Commit | Line | Data |
---|---|---|
83144186 RW |
1 | Freezing of tasks |
2 | (C) 2007 Rafael J. Wysocki <rjw@sisk.pl>, GPL | |
3 | ||
4 | I. What is the freezing of tasks? | |
5 | ||
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 | |
8 | architectures). | |
9 | ||
10 | II. How does it work? | |
11 | ||
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). | |
18 | ||
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 | sends a fake signal to each of them. A task that receives such a signal and has | |
23 | TIF_FREEZE set, should react to it by calling the refrigerator() function | |
24 | (defined in kernel/power/process.c), which sets the task's PF_FROZEN flag, | |
25 | changes its state to TASK_UNINTERRUPTIBLE and makes it loop until PF_FROZEN is | |
26 | cleared for it. Then, we say that the task is 'frozen' and therefore the set of | |
27 | functions handling this mechanism is called 'the freezer' (these functions are | |
28 | defined in kernel/power/process.c and include/linux/freezer.h). User space | |
29 | processes are generally frozen before kernel threads. | |
30 | ||
31 | It is not recommended to call refrigerator() directly. Instead, it is | |
32 | recommended to use the try_to_freeze() function (defined in | |
33 | include/linux/freezer.h), that checks the task's TIF_FREEZE flag and makes the | |
34 | task enter refrigerator() if the flag is set. | |
35 | ||
36 | For user space processes try_to_freeze() is called automatically from the | |
37 | signal-handling code, but the freezable kernel threads need to call it | |
38 | explicitly in suitable places. The code to do this may look like the following: | |
39 | ||
40 | do { | |
41 | hub_events(); | |
42 | wait_event_interruptible(khubd_wait, | |
43 | !list_empty(&hub_event_list)); | |
44 | try_to_freeze(); | |
45 | } while (!signal_pending(current)); | |
46 | ||
47 | (from drivers/usb/core/hub.c::hub_thread()). | |
48 | ||
49 | If a freezable kernel thread fails to call try_to_freeze() after the freezer has | |
50 | set TIF_FREEZE for it, the freezing of tasks will fail and the entire | |
51 | hibernation operation will be cancelled. For this reason, freezable kernel | |
52 | threads must call try_to_freeze() somewhere. | |
53 | ||
54 | After the system memory state has been restored from a hibernation image and | |
55 | devices have been reinitialized, the function thaw_processes() is called in | |
56 | order to clear the PF_FROZEN flag for each frozen task. Then, the tasks that | |
57 | have been frozen leave refrigerator() and continue running. | |
58 | ||
59 | III. Which kernel threads are freezable? | |
60 | ||
61 | Kernel threads are not freezable by default. However, a kernel thread may clear | |
62 | PF_NOFREEZE for itself by calling set_freezable() (the resetting of PF_NOFREEZE | |
63 | directly is strongly discouraged). From this point it is regarded as freezable | |
64 | and must call try_to_freeze() in a suitable place. | |
65 | ||
66 | IV. Why do we do that? | |
67 | ||
68 | Generally speaking, there is a couple of reasons to use the freezing of tasks: | |
69 | ||
70 | 1. The principal reason is to prevent filesystems from being damaged after | |
71 | hibernation. At the moment we have no simple means of checkpointing | |
72 | filesystems, so if there are any modifications made to filesystem data and/or | |
73 | metadata on disks, we cannot bring them back to the state from before the | |
74 | modifications. At the same time each hibernation image contains some | |
75 | filesystem-related information that must be consistent with the state of the | |
76 | on-disk data and metadata after the system memory state has been restored from | |
77 | the image (otherwise the filesystems will be damaged in a nasty way, usually | |
78 | making them almost impossible to repair). We therefore freeze tasks that might | |
79 | cause the on-disk filesystems' data and metadata to be modified after the | |
80 | hibernation image has been created and before the system is finally powered off. | |
81 | The majority of these are user space processes, but if any of the kernel threads | |
82 | may cause something like this to happen, they have to be freezable. | |
83 | ||
84 | 2. The second reason is to prevent user space processes and some kernel threads | |
85 | from interfering with the suspending and resuming of devices. A user space | |
86 | process running on a second CPU while we are suspending devices may, for | |
87 | example, be troublesome and without the freezing of tasks we would need some | |
88 | safeguards against race conditions that might occur in such a case. | |
89 | ||
90 | Although Linus Torvalds doesn't like the freezing of tasks, he said this in one | |
91 | of the discussions on LKML (http://lkml.org/lkml/2007/4/27/608): | |
92 | ||
93 | "RJW:> Why we freeze tasks at all or why we freeze kernel threads? | |
94 | ||
95 | Linus: In many ways, 'at all'. | |
96 | ||
97 | I _do_ realize the IO request queue issues, and that we cannot actually do | |
98 | s2ram with some devices in the middle of a DMA. So we want to be able to | |
99 | avoid *that*, there's no question about that. And I suspect that stopping | |
100 | user threads and then waiting for a sync is practically one of the easier | |
101 | ways to do so. | |
102 | ||
103 | So in practice, the 'at all' may become a 'why freeze kernel threads?' and | |
104 | freezing user threads I don't find really objectionable." | |
105 | ||
106 | Still, there are kernel threads that may want to be freezable. For example, if | |
107 | a kernel that belongs to a device driver accesses the device directly, it in | |
108 | principle needs to know when the device is suspended, so that it doesn't try to | |
109 | access it at that time. However, if the kernel thread is freezable, it will be | |
110 | frozen before the driver's .suspend() callback is executed and it will be | |
111 | thawed after the driver's .resume() callback has run, so it won't be accessing | |
112 | the device while it's suspended. | |
113 | ||
114 | 3. Another reason for freezing tasks is to prevent user space processes from | |
115 | realizing that hibernation (or suspend) operation takes place. Ideally, user | |
116 | space processes should not notice that such a system-wide operation has occurred | |
117 | and should continue running without any problems after the restore (or resume | |
118 | from suspend). Unfortunately, in the most general case this is quite difficult | |
119 | to achieve without the freezing of tasks. Consider, for example, a process | |
120 | that depends on all CPUs being online while it's running. Since we need to | |
121 | disable nonboot CPUs during the hibernation, if this process is not frozen, it | |
122 | may notice that the number of CPUs has changed and may start to work incorrectly | |
123 | because of that. | |
124 | ||
125 | V. Are there any problems related to the freezing of tasks? | |
126 | ||
127 | Yes, there are. | |
128 | ||
129 | First of all, the freezing of kernel threads may be tricky if they depend one | |
130 | on another. For example, if kernel thread A waits for a completion (in the | |
131 | TASK_UNINTERRUPTIBLE state) that needs to be done by freezable kernel thread B | |
132 | and B is frozen in the meantime, then A will be blocked until B is thawed, which | |
133 | may be undesirable. That's why kernel threads are not freezable by default. | |
134 | ||
135 | Second, there are the following two problems related to the freezing of user | |
136 | space processes: | |
137 | 1. Putting processes into an uninterruptible sleep distorts the load average. | |
138 | 2. Now that we have FUSE, plus the framework for doing device drivers in | |
139 | userspace, it gets even more complicated because some userspace processes are | |
140 | now doing the sorts of things that kernel threads do | |
141 | (https://lists.linux-foundation.org/pipermail/linux-pm/2007-May/012309.html). | |
142 | ||
143 | The problem 1. seems to be fixable, although it hasn't been fixed so far. The | |
144 | other one is more serious, but it seems that we can work around it by using | |
145 | hibernation (and suspend) notifiers (in that case, though, we won't be able to | |
146 | avoid the realization by the user space processes that the hibernation is taking | |
147 | place). | |
148 | ||
149 | There are also problems that the freezing of tasks tends to expose, although | |
150 | they are not directly related to it. For example, if request_firmware() is | |
151 | called from a device driver's .resume() routine, it will timeout and eventually | |
152 | fail, because the user land process that should respond to the request is frozen | |
153 | at this point. So, seemingly, the failure is due to the freezing of tasks. | |
154 | Suppose, however, that the firmware file is located on a filesystem accessible | |
155 | only through another device that hasn't been resumed yet. In that case, | |
156 | request_firmware() will fail regardless of whether or not the freezing of tasks | |
157 | is used. Consequently, the problem is not really related to the freezing of | |
fccdb5ae ON |
158 | tasks, since it generally exists anyway. |
159 | ||
160 | A driver must have all firmwares it may need in RAM before suspend() is called. | |
161 | If keeping them is not practical, for example due to their size, they must be | |
162 | requested early enough using the suspend notifier API described in notifiers.txt. |