2 * Context tracking: Probe on high level context boundaries such as kernel
3 * and userspace. This includes syscalls and exceptions entry/exit.
5 * This is used by RCU to remove its dependency on the timer tick while a CPU
8 * Started by Frederic Weisbecker:
10 * Copyright (C) 2012 Red Hat, Inc., Frederic Weisbecker <fweisbec@redhat.com>
12 * Many thanks to Gilad Ben-Yossef, Paul McKenney, Ingo Molnar, Andrew Morton,
13 * Steven Rostedt, Peter Zijlstra for suggestions and improvements.
17 #include <linux/context_tracking.h>
18 #include <linux/rcupdate.h>
19 #include <linux/sched.h>
20 #include <linux/hardirq.h>
21 #include <linux/export.h>
22 #include <linux/kprobes.h>
24 #define CREATE_TRACE_POINTS
25 #include <trace/events/context_tracking.h>
27 struct static_key context_tracking_enabled = STATIC_KEY_INIT_FALSE;
28 EXPORT_SYMBOL_GPL(context_tracking_enabled);
30 DEFINE_PER_CPU(struct context_tracking, context_tracking);
31 EXPORT_SYMBOL_GPL(context_tracking);
33 void context_tracking_cpu_set(int cpu)
35 if (!per_cpu(context_tracking.active, cpu)) {
36 per_cpu(context_tracking.active, cpu) = true;
37 static_key_slow_inc(&context_tracking_enabled);
42 * context_tracking_user_enter - Inform the context tracking that the CPU is going to
43 * enter userspace mode.
45 * This function must be called right before we switch from the kernel
46 * to userspace, when it's guaranteed the remaining kernel instructions
47 * to execute won't use any RCU read side critical section because this
48 * function sets RCU in extended quiescent state.
50 void context_tracking_user_enter(void)
55 * Repeat the user_enter() check here because some archs may be calling
56 * this from asm and if no CPU needs context tracking, they shouldn't
57 * go further. Repeat the check here until they support the inline static
60 if (!context_tracking_is_enabled())
64 * Some contexts may involve an exception occuring in an irq,
65 * leading to that nesting:
66 * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
67 * This would mess up the dyntick_nesting count though. And rcu_irq_*()
68 * helpers are enough to protect RCU uses inside the exception. So
69 * just return immediately if we detect we are in an IRQ.
74 /* Kernel threads aren't supposed to go to userspace */
75 WARN_ON_ONCE(!current->mm);
77 local_irq_save(flags);
78 if ( __this_cpu_read(context_tracking.state) != IN_USER) {
79 if (__this_cpu_read(context_tracking.active)) {
82 * At this stage, only low level arch entry code remains and
83 * then we'll run in userspace. We can assume there won't be
84 * any RCU read-side critical section until the next call to
85 * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency
88 vtime_user_enter(current);
92 * Even if context tracking is disabled on this CPU, because it's outside
93 * the full dynticks mask for example, we still have to keep track of the
94 * context transitions and states to prevent inconsistency on those of
96 * If a task triggers an exception in userspace, sleep on the exception
97 * handler and then migrate to another CPU, that new CPU must know where
98 * the exception returns by the time we call exception_exit().
99 * This information can only be provided by the previous CPU when it called
101 * OTOH we can spare the calls to vtime and RCU when context_tracking.active
102 * is false because we know that CPU is not tickless.
104 __this_cpu_write(context_tracking.state, IN_USER);
106 local_irq_restore(flags);
108 NOKPROBE_SYMBOL(context_tracking_user_enter);
111 * context_tracking_user_exit - Inform the context tracking that the CPU is
112 * exiting userspace mode and entering the kernel.
114 * This function must be called after we entered the kernel from userspace
115 * before any use of RCU read side critical section. This potentially include
116 * any high level kernel code like syscalls, exceptions, signal handling, etc...
118 * This call supports re-entrancy. This way it can be called from any exception
119 * handler without needing to know if we came from userspace or not.
121 void context_tracking_user_exit(void)
125 if (!context_tracking_is_enabled())
131 local_irq_save(flags);
132 if (__this_cpu_read(context_tracking.state) == IN_USER) {
133 if (__this_cpu_read(context_tracking.active)) {
135 * We are going to run code that may use RCU. Inform
136 * RCU core about that (ie: we may need the tick again).
139 vtime_user_exit(current);
142 __this_cpu_write(context_tracking.state, IN_KERNEL);
144 local_irq_restore(flags);
146 NOKPROBE_SYMBOL(context_tracking_user_exit);
149 * __context_tracking_task_switch - context switch the syscall callbacks
150 * @prev: the task that is being switched out
151 * @next: the task that is being switched in
153 * The context tracking uses the syscall slow path to implement its user-kernel
154 * boundaries probes on syscalls. This way it doesn't impact the syscall fast
155 * path on CPUs that don't do context tracking.
157 * But we need to clear the flag on the previous task because it may later
158 * migrate to some CPU that doesn't do the context tracking. As such the TIF
159 * flag may not be desired there.
161 void __context_tracking_task_switch(struct task_struct *prev,
162 struct task_struct *next)
164 clear_tsk_thread_flag(prev, TIF_NOHZ);
165 set_tsk_thread_flag(next, TIF_NOHZ);
168 #ifdef CONFIG_CONTEXT_TRACKING_FORCE
169 void __init context_tracking_init(void)
173 for_each_possible_cpu(cpu)
174 context_tracking_cpu_set(cpu);