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 DEFINE_STATIC_KEY_FALSE(context_tracking_enabled);
  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 static bool context_tracking_recursion_enter(void)
  34 {
  35         int recursion;
  36
  37         recursion = __this_cpu_inc_return(context_tracking.recursion);
  38         if (recursion == 1)
  39                 return true;
  40
  41         WARN_ONCE((recursion < 1), "Invalid context tracking recursion value %d\n", recursion);
  42         __this_cpu_dec(context_tracking.recursion);
  43
  44         return false;
  45 }
  46
  47 static void context_tracking_recursion_exit(void)
  48 {
  49         __this_cpu_dec(context_tracking.recursion);
  50 }
  51
  52 /**
  53  * context_tracking_enter - Inform the context tracking that the CPU is going
  54  *                          enter user or guest space mode.
  55  *
  56  * This function must be called right before we switch from the kernel
  57  * to user or guest space, when it's guaranteed the remaining kernel
  58  * instructions to execute won't use any RCU read side critical section
  59  * because this function sets RCU in extended quiescent state.
  60  */
  61 void __context_tracking_enter(enum ctx_state state)
  62 {
  63         /* Kernel threads aren't supposed to go to userspace */
  64         WARN_ON_ONCE(!current->mm);
  65
  66         if (!context_tracking_recursion_enter())
  67                 return;
  68
  69         if ( __this_cpu_read(context_tracking.state) != state) {
  70                 if (__this_cpu_read(context_tracking.active)) {
  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                         if (state == CONTEXT_USER) {
  79                                 trace_user_enter(0);
  80                                 vtime_user_enter(current);
  81                         }
  82                         rcu_user_enter();
  83                 }
  84                 /*
  85                  * Even if context tracking is disabled on this CPU, because it's outside
  86                  * the full dynticks mask for example, we still have to keep track of the
  87                  * context transitions and states to prevent inconsistency on those of
  88                  * other CPUs.
  89                  * If a task triggers an exception in userspace, sleep on the exception
  90                  * handler and then migrate to another CPU, that new CPU must know where
  91                  * the exception returns by the time we call exception_exit().
  92                  * This information can only be provided by the previous CPU when it called
  93                  * exception_enter().
  94                  * OTOH we can spare the calls to vtime and RCU when context_tracking.active
  95                  * is false because we know that CPU is not tickless.
  96                  */
  97                 __this_cpu_write(context_tracking.state, state);
  98         }
  99         context_tracking_recursion_exit();
 100 }
 101 NOKPROBE_SYMBOL(__context_tracking_enter);
 102 EXPORT_SYMBOL_GPL(__context_tracking_enter);
 103
 104 void context_tracking_enter(enum ctx_state state)
 105 {
 106         unsigned long flags;
 107
 108         /*
 109          * Some contexts may involve an exception occuring in an irq,
 110          * leading to that nesting:
 111          * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
 112          * This would mess up the dyntick_nesting count though. And rcu_irq_*()
 113          * helpers are enough to protect RCU uses inside the exception. So
 114          * just return immediately if we detect we are in an IRQ.
 115          */
 116         if (in_interrupt())
 117                 return;
 118
 119         local_irq_save(flags);
 120         __context_tracking_enter(state);
 121         local_irq_restore(flags);
 122 }
 123 NOKPROBE_SYMBOL(context_tracking_enter);
 124 EXPORT_SYMBOL_GPL(context_tracking_enter);
 125
 126 void context_tracking_user_enter(void)
 127 {
 128         user_enter();
 129 }
 130 NOKPROBE_SYMBOL(context_tracking_user_enter);
 131
 132 /**
 133  * context_tracking_exit - Inform the context tracking that the CPU is
 134  *                         exiting user or guest mode and entering the kernel.
 135  *
 136  * This function must be called after we entered the kernel from user or
 137  * guest space before any use of RCU read side critical section. This
 138  * potentially include any high level kernel code like syscalls, exceptions,
 139  * signal handling, etc...
 140  *
 141  * This call supports re-entrancy. This way it can be called from any exception
 142  * handler without needing to know if we came from userspace or not.
 143  */
 144 void __context_tracking_exit(enum ctx_state state)
 145 {
 146         if (!context_tracking_recursion_enter())
 147                 return;
 148
 149         if (__this_cpu_read(context_tracking.state) == state) {
 150                 if (__this_cpu_read(context_tracking.active)) {
 151                         /*
 152                          * We are going to run code that may use RCU. Inform
 153                          * RCU core about that (ie: we may need the tick again).
 154                          */
 155                         rcu_user_exit();
 156                         if (state == CONTEXT_USER) {
 157                                 vtime_user_exit(current);
 158                                 trace_user_exit(0);
 159                         }
 160                 }
 161                 __this_cpu_write(context_tracking.state, CONTEXT_KERNEL);
 162         }
 163         context_tracking_recursion_exit();
 164 }
 165 NOKPROBE_SYMBOL(__context_tracking_exit);
 166 EXPORT_SYMBOL_GPL(__context_tracking_exit);
 167
 168 void context_tracking_exit(enum ctx_state state)
 169 {
 170         unsigned long flags;
 171
 172         if (in_interrupt())
 173                 return;
 174
 175         local_irq_save(flags);
 176         __context_tracking_exit(state);
 177         local_irq_restore(flags);
 178 }
 179 NOKPROBE_SYMBOL(context_tracking_exit);
 180 EXPORT_SYMBOL_GPL(context_tracking_exit);
 181
 182 void context_tracking_user_exit(void)
 183 {
 184         user_exit();
 185 }
 186 NOKPROBE_SYMBOL(context_tracking_user_exit);
 187
 188 void __init context_tracking_cpu_set(int cpu)
 189 {
 190         static __initdata bool initialized = false;
 191
 192         if (!per_cpu(context_tracking.active, cpu)) {
 193                 per_cpu(context_tracking.active, cpu) = true;
 194                 static_branch_inc(&context_tracking_enabled);
 195         }
 196
 197         if (initialized)
 198                 return;
 199
 200         /*
 201          * Set TIF_NOHZ to init/0 and let it propagate to all tasks through fork
 202          * This assumes that init is the only task at this early boot stage.
 203          */
 204         set_tsk_thread_flag(&init_task, TIF_NOHZ);
 205         WARN_ON_ONCE(!tasklist_empty());
 206
 207         initialized = true;
 208 }
 209
 210 #ifdef CONFIG_CONTEXT_TRACKING_FORCE
 211 void __init context_tracking_init(void)
 212 {
 213         int cpu;
 214
 215         for_each_possible_cpu(cpu)
 216                 context_tracking_cpu_set(cpu);
 217 }
 218 #endif