x86_xpmap.c revision 1.20
1/* $NetBSD: x86_xpmap.c,v 1.20 2010/07/15 23:20:34 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 1.20 2010/07/15 23:20:34 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; 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 if (xpq_idx != 0 && 177 HYPERVISOR_mmu_update_self(xpq_queue, xpq_idx, &ok) < 0) { 178 printf("xpq_flush_queue: %d entries \n", xpq_idx); 179 for (i = 0; i < xpq_idx; i++) 180 printf("0x%016" PRIx64 ": 0x%016" PRIx64 "\n", 181 xpq_queue[i].ptr, xpq_queue[i].val); 182 panic("HYPERVISOR_mmu_update failed\n"); 183 } 184 xpq_idx = 0; 185} 186 187static inline void 188xpq_increment_idx(void) 189{ 190 191 xpq_idx++; 192 if (__predict_false(xpq_idx == XPQUEUE_SIZE)) 193 xpq_flush_queue(); 194} 195 196void 197xpq_queue_machphys_update(paddr_t ma, paddr_t pa) 198{ 199 XENPRINTK2(("xpq_queue_machphys_update ma=0x%" PRIx64 " pa=0x%" PRIx64 200 "\n", (int64_t)ma, (int64_t)pa)); 201 xpq_queue[xpq_idx].ptr = ma | MMU_MACHPHYS_UPDATE; 202 xpq_queue[xpq_idx].val = (pa - XPMAP_OFFSET) >> PAGE_SHIFT; 203 xpq_increment_idx(); 204#ifdef XENDEBUG_SYNC 205 xpq_flush_queue(); 206#endif 207} 208 209void 210xpq_queue_pte_update(paddr_t ptr, pt_entry_t val) 211{ 212 213 KASSERT((ptr & 3) == 0); 214 xpq_queue[xpq_idx].ptr = (paddr_t)ptr | MMU_NORMAL_PT_UPDATE; 215 xpq_queue[xpq_idx].val = val; 216 xpq_increment_idx(); 217#ifdef XENDEBUG_SYNC 218 xpq_flush_queue(); 219#endif 220} 221 222void 223xpq_queue_pt_switch(paddr_t pa) 224{ 225 struct mmuext_op op; 226 xpq_flush_queue(); 227 228 XENPRINTK2(("xpq_queue_pt_switch: 0x%" PRIx64 " 0x%" PRIx64 "\n", 229 (int64_t)pa, (int64_t)pa)); 230 op.cmd = MMUEXT_NEW_BASEPTR; 231 op.arg1.mfn = pa >> PAGE_SHIFT; 232 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 233 panic("xpq_queue_pt_switch"); 234} 235 236void 237xpq_queue_pin_table(paddr_t pa) 238{ 239 struct mmuext_op op; 240 xpq_flush_queue(); 241 242 XENPRINTK2(("xpq_queue_pin_table: 0x%" PRIx64 " 0x%" PRIx64 "\n", 243 (int64_t)pa, (int64_t)pa)); 244 op.arg1.mfn = pa >> PAGE_SHIFT; 245 246#if defined(__x86_64__) 247 op.cmd = MMUEXT_PIN_L4_TABLE; 248#else 249 op.cmd = MMUEXT_PIN_L2_TABLE; 250#endif 251 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 252 panic("xpq_queue_pin_table"); 253} 254 255#ifdef PAE 256static void 257xpq_queue_pin_l3_table(paddr_t pa) 258{ 259 struct mmuext_op op; 260 xpq_flush_queue(); 261 262 XENPRINTK2(("xpq_queue_pin_l2_table: 0x%" PRIx64 " 0x%" PRIx64 "\n", 263 (int64_t)pa, (int64_t)pa)); 264 op.arg1.mfn = pa >> PAGE_SHIFT; 265 266 op.cmd = MMUEXT_PIN_L3_TABLE; 267 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 268 panic("xpq_queue_pin_table"); 269} 270#endif 271 272void 273xpq_queue_unpin_table(paddr_t pa) 274{ 275 struct mmuext_op op; 276 xpq_flush_queue(); 277 278 XENPRINTK2(("xpq_queue_unpin_table: 0x%" PRIx64 " 0x%" PRIx64 "\n", 279 (int64_t)pa, (int64_t)pa)); 280 op.arg1.mfn = pa >> PAGE_SHIFT; 281 op.cmd = MMUEXT_UNPIN_TABLE; 282 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 283 panic("xpq_queue_unpin_table"); 284} 285 286void 287xpq_queue_set_ldt(vaddr_t va, uint32_t entries) 288{ 289 struct mmuext_op op; 290 xpq_flush_queue(); 291 292 XENPRINTK2(("xpq_queue_set_ldt\n")); 293 KASSERT(va == (va & ~PAGE_MASK)); 294 op.cmd = MMUEXT_SET_LDT; 295 op.arg1.linear_addr = va; 296 op.arg2.nr_ents = entries; 297 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 298 panic("xpq_queue_set_ldt"); 299} 300 301void 302xpq_queue_tlb_flush(void) 303{ 304 struct mmuext_op op; 305 xpq_flush_queue(); 306 307 XENPRINTK2(("xpq_queue_tlb_flush\n")); 308 op.cmd = MMUEXT_TLB_FLUSH_LOCAL; 309 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 310 panic("xpq_queue_tlb_flush"); 311} 312 313void 314xpq_flush_cache(void) 315{ 316 struct mmuext_op op; 317 int s = splvm(); 318 xpq_flush_queue(); 319 320 XENPRINTK2(("xpq_queue_flush_cache\n")); 321 op.cmd = MMUEXT_FLUSH_CACHE; 322 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 323 panic("xpq_flush_cache"); 324 splx(s); 325} 326 327void 328xpq_queue_invlpg(vaddr_t va) 329{ 330 struct mmuext_op op; 331 xpq_flush_queue(); 332 333 XENPRINTK2(("xpq_queue_invlpg %#" PRIxVADDR "\n", va)); 334 op.cmd = MMUEXT_INVLPG_LOCAL; 335 op.arg1.linear_addr = (va & ~PAGE_MASK); 336 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 337 panic("xpq_queue_invlpg"); 338} 339 340int 341xpq_update_foreign(paddr_t ptr, pt_entry_t val, int dom) 342{ 343 mmu_update_t op; 344 int ok; 345 xpq_flush_queue(); 346 347 op.ptr = ptr; 348 op.val = val; 349 if (HYPERVISOR_mmu_update(&op, 1, &ok, dom) < 0) 350 return EFAULT; 351 return (0); 352} 353 354#ifdef XENDEBUG 355void 356xpq_debug_dump(void) 357{ 358 int i; 359 360 XENPRINTK2(("idx: %d\n", xpq_idx)); 361 for (i = 0; i < xpq_idx; i++) { 362 snprintf(XBUF, sizeof(XBUF), "%" PRIx64 " %08" PRIx64, 363 xpq_queue[i].ptr, xpq_queue[i].val); 364 if (++i < xpq_idx) 365 snprintf(XBUF + strlen(XBUF), 366 sizeof(XBUF) - strlen(XBUF), 367 "%" PRIx64 " %08" PRIx64, 368 xpq_queue[i].ptr, xpq_queue[i].val); 369 if (++i < xpq_idx) 370 snprintf(XBUF + strlen(XBUF), 371 sizeof(XBUF) - strlen(XBUF), 372 "%" PRIx64 " %08" PRIx64, 373 xpq_queue[i].ptr, xpq_queue[i].val); 374 if (++i < xpq_idx) 375 snprintf(XBUF + strlen(XBUF), 376 sizeof(XBUF) - strlen(XBUF), 377 "%" PRIx64 " %08" PRIx64, 378 xpq_queue[i].ptr, xpq_queue[i].val); 379 XENPRINTK2(("%d: %s\n", xpq_idx, XBUF)); 380 } 381} 382#endif 383 384 385extern volatile struct xencons_interface *xencons_interface; /* XXX */ 386extern struct xenstore_domain_interface *xenstore_interface; /* XXX */ 387 388static void xen_bt_set_readonly (vaddr_t); 389static void xen_bootstrap_tables (vaddr_t, vaddr_t, int, int, int); 390 391/* How many PDEs ? */ 392#if L2_SLOT_KERNBASE > 0 393#define TABLE_L2_ENTRIES (2 * (NKL2_KIMG_ENTRIES + 1)) 394#else 395#define TABLE_L2_ENTRIES (NKL2_KIMG_ENTRIES + 1) 396#endif 397 398/* 399 * Construct and switch to new pagetables 400 * first_avail is the first vaddr we can use after 401 * we get rid of Xen pagetables 402 */ 403 404vaddr_t xen_pmap_bootstrap (void); 405 406/* 407 * Function to get rid of Xen bootstrap tables 408 */ 409 410/* How many PDP do we need: */ 411#ifdef PAE 412/* 413 * For PAE, we consider a single contigous L2 "superpage" of 4 pages, 414 * all of them mapped by the L3 page. We also need a shadow page 415 * for L3[3]. 416 */ 417static const int l2_4_count = 6; 418#else 419static const int l2_4_count = PTP_LEVELS - 1; 420#endif 421 422vaddr_t 423xen_pmap_bootstrap(void) 424{ 425 int count, oldcount; 426 long mapsize; 427 vaddr_t bootstrap_tables, init_tables; 428 429 xpmap_phys_to_machine_mapping = 430 (unsigned long *)xen_start_info.mfn_list; 431 init_tables = xen_start_info.pt_base; 432 __PRINTK(("xen_arch_pmap_bootstrap init_tables=0x%lx\n", init_tables)); 433 434 /* Space after Xen boostrap tables should be free */ 435 bootstrap_tables = xen_start_info.pt_base + 436 (xen_start_info.nr_pt_frames * PAGE_SIZE); 437 438 /* 439 * Calculate how many space we need 440 * first everything mapped before the Xen bootstrap tables 441 */ 442 mapsize = init_tables - KERNTEXTOFF; 443 /* after the tables we'll have: 444 * - UAREA 445 * - dummy user PGD (x86_64) 446 * - HYPERVISOR_shared_info 447 * - ISA I/O mem (if needed) 448 */ 449 mapsize += UPAGES * NBPG; 450#ifdef __x86_64__ 451 mapsize += NBPG; 452#endif 453 mapsize += NBPG; 454 455#ifdef DOM0OPS 456 if (xendomain_is_dom0()) { 457 /* space for ISA I/O mem */ 458 mapsize += IOM_SIZE; 459 } 460#endif 461 /* at this point mapsize doens't include the table size */ 462 463#ifdef __x86_64__ 464 count = TABLE_L2_ENTRIES; 465#else 466 count = (mapsize + (NBPD_L2 -1)) >> L2_SHIFT; 467#endif /* __x86_64__ */ 468 469 /* now compute how many L2 pages we need exactly */ 470 XENPRINTK(("bootstrap_final mapsize 0x%lx count %d\n", mapsize, count)); 471 while (mapsize + (count + l2_4_count) * PAGE_SIZE + KERNTEXTOFF > 472 ((long)count << L2_SHIFT) + KERNBASE) { 473 count++; 474 } 475#ifndef __x86_64__ 476 /* 477 * one more L2 page: we'll alocate several pages after kva_start 478 * in pmap_bootstrap() before pmap_growkernel(), which have not been 479 * counted here. It's not a big issue to allocate one more L2 as 480 * pmap_growkernel() will be called anyway. 481 */ 482 count++; 483 nkptp[1] = count; 484#endif 485 486 /* 487 * install bootstrap pages. We may need more L2 pages than will 488 * have the final table here, as it's installed after the final table 489 */ 490 oldcount = count; 491 492bootstrap_again: 493 XENPRINTK(("bootstrap_again oldcount %d\n", oldcount)); 494 /* 495 * Xen space we'll reclaim may not be enough for our new page tables, 496 * move bootstrap tables if necessary 497 */ 498 if (bootstrap_tables < init_tables + ((count + l2_4_count) * PAGE_SIZE)) 499 bootstrap_tables = init_tables + 500 ((count + l2_4_count) * PAGE_SIZE); 501 /* make sure we have enough to map the bootstrap_tables */ 502 if (bootstrap_tables + ((oldcount + l2_4_count) * PAGE_SIZE) > 503 ((long)oldcount << L2_SHIFT) + KERNBASE) { 504 oldcount++; 505 goto bootstrap_again; 506 } 507 508 /* Create temporary tables */ 509 xen_bootstrap_tables(xen_start_info.pt_base, bootstrap_tables, 510 xen_start_info.nr_pt_frames, oldcount, 0); 511 512 /* Create final tables */ 513 xen_bootstrap_tables(bootstrap_tables, init_tables, 514 oldcount + l2_4_count, count, 1); 515 516 /* zero out free space after tables */ 517 memset((void *)(init_tables + ((count + l2_4_count) * PAGE_SIZE)), 0, 518 (UPAGES + 1) * NBPG); 519 return (init_tables + ((count + l2_4_count) * PAGE_SIZE)); 520} 521 522 523/* 524 * Build a new table and switch to it 525 * old_count is # of old tables (including PGD, PDTPE and PDE) 526 * new_count is # of new tables (PTE only) 527 * we assume areas don't overlap 528 */ 529 530 531static void 532xen_bootstrap_tables (vaddr_t old_pgd, vaddr_t new_pgd, 533 int old_count, int new_count, int final) 534{ 535 pd_entry_t *pdtpe, *pde, *pte; 536 pd_entry_t *cur_pgd, *bt_pgd; 537 paddr_t addr; 538 vaddr_t page, avail, text_end, map_end; 539 int i; 540 extern char __data_start; 541 542 __PRINTK(("xen_bootstrap_tables(%#" PRIxVADDR ", %#" PRIxVADDR "," 543 " %d, %d)\n", 544 old_pgd, new_pgd, old_count, new_count)); 545 text_end = ((vaddr_t)&__data_start) & ~PAGE_MASK; 546 /* 547 * size of R/W area after kernel text: 548 * xencons_interface (if present) 549 * xenstore_interface (if present) 550 * table pages (new_count + l2_4_count entries) 551 * extra mappings (only when final is true): 552 * UAREA 553 * dummy user PGD (x86_64 only)/gdt page (i386 only) 554 * HYPERVISOR_shared_info 555 * ISA I/O mem (if needed) 556 */ 557 map_end = new_pgd + ((new_count + l2_4_count) * NBPG); 558 if (final) { 559 map_end += (UPAGES + 1) * NBPG; 560 HYPERVISOR_shared_info = (shared_info_t *)map_end; 561 map_end += NBPG; 562 } 563 /* 564 * we always set atdevbase, as it's used by init386 to find the first 565 * available VA. map_end is updated only if we are dom0, so 566 * atdevbase -> atdevbase + IOM_SIZE will be mapped only in 567 * this case. 568 */ 569 if (final) 570 atdevbase = map_end; 571#ifdef DOM0OPS 572 if (final && xendomain_is_dom0()) { 573 /* ISA I/O mem */ 574 map_end += IOM_SIZE; 575 } 576#endif /* DOM0OPS */ 577 578 __PRINTK(("xen_bootstrap_tables text_end 0x%lx map_end 0x%lx\n", 579 text_end, map_end)); 580 __PRINTK(("console %#lx ", xen_start_info.console_mfn)); 581 __PRINTK(("xenstore %#" PRIx32 "\n", xen_start_info.store_mfn)); 582 583 /* 584 * Create bootstrap page tables 585 * What we need: 586 * - a PGD (level 4) 587 * - a PDTPE (level 3) 588 * - a PDE (level2) 589 * - some PTEs (level 1) 590 */ 591 592 cur_pgd = (pd_entry_t *) old_pgd; 593 bt_pgd = (pd_entry_t *) new_pgd; 594 memset (bt_pgd, 0, PAGE_SIZE); 595 avail = new_pgd + PAGE_SIZE; 596#if PTP_LEVELS > 3 597 /* Install level 3 */ 598 pdtpe = (pd_entry_t *) avail; 599 memset (pdtpe, 0, PAGE_SIZE); 600 avail += PAGE_SIZE; 601 602 addr = ((u_long) pdtpe) - KERNBASE; 603 bt_pgd[pl4_pi(KERNTEXTOFF)] = 604 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 605 606 __PRINTK(("L3 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 607 " -> L4[%#x]\n", 608 pdtpe, addr, bt_pgd[pl4_pi(KERNTEXTOFF)], pl4_pi(KERNTEXTOFF))); 609#else 610 pdtpe = bt_pgd; 611#endif /* PTP_LEVELS > 3 */ 612 613#if PTP_LEVELS > 2 614 /* Level 2 */ 615 pde = (pd_entry_t *) avail; 616 memset(pde, 0, PAGE_SIZE); 617 avail += PAGE_SIZE; 618 619 addr = ((u_long) pde) - KERNBASE; 620 pdtpe[pl3_pi(KERNTEXTOFF)] = 621 xpmap_ptom_masked(addr) | PG_k | PG_V | PG_RW; 622 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 623 " -> L3[%#x]\n", 624 pde, addr, pdtpe[pl3_pi(KERNTEXTOFF)], pl3_pi(KERNTEXTOFF))); 625#elif defined(PAE) 626 /* our PAE-style level 2: 5 contigous pages (4 L2 + 1 shadow) */ 627 pde = (pd_entry_t *) avail; 628 memset(pde, 0, PAGE_SIZE * 5); 629 avail += PAGE_SIZE * 5; 630 addr = ((u_long) pde) - KERNBASE; 631 /* 632 * enter L2 pages in the L3. 633 * The real L2 kernel PD will be the last one (so that 634 * pde[L2_SLOT_KERN] always point to the shadow). 635 */ 636 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 637 /* 638 * Xen doens't want R/W mappings in L3 entries, it'll add it 639 * itself. 640 */ 641 pdtpe[i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 642 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 643 " -> L3[%#x]\n", 644 (vaddr_t)pde + PAGE_SIZE * i, addr, pdtpe[i], i)); 645 } 646 addr += PAGE_SIZE; 647 pdtpe[3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 648 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 649 " -> L3[%#x]\n", 650 (vaddr_t)pde + PAGE_SIZE * 4, addr, pdtpe[3], 3)); 651 652#else /* PAE */ 653 pde = bt_pgd; 654#endif /* PTP_LEVELS > 2 */ 655 656 /* Level 1 */ 657 page = KERNTEXTOFF; 658 for (i = 0; i < new_count; i ++) { 659 vaddr_t cur_page = page; 660 661 pte = (pd_entry_t *) avail; 662 avail += PAGE_SIZE; 663 664 memset(pte, 0, PAGE_SIZE); 665 while (pl2_pi(page) == pl2_pi (cur_page)) { 666 if (page >= map_end) { 667 /* not mapped at all */ 668 pte[pl1_pi(page)] = 0; 669 page += PAGE_SIZE; 670 continue; 671 } 672 pte[pl1_pi(page)] = xpmap_ptom_masked(page - KERNBASE); 673 if (page == (vaddr_t)HYPERVISOR_shared_info) { 674 pte[pl1_pi(page)] = xen_start_info.shared_info; 675 __PRINTK(("HYPERVISOR_shared_info " 676 "va %#lx pte %#" PRIxPADDR "\n", 677 HYPERVISOR_shared_info, pte[pl1_pi(page)])); 678 } 679 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 680 == xen_start_info.console.domU.mfn) { 681 xencons_interface = (void *)page; 682 pte[pl1_pi(page)] = xen_start_info.console_mfn; 683 pte[pl1_pi(page)] <<= PAGE_SHIFT; 684 __PRINTK(("xencons_interface " 685 "va %#lx pte %#" PRIxPADDR "\n", 686 xencons_interface, pte[pl1_pi(page)])); 687 } 688 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 689 == xen_start_info.store_mfn) { 690 xenstore_interface = (void *)page; 691 pte[pl1_pi(page)] = xen_start_info.store_mfn; 692 pte[pl1_pi(page)] <<= PAGE_SHIFT; 693 __PRINTK(("xenstore_interface " 694 "va %#lx pte %#" PRIxPADDR "\n", 695 xenstore_interface, pte[pl1_pi(page)])); 696 } 697#ifdef DOM0OPS 698 if (page >= (vaddr_t)atdevbase && 699 page < (vaddr_t)atdevbase + IOM_SIZE) { 700 pte[pl1_pi(page)] = 701 IOM_BEGIN + (page - (vaddr_t)atdevbase); 702 } 703#endif 704 pte[pl1_pi(page)] |= PG_k | PG_V; 705 if (page < text_end) { 706 /* map kernel text RO */ 707 pte[pl1_pi(page)] |= 0; 708 } else if (page >= old_pgd 709 && page < old_pgd + (old_count * PAGE_SIZE)) { 710 /* map old page tables RO */ 711 pte[pl1_pi(page)] |= 0; 712 } else if (page >= new_pgd && 713 page < new_pgd + ((new_count + l2_4_count) * PAGE_SIZE)) { 714 /* map new page tables RO */ 715 pte[pl1_pi(page)] |= 0; 716 } else { 717 /* map page RW */ 718 pte[pl1_pi(page)] |= PG_RW; 719 } 720 721 if ((page >= old_pgd && page < old_pgd + (old_count * PAGE_SIZE)) 722 || page >= new_pgd) { 723 __PRINTK(("va %#lx pa %#lx " 724 "entry 0x%" PRIxPADDR " -> L1[%#x]\n", 725 page, page - KERNBASE, 726 pte[pl1_pi(page)], pl1_pi(page))); 727 } 728 page += PAGE_SIZE; 729 } 730 731 addr = ((u_long) pte) - KERNBASE; 732 pde[pl2_pi(cur_page)] = 733 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 734 __PRINTK(("L1 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 735 " -> L2[%#x]\n", 736 pte, addr, pde[pl2_pi(cur_page)], pl2_pi(cur_page))); 737 /* Mark readonly */ 738 xen_bt_set_readonly((vaddr_t) pte); 739 } 740 741 /* Install recursive page tables mapping */ 742#ifdef PAE 743 /* 744 * we need a shadow page for the kernel's L2 page 745 * The real L2 kernel PD will be the last one (so that 746 * pde[L2_SLOT_KERN] always point to the shadow. 747 */ 748 memcpy(&pde[L2_SLOT_KERN + NPDPG], &pde[L2_SLOT_KERN], PAGE_SIZE); 749 pmap_kl2pd = &pde[L2_SLOT_KERN + NPDPG]; 750 pmap_kl2paddr = (u_long)pmap_kl2pd - KERNBASE; 751 752 /* 753 * We don't enter a recursive entry from the L3 PD. Instead, 754 * we enter the first 4 L2 pages, which includes the kernel's L2 755 * shadow. But we have to entrer the shadow after switching 756 * %cr3, or Xen will refcount some PTE with the wrong type. 757 */ 758 addr = (u_long)pde - KERNBASE; 759 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 760 pde[PDIR_SLOT_PTE + i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 761 __PRINTK(("pde[%d] va %#" PRIxVADDR " pa %#" PRIxPADDR 762 " entry %#" PRIxPADDR "\n", 763 (int)(PDIR_SLOT_PTE + i), pde + PAGE_SIZE * i, 764 addr, pde[PDIR_SLOT_PTE + i])); 765 } 766#if 0 767 addr += PAGE_SIZE; /* point to shadow L2 */ 768 pde[PDIR_SLOT_PTE + 3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 769 __PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n", 770 (int)(PDIR_SLOT_PTE + 3), pde + PAGE_SIZE * 4, (long)addr, 771 (int64_t)pde[PDIR_SLOT_PTE + 3])); 772#endif 773 /* Mark tables RO, and pin the kernel's shadow as L2 */ 774 addr = (u_long)pde - KERNBASE; 775 for (i = 0; i < 5; i++, addr += PAGE_SIZE) { 776 xen_bt_set_readonly(((vaddr_t)pde) + PAGE_SIZE * i); 777 if (i == 2 || i == 3) 778 continue; 779#if 0 780 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", i, (int64_t)addr)); 781 xpq_queue_pin_table(xpmap_ptom_masked(addr)); 782#endif 783 } 784 if (final) { 785 addr = (u_long)pde - KERNBASE + 3 * PAGE_SIZE; 786 __PRINTK(("pin L2 %d addr %#" PRIxPADDR "\n", 2, addr)); 787 xpq_queue_pin_table(xpmap_ptom_masked(addr)); 788 } 789#if 0 790 addr = (u_long)pde - KERNBASE + 2 * PAGE_SIZE; 791 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr)); 792 xpq_queue_pin_table(xpmap_ptom_masked(addr)); 793#endif 794#else /* PAE */ 795 /* recursive entry in higher-level PD */ 796 bt_pgd[PDIR_SLOT_PTE] = 797 xpmap_ptom_masked(new_pgd - KERNBASE) | PG_k | PG_V; 798 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] va %#" PRIxVADDR " pa %#" PRIxPADDR 799 " entry %#" PRIxPADDR "\n", new_pgd, (paddr_t)new_pgd - KERNBASE, 800 bt_pgd[PDIR_SLOT_PTE])); 801 /* Mark tables RO */ 802 xen_bt_set_readonly((vaddr_t) pde); 803#endif 804#if PTP_LEVELS > 2 || defined(PAE) 805 xen_bt_set_readonly((vaddr_t) pdtpe); 806#endif 807#if PTP_LEVELS > 3 808 xen_bt_set_readonly(new_pgd); 809#endif 810 /* Pin the PGD */ 811 __PRINTK(("pin PGD\n")); 812#ifdef PAE 813 xpq_queue_pin_l3_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 814#else 815 xpq_queue_pin_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 816#endif 817#ifdef __i386__ 818 /* Save phys. addr of PDP, for libkvm. */ 819 PDPpaddr = (long)pde - KERNBASE; 820#ifdef PAE 821 /* also save the address of the L3 page */ 822 pmap_l3pd = pdtpe; 823 pmap_l3paddr = (new_pgd - KERNBASE); 824#endif /* PAE */ 825#endif /* i386 */ 826 /* Switch to new tables */ 827 __PRINTK(("switch to PGD\n")); 828 xpq_queue_pt_switch(xpmap_ptom_masked(new_pgd - KERNBASE)); 829 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] now entry %#" PRIxPADDR "\n", 830 bt_pgd[PDIR_SLOT_PTE])); 831#ifdef PAE 832 if (final) { 833 /* now enter kernel's PTE mappings */ 834 addr = (u_long)pde - KERNBASE + PAGE_SIZE * 3; 835 xpq_queue_pte_update( 836 xpmap_ptom(((vaddr_t)&pde[PDIR_SLOT_PTE + 3]) - KERNBASE), 837 xpmap_ptom_masked(addr) | PG_k | PG_V); 838 xpq_flush_queue(); 839 } 840#endif 841 842 843 844 /* Now we can safely reclaim space taken by old tables */ 845 846 __PRINTK(("unpin old PGD\n")); 847 /* Unpin old PGD */ 848 xpq_queue_unpin_table(xpmap_ptom_masked(old_pgd - KERNBASE)); 849 /* Mark old tables RW */ 850 page = old_pgd; 851 addr = (paddr_t) pde[pl2_pi(page)] & PG_FRAME; 852 addr = xpmap_mtop(addr); 853 pte = (pd_entry_t *) ((u_long)addr + KERNBASE); 854 pte += pl1_pi(page); 855 __PRINTK(("*pde %#" PRIxPADDR " addr %#" PRIxPADDR " pte %#lx\n", 856 pde[pl2_pi(page)], addr, (long)pte)); 857 while (page < old_pgd + (old_count * PAGE_SIZE) && page < map_end) { 858 addr = xpmap_ptom(((u_long) pte) - KERNBASE); 859 XENPRINTK(("addr %#" PRIxPADDR " pte %#lx " 860 "*pte %#" PRIxPADDR "\n", 861 addr, (long)pte, *pte)); 862 xpq_queue_pte_update(addr, *pte | PG_RW); 863 page += PAGE_SIZE; 864 /* 865 * Our ptes are contiguous 866 * so it's safe to just "++" here 867 */ 868 pte++; 869 } 870 xpq_flush_queue(); 871} 872 873 874/* 875 * Bootstrap helper functions 876 */ 877 878/* 879 * Mark a page readonly 880 * XXX: assuming vaddr = paddr + KERNBASE 881 */ 882 883static void 884xen_bt_set_readonly (vaddr_t page) 885{ 886 pt_entry_t entry; 887 888 entry = xpmap_ptom_masked(page - KERNBASE); 889 entry |= PG_k | PG_V; 890 891 HYPERVISOR_update_va_mapping (page, entry, UVMF_INVLPG); 892} 893 894#ifdef __x86_64__ 895void 896xen_set_user_pgd(paddr_t page) 897{ 898 struct mmuext_op op; 899 int s = splvm(); 900 901 xpq_flush_queue(); 902 op.cmd = MMUEXT_NEW_USER_BASEPTR; 903 op.arg1.mfn = xpmap_phys_to_machine_mapping[page >> PAGE_SHIFT]; 904 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 905 panic("xen_set_user_pgd: failed to install new user page" 906 " directory %#" PRIxPADDR, page); 907 splx(s); 908} 909#endif /* __x86_64__ */ 910