1 /* $NetBSD: x86_xpmap.c,v 1.28 2011/06/15 20:50:02 rmind 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 1.28 2011/06/15 20:50:02 rmind 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 99 static 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 114 115 volatile shared_info_t *HYPERVISOR_shared_info; 116 /* Xen requires the start_info struct to be page aligned */ 117 union start_info_union start_info_union __aligned(PAGE_SIZE); 118 unsigned long *xpmap_phys_to_machine_mapping; 119 120 void 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) 124 125 void 126 xen_failsafe_handler(void) 127 { 128 129 panic("xen_failsafe_handler called!\n"); 130 } 131 132 133 void 134 xen_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 160 void xpq_debug_dump(void); 161 #endif 162 163 #define XPQUEUE_SIZE 2048 164 static mmu_update_t xpq_queue[XPQUEUE_SIZE]; 165 static int xpq_idx = 0; 166 167 void 168 xpq_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 } 189 190 static inline void 191 xpq_increment_idx(void) 192 { 193 194 xpq_idx++; 195 if (__predict_false(xpq_idx == XPQUEUE_SIZE)) 196 xpq_flush_queue(); 197 } 198 199 void 200 xpq_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 } 211 212 void 213 xpq_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 } 224 225 void 226 xpq_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 } 238 239 void 240 xpq_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 } 254 255 void 256 xpq_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 } 267 268 void 269 xpq_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 } 282 283 void 284 xpq_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 } 294 295 void 296 xpq_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 } 308 309 void 310 xpq_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 } 321 322 int 323 xpq_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 337 void 338 xpq_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 365 366 367 extern volatile struct xencons_interface *xencons_interface; /* XXX */ 368 extern struct xenstore_domain_interface *xenstore_interface; /* XXX */ 369 370 static void xen_bt_set_readonly (vaddr_t); 371 static 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 */ 385 386 vaddr_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 */ 399 static const int l2_4_count = 6; 400 #else 401 static const int l2_4_count = PTP_LEVELS - 1; 402 #endif 403 404 vaddr_t 405 xen_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; 473 474 bootstrap_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); 501 502 /* Finally, flush TLB. */ 503 xpq_queue_tlb_flush(); 504 505 return (init_tables + ((count + l2_4_count) * PAGE_SIZE)); 506 } 507 508 509 /* 510 * Build a new table and switch to it 511 * old_count is # of old tables (including PGD, PDTPE and PDE) 512 * new_count is # of new tables (PTE only) 513 * we assume areas don't overlap 514 */ 515 static void 516 xen_bootstrap_tables (vaddr_t old_pgd, vaddr_t new_pgd, 517 int old_count, int new_count, int final) 518 { 519 pd_entry_t *pdtpe, *pde, *pte; 520 pd_entry_t *cur_pgd, *bt_pgd; 521 paddr_t addr; 522 vaddr_t page, avail, text_end, map_end; 523 int i; 524 extern char __data_start; 525 526 __PRINTK(("xen_bootstrap_tables(%#" PRIxVADDR ", %#" PRIxVADDR "," 527 " %d, %d)\n", 528 old_pgd, new_pgd, old_count, new_count)); 529 text_end = ((vaddr_t)&__data_start) & ~PAGE_MASK; 530 /* 531 * size of R/W area after kernel text: 532 * xencons_interface (if present) 533 * xenstore_interface (if present) 534 * table pages (new_count + l2_4_count entries) 535 * extra mappings (only when final is true): 536 * UAREA 537 * dummy user PGD (x86_64 only)/gdt page (i386 only) 538 * HYPERVISOR_shared_info 539 * ISA I/O mem (if needed) 540 */ 541 map_end = new_pgd + ((new_count + l2_4_count) * NBPG); 542 if (final) { 543 map_end += (UPAGES + 1) * NBPG; 544 HYPERVISOR_shared_info = (shared_info_t *)map_end; 545 map_end += NBPG; 546 } 547 /* 548 * we always set atdevbase, as it's used by init386 to find the first 549 * available VA. map_end is updated only if we are dom0, so 550 * atdevbase -> atdevbase + IOM_SIZE will be mapped only in 551 * this case. 552 */ 553 if (final) 554 atdevbase = map_end; 555 #ifdef DOM0OPS 556 if (final && xendomain_is_dom0()) { 557 /* ISA I/O mem */ 558 map_end += IOM_SIZE; 559 } 560 #endif /* DOM0OPS */ 561 562 __PRINTK(("xen_bootstrap_tables text_end 0x%lx map_end 0x%lx\n", 563 text_end, map_end)); 564 __PRINTK(("console %#lx ", xen_start_info.console_mfn)); 565 __PRINTK(("xenstore %#" PRIx32 "\n", xen_start_info.store_mfn)); 566 567 /* 568 * Create bootstrap page tables 569 * What we need: 570 * - a PGD (level 4) 571 * - a PDTPE (level 3) 572 * - a PDE (level2) 573 * - some PTEs (level 1) 574 */ 575 576 cur_pgd = (pd_entry_t *) old_pgd; 577 bt_pgd = (pd_entry_t *) new_pgd; 578 memset (bt_pgd, 0, PAGE_SIZE); 579 avail = new_pgd + PAGE_SIZE; 580 #if PTP_LEVELS > 3 581 /* Install level 3 */ 582 pdtpe = (pd_entry_t *) avail; 583 memset (pdtpe, 0, PAGE_SIZE); 584 avail += PAGE_SIZE; 585 586 addr = ((u_long) pdtpe) - KERNBASE; 587 bt_pgd[pl4_pi(KERNTEXTOFF)] = 588 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 589 590 __PRINTK(("L3 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 591 " -> L4[%#x]\n", 592 pdtpe, addr, bt_pgd[pl4_pi(KERNTEXTOFF)], pl4_pi(KERNTEXTOFF))); 593 #else 594 pdtpe = bt_pgd; 595 #endif /* PTP_LEVELS > 3 */ 596 597 #if PTP_LEVELS > 2 598 /* Level 2 */ 599 pde = (pd_entry_t *) avail; 600 memset(pde, 0, PAGE_SIZE); 601 avail += PAGE_SIZE; 602 603 addr = ((u_long) pde) - KERNBASE; 604 pdtpe[pl3_pi(KERNTEXTOFF)] = 605 xpmap_ptom_masked(addr) | PG_k | PG_V | PG_RW; 606 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 607 " -> L3[%#x]\n", 608 pde, addr, pdtpe[pl3_pi(KERNTEXTOFF)], pl3_pi(KERNTEXTOFF))); 609 #elif defined(PAE) 610 /* our PAE-style level 2: 5 contigous pages (4 L2 + 1 shadow) */ 611 pde = (pd_entry_t *) avail; 612 memset(pde, 0, PAGE_SIZE * 5); 613 avail += PAGE_SIZE * 5; 614 addr = ((u_long) pde) - KERNBASE; 615 /* 616 * enter L2 pages in the L3. 617 * The real L2 kernel PD will be the last one (so that 618 * pde[L2_SLOT_KERN] always point to the shadow). 619 */ 620 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 621 /* 622 * Xen doesn't want R/W mappings in L3 entries, it'll add it 623 * itself. 624 */ 625 pdtpe[i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 626 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 627 " -> L3[%#x]\n", 628 (vaddr_t)pde + PAGE_SIZE * i, addr, pdtpe[i], i)); 629 } 630 addr += PAGE_SIZE; 631 pdtpe[3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 632 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 633 " -> L3[%#x]\n", 634 (vaddr_t)pde + PAGE_SIZE * 4, addr, pdtpe[3], 3)); 635 636 #else /* PAE */ 637 pde = bt_pgd; 638 #endif /* PTP_LEVELS > 2 */ 639 640 /* Level 1 */ 641 page = KERNTEXTOFF; 642 for (i = 0; i < new_count; i ++) { 643 vaddr_t cur_page = page; 644 645 pte = (pd_entry_t *) avail; 646 avail += PAGE_SIZE; 647 648 memset(pte, 0, PAGE_SIZE); 649 while (pl2_pi(page) == pl2_pi (cur_page)) { 650 if (page >= map_end) { 651 /* not mapped at all */ 652 pte[pl1_pi(page)] = 0; 653 page += PAGE_SIZE; 654 continue; 655 } 656 pte[pl1_pi(page)] = xpmap_ptom_masked(page - KERNBASE); 657 if (page == (vaddr_t)HYPERVISOR_shared_info) { 658 pte[pl1_pi(page)] = xen_start_info.shared_info; 659 __PRINTK(("HYPERVISOR_shared_info " 660 "va %#lx pte %#" PRIxPADDR "\n", 661 HYPERVISOR_shared_info, pte[pl1_pi(page)])); 662 } 663 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 664 == xen_start_info.console.domU.mfn) { 665 xencons_interface = (void *)page; 666 pte[pl1_pi(page)] = xen_start_info.console_mfn; 667 pte[pl1_pi(page)] <<= PAGE_SHIFT; 668 __PRINTK(("xencons_interface " 669 "va %#lx pte %#" PRIxPADDR "\n", 670 xencons_interface, pte[pl1_pi(page)])); 671 } 672 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 673 == xen_start_info.store_mfn) { 674 xenstore_interface = (void *)page; 675 pte[pl1_pi(page)] = xen_start_info.store_mfn; 676 pte[pl1_pi(page)] <<= PAGE_SHIFT; 677 __PRINTK(("xenstore_interface " 678 "va %#lx pte %#" PRIxPADDR "\n", 679 xenstore_interface, pte[pl1_pi(page)])); 680 } 681 #ifdef DOM0OPS 682 if (page >= (vaddr_t)atdevbase && 683 page < (vaddr_t)atdevbase + IOM_SIZE) { 684 pte[pl1_pi(page)] = 685 IOM_BEGIN + (page - (vaddr_t)atdevbase); 686 } 687 #endif 688 pte[pl1_pi(page)] |= PG_k | PG_V; 689 if (page < text_end) { 690 /* map kernel text RO */ 691 pte[pl1_pi(page)] |= 0; 692 } else if (page >= old_pgd 693 && page < old_pgd + (old_count * PAGE_SIZE)) { 694 /* map old page tables RO */ 695 pte[pl1_pi(page)] |= 0; 696 } else if (page >= new_pgd && 697 page < new_pgd + ((new_count + l2_4_count) * PAGE_SIZE)) { 698 /* map new page tables RO */ 699 pte[pl1_pi(page)] |= 0; 700 } else { 701 /* map page RW */ 702 pte[pl1_pi(page)] |= PG_RW; 703 } 704 705 if ((page >= old_pgd && page < old_pgd + (old_count * PAGE_SIZE)) 706 || page >= new_pgd) { 707 __PRINTK(("va %#lx pa %#lx " 708 "entry 0x%" PRIxPADDR " -> L1[%#x]\n", 709 page, page - KERNBASE, 710 pte[pl1_pi(page)], pl1_pi(page))); 711 } 712 page += PAGE_SIZE; 713 } 714 715 addr = ((u_long) pte) - KERNBASE; 716 pde[pl2_pi(cur_page)] = 717 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 718 __PRINTK(("L1 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 719 " -> L2[%#x]\n", 720 pte, addr, pde[pl2_pi(cur_page)], pl2_pi(cur_page))); 721 /* Mark readonly */ 722 xen_bt_set_readonly((vaddr_t) pte); 723 } 724 725 /* Install recursive page tables mapping */ 726 #ifdef PAE 727 /* 728 * we need a shadow page for the kernel's L2 page 729 * The real L2 kernel PD will be the last one (so that 730 * pde[L2_SLOT_KERN] always point to the shadow. 731 */ 732 memcpy(&pde[L2_SLOT_KERN + NPDPG], &pde[L2_SLOT_KERN], PAGE_SIZE); 733 pmap_kl2pd = &pde[L2_SLOT_KERN + NPDPG]; 734 pmap_kl2paddr = (u_long)pmap_kl2pd - KERNBASE; 735 736 /* 737 * We don't enter a recursive entry from the L3 PD. Instead, 738 * we enter the first 4 L2 pages, which includes the kernel's L2 739 * shadow. But we have to entrer the shadow after switching 740 * %cr3, or Xen will refcount some PTE with the wrong type. 741 */ 742 addr = (u_long)pde - KERNBASE; 743 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 744 pde[PDIR_SLOT_PTE + i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 745 __PRINTK(("pde[%d] va %#" PRIxVADDR " pa %#" PRIxPADDR 746 " entry %#" PRIxPADDR "\n", 747 (int)(PDIR_SLOT_PTE + i), pde + PAGE_SIZE * i, 748 addr, pde[PDIR_SLOT_PTE + i])); 749 } 750 #if 0 751 addr += PAGE_SIZE; /* point to shadow L2 */ 752 pde[PDIR_SLOT_PTE + 3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 753 __PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n", 754 (int)(PDIR_SLOT_PTE + 3), pde + PAGE_SIZE * 4, (long)addr, 755 (int64_t)pde[PDIR_SLOT_PTE + 3])); 756 #endif 757 /* Mark tables RO, and pin the kernel's shadow as L2 */ 758 addr = (u_long)pde - KERNBASE; 759 for (i = 0; i < 5; i++, addr += PAGE_SIZE) { 760 xen_bt_set_readonly(((vaddr_t)pde) + PAGE_SIZE * i); 761 if (i == 2 || i == 3) 762 continue; 763 #if 0 764 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", i, (int64_t)addr)); 765 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 766 #endif 767 } 768 if (final) { 769 addr = (u_long)pde - KERNBASE + 3 * PAGE_SIZE; 770 __PRINTK(("pin L2 %d addr %#" PRIxPADDR "\n", 2, addr)); 771 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 772 } 773 #if 0 774 addr = (u_long)pde - KERNBASE + 2 * PAGE_SIZE; 775 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr)); 776 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 777 #endif 778 #else /* PAE */ 779 /* recursive entry in higher-level PD */ 780 bt_pgd[PDIR_SLOT_PTE] = 781 xpmap_ptom_masked(new_pgd - KERNBASE) | PG_k | PG_V; 782 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] va %#" PRIxVADDR " pa %#" PRIxPADDR 783 " entry %#" PRIxPADDR "\n", new_pgd, (paddr_t)new_pgd - KERNBASE, 784 bt_pgd[PDIR_SLOT_PTE])); 785 /* Mark tables RO */ 786 xen_bt_set_readonly((vaddr_t) pde); 787 #endif 788 #if PTP_LEVELS > 2 || defined(PAE) 789 xen_bt_set_readonly((vaddr_t) pdtpe); 790 #endif 791 #if PTP_LEVELS > 3 792 xen_bt_set_readonly(new_pgd); 793 #endif 794 /* Pin the PGD */ 795 __PRINTK(("pin PGD: %"PRIxVADDR"\n", new_pgd - KERNBASE)); 796 #ifdef __x86_64__ 797 xpq_queue_pin_l4_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 798 #elif PAE 799 xpq_queue_pin_l3_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 800 #else 801 xpq_queue_pin_l2_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 802 #endif 803 804 /* Save phys. addr of PDP, for libkvm. */ 805 #ifdef PAE 806 PDPpaddr = (u_long)pde - KERNBASE; /* PDP is the L2 with PAE */ 807 #else 808 PDPpaddr = (u_long)new_pgd - KERNBASE; 809 #endif 810 811 /* Switch to new tables */ 812 __PRINTK(("switch to PGD\n")); 813 xpq_queue_pt_switch(xpmap_ptom_masked(new_pgd - KERNBASE)); 814 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] now entry %#" PRIxPADDR "\n", 815 bt_pgd[PDIR_SLOT_PTE])); 816 817 #ifdef PAE 818 if (final) { 819 /* save the address of the L3 page */ 820 cpu_info_primary.ci_pae_l3_pdir = pdtpe; 821 cpu_info_primary.ci_pae_l3_pdirpa = (new_pgd - KERNBASE); 822 823 /* now enter kernel's PTE mappings */ 824 addr = (u_long)pde - KERNBASE + PAGE_SIZE * 3; 825 xpq_queue_pte_update( 826 xpmap_ptom(((vaddr_t)&pde[PDIR_SLOT_PTE + 3]) - KERNBASE), 827 xpmap_ptom_masked(addr) | PG_k | PG_V); 828 xpq_flush_queue(); 829 } 830 #endif 831 832 /* Now we can safely reclaim space taken by old tables */ 833 834 __PRINTK(("unpin old PGD\n")); 835 /* Unpin old PGD */ 836 xpq_queue_unpin_table(xpmap_ptom_masked(old_pgd - KERNBASE)); 837 /* Mark old tables RW */ 838 page = old_pgd; 839 addr = (paddr_t) pde[pl2_pi(page)] & PG_FRAME; 840 addr = xpmap_mtop(addr); 841 pte = (pd_entry_t *) ((u_long)addr + KERNBASE); 842 pte += pl1_pi(page); 843 __PRINTK(("*pde %#" PRIxPADDR " addr %#" PRIxPADDR " pte %#lx\n", 844 pde[pl2_pi(page)], addr, (long)pte)); 845 while (page < old_pgd + (old_count * PAGE_SIZE) && page < map_end) { 846 addr = xpmap_ptom(((u_long) pte) - KERNBASE); 847 XENPRINTK(("addr %#" PRIxPADDR " pte %#lx " 848 "*pte %#" PRIxPADDR "\n", 849 addr, (long)pte, *pte)); 850 xpq_queue_pte_update(addr, *pte | PG_RW); 851 page += PAGE_SIZE; 852 /* 853 * Our ptes are contiguous 854 * so it's safe to just "++" here 855 */ 856 pte++; 857 } 858 xpq_flush_queue(); 859 } 860 861 862 /* 863 * Bootstrap helper functions 864 */ 865 866 /* 867 * Mark a page readonly 868 * XXX: assuming vaddr = paddr + KERNBASE 869 */ 870 871 static void 872 xen_bt_set_readonly (vaddr_t page) 873 { 874 pt_entry_t entry; 875 876 entry = xpmap_ptom_masked(page - KERNBASE); 877 entry |= PG_k | PG_V; 878 879 HYPERVISOR_update_va_mapping (page, entry, UVMF_INVLPG); 880 } 881 882 #ifdef __x86_64__ 883 void 884 xen_set_user_pgd(paddr_t page) 885 { 886 struct mmuext_op op; 887 int s = splvm(); 888 889 xpq_flush_queue(); 890 op.cmd = MMUEXT_NEW_USER_BASEPTR; 891 op.arg1.mfn = xpmap_phys_to_machine_mapping[page >> PAGE_SHIFT]; 892 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 893 panic("xen_set_user_pgd: failed to install new user page" 894 " directory %#" PRIxPADDR, page); 895 splx(s); 896 } 897 #endif /* __x86_64__ */ 898