]>
Commit | Line | Data |
---|---|---|
4eacdf18 FW |
1 | /* |
2 | * Context tracking: Probe on high level context boundaries such as kernel | |
3 | * and userspace. This includes syscalls and exceptions entry/exit. | |
4 | * | |
5 | * This is used by RCU to remove its dependency on the timer tick while a CPU | |
6 | * runs in userspace. | |
7 | * | |
8 | * Started by Frederic Weisbecker: | |
9 | * | |
10 | * Copyright (C) 2012 Red Hat, Inc., Frederic Weisbecker <fweisbec@redhat.com> | |
11 | * | |
12 | * Many thanks to Gilad Ben-Yossef, Paul McKenney, Ingo Molnar, Andrew Morton, | |
13 | * Steven Rostedt, Peter Zijlstra for suggestions and improvements. | |
14 | * | |
15 | */ | |
16 | ||
91d1aa43 FW |
17 | #include <linux/context_tracking.h> |
18 | #include <linux/rcupdate.h> | |
19 | #include <linux/sched.h> | |
91d1aa43 | 20 | #include <linux/hardirq.h> |
6a61671b | 21 | #include <linux/export.h> |
91d1aa43 | 22 | |
65f382fd FW |
23 | struct static_key context_tracking_enabled = STATIC_KEY_INIT_FALSE; |
24 | ||
25 | DEFINE_PER_CPU(struct context_tracking, context_tracking); | |
91d1aa43 | 26 | |
2e709338 FW |
27 | void context_tracking_cpu_set(int cpu) |
28 | { | |
65f382fd FW |
29 | if (!per_cpu(context_tracking.active, cpu)) { |
30 | per_cpu(context_tracking.active, cpu) = true; | |
31 | static_key_slow_inc(&context_tracking_enabled); | |
32 | } | |
2e709338 FW |
33 | } |
34 | ||
4eacdf18 FW |
35 | /** |
36 | * user_enter - Inform the context tracking that the CPU is going to | |
37 | * enter userspace mode. | |
38 | * | |
39 | * This function must be called right before we switch from the kernel | |
40 | * to userspace, when it's guaranteed the remaining kernel instructions | |
41 | * to execute won't use any RCU read side critical section because this | |
42 | * function sets RCU in extended quiescent state. | |
43 | */ | |
91d1aa43 FW |
44 | void user_enter(void) |
45 | { | |
46 | unsigned long flags; | |
47 | ||
48 | /* | |
49 | * Some contexts may involve an exception occuring in an irq, | |
50 | * leading to that nesting: | |
51 | * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit() | |
52 | * This would mess up the dyntick_nesting count though. And rcu_irq_*() | |
53 | * helpers are enough to protect RCU uses inside the exception. So | |
54 | * just return immediately if we detect we are in an IRQ. | |
55 | */ | |
56 | if (in_interrupt()) | |
57 | return; | |
58 | ||
4eacdf18 | 59 | /* Kernel threads aren't supposed to go to userspace */ |
91d1aa43 FW |
60 | WARN_ON_ONCE(!current->mm); |
61 | ||
62 | local_irq_save(flags); | |
d65ec121 FW |
63 | if ( __this_cpu_read(context_tracking.state) != IN_USER) { |
64 | if (__this_cpu_read(context_tracking.active)) { | |
65 | /* | |
66 | * At this stage, only low level arch entry code remains and | |
67 | * then we'll run in userspace. We can assume there won't be | |
68 | * any RCU read-side critical section until the next call to | |
69 | * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency | |
70 | * on the tick. | |
71 | */ | |
72 | vtime_user_enter(current); | |
73 | rcu_user_enter(); | |
74 | } | |
4eacdf18 | 75 | /* |
d65ec121 FW |
76 | * Even if context tracking is disabled on this CPU, because it's outside |
77 | * the full dynticks mask for example, we still have to keep track of the | |
78 | * context transitions and states to prevent inconsistency on those of | |
79 | * other CPUs. | |
80 | * If a task triggers an exception in userspace, sleep on the exception | |
81 | * handler and then migrate to another CPU, that new CPU must know where | |
82 | * the exception returns by the time we call exception_exit(). | |
83 | * This information can only be provided by the previous CPU when it called | |
84 | * exception_enter(). | |
85 | * OTOH we can spare the calls to vtime and RCU when context_tracking.active | |
86 | * is false because we know that CPU is not tickless. | |
4eacdf18 | 87 | */ |
abf917cd | 88 | __this_cpu_write(context_tracking.state, IN_USER); |
91d1aa43 FW |
89 | } |
90 | local_irq_restore(flags); | |
91 | } | |
92 | ||
29bb9e5a SR |
93 | #ifdef CONFIG_PREEMPT |
94 | /** | |
95 | * preempt_schedule_context - preempt_schedule called by tracing | |
96 | * | |
97 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
98 | * recursion and tracing preempt enabling caused by the tracing | |
99 | * infrastructure itself. But as tracing can happen in areas coming | |
100 | * from userspace or just about to enter userspace, a preempt enable | |
101 | * can occur before user_exit() is called. This will cause the scheduler | |
102 | * to be called when the system is still in usermode. | |
103 | * | |
104 | * To prevent this, the preempt_enable_notrace will use this function | |
105 | * instead of preempt_schedule() to exit user context if needed before | |
106 | * calling the scheduler. | |
107 | */ | |
108 | void __sched notrace preempt_schedule_context(void) | |
109 | { | |
29bb9e5a SR |
110 | enum ctx_state prev_ctx; |
111 | ||
fbb00b56 | 112 | if (likely(!preemptible())) |
29bb9e5a SR |
113 | return; |
114 | ||
115 | /* | |
116 | * Need to disable preemption in case user_exit() is traced | |
117 | * and the tracer calls preempt_enable_notrace() causing | |
118 | * an infinite recursion. | |
119 | */ | |
120 | preempt_disable_notrace(); | |
121 | prev_ctx = exception_enter(); | |
122 | preempt_enable_no_resched_notrace(); | |
123 | ||
124 | preempt_schedule(); | |
125 | ||
126 | preempt_disable_notrace(); | |
127 | exception_exit(prev_ctx); | |
128 | preempt_enable_notrace(); | |
129 | } | |
130 | EXPORT_SYMBOL_GPL(preempt_schedule_context); | |
131 | #endif /* CONFIG_PREEMPT */ | |
4eacdf18 FW |
132 | |
133 | /** | |
134 | * user_exit - Inform the context tracking that the CPU is | |
135 | * exiting userspace mode and entering the kernel. | |
136 | * | |
137 | * This function must be called after we entered the kernel from userspace | |
138 | * before any use of RCU read side critical section. This potentially include | |
139 | * any high level kernel code like syscalls, exceptions, signal handling, etc... | |
140 | * | |
141 | * This call supports re-entrancy. This way it can be called from any exception | |
142 | * handler without needing to know if we came from userspace or not. | |
143 | */ | |
91d1aa43 FW |
144 | void user_exit(void) |
145 | { | |
146 | unsigned long flags; | |
147 | ||
91d1aa43 FW |
148 | if (in_interrupt()) |
149 | return; | |
150 | ||
151 | local_irq_save(flags); | |
152 | if (__this_cpu_read(context_tracking.state) == IN_USER) { | |
d65ec121 FW |
153 | if (__this_cpu_read(context_tracking.active)) { |
154 | /* | |
155 | * We are going to run code that may use RCU. Inform | |
156 | * RCU core about that (ie: we may need the tick again). | |
157 | */ | |
158 | rcu_user_exit(); | |
159 | vtime_user_exit(current); | |
160 | } | |
abf917cd | 161 | __this_cpu_write(context_tracking.state, IN_KERNEL); |
91d1aa43 FW |
162 | } |
163 | local_irq_restore(flags); | |
164 | } | |
165 | ||
2d854e57 | 166 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN |
6a61671b FW |
167 | void guest_enter(void) |
168 | { | |
169 | if (vtime_accounting_enabled()) | |
170 | vtime_guest_enter(current); | |
171 | else | |
2d854e57 | 172 | current->flags |= PF_VCPU; |
6a61671b FW |
173 | } |
174 | EXPORT_SYMBOL_GPL(guest_enter); | |
175 | ||
176 | void guest_exit(void) | |
177 | { | |
178 | if (vtime_accounting_enabled()) | |
179 | vtime_guest_exit(current); | |
180 | else | |
2d854e57 | 181 | current->flags &= ~PF_VCPU; |
6a61671b FW |
182 | } |
183 | EXPORT_SYMBOL_GPL(guest_exit); | |
2d854e57 | 184 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |
6a61671b | 185 | |
4eacdf18 FW |
186 | |
187 | /** | |
188 | * context_tracking_task_switch - context switch the syscall callbacks | |
189 | * @prev: the task that is being switched out | |
190 | * @next: the task that is being switched in | |
191 | * | |
192 | * The context tracking uses the syscall slow path to implement its user-kernel | |
193 | * boundaries probes on syscalls. This way it doesn't impact the syscall fast | |
194 | * path on CPUs that don't do context tracking. | |
195 | * | |
196 | * But we need to clear the flag on the previous task because it may later | |
197 | * migrate to some CPU that doesn't do the context tracking. As such the TIF | |
198 | * flag may not be desired there. | |
199 | */ | |
91d1aa43 FW |
200 | void context_tracking_task_switch(struct task_struct *prev, |
201 | struct task_struct *next) | |
202 | { | |
d65ec121 FW |
203 | clear_tsk_thread_flag(prev, TIF_NOHZ); |
204 | set_tsk_thread_flag(next, TIF_NOHZ); | |
91d1aa43 | 205 | } |
65f382fd FW |
206 | |
207 | #ifdef CONFIG_CONTEXT_TRACKING_FORCE | |
208 | void __init context_tracking_init(void) | |
209 | { | |
210 | int cpu; | |
211 | ||
212 | for_each_possible_cpu(cpu) | |
213 | context_tracking_cpu_set(cpu); | |
214 | } | |
215 | #endif |