1 =================
2 Freezing of tasks
3 =================
4
5 (C) 2007 Rafael J. Wysocki <rjw@sisk.pl>, GPL
6
7 I. What is the freezing of tasks?
8 =================================
9
10 The freezing of tasks is a mechanism by which user space processes and some
11 kernel threads are controlled during hibernation or system-wide suspend (on some
12 architectures).
13
14 II. How does it work?
15 =====================
16
17 There is one per-task flag (PF_NOFREEZE) and three per-task states
18 (TASK_FROZEN, TASK_FREEZABLE and __TASK_FREEZABLE_UNSAFE) used for that.
19 The tasks that have PF_NOFREEZE unset (all user space tasks and some kernel
20 threads) are regarded as 'freezable' and treated in a special way before the
21 system enters a sleep state as well as before a hibernation image is created
22 (hibernation is directly covered by what follows, but the description applies
23 to system-wide suspend too).
24
25 Namely, as the first step of the hibernation procedure the function
26 freeze_processes() (defined in kernel/power/process.c) is called. A system-wide
27 static key freezer_active (as opposed to a per-task flag or state) is used to
28 indicate whether the system is to undergo a freezing operation. And
29 freeze_processes() sets this static key. After this, it executes
30 try_to_freeze_tasks() that sends a fake signal to all user space processes, and
31 wakes up all the kernel threads. All freezable tasks must react to that by
32 calling try_to_freeze(), which results in a call to __refrigerator() (defined
33 in kernel/freezer.c), which changes the task's state to TASK_FROZEN, and makes
34 it loop until it is woken by an explicit TASK_FROZEN wakeup. Then, that task
35 is regarded as 'frozen' and so the set of functions handling this mechanism is
36 referred to as 'the freezer' (these functions are defined in
37 kernel/power/process.c, kernel/freezer.c & include/linux/freezer.h). User space
38 tasks are generally frozen before kernel threads.
39
40 __refrigerator() must not be called directly. Instead, use the
41 try_to_freeze() function (defined in include/linux/freezer.h), that checks
42 if the task is to be frozen and makes the task enter __refrigerator().
43
44 For user space processes try_to_freeze() is called automatically from the
45 signal-handling code, but the freezable kernel threads need to call it
46 explicitly in suitable places or use the wait_event_freezable() or
47 wait_event_freezable_timeout() macros (defined in include/linux/wait.h)
48 that put the task to sleep (TASK_INTERRUPTIBLE) or freeze it (TASK_FROZEN) if
49 freezer_active is set. The main loop of a freezable kernel thread may look
50 like the following one::
51
52 set_freezable();
53
54 while (true) {
55 struct task_struct *tsk = NULL;
56
57 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
58 spin_lock_irq(&oom_reaper_lock);
59 if (oom_reaper_list != NULL) {
60 tsk = oom_reaper_list;
61 oom_reaper_list = tsk->oom_reaper_list;
62 }
63 spin_unlock_irq(&oom_reaper_lock);
64
65 if (tsk)
66 oom_reap_task(tsk);
67 }
68
69 (from mm/oom_kill.c::oom_reaper()).
70
71 If a freezable kernel thread is not put to the frozen state after the freezer
72 has initiated a freezing operation, the freezing of tasks will fail and the
73 entire system-wide transition will be cancelled. For this reason, freezable
74 kernel threads must call try_to_freeze() somewhere or use one of the
75 wait_event_freezable() and wait_event_freezable_timeout() macros.
76
77 After the system memory state has been restored from a hibernation image and
78 devices have been reinitialized, the function thaw_processes() is called in
79 order to wake up each frozen task. Then, the tasks that have been frozen leave
80 __refrigerator() and continue running.
81
82
83 Rationale behind the functions dealing with freezing and thawing of tasks
84 -------------------------------------------------------------------------
85
86 freeze_processes():
87 - freezes only userspace tasks
88
89 freeze_kernel_threads():
90 - freezes all tasks (including kernel threads) because we can't freeze
91 kernel threads without freezing userspace tasks
92
93 thaw_kernel_threads():
94 - thaws only kernel threads; this is particularly useful if we need to do
95 anything special in between thawing of kernel threads and thawing of
96 userspace tasks, or if we want to postpone the thawing of userspace tasks
97
98 thaw_processes():
99 - thaws all tasks (including kernel threads) because we can't thaw userspace
100 tasks without thawing kernel threads
101
102
103 III. Which kernel threads are freezable?
104 ========================================
105
106 Kernel threads are not freezable by default. However, a kernel thread may clear
107 PF_NOFREEZE for itself by calling set_freezable() (the resetting of PF_NOFREEZE
108 directly is not allowed). From this point it is regarded as freezable
109 and must call try_to_freeze() or variants of wait_event_freezable() in a
110 suitable place.
111
112 IV. Why do we do that?
113 ======================
114
115 Generally speaking, there is a couple of reasons to use the freezing of tasks:
116
117 1. The principal reason is to prevent filesystems from being damaged after
118 hibernation. At the moment we have no simple means of checkpointing
119 filesystems, so if there are any modifications made to filesystem data and/or
120 metadata on disks, we cannot bring them back to the state from before the
121 modifications. At the same time each hibernation image contains some
122 filesystem-related information that must be consistent with the state of the
123 on-disk data and metadata after the system memory state has been restored
124 from the image (otherwise the filesystems will be damaged in a nasty way,
125 usually making them almost impossible to repair). We therefore freeze
126 tasks that might cause the on-disk filesystems' data and metadata to be
127 modified after the hibernation image has been created and before the
128 system is finally powered off. The majority of these are user space
129 processes, but if any of the kernel threads may cause something like this
130 to happen, they have to be freezable.
131
132 2. Next, to create the hibernation image we need to free a sufficient amount of
133 memory (approximately 50% of available RAM) and we need to do that before
134 devices are deactivated, because we generally need them for swapping out.
135 Then, after the memory for the image has been freed, we don't want tasks
136 to allocate additional memory and we prevent them from doing that by
137 freezing them earlier. [Of course, this also means that device drivers
138 should not allocate substantial amounts of memory from their .suspend()
139 callbacks before hibernation, but this is a separate issue.]
140
141 3. The third reason is to prevent user space processes and some kernel threads
142 from interfering with the suspending and resuming of devices. A user space
143 process running on a second CPU while we are suspending devices may, for
144 example, be troublesome and without the freezing of tasks we would need some
145 safeguards against race conditions that might occur in such a case.
146
147 Although Linus Torvalds doesn't like the freezing of tasks, he said this in one
148 of the discussions on LKML (https://lore.kernel.org/r/alpine.LFD.0.98.0704271801020.9964@woody.linux-foundation.org):
149
150 "RJW:> Why we freeze tasks at all or why we freeze kernel threads?
151
152 Linus: In many ways, 'at all'.
153
154 I **do** realize the IO request queue issues, and that we cannot actually do
155 s2ram with some devices in the middle of a DMA. So we want to be able to
156 avoid *that*, there's no question about that. And I suspect that stopping
157 user threads and then waiting for a sync is practically one of the easier
158 ways to do so.
159
160 So in practice, the 'at all' may become a 'why freeze kernel threads?' and
161 freezing user threads I don't find really objectionable."
162
163 Still, there are kernel threads that may want to be freezable. For example, if
164 a kernel thread that belongs to a device driver accesses the device directly, it
165 in principle needs to know when the device is suspended, so that it doesn't try
166 to access it at that time. However, if the kernel thread is freezable, it will
167 be frozen before the driver's .suspend() callback is executed and it will be
168 thawed after the driver's .resume() callback has run, so it won't be accessing
169 the device while it's suspended.
170
171 4. Another reason for freezing tasks is to prevent user space processes from
172 realizing that hibernation (or suspend) operation takes place. Ideally, user
173 space processes should not notice that such a system-wide operation has
174 occurred and should continue running without any problems after the restore
175 (or resume from suspend). Unfortunately, in the most general case this
176 is quite difficult to achieve without the freezing of tasks. Consider,
177 for example, a process that depends on all CPUs being online while it's
178 running. Since we need to disable nonboot CPUs during the hibernation,
179 if this process is not frozen, it may notice that the number of CPUs has
180 changed and may start to work incorrectly because of that.
181
182 V. Are there any problems related to the freezing of tasks?
183 ===========================================================
184
185 Yes, there are.
186
187 First of all, the freezing of kernel threads may be tricky if they depend one
188 on another. For example, if kernel thread A waits for a completion (in the
189 TASK_UNINTERRUPTIBLE state) that needs to be done by freezable kernel thread B
190 and B is frozen in the meantime, then A will be blocked until B is thawed, which
191 may be undesirable. That's why kernel threads are not freezable by default.
192
193 Second, there are the following two problems related to the freezing of user
194 space processes:
195
196 1. Putting processes into an uninterruptible sleep distorts the load average.
197 2. Now that we have FUSE, plus the framework for doing device drivers in
198 userspace, it gets even more complicated because some userspace processes are
199 now doing the sorts of things that kernel threads do
200 (https://lists.linux-foundation.org/pipermail/linux-pm/2007-May/012309.html).
201
202 The problem 1. seems to be fixable, although it hasn't been fixed so far. The
203 other one is more serious, but it seems that we can work around it by using
204 hibernation (and suspend) notifiers (in that case, though, we won't be able to
205 avoid the realization by the user space processes that the hibernation is taking
206 place).
207
208 There are also problems that the freezing of tasks tends to expose, although
209 they are not directly related to it. For example, if request_firmware() is
210 called from a device driver's .resume() routine, it will timeout and eventually
211 fail, because the user land process that should respond to the request is frozen
212 at this point. So, seemingly, the failure is due to the freezing of tasks.
213 Suppose, however, that the firmware file is located on a filesystem accessible
214 only through another device that hasn't been resumed yet. In that case,
215 request_firmware() will fail regardless of whether or not the freezing of tasks
216 is used. Consequently, the problem is not really related to the freezing of
217 tasks, since it generally exists anyway.
218
219 A driver must have all firmwares it may need in RAM before suspend() is called.
220 If keeping them is not practical, for example due to their size, they must be
221 requested early enough using the suspend notifier API described in
222 Documentation/driver-api/pm/notifiers.rst.
223
224 VI. Are there any precautions to be taken to prevent freezing failures?
225 =======================================================================
226
227 Yes, there are.
228
229 First of all, grabbing the 'system_transition_mutex' lock to mutually exclude a
230 piece of code from system-wide sleep such as suspend/hibernation is not
231 encouraged. If possible, that piece of code must instead hook onto the
232 suspend/hibernation notifiers to achieve mutual exclusion. Look at the
233 CPU-Hotplug code (kernel/cpu.c) for an example.
234
235 However, if that is not feasible, and grabbing 'system_transition_mutex' is
236 deemed necessary, it is strongly discouraged to directly call
237 mutex_[un]lock(&system_transition_mutex) since that could lead to freezing
238 failures, because if the suspend/hibernate code successfully acquired the
239 'system_transition_mutex' lock, and hence that other entity failed to acquire
240 the lock, then that task would get blocked in TASK_UNINTERRUPTIBLE state. As a
241 consequence, the freezer would not be able to freeze that task, leading to
242 freezing failure.
243
244 However, the [un]lock_system_sleep() APIs are safe to use in this scenario,
245 since they ask the freezer to skip freezing this task, since it is anyway
246 "frozen enough" as it is blocked on 'system_transition_mutex', which will be
247 released only after the entire suspend/hibernation sequence is complete. So, to
248 summarize, use [un]lock_system_sleep() instead of directly using
249 mutex_[un]lock(&system_transition_mutex). That would prevent freezing failures.
250
251 V. Miscellaneous
252 ================
253
254 /sys/power/pm_freeze_timeout controls how long it will cost at most to freeze
255 all user space processes or all freezable kernel threads, in unit of
256 millisecond. The default value is 20000, with range of unsigned integer.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.