PM QoS: Use spinlock in the per-device PM QoS constraints code

The per-device PM QoS locking requires a spinlock to be used. The reasons
are:
 - an alignement with the PM QoS core code, which is used by the per-device
   PM QoS code for the constraints lists management. The PM QoS core code
   uses spinlocks to protect the constraints lists,
 - some drivers need to use the per-device PM QoS functionality from
   interrupt context or spinlock protected context.
   An example of such a driver is the OMAP HSI (high-speed synchronous serial
   interface) driver which needs to control the IP block idle state
   depending on the FIFO empty state, from interrupt context.

Reported-by: Djamil Elaidi <d-elaidi@ti.com>
Signed-off-by: Jean Pihet <j-pihet@ti.com>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>

1 files changed, 41 insertions, 26 deletions

diff --git a/drivers/base/power/qos.c b/drivers/base/power/qos.c
index 74a67e0019a2..968a77145e81 100644
--- a/drivers/base/power/qos.c
+++ b/drivers/base/power/qos.c
@@ -24,26 +24,32 @@
  *  . a system-wide notification callback using the dev_pm_qos_*_global_notifier
  *    API. The notification chain data is stored in a static variable.
  *
- * Note about the per-device constraint data struct allocation:
- * . The per-device constraints data struct ptr is tored into the device
+ * Notes about the per-device constraint data struct allocation:
+ * . The per-device constraints data struct ptr is stored into the device
  *    dev_pm_info.
  * . To minimize the data usage by the per-device constraints, the data struct
- *   is only allocated at the first call to dev_pm_qos_add_request.
+ *    is only allocated at the first call to dev_pm_qos_add_request.
  * . The data is later free'd when the device is removed from the system.
- *  . A global mutex protects the constraints users from the data being
- *     allocated and free'd.
+ *
+ * Notes about locking:
+ * . The dev->power.lock lock protects the constraints list
+ *    (dev->power.constraints) allocation and free, as triggered by the
+ *   driver core code at device insertion and removal,
+ * . A global lock dev_pm_qos_lock protects the constraints list entries
+ *    from any modification and the notifiers registration and unregistration.
+ * . For both locks a spinlock is needed since this code can be called from
+ *    interrupt context or spinlock protected context.
  */
 
 #include <linux/pm_qos.h>
 #include <linux/spinlock.h>
 #include <linux/slab.h>
 #include <linux/device.h>
-#include <linux/mutex.h>
 #include <linux/export.h>
 
 #include "power.h"
 
-static DEFINE_MUTEX(dev_pm_qos_mtx);
+static DEFINE_SPINLOCK(dev_pm_qos_lock);
 
 static BLOCKING_NOTIFIER_HEAD(dev_pm_notifiers);
 
@@ -110,18 +116,19 @@ static int apply_constraint(struct dev_pm_qos_request *req,
  * @dev: device to allocate data for
  *
  * Called at the first call to add_request, for constraint data allocation
- * Must be called with the dev_pm_qos_mtx mutex held
+ * Must be called with the dev_pm_qos_lock lock held
  */
 static int dev_pm_qos_constraints_allocate(struct device *dev)
 {
         struct pm_qos_constraints *c;
         struct blocking_notifier_head *n;
+        unsigned long flags;
 
-        c = kzalloc(sizeof(*c), GFP_KERNEL);
+        c = kzalloc(sizeof(*c), GFP_ATOMIC);
         if (!c)
                 return -ENOMEM;
 
-        n = kzalloc(sizeof(*n), GFP_KERNEL);
+        n = kzalloc(sizeof(*n), GFP_ATOMIC);
         if (!n) {
                 kfree(c);
                 return -ENOMEM;
@@ -134,9 +141,9 @@ static int dev_pm_qos_constraints_allocate(struct device *dev)
         c->type = PM_QOS_MIN;
         c->notifiers = n;
 
-        spin_lock_irq(&dev->power.lock);
+        spin_lock_irqsave(&dev->power.lock, flags);
         dev->power.constraints = c;
-        spin_unlock_irq(&dev->power.lock);
+        spin_unlock_irqrestore(&dev->power.lock, flags);
 
         return 0;
 }
@@ -150,10 +157,12 @@ static int dev_pm_qos_constraints_allocate(struct device *dev)
  */
 void dev_pm_qos_constraints_init(struct device *dev)
 {
-        mutex_lock(&dev_pm_qos_mtx);
+        unsigned long flags;
+
+        spin_lock_irqsave(&dev_pm_qos_lock, flags);
         dev->power.constraints = NULL;
         dev->power.power_state = PMSG_ON;
-        mutex_unlock(&dev_pm_qos_mtx);
+        spin_unlock_irqrestore(&dev_pm_qos_lock, flags);
 }
 
 /**
@@ -166,6 +175,7 @@ void dev_pm_qos_constraints_destroy(struct device *dev)
 {
         struct dev_pm_qos_request *req, *tmp;
         struct pm_qos_constraints *c;
+        unsigned long flags;
 
         /*
          * If the device's PM QoS resume latency limit has been exposed to user
@@ -173,7 +183,7 @@ void dev_pm_qos_constraints_destroy(struct device *dev)
          */
         dev_pm_qos_hide_latency_limit(dev);
 
-        mutex_lock(&dev_pm_qos_mtx);
+        spin_lock_irqsave(&dev_pm_qos_lock, flags);
 
         dev->power.power_state = PMSG_INVALID;
         c = dev->power.constraints;
@@ -198,7 +208,7 @@ void dev_pm_qos_constraints_destroy(struct device *dev)
         kfree(c);
 
  out:
-        mutex_unlock(&dev_pm_qos_mtx);
+        spin_unlock_irqrestore(&dev_pm_qos_lock, flags);
 }
 
 /**
@@ -223,6 +233,7 @@ int dev_pm_qos_add_request(struct device *dev, struct dev_pm_qos_request *req,
                            s32 value)
 {
         int ret = 0;
+        unsigned long flags;
 
         if (!dev || !req) /*guard against callers passing in null */
                 return -EINVAL;
@@ -233,7 +244,7 @@ int dev_pm_qos_add_request(struct device *dev, struct dev_pm_qos_request *req,
 
         req->dev = dev;
 
-        mutex_lock(&dev_pm_qos_mtx);
+        spin_lock_irqsave(&dev_pm_qos_lock, flags);
 
         if (!dev->power.constraints) {
                 if (dev->power.power_state.event == PM_EVENT_INVALID) {
@@ -255,7 +266,7 @@ int dev_pm_qos_add_request(struct device *dev, struct dev_pm_qos_request *req,
                 ret = apply_constraint(req, PM_QOS_ADD_REQ, value);
 
  out:
-        mutex_unlock(&dev_pm_qos_mtx);
+        spin_unlock_irqrestore(&dev_pm_qos_lock, flags);
 
         return ret;
 }
@@ -280,6 +291,7 @@ int dev_pm_qos_update_request(struct dev_pm_qos_request *req,
                               s32 new_value)
 {
         int ret = 0;
+        unsigned long flags;
 
         if (!req) /*guard against callers passing in null */
                 return -EINVAL;
@@ -288,7 +300,7 @@ int dev_pm_qos_update_request(struct dev_pm_qos_request *req,
                  "%s() called for unknown object\n", __func__))
                 return -EINVAL;
 
-        mutex_lock(&dev_pm_qos_mtx);
+        spin_lock_irqsave(&dev_pm_qos_lock, flags);
 
         if (req->dev->power.constraints) {
                 if (new_value != req->node.prio)
@@ -299,7 +311,7 @@ int dev_pm_qos_update_request(struct dev_pm_qos_request *req,
                 ret = -ENODEV;
         }
 
-        mutex_unlock(&dev_pm_qos_mtx);
+        spin_unlock_irqrestore(&dev_pm_qos_lock, flags);
         return ret;
 }
 EXPORT_SYMBOL_GPL(dev_pm_qos_update_request);
@@ -319,6 +331,7 @@ EXPORT_SYMBOL_GPL(dev_pm_qos_update_request);
 int dev_pm_qos_remove_request(struct dev_pm_qos_request *req)
 {
         int ret = 0;
+        unsigned long flags;
 
         if (!req) /*guard against callers passing in null */
                 return -EINVAL;
@@ -327,7 +340,7 @@ int dev_pm_qos_remove_request(struct dev_pm_qos_request *req)
                  "%s() called for unknown object\n", __func__))
                 return -EINVAL;
 
-        mutex_lock(&dev_pm_qos_mtx);
+        spin_lock_irqsave(&dev_pm_qos_lock, flags);
 
         if (req->dev->power.constraints) {
                 ret = apply_constraint(req, PM_QOS_REMOVE_REQ,
@@ -338,7 +351,7 @@ int dev_pm_qos_remove_request(struct dev_pm_qos_request *req)
                 ret = -ENODEV;
         }
 
-        mutex_unlock(&dev_pm_qos_mtx);
+        spin_unlock_irqrestore(&dev_pm_qos_lock, flags);
         return ret;
 }
 EXPORT_SYMBOL_GPL(dev_pm_qos_remove_request);
@@ -359,8 +372,9 @@ EXPORT_SYMBOL_GPL(dev_pm_qos_remove_request);
 int dev_pm_qos_add_notifier(struct device *dev, struct notifier_block *notifier)
 {
         int ret = 0;
+        unsigned long flags;
 
-        mutex_lock(&dev_pm_qos_mtx);
+        spin_lock_irqsave(&dev_pm_qos_lock, flags);
 
         if (!dev->power.constraints)
                 ret = dev->power.power_state.event != PM_EVENT_INVALID ?
@@ -370,7 +384,7 @@ int dev_pm_qos_add_notifier(struct device *dev, struct notifier_block *notifier)
                 ret = blocking_notifier_chain_register(
                                 dev->power.constraints->notifiers, notifier);
 
-        mutex_unlock(&dev_pm_qos_mtx);
+        spin_unlock_irqrestore(&dev_pm_qos_lock, flags);
         return ret;
 }
 EXPORT_SYMBOL_GPL(dev_pm_qos_add_notifier);
@@ -389,8 +403,9 @@ int dev_pm_qos_remove_notifier(struct device *dev,
                                struct notifier_block *notifier)
 {
         int retval = 0;
+        unsigned long flags;
 
-        mutex_lock(&dev_pm_qos_mtx);
+        spin_lock_irqsave(&dev_pm_qos_lock, flags);
 
         /* Silently return if the constraints object is not present. */
         if (dev->power.constraints)
@@ -398,7 +413,7 @@ int dev_pm_qos_remove_notifier(struct device *dev,
                                 dev->power.constraints->notifiers,
                                 notifier);
 
-        mutex_unlock(&dev_pm_qos_mtx);
+        spin_unlock_irqrestore(&dev_pm_qos_lock, flags);
         return retval;
 }
 EXPORT_SYMBOL_GPL(dev_pm_qos_remove_notifier);