Skip to contentSkip to conversation

Suspend notifiers

RFC PATCH In review Rafael J. Wysocki 3 participants 10 messages Apr 21, 2007 – Apr 23, 2007

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover Suspend notifiers

Hi,

The following two patches are intended to address the problem that some drivers allocate a lot of memory in their .suspend() routines causing swsusp to fail (and generally may want to do some things that shouldn't be done in .suspend()/.resume()).

Greetings,
Rafael

Signature · 3 lines
If you don't have the time to read,
you don't have the time or the tools to write.
		- Stephen King
#1 lore raw

RFC PATCH 1/2 swsusp: Free more memory

From: Rafael J. Wysocki <[email protected]>

Move the definition of PAGES_FOR_IO to kernel/power/power.h and introduce SPARE_PAGES representing the number of pages that should be freed by the swsusp's memory shrinker in addition to PAGES_FOR_IO so that device drivers can allocate some memory (up to 1 MB total) in their .suspend() routines without causing the suspend to fail.

---
 include/linux/suspend.h |    6 ------
 kernel/power/power.h    |   12 +++++++++++-
 kernel/power/swsusp.c   |    2 +-
 3 files changed, 12 insertions(+), 8 deletions(-)
include/linux/suspend.h +0 -6
Index: linux-2.6.21-rc6-mm1/include/linux/suspend.h
===================================================================
@@ -60,10 +60,4 @@ struct saved_context;
 void __save_processor_state(struct saved_context *ctxt);
 void __restore_processor_state(struct saved_context *ctxt);
 
-/*
- * XXX: We try to keep some more pages free so that I/O operations succeed
- * without paging. Might this be more?
- */
-#define PAGES_FOR_IO	1024
-
 #endif /* _LINUX_SWSUSP_H */
kernel/power/power.h +11 -1
Index: linux-2.6.21-rc6-mm1/kernel/power/power.h
===================================================================
@@ -14,8 +14,18 @@ struct swsusp_info {
 
 
 #ifdef CONFIG_SOFTWARE_SUSPEND
-extern int pm_suspend_disk(void);
+/*
+ * Keep some memory free so that I/O operations can succeed without paging
+ * [Might this be more than 4 MB?]
+ */
+#define PAGES_FOR_IO	((4096 * 1024) >> PAGE_SHIFT)
+/*
+ * Keep 1 MB of memory free so that device drivers can allocate some pages in
+ * their .suspend() routines without breaking the suspend to disk.
+ */
+#define SPARE_PAGES	((1024 * 1024) >> PAGE_SHIFT)
 
+extern int pm_suspend_disk(void);
 #else
 static inline int pm_suspend_disk(void)
 {
kernel/power/swsusp.c +1 -1
Index: linux-2.6.21-rc6-mm1/kernel/power/swsusp.c
===================================================================
@@ -233,7 +233,7 @@ int swsusp_shrink_memory(void)
 		long size, highmem_size;
 
 		highmem_size = count_highmem_pages();
-		size = count_data_pages() + PAGES_FOR_IO;
+		size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES;
 		tmp = size;
 		size += highmem_size;
 		for_each_zone (zone)
#2 lore raw

Re: [RFC][PATCH -mm 1/2] swsusp: Free more memory

Hi.

Quoted 7 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Move the definition of PAGES_FOR_IO to kernel/power/power.h and introduce SPARE_PAGES representing the number of pages that should be freed by the swsusp's memory shrinker in addition to PAGES_FOR_IO so that device drivers can allocate some memory (up to 1 MB total) in their .suspend() routines without causing the suspend to fail.

1MB is not going to be enough for people with DRI. I have a cycle on my machine recording 3510 extra pages (13 meg) allocated during the atomic copy, and I know there are others with larger numbers.

Nigel

#3 lore raw

Re: [RFC][PATCH -mm 1/2] swsusp: Free more memory

Quoted 11 lines · Nigel Cunningham

Hi.

Quoted 7 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Move the definition of PAGES_FOR_IO to kernel/power/power.h and introduce SPARE_PAGES representing the number of pages that should be freed by the swsusp's memory shrinker in addition to PAGES_FOR_IO so that device drivers can allocate some memory (up to 1 MB total) in their .suspend() routines without causing the suspend to fail.

Quoted 3 lines

1MB is not going to be enough for people with DRI. I have a cycle on my machine recording 3510 extra pages (13 meg) allocated during the atomic copy, and I know there are others with larger numbers.

I know. The other mechanism (ie. notifiers) is for them.

The idea of this patch is to make it possible to _avoid_ using the notifiers if you need only a limited amount of memory.

Greetings,
Rafael

#4 lore raw

Re: [RFC][PATCH -mm 1/2] swsusp: Free more memory

Hi.

Quoted 12 lines · Rafael J. Wysocki
Quoted 11 lines · Nigel Cunningham

Hi.

Quoted 7 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Move the definition of PAGES_FOR_IO to kernel/power/power.h and introduce SPARE_PAGES representing the number of pages that should be freed by the swsusp's memory shrinker in addition to PAGES_FOR_IO so that device drivers can allocate some memory (up to 1 MB total) in their .suspend() routines without causing the suspend to fail.

Quoted 8 lines
Quoted 3 lines

1MB is not going to be enough for people with DRI. I have a cycle on my machine recording 3510 extra pages (13 meg) allocated during the atomic copy, and I know there are others with larger numbers.

I know. The other mechanism (ie. notifiers) is for them.

The idea of this patch is to make it possible to _avoid_ using the notifiers if you need only a limited amount of memory.

That's fine when the notifiers are actually used, but for the moment....

Regards,

Nigel

#5 lore raw

Re: [RFC][PATCH -mm 1/2] swsusp: Free more memory

Quoted 19 lines · Nigel Cunningham

Hi.

Quoted 16 lines · Rafael J. Wysocki
Quoted 14 lines · Nigel Cunningham

Hi.

Quoted 7 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Move the definition of PAGES_FOR_IO to kernel/power/power.h and introduce SPARE_PAGES representing the number of pages that should be freed by the swsusp's memory shrinker in addition to PAGES_FOR_IO so that device drivers can allocate some memory (up to 1 MB total) in their .suspend() routines without causing the suspend to fail.

1MB is not going to be enough for people with DRI. I have a cycle on my machine recording 3510 extra pages (13 meg) allocated during the atomic copy, and I know there are others with larger numbers.

Quoted 6 lines
Quoted 4 lines

I know. The other mechanism (ie. notifiers) is for them.

The idea of this patch is to make it possible to _avoid_ using the notifiers if you need only a limited amount of memory.

That's fine when the notifiers are actually used, but for the moment....

I know, but if I make SPARE_PAGES large enough for everyone, then no one will have any reason to use the notifiers. :-)

Greetings,
Rafael

#6 lore raw

Re: [RFC][PATCH -mm 1/2] swsusp: Free more memory

In reply to Rafael J. Wysocki

Hi!

Quoted 7 lines

Move the definition of PAGES_FOR_IO to kernel/power/power.h and introduce SPARE_PAGES representing the number of pages that should be freed by the swsusp's memory shrinker in addition to PAGES_FOR_IO so that device drivers can allocate some memory (up to 1 MB total) in their .suspend() routines without causing the suspend to fail.

ACK.

Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#7 lore raw

RFC PATCH 2/2 PM: Introduce suspend notifiers

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Make it possible to register suspend notifiers so that subsystems can perform suspend-related operations which should not be carried out by device drivers' .suspend() and .resume() routines.

---
 Documentation/power/suspend-notifiers.txt |   41 ++++++++++++++++++++++++
 include/linux/notifier.h                  |    6 +++
 include/linux/suspend.h                   |   27 +++++++++++++--
 kernel/power/Makefile                     |    2 -
 kernel/power/disk.c                       |   21 ++++++++++++
 kernel/power/main.c                       |   13 +++++++
 kernel/power/notify.c                     |   51 ++++++++++++++++++++++++++++++
 kernel/power/power.h                      |    3 +
 kernel/power/user.c                       |   19 +++++++++++
 9 files changed, 178 insertions(+), 5 deletions(-)
kernel/power/notify.c new file +51 -0
Index: linux-2.6.21-rc6-mm1/kernel/power/notify.c
===================================================================
@@ -0,0 +1,51 @@
+/*
+ * linux/kernel/power/notify.c
+ *
+ * This file contains functions used for registering and calling suspend
+ * notifiers that can be used by subsystems for carrying out some special
+ * suspend-related operations.
+ *
+ * Copyright (C) 2007 Rafael J. Wysocki <[email protected]>
+ *
+ * This file is released under the GPLv2.
+ *
+ */
+
+#include <linux/smp.h>
+#include <linux/notifier.h>
+#include <linux/module.h>
+#include <linux/suspend.h>
+
+static DEFINE_MUTEX(suspend_notifier_lock);
+
+static RAW_NOTIFIER_HEAD(suspend_chain);
+
+int register_suspend_notifier(struct notifier_block *nb)
+{
+	int ret;
+	mutex_lock(&suspend_notifier_lock);
+	ret = raw_notifier_chain_register(&suspend_chain, nb);
+	mutex_unlock(&suspend_notifier_lock);
+	return ret;
+}
+EXPORT_SYMBOL(register_suspend_notifier);
+
+void unregister_suspend_notifier(struct notifier_block *nb)
+{
+	mutex_lock(&suspend_notifier_lock);
+	raw_notifier_chain_unregister(&suspend_chain, nb);
+	mutex_unlock(&suspend_notifier_lock);
+}
+EXPORT_SYMBOL(unregister_suspend_notifier);
+
+int suspend_notifier_call_chain(unsigned long val)
+{
+	int error = 0;
+
+	mutex_lock(&suspend_notifier_lock);
+	if (raw_notifier_call_chain(&suspend_chain, val, NULL) == NOTIFY_BAD)
+		error = -EINVAL;
+
+	mutex_unlock(&suspend_notifier_lock);
+	return error;
+}
include/linux/suspend.h +23 -4
Index: linux-2.6.21-rc6-mm1/include/linux/suspend.h
===================================================================
@@ -24,15 +24,16 @@ struct pbe {
 extern void drain_local_pages(void);
 extern void mark_free_pages(struct zone *zone);
 
-#if defined(CONFIG_PM) && defined(CONFIG_VT) && defined(CONFIG_VT_CONSOLE)
+#ifdef CONFIG_PM
+#if defined(CONFIG_VT) && defined(CONFIG_VT_CONSOLE)
 extern int pm_prepare_console(void);
 extern void pm_restore_console(void);
 #else
 static inline int pm_prepare_console(void) { return 0; }
 static inline void pm_restore_console(void) {}
-#endif
+#endif /* defined(CONFIG_VT) && defined(CONFIG_VT_CONSOLE) */
 
-#if defined(CONFIG_PM) && defined(CONFIG_SOFTWARE_SUSPEND)
+#ifdef CONFIG_SOFTWARE_SUSPEND
 /* kernel/power/swsusp.c */
 extern int software_suspend(void);
 /* kernel/power/snapshot.c */
@@ -52,7 +53,7 @@ static inline void register_nosave_regio
 static inline int swsusp_page_is_forbidden(struct page *p) { return 0; }
 static inline void swsusp_set_page_free(struct page *p) {}
 static inline void swsusp_unset_page_free(struct page *p) {}
-#endif /* defined(CONFIG_PM) && defined(CONFIG_SOFTWARE_SUSPEND) */
+#endif /* CONFIG_SOFTWARE_SUSPEND */
 
 void save_processor_state(void);
 void restore_processor_state(void);
@@ -60,4 +61,22 @@ struct saved_context;
 void __save_processor_state(struct saved_context *ctxt);
 void __restore_processor_state(struct saved_context *ctxt);
 
+int register_suspend_notifier(struct notifier_block *nb);
+void unregister_suspend_notifier(struct notifier_block *nb);
+
+#define suspend_notifier(fn, pri) {				\
+	static struct notifier_block fn##_nb =			\
+		{ .notifier_call = fn, .priority = pri };	\
+	register_suspend_notifier(&fn##_nb);			\
+}
+#else /* CONFIG_PM */
+static inline int register_suspend_notifier(struct notifier_block *nb) {
+	return 0;
+}
+static inline void unregister_suspend_notifier(struct notifier_block *nb) {
+}
+
+#define suspend_notifier(fn, pri)	do { (void)(fn); } while (0)
+#endif
+
 #endif /* _LINUX_SWSUSP_H */
kernel/power/Makefile +1 -1
Index: linux-2.6.21-rc6-mm1/kernel/power/Makefile
===================================================================
@@ -3,7 +3,7 @@ ifeq ($(CONFIG_PM_DEBUG),y)
 EXTRA_CFLAGS	+=	-DDEBUG
 endif
 
-obj-y				:= main.o process.o console.o
+obj-y				:= main.o process.o console.o notify.o
 obj-$(CONFIG_PM_LEGACY)		+= pm.o
 obj-$(CONFIG_SOFTWARE_SUSPEND)	+= swsusp.o disk.o snapshot.o swap.o user.o
 
kernel/power/power.h +3 -0
Index: linux-2.6.21-rc6-mm1/kernel/power/power.h
===================================================================
@@ -170,3 +170,6 @@ extern int suspend_enter(suspend_state_t
 struct timeval;
 extern void swsusp_show_speed(struct timeval *, struct timeval *,
 				unsigned int, char *);
+
+/* kernel/power/notify.c */
+extern int suspend_notifier_call_chain(unsigned long val);
include/linux/notifier.h +6 -0
Index: linux-2.6.21-rc6-mm1/include/linux/notifier.h
===================================================================
@@ -197,5 +197,11 @@ extern int __srcu_notifier_call_chain(st
 #define CPU_LOCK_ACQUIRE	0x0008 /* Acquire all hotcpu locks */
 #define CPU_LOCK_RELEASE	0x0009 /* Release all hotcpu locks */
 
+#define SUSPEND_PREPARE		0x0002 /* Going to freeze tasks */
+#define SUSPEND_ENTER_PREPARE	0x0003 /* Tasks are frozen, we are suspending */
+#define SUSPEND_THAW_PREPARE	0x0004 /* Going to thaw frozen tasks */
+#define SUSPEND_FINISHED	0x0005 /* Tasks have been thawed */
+#define SUSPEND_RESTORE_PREPARE	0x0006 /* STD restore is going to happen */
+
 #endif /* __KERNEL__ */
 #endif /* _LINUX_NOTIFIER_H */
kernel/power/disk.c +21 -0
Index: linux-2.6.21-rc6-mm1/kernel/power/disk.c
===================================================================
@@ -114,6 +114,7 @@ static int prepare_processes(void)
 	pm_prepare_console();
 	if (freeze_processes()) {
 		error = -EBUSY;
+		suspend_notifier_call_chain(SUSPEND_THAW_PREPARE);
 		unprepare_processes();
 	}
 	return error;
@@ -139,6 +140,10 @@ int pm_suspend_disk(void)
 	if (error)
 		goto Exit;
 
+	error = suspend_notifier_call_chain(SUSPEND_PREPARE);
+	if (error)
+		goto Finish;
+
 	error = prepare_processes();
 	if (error)
 		goto Finish;
@@ -149,6 +154,10 @@ int pm_suspend_disk(void)
 		goto Thaw;
 	}
 
+	error = suspend_notifier_call_chain(SUSPEND_ENTER_PREPARE);
+	if (error)
+		goto Thaw;
+
 	/* Free memory before shutting down devices. */
 	error = swsusp_shrink_memory();
 	if (error)
@@ -205,8 +214,10 @@ int pm_suspend_disk(void)
 	device_resume();
 	resume_console();
  Thaw:
+	suspend_notifier_call_chain(SUSPEND_THAW_PREPARE);
 	unprepare_processes();
  Finish:
+	suspend_notifier_call_chain(SUSPEND_FINISHED);
 	free_basic_memory_bitmaps();
  Exit:
 	atomic_inc(&snapshot_device_available);
@@ -267,6 +278,10 @@ static int software_resume(void)
 	if (error)
 		goto Finish;
 
+	error = suspend_notifier_call_chain(SUSPEND_PREPARE);
+	if (error)
+		goto Done;
+
 	pr_debug("PM: Preparing processes for restore.\n");
 	error = prepare_processes();
 	if (error) {
@@ -282,6 +297,10 @@ static int software_resume(void)
 		goto Thaw;
 	}
 
+	error = suspend_notifier_call_chain(SUSPEND_RESTORE_PREPARE);
+	if (error)
+		goto Thaw;
+
 	pr_debug("PM: Preparing devices for restore.\n");
 
 	suspend_console();
@@ -300,8 +319,10 @@ static int software_resume(void)
 	resume_console();
  Thaw:
 	printk(KERN_ERR "PM: Restore failed, recovering.\n");
+	suspend_notifier_call_chain(SUSPEND_THAW_PREPARE);
 	unprepare_processes();
  Done:
+	suspend_notifier_call_chain(SUSPEND_FINISHED);
 	free_basic_memory_bitmaps();
  Finish:
 	atomic_inc(&snapshot_device_available);
kernel/power/main.c +13 -0
Index: linux-2.6.21-rc6-mm1/kernel/power/main.c
===================================================================
@@ -86,11 +86,19 @@ static int suspend_prepare(suspend_state
 
 	pm_prepare_console();
 
+	error = suspend_notifier_call_chain(SUSPEND_PREPARE);
+	if (error)
+		goto Finish;
+
 	if (freeze_processes()) {
 		error = -EAGAIN;
 		goto Thaw;
 	}
 
+	error = suspend_notifier_call_chain(SUSPEND_ENTER_PREPARE);
+	if (error)
+		goto Thaw;
+
 	if ((free_pages = global_page_state(NR_FREE_PAGES))
 			< FREE_PAGE_NUMBER) {
 		pr_debug("PM: free some memory\n");
@@ -123,8 +131,11 @@ static int suspend_prepare(suspend_state
 	device_resume();
 	resume_console();
  Thaw:
+	suspend_notifier_call_chain(SUSPEND_THAW_PREPARE);
 	thaw_processes();
+ Finish:
 	pm_restore_console();
+	suspend_notifier_call_chain(SUSPEND_FINISHED);
 	return error;
 }
 
@@ -162,8 +173,10 @@ static void suspend_finish(suspend_state
 	pm_finish(state);
 	device_resume();
 	resume_console();
+	suspend_notifier_call_chain(SUSPEND_THAW_PREPARE);
 	thaw_processes();
 	pm_restore_console();
+	suspend_notifier_call_chain(SUSPEND_FINISHED);
 }
 
 
kernel/power/user.c +19 -0
Index: linux-2.6.21-rc6-mm1/kernel/power/user.c
===================================================================
@@ -148,6 +148,10 @@ static inline int snapshot_suspend(int p
 	int error;
 
 	mutex_lock(&pm_mutex);
+	error = suspend_notifier_call_chain(SUSPEND_ENTER_PREPARE);
+	if (error)
+		goto Finish;
+
 	/* Free memory before shutting down devices. */
 	error = swsusp_shrink_memory();
 	if (error)
@@ -186,6 +190,10 @@ static inline int snapshot_restore(int p
 
 	mutex_lock(&pm_mutex);
 	pm_prepare_console();
+	error = suspend_notifier_call_chain(SUSPEND_RESTORE_PREPARE);
+	if (error)
+		goto Finish;
+
 	if (platform_suspend) {
 		error = platform_prepare();
 		if (error)
@@ -236,7 +244,12 @@ static int snapshot_ioctl(struct inode *
 		if (data->frozen)
 			break;
 		mutex_lock(&pm_mutex);
+		error = suspend_notifier_call_chain(SUSPEND_PREPARE);
+		if (error)
+			break;
+
 		if (freeze_processes()) {
+			suspend_notifier_call_chain(SUSPEND_THAW_PREPARE);
 			thaw_processes();
 			error = -EBUSY;
 		}
@@ -249,7 +262,9 @@ static int snapshot_ioctl(struct inode *
 		if (!data->frozen)
 			break;
 		mutex_lock(&pm_mutex);
+		suspend_notifier_call_chain(SUSPEND_THAW_PREPARE);
 		thaw_processes();
+		suspend_notifier_call_chain(SUSPEND_FINISHED);
 		mutex_unlock(&pm_mutex);
 		data->frozen = 0;
 		break;
@@ -350,6 +365,10 @@ static int snapshot_ioctl(struct inode *
 			break;
 		}
 
+		error = suspend_notifier_call_chain(SUSPEND_ENTER_PREPARE);
+		if (error)
+			break;
+
 		if (pm_ops->prepare) {
 			error = pm_ops->prepare(PM_SUSPEND_MEM);
 			if (error)
Documentation/power/suspend-notifiers.txt new file +41 -0

41 lines not on this page · raw message

#8 lore raw

Re: [RFC][PATCH -mm 2/2] PM: Introduce suspend notifiers

Hi!

Quoted 20 lines

Index: linux-2.6.21-rc6-mm1/Documentation/power/suspend-notifiers.txt
===================================================================

Documentation/power/suspend-notifiers.txt new file +15 -0
@@ -0,0 +1,41 @@
+Suspend notifiers
+	(C) 2007 Rafael J. Wysocki <[email protected]>, GPL
+
+There are some operations that device drivers may want to carry out in their
+.suspend() routines, but shouldn't, because they can cause a suspend to fail.
+For example, a driver may want to allocate a substantial amount of memory
+(like 50 MB) in .suspend(), but that shouldn't be done after the swsusp's memory
+shrinker has run.  Also, there may be some operations, that subsystems may want
+to carry out before a suspend or after a resume, requiring the system to be
+fully functional, so the drivers' .suspend() and .resume() routines are not
+suitable for this purpose.
+
+The subsystems that have such needs can register suspend notifiers that will be
+notified of the following events by the suspend core:
+

Documentation/power/suspend-notifiers.txt line 25

Quoted 10 lines
Documentation/power/suspend-notifiers.txt new file +10 -0
+SUSPEND_PREPARE		the system is going to suspend, tasks will be frozen
+			immediately
+
+SUSPEND_ENTER_PREPARE	tasks have been frozen, memory is going to be freed
+			and devices are going to be suspended
+
+SUSPEND_THAW_PREPARE	devices have been resumed, tasks will be thawed
+			immediately
+
+SUSPEND_FINISHED	the resume is complete, the system is fully functional

+ This hook is called even if suspend fails, so that we do not
+ leak memory.

?

Looks ok to me.

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#9 lore raw

Re: [RFC][PATCH -mm 2/2] PM: Introduce suspend notifiers

Hi,

Quoted 38 lines · Pavel Machek

Hi!

Quoted 20 lines

Index: linux-2.6.21-rc6-mm1/Documentation/power/suspend-notifiers.txt
===================================================================

Documentation/power/suspend-notifiers.txt new file +15 -0
@@ -0,0 +1,41 @@
+Suspend notifiers
+	(C) 2007 Rafael J. Wysocki <[email protected]>, GPL
+
+There are some operations that device drivers may want to carry out in their
+.suspend() routines, but shouldn't, because they can cause a suspend to fail.
+For example, a driver may want to allocate a substantial amount of memory
+(like 50 MB) in .suspend(), but that shouldn't be done after the swsusp's memory
+shrinker has run.  Also, there may be some operations, that subsystems may want
+to carry out before a suspend or after a resume, requiring the system to be
+fully functional, so the drivers' .suspend() and .resume() routines are not
+suitable for this purpose.
+
+The subsystems that have such needs can register suspend notifiers that will be
+notified of the following events by the suspend core:
+

Documentation/power/suspend-notifiers.txt line 25

Quoted 10 lines
Documentation/power/suspend-notifiers.txt new file +10 -0
+SUSPEND_PREPARE		the system is going to suspend, tasks will be frozen
+			immediately
+
+SUSPEND_ENTER_PREPARE	tasks have been frozen, memory is going to be freed
+			and devices are going to be suspended
+
+SUSPEND_THAW_PREPARE	devices have been resumed, tasks will be thawed
+			immediately
+
+SUSPEND_FINISHED	the resume is complete, the system is fully functional

+ This hook is called even if suspend fails, so that we do not
+ leak memory.

Quoted 1 line

?

Yes, I'll add something like this.