Skip to contentSkip to conversation

Fixes for hybrid sleep

PATCH Superseded Mario Limonciello (AMD) 2 participants 9 messages Sep 25, 2025

Subsystems (4)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover Fixes for hybrid sleep

Ionut Nechita reported recently a hibernate failure, but in debugging the issue it's actually not a hibernate failure; but a hybrid sleep failure.

Multiple changes related to the change of when swap is disabled in the suspend sequence contribute to the failure. See the individual patches for details.

NOTE: I realize this is super late in the cycle, so sorry about that, but I debugged it as fast as I could as soon as I heard about it. If it needs to push out to the next cycle it is what it is.

As it touches two subsystems it either needs to go through linux-pm or drm. Patch 3 has an Ack from Alex, this should merge through linux-pm.

---
v2:
* Fix LKP robot errors without CONFIG_SUSPEND
* Add tags

Mario Limonciello (AMD) (3):
PM: hibernate: Fix hybrid-sleep
PM: hibernate: Add pm_hibernation_mode_is_suspend() drm/amd: Fix hybrid sleep

 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c |  2 +-
 include/linux/suspend.h                 |  2 ++
 kernel/power/hibernate.c                | 22 +++++++++++++++++++++-
 3 files changed, 24 insertions(+), 2 deletions(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH v2 1/3 PM: hibernate: Fix hybrid-sleep

Hybrid sleep will hibernate the system followed by running through the suspend routine. Since both the hibernate and the suspend routine will call pm_restrict_gfp_mask(), pm_restore_gfp_mask() must be called before starting the suspend sequence.

Add an explicit call to pm_restore_gfp_mask() to power_down() before the suspend sequence starts. Don't call pm_restore_gfp_mask() when exiting suspend sequence it is already called:

```
power_down()
->suspend_devices_and_enter()
-->dpm_resume_end()
```

---
v2:
* Move under CONFIG_SUSPEND scope (LKP robot)
* Add tags

---
 kernel/power/hibernate.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
kernel/power/hibernate.c +10 -1
index 2f66ab4538231..52c1818749724 100644
@@ -695,6 +695,7 @@ static void power_down(void)
 
 #ifdef CONFIG_SUSPEND
 	if (hibernation_mode == HIBERNATION_SUSPEND) {
+		pm_restore_gfp_mask();
 		error = suspend_devices_and_enter(mem_sleep_current);
 		if (error) {
 			hibernation_mode = hibernation_ops ?
@@ -862,7 +863,15 @@ int hibernate(void)
 				power_down();
 		}
 		in_suspend = 0;
-		pm_restore_gfp_mask();
+		switch (hibernation_mode) {
+#ifdef CONFIG_SUSPEND
+		case HIBERNATION_SUSPEND:
+			break;
+#endif
+		default:
+			pm_restore_gfp_mask();
+			break;
+		}
 	} else {
 		pm_pr_dbg("Hibernation image restored successfully.\n");
 	}
Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH v2 1/3] PM: hibernate: Fix hybrid-sleep

Quoted 47 lines · Mario Limonciello (AMD)

Hybrid sleep will hibernate the system followed by running through the suspend routine. Since both the hibernate and the suspend routine will call pm_restrict_gfp_mask(), pm_restore_gfp_mask() must be called before starting the suspend sequence.

Add an explicit call to pm_restore_gfp_mask() to power_down() before the suspend sequence starts. Don't call pm_restore_gfp_mask() when exiting suspend sequence it is already called:

```
power_down()
->suspend_devices_and_enter()
-->dpm_resume_end()
```

---
v2:
* Move under CONFIG_SUSPEND scope (LKP robot)
* Add tags

---
 kernel/power/hibernate.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
kernel/power/hibernate.c +1 -0
index 2f66ab4538231..52c1818749724 100644
@@ -695,6 +695,7 @@ static void power_down(void)
 
 #ifdef CONFIG_SUSPEND
        if (hibernation_mode == HIBERNATION_SUSPEND) {
+               pm_restore_gfp_mask();
                error = suspend_devices_and_enter(mem_sleep_current);
                if (error) {
                        hibernation_mode = hibernation_ops ?
@@ -862,7 +863,15 @@ int hibernate(void)
                                power_down();
                }
                in_suspend = 0;

kernel/power/hibernate.c int hibernate(void) · line 874

Quoted 10 lines
kernel/power/hibernate.c +9 -1
-               pm_restore_gfp_mask();
+               switch (hibernation_mode) {
+#ifdef CONFIG_SUSPEND
+               case HIBERNATION_SUSPEND:
+                       break;
+#endif
+               default:
+                       pm_restore_gfp_mask();
+                       break;
+               }

You're breaking HIBERNATION_TEST_RESUME here AFAICS and power_down() doesn't return.

Quoted 4 lines
kernel/power/hibernate.c +0 -1
        } else {
                pm_pr_dbg("Hibernation image restored successfully.\n");
        }
--
#3 lore raw

Re: [PATCH v2 1/3] PM: hibernate: Fix hybrid-sleep

Quoted 54 lines · Rafael J. Wysocki
Quoted 51 lines · Mario Limonciello (AMD)

Hybrid sleep will hibernate the system followed by running through the suspend routine. Since both the hibernate and the suspend routine will call pm_restrict_gfp_mask(), pm_restore_gfp_mask() must be called before starting the suspend sequence.

Add an explicit call to pm_restore_gfp_mask() to power_down() before the suspend sequence starts. Don't call pm_restore_gfp_mask() when exiting suspend sequence it is already called:

```
power_down()
->suspend_devices_and_enter()
-->dpm_resume_end()
```

---
v2:
* Move under CONFIG_SUSPEND scope (LKP robot)
* Add tags

---
 kernel/power/hibernate.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
kernel/power/hibernate.c +4 -1
index 2f66ab4538231..52c1818749724 100644
@@ -695,6 +695,7 @@ static void power_down(void)
 
 #ifdef CONFIG_SUSPEND
        if (hibernation_mode == HIBERNATION_SUSPEND) {
+               pm_restore_gfp_mask();
                error = suspend_devices_and_enter(mem_sleep_current);
                if (error) {
                        hibernation_mode = hibernation_ops ?
@@ -862,7 +863,15 @@ int hibernate(void)
                                power_down();
                }
                in_suspend = 0;
-               pm_restore_gfp_mask();
+               switch (hibernation_mode) {
+#ifdef CONFIG_SUSPEND
+               case HIBERNATION_SUSPEND:
Quoted 8 lines

kernel/power/hibernate.c int hibernate(void) · line 874

Quoted 6 lines
kernel/power/hibernate.c +6 -0
+                       break;
+#endif
+               default:
+                       pm_restore_gfp_mask();
+                       break;
+               }

You're breaking HIBERNATION_TEST_RESUME here AFAICS

Well, not really because of the hibernation_mode check.

Quoted 1 line

and power_down() doesn't return.

But this still is true.

Quoted 5 lines
Quoted 5 lines
kernel/power/hibernate.c +0 -1
        } else {
                pm_pr_dbg("Hibernation image restored successfully.\n");
        }
--
#4 lore raw

Re: [PATCH v2 1/3] PM: hibernate: Fix hybrid-sleep

Quoted 65 lines · Rafael J. Wysocki
Quoted 54 lines · Rafael J. Wysocki
Quoted 51 lines · Mario Limonciello (AMD)

Hybrid sleep will hibernate the system followed by running through the suspend routine. Since both the hibernate and the suspend routine will call pm_restrict_gfp_mask(), pm_restore_gfp_mask() must be called before starting the suspend sequence.

Add an explicit call to pm_restore_gfp_mask() to power_down() before the suspend sequence starts. Don't call pm_restore_gfp_mask() when exiting suspend sequence it is already called:

```
power_down()
->suspend_devices_and_enter()
-->dpm_resume_end()
```

---
v2:
* Move under CONFIG_SUSPEND scope (LKP robot)
* Add tags

---
 kernel/power/hibernate.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
kernel/power/hibernate.c +4 -1
index 2f66ab4538231..52c1818749724 100644
@@ -695,6 +695,7 @@ static void power_down(void)
 
 #ifdef CONFIG_SUSPEND
        if (hibernation_mode == HIBERNATION_SUSPEND) {
+               pm_restore_gfp_mask();
                error = suspend_devices_and_enter(mem_sleep_current);
                if (error) {
                        hibernation_mode = hibernation_ops ?
@@ -862,7 +863,15 @@ int hibernate(void)
                                power_down();
                }
                in_suspend = 0;
-               pm_restore_gfp_mask();
+               switch (hibernation_mode) {
+#ifdef CONFIG_SUSPEND
+               case HIBERNATION_SUSPEND:
Quoted 8 lines

kernel/power/hibernate.c int hibernate(void) · line 874

Quoted 6 lines
kernel/power/hibernate.c +6 -0
+                       break;
+#endif
+               default:
+                       pm_restore_gfp_mask();
+                       break;
+               }

You're breaking HIBERNATION_TEST_RESUME here AFAICS

Quoted 5 lines

Well, not really because of the hibernation_mode check.

Quoted 1 line

and power_down() doesn't return.

But this still is true.

Well, except when it does HIBERNATION_SUSPEND.

But can you just make power_down() call pm_restrict_gfp_mask() before returning and leave the code in hibernate() as is?

#5 lore raw

Re: [PATCH v2 1/3] PM: hibernate: Fix hybrid-sleep

Quoted 68 lines · Rafael J. Wysocki
Quoted 67 lines · Rafael J. Wysocki
Quoted 54 lines · Rafael J. Wysocki
Quoted 51 lines · Mario Limonciello (AMD)

Hybrid sleep will hibernate the system followed by running through the suspend routine. Since both the hibernate and the suspend routine will call pm_restrict_gfp_mask(), pm_restore_gfp_mask() must be called before starting the suspend sequence.

Add an explicit call to pm_restore_gfp_mask() to power_down() before the suspend sequence starts. Don't call pm_restore_gfp_mask() when exiting suspend sequence it is already called:

```
power_down()
->suspend_devices_and_enter()
-->dpm_resume_end()
```

---
v2:
* Move under CONFIG_SUSPEND scope (LKP robot)
* Add tags

---
  kernel/power/hibernate.c | 11 ++++++++++-
  1 file changed, 10 insertions(+), 1 deletion(-)
kernel/power/hibernate.c +4 -1
index 2f66ab4538231..52c1818749724 100644
@@ -695,6 +695,7 @@ static void power_down(void)
 
  #ifdef CONFIG_SUSPEND
         if (hibernation_mode == HIBERNATION_SUSPEND) {
+               pm_restore_gfp_mask();
                 error = suspend_devices_and_enter(mem_sleep_current);
                 if (error) {
                         hibernation_mode = hibernation_ops ?
@@ -862,7 +863,15 @@ int hibernate(void)
                                 power_down();
                 }
                 in_suspend = 0;
-               pm_restore_gfp_mask();
+               switch (hibernation_mode) {
+#ifdef CONFIG_SUSPEND
+               case HIBERNATION_SUSPEND:
Quoted 8 lines

kernel/power/hibernate.c int hibernate(void) · line 874

Quoted 6 lines
kernel/power/hibernate.c +6 -0
+                       break;
+#endif
+               default:
+                       pm_restore_gfp_mask();
+                       break;
+               }

You're breaking HIBERNATION_TEST_RESUME here AFAICS

Well, not really because of the hibernation_mode check.

Quoted 8 lines
Quoted 3 lines
Quoted 1 line

and power_down() doesn't return.

But this still is true.

Well, except when it does HIBERNATION_SUSPEND.

But can you just make power_down() call pm_restrict_gfp_mask() before returning and leave the code in hibernate() as is?

Ah good suggestion. I believe so. I'll test it and get out updated patches if that works.

#6 lore raw

PATCH v2 2/3 PM: hibernate: Add pm_hibernation_mode_is_suspend()

In reply to Mario Limonciello (AMD)

Some drivers have different flows for hibernation and suspend. If the driver opportunistically will skip thaw() then it needs a hint to know what is happening after the hibernate.

Introduce a new symbol pm_hibernation_mode_is_suspend() that drivers can call to determine if suspending the system for this purpose.

---
v2:
* Move under CONFIG_SUSPEND scope (LKP robot)
* Add tags

---
 include/linux/suspend.h  |  2 ++
 kernel/power/hibernate.c | 11 +++++++++++
 2 files changed, 13 insertions(+)
include/linux/suspend.h +2 -0
index 317ae31e89b37..0664c685f0b24 100644
@@ -276,6 +276,7 @@ extern void arch_suspend_enable_irqs(void);
 
 extern int pm_suspend(suspend_state_t state);
 extern bool sync_on_suspend_enabled;
+bool pm_hibernation_mode_is_suspend(void);
 #else /* !CONFIG_SUSPEND */
 #define suspend_valid_only_mem	NULL
 
@@ -288,6 +289,7 @@ static inline bool pm_suspend_via_firmware(void) { return false; }
 static inline bool pm_resume_via_firmware(void) { return false; }
 static inline bool pm_suspend_no_platform(void) { return false; }
 static inline bool pm_suspend_default_s2idle(void) { return false; }
+static inline bool pm_hibernation_mode_is_suspend(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; }
kernel/power/hibernate.c +11 -0
index 52c1818749724..09f5271c98f35 100644
@@ -80,6 +80,17 @@ static const struct platform_hibernation_ops *hibernation_ops;
 
 static atomic_t hibernate_atomic = ATOMIC_INIT(1);
 
+#ifdef CONFIG_SUSPEND
+/**
+ * pm_hibernation_mode_is_suspend - Check if hibernation has been set to suspend
+ */
+bool pm_hibernation_mode_is_suspend(void)
+{
+	return hibernation_mode == HIBERNATION_SUSPEND;
+}
+EXPORT_SYMBOL_GPL(pm_hibernation_mode_is_suspend);
+#endif
+
 bool hibernate_acquire(void)
 {
 	return atomic_add_unless(&hibernate_atomic, -1, 0);
Signature · 1 line
2.43.0
#7 lore raw

PATCH v2 3/3 drm/amd: Fix hybrid sleep

In reply to Mario Limonciello (AMD)

[Why]
commit 530694f54dd5e ("drm/amdgpu: do not resume device in thaw for normal hibernation") optimized the flow for systems that are going into S4 where the power would be turned off. Basically the thaw() callback wouldn't resume the device if the hibernation image was successfully created since the system would be powered off.

This however isn't the correct flow for a system entering into s0i3 after the hibernation image is created. Some of the amdgpu callbacks have different behavior depending upon the intended state of the suspend.

[How]
Use pm_hibernation_mode_is_suspend() as an input to decide whether to run resume during thaw() callback.

---
v2:
* Add tags

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +1 -1
index 395c6be901ce7..dcea66aadfa33 100644
@@ -2665,7 +2665,7 @@ static int amdgpu_pmops_thaw(struct device *dev)
 	struct drm_device *drm_dev = dev_get_drvdata(dev);
 
 	/* do not resume device if it's normal hibernation */
-	if (!pm_hibernate_is_recovering())
+	if (!pm_hibernate_is_recovering() && !pm_hibernation_mode_is_suspend())
 		return 0;
 
 	return amdgpu_device_resume(drm_dev, true);
Signature · 1 line
2.43.0
#8 lore raw
In reply to Mario Limonciello (AMD)
Quoted 5 lines · Mario Limonciello (AMD)

Ionut Nechita reported recently a hibernate failure, but in debugging the issue it's actually not a hibernate failure; but a hybrid sleep failure.

Quoted 7 lines

Multiple changes related to the change of when swap is disabled in the suspend sequence contribute to the failure. See the individual patches for details.

NOTE: I realize this is super late in the cycle, so sorry about that,

No worries.

Quoted 2 lines

but I debugged it as fast as I could as soon as I heard about it. If it needs to push out to the next cycle it is what it is.

Well, it can go into 6.18.

Fortunately, hybrid sleep isn't too popular with Linux users.

Quoted 3 lines

As it touches two subsystems it either needs to go through linux-pm or drm. Patch 3 has an Ack from Alex, this should merge through linux-pm.

Sure.

Thanks!