1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Context tracking: Probe on high level context boundaries such as kernel
4  * and userspace. This includes syscalls and exceptions entry/exit.
5  *
6  * This is used by RCU to remove its dependency on the timer tick while a CPU
7  * runs in userspace.
8  *
9  *  Started by Frederic Weisbecker:
10  *
11  * Copyright (C) 2012 Red Hat, Inc., Frederic Weisbecker <fweisbec@redhat.com>
12  *
13  * Many thanks to Gilad Ben-Yossef, Paul McKenney, Ingo Molnar, Andrew Morton,
14  * Steven Rostedt, Peter Zijlstra for suggestions and improvements.
15  *
16  */
17 
18 #include <linux/context_tracking.h>
19 #include <linux/rcupdate.h>
20 #include <linux/sched.h>
21 #include <linux/hardirq.h>
22 #include <linux/export.h>
23 #include <linux/kprobes.h>
24 
25 #define CREATE_TRACE_POINTS
26 #include <trace/events/context_tracking.h>
27 
28 DEFINE_STATIC_KEY_FALSE(context_tracking_key);
29 EXPORT_SYMBOL_GPL(context_tracking_key);
30 
31 DEFINE_PER_CPU(struct context_tracking, context_tracking);
32 EXPORT_SYMBOL_GPL(context_tracking);
33 
context_tracking_recursion_enter(void)34 static noinstr bool context_tracking_recursion_enter(void)
35 {
36 	int recursion;
37 
38 	recursion = __this_cpu_inc_return(context_tracking.recursion);
39 	if (recursion == 1)
40 		return true;
41 
42 	WARN_ONCE((recursion < 1), "Invalid context tracking recursion value %d\n", recursion);
43 	__this_cpu_dec(context_tracking.recursion);
44 
45 	return false;
46 }
47 
context_tracking_recursion_exit(void)48 static __always_inline void context_tracking_recursion_exit(void)
49 {
50 	__this_cpu_dec(context_tracking.recursion);
51 }
52 
53 /**
54  * context_tracking_enter - Inform the context tracking that the CPU is going
55  *                          enter user or guest space mode.
56  *
57  * This function must be called right before we switch from the kernel
58  * to user or guest space, when it's guaranteed the remaining kernel
59  * instructions to execute won't use any RCU read side critical section
60  * because this function sets RCU in extended quiescent state.
61  */
__context_tracking_enter(enum ctx_state state)62 void noinstr __context_tracking_enter(enum ctx_state state)
63 {
64 	/* Kernel threads aren't supposed to go to userspace */
65 	WARN_ON_ONCE(!current->mm);
66 
67 	if (!context_tracking_recursion_enter())
68 		return;
69 
70 	if ( __this_cpu_read(context_tracking.state) != state) {
71 		if (__this_cpu_read(context_tracking.active)) {
72 			/*
73 			 * At this stage, only low level arch entry code remains and
74 			 * then we'll run in userspace. We can assume there won't be
75 			 * any RCU read-side critical section until the next call to
76 			 * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency
77 			 * on the tick.
78 			 */
79 			if (state == CONTEXT_USER) {
80 				instrumentation_begin();
81 				trace_user_enter(0);
82 				vtime_user_enter(current);
83 				instrumentation_end();
84 			}
85 			rcu_user_enter();
86 		}
87 		/*
88 		 * Even if context tracking is disabled on this CPU, because it's outside
89 		 * the full dynticks mask for example, we still have to keep track of the
90 		 * context transitions and states to prevent inconsistency on those of
91 		 * other CPUs.
92 		 * If a task triggers an exception in userspace, sleep on the exception
93 		 * handler and then migrate to another CPU, that new CPU must know where
94 		 * the exception returns by the time we call exception_exit().
95 		 * This information can only be provided by the previous CPU when it called
96 		 * exception_enter().
97 		 * OTOH we can spare the calls to vtime and RCU when context_tracking.active
98 		 * is false because we know that CPU is not tickless.
99 		 */
100 		__this_cpu_write(context_tracking.state, state);
101 	}
102 	context_tracking_recursion_exit();
103 }
104 EXPORT_SYMBOL_GPL(__context_tracking_enter);
105 
context_tracking_enter(enum ctx_state state)106 void context_tracking_enter(enum ctx_state state)
107 {
108 	unsigned long flags;
109 
110 	/*
111 	 * Some contexts may involve an exception occuring in an irq,
112 	 * leading to that nesting:
113 	 * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
114 	 * This would mess up the dyntick_nesting count though. And rcu_irq_*()
115 	 * helpers are enough to protect RCU uses inside the exception. So
116 	 * just return immediately if we detect we are in an IRQ.
117 	 */
118 	if (in_interrupt())
119 		return;
120 
121 	local_irq_save(flags);
122 	__context_tracking_enter(state);
123 	local_irq_restore(flags);
124 }
125 NOKPROBE_SYMBOL(context_tracking_enter);
126 EXPORT_SYMBOL_GPL(context_tracking_enter);
127 
context_tracking_user_enter(void)128 void context_tracking_user_enter(void)
129 {
130 	user_enter();
131 }
132 NOKPROBE_SYMBOL(context_tracking_user_enter);
133 
134 /**
135  * context_tracking_exit - Inform the context tracking that the CPU is
136  *                         exiting user or guest mode and entering the kernel.
137  *
138  * This function must be called after we entered the kernel from user or
139  * guest space before any use of RCU read side critical section. This
140  * potentially include any high level kernel code like syscalls, exceptions,
141  * signal handling, etc...
142  *
143  * This call supports re-entrancy. This way it can be called from any exception
144  * handler without needing to know if we came from userspace or not.
145  */
__context_tracking_exit(enum ctx_state state)146 void noinstr __context_tracking_exit(enum ctx_state state)
147 {
148 	if (!context_tracking_recursion_enter())
149 		return;
150 
151 	if (__this_cpu_read(context_tracking.state) == state) {
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 			if (state == CONTEXT_USER) {
159 				instrumentation_begin();
160 				vtime_user_exit(current);
161 				trace_user_exit(0);
162 				instrumentation_end();
163 			}
164 		}
165 		__this_cpu_write(context_tracking.state, CONTEXT_KERNEL);
166 	}
167 	context_tracking_recursion_exit();
168 }
169 EXPORT_SYMBOL_GPL(__context_tracking_exit);
170 
context_tracking_exit(enum ctx_state state)171 void context_tracking_exit(enum ctx_state state)
172 {
173 	unsigned long flags;
174 
175 	if (in_interrupt())
176 		return;
177 
178 	local_irq_save(flags);
179 	__context_tracking_exit(state);
180 	local_irq_restore(flags);
181 }
182 NOKPROBE_SYMBOL(context_tracking_exit);
183 EXPORT_SYMBOL_GPL(context_tracking_exit);
184 
context_tracking_user_exit(void)185 void context_tracking_user_exit(void)
186 {
187 	user_exit();
188 }
189 NOKPROBE_SYMBOL(context_tracking_user_exit);
190 
context_tracking_cpu_set(int cpu)191 void __init context_tracking_cpu_set(int cpu)
192 {
193 	static __initdata bool initialized = false;
194 
195 	if (!per_cpu(context_tracking.active, cpu)) {
196 		per_cpu(context_tracking.active, cpu) = true;
197 		static_branch_inc(&context_tracking_key);
198 	}
199 
200 	if (initialized)
201 		return;
202 
203 #ifdef CONFIG_HAVE_TIF_NOHZ
204 	/*
205 	 * Set TIF_NOHZ to init/0 and let it propagate to all tasks through fork
206 	 * This assumes that init is the only task at this early boot stage.
207 	 */
208 	set_tsk_thread_flag(&init_task, TIF_NOHZ);
209 #endif
210 	WARN_ON_ONCE(!tasklist_empty());
211 
212 	initialized = true;
213 }
214 
215 #ifdef CONFIG_CONTEXT_TRACKING_FORCE
context_tracking_init(void)216 void __init context_tracking_init(void)
217 {
218 	int cpu;
219 
220 	for_each_possible_cpu(cpu)
221 		context_tracking_cpu_set(cpu);
222 }
223 #endif
224