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 four per-task flags used for that, PF_NOFREEZE, PF_FROZEN, TIF_FREEZE
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. It executes
21 try_to_freeze_tasks() that sets TIF_FREEZE for all of the freezable tasks and
22 either wakes them up, if they are kernel threads, or sends fake signals to them,
23 if they are user space processes. A task that has TIF_FREEZE set, should react
24 to it by calling the function called __refrigerator() (defined in
25 kernel/power/process.c), which sets the task's PF_FROZEN flag, changes its state
26 to TASK_UNINTERRUPTIBLE and makes it loop until PF_FROZEN is cleared for it.
27 Then, we say that the task is 'frozen' and therefore the set of functions
28 handling this mechanism is referred to as 'the freezer' (these functions are
29 defined in kernel/power/process.c and include/linux/freezer.h). User space
30 processes are generally frozen before kernel threads.
32 __refrigerator() must not be called directly. Instead, use the
33 try_to_freeze() function (defined in include/linux/freezer.h), that checks
34 the task's TIF_FREEZE flag and makes the task enter __refrigerator() if the
37 For user space processes try_to_freeze() is called automatically from the
38 signal-handling code, but the freezable kernel threads need to call it
39 explicitly in suitable places or use the wait_event_freezable() or
40 wait_event_freezable_timeout() macros (defined in include/linux/freezer.h)
41 that combine interruptible sleep with checking if TIF_FREEZE is set and calling
42 try_to_freeze(). The main loop of a freezable kernel thread may look like the
48 wait_event_freezable(khubd_wait,
49 !list_empty(&hub_event_list) ||
50 kthread_should_stop());
51 } while (!kthread_should_stop() || !list_empty(&hub_event_list));
53 (from drivers/usb/core/hub.c::hub_thread()).
55 If a freezable kernel thread fails to call try_to_freeze() after the freezer has
56 set TIF_FREEZE for it, the freezing of tasks will fail and the entire
57 hibernation operation will be cancelled. For this reason, freezable kernel
58 threads must call try_to_freeze() somewhere or use one of the
59 wait_event_freezable() and wait_event_freezable_timeout() macros.
61 After the system memory state has been restored from a hibernation image and
62 devices have been reinitialized, the function thaw_processes() is called in
63 order to clear the PF_FROZEN flag for each frozen task. Then, the tasks that
64 have been frozen leave __refrigerator() and continue running.
66 III. Which kernel threads are freezable?
68 Kernel threads are not freezable by default. However, a kernel thread may clear
69 PF_NOFREEZE for itself by calling set_freezable() (the resetting of PF_NOFREEZE
70 directly is not allowed). From this point it is regarded as freezable
71 and must call try_to_freeze() in a suitable place.
73 IV. Why do we do that?
75 Generally speaking, there is a couple of reasons to use the freezing of tasks:
77 1. The principal reason is to prevent filesystems from being damaged after
78 hibernation. At the moment we have no simple means of checkpointing
79 filesystems, so if there are any modifications made to filesystem data and/or
80 metadata on disks, we cannot bring them back to the state from before the
81 modifications. At the same time each hibernation image contains some
82 filesystem-related information that must be consistent with the state of the
83 on-disk data and metadata after the system memory state has been restored from
84 the image (otherwise the filesystems will be damaged in a nasty way, usually
85 making them almost impossible to repair). We therefore freeze tasks that might
86 cause the on-disk filesystems' data and metadata to be modified after the
87 hibernation image has been created and before the system is finally powered off.
88 The majority of these are user space processes, but if any of the kernel threads
89 may cause something like this to happen, they have to be freezable.
91 2. Next, to create the hibernation image we need to free a sufficient amount of
92 memory (approximately 50% of available RAM) and we need to do that before
93 devices are deactivated, because we generally need them for swapping out. Then,
94 after the memory for the image has been freed, we don't want tasks to allocate
95 additional memory and we prevent them from doing that by freezing them earlier.
96 [Of course, this also means that device drivers should not allocate substantial
97 amounts of memory from their .suspend() callbacks before hibernation, but this
100 3. The third reason is to prevent user space processes and some kernel threads
101 from interfering with the suspending and resuming of devices. A user space
102 process running on a second CPU while we are suspending devices may, for
103 example, be troublesome and without the freezing of tasks we would need some
104 safeguards against race conditions that might occur in such a case.
106 Although Linus Torvalds doesn't like the freezing of tasks, he said this in one
107 of the discussions on LKML (http://lkml.org/lkml/2007/4/27/608):
109 "RJW:> Why we freeze tasks at all or why we freeze kernel threads?
111 Linus: In many ways, 'at all'.
113 I _do_ realize the IO request queue issues, and that we cannot actually do
114 s2ram with some devices in the middle of a DMA. So we want to be able to
115 avoid *that*, there's no question about that. And I suspect that stopping
116 user threads and then waiting for a sync is practically one of the easier
119 So in practice, the 'at all' may become a 'why freeze kernel threads?' and
120 freezing user threads I don't find really objectionable."
122 Still, there are kernel threads that may want to be freezable. For example, if
123 a kernel that belongs to a device driver accesses the device directly, it in
124 principle needs to know when the device is suspended, so that it doesn't try to
125 access it at that time. However, if the kernel thread is freezable, it will be
126 frozen before the driver's .suspend() callback is executed and it will be
127 thawed after the driver's .resume() callback has run, so it won't be accessing
128 the device while it's suspended.
130 4. Another reason for freezing tasks is to prevent user space processes from
131 realizing that hibernation (or suspend) operation takes place. Ideally, user
132 space processes should not notice that such a system-wide operation has occurred
133 and should continue running without any problems after the restore (or resume
134 from suspend). Unfortunately, in the most general case this is quite difficult
135 to achieve without the freezing of tasks. Consider, for example, a process
136 that depends on all CPUs being online while it's running. Since we need to
137 disable nonboot CPUs during the hibernation, if this process is not frozen, it
138 may notice that the number of CPUs has changed and may start to work incorrectly
141 V. Are there any problems related to the freezing of tasks?
145 First of all, the freezing of kernel threads may be tricky if they depend one
146 on another. For example, if kernel thread A waits for a completion (in the
147 TASK_UNINTERRUPTIBLE state) that needs to be done by freezable kernel thread B
148 and B is frozen in the meantime, then A will be blocked until B is thawed, which
149 may be undesirable. That's why kernel threads are not freezable by default.
151 Second, there are the following two problems related to the freezing of user
153 1. Putting processes into an uninterruptible sleep distorts the load average.
154 2. Now that we have FUSE, plus the framework for doing device drivers in
155 userspace, it gets even more complicated because some userspace processes are
156 now doing the sorts of things that kernel threads do
157 (https://lists.linux-foundation.org/pipermail/linux-pm/2007-May/012309.html).
159 The problem 1. seems to be fixable, although it hasn't been fixed so far. The
160 other one is more serious, but it seems that we can work around it by using
161 hibernation (and suspend) notifiers (in that case, though, we won't be able to
162 avoid the realization by the user space processes that the hibernation is taking
165 There are also problems that the freezing of tasks tends to expose, although
166 they are not directly related to it. For example, if request_firmware() is
167 called from a device driver's .resume() routine, it will timeout and eventually
168 fail, because the user land process that should respond to the request is frozen
169 at this point. So, seemingly, the failure is due to the freezing of tasks.
170 Suppose, however, that the firmware file is located on a filesystem accessible
171 only through another device that hasn't been resumed yet. In that case,
172 request_firmware() will fail regardless of whether or not the freezing of tasks
173 is used. Consequently, the problem is not really related to the freezing of
174 tasks, since it generally exists anyway.
176 A driver must have all firmwares it may need in RAM before suspend() is called.
177 If keeping them is not practical, for example due to their size, they must be
178 requested early enough using the suspend notifier API described in notifiers.txt.