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                         flush_cache_page(vma, address);
 114                         set_pte(page_table, __pte(entry));
 115                         flush_tlb_page(vma, address);
 116                         tsk->nswap++;
 117                         rw_swap_page(WRITE, entry, (char *) page, wait);
 118                 }
 119                 free_page(page);
 120                 return 1;       /* we slept: the process may not exist any more */
 121         }
 122         if ((entry = find_in_swap_cache(MAP_NR(page))))  {
 123                 if (page_map->count != 1) {
 124                         set_pte(page_table, pte_mkdirty(pte));
 125                         printk("Aiee.. duplicated cached swap-cache entry\n");
 126                         return 0;
 127                 }
 128                 vma->vm_mm->rss--;
 129                 flush_cache_page(vma, address);
 130                 set_pte(page_table, __pte(entry));
 131                 flush_tlb_page(vma, address);
 132                 free_page(page);
 133                 return 1;
 134         } 
 135         vma->vm_mm->rss--;
 136         flush_cache_page(vma, address);
 137         pte_clear(page_table);
 138         flush_tlb_page(vma, address);
 139         entry = page_unuse(page);
 140         free_page(page);
 141         return entry;
 142 }
 143 
 144 /*
 145  * A new implementation of swap_out().  We do not swap complete processes,
 146  * but only a small number of blocks, before we continue with the next
 147  * process.  The number of blocks actually swapped is determined on the
 148  * number of page faults, that this process actually had in the last time,
 149  * so we won't swap heavily used processes all the time ...
 150  *
 151  * Note: the priority argument is a hint on much CPU to waste with the
 152  *       swap block search, not a hint, of how much blocks to swap with
 153  *       each process.
 154  *
 155  * (C) 1993 Kai Petzke, wpp@marie.physik.tu-berlin.de
 156  */
 157 
 158 static inline int swap_out_pmd(struct task_struct * tsk, struct vm_area_struct * vma,
     /*  */
 159         pmd_t *dir, unsigned long address, unsigned long end, int dma, int wait)
 160 {
 161         pte_t * pte;
 162         unsigned long pmd_end;
 163 
 164         if (pmd_none(*dir))
 165                 return 0;
 166         if (pmd_bad(*dir)) {
 167                 printk("swap_out_pmd: bad pmd (%08lx)\n", pmd_val(*dir));
 168                 pmd_clear(dir);
 169                 return 0;
 170         }
 171         
 172         pte = pte_offset(dir, address);
 173         
 174         pmd_end = (address + PMD_SIZE) & PMD_MASK;
 175         if (end > pmd_end)
 176                 end = pmd_end;
 177 
 178         do {
 179                 int result;
 180                 tsk->swap_address = address + PAGE_SIZE;
 181                 result = try_to_swap_out(tsk, vma, address, pte, dma, wait);
 182                 if (result)
 183                         return result;
 184                 address += PAGE_SIZE;
 185                 pte++;
 186         } while (address < end);
 187         return 0;
 188 }
 189 
 190 static inline int swap_out_pgd(struct task_struct * tsk, struct vm_area_struct * vma,
     /*  */
 191         pgd_t *dir, unsigned long address, unsigned long end, int dma, int wait)
 192 {
 193         pmd_t * pmd;
 194         unsigned long pgd_end;
 195 
 196         if (pgd_none(*dir))
 197                 return 0;
 198         if (pgd_bad(*dir)) {
 199                 printk("swap_out_pgd: bad pgd (%08lx)\n", pgd_val(*dir));
 200                 pgd_clear(dir);
 201                 return 0;
 202         }
 203 
 204         pmd = pmd_offset(dir, address);
 205 
 206         pgd_end = (address + PGDIR_SIZE) & PGDIR_MASK;  
 207         if (end > pgd_end)
 208                 end = pgd_end;
 209         
 210         do {
 211                 int result = swap_out_pmd(tsk, vma, pmd, address, end, dma, wait);
 212                 if (result)
 213                         return result;
 214                 address = (address + PMD_SIZE) & PMD_MASK;
 215                 pmd++;
 216         } while (address < end);
 217         return 0;
 218 }
 219 
 220 static int swap_out_vma(struct task_struct * tsk, struct vm_area_struct * vma,
     /*  */
 221         pgd_t *pgdir, unsigned long start, int dma, int wait)
 222 {
 223         unsigned long end;
 224 
 225         /* Don't swap out areas like shared memory which have their
 226             own separate swapping mechanism or areas which are locked down */
 227         if (vma->vm_flags & (VM_SHM | VM_LOCKED))
 228                 return 0;
 229 
 230         end = vma->vm_end;
 231         while (start < end) {
 232                 int result = swap_out_pgd(tsk, vma, pgdir, start, end, dma, wait);
 233                 if (result)
 234                         return result;
 235                 start = (start + PGDIR_SIZE) & PGDIR_MASK;
 236                 pgdir++;
 237         }
 238         return 0;
 239 }
 240 
 241 static int swap_out_process(struct task_struct * p, int dma, int wait)
     /*  */
 242 {
 243         unsigned long address;
 244         struct vm_area_struct* vma;
 245 
 246         /*
 247          * Go through process' page directory.
 248          */
 249         address = p->swap_address;
 250         p->swap_address = 0;
 251 
 252         /*
 253          * Find the proper vm-area
 254          */
 255         vma = find_vma(p, address);
 256         if (!vma)
 257                 return 0;
 258         if (address < vma->vm_start)
 259                 address = vma->vm_start;
 260 
 261         for (;;) {
 262                 int result = swap_out_vma(p, vma, pgd_offset(p->mm, address), address, dma, wait);
 263                 if (result)
 264                         return result;
 265                 vma = vma->vm_next;
 266                 if (!vma)
 267                         break;
 268                 address = vma->vm_start;
 269         }
 270         p->swap_address = 0;
 271         return 0;
 272 }
 273 
 274 static int swap_out(unsigned int priority, int dma, int wait)
     /*  */
 275 {
 276         static int swap_task;
 277         int loop, counter;
 278         struct task_struct *p;
 279 
 280         counter = ((PAGEOUT_WEIGHT * nr_tasks) >> 10) >> priority;
 281         for(; counter >= 0; counter--) {
 282                 /*
 283                  * Check that swap_task is suitable for swapping.  If not, look for
 284                  * the next suitable process.
 285                  */
 286                 loop = 0;
 287                 while(1) {
 288                         if (swap_task >= NR_TASKS) {
 289                                 swap_task = 1;
 290                                 if (loop)
 291                                         /* all processes are unswappable or already swapped out */
 292                                         return 0;
 293                                 loop = 1;
 294                         }
 295 
 296                         p = task[swap_task];
 297                         if (p && p->swappable && p->mm->rss)
 298                                 break;
 299 
 300                         swap_task++;
 301                 }
 302 
 303                 /*
 304                  * Determine the number of pages to swap from this process.
 305                  */
 306                 if (!p->swap_cnt) {
 307                         /* Normalise the number of pages swapped by
 308                            multiplying by (RSS / 1MB) */
 309                         p->swap_cnt = AGE_CLUSTER_SIZE(p->mm->rss);
 310                 }
 311                 if (!--p->swap_cnt)
 312                         swap_task++;
 313                 switch (swap_out_process(p, dma, wait)) {
 314                         case 0:
 315                                 if (p->swap_cnt)
 316                                         swap_task++;
 317                                 break;
 318                         case 1:
 319                                 return 1;
 320                         default:
 321                                 break;
 322                 }
 323         }
 324         return 0;
 325 }
 326 
 327 /*
 328  * We are much more aggressive about trying to swap out than we used
 329  * to be.  This works out OK, because we now do proper aging on page
 330  * contents. 
 331  */
 332 int try_to_free_page(int priority, int dma, int wait)
     /*  */
 333 {
 334         static int state = 0;
 335         int i=6;
 336 
 337         switch (state) {
 338                 do {
 339                 case 0:
 340                         if (shrink_mmap(i, dma))
 341                                 return 1;
 342                         state = 1;
 343                 case 1:
 344                         if (shm_swap(i, dma))
 345                                 return 1;
 346                         state = 2;
 347                 default:
 348                         if (swap_out(i, dma, wait))
 349                                 return 1;
 350                         state = 0;
 351                 } while (i--);
 352         }
 353         return 0;
 354 }
 355 
 356 
 357 /*
 358  * The background pageout daemon.
 359  * Started as a kernel thread from the init process.
 360  */
 361 int kswapd(void *unused)
     /*  */
 362 {
 363         int i;
 364         char *revision="$Revision: 1.4.2.2 $", *s, *e;
 365         
 366         current->session = 1;
 367         current->pgrp = 1;
 368         sprintf(current->comm, "kswapd");
 369         current->blocked = ~0UL;
 370         
 371         /*
 372          *      As a kernel thread we want to tamper with system buffers
 373          *      and other internals and thus be subject to the SMP locking
 374          *      rules. (On a uniprocessor box this does nothing).
 375          */
 376          
 377 #ifdef __SMP__
 378         lock_kernel();
 379         syscall_count++;
 380 #endif
 381 
 382         /* Give kswapd a realtime priority. */
 383         current->policy = SCHED_FIFO;
 384         current->priority = 32;  /* Fixme --- we need to standardise our
 385                                     namings for POSIX.4 realtime scheduling
 386                                     priorities.  */
 387 
 388         init_swap_timer();
 389         
 390         if ((s = strchr(revision, ':')) &&
 391             (e = strchr(s, '$')))
 392                 s++, i = e - s;
 393         else
 394                 s = revision, i = -1;
 395         printk ("Started kswapd v%.*s\n", i, s);
 396 
 397         while (1) {
 398                 kswapd_awake = 0;
 399                 current->signal = 0;
 400                 interruptible_sleep_on(&kswapd_wait);
 401                 kswapd_awake = 1;
 402                 swapstats.wakeups++;
 403                 /* Do the background pageout: */
 404                 for (i=0; i < kswapd_ctl.maxpages; i++)
 405                         try_to_free_page(GFP_KERNEL, 0, 0);
 406         }
 407 }
 408 
 409 /* 
 410  * The swap_tick function gets called on every clock tick.
 411  */
 412 
 413 void swap_tick(void)
     /*  */
 414 {
 415         if ((nr_free_pages + nr_async_pages) < free_pages_low ||
 416             ((nr_free_pages + nr_async_pages) < free_pages_high && 
 417              jiffies >= next_swap_jiffies)) {
 418                 if (!kswapd_awake && kswapd_ctl.maxpages > 0) {
 419                         wake_up(&kswapd_wait);
 420                         need_resched = 1;
 421                         kswapd_awake = 1;
 422                 }
 423                 next_swap_jiffies = jiffies + swapout_interval;
 424         }
 425         timer_active |= (1<<SWAP_TIMER);
 426 }
 427 
 428 
 429 /* 
 430  * Initialise the swap timer
 431  */
 432 
 433 void init_swap_timer(void)
     /*  */
 434 {
 435         timer_table[SWAP_TIMER].expires = 0;
 436         timer_table[SWAP_TIMER].fn = swap_tick;
 437         timer_active |= (1<<SWAP_TIMER);
 438 }

/* */