1 /* $NetBSD: uvm_anon.c,v 1.10 2000/11/25 06:27:59 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 <uvm/uvm.h> 49 #include <uvm/uvm_swap.h> 50 51 /* 52 * anonblock_list: global list of anon blocks, 53 * locked by swap_syscall_lock (since we never remove 54 * anything from this list and we only add to it via swapctl(2)). 55 */ 56 57 struct uvm_anonblock { 58 LIST_ENTRY(uvm_anonblock) list; 59 int count; 60 struct vm_anon *anons; 61 }; 62 static LIST_HEAD(anonlist, uvm_anonblock) anonblock_list; 63 64 65 static boolean_t anon_pagein __P((struct vm_anon *)); 66 67 68 /* 69 * allocate anons 70 */ 71 void 72 uvm_anon_init() 73 { 74 int nanon = uvmexp.free - (uvmexp.free / 16); /* XXXCDC ??? */ 75 76 simple_lock_init(&uvm.afreelock); 77 LIST_INIT(&anonblock_list); 78 79 /* 80 * Allocate the initial anons. 81 */ 82 uvm_anon_add(nanon); 83 } 84 85 /* 86 * add some more anons to the free pool. called when we add 87 * more swap space. 88 * 89 * => swap_syscall_lock should be held (protects anonblock_list). 90 */ 91 void 92 uvm_anon_add(count) 93 int count; 94 { 95 struct uvm_anonblock *anonblock; 96 struct vm_anon *anon; 97 int lcv, needed; 98 99 simple_lock(&uvm.afreelock); 100 uvmexp.nanonneeded += count; 101 needed = uvmexp.nanonneeded - uvmexp.nanon; 102 simple_unlock(&uvm.afreelock); 103 104 if (needed <= 0) { 105 return; 106 } 107 108 MALLOC(anonblock, void *, sizeof(*anonblock), M_UVMAMAP, M_WAITOK); 109 anon = (void *)uvm_km_alloc(kernel_map, sizeof(*anon) * needed); 110 111 /* XXX Should wait for VM to free up. */ 112 if (anonblock == NULL || anon == NULL) { 113 printf("uvm_anon_add: can not allocate %d anons\n", needed); 114 panic("uvm_anon_add"); 115 } 116 117 anonblock->count = needed; 118 anonblock->anons = anon; 119 LIST_INSERT_HEAD(&anonblock_list, anonblock, list); 120 memset(anon, 0, sizeof(*anon) * needed); 121 122 simple_lock(&uvm.afreelock); 123 uvmexp.nanon += needed; 124 uvmexp.nfreeanon += needed; 125 for (lcv = 0; lcv < needed; lcv++) { 126 simple_lock_init(&anon->an_lock); 127 anon[lcv].u.an_nxt = uvm.afree; 128 uvm.afree = &anon[lcv]; 129 simple_lock_init(&uvm.afree->an_lock); 130 } 131 simple_unlock(&uvm.afreelock); 132 } 133 134 /* 135 * remove anons from the free pool. 136 */ 137 void 138 uvm_anon_remove(count) 139 int count; 140 { 141 /* 142 * we never actually free any anons, to avoid allocation overhead. 143 * XXX someday we might want to try to free anons. 144 */ 145 146 simple_lock(&uvm.afreelock); 147 uvmexp.nanonneeded -= count; 148 simple_unlock(&uvm.afreelock); 149 } 150 151 /* 152 * allocate an anon 153 */ 154 struct vm_anon * 155 uvm_analloc() 156 { 157 struct vm_anon *a; 158 159 simple_lock(&uvm.afreelock); 160 a = uvm.afree; 161 if (a) { 162 uvm.afree = a->u.an_nxt; 163 uvmexp.nfreeanon--; 164 a->an_ref = 1; 165 a->an_swslot = 0; 166 a->u.an_page = NULL; /* so we can free quickly */ 167 } 168 simple_unlock(&uvm.afreelock); 169 return(a); 170 } 171 172 /* 173 * uvm_anfree: free a single anon structure 174 * 175 * => caller must remove anon from its amap before calling (if it was in 176 * an amap). 177 * => anon must be unlocked and have a zero reference count. 178 * => we may lock the pageq's. 179 */ 180 void 181 uvm_anfree(anon) 182 struct vm_anon *anon; 183 { 184 struct vm_page *pg; 185 UVMHIST_FUNC("uvm_anfree"); UVMHIST_CALLED(maphist); 186 UVMHIST_LOG(maphist,"(anon=0x%x)", anon, 0,0,0); 187 188 /* 189 * get page 190 */ 191 192 pg = anon->u.an_page; 193 194 /* 195 * if there is a resident page and it is loaned, then anon may not 196 * own it. call out to uvm_anon_lockpage() to ensure the real owner 197 * of the page has been identified and locked. 198 */ 199 200 if (pg && pg->loan_count) 201 pg = uvm_anon_lockloanpg(anon); 202 203 /* 204 * if we have a resident page, we must dispose of it before freeing 205 * the anon. 206 */ 207 208 if (pg) { 209 210 /* 211 * if the page is owned by a uobject (now locked), then we must 212 * kill the loan on the page rather than free it. 213 */ 214 215 if (pg->uobject) { 216 uvm_lock_pageq(); 217 KASSERT(pg->loan_count > 0); 218 pg->loan_count--; 219 pg->uanon = NULL; 220 uvm_unlock_pageq(); 221 simple_unlock(&pg->uobject->vmobjlock); 222 } else { 223 224 /* 225 * page has no uobject, so we must be the owner of it. 226 * 227 * if page is busy then we just mark it as released 228 * (who ever has it busy must check for this when they 229 * wake up). if the page is not busy then we can 230 * free it now. 231 */ 232 233 if ((pg->flags & PG_BUSY) != 0) { 234 /* tell them to dump it when done */ 235 pg->flags |= PG_RELEASED; 236 UVMHIST_LOG(maphist, 237 " anon 0x%x, page 0x%x: BUSY (released!)", 238 anon, pg, 0, 0); 239 return; 240 } 241 pmap_page_protect(pg, VM_PROT_NONE); 242 uvm_lock_pageq(); /* lock out pagedaemon */ 243 uvm_pagefree(pg); /* bye bye */ 244 uvm_unlock_pageq(); /* free the daemon */ 245 UVMHIST_LOG(maphist,"anon 0x%x, page 0x%x: freed now!", 246 anon, pg, 0, 0); 247 } 248 } 249 250 /* 251 * free any swap resources. 252 */ 253 uvm_anon_dropswap(anon); 254 255 /* 256 * now that we've stripped the data areas from the anon, free the anon 257 * itself! 258 */ 259 simple_lock(&uvm.afreelock); 260 anon->u.an_nxt = uvm.afree; 261 uvm.afree = anon; 262 uvmexp.nfreeanon++; 263 simple_unlock(&uvm.afreelock); 264 UVMHIST_LOG(maphist,"<- done!",0,0,0,0); 265 } 266 267 /* 268 * uvm_anon_dropswap: release any swap resources from this anon. 269 * 270 * => anon must be locked or have a reference count of 0. 271 */ 272 void 273 uvm_anon_dropswap(anon) 274 struct vm_anon *anon; 275 { 276 UVMHIST_FUNC("uvm_anon_dropswap"); UVMHIST_CALLED(maphist); 277 if (anon->an_swslot == 0) { 278 return; 279 } 280 281 UVMHIST_LOG(maphist,"freeing swap for anon %p, paged to swslot 0x%x", 282 anon, anon->an_swslot, 0, 0); 283 uvm_swap_free(anon->an_swslot, 1); 284 anon->an_swslot = 0; 285 286 if (anon->u.an_page == NULL) { 287 /* this page is no longer only in swap. */ 288 simple_lock(&uvm.swap_data_lock); 289 uvmexp.swpgonly--; 290 simple_unlock(&uvm.swap_data_lock); 291 } 292 } 293 294 /* 295 * uvm_anon_lockloanpg: given a locked anon, lock its resident page 296 * 297 * => anon is locked by caller 298 * => on return: anon is locked 299 * if there is a resident page: 300 * if it has a uobject, it is locked by us 301 * if it is ownerless, we take over as owner 302 * we return the resident page (it can change during 303 * this function) 304 * => note that the only time an anon has an ownerless resident page 305 * is if the page was loaned from a uvm_object and the uvm_object 306 * disowned it 307 * => this only needs to be called when you want to do an operation 308 * on an anon's resident page and that page has a non-zero loan 309 * count. 310 */ 311 struct vm_page * 312 uvm_anon_lockloanpg(anon) 313 struct vm_anon *anon; 314 { 315 struct vm_page *pg; 316 boolean_t locked = FALSE; 317 318 /* 319 * loop while we have a resident page that has a non-zero loan count. 320 * if we successfully get our lock, we will "break" the loop. 321 * note that the test for pg->loan_count is not protected -- this 322 * may produce false positive results. note that a false positive 323 * result may cause us to do more work than we need to, but it will 324 * not produce an incorrect result. 325 */ 326 327 while (((pg = anon->u.an_page) != NULL) && pg->loan_count != 0) { 328 329 /* 330 * quickly check to see if the page has an object before 331 * bothering to lock the page queues. this may also produce 332 * a false positive result, but that's ok because we do a real 333 * check after that. 334 * 335 * XXX: quick check -- worth it? need volatile? 336 */ 337 338 if (pg->uobject) { 339 340 uvm_lock_pageq(); 341 if (pg->uobject) { /* the "real" check */ 342 locked = 343 simple_lock_try(&pg->uobject->vmobjlock); 344 } else { 345 /* object disowned before we got PQ lock */ 346 locked = TRUE; 347 } 348 uvm_unlock_pageq(); 349 350 /* 351 * if we didn't get a lock (try lock failed), then we 352 * toggle our anon lock and try again 353 */ 354 355 if (!locked) { 356 simple_unlock(&anon->an_lock); 357 358 /* 359 * someone locking the object has a chance to 360 * lock us right now 361 */ 362 363 simple_lock(&anon->an_lock); 364 continue; 365 } 366 } 367 368 /* 369 * if page is un-owned [i.e. the object dropped its ownership], 370 * then we can take over as owner! 371 */ 372 373 if (pg->uobject == NULL && (pg->pqflags & PQ_ANON) == 0) { 374 uvm_lock_pageq(); 375 pg->pqflags |= PQ_ANON; /* take ownership... */ 376 pg->loan_count--; /* ... and drop our loan */ 377 uvm_unlock_pageq(); 378 } 379 380 /* 381 * we did it! break the loop 382 */ 383 384 break; 385 } 386 return(pg); 387 } 388 389 390 391 /* 392 * page in every anon that is paged out to a range of swslots. 393 * 394 * swap_syscall_lock should be held (protects anonblock_list). 395 */ 396 397 boolean_t 398 anon_swap_off(startslot, endslot) 399 int startslot, endslot; 400 { 401 struct uvm_anonblock *anonblock; 402 403 for (anonblock = LIST_FIRST(&anonblock_list); 404 anonblock != NULL; 405 anonblock = LIST_NEXT(anonblock, list)) { 406 int i; 407 408 /* 409 * loop thru all the anons in the anonblock, 410 * paging in where needed. 411 */ 412 413 for (i = 0; i < anonblock->count; i++) { 414 struct vm_anon *anon = &anonblock->anons[i]; 415 int slot; 416 417 /* 418 * lock anon to work on it. 419 */ 420 421 simple_lock(&anon->an_lock); 422 423 /* 424 * is this anon's swap slot in range? 425 */ 426 427 slot = anon->an_swslot; 428 if (slot >= startslot && slot < endslot) { 429 boolean_t rv; 430 431 /* 432 * yup, page it in. 433 */ 434 435 /* locked: anon */ 436 rv = anon_pagein(anon); 437 /* unlocked: anon */ 438 439 if (rv) { 440 return rv; 441 } 442 } else { 443 444 /* 445 * nope, unlock and proceed. 446 */ 447 448 simple_unlock(&anon->an_lock); 449 } 450 } 451 } 452 return FALSE; 453 } 454 455 456 /* 457 * fetch an anon's page. 458 * 459 * => anon must be locked, and is unlocked upon return. 460 * => returns TRUE if pagein was aborted due to lack of memory. 461 */ 462 463 static boolean_t 464 anon_pagein(anon) 465 struct vm_anon *anon; 466 { 467 struct vm_page *pg; 468 struct uvm_object *uobj; 469 int rv; 470 471 /* locked: anon */ 472 rv = uvmfault_anonget(NULL, NULL, anon); 473 /* 474 * if rv == VM_PAGER_OK, anon is still locked, else anon 475 * is unlocked 476 */ 477 478 switch (rv) { 479 case VM_PAGER_OK: 480 break; 481 482 case VM_PAGER_ERROR: 483 case VM_PAGER_REFAULT: 484 485 /* 486 * nothing more to do on errors. 487 * VM_PAGER_REFAULT can only mean that the anon was freed, 488 * so again there's nothing to do. 489 */ 490 491 return FALSE; 492 493 default: 494 #ifdef DIAGNOSTIC 495 panic("anon_pagein: uvmfault_anonget -> %d", rv); 496 #else 497 return FALSE; 498 #endif 499 } 500 501 /* 502 * ok, we've got the page now. 503 * mark it as dirty, clear its swslot and un-busy it. 504 */ 505 506 pg = anon->u.an_page; 507 uobj = pg->uobject; 508 uvm_swap_free(anon->an_swslot, 1); 509 anon->an_swslot = 0; 510 pg->flags &= ~(PG_CLEAN); 511 512 /* 513 * deactivate the page (to put it on a page queue) 514 */ 515 516 pmap_clear_reference(pg); 517 pmap_page_protect(pg, VM_PROT_NONE); 518 uvm_lock_pageq(); 519 uvm_pagedeactivate(pg); 520 uvm_unlock_pageq(); 521 522 /* 523 * unlock the anon and we're done. 524 */ 525 526 simple_unlock(&anon->an_lock); 527 if (uobj) { 528 simple_unlock(&uobj->vmobjlock); 529 } 530 return FALSE; 531 } 532