xref: /openbmc/linux/kernel/context_tracking.c (revision 0da85d1e)
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 
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>
23 
24 #define CREATE_TRACE_POINTS
25 #include <trace/events/context_tracking.h>
26 
27 struct static_key context_tracking_enabled = STATIC_KEY_INIT_FALSE;
28 EXPORT_SYMBOL_GPL(context_tracking_enabled);
29 
30 DEFINE_PER_CPU(struct context_tracking, context_tracking);
31 EXPORT_SYMBOL_GPL(context_tracking);
32 
33 void context_tracking_cpu_set(int cpu)
34 {
35 	if (!per_cpu(context_tracking.active, cpu)) {
36 		per_cpu(context_tracking.active, cpu) = true;
37 		static_key_slow_inc(&context_tracking_enabled);
38 	}
39 }
40 
41 /**
42  * context_tracking_enter - Inform the context tracking that the CPU is going
43  *                          enter user or guest space mode.
44  *
45  * This function must be called right before we switch from the kernel
46  * to user or guest space, when it's guaranteed the remaining kernel
47  * instructions to execute won't use any RCU read side critical section
48  * because this function sets RCU in extended quiescent state.
49  */
50 void context_tracking_enter(enum ctx_state state)
51 {
52 	unsigned long flags;
53 
54 	/*
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
58 	 * key check.
59 	 */
60 	if (!context_tracking_is_enabled())
61 		return;
62 
63 	/*
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.
70 	 */
71 	if (in_interrupt())
72 		return;
73 
74 	/* Kernel threads aren't supposed to go to userspace */
75 	WARN_ON_ONCE(!current->mm);
76 
77 	local_irq_save(flags);
78 	if ( __this_cpu_read(context_tracking.state) != state) {
79 		if (__this_cpu_read(context_tracking.active)) {
80 			/*
81 			 * At this stage, only low level arch entry code remains and
82 			 * then we'll run in userspace. We can assume there won't be
83 			 * any RCU read-side critical section until the next call to
84 			 * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency
85 			 * on the tick.
86 			 */
87 			if (state == CONTEXT_USER) {
88 				trace_user_enter(0);
89 				vtime_user_enter(current);
90 			}
91 			rcu_user_enter();
92 		}
93 		/*
94 		 * Even if context tracking is disabled on this CPU, because it's outside
95 		 * the full dynticks mask for example, we still have to keep track of the
96 		 * context transitions and states to prevent inconsistency on those of
97 		 * other CPUs.
98 		 * If a task triggers an exception in userspace, sleep on the exception
99 		 * handler and then migrate to another CPU, that new CPU must know where
100 		 * the exception returns by the time we call exception_exit().
101 		 * This information can only be provided by the previous CPU when it called
102 		 * exception_enter().
103 		 * OTOH we can spare the calls to vtime and RCU when context_tracking.active
104 		 * is false because we know that CPU is not tickless.
105 		 */
106 		__this_cpu_write(context_tracking.state, state);
107 	}
108 	local_irq_restore(flags);
109 }
110 NOKPROBE_SYMBOL(context_tracking_enter);
111 EXPORT_SYMBOL_GPL(context_tracking_enter);
112 
113 void context_tracking_user_enter(void)
114 {
115 	context_tracking_enter(CONTEXT_USER);
116 }
117 NOKPROBE_SYMBOL(context_tracking_user_enter);
118 
119 /**
120  * context_tracking_exit - Inform the context tracking that the CPU is
121  *                         exiting user or guest mode and entering the kernel.
122  *
123  * This function must be called after we entered the kernel from user or
124  * guest space before any use of RCU read side critical section. This
125  * potentially include any high level kernel code like syscalls, exceptions,
126  * signal handling, etc...
127  *
128  * This call supports re-entrancy. This way it can be called from any exception
129  * handler without needing to know if we came from userspace or not.
130  */
131 void context_tracking_exit(enum ctx_state state)
132 {
133 	unsigned long flags;
134 
135 	if (!context_tracking_is_enabled())
136 		return;
137 
138 	if (in_interrupt())
139 		return;
140 
141 	local_irq_save(flags);
142 	if (__this_cpu_read(context_tracking.state) == state) {
143 		if (__this_cpu_read(context_tracking.active)) {
144 			/*
145 			 * We are going to run code that may use RCU. Inform
146 			 * RCU core about that (ie: we may need the tick again).
147 			 */
148 			rcu_user_exit();
149 			if (state == CONTEXT_USER) {
150 				vtime_user_exit(current);
151 				trace_user_exit(0);
152 			}
153 		}
154 		__this_cpu_write(context_tracking.state, CONTEXT_KERNEL);
155 	}
156 	local_irq_restore(flags);
157 }
158 NOKPROBE_SYMBOL(context_tracking_exit);
159 EXPORT_SYMBOL_GPL(context_tracking_exit);
160 
161 void context_tracking_user_exit(void)
162 {
163 	context_tracking_exit(CONTEXT_USER);
164 }
165 NOKPROBE_SYMBOL(context_tracking_user_exit);
166 
167 /**
168  * __context_tracking_task_switch - context switch the syscall callbacks
169  * @prev: the task that is being switched out
170  * @next: the task that is being switched in
171  *
172  * The context tracking uses the syscall slow path to implement its user-kernel
173  * boundaries probes on syscalls. This way it doesn't impact the syscall fast
174  * path on CPUs that don't do context tracking.
175  *
176  * But we need to clear the flag on the previous task because it may later
177  * migrate to some CPU that doesn't do the context tracking. As such the TIF
178  * flag may not be desired there.
179  */
180 void __context_tracking_task_switch(struct task_struct *prev,
181 				    struct task_struct *next)
182 {
183 	clear_tsk_thread_flag(prev, TIF_NOHZ);
184 	set_tsk_thread_flag(next, TIF_NOHZ);
185 }
186 
187 #ifdef CONFIG_CONTEXT_TRACKING_FORCE
188 void __init context_tracking_init(void)
189 {
190 	int cpu;
191 
192 	for_each_possible_cpu(cpu)
193 		context_tracking_cpu_set(cpu);
194 }
195 #endif
196