uvm_pager.c revision 1.85
1 /* $NetBSD: uvm_pager.c,v 1.85 2007/07/29 13:31:17 ad 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 * from: Id: uvm_pager.c,v 1.1.2.23 1998/02/02 20:38:06 chuck Exp 35 */ 36 37 /* 38 * uvm_pager.c: generic functions used to assist the pagers. 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: uvm_pager.c,v 1.85 2007/07/29 13:31:17 ad Exp $"); 43 44 #include "opt_uvmhist.h" 45 #include "opt_readahead.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/proc.h> 50 #include <sys/malloc.h> 51 #include <sys/pool.h> 52 #include <sys/vnode.h> 53 54 #include <uvm/uvm.h> 55 56 struct pool *uvm_aiobuf_pool; 57 58 /* 59 * list of uvm pagers in the system 60 */ 61 62 struct uvm_pagerops * const uvmpagerops[] = { 63 &aobj_pager, 64 &uvm_deviceops, 65 &uvm_vnodeops, 66 &ubc_pager, 67 }; 68 69 /* 70 * the pager map: provides KVA for I/O 71 */ 72 73 struct vm_map *pager_map; /* XXX */ 74 kmutex_t pager_map_wanted_lock; 75 bool pager_map_wanted; /* locked by pager map */ 76 static vaddr_t emergva; 77 static bool emerginuse; 78 79 /* 80 * uvm_pager_init: init pagers (at boot time) 81 */ 82 83 void 84 uvm_pager_init(void) 85 { 86 u_int lcv; 87 vaddr_t sva, eva; 88 89 /* 90 * init pager map 91 */ 92 93 sva = 0; 94 pager_map = uvm_km_suballoc(kernel_map, &sva, &eva, PAGER_MAP_SIZE, 0, 95 false, NULL); 96 mutex_init(&pager_map_wanted_lock, MUTEX_DEFAULT, IPL_NONE); 97 pager_map_wanted = false; 98 emergva = uvm_km_alloc(kernel_map, round_page(MAXPHYS), 0, 99 UVM_KMF_VAONLY); 100 #if defined(DEBUG) 101 if (emergva == 0) 102 panic("emergva"); 103 #endif 104 emerginuse = false; 105 106 /* 107 * init ASYNC I/O queue 108 */ 109 110 TAILQ_INIT(&uvm.aio_done); 111 112 /* 113 * call pager init functions 114 */ 115 for (lcv = 0 ; lcv < sizeof(uvmpagerops)/sizeof(struct uvm_pagerops *); 116 lcv++) { 117 if (uvmpagerops[lcv]->pgo_init) 118 uvmpagerops[lcv]->pgo_init(); 119 } 120 } 121 122 /* 123 * uvm_pagermapin: map pages into KVA (pager_map) for I/O that needs mappings 124 * 125 * we basically just map in a blank map entry to reserve the space in the 126 * map and then use pmap_enter() to put the mappings in by hand. 127 */ 128 129 vaddr_t 130 uvm_pagermapin(struct vm_page **pps, int npages, int flags) 131 { 132 vsize_t size; 133 vaddr_t kva; 134 vaddr_t cva; 135 struct vm_page *pp; 136 vm_prot_t prot; 137 const bool pdaemon = curlwp == uvm.pagedaemon_lwp; 138 UVMHIST_FUNC("uvm_pagermapin"); UVMHIST_CALLED(maphist); 139 140 UVMHIST_LOG(maphist,"(pps=0x%x, npages=%d)", pps, npages,0,0); 141 142 /* 143 * compute protection. outgoing I/O only needs read 144 * access to the page, whereas incoming needs read/write. 145 */ 146 147 prot = VM_PROT_READ; 148 if (flags & UVMPAGER_MAPIN_READ) 149 prot |= VM_PROT_WRITE; 150 151 ReStart: 152 size = npages << PAGE_SHIFT; 153 kva = 0; /* let system choose VA */ 154 155 if (uvm_map(pager_map, &kva, size, NULL, UVM_UNKNOWN_OFFSET, 0, 156 UVM_FLAG_NOMERGE | (pdaemon ? UVM_FLAG_NOWAIT : 0)) != 0) { 157 if (pdaemon) { 158 mutex_enter(&pager_map_wanted_lock); 159 if (emerginuse) { 160 mtsleep(&emergva, PVM | PNORELOCK, "emergva", 161 0, &pager_map_wanted_lock); 162 goto ReStart; 163 } 164 emerginuse = true; 165 mutex_exit(&pager_map_wanted_lock); 166 kva = emergva; 167 /* The shift implicitly truncates to PAGE_SIZE */ 168 KASSERT(npages <= (MAXPHYS >> PAGE_SHIFT)); 169 goto enter; 170 } 171 if ((flags & UVMPAGER_MAPIN_WAITOK) == 0) { 172 UVMHIST_LOG(maphist,"<- NOWAIT failed", 0,0,0,0); 173 return(0); 174 } 175 mutex_enter(&pager_map_wanted_lock); 176 pager_map_wanted = true; 177 UVMHIST_LOG(maphist, " SLEEPING on pager_map",0,0,0,0); 178 mtsleep(pager_map, PVM | PNORELOCK, "pager_map", 0, 179 &pager_map_wanted_lock); 180 goto ReStart; 181 } 182 183 enter: 184 /* got it */ 185 for (cva = kva ; size != 0 ; size -= PAGE_SIZE, cva += PAGE_SIZE) { 186 pp = *pps++; 187 KASSERT(pp); 188 KASSERT(pp->flags & PG_BUSY); 189 pmap_kenter_pa(cva, VM_PAGE_TO_PHYS(pp), prot); 190 } 191 pmap_update(vm_map_pmap(pager_map)); 192 193 UVMHIST_LOG(maphist, "<- done (KVA=0x%x)", kva,0,0,0); 194 return(kva); 195 } 196 197 /* 198 * uvm_pagermapout: remove pager_map mapping 199 * 200 * we remove our mappings by hand and then remove the mapping (waking 201 * up anyone wanting space). 202 */ 203 204 void 205 uvm_pagermapout(vaddr_t kva, int npages) 206 { 207 vsize_t size = npages << PAGE_SHIFT; 208 struct vm_map_entry *entries; 209 UVMHIST_FUNC("uvm_pagermapout"); UVMHIST_CALLED(maphist); 210 211 UVMHIST_LOG(maphist, " (kva=0x%x, npages=%d)", kva, npages,0,0); 212 213 /* 214 * duplicate uvm_unmap, but add in pager_map_wanted handling. 215 */ 216 217 pmap_kremove(kva, npages << PAGE_SHIFT); 218 if (kva == emergva) { 219 mutex_enter(&pager_map_wanted_lock); 220 emerginuse = false; 221 wakeup(&emergva); 222 mutex_exit(&pager_map_wanted_lock); 223 return; 224 } 225 226 vm_map_lock(pager_map); 227 uvm_unmap_remove(pager_map, kva, kva + size, &entries, NULL, 0); 228 mutex_enter(&pager_map_wanted_lock); 229 if (pager_map_wanted) { 230 pager_map_wanted = false; 231 wakeup(pager_map); 232 } 233 mutex_exit(&pager_map_wanted_lock); 234 vm_map_unlock(pager_map); 235 if (entries) 236 uvm_unmap_detach(entries, 0); 237 pmap_update(pmap_kernel()); 238 UVMHIST_LOG(maphist,"<- done",0,0,0,0); 239 } 240 241 /* 242 * interrupt-context iodone handler for nested i/o bufs. 243 * 244 * => must be at splbio(). 245 */ 246 247 void 248 uvm_aio_biodone1(struct buf *bp) 249 { 250 struct buf *mbp = bp->b_private; 251 252 KASSERT(mbp != bp); 253 if (bp->b_error != 0) 254 mbp->b_error = bp->b_error; 255 mbp->b_resid -= bp->b_bcount; 256 putiobuf(bp); 257 if (mbp->b_resid == 0) { 258 biodone(mbp); 259 } 260 } 261 262 /* 263 * interrupt-context iodone handler for single-buf i/os 264 * or the top-level buf of a nested-buf i/o. 265 * 266 * => must be at splbio(). 267 */ 268 269 void 270 uvm_aio_biodone(struct buf *bp) 271 { 272 /* reset b_iodone for when this is a single-buf i/o. */ 273 bp->b_iodone = uvm_aio_aiodone; 274 275 workqueue_enqueue(uvm.aiodone_queue, &bp->b_work, NULL); 276 } 277 278 /* 279 * uvm_aio_aiodone: do iodone processing for async i/os. 280 * this should be called in thread context, not interrupt context. 281 */ 282 283 void 284 uvm_aio_aiodone(struct buf *bp) 285 { 286 int npages = bp->b_bufsize >> PAGE_SHIFT; 287 struct vm_page *pg, *pgs[npages]; 288 struct uvm_object *uobj; 289 struct simplelock *slock; 290 int s, i, error, swslot; 291 bool write, swap; 292 UVMHIST_FUNC("uvm_aio_aiodone"); UVMHIST_CALLED(ubchist); 293 UVMHIST_LOG(ubchist, "bp %p", bp, 0,0,0); 294 295 error = bp->b_error; 296 write = (bp->b_flags & B_READ) == 0; 297 /* XXXUBC B_NOCACHE is for swap pager, should be done differently */ 298 if (write && !(bp->b_flags & B_NOCACHE) && bioops.io_pageiodone) { 299 (*bioops.io_pageiodone)(bp); 300 } 301 302 uobj = NULL; 303 for (i = 0; i < npages; i++) { 304 pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + (i << PAGE_SHIFT)); 305 UVMHIST_LOG(ubchist, "pgs[%d] = %p", i, pgs[i],0,0); 306 } 307 uvm_pagermapout((vaddr_t)bp->b_data, npages); 308 309 swslot = 0; 310 slock = NULL; 311 pg = pgs[0]; 312 swap = (pg->uanon != NULL && pg->uobject == NULL) || 313 (pg->pqflags & PQ_AOBJ) != 0; 314 if (!swap) { 315 uobj = pg->uobject; 316 slock = &uobj->vmobjlock; 317 simple_lock(slock); 318 uvm_lock_pageq(); 319 } else { 320 #if defined(VMSWAP) 321 if (error) { 322 if (pg->uobject != NULL) { 323 swslot = uao_find_swslot(pg->uobject, 324 pg->offset >> PAGE_SHIFT); 325 } else { 326 KASSERT(pg->uanon != NULL); 327 swslot = pg->uanon->an_swslot; 328 } 329 KASSERT(swslot); 330 } 331 #else /* defined(VMSWAP) */ 332 panic("%s: swap", __func__); 333 #endif /* defined(VMSWAP) */ 334 } 335 for (i = 0; i < npages; i++) { 336 pg = pgs[i]; 337 KASSERT(swap || pg->uobject == uobj); 338 UVMHIST_LOG(ubchist, "pg %p", pg, 0,0,0); 339 340 #if defined(VMSWAP) 341 /* 342 * for swap i/os, lock each page's object (or anon) 343 * individually since each page may need a different lock. 344 */ 345 346 if (swap) { 347 if (pg->uobject != NULL) { 348 slock = &pg->uobject->vmobjlock; 349 } else { 350 slock = &pg->uanon->an_lock; 351 } 352 simple_lock(slock); 353 uvm_lock_pageq(); 354 } 355 #endif /* defined(VMSWAP) */ 356 357 /* 358 * process errors. for reads, just mark the page to be freed. 359 * for writes, if the error was ENOMEM, we assume this was 360 * a transient failure so we mark the page dirty so that 361 * we'll try to write it again later. for all other write 362 * errors, we assume the error is permanent, thus the data 363 * in the page is lost. bummer. 364 */ 365 366 if (error) { 367 int slot; 368 if (!write) { 369 pg->flags |= PG_RELEASED; 370 continue; 371 } else if (error == ENOMEM) { 372 if (pg->flags & PG_PAGEOUT) { 373 pg->flags &= ~PG_PAGEOUT; 374 uvmexp.paging--; 375 } 376 pg->flags &= ~PG_CLEAN; 377 uvm_pageactivate(pg); 378 slot = 0; 379 } else 380 slot = SWSLOT_BAD; 381 382 #if defined(VMSWAP) 383 if (swap) { 384 if (pg->uobject != NULL) { 385 int oldslot; 386 oldslot = uao_set_swslot(pg->uobject, 387 pg->offset >> PAGE_SHIFT, slot); 388 KASSERT(oldslot == swslot + i); 389 } else { 390 KASSERT(pg->uanon->an_swslot == 391 swslot + i); 392 pg->uanon->an_swslot = slot; 393 } 394 } 395 #endif /* defined(VMSWAP) */ 396 } 397 398 /* 399 * if the page is PG_FAKE, this must have been a read to 400 * initialize the page. clear PG_FAKE and activate the page. 401 * we must also clear the pmap "modified" flag since it may 402 * still be set from the page's previous identity. 403 */ 404 405 if (pg->flags & PG_FAKE) { 406 KASSERT(!write); 407 pg->flags &= ~PG_FAKE; 408 #if defined(READAHEAD_STATS) 409 pg->pqflags |= PQ_READAHEAD; 410 uvm_ra_total.ev_count++; 411 #endif /* defined(READAHEAD_STATS) */ 412 KASSERT((pg->flags & PG_CLEAN) != 0); 413 uvm_pageenqueue(pg); 414 pmap_clear_modify(pg); 415 } 416 417 /* 418 * do accounting for pagedaemon i/o and arrange to free 419 * the pages instead of just unbusying them. 420 */ 421 422 if (pg->flags & PG_PAGEOUT) { 423 pg->flags &= ~PG_PAGEOUT; 424 uvmexp.paging--; 425 uvmexp.pdfreed++; 426 pg->flags |= PG_RELEASED; 427 } 428 429 #if defined(VMSWAP) 430 /* 431 * for swap pages, unlock everything for this page now. 432 */ 433 434 if (swap) { 435 if (pg->uobject == NULL && pg->uanon->an_ref == 0 && 436 (pg->flags & PG_RELEASED) != 0) { 437 uvm_unlock_pageq(); 438 uvm_anon_release(pg->uanon); 439 } else { 440 uvm_page_unbusy(&pg, 1); 441 uvm_unlock_pageq(); 442 simple_unlock(slock); 443 } 444 } 445 #endif /* defined(VMSWAP) */ 446 } 447 if (!swap) { 448 uvm_page_unbusy(pgs, npages); 449 uvm_unlock_pageq(); 450 simple_unlock(slock); 451 } else { 452 #if defined(VMSWAP) 453 KASSERT(write); 454 455 /* these pages are now only in swap. */ 456 mutex_enter(&uvm_swap_data_lock); 457 KASSERT(uvmexp.swpgonly + npages <= uvmexp.swpginuse); 458 if (error != ENOMEM) 459 uvmexp.swpgonly += npages; 460 mutex_exit(&uvm_swap_data_lock); 461 if (error) { 462 if (error != ENOMEM) 463 uvm_swap_markbad(swslot, npages); 464 else 465 uvm_swap_free(swslot, npages); 466 } 467 uvmexp.pdpending--; 468 #endif /* defined(VMSWAP) */ 469 } 470 s = splbio(); 471 if (write && (bp->b_flags & B_AGE) != 0) { 472 vwakeup(bp); 473 } 474 putiobuf(bp); 475 splx(s); 476 } 477 478 /* 479 * uvm_pageratop: convert KVAs in the pager map back to their page 480 * structures. 481 */ 482 483 struct vm_page * 484 uvm_pageratop(vaddr_t kva) 485 { 486 struct vm_page *pg; 487 paddr_t pa; 488 bool rv; 489 490 rv = pmap_extract(pmap_kernel(), kva, &pa); 491 KASSERT(rv); 492 pg = PHYS_TO_VM_PAGE(pa); 493 KASSERT(pg != NULL); 494 return (pg); 495 } 496
Indexes created Wed Sep 24 05:09:52 GMT 2025