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 #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_user_enter - Inform the context tracking that the CPU is going to
  43  *                               enter userspace mode.
  44  *
  45  * This function must be called right before we switch from the kernel
  46  * to userspace, when it's guaranteed the remaining kernel instructions
  47  * to execute won't use any RCU read side critical section because this
  48  * function sets RCU in extended quiescent state.
  49  */
  50 void context_tracking_user_enter(void)
  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) != IN_USER) {
  79                 if (__this_cpu_read(context_tracking.active)) {
  80                         trace_user_enter(0);
  81                         /*
  82                          * At this stage, only low level arch entry code remains and
  83                          * then we'll run in userspace. We can assume there won't be
  84                          * any RCU read-side critical section until the next call to
  85                          * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency
  86                          * on the tick.
  87                          */
  88                         vtime_user_enter(current);
  89                         rcu_user_enter();
  90                 }
  91                 /*
  92                  * Even if context tracking is disabled on this CPU, because it's outside
  93                  * the full dynticks mask for example, we still have to keep track of the
  94                  * context transitions and states to prevent inconsistency on those of
  95                  * other CPUs.
  96                  * If a task triggers an exception in userspace, sleep on the exception
  97                  * handler and then migrate to another CPU, that new CPU must know where
  98                  * the exception returns by the time we call exception_exit().
  99                  * This information can only be provided by the previous CPU when it called
 100                  * exception_enter().
 101                  * OTOH we can spare the calls to vtime and RCU when context_tracking.active
 102                  * is false because we know that CPU is not tickless.
 103                  */
 104                 __this_cpu_write(context_tracking.state, IN_USER);
 105         }
 106         local_irq_restore(flags);
 107 }
 108 NOKPROBE_SYMBOL(context_tracking_user_enter);
 109
 110 #ifdef CONFIG_PREEMPT
 111 /**
 112  * preempt_schedule_context - preempt_schedule called by tracing
 113  *
 114  * The tracing infrastructure uses preempt_enable_notrace to prevent
 115  * recursion and tracing preempt enabling caused by the tracing
 116  * infrastructure itself. But as tracing can happen in areas coming
 117  * from userspace or just about to enter userspace, a preempt enable
 118  * can occur before user_exit() is called. This will cause the scheduler
 119  * to be called when the system is still in usermode.
 120  *
 121  * To prevent this, the preempt_enable_notrace will use this function
 122  * instead of preempt_schedule() to exit user context if needed before
 123  * calling the scheduler.
 124  */
 125 asmlinkage __visible void __sched notrace preempt_schedule_context(void)
 126 {
 127         enum ctx_state prev_ctx;
 128
 129         if (likely(!preemptible()))
 130                 return;
 131
 132         /*
 133          * Need to disable preemption in case user_exit() is traced
 134          * and the tracer calls preempt_enable_notrace() causing
 135          * an infinite recursion.
 136          */
 137         preempt_disable_notrace();
 138         prev_ctx = exception_enter();
 139         preempt_enable_no_resched_notrace();
 140
 141         preempt_schedule();
 142
 143         preempt_disable_notrace();
 144         exception_exit(prev_ctx);
 145         preempt_enable_notrace();
 146 }
 147 EXPORT_SYMBOL_GPL(preempt_schedule_context);
 148 #endif /* CONFIG_PREEMPT */
 149
 150 /**
 151  * context_tracking_user_exit - Inform the context tracking that the CPU is
 152  *                              exiting userspace mode and entering the kernel.
 153  *
 154  * This function must be called after we entered the kernel from userspace
 155  * before any use of RCU read side critical section. This potentially include
 156  * any high level kernel code like syscalls, exceptions, signal handling, etc...
 157  *
 158  * This call supports re-entrancy. This way it can be called from any exception
 159  * handler without needing to know if we came from userspace or not.
 160  */
 161 void context_tracking_user_exit(void)
 162 {
 163         unsigned long flags;
 164
 165         if (!context_tracking_is_enabled())
 166                 return;
 167
 168         if (in_interrupt())
 169                 return;
 170
 171         local_irq_save(flags);
 172         if (__this_cpu_read(context_tracking.state) == IN_USER) {
 173                 if (__this_cpu_read(context_tracking.active)) {
 174                         /*
 175                          * We are going to run code that may use RCU. Inform
 176                          * RCU core about that (ie: we may need the tick again).
 177                          */
 178                         rcu_user_exit();
 179                         vtime_user_exit(current);
 180                         trace_user_exit(0);
 181                 }
 182                 __this_cpu_write(context_tracking.state, IN_KERNEL);
 183         }
 184         local_irq_restore(flags);
 185 }
 186 NOKPROBE_SYMBOL(context_tracking_user_exit);
 187
 188 /**
 189  * __context_tracking_task_switch - context switch the syscall callbacks
 190  * @prev: the task that is being switched out
 191  * @next: the task that is being switched in
 192  *
 193  * The context tracking uses the syscall slow path to implement its user-kernel
 194  * boundaries probes on syscalls. This way it doesn't impact the syscall fast
 195  * path on CPUs that don't do context tracking.
 196  *
 197  * But we need to clear the flag on the previous task because it may later
 198  * migrate to some CPU that doesn't do the context tracking. As such the TIF
 199  * flag may not be desired there.
 200  */
 201 void __context_tracking_task_switch(struct task_struct *prev,
 202                                     struct task_struct *next)
 203 {
 204         clear_tsk_thread_flag(prev, TIF_NOHZ);
 205         set_tsk_thread_flag(next, TIF_NOHZ);
 206 }
 207
 208 #ifdef CONFIG_CONTEXT_TRACKING_FORCE
 209 void __init context_tracking_init(void)
 210 {
 211         int cpu;
 212
 213         for_each_possible_cpu(cpu)
 214                 context_tracking_cpu_set(cpu);
 215 }
 216 #endif