2 (C) 2007 Rafael J. Wysocki <rjw@sisk.pl>, GPL
4 I. What is the freezing of tasks?
6 The freezing of tasks is a mechanism by which user space processes and some
7 kernel threads are controlled during hibernation or system-wide suspend (on some
12 There are three per-task flags used for that, PF_NOFREEZE, PF_FROZEN
13 and PF_FREEZER_SKIP (the last one is auxiliary). The tasks that have
14 PF_NOFREEZE unset (all user space processes and some kernel threads) are
15 regarded as 'freezable' and treated in a special way before the system enters a
16 suspend state as well as before a hibernation image is created (in what follows
17 we only consider hibernation, but the description also applies to suspend).
19 Namely, as the first step of the hibernation procedure the function
20 freeze_processes() (defined in kernel/power/process.c) is called. A system-wide
21 variable system_freezing_cnt (as opposed to a per-task flag) is used to indicate
22 whether the system is to undergo a freezing operation. And freeze_processes()
23 sets this variable. After this, it executes try_to_freeze_tasks() that sends a
24 fake signal to all user space processes, and wakes up all the kernel threads.
25 All freezable tasks must react to that by calling try_to_freeze(), which
26 results in a call to __refrigerator() (defined in kernel/freezer.c), which sets
27 the task's PF_FROZEN flag, changes its state to TASK_UNINTERRUPTIBLE and makes
28 it loop until PF_FROZEN is cleared for it. Then, we say that the task is
29 'frozen' and therefore the set of functions handling this mechanism is referred
30 to as 'the freezer' (these functions are defined in kernel/power/process.c,
31 kernel/freezer.c & include/linux/freezer.h). User space processes are generally
32 frozen before kernel threads.
34 __refrigerator() must not be called directly. Instead, use the
35 try_to_freeze() function (defined in include/linux/freezer.h), that checks
36 if the task is to be frozen and makes the task enter __refrigerator().
38 For user space processes try_to_freeze() is called automatically from the
39 signal-handling code, but the freezable kernel threads need to call it
40 explicitly in suitable places or use the wait_event_freezable() or
41 wait_event_freezable_timeout() macros (defined in include/linux/freezer.h)
42 that combine interruptible sleep with checking if the task is to be frozen and
43 calling try_to_freeze(). The main loop of a freezable kernel thread may look
44 like the following one:
49 wait_event_freezable(khubd_wait,
50 !list_empty(&hub_event_list) ||
51 kthread_should_stop());
52 } while (!kthread_should_stop() || !list_empty(&hub_event_list));
54 (from drivers/usb/core/hub.c::hub_thread()).
56 If a freezable kernel thread fails to call try_to_freeze() after the freezer has
57 initiated a freezing operation, the freezing of tasks will fail and the entire
58 hibernation operation will be cancelled. For this reason, freezable kernel
59 threads must call try_to_freeze() somewhere or use one of the
60 wait_event_freezable() and wait_event_freezable_timeout() macros.
62 After the system memory state has been restored from a hibernation image and
63 devices have been reinitialized, the function thaw_processes() is called in
64 order to clear the PF_FROZEN flag for each frozen task. Then, the tasks that
65 have been frozen leave __refrigerator() and continue running.
68 Rationale behind the functions dealing with freezing and thawing of tasks:
69 -------------------------------------------------------------------------
72 - freezes only userspace tasks
74 freeze_kernel_threads():
75 - freezes all tasks (including kernel threads) because we can't freeze
76 kernel threads without freezing userspace tasks
78 thaw_kernel_threads():
79 - thaws only kernel threads; this is particularly useful if we need to do
80 anything special in between thawing of kernel threads and thawing of
81 userspace tasks, or if we want to postpone the thawing of userspace tasks
84 - thaws all tasks (including kernel threads) because we can't thaw userspace
85 tasks without thawing kernel threads
88 III. Which kernel threads are freezable?
90 Kernel threads are not freezable by default. However, a kernel thread may clear
91 PF_NOFREEZE for itself by calling set_freezable() (the resetting of PF_NOFREEZE
92 directly is not allowed). From this point it is regarded as freezable
93 and must call try_to_freeze() in a suitable place.
95 IV. Why do we do that?
97 Generally speaking, there is a couple of reasons to use the freezing of tasks:
99 1. The principal reason is to prevent filesystems from being damaged after
100 hibernation. At the moment we have no simple means of checkpointing
101 filesystems, so if there are any modifications made to filesystem data and/or
102 metadata on disks, we cannot bring them back to the state from before the
103 modifications. At the same time each hibernation image contains some
104 filesystem-related information that must be consistent with the state of the
105 on-disk data and metadata after the system memory state has been restored from
106 the image (otherwise the filesystems will be damaged in a nasty way, usually
107 making them almost impossible to repair). We therefore freeze tasks that might
108 cause the on-disk filesystems' data and metadata to be modified after the
109 hibernation image has been created and before the system is finally powered off.
110 The majority of these are user space processes, but if any of the kernel threads
111 may cause something like this to happen, they have to be freezable.
113 2. Next, to create the hibernation image we need to free a sufficient amount of
114 memory (approximately 50% of available RAM) and we need to do that before
115 devices are deactivated, because we generally need them for swapping out. Then,
116 after the memory for the image has been freed, we don't want tasks to allocate
117 additional memory and we prevent them from doing that by freezing them earlier.
118 [Of course, this also means that device drivers should not allocate substantial
119 amounts of memory from their .suspend() callbacks before hibernation, but this
120 is a separate issue.]
122 3. The third reason is to prevent user space processes and some kernel threads
123 from interfering with the suspending and resuming of devices. A user space
124 process running on a second CPU while we are suspending devices may, for
125 example, be troublesome and without the freezing of tasks we would need some
126 safeguards against race conditions that might occur in such a case.
128 Although Linus Torvalds doesn't like the freezing of tasks, he said this in one
129 of the discussions on LKML (http://lkml.org/lkml/2007/4/27/608):
131 "RJW:> Why we freeze tasks at all or why we freeze kernel threads?
133 Linus: In many ways, 'at all'.
135 I _do_ realize the IO request queue issues, and that we cannot actually do
136 s2ram with some devices in the middle of a DMA. So we want to be able to
137 avoid *that*, there's no question about that. And I suspect that stopping
138 user threads and then waiting for a sync is practically one of the easier
141 So in practice, the 'at all' may become a 'why freeze kernel threads?' and
142 freezing user threads I don't find really objectionable."
144 Still, there are kernel threads that may want to be freezable. For example, if
145 a kernel thread that belongs to a device driver accesses the device directly, it
146 in principle needs to know when the device is suspended, so that it doesn't try
147 to access it at that time. However, if the kernel thread is freezable, it will
148 be frozen before the driver's .suspend() callback is executed and it will be
149 thawed after the driver's .resume() callback has run, so it won't be accessing
150 the device while it's suspended.
152 4. Another reason for freezing tasks is to prevent user space processes from
153 realizing that hibernation (or suspend) operation takes place. Ideally, user
154 space processes should not notice that such a system-wide operation has occurred
155 and should continue running without any problems after the restore (or resume
156 from suspend). Unfortunately, in the most general case this is quite difficult
157 to achieve without the freezing of tasks. Consider, for example, a process
158 that depends on all CPUs being online while it's running. Since we need to
159 disable nonboot CPUs during the hibernation, if this process is not frozen, it
160 may notice that the number of CPUs has changed and may start to work incorrectly
163 V. Are there any problems related to the freezing of tasks?
167 First of all, the freezing of kernel threads may be tricky if they depend one
168 on another. For example, if kernel thread A waits for a completion (in the
169 TASK_UNINTERRUPTIBLE state) that needs to be done by freezable kernel thread B
170 and B is frozen in the meantime, then A will be blocked until B is thawed, which
171 may be undesirable. That's why kernel threads are not freezable by default.
173 Second, there are the following two problems related to the freezing of user
175 1. Putting processes into an uninterruptible sleep distorts the load average.
176 2. Now that we have FUSE, plus the framework for doing device drivers in
177 userspace, it gets even more complicated because some userspace processes are
178 now doing the sorts of things that kernel threads do
179 (https://lists.linux-foundation.org/pipermail/linux-pm/2007-May/012309.html).
181 The problem 1. seems to be fixable, although it hasn't been fixed so far. The
182 other one is more serious, but it seems that we can work around it by using
183 hibernation (and suspend) notifiers (in that case, though, we won't be able to
184 avoid the realization by the user space processes that the hibernation is taking
187 There are also problems that the freezing of tasks tends to expose, although
188 they are not directly related to it. For example, if request_firmware() is
189 called from a device driver's .resume() routine, it will timeout and eventually
190 fail, because the user land process that should respond to the request is frozen
191 at this point. So, seemingly, the failure is due to the freezing of tasks.
192 Suppose, however, that the firmware file is located on a filesystem accessible
193 only through another device that hasn't been resumed yet. In that case,
194 request_firmware() will fail regardless of whether or not the freezing of tasks
195 is used. Consequently, the problem is not really related to the freezing of
196 tasks, since it generally exists anyway.
198 A driver must have all firmwares it may need in RAM before suspend() is called.
199 If keeping them is not practical, for example due to their size, they must be
200 requested early enough using the suspend notifier API described in
201 Documentation/driver-api/pm/notifiers.rst.
203 VI. Are there any precautions to be taken to prevent freezing failures?
207 First of all, grabbing the 'system_transition_mutex' lock to mutually exclude a piece of code
208 from system-wide sleep such as suspend/hibernation is not encouraged.
209 If possible, that piece of code must instead hook onto the suspend/hibernation
210 notifiers to achieve mutual exclusion. Look at the CPU-Hotplug code
211 (kernel/cpu.c) for an example.
213 However, if that is not feasible, and grabbing 'system_transition_mutex' is deemed necessary,
214 it is strongly discouraged to directly call mutex_[un]lock(&system_transition_mutex) since
215 that could lead to freezing failures, because if the suspend/hibernate code
216 successfully acquired the 'system_transition_mutex' lock, and hence that other entity failed
217 to acquire the lock, then that task would get blocked in TASK_UNINTERRUPTIBLE
218 state. As a consequence, the freezer would not be able to freeze that task,
219 leading to freezing failure.
221 However, the [un]lock_system_sleep() APIs are safe to use in this scenario,
222 since they ask the freezer to skip freezing this task, since it is anyway
223 "frozen enough" as it is blocked on 'system_transition_mutex', which will be released
224 only after the entire suspend/hibernation sequence is complete.
225 So, to summarize, use [un]lock_system_sleep() instead of directly using
226 mutex_[un]lock(&system_transition_mutex). That would prevent freezing failures.
229 /sys/power/pm_freeze_timeout controls how long it will cost at most to freeze
230 all user space processes or all freezable kernel threads, in unit of millisecond.
231 The default value is 20000, with range of unsigned integer.