

2.20.1

#3 lore raw

Re: [RFC PATCH v1 1/2] cpuidle: allow idle state to be found as deepest state for s2idle only

In reply to [email protected]
Quoted 7 lines · On Mon, Apr 13, 2020 at 03:00:13PM +0800, [email protected] wrote:

From: Chunyan Zhang <[email protected]>

Add a new flag CPUIDLE_FLAG_S2IDLE to allow c-state to be found as deepest state for s2idle only, so that users can add a new c-state for using s2idle and don't worry disturbing other use cases such as play_idle() which probably don't want to enter into so much deep idle state since devices are not suspended for that kind of cases.

Can you please elaborate on this?

Why exactly are these states not suited for regular cpu idle? What problems do they cause? e.g. long wakeup latency?

The flag and the for-s2-idle-only DT property are encoding a policy rarher than a property, and as such I don't think this is the right way to describe this in the DT. However, if there might be porperties of the idle state that we could describe so that the OS can come to the same conclusion.

Thanks,
Mark.

Quoted 51 lines
---
 drivers/cpuidle/cpuidle.c        | 3 ++-
 drivers/cpuidle/dt_idle_states.c | 3 +++
 include/linux/cpuidle.h          | 1 +
 3 files changed, 6 insertions(+), 1 deletion(-)
drivers/cpuidle/cpuidle.c +2 -1
index de81298051b3..bb61f0c271d2 100644
@@ -89,7 +89,8 @@ static int find_deepest_state(struct cpuidle_driver *drv,
 		    s->exit_latency_ns <= latency_req ||
 		    s->exit_latency_ns > max_latency_ns ||
 		    (s->flags & forbidden_flags) ||
-		    (s2idle && !s->enter_s2idle))
+		    (s2idle && !s->enter_s2idle) ||
+		    (!s2idle && (s->flags & CPUIDLE_FLAG_S2ILDE)))
 			continue;
 
 		latency_req = s->exit_latency_ns;
drivers/cpuidle/dt_idle_states.c +3 -0
index 252f2a9686a6..530db2726c05 100644
@@ -80,6 +80,9 @@ static int init_state_node(struct cpuidle_state *idle_state,
 	idle_state->flags = 0;
 	if (of_property_read_bool(state_node, "local-timer-stop"))
 		idle_state->flags |= CPUIDLE_FLAG_TIMER_STOP;
+
+	if (of_property_read_bool(state_node, "for-s2idle-only"))
+		idle_state->flags |= CPUIDLE_FLAG_S2ILDE;
 	/*
 	 * TODO:
 	 *	replace with kstrdup and pointer assignment when name
include/linux/cpuidle.h +1 -0
index ec2ef63771f0..08da701f74cd 100644
@@ -78,6 +78,7 @@ struct cpuidle_state {
 #define CPUIDLE_FLAG_TIMER_STOP BIT(2) /* timer is stopped on this state */
 #define CPUIDLE_FLAG_UNUSABLE	BIT(3) /* avoid using this state */
 #define CPUIDLE_FLAG_OFF	BIT(4) /* disable this state by default */
+#define CPUIDLE_FLAG_S2ILDE	BIT(5) /* state is used for s2idle only */
 
 struct cpuidle_device_kobj;
 struct cpuidle_state_kobj;
Signature · 1 line
2.20.1
#4 lore raw

Re: [RFC PATCH v1 1/2] cpuidle: allow idle state to be found as deepest state for s2idle only

Hi Mark,

Many thanks for your comments.

Quoted 11 lines · Mark Rutland
Quoted 7 lines · On Mon, Apr 13, 2020 at 03:00:13PM +0800, [email protected] wrote:

From: Chunyan Zhang <[email protected]>

Add a new flag CPUIDLE_FLAG_S2IDLE to allow c-state to be found as deepest state for s2idle only, so that users can add a new c-state for using s2idle and don't worry disturbing other use cases such as play_idle() which probably don't want to enter into so much deep idle state since devices are not suspended for that kind of cases.

Can you please elaborate on this?

Ok.

The thing was, I added a new c-state (named DOMAIN_PD for example) in DT for using s2idle, and the target power level indicated in DOMAIN_PD would be deeper than the level for regular cpuidle (for example level-0 is for regular cpuidle; level-1 is for system suspend and power domain would be shutdown as well as all cores , DOMAIN_PD uses level-1). I worried that would cause the deeper power domain to be shutdown if DOMAIN_PD was selected by play_idle().

But after have another look at PSCI in ATF, I consider that it probably is not a problem which would really happen. Since play_idle() wouldn't occur on all cpus at the same time, although play_idle() could use DOMAIN_PD, the system wouldn't enter into that so deep power level.

Hope I've explained the things clearly :)

In a word, this patch seems not needed for now.

Thanks again,
Chunyan

Quoted 64 lines

Why exactly are these states not suited for regular cpu idle? What problems do they cause? e.g. long wakeup latency?

The flag and the for-s2-idle-only DT property are encoding a policy rarher than a property, and as such I don't think this is the right way to describe this in the DT. However, if there might be porperties of the idle state that we could describe so that the OS can come to the same conclusion.

Thanks,
Mark.

Quoted 51 lines
---
 drivers/cpuidle/cpuidle.c        | 3 ++-
 drivers/cpuidle/dt_idle_states.c | 3 +++
 include/linux/cpuidle.h          | 1 +
 3 files changed, 6 insertions(+), 1 deletion(-)
drivers/cpuidle/cpuidle.c +2 -1
index de81298051b3..bb61f0c271d2 100644
@@ -89,7 +89,8 @@ static int find_deepest_state(struct cpuidle_driver *drv,
                  s->exit_latency_ns <= latency_req ||
                  s->exit_latency_ns > max_latency_ns ||
                  (s->flags & forbidden_flags) ||
-                 (s2idle && !s->enter_s2idle))
+                 (s2idle && !s->enter_s2idle) ||
+                 (!s2idle && (s->flags & CPUIDLE_FLAG_S2ILDE)))
                      continue;
 
              latency_req = s->exit_latency_ns;
drivers/cpuidle/dt_idle_states.c +3 -0
index 252f2a9686a6..530db2726c05 100644
@@ -80,6 +80,9 @@ static int init_state_node(struct cpuidle_state *idle_state,
      idle_state->flags = 0;
      if (of_property_read_bool(state_node, "local-timer-stop"))
              idle_state->flags |= CPUIDLE_FLAG_TIMER_STOP;
+
+     if (of_property_read_bool(state_node, "for-s2idle-only"))
+             idle_state->flags |= CPUIDLE_FLAG_S2ILDE;
      /*
       * TODO:
       *      replace with kstrdup and pointer assignment when name
include/linux/cpuidle.h +1 -1
index ec2ef63771f0..08da701f74cd 100644
@@ -78,6 +78,7 @@ struct cpuidle_state {
 #define CPUIDLE_FLAG_TIMER_STOP BIT(2) /* timer is stopped on this state */
 #define CPUIDLE_FLAG_UNUSABLE        BIT(3) /* avoid using this state */
 #define CPUIDLE_FLAG_OFF     BIT(4) /* disable this state by default */
+#define CPUIDLE_FLAG_S2ILDE  BIT(5) /* state is used for s2idle only */
 
 struct cpuidle_device_kobj;
 struct cpuidle_state_kobj;
--

2.20.1

#5 lore raw

RFC PATCH 2/2 dt-bindings: arm: Add description to the new property for-s2idle-only

In reply to [email protected]

From: Chunyan Zhang <[email protected]>

Add a new property for-s2idle-only. The idle-state marked with this property will be set with CPUIDLE_FLAG_S2IDLE during initialization and it would be expected to be found as deepest state for s2idle rather than other cases like play_idle().

---
 Documentation/devicetree/bindings/arm/idle-states.yaml | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
Documentation/devicetree/bindings/arm/idle-states.yaml +9 -1
index ea805c1e6b20..cec47b3a447f 100644
@@ -263,7 +263,6 @@ patternProperties:
     description: |
       Each state node represents an idle state description and must be defined
       as follows.
-
       The idle state entered by executing the wfi instruction (idle_standby
       SBSA,[3][4]) is considered standard on all ARM platforms and therefore
       must not be listed.
@@ -283,6 +282,15 @@ patternProperties:
              lost on state entry, otherwise it is retained.
         type: boolean
 
+      for-s2idle-only:
+        description:
+          This indicates that the state only can be found as deepest state
+          for s2idle rather than other cases like play_idle(). In general,
+          the state having this property should have longer min-residency
+          than the cpuidle target min-residency which CPU QoS constraints
+          defines, to avoid being used by runtime cpuidle.
+        type: boolean
+
       entry-latency-us:
         description:
           Worst case latency in microseconds required to enter the idle state.
Signature · 1 line
2.20.1
#6 lore raw

Re: [RFC PATCH v1 2/2] dt-bindings: arm: Add description to the new property for-s2idle-only

Quoted 26 lines · On Mon, Apr 13, 2020 at 03:00:14PM +0800, [email protected] wrote:

From: Chunyan Zhang <[email protected]>

Add a new property for-s2idle-only. The idle-state marked with this property will be set with CPUIDLE_FLAG_S2IDLE during initialization and it would be expected to be found as deepest state for s2idle rather than other cases like play_idle().

---
 Documentation/devicetree/bindings/arm/idle-states.yaml | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
Documentation/devicetree/bindings/arm/idle-states.yaml +0 -1
index ea805c1e6b20..cec47b3a447f 100644
@@ -263,7 +263,6 @@ patternProperties:
     description: |
       Each state node represents an idle state description and must be defined
       as follows.
-
       The idle state entered by executing the wfi instruction (idle_standby
       SBSA,[3][4]) is considered standard on all ARM platforms and therefore
       must not be listed.
@@ -283,6 +282,15 @@ patternProperties:
              lost on state entry, otherwise it is retained.

Documentation/devicetree/bindings/arm/idle-states.yaml patternProperties: · line 292

Quoted 10 lines
Documentation/devicetree/bindings/arm/idle-states.yaml +8 -0
         type: boolean
 
+      for-s2idle-only:
+        description:
+          This indicates that the state only can be found as deepest state
+          for s2idle rather than other cases like play_idle(). In general,
+          the state having this property should have longer min-residency
+          than the cpuidle target min-residency which CPU QoS constraints
+          defines, to avoid being used by runtime cpuidle.
+        type: boolean

This is very Linux-specific, and is encoding a policy ratehr than a property of the state.

As on patch 1, can you please describe the expected properties of this idle state, with some rationale as to why it's not suited for idle in general? A real example in the commit message would be very helpful.

Thanks,
Mark.