1 /* $NetBSD: uvm_anon.c,v 1.21 2001/11/10 07:36:59 lukem 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 <sys/cdefs.h> 40 __KERNEL_RCSID(0, "$NetBSD: uvm_anon.c,v 1.21 2001/11/10 07:36:59 lukem Exp $"); 41 42 #include "opt_uvmhist.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/proc.h> 47 #include <sys/malloc.h> 48 #include <sys/pool.h> 49 #include <sys/kernel.h> 50 51 #include <uvm/uvm.h> 52 #include <uvm/uvm_swap.h> 53 54 /* 55 * anonblock_list: global list of anon blocks, 56 * locked by swap_syscall_lock (since we never remove 57 * anything from this list and we only add to it via swapctl(2)). 58 */ 59 60 struct uvm_anonblock { 61 LIST_ENTRY(uvm_anonblock) list; 62 int count; 63 struct vm_anon *anons; 64 }; 65 static LIST_HEAD(anonlist, uvm_anonblock) anonblock_list; 66 67 68 static boolean_t anon_pagein __P((struct vm_anon *)); 69 70 71 /* 72 * allocate anons 73 */ 74 void 75 uvm_anon_init() 76 { 77 int nanon = uvmexp.free - (uvmexp.free / 16); /* XXXCDC ??? */ 78 79 simple_lock_init(&uvm.afreelock); 80 LIST_INIT(&anonblock_list); 81 82 /* 83 * Allocate the initial anons. 84 */ 85 uvm_anon_add(nanon); 86 } 87 88 /* 89 * add some more anons to the free pool. called when we add 90 * more swap space. 91 * 92 * => swap_syscall_lock should be held (protects anonblock_list). 93 */ 94 int 95 uvm_anon_add(count) 96 int count; 97 { 98 struct uvm_anonblock *anonblock; 99 struct vm_anon *anon; 100 int lcv, needed; 101 102 simple_lock(&uvm.afreelock); 103 uvmexp.nanonneeded += count; 104 needed = uvmexp.nanonneeded - uvmexp.nanon; 105 simple_unlock(&uvm.afreelock); 106 107 if (needed <= 0) { 108 return 0; 109 } 110 anon = (void *)uvm_km_alloc(kernel_map, sizeof(*anon) * needed); 111 if (anon == NULL) { 112 simple_lock(&uvm.afreelock); 113 uvmexp.nanonneeded -= count; 114 simple_unlock(&uvm.afreelock); 115 return ENOMEM; 116 } 117 MALLOC(anonblock, void *, sizeof(*anonblock), M_UVMAMAP, M_WAITOK); 118 119 anonblock->count = needed; 120 anonblock->anons = anon; 121 LIST_INSERT_HEAD(&anonblock_list, anonblock, list); 122 memset(anon, 0, sizeof(*anon) * needed); 123 124 simple_lock(&uvm.afreelock); 125 uvmexp.nanon += needed; 126 uvmexp.nfreeanon += needed; 127 for (lcv = 0; lcv < needed; lcv++) { 128 simple_lock_init(&anon->an_lock); 129 anon[lcv].u.an_nxt = uvm.afree; 130 uvm.afree = &anon[lcv]; 131 simple_lock_init(&uvm.afree->an_lock); 132 } 133 simple_unlock(&uvm.afreelock); 134 return 0; 135 } 136 137 /* 138 * remove anons from the free pool. 139 */ 140 void 141 uvm_anon_remove(count) 142 int count; 143 { 144 /* 145 * we never actually free any anons, to avoid allocation overhead. 146 * XXX someday we might want to try to free anons. 147 */ 148 149 simple_lock(&uvm.afreelock); 150 uvmexp.nanonneeded -= count; 151 simple_unlock(&uvm.afreelock); 152 } 153 154 /* 155 * allocate an anon 156 * 157 * => new anon is returned locked! 158 */ 159 struct vm_anon * 160 uvm_analloc() 161 { 162 struct vm_anon *a; 163 164 simple_lock(&uvm.afreelock); 165 a = uvm.afree; 166 if (a) { 167 uvm.afree = a->u.an_nxt; 168 uvmexp.nfreeanon--; 169 a->an_ref = 1; 170 a->an_swslot = 0; 171 a->u.an_page = NULL; /* so we can free quickly */ 172 LOCK_ASSERT(simple_lock_held(&a->an_lock) == 0); 173 simple_lock(&a->an_lock); 174 } 175 simple_unlock(&uvm.afreelock); 176 return(a); 177 } 178 179 /* 180 * uvm_anfree: free a single anon structure 181 * 182 * => caller must remove anon from its amap before calling (if it was in 183 * an amap). 184 * => anon must be unlocked and have a zero reference count. 185 * => we may lock the pageq's. 186 */ 187 188 void 189 uvm_anfree(anon) 190 struct vm_anon *anon; 191 { 192 struct vm_page *pg; 193 UVMHIST_FUNC("uvm_anfree"); UVMHIST_CALLED(maphist); 194 UVMHIST_LOG(maphist,"(anon=0x%x)", anon, 0,0,0); 195 196 KASSERT(anon->an_ref == 0); 197 LOCK_ASSERT(!simple_lock_held(&anon->an_lock)); 198 199 /* 200 * get page 201 */ 202 203 pg = anon->u.an_page; 204 205 /* 206 * if there is a resident page and it is loaned, then anon may not 207 * own it. call out to uvm_anon_lockpage() to ensure the real owner 208 * of the page has been identified and locked. 209 */ 210 211 if (pg && pg->loan_count) 212 pg = uvm_anon_lockloanpg(anon); 213 214 /* 215 * if we have a resident page, we must dispose of it before freeing 216 * the anon. 217 */ 218 219 if (pg) { 220 221 /* 222 * if the page is owned by a uobject (now locked), then we must 223 * kill the loan on the page rather than free it. 224 */ 225 226 if (pg->uobject) { 227 uvm_lock_pageq(); 228 KASSERT(pg->loan_count > 0); 229 pg->loan_count--; 230 pg->uanon = NULL; 231 uvm_unlock_pageq(); 232 simple_unlock(&pg->uobject->vmobjlock); 233 } else { 234 235 /* 236 * page has no uobject, so we must be the owner of it. 237 * if page is busy then we wait until it is not busy, 238 * and then free it. 239 */ 240 241 KASSERT((pg->flags & PG_RELEASED) == 0); 242 simple_lock(&anon->an_lock); 243 pmap_page_protect(pg, VM_PROT_NONE); 244 while ((pg = anon->u.an_page) && 245 (pg->flags & PG_BUSY) != 0) { 246 pg->flags |= PG_WANTED; 247 UVM_UNLOCK_AND_WAIT(pg, &anon->an_lock, 0, 248 "anfree", 0); 249 simple_lock(&anon->an_lock); 250 } 251 if (pg) { 252 uvm_lock_pageq(); 253 uvm_pagefree(pg); 254 uvm_unlock_pageq(); 255 } 256 simple_unlock(&anon->an_lock); 257 UVMHIST_LOG(maphist, "anon 0x%x, page 0x%x: " 258 "freed now!", anon, pg, 0, 0); 259 } 260 } 261 if (pg == NULL && anon->an_swslot != 0) { 262 /* this page is no longer only in swap. */ 263 simple_lock(&uvm.swap_data_lock); 264 KASSERT(uvmexp.swpgonly > 0); 265 uvmexp.swpgonly--; 266 simple_unlock(&uvm.swap_data_lock); 267 } 268 269 /* 270 * free any swap resources. 271 */ 272 273 uvm_anon_dropswap(anon); 274 275 /* 276 * now that we've stripped the data areas from the anon, 277 * free the anon itself. 278 */ 279 280 simple_lock(&uvm.afreelock); 281 anon->u.an_nxt = uvm.afree; 282 uvm.afree = anon; 283 uvmexp.nfreeanon++; 284 simple_unlock(&uvm.afreelock); 285 UVMHIST_LOG(maphist,"<- done!",0,0,0,0); 286 } 287 288 /* 289 * uvm_anon_dropswap: release any swap resources from this anon. 290 * 291 * => anon must be locked or have a reference count of 0. 292 */ 293 void 294 uvm_anon_dropswap(anon) 295 struct vm_anon *anon; 296 { 297 UVMHIST_FUNC("uvm_anon_dropswap"); UVMHIST_CALLED(maphist); 298 299 if (anon->an_swslot == 0) 300 return; 301 302 UVMHIST_LOG(maphist,"freeing swap for anon %p, paged to swslot 0x%x", 303 anon, anon->an_swslot, 0, 0); 304 uvm_swap_free(anon->an_swslot, 1); 305 anon->an_swslot = 0; 306 } 307 308 /* 309 * uvm_anon_lockloanpg: given a locked anon, lock its resident page 310 * 311 * => anon is locked by caller 312 * => on return: anon is locked 313 * if there is a resident page: 314 * if it has a uobject, it is locked by us 315 * if it is ownerless, we take over as owner 316 * we return the resident page (it can change during 317 * this function) 318 * => note that the only time an anon has an ownerless resident page 319 * is if the page was loaned from a uvm_object and the uvm_object 320 * disowned it 321 * => this only needs to be called when you want to do an operation 322 * on an anon's resident page and that page has a non-zero loan 323 * count. 324 */ 325 struct vm_page * 326 uvm_anon_lockloanpg(anon) 327 struct vm_anon *anon; 328 { 329 struct vm_page *pg; 330 boolean_t locked = FALSE; 331 332 LOCK_ASSERT(simple_lock_held(&anon->an_lock)); 333 334 /* 335 * loop while we have a resident page that has a non-zero loan count. 336 * if we successfully get our lock, we will "break" the loop. 337 * note that the test for pg->loan_count is not protected -- this 338 * may produce false positive results. note that a false positive 339 * result may cause us to do more work than we need to, but it will 340 * not produce an incorrect result. 341 */ 342 343 while (((pg = anon->u.an_page) != NULL) && pg->loan_count != 0) { 344 345 /* 346 * quickly check to see if the page has an object before 347 * bothering to lock the page queues. this may also produce 348 * a false positive result, but that's ok because we do a real 349 * check after that. 350 */ 351 352 if (pg->uobject) { 353 uvm_lock_pageq(); 354 if (pg->uobject) { 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 /* 372 * someone locking the object has a chance to 373 * lock us right now 374 */ 375 376 simple_lock(&anon->an_lock); 377 continue; 378 } 379 } 380 381 /* 382 * if page is un-owned [i.e. the object dropped its ownership], 383 * then we can take over as owner! 384 */ 385 386 if (pg->uobject == NULL && (pg->pqflags & PQ_ANON) == 0) { 387 uvm_lock_pageq(); 388 pg->pqflags |= PQ_ANON; 389 pg->loan_count--; 390 uvm_unlock_pageq(); 391 } 392 break; 393 } 394 return(pg); 395 } 396 397 398 399 /* 400 * page in every anon that is paged out to a range of swslots. 401 * 402 * swap_syscall_lock should be held (protects anonblock_list). 403 */ 404 405 boolean_t 406 anon_swap_off(startslot, endslot) 407 int startslot, endslot; 408 { 409 struct uvm_anonblock *anonblock; 410 411 LIST_FOREACH(anonblock, &anonblock_list, list) { 412 int i; 413 414 /* 415 * loop thru all the anons in the anonblock, 416 * paging in where needed. 417 */ 418 419 for (i = 0; i < anonblock->count; i++) { 420 struct vm_anon *anon = &anonblock->anons[i]; 421 int slot; 422 423 /* 424 * lock anon to work on it. 425 */ 426 427 simple_lock(&anon->an_lock); 428 429 /* 430 * is this anon's swap slot in range? 431 */ 432 433 slot = anon->an_swslot; 434 if (slot >= startslot && slot < endslot) { 435 boolean_t rv; 436 437 /* 438 * yup, page it in. 439 */ 440 441 /* locked: anon */ 442 rv = anon_pagein(anon); 443 /* unlocked: anon */ 444 445 if (rv) { 446 return rv; 447 } 448 } else { 449 450 /* 451 * nope, unlock and proceed. 452 */ 453 454 simple_unlock(&anon->an_lock); 455 } 456 } 457 } 458 return FALSE; 459 } 460 461 462 /* 463 * fetch an anon's page. 464 * 465 * => anon must be locked, and is unlocked upon return. 466 * => returns TRUE if pagein was aborted due to lack of memory. 467 */ 468 469 static boolean_t 470 anon_pagein(anon) 471 struct vm_anon *anon; 472 { 473 struct vm_page *pg; 474 struct uvm_object *uobj; 475 int rv; 476 477 /* locked: anon */ 478 LOCK_ASSERT(simple_lock_held(&anon->an_lock)); 479 480 rv = uvmfault_anonget(NULL, NULL, anon); 481 482 /* 483 * if rv == 0, anon is still locked, else anon 484 * is unlocked 485 */ 486 487 switch (rv) { 488 case 0: 489 break; 490 491 case EIO: 492 case ERESTART: 493 494 /* 495 * nothing more to do on errors. 496 * ERESTART can only mean that the anon was freed, 497 * so again there's nothing to do. 498 */ 499 500 return FALSE; 501 } 502 503 /* 504 * ok, we've got the page now. 505 * mark it as dirty, clear its swslot and un-busy it. 506 */ 507 508 pg = anon->u.an_page; 509 uobj = pg->uobject; 510 uvm_swap_free(anon->an_swslot, 1); 511 anon->an_swslot = 0; 512 pg->flags &= ~(PG_CLEAN); 513 514 /* 515 * deactivate the page (to put it on a page queue) 516 */ 517 518 pmap_clear_reference(pg); 519 uvm_lock_pageq(); 520 uvm_pagedeactivate(pg); 521 uvm_unlock_pageq(); 522 523 /* 524 * unlock the anon and we're done. 525 */ 526 527 simple_unlock(&anon->an_lock); 528 if (uobj) { 529 simple_unlock(&uobj->vmobjlock); 530 } 531 return FALSE; 532 } 533