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 CREATE_TRACE_POINTS
  24 #include <trace/events/context_tracking.h>
  25
  26 struct static_key context_tracking_enabled = STATIC_KEY_INIT_FALSE;
  27 EXPORT_SYMBOL_GPL(context_tracking_enabled);
  28
  29 DEFINE_PER_CPU(struct context_tracking, context_tracking);
  30 EXPORT_SYMBOL_GPL(context_tracking);
  31
  32 void 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  */
  49 void context_tracking_user_enter(void)
  50 {
  51         unsigned long flags;
  52
  53         /*
  54          * Some contexts may involve an exception occuring in an irq,
  55          * leading to that nesting:
  56          * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
  57          * This would mess up the dyntick_nesting count though. And rcu_irq_*()
  58          * helpers are enough to protect RCU uses inside the exception. So
  59          * just return immediately if we detect we are in an IRQ.
  60          */
  61         if (in_interrupt())
  62                 return;
  63
  64         /* Kernel threads aren't supposed to go to userspace */
  65         WARN_ON_ONCE(!current->mm);
  66
  67         local_irq_save(flags);
  68         if ( __this_cpu_read(context_tracking.state) != IN_USER) {
  69                 if (__this_cpu_read(context_tracking.active)) {
  70                         trace_user_enter(0);
  71                         /*
  72                          * At this stage, only low level arch entry code remains and
  73                          * then we'll run in userspace. We can assume there won't be
  74                          * any RCU read-side critical section until the next call to
  75                          * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency
  76                          * on the tick.
  77                          */
  78                         vtime_user_enter(current);
  79                         rcu_user_enter();
  80                 }
  81                 /*
  82                  * Even if context tracking is disabled on this CPU, because it's outside
  83                  * the full dynticks mask for example, we still have to keep track of the
  84                  * context transitions and states to prevent inconsistency on those of
  85                  * other CPUs.
  86                  * If a task triggers an exception in userspace, sleep on the exception
  87                  * handler and then migrate to another CPU, that new CPU must know where
  88                  * the exception returns by the time we call exception_exit().
  89                  * This information can only be provided by the previous CPU when it called
  90                  * exception_enter().
  91                  * OTOH we can spare the calls to vtime and RCU when context_tracking.active
  92                  * is false because we know that CPU is not tickless.
  93                  */
  94                 __this_cpu_write(context_tracking.state, IN_USER);
  95         }
  96         local_irq_restore(flags);
  97 }
  98
  99 #ifdef CONFIG_PREEMPT
 100 /**
 101  * preempt_schedule_context - preempt_schedule called by tracing
 102  *
 103  * The tracing infrastructure uses preempt_enable_notrace to prevent
 104  * recursion and tracing preempt enabling caused by the tracing
 105  * infrastructure itself. But as tracing can happen in areas coming
 106  * from userspace or just about to enter userspace, a preempt enable
 107  * can occur before user_exit() is called. This will cause the scheduler
 108  * to be called when the system is still in usermode.
 109  *
 110  * To prevent this, the preempt_enable_notrace will use this function
 111  * instead of preempt_schedule() to exit user context if needed before
 112  * calling the scheduler.
 113  */
 114 void __sched notrace preempt_schedule_context(void)
 115 {
 116         enum ctx_state prev_ctx;
 117
 118         if (likely(!preemptible()))
 119                 return;
 120
 121         /*
 122          * Need to disable preemption in case user_exit() is traced
 123          * and the tracer calls preempt_enable_notrace() causing
 124          * an infinite recursion.
 125          */
 126         preempt_disable_notrace();
 127         prev_ctx = exception_enter();
 128         preempt_enable_no_resched_notrace();
 129
 130         preempt_schedule();
 131
 132         preempt_disable_notrace();
 133         exception_exit(prev_ctx);
 134         preempt_enable_notrace();
 135 }
 136 EXPORT_SYMBOL_GPL(preempt_schedule_context);
 137 #endif /* CONFIG_PREEMPT */
 138
 139 /**
 140  * context_tracking_user_exit - Inform the context tracking that the CPU is
 141  *                              exiting userspace mode and entering the kernel.
 142  *
 143  * This function must be called after we entered the kernel from userspace
 144  * before any use of RCU read side critical section. This potentially include
 145  * any high level kernel code like syscalls, exceptions, signal handling, etc...
 146  *
 147  * This call supports re-entrancy. This way it can be called from any exception
 148  * handler without needing to know if we came from userspace or not.
 149  */
 150 void context_tracking_user_exit(void)
 151 {
 152         unsigned long flags;
 153
 154         if (in_interrupt())
 155                 return;
 156
 157         local_irq_save(flags);
 158         if (__this_cpu_read(context_tracking.state) == IN_USER) {
 159                 if (__this_cpu_read(context_tracking.active)) {
 160                         /*
 161                          * We are going to run code that may use RCU. Inform
 162                          * RCU core about that (ie: we may need the tick again).
 163                          */
 164                         rcu_user_exit();
 165                         vtime_user_exit(current);
 166                         trace_user_exit(0);
 167                 }
 168                 __this_cpu_write(context_tracking.state, IN_KERNEL);
 169         }
 170         local_irq_restore(flags);
 171 }
 172
 173 /**
 174  * __context_tracking_task_switch - context switch the syscall callbacks
 175  * @prev: the task that is being switched out
 176  * @next: the task that is being switched in
 177  *
 178  * The context tracking uses the syscall slow path to implement its user-kernel
 179  * boundaries probes on syscalls. This way it doesn't impact the syscall fast
 180  * path on CPUs that don't do context tracking.
 181  *
 182  * But we need to clear the flag on the previous task because it may later
 183  * migrate to some CPU that doesn't do the context tracking. As such the TIF
 184  * flag may not be desired there.
 185  */
 186 void __context_tracking_task_switch(struct task_struct *prev,
 187                                     struct task_struct *next)
 188 {
 189         clear_tsk_thread_flag(prev, TIF_NOHZ);
 190         set_tsk_thread_flag(next, TIF_NOHZ);
 191 }
 192
 193 #ifdef CONFIG_CONTEXT_TRACKING_FORCE
 194 void __init context_tracking_init(void)
 195 {
 196         int cpu;
 197
 198         for_each_possible_cpu(cpu)
 199                 context_tracking_cpu_set(cpu);
 200 }
 201 #endif