1 /* $NetBSD: vm.c,v 1.24 2007/12/01 10:45:42 yamt 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 struct uvmexp uvmexp; 83 struct uvm uvm; 84 85 struct vmspace rump_vmspace; 86 struct vm_map rump_vmmap; 87 88 /* 89 * vm pages 90 */ 91 92 /* called with the object locked */ 93 struct vm_page * 94 rumpvm_makepage(struct uvm_object *uobj, voff_t off) 95 { 96 struct vm_page *pg; 97 98 pg = rumpuser_malloc(sizeof(struct vm_page), 0); 99 memset(pg, 0, sizeof(struct vm_page)); 100 pg->offset = off; 101 pg->uobject = uobj; 102 103 pg->uanon = (void *)rumpuser_malloc(PAGE_SIZE, 0); 104 memset((void *)pg->uanon, 0, PAGE_SIZE); 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 rumpuser_free((void *)pg->uanon); 127 rumpuser_free(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 = rumpuser_malloc(sizeof(struct rumpva), 0); 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 rumpuser_free(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 simple_unlock(&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 simple_unlock(&uobj->vmobjlock); 236 return 0; 237 } 238 239 while ((pg = TAILQ_FIRST(&uobj->memq)) != NULL) 240 uvm_pagefree(pg); 241 simple_unlock(&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 = rumpuser_malloc(sizeof(struct uvm_object), 0); 252 memset(uobj, 0, sizeof(struct uvm_object)); 253 uobj->pgops = &aobj_pager; 254 TAILQ_INIT(&uobj->memq); 255 simple_lock_init(&uobj->vmobjlock); 256 257 return uobj; 258 } 259 260 void 261 uao_detach(struct uvm_object *uobj) 262 { 263 264 ao_put(uobj, 0, 0, PGO_ALLPAGES | PGO_FREE); 265 rumpuser_free(uobj); 266 } 267 268 /* 269 * UBC 270 */ 271 272 struct ubc_window { 273 struct uvm_object *uwin_obj; 274 voff_t uwin_off; 275 uint8_t *uwin_mem; 276 size_t uwin_mapsize; 277 278 LIST_ENTRY(ubc_window) uwin_entries; 279 }; 280 281 static LIST_HEAD(, ubc_window) uwinlst = LIST_HEAD_INITIALIZER(uwinlst); 282 static kmutex_t uwinmtx; 283 284 int 285 rump_ubc_magic_uiomove(void *va, size_t n, struct uio *uio, int *rvp, 286 struct ubc_window *uwinp) 287 { 288 struct vm_page **pgs; 289 int npages = len2npages(uio->uio_offset, n); 290 size_t allocsize; 291 int i, rv; 292 293 if (uwinp == NULL) { 294 mutex_enter(&uwinmtx); 295 LIST_FOREACH(uwinp, &uwinlst, uwin_entries) 296 if ((uint8_t *)va >= uwinp->uwin_mem 297 && (uint8_t *)va 298 < (uwinp->uwin_mem + uwinp->uwin_mapsize)) 299 break; 300 mutex_exit(&uwinmtx); 301 if (uwinp == NULL) { 302 KASSERT(rvp != NULL); 303 return 0; 304 } 305 } 306 307 allocsize = npages * sizeof(pgs); 308 pgs = kmem_zalloc(allocsize, KM_SLEEP); 309 simple_lock(&uwinp->uwin_obj->vmobjlock); 310 rv = uwinp->uwin_obj->pgops->pgo_get(uwinp->uwin_obj, 311 uwinp->uwin_off + ((uint8_t *)va - uwinp->uwin_mem), 312 pgs, &npages, 0, 0, 0, 0); 313 if (rv) 314 goto out; 315 316 for (i = 0; i < npages; i++) { 317 size_t xfersize; 318 off_t pageoff; 319 320 pageoff = uio->uio_offset & PAGE_MASK; 321 xfersize = MIN(MIN(n, PAGE_SIZE), PAGE_SIZE-pageoff); 322 uiomove((uint8_t *)pgs[i]->uanon + pageoff, xfersize, uio); 323 if (uio->uio_rw == UIO_WRITE) 324 pgs[i]->flags &= ~PG_CLEAN; 325 n -= xfersize; 326 } 327 uvm_page_unbusy(pgs, npages); 328 329 out: 330 kmem_free(pgs, allocsize); 331 if (rvp) 332 *rvp = rv; 333 return 1; 334 } 335 336 static struct ubc_window * 337 uwin_alloc(struct uvm_object *uobj, voff_t off, vsize_t len) 338 { 339 struct ubc_window *uwinp; /* pronounced: you wimp! */ 340 341 uwinp = kmem_alloc(sizeof(struct ubc_window), KM_SLEEP); 342 uwinp->uwin_obj = uobj; 343 uwinp->uwin_off = off; 344 uwinp->uwin_mapsize = len; 345 uwinp->uwin_mem = kmem_alloc(len, KM_SLEEP); 346 347 return uwinp; 348 } 349 350 static void 351 uwin_free(struct ubc_window *uwinp) 352 { 353 354 kmem_free(uwinp->uwin_mem, uwinp->uwin_mapsize); 355 kmem_free(uwinp, sizeof(struct ubc_window)); 356 } 357 358 void * 359 ubc_alloc(struct uvm_object *uobj, voff_t offset, vsize_t *lenp, int advice, 360 int flags) 361 { 362 struct ubc_window *uwinp; 363 364 uwinp = uwin_alloc(uobj, offset, *lenp); 365 mutex_enter(&uwinmtx); 366 LIST_INSERT_HEAD(&uwinlst, uwinp, uwin_entries); 367 mutex_exit(&uwinmtx); 368 369 DPRINTF(("UBC_ALLOC offset 0x%llx, uwin %p, mem %p\n", 370 (unsigned long long)offset, uwinp, uwinp->uwin_mem)); 371 372 return uwinp->uwin_mem; 373 } 374 375 void 376 ubc_release(void *va, int flags) 377 { 378 struct ubc_window *uwinp; 379 380 mutex_enter(&uwinmtx); 381 LIST_FOREACH(uwinp, &uwinlst, uwin_entries) 382 if ((uint8_t *)va >= uwinp->uwin_mem 383 && (uint8_t *)va < (uwinp->uwin_mem + uwinp->uwin_mapsize)) 384 break; 385 mutex_exit(&uwinmtx); 386 if (uwinp == NULL) 387 panic("%s: releasing invalid window at %p", __func__, va); 388 389 LIST_REMOVE(uwinp, uwin_entries); 390 uwin_free(uwinp); 391 } 392 393 int 394 ubc_uiomove(struct uvm_object *uobj, struct uio *uio, vsize_t todo, 395 int advice, int flags) 396 { 397 struct ubc_window *uwinp; 398 vsize_t len; 399 400 while (todo > 0) { 401 len = todo; 402 403 uwinp = uwin_alloc(uobj, uio->uio_offset, len); 404 rump_ubc_magic_uiomove(uwinp->uwin_mem, len, uio, NULL, uwinp); 405 uwin_free(uwinp); 406 407 todo -= len; 408 } 409 return 0; 410 } 411 412 413 /* 414 * Misc routines 415 */ 416 417 void 418 rumpvm_init() 419 { 420 421 uvmexp.free = 1024*1024; /* XXX */ 422 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */ 423 424 mutex_init(&rvamtx, MUTEX_DEFAULT, 0); 425 mutex_init(&uwinmtx, 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 simple_unlock(&uobj->vmobjlock); 465 return pg; 466 } 467 } 468 469 return NULL; 470 } 471 472 struct vm_page * 473 uvm_pageratop(vaddr_t va) 474 { 475 struct rumpva *rva; 476 477 mutex_enter(&rvamtx); 478 LIST_FOREACH(rva, &rvahead, entries) 479 if (rva->addr == va) 480 break; 481 mutex_exit(&rvamtx); 482 483 if (rva == NULL) 484 panic("%s: va %llu", __func__, (unsigned long long)va); 485 486 return rva->pg; 487 } 488 489 void 490 uvm_page_unbusy(struct vm_page **pgs, int npgs) 491 { 492 struct vm_page *pg; 493 int i; 494 495 for (i = 0; i < npgs; i++) { 496 pg = pgs[i]; 497 if (pg == NULL) 498 continue; 499 500 KASSERT(pg->flags & PG_BUSY); 501 if (pg->flags & PG_WANTED) 502 wakeup(pg); 503 pg->flags &= ~(PG_WANTED|PG_BUSY); 504 } 505 } 506 507 void 508 uvm_estimatepageable(int *active, int *inactive) 509 { 510 511 /* XXX: guessing game */ 512 *active = 1024; 513 *inactive = 1024; 514 } 515 516 void 517 uvm_aio_biodone1(struct buf *bp) 518 { 519 520 panic("%s: unimplemented", __func__); 521 } 522 523 void 524 uvm_aio_biodone(struct buf *bp) 525 { 526 527 uvm_aio_aiodone(bp); 528 } 529 530 void 531 uvm_aio_aiodone(struct buf *bp) 532 { 533 534 if ((bp->b_flags & (B_READ | B_NOCACHE)) == 0 && bioopsp) 535 bioopsp->io_pageiodone(bp); 536 } 537 538 void 539 uvm_vnp_setsize(struct vnode *vp, voff_t newsize) 540 { 541 542 vp->v_size = vp->v_writesize = newsize; 543 } 544 545 void 546 uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize) 547 { 548 549 vp->v_writesize = newsize; 550 } 551 552 void 553 uvm_vnp_zerorange(struct vnode *vp, off_t off, size_t len) 554 { 555 struct uvm_object *uobj = &vp->v_uobj; 556 struct vm_page **pgs; 557 int maxpages = MIN(32, round_page(len) >> PAGE_SHIFT); 558 int rv, npages, i; 559 560 pgs = kmem_zalloc(maxpages * sizeof(pgs), KM_SLEEP); 561 while (len) { 562 npages = MIN(maxpages, round_page(len) >> PAGE_SHIFT); 563 memset(pgs, 0, npages * sizeof(struct vm_page *)); 564 simple_lock(&uobj->vmobjlock); 565 rv = uobj->pgops->pgo_get(uobj, off, pgs, &npages, 0, 0, 0, 0); 566 assert(npages > 0); 567 568 for (i = 0; i < npages; i++) { 569 uint8_t *start; 570 size_t chunkoff, chunklen; 571 572 chunkoff = off & PAGE_MASK; 573 chunklen = MIN(PAGE_SIZE - chunkoff, len); 574 start = (uint8_t *)pgs[i]->uanon + chunkoff; 575 576 memset(start, 0, chunklen); 577 pgs[i]->flags &= PG_CLEAN; 578 579 off += chunklen; 580 len -= chunklen; 581 } 582 uvm_page_unbusy(pgs, npages); 583 } 584 kmem_free(pgs, maxpages * sizeof(pgs)); 585 586 return; 587 } 588 589 struct uvm_ractx * 590 uvm_ra_allocctx() 591 { 592 593 return NULL; 594 } 595 596 void 597 uvm_ra_freectx(struct uvm_ractx *ra) 598 { 599 600 return; 601 } 602 603 bool 604 uvn_clean_p(struct uvm_object *uobj) 605 { 606 struct vnode *vp = (void *)uobj; 607 608 return (vp->v_iflag & VI_ONWORKLST) == 0; 609 } 610 611 /* 612 * Kmem 613 */ 614 615 void * 616 kmem_alloc(size_t size, km_flag_t kmflag) 617 { 618 619 return rumpuser_malloc(size, kmflag == KM_NOSLEEP); 620 } 621 622 void * 623 kmem_zalloc(size_t size, km_flag_t kmflag) 624 { 625 void *rv; 626 627 rv = kmem_alloc(size, kmflag); 628 if (rv) 629 memset(rv, 0, size); 630 631 return rv; 632 } 633 634 void 635 kmem_free(void *p, size_t size) 636 { 637 638 rumpuser_free(p); 639 } 640 641 /* 642 * UVM km 643 */ 644 645 vaddr_t 646 uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags) 647 { 648 void *rv; 649 650 rv = rumpuser_malloc(size, flags & (UVM_KMF_CANFAIL | UVM_KMF_NOWAIT)); 651 if (rv && flags & UVM_KMF_ZERO) 652 memset(rv, 0, size); 653 654 return (vaddr_t)rv; 655 } 656 657 void 658 uvm_km_free(struct vm_map *map, vaddr_t vaddr, vsize_t size, uvm_flag_t flags) 659 { 660 661 rumpuser_free((void *)vaddr); 662 } 663 664 struct vm_map * 665 uvm_km_suballoc(struct vm_map *map, vaddr_t *minaddr, vaddr_t *maxaddr, 666 vsize_t size, int pageable, bool fixed, struct vm_map_kernel *submap) 667 { 668 669 return (struct vm_map *)417416; 670 } 671