root/kernel/sched.c

/* */

DEFINITIONS

1. math_state_restore
2. schedule
3. sys_pause
4. wake_up
5. wake_up_interruptible
6. __sleep_on
7. interruptible_sleep_on
8. sleep_on
9. ticks_to_floppy_on
10. floppy_off
11. do_floppy_timer
12. add_timer
13. count_active_tasks
14. calc_load
15. do_timer
16. sys_alarm
17. sys_getpid
18. sys_getppid
19. sys_getuid
20. sys_geteuid
21. sys_getgid
22. sys_getegid
23. sys_nice
24. show_task
25. show_state
26. sched_init

   1 /*
   2  *  linux/kernel/sched.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6 
   7 /*
   8  * 'sched.c' is the main kernel file. It contains scheduling primitives
   9  * (sleep_on, wakeup, schedule etc) as well as a number of simple system
  10  * call functions (type getpid(), which just extracts a field from
  11  * current-task
  12  */
  13 
  14 #define TIMER_IRQ 0
  15 
  16 #include <linux/config.h>
  17 #include <linux/signal.h>
  18 #include <linux/sched.h>
  19 #include <linux/timer.h>
  20 #include <linux/kernel.h>
  21 #include <linux/sys.h>
  22 #include <linux/fdreg.h>
  23 #include <linux/errno.h>
  24 #include <linux/time.h>
  25 #include <linux/ptrace.h>
  26 
  27 #include <asm/system.h>
  28 #include <asm/io.h>
  29 #include <asm/segment.h>
  30 
  31 int need_resched = 0;
  32 int hard_math = 0;              /* set by boot/head.S */
  33 
  34 unsigned long * prof_buffer = NULL;
  35 unsigned long prof_len = 0;
  36 
  37 #define _S(nr) (1<<((nr)-1))
  38 #define _BLOCKABLE (~(_S(SIGKILL) | _S(SIGSTOP)))
  39 
  40 #define LATCH (1193180/HZ)
  41 
  42 extern void mem_use(void);
  43 
  44 extern int timer_interrupt(void);
  45 extern int system_call(void);
  46 
  47 static unsigned long init_kernel_stack[1024];
  48 static struct task_struct init_task = INIT_TASK;
  49 
  50 unsigned long volatile jiffies=0;
  51 unsigned long startup_time=0;
  52 int jiffies_offset = 0;         /* # clock ticks to add to get "true
  53                                    time".  Should always be less than
  54                                    1 second's worth.  For time fanatics
  55                                    who like to syncronize their machines
  56                                    to WWV :-) */
  57 
  58 struct task_struct *current = &init_task;
  59 struct task_struct *last_task_used_math = NULL;
  60 
  61 struct task_struct * task[NR_TASKS] = {&init_task, };
  62 
  63 long user_stack [ PAGE_SIZE>>2 ] ;
  64 
  65 struct {
  66         long * a;
  67         short b;
  68         } stack_start = { & user_stack [PAGE_SIZE>>2] , 0x10 };
  69 /*
  70  *  'math_state_restore()' saves the current math information in the
  71  * old math state array, and gets the new ones from the current task
  72  */
  73 void math_state_restore()
     /*  */
  74 {
  75         if (last_task_used_math == current)
  76                 return;
  77         if (last_task_used_math) {
  78                 __asm__("fnsave %0"::"m" (last_task_used_math->tss.i387));
  79         }
  80         __asm__("fwait");
  81         last_task_used_math = current;
  82         if (current->used_math) {
  83                 __asm__("frstor %0"::"m" (current->tss.i387));
  84         } else {
  85                 __asm__("fninit"::);
  86                 current->used_math=1;
  87         }
  88 }
  89 
  90 /*
  91  *  'schedule()' is the scheduler function. It's a very simple and nice
  92  * scheduler: it's not perfect, but certainly works for most things.
  93  * The one thing you might take a look at is the signal-handler code here.
  94  *
  95  *   NOTE!!  Task 0 is the 'idle' task, which gets called when no other
  96  * tasks can run. It can not be killed, and it cannot sleep. The 'state'
  97  * information in task[0] is never used.
  98  */
  99 void schedule(void)
     /*  */
 100 {
 101         int i,next,c;
 102         struct task_struct ** p;
 103 
 104 /* check alarm, wake up any interruptible tasks that have got a signal */
 105 
 106         need_resched = 0;
 107         for(p = &LAST_TASK ; p > &FIRST_TASK ; --p) {
 108                 if (!*p || ((*p)->state != TASK_INTERRUPTIBLE))
 109                         continue;
 110                 if ((*p)->timeout && (*p)->timeout < jiffies) {
 111                         (*p)->timeout = 0;
 112                         (*p)->state = TASK_RUNNING;
 113                 } else if ((*p)->signal & ~(*p)->blocked)
 114                         (*p)->state = TASK_RUNNING;
 115         }
 116 
 117 /* this is the scheduler proper: */
 118 
 119         while (1) {
 120                 c = -1;
 121                 next = 0;
 122                 i = NR_TASKS;
 123                 p = &task[NR_TASKS];
 124                 while (--i) {
 125                         if (!*--p)
 126                                 continue;
 127                         if ((*p)->state == TASK_RUNNING && (*p)->counter > c)
 128                                 c = (*p)->counter, next = i;
 129                 }
 130                 if (c)
 131                         break;
 132                 for(p = &LAST_TASK ; p > &FIRST_TASK ; --p)
 133                         if (*p)
 134                                 (*p)->counter = ((*p)->counter >> 1) +
 135                                                 (*p)->priority;
 136         }
 137         sti();
 138         switch_to(next);
 139 }
 140 
 141 int sys_pause(void)
     /*  */
 142 {
 143         unsigned long old_blocked;
 144         unsigned long mask;
 145         struct sigaction * sa = current->sigaction;
 146 
 147         old_blocked = current->blocked;
 148         for (mask=1 ; mask ; sa++,mask += mask)
 149                 if (sa->sa_handler == SIG_IGN)
 150                         current->blocked |= mask;
 151         current->state = TASK_INTERRUPTIBLE;
 152         schedule();
 153         current->blocked = old_blocked;
 154         return -EINTR;
 155 }
 156 
 157 /*
 158  * wake_up doesn't wake up stopped processes - they have to be awakened
 159  * with signals or similar.
 160  *
 161  * Note that this doesn't need cli-sti pairs: interrupts may not change
 162  * the wait-queue structures directly, but only call wake_up() to wake
 163  * a process. The process itself must remove the queue once it has woken.
 164  */
 165 void wake_up(struct wait_queue **q)
     /*  */
 166 {
 167         struct wait_queue *tmp;
 168         struct task_struct * p;
 169 
 170         if (!q || !(tmp = *q))
 171                 return;
 172         do {
 173                 if ((p = tmp->task) != NULL) {
 174                         if ((p->state == TASK_UNINTERRUPTIBLE) ||
 175                             (p->state == TASK_INTERRUPTIBLE)) {
 176                                 p->state = TASK_RUNNING;
 177                                 if (p->counter > current->counter)
 178                                         need_resched = 1;
 179                         }
 180                 }
 181                 if (!tmp->next) {
 182                         printk("wait_queue is bad (eip = %08x)\n",((unsigned long *) q)[-1]);
 183                         printk("        q = %08x\n",q);
 184                         printk("       *q = %08x\n",*q);
 185                         printk("      tmp = %08x\n",tmp);
 186                         break;
 187                 }
 188                 tmp = tmp->next;
 189         } while (tmp != *q);
 190 }
 191 
 192 void wake_up_interruptible(struct wait_queue **q)
     /*  */
 193 {
 194         struct wait_queue *tmp;
 195         struct task_struct * p;
 196 
 197         if (!q || !(tmp = *q))
 198                 return;
 199         do {
 200                 if ((p = tmp->task) != NULL) {
 201                         if (p->state == TASK_INTERRUPTIBLE) {
 202                                 p->state = TASK_RUNNING;
 203                                 if (p->counter > current->counter)
 204                                         need_resched = 1;
 205                         }
 206                 }
 207                 if (!tmp->next) {
 208                         printk("wait_queue is bad (eip = %08x)\n",((unsigned long *) q)[-1]);
 209                         printk("        q = %08x\n",q);
 210                         printk("       *q = %08x\n",*q);
 211                         printk("      tmp = %08x\n",tmp);
 212                         break;
 213                 }
 214                 tmp = tmp->next;
 215         } while (tmp != *q);
 216 }
 217 
 218 static inline void __sleep_on(struct wait_queue **p, int state)
     /*  */
 219 {
 220         unsigned long flags;
 221         struct wait_queue wait = { current, NULL };
 222 
 223         if (!p)
 224                 return;
 225         if (current == task[0])
 226                 panic("task[0] trying to sleep");
 227         current->state = state;
 228         add_wait_queue(p, &wait);
 229         save_flags(flags);
 230         sti();
 231         schedule();
 232         remove_wait_queue(p, &wait);
 233         restore_flags(flags);
 234 }
 235 
 236 void interruptible_sleep_on(struct wait_queue **p)
     /*  */
 237 {
 238         __sleep_on(p,TASK_INTERRUPTIBLE);
 239 }
 240 
 241 void sleep_on(struct wait_queue **p)
     /*  */
 242 {
 243         __sleep_on(p,TASK_UNINTERRUPTIBLE);
 244 }
 245 
 246 /*
 247  * OK, here are some floppy things that shouldn't be in the kernel
 248  * proper. They are here because the floppy needs a timer, and this
 249  * was the easiest way of doing it.
 250  */
 251 static struct wait_queue * wait_motor[4] = {NULL,NULL,NULL,NULL};
 252 static int  mon_timer[4]={0,0,0,0};
 253 static int moff_timer[4]={0,0,0,0};
 254 unsigned char current_DOR = 0x0C;
 255 
 256 int ticks_to_floppy_on(unsigned int nr)
     /*  */
 257 {
 258         extern unsigned char selected;
 259         unsigned char mask = 0x10 << nr;
 260 
 261         if (nr>3)
 262                 panic("floppy_on: nr>3");
 263         moff_timer[nr]=10000;           /* 100 s = very big :-) */
 264         cli();                          /* use floppy_off to turn it off */
 265         mask |= current_DOR;
 266         if (!selected) {
 267                 mask &= 0xFC;
 268                 mask |= nr;
 269         }
 270         if (mask != current_DOR) {
 271                 outb(mask,FD_DOR);
 272                 if ((mask ^ current_DOR) & 0xf0)
 273                         mon_timer[nr] = HZ/2;
 274                 else if (mon_timer[nr] < 2)
 275                         mon_timer[nr] = 2;
 276                 current_DOR = mask;
 277         }
 278         sti();
 279         return mon_timer[nr];
 280 }
 281 
 282 void floppy_off(unsigned int nr)
     /*  */
 283 {
 284         moff_timer[nr]=3*HZ;
 285 }
 286 
 287 void do_floppy_timer(void)
     /*  */
 288 {
 289         int i;
 290         unsigned char mask = 0x10;
 291 
 292         for (i=0 ; i<4 ; i++,mask <<= 1) {
 293                 if (!(mask & current_DOR))
 294                         continue;
 295                 if (mon_timer[i]) {
 296                         if (!--mon_timer[i])
 297                                 wake_up(i+wait_motor);
 298                 } else if (!moff_timer[i]) {
 299                         current_DOR &= ~mask;
 300                         outb(current_DOR,FD_DOR);
 301                 } else
 302                         moff_timer[i]--;
 303         }
 304 }
 305 
 306 #define TIME_REQUESTS 64
 307 
 308 static struct timer_list {
 309         long jiffies;
 310         void (*fn)();
 311         struct timer_list * next;
 312 } timer_list[TIME_REQUESTS] = { { 0, NULL, NULL }, };
 313 
 314 static struct timer_list * next_timer = NULL;
 315 
 316 void add_timer(long jiffies, void (*fn)(void))
     /*  */
 317 {
 318         struct timer_list * p;
 319 
 320         if (!fn)
 321                 return;
 322         cli();
 323         if (jiffies <= 0)
 324                 (fn)();
 325         else {
 326                 for (p = timer_list ; p < timer_list + TIME_REQUESTS ; p++)
 327                         if (!p->fn)
 328                                 break;
 329                 if (p >= timer_list + TIME_REQUESTS)
 330                         panic("No more time requests free");
 331                 p->fn = fn;
 332                 p->jiffies = jiffies;
 333                 p->next = next_timer;
 334                 next_timer = p;
 335                 while (p->next && p->next->jiffies < p->jiffies) {
 336                         p->jiffies -= p->next->jiffies;
 337                         fn = p->fn;
 338                         p->fn = p->next->fn;
 339                         p->next->fn = fn;
 340                         jiffies = p->jiffies;
 341                         p->jiffies = p->next->jiffies;
 342                         p->next->jiffies = jiffies;
 343                         p = p->next;
 344                 }
 345         }
 346         sti();
 347 }
 348 
 349 unsigned long timer_active = 0;
 350 struct timer_struct timer_table[32];
 351 
 352 /*
 353  * Hmm.. Changed this, as the GNU make sources (load.c) seems to
 354  * imply that avenrun[] is the standard name for this kind of thing.
 355  * Nothing else seems to be standardized: the fractional size etc
 356  * all seem to differ on different machines.
 357  */
 358 unsigned long avenrun[3] = { 0,0,0 };
 359 
 360 /*
 361  * Nr of active tasks - counted in fixed-point numbers
 362  */
 363 static unsigned long count_active_tasks(void)
     /*  */
 364 {
 365         struct task_struct **p;
 366         unsigned long nr = 0;
 367 
 368         for(p = &LAST_TASK; p > &FIRST_TASK; --p)
 369                 if (*p && (*p)->state == TASK_RUNNING)
 370                         nr += FIXED_1;
 371         return nr;
 372 }
 373 
 374 static inline void calc_load(void)
     /*  */
 375 {
 376         unsigned long active_tasks; /* fixed-point */
 377         static int count = LOAD_FREQ;
 378 
 379         if (count-- > 0)
 380                 return;
 381         count = LOAD_FREQ;
 382         active_tasks = count_active_tasks();
 383         CALC_LOAD(avenrun[0], EXP_1, active_tasks);
 384         CALC_LOAD(avenrun[1], EXP_5, active_tasks);
 385         CALC_LOAD(avenrun[2], EXP_15, active_tasks);
 386 }
 387 
 388 /*
 389  * The int argument is really a (struct pt_regs *), in case the
 390  * interrupt wants to know from where it was called. The timer
 391  * irq uses this to decide if it should update the user or system
 392  * times.
 393  */
 394 static void do_timer(struct pt_regs * regs)
     /*  */
 395 {
 396         unsigned long mask;
 397         struct timer_struct *tp = timer_table+0;
 398         struct task_struct ** task_p;
 399 
 400         jiffies++;
 401         calc_load();
 402         if ((VM_MASK & regs->eflags) || (3 & regs->cs)) {
 403                 current->utime++;
 404                 /* Update ITIMER_VIRT for current task if not in a system call */
 405                 if (current->it_virt_value && !(--current->it_virt_value)) {
 406                         current->it_virt_value = current->it_virt_incr;
 407                         send_sig(SIGVTALRM,current,1);
 408                 }
 409         } else {
 410                 current->stime++;
 411 #ifdef CONFIG_PROFILE
 412                 if (prof_buffer && current != task[0]) {
 413                         unsigned long eip = regs->eip;
 414                         eip >>= 2;
 415                         if (eip < prof_len)
 416                                 prof_buffer[eip]++;
 417                 }
 418 #endif
 419         }
 420         if (current == task[0] || (--current->counter)<=0) {
 421                 current->counter=0;
 422                 need_resched = 1;
 423         }
 424         /* Update ITIMER_REAL for every task */
 425         for (task_p = &LAST_TASK; task_p >= &FIRST_TASK; task_p--)
 426                 if (*task_p && (*task_p)->it_real_value
 427                         && !(--(*task_p)->it_real_value)) {
 428                         send_sig(SIGALRM,*task_p,1);
 429                         (*task_p)->it_real_value = (*task_p)->it_real_incr;
 430                         need_resched = 1;
 431                 }
 432         /* Update ITIMER_PROF for the current task */
 433         if (current->it_prof_value && !(--current->it_prof_value)) {
 434                 current->it_prof_value = current->it_prof_incr;
 435                 send_sig(SIGPROF,current,1);
 436         }
 437         for (mask = 1 ; mask ; tp++,mask += mask) {
 438                 if (mask > timer_active)
 439                         break;
 440                 if (!(mask & timer_active))
 441                         continue;
 442                 if (tp->expires > jiffies)
 443                         continue;
 444                 timer_active &= ~mask;
 445                 tp->fn();
 446                 sti();
 447         }
 448         if (next_timer) {
 449                 next_timer->jiffies--;
 450                 while (next_timer && next_timer->jiffies <= 0) {
 451                         void (*fn)(void);
 452                         
 453                         fn = next_timer->fn;
 454                         next_timer->fn = NULL;
 455                         next_timer = next_timer->next;
 456                         (fn)();
 457                 }
 458         }
 459         if (current_DOR & 0xf0)
 460                 do_floppy_timer();
 461 }
 462 
 463 int sys_alarm(long seconds)
     /*  */
 464 {
 465         extern int _setitimer(int, struct itimerval *, struct itimerval *);
 466         struct itimerval new, old;
 467 
 468         new.it_interval.tv_sec = new.it_interval.tv_usec = 0;
 469         new.it_value.tv_sec = seconds;
 470         new.it_value.tv_usec = 0;
 471         _setitimer(ITIMER_REAL, &new, &old);
 472         return(old.it_value.tv_sec + (old.it_value.tv_usec / 1000000));
 473 }
 474 
 475 int sys_getpid(void)
     /*  */
 476 {
 477         return current->pid;
 478 }
 479 
 480 int sys_getppid(void)
     /*  */
 481 {
 482         return current->p_pptr->pid;
 483 }
 484 
 485 int sys_getuid(void)
     /*  */
 486 {
 487         return current->uid;
 488 }
 489 
 490 int sys_geteuid(void)
     /*  */
 491 {
 492         return current->euid;
 493 }
 494 
 495 int sys_getgid(void)
     /*  */
 496 {
 497         return current->gid;
 498 }
 499 
 500 int sys_getegid(void)
     /*  */
 501 {
 502         return current->egid;
 503 }
 504 
 505 int sys_nice(long increment)
     /*  */
 506 {
 507         if (increment < 0 && !suser())
 508                 return -EPERM;
 509         if (increment >= current->priority)
 510                 increment = current->priority-1;
 511         current->priority -= increment;
 512         return 0;
 513 }
 514 
 515 static void show_task(int nr,struct task_struct * p)
     /*  */
 516 {
 517         int i, j;
 518         unsigned char * stack;
 519 
 520         printk("%d: pid=%d, state=%d, father=%d, child=%d, ",(p == current)?-nr:nr,p->pid,
 521                 p->state, p->p_pptr->pid, p->p_cptr ? p->p_cptr->pid : -1);
 522         i = 0;
 523         j = 4096;
 524         if (!(stack = (char *) p->kernel_stack_page)) {
 525                 stack = (char *) init_kernel_stack;
 526                 j = sizeof(init_kernel_stack);
 527         }
 528         while (i<j && !*(stack++))
 529                 i++;
 530         printk("%d/%d chars free in kstack\n\r",i,j);
 531         printk("   PC=%08X.", *(1019 + (unsigned long *) p));
 532         if (p->p_ysptr || p->p_osptr) 
 533                 printk("   Younger sib=%d, older sib=%d\n\r", 
 534                         p->p_ysptr ? p->p_ysptr->pid : -1,
 535                         p->p_osptr ? p->p_osptr->pid : -1);
 536         else
 537                 printk("\n\r");
 538 }
 539 
 540 void show_state(void)
     /*  */
 541 {
 542         int i;
 543 
 544         printk("\rTask-info:\n\r");
 545         for (i=0 ; i<NR_TASKS ; i++)
 546                 if (task[i])
 547                         show_task(i,task[i]);
 548 }
 549 
 550 void sched_init(void)
     /*  */
 551 {
 552         int i;
 553         struct desc_struct * p;
 554 
 555         if (sizeof(struct sigaction) != 16)
 556                 panic("Struct sigaction MUST be 16 bytes");
 557         set_tss_desc(gdt+FIRST_TSS_ENTRY,&init_task.tss);
 558         set_ldt_desc(gdt+FIRST_LDT_ENTRY,&init_task.ldt);
 559         set_system_gate(0x80,&system_call);
 560         p = gdt+2+FIRST_TSS_ENTRY;
 561         for(i=1 ; i<NR_TASKS ; i++) {
 562                 task[i] = NULL;
 563                 p->a=p->b=0;
 564                 p++;
 565                 p->a=p->b=0;
 566                 p++;
 567         }
 568 /* Clear NT, so that we won't have troubles with that later on */
 569         __asm__("pushfl ; andl $0xffffbfff,(%esp) ; popfl");
 570         load_TR(0);
 571         load_ldt(0);
 572         outb_p(0x36,0x43);              /* binary, mode 3, LSB/MSB, ch 0 */
 573         outb_p(LATCH & 0xff , 0x40);    /* LSB */
 574         outb(LATCH >> 8 , 0x40);        /* MSB */
 575         request_irq(TIMER_IRQ,(void (*)(int)) do_timer);
 576 }

/* */