thermal: core/ACPI: Fix processor cooling device regression

From: Rafael J. Wysocki <[email protected]>

The cpufreq policy notifier in the ACPI processor driver may as well be registered before the driver itself, which causes acpi_processor_cpufreq_init to be true (unless the notifier registration fails, which is unlikely at that point) when the ACPI CPU thermal cooling devices are registered, so the processor_get_max_state() result does not change while acpi_processor_driver_init() is running.

Change the ordering in acpi_processor_driver_init() accordingly to prevent the max_state value from remaining 0 permanently for all ACPI CPU cooling devices due to setting acpi_processor_cpufreq_init too late. [Note that processor_get_max_state() may still return different values at different times after this change, depending on the cpufreq driver registration time, but that issue needs to be addressed separately.]

Fixes a365105c685c("thermal: sysfs: Reuse cdev->max_state")

Reported-by Wang, Quanxian <[email protected]>

Link:
https://lore.kernel.org/linux-pm/[email protected]/

Signed-off-by Rafael J. Wysocki <[email protected]>

---

v1 -> v2: Expand changelog to explain that this particular patch addresses
part of the issue.

---
 drivers/acpi/processor_driver.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

Index: linux-pm/drivers/acpi/processor_driver.c
===================================================================

drivers/acpi/processor_driver.c +6 -6

@@ -263,6 +263,12 @@ static int __init acpi_processor_driver_
 	if (acpi_disabled)
 		return 0;
 
+	if (!cpufreq_register_notifier(&acpi_processor_notifier_block,
+				       CPUFREQ_POLICY_NOTIFIER)) {
+		acpi_processor_cpufreq_init = true;
+		acpi_processor_ignore_ppc_init();
+	}
+
 	result = driver_register(&acpi_processor_driver);
 	if (result < 0)
 		return result;

@@ -276,12 +282,6 @@ static int __init acpi_processor_driver_
 	cpuhp_setup_state_nocalls(CPUHP_ACPI_CPUDRV_DEAD, "acpi/cpu-
 drv:dead",
 				  NULL, acpi_soft_cpu_dead);
 
-	if (!cpufreq_register_notifier(&acpi_processor_notifier_block,
-				       CPUFREQ_POLICY_NOTIFIER)) {
-		acpi_processor_cpufreq_init = true;
-		acpi_processor_ignore_ppc_init();
-	}
-
 	acpi_processor_throttling_init();
 	return 0;

err:

From: Rafael J. Wysocki <[email protected]>

Introduce a helper function, thermal_cooling_device_present(), for checking if the given cooling device is in the list of registered cooling devices to avoid some code duplication in a subsequent patch.

No expected functional impact.

Signed-off-by Rafael J. Wysocki <[email protected]>

---

v1 -> v2: No changes

---
 drivers/thermal/thermal_core.c |   21 +++++++++++++++------
 1 file changed, 15 insertions(+), 6 deletions(-)

Index: linux-pm/drivers/thermal/thermal_core.c
===================================================================

drivers/thermal/thermal_core.c +12 -0

@@ -1045,6 +1045,18 @@ devm_thermal_of_cooling_device_register(
 }
 EXPORT_SYMBOL_GPL(devm_thermal_of_cooling_device_register);
 
+static bool thermal_cooling_device_present(struct
 thermal_cooling_device *cdev)
+{
+	struct thermal_cooling_device *pos = NULL;
+
+	list_for_each_entry(pos, &thermal_cdev_list, node) {
+		if (pos == cdev)
+			return true;
+	}
+
+	return false;
+}
+
 static void __unbind(struct thermal_zone_device *tz, int mask,
 		     struct thermal_cooling_device *cdev)

{

drivers/thermal/thermal_core.c +3 -6

@@ -1067,20 +1079,17 @@ void thermal_cooling_device_unregister(s
 	int i;
 	const struct thermal_zone_params *tzp;
 	struct thermal_zone_device *tz;
-	struct thermal_cooling_device *pos = NULL;
 
 	if (!cdev)
 		return;
 
 	mutex_lock(&thermal_list_lock);
-	list_for_each_entry(pos, &thermal_cdev_list, node)
-		if (pos == cdev)
-			break;
-	if (pos != cdev) {
-		/* thermal cooling device not found */
+
+	if (!thermal_cooling_device_present(cdev)) {
 		mutex_unlock(&thermal_list_lock);
 		return;
 	}
+
 	list_del(&cdev->node);
 
 	/* Unbind all thermal zones associated with 'this' cdev */

From: Rafael J. Wysocki <[email protected]>

Introduce a core thermal API function, thermal_cooling_device_update(),
for updating the max_state value for a cooling device and rearranging its statistics in sysfs after a possible change of its ->get_max_state()
callback return value.

That callback is now invoked only once, during cooling device registration, to populate the max_state field in the cooling device object, so if its return value changes, it needs to be invoked again and the new return value needs to be stored as max_state. Moreover, the statistics presented in sysfs need to be rearranged in general, because there may not be enough room in them to store data for all of the possible states (in the case when max_state grows).

The new function takes care of that (and some other minor things related to it), but some extra locking and lockdep annotations are added in several places too to protect against crashes in the cases when the statistics are not present or when a stale max_state value might be used by sysfs attributes.

Note that the actual user of the new function will be added separately.

Link:
https://lore.kernel.org/linux-pm/[email protected]/

Signed-off-by Rafael J. Wysocki <[email protected]>

---

v1 -> v2:
* Make thermal_cooling_device_update() do the full
IS_ERR_OR_NULL() check
on cdev, so that its callers don't need to do it.
* Remove ->set_cur_state() callback invocation from thermal_cooling_device_update(), because it is redundant (and
it doesn't
make sense to pass a state value exceeding max_state to it
anyway).

---
 drivers/thermal/thermal_core.c  |   40 ++++++++++++++++++++++
 drivers/thermal/thermal_core.h  |    1 
 drivers/thermal/thermal_sysfs.c |   72
+++++++++++++++++++++++++++++++++++-----
 include/linux/thermal.h         |    1 
 4 files changed, 106 insertions(+), 8 deletions(-)

Index: linux-pm/drivers/thermal/thermal_core.c
===================================================================

drivers/thermal/thermal_core.c +9 -0

@@ -1057,6 +1057,46 @@ static bool thermal_cooling_device_prese
 	return false;
 }
 
+void thermal_cooling_device_update(struct thermal_cooling_device
 *cdev)
+{
+	unsigned long state;
+
+	if (IS_ERR_OR_NULL(cdev))
+		return;
+
+	/*
+	 * Hold thermal_list_lock throughout the update to prevent the

device

drivers/thermal/thermal_core.c +9 -0

+	 * from going away while being updated.
+	 */
+	mutex_lock(&thermal_list_lock);
+
+	if (!thermal_cooling_device_present(cdev))
+		goto unlock_list;
+
+	/*
+	 * Update under the cdev lock to prevent the state from being

set beyond

drivers/thermal/thermal_core.c +10 -0

+	 * the new limit concurrently.
+	 */
+	mutex_lock(&cdev->lock);
+
+	if (cdev->ops->get_max_state(cdev, &cdev->max_state))
+		goto unlock;
+
+	thermal_cooling_device_stats_reinit(cdev);
+
+	if (cdev->ops->get_cur_state(cdev, &state) || state > cdev-

Quoted 1 line

max_state)

drivers/thermal/thermal_core.c +12 -0

+		goto unlock;
+
+	thermal_cooling_device_stats_update(cdev, state);
+
+unlock:
+	mutex_unlock(&cdev->lock);
+
+unlock_list:
+	mutex_unlock(&thermal_list_lock);
+}
+EXPORT_SYMBOL_GPL(thermal_cooling_device_update);
+
 static void __unbind(struct thermal_zone_device *tz, int mask,
 		     struct thermal_cooling_device *cdev)
 {

Index: linux-pm/drivers/thermal/thermal_sysfs.c
===================================================================

drivers/thermal/thermal_sysfs.c +45 -5

@@ -685,6 +685,8 @@ void thermal_cooling_device_stats_update
 {
 	struct cooling_dev_stats *stats = cdev->stats;
 
+	lockdep_assert_held(&cdev->lock);
+
 	if (!stats)
 		return;
 

@@ -706,13 +708,22 @@ static ssize_t total_trans_show(struct d
 				struct device_attribute *attr, char
 *buf)
 {
 	struct thermal_cooling_device *cdev = to_cooling_device(dev);
-	struct cooling_dev_stats *stats = cdev->stats;
+	struct cooling_dev_stats *stats;
 	int ret;
 
+	mutex_lock(&cdev->lock);
+
+	stats = cdev->stats;
+	if (!stats)
+		goto unlock;
+
 	spin_lock(&stats->lock);
 	ret = sprintf(buf, "%u\n", stats->total_trans);
 	spin_unlock(&stats->lock);
 
+unlock:
+	mutex_unlock(&cdev->lock);
+
 	return ret;
 }

@@ -721,11 +732,18 @@ time_in_state_ms_show(struct device *dev
 		      char *buf)
 {
 	struct thermal_cooling_device *cdev = to_cooling_device(dev);
-	struct cooling_dev_stats *stats = cdev->stats;
+	struct cooling_dev_stats *stats;
 	ssize_t len = 0;
 	int i;
 
+	mutex_lock(&cdev->lock);
+
+	stats = cdev->stats;
+	if (!stats)
+		goto unlock;
+
 	spin_lock(&stats->lock);
+
 	update_time_in_state(stats);
 
 	for (i = 0; i <= cdev->max_state; i++) {

@@ -734,6 +752,9 @@ time_in_state_ms_show(struct device *dev
 	}
 	spin_unlock(&stats->lock);
 
+unlock:
+	mutex_unlock(&cdev->lock);
+
 	return len;
 }
 

@@ -742,8 +763,16 @@ reset_store(struct device *dev, struct d
 	    size_t count)
 {
 	struct thermal_cooling_device *cdev = to_cooling_device(dev);
-	struct cooling_dev_stats *stats = cdev->stats;
-	int i, states = cdev->max_state + 1;
+	struct cooling_dev_stats *stats;
+	int i, states;
+
+	mutex_lock(&cdev->lock);
+
+	stats = cdev->stats;
+	if (!stats)
+		goto unlock;
+
+	states = cdev->max_state + 1;
 
 	spin_lock(&stats->lock);
 

@@ -757,6 +786,9 @@ reset_store(struct device *dev, struct d
 
 	spin_unlock(&stats->lock);
 
+unlock:
+	mutex_unlock(&cdev->lock);
+
 	return count;
 }
 

@@ -764,10 +796,18 @@ static ssize_t trans_table_show(struct d
 				struct device_attribute *attr, char
 *buf)
 {
 	struct thermal_cooling_device *cdev = to_cooling_device(dev);
-	struct cooling_dev_stats *stats = cdev->stats;
+	struct cooling_dev_stats *stats;
 	ssize_t len = 0;
 	int i, j;
 
+	mutex_lock(&cdev->lock);
+
+	stats = cdev->stats;
+	if (!stats) {
+		len = -ENODATA;
+		goto unlock;
+	}
+
 	len += snprintf(buf + len, PAGE_SIZE - len, "
 From  :    To\n");

drivers/thermal/thermal_sysfs.c +13 -3

 	len += snprintf(buf + len, PAGE_SIZE - len, "       : ");
 	for (i = 0; i <= cdev->max_state; i++) {

@@ -775,8 +815,10 @@ static ssize_t trans_table_show(struct d
 			break;
 		len += snprintf(buf + len, PAGE_SIZE - len,
 "state%2u  ", i);
 	}
-	if (len >= PAGE_SIZE)
-		return PAGE_SIZE;
+	if (len >= PAGE_SIZE) {
+		len = PAGE_SIZE;
+		goto unlock;
+	}
 
 	len += snprintf(buf + len, PAGE_SIZE - len, "\n");

@@ -799,8 +841,12 @@ static ssize_t trans_table_show(struct d
 
 	if (len >= PAGE_SIZE) {
 		pr_warn_once("Thermal transition table exceeds
 PAGE_SIZE. Disabling\n");
-		return -EFBIG;
+		len = -EFBIG;
 	}
+
+unlock:
+	mutex_unlock(&cdev->lock);
+
 	return len;
 }

@@ -830,6 +876,8 @@ static void cooling_device_stats_setup(s
 	unsigned long states = cdev->max_state + 1;
 	int var;
 
+	lockdep_assert_held(&cdev->lock);
+
 	var = sizeof(*stats);
 	var += sizeof(*stats->time_in_state) * states;
 	var += sizeof(*stats->trans_table) * states * states;

@@ -855,6 +903,8 @@ out:
 
 static void cooling_device_stats_destroy(struct
 thermal_cooling_device *cdev)
 {
+	lockdep_assert_held(&cdev->lock);
+
 	kfree(cdev->stats);
 	cdev->stats = NULL;

}

drivers/thermal/thermal_sysfs.c +6 -0

@@ -879,6 +929,12 @@ void thermal_cooling_device_destroy_sysf
 	cooling_device_stats_destroy(cdev);
 }
 
+void thermal_cooling_device_stats_reinit(struct
 thermal_cooling_device *cdev)
+{
+	cooling_device_stats_destroy(cdev);
+	cooling_device_stats_setup(cdev);
+}
+
 /* these helper will be used only at the time of bindig */
 ssize_t

trip_point_show(struct device *dev, struct device_attribute *attr,
char *buf)
Index: linux-pm/drivers/thermal/thermal_core.h
===================================================================

drivers/thermal/thermal_core.h +1 -0

@@ -127,6 +127,7 @@ int thermal_zone_create_device_groups(st
 void thermal_zone_destroy_device_groups(struct thermal_zone_device
 *);
 void thermal_cooling_device_setup_sysfs(struct
 thermal_cooling_device *);
 void thermal_cooling_device_destroy_sysfs(struct
 thermal_cooling_device *cdev);
+void thermal_cooling_device_stats_reinit(struct

thermal_cooling_device *cdev);
/* used only at binding time */
ssize_t trip_point_show(struct device *, struct device_attribute *,
char *);
ssize_t weight_show(struct device *, struct device_attribute *, char
*);
Index: linux-pm/include/linux/thermal.h
===================================================================

include/linux/thermal.h +1 -0

@@ -388,6 +388,7 @@ devm_thermal_of_cooling_device_register(
 				struct device_node *np,
 				char *type, void *devdata,
 				const struct thermal_cooling_device_ops
 *ops);
+void thermal_cooling_device_update(struct thermal_cooling_device *);
 void thermal_cooling_device_unregister(struct thermal_cooling_device
 *);

struct thermal_zone_device *thermal_zone_get_zone_by_name(const char
*name);
int thermal_zone_get_temp(struct thermal_zone_device *tz, int
*temp);

From: Rafael J. Wysocki <[email protected]>

When a cpufreq policy appears or goes away, the CPU cooling devices for
the CPUs covered by that policy need to be updated so that the new processor_get_max_state() value is stored as max_state and the statistics in sysfs are rearranged for each of them.

Do that accordingly in acpi_thermal_cpufreq_init() and acpi_thermal_cpufreq_exit().

Fixes a365105c685c("thermal: sysfs: Reuse cdev->max_state")

Reported-by Wang, Quanxian <[email protected]>

Link:
https://lore.kernel.org/linux-pm/[email protected]/

Signed-off-by Rafael J. Wysocki <[email protected]>

---

v1 -> v2: Remove the now redundant IS_ERR() checks on cdev before calling
thermal_cooling_device_update().

---
 drivers/acpi/processor_thermal.c |   14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)

Index: linux-pm/drivers/acpi/processor_thermal.c
===================================================================

drivers/acpi/processor_thermal.c +11 -3

@@ -140,9 +140,13 @@ void acpi_thermal_cpufreq_init(struct cp
 		ret = freq_qos_add_request(&policy->constraints,
 					   &pr->thermal_req,
 					   FREQ_QOS_MAX, INT_MAX);
-		if (ret < 0)
+		if (ret < 0) {
 			pr_err("Failed to add freq constraint for CPU%d
 (%d)\n",
 			       cpu, ret);
+			continue;
+		}
+
+		thermal_cooling_device_update(pr->cdev);
 	}
 }

@@ -153,8 +157,12 @@ void acpi_thermal_cpufreq_exit(struct cp
 	for_each_cpu(cpu, policy->related_cpus) {
 		struct acpi_processor *pr = per_cpu(processors, cpu);
 
-		if (pr)
-			freq_qos_remove_request(&pr->thermal_req);
+		if (!pr)
+			continue;
+
+		freq_qos_remove_request(&pr->thermal_req);
+
+		thermal_cooling_device_update(pr->cdev);
 	}
 }
 #else				/* ! CONFIG_CPU_FREQ */

Hi All,

The first revision of this patch series was posted as

https://lore.kernel.org/linux-pm/2148907.irdbgypaU6@kreacher/

As reported by Rui in this thread:

Link:
https://lore.kernel.org/linux-pm/[email protected]/

some recent changes in the thermal core cause the CPU cooling devices registered by the ACPI processor driver to become unusable in some cases
and somewhat crippled in general.

The problem is that the ACPI processor driver changes its ->get_max_state()
callback return value depending on whether or not cpufreq is available and
there is a cpufreq policy for a given CPU. However, the thermal core has
always assumed that the return value of that callback will not change, which
in fact is relied on by the cooling device statistics code. In particular,
when the ->get_max_state() grows, the memory buffer allocated for storing the
statistics will be too small and corruption may ensue as a result.

For this reason, the issue needs to be addressed in the ACPI processor driver
and not in the thermal core, but the core needs to help somewhat too. Namely,
it needs to provide a helper allowing an interested driver to update the
max_state value for an already registered cooling device in certain situations
which will also cause the statistics to be rebuilt.

This series implements the above and for details please refer to the individual
patch chagelogs.

Thanks!

Hi, Rafael,

The only concern to me is that, in thermal_cooling_device_update(), we should handle the cases that the cooling device is current used by one/more thermal zone. say, something like

list_for_each_entry(pos, &cdev->thermal_instances, cdev_node) {
        /* e.g. what to do if tz1 set it to state 1 previously */
}

I have not got a clear idea what we should do here.

But given that I have confirmed that this patch series fixes the original problem, and the ACPI passive cooling is unlikely to be triggered before CPUFREQ_CREATE_POLICY notification, probably we can address that problem later.

Tested-by Zhang Rui <[email protected]>

Reviewed-by Zhang Rui <[email protected]>

Quoted 7 lines · Rui

Hi, Rafael,

The only concern to me is that, in thermal_cooling_device_update(), we
should handle the cases that the cooling device is current used by one/more thermal zone. say, something like

Quoted 4 lines

list_for_each_entry(pos, &cdev->thermal_instances, cdev_node) {
        /* e.g. what to do if tz1 set it to state 1 previously */
}

I have not got a clear idea what we should do here.

For each instance, set upper to max_state if above it and set target to upper if above it I'd say.

I guess otherwise there may be some confusion in principle and I have missed that piece, so thanks for pointing it out!

Quoted 8 lines

But given that I have confirmed that this patch series fixes the original problem, and the ACPI passive cooling is unlikely to be triggered before CPUFREQ_CREATE_POLICY notification, probably we can
address that problem later.

Tested-by Zhang Rui <[email protected]>

Reviewed-by Zhang Rui <[email protected]>