vm.c revision 1.28
1 /* $NetBSD: vm.c,v 1.28 2008/01/27 00:16:22 pooka Exp $ */ 2 3 /* 4 * Copyright (c) 2007 Antti Kantee. All Rights Reserved. 5 * 6 * Development of this software was supported by Google Summer of Code. 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 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS 18 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 19 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 20 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 23 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 /* 31 * Virtual memory emulation routines. Contents: 32 * + UBC 33 * + anon objects & pager 34 * + vnode objects & pager 35 * + misc support routines 36 * + kmem 37 */ 38 39 /* 40 * XXX: we abuse pg->uanon for the virtual address of the storage 41 * for each page. phys_addr would fit the job description better, 42 * except that it will create unnecessary lossage on some platforms 43 * due to not being a pointer type. 44 */ 45 46 #include <sys/param.h> 47 #include <sys/null.h> 48 #include <sys/vnode.h> 49 #include <sys/buf.h> 50 #include <sys/kmem.h> 51 52 #include <uvm/uvm.h> 53 #include <uvm/uvm_prot.h> 54 #include <uvm/uvm_readahead.h> 55 56 #include <machine/pmap.h> 57 58 #include "rump_private.h" 59 #include "rumpuser.h" 60 61 /* dumdidumdum */ 62 #define len2npages(off, len) \ 63 (((((len) + PAGE_MASK) & ~(PAGE_MASK)) >> PAGE_SHIFT) \ 64 + (((off & PAGE_MASK) + (len & PAGE_MASK)) > PAGE_SIZE)) 65 66 static int vn_get(struct uvm_object *, voff_t, struct vm_page **, 67 int *, int, vm_prot_t, int, int); 68 static int vn_put(struct uvm_object *, voff_t, voff_t, int); 69 static int ao_get(struct uvm_object *, voff_t, struct vm_page **, 70 int *, int, vm_prot_t, int, int); 71 static int ao_put(struct uvm_object *, voff_t, voff_t, int); 72 73 const struct uvm_pagerops uvm_vnodeops = { 74 .pgo_get = vn_get, 75 .pgo_put = vn_put, 76 }; 77 const struct uvm_pagerops aobj_pager = { 78 .pgo_get = ao_get, 79 .pgo_put = ao_put, 80 }; 81 82 kmutex_t uvm_pageqlock; 83 84 struct uvmexp uvmexp; 85 struct uvm uvm; 86 87 struct vmspace rump_vmspace; 88 struct vm_map rump_vmmap; 89 90 /* 91 * vm pages 92 */ 93 94 /* called with the object locked */ 95 struct vm_page * 96 rumpvm_makepage(struct uvm_object *uobj, voff_t off) 97 { 98 struct vm_page *pg; 99 100 pg = kmem_zalloc(sizeof(struct vm_page), KM_SLEEP); 101 pg->offset = off; 102 pg->uobject = uobj; 103 104 pg->uanon = (void *)kmem_zalloc(PAGE_SIZE, KM_SLEEP); 105 pg->flags = PG_CLEAN|PG_BUSY|PG_FAKE; 106 107 TAILQ_INSERT_TAIL(&uobj->memq, pg, listq); 108 109 return pg; 110 } 111 112 /* 113 * Release a page. 114 * 115 * Called with the vm object locked. 116 */ 117 void 118 uvm_pagefree(struct vm_page *pg) 119 { 120 struct uvm_object *uobj = pg->uobject; 121 122 if (pg->flags & PG_WANTED) 123 wakeup(pg); 124 125 TAILQ_REMOVE(&uobj->memq, pg, listq); 126 kmem_free((void *)pg->uanon, PAGE_SIZE); 127 kmem_free(pg, sizeof(*pg)); 128 } 129 130 struct rumpva { 131 vaddr_t addr; 132 struct vm_page *pg; 133 134 LIST_ENTRY(rumpva) entries; 135 }; 136 static LIST_HEAD(, rumpva) rvahead = LIST_HEAD_INITIALIZER(rvahead); 137 static kmutex_t rvamtx; 138 139 void 140 rumpvm_enterva(vaddr_t addr, struct vm_page *pg) 141 { 142 struct rumpva *rva; 143 144 rva = kmem_alloc(sizeof(struct rumpva), KM_SLEEP); 145 rva->addr = addr; 146 rva->pg = pg; 147 mutex_enter(&rvamtx); 148 LIST_INSERT_HEAD(&rvahead, rva, entries); 149 mutex_exit(&rvamtx); 150 } 151 152 void 153 rumpvm_flushva() 154 { 155 struct rumpva *rva; 156 157 mutex_enter(&rvamtx); 158 while ((rva = LIST_FIRST(&rvahead)) != NULL) { 159 LIST_REMOVE(rva, entries); 160 kmem_free(rva, sizeof(*rva)); 161 } 162 mutex_exit(&rvamtx); 163 } 164 165 /* 166 * vnode pager 167 */ 168 169 static int 170 vn_get(struct uvm_object *uobj, voff_t off, struct vm_page **pgs, 171 int *npages, int centeridx, vm_prot_t access_type, 172 int advice, int flags) 173 { 174 struct vnode *vp = (struct vnode *)uobj; 175 176 return VOP_GETPAGES(vp, off, pgs, npages, centeridx, access_type, 177 advice, flags); 178 } 179 180 static int 181 vn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags) 182 { 183 struct vnode *vp = (struct vnode *)uobj; 184 185 return VOP_PUTPAGES(vp, offlo, offhi, flags); 186 } 187 188 /* 189 * Anon object stuff 190 */ 191 192 static int 193 ao_get(struct uvm_object *uobj, voff_t off, struct vm_page **pgs, 194 int *npages, int centeridx, vm_prot_t access_type, 195 int advice, int flags) 196 { 197 struct vm_page *pg; 198 int i; 199 200 if (centeridx) 201 panic("%s: centeridx != 0 not supported", __func__); 202 203 /* loop over pages */ 204 off = trunc_page(off); 205 for (i = 0; i < *npages; i++) { 206 retrylookup: 207 pg = uvm_pagelookup(uobj, off + (i << PAGE_SHIFT)); 208 if (pg) { 209 if (pg->flags & PG_BUSY) { 210 pg->flags |= PG_WANTED; 211 UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0, 212 "aogetpg", 0); 213 goto retrylookup; 214 } 215 pg->flags |= PG_BUSY; 216 pgs[i] = pg; 217 } else { 218 pg = rumpvm_makepage(uobj, off + (i << PAGE_SHIFT)); 219 pgs[i] = pg; 220 } 221 } 222 mutex_exit(&uobj->vmobjlock); 223 224 return 0; 225 226 } 227 228 static int 229 ao_put(struct uvm_object *uobj, voff_t start, voff_t stop, int flags) 230 { 231 struct vm_page *pg; 232 233 /* we only free all pages for now */ 234 if ((flags & PGO_FREE) == 0 || (flags & PGO_ALLPAGES) == 0) { 235 mutex_exit(&uobj->vmobjlock); 236 return 0; 237 } 238 239 while ((pg = TAILQ_FIRST(&uobj->memq)) != NULL) 240 uvm_pagefree(pg); 241 mutex_exit(&uobj->vmobjlock); 242 243 return 0; 244 } 245 246 struct uvm_object * 247 uao_create(vsize_t size, int flags) 248 { 249 struct uvm_object *uobj; 250 251 uobj = kmem_zalloc(sizeof(struct uvm_object), KM_SLEEP); 252 uobj->pgops = &aobj_pager; 253 TAILQ_INIT(&uobj->memq); 254 mutex_init(&uobj->vmobjlock, MUTEX_DEFAULT, IPL_NONE); 255 256 return uobj; 257 } 258 259 void 260 uao_detach(struct uvm_object *uobj) 261 { 262 263 ao_put(uobj, 0, 0, PGO_ALLPAGES | PGO_FREE); 264 kmem_free(uobj, sizeof(*uobj)); 265 } 266 267 /* 268 * UBC 269 */ 270 271 struct ubc_window { 272 struct uvm_object *uwin_obj; 273 voff_t uwin_off; 274 uint8_t *uwin_mem; 275 size_t uwin_mapsize; 276 277 LIST_ENTRY(ubc_window) uwin_entries; 278 }; 279 280 static LIST_HEAD(, ubc_window) uwinlst = LIST_HEAD_INITIALIZER(uwinlst); 281 static kmutex_t uwinmtx; 282 283 int 284 rump_ubc_magic_uiomove(void *va, size_t n, struct uio *uio, int *rvp, 285 struct ubc_window *uwinp) 286 { 287 struct vm_page **pgs; 288 int npages = len2npages(uio->uio_offset, n); 289 size_t allocsize; 290 int i, rv; 291 292 if (uwinp == NULL) { 293 mutex_enter(&uwinmtx); 294 LIST_FOREACH(uwinp, &uwinlst, uwin_entries) 295 if ((uint8_t *)va >= uwinp->uwin_mem 296 && (uint8_t *)va 297 < (uwinp->uwin_mem + uwinp->uwin_mapsize)) 298 break; 299 mutex_exit(&uwinmtx); 300 if (uwinp == NULL) { 301 KASSERT(rvp != NULL); 302 return 0; 303 } 304 } 305 306 allocsize = npages * sizeof(pgs); 307 pgs = kmem_zalloc(allocsize, KM_SLEEP); 308 mutex_enter(&uwinp->uwin_obj->vmobjlock); 309 rv = uwinp->uwin_obj->pgops->pgo_get(uwinp->uwin_obj, 310 uwinp->uwin_off + ((uint8_t *)va - uwinp->uwin_mem), 311 pgs, &npages, 0, 0, 0, 0); 312 if (rv) 313 goto out; 314 315 for (i = 0; i < npages; i++) { 316 size_t xfersize; 317 off_t pageoff; 318 319 pageoff = uio->uio_offset & PAGE_MASK; 320 xfersize = MIN(MIN(n, PAGE_SIZE), PAGE_SIZE-pageoff); 321 uiomove((uint8_t *)pgs[i]->uanon + pageoff, xfersize, uio); 322 if (uio->uio_rw == UIO_WRITE) 323 pgs[i]->flags &= ~PG_CLEAN; 324 n -= xfersize; 325 } 326 uvm_page_unbusy(pgs, npages); 327 328 out: 329 kmem_free(pgs, allocsize); 330 if (rvp) 331 *rvp = rv; 332 return 1; 333 } 334 335 static struct ubc_window * 336 uwin_alloc(struct uvm_object *uobj, voff_t off, vsize_t len) 337 { 338 struct ubc_window *uwinp; /* pronounced: you wimp! */ 339 340 uwinp = kmem_alloc(sizeof(struct ubc_window), KM_SLEEP); 341 uwinp->uwin_obj = uobj; 342 uwinp->uwin_off = off; 343 uwinp->uwin_mapsize = len; 344 uwinp->uwin_mem = kmem_alloc(len, KM_SLEEP); 345 346 return uwinp; 347 } 348 349 static void 350 uwin_free(struct ubc_window *uwinp) 351 { 352 353 kmem_free(uwinp->uwin_mem, uwinp->uwin_mapsize); 354 kmem_free(uwinp, sizeof(struct ubc_window)); 355 } 356 357 void * 358 ubc_alloc(struct uvm_object *uobj, voff_t offset, vsize_t *lenp, int advice, 359 int flags) 360 { 361 struct ubc_window *uwinp; 362 363 uwinp = uwin_alloc(uobj, offset, *lenp); 364 mutex_enter(&uwinmtx); 365 LIST_INSERT_HEAD(&uwinlst, uwinp, uwin_entries); 366 mutex_exit(&uwinmtx); 367 368 DPRINTF(("UBC_ALLOC offset 0x%llx, uwin %p, mem %p\n", 369 (unsigned long long)offset, uwinp, uwinp->uwin_mem)); 370 371 return uwinp->uwin_mem; 372 } 373 374 void 375 ubc_release(void *va, int flags) 376 { 377 struct ubc_window *uwinp; 378 379 mutex_enter(&uwinmtx); 380 LIST_FOREACH(uwinp, &uwinlst, uwin_entries) 381 if ((uint8_t *)va >= uwinp->uwin_mem 382 && (uint8_t *)va < (uwinp->uwin_mem + uwinp->uwin_mapsize)) 383 break; 384 mutex_exit(&uwinmtx); 385 if (uwinp == NULL) 386 panic("%s: releasing invalid window at %p", __func__, va); 387 388 LIST_REMOVE(uwinp, uwin_entries); 389 uwin_free(uwinp); 390 } 391 392 int 393 ubc_uiomove(struct uvm_object *uobj, struct uio *uio, vsize_t todo, 394 int advice, int flags) 395 { 396 struct ubc_window *uwinp; 397 vsize_t len; 398 399 while (todo > 0) { 400 len = todo; 401 402 uwinp = uwin_alloc(uobj, uio->uio_offset, len); 403 rump_ubc_magic_uiomove(uwinp->uwin_mem, len, uio, NULL, uwinp); 404 uwin_free(uwinp); 405 406 todo -= len; 407 } 408 return 0; 409 } 410 411 412 /* 413 * Misc routines 414 */ 415 416 void 417 rumpvm_init() 418 { 419 420 uvmexp.free = 1024*1024; /* XXX */ 421 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */ 422 423 mutex_init(&rvamtx, MUTEX_DEFAULT, 0); 424 mutex_init(&uwinmtx, MUTEX_DEFAULT, 0); 425 mutex_init(&uvm_pageqlock, MUTEX_DEFAULT, 0); 426 } 427 428 void 429 uvm_pageactivate(struct vm_page *pg) 430 { 431 432 /* nada */ 433 } 434 435 void 436 uvm_pagewire(struct vm_page *pg) 437 { 438 439 /* nada */ 440 } 441 442 void 443 uvm_pageunwire(struct vm_page *pg) 444 { 445 446 /* nada */ 447 } 448 449 vaddr_t 450 uvm_pagermapin(struct vm_page **pps, int npages, int flags) 451 { 452 453 panic("%s: unimplemented", __func__); 454 } 455 456 /* Called with the vm object locked */ 457 struct vm_page * 458 uvm_pagelookup(struct uvm_object *uobj, voff_t off) 459 { 460 struct vm_page *pg; 461 462 TAILQ_FOREACH(pg, &uobj->memq, listq) { 463 if (pg->offset == off) { 464 return pg; 465 } 466 } 467 468 return NULL; 469 } 470 471 struct vm_page * 472 uvm_pageratop(vaddr_t va) 473 { 474 struct rumpva *rva; 475 476 mutex_enter(&rvamtx); 477 LIST_FOREACH(rva, &rvahead, entries) 478 if (rva->addr == va) 479 break; 480 mutex_exit(&rvamtx); 481 482 if (rva == NULL) 483 panic("%s: va %llu", __func__, (unsigned long long)va); 484 485 return rva->pg; 486 } 487 488 void 489 uvm_page_unbusy(struct vm_page **pgs, int npgs) 490 { 491 struct vm_page *pg; 492 int i; 493 494 for (i = 0; i < npgs; i++) { 495 pg = pgs[i]; 496 if (pg == NULL) 497 continue; 498 499 KASSERT(pg->flags & PG_BUSY); 500 if (pg->flags & PG_WANTED) 501 wakeup(pg); 502 pg->flags &= ~(PG_WANTED|PG_BUSY); 503 } 504 } 505 506 void 507 uvm_estimatepageable(int *active, int *inactive) 508 { 509 510 /* XXX: guessing game */ 511 *active = 1024; 512 *inactive = 1024; 513 } 514 515 void 516 uvm_aio_biodone1(struct buf *bp) 517 { 518 519 panic("%s: unimplemented", __func__); 520 } 521 522 void 523 uvm_aio_biodone(struct buf *bp) 524 { 525 526 uvm_aio_aiodone(bp); 527 } 528 529 void 530 uvm_aio_aiodone(struct buf *bp) 531 { 532 533 if (((bp->b_flags | bp->b_cflags) & (B_READ | BC_NOCACHE)) == 0 && bioopsp) 534 bioopsp->io_pageiodone(bp); 535 } 536 537 void 538 uvm_vnp_setsize(struct vnode *vp, voff_t newsize) 539 { 540 541 vp->v_size = vp->v_writesize = newsize; 542 } 543 544 void 545 uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize) 546 { 547 548 vp->v_writesize = newsize; 549 } 550 551 void 552 uvm_vnp_zerorange(struct vnode *vp, off_t off, size_t len) 553 { 554 struct uvm_object *uobj = &vp->v_uobj; 555 struct vm_page **pgs; 556 int maxpages = MIN(32, round_page(len) >> PAGE_SHIFT); 557 int rv, npages, i; 558 559 pgs = kmem_zalloc(maxpages * sizeof(pgs), KM_SLEEP); 560 while (len) { 561 npages = MIN(maxpages, round_page(len) >> PAGE_SHIFT); 562 memset(pgs, 0, npages * sizeof(struct vm_page *)); 563 mutex_enter(&uobj->vmobjlock); 564 rv = uobj->pgops->pgo_get(uobj, off, pgs, &npages, 0, 0, 0, 0); 565 assert(npages > 0); 566 567 for (i = 0; i < npages; i++) { 568 uint8_t *start; 569 size_t chunkoff, chunklen; 570 571 chunkoff = off & PAGE_MASK; 572 chunklen = MIN(PAGE_SIZE - chunkoff, len); 573 start = (uint8_t *)pgs[i]->uanon + chunkoff; 574 575 memset(start, 0, chunklen); 576 pgs[i]->flags &= PG_CLEAN; 577 578 off += chunklen; 579 len -= chunklen; 580 } 581 uvm_page_unbusy(pgs, npages); 582 } 583 kmem_free(pgs, maxpages * sizeof(pgs)); 584 585 return; 586 } 587 588 struct uvm_ractx * 589 uvm_ra_allocctx() 590 { 591 592 return NULL; 593 } 594 595 void 596 uvm_ra_freectx(struct uvm_ractx *ra) 597 { 598 599 return; 600 } 601 602 bool 603 uvn_clean_p(struct uvm_object *uobj) 604 { 605 struct vnode *vp = (void *)uobj; 606 607 return (vp->v_iflag & VI_ONWORKLST) == 0; 608 } 609 610 /* 611 * Kmem 612 */ 613 614 void * 615 kmem_alloc(size_t size, km_flag_t kmflag) 616 { 617 618 return rumpuser_malloc(size, kmflag == KM_NOSLEEP); 619 } 620 621 void * 622 kmem_zalloc(size_t size, km_flag_t kmflag) 623 { 624 void *rv; 625 626 rv = kmem_alloc(size, kmflag); 627 if (rv) 628 memset(rv, 0, size); 629 630 return rv; 631 } 632 633 void 634 kmem_free(void *p, size_t size) 635 { 636 637 rumpuser_free(p); 638 } 639 640 /* 641 * UVM km 642 */ 643 644 vaddr_t 645 uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags) 646 { 647 void *rv; 648 649 rv = rumpuser_malloc(size, flags & (UVM_KMF_CANFAIL | UVM_KMF_NOWAIT)); 650 if (rv && flags & UVM_KMF_ZERO) 651 memset(rv, 0, size); 652 653 return (vaddr_t)rv; 654 } 655 656 void 657 uvm_km_free(struct vm_map *map, vaddr_t vaddr, vsize_t size, uvm_flag_t flags) 658 { 659 660 rumpuser_free((void *)vaddr); 661 } 662 663 struct vm_map * 664 uvm_km_suballoc(struct vm_map *map, vaddr_t *minaddr, vaddr_t *maxaddr, 665 vsize_t size, int pageable, bool fixed, struct vm_map_kernel *submap) 666 { 667 668 return (struct vm_map *)417416; 669 } 670 671 void 672 uvm_pageout_start(int npages) 673 { 674 675 uvmexp.paging += npages; 676 } 677 678 void 679 uvm_pageout_done(int npages) 680 { 681 682 uvmexp.paging -= npages; 683 684 /* 685 * wake up either of pagedaemon or LWPs waiting for it. 686 */ 687 688 if (uvmexp.free <= uvmexp.reserve_kernel) { 689 wakeup(&uvm.pagedaemon); 690 } else { 691 wakeup(&uvmexp.free); 692 } 693 } 694
Indexes created Mon Sep 29 21:09:56 GMT 2025