root/mm/vmscan.c

/* */

DEFINITIONS

1. try_to_swap_out
2. swap_out_pmd
3. swap_out_pgd
4. swap_out_vma
5. swap_out_process
6. swap_out
7. try_to_free_page
8. kswapd
9. swap_tick
10. init_swap_timer

   1 /*
   2  *  linux/mm/vmscan.c
   3  *
   4  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   5  *
   6  *  Swap reorganised 29.12.95, Stephen Tweedie.
   7  *  kswapd added: 7.1.96  sct
   8  *  Version: $Id: vmscan.c,v 1.4.2.2 1996/01/20 18:22:47 linux Exp $
   9  */
  10 
  11 #include <linux/mm.h>
  12 #include <linux/sched.h>
  13 #include <linux/head.h>
  14 #include <linux/kernel.h>
  15 #include <linux/kernel_stat.h>
  16 #include <linux/errno.h>
  17 #include <linux/string.h>
  18 #include <linux/stat.h>
  19 #include <linux/swap.h>
  20 #include <linux/fs.h>
  21 #include <linux/swapctl.h>
  22 #include <linux/smp_lock.h>
  23 
  24 #include <asm/dma.h>
  25 #include <asm/system.h> /* for cli()/sti() */
  26 #include <asm/segment.h> /* for memcpy_to/fromfs */
  27 #include <asm/bitops.h>
  28 #include <asm/pgtable.h>
  29 
  30 /* 
  31  * When are we next due for a page scan? 
  32  */
  33 static int next_swap_jiffies = 0;
  34 
  35 /* 
  36  * How often do we do a pageout scan during normal conditions?
  37  * Default is four times a second.
  38  */
  39 int swapout_interval = HZ / 4;
  40 
  41 /* 
  42  * The wait queue for waking up the pageout daemon:
  43  */
  44 static struct wait_queue * kswapd_wait = NULL;
  45 
  46 /* 
  47  * We avoid doing a reschedule if the pageout daemon is already awake;
  48  */
  49 static int kswapd_awake = 0;
  50 
  51 /*
  52  * sysctl-modifiable parameters to control the aggressiveness of the
  53  * page-searching within the kswapd page recovery daemon.
  54  */
  55 kswapd_control_t kswapd_ctl = {4, -1, -1, -1, -1};
  56 
  57 static void init_swap_timer(void);
  58 
  59 /*
  60  * The swap-out functions return 1 if they successfully
  61  * threw something out, and we got a free page. It returns
  62  * zero if it couldn't do anything, and any other value
  63  * indicates it decreased rss, but the page was shared.
  64  *
  65  * NOTE! If it sleeps, it *must* return 1 to make sure we
  66  * don't continue with the swap-out. Otherwise we may be
  67  * using a process that no longer actually exists (it might
  68  * have died while we slept).
  69  */
  70 static inline int try_to_swap_out(struct task_struct * tsk, struct vm_area_struct* vma,
     /*  */
  71         unsigned long address, pte_t * page_table, int dma, int wait)
  72 {
  73         pte_t pte;
  74         unsigned long entry;
  75         unsigned long page;
  76         struct page * page_map;
  77 
  78         pte = *page_table;
  79         if (!pte_present(pte))
  80                 return 0;
  81         page = pte_page(pte);
  82         if (MAP_NR(page) >= MAP_NR(high_memory))
  83                 return 0;
  84 
  85         page_map = mem_map + MAP_NR(page);
  86         if (page_map->reserved || page_map->locked ||
  87             (dma && !page_map->dma))
  88                 return 0;
  89         /* Deal with page aging.  Pages age from being unused; they
  90          * rejuvinate on being accessed.  Only swap old pages (age==0
  91          * is oldest). */
  92         if ((pte_dirty(pte) && delete_from_swap_cache(MAP_NR(page))) 
  93             || pte_young(pte))  {
  94                 set_pte(page_table, pte_mkold(pte));
  95                 touch_page(page_map);
  96                 return 0;
  97         }
  98         age_page(page_map);
  99         if (page_map->age)
 100                 return 0;
 101         if (pte_dirty(pte)) {
 102                 if (vma->vm_ops && vma->vm_ops->swapout) {
 103                         pid_t pid = tsk->pid;
 104                         vma->vm_mm->rss--;
 105                         if (vma->vm_ops->swapout(vma, address - vma->vm_start + vma->vm_offset, page_table))
 106                                 kill_proc(pid, SIGBUS, 1);
 107                 } else {
 108                         if (page_map->count != 1)
 109                                 return 0;
 110                         if (!(entry = get_swap_page()))
 111                                 return 0;
 112                         vma->vm_mm->rss--;
 113                         set_pte(page_table, __pte(entry));
 114                         invalidate_page(vma, address);
 115                         tsk->nswap++;
 116                         rw_swap_page(WRITE, entry, (char *) page, wait);
 117                 }
 118                 free_page(page);
 119                 return 1;       /* we slept: the process may not exist any more */
 120         }
 121         if ((entry = find_in_swap_cache(MAP_NR(page))))  {
 122                 if (page_map->count != 1) {
 123                         set_pte(page_table, pte_mkdirty(pte));
 124                         printk("Aiee.. duplicated cached swap-cache entry\n");
 125                         return 0;
 126                 }
 127                 vma->vm_mm->rss--;
 128                 set_pte(page_table, __pte(entry));
 129                 invalidate_page(vma, address);
 130                 free_page(page);
 131                 return 1;
 132         } 
 133         vma->vm_mm->rss--;
 134         pte_clear(page_table);
 135         invalidate_page(vma, address);
 136         entry = page_unuse(page);
 137         free_page(page);
 138         return entry;
 139 }
 140 
 141 /*
 142  * A new implementation of swap_out().  We do not swap complete processes,
 143  * but only a small number of blocks, before we continue with the next
 144  * process.  The number of blocks actually swapped is determined on the
 145  * number of page faults, that this process actually had in the last time,
 146  * so we won't swap heavily used processes all the time ...
 147  *
 148  * Note: the priority argument is a hint on much CPU to waste with the
 149  *       swap block search, not a hint, of how much blocks to swap with
 150  *       each process.
 151  *
 152  * (C) 1993 Kai Petzke, wpp@marie.physik.tu-berlin.de
 153  */
 154 
 155 static inline int swap_out_pmd(struct task_struct * tsk, struct vm_area_struct * vma,
     /*  */
 156         pmd_t *dir, unsigned long address, unsigned long end, int dma, int wait)
 157 {
 158         pte_t * pte;
 159         unsigned long pmd_end;
 160 
 161         if (pmd_none(*dir))
 162                 return 0;
 163         if (pmd_bad(*dir)) {
 164                 printk("swap_out_pmd: bad pmd (%08lx)\n", pmd_val(*dir));
 165                 pmd_clear(dir);
 166                 return 0;
 167         }
 168         
 169         pte = pte_offset(dir, address);
 170         
 171         pmd_end = (address + PMD_SIZE) & PMD_MASK;
 172         if (end > pmd_end)
 173                 end = pmd_end;
 174 
 175         do {
 176                 int result;
 177                 tsk->swap_address = address + PAGE_SIZE;
 178                 result = try_to_swap_out(tsk, vma, address, pte, dma, wait);
 179                 if (result)
 180                         return result;
 181                 address += PAGE_SIZE;
 182                 pte++;
 183         } while (address < end);
 184         return 0;
 185 }
 186 
 187 static inline int swap_out_pgd(struct task_struct * tsk, struct vm_area_struct * vma,
     /*  */
 188         pgd_t *dir, unsigned long address, unsigned long end, int dma, int wait)
 189 {
 190         pmd_t * pmd;
 191         unsigned long pgd_end;
 192 
 193         if (pgd_none(*dir))
 194                 return 0;
 195         if (pgd_bad(*dir)) {
 196                 printk("swap_out_pgd: bad pgd (%08lx)\n", pgd_val(*dir));
 197                 pgd_clear(dir);
 198                 return 0;
 199         }
 200 
 201         pmd = pmd_offset(dir, address);
 202 
 203         pgd_end = (address + PGDIR_SIZE) & PGDIR_MASK;  
 204         if (end > pgd_end)
 205                 end = pgd_end;
 206         
 207         do {
 208                 int result = swap_out_pmd(tsk, vma, pmd, address, end, dma, wait);
 209                 if (result)
 210                         return result;
 211                 address = (address + PMD_SIZE) & PMD_MASK;
 212                 pmd++;
 213         } while (address < end);
 214         return 0;
 215 }
 216 
 217 static int swap_out_vma(struct task_struct * tsk, struct vm_area_struct * vma,
     /*  */
 218         pgd_t *pgdir, unsigned long start, int dma, int wait)
 219 {
 220         unsigned long end;
 221 
 222         /* Don't swap out areas like shared memory which have their
 223             own separate swapping mechanism or areas which are locked down */
 224         if (vma->vm_flags & (VM_SHM | VM_LOCKED))
 225                 return 0;
 226 
 227         end = vma->vm_end;
 228         while (start < end) {
 229                 int result = swap_out_pgd(tsk, vma, pgdir, start, end, dma, wait);
 230                 if (result)
 231                         return result;
 232                 start = (start + PGDIR_SIZE) & PGDIR_MASK;
 233                 pgdir++;
 234         }
 235         return 0;
 236 }
 237 
 238 static int swap_out_process(struct task_struct * p, int dma, int wait)
     /*  */
 239 {
 240         unsigned long address;
 241         struct vm_area_struct* vma;
 242 
 243         /*
 244          * Go through process' page directory.
 245          */
 246         address = p->swap_address;
 247         p->swap_address = 0;
 248 
 249         /*
 250          * Find the proper vm-area
 251          */
 252         vma = find_vma(p, address);
 253         if (!vma)
 254                 return 0;
 255         if (address < vma->vm_start)
 256                 address = vma->vm_start;
 257 
 258         for (;;) {
 259                 int result = swap_out_vma(p, vma, pgd_offset(p->mm, address), address, dma, wait);
 260                 if (result)
 261                         return result;
 262                 vma = vma->vm_next;
 263                 if (!vma)
 264                         break;
 265                 address = vma->vm_start;
 266         }
 267         p->swap_address = 0;
 268         return 0;
 269 }
 270 
 271 static int swap_out(unsigned int priority, int dma, int wait)
     /*  */
 272 {
 273         static int swap_task;
 274         int loop, counter;
 275         struct task_struct *p;
 276 
 277         counter = ((PAGEOUT_WEIGHT * nr_tasks) >> 10) >> priority;
 278         for(; counter >= 0; counter--) {
 279                 /*
 280                  * Check that swap_task is suitable for swapping.  If not, look for
 281                  * the next suitable process.
 282                  */
 283                 loop = 0;
 284                 while(1) {
 285                         if (swap_task >= NR_TASKS) {
 286                                 swap_task = 1;
 287                                 if (loop)
 288                                         /* all processes are unswappable or already swapped out */
 289                                         return 0;
 290                                 loop = 1;
 291                         }
 292 
 293                         p = task[swap_task];
 294                         if (p && p->swappable && p->mm->rss)
 295                                 break;
 296 
 297                         swap_task++;
 298                 }
 299 
 300                 /*
 301                  * Determine the number of pages to swap from this process.
 302                  */
 303                 if (!p->swap_cnt) {
 304                         /* Normalise the number of pages swapped by
 305                            multiplying by (RSS / 1MB) */
 306                         p->swap_cnt = AGE_CLUSTER_SIZE(p->mm->rss);
 307                 }
 308                 if (!--p->swap_cnt)
 309                         swap_task++;
 310                 switch (swap_out_process(p, dma, wait)) {
 311                         case 0:
 312                                 if (p->swap_cnt)
 313                                         swap_task++;
 314                                 break;
 315                         case 1:
 316                                 return 1;
 317                         default:
 318                                 break;
 319                 }
 320         }
 321         return 0;
 322 }
 323 
 324 /*
 325  * We are much more aggressive about trying to swap out than we used
 326  * to be.  This works out OK, because we now do proper aging on page
 327  * contents. 
 328  */
 329 int try_to_free_page(int priority, int dma, int wait)
     /*  */
 330 {
 331         static int state = 0;
 332         int i=6;
 333 
 334         switch (state) {
 335                 do {
 336                 case 0:
 337                         if (shrink_mmap(i, dma))
 338                                 return 1;
 339                         state = 1;
 340                 case 1:
 341                         if (shm_swap(i, dma))
 342                                 return 1;
 343                         state = 2;
 344                 default:
 345                         if (swap_out(i, dma, wait))
 346                                 return 1;
 347                         state = 0;
 348                 } while (i--);
 349         }
 350         return 0;
 351 }
 352 
 353 
 354 /*
 355  * The background pageout daemon.
 356  * Started as a kernel thread from the init process.
 357  */
 358 int kswapd(void *unused)
     /*  */
 359 {
 360         int i;
 361         char *revision="$Revision: 1.4.2.2 $", *s, *e;
 362         
 363         current->session = 1;
 364         current->pgrp = 1;
 365         sprintf(current->comm, "kswapd");
 366         current->blocked = ~0UL;
 367         
 368         /*
 369          *      As a kernel thread we want to tamper with system buffers
 370          *      and other internals and thus be subject to the SMP locking
 371          *      rules. (On a uniprocessor box this does nothing).
 372          */
 373          
 374 #ifdef __SMP__
 375         lock_kernel();
 376         syscall_count++;
 377 #endif
 378 
 379         /* Give kswapd a realtime priority. */
 380         current->policy = SCHED_FIFO;
 381         current->priority = 32;  /* Fixme --- we need to standardise our
 382                                     namings for POSIX.4 realtime scheduling
 383                                     priorities.  */
 384 
 385         init_swap_timer();
 386         
 387         if ((s = strchr(revision, ':')) &&
 388             (e = strchr(s, '$')))
 389                 s++, i = e - s;
 390         else
 391                 s = revision, i = -1;
 392         printk ("Started kswapd v%.*s\n", i, s);
 393 
 394         while (1) {
 395                 kswapd_awake = 0;
 396                 current->signal = 0;
 397                 interruptible_sleep_on(&kswapd_wait);
 398                 kswapd_awake = 1;
 399                 swapstats.wakeups++;
 400                 /* Do the background pageout: */
 401                 for (i=0; i < kswapd_ctl.maxpages; i++)
 402                         try_to_free_page(GFP_KERNEL, 0, 0);
 403         }
 404 }
 405 
 406 /* 
 407  * The swap_tick function gets called on every clock tick.
 408  */
 409 
 410 void swap_tick(void)
     /*  */
 411 {
 412         if ((nr_free_pages + nr_async_pages) < free_pages_low ||
 413             ((nr_free_pages + nr_async_pages) < free_pages_high && 
 414              jiffies >= next_swap_jiffies)) {
 415                 if (!kswapd_awake && kswapd_ctl.maxpages > 0) {
 416                         wake_up(&kswapd_wait);
 417                         need_resched = 1;
 418                         kswapd_awake = 1;
 419                 }
 420                 next_swap_jiffies = jiffies + swapout_interval;
 421         }
 422         timer_active |= (1<<SWAP_TIMER);
 423 }
 424 
 425 
 426 /* 
 427  * Initialise the swap timer
 428  */
 429 
 430 void init_swap_timer(void)
     /*  */
 431 {
 432         timer_table[SWAP_TIMER].expires = 0;
 433         timer_table[SWAP_TIMER].fn = swap_tick;
 434         timer_active |= (1<<SWAP_TIMER);
 435 }

/* */