1 /* $NetBSD: uvm_anon.c,v 1.5 2000/01/11 06:57:49 chs Exp $ */ 2 3 /* 4 * 5 * Copyright (c) 1997 Charles D. Cranor and Washington University. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by Charles D. Cranor and 19 * Washington University. 20 * 4. The name of the author may not be used to endorse or promote products 21 * derived from this software without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 26 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 28 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 32 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 /* 36 * uvm_anon.c: uvm anon ops 37 */ 38 39 #include "opt_uvmhist.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/proc.h> 44 #include <sys/malloc.h> 45 #include <sys/pool.h> 46 #include <sys/kernel.h> 47 48 #include <vm/vm.h> 49 #include <vm/vm_page.h> 50 #include <vm/vm_kern.h> 51 52 #include <uvm/uvm.h> 53 #include <uvm/uvm_swap.h> 54 55 /* 56 * anonblock_list: global list of anon blocks, 57 * locked by swap_syscall_lock (since we never remove 58 * anything from this list and we only add to it via swapctl(2)). 59 */ 60 61 struct uvm_anonblock { 62 LIST_ENTRY(uvm_anonblock) list; 63 int count; 64 struct vm_anon *anons; 65 }; 66 static LIST_HEAD(anonlist, uvm_anonblock) anonblock_list; 67 68 69 static boolean_t anon_pagein __P((struct vm_anon *)); 70 71 72 /* 73 * allocate anons 74 */ 75 void 76 uvm_anon_init() 77 { 78 int nanon = uvmexp.free - (uvmexp.free / 16); /* XXXCDC ??? */ 79 80 simple_lock_init(&uvm.afreelock); 81 LIST_INIT(&anonblock_list); 82 83 /* 84 * Allocate the initial anons. 85 */ 86 uvm_anon_add(nanon); 87 } 88 89 /* 90 * add some more anons to the free pool. called when we add 91 * more swap space. 92 * 93 * => swap_syscall_lock should be held (protects anonblock_list). 94 */ 95 void 96 uvm_anon_add(count) 97 int count; 98 { 99 struct uvm_anonblock *anonblock; 100 struct vm_anon *anon; 101 int lcv, needed; 102 103 simple_lock(&uvm.afreelock); 104 uvmexp.nanonneeded += count; 105 needed = uvmexp.nanonneeded - uvmexp.nanon; 106 simple_unlock(&uvm.afreelock); 107 108 if (needed <= 0) { 109 return; 110 } 111 112 MALLOC(anonblock, void *, sizeof(*anonblock), M_UVMAMAP, M_WAITOK); 113 anon = (void *)uvm_km_alloc(kernel_map, sizeof(*anon) * needed); 114 115 /* XXX Should wait for VM to free up. */ 116 if (anonblock == NULL || anon == NULL) { 117 printf("uvm_anon_add: can not allocate %d anons\n", needed); 118 panic("uvm_anon_add"); 119 } 120 121 anonblock->count = needed; 122 anonblock->anons = anon; 123 LIST_INSERT_HEAD(&anonblock_list, anonblock, list); 124 memset(anon, 0, sizeof(*anon) * needed); 125 126 simple_lock(&uvm.afreelock); 127 uvmexp.nanon += needed; 128 uvmexp.nfreeanon += needed; 129 for (lcv = 0; lcv < needed; lcv++) { 130 simple_lock_init(&anon->an_lock); 131 anon[lcv].u.an_nxt = uvm.afree; 132 uvm.afree = &anon[lcv]; 133 simple_lock_init(&uvm.afree->an_lock); 134 } 135 simple_unlock(&uvm.afreelock); 136 } 137 138 /* 139 * remove anons from the free pool. 140 */ 141 void 142 uvm_anon_remove(count) 143 int count; 144 { 145 /* 146 * we never actually free any anons, to avoid allocation overhead. 147 * XXX someday we might want to try to free anons. 148 */ 149 150 simple_lock(&uvm.afreelock); 151 uvmexp.nanonneeded -= count; 152 simple_unlock(&uvm.afreelock); 153 } 154 155 /* 156 * allocate an anon 157 */ 158 struct vm_anon * 159 uvm_analloc() 160 { 161 struct vm_anon *a; 162 163 simple_lock(&uvm.afreelock); 164 a = uvm.afree; 165 if (a) { 166 uvm.afree = a->u.an_nxt; 167 uvmexp.nfreeanon--; 168 a->an_ref = 1; 169 a->an_swslot = 0; 170 a->u.an_page = NULL; /* so we can free quickly */ 171 } 172 simple_unlock(&uvm.afreelock); 173 return(a); 174 } 175 176 /* 177 * uvm_anfree: free a single anon structure 178 * 179 * => caller must remove anon from its amap before calling (if it was in 180 * an amap). 181 * => anon must be unlocked and have a zero reference count. 182 * => we may lock the pageq's. 183 */ 184 void 185 uvm_anfree(anon) 186 struct vm_anon *anon; 187 { 188 struct vm_page *pg; 189 UVMHIST_FUNC("uvm_anfree"); UVMHIST_CALLED(maphist); 190 UVMHIST_LOG(maphist,"(anon=0x%x)", anon, 0,0,0); 191 192 /* 193 * get page 194 */ 195 196 pg = anon->u.an_page; 197 198 /* 199 * if there is a resident page and it is loaned, then anon may not 200 * own it. call out to uvm_anon_lockpage() to ensure the real owner 201 * of the page has been identified and locked. 202 */ 203 204 if (pg && pg->loan_count) 205 pg = uvm_anon_lockloanpg(anon); 206 207 /* 208 * if we have a resident page, we must dispose of it before freeing 209 * the anon. 210 */ 211 212 if (pg) { 213 214 /* 215 * if the page is owned by a uobject (now locked), then we must 216 * kill the loan on the page rather than free it. 217 */ 218 219 if (pg->uobject) { 220 221 /* kill loan */ 222 uvm_lock_pageq(); 223 #ifdef DIAGNOSTIC 224 if (pg->loan_count < 1) 225 panic("uvm_anfree: obj owned page " 226 "with no loan count"); 227 #endif 228 pg->loan_count--; 229 pg->uanon = NULL; 230 uvm_unlock_pageq(); 231 simple_unlock(&pg->uobject->vmobjlock); 232 233 } else { 234 235 /* 236 * page has no uobject, so we must be the owner of it. 237 * 238 * if page is busy then we just mark it as released 239 * (who ever has it busy must check for this when they 240 * wake up). if the page is not busy then we can 241 * free it now. 242 */ 243 244 if ((pg->flags & PG_BUSY) != 0) { 245 /* tell them to dump it when done */ 246 pg->flags |= PG_RELEASED; 247 UVMHIST_LOG(maphist, 248 " anon 0x%x, page 0x%x: BUSY (released!)", 249 anon, pg, 0, 0); 250 return; 251 } 252 253 pmap_page_protect(pg, VM_PROT_NONE); 254 uvm_lock_pageq(); /* lock out pagedaemon */ 255 uvm_pagefree(pg); /* bye bye */ 256 uvm_unlock_pageq(); /* free the daemon */ 257 258 UVMHIST_LOG(maphist," anon 0x%x, page 0x%x: freed now!", 259 anon, pg, 0, 0); 260 } 261 } 262 263 /* 264 * free any swap resources. 265 */ 266 uvm_anon_dropswap(anon); 267 268 /* 269 * now that we've stripped the data areas from the anon, free the anon 270 * itself! 271 */ 272 simple_lock(&uvm.afreelock); 273 anon->u.an_nxt = uvm.afree; 274 uvm.afree = anon; 275 uvmexp.nfreeanon++; 276 simple_unlock(&uvm.afreelock); 277 UVMHIST_LOG(maphist,"<- done!",0,0,0,0); 278 } 279 280 /* 281 * uvm_anon_dropswap: release any swap resources from this anon. 282 * 283 * => anon must be locked or have a reference count of 0. 284 */ 285 void 286 uvm_anon_dropswap(anon) 287 struct vm_anon *anon; 288 { 289 UVMHIST_FUNC("uvm_anon_dropswap"); UVMHIST_CALLED(maphist); 290 if (anon->an_swslot == 0) { 291 return; 292 } 293 294 UVMHIST_LOG(maphist,"freeing swap for anon %p, paged to swslot 0x%x", 295 anon, anon->an_swslot, 0, 0); 296 uvm_swap_free(anon->an_swslot, 1); 297 anon->an_swslot = 0; 298 299 if (anon->u.an_page == NULL) { 300 /* this page is no longer only in swap. */ 301 simple_lock(&uvm.swap_data_lock); 302 uvmexp.swpgonly--; 303 simple_unlock(&uvm.swap_data_lock); 304 } 305 } 306 307 /* 308 * uvm_anon_lockloanpg: given a locked anon, lock its resident page 309 * 310 * => anon is locked by caller 311 * => on return: anon is locked 312 * if there is a resident page: 313 * if it has a uobject, it is locked by us 314 * if it is ownerless, we take over as owner 315 * we return the resident page (it can change during 316 * this function) 317 * => note that the only time an anon has an ownerless resident page 318 * is if the page was loaned from a uvm_object and the uvm_object 319 * disowned it 320 * => this only needs to be called when you want to do an operation 321 * on an anon's resident page and that page has a non-zero loan 322 * count. 323 */ 324 struct vm_page * 325 uvm_anon_lockloanpg(anon) 326 struct vm_anon *anon; 327 { 328 struct vm_page *pg; 329 boolean_t locked = FALSE; 330 331 /* 332 * loop while we have a resident page that has a non-zero loan count. 333 * if we successfully get our lock, we will "break" the loop. 334 * note that the test for pg->loan_count is not protected -- this 335 * may produce false positive results. note that a false positive 336 * result may cause us to do more work than we need to, but it will 337 * not produce an incorrect result. 338 */ 339 340 while (((pg = anon->u.an_page) != NULL) && pg->loan_count != 0) { 341 342 /* 343 * quickly check to see if the page has an object before 344 * bothering to lock the page queues. this may also produce 345 * a false positive result, but that's ok because we do a real 346 * check after that. 347 * 348 * XXX: quick check -- worth it? need volatile? 349 */ 350 351 if (pg->uobject) { 352 353 uvm_lock_pageq(); 354 if (pg->uobject) { /* the "real" check */ 355 locked = 356 simple_lock_try(&pg->uobject->vmobjlock); 357 } else { 358 /* object disowned before we got PQ lock */ 359 locked = TRUE; 360 } 361 uvm_unlock_pageq(); 362 363 /* 364 * if we didn't get a lock (try lock failed), then we 365 * toggle our anon lock and try again 366 */ 367 368 if (!locked) { 369 simple_unlock(&anon->an_lock); 370 /* 371 * someone locking the object has a chance to 372 * lock us right now 373 */ 374 simple_lock(&anon->an_lock); 375 continue; /* start over */ 376 } 377 } 378 379 /* 380 * if page is un-owned [i.e. the object dropped its ownership], 381 * then we can take over as owner! 382 */ 383 384 if (pg->uobject == NULL && (pg->pqflags & PQ_ANON) == 0) { 385 uvm_lock_pageq(); 386 pg->pqflags |= PQ_ANON; /* take ownership... */ 387 pg->loan_count--; /* ... and drop our loan */ 388 uvm_unlock_pageq(); 389 } 390 391 /* 392 * we did it! break the loop 393 */ 394 break; 395 } 396 397 /* 398 * done! 399 */ 400 401 return(pg); 402 } 403 404 405 406 /* 407 * page in every anon that is paged out to a range of swslots. 408 * 409 * swap_syscall_lock should be held (protects anonblock_list). 410 */ 411 412 boolean_t 413 anon_swap_off(startslot, endslot) 414 int startslot, endslot; 415 { 416 struct uvm_anonblock *anonblock; 417 418 for (anonblock = LIST_FIRST(&anonblock_list); 419 anonblock != NULL; 420 anonblock = LIST_NEXT(anonblock, list)) { 421 int i; 422 423 /* 424 * loop thru all the anons in the anonblock, 425 * paging in where needed. 426 */ 427 428 for (i = 0; i < anonblock->count; i++) { 429 struct vm_anon *anon = &anonblock->anons[i]; 430 int slot; 431 432 /* 433 * lock anon to work on it. 434 */ 435 436 simple_lock(&anon->an_lock); 437 438 /* 439 * is this anon's swap slot in range? 440 */ 441 442 slot = anon->an_swslot; 443 if (slot >= startslot && slot < endslot) { 444 boolean_t rv; 445 446 /* 447 * yup, page it in. 448 */ 449 450 /* locked: anon */ 451 rv = anon_pagein(anon); 452 /* unlocked: anon */ 453 454 if (rv) { 455 return rv; 456 } 457 } else { 458 459 /* 460 * nope, unlock and proceed. 461 */ 462 463 simple_unlock(&anon->an_lock); 464 } 465 } 466 } 467 return FALSE; 468 } 469 470 471 /* 472 * fetch an anon's page. 473 * 474 * => anon must be locked, and is unlocked upon return. 475 * => returns TRUE if pagein was aborted due to lack of memory. 476 */ 477 478 static boolean_t 479 anon_pagein(anon) 480 struct vm_anon *anon; 481 { 482 struct vm_page *pg; 483 struct uvm_object *uobj; 484 int rv; 485 UVMHIST_FUNC("anon_pagein"); UVMHIST_CALLED(pdhist); 486 487 /* locked: anon */ 488 rv = uvmfault_anonget(NULL, NULL, anon); 489 /* unlocked: anon */ 490 491 switch (rv) { 492 case VM_PAGER_OK: 493 break; 494 495 case VM_PAGER_ERROR: 496 case VM_PAGER_REFAULT: 497 498 /* 499 * nothing more to do on errors. 500 * VM_PAGER_REFAULT can only mean that the anon was freed, 501 * so again there's nothing to do. 502 */ 503 504 return FALSE; 505 506 #ifdef DIAGNOSTIC 507 default: 508 panic("anon_pagein: uvmfault_anonget -> %d", rv); 509 #endif 510 } 511 512 /* 513 * ok, we've got the page now. 514 * mark it as dirty, clear its swslot and un-busy it. 515 */ 516 517 pg = anon->u.an_page; 518 uobj = pg->uobject; 519 uvm_swap_free(anon->an_swslot, 1); 520 anon->an_swslot = 0; 521 pg->flags &= ~(PG_CLEAN); 522 523 /* 524 * deactivate the page (to put it on a page queue) 525 */ 526 527 pmap_clear_reference(pg); 528 pmap_page_protect(pg, VM_PROT_NONE); 529 uvm_lock_pageq(); 530 uvm_pagedeactivate(pg); 531 uvm_unlock_pageq(); 532 533 /* 534 * unlock the anon and we're done. 535 */ 536 537 simple_unlock(&anon->an_lock); 538 if (uobj) { 539 simple_unlock(&uobj->vmobjlock); 540 } 541 return FALSE; 542 } 543