4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and
7 * make semaphores SMP safe
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
9 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
10 * "A Kernel Model for Precision Timekeeping" by Dave Mills
11 * 1998-11-19 Implemented schedule_timeout() and related stuff
13 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
14 * serialize accesses to xtime/lost_ticks).
15 * Copyright (C) 1998 Andrea Arcangeli
16 * 1998-12-28 Implemented better SMP scheduling by Ingo Molnar
17 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
21 * 'sched.c' is the main kernel file. It contains scheduling primitives
22 * (sleep_on, wakeup, schedule etc) as well as a number of simple system
23 * call functions (type getpid()), which just extract a field from
28 #include <linux/kernel_stat.h>
29 #include <linux/fdreg.h>
30 #include <linux/delay.h>
31 #include <linux/interrupt.h>
32 #include <linux/smp_lock.h>
33 #include <linux/init.h>
36 #include <asm/uaccess.h>
37 #include <asm/pgtable.h>
38 #include <asm/mmu_context.h>
39 #include <asm/semaphore-helper.h>
41 #include <linux/timex.h>
47 unsigned securebits
= SECUREBITS_DEFAULT
; /* systemwide security settings */
49 long tick
= (1000000 + HZ
/2) / HZ
; /* timer interrupt period */
51 /* The current time */
52 volatile struct timeval xtime
__attribute__ ((aligned (16)));
54 /* Don't completely fail for HZ > 500. */
55 int tickadj
= 500/HZ
? : 1; /* microsecs */
57 DECLARE_TASK_QUEUE(tq_timer
);
58 DECLARE_TASK_QUEUE(tq_immediate
);
59 DECLARE_TASK_QUEUE(tq_scheduler
);
62 * phase-lock loop variables
64 /* TIME_ERROR prevents overwriting the CMOS clock */
65 int time_state
= TIME_OK
; /* clock synchronization status */
66 int time_status
= STA_UNSYNC
; /* clock status bits */
67 long time_offset
= 0; /* time adjustment (us) */
68 long time_constant
= 2; /* pll time constant */
69 long time_tolerance
= MAXFREQ
; /* frequency tolerance (ppm) */
70 long time_precision
= 1; /* clock precision (us) */
71 long time_maxerror
= NTP_PHASE_LIMIT
; /* maximum error (us) */
72 long time_esterror
= NTP_PHASE_LIMIT
; /* estimated error (us) */
73 long time_phase
= 0; /* phase offset (scaled us) */
74 long time_freq
= ((1000000 + HZ
/2) % HZ
- HZ
/2) << SHIFT_USEC
; /* frequency offset (scaled ppm) */
75 long time_adj
= 0; /* tick adjust (scaled 1 / HZ) */
76 long time_reftime
= 0; /* time at last adjustment (s) */
79 long time_adjust_step
= 0;
81 unsigned long event
= 0;
83 extern int do_setitimer(int, struct itimerval
*, struct itimerval
*);
84 unsigned int * prof_buffer
= NULL
;
85 unsigned long prof_len
= 0;
86 unsigned long prof_shift
= 0;
88 extern void mem_use(void);
90 unsigned long volatile jiffies
=0;
93 * Init task must be ok at boot for the ix86 as we will check its signals
94 * via the SMP irq return path.
97 struct task_struct
* init_tasks
[NR_CPUS
] = {&init_task
, };
100 * The tasklist_lock protects the linked list of processes.
102 * The scheduler lock is protecting against multiple entry
103 * into the scheduling code, and doesn't need to worry
104 * about interrupts (because interrupts cannot call the
107 * The run-queue lock locks the parts that actually access
108 * and change the run-queues, and have to be interrupt-safe.
110 spinlock_t runqueue_lock
= SPIN_LOCK_UNLOCKED
; /* second */
111 rwlock_t tasklist_lock
= RW_LOCK_UNLOCKED
; /* third */
113 static LIST_HEAD(runqueue_head
);
116 * We align per-CPU scheduling data on cacheline boundaries,
117 * to prevent cacheline ping-pong.
120 struct schedule_data
{
121 struct task_struct
* curr
;
122 cycles_t last_schedule
;
124 char __pad
[SMP_CACHE_BYTES
];
125 } aligned_data
[NR_CPUS
] __cacheline_aligned
= { {{&init_task
,0}}};
127 #define cpu_curr(cpu) aligned_data[(cpu)].schedule_data.curr
129 struct kernel_stat kstat
= { 0 };
133 #define idle_task(cpu) (init_tasks[cpu_number_map[(cpu)]])
134 #define can_schedule(p) (!(p)->has_cpu)
138 #define idle_task(cpu) (&init_task)
139 #define can_schedule(p) (1)
143 void scheduling_functions_start_here(void) { }
146 * This is the function that decides how desirable a process is..
147 * You can weigh different processes against each other depending
148 * on what CPU they've run on lately etc to try to handle cache
149 * and TLB miss penalties.
152 * -1000: never select this
153 * 0: out of time, recalculate counters (but it might still be
155 * +ve: "goodness" value (the larger, the better)
156 * +1000: realtime process, select this.
159 static inline int goodness(struct task_struct
* p
, int this_cpu
, struct mm_struct
*this_mm
)
164 * Realtime process, select the first one on the
165 * runqueue (taking priorities within processes
168 if (p
->policy
!= SCHED_OTHER
) {
169 weight
= 1000 + p
->rt_priority
;
174 * Give the process a first-approximation goodness value
175 * according to the number of clock-ticks it has left.
177 * Don't do any other calculations if the time slice is
185 /* Give a largish advantage to the same processor... */
186 /* (this is equivalent to penalizing other processors) */
187 if (p
->processor
== this_cpu
)
188 weight
+= PROC_CHANGE_PENALTY
;
191 /* .. and a slight advantage to the current MM */
192 if (p
->mm
== this_mm
)
194 weight
+= p
->priority
;
201 * subtle. We want to discard a yielded process only if it's being
202 * considered for a reschedule. Wakeup-time 'queries' of the scheduling
203 * state do not count. Another optimization we do: sched_yield()-ed
204 * processes are runnable (and thus will be considered for scheduling)
205 * right when they are calling schedule(). So the only place we need
206 * to care about SCHED_YIELD is when we calculate the previous process'
209 static inline int prev_goodness(struct task_struct
* p
, int this_cpu
, struct mm_struct
*this_mm
)
211 if (p
->policy
& SCHED_YIELD
) {
212 p
->policy
&= ~SCHED_YIELD
;
215 return goodness(p
, this_cpu
, this_mm
);
219 * the 'goodness value' of replacing a process on a given CPU.
220 * positive value means 'replace', zero or negative means 'dont'.
222 static inline int preemption_goodness(struct task_struct
* prev
, struct task_struct
* p
, int cpu
)
224 return goodness(p
, cpu
, prev
->mm
) - goodness(prev
, cpu
, prev
->mm
);
228 * If there is a dependency between p1 and p2,
229 * don't be too eager to go into the slow schedule.
230 * In particular, if p1 and p2 both want the kernel
231 * lock, there is no point in trying to make them
232 * extremely parallel..
234 * (No lock - lock_depth < 0)
236 * There are two additional metrics here:
238 * first, a 'cutoff' interval, currently 0-200 usecs on
239 * x86 CPUs, depending on the size of the 'SMP-local cache'.
240 * If the current process has longer average timeslices than
241 * this, then we utilize the idle CPU.
243 * second, if the wakeup comes from a process context,
244 * then the two processes are 'related'. (they form a
247 * An idle CPU is almost always a bad thing, thus we skip
248 * the idle-CPU utilization only if both these conditions
249 * are true. (ie. a 'process-gang' rescheduling with rather
250 * high frequency should stay on the same CPU).
252 * [We can switch to something more finegrained in 2.3.]
254 * do not 'guess' if the to-be-scheduled task is RT.
256 #define related(p1,p2) (((p1)->lock_depth >= 0) && (p2)->lock_depth >= 0) && \
257 (((p2)->policy == SCHED_OTHER) && ((p1)->avg_slice < cacheflush_time))
259 static inline void reschedule_idle_slow(struct task_struct
* p
)
263 * (see reschedule_idle() for an explanation first ...)
267 * We try to find another (idle) CPU for this woken-up process.
269 * On SMP, we mostly try to see if the CPU the task used
270 * to run on is idle.. but we will use another idle CPU too,
271 * at this point we already know that this CPU is not
272 * willing to reschedule in the near future.
274 * An idle CPU is definitely wasted, especially if this CPU is
275 * running long-timeslice processes. The following algorithm is
276 * pretty good at finding the best idle CPU to send this process
279 * [We can try to preempt low-priority processes on other CPUs in
280 * 2.3. Also we can try to use the avg_slice value to predict
281 * 'likely reschedule' events even on other CPUs.]
283 int this_cpu
= smp_processor_id(), target_cpu
;
284 struct task_struct
*tsk
, *target_tsk
;
285 int cpu
, best_cpu
, weight
, best_weight
, i
;
288 best_weight
= 0; /* prevents negative weight */
290 spin_lock_irqsave(&runqueue_lock
, flags
);
293 * shortcut if the woken up task's last CPU is
296 best_cpu
= p
->processor
;
297 target_tsk
= idle_task(best_cpu
);
298 if (cpu_curr(best_cpu
) == target_tsk
)
302 for (i
= 0; i
< smp_num_cpus
; i
++) {
303 cpu
= cpu_logical_map(i
);
307 weight
= preemption_goodness(tsk
, p
, cpu
);
308 if (weight
> best_weight
) {
309 best_weight
= weight
;
315 * found any suitable CPU?
321 target_cpu
= target_tsk
->processor
;
322 target_tsk
->need_resched
= 1;
323 spin_unlock_irqrestore(&runqueue_lock
, flags
);
325 * the APIC stuff can go outside of the lock because
326 * it uses no task information, only CPU#.
328 if (target_cpu
!= this_cpu
)
329 smp_send_reschedule(target_cpu
);
332 spin_unlock_irqrestore(&runqueue_lock
, flags
);
335 int this_cpu
= smp_processor_id();
336 struct task_struct
*tsk
;
338 tsk
= cpu_curr(this_cpu
);
339 if (preemption_goodness(tsk
, p
, this_cpu
) > 0)
340 tsk
->need_resched
= 1;
344 static void reschedule_idle(struct task_struct
* p
)
347 int cpu
= smp_processor_id();
349 * ("wakeup()" should not be called before we've initialized
351 * Basically a not-yet initialized SMP subsystem can be
352 * considered as a not-yet working scheduler, simply dont use
353 * it before it's up and running ...)
355 * SMP rescheduling is done in 2 passes:
356 * - pass #1: faster: 'quick decisions'
357 * - pass #2: slower: 'lets try and find a suitable CPU'
361 * Pass #1. (subtle. We might be in the middle of __switch_to, so
362 * to preserve scheduling atomicity we have to use cpu_curr)
364 if ((p
->processor
== cpu
) && related(cpu_curr(cpu
), p
))
370 reschedule_idle_slow(p
);
376 * This has to add the process to the _beginning_ of the
377 * run-queue, not the end. See the comment about "This is
378 * subtle" in the scheduler proper..
380 static inline void add_to_runqueue(struct task_struct
* p
)
382 list_add(&p
->run_list
, &runqueue_head
);
386 static inline void move_last_runqueue(struct task_struct
* p
)
388 list_del(&p
->run_list
);
389 list_add_tail(&p
->run_list
, &runqueue_head
);
392 static inline void move_first_runqueue(struct task_struct
* p
)
394 list_del(&p
->run_list
);
395 list_add(&p
->run_list
, &runqueue_head
);
399 * Wake up a process. Put it on the run-queue if it's not
400 * already there. The "current" process is always on the
401 * run-queue (except when the actual re-schedule is in
402 * progress), and as such you're allowed to do the simpler
403 * "current->state = TASK_RUNNING" to mark yourself runnable
404 * without the overhead of this.
406 void wake_up_process(struct task_struct
* p
)
411 * We want the common case fall through straight, thus the goto.
413 spin_lock_irqsave(&runqueue_lock
, flags
);
414 p
->state
= TASK_RUNNING
;
415 if (task_on_runqueue(p
))
418 spin_unlock_irqrestore(&runqueue_lock
, flags
);
423 spin_unlock_irqrestore(&runqueue_lock
, flags
);
426 static void process_timeout(unsigned long __data
)
428 struct task_struct
* p
= (struct task_struct
*) __data
;
438 #define TVN_SIZE (1 << TVN_BITS)
439 #define TVR_SIZE (1 << TVR_BITS)
440 #define TVN_MASK (TVN_SIZE - 1)
441 #define TVR_MASK (TVR_SIZE - 1)
445 struct timer_list
*vec
[TVN_SIZE
];
448 struct timer_vec_root
{
450 struct timer_list
*vec
[TVR_SIZE
];
453 static struct timer_vec tv5
= { 0 };
454 static struct timer_vec tv4
= { 0 };
455 static struct timer_vec tv3
= { 0 };
456 static struct timer_vec tv2
= { 0 };
457 static struct timer_vec_root tv1
= { 0 };
459 static struct timer_vec
* const tvecs
[] = {
460 (struct timer_vec
*)&tv1
, &tv2
, &tv3
, &tv4
, &tv5
463 #define NOOF_TVECS (sizeof(tvecs) / sizeof(tvecs[0]))
465 static unsigned long timer_jiffies
= 0;
467 static inline void insert_timer(struct timer_list
*timer
,
468 struct timer_list
**vec
, int idx
)
470 if ((timer
->next
= vec
[idx
]))
471 vec
[idx
]->prev
= timer
;
473 timer
->prev
= (struct timer_list
*)&vec
[idx
];
476 static inline void internal_add_timer(struct timer_list
*timer
)
479 * must be cli-ed when calling this
481 unsigned long expires
= timer
->expires
;
482 unsigned long idx
= expires
- timer_jiffies
;
484 if (idx
< TVR_SIZE
) {
485 int i
= expires
& TVR_MASK
;
486 insert_timer(timer
, tv1
.vec
, i
);
487 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
488 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
489 insert_timer(timer
, tv2
.vec
, i
);
490 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
491 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
492 insert_timer(timer
, tv3
.vec
, i
);
493 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
494 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
495 insert_timer(timer
, tv4
.vec
, i
);
496 } else if ((signed long) idx
< 0) {
497 /* can happen if you add a timer with expires == jiffies,
498 * or you set a timer to go off in the past
500 insert_timer(timer
, tv1
.vec
, tv1
.index
);
501 } else if (idx
<= 0xffffffffUL
) {
502 int i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
503 insert_timer(timer
, tv5
.vec
, i
);
505 /* Can only get here on architectures with 64-bit jiffies */
506 timer
->next
= timer
->prev
= timer
;
510 spinlock_t timerlist_lock
= SPIN_LOCK_UNLOCKED
;
512 void add_timer(struct timer_list
*timer
)
516 spin_lock_irqsave(&timerlist_lock
, flags
);
519 internal_add_timer(timer
);
521 spin_unlock_irqrestore(&timerlist_lock
, flags
);
525 printk("bug: kernel timer added twice at %p.\n",
526 __builtin_return_address(0));
530 static inline int detach_timer(struct timer_list
*timer
)
532 struct timer_list
*prev
= timer
->prev
;
534 struct timer_list
*next
= timer
->next
;
543 void mod_timer(struct timer_list
*timer
, unsigned long expires
)
547 spin_lock_irqsave(&timerlist_lock
, flags
);
548 timer
->expires
= expires
;
550 internal_add_timer(timer
);
551 spin_unlock_irqrestore(&timerlist_lock
, flags
);
554 int del_timer(struct timer_list
* timer
)
559 spin_lock_irqsave(&timerlist_lock
, flags
);
560 ret
= detach_timer(timer
);
561 timer
->next
= timer
->prev
= 0;
562 spin_unlock_irqrestore(&timerlist_lock
, flags
);
566 signed long schedule_timeout(signed long timeout
)
568 struct timer_list timer
;
569 unsigned long expire
;
573 case MAX_SCHEDULE_TIMEOUT
:
575 * These two special cases are useful to be comfortable
576 * in the caller. Nothing more. We could take
577 * MAX_SCHEDULE_TIMEOUT from one of the negative value
578 * but I' d like to return a valid offset (>=0) to allow
579 * the caller to do everything it want with the retval.
585 * Another bit of PARANOID. Note that the retval will be
586 * 0 since no piece of kernel is supposed to do a check
587 * for a negative retval of schedule_timeout() (since it
588 * should never happens anyway). You just have the printk()
589 * that will tell you if something is gone wrong and where.
593 printk(KERN_ERR
"schedule_timeout: wrong timeout "
594 "value %lx from %p\n", timeout
,
595 __builtin_return_address(0));
600 expire
= timeout
+ jiffies
;
603 timer
.expires
= expire
;
604 timer
.data
= (unsigned long) current
;
605 timer
.function
= process_timeout
;
611 timeout
= expire
- jiffies
;
614 return timeout
< 0 ? 0 : timeout
;
618 * schedule_tail() is getting called from the fork return path. This
619 * cleans up all remaining scheduler things, without impacting the
622 static inline void __schedule_tail(struct task_struct
*prev
)
624 if (!current
->active_mm
) BUG();
627 struct mm_struct
*mm
= prev
->active_mm
;
629 prev
->active_mm
= NULL
;
634 if ((prev
->state
== TASK_RUNNING
) &&
635 (prev
!= idle_task(smp_processor_id())))
636 reschedule_idle(prev
);
642 void schedule_tail(struct task_struct
*prev
)
644 __schedule_tail(prev
);
648 * 'schedule()' is the scheduler function. It's a very simple and nice
649 * scheduler: it's not perfect, but certainly works for most things.
651 * The goto is "interesting".
653 * NOTE!! Task 0 is the 'idle' task, which gets called when no other
654 * tasks can run. It can not be killed, and it cannot sleep. The 'state'
655 * information in task[0] is never used.
657 asmlinkage
void schedule(void)
659 struct schedule_data
* sched_data
;
660 struct task_struct
*prev
, *next
, *p
;
661 struct list_head
*tmp
;
664 if (!current
->active_mm
) BUG();
666 goto handle_tq_scheduler
;
670 this_cpu
= prev
->processor
;
673 goto scheduling_in_interrupt
;
675 release_kernel_lock(prev
, this_cpu
);
677 /* Do "administrative" work here while we don't hold any locks */
678 if (bh_mask
& bh_active
)
683 * 'sched_data' is protected by the fact that we can run
684 * only one process per CPU.
686 sched_data
= & aligned_data
[this_cpu
].schedule_data
;
688 spin_lock_irq(&runqueue_lock
);
690 /* move an exhausted RR process to be last.. */
691 if (prev
->policy
== SCHED_RR
)
695 switch (prev
->state
) {
696 case TASK_INTERRUPTIBLE
:
697 if (signal_pending(prev
)) {
698 prev
->state
= TASK_RUNNING
;
702 del_from_runqueue(prev
);
705 prev
->need_resched
= 0;
708 * this is the scheduler proper:
713 * Default process to select..
715 next
= idle_task(this_cpu
);
717 if (prev
->state
== TASK_RUNNING
)
721 tmp
= runqueue_head
.next
;
722 while (tmp
!= &runqueue_head
) {
723 p
= list_entry(tmp
, struct task_struct
, run_list
);
724 if (can_schedule(p
)) {
725 int weight
= goodness(p
, this_cpu
, prev
->active_mm
);
727 c
= weight
, next
= p
;
732 /* Do we need to re-calculate counters? */
736 * from this point on nothing can prevent us from
737 * switching to the next task, save this fact in
740 sched_data
->curr
= next
;
743 next
->processor
= this_cpu
;
745 spin_unlock_irq(&runqueue_lock
);
752 * maintain the per-process 'average timeslice' value.
753 * (this has to be recalculated even if we reschedule to
754 * the same process) Currently this is only used on SMP,
755 * and it's approximate, so we do not have to maintain
756 * it while holding the runqueue spinlock.
759 cycles_t t
, this_slice
;
762 this_slice
= t
- sched_data
->last_schedule
;
763 sched_data
->last_schedule
= t
;
766 * Exponentially fading average calculation, with
767 * some weight so it doesnt get fooled easily by
768 * smaller irregularities.
770 prev
->avg_slice
= (this_slice
*1 + prev
->avg_slice
*1)/2;
774 * We drop the scheduler lock early (it's a global spinlock),
775 * thus we have to lock the previous process from getting
776 * rescheduled during switch_to().
781 kstat
.context_swtch
++;
783 * there are 3 processes which are affected by a context switch:
785 * prev == .... ==> (last => next)
787 * It's the 'much more previous' 'prev' that is on next's stack,
788 * but prev is set to (the just run) 'last' process by switch_to().
789 * This might sound slightly confusing but makes tons of sense.
792 struct mm_struct
*mm
= next
->mm
;
794 mm
= prev
->active_mm
;
795 set_mmu_context(prev
,next
);
796 if (next
->active_mm
) BUG();
797 next
->active_mm
= mm
;
798 atomic_inc(&mm
->mm_count
);
802 get_mmu_context(next
);
803 switch_to(prev
, next
, prev
);
804 __schedule_tail(prev
);
807 reacquire_kernel_lock(current
);
812 struct task_struct
*p
;
813 spin_unlock_irq(&runqueue_lock
);
814 read_lock(&tasklist_lock
);
816 p
->counter
= (p
->counter
>> 1) + p
->priority
;
817 read_unlock(&tasklist_lock
);
818 spin_lock_irq(&runqueue_lock
);
820 goto repeat_schedule
;
823 c
= prev_goodness(prev
, this_cpu
, prev
->active_mm
);
825 goto still_running_back
;
832 run_task_queue(&tq_scheduler
);
833 goto tq_scheduler_back
;
836 if (!prev
->counter
) {
837 prev
->counter
= prev
->priority
;
838 move_last_runqueue(prev
);
842 scheduling_in_interrupt
:
843 printk("Scheduling in interrupt\n");
848 void __wake_up(wait_queue_head_t
*q
, unsigned int mode
)
850 struct list_head
*tmp
, *head
;
851 struct task_struct
*p
;
857 wq_write_lock_irqsave(&q
->lock
, flags
);
860 CHECK_MAGIC_WQHEAD(q
);
863 head
= &q
->task_list
;
865 if (!head
->next
|| !head
->prev
)
869 while (tmp
!= head
) {
871 wait_queue_t
*curr
= list_entry(tmp
, wait_queue_t
, task_list
);
876 CHECK_MAGIC(curr
->__magic
);
882 curr
->__waker
= (long)__builtin_return_address(0);
885 if (state
& TASK_EXCLUSIVE
)
889 wq_write_unlock_irqrestore(&q
->lock
, flags
);
895 * Semaphores are implemented using a two-way counter:
896 * The "count" variable is decremented for each process
897 * that tries to sleep, while the "waking" variable is
898 * incremented when the "up()" code goes to wake up waiting
901 * Notably, the inline "up()" and "down()" functions can
902 * efficiently test if they need to do any extra work (up
903 * needs to do something only if count was negative before
904 * the increment operation.
906 * waking_non_zero() (from asm/semaphore.h) must execute
909 * When __up() is called, the count was negative before
910 * incrementing it, and we need to wake up somebody.
912 * This routine adds one to the count of processes that need to
913 * wake up and exit. ALL waiting processes actually wake up but
914 * only the one that gets to the "waking" field first will gate
915 * through and acquire the semaphore. The others will go back
918 * Note that these functions are only called when there is
919 * contention on the lock, and as such all this is the
920 * "non-critical" part of the whole semaphore business. The
921 * critical part is the inline stuff in <asm/semaphore.h>
922 * where we want to avoid any extra jumps and calls.
924 void __up(struct semaphore
*sem
)
931 * Perform the "down" function. Return zero for semaphore acquired,
932 * return negative for signalled out of the function.
934 * If called from __down, the return is ignored and the wait loop is
935 * not interruptible. This means that a task waiting on a semaphore
936 * using "down()" cannot be killed until someone does an "up()" on
939 * If called from __down_interruptible, the return value gets checked
940 * upon return. If the return value is negative then the task continues
941 * with the negative value in the return register (it can be tested by
944 * Either form may be used in conjunction with "up()".
949 struct task_struct *tsk = current; \
951 init_waitqueue_entry(&wait, tsk);
953 #define DOWN_HEAD(task_state) \
956 tsk->state = (task_state); \
957 add_wait_queue(&sem->wait, &wait); \
960 * Ok, we're set up. sem->count is known to be less than zero \
963 * We can let go the lock for purposes of waiting. \
964 * We re-acquire it after awaking so as to protect \
965 * all semaphore operations. \
967 * If "up()" is called before we call waking_non_zero() then \
968 * we will catch it right away. If it is called later then \
969 * we will have to go through a wakeup cycle to catch it. \
971 * Multiple waiters contend for the semaphore lock to see \
972 * who gets to gate through and who has to wait some more. \
976 #define DOWN_TAIL(task_state) \
977 tsk->state = (task_state); \
979 tsk->state = TASK_RUNNING; \
980 remove_wait_queue(&sem->wait, &wait);
982 void __down(struct semaphore
* sem
)
985 DOWN_HEAD(TASK_UNINTERRUPTIBLE
)
986 if (waking_non_zero(sem
))
989 DOWN_TAIL(TASK_UNINTERRUPTIBLE
)
992 int __down_interruptible(struct semaphore
* sem
)
996 DOWN_HEAD(TASK_INTERRUPTIBLE
)
998 ret
= waking_non_zero_interruptible(sem
, tsk
);
1002 /* ret != 0 only if we get interrupted -arca */
1007 DOWN_TAIL(TASK_INTERRUPTIBLE
)
1011 int __down_trylock(struct semaphore
* sem
)
1013 return waking_non_zero_trylock(sem
);
1016 #define SLEEP_ON_VAR \
1017 unsigned long flags; \
1018 wait_queue_t wait; \
1019 init_waitqueue_entry(&wait, current);
1021 #define SLEEP_ON_HEAD \
1022 wq_write_lock_irqsave(&q->lock,flags); \
1023 __add_wait_queue(q, &wait); \
1024 wq_write_unlock(&q->lock);
1026 #define SLEEP_ON_TAIL \
1027 wq_write_lock_irq(&q->lock); \
1028 __remove_wait_queue(q, &wait); \
1029 wq_write_unlock_irqrestore(&q->lock,flags);
1031 void interruptible_sleep_on(wait_queue_head_t
*q
)
1035 current
->state
= TASK_INTERRUPTIBLE
;
1042 long interruptible_sleep_on_timeout(wait_queue_head_t
*q
, long timeout
)
1046 current
->state
= TASK_INTERRUPTIBLE
;
1049 timeout
= schedule_timeout(timeout
);
1055 void sleep_on(wait_queue_head_t
*q
)
1059 current
->state
= TASK_UNINTERRUPTIBLE
;
1066 long sleep_on_timeout(wait_queue_head_t
*q
, long timeout
)
1070 current
->state
= TASK_UNINTERRUPTIBLE
;
1073 timeout
= schedule_timeout(timeout
);
1079 void scheduling_functions_end_here(void) { }
1081 static inline void cascade_timers(struct timer_vec
*tv
)
1083 /* cascade all the timers from tv up one level */
1084 struct timer_list
*timer
;
1085 timer
= tv
->vec
[tv
->index
];
1087 * We are removing _all_ timers from the list, so we don't have to
1088 * detach them individually, just clear the list afterwards.
1091 struct timer_list
*tmp
= timer
;
1092 timer
= timer
->next
;
1093 internal_add_timer(tmp
);
1095 tv
->vec
[tv
->index
] = NULL
;
1096 tv
->index
= (tv
->index
+ 1) & TVN_MASK
;
1099 static inline void run_timer_list(void)
1101 spin_lock_irq(&timerlist_lock
);
1102 while ((long)(jiffies
- timer_jiffies
) >= 0) {
1103 struct timer_list
*timer
;
1107 cascade_timers(tvecs
[n
]);
1108 } while (tvecs
[n
]->index
== 1 && ++n
< NOOF_TVECS
);
1110 while ((timer
= tv1
.vec
[tv1
.index
])) {
1111 void (*fn
)(unsigned long) = timer
->function
;
1112 unsigned long data
= timer
->data
;
1113 detach_timer(timer
);
1114 timer
->next
= timer
->prev
= NULL
;
1115 spin_unlock_irq(&timerlist_lock
);
1117 spin_lock_irq(&timerlist_lock
);
1120 tv1
.index
= (tv1
.index
+ 1) & TVR_MASK
;
1122 spin_unlock_irq(&timerlist_lock
);
1126 static inline void run_old_timers(void)
1128 struct timer_struct
*tp
;
1131 for (mask
= 1, tp
= timer_table
+0 ; mask
; tp
++,mask
+= mask
) {
1132 if (mask
> timer_active
)
1134 if (!(mask
& timer_active
))
1136 if (time_after(tp
->expires
, jiffies
))
1138 timer_active
&= ~mask
;
1144 spinlock_t tqueue_lock
;
1146 void tqueue_bh(void)
1148 run_task_queue(&tq_timer
);
1151 void immediate_bh(void)
1153 run_task_queue(&tq_immediate
);
1156 unsigned long timer_active
= 0;
1157 struct timer_struct timer_table
[32];
1160 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1161 * imply that avenrun[] is the standard name for this kind of thing.
1162 * Nothing else seems to be standardized: the fractional size etc
1163 * all seem to differ on different machines.
1165 unsigned long avenrun
[3] = { 0,0,0 };
1168 * Nr of active tasks - counted in fixed-point numbers
1170 static unsigned long count_active_tasks(void)
1172 struct task_struct
*p
;
1173 unsigned long nr
= 0;
1175 read_lock(&tasklist_lock
);
1177 if ((p
->state
== TASK_RUNNING
||
1178 (p
->state
& TASK_UNINTERRUPTIBLE
) ||
1179 (p
->state
& TASK_SWAPPING
)))
1182 read_unlock(&tasklist_lock
);
1186 static inline void calc_load(unsigned long ticks
)
1188 unsigned long active_tasks
; /* fixed-point */
1189 static int count
= LOAD_FREQ
;
1194 active_tasks
= count_active_tasks();
1195 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
1196 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
1197 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
1202 * this routine handles the overflow of the microsecond field
1204 * The tricky bits of code to handle the accurate clock support
1205 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
1206 * They were originally developed for SUN and DEC kernels.
1207 * All the kudos should go to Dave for this stuff.
1210 static void second_overflow(void)
1214 /* Bump the maxerror field */
1215 time_maxerror
+= time_tolerance
>> SHIFT_USEC
;
1216 if ( time_maxerror
> NTP_PHASE_LIMIT
) {
1217 time_maxerror
= NTP_PHASE_LIMIT
;
1218 time_status
|= STA_UNSYNC
;
1222 * Leap second processing. If in leap-insert state at
1223 * the end of the day, the system clock is set back one
1224 * second; if in leap-delete state, the system clock is
1225 * set ahead one second. The microtime() routine or
1226 * external clock driver will insure that reported time
1227 * is always monotonic. The ugly divides should be
1230 switch (time_state
) {
1233 if (time_status
& STA_INS
)
1234 time_state
= TIME_INS
;
1235 else if (time_status
& STA_DEL
)
1236 time_state
= TIME_DEL
;
1240 if (xtime
.tv_sec
% 86400 == 0) {
1242 time_state
= TIME_OOP
;
1243 printk(KERN_NOTICE
"Clock: inserting leap second 23:59:60 UTC\n");
1248 if ((xtime
.tv_sec
+ 1) % 86400 == 0) {
1250 time_state
= TIME_WAIT
;
1251 printk(KERN_NOTICE
"Clock: deleting leap second 23:59:59 UTC\n");
1256 time_state
= TIME_WAIT
;
1260 if (!(time_status
& (STA_INS
| STA_DEL
)))
1261 time_state
= TIME_OK
;
1265 * Compute the phase adjustment for the next second. In
1266 * PLL mode, the offset is reduced by a fixed factor
1267 * times the time constant. In FLL mode the offset is
1268 * used directly. In either mode, the maximum phase
1269 * adjustment for each second is clamped so as to spread
1270 * the adjustment over not more than the number of
1271 * seconds between updates.
1273 if (time_offset
< 0) {
1274 ltemp
= -time_offset
;
1275 if (!(time_status
& STA_FLL
))
1276 ltemp
>>= SHIFT_KG
+ time_constant
;
1277 if (ltemp
> (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
)
1278 ltemp
= (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
;
1279 time_offset
+= ltemp
;
1280 time_adj
= -ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
1282 ltemp
= time_offset
;
1283 if (!(time_status
& STA_FLL
))
1284 ltemp
>>= SHIFT_KG
+ time_constant
;
1285 if (ltemp
> (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
)
1286 ltemp
= (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
;
1287 time_offset
-= ltemp
;
1288 time_adj
= ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
1292 * Compute the frequency estimate and additional phase
1293 * adjustment due to frequency error for the next
1294 * second. When the PPS signal is engaged, gnaw on the
1295 * watchdog counter and update the frequency computed by
1296 * the pll and the PPS signal.
1299 if (pps_valid
== PPS_VALID
) { /* PPS signal lost */
1300 pps_jitter
= MAXTIME
;
1301 pps_stabil
= MAXFREQ
;
1302 time_status
&= ~(STA_PPSSIGNAL
| STA_PPSJITTER
|
1303 STA_PPSWANDER
| STA_PPSERROR
);
1305 ltemp
= time_freq
+ pps_freq
;
1307 time_adj
-= -ltemp
>>
1308 (SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
);
1310 time_adj
+= ltemp
>>
1311 (SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
);
1314 /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
1315 * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
1318 time_adj
-= (-time_adj
>> 2) + (-time_adj
>> 5);
1320 time_adj
+= (time_adj
>> 2) + (time_adj
>> 5);
1324 /* in the NTP reference this is called "hardclock()" */
1325 static void update_wall_time_one_tick(void)
1327 if ( (time_adjust_step
= time_adjust
) != 0 ) {
1328 /* We are doing an adjtime thing.
1330 * Prepare time_adjust_step to be within bounds.
1331 * Note that a positive time_adjust means we want the clock
1334 * Limit the amount of the step to be in the range
1335 * -tickadj .. +tickadj
1337 if (time_adjust
> tickadj
)
1338 time_adjust_step
= tickadj
;
1339 else if (time_adjust
< -tickadj
)
1340 time_adjust_step
= -tickadj
;
1342 /* Reduce by this step the amount of time left */
1343 time_adjust
-= time_adjust_step
;
1345 xtime
.tv_usec
+= tick
+ time_adjust_step
;
1347 * Advance the phase, once it gets to one microsecond, then
1348 * advance the tick more.
1350 time_phase
+= time_adj
;
1351 if (time_phase
<= -FINEUSEC
) {
1352 long ltemp
= -time_phase
>> SHIFT_SCALE
;
1353 time_phase
+= ltemp
<< SHIFT_SCALE
;
1354 xtime
.tv_usec
-= ltemp
;
1356 else if (time_phase
>= FINEUSEC
) {
1357 long ltemp
= time_phase
>> SHIFT_SCALE
;
1358 time_phase
-= ltemp
<< SHIFT_SCALE
;
1359 xtime
.tv_usec
+= ltemp
;
1364 * Using a loop looks inefficient, but "ticks" is
1365 * usually just one (we shouldn't be losing ticks,
1366 * we're doing this this way mainly for interrupt
1367 * latency reasons, not because we think we'll
1368 * have lots of lost timer ticks
1370 static void update_wall_time(unsigned long ticks
)
1374 update_wall_time_one_tick();
1377 if (xtime
.tv_usec
>= 1000000) {
1378 xtime
.tv_usec
-= 1000000;
1384 static inline void do_process_times(struct task_struct
*p
,
1385 unsigned long user
, unsigned long system
)
1389 psecs
= (p
->times
.tms_utime
+= user
);
1390 psecs
+= (p
->times
.tms_stime
+= system
);
1391 if (psecs
/ HZ
> p
->rlim
[RLIMIT_CPU
].rlim_cur
) {
1392 /* Send SIGXCPU every second.. */
1394 send_sig(SIGXCPU
, p
, 1);
1395 /* and SIGKILL when we go over max.. */
1396 if (psecs
/ HZ
> p
->rlim
[RLIMIT_CPU
].rlim_max
)
1397 send_sig(SIGKILL
, p
, 1);
1401 static inline void do_it_virt(struct task_struct
* p
, unsigned long ticks
)
1403 unsigned long it_virt
= p
->it_virt_value
;
1406 if (it_virt
<= ticks
) {
1407 it_virt
= ticks
+ p
->it_virt_incr
;
1408 send_sig(SIGVTALRM
, p
, 1);
1410 p
->it_virt_value
= it_virt
- ticks
;
1414 static inline void do_it_prof(struct task_struct
* p
, unsigned long ticks
)
1416 unsigned long it_prof
= p
->it_prof_value
;
1419 if (it_prof
<= ticks
) {
1420 it_prof
= ticks
+ p
->it_prof_incr
;
1421 send_sig(SIGPROF
, p
, 1);
1423 p
->it_prof_value
= it_prof
- ticks
;
1427 void update_one_process(struct task_struct
*p
,
1428 unsigned long ticks
, unsigned long user
, unsigned long system
, int cpu
)
1430 p
->per_cpu_utime
[cpu
] += user
;
1431 p
->per_cpu_stime
[cpu
] += system
;
1432 do_process_times(p
, user
, system
);
1433 do_it_virt(p
, user
);
1434 do_it_prof(p
, ticks
);
1437 static void update_process_times(unsigned long ticks
, unsigned long system
)
1440 * SMP does this on a per-CPU basis elsewhere
1443 struct task_struct
* p
= current
;
1444 unsigned long user
= ticks
- system
;
1446 p
->counter
-= ticks
;
1447 if (p
->counter
<= 0) {
1449 p
->need_resched
= 1;
1451 if (p
->priority
< DEF_PRIORITY
)
1452 kstat
.cpu_nice
+= user
;
1454 kstat
.cpu_user
+= user
;
1455 kstat
.cpu_system
+= system
;
1457 update_one_process(p
, ticks
, user
, system
, 0);
1461 volatile unsigned long lost_ticks
= 0;
1462 static unsigned long lost_ticks_system
= 0;
1465 * This spinlock protect us from races in SMP while playing with xtime. -arca
1467 rwlock_t xtime_lock
= RW_LOCK_UNLOCKED
;
1469 static inline void update_times(void)
1471 unsigned long ticks
;
1474 * update_times() is run from the raw timer_bh handler so we
1475 * just know that the irqs are locally enabled and so we don't
1476 * need to save/restore the flags of the local CPU here. -arca
1478 write_lock_irq(&xtime_lock
);
1484 unsigned long system
;
1485 system
= xchg(&lost_ticks_system
, 0);
1488 update_wall_time(ticks
);
1489 write_unlock_irq(&xtime_lock
);
1491 update_process_times(ticks
, system
);
1494 write_unlock_irq(&xtime_lock
);
1497 static void timer_bh(void)
1504 void do_timer(struct pt_regs
* regs
)
1506 (*(unsigned long *)&jiffies
)++;
1509 if (!user_mode(regs
))
1510 lost_ticks_system
++;
1518 * For backwards compatibility? This can be done in libc so Alpha
1519 * and all newer ports shouldn't need it.
1521 asmlinkage
unsigned int sys_alarm(unsigned int seconds
)
1523 struct itimerval it_new
, it_old
;
1524 unsigned int oldalarm
;
1526 it_new
.it_interval
.tv_sec
= it_new
.it_interval
.tv_usec
= 0;
1527 it_new
.it_value
.tv_sec
= seconds
;
1528 it_new
.it_value
.tv_usec
= 0;
1529 do_setitimer(ITIMER_REAL
, &it_new
, &it_old
);
1530 oldalarm
= it_old
.it_value
.tv_sec
;
1531 /* ehhh.. We can't return 0 if we have an alarm pending.. */
1532 /* And we'd better return too much than too little anyway */
1533 if (it_old
.it_value
.tv_usec
)
1539 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1540 * should be moved into arch/i386 instead?
1543 asmlinkage
int sys_getpid(void)
1545 /* This is SMP safe - current->pid doesn't change */
1546 return current
->pid
;
1550 * This is not strictly SMP safe: p_opptr could change
1551 * from under us. However, rather than getting any lock
1552 * we can use an optimistic algorithm: get the parent
1553 * pid, and go back and check that the parent is still
1554 * the same. If it has changed (which is extremely unlikely
1555 * indeed), we just try again..
1557 * NOTE! This depends on the fact that even if we _do_
1558 * get an old value of "parent", we can happily dereference
1559 * the pointer: we just can't necessarily trust the result
1560 * until we know that the parent pointer is valid.
1562 * The "mb()" macro is a memory barrier - a synchronizing
1563 * event. It also makes sure that gcc doesn't optimize
1564 * away the necessary memory references.. The barrier doesn't
1565 * have to have all that strong semantics: on x86 we don't
1566 * really require a synchronizing instruction, for example.
1567 * The barrier is more important for code generation than
1568 * for any real memory ordering semantics (even if there is
1569 * a small window for a race, using the old pointer is
1570 * harmless for a while).
1572 asmlinkage
int sys_getppid(void)
1575 struct task_struct
* me
= current
;
1576 struct task_struct
* parent
;
1578 parent
= me
->p_opptr
;
1583 struct task_struct
*old
= parent
;
1585 parent
= me
->p_opptr
;
1595 asmlinkage
int sys_getuid(void)
1597 /* Only we change this so SMP safe */
1598 return current
->uid
;
1601 asmlinkage
int sys_geteuid(void)
1603 /* Only we change this so SMP safe */
1604 return current
->euid
;
1607 asmlinkage
int sys_getgid(void)
1609 /* Only we change this so SMP safe */
1610 return current
->gid
;
1613 asmlinkage
int sys_getegid(void)
1615 /* Only we change this so SMP safe */
1616 return current
->egid
;
1620 * This has been replaced by sys_setpriority. Maybe it should be
1621 * moved into the arch dependent tree for those ports that require
1622 * it for backward compatibility?
1625 asmlinkage
int sys_nice(int increment
)
1627 unsigned long newprio
;
1631 * Setpriority might change our priority at the same moment.
1632 * We don't have to worry. Conceptually one call occurs first
1633 * and we have a single winner.
1636 newprio
= increment
;
1637 if (increment
< 0) {
1638 if (!capable(CAP_SYS_NICE
))
1640 newprio
= -increment
;
1647 * do a "normalization" of the priority (traditionally
1648 * Unix nice values are -20 to 20; Linux doesn't really
1649 * use that kind of thing, but uses the length of the
1650 * timeslice instead (default 200 ms). The rounding is
1651 * why we want to avoid negative values.
1653 newprio
= (newprio
* DEF_PRIORITY
+ 10) / 20;
1654 increment
= newprio
;
1656 increment
= -increment
;
1658 * Current->priority can change between this point
1659 * and the assignment. We are assigning not doing add/subs
1660 * so thats ok. Conceptually a process might just instantaneously
1661 * read the value we stomp over. I don't think that is an issue
1662 * unless posix makes it one. If so we can loop on changes
1663 * to current->priority.
1665 newprio
= current
->priority
- increment
;
1666 if ((signed) newprio
< 1)
1668 if (newprio
> DEF_PRIORITY
*2)
1669 newprio
= DEF_PRIORITY
*2;
1670 current
->priority
= newprio
;
1676 static inline struct task_struct
*find_process_by_pid(pid_t pid
)
1678 struct task_struct
*tsk
= current
;
1681 tsk
= find_task_by_pid(pid
);
1685 static int setscheduler(pid_t pid
, int policy
,
1686 struct sched_param
*param
)
1688 struct sched_param lp
;
1689 struct task_struct
*p
;
1693 if (!param
|| pid
< 0)
1697 if (copy_from_user(&lp
, param
, sizeof(struct sched_param
)))
1701 * We play safe to avoid deadlocks.
1703 spin_lock_irq(&runqueue_lock
);
1704 read_lock(&tasklist_lock
);
1706 p
= find_process_by_pid(pid
);
1716 if (policy
!= SCHED_FIFO
&& policy
!= SCHED_RR
&&
1717 policy
!= SCHED_OTHER
)
1722 * Valid priorities for SCHED_FIFO and SCHED_RR are 1..99, valid
1723 * priority for SCHED_OTHER is 0.
1726 if (lp
.sched_priority
< 0 || lp
.sched_priority
> 99)
1728 if ((policy
== SCHED_OTHER
) != (lp
.sched_priority
== 0))
1732 if ((policy
== SCHED_FIFO
|| policy
== SCHED_RR
) &&
1733 !capable(CAP_SYS_NICE
))
1735 if ((current
->euid
!= p
->euid
) && (current
->euid
!= p
->uid
) &&
1736 !capable(CAP_SYS_NICE
))
1741 p
->rt_priority
= lp
.sched_priority
;
1742 if (task_on_runqueue(p
))
1743 move_first_runqueue(p
);
1745 current
->need_resched
= 1;
1748 read_unlock(&tasklist_lock
);
1749 spin_unlock_irq(&runqueue_lock
);
1755 asmlinkage
int sys_sched_setscheduler(pid_t pid
, int policy
,
1756 struct sched_param
*param
)
1758 return setscheduler(pid
, policy
, param
);
1761 asmlinkage
int sys_sched_setparam(pid_t pid
, struct sched_param
*param
)
1763 return setscheduler(pid
, -1, param
);
1766 asmlinkage
int sys_sched_getscheduler(pid_t pid
)
1768 struct task_struct
*p
;
1775 read_lock(&tasklist_lock
);
1778 p
= find_process_by_pid(pid
);
1785 read_unlock(&tasklist_lock
);
1791 asmlinkage
int sys_sched_getparam(pid_t pid
, struct sched_param
*param
)
1793 struct task_struct
*p
;
1794 struct sched_param lp
;
1798 if (!param
|| pid
< 0)
1801 read_lock(&tasklist_lock
);
1802 p
= find_process_by_pid(pid
);
1806 lp
.sched_priority
= p
->rt_priority
;
1807 read_unlock(&tasklist_lock
);
1810 * This one might sleep, we cannot do it with a spinlock held ...
1812 retval
= copy_to_user(param
, &lp
, sizeof(*param
)) ? -EFAULT
: 0;
1818 read_unlock(&tasklist_lock
);
1822 asmlinkage
int sys_sched_yield(void)
1824 spin_lock_irq(&runqueue_lock
);
1825 if (current
->policy
== SCHED_OTHER
)
1826 current
->policy
|= SCHED_YIELD
;
1827 current
->need_resched
= 1;
1828 move_last_runqueue(current
);
1829 spin_unlock_irq(&runqueue_lock
);
1833 asmlinkage
int sys_sched_get_priority_max(int policy
)
1849 asmlinkage
int sys_sched_get_priority_min(int policy
)
1864 asmlinkage
int sys_sched_rr_get_interval(pid_t pid
, struct timespec
*interval
)
1870 if (copy_to_user(interval
, &t
, sizeof(struct timespec
)))
1875 asmlinkage
int sys_nanosleep(struct timespec
*rqtp
, struct timespec
*rmtp
)
1878 unsigned long expire
;
1880 if(copy_from_user(&t
, rqtp
, sizeof(struct timespec
)))
1883 if (t
.tv_nsec
>= 1000000000L || t
.tv_nsec
< 0 || t
.tv_sec
< 0)
1887 if (t
.tv_sec
== 0 && t
.tv_nsec
<= 2000000L &&
1888 current
->policy
!= SCHED_OTHER
)
1891 * Short delay requests up to 2 ms will be handled with
1892 * high precision by a busy wait for all real-time processes.
1894 * Its important on SMP not to do this holding locks.
1896 udelay((t
.tv_nsec
+ 999) / 1000);
1900 expire
= timespec_to_jiffies(&t
) + (t
.tv_sec
|| t
.tv_nsec
);
1902 current
->state
= TASK_INTERRUPTIBLE
;
1903 expire
= schedule_timeout(expire
);
1907 jiffies_to_timespec(expire
, &t
);
1908 if (copy_to_user(rmtp
, &t
, sizeof(struct timespec
)))
1916 static void show_task(struct task_struct
* p
)
1918 unsigned long free
= 0;
1920 static const char * stat_nam
[] = { "R", "S", "D", "Z", "T", "W" };
1922 printk("%-8s ", p
->comm
);
1923 state
= p
->state
? ffz(~p
->state
) + 1 : 0;
1924 if (((unsigned) state
) < sizeof(stat_nam
)/sizeof(char *))
1925 printk(stat_nam
[state
]);
1928 #if (BITS_PER_LONG == 32)
1930 printk(" current ");
1932 printk(" %08lX ", thread_saved_pc(&p
->thread
));
1935 printk(" current task ");
1937 printk(" %016lx ", thread_saved_pc(&p
->thread
));
1940 unsigned long * n
= (unsigned long *) (p
+1);
1943 free
= (unsigned long) n
- (unsigned long)(p
+1);
1945 printk("%5lu %5d %6d ", free
, p
->pid
, p
->p_pptr
->pid
);
1947 printk("%5d ", p
->p_cptr
->pid
);
1951 printk(" (L-TLB) ");
1953 printk(" (NOTLB) ");
1955 printk("%7d", p
->p_ysptr
->pid
);
1959 printk(" %5d\n", p
->p_osptr
->pid
);
1964 struct signal_queue
*q
;
1965 char s
[sizeof(sigset_t
)*2+1], b
[sizeof(sigset_t
)*2+1];
1967 render_sigset_t(&p
->signal
, s
);
1968 render_sigset_t(&p
->blocked
, b
);
1969 printk(" sig: %d %s %s :", signal_pending(p
), s
, b
);
1970 for (q
= p
->sigqueue
; q
; q
= q
->next
)
1971 printk(" %d", q
->info
.si_signo
);
1976 char * render_sigset_t(sigset_t
*set
, char *buffer
)
1981 if (sigismember(set
, i
+1)) x
|= 1;
1982 if (sigismember(set
, i
+2)) x
|= 2;
1983 if (sigismember(set
, i
+3)) x
|= 4;
1984 if (sigismember(set
, i
+4)) x
|= 8;
1985 *buffer
++ = (x
< 10 ? '0' : 'a' - 10) + x
;
1991 void show_state(void)
1993 struct task_struct
*p
;
1995 #if (BITS_PER_LONG == 32)
1998 printk(" task PC stack pid father child younger older\n");
2002 printk(" task PC stack pid father child younger older\n");
2004 read_lock(&tasklist_lock
);
2007 read_unlock(&tasklist_lock
);
2010 void __init
init_idle(void)
2013 struct schedule_data
* sched_data
;
2014 sched_data
= &aligned_data
[smp_processor_id()].schedule_data
;
2016 if (current
!= &init_task
&& task_on_runqueue(current
)) {
2017 printk("UGH! (%d:%d) was on the runqueue, removing.\n",
2018 smp_processor_id(), current
->pid
);
2019 del_from_runqueue(current
);
2022 sched_data
->curr
= current
;
2023 sched_data
->last_schedule
= t
;
2026 void __init
sched_init(void)
2029 * We have to do a little magic to get the first
2030 * process right in SMP mode.
2032 int cpu
=hard_smp_processor_id();
2035 init_task
.processor
=cpu
;
2037 for(nr
= 0; nr
< PIDHASH_SZ
; nr
++)
2040 init_bh(TIMER_BH
, timer_bh
);
2041 init_bh(TQUEUE_BH
, tqueue_bh
);
2042 init_bh(IMMEDIATE_BH
, immediate_bh
);
2045 * The boot idle thread does lazy MMU switching as well:
2047 atomic_inc(&init_mm
.mm_count
);