PM / hibernate: Introduce snapshot test mode for hibernation

This mode is used to verify if the snapshot data written to the swap device can be successfully restored to the memory. It is useful to ease the debugging process on hibernation, since this mode can not only bypass the BIOSes/bootloader, but also the system re-initialization.

For example:

echo snapshot > /sys/power/pm_test
echo disk > /sys/power/state

[  267.959784] PM: Image saving progress:  80%
[  268.038669] PM: Image saving progress:  90%
[  268.111745] PM: Image saving progress: 100%
[  268.129269] PM: Image saving done.
[  268.133485] PM: Wrote 518612 kbytes in 0.75 seconds (691.48 MB/s)
[  268.140564] PM: S|
[  268.160067] hibernation debug: Waiting for 5 seconds.
...
[  273.508638] PM: Looking for hibernation image.
[  273.516583] PM: Image signature found, resuming
[  273.926547] PM: Loading and decompressing image data (129653 pages)...
[  274.122452] PM: Image loading progress:   0%
[  274.322127] PM: Image loading progress:  10%

...

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

Signed-off-by Chen Yu <[email protected]>

---
v3:
- According to the discussion, there was a potential risk in previous version that might break the filesystem. This version avoids that issue by restoring the pages with user/kernel threads kept in frozen. Also updated the document.

---
  kernel/power/hibernate.c | 49 +++++++++++++++++++++++++++++++++++++++++++++++-
  kernel/power/main.c      |  3 +++
  kernel/power/power.h     |  3 +++
  kernel/power/suspend.c   |  7 +++++++
  kernel/power/swap.c      |  7 +++++++
  5 files changed, 68 insertions(+), 1 deletion(-)

kernel/power/hibernate.c +48 -1

index 51441d8..c0c8fe3 100644
@@ -643,11 +643,53 @@ static void power_down(void)
  }
 
  /**
+ * software_resume_unthaw - Resume from a saved hibernation image with threads frozen.
+ *
+ * This routine is similar to software_resume, except that this one tries
+ * to resume from hibernation with user/kernel threads frozen.
+ * It is mainly used for snapshot test mode because it is important to keep the
+ * threads frozen, otherwise the filesystem might be broken due to inconsistent
+ * disk-data/metadata across hibernate.
+ */
+static int software_resume_unthaw(void)
+{
+	int error;
+	unsigned int flags;
+
+	pr_debug("PM: Hibernation image partition %d:%d present\n",
+		MAJOR(swsusp_resume_device), MINOR(swsusp_resume_device));
+
+	pr_debug("PM: Looking for hibernation image.\n");
+	error = swsusp_check();
+	if (error)
+		return error;
+
+	lock_device_hotplug();
+	error = create_basic_memory_bitmaps();
+	if (error)
+		goto Unlock;
+
+	error = swsusp_read(&flags);
+	swsusp_close(FMODE_READ);
+	if (!error)
+		hibernation_restore(flags & SF_PLATFORM_MODE);
+
+	printk(KERN_ERR "PM: Failed to load hibernation image, recovering.\n");
+	swsusp_free();
+	free_basic_memory_bitmaps();
+ Unlock:
+	unlock_device_hotplug();
+
+	return error;
+}
+
+/**
   * hibernate - Carry out system hibernation, including saving the image.
   */
  int hibernate(void)
  {
  	int error, nr_calls = 0;
+	bool snapshot_test = false;
 
  	if (!hibernation_available()) {
  		pr_debug("PM: Hibernation not available.\n");

@@ -699,7 +741,9 @@ int hibernate(void)
  		pr_debug("PM: writing image.\n");
  		error = swsusp_write(flags);
  		swsusp_free();
-		if (!error)
+		if (hibernation_test(TEST_SNAPSHOT))
+			snapshot_test = true;
+		if (!error && !snapshot_test)
  			power_down();
  		in_suspend = 0;
  		pm_restore_gfp_mask();

@@ -711,6 +755,8 @@ int hibernate(void)
  	free_basic_memory_bitmaps();
   Thaw:
  	unlock_device_hotplug();
+	if (snapshot_test)
+		software_resume_unthaw();
  	thaw_processes();
 
  	/* Don't bother checking whether freezer_test_done is true */

@@ -721,6 +767,7 @@ int hibernate(void)
  	atomic_inc(&snapshot_device_available);
   Unlock:
  	unlock_system_sleep();
+
  	return error;
  }
 

kernel/power/main.c +3 -0

index 5ea50b1..80fe48e 100644
@@ -83,6 +83,9 @@ int pm_test_level = TEST_NONE;
 
  static const char * const pm_tests[__TEST_AFTER_LAST] = {
  	[TEST_NONE] = "none",
+#ifdef CONFIG_HIBERNATION
+	[TEST_SNAPSHOT] = "snapshot",
+#endif
  	[TEST_CORE] = "core",
  	[TEST_CPUS] = "processors",
  	[TEST_PLATFORM] = "platform",

kernel/power/power.h +3 -0

index 064963e..101d636 100644
@@ -225,6 +225,9 @@ static inline int restore_highmem(void) { return 0; }
  enum {
  	/* keep first */
  	TEST_NONE,
+#ifdef CONFIG_HIBERNATION
+	TEST_SNAPSHOT,
+#endif
  	TEST_CORE,
  	TEST_CPUS,
  	TEST_PLATFORM,

kernel/power/suspend.c +7 -0

index 0acab9d..7c0d5e9 100644
@@ -479,6 +479,13 @@ static int enter_state(suspend_state_t state)
  			return -EAGAIN;
  		}
  #endif
+	} else if (state == PM_SUSPEND_MEM) {
+#ifdef CONFIG_PM_DEBUG
+		if (pm_test_level != TEST_NONE && pm_test_level < TEST_CORES) {
+			pr_warn("PM: Unsupported test mode for suspend to ram, please choose none/freezer/devices/platform/processors/core.\n");
+			return -EAGAIN;
+		}
+#endif
  	} else if (!valid_state(state)) {
  		return -EINVAL;
  	}

kernel/power/swap.c +7 -0

index 160e100..facd71b 100644
@@ -348,6 +348,13 @@ static int swsusp_swap_check(void)
  	if (res < 0)
  		blkdev_put(hib_resume_bdev, FMODE_WRITE);
 
+	/*
+	 * Update the resume device to the one actually used,
+	 * so software_resume() can use it in case it is invoked
+	 * from hibernate() to test the snapshot.
+	 */
+	swsusp_resume_device = hib_resume_bdev->bd_dev;
+
  	return res;
  }