PM / Domains: Updates to IRQ safe PM domains
Subsystems (7)
cover PM / Domains: Updates to IRQ safe PM domains
Hi,
This follows the discussions at Linux Plumbers Conference on PM domains and idle states. In the context of CPU hierarchy, the CPU's idle states may be presented using the property domain-idle-states if the domain is considered as the boundary that powers on/off an individual CPU.
To that effect, domain idle states may not be specified but not compatible with what PM domains expects. Hence ignore the idle states that are not "domain-idle-state" compatible.
The second patch covers the case, where the hierarchy of PM domains is not powered off in the same context. Currently, a workqueue is scheduled to power off the parent domains in the hierarchy. In case of CPU PM domains, this will not work, since we are running with interrupts disabled. Hence for IRQ safe PM domains, power off the parent domains in the same context as the caller.
I understand Ulf is working on some argument changes in the same area of functions. Will work with him to resolve conflicts before the merge. This is based on the tip of Rafel's next.
Thanks,
Lina
Lina Iyer (2):
PM / Domains: Ignore domain-idle-states that are not compatible PM / Domains: Support in context powering off parent domain
.../devicetree/bindings/power/power_domain.txt | 4 ++- drivers/base/power/domain.c | 37 +++++++++++++++------- 2 files changed, 28 insertions(+), 13 deletions(-)
Signature · 1 line
2.7.4
PATCH 1/1 PM / Domains: Fix compatible for domain idle state
Re-using idle state definition provided by arm,idle-state for domain idle states creates a lot of confusion and limits further evolution of the domain idle definition. To keep things clear and simple, define a idle states for domain using a new compatible "domain-idle-state".
Fix existing PM domains code to look for the newly defined compatible.
--- .../bindings/power/domain-idle-state.txt | 33 ++++++++++++++++++++++ .../devicetree/bindings/power/power_domain.txt | 8 +++--- drivers/base/power/domain.c | 2 +- 3 files changed, 38 insertions(+), 5 deletions(-) create mode 100644 Documentation/devicetree/bindings/power/domain-idle-state.txt
Documentation/devicetree/bindings/power/domain-idle-state.txt new file +33 -0
@@ -0,0 +1,33 @@ +PM Domain Idle State Node: + +A domain idle state node represents the state parameters that will be used to +select the state when there are no active components in the domain. + +The state node has the following parameters - + +- compatible: + Usage: Required + Value type: <string> + Definition: Must be "domain-idle-state". + +- entry-latency-us + Usage: Required + Value type: <prop-encoded-array> + Definition: u32 value representing worst case latency in + microseconds required to enter the idle state. + The exit-latency-us duration may be guaranteed + only after entry-latency-us has passed. + +- exit-latency-us + Usage: Required + Value type: <prop-encoded-array> + Definition: u32 value representing worst case latency + in microseconds required to exit the idle state. + +- min-residency-us + Usage: Required + Value type: <prop-encoded-array> + Definition: u32 value representing minimum residency duration + in microseconds after which the idle state will yield + power benefits after overcoming the overhead in entering +i the idle state.
Documentation/devicetree/bindings/power/power_domain.txt +4 -4
@@ -31,7 +31,7 @@ Optional properties: - domain-idle-states : A phandle of an idle-state that shall be soaked into a generic domain power state. The idle state definitions are - compatible with arm,idle-state specified in [1]. + compatible with domain-idle-state specified in [1]. The domain-idle-state property reflects the idle state of this PM domain and not the idle states of the devices or sub-domains in the PM domain. Devices and sub-domains have their own idle-states independent of the parent
@@ -85,7 +85,7 @@ Example 3: }; DOMAIN_RET: state at 0 { - compatible = "arm,idle-state"; + compatible = "domain-idle-state"; reg = <0x0>; entry-latency-us = <1000>; exit-latency-us = <2000>;
@@ -93,7 +93,7 @@ Example 3: }; DOMAIN_PWR_DN: state at 1 { - compatible = "arm,idle-state"; + compatible = "domain-idle-state"; reg = <0x1>; entry-latency-us = <5000>; exit-latency-us = <8000>;
@@ -118,4 +118,4 @@ The node above defines a typical PM domain consumer device, which is located inside a PM domain with index 0 of a power controller represented by a node with the label "power". -[1]. Documentation/devicetree/bindings/arm/idle-states.txt +[1]. Documentation/devicetree/bindings/power/domain-idle-state.txt
drivers/base/power/domain.c +1 -1
@@ -2048,7 +2048,7 @@ int genpd_dev_pm_attach(struct device *dev)EXPORT_SYMBOL_GPL (genpd_dev_pm_attach);static const struct of_device_id idle_state_match[] = { - { .compatible ="arm,idle-state" , }, + { .compatible ="domain-idle-state" , }, { } };
Signature · 1 line
2.7.4
PATCH 1/1 PM / Domains: Fix compatible for domain idle state
Quoted 10 lines · Lina Iyer
Re-using idle state definition provided by arm,idle-state for domain idle states creates a lot of confusion and limits further evolution of the domain idle definition. To keep things clear and simple, define a idle states for domain using a new compatible "domain-idle-state".
Fix existing PM domains code to look for the newly defined compatible.
Sorry, this has already been applied. Got tagged along with the email. Kindly ignore.
Thanks,
Lina
Quoted 99 lines
---
.../bindings/power/domain-idle-state.txt | 33 ++++++++++++++++++++++ .../devicetree/bindings/power/power_domain.txt | 8 +++--- drivers/base/power/domain.c | 2 +- 3 files changed, 38 insertions(+), 5 deletions(-)create mode 100644
Documentation/devicetree/bindings/power/domain-idle-state.txtDocumentation/devicetree/bindings/power/domain-idle-state.txt new file +33
-0@@ -0,0 +1,33 @@ +PM Domain Idle State Node: + +A domain idle state node represents the state parameters that will be used to +select the state when there are no active components in the domain. + +The state node has the following parameters - + +- compatible: + Usage: Required + Value type: <string> + Definition: Must be "domain-idle-state". + +- entry-latency-us + Usage: Required + Value type: <prop-encoded-array> + Definition: u32 value representing worst case latency in + microseconds required to enter the idle state. + The exit-latency-us duration may be guaranteed + only after entry-latency-us has passed. + +- exit-latency-us + Usage: Required + Value type: <prop-encoded-array> + Definition: u32 value representing worst case latency + in microseconds required to exit the idle state. + +- min-residency-us + Usage: Required + Value type: <prop-encoded-array> + Definition: u32 value representing minimum residency duration + in microseconds after which the idle state will yield + power benefits after overcoming the overhead in entering +i the idle state.Documentation/devicetree/bindings/power/power_domain.txt +3
-4@@ -31,7 +31,7 @@ Optional properties: - domain-idle-states : A phandle of an idle-state that shall be soaked into a generic domain power state. The idle state definitions are - compatible with arm,idle-state specified in [1]. + compatible with domain-idle-state specified in [1]. The domain-idle-state property reflects the idle state of this PM domain and not the idle states of the devices or sub-domains in the PM domain. Devices and sub-domains have their own idle-states independent of the parent@@ -85,7 +85,7 @@ Example 3: }; DOMAIN_RET: state at 0 { - compatible = "arm,idle-state"; + compatible = "domain-idle-state"; reg = <0x0>; entry-latency-us = <1000>; exit-latency-us = <2000>;@@ -93,7 +93,7 @@ Example 3: }; DOMAIN_PWR_DN: state at 1 { - compatible = "arm,idle-state"; + compatible = "domain-idle-state"; reg = <0x1>; entry-latency-us = <5000>; exit-latency-us = <8000>;@@ -118,4 +118,4 @@ The node above defines a typical PM domain consumer device, which is locatedinside a PM domain with index 0 of a power controller represented by a node with the label "power".
-[1].
Documentation/devicetree/bindings/arm/idle-states.txt+[1].Documentation/devicetree/bindings/power/domain-idle-state.txtdrivers/base/power/domain.c +1
-1@@ -2048,7 +2048,7 @@ int genpd_dev_pm_attach(struct device *dev)EXPORT_SYMBOL_GPL (genpd_dev_pm_attach);static const struct of_device_id idle_state_match[] = { - { .compatible ="arm,idle-state" , }, + { .compatible ="domain-idle-state" , }, { } };Signature · 1 line
2.7.4
PATCH 1/2 PM / Domains: Ignore domain-idle-states that are not compatible
In reply to Lina Iyerdomain-idle-states property may have phandles to idle state bindings that may not be compatible with idle state definition defined in [1]. Such phandles would just be ignored and not throw and error when read by the domain core.
--- Documentation/devicetree/bindings/power/power_domain.txt | 4 +++- drivers/base/power/domain.c | 16 +++++++++------- 2 files changed, 12 insertions(+), 8 deletions(-)
Documentation/devicetree/bindings/power/power_domain.txt +3 -1
@@ -31,7 +31,9 @@ Optional properties: - domain-idle-states : A phandle of an idle-state that shall be soaked into a generic domain power state. The idle state definitions are - compatible with domain-idle-state specified in [1]. + compatible with domain-idle-state specified in [1]. phandles + that are not compatible with domain-idle-state will be + ignored. The domain-idle-state property reflects the idle state of this PM domain and not the idle states of the devices or sub-domains in the PM domain. Devices and sub-domains have their own idle-states independent of the parent
drivers/base/power/domain.c +9 -7
@@ -2065,11 +2065,6 @@ static int genpd_parse_state(struct genpd_power_state *genpd_state,int err;u32 residency;u32 entry_latency, exit_latency; -const struct of_device_id *match_id; - - match_id =of_match_node (idle_state_match, state_node); -if (!match_id) -return -EINVAL ; err =of_property_read_u32 (state_node,"entry-latency-us" , &entry_latency);
@@ -2118,6 +2113,7 @@ int of_genpd_parse_idle_states(struct device_node *dn,int err, ret;int count;struct of_phandle_iterator it; +const struct of_device_id *match_id; count =of_count_phandle_with_args (dn,"domain-idle-states" ,NULL );if (count <=0 )
@@ -2130,6 +2126,9 @@ int of_genpd_parse_idle_states(struct device_node *dn,/* Loop over the phandles until all the requested entry is found */ of_for_each_phandle (&it, err, dn,"domain-idle-states" ,NULL ,0 ) { np = it.node; + match_id =of_match_node (idle_state_match, np); +if (!match_id) +continue ; ret =genpd_parse_state (&st[i++], np);if (ret) { pr_err
@@ -2141,8 +2140,11 @@ int of_genpd_parse_idle_states(struct device_node *dn, } } - *n = count; - *states = st; + *n = i; +if (!i) +kfree (st); +else + *states = st;return 0 ; }
Signature · 1 line
2.7.4
PATCH 1/2 PM / Domains: Ignore domain-idle-states that are not compatible
Quoted 2 lines · Lina Iyer
domain-idle-states property may have phandles to idle state bindings that may not be compatible with idle state definition defined in [1].
I don't find the reference to [1] in the change-log, could you please add it.
Quoted 2 lines
Such phandles would just be ignored and not throw and error when read by the domain core.
Perhaps, you could also share a minimal snipped from a DTS as it helps to describe why and what this change is needed.
Quoted 26 lines
--- Documentation/devicetree/bindings/power/power_domain.txt | 4 +++- drivers/base/power/domain.c | 16 +++++++++------- 2 files changed, 12 insertions(+), 8 deletions(-)Documentation/devicetree/bindings/power/power_domain.txt +3
-1@@ -31,7 +31,9 @@ Optional properties: - domain-idle-states : A phandle of an idle-state that shall be soaked into a generic domain power state. The idle state definitions are - compatible with domain-idle-state specified in [1]. + compatible with domain-idle-state specified in [1]. phandles + that are not compatible with domain-idle-state will be + ignored. The domain-idle-state property reflects the idle state of this PM domain and not the idle states of the devices or sub-domains in the PM domain. Devices and sub-domains have their own idle-states independent of the parentdrivers/base/power/domain.c +0
-0
drivers/base/power/domain.c
Quoted 1 line
drivers/base/power/domain.c +0
-1--- a/drivers/base/power/domain.c
Always split DT documentation changes from the code changes and make the DT doc changes precede the code changes in the series of patches.
Quoted 22 lines
+++
b/drivers/base/power/domain.cdrivers/base/power/domain.c +1
-5@@ -2065,11 +2065,6 @@ static int genpd_parse_state(struct genpd_power_state *genpd_state,int err;u32 residency;u32 entry_latency, exit_latency; -const struct of_device_id *match_id; - - match_id =of_match_node (idle_state_match, state_node); -if (!match_id) -return -EINVAL ; err =of_property_read_u32 (state_node,"entry-latency-us" , &entry_latency);@@ -2118,6 +2113,7 @@ int of_genpd_parse_idle_states(struct device_node *dn,int err, ret;int count;struct of_phandle_iterator it; +const struct of_device_id *match_id; count =of_count_phandle_with_args (dn,"domain-idle-states" ,NULL );if (count <=0 )
drivers/base/power/domain.c int of_genpd_parse_idle_states(struct device_node *dn, · line 2131
Quoted 6 lines
drivers/base/power/domain.c +3
-0@@ -2130,6 +2126,9 @@ int of_genpd_parse_idle_states(struct device_node *dn,/* Loop over the phandles until all the requested entry is found */ of_for_each_phandle (&it, err, dn,"domain-idle-states" ,NULL ,0 ) { np = it.node; + match_id =of_match_node (idle_state_match, np); +if (!match_id) +continue ;
Earlier we have allocated "count" numbers of struct genpd_power_state, by using a kcalloc().
This change may lead to that we could have allocated more memory than actually needed - because there may be some nodes that doesn't match.
Perhaps it's better to do a pre-iteration to find the real numbers of how many struct genpd_power_state we actually need to allocate!?
Quoted 18 lines
drivers/base/power/domain.c +5
-3ret =genpd_parse_state (&st[i++], np);if (ret) { pr_err@@ -2141,8 +2140,11 @@ int of_genpd_parse_idle_states(struct device_node *dn, } } - *n = count; - *states = st; + *n = i; +if (!i) +kfree (st); +else + *states = st;return 0 ; } --
Quoted 2 lines
2.7.4
Kind regards
Uffe
PATCH 2/2 PM / Domains: Support in context powering off parent domain
In reply to Lina IyerPowering off a domain schedules a work to opportunistically power off the parent domains. Domains that are IRQ safe may have parents that are also IRQ safe. It would be beneficial to power off such IRQ safe parents in the same context as well.
--- drivers/base/power/domain.c | 21 ++++++++++++++++----- 1 file changed, 16 insertions(+), 5 deletions(-)
drivers/base/power/domain.c +16 -5
@@ -375,7 +375,8 @@ static int genpd_dev_pm_qos_notifier(struct notifier_block *nb,* If all of the @genpd's devices have been suspended and all of its subdomains * have been powered down, remove power from @genpd. */ -static int genpd_power_off (struct generic_pm_domain *genpd,bool is_async) +static int genpd_power_off (struct generic_pm_domain *genpd,bool is_async, +unsigned int depth) {struct pm_domain_data *pdd;struct gpd_link *link;
@@ -442,7 +443,17 @@ static int genpd_power_off(struct generic_pm_domain *genpd, bool is_async)list_for_each_entry (link, &genpd->slave_links, slave_node) {genpd_sd_counter_dec (link->master); -genpd_queue_power_off_work (link->master); +/* +* Power off the parent in the same context if the parent +* domain is also IRQ safe. +*/ +if (genpd_is_irq_safe (genpd) && +genpd_is_irq_safe (link->master)) { +genpd_lock_nested (link->master, depth +1 ); +genpd_power_off (link->master, false, depth +1 ); +genpd_unlock (link->master); + }else +genpd_queue_power_off_work (link->master); }return 0 ;
@@ -459,7 +470,7 @@ static void genpd_power_off_work_fn(struct work_struct *work) genpd = container_of(work, struct generic_pm_domain, power_off_work);genpd_lock (genpd); -genpd_power_off (genpd, true); +genpd_power_off (genpd, true,0 );genpd_unlock (genpd); }
@@ -578,7 +589,7 @@ static int genpd_runtime_suspend(struct device *dev)return 0 ;genpd_lock (genpd); -genpd_power_off (genpd, false); +genpd_power_off (genpd, false,0 );genpd_unlock (genpd);return 0 ;
@@ -658,7 +669,7 @@ static int genpd_runtime_resume(struct device *dev)if (!pm_runtime_is_irq_safe (dev) || (pm_runtime_is_irq_safe (dev) &&genpd_is_irq_safe (genpd))) {genpd_lock (genpd); -genpd_power_off (genpd,0 ); +genpd_power_off (genpd,0 ,0 );genpd_unlock (genpd); }
Signature · 1 line
2.7.4
PATCH 2/2 PM / Domains: Support in context powering off parent domain
Quoted 27 lines · Lina Iyer
Powering off a domain schedules a work to opportunistically power off the parent domains. Domains that are IRQ safe may have parents that are also IRQ safe. It would be beneficial to power off such IRQ safe parents in the same context as well.
Signed-off-by Lina Iyer <[email protected]>--- drivers/base/power/domain.c | 21 ++++++++++++++++----- 1 file changed, 16 insertions(+), 5 deletions(-)drivers/base/power/domain.c +2
-1@@ -375,7 +375,8 @@ static int genpd_dev_pm_qos_notifier(struct notifier_block *nb,* If all of the @genpd's devices have been suspended and all of its subdomains * have been powered down, remove power from @genpd. */ -static int genpd_power_off (struct generic_pm_domain *genpd,bool is_async) +static int genpd_power_off (struct generic_pm_domain *genpd,bool is_async, +unsigned int depth) {struct pm_domain_data *pdd;struct gpd_link *link;@@ -442,7 +443,17 @@ static int genpd_power_off(struct generic_pm_domain *genpd, bool is_async)list_for_each_entry (link, &genpd->slave_links, slave_node) {
drivers/base/power/domain.c static int genpd_power_off(struct generic_pm_domain *genpd, bool is_async) · line 453
Quoted 10 lines
drivers/base/power/domain.c +8
-1genpd_sd_counter_dec (link->master); -genpd_queue_power_off_work (link->master); +/* +* Power off the parent in the same context if the parent +* domain is also IRQ safe. +*/ +if (genpd_is_irq_safe (genpd) && +genpd_is_irq_safe (link->master)) { +genpd_lock_nested (link->master, depth +1 ); +genpd_power_off (link->master, false, depth +1 );
This doesn't work. You must not call genpd_power_off() using "false" here.
That's because "true" in the recursive call, for the master domain tells genpd_power_off() that is has been called from genpd's ->runtime_suspend() callback. That means genpd_power_off() thinks it's okay to allow *one* device in the domain to not be runtime suspended when allowing a power off to be done. This assumption is not correct for the master domain.
Quoted 34 lines
drivers/base/power/domain.c +6
-4+genpd_unlock (link->master); + }else +genpd_queue_power_off_work (link->master); }return 0 ;@@ -459,7 +470,7 @@ static void genpd_power_off_work_fn(struct work_struct *work) genpd = container_of(work, struct generic_pm_domain, power_off_work);genpd_lock (genpd); -genpd_power_off (genpd, true); +genpd_power_off (genpd, true,0 );genpd_unlock (genpd); }@@ -578,7 +589,7 @@ static int genpd_runtime_suspend(struct device *dev)return 0 ;genpd_lock (genpd); -genpd_power_off (genpd, false); +genpd_power_off (genpd, false,0 );genpd_unlock (genpd);return 0 ;@@ -658,7 +669,7 @@ static int genpd_runtime_resume(struct device *dev)if (!pm_runtime_is_irq_safe (dev) || (pm_runtime_is_irq_safe (dev) &&genpd_is_irq_safe (genpd))) {genpd_lock (genpd); -genpd_power_off (genpd,0 ); +genpd_power_off (genpd,0 ,0 );genpd_unlock (genpd); } --
Quoted 2 lines
2.7.4
Some more thoughts..
Actually, I have been thinking of changing genpd to avoid queuing power off works, no matter if the PM domain are IRQ safe or not. There are several reasons, but primarily it helps to avoid wasting power.
Currently I don't see any reasons to why such change shouldn't be feasible. As a matter of fact, changing this became possible while we removed the intermediate states in genpd in commit ba2bbfbf6307 ("PM /
Domains: Remove intermediate states from the power off sequence").
Allow me to help out and cook a patch for this, it's already in the pipe. :-)
Kind regards
Uffe
PATCH 2/2 PM / Domains: Support in context powering off parent domain
Quoted 41 lines · Ulf Hansson
Quoted 27 lines · Lina Iyer
Powering off a domain schedules a work to opportunistically power off the parent domains. Domains that are IRQ safe may have parents that are also IRQ safe. It would be beneficial to power off such IRQ safe parents in the same context as well.
Signed-off-by Lina Iyer <[email protected]>--- drivers/base/power/domain.c | 21 ++++++++++++++++----- 1 file changed, 16 insertions(+), 5 deletions(-)drivers/base/power/domain.c +2
-1@@ -375,7 +375,8 @@ static int genpd_dev_pm_qos_notifier(struct notifier_block *nb,* If all of the @genpd's devices have been suspended and all of its subdomains * have been powered down, remove power from @genpd. */ -static int genpd_power_off (struct generic_pm_domain *genpd,bool is_async) +static int genpd_power_off (struct generic_pm_domain *genpd,bool is_async, +unsigned int depth) {struct pm_domain_data *pdd;struct gpd_link *link;@@ -442,7 +443,17 @@ static int genpd_power_off(struct generic_pm_domain *genpd, bool is_async)list_for_each_entry (link, &genpd->slave_links, slave_node) {drivers/base/power/domain.c static int genpd_power_off(struct generic_pm_domain *genpd, bool is_async) · line 453
Quoted 10 lines
drivers/base/power/domain.c +8
-1genpd_sd_counter_dec (link->master); -genpd_queue_power_off_work (link->master); +/* +* Power off the parent in the same context if the parent +* domain is also IRQ safe. +*/ +if (genpd_is_irq_safe (genpd) && +genpd_is_irq_safe (link->master)) { +genpd_lock_nested (link->master, depth +1 ); +genpd_power_off (link->master, false, depth +1 );This doesn't work. You must not call
genpd_power_off()using "false" here.
Quoted 1 line
That's because "true" in the recursive call, for the master domain
/s/true/false
Quoted 58 lines
tells
genpd_power_off()that is has been called from genpd's ->runtime_suspend() callback. That meansgenpd_power_off()thinks it's okay to allow *one* device in the domain to not be runtime suspended when allowing a power off to be done. This assumption is not correct for the master domain.Quoted 34 lines
drivers/base/power/domain.c +6
-4+genpd_unlock (link->master); + }else +genpd_queue_power_off_work (link->master); }return 0 ;@@ -459,7 +470,7 @@ static void genpd_power_off_work_fn(struct work_struct *work) genpd = container_of(work, struct generic_pm_domain, power_off_work);genpd_lock (genpd); -genpd_power_off (genpd, true); +genpd_power_off (genpd, true,0 );genpd_unlock (genpd); }@@ -578,7 +589,7 @@ static int genpd_runtime_suspend(struct device *dev)return 0 ;genpd_lock (genpd); -genpd_power_off (genpd, false); +genpd_power_off (genpd, false,0 );genpd_unlock (genpd);return 0 ;@@ -658,7 +669,7 @@ static int genpd_runtime_resume(struct device *dev)if (!pm_runtime_is_irq_safe (dev) || (pm_runtime_is_irq_safe (dev) &&genpd_is_irq_safe (genpd))) {genpd_lock (genpd); -genpd_power_off (genpd,0 ); +genpd_power_off (genpd,0 ,0 );genpd_unlock (genpd); } --Quoted 2 lines
2.7.4
Some more thoughts..
Actually, I have been thinking of changing genpd to avoid queuing power off works, no matter if the PM domain are IRQ safe or not. There are several reasons, but primarily it helps to avoid wasting power.
Currently I don't see any reasons to why such change shouldn't be feasible. As a matter of fact, changing this became possible while we removed the intermediate states in genpd in commit
ba2bbfbf6307("PM /
Domains: Remove intermediate states from the power off sequence").Allow me to help out and cook a patch for this, it's already in the pipe. :-)
Kind regards
Uffe