vm.c revision 1.21
1 /* $NetBSD: vm.c,v 1.21 2007/11/04 19:52:14 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 struct uvm_pagerops uvm_vnodeops; 67 struct uvm_pagerops aobj_pager; 68 struct uvmexp uvmexp; 69 struct uvm uvm; 70 71 struct vmspace rump_vmspace; 72 struct vm_map rump_vmmap; 73 74 /* 75 * vm pages 76 */ 77 78 struct vm_page * 79 rumpvm_makepage(struct uvm_object *uobj, voff_t off) 80 { 81 struct vm_page *pg; 82 83 pg = rumpuser_malloc(sizeof(struct vm_page), 0); 84 memset(pg, 0, sizeof(struct vm_page)); 85 pg->offset = off; 86 pg->uobject = uobj; 87 88 pg->uanon = (void *)rumpuser_malloc(PAGE_SIZE, 0); 89 memset((void *)pg->uanon, 0, PAGE_SIZE); 90 pg->flags = PG_CLEAN; 91 92 simple_lock(&uobj->vmobjlock); 93 TAILQ_INSERT_TAIL(&uobj->memq, pg, listq); 94 simple_unlock(&uobj->vmobjlock); 95 96 return pg; 97 } 98 99 /* 100 * Release a page. 101 * 102 * Called with the vm object locked. Returns with the lock released. 103 */ 104 void 105 rumpvm_freepage(struct vm_page *pg) 106 { 107 struct uvm_object *uobj = pg->uobject; 108 109 TAILQ_REMOVE(&uobj->memq, pg, listq); 110 simple_unlock(&uobj->vmobjlock); 111 112 rumpuser_free((void *)pg->uanon); 113 rumpuser_free(pg); 114 } 115 116 struct rumpva { 117 vaddr_t addr; 118 struct vm_page *pg; 119 120 LIST_ENTRY(rumpva) entries; 121 }; 122 static LIST_HEAD(, rumpva) rvahead = LIST_HEAD_INITIALIZER(rvahead); 123 static kmutex_t rvamtx; 124 125 void 126 rumpvm_enterva(vaddr_t addr, struct vm_page *pg) 127 { 128 struct rumpva *rva; 129 130 rva = rumpuser_malloc(sizeof(struct rumpva), 0); 131 rva->addr = addr; 132 rva->pg = pg; 133 mutex_enter(&rvamtx); 134 LIST_INSERT_HEAD(&rvahead, rva, entries); 135 mutex_exit(&rvamtx); 136 } 137 138 void 139 rumpvm_flushva() 140 { 141 struct rumpva *rva; 142 143 mutex_enter(&rvamtx); 144 while ((rva = LIST_FIRST(&rvahead)) != NULL) { 145 LIST_REMOVE(rva, entries); 146 rumpuser_free(rva); 147 } 148 mutex_exit(&rvamtx); 149 } 150 151 /* 152 * vnode pager 153 */ 154 155 static int 156 vn_get(struct uvm_object *uobj, voff_t off, struct vm_page **pgs, 157 int *npages, int centeridx, vm_prot_t access_type, 158 int advice, int flags) 159 { 160 struct vnode *vp = (struct vnode *)uobj; 161 162 return VOP_GETPAGES(vp, off, pgs, npages, centeridx, access_type, 163 advice, flags); 164 } 165 166 static int 167 vn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags) 168 { 169 struct vnode *vp = (struct vnode *)uobj; 170 171 return VOP_PUTPAGES(vp, offlo, offhi, flags); 172 } 173 174 /* 175 * Anon object stuff 176 */ 177 178 static int 179 ao_get(struct uvm_object *uobj, voff_t off, struct vm_page **pgs, 180 int *npages, int centeridx, vm_prot_t access_type, 181 int advice, int flags) 182 { 183 struct vm_page *pg; 184 int i; 185 186 if (centeridx) 187 panic("%s: centeridx != 0 not supported", __func__); 188 189 /* loop over pages */ 190 off = trunc_page(off); 191 for (i = 0; i < *npages; i++) { 192 pg = uvm_pagelookup(uobj, off + (i << PAGE_SHIFT)); 193 if (pg) { 194 pgs[i] = pg; 195 } else { 196 pg = rumpvm_makepage(uobj, off + (i << PAGE_SHIFT)); 197 pgs[i] = pg; 198 } 199 } 200 201 return 0; 202 203 } 204 205 static int 206 ao_put(struct uvm_object *uobj, voff_t start, voff_t stop, int flags) 207 { 208 struct vm_page *pg; 209 210 /* we only free all pages for now */ 211 if ((flags & PGO_FREE) == 0 || (flags & PGO_ALLPAGES) == 0) 212 return 0; 213 214 simple_lock(&uobj->vmobjlock); 215 while ((pg = TAILQ_FIRST(&uobj->memq)) != NULL) 216 rumpvm_freepage(pg); 217 simple_unlock(&uobj->vmobjlock); 218 219 return 0; 220 } 221 222 struct uvm_object * 223 uao_create(vsize_t size, int flags) 224 { 225 struct uvm_object *uobj; 226 227 uobj = rumpuser_malloc(sizeof(struct uvm_object), 0); 228 memset(uobj, 0, sizeof(struct uvm_object)); 229 uobj->pgops = &aobj_pager; 230 TAILQ_INIT(&uobj->memq); 231 232 return uobj; 233 } 234 235 void 236 uao_detach(struct uvm_object *uobj) 237 { 238 239 ao_put(uobj, 0, 0, PGO_ALLPAGES | PGO_FREE); 240 rumpuser_free(uobj); 241 } 242 243 /* 244 * UBC 245 */ 246 247 struct ubc_window { 248 struct uvm_object *uwin_obj; 249 voff_t uwin_off; 250 uint8_t *uwin_mem; 251 size_t uwin_mapsize; 252 253 LIST_ENTRY(ubc_window) uwin_entries; 254 }; 255 256 static LIST_HEAD(, ubc_window) uwinlst = LIST_HEAD_INITIALIZER(uwinlst); 257 static kmutex_t uwinmtx; 258 259 int 260 rump_ubc_magic_uiomove(void *va, size_t n, struct uio *uio, int *rvp, 261 struct ubc_window *uwinp) 262 { 263 struct vm_page **pgs; 264 int npages = len2npages(uio->uio_offset, n); 265 int i, rv; 266 267 if (uwinp == NULL) { 268 mutex_enter(&uwinmtx); 269 LIST_FOREACH(uwinp, &uwinlst, uwin_entries) 270 if ((uint8_t *)va >= uwinp->uwin_mem 271 && (uint8_t *)va 272 < (uwinp->uwin_mem + uwinp->uwin_mapsize)) 273 break; 274 mutex_exit(&uwinmtx); 275 if (uwinp == NULL) { 276 KASSERT(rvp != NULL); 277 return 0; 278 } 279 } 280 281 pgs = rumpuser_malloc(npages * sizeof(pgs), 0); 282 memset(pgs, 0, sizeof(pgs)); 283 rv = uwinp->uwin_obj->pgops->pgo_get(uwinp->uwin_obj, 284 uwinp->uwin_off + ((uint8_t *)va - uwinp->uwin_mem), 285 pgs, &npages, 0, 0, 0, 0); 286 if (rv) 287 goto out; 288 289 for (i = 0; i < npages; i++) { 290 size_t xfersize; 291 off_t pageoff; 292 293 pageoff = uio->uio_offset & PAGE_MASK; 294 xfersize = MIN(MIN(n, PAGE_SIZE), PAGE_SIZE-pageoff); 295 uiomove((uint8_t *)pgs[i]->uanon + pageoff, xfersize, uio); 296 if (uio->uio_rw == UIO_WRITE) 297 pgs[i]->flags &= ~PG_CLEAN; 298 n -= xfersize; 299 } 300 301 out: 302 rumpuser_free(pgs); 303 if (rvp) 304 *rvp = rv; 305 return 1; 306 } 307 308 static struct ubc_window * 309 uwin_alloc(struct uvm_object *uobj, voff_t off, vsize_t len) 310 { 311 struct ubc_window *uwinp; /* pronounced: you wimp! */ 312 313 uwinp = kmem_alloc(sizeof(struct ubc_window), KM_SLEEP); 314 uwinp->uwin_obj = uobj; 315 uwinp->uwin_off = off; 316 uwinp->uwin_mapsize = len; 317 uwinp->uwin_mem = kmem_alloc(len, KM_SLEEP); 318 319 return uwinp; 320 } 321 322 static void 323 uwin_free(struct ubc_window *uwinp) 324 { 325 326 kmem_free(uwinp->uwin_mem, uwinp->uwin_mapsize); 327 kmem_free(uwinp, sizeof(struct ubc_window)); 328 } 329 330 void * 331 ubc_alloc(struct uvm_object *uobj, voff_t offset, vsize_t *lenp, int advice, 332 int flags) 333 { 334 struct ubc_window *uwinp; 335 336 uwinp = uwin_alloc(uobj, offset, *lenp); 337 mutex_enter(&uwinmtx); 338 LIST_INSERT_HEAD(&uwinlst, uwinp, uwin_entries); 339 mutex_exit(&uwinmtx); 340 341 DPRINTF(("UBC_ALLOC offset 0x%llx, uwin %p, mem %p\n", 342 (unsigned long long)offset, uwinp, uwinp->uwin_mem)); 343 344 return uwinp->uwin_mem; 345 } 346 347 void 348 ubc_release(void *va, int flags) 349 { 350 struct ubc_window *uwinp; 351 352 mutex_enter(&uwinmtx); 353 LIST_FOREACH(uwinp, &uwinlst, uwin_entries) 354 if ((uint8_t *)va >= uwinp->uwin_mem 355 && (uint8_t *)va < (uwinp->uwin_mem + uwinp->uwin_mapsize)) 356 break; 357 mutex_exit(&uwinmtx); 358 if (uwinp == NULL) 359 panic("%s: releasing invalid window at %p", __func__, va); 360 361 LIST_REMOVE(uwinp, uwin_entries); 362 uwin_free(uwinp); 363 } 364 365 int 366 ubc_uiomove(struct uvm_object *uobj, struct uio *uio, vsize_t todo, 367 int advice, int flags) 368 { 369 struct ubc_window *uwinp; 370 vsize_t len; 371 372 while (todo > 0) { 373 len = todo; 374 375 uwinp = uwin_alloc(uobj, uio->uio_offset, len); 376 rump_ubc_magic_uiomove(uwinp->uwin_mem, len, uio, NULL, uwinp); 377 uwin_free(uwinp); 378 379 todo -= len; 380 } 381 return 0; 382 } 383 384 385 /* 386 * Misc routines 387 */ 388 389 void 390 rumpvm_init() 391 { 392 393 uvm_vnodeops.pgo_get = vn_get; 394 uvm_vnodeops.pgo_put = vn_put; 395 aobj_pager.pgo_get = ao_get; 396 aobj_pager.pgo_put = ao_put; 397 398 uvmexp.free = 1024*1024; /* XXX */ 399 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */ 400 401 mutex_init(&rvamtx, MUTEX_DEFAULT, 0); 402 mutex_init(&uwinmtx, MUTEX_DEFAULT, 0); 403 } 404 405 void 406 uvm_pageactivate(struct vm_page *pg) 407 { 408 409 /* nada */ 410 } 411 412 void 413 uvm_page_unbusy(struct vm_page **pgs, int npgs) 414 { 415 416 /* nada */ 417 } 418 419 void 420 uvm_pagewire(struct vm_page *pg) 421 { 422 423 /* nada */ 424 } 425 426 void 427 uvm_pageunwire(struct vm_page *pg) 428 { 429 430 /* nada */ 431 } 432 433 vaddr_t 434 uvm_pagermapin(struct vm_page **pps, int npages, int flags) 435 { 436 437 panic("%s: unimplemented", __func__); 438 } 439 440 struct vm_page * 441 uvm_pagelookup(struct uvm_object *uobj, voff_t off) 442 { 443 struct vm_page *pg; 444 445 simple_lock(&uobj->vmobjlock); 446 TAILQ_FOREACH(pg, &uobj->memq, listq) { 447 if (pg->offset == off) { 448 simple_unlock(&uobj->vmobjlock); 449 return pg; 450 } 451 } 452 simple_unlock(&uobj->vmobjlock); 453 454 return NULL; 455 } 456 457 struct vm_page * 458 uvm_pageratop(vaddr_t va) 459 { 460 struct rumpva *rva; 461 462 mutex_enter(&rvamtx); 463 LIST_FOREACH(rva, &rvahead, entries) 464 if (rva->addr == va) 465 break; 466 mutex_exit(&rvamtx); 467 468 if (rva == NULL) 469 panic("%s: va %llu", __func__, (unsigned long long)va); 470 471 return rva->pg; 472 } 473 474 void 475 uvm_estimatepageable(int *active, int *inactive) 476 { 477 478 /* XXX: guessing game */ 479 *active = 1024; 480 *inactive = 1024; 481 } 482 483 void 484 uvm_aio_biodone1(struct buf *bp) 485 { 486 487 panic("%s: unimplemented", __func__); 488 } 489 490 void 491 uvm_aio_biodone(struct buf *bp) 492 { 493 494 uvm_aio_aiodone(bp); 495 } 496 497 void 498 uvm_aio_aiodone(struct buf *bp) 499 { 500 501 if ((bp->b_flags & (B_READ | B_NOCACHE)) == 0 && bioopsp) 502 bioopsp->io_pageiodone(bp); 503 } 504 505 void 506 uvm_vnp_setsize(struct vnode *vp, voff_t newsize) 507 { 508 509 vp->v_size = vp->v_writesize = newsize; 510 } 511 512 void 513 uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize) 514 { 515 516 vp->v_writesize = newsize; 517 } 518 519 void 520 uvm_vnp_zerorange(struct vnode *vp, off_t off, size_t len) 521 { 522 struct uvm_object *uobj = &vp->v_uobj; 523 struct vm_page **pgs; 524 int maxpages = MIN(32, round_page(len) >> PAGE_SHIFT); 525 int rv, npages, i; 526 527 pgs = rumpuser_malloc(maxpages * sizeof(pgs), 0); 528 while (len) { 529 npages = MIN(maxpages, round_page(len) >> PAGE_SHIFT); 530 memset(pgs, 0, npages * sizeof(struct vm_page *)); 531 rv = uobj->pgops->pgo_get(uobj, off, pgs, &npages, 0, 0, 0, 0); 532 assert(npages > 0); 533 534 for (i = 0; i < npages; i++) { 535 uint8_t *start; 536 size_t chunkoff, chunklen; 537 538 chunkoff = off & PAGE_MASK; 539 chunklen = MIN(PAGE_SIZE - chunkoff, len); 540 start = (uint8_t *)pgs[i]->uanon + chunkoff; 541 542 memset(start, 0, chunklen); 543 pgs[i]->flags &= PG_CLEAN; 544 545 off += chunklen; 546 len -= chunklen; 547 } 548 } 549 rumpuser_free(pgs); 550 551 return; 552 } 553 554 struct uvm_ractx * 555 uvm_ra_allocctx() 556 { 557 558 return NULL; 559 } 560 561 void 562 uvm_ra_freectx(struct uvm_ractx *ra) 563 { 564 565 return; 566 } 567 568 bool 569 uvn_clean_p(struct uvm_object *uobj) 570 { 571 struct vnode *vp = (void *)uobj; 572 573 return (vp->v_iflag & VI_ONWORKLST) == 0; 574 } 575 576 /* 577 * Kmem 578 */ 579 580 void * 581 kmem_alloc(size_t size, km_flag_t kmflag) 582 { 583 584 return rumpuser_malloc(size, kmflag == KM_NOSLEEP); 585 } 586 587 void * 588 kmem_zalloc(size_t size, km_flag_t kmflag) 589 { 590 void *rv; 591 592 rv = kmem_alloc(size, kmflag); 593 if (rv) 594 memset(rv, 0, size); 595 596 return rv; 597 } 598 599 void 600 kmem_free(void *p, size_t size) 601 { 602 603 rumpuser_free(p); 604 } 605 606 /* 607 * UVM km 608 */ 609 610 vaddr_t 611 uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags) 612 { 613 void *rv; 614 615 rv = rumpuser_malloc(size, flags & (UVM_KMF_CANFAIL | UVM_KMF_NOWAIT)); 616 if (rv && flags & UVM_KMF_ZERO) 617 memset(rv, 0, size); 618 619 return (vaddr_t)rv; 620 } 621 622 void 623 uvm_km_free(struct vm_map *map, vaddr_t vaddr, vsize_t size, uvm_flag_t flags) 624 { 625 626 rumpuser_free((void *)vaddr); 627 } 628 629 struct vm_map * 630 uvm_km_suballoc(struct vm_map *map, vaddr_t *minaddr, vaddr_t *maxaddr, 631 vsize_t size, int pageable, bool fixed, struct vm_map_kernel *submap) 632 { 633 634 return (struct vm_map *)417416; 635 } 636
Indexes created Sat Sep 27 22:09:54 GMT 2025