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
 23#define CREATE_TRACE_POINTS
 24#include <trace/events/context_tracking.h>
 25
 26struct static_key context_tracking_enabled = STATIC_KEY_INIT_FALSE;
 27EXPORT_SYMBOL_GPL(context_tracking_enabled);
 28
 29DEFINE_PER_CPU(struct context_tracking, context_tracking);
 30EXPORT_SYMBOL_GPL(context_tracking);
 31
 32void context_tracking_cpu_set(int cpu)
 33{
 34	if (!per_cpu(context_tracking.active, cpu)) {
 35		per_cpu(context_tracking.active, cpu) = true;
 36		static_key_slow_inc(&context_tracking_enabled);
 37	}
 38}
 39
 40/**
 41 * context_tracking_user_enter - Inform the context tracking that the CPU is going to
 42 *                               enter userspace mode.
 43 *
 44 * This function must be called right before we switch from the kernel
 45 * to userspace, when it's guaranteed the remaining kernel instructions
 46 * to execute won't use any RCU read side critical section because this
 47 * function sets RCU in extended quiescent state.
 48 */
 49void context_tracking_user_enter(void)
 50{
 51	unsigned long flags;
 52
 53	/*
 54	 * Repeat the user_enter() check here because some archs may be calling
 55	 * this from asm and if no CPU needs context tracking, they shouldn't
 56	 * go further. Repeat the check here until they support the inline static
 57	 * key check.
 58	 */
 59	if (!context_tracking_is_enabled())
 60		return;
 61
 62	/*
 63	 * Some contexts may involve an exception occuring in an irq,
 64	 * leading to that nesting:
 65	 * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
 66	 * This would mess up the dyntick_nesting count though. And rcu_irq_*()
 67	 * helpers are enough to protect RCU uses inside the exception. So
 68	 * just return immediately if we detect we are in an IRQ.
 69	 */
 70	if (in_interrupt())
 71		return;
 72
 73	/* Kernel threads aren't supposed to go to userspace */
 74	WARN_ON_ONCE(!current->mm);
 75
 76	local_irq_save(flags);
 77	if ( __this_cpu_read(context_tracking.state) != IN_USER) {
 78		if (__this_cpu_read(context_tracking.active)) {
 79			trace_user_enter(0);
 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			vtime_user_enter(current);
 88			rcu_user_enter();
 89		}
 90		/*
 91		 * Even if context tracking is disabled on this CPU, because it's outside
 92		 * the full dynticks mask for example, we still have to keep track of the
 93		 * context transitions and states to prevent inconsistency on those of
 94		 * other CPUs.
 95		 * If a task triggers an exception in userspace, sleep on the exception
 96		 * handler and then migrate to another CPU, that new CPU must know where
 97		 * the exception returns by the time we call exception_exit().
 98		 * This information can only be provided by the previous CPU when it called
 99		 * exception_enter().
100		 * OTOH we can spare the calls to vtime and RCU when context_tracking.active
101		 * is false because we know that CPU is not tickless.
102		 */
103		__this_cpu_write(context_tracking.state, IN_USER);
104	}
105	local_irq_restore(flags);
106}
107
108#ifdef CONFIG_PREEMPT
109/**
110 * preempt_schedule_context - preempt_schedule called by tracing
111 *
112 * The tracing infrastructure uses preempt_enable_notrace to prevent
113 * recursion and tracing preempt enabling caused by the tracing
114 * infrastructure itself. But as tracing can happen in areas coming
115 * from userspace or just about to enter userspace, a preempt enable
116 * can occur before user_exit() is called. This will cause the scheduler
117 * to be called when the system is still in usermode.
118 *
119 * To prevent this, the preempt_enable_notrace will use this function
120 * instead of preempt_schedule() to exit user context if needed before
121 * calling the scheduler.
122 */
123asmlinkage __visible void __sched notrace preempt_schedule_context(void)
124{
125	enum ctx_state prev_ctx;
126
127	if (likely(!preemptible()))
128		return;
129
130	/*
131	 * Need to disable preemption in case user_exit() is traced
132	 * and the tracer calls preempt_enable_notrace() causing
133	 * an infinite recursion.
134	 */
135	preempt_disable_notrace();
136	prev_ctx = exception_enter();
137	preempt_enable_no_resched_notrace();
138
139	preempt_schedule();
140
141	preempt_disable_notrace();
142	exception_exit(prev_ctx);
143	preempt_enable_notrace();
144}
145EXPORT_SYMBOL_GPL(preempt_schedule_context);
146#endif /* CONFIG_PREEMPT */
147
148/**
149 * context_tracking_user_exit - Inform the context tracking that the CPU is
150 *                              exiting userspace mode and entering the kernel.
151 *
152 * This function must be called after we entered the kernel from userspace
153 * before any use of RCU read side critical section. This potentially include
154 * any high level kernel code like syscalls, exceptions, signal handling, etc...
155 *
156 * This call supports re-entrancy. This way it can be called from any exception
157 * handler without needing to know if we came from userspace or not.
158 */
159void context_tracking_user_exit(void)
160{
161	unsigned long flags;
162
163	if (!context_tracking_is_enabled())
164		return;
165
166	if (in_interrupt())
167		return;
168
169	local_irq_save(flags);
170	if (__this_cpu_read(context_tracking.state) == IN_USER) {
171		if (__this_cpu_read(context_tracking.active)) {
172			/*
173			 * We are going to run code that may use RCU. Inform
174			 * RCU core about that (ie: we may need the tick again).
175			 */
176			rcu_user_exit();
177			vtime_user_exit(current);
178			trace_user_exit(0);
179		}
180		__this_cpu_write(context_tracking.state, IN_KERNEL);
181	}
182	local_irq_restore(flags);
183}
184
185/**
186 * __context_tracking_task_switch - context switch the syscall callbacks
187 * @prev: the task that is being switched out
188 * @next: the task that is being switched in
189 *
190 * The context tracking uses the syscall slow path to implement its user-kernel
191 * boundaries probes on syscalls. This way it doesn't impact the syscall fast
192 * path on CPUs that don't do context tracking.
193 *
194 * But we need to clear the flag on the previous task because it may later
195 * migrate to some CPU that doesn't do the context tracking. As such the TIF
196 * flag may not be desired there.
197 */
198void __context_tracking_task_switch(struct task_struct *prev,
199				    struct task_struct *next)
200{
201	clear_tsk_thread_flag(prev, TIF_NOHZ);
202	set_tsk_thread_flag(next, TIF_NOHZ);
203}
204
205#ifdef CONFIG_CONTEXT_TRACKING_FORCE
206void __init context_tracking_init(void)
207{
208	int cpu;
209
210	for_each_possible_cpu(cpu)
211		context_tracking_cpu_set(cpu);
212}
213#endif