include/linux/preempt.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef __LINUX_PREEMPT_H
   3 #define __LINUX_PREEMPT_H
   4
   5 /*
   6  * include/linux/preempt.h - macros for accessing and manipulating
   7  * preempt_count (used for kernel preemption, interrupt count, etc.)
   8  */
   9
  10 #include <linux/linkage.h>
  11 #include <linux/list.h>
  12
  13 /*
  14  * We put the hardirq and softirq counter into the preemption
  15  * counter. The bitmask has the following meaning:
  16  *
  17  * - bits 0-7 are the preemption count (max preemption depth: 256)
  18  * - bits 8-15 are the softirq count (max # of softirqs: 256)
  19  *
  20  * The hardirq count could in theory be the same as the number of
  21  * interrupts in the system, but we run all interrupt handlers with
  22  * interrupts disabled, so we cannot have nesting interrupts. Though
  23  * there are a few palaeontologic drivers which reenable interrupts in
  24  * the handler, so we need more than one bit here.
  25  *
  26  *         PREEMPT_MASK:        0x000000ff
  27  *         SOFTIRQ_MASK:        0x0000ff00
  28  *         HARDIRQ_MASK:        0x000f0000
  29  *             NMI_MASK:        0x00f00000
  30  * PREEMPT_NEED_RESCHED:        0x80000000
  31  */
  32 #define PREEMPT_BITS    8
  33 #define SOFTIRQ_BITS    8
  34 #define HARDIRQ_BITS    4
  35 #define NMI_BITS        4
  36
  37 #define PREEMPT_SHIFT   0
  38 #define SOFTIRQ_SHIFT   (PREEMPT_SHIFT + PREEMPT_BITS)
  39 #define HARDIRQ_SHIFT   (SOFTIRQ_SHIFT + SOFTIRQ_BITS)
  40 #define NMI_SHIFT       (HARDIRQ_SHIFT + HARDIRQ_BITS)
  41
  42 #define __IRQ_MASK(x)   ((1UL << (x))-1)
  43
  44 #define PREEMPT_MASK    (__IRQ_MASK(PREEMPT_BITS) << PREEMPT_SHIFT)
  45 #define SOFTIRQ_MASK    (__IRQ_MASK(SOFTIRQ_BITS) << SOFTIRQ_SHIFT)
  46 #define HARDIRQ_MASK    (__IRQ_MASK(HARDIRQ_BITS) << HARDIRQ_SHIFT)
  47 #define NMI_MASK        (__IRQ_MASK(NMI_BITS)     << NMI_SHIFT)
  48
  49 #define PREEMPT_OFFSET  (1UL << PREEMPT_SHIFT)
  50 #define SOFTIRQ_OFFSET  (1UL << SOFTIRQ_SHIFT)
  51 #define HARDIRQ_OFFSET  (1UL << HARDIRQ_SHIFT)
  52 #define NMI_OFFSET      (1UL << NMI_SHIFT)
  53
  54 #define SOFTIRQ_DISABLE_OFFSET  (2 * SOFTIRQ_OFFSET)
  55
  56 #define PREEMPT_DISABLED        (PREEMPT_DISABLE_OFFSET + PREEMPT_ENABLED)
  57
  58 /*
  59  * Disable preemption until the scheduler is running -- use an unconditional
  60  * value so that it also works on !PREEMPT_COUNT kernels.
  61  *
  62  * Reset by start_kernel()->sched_init()->init_idle()->init_idle_preempt_count().
  63  */
  64 #define INIT_PREEMPT_COUNT      PREEMPT_OFFSET
  65
  66 /*
  67  * Initial preempt_count value; reflects the preempt_count schedule invariant
  68  * which states that during context switches:
  69  *
  70  *    preempt_count() == 2*PREEMPT_DISABLE_OFFSET
  71  *
  72  * Note: PREEMPT_DISABLE_OFFSET is 0 for !PREEMPT_COUNT kernels.
  73  * Note: See finish_task_switch().
  74  */
  75 #define FORK_PREEMPT_COUNT      (2*PREEMPT_DISABLE_OFFSET + PREEMPT_ENABLED)
  76
  77 /* preempt_count() and related functions, depends on PREEMPT_NEED_RESCHED */
  78 #include <asm/preempt.h>
  79
  80 #define hardirq_count() (preempt_count() & HARDIRQ_MASK)
  81 #define softirq_count() (preempt_count() & SOFTIRQ_MASK)
  82 #define irq_count()     (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_MASK \
  83                                  | NMI_MASK))
  84
  85 /*
  86  * Are we doing bottom half or hardware interrupt processing?
  87  *
  88  * in_irq()       - We're in (hard) IRQ context
  89  * in_softirq()   - We have BH disabled, or are processing softirqs
  90  * in_interrupt() - We're in NMI,IRQ,SoftIRQ context or have BH disabled
  91  * in_serving_softirq() - We're in softirq context
  92  * in_nmi()       - We're in NMI context
  93  * in_task()      - We're in task context
  94  *
  95  * Note: due to the BH disabled confusion: in_softirq(),in_interrupt() really
  96  *       should not be used in new code.
  97  */
  98 #define in_irq()                (hardirq_count())
  99 #define in_softirq()            (softirq_count())
 100 #define in_interrupt()          (irq_count())
 101 #define in_serving_softirq()    (softirq_count() & SOFTIRQ_OFFSET)
 102 #define in_nmi()                (preempt_count() & NMI_MASK)
 103 #define in_task()               (!(preempt_count() & \
 104                                    (NMI_MASK | HARDIRQ_MASK | SOFTIRQ_OFFSET)))
 105
 106 /*
 107  * The preempt_count offset after preempt_disable();
 108  */
 109 #if defined(CONFIG_PREEMPT_COUNT)
 110 # define PREEMPT_DISABLE_OFFSET PREEMPT_OFFSET
 111 #else
 112 # define PREEMPT_DISABLE_OFFSET 0
 113 #endif
 114
 115 /*
 116  * The preempt_count offset after spin_lock()
 117  */
 118 #define PREEMPT_LOCK_OFFSET     PREEMPT_DISABLE_OFFSET
 119
 120 /*
 121  * The preempt_count offset needed for things like:
 122  *
 123  *  spin_lock_bh()
 124  *
 125  * Which need to disable both preemption (CONFIG_PREEMPT_COUNT) and
 126  * softirqs, such that unlock sequences of:
 127  *
 128  *  spin_unlock();
 129  *  local_bh_enable();
 130  *
 131  * Work as expected.
 132  */
 133 #define SOFTIRQ_LOCK_OFFSET (SOFTIRQ_DISABLE_OFFSET + PREEMPT_LOCK_OFFSET)
 134
 135 /*
 136  * Are we running in atomic context?  WARNING: this macro cannot
 137  * always detect atomic context; in particular, it cannot know about
 138  * held spinlocks in non-preemptible kernels.  Thus it should not be
 139  * used in the general case to determine whether sleeping is possible.
 140  * Do not use in_atomic() in driver code.
 141  */
 142 #define in_atomic()     (preempt_count() != 0)
 143
 144 /*
 145  * Check whether we were atomic before we did preempt_disable():
 146  * (used by the scheduler)
 147  */
 148 #define in_atomic_preempt_off() (preempt_count() != PREEMPT_DISABLE_OFFSET)
 149
 150 #if defined(CONFIG_DEBUG_PREEMPT) || defined(CONFIG_TRACE_PREEMPT_TOGGLE)
 151 extern void preempt_count_add(int val);
 152 extern void preempt_count_sub(int val);
 153 #define preempt_count_dec_and_test() \
 154         ({ preempt_count_sub(1); should_resched(0); })
 155 #else
 156 #define preempt_count_add(val)  __preempt_count_add(val)
 157 #define preempt_count_sub(val)  __preempt_count_sub(val)
 158 #define preempt_count_dec_and_test() __preempt_count_dec_and_test()
 159 #endif
 160
 161 #define __preempt_count_inc() __preempt_count_add(1)
 162 #define __preempt_count_dec() __preempt_count_sub(1)
 163
 164 #define preempt_count_inc() preempt_count_add(1)
 165 #define preempt_count_dec() preempt_count_sub(1)
 166
 167 #ifdef CONFIG_PREEMPT_COUNT
 168
 169 #define preempt_disable() \
 170 do { \
 171         preempt_count_inc(); \
 172         barrier(); \
 173 } while (0)
 174
 175 #define sched_preempt_enable_no_resched() \
 176 do { \
 177         barrier(); \
 178         preempt_count_dec(); \
 179 } while (0)
 180
 181 #define preempt_enable_no_resched() sched_preempt_enable_no_resched()
 182
 183 #define preemptible()   (preempt_count() == 0 && !irqs_disabled())
 184
 185 #ifdef CONFIG_PREEMPTION
 186 #define preempt_enable() \
 187 do { \
 188         barrier(); \
 189         if (unlikely(preempt_count_dec_and_test())) \
 190                 __preempt_schedule(); \
 191 } while (0)
 192
 193 #define preempt_enable_notrace() \
 194 do { \
 195         barrier(); \
 196         if (unlikely(__preempt_count_dec_and_test())) \
 197                 __preempt_schedule_notrace(); \
 198 } while (0)
 199
 200 #define preempt_check_resched() \
 201 do { \
 202         if (should_resched(0)) \
 203                 __preempt_schedule(); \
 204 } while (0)
 205
 206 #else /* !CONFIG_PREEMPTION */
 207 #define preempt_enable() \
 208 do { \
 209         barrier(); \
 210         preempt_count_dec(); \
 211 } while (0)
 212
 213 #define preempt_enable_notrace() \
 214 do { \
 215         barrier(); \
 216         __preempt_count_dec(); \
 217 } while (0)
 218
 219 #define preempt_check_resched() do { } while (0)
 220 #endif /* CONFIG_PREEMPTION */
 221
 222 #define preempt_disable_notrace() \
 223 do { \
 224         __preempt_count_inc(); \
 225         barrier(); \
 226 } while (0)
 227
 228 #define preempt_enable_no_resched_notrace() \
 229 do { \
 230         barrier(); \
 231         __preempt_count_dec(); \
 232 } while (0)
 233
 234 #else /* !CONFIG_PREEMPT_COUNT */
 235
 236 /*
 237  * Even if we don't have any preemption, we need preempt disable/enable
 238  * to be barriers, so that we don't have things like get_user/put_user
 239  * that can cause faults and scheduling migrate into our preempt-protected
 240  * region.
 241  */
 242 #define preempt_disable()                       barrier()
 243 #define sched_preempt_enable_no_resched()       barrier()
 244 #define preempt_enable_no_resched()             barrier()
 245 #define preempt_enable()                        barrier()
 246 #define preempt_check_resched()                 do { } while (0)
 247
 248 #define preempt_disable_notrace()               barrier()
 249 #define preempt_enable_no_resched_notrace()     barrier()
 250 #define preempt_enable_notrace()                barrier()
 251 #define preemptible()                           0
 252
 253 #endif /* CONFIG_PREEMPT_COUNT */
 254
 255 #ifdef MODULE
 256 /*
 257  * Modules have no business playing preemption tricks.
 258  */
 259 #undef sched_preempt_enable_no_resched
 260 #undef preempt_enable_no_resched
 261 #undef preempt_enable_no_resched_notrace
 262 #undef preempt_check_resched
 263 #endif
 264
 265 #define preempt_set_need_resched() \
 266 do { \
 267         set_preempt_need_resched(); \
 268 } while (0)
 269 #define preempt_fold_need_resched() \
 270 do { \
 271         if (tif_need_resched()) \
 272                 set_preempt_need_resched(); \
 273 } while (0)
 274
 275 #ifdef CONFIG_PREEMPT_NOTIFIERS
 276
 277 struct preempt_notifier;
 278
 279 /**
 280  * preempt_ops - notifiers called when a task is preempted and rescheduled
 281  * @sched_in: we're about to be rescheduled:
 282  *    notifier: struct preempt_notifier for the task being scheduled
 283  *    cpu:  cpu we're scheduled on
 284  * @sched_out: we've just been preempted
 285  *    notifier: struct preempt_notifier for the task being preempted
 286  *    next: the task that's kicking us out
 287  *
 288  * Please note that sched_in and out are called under different
 289  * contexts.  sched_out is called with rq lock held and irq disabled
 290  * while sched_in is called without rq lock and irq enabled.  This
 291  * difference is intentional and depended upon by its users.
 292  */
 293 struct preempt_ops {
 294         void (*sched_in)(struct preempt_notifier *notifier, int cpu);
 295         void (*sched_out)(struct preempt_notifier *notifier,
 296                           struct task_struct *next);
 297 };
 298
 299 /**
 300  * preempt_notifier - key for installing preemption notifiers
 301  * @link: internal use
 302  * @ops: defines the notifier functions to be called
 303  *
 304  * Usually used in conjunction with container_of().
 305  */
 306 struct preempt_notifier {
 307         struct hlist_node link;
 308         struct preempt_ops *ops;
 309 };
 310
 311 void preempt_notifier_inc(void);
 312 void preempt_notifier_dec(void);
 313 void preempt_notifier_register(struct preempt_notifier *notifier);
 314 void preempt_notifier_unregister(struct preempt_notifier *notifier);
 315
 316 static inline void preempt_notifier_init(struct preempt_notifier *notifier,
 317                                      struct preempt_ops *ops)
 318 {
 319         INIT_HLIST_NODE(&notifier->link);
 320         notifier->ops = ops;
 321 }
 322
 323 #endif
 324
 325 /**
 326  * migrate_disable - Prevent migration of the current task
 327  *
 328  * Maps to preempt_disable() which also disables preemption. Use
 329  * migrate_disable() to annotate that the intent is to prevent migration,
 330  * but not necessarily preemption.
 331  *
 332  * Can be invoked nested like preempt_disable() and needs the corresponding
 333  * number of migrate_enable() invocations.
 334  */
 335 static __always_inline void migrate_disable(void)
 336 {
 337         preempt_disable();
 338 }
 339
 340 /**
 341  * migrate_enable - Allow migration of the current task
 342  *
 343  * Counterpart to migrate_disable().
 344  *
 345  * As migrate_disable() can be invoked nested, only the outermost invocation
 346  * reenables migration.
 347  *
 348  * Currently mapped to preempt_enable().
 349  */
 350 static __always_inline void migrate_enable(void)
 351 {
 352         preempt_enable();
 353 }
 354
 355 #endif /* __LINUX_PREEMPT_H */