Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 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 | |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 17 | #include <linux/context_tracking.h> |
Frederic Weisbecker | 6a61671 | 2012-12-16 20:00:34 +0100 | [diff] [blame] | 18 | #include <linux/kvm_host.h> |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 19 | #include <linux/rcupdate.h> |
| 20 | #include <linux/sched.h> |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 21 | #include <linux/hardirq.h> |
Frederic Weisbecker | 6a61671 | 2012-12-16 20:00:34 +0100 | [diff] [blame] | 22 | #include <linux/export.h> |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 23 | |
Frederic Weisbecker | 95a79fd | 2013-01-07 18:12:14 +0100 | [diff] [blame] | 24 | DEFINE_PER_CPU(struct context_tracking, context_tracking) = { |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 25 | #ifdef CONFIG_CONTEXT_TRACKING_FORCE |
| 26 | .active = true, |
| 27 | #endif |
| 28 | }; |
| 29 | |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 30 | /** |
| 31 | * user_enter - Inform the context tracking that the CPU is going to |
| 32 | * enter userspace mode. |
| 33 | * |
| 34 | * This function must be called right before we switch from the kernel |
| 35 | * to userspace, when it's guaranteed the remaining kernel instructions |
| 36 | * to execute won't use any RCU read side critical section because this |
| 37 | * function sets RCU in extended quiescent state. |
| 38 | */ |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 39 | void user_enter(void) |
| 40 | { |
| 41 | unsigned long flags; |
| 42 | |
| 43 | /* |
| 44 | * Some contexts may involve an exception occuring in an irq, |
| 45 | * leading to that nesting: |
| 46 | * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit() |
| 47 | * This would mess up the dyntick_nesting count though. And rcu_irq_*() |
| 48 | * helpers are enough to protect RCU uses inside the exception. So |
| 49 | * just return immediately if we detect we are in an IRQ. |
| 50 | */ |
| 51 | if (in_interrupt()) |
| 52 | return; |
| 53 | |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 54 | /* Kernel threads aren't supposed to go to userspace */ |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 55 | WARN_ON_ONCE(!current->mm); |
| 56 | |
| 57 | local_irq_save(flags); |
| 58 | if (__this_cpu_read(context_tracking.active) && |
| 59 | __this_cpu_read(context_tracking.state) != IN_USER) { |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 60 | /* |
| 61 | * At this stage, only low level arch entry code remains and |
| 62 | * then we'll run in userspace. We can assume there won't be |
| 63 | * any RCU read-side critical section until the next call to |
| 64 | * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency |
| 65 | * on the tick. |
| 66 | */ |
Frederic Weisbecker | abf917c | 2012-07-25 07:56:04 +0200 | [diff] [blame] | 67 | vtime_user_enter(current); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 68 | rcu_user_enter(); |
Frederic Weisbecker | abf917c | 2012-07-25 07:56:04 +0200 | [diff] [blame] | 69 | __this_cpu_write(context_tracking.state, IN_USER); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 70 | } |
| 71 | local_irq_restore(flags); |
| 72 | } |
| 73 | |
Steven Rostedt | 29bb9e5 | 2013-05-24 15:23:40 -0400 | [diff] [blame^] | 74 | #ifdef CONFIG_PREEMPT |
| 75 | /** |
| 76 | * preempt_schedule_context - preempt_schedule called by tracing |
| 77 | * |
| 78 | * The tracing infrastructure uses preempt_enable_notrace to prevent |
| 79 | * recursion and tracing preempt enabling caused by the tracing |
| 80 | * infrastructure itself. But as tracing can happen in areas coming |
| 81 | * from userspace or just about to enter userspace, a preempt enable |
| 82 | * can occur before user_exit() is called. This will cause the scheduler |
| 83 | * to be called when the system is still in usermode. |
| 84 | * |
| 85 | * To prevent this, the preempt_enable_notrace will use this function |
| 86 | * instead of preempt_schedule() to exit user context if needed before |
| 87 | * calling the scheduler. |
| 88 | */ |
| 89 | void __sched notrace preempt_schedule_context(void) |
| 90 | { |
| 91 | struct thread_info *ti = current_thread_info(); |
| 92 | enum ctx_state prev_ctx; |
| 93 | |
| 94 | if (likely(ti->preempt_count || irqs_disabled())) |
| 95 | return; |
| 96 | |
| 97 | /* |
| 98 | * Need to disable preemption in case user_exit() is traced |
| 99 | * and the tracer calls preempt_enable_notrace() causing |
| 100 | * an infinite recursion. |
| 101 | */ |
| 102 | preempt_disable_notrace(); |
| 103 | prev_ctx = exception_enter(); |
| 104 | preempt_enable_no_resched_notrace(); |
| 105 | |
| 106 | preempt_schedule(); |
| 107 | |
| 108 | preempt_disable_notrace(); |
| 109 | exception_exit(prev_ctx); |
| 110 | preempt_enable_notrace(); |
| 111 | } |
| 112 | EXPORT_SYMBOL_GPL(preempt_schedule_context); |
| 113 | #endif /* CONFIG_PREEMPT */ |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 114 | |
| 115 | /** |
| 116 | * user_exit - Inform the context tracking that the CPU is |
| 117 | * exiting userspace mode and entering the kernel. |
| 118 | * |
| 119 | * This function must be called after we entered the kernel from userspace |
| 120 | * before any use of RCU read side critical section. This potentially include |
| 121 | * any high level kernel code like syscalls, exceptions, signal handling, etc... |
| 122 | * |
| 123 | * This call supports re-entrancy. This way it can be called from any exception |
| 124 | * handler without needing to know if we came from userspace or not. |
| 125 | */ |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 126 | void user_exit(void) |
| 127 | { |
| 128 | unsigned long flags; |
| 129 | |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 130 | if (in_interrupt()) |
| 131 | return; |
| 132 | |
| 133 | local_irq_save(flags); |
| 134 | if (__this_cpu_read(context_tracking.state) == IN_USER) { |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 135 | /* |
| 136 | * We are going to run code that may use RCU. Inform |
| 137 | * RCU core about that (ie: we may need the tick again). |
| 138 | */ |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 139 | rcu_user_exit(); |
Frederic Weisbecker | abf917c | 2012-07-25 07:56:04 +0200 | [diff] [blame] | 140 | vtime_user_exit(current); |
| 141 | __this_cpu_write(context_tracking.state, IN_KERNEL); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 142 | } |
| 143 | local_irq_restore(flags); |
| 144 | } |
| 145 | |
Frederic Weisbecker | 6a61671 | 2012-12-16 20:00:34 +0100 | [diff] [blame] | 146 | void guest_enter(void) |
| 147 | { |
| 148 | if (vtime_accounting_enabled()) |
| 149 | vtime_guest_enter(current); |
| 150 | else |
| 151 | __guest_enter(); |
| 152 | } |
| 153 | EXPORT_SYMBOL_GPL(guest_enter); |
| 154 | |
| 155 | void guest_exit(void) |
| 156 | { |
| 157 | if (vtime_accounting_enabled()) |
| 158 | vtime_guest_exit(current); |
| 159 | else |
| 160 | __guest_exit(); |
| 161 | } |
| 162 | EXPORT_SYMBOL_GPL(guest_exit); |
| 163 | |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 164 | |
| 165 | /** |
| 166 | * context_tracking_task_switch - context switch the syscall callbacks |
| 167 | * @prev: the task that is being switched out |
| 168 | * @next: the task that is being switched in |
| 169 | * |
| 170 | * The context tracking uses the syscall slow path to implement its user-kernel |
| 171 | * boundaries probes on syscalls. This way it doesn't impact the syscall fast |
| 172 | * path on CPUs that don't do context tracking. |
| 173 | * |
| 174 | * But we need to clear the flag on the previous task because it may later |
| 175 | * migrate to some CPU that doesn't do the context tracking. As such the TIF |
| 176 | * flag may not be desired there. |
| 177 | */ |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 178 | void context_tracking_task_switch(struct task_struct *prev, |
| 179 | struct task_struct *next) |
| 180 | { |
| 181 | if (__this_cpu_read(context_tracking.active)) { |
| 182 | clear_tsk_thread_flag(prev, TIF_NOHZ); |
| 183 | set_tsk_thread_flag(next, TIF_NOHZ); |
| 184 | } |
| 185 | } |