kernel/context_tracking.c

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