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 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 unsigned long flags; 64 65 /* 66 * Repeat the user_enter() check here because some archs may be calling 67 * this from asm and if no CPU needs context tracking, they shouldn't 68 * go further. Repeat the check here until they support the inline static 69 * key check. 70 */ 71 if (!context_tracking_is_enabled()) 72 return; 73 74 /* 75 * Some contexts may involve an exception occuring in an irq, 76 * leading to that nesting: 77 * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit() 78 * This would mess up the dyntick_nesting count though. And rcu_irq_*() 79 * helpers are enough to protect RCU uses inside the exception. So 80 * just return immediately if we detect we are in an IRQ. 81 */ 82 if (in_interrupt()) 83 return; 84 85 /* Kernel threads aren't supposed to go to userspace */ 86 WARN_ON_ONCE(!current->mm); 87 88 local_irq_save(flags); 89 if (!context_tracking_recursion_enter()) 90 goto out_irq_restore; 91 92 if ( __this_cpu_read(context_tracking.state) != state) { 93 if (__this_cpu_read(context_tracking.active)) { 94 /* 95 * At this stage, only low level arch entry code remains and 96 * then we'll run in userspace. We can assume there won't be 97 * any RCU read-side critical section until the next call to 98 * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency 99 * on the tick. 100 */ 101 if (state == CONTEXT_USER) { 102 trace_user_enter(0); 103 vtime_user_enter(current); 104 } 105 rcu_user_enter(); 106 } 107 /* 108 * Even if context tracking is disabled on this CPU, because it's outside 109 * the full dynticks mask for example, we still have to keep track of the 110 * context transitions and states to prevent inconsistency on those of 111 * other CPUs. 112 * If a task triggers an exception in userspace, sleep on the exception 113 * handler and then migrate to another CPU, that new CPU must know where 114 * the exception returns by the time we call exception_exit(). 115 * This information can only be provided by the previous CPU when it called 116 * exception_enter(). 117 * OTOH we can spare the calls to vtime and RCU when context_tracking.active 118 * is false because we know that CPU is not tickless. 119 */ 120 __this_cpu_write(context_tracking.state, state); 121 } 122 context_tracking_recursion_exit(); 123 out_irq_restore: 124 local_irq_restore(flags); 125 } 126 NOKPROBE_SYMBOL(context_tracking_enter); 127 EXPORT_SYMBOL_GPL(context_tracking_enter); 128 129 void context_tracking_user_enter(void) 130 { 131 context_tracking_enter(CONTEXT_USER); 132 } 133 NOKPROBE_SYMBOL(context_tracking_user_enter); 134 135 /** 136 * context_tracking_exit - Inform the context tracking that the CPU is 137 * exiting user or guest mode and entering the kernel. 138 * 139 * This function must be called after we entered the kernel from user or 140 * guest space before any use of RCU read side critical section. This 141 * potentially include any high level kernel code like syscalls, exceptions, 142 * signal handling, etc... 143 * 144 * This call supports re-entrancy. This way it can be called from any exception 145 * handler without needing to know if we came from userspace or not. 146 */ 147 void context_tracking_exit(enum ctx_state state) 148 { 149 unsigned long flags; 150 151 if (!context_tracking_is_enabled()) 152 return; 153 154 if (in_interrupt()) 155 return; 156 157 local_irq_save(flags); 158 if (!context_tracking_recursion_enter()) 159 goto out_irq_restore; 160 161 if (__this_cpu_read(context_tracking.state) == state) { 162 if (__this_cpu_read(context_tracking.active)) { 163 /* 164 * We are going to run code that may use RCU. Inform 165 * RCU core about that (ie: we may need the tick again). 166 */ 167 rcu_user_exit(); 168 if (state == CONTEXT_USER) { 169 vtime_user_exit(current); 170 trace_user_exit(0); 171 } 172 } 173 __this_cpu_write(context_tracking.state, CONTEXT_KERNEL); 174 } 175 context_tracking_recursion_exit(); 176 out_irq_restore: 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 context_tracking_exit(CONTEXT_USER); 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_key_slow_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 219