4.3BSD - /usr/src/sys/sys/vm_page.c

   1: /*
   2:  * Copyright (c) 1982, 1986 Regents of the University of California.
   3:  * All rights reserved.  The Berkeley software License Agreement
   4:  * specifies the terms and conditions for redistribution.
   5:  *
   6:  *	@(#)vm_page.c	7.1 (Berkeley) 6/5/86
   7:  */
   8: 
   9: #include "../machine/reg.h"
  10: #include "../machine/pte.h"
  11: 
  12: #include "param.h"
  13: #include "systm.h"
  14: #include "inode.h"
  15: #include "dir.h"
  16: #include "user.h"
  17: #include "proc.h"
  18: #include "buf.h"
  19: #include "text.h"
  20: #include "cmap.h"
  21: #include "vm.h"
  22: #include "file.h"
  23: #include "trace.h"
  24: 
  25: int nohash = 0;
  26: /*
  27:  * Handle a page fault.
  28:  *
  29:  * Basic outline
  30:  *	If page is allocated, but just not valid:
  31:  *		Wait if intransit, else just revalidate
  32:  *		Done
  33:  *	Compute <dev,bn> from which page operation would take place
  34:  *	If page is text page, and filling from file system or swap space:
  35:  *		If in free list cache, reattach it and then done
  36:  *	Allocate memory for page in
  37:  *		If block here, restart because we could have swapped, etc.
  38:  *	Lock process from swapping for duration
  39:  *	Update pte's to reflect that page is intransit.
  40:  *	If page is zero fill on demand:
  41:  *		Clear pages and flush free list cache of stale cacheing
  42:  *		for this swap page (e.g. before initializing again due
  43:  *		to 407/410 exec).
  44:  *	If page is fill from file and in buffer cache:
  45:  *		Copy the page from the buffer cache.
  46:  *	If not a fill on demand:
  47:  *		Determine swap address and cluster to page in
  48:  *	Do the swap to bring the page in
  49:  *	Instrument the pagein
  50:  *	After swap validate the required new page
  51:  *	Leave prepaged pages reclaimable (not valid)
  52:  *	Update shared copies of text page tables
  53:  *	Complete bookkeeping on pages brought in:
  54:  *		No longer intransit
  55:  *		Hash text pages into core hash structure
  56:  *		Unlock pages (modulo raw i/o requirements)
  57:  *		Flush translation buffer
  58:  *	Process pagein is done
  59:  */
  60: #ifdef TRACE
  61: #define pgtrace(e)  trace(e,v,u.u_procp->p_pid)
  62: #else
  63: #define pgtrace(e)
  64: #endif
  65: 
  66: int preptofree = 1;     /* send pre-paged pages to free list */
  67: 
  68: pagein(virtaddr, dlyu)
  69:     unsigned virtaddr;
  70:     int dlyu;
  71: {
  72:     register struct proc *p;
  73:     register struct pte *pte;
  74:     register unsigned v;
  75:     unsigned pf;
  76:     int type, fileno;
  77:     struct pte opte;
  78:     dev_t dev;
  79:     register int i;
  80:     int klsize;
  81:     unsigned vsave;
  82:     struct cmap *c;
  83:     int j;
  84:     daddr_t bn, bncache, bnswap;
  85:     int si, sk;
  86:     int swerror = 0;
  87: #ifdef PGINPROF
  88: #include "../vax/mtpr.h"
  89:     int otime, olbolt, oicr, s;
  90:     long a;
  91: 
  92:     s = splclock();
  93:     otime = time, olbolt = lbolt, oicr = mfpr(ICR);
  94: #endif
  95:     cnt.v_faults++;
  96:     /*
  97: 	 * Classify faulted page into a segment and get a pte
  98: 	 * for the faulted page.
  99: 	 */
 100:     vsave = v = clbase(btop(virtaddr));
 101:     p = u.u_procp;
 102:     if (isatsv(p, v))
 103:         type = CTEXT;
 104:     else if (isassv(p, v))
 105:         type = CSTACK;
 106:     else
 107:         type = CDATA;
 108:     pte = vtopte(p, v);
 109:     if (pte->pg_v)
 110:         panic("pagein");
 111: 
 112:     /*
 113: 	 * If page is reclaimable, reclaim it.
 114: 	 * If page is text and intransit, sleep while it is intransit,
 115: 	 * If it is valid after the sleep, we are done.
 116: 	 * Otherwise we have to start checking again, since page could
 117: 	 * even be reclaimable now (we may have swapped for a long time).
 118: 	 */
 119: restart:
 120:     if (pte->pg_fod == 0 && pte->pg_pfnum) {
 121:         if (type == CTEXT && cmap[pgtocm(pte->pg_pfnum)].c_intrans) {
 122:             pgtrace(TR_INTRANS);
 123:             sleep((caddr_t)p->p_textp, PSWP+1);
 124:             pgtrace(TR_EINTRANS);
 125:             pte = vtopte(p, v);
 126:             if (pte->pg_v) {
 127: valid:
 128:                 if (dlyu) {
 129:                     c = &cmap[pgtocm(pte->pg_pfnum)];
 130:                     if (c->c_lock) {
 131:                         c->c_want = 1;
 132:                         sleep((caddr_t)c, PSWP+1);
 133:                         goto restart;
 134:                     }
 135:                     c->c_lock = 1;
 136:                 }
 137:                 newptes(pte, v, CLSIZE);
 138:                 cnt.v_intrans++;
 139:                 return;
 140:             }
 141:             goto restart;
 142:         }
 143:         /*
 144: 		 * If page is in the free list, then take
 145: 		 * it back into the resident set, updating
 146: 		 * the size recorded for the resident set.
 147: 		 */
 148:         si = splimp();
 149:         c = &cmap[pgtocm(pte->pg_pfnum)];
 150:         if (c->c_free) {
 151:             pgtrace(TR_FRECLAIM);
 152:             munlink(c);
 153:             cnt.v_pgfrec++;
 154:             if (type == CTEXT)
 155:                 p->p_textp->x_rssize += CLSIZE;
 156:             else
 157:                 p->p_rssize += CLSIZE;
 158:         } else
 159:             pgtrace(TR_RECLAIM);
 160:         splx(si);
 161:         pte->pg_v = 1;
 162:         if (anycl(pte, pg_m))
 163:             pte->pg_m = 1;
 164:         distcl(pte);
 165:         if (type == CTEXT)
 166:             distpte(p->p_textp, (unsigned)vtotp(p, v), pte);
 167:         u.u_ru.ru_minflt++;
 168:         cnt.v_pgrec++;
 169:         if (dlyu) {
 170:             c = &cmap[pgtocm(pte->pg_pfnum)];
 171:             if (c->c_lock) {
 172:                 c->c_want = 1;
 173:                 sleep((caddr_t)c, PSWP+1);
 174:                 goto restart;
 175:             }
 176:             c->c_lock = 1;
 177:         }
 178:         newptes(pte, v, CLSIZE);
 179: #ifdef PGINPROF
 180:         a = vmtime(otime, olbolt, oicr);
 181:         rectime += a;
 182:         if (a >= 0)
 183:             vmfltmon(rmon, a, rmonmin, rres, NRMON);
 184:         splx(s);
 185: #endif
 186:         return;
 187:     }
 188: #ifdef PGINPROF
 189:     splx(s);
 190: #endif
 191:     /*
 192: 	 * <dev,bn> is where data comes from/goes to.
 193: 	 * <dev,bncache> is where data is cached from/to.
 194: 	 * <swapdev,bnswap> is where data will eventually go.
 195: 	 */
 196:     if (pte->pg_fod == 0) {
 197:         fileno = -1;
 198:         bnswap = bncache = bn = vtod(p, v, &u.u_dmap, &u.u_smap);
 199:         dev = swapdev;
 200:     } else {
 201:         fileno = ((struct fpte *)pte)->pg_fileno;
 202:         bn = ((struct fpte *)pte)->pg_blkno;
 203:         bnswap = vtod(p, v, &u.u_dmap, &u.u_smap);
 204:         if (fileno > PG_FMAX)
 205:             panic("pagein pg_fileno");
 206:         if (fileno == PG_FTEXT) {
 207:             if (p->p_textp == 0)
 208:                 panic("pagein PG_FTEXT");
 209:             dev = p->p_textp->x_iptr->i_dev;
 210:             bncache = bn;
 211:         } else if (fileno == PG_FZERO) {
 212:             dev = swapdev;
 213:             bncache = bnswap;
 214:         } else {
 215:             panic("pagein");    /* can't happen */
 216:         }
 217:     }
 218:     klsize = 1;
 219:     opte = *pte;
 220: 
 221:     /*
 222: 	 * Check for text detached but in free list.
 223: 	 * This can happen only if the page is filling
 224: 	 * from a inode or from the swap device, (e.g. not when reading
 225: 	 * in 407/410 execs to a zero fill page.)
 226: 	 * Honor lock bit to avoid races with pageouts.
 227: 	 */
 228:     if (type == CTEXT && fileno != PG_FZERO && !nohash) {
 229:         si = splimp();
 230:         while ((c = mfind(dev, bncache)) != 0) {
 231:             if (c->c_lock == 0)
 232:                 break;
 233:             MLOCK(c);
 234:             MUNLOCK(c);
 235:         }
 236:         if (c) {
 237:             if (c->c_type != CTEXT || c->c_gone == 0 ||
 238:                 c->c_free == 0)
 239:                 panic("pagein mfind");
 240:             p->p_textp->x_rssize += CLSIZE;
 241:             /*
 242: 			 * Following code mimics memall().
 243: 			 */
 244:             munlink(c);
 245:             pf = cmtopg(c - cmap);
 246:             for (j = 0; j < CLSIZE; j++) {
 247:                 *(int *)pte = pf++;
 248:                 pte->pg_prot = opte.pg_prot;
 249:                 pte++;
 250:             }
 251:             pte -= CLSIZE;
 252:             c->c_free = 0;
 253:             c->c_gone = 0;
 254:             if (c->c_intrans || c->c_want)
 255:                 panic("pagein intrans|want");
 256:             c->c_lock = 1;
 257:             if (c->c_page != vtotp(p, v))
 258:                 panic("pagein c_page chgd");
 259:             c->c_ndx = p->p_textp - &text[0];
 260:             if (dev == swapdev) {
 261:                 cnt.v_xsfrec++;
 262:                 pgtrace(TR_XSFREC);
 263:             } else {
 264:                 cnt.v_xifrec++;
 265:                 pgtrace(TR_XIFREC);
 266:             }
 267:             cnt.v_pgrec++;
 268:             u.u_ru.ru_minflt++;
 269:             if (dev != swapdev) {
 270:                 c = mfind(swapdev, bnswap);
 271:                 if (c)
 272:                     munhash(swapdev, bnswap);
 273:                 pte->pg_m = 1;
 274:             }
 275:             splx(si);
 276:             goto skipswap;
 277:         }
 278:         splx(si);
 279:     }
 280: 
 281:     /*
 282: 	 * Wasn't reclaimable or reattachable.
 283: 	 * Have to prepare to bring the page in.
 284: 	 * We allocate the page before locking so we will
 285: 	 * be swappable if there is no free memory.
 286: 	 * If we block we have to start over, since anything
 287: 	 * could have happened.
 288: 	 */
 289:     sk = splimp();      /* lock memalls from here into kluster */
 290:     if (freemem < CLSIZE * KLMAX) {
 291:         pgtrace(TR_WAITMEM);
 292:         while (freemem < CLSIZE * KLMAX)
 293:             sleep((caddr_t)&freemem, PSWP+2);
 294:         pgtrace(TR_EWAITMEM);
 295:         splx(sk);
 296:         pte = vtopte(p, v);
 297:         if (pte->pg_v)
 298:             goto valid;
 299:         goto restart;
 300:     }
 301: 
 302:     /*
 303: 	 * Now can get memory and committed to bringing in the page.
 304: 	 * Lock this process, get a page,
 305: 	 * construct the new pte, and increment
 306: 	 * the (process or text) resident set size.
 307: 	 */
 308:     p->p_flag |= SPAGE;
 309:     if (memall(pte, CLSIZE, p, type) == 0)
 310:         panic("pagein memall");
 311:     pte->pg_prot = opte.pg_prot;
 312:     pf = pte->pg_pfnum;
 313:     cmap[pgtocm(pf)].c_intrans = 1;
 314:     distcl(pte);
 315:     if (type == CTEXT) {
 316:         p->p_textp->x_rssize += CLSIZE;
 317:         distpte(p->p_textp, (unsigned)vtotp(p, v), pte);
 318:     } else
 319:         p->p_rssize += CLSIZE;
 320: 
 321:     /*
 322: 	 * Two cases: either fill on demand (zero, or from file or text)
 323: 	 * or from swap space.
 324: 	 */
 325:     if (opte.pg_fod) {
 326:         pte->pg_m = 1;
 327:         if (fileno == PG_FZERO || fileno == PG_FTEXT) {
 328:             /*
 329: 			 * Flush any previous text page use of this
 330: 			 * swap device block.
 331: 			 */
 332:             si = splimp();
 333:             if (type == CTEXT) {
 334:                 c = mfind(swapdev, bnswap);
 335:                 if (c)
 336:                     munhash(swapdev, bnswap);
 337:             }
 338:             splx(si);
 339:             /*
 340: 			 * If zero fill, short-circuit hard work
 341: 			 * by just clearing pages.
 342: 			 */
 343:             if (fileno == PG_FZERO) {
 344:                 pgtrace(TR_ZFOD);
 345:                 for (i = 0; i < CLSIZE; i++)
 346:                     clearseg(pf+i);
 347:                 if (type != CTEXT)
 348:                     cnt.v_zfod += CLSIZE;
 349:                 splx(sk);
 350:                 goto skipswap;
 351:             }
 352:             pgtrace(TR_EXFOD);
 353:             cnt.v_exfod += CLSIZE;
 354:         } else
 355:             panic("pagein vread");
 356:         /*
 357: 		 * Fill from inode.  Try to find adjacent
 358: 		 * pages to bring in also.
 359: 		 */
 360:         v = fodkluster(p, v, pte, &klsize, dev, &bn);
 361:         bncache = bn;
 362:         splx(sk);
 363:         /*
 364: 		 * Blocks of an executable may still be in the buffer
 365: 		 * cache, so we explicitly flush them out to disk
 366: 		 * so that the proper data will be paged in.
 367: 		 */
 368:         blkflush(dev, bn, (long)CLSIZE*NBPG);
 369: #ifdef TRACE
 370:         if (type != CTEXT)
 371:             trace(TR_XFODMISS, dev, bn);
 372: #endif
 373:     } else {
 374:         if (opte.pg_pfnum)
 375:             panic("pagein pfnum");
 376:         pgtrace(TR_SWAPIN);
 377:         /*
 378: 		 * Fill from swap area.  Try to find adjacent
 379: 		 * pages to bring in also.
 380: 		 */
 381:         v = kluster(p, v, pte, B_READ, &klsize,
 382:             (type == CTEXT) ? kltxt :
 383:             ((p->p_flag & SSEQL) ? klseql : klin), bn);
 384:         splx(sk);
 385:         /* THIS COULD BE COMPUTED INCREMENTALLY... */
 386:         bncache = bn = vtod(p, v, &u.u_dmap, &u.u_smap);
 387:     }
 388: 
 389:     distcl(pte);
 390:     swerror = swap(p, bn, ptob(v), klsize * ctob(CLSIZE),
 391:         B_READ, B_PGIN, dev, 0);
 392: #ifdef TRACE
 393:     trace(TR_PGINDONE, vsave, u.u_procp->p_pid);
 394: #endif
 395: 
 396:     /*
 397: 	 * Instrumentation.
 398: 	 */
 399:     u.u_ru.ru_majflt++;
 400:     cnt.v_pgin++;
 401:     cnt.v_pgpgin += klsize * CLSIZE;
 402: #ifdef PGINPROF
 403:     a = vmtime(otime, olbolt, oicr) / 100;
 404:     pgintime += a;
 405:     if (a >= 0)
 406:         vmfltmon(pmon, a, pmonmin, pres, NPMON);
 407: #endif
 408: 
 409: skipswap:
 410:     /*
 411: 	 * Fix page table entries.
 412: 	 *
 413: 	 * Only page requested in is validated, and rest of pages
 414: 	 * can be ``reclaimed''.  This allows system to reclaim prepaged pages
 415: 	 * quickly if they are not used and memory is tight.
 416: 	 */
 417:     pte = vtopte(p, vsave);
 418:     pte->pg_v = 1;
 419:     distcl(pte);
 420:     if (type == CTEXT) {
 421:         if (swerror == 0) {
 422:             distpte(p->p_textp, (unsigned)vtotp(p, vsave), pte);
 423:             if (opte.pg_fod)
 424:                 p->p_textp->x_flag |= XWRIT;
 425:         }
 426:         wakeup((caddr_t)p->p_textp);
 427:     }
 428: 
 429:     /*
 430: 	 * Memall returned page(s) locked.  Unlock all
 431: 	 * pages in cluster.  If locking pages for raw i/o
 432: 	 * leave the page which was required to be paged in locked,
 433: 	 * but still unlock others.
 434: 	 * If text pages, hash into the cmap situation table.
 435: 	 */
 436:     pte = vtopte(p, v);
 437:     for (i = 0; i < klsize; i++) {
 438:         c = &cmap[pgtocm(pte->pg_pfnum)];
 439:         c->c_intrans = 0;
 440:         if (type == CTEXT && c->c_blkno == 0 && bncache && !nohash &&
 441:             !swerror) {
 442:             mhash(c, dev, bncache);
 443:             bncache += btodb(CLBYTES);
 444:         }
 445:         if (v != vsave || !dlyu)
 446:             MUNLOCK(c);
 447:         if (v != vsave && type != CTEXT && preptofree &&
 448:             opte.pg_fod == 0) {
 449:             /*
 450: 			 * Throw pre-paged data/stack pages at the
 451: 			 * bottom of the free list.
 452: 			 */
 453:             p->p_rssize -= CLSIZE;
 454:             memfree(pte, CLSIZE, 0);
 455:         }
 456:         newptes(pte, v, CLSIZE);
 457:         v += CLSIZE;
 458:         pte += CLSIZE;
 459:     }
 460: 
 461:     /*
 462: 	 * All done.
 463: 	 */
 464:     p->p_flag &= ~SPAGE;
 465: 
 466:     /*
 467: 	 * If process is declared fifo, memory is tight,
 468: 	 * and this was a data page-in, free memory
 469: 	 * klsdist pagein clusters away from the current fault.
 470: 	 */
 471:     if ((p->p_flag&SSEQL) && freemem < lotsfree && type == CDATA) {
 472:         int k = (vtodp(p, vsave) / CLSIZE) / klseql;
 473: #ifdef notdef
 474:         if (vsave > u.u_vsave)
 475:             k -= klsdist;
 476:         else
 477:             k += klsdist;
 478:         dpageout(p, k * klseql * CLSIZE, klout*CLSIZE);
 479:         u.u_vsave = vsave;
 480: #else
 481:         dpageout(p, (k - klsdist) * klseql * CLSIZE, klout*CLSIZE);
 482:         dpageout(p, (k + klsdist) * klseql * CLSIZE, klout*CLSIZE);
 483: #endif
 484:     }
 485: }
 486: 
 487: /*
 488:  * Take away n pages of data space
 489:  * starting at data page dp.
 490:  * Used to take pages away from sequential processes.
 491:  * Mimics pieces of code in pageout() below.
 492:  */
 493: dpageout(p, dp, n)
 494:     struct proc *p;
 495:     int dp, n;
 496: {
 497:     register struct cmap *c;
 498:     int i, klsize;
 499:     register struct pte *pte;
 500:     unsigned v;
 501:     daddr_t daddr;
 502: 
 503:     if (dp < 0) {
 504:         n += dp;
 505:         dp = 0;
 506:     }
 507:     if (dp + n > p->p_dsize)
 508:         n = p->p_dsize - dp;
 509:     for (i = 0; i < n; i += CLSIZE, dp += CLSIZE) {
 510:         pte = dptopte(p, dp);
 511:         if (pte->pg_fod || pte->pg_pfnum == 0)
 512:             continue;
 513:         c = &cmap[pgtocm(pte->pg_pfnum)];
 514:         if (c->c_lock || c->c_free)
 515:             continue;
 516:         if (pte->pg_v) {
 517:             pte->pg_v = 0;
 518:             if (anycl(pte, pg_m))
 519:                 pte->pg_m = 1;
 520:             distcl(pte);
 521:         }
 522:         if (dirtycl(pte)) {
 523:             if (bswlist.av_forw == NULL)
 524:                 continue;
 525:             MLOCK(c);
 526:             pte->pg_m = 0;
 527:             distcl(pte);
 528:             p->p_poip++;
 529:             v = kluster(p, dptov(p, dp), pte, B_WRITE,
 530:                 &klsize, klout, (daddr_t)0);
 531:             /* THIS ASSUMES THAT p == u.u_procp */
 532:             daddr = vtod(p, v, &u.u_dmap, &u.u_smap);
 533:             (void)swap(p, daddr, ptob(v), klsize * ctob(CLSIZE),
 534:                 B_WRITE, B_DIRTY, swapdev, pte->pg_pfnum);
 535:         } else {
 536:             if (c->c_gone == 0)
 537:                 p->p_rssize -= CLSIZE;
 538:             memfree(pte, CLSIZE, 0);
 539:             cnt.v_seqfree += CLSIZE;
 540:         }
 541:     }
 542: }
 543: 
 544: unsigned maxdmap;
 545: unsigned maxtsize;
 546: 
 547: /*
 548:  * Setup the paging constants for the clock algorithm.
 549:  * Called after the system is initialized and the amount of memory
 550:  * and number of paging devices is known.
 551:  *
 552:  * Threshold constants are defined in ../machine/vmparam.h.
 553:  */
 554: vminit()
 555: {
 556: 
 557:     /*
 558: 	 * Lotsfree is threshold where paging daemon turns on.
 559: 	 */
 560:     if (lotsfree == 0) {
 561:         lotsfree = LOTSFREE / NBPG;
 562:         if (lotsfree > LOOPPAGES / LOTSFREEFRACT)
 563:             lotsfree = LOOPPAGES / LOTSFREEFRACT;
 564:     }
 565:     /*
 566: 	 * Desfree is amount of memory desired free.
 567: 	 * If less than this for extended period, do swapping.
 568: 	 */
 569:     if (desfree == 0) {
 570:         desfree = DESFREE / NBPG;
 571:         if (desfree > LOOPPAGES / DESFREEFRACT)
 572:             desfree = LOOPPAGES / DESFREEFRACT;
 573:     }
 574: 
 575:     /*
 576: 	 * Minfree is minimal amount of free memory which is tolerable.
 577: 	 */
 578:     if (minfree == 0) {
 579:         minfree = MINFREE / NBPG;
 580:         if (minfree > desfree / MINFREEFRACT)
 581:             minfree = desfree / MINFREEFRACT;
 582:     }
 583: 
 584:     /*
 585: 	 * Maxpgio thresholds how much paging is acceptable.
 586: 	 * This figures that 2/3 busy on an arm is all that is
 587: 	 * tolerable for paging.  We assume one operation per disk rev.
 588: 	 */
 589:     if (maxpgio == 0)
 590:         maxpgio = (DISKRPM * 2) / 3;
 591: 
 592:     /*
 593: 	 * Clock to scan using max of ~~10% of processor time for sampling,
 594: 	 *     this estimated to allow maximum of 200 samples per second.
 595: 	 * This yields a ``fastscan'' of roughly (with CLSIZE=2):
 596: 	 *	<=1m	2m	3m	4m	8m
 597: 	 * 	5s	10s	15s	20s	40s
 598: 	 */
 599:     if (fastscan == 0)
 600:         fastscan = 200;
 601:     if (fastscan > LOOPPAGES / 5)
 602:         fastscan = LOOPPAGES / 5;
 603: 
 604:     /*
 605: 	 * Set slow scan time to 1/2 the fast scan time.
 606: 	 */
 607:     if (slowscan == 0)
 608:         slowscan = fastscan / 2;
 609: 
 610:     /*
 611: 	 * Calculate the swap allocation constants.
 612: 	 */
 613:         if (dmmin == 0)
 614:                 dmmin = DMMIN;
 615:         if (dmmax == 0) {
 616:                 dmmax = DMMAX;
 617:         while (dmapsize(dmmin, dmmax / 2) >= MAXDSIZ && dmmax > dmmin)
 618:             dmmax /= 2;
 619:     }
 620:     maxdmap = dmapsize(dmmin, dmmax);
 621:         if (dmtext == 0)
 622:                 dmtext = DMTEXT;
 623:         if (dmtext > dmmax)
 624:                 dmtext = dmmax;
 625:     if (maxtsize == 0)
 626:         maxtsize = MAXTSIZ;
 627:     if (maxtsize > dtob(NXDAD * dmtext))
 628:         maxtsize = dtob(NXDAD * dmtext);
 629: 
 630:     /*
 631: 	 * Set up the initial limits on process VM.
 632: 	 * Set the maximum resident set size to be all
 633: 	 * of (reasonably) available memory.  This causes
 634: 	 * any single, large process to start random page
 635: 	 * replacement once it fills memory.
 636: 	 */
 637:         u.u_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
 638:         u.u_rlimit[RLIMIT_STACK].rlim_max = MIN(MAXSSIZ, maxdmap);
 639:         u.u_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
 640:         u.u_rlimit[RLIMIT_DATA].rlim_max = MIN(MAXDSIZ, maxdmap);
 641:     u.u_rlimit[RLIMIT_RSS].rlim_cur = u.u_rlimit[RLIMIT_RSS].rlim_max =
 642:         ctob(LOOPPAGES - desfree);
 643:     proc[0].p_maxrss = LOOPPAGES - desfree;
 644: }
 645: 
 646: dmapsize(dmin, dmax)
 647:     int dmin, dmax;
 648: {
 649:     register int i, blk, size = 0;
 650: 
 651:     blk = dmin;
 652:     for (i = 0; i < NDMAP; i++) {
 653:         size += blk;
 654:         if (blk < dmax)
 655:             blk *= 2;
 656:     }
 657:     return (dtob(size));
 658: }
 659: 
 660: int pushes;
 661: 
 662: #define FRONT   1
 663: #define BACK    2
 664: 
 665: /*
 666:  * The page out daemon, which runs as process 2.
 667:  *
 668:  * As long as there are at least lotsfree pages,
 669:  * this process is not run.  When the number of free
 670:  * pages stays in the range desfree to lotsfree,
 671:  * this daemon runs through the pages in the loop
 672:  * at a rate determined in vmsched().  Pageout manages
 673:  * two hands on the clock.  The front hand moves through
 674:  * memory, clearing the valid bit (simulating a reference bit),
 675:  * and stealing pages from procs that are over maxrss.
 676:  * The back hand travels a distance behind the front hand,
 677:  * freeing the pages that have not been referenced in the time
 678:  * since the front hand passed.  If modified, they are pushed to
 679:  * swap before being freed.
 680:  */
 681: pageout()
 682: {
 683:     register int count;
 684:     register int maxhand = pgtocm(maxfree);
 685:     register int fronthand, backhand;
 686: 
 687:     /*
 688: 	 * Set the two clock hands to be separated by a reasonable amount,
 689: 	 * but no more than 360 degrees apart.
 690: 	 */
 691:     backhand = 0 / CLBYTES;
 692:     fronthand = HANDSPREAD / CLBYTES;
 693:     if (fronthand >= maxhand)
 694:         fronthand = maxhand - 1;
 695: 
 696: loop:
 697:     /*
 698: 	 * Before sleeping, look to see if there are any swap I/O headers
 699: 	 * in the ``cleaned'' list that correspond to dirty
 700: 	 * pages that have been pushed asynchronously. If so,
 701: 	 * empty the list by calling cleanup().
 702: 	 *
 703: 	 * N.B.: We guarantee never to block while the cleaned list is nonempty.
 704: 	 */
 705:     (void) splbio();
 706:     if (bclnlist != NULL) {
 707:         (void) spl0();
 708:         cleanup();
 709:         goto loop;
 710:     }
 711:     sleep((caddr_t)&proc[2], PSWP+1);
 712:     (void) spl0();
 713:     count = 0;
 714:     pushes = 0;
 715:     while (nscan < desscan && freemem < lotsfree) {
 716:         /*
 717: 		 * If checkpage manages to add a page to the free list,
 718: 		 * we give ourselves another couple of trips around the loop.
 719: 		 */
 720:         if (checkpage(fronthand, FRONT))
 721:             count = 0;
 722:         if (checkpage(backhand, BACK))
 723:             count = 0;
 724:         cnt.v_scan++;
 725:         nscan++;
 726:         if (++fronthand >= maxhand) {
 727:             fronthand = 0;
 728:             cnt.v_rev++;
 729:             if (count > 2) {
 730:                 /*
 731: 				 * Extremely unlikely, but we went around
 732: 				 * the loop twice and didn't get anywhere.
 733: 				 * Don't cycle, stop till the next clock tick.
 734: 				 */
 735:                 goto loop;
 736:             }
 737:             count++;
 738:         }
 739:         if (++backhand >= maxhand)
 740:             backhand = 0;
 741:     }
 742:     goto loop;
 743: }
 744: 
 745: /*
 746:  * An iteration of the clock pointer (hand) around the loop.
 747:  * Look at the page at hand.  If it is a
 748:  * locked (for physical i/o e.g.), system (u., page table)
 749:  * or free, then leave it alone.
 750:  * Otherwise, if we are running the front hand,
 751:  * invalidate the page for simulation of the reference bit.
 752:  * If the proc is over maxrss, we take it.
 753:  * If running the back hand, check whether the page
 754:  * has been reclaimed.  If not, free the page,
 755:  * pushing it to disk first if necessary.
 756:  */
 757: checkpage(hand, whichhand)
 758:     int hand, whichhand;
 759: {
 760:     register struct proc *rp;
 761:     register struct text *xp;
 762:     register struct cmap *c;
 763:     register struct pte *pte;
 764:     swblk_t daddr;
 765:     unsigned v;
 766:     int klsize;
 767: 
 768: top:
 769:     /*
 770: 	 * Find a process and text pointer for the
 771: 	 * page, and a virtual page number in either the
 772: 	 * process or the text image.
 773: 	 */
 774:     c = &cmap[hand];
 775:     if (c->c_lock || c->c_free)
 776:         return (0);
 777:     switch (c->c_type) {
 778: 
 779:     case CSYS:
 780:         return (0);
 781: 
 782:     case CTEXT:
 783:         xp = &text[c->c_ndx];
 784:         rp = xp->x_caddr;
 785:         v = tptov(rp, c->c_page);
 786:         pte = tptopte(rp, c->c_page);
 787:         break;
 788: 
 789:     case CDATA:
 790:     case CSTACK:
 791:         rp = &proc[c->c_ndx];
 792:         while (rp->p_flag & SNOVM)
 793:             rp = rp->p_xlink;
 794:         xp = rp->p_textp;
 795:         if (c->c_type == CDATA) {
 796:             v = dptov(rp, c->c_page);
 797:             pte = dptopte(rp, c->c_page);
 798:         } else {
 799:             v = sptov(rp, c->c_page);
 800:             pte = sptopte(rp, c->c_page);
 801:         }
 802:         break;
 803:     }
 804: 
 805:     if (pte->pg_pfnum != cmtopg(hand))
 806:         panic("bad c_page");
 807: 
 808:     /*
 809: 	 * If page is valid; make invalid but reclaimable.
 810: 	 * If this pte is not valid, then it must be reclaimable
 811: 	 * and we can add it to the free list.
 812: 	 */
 813:     if (pte->pg_v) {
 814:         if (whichhand == BACK)
 815:             return(0);
 816:         pte->pg_v = 0;
 817:         if (anycl(pte, pg_m))
 818:             pte->pg_m = 1;
 819:         distcl(pte);
 820:         if (c->c_type == CTEXT)
 821:             distpte(xp, (unsigned)vtotp(rp, v), pte);
 822:         if ((rp->p_flag & (SSEQL|SUANOM)) == 0 &&
 823:             rp->p_rssize <= rp->p_maxrss)
 824:             return (0);
 825:     }
 826:     if (c->c_type != CTEXT) {
 827:         /*
 828: 		 * Guarantee a minimal investment in data
 829: 		 * space for jobs in balance set.
 830: 		 */
 831:         if (rp->p_rssize < saferss - rp->p_slptime)
 832:             return (0);
 833:     }
 834: 
 835:     /*
 836: 	 * If the page is currently dirty, we
 837: 	 * have to arrange to have it cleaned before it
 838: 	 * can be freed.  We mark it clean immediately.
 839: 	 * If it is reclaimed while being pushed, then modified
 840: 	 * again, we are assured of the correct order of
 841: 	 * writes because we lock the page during the write.
 842: 	 * This guarantees that a swap() of this process (and
 843: 	 * thus this page), initiated in parallel, will,
 844: 	 * in fact, push the page after us.
 845: 	 *
 846: 	 * The most general worst case here would be for
 847: 	 * a reclaim, a modify and a swapout to occur
 848: 	 * all before the single page transfer completes.
 849: 	 */
 850:     if (dirtycl(pte)) {
 851:         /*
 852: 		 * If the process is being swapped out
 853: 		 * or about to exit, do not bother with its
 854: 		 * dirty pages
 855: 		 */
 856:         if (rp->p_flag & (SLOCK|SWEXIT))
 857:             return (0);
 858:         /*
 859: 		 * Limit pushes to avoid saturating
 860: 		 * pageout device.
 861: 		 */
 862:         if (pushes > maxpgio / RATETOSCHEDPAGING)
 863:             return (0);
 864:         pushes++;
 865: 
 866:         /*
 867: 		 * Now carefully make sure that there will
 868: 		 * be a header available for the push so that
 869: 		 * we will not block waiting for a header in
 870: 		 * swap().  The reason this is important is
 871: 		 * that we (proc[2]) are the one who cleans
 872: 		 * dirty swap headers and we could otherwise
 873: 		 * deadlock waiting for ourselves to clean
 874: 		 * swap headers.  The sleep here on &proc[2]
 875: 		 * is actually (effectively) a sleep on both
 876: 		 * ourselves and &bswlist, and this is known
 877: 		 * to swdone and swap in vm_swp.c.  That is,
 878: 		 * &proc[2] will be awakened both when dirty
 879: 		 * headers show up and also to get the pageout
 880: 		 * daemon moving.
 881: 		 */
 882: loop2:
 883:         (void) splbio();
 884:         if (bclnlist != NULL) {
 885:             (void) spl0();
 886:             cleanup();
 887:             goto loop2;
 888:         }
 889:         if (bswlist.av_forw == NULL) {
 890:             bswlist.b_flags |= B_WANTED;
 891:             sleep((caddr_t)&proc[2], PSWP+2);
 892:             (void) spl0();
 893:             /*
 894: 			 * Page disposition may have changed
 895: 			 * since process may have exec'ed,
 896: 			 * forked, exited or just about
 897: 			 * anything else... try this page
 898: 			 * frame again, from the top.
 899: 			 */
 900:             goto top;
 901:         }
 902:         (void) spl0();
 903: 
 904:         MLOCK(c);
 905:         uaccess(rp, Pushmap, &pushutl);
 906:         /*
 907: 		 * Now committed to pushing the page...
 908: 		 */
 909:         pte->pg_m = 0;
 910:         distcl(pte);
 911:         if (c->c_type == CTEXT)  {
 912:             xp->x_poip++;
 913:             distpte(xp, (unsigned)vtotp(rp, v), pte);
 914:         } else
 915:             rp->p_poip++;
 916:         v = kluster(rp, v, pte, B_WRITE, &klsize, klout, (daddr_t)0);
 917:         if (klsize == 0)
 918:             panic("pageout klsize");
 919:         daddr = vtod(rp, v, &pushutl.u_dmap, &pushutl.u_smap);
 920:         (void)swap(rp, daddr, ptob(v), klsize * ctob(CLSIZE),
 921:             B_WRITE, B_DIRTY, swapdev, pte->pg_pfnum);
 922:         /*
 923: 		 * The cleaning of this page will be
 924: 		 * completed later, in cleanup() called
 925: 		 * (synchronously) by us (proc[2]).  In
 926: 		 * the meantime, the page frame is locked
 927: 		 * so no havoc can result.
 928: 		 */
 929:         return (1); /* well, it'll be free soon */
 930: 
 931:     }
 932:     /*
 933: 	 * Decrement the resident set size of the current
 934: 	 * text object/process, and put the page in the
 935: 	 * free list. Note that we don't give memfree the
 936: 	 * pte as its argument, since we don't want to destroy
 937: 	 * the pte.  If it hasn't already been discarded
 938: 	 * it may yet have a chance to be reclaimed from
 939: 	 * the free list.
 940: 	 */
 941:     if (c->c_gone == 0)
 942:         if (c->c_type == CTEXT)
 943:             xp->x_rssize -= CLSIZE;
 944:         else
 945:             rp->p_rssize -= CLSIZE;
 946:     memfree(pte, CLSIZE, 0);
 947:     cnt.v_dfree += CLSIZE;
 948:     return (1);     /* freed a page! */
 949: }
 950: 
 951: /*
 952:  * Process the ``cleaned'' list.
 953:  *
 954:  * Scan through the linked list of swap I/O headers
 955:  * and free the corresponding pages that have been
 956:  * cleaned by being written back to the paging area.
 957:  * If the page has been reclaimed during this time,
 958:  * we do not free the page.  As they are processed,
 959:  * the swap I/O headers are removed from the cleaned
 960:  * list and inserted into the free list.
 961:  */
 962: cleanup()
 963: {
 964:     register struct buf *bp;
 965:     register struct proc *rp;
 966:     register struct text *xp;
 967:     register struct cmap *c;
 968:     register struct pte *pte;
 969:     struct pte *upte;
 970:     unsigned pf;
 971:     register int i;
 972:     int s, center;
 973: 
 974:     for (;;) {
 975:         s = splbio();
 976:         if ((bp = bclnlist) == 0)
 977:             break;
 978:         bclnlist = bp->av_forw;
 979:         splx(s);
 980:         pte = vtopte(&proc[2], btop(bp->b_un.b_addr));
 981:         center = 0;
 982:         for (i = 0; i < bp->b_bcount; i += CLSIZE * NBPG) {
 983:             pf = pte->pg_pfnum;
 984:             c = &cmap[pgtocm(pf)];
 985:             MUNLOCK(c);
 986:             if (pf != bp->b_pfcent) {
 987:                 if (c->c_gone) {
 988:                     memfree(pte, CLSIZE, 0);
 989:                     cnt.v_dfree += CLSIZE;
 990:                 }
 991:                 goto skip;
 992:             }
 993:             center++;
 994:             switch (c->c_type) {
 995: 
 996:             case CSYS:
 997:                 panic("cleanup CSYS");
 998: 
 999:             case CTEXT:
1000:                 xp = &text[c->c_ndx];
1001:                 xp->x_poip--;
1002:                 if (xp->x_poip == 0)
1003:                     wakeup((caddr_t)&xp->x_poip);
1004:                 break;
1005: 
1006:             case CDATA:
1007:             case CSTACK:
1008:                 rp = &proc[c->c_ndx];
1009:                 while (rp->p_flag & SNOVM)
1010:                     rp = rp->p_xlink;
1011:                 rp->p_poip--;
1012:                 if (rp->p_poip == 0)
1013:                     wakeup((caddr_t)&rp->p_poip);
1014:                 break;
1015:             }
1016:             if (c->c_gone == 0) {
1017:                 switch (c->c_type) {
1018: 
1019:                 case CTEXT:
1020:                     upte = tptopte(xp->x_caddr, c->c_page);
1021:                     break;
1022: 
1023:                 case CDATA:
1024:                     upte = dptopte(rp, c->c_page);
1025:                     break;
1026: 
1027:                 case CSTACK:
1028:                     upte = sptopte(rp, c->c_page);
1029:                     break;
1030:                 }
1031:                 if (upte->pg_v)
1032:                     goto skip;
1033:                 if (c->c_type == CTEXT)
1034:                     xp->x_rssize -= CLSIZE;
1035:                 else
1036:                     rp->p_rssize -= CLSIZE;
1037:             }
1038:             memfree(pte, CLSIZE, 0);
1039:             cnt.v_dfree += CLSIZE;
1040: skip:
1041:             pte += CLSIZE;
1042:         }
1043:         if (center != 1)
1044:             panic("cleanup center");
1045:         bp->b_flags = 0;
1046:         bp->av_forw = bswlist.av_forw;
1047:         bswlist.av_forw = bp;
1048:         if (bswlist.b_flags & B_WANTED) {
1049:             bswlist.b_flags &= ~B_WANTED;
1050:             wakeup((caddr_t)&bswlist);
1051:         }
1052:     }
1053:     splx(s);
1054: }
1055: 
1056: /*
1057:  * Kluster locates pages adjacent to the argument pages
1058:  * that are immediately available to include in the pagein/pageout,
1059:  * and given the availability of memory includes them.
1060:  * It knows that the process image is contiguous in chunks;
1061:  * an assumption here is that CLSIZE * KLMAX is a divisor of dmmin,
1062:  * so that by looking at KLMAX chunks of pages, all such will
1063:  * necessarily be mapped swap contiguous.
1064:  */
1065: int noklust;
1066: int klicnt[KLMAX];
1067: int klocnt[KLMAX];
1068: 
1069: kluster(p, v, pte0, rw, pkl, klsize, bn0)
1070:     register struct proc *p;
1071:     unsigned v;
1072:     struct pte *pte0;
1073:     int rw;
1074:     register int *pkl;
1075:     int klsize;
1076:     daddr_t bn0;
1077: {
1078:     int type, cl, clmax;
1079:     int kloff, k, klmax;
1080:     register struct pte *pte;
1081:     int klback, klforw;
1082:     int i;
1083:     unsigned v0;
1084:     daddr_t bn;
1085:     register struct cmap *c;
1086: 
1087:     if (rw == B_READ)
1088:         klicnt[0]++;
1089:     else
1090:         klocnt[0]++;
1091:     *pkl = 1;
1092:     if (noklust || klsize <= 1 || klsize > KLMAX || (klsize & (klsize - 1)))
1093:         return (v);
1094:     if (rw == B_READ && freemem < CLSIZE * KLMAX)
1095:         return (v);
1096:     if (isassv(p, v)) {
1097:         type = CSTACK;
1098:         cl = vtosp(p, v) / CLSIZE;
1099:         clmax = p->p_ssize / CLSIZE;
1100:     } else if (isadsv(p, v)) {
1101:         type = CDATA;
1102:         cl = vtodp(p, v) / CLSIZE;
1103:         clmax = p->p_dsize / CLSIZE;
1104:     } else {
1105:         type = CTEXT;
1106:         cl = vtotp(p, v) / CLSIZE;
1107:         clmax = p->p_textp->x_size / CLSIZE;
1108:     }
1109:     kloff = cl & (klsize - 1);
1110:     pte = pte0;
1111:     bn = bn0;
1112:     for (k = kloff; --k >= 0;) {
1113:         if (type == CSTACK)
1114:             pte += CLSIZE;
1115:         else
1116:             pte -= CLSIZE;
1117:         if (type == CTEXT && rw == B_READ && bn) {
1118:             bn -= btodb(CLBYTES);
1119:             if (mfind(swapdev, bn))
1120:                 break;
1121:         }
1122:         if (!klok(pte, rw))
1123:             break;
1124:     }
1125:     klback = (kloff - k) - 1;
1126:     pte = pte0;
1127:     if ((cl - kloff) + klsize > clmax)
1128:         klmax = clmax - (cl - kloff);
1129:     else
1130:         klmax = klsize;
1131:     bn = bn0;
1132:     for (k = kloff; ++k < klmax;) {
1133:         if (type == CSTACK)
1134:             pte -= CLSIZE;
1135:         else
1136:             pte += CLSIZE;
1137:         if (type == CTEXT && rw == B_READ && bn) {
1138:             bn += btodb(CLBYTES);
1139:             if (mfind(swapdev, bn))
1140:                 break;
1141:         }
1142:         if (!klok(pte, rw))
1143:             break;
1144:     }
1145:     klforw = (k - kloff) - 1;
1146:     if (klforw + klback == 0)
1147:         return (v);
1148:     pte = pte0;
1149:     if (type == CSTACK) {
1150:         pte -= klforw * CLSIZE;
1151:         v -= klforw * CLSIZE;
1152:     } else {
1153:         pte -= klback * CLSIZE;
1154:         v -= klback * CLSIZE;
1155:     }
1156:     *pkl = klforw + klback + 1;
1157:     if (rw == B_READ)
1158:         klicnt[0]--, klicnt[*pkl - 1]++;
1159:     else
1160:         klocnt[0]--, klocnt[*pkl - 1]++;
1161:     v0 = v;
1162:     for (i = 0; i < *pkl; i++) {
1163:         if (pte == pte0)
1164:             goto cont;
1165:         if (rw == B_WRITE) {
1166:             c = &cmap[pgtocm(pte->pg_pfnum)];
1167:             MLOCK(c);
1168:             pte->pg_m = 0;
1169:             distcl(pte);
1170:             if (type == CTEXT)
1171:                 distpte(p->p_textp, (unsigned)vtotp(p, v), pte);
1172:         } else {
1173:             struct pte opte;
1174: 
1175:             opte = *pte;
1176:             if (memall(pte, CLSIZE, p, type) == 0)
1177:                 panic("kluster");
1178:             pte->pg_prot = opte.pg_prot;
1179:             cmap[pgtocm(pte->pg_pfnum)].c_intrans = 1;
1180:             distcl(pte);
1181:             if (type == CTEXT) {
1182:                 p->p_textp->x_rssize += CLSIZE;
1183:                 distpte(p->p_textp, (unsigned)vtotp(p, v), pte);
1184:             } else
1185:                 p->p_rssize += CLSIZE;
1186:             distcl(pte);
1187:         }
1188: cont:
1189:         pte += CLSIZE;
1190:         v += CLSIZE;
1191:     }
1192:     return (v0);
1193: }
1194: 
1195: klok(pte, rw)
1196:     register struct pte *pte;
1197:     int rw;
1198: {
1199:     register struct cmap *c;
1200: 
1201:     if (rw == B_WRITE) {
1202:         if (pte->pg_fod)
1203:             return (0);
1204:         if (pte->pg_pfnum == 0)
1205:             return (0);
1206:         c = &cmap[pgtocm(pte->pg_pfnum)];
1207:         if (c->c_lock || c->c_intrans)
1208:             return (0);
1209:         if (!dirtycl(pte))
1210:             return (0);
1211:         return (1);
1212:     } else {
1213:         if (pte->pg_fod)
1214:             return (0);
1215:         if (pte->pg_pfnum)
1216:             return (0);
1217:         return (1);
1218:     }
1219: }
1220: 
1221: /*
1222:  * Fodkluster locates pages adjacent to the argument pages
1223:  * that are immediately available to include in the pagein,
1224:  * and given the availability of memory includes them.
1225:  * It wants to page in a file system block if it can.
1226:  */
1227: int nofodklust;
1228: int fodklcnt[KLMAX];
1229: 
1230: fodkluster(p, v0, pte0, pkl, dev, pbn)
1231:     register struct proc *p;
1232:     unsigned v0;
1233:     struct pte *pte0;
1234:     int *pkl;
1235:     dev_t dev;
1236:     daddr_t *pbn;
1237: {
1238:     register struct pte *pte;
1239:     register struct fpte *fpte;
1240:     struct cmap *c;
1241:     register daddr_t bn;
1242:     daddr_t bnswap;
1243:     unsigned v, vmin, vmax;
1244:     register int klsize;
1245:     int klback, type, i;
1246: 
1247:     if (nofodklust)
1248:         return (v0);
1249:     fodklcnt[0]++;
1250:     *pkl = 1;
1251:     if (freemem < KLMAX)
1252:         return (v0);
1253:     if (isatsv(p, v0)) {
1254:         type = CTEXT;
1255:         vmin = tptov(p, 0);
1256:         vmax = tptov(p, clrnd(p->p_tsize) - CLSIZE);
1257:     } else {
1258:         type = CDATA;
1259:         vmin = dptov(p, 0);
1260:         vmax = dptov(p, clrnd(p->p_dsize) - CLSIZE);
1261:     }
1262:     fpte = (struct fpte *)pte0;
1263:     bn = *pbn;
1264:     v = v0;
1265:     for (klsize = 1; klsize < KLMAX; klsize++) {
1266:         v -= CLSIZE;
1267:         if (v < vmin)
1268:             break;
1269:         fpte -= CLSIZE;
1270:         if (fpte->pg_fod == 0)
1271:             break;
1272:         bn -= btodb(CLBYTES);
1273:         if (fpte->pg_blkno != bn)
1274:             break;
1275:         if (type == CTEXT) {
1276:             if (mfind(dev, bn))
1277:                 break;
1278:             /*
1279: 			 * Flush any previous text page use of this
1280: 			 * swap device block.
1281: 			 */
1282:             bnswap = vtod(p, v, &u.u_dmap, &u.u_smap);
1283:             c = mfind(swapdev, bnswap);
1284:             if (c)
1285:                 munhash(swapdev, bnswap);
1286:         }
1287:     }
1288:     klback = klsize - 1;
1289:     fpte = (struct fpte *)pte0;
1290:     bn = *pbn;
1291:     v = v0;
1292:     for (; klsize < KLMAX; klsize++) {
1293:         v += CLSIZE;
1294:         if (v > vmax)
1295:             break;
1296:         fpte += CLSIZE;
1297:         if (fpte->pg_fod == 0)
1298:             break;
1299:         bn += btodb(CLBYTES);
1300:         if (fpte->pg_blkno != bn)
1301:             break;
1302:         if (type == CTEXT) {
1303:             if (mfind(dev, bn))
1304:                 break;
1305:             /*
1306: 			 * Flush any previous text page use of this
1307: 			 * swap device block.
1308: 			 */
1309:             bnswap = vtod(p, v, &u.u_dmap, &u.u_smap);
1310:             c = mfind(swapdev, bnswap);
1311:             if (c)
1312:                 munhash(swapdev, bnswap);
1313:         }
1314:     }
1315:     if (klsize == 1)
1316:         return (v0);
1317:     pte = pte0;
1318:     pte -= klback * CLSIZE;
1319:     v0 -= klback * CLSIZE;
1320:     *pbn -= klback * btodb(CLBYTES);
1321:     *pkl = klsize;
1322:     fodklcnt[0]--; fodklcnt[klsize - 1]++;
1323:     v = v0;
1324:     for (i = 0; i < klsize; i++) {
1325:         if (pte != pte0) {
1326:             struct pte opte;
1327:             int pf;
1328: 
1329:             opte = *pte;
1330:             if (memall(pte, CLSIZE, p, type) == 0)
1331:                 panic("fodkluster");
1332:             pte->pg_prot = opte.pg_prot;
1333:             pf = pte->pg_pfnum;
1334:             pte->pg_m = 1;
1335:             cmap[pgtocm(pf)].c_intrans = 1;
1336:             distcl(pte);
1337:             if (type == CTEXT) {
1338:                 p->p_textp->x_rssize += CLSIZE;
1339:                 distpte(p->p_textp, (unsigned)vtotp(p, v), pte);
1340:             } else
1341:                 p->p_rssize += CLSIZE;
1342:             distcl(pte);
1343:         }
1344:         pte += CLSIZE;
1345:         v += CLSIZE;
1346:     }
1347:     return (v0);
1348: }

Defined functions

Defined variables

Defined macros