1Freezing of tasks
2	(C) 2007 Rafael J. Wysocki <rjw@sisk.pl>, GPL
3
4I. What is the freezing of tasks?
5
6The freezing of tasks is a mechanism by which user space processes and some
7kernel threads are controlled during hibernation or system-wide suspend (on some
8architectures).
9
10II. How does it work?
11
12There are three per-task flags used for that, PF_NOFREEZE, PF_FROZEN
13and PF_FREEZER_SKIP (the last one is auxiliary).  The tasks that have
14PF_NOFREEZE unset (all user space processes and some kernel threads) are
15regarded as 'freezable' and treated in a special way before the system enters a
16suspend state as well as before a hibernation image is created (in what follows
17we only consider hibernation, but the description also applies to suspend).
18
19Namely, as the first step of the hibernation procedure the function
20freeze_processes() (defined in kernel/power/process.c) is called.  A system-wide
21variable system_freezing_cnt (as opposed to a per-task flag) is used to indicate
22whether the system is to undergo a freezing operation. And freeze_processes()
23sets this variable.  After this, it executes try_to_freeze_tasks() that sends a
24fake signal to all user space processes, and wakes up all the kernel threads.
25All freezable tasks must react to that by calling try_to_freeze(), which
26results in a call to __refrigerator() (defined in kernel/freezer.c), which sets
27the task's PF_FROZEN flag, changes its state to TASK_UNINTERRUPTIBLE and makes
28it 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
30to as 'the freezer' (these functions are defined in kernel/power/process.c,
31kernel/freezer.c & include/linux/freezer.h). User space processes are generally
32frozen before kernel threads.
33
34__refrigerator() must not be called directly.  Instead, use the
35try_to_freeze() function (defined in include/linux/freezer.h), that checks
36if the task is to be frozen and makes the task enter __refrigerator().
37
38For user space processes try_to_freeze() is called automatically from the
39signal-handling code, but the freezable kernel threads need to call it
40explicitly in suitable places or use the wait_event_freezable() or
41wait_event_freezable_timeout() macros (defined in include/linux/freezer.h)
42that combine interruptible sleep with checking if the task is to be frozen and
43calling try_to_freeze().  The main loop of a freezable kernel thread may look
44like the following one:
45
46	set_freezable();
47	do {
48		hub_events();
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));
53
54(from drivers/usb/core/hub.c::hub_thread()).
55
56If a freezable kernel thread fails to call try_to_freeze() after the freezer has
57initiated a freezing operation, the freezing of tasks will fail and the entire
58hibernation operation will be cancelled.  For this reason, freezable kernel
59threads must call try_to_freeze() somewhere or use one of the
60wait_event_freezable() and wait_event_freezable_timeout() macros.
61
62After the system memory state has been restored from a hibernation image and
63devices have been reinitialized, the function thaw_processes() is called in
64order to clear the PF_FROZEN flag for each frozen task.  Then, the tasks that
65have been frozen leave __refrigerator() and continue running.
66
67
68Rationale behind the functions dealing with freezing and thawing of tasks:
69-------------------------------------------------------------------------
70
71freeze_processes():
72  - freezes only userspace tasks
73
74freeze_kernel_threads():
75  - freezes all tasks (including kernel threads) because we can't freeze
76    kernel threads without freezing userspace tasks
77
78thaw_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
82
83thaw_processes():
84  - thaws all tasks (including kernel threads) because we can't thaw userspace
85    tasks without thawing kernel threads
86
87
88III. Which kernel threads are freezable?
89
90Kernel threads are not freezable by default.  However, a kernel thread may clear
91PF_NOFREEZE for itself by calling set_freezable() (the resetting of PF_NOFREEZE
92directly is not allowed).  From this point it is regarded as freezable
93and must call try_to_freeze() in a suitable place.
94
95IV. Why do we do that?
96
97Generally speaking, there is a couple of reasons to use the freezing of tasks:
98
991. The principal reason is to prevent filesystems from being damaged after
100hibernation.  At the moment we have no simple means of checkpointing
101filesystems, so if there are any modifications made to filesystem data and/or
102metadata on disks, we cannot bring them back to the state from before the
103modifications.  At the same time each hibernation image contains some
104filesystem-related information that must be consistent with the state of the
105on-disk data and metadata after the system memory state has been restored from
106the image (otherwise the filesystems will be damaged in a nasty way, usually
107making them almost impossible to repair).  We therefore freeze tasks that might
108cause the on-disk filesystems' data and metadata to be modified after the
109hibernation image has been created and before the system is finally powered off.
110The majority of these are user space processes, but if any of the kernel threads
111may cause something like this to happen, they have to be freezable.
112
1132. Next, to create the hibernation image we need to free a sufficient amount of
114memory (approximately 50% of available RAM) and we need to do that before
115devices are deactivated, because we generally need them for swapping out.  Then,
116after the memory for the image has been freed, we don't want tasks to allocate
117additional 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
119amounts of memory from their .suspend() callbacks before hibernation, but this
120is a separate issue.]
121
1223. The third reason is to prevent user space processes and some kernel threads
123from interfering with the suspending and resuming of devices.  A user space
124process running on a second CPU while we are suspending devices may, for
125example, be troublesome and without the freezing of tasks we would need some
126safeguards against race conditions that might occur in such a case.
127
128Although Linus Torvalds doesn't like the freezing of tasks, he said this in one
129of the discussions on LKML (http://lkml.org/lkml/2007/4/27/608):
130
131"RJW:> Why we freeze tasks at all or why we freeze kernel threads?
132
133Linus: In many ways, 'at all'.
134
135I _do_ realize the IO request queue issues, and that we cannot actually do
136s2ram with some devices in the middle of a DMA.  So we want to be able to
137avoid *that*, there's no question about that.  And I suspect that stopping
138user threads and then waiting for a sync is practically one of the easier
139ways to do so.
140
141So in practice, the 'at all' may become a 'why freeze kernel threads?' and
142freezing user threads I don't find really objectionable."
143
144Still, there are kernel threads that may want to be freezable.  For example, if
145a kernel thread that belongs to a device driver accesses the device directly, it
146in principle needs to know when the device is suspended, so that it doesn't try
147to access it at that time.  However, if the kernel thread is freezable, it will
148be frozen before the driver's .suspend() callback is executed and it will be
149thawed after the driver's .resume() callback has run, so it won't be accessing
150the device while it's suspended.
151
1524. Another reason for freezing tasks is to prevent user space processes from
153realizing that hibernation (or suspend) operation takes place.  Ideally, user
154space processes should not notice that such a system-wide operation has occurred
155and should continue running without any problems after the restore (or resume
156from suspend).  Unfortunately, in the most general case this is quite difficult
157to achieve without the freezing of tasks.  Consider, for example, a process
158that depends on all CPUs being online while it's running.  Since we need to
159disable nonboot CPUs during the hibernation, if this process is not frozen, it
160may notice that the number of CPUs has changed and may start to work incorrectly
161because of that.
162
163V. Are there any problems related to the freezing of tasks?
164
165Yes, there are.
166
167First of all, the freezing of kernel threads may be tricky if they depend one
168on another.  For example, if kernel thread A waits for a completion (in the
169TASK_UNINTERRUPTIBLE state) that needs to be done by freezable kernel thread B
170and B is frozen in the meantime, then A will be blocked until B is thawed, which
171may be undesirable.  That's why kernel threads are not freezable by default.
172
173Second, there are the following two problems related to the freezing of user
174space processes:
1751. Putting processes into an uninterruptible sleep distorts the load average.
1762. Now that we have FUSE, plus the framework for doing device drivers in
177userspace, it gets even more complicated because some userspace processes are
178now doing the sorts of things that kernel threads do
179(https://lists.linux-foundation.org/pipermail/linux-pm/2007-May/012309.html).
180
181The problem 1. seems to be fixable, although it hasn't been fixed so far.  The
182other one is more serious, but it seems that we can work around it by using
183hibernation (and suspend) notifiers (in that case, though, we won't be able to
184avoid the realization by the user space processes that the hibernation is taking
185place).
186
187There are also problems that the freezing of tasks tends to expose, although
188they are not directly related to it.  For example, if request_firmware() is
189called from a device driver's .resume() routine, it will timeout and eventually
190fail, because the user land process that should respond to the request is frozen
191at this point.  So, seemingly, the failure is due to the freezing of tasks.
192Suppose, however, that the firmware file is located on a filesystem accessible
193only through another device that hasn't been resumed yet.  In that case,
194request_firmware() will fail regardless of whether or not the freezing of tasks
195is used.  Consequently, the problem is not really related to the freezing of
196tasks, since it generally exists anyway.
197
198A driver must have all firmwares it may need in RAM before suspend() is called.
199If keeping them is not practical, for example due to their size, they must be
200requested early enough using the suspend notifier API described in notifiers.txt.
201
202VI. Are there any precautions to be taken to prevent freezing failures?
203
204Yes, there are.
205
206First of all, grabbing the 'pm_mutex' lock to mutually exclude a piece of code
207from system-wide sleep such as suspend/hibernation is not encouraged.
208If possible, that piece of code must instead hook onto the suspend/hibernation
209notifiers to achieve mutual exclusion. Look at the CPU-Hotplug code
210(kernel/cpu.c) for an example.
211
212However, if that is not feasible, and grabbing 'pm_mutex' is deemed necessary,
213it is strongly discouraged to directly call mutex_[un]lock(&pm_mutex) since
214that could lead to freezing failures, because if the suspend/hibernate code
215successfully acquired the 'pm_mutex' lock, and hence that other entity failed
216to acquire the lock, then that task would get blocked in TASK_UNINTERRUPTIBLE
217state. As a consequence, the freezer would not be able to freeze that task,
218leading to freezing failure.
219
220However, the [un]lock_system_sleep() APIs are safe to use in this scenario,
221since they ask the freezer to skip freezing this task, since it is anyway
222"frozen enough" as it is blocked on 'pm_mutex', which will be released
223only after the entire suspend/hibernation sequence is complete.
224So, to summarize, use [un]lock_system_sleep() instead of directly using
225mutex_[un]lock(&pm_mutex). That would prevent freezing failures.
226
227V. Miscellaneous
228/sys/power/pm_freeze_timeout controls how long it will cost at most to freeze
229all user space processes or all freezable kernel threads, in unit of millisecond.
230The default value is 20000, with range of unsigned integer.
231