PM: suspend: Add sysfs attribute to control the "sync on suspend" behavior

The sysfs attribute `/sys/power/sync_on_suspend` controls, whether or not
filesystems are synced by the kernel before system suspend.

Congruously, the behaviour of build-time switch CONFIG_SUSPEND_SKIP_SYNC
is slightly changed: It now defines the run-tim default for the new sysfs
attribute `/sys/power/sync_on_suspend`.

The run-time attribute is added because the existing corresponding
build-time Kconfig flag for (`CONFIG_SUSPEND_SKIP_SYNC`) is not flexible
enough. E.g. Linux distributions that provide pre-compiled kernels
usually want to stick with the default (sync filesystems before suspend)
but under special conditions this needs to be changed.

One example for such a special condition is user-space handling of
suspending block devices (e.g. using `cryptsetup luksSuspend` or `dmsetup
suspend`) before system suspend. The Kernel trying to sync filesystems
after the underlying block device already got suspended obviously leads
to dead-locks. Be aware that you have to take care of the filesystem sync
yourself before suspending the system in those scenarios.

Signed-off-by: Jonas Meurer <jonas@freesources.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Documentation/ABI/testing/sysfs-power

@@ -407,3 +407,16 @@ Contact:	Kalesh Singh <kaleshsingh96@gmail.com>
 Description:
 		The /sys/power/suspend_stats/last_failed_step file contains
 		the last failed step in the suspend/resume path.
+
+What:		/sys/power/sync_on_suspend
+Date:		October 2019
+Contact:	Jonas Meurer <jonas@freesources.org>
+Description:
+		This file controls whether or not the kernel will sync()
+		filesystems during system suspend (after freezing user space
+		and before suspending devices).
+
+		Writing a "1" to this file enables the sync() and writing a "0"
+		disables it.  Reads from the file return the current value.
+		The default is "1" if the build-time "SUSPEND_SKIP_SYNC" config
+		flag is unset, or "0" otherwise.

include/linux/suspend.h

@@ -329,6 +329,7 @@ extern void arch_suspend_disable_irqs(void);
 extern void arch_suspend_enable_irqs(void);
 
 extern int pm_suspend(suspend_state_t state);
+extern bool sync_on_suspend_enabled;
 #else /* !CONFIG_SUSPEND */
 #define suspend_valid_only_mem	NULL
 
@@ -342,6 +343,7 @@ static inline bool pm_suspend_default_s2idle(void) { return false; }
 
 static inline void suspend_set_ops(const struct platform_suspend_ops *ops) {}
 static inline int pm_suspend(suspend_state_t state) { return -ENOSYS; }
+static inline bool sync_on_suspend_enabled(void) { return true; }
 static inline bool idle_should_enter_s2idle(void) { return false; }
 static inline void __init pm_states_init(void) {}
 static inline void s2idle_set_ops(const struct platform_s2idle_ops *ops) {}

kernel/power/Kconfig

@@ -27,7 +27,10 @@ config SUSPEND_SKIP_SYNC
 	  Skip the kernel sys_sync() before freezing user processes.
 	  Some systems prefer not to pay this cost on every invocation
 	  of suspend, or they are content with invoking sync() from
-	  user-space before invoking suspend.  Say Y if that's your case.
+	  user-space before invoking suspend.  There's a run-time switch
+	  at '/sys/power/sync_on_suspend' to configure this behaviour.
+	  This setting changes the default for the run-tim switch. Say Y
+	  to change the default to disable the kernel sys_sync().
 
 config HIBERNATE_CALLBACKS
 	bool

kernel/power/main.c

@@ -190,6 +190,38 @@ static ssize_t mem_sleep_store(struct kobject *kobj, struct kobj_attribute *attr
 }
 
 power_attr(mem_sleep);
+
+/*
+ * sync_on_suspend: invoke ksys_sync_helper() before suspend.
+ *
+ * show() returns whether ksys_sync_helper() is invoked before suspend.
+ * store() accepts 0 or 1.  0 disables ksys_sync_helper() and 1 enables it.
+ */
+bool sync_on_suspend_enabled = !IS_ENABLED(CONFIG_SUSPEND_SKIP_SYNC);
+
+static ssize_t sync_on_suspend_show(struct kobject *kobj,
+				   struct kobj_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", sync_on_suspend_enabled);
+}
+
+static ssize_t sync_on_suspend_store(struct kobject *kobj,
+				    struct kobj_attribute *attr,
+				    const char *buf, size_t n)
+{
+	unsigned long val;
+
+	if (kstrtoul(buf, 10, &val))
+		return -EINVAL;
+
+	if (val > 1)
+		return -EINVAL;
+
+	sync_on_suspend_enabled = !!val;
+	return n;
+}
+
+power_attr(sync_on_suspend);
 #endif /* CONFIG_SUSPEND */
 
 #ifdef CONFIG_PM_SLEEP_DEBUG
@@ -855,6 +887,7 @@ static struct attribute * g[] = {
 	&wakeup_count_attr.attr,
 #ifdef CONFIG_SUSPEND
 	&mem_sleep_attr.attr,
+	&sync_on_suspend_attr.attr,
 #endif
 #ifdef CONFIG_PM_AUTOSLEEP
 	&autosleep_attr.attr,

kernel/power/suspend.c

@@ -564,7 +564,7 @@ static int enter_state(suspend_state_t state)
 	if (state == PM_SUSPEND_TO_IDLE)
 		s2idle_begin();
 
-	if (!IS_ENABLED(CONFIG_SUSPEND_SKIP_SYNC)) {
+	if (sync_on_suspend_enabled) {
 		trace_suspend_resume(TPS("sync_filesystems"), 0, true);
 		ksys_sync_helper();
 		trace_suspend_resume(TPS("sync_filesystems"), 0, false);