PM: sleep: Fix possible deadlocks in core system-wide PM code
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Wed, 27 Dec 2023 20:41:06 +0000 (21:41 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 1 Feb 2024 00:18:49 +0000 (16:18 -0800)
commit 7839d0078e0d5e6cc2fa0b0dfbee71de74f1e557 upstream.

It is reported that in low-memory situations the system-wide resume core
code deadlocks, because async_schedule_dev() executes its argument
function synchronously if it cannot allocate memory (and not only in
that case) and that function attempts to acquire a mutex that is already
held.  Executing the argument function synchronously from within
dpm_async_fn() may also be problematic for ordering reasons (it may
cause a consumer device's resume callback to be invoked before a
requisite supplier device's one, for example).

Address this by changing the code in question to use
async_schedule_dev_nocall() for scheduling the asynchronous
execution of device suspend and resume functions and to directly
run them synchronously if async_schedule_dev_nocall() returns false.

Link: https://lore.kernel.org/linux-pm/ZYvjiqX6EsL15moe@perf/
Reported-by: Youngmin Nam <youngmin.nam@samsung.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Tested-by: Youngmin Nam <youngmin.nam@samsung.com>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Cc: 5.7+ <stable@vger.kernel.org> # 5.7+: 6aa09a5bccd8 async: Split async_schedule_node_domain()
Cc: 5.7+ <stable@vger.kernel.org> # 5.7+: 7d4b5d7a37bd async: Introduce async_schedule_dev_nocall()
Cc: 5.7+ <stable@vger.kernel.org> # 5.7+
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/base/power/main.c

index f85f351..9c5a5f4 100644 (file)
@@ -579,7 +579,7 @@ bool dev_pm_skip_resume(struct device *dev)
 }
 
 /**
- * device_resume_noirq - Execute a "noirq resume" callback for given device.
+ * __device_resume_noirq - Execute a "noirq resume" callback for given device.
  * @dev: Device to handle.
  * @state: PM transition of the system being carried out.
  * @async: If true, the device is being resumed asynchronously.
@@ -587,7 +587,7 @@ bool dev_pm_skip_resume(struct device *dev)
  * The driver of @dev will not receive interrupts while this function is being
  * executed.
  */
-static int device_resume_noirq(struct device *dev, pm_message_t state, bool async)
+static void __device_resume_noirq(struct device *dev, pm_message_t state, bool async)
 {
        pm_callback_t callback = NULL;
        const char *info = NULL;
@@ -655,7 +655,13 @@ Skip:
 Out:
        complete_all(&dev->power.completion);
        TRACE_RESUME(error);
-       return error;
+
+       if (error) {
+               suspend_stats.failed_resume_noirq++;
+               dpm_save_failed_step(SUSPEND_RESUME_NOIRQ);
+               dpm_save_failed_dev(dev_name(dev));
+               pm_dev_err(dev, state, async ? " async noirq" : " noirq", error);
+       }
 }
 
 static bool is_async(struct device *dev)
@@ -668,11 +674,15 @@ static bool dpm_async_fn(struct device *dev, async_func_t func)
 {
        reinit_completion(&dev->power.completion);
 
-       if (is_async(dev)) {
-               get_device(dev);
-               async_schedule_dev(func, dev);
+       if (!is_async(dev))
+               return false;
+
+       get_device(dev);
+
+       if (async_schedule_dev_nocall(func, dev))
                return true;
-       }
+
+       put_device(dev);
 
        return false;
 }
@@ -680,15 +690,19 @@ static bool dpm_async_fn(struct device *dev, async_func_t func)
 static void async_resume_noirq(void *data, async_cookie_t cookie)
 {
        struct device *dev = data;
-       int error;
-
-       error = device_resume_noirq(dev, pm_transition, true);
-       if (error)
-               pm_dev_err(dev, pm_transition, " async", error);
 
+       __device_resume_noirq(dev, pm_transition, true);
        put_device(dev);
 }
 
+static void device_resume_noirq(struct device *dev)
+{
+       if (dpm_async_fn(dev, async_resume_noirq))
+               return;
+
+       __device_resume_noirq(dev, pm_transition, false);
+}
+
 static void dpm_noirq_resume_devices(pm_message_t state)
 {
        struct device *dev;
@@ -698,14 +712,6 @@ static void dpm_noirq_resume_devices(pm_message_t state)
        mutex_lock(&dpm_list_mtx);
        pm_transition = state;
 
-       /*
-        * Advanced the async threads upfront,
-        * in case the starting of async threads is
-        * delayed by non-async resuming devices.
-        */
-       list_for_each_entry(dev, &dpm_noirq_list, power.entry)
-               dpm_async_fn(dev, async_resume_noirq);
-
        while (!list_empty(&dpm_noirq_list)) {
                dev = to_device(dpm_noirq_list.next);
                get_device(dev);
@@ -713,17 +719,7 @@ static void dpm_noirq_resume_devices(pm_message_t state)
 
                mutex_unlock(&dpm_list_mtx);
 
-               if (!is_async(dev)) {
-                       int error;
-
-                       error = device_resume_noirq(dev, state, false);
-                       if (error) {
-                               suspend_stats.failed_resume_noirq++;
-                               dpm_save_failed_step(SUSPEND_RESUME_NOIRQ);
-                               dpm_save_failed_dev(dev_name(dev));
-                               pm_dev_err(dev, state, " noirq", error);
-                       }
-               }
+               device_resume_noirq(dev);
 
                put_device(dev);
 
@@ -751,14 +747,14 @@ void dpm_resume_noirq(pm_message_t state)
 }
 
 /**
- * device_resume_early - Execute an "early resume" callback for given device.
+ * __device_resume_early - Execute an "early resume" callback for given device.
  * @dev: Device to handle.
  * @state: PM transition of the system being carried out.
  * @async: If true, the device is being resumed asynchronously.
  *
  * Runtime PM is disabled for @dev while this function is being executed.
  */
-static int device_resume_early(struct device *dev, pm_message_t state, bool async)
+static void __device_resume_early(struct device *dev, pm_message_t state, bool async)
 {
        pm_callback_t callback = NULL;
        const char *info = NULL;
@@ -811,21 +807,31 @@ Out:
 
        pm_runtime_enable(dev);
        complete_all(&dev->power.completion);
-       return error;
+
+       if (error) {
+               suspend_stats.failed_resume_early++;
+               dpm_save_failed_step(SUSPEND_RESUME_EARLY);
+               dpm_save_failed_dev(dev_name(dev));
+               pm_dev_err(dev, state, async ? " async early" : " early", error);
+       }
 }
 
 static void async_resume_early(void *data, async_cookie_t cookie)
 {
        struct device *dev = data;
-       int error;
-
-       error = device_resume_early(dev, pm_transition, true);
-       if (error)
-               pm_dev_err(dev, pm_transition, " async", error);
 
+       __device_resume_early(dev, pm_transition, true);
        put_device(dev);
 }
 
+static void device_resume_early(struct device *dev)
+{
+       if (dpm_async_fn(dev, async_resume_early))
+               return;
+
+       __device_resume_early(dev, pm_transition, false);
+}
+
 /**
  * dpm_resume_early - Execute "early resume" callbacks for all devices.
  * @state: PM transition of the system being carried out.
@@ -839,14 +845,6 @@ void dpm_resume_early(pm_message_t state)
        mutex_lock(&dpm_list_mtx);
        pm_transition = state;
 
-       /*
-        * Advanced the async threads upfront,
-        * in case the starting of async threads is
-        * delayed by non-async resuming devices.
-        */
-       list_for_each_entry(dev, &dpm_late_early_list, power.entry)
-               dpm_async_fn(dev, async_resume_early);
-
        while (!list_empty(&dpm_late_early_list)) {
                dev = to_device(dpm_late_early_list.next);
                get_device(dev);
@@ -854,17 +852,7 @@ void dpm_resume_early(pm_message_t state)
 
                mutex_unlock(&dpm_list_mtx);
 
-               if (!is_async(dev)) {
-                       int error;
-
-                       error = device_resume_early(dev, state, false);
-                       if (error) {
-                               suspend_stats.failed_resume_early++;
-                               dpm_save_failed_step(SUSPEND_RESUME_EARLY);
-                               dpm_save_failed_dev(dev_name(dev));
-                               pm_dev_err(dev, state, " early", error);
-                       }
-               }
+               device_resume_early(dev);
 
                put_device(dev);
 
@@ -888,12 +876,12 @@ void dpm_resume_start(pm_message_t state)
 EXPORT_SYMBOL_GPL(dpm_resume_start);
 
 /**
- * device_resume - Execute "resume" callbacks for given device.
+ * __device_resume - Execute "resume" callbacks for given device.
  * @dev: Device to handle.
  * @state: PM transition of the system being carried out.
  * @async: If true, the device is being resumed asynchronously.
  */
-static int device_resume(struct device *dev, pm_message_t state, bool async)
+static void __device_resume(struct device *dev, pm_message_t state, bool async)
 {
        pm_callback_t callback = NULL;
        const char *info = NULL;
@@ -975,20 +963,30 @@ static int device_resume(struct device *dev, pm_message_t state, bool async)
 
        TRACE_RESUME(error);
 
-       return error;
+       if (error) {
+               suspend_stats.failed_resume++;
+               dpm_save_failed_step(SUSPEND_RESUME);
+               dpm_save_failed_dev(dev_name(dev));
+               pm_dev_err(dev, state, async ? " async" : "", error);
+       }
 }
 
 static void async_resume(void *data, async_cookie_t cookie)
 {
        struct device *dev = data;
-       int error;
 
-       error = device_resume(dev, pm_transition, true);
-       if (error)
-               pm_dev_err(dev, pm_transition, " async", error);
+       __device_resume(dev, pm_transition, true);
        put_device(dev);
 }
 
+static void device_resume(struct device *dev)
+{
+       if (dpm_async_fn(dev, async_resume))
+               return;
+
+       __device_resume(dev, pm_transition, false);
+}
+
 /**
  * dpm_resume - Execute "resume" callbacks for non-sysdev devices.
  * @state: PM transition of the system being carried out.
@@ -1008,27 +1006,17 @@ void dpm_resume(pm_message_t state)
        pm_transition = state;
        async_error = 0;
 
-       list_for_each_entry(dev, &dpm_suspended_list, power.entry)
-               dpm_async_fn(dev, async_resume);
-
        while (!list_empty(&dpm_suspended_list)) {
                dev = to_device(dpm_suspended_list.next);
+
                get_device(dev);
-               if (!is_async(dev)) {
-                       int error;
 
-                       mutex_unlock(&dpm_list_mtx);
+               mutex_unlock(&dpm_list_mtx);
+
+               device_resume(dev);
 
-                       error = device_resume(dev, state, false);
-                       if (error) {
-                               suspend_stats.failed_resume++;
-                               dpm_save_failed_step(SUSPEND_RESUME);
-                               dpm_save_failed_dev(dev_name(dev));
-                               pm_dev_err(dev, state, "", error);
-                       }
+               mutex_lock(&dpm_list_mtx);
 
-                       mutex_lock(&dpm_list_mtx);
-               }
                if (!list_empty(&dev->power.entry))
                        list_move_tail(&dev->power.entry, &dpm_prepared_list);