x86_xpmap.c revision 1.25
1/* $NetBSD: x86_xpmap.c,v 1.25 2011/03/29 23:51:32 jym Exp $ */ 2 3/* 4 * Copyright (c) 2006 Mathieu Ropert <mro@adviseo.fr> 5 * 6 * Permission to use, copy, modify, and distribute this software for any 7 * purpose with or without fee is hereby granted, provided that the above 8 * copyright notice and this permission notice appear in all copies. 9 * 10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 13 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 14 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 15 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 16 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 17 */ 18 19/* 20 * Copyright (c) 2006, 2007 Manuel Bouyer. 21 * 22 * Redistribution and use in source and binary forms, with or without 23 * modification, are permitted provided that the following conditions 24 * are met: 25 * 1. Redistributions of source code must retain the above copyright 26 * notice, this list of conditions and the following disclaimer. 27 * 2. Redistributions in binary form must reproduce the above copyright 28 * notice, this list of conditions and the following disclaimer in the 29 * documentation and/or other materials provided with the distribution. 30 * 31 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 32 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 33 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 34 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 35 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 36 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 37 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 38 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 39 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 40 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 41 * 42 */ 43 44/* 45 * 46 * Copyright (c) 2004 Christian Limpach. 47 * All rights reserved. 48 * 49 * Redistribution and use in source and binary forms, with or without 50 * modification, are permitted provided that the following conditions 51 * are met: 52 * 1. Redistributions of source code must retain the above copyright 53 * notice, this list of conditions and the following disclaimer. 54 * 2. Redistributions in binary form must reproduce the above copyright 55 * notice, this list of conditions and the following disclaimer in the 56 * documentation and/or other materials provided with the distribution. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 59 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 60 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 61 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 62 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 63 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 64 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 65 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 66 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 67 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 68 */ 69 70 71#include <sys/cdefs.h> 72__KERNEL_RCSID(0, "$NetBSD: x86_xpmap.c,v 1.25 2011/03/29 23:51:32 jym Exp $"); 73 74#include "opt_xen.h" 75#include "opt_ddb.h" 76#include "ksyms.h" 77 78#include <sys/param.h> 79#include <sys/systm.h> 80 81#include <uvm/uvm.h> 82 83#include <machine/pmap.h> 84#include <machine/gdt.h> 85#include <xen/xenfunc.h> 86 87#include <dev/isa/isareg.h> 88#include <machine/isa_machdep.h> 89 90#undef XENDEBUG 91/* #define XENDEBUG_SYNC */ 92/* #define XENDEBUG_LOW */ 93 94#ifdef XENDEBUG 95#define XENPRINTF(x) printf x 96#define XENPRINTK(x) printk x 97#define XENPRINTK2(x) /* printk x */ 98 99static char XBUF[256]; 100#else 101#define XENPRINTF(x) 102#define XENPRINTK(x) 103#define XENPRINTK2(x) 104#endif 105#define PRINTF(x) printf x 106#define PRINTK(x) printk x 107 108/* on x86_64 kernel runs in ring 3 */ 109#ifdef __x86_64__ 110#define PG_k PG_u 111#else 112#define PG_k 0 113#endif 114 115volatile shared_info_t *HYPERVISOR_shared_info; 116/* Xen requires the start_info struct to be page aligned */ 117union start_info_union start_info_union __aligned(PAGE_SIZE); 118unsigned long *xpmap_phys_to_machine_mapping; 119 120void xen_failsafe_handler(void); 121 122#define HYPERVISOR_mmu_update_self(req, count, success_count) \ 123 HYPERVISOR_mmu_update((req), (count), (success_count), DOMID_SELF) 124 125void 126xen_failsafe_handler(void) 127{ 128 129 panic("xen_failsafe_handler called!\n"); 130} 131 132 133void 134xen_set_ldt(vaddr_t base, uint32_t entries) 135{ 136 vaddr_t va; 137 vaddr_t end; 138 pt_entry_t *ptp; 139 int s; 140 141#ifdef __x86_64__ 142 end = base + (entries << 3); 143#else 144 end = base + entries * sizeof(union descriptor); 145#endif 146 147 for (va = base; va < end; va += PAGE_SIZE) { 148 KASSERT(va >= VM_MIN_KERNEL_ADDRESS); 149 ptp = kvtopte(va); 150 XENPRINTF(("xen_set_ldt %#" PRIxVADDR " %d %p\n", 151 base, entries, ptp)); 152 pmap_pte_clearbits(ptp, PG_RW); 153 } 154 s = splvm(); 155 xpq_queue_set_ldt(base, entries); 156 splx(s); 157} 158 159#ifdef XENDEBUG 160void xpq_debug_dump(void); 161#endif 162 163#define XPQUEUE_SIZE 2048 164static mmu_update_t xpq_queue[XPQUEUE_SIZE]; 165static int xpq_idx = 0; 166 167void 168xpq_flush_queue(void) 169{ 170 int i, ok, ret; 171 172 XENPRINTK2(("flush queue %p entries %d\n", xpq_queue, xpq_idx)); 173 for (i = 0; i < xpq_idx; i++) 174 XENPRINTK2(("%d: 0x%08" PRIx64 " 0x%08" PRIx64 "\n", i, 175 xpq_queue[i].ptr, xpq_queue[i].val)); 176 177 ret = HYPERVISOR_mmu_update_self(xpq_queue, xpq_idx, &ok); 178 179 if (xpq_idx != 0 && ret < 0) { 180 printf("xpq_flush_queue: %d entries (%d successful)\n", 181 xpq_idx, ok); 182 for (i = 0; i < xpq_idx; i++) 183 printf("0x%016" PRIx64 ": 0x%016" PRIx64 "\n", 184 xpq_queue[i].ptr, xpq_queue[i].val); 185 panic("HYPERVISOR_mmu_update failed, ret: %d\n", ret); 186 } 187 xpq_idx = 0; 188} 189 190static inline void 191xpq_increment_idx(void) 192{ 193 194 xpq_idx++; 195 if (__predict_false(xpq_idx == XPQUEUE_SIZE)) 196 xpq_flush_queue(); 197} 198 199void 200xpq_queue_machphys_update(paddr_t ma, paddr_t pa) 201{ 202 XENPRINTK2(("xpq_queue_machphys_update ma=0x%" PRIx64 " pa=0x%" PRIx64 203 "\n", (int64_t)ma, (int64_t)pa)); 204 xpq_queue[xpq_idx].ptr = ma | MMU_MACHPHYS_UPDATE; 205 xpq_queue[xpq_idx].val = (pa - XPMAP_OFFSET) >> PAGE_SHIFT; 206 xpq_increment_idx(); 207#ifdef XENDEBUG_SYNC 208 xpq_flush_queue(); 209#endif 210} 211 212void 213xpq_queue_pte_update(paddr_t ptr, pt_entry_t val) 214{ 215 216 KASSERT((ptr & 3) == 0); 217 xpq_queue[xpq_idx].ptr = (paddr_t)ptr | MMU_NORMAL_PT_UPDATE; 218 xpq_queue[xpq_idx].val = val; 219 xpq_increment_idx(); 220#ifdef XENDEBUG_SYNC 221 xpq_flush_queue(); 222#endif 223} 224 225void 226xpq_queue_pt_switch(paddr_t pa) 227{ 228 struct mmuext_op op; 229 xpq_flush_queue(); 230 231 XENPRINTK2(("xpq_queue_pt_switch: 0x%" PRIx64 " 0x%" PRIx64 "\n", 232 (int64_t)pa, (int64_t)pa)); 233 op.cmd = MMUEXT_NEW_BASEPTR; 234 op.arg1.mfn = pa >> PAGE_SHIFT; 235 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 236 panic("xpq_queue_pt_switch"); 237} 238 239void 240xpq_queue_pin_table(paddr_t pa, int lvl) 241{ 242 struct mmuext_op op; 243 xpq_flush_queue(); 244 245 XENPRINTK2(("xpq_queue_pin_l%d_table: %#" PRIxPADDR "\n", 246 lvl + 1, pa)); 247 248 op.arg1.mfn = pa >> PAGE_SHIFT; 249 op.cmd = lvl; 250 251 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 252 panic("xpq_queue_pin_table"); 253} 254 255void 256xpq_queue_unpin_table(paddr_t pa) 257{ 258 struct mmuext_op op; 259 xpq_flush_queue(); 260 261 XENPRINTK2(("xpq_queue_unpin_table: %#" PRIxPADDR "\n", pa)); 262 op.arg1.mfn = pa >> PAGE_SHIFT; 263 op.cmd = MMUEXT_UNPIN_TABLE; 264 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 265 panic("xpq_queue_unpin_table"); 266} 267 268void 269xpq_queue_set_ldt(vaddr_t va, uint32_t entries) 270{ 271 struct mmuext_op op; 272 xpq_flush_queue(); 273 274 XENPRINTK2(("xpq_queue_set_ldt\n")); 275 KASSERT(va == (va & ~PAGE_MASK)); 276 op.cmd = MMUEXT_SET_LDT; 277 op.arg1.linear_addr = va; 278 op.arg2.nr_ents = entries; 279 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 280 panic("xpq_queue_set_ldt"); 281} 282 283void 284xpq_queue_tlb_flush(void) 285{ 286 struct mmuext_op op; 287 xpq_flush_queue(); 288 289 XENPRINTK2(("xpq_queue_tlb_flush\n")); 290 op.cmd = MMUEXT_TLB_FLUSH_LOCAL; 291 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 292 panic("xpq_queue_tlb_flush"); 293} 294 295void 296xpq_flush_cache(void) 297{ 298 struct mmuext_op op; 299 int s = splvm(); 300 xpq_flush_queue(); 301 302 XENPRINTK2(("xpq_queue_flush_cache\n")); 303 op.cmd = MMUEXT_FLUSH_CACHE; 304 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 305 panic("xpq_flush_cache"); 306 splx(s); 307} 308 309void 310xpq_queue_invlpg(vaddr_t va) 311{ 312 struct mmuext_op op; 313 xpq_flush_queue(); 314 315 XENPRINTK2(("xpq_queue_invlpg %#" PRIxVADDR "\n", va)); 316 op.cmd = MMUEXT_INVLPG_LOCAL; 317 op.arg1.linear_addr = (va & ~PAGE_MASK); 318 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 319 panic("xpq_queue_invlpg"); 320} 321 322int 323xpq_update_foreign(paddr_t ptr, pt_entry_t val, int dom) 324{ 325 mmu_update_t op; 326 int ok; 327 xpq_flush_queue(); 328 329 op.ptr = ptr; 330 op.val = val; 331 if (HYPERVISOR_mmu_update(&op, 1, &ok, dom) < 0) 332 return EFAULT; 333 return (0); 334} 335 336#ifdef XENDEBUG 337void 338xpq_debug_dump(void) 339{ 340 int i; 341 342 XENPRINTK2(("idx: %d\n", xpq_idx)); 343 for (i = 0; i < xpq_idx; i++) { 344 snprintf(XBUF, sizeof(XBUF), "%" PRIx64 " %08" PRIx64, 345 xpq_queue[i].ptr, xpq_queue[i].val); 346 if (++i < xpq_idx) 347 snprintf(XBUF + strlen(XBUF), 348 sizeof(XBUF) - strlen(XBUF), 349 "%" PRIx64 " %08" PRIx64, 350 xpq_queue[i].ptr, xpq_queue[i].val); 351 if (++i < xpq_idx) 352 snprintf(XBUF + strlen(XBUF), 353 sizeof(XBUF) - strlen(XBUF), 354 "%" PRIx64 " %08" PRIx64, 355 xpq_queue[i].ptr, xpq_queue[i].val); 356 if (++i < xpq_idx) 357 snprintf(XBUF + strlen(XBUF), 358 sizeof(XBUF) - strlen(XBUF), 359 "%" PRIx64 " %08" PRIx64, 360 xpq_queue[i].ptr, xpq_queue[i].val); 361 XENPRINTK2(("%d: %s\n", xpq_idx, XBUF)); 362 } 363} 364#endif 365 366 367extern volatile struct xencons_interface *xencons_interface; /* XXX */ 368extern struct xenstore_domain_interface *xenstore_interface; /* XXX */ 369 370static void xen_bt_set_readonly (vaddr_t); 371static void xen_bootstrap_tables (vaddr_t, vaddr_t, int, int, int); 372 373/* How many PDEs ? */ 374#if L2_SLOT_KERNBASE > 0 375#define TABLE_L2_ENTRIES (2 * (NKL2_KIMG_ENTRIES + 1)) 376#else 377#define TABLE_L2_ENTRIES (NKL2_KIMG_ENTRIES + 1) 378#endif 379 380/* 381 * Construct and switch to new pagetables 382 * first_avail is the first vaddr we can use after 383 * we get rid of Xen pagetables 384 */ 385 386vaddr_t xen_pmap_bootstrap (void); 387 388/* 389 * Function to get rid of Xen bootstrap tables 390 */ 391 392/* How many PDP do we need: */ 393#ifdef PAE 394/* 395 * For PAE, we consider a single contigous L2 "superpage" of 4 pages, 396 * all of them mapped by the L3 page. We also need a shadow page 397 * for L3[3]. 398 */ 399static const int l2_4_count = 6; 400#else 401static const int l2_4_count = PTP_LEVELS - 1; 402#endif 403 404vaddr_t 405xen_pmap_bootstrap(void) 406{ 407 int count, oldcount; 408 long mapsize; 409 vaddr_t bootstrap_tables, init_tables; 410 411 xpmap_phys_to_machine_mapping = 412 (unsigned long *)xen_start_info.mfn_list; 413 init_tables = xen_start_info.pt_base; 414 __PRINTK(("xen_arch_pmap_bootstrap init_tables=0x%lx\n", init_tables)); 415 416 /* Space after Xen boostrap tables should be free */ 417 bootstrap_tables = xen_start_info.pt_base + 418 (xen_start_info.nr_pt_frames * PAGE_SIZE); 419 420 /* 421 * Calculate how many space we need 422 * first everything mapped before the Xen bootstrap tables 423 */ 424 mapsize = init_tables - KERNTEXTOFF; 425 /* after the tables we'll have: 426 * - UAREA 427 * - dummy user PGD (x86_64) 428 * - HYPERVISOR_shared_info 429 * - ISA I/O mem (if needed) 430 */ 431 mapsize += UPAGES * NBPG; 432#ifdef __x86_64__ 433 mapsize += NBPG; 434#endif 435 mapsize += NBPG; 436 437#ifdef DOM0OPS 438 if (xendomain_is_dom0()) { 439 /* space for ISA I/O mem */ 440 mapsize += IOM_SIZE; 441 } 442#endif 443 /* at this point mapsize doens't include the table size */ 444 445#ifdef __x86_64__ 446 count = TABLE_L2_ENTRIES; 447#else 448 count = (mapsize + (NBPD_L2 -1)) >> L2_SHIFT; 449#endif /* __x86_64__ */ 450 451 /* now compute how many L2 pages we need exactly */ 452 XENPRINTK(("bootstrap_final mapsize 0x%lx count %d\n", mapsize, count)); 453 while (mapsize + (count + l2_4_count) * PAGE_SIZE + KERNTEXTOFF > 454 ((long)count << L2_SHIFT) + KERNBASE) { 455 count++; 456 } 457#ifndef __x86_64__ 458 /* 459 * one more L2 page: we'll alocate several pages after kva_start 460 * in pmap_bootstrap() before pmap_growkernel(), which have not been 461 * counted here. It's not a big issue to allocate one more L2 as 462 * pmap_growkernel() will be called anyway. 463 */ 464 count++; 465 nkptp[1] = count; 466#endif 467 468 /* 469 * install bootstrap pages. We may need more L2 pages than will 470 * have the final table here, as it's installed after the final table 471 */ 472 oldcount = count; 473 474bootstrap_again: 475 XENPRINTK(("bootstrap_again oldcount %d\n", oldcount)); 476 /* 477 * Xen space we'll reclaim may not be enough for our new page tables, 478 * move bootstrap tables if necessary 479 */ 480 if (bootstrap_tables < init_tables + ((count + l2_4_count) * PAGE_SIZE)) 481 bootstrap_tables = init_tables + 482 ((count + l2_4_count) * PAGE_SIZE); 483 /* make sure we have enough to map the bootstrap_tables */ 484 if (bootstrap_tables + ((oldcount + l2_4_count) * PAGE_SIZE) > 485 ((long)oldcount << L2_SHIFT) + KERNBASE) { 486 oldcount++; 487 goto bootstrap_again; 488 } 489 490 /* Create temporary tables */ 491 xen_bootstrap_tables(xen_start_info.pt_base, bootstrap_tables, 492 xen_start_info.nr_pt_frames, oldcount, 0); 493 494 /* Create final tables */ 495 xen_bootstrap_tables(bootstrap_tables, init_tables, 496 oldcount + l2_4_count, count, 1); 497 498 /* zero out free space after tables */ 499 memset((void *)(init_tables + ((count + l2_4_count) * PAGE_SIZE)), 0, 500 (UPAGES + 1) * NBPG); 501 return (init_tables + ((count + l2_4_count) * PAGE_SIZE)); 502} 503 504 505/* 506 * Build a new table and switch to it 507 * old_count is # of old tables (including PGD, PDTPE and PDE) 508 * new_count is # of new tables (PTE only) 509 * we assume areas don't overlap 510 */ 511 512 513static void 514xen_bootstrap_tables (vaddr_t old_pgd, vaddr_t new_pgd, 515 int old_count, int new_count, int final) 516{ 517 pd_entry_t *pdtpe, *pde, *pte; 518 pd_entry_t *cur_pgd, *bt_pgd; 519 paddr_t addr; 520 vaddr_t page, avail, text_end, map_end; 521 int i; 522 extern char __data_start; 523 524 __PRINTK(("xen_bootstrap_tables(%#" PRIxVADDR ", %#" PRIxVADDR "," 525 " %d, %d)\n", 526 old_pgd, new_pgd, old_count, new_count)); 527 text_end = ((vaddr_t)&__data_start) & ~PAGE_MASK; 528 /* 529 * size of R/W area after kernel text: 530 * xencons_interface (if present) 531 * xenstore_interface (if present) 532 * table pages (new_count + l2_4_count entries) 533 * extra mappings (only when final is true): 534 * UAREA 535 * dummy user PGD (x86_64 only)/gdt page (i386 only) 536 * HYPERVISOR_shared_info 537 * ISA I/O mem (if needed) 538 */ 539 map_end = new_pgd + ((new_count + l2_4_count) * NBPG); 540 if (final) { 541 map_end += (UPAGES + 1) * NBPG; 542 HYPERVISOR_shared_info = (shared_info_t *)map_end; 543 map_end += NBPG; 544 } 545 /* 546 * we always set atdevbase, as it's used by init386 to find the first 547 * available VA. map_end is updated only if we are dom0, so 548 * atdevbase -> atdevbase + IOM_SIZE will be mapped only in 549 * this case. 550 */ 551 if (final) 552 atdevbase = map_end; 553#ifdef DOM0OPS 554 if (final && xendomain_is_dom0()) { 555 /* ISA I/O mem */ 556 map_end += IOM_SIZE; 557 } 558#endif /* DOM0OPS */ 559 560 __PRINTK(("xen_bootstrap_tables text_end 0x%lx map_end 0x%lx\n", 561 text_end, map_end)); 562 __PRINTK(("console %#lx ", xen_start_info.console_mfn)); 563 __PRINTK(("xenstore %#" PRIx32 "\n", xen_start_info.store_mfn)); 564 565 /* 566 * Create bootstrap page tables 567 * What we need: 568 * - a PGD (level 4) 569 * - a PDTPE (level 3) 570 * - a PDE (level2) 571 * - some PTEs (level 1) 572 */ 573 574 cur_pgd = (pd_entry_t *) old_pgd; 575 bt_pgd = (pd_entry_t *) new_pgd; 576 memset (bt_pgd, 0, PAGE_SIZE); 577 avail = new_pgd + PAGE_SIZE; 578#if PTP_LEVELS > 3 579 /* Install level 3 */ 580 pdtpe = (pd_entry_t *) avail; 581 memset (pdtpe, 0, PAGE_SIZE); 582 avail += PAGE_SIZE; 583 584 addr = ((u_long) pdtpe) - KERNBASE; 585 bt_pgd[pl4_pi(KERNTEXTOFF)] = 586 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 587 588 __PRINTK(("L3 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 589 " -> L4[%#x]\n", 590 pdtpe, addr, bt_pgd[pl4_pi(KERNTEXTOFF)], pl4_pi(KERNTEXTOFF))); 591#else 592 pdtpe = bt_pgd; 593#endif /* PTP_LEVELS > 3 */ 594 595#if PTP_LEVELS > 2 596 /* Level 2 */ 597 pde = (pd_entry_t *) avail; 598 memset(pde, 0, PAGE_SIZE); 599 avail += PAGE_SIZE; 600 601 addr = ((u_long) pde) - KERNBASE; 602 pdtpe[pl3_pi(KERNTEXTOFF)] = 603 xpmap_ptom_masked(addr) | PG_k | PG_V | PG_RW; 604 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 605 " -> L3[%#x]\n", 606 pde, addr, pdtpe[pl3_pi(KERNTEXTOFF)], pl3_pi(KERNTEXTOFF))); 607#elif defined(PAE) 608 /* our PAE-style level 2: 5 contigous pages (4 L2 + 1 shadow) */ 609 pde = (pd_entry_t *) avail; 610 memset(pde, 0, PAGE_SIZE * 5); 611 avail += PAGE_SIZE * 5; 612 addr = ((u_long) pde) - KERNBASE; 613 /* 614 * enter L2 pages in the L3. 615 * The real L2 kernel PD will be the last one (so that 616 * pde[L2_SLOT_KERN] always point to the shadow). 617 */ 618 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 619 /* 620 * Xen doesn't want R/W mappings in L3 entries, it'll add it 621 * itself. 622 */ 623 pdtpe[i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 624 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 625 " -> L3[%#x]\n", 626 (vaddr_t)pde + PAGE_SIZE * i, addr, pdtpe[i], i)); 627 } 628 addr += PAGE_SIZE; 629 pdtpe[3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 630 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 631 " -> L3[%#x]\n", 632 (vaddr_t)pde + PAGE_SIZE * 4, addr, pdtpe[3], 3)); 633 634#else /* PAE */ 635 pde = bt_pgd; 636#endif /* PTP_LEVELS > 2 */ 637 638 /* Level 1 */ 639 page = KERNTEXTOFF; 640 for (i = 0; i < new_count; i ++) { 641 vaddr_t cur_page = page; 642 643 pte = (pd_entry_t *) avail; 644 avail += PAGE_SIZE; 645 646 memset(pte, 0, PAGE_SIZE); 647 while (pl2_pi(page) == pl2_pi (cur_page)) { 648 if (page >= map_end) { 649 /* not mapped at all */ 650 pte[pl1_pi(page)] = 0; 651 page += PAGE_SIZE; 652 continue; 653 } 654 pte[pl1_pi(page)] = xpmap_ptom_masked(page - KERNBASE); 655 if (page == (vaddr_t)HYPERVISOR_shared_info) { 656 pte[pl1_pi(page)] = xen_start_info.shared_info; 657 __PRINTK(("HYPERVISOR_shared_info " 658 "va %#lx pte %#" PRIxPADDR "\n", 659 HYPERVISOR_shared_info, pte[pl1_pi(page)])); 660 } 661 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 662 == xen_start_info.console.domU.mfn) { 663 xencons_interface = (void *)page; 664 pte[pl1_pi(page)] = xen_start_info.console_mfn; 665 pte[pl1_pi(page)] <<= PAGE_SHIFT; 666 __PRINTK(("xencons_interface " 667 "va %#lx pte %#" PRIxPADDR "\n", 668 xencons_interface, pte[pl1_pi(page)])); 669 } 670 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 671 == xen_start_info.store_mfn) { 672 xenstore_interface = (void *)page; 673 pte[pl1_pi(page)] = xen_start_info.store_mfn; 674 pte[pl1_pi(page)] <<= PAGE_SHIFT; 675 __PRINTK(("xenstore_interface " 676 "va %#lx pte %#" PRIxPADDR "\n", 677 xenstore_interface, pte[pl1_pi(page)])); 678 } 679#ifdef DOM0OPS 680 if (page >= (vaddr_t)atdevbase && 681 page < (vaddr_t)atdevbase + IOM_SIZE) { 682 pte[pl1_pi(page)] = 683 IOM_BEGIN + (page - (vaddr_t)atdevbase); 684 } 685#endif 686 pte[pl1_pi(page)] |= PG_k | PG_V; 687 if (page < text_end) { 688 /* map kernel text RO */ 689 pte[pl1_pi(page)] |= 0; 690 } else if (page >= old_pgd 691 && page < old_pgd + (old_count * PAGE_SIZE)) { 692 /* map old page tables RO */ 693 pte[pl1_pi(page)] |= 0; 694 } else if (page >= new_pgd && 695 page < new_pgd + ((new_count + l2_4_count) * PAGE_SIZE)) { 696 /* map new page tables RO */ 697 pte[pl1_pi(page)] |= 0; 698 } else { 699 /* map page RW */ 700 pte[pl1_pi(page)] |= PG_RW; 701 } 702 703 if ((page >= old_pgd && page < old_pgd + (old_count * PAGE_SIZE)) 704 || page >= new_pgd) { 705 __PRINTK(("va %#lx pa %#lx " 706 "entry 0x%" PRIxPADDR " -> L1[%#x]\n", 707 page, page - KERNBASE, 708 pte[pl1_pi(page)], pl1_pi(page))); 709 } 710 page += PAGE_SIZE; 711 } 712 713 addr = ((u_long) pte) - KERNBASE; 714 pde[pl2_pi(cur_page)] = 715 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 716 __PRINTK(("L1 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 717 " -> L2[%#x]\n", 718 pte, addr, pde[pl2_pi(cur_page)], pl2_pi(cur_page))); 719 /* Mark readonly */ 720 xen_bt_set_readonly((vaddr_t) pte); 721 } 722 723 /* Install recursive page tables mapping */ 724#ifdef PAE 725 /* 726 * we need a shadow page for the kernel's L2 page 727 * The real L2 kernel PD will be the last one (so that 728 * pde[L2_SLOT_KERN] always point to the shadow. 729 */ 730 memcpy(&pde[L2_SLOT_KERN + NPDPG], &pde[L2_SLOT_KERN], PAGE_SIZE); 731 pmap_kl2pd = &pde[L2_SLOT_KERN + NPDPG]; 732 pmap_kl2paddr = (u_long)pmap_kl2pd - KERNBASE; 733 734 /* 735 * We don't enter a recursive entry from the L3 PD. Instead, 736 * we enter the first 4 L2 pages, which includes the kernel's L2 737 * shadow. But we have to entrer the shadow after switching 738 * %cr3, or Xen will refcount some PTE with the wrong type. 739 */ 740 addr = (u_long)pde - KERNBASE; 741 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 742 pde[PDIR_SLOT_PTE + i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 743 __PRINTK(("pde[%d] va %#" PRIxVADDR " pa %#" PRIxPADDR 744 " entry %#" PRIxPADDR "\n", 745 (int)(PDIR_SLOT_PTE + i), pde + PAGE_SIZE * i, 746 addr, pde[PDIR_SLOT_PTE + i])); 747 } 748#if 0 749 addr += PAGE_SIZE; /* point to shadow L2 */ 750 pde[PDIR_SLOT_PTE + 3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 751 __PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n", 752 (int)(PDIR_SLOT_PTE + 3), pde + PAGE_SIZE * 4, (long)addr, 753 (int64_t)pde[PDIR_SLOT_PTE + 3])); 754#endif 755 /* Mark tables RO, and pin the kernel's shadow as L2 */ 756 addr = (u_long)pde - KERNBASE; 757 for (i = 0; i < 5; i++, addr += PAGE_SIZE) { 758 xen_bt_set_readonly(((vaddr_t)pde) + PAGE_SIZE * i); 759 if (i == 2 || i == 3) 760 continue; 761#if 0 762 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", i, (int64_t)addr)); 763 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 764#endif 765 } 766 if (final) { 767 addr = (u_long)pde - KERNBASE + 3 * PAGE_SIZE; 768 __PRINTK(("pin L2 %d addr %#" PRIxPADDR "\n", 2, addr)); 769 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 770 } 771#if 0 772 addr = (u_long)pde - KERNBASE + 2 * PAGE_SIZE; 773 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr)); 774 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 775#endif 776#else /* PAE */ 777 /* recursive entry in higher-level PD */ 778 bt_pgd[PDIR_SLOT_PTE] = 779 xpmap_ptom_masked(new_pgd - KERNBASE) | PG_k | PG_V; 780 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] va %#" PRIxVADDR " pa %#" PRIxPADDR 781 " entry %#" PRIxPADDR "\n", new_pgd, (paddr_t)new_pgd - KERNBASE, 782 bt_pgd[PDIR_SLOT_PTE])); 783 /* Mark tables RO */ 784 xen_bt_set_readonly((vaddr_t) pde); 785#endif 786#if PTP_LEVELS > 2 || defined(PAE) 787 xen_bt_set_readonly((vaddr_t) pdtpe); 788#endif 789#if PTP_LEVELS > 3 790 xen_bt_set_readonly(new_pgd); 791#endif 792 /* Pin the PGD */ 793 __PRINTK(("pin PGD\n")); 794#ifdef __x86_64__ 795 xpq_queue_pin_l4_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 796#elif PAE 797 xpq_queue_pin_l3_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 798#else 799 xpq_queue_pin_l2_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 800#endif 801 802 /* Save phys. addr of PDP, for libkvm. */ 803#ifdef PAE 804 PDPpaddr = (u_long)pde - KERNBASE; /* PDP is the L2 with PAE */ 805#else 806 PDPpaddr = (u_long)new_pgd - KERNBASE; 807#endif 808 809 /* Switch to new tables */ 810 __PRINTK(("switch to PGD\n")); 811 xpq_queue_pt_switch(xpmap_ptom_masked(new_pgd - KERNBASE)); 812 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] now entry %#" PRIxPADDR "\n", 813 bt_pgd[PDIR_SLOT_PTE])); 814 815#ifdef PAE 816 if (final) { 817 /* save the address of the L3 page */ 818 cpu_info_primary.ci_pae_l3_pdir = pdtpe; 819 cpu_info_primary.ci_pae_l3_pdirpa = (new_pgd - KERNBASE); 820 821 /* now enter kernel's PTE mappings */ 822 addr = (u_long)pde - KERNBASE + PAGE_SIZE * 3; 823 xpq_queue_pte_update( 824 xpmap_ptom(((vaddr_t)&pde[PDIR_SLOT_PTE + 3]) - KERNBASE), 825 xpmap_ptom_masked(addr) | PG_k | PG_V); 826 xpq_flush_queue(); 827 } 828#endif 829 830 /* Now we can safely reclaim space taken by old tables */ 831 832 __PRINTK(("unpin old PGD\n")); 833 /* Unpin old PGD */ 834 xpq_queue_unpin_table(xpmap_ptom_masked(old_pgd - KERNBASE)); 835 /* Mark old tables RW */ 836 page = old_pgd; 837 addr = (paddr_t) pde[pl2_pi(page)] & PG_FRAME; 838 addr = xpmap_mtop(addr); 839 pte = (pd_entry_t *) ((u_long)addr + KERNBASE); 840 pte += pl1_pi(page); 841 __PRINTK(("*pde %#" PRIxPADDR " addr %#" PRIxPADDR " pte %#lx\n", 842 pde[pl2_pi(page)], addr, (long)pte)); 843 while (page < old_pgd + (old_count * PAGE_SIZE) && page < map_end) { 844 addr = xpmap_ptom(((u_long) pte) - KERNBASE); 845 XENPRINTK(("addr %#" PRIxPADDR " pte %#lx " 846 "*pte %#" PRIxPADDR "\n", 847 addr, (long)pte, *pte)); 848 xpq_queue_pte_update(addr, *pte | PG_RW); 849 page += PAGE_SIZE; 850 /* 851 * Our ptes are contiguous 852 * so it's safe to just "++" here 853 */ 854 pte++; 855 } 856 xpq_flush_queue(); 857} 858 859 860/* 861 * Bootstrap helper functions 862 */ 863 864/* 865 * Mark a page readonly 866 * XXX: assuming vaddr = paddr + KERNBASE 867 */ 868 869static void 870xen_bt_set_readonly (vaddr_t page) 871{ 872 pt_entry_t entry; 873 874 entry = xpmap_ptom_masked(page - KERNBASE); 875 entry |= PG_k | PG_V; 876 877 HYPERVISOR_update_va_mapping (page, entry, UVMF_INVLPG); 878} 879 880#ifdef __x86_64__ 881void 882xen_set_user_pgd(paddr_t page) 883{ 884 struct mmuext_op op; 885 int s = splvm(); 886 887 xpq_flush_queue(); 888 op.cmd = MMUEXT_NEW_USER_BASEPTR; 889 op.arg1.mfn = xpmap_phys_to_machine_mapping[page >> PAGE_SHIFT]; 890 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 891 panic("xen_set_user_pgd: failed to install new user page" 892 " directory %#" PRIxPADDR, page); 893 splx(s); 894} 895#endif /* __x86_64__ */ 896