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 | |
95a79fd4 | 23 | DEFINE_PER_CPU(struct context_tracking, context_tracking) = { |
91d1aa43 FW |
24 | #ifdef CONFIG_CONTEXT_TRACKING_FORCE |
25 | .active = true, | |
26 | #endif | |
27 | }; | |
28 | ||
2e709338 FW |
29 | void context_tracking_cpu_set(int cpu) |
30 | { | |
31 | per_cpu(context_tracking.active, cpu) = true; | |
32 | } | |
33 | ||
4eacdf18 FW |
34 | /** |
35 | * user_enter - Inform the context tracking that the CPU is going to | |
36 | * enter userspace mode. | |
37 | * | |
38 | * This function must be called right before we switch from the kernel | |
39 | * to userspace, when it's guaranteed the remaining kernel instructions | |
40 | * to execute won't use any RCU read side critical section because this | |
41 | * function sets RCU in extended quiescent state. | |
42 | */ | |
91d1aa43 FW |
43 | void user_enter(void) |
44 | { | |
45 | unsigned long flags; | |
46 | ||
47 | /* | |
48 | * Some contexts may involve an exception occuring in an irq, | |
49 | * leading to that nesting: | |
50 | * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit() | |
51 | * This would mess up the dyntick_nesting count though. And rcu_irq_*() | |
52 | * helpers are enough to protect RCU uses inside the exception. So | |
53 | * just return immediately if we detect we are in an IRQ. | |
54 | */ | |
55 | if (in_interrupt()) | |
56 | return; | |
57 | ||
4eacdf18 | 58 | /* Kernel threads aren't supposed to go to userspace */ |
91d1aa43 FW |
59 | WARN_ON_ONCE(!current->mm); |
60 | ||
61 | local_irq_save(flags); | |
d65ec121 FW |
62 | if ( __this_cpu_read(context_tracking.state) != IN_USER) { |
63 | if (__this_cpu_read(context_tracking.active)) { | |
64 | /* | |
65 | * At this stage, only low level arch entry code remains and | |
66 | * then we'll run in userspace. We can assume there won't be | |
67 | * any RCU read-side critical section until the next call to | |
68 | * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency | |
69 | * on the tick. | |
70 | */ | |
71 | vtime_user_enter(current); | |
72 | rcu_user_enter(); | |
73 | } | |
4eacdf18 | 74 | /* |
d65ec121 FW |
75 | * Even if context tracking is disabled on this CPU, because it's outside |
76 | * the full dynticks mask for example, we still have to keep track of the | |
77 | * context transitions and states to prevent inconsistency on those of | |
78 | * other CPUs. | |
79 | * If a task triggers an exception in userspace, sleep on the exception | |
80 | * handler and then migrate to another CPU, that new CPU must know where | |
81 | * the exception returns by the time we call exception_exit(). | |
82 | * This information can only be provided by the previous CPU when it called | |
83 | * exception_enter(). | |
84 | * OTOH we can spare the calls to vtime and RCU when context_tracking.active | |
85 | * is false because we know that CPU is not tickless. | |
4eacdf18 | 86 | */ |
abf917cd | 87 | __this_cpu_write(context_tracking.state, IN_USER); |
91d1aa43 FW |
88 | } |
89 | local_irq_restore(flags); | |
90 | } | |
91 | ||
29bb9e5a SR |
92 | #ifdef CONFIG_PREEMPT |
93 | /** | |
94 | * preempt_schedule_context - preempt_schedule called by tracing | |
95 | * | |
96 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
97 | * recursion and tracing preempt enabling caused by the tracing | |
98 | * infrastructure itself. But as tracing can happen in areas coming | |
99 | * from userspace or just about to enter userspace, a preempt enable | |
100 | * can occur before user_exit() is called. This will cause the scheduler | |
101 | * to be called when the system is still in usermode. | |
102 | * | |
103 | * To prevent this, the preempt_enable_notrace will use this function | |
104 | * instead of preempt_schedule() to exit user context if needed before | |
105 | * calling the scheduler. | |
106 | */ | |
107 | void __sched notrace preempt_schedule_context(void) | |
108 | { | |
29bb9e5a SR |
109 | enum ctx_state prev_ctx; |
110 | ||
fbb00b56 | 111 | if (likely(!preemptible())) |
29bb9e5a SR |
112 | return; |
113 | ||
114 | /* | |
115 | * Need to disable preemption in case user_exit() is traced | |
116 | * and the tracer calls preempt_enable_notrace() causing | |
117 | * an infinite recursion. | |
118 | */ | |
119 | preempt_disable_notrace(); | |
120 | prev_ctx = exception_enter(); | |
121 | preempt_enable_no_resched_notrace(); | |
122 | ||
123 | preempt_schedule(); | |
124 | ||
125 | preempt_disable_notrace(); | |
126 | exception_exit(prev_ctx); | |
127 | preempt_enable_notrace(); | |
128 | } | |
129 | EXPORT_SYMBOL_GPL(preempt_schedule_context); | |
130 | #endif /* CONFIG_PREEMPT */ | |
4eacdf18 FW |
131 | |
132 | /** | |
133 | * user_exit - Inform the context tracking that the CPU is | |
134 | * exiting userspace mode and entering the kernel. | |
135 | * | |
136 | * This function must be called after we entered the kernel from userspace | |
137 | * before any use of RCU read side critical section. This potentially include | |
138 | * any high level kernel code like syscalls, exceptions, signal handling, etc... | |
139 | * | |
140 | * This call supports re-entrancy. This way it can be called from any exception | |
141 | * handler without needing to know if we came from userspace or not. | |
142 | */ | |
91d1aa43 FW |
143 | void user_exit(void) |
144 | { | |
145 | unsigned long flags; | |
146 | ||
91d1aa43 FW |
147 | if (in_interrupt()) |
148 | return; | |
149 | ||
150 | local_irq_save(flags); | |
151 | if (__this_cpu_read(context_tracking.state) == IN_USER) { | |
d65ec121 FW |
152 | if (__this_cpu_read(context_tracking.active)) { |
153 | /* | |
154 | * We are going to run code that may use RCU. Inform | |
155 | * RCU core about that (ie: we may need the tick again). | |
156 | */ | |
157 | rcu_user_exit(); | |
158 | vtime_user_exit(current); | |
159 | } | |
abf917cd | 160 | __this_cpu_write(context_tracking.state, IN_KERNEL); |
91d1aa43 FW |
161 | } |
162 | local_irq_restore(flags); | |
163 | } | |
164 | ||
2d854e57 | 165 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN |
6a61671b FW |
166 | void guest_enter(void) |
167 | { | |
168 | if (vtime_accounting_enabled()) | |
169 | vtime_guest_enter(current); | |
170 | else | |
2d854e57 | 171 | current->flags |= PF_VCPU; |
6a61671b FW |
172 | } |
173 | EXPORT_SYMBOL_GPL(guest_enter); | |
174 | ||
175 | void guest_exit(void) | |
176 | { | |
177 | if (vtime_accounting_enabled()) | |
178 | vtime_guest_exit(current); | |
179 | else | |
2d854e57 | 180 | current->flags &= ~PF_VCPU; |
6a61671b FW |
181 | } |
182 | EXPORT_SYMBOL_GPL(guest_exit); | |
2d854e57 | 183 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |
6a61671b | 184 | |
4eacdf18 FW |
185 | |
186 | /** | |
187 | * context_tracking_task_switch - context switch the syscall callbacks | |
188 | * @prev: the task that is being switched out | |
189 | * @next: the task that is being switched in | |
190 | * | |
191 | * The context tracking uses the syscall slow path to implement its user-kernel | |
192 | * boundaries probes on syscalls. This way it doesn't impact the syscall fast | |
193 | * path on CPUs that don't do context tracking. | |
194 | * | |
195 | * But we need to clear the flag on the previous task because it may later | |
196 | * migrate to some CPU that doesn't do the context tracking. As such the TIF | |
197 | * flag may not be desired there. | |
198 | */ | |
91d1aa43 FW |
199 | void context_tracking_task_switch(struct task_struct *prev, |
200 | struct task_struct *next) | |
201 | { | |
d65ec121 FW |
202 | clear_tsk_thread_flag(prev, TIF_NOHZ); |
203 | set_tsk_thread_flag(next, TIF_NOHZ); | |
91d1aa43 | 204 | } |