x86_xpmap.c revision 1.36
1/* $NetBSD: x86_xpmap.c,v 1.36 2011/11/06 15:18:19 cherry 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 1.36 2011/11/06 15:18:19 cherry 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#include <sys/simplelock.h> 81 82#include <uvm/uvm.h> 83 84#include <machine/pmap.h> 85#include <machine/gdt.h> 86#include <xen/xenfunc.h> 87 88#include <dev/isa/isareg.h> 89#include <machine/isa_machdep.h> 90 91#undef XENDEBUG 92/* #define XENDEBUG_SYNC */ 93/* #define XENDEBUG_LOW */ 94 95#ifdef XENDEBUG 96#define XENPRINTF(x) printf x 97#define XENPRINTK(x) printk x 98#define XENPRINTK2(x) /* printk x */ 99 100static char XBUF[256]; 101#else 102#define XENPRINTF(x) 103#define XENPRINTK(x) 104#define XENPRINTK2(x) 105#endif 106#define PRINTF(x) printf x 107#define PRINTK(x) printk x 108 109/* on x86_64 kernel runs in ring 3 */ 110#ifdef __x86_64__ 111#define PG_k PG_u 112#else 113#define PG_k 0 114#endif 115 116volatile shared_info_t *HYPERVISOR_shared_info; 117/* Xen requires the start_info struct to be page aligned */ 118union start_info_union start_info_union __aligned(PAGE_SIZE); 119unsigned long *xpmap_phys_to_machine_mapping; 120 121void xen_failsafe_handler(void); 122 123#define HYPERVISOR_mmu_update_self(req, count, success_count) \ 124 HYPERVISOR_mmu_update((req), (count), (success_count), DOMID_SELF) 125 126void 127xen_failsafe_handler(void) 128{ 129 130 panic("xen_failsafe_handler called!\n"); 131} 132 133 134void 135xen_set_ldt(vaddr_t base, uint32_t entries) 136{ 137 vaddr_t va; 138 vaddr_t end; 139 pt_entry_t *ptp; 140 int s; 141 142#ifdef __x86_64__ 143 end = base + (entries << 3); 144#else 145 end = base + entries * sizeof(union descriptor); 146#endif 147 148 for (va = base; va < end; va += PAGE_SIZE) { 149 KASSERT(va >= VM_MIN_KERNEL_ADDRESS); 150 ptp = kvtopte(va); 151 XENPRINTF(("xen_set_ldt %#" PRIxVADDR " %d %p\n", 152 base, entries, ptp)); 153 pmap_pte_clearbits(ptp, PG_RW); 154 } 155 s = splvm(); 156 xpq_queue_set_ldt(base, entries); 157 splx(s); 158} 159 160#ifdef XENDEBUG 161void xpq_debug_dump(void); 162#endif 163 164#define XPQUEUE_SIZE 2048 165static mmu_update_t xpq_queue_array[MAXCPUS][XPQUEUE_SIZE]; 166static int xpq_idx_array[MAXCPUS]; 167 168extern struct cpu_info * (*xpq_cpu)(void); 169 170void 171xpq_flush_queue(void) 172{ 173 int i, ok = 0, ret; 174 175 mmu_update_t *xpq_queue = xpq_queue_array[xpq_cpu()->ci_cpuid]; 176 int xpq_idx = xpq_idx_array[xpq_cpu()->ci_cpuid]; 177 178 XENPRINTK2(("flush queue %p entries %d\n", xpq_queue, xpq_idx)); 179 for (i = 0; i < xpq_idx; i++) 180 XENPRINTK2(("%d: 0x%08" PRIx64 " 0x%08" PRIx64 "\n", i, 181 xpq_queue[i].ptr, xpq_queue[i].val)); 182 183retry: 184 ret = HYPERVISOR_mmu_update_self(xpq_queue, xpq_idx, &ok); 185 186 if (xpq_idx != 0 && ret < 0) { 187 printf("xpq_flush_queue: %d entries (%d successful)\n", 188 xpq_idx, ok); 189 190 if (ok != 0) { 191 xpq_queue += ok; 192 xpq_idx -= ok; 193 ok = 0; 194 goto retry; 195 } 196 197 for (i = 0; i < xpq_idx; i++) 198 printf("0x%016" PRIx64 ": 0x%016" PRIx64 "\n", 199 xpq_queue[i].ptr, xpq_queue[i].val); 200 panic("HYPERVISOR_mmu_update failed, ret: %d\n", ret); 201 } 202 xpq_idx_array[xpq_cpu()->ci_cpuid] = 0; 203} 204 205static inline void 206xpq_increment_idx(void) 207{ 208 209 if (__predict_false(++xpq_idx_array[xpq_cpu()->ci_cpuid] == XPQUEUE_SIZE)) 210 xpq_flush_queue(); 211} 212 213void 214xpq_queue_machphys_update(paddr_t ma, paddr_t pa) 215{ 216 217 mmu_update_t *xpq_queue = xpq_queue_array[xpq_cpu()->ci_cpuid]; 218 int xpq_idx = xpq_idx_array[xpq_cpu()->ci_cpuid]; 219 220 XENPRINTK2(("xpq_queue_machphys_update ma=0x%" PRIx64 " pa=0x%" PRIx64 221 "\n", (int64_t)ma, (int64_t)pa)); 222 223 xpq_queue[xpq_idx].ptr = ma | MMU_MACHPHYS_UPDATE; 224 xpq_queue[xpq_idx].val = (pa - XPMAP_OFFSET) >> PAGE_SHIFT; 225 xpq_increment_idx(); 226#ifdef XENDEBUG_SYNC 227 xpq_flush_queue(); 228#endif 229} 230 231void 232xpq_queue_pte_update(paddr_t ptr, pt_entry_t val) 233{ 234 235 mmu_update_t *xpq_queue = xpq_queue_array[xpq_cpu()->ci_cpuid]; 236 int xpq_idx = xpq_idx_array[xpq_cpu()->ci_cpuid]; 237 238 KASSERT((ptr & 3) == 0); 239 xpq_queue[xpq_idx].ptr = (paddr_t)ptr | MMU_NORMAL_PT_UPDATE; 240 xpq_queue[xpq_idx].val = val; 241 xpq_increment_idx(); 242#ifdef XENDEBUG_SYNC 243 xpq_flush_queue(); 244#endif 245} 246 247void 248xpq_queue_pt_switch(paddr_t pa) 249{ 250 struct mmuext_op op; 251 xpq_flush_queue(); 252 253 XENPRINTK2(("xpq_queue_pt_switch: 0x%" PRIx64 " 0x%" PRIx64 "\n", 254 (int64_t)pa, (int64_t)pa)); 255 op.cmd = MMUEXT_NEW_BASEPTR; 256 op.arg1.mfn = pa >> PAGE_SHIFT; 257 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 258 panic("xpq_queue_pt_switch"); 259} 260 261void 262xpq_queue_pin_table(paddr_t pa, int lvl) 263{ 264 struct mmuext_op op; 265 266 xpq_flush_queue(); 267 268 XENPRINTK2(("xpq_queue_pin_l%d_table: %#" PRIxPADDR "\n", 269 lvl + 1, pa)); 270 271 op.arg1.mfn = pa >> PAGE_SHIFT; 272 op.cmd = lvl; 273 274 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 275 panic("xpq_queue_pin_table"); 276} 277 278void 279xpq_queue_unpin_table(paddr_t pa) 280{ 281 struct mmuext_op op; 282 283 xpq_flush_queue(); 284 285 XENPRINTK2(("xpq_queue_unpin_table: %#" PRIxPADDR "\n", pa)); 286 op.arg1.mfn = pa >> PAGE_SHIFT; 287 op.cmd = MMUEXT_UNPIN_TABLE; 288 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 289 panic("xpq_queue_unpin_table"); 290} 291 292void 293xpq_queue_set_ldt(vaddr_t va, uint32_t entries) 294{ 295 struct mmuext_op op; 296 297 xpq_flush_queue(); 298 299 XENPRINTK2(("xpq_queue_set_ldt\n")); 300 KASSERT(va == (va & ~PAGE_MASK)); 301 op.cmd = MMUEXT_SET_LDT; 302 op.arg1.linear_addr = va; 303 op.arg2.nr_ents = entries; 304 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 305 panic("xpq_queue_set_ldt"); 306} 307 308void 309xpq_queue_tlb_flush(void) 310{ 311 struct mmuext_op op; 312 313 xpq_flush_queue(); 314 315 XENPRINTK2(("xpq_queue_tlb_flush\n")); 316 op.cmd = MMUEXT_TLB_FLUSH_LOCAL; 317 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 318 panic("xpq_queue_tlb_flush"); 319} 320 321void 322xpq_flush_cache(void) 323{ 324 struct mmuext_op op; 325 int s = splvm(), err; 326 327 xpq_flush_queue(); 328 329 XENPRINTK2(("xpq_queue_flush_cache\n")); 330 op.cmd = MMUEXT_FLUSH_CACHE; 331 if ((err = HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) < 0) { 332 panic("xpq_flush_cache, err %d", err); 333 } 334 splx(s); /* XXX: removeme */ 335} 336 337void 338xpq_queue_invlpg(vaddr_t va) 339{ 340 struct mmuext_op op; 341 xpq_flush_queue(); 342 343 XENPRINTK2(("xpq_queue_invlpg %#" PRIxVADDR "\n", va)); 344 op.cmd = MMUEXT_INVLPG_LOCAL; 345 op.arg1.linear_addr = (va & ~PAGE_MASK); 346 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 347 panic("xpq_queue_invlpg"); 348} 349 350void 351xen_mcast_invlpg(vaddr_t va, uint32_t cpumask) 352{ 353 mmuext_op_t op; 354 355 /* Flush pending page updates */ 356 xpq_flush_queue(); 357 358 op.cmd = MMUEXT_INVLPG_MULTI; 359 op.arg1.linear_addr = va; 360 op.arg2.vcpumask = &cpumask; 361 362 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 363 panic("xpq_queue_invlpg_all"); 364 } 365 366 return; 367} 368 369void 370xen_bcast_invlpg(vaddr_t va) 371{ 372 mmuext_op_t op; 373 374 /* Flush pending page updates */ 375 xpq_flush_queue(); 376 377 op.cmd = MMUEXT_INVLPG_ALL; 378 op.arg1.linear_addr = va; 379 380 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 381 panic("xpq_queue_invlpg_all"); 382 } 383 384 return; 385} 386 387/* This is a synchronous call. */ 388void 389xen_mcast_tlbflush(uint32_t cpumask) 390{ 391 mmuext_op_t op; 392 393 /* Flush pending page updates */ 394 xpq_flush_queue(); 395 396 op.cmd = MMUEXT_TLB_FLUSH_MULTI; 397 op.arg2.vcpumask = &cpumask; 398 399 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 400 panic("xpq_queue_invlpg_all"); 401 } 402 403 return; 404} 405 406/* This is a synchronous call. */ 407void 408xen_bcast_tlbflush(void) 409{ 410 mmuext_op_t op; 411 412 /* Flush pending page updates */ 413 xpq_flush_queue(); 414 415 op.cmd = MMUEXT_TLB_FLUSH_ALL; 416 417 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 418 panic("xpq_queue_invlpg_all"); 419 } 420 421 return; 422} 423 424/* This is a synchronous call. */ 425void 426xen_vcpu_mcast_invlpg(vaddr_t sva, vaddr_t eva, uint32_t cpumask) 427{ 428 KASSERT(eva > sva); 429 430 /* Flush pending page updates */ 431 xpq_flush_queue(); 432 433 /* Align to nearest page boundary */ 434 sva &= ~PAGE_MASK; 435 eva &= ~PAGE_MASK; 436 437 for ( ; sva <= eva; sva += PAGE_SIZE) { 438 xen_mcast_invlpg(sva, cpumask); 439 } 440 441 return; 442} 443 444/* This is a synchronous call. */ 445void 446xen_vcpu_bcast_invlpg(vaddr_t sva, vaddr_t eva) 447{ 448 KASSERT(eva > sva); 449 450 /* Flush pending page updates */ 451 xpq_flush_queue(); 452 453 /* Align to nearest page boundary */ 454 sva &= ~PAGE_MASK; 455 eva &= ~PAGE_MASK; 456 457 for ( ; sva <= eva; sva += PAGE_SIZE) { 458 xen_bcast_invlpg(sva); 459 } 460 461 return; 462} 463 464int 465xpq_update_foreign(paddr_t ptr, pt_entry_t val, int dom) 466{ 467 mmu_update_t op; 468 int ok; 469 470 xpq_flush_queue(); 471 472 op.ptr = ptr; 473 op.val = val; 474 if (HYPERVISOR_mmu_update(&op, 1, &ok, dom) < 0) 475 return EFAULT; 476 return (0); 477} 478 479#ifdef XENDEBUG 480void 481xpq_debug_dump(void) 482{ 483 int i; 484 485 mmu_update_t *xpq_queue = xpq_queue_array[xpq_cpu()->ci_cpuid]; 486 int xpq_idx = xpq_idx_array[xpq_cpu()->ci_cpuid]; 487 488 XENPRINTK2(("idx: %d\n", xpq_idx)); 489 for (i = 0; i < xpq_idx; i++) { 490 snprintf(XBUF, sizeof(XBUF), "%" PRIx64 " %08" PRIx64, 491 xpq_queue[i].ptr, xpq_queue[i].val); 492 if (++i < xpq_idx) 493 snprintf(XBUF + strlen(XBUF), 494 sizeof(XBUF) - strlen(XBUF), 495 "%" PRIx64 " %08" PRIx64, 496 xpq_queue[i].ptr, xpq_queue[i].val); 497 if (++i < xpq_idx) 498 snprintf(XBUF + strlen(XBUF), 499 sizeof(XBUF) - strlen(XBUF), 500 "%" PRIx64 " %08" PRIx64, 501 xpq_queue[i].ptr, xpq_queue[i].val); 502 if (++i < xpq_idx) 503 snprintf(XBUF + strlen(XBUF), 504 sizeof(XBUF) - strlen(XBUF), 505 "%" PRIx64 " %08" PRIx64, 506 xpq_queue[i].ptr, xpq_queue[i].val); 507 XENPRINTK2(("%d: %s\n", xpq_idx, XBUF)); 508 } 509} 510#endif 511 512 513extern volatile struct xencons_interface *xencons_interface; /* XXX */ 514extern struct xenstore_domain_interface *xenstore_interface; /* XXX */ 515 516static void xen_bt_set_readonly (vaddr_t); 517static void xen_bootstrap_tables (vaddr_t, vaddr_t, int, int, int); 518 519/* How many PDEs ? */ 520#if L2_SLOT_KERNBASE > 0 521#define TABLE_L2_ENTRIES (2 * (NKL2_KIMG_ENTRIES + 1)) 522#else 523#define TABLE_L2_ENTRIES (NKL2_KIMG_ENTRIES + 1) 524#endif 525 526/* 527 * Construct and switch to new pagetables 528 * first_avail is the first vaddr we can use after 529 * we get rid of Xen pagetables 530 */ 531 532vaddr_t xen_pmap_bootstrap (void); 533 534/* 535 * Function to get rid of Xen bootstrap tables 536 */ 537 538/* How many PDP do we need: */ 539#ifdef PAE 540/* 541 * For PAE, we consider a single contigous L2 "superpage" of 4 pages, 542 * all of them mapped by the L3 page. We also need a shadow page 543 * for L3[3]. 544 */ 545static const int l2_4_count = 6; 546#elif defined(__x86_64__) 547static const int l2_4_count = PTP_LEVELS; 548#else 549static const int l2_4_count = PTP_LEVELS - 1; 550#endif 551 552vaddr_t 553xen_pmap_bootstrap(void) 554{ 555 int count, oldcount; 556 long mapsize; 557 vaddr_t bootstrap_tables, init_tables; 558 559 memset(xpq_idx_array, 0, sizeof xpq_idx_array); 560 561 xpmap_phys_to_machine_mapping = 562 (unsigned long *)xen_start_info.mfn_list; 563 init_tables = xen_start_info.pt_base; 564 __PRINTK(("xen_arch_pmap_bootstrap init_tables=0x%lx\n", init_tables)); 565 566 /* Space after Xen boostrap tables should be free */ 567 bootstrap_tables = xen_start_info.pt_base + 568 (xen_start_info.nr_pt_frames * PAGE_SIZE); 569 570 /* 571 * Calculate how many space we need 572 * first everything mapped before the Xen bootstrap tables 573 */ 574 mapsize = init_tables - KERNTEXTOFF; 575 /* after the tables we'll have: 576 * - UAREA 577 * - dummy user PGD (x86_64) 578 * - HYPERVISOR_shared_info 579 * - ISA I/O mem (if needed) 580 */ 581 mapsize += UPAGES * NBPG; 582#ifdef __x86_64__ 583 mapsize += NBPG; 584#endif 585 mapsize += NBPG; 586 587#ifdef DOM0OPS 588 if (xendomain_is_dom0()) { 589 /* space for ISA I/O mem */ 590 mapsize += IOM_SIZE; 591 } 592#endif 593 /* at this point mapsize doens't include the table size */ 594 595#ifdef __x86_64__ 596 count = TABLE_L2_ENTRIES; 597#else 598 count = (mapsize + (NBPD_L2 -1)) >> L2_SHIFT; 599#endif /* __x86_64__ */ 600 601 /* now compute how many L2 pages we need exactly */ 602 XENPRINTK(("bootstrap_final mapsize 0x%lx count %d\n", mapsize, count)); 603 while (mapsize + (count + l2_4_count) * PAGE_SIZE + KERNTEXTOFF > 604 ((long)count << L2_SHIFT) + KERNBASE) { 605 count++; 606 } 607#ifndef __x86_64__ 608 /* 609 * one more L2 page: we'll alocate several pages after kva_start 610 * in pmap_bootstrap() before pmap_growkernel(), which have not been 611 * counted here. It's not a big issue to allocate one more L2 as 612 * pmap_growkernel() will be called anyway. 613 */ 614 count++; 615 nkptp[1] = count; 616#endif 617 618 /* 619 * install bootstrap pages. We may need more L2 pages than will 620 * have the final table here, as it's installed after the final table 621 */ 622 oldcount = count; 623 624bootstrap_again: 625 XENPRINTK(("bootstrap_again oldcount %d\n", oldcount)); 626 /* 627 * Xen space we'll reclaim may not be enough for our new page tables, 628 * move bootstrap tables if necessary 629 */ 630 if (bootstrap_tables < init_tables + ((count + l2_4_count) * PAGE_SIZE)) 631 bootstrap_tables = init_tables + 632 ((count + l2_4_count) * PAGE_SIZE); 633 /* make sure we have enough to map the bootstrap_tables */ 634 if (bootstrap_tables + ((oldcount + l2_4_count) * PAGE_SIZE) > 635 ((long)oldcount << L2_SHIFT) + KERNBASE) { 636 oldcount++; 637 goto bootstrap_again; 638 } 639 640 /* Create temporary tables */ 641 xen_bootstrap_tables(xen_start_info.pt_base, bootstrap_tables, 642 xen_start_info.nr_pt_frames, oldcount, 0); 643 644 /* Create final tables */ 645 xen_bootstrap_tables(bootstrap_tables, init_tables, 646 oldcount + l2_4_count, count, 1); 647 648 /* zero out free space after tables */ 649 memset((void *)(init_tables + ((count + l2_4_count) * PAGE_SIZE)), 0, 650 (UPAGES + 1) * NBPG); 651 652 /* Finally, flush TLB. */ 653 xpq_queue_tlb_flush(); 654 655 return (init_tables + ((count + l2_4_count) * PAGE_SIZE)); 656} 657 658/* 659 * Build a new table and switch to it 660 * old_count is # of old tables (including PGD, PDTPE and PDE) 661 * new_count is # of new tables (PTE only) 662 * we assume areas don't overlap 663 */ 664static void 665xen_bootstrap_tables (vaddr_t old_pgd, vaddr_t new_pgd, 666 int old_count, int new_count, int final) 667{ 668 pd_entry_t *pdtpe, *pde, *pte; 669 pd_entry_t *cur_pgd, *bt_pgd; 670 paddr_t addr; 671 vaddr_t page, avail, text_end, map_end; 672 int i; 673 extern char __data_start; 674 675 __PRINTK(("xen_bootstrap_tables(%#" PRIxVADDR ", %#" PRIxVADDR "," 676 " %d, %d)\n", 677 old_pgd, new_pgd, old_count, new_count)); 678 text_end = ((vaddr_t)&__data_start) & ~PAGE_MASK; 679 /* 680 * size of R/W area after kernel text: 681 * xencons_interface (if present) 682 * xenstore_interface (if present) 683 * table pages (new_count + l2_4_count entries) 684 * extra mappings (only when final is true): 685 * UAREA 686 * dummy user PGD (x86_64 only)/gdt page (i386 only) 687 * HYPERVISOR_shared_info 688 * ISA I/O mem (if needed) 689 */ 690 map_end = new_pgd + ((new_count + l2_4_count) * NBPG); 691 if (final) { 692 map_end += (UPAGES + 1) * NBPG; 693 HYPERVISOR_shared_info = (shared_info_t *)map_end; 694 map_end += NBPG; 695 } 696 /* 697 * we always set atdevbase, as it's used by init386 to find the first 698 * available VA. map_end is updated only if we are dom0, so 699 * atdevbase -> atdevbase + IOM_SIZE will be mapped only in 700 * this case. 701 */ 702 if (final) 703 atdevbase = map_end; 704#ifdef DOM0OPS 705 if (final && xendomain_is_dom0()) { 706 /* ISA I/O mem */ 707 map_end += IOM_SIZE; 708 } 709#endif /* DOM0OPS */ 710 711 __PRINTK(("xen_bootstrap_tables text_end 0x%lx map_end 0x%lx\n", 712 text_end, map_end)); 713 __PRINTK(("console %#lx ", xen_start_info.console_mfn)); 714 __PRINTK(("xenstore %#" PRIx32 "\n", xen_start_info.store_mfn)); 715 716 /* 717 * Create bootstrap page tables 718 * What we need: 719 * - a PGD (level 4) 720 * - a PDTPE (level 3) 721 * - a PDE (level2) 722 * - some PTEs (level 1) 723 */ 724 725 cur_pgd = (pd_entry_t *) old_pgd; 726 bt_pgd = (pd_entry_t *) new_pgd; 727 memset (bt_pgd, 0, PAGE_SIZE); 728 avail = new_pgd + PAGE_SIZE; 729#if PTP_LEVELS > 3 730 /* per-cpu L4 PD */ 731 pd_entry_t *bt_cpu_pgd = bt_pgd; 732 /* pmap_kernel() "shadow" L4 PD */ 733 bt_pgd = (pd_entry_t *) avail; 734 memset(bt_pgd, 0, PAGE_SIZE); 735 avail += PAGE_SIZE; 736 737 /* Install level 3 */ 738 pdtpe = (pd_entry_t *) avail; 739 memset (pdtpe, 0, PAGE_SIZE); 740 avail += PAGE_SIZE; 741 742 addr = ((u_long) pdtpe) - KERNBASE; 743 bt_pgd[pl4_pi(KERNTEXTOFF)] = bt_cpu_pgd[pl4_pi(KERNTEXTOFF)] = 744 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 745 746 __PRINTK(("L3 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 747 " -> L4[%#x]\n", 748 pdtpe, addr, bt_pgd[pl4_pi(KERNTEXTOFF)], pl4_pi(KERNTEXTOFF))); 749#else 750 pdtpe = bt_pgd; 751#endif /* PTP_LEVELS > 3 */ 752 753#if PTP_LEVELS > 2 754 /* Level 2 */ 755 pde = (pd_entry_t *) avail; 756 memset(pde, 0, PAGE_SIZE); 757 avail += PAGE_SIZE; 758 759 addr = ((u_long) pde) - KERNBASE; 760 pdtpe[pl3_pi(KERNTEXTOFF)] = 761 xpmap_ptom_masked(addr) | PG_k | PG_V | PG_RW; 762 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 763 " -> L3[%#x]\n", 764 pde, addr, pdtpe[pl3_pi(KERNTEXTOFF)], pl3_pi(KERNTEXTOFF))); 765#elif defined(PAE) 766 /* our PAE-style level 2: 5 contigous pages (4 L2 + 1 shadow) */ 767 pde = (pd_entry_t *) avail; 768 memset(pde, 0, PAGE_SIZE * 5); 769 avail += PAGE_SIZE * 5; 770 addr = ((u_long) pde) - KERNBASE; 771 /* 772 * enter L2 pages in the L3. 773 * The real L2 kernel PD will be the last one (so that 774 * pde[L2_SLOT_KERN] always point to the shadow). 775 */ 776 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 777 /* 778 * Xen doesn't want R/W mappings in L3 entries, it'll add it 779 * itself. 780 */ 781 pdtpe[i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 782 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 783 " -> L3[%#x]\n", 784 (vaddr_t)pde + PAGE_SIZE * i, addr, pdtpe[i], i)); 785 } 786 addr += PAGE_SIZE; 787 pdtpe[3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 788 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 789 " -> L3[%#x]\n", 790 (vaddr_t)pde + PAGE_SIZE * 4, addr, pdtpe[3], 3)); 791 792#else /* PAE */ 793 pde = bt_pgd; 794#endif /* PTP_LEVELS > 2 */ 795 796 /* Level 1 */ 797 page = KERNTEXTOFF; 798 for (i = 0; i < new_count; i ++) { 799 vaddr_t cur_page = page; 800 801 pte = (pd_entry_t *) avail; 802 avail += PAGE_SIZE; 803 804 memset(pte, 0, PAGE_SIZE); 805 while (pl2_pi(page) == pl2_pi (cur_page)) { 806 if (page >= map_end) { 807 /* not mapped at all */ 808 pte[pl1_pi(page)] = 0; 809 page += PAGE_SIZE; 810 continue; 811 } 812 pte[pl1_pi(page)] = xpmap_ptom_masked(page - KERNBASE); 813 if (page == (vaddr_t)HYPERVISOR_shared_info) { 814 pte[pl1_pi(page)] = xen_start_info.shared_info; 815 __PRINTK(("HYPERVISOR_shared_info " 816 "va %#lx pte %#" PRIxPADDR "\n", 817 HYPERVISOR_shared_info, pte[pl1_pi(page)])); 818 } 819 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 820 == xen_start_info.console.domU.mfn) { 821 xencons_interface = (void *)page; 822 pte[pl1_pi(page)] = xen_start_info.console_mfn; 823 pte[pl1_pi(page)] <<= PAGE_SHIFT; 824 __PRINTK(("xencons_interface " 825 "va %#lx pte %#" PRIxPADDR "\n", 826 xencons_interface, pte[pl1_pi(page)])); 827 } 828 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 829 == xen_start_info.store_mfn) { 830 xenstore_interface = (void *)page; 831 pte[pl1_pi(page)] = xen_start_info.store_mfn; 832 pte[pl1_pi(page)] <<= PAGE_SHIFT; 833 __PRINTK(("xenstore_interface " 834 "va %#lx pte %#" PRIxPADDR "\n", 835 xenstore_interface, pte[pl1_pi(page)])); 836 } 837#ifdef DOM0OPS 838 if (page >= (vaddr_t)atdevbase && 839 page < (vaddr_t)atdevbase + IOM_SIZE) { 840 pte[pl1_pi(page)] = 841 IOM_BEGIN + (page - (vaddr_t)atdevbase); 842 } 843#endif 844 pte[pl1_pi(page)] |= PG_k | PG_V; 845 if (page < text_end) { 846 /* map kernel text RO */ 847 pte[pl1_pi(page)] |= 0; 848 } else if (page >= old_pgd 849 && page < old_pgd + (old_count * PAGE_SIZE)) { 850 /* map old page tables RO */ 851 pte[pl1_pi(page)] |= 0; 852 } else if (page >= new_pgd && 853 page < new_pgd + ((new_count + l2_4_count) * PAGE_SIZE)) { 854 /* map new page tables RO */ 855 pte[pl1_pi(page)] |= 0; 856 } else { 857 /* map page RW */ 858 pte[pl1_pi(page)] |= PG_RW; 859 } 860 861 if ((page >= old_pgd && page < old_pgd + (old_count * PAGE_SIZE)) 862 || page >= new_pgd) { 863 __PRINTK(("va %#lx pa %#lx " 864 "entry 0x%" PRIxPADDR " -> L1[%#x]\n", 865 page, page - KERNBASE, 866 pte[pl1_pi(page)], pl1_pi(page))); 867 } 868 page += PAGE_SIZE; 869 } 870 871 addr = ((u_long) pte) - KERNBASE; 872 pde[pl2_pi(cur_page)] = 873 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 874 __PRINTK(("L1 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 875 " -> L2[%#x]\n", 876 pte, addr, pde[pl2_pi(cur_page)], pl2_pi(cur_page))); 877 /* Mark readonly */ 878 xen_bt_set_readonly((vaddr_t) pte); 879 } 880 881 /* Install recursive page tables mapping */ 882#ifdef PAE 883 /* 884 * we need a shadow page for the kernel's L2 page 885 * The real L2 kernel PD will be the last one (so that 886 * pde[L2_SLOT_KERN] always point to the shadow. 887 */ 888 memcpy(&pde[L2_SLOT_KERN + NPDPG], &pde[L2_SLOT_KERN], PAGE_SIZE); 889 cpu_info_primary.ci_kpm_pdir = &pde[L2_SLOT_KERN + NPDPG]; 890 cpu_info_primary.ci_kpm_pdirpa = 891 (vaddr_t) cpu_info_primary.ci_kpm_pdir - KERNBASE; 892 893 /* 894 * We don't enter a recursive entry from the L3 PD. Instead, 895 * we enter the first 4 L2 pages, which includes the kernel's L2 896 * shadow. But we have to entrer the shadow after switching 897 * %cr3, or Xen will refcount some PTE with the wrong type. 898 */ 899 addr = (u_long)pde - KERNBASE; 900 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 901 pde[PDIR_SLOT_PTE + i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 902 __PRINTK(("pde[%d] va %#" PRIxVADDR " pa %#" PRIxPADDR 903 " entry %#" PRIxPADDR "\n", 904 (int)(PDIR_SLOT_PTE + i), pde + PAGE_SIZE * i, 905 addr, pde[PDIR_SLOT_PTE + i])); 906 } 907#if 0 908 addr += PAGE_SIZE; /* point to shadow L2 */ 909 pde[PDIR_SLOT_PTE + 3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 910 __PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n", 911 (int)(PDIR_SLOT_PTE + 3), pde + PAGE_SIZE * 4, (long)addr, 912 (int64_t)pde[PDIR_SLOT_PTE + 3])); 913#endif 914 /* Mark tables RO, and pin the kernel's shadow as L2 */ 915 addr = (u_long)pde - KERNBASE; 916 for (i = 0; i < 5; i++, addr += PAGE_SIZE) { 917 xen_bt_set_readonly(((vaddr_t)pde) + PAGE_SIZE * i); 918 if (i == 2 || i == 3) 919 continue; 920#if 0 921 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", i, (int64_t)addr)); 922 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 923#endif 924 } 925 if (final) { 926 addr = (u_long)pde - KERNBASE + 3 * PAGE_SIZE; 927 __PRINTK(("pin L2 %d addr %#" PRIxPADDR "\n", 2, addr)); 928 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 929 } 930#if 0 931 addr = (u_long)pde - KERNBASE + 2 * PAGE_SIZE; 932 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr)); 933 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 934#endif 935#else /* PAE */ 936 /* recursive entry in higher-level per-cpu PD and pmap_kernel() */ 937 bt_pgd[PDIR_SLOT_PTE] = xpmap_ptom_masked((paddr_t)bt_pgd - KERNBASE) | PG_k | PG_V; 938#ifdef __x86_64__ 939 bt_cpu_pgd[PDIR_SLOT_PTE] = 940 xpmap_ptom_masked((paddr_t)bt_cpu_pgd - KERNBASE) | PG_k | PG_V; 941#endif /* __x86_64__ */ 942 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] va %#" PRIxVADDR " pa %#" PRIxPADDR 943 " entry %#" PRIxPADDR "\n", new_pgd, (paddr_t)new_pgd - KERNBASE, 944 bt_pgd[PDIR_SLOT_PTE])); 945 /* Mark tables RO */ 946 xen_bt_set_readonly((vaddr_t) pde); 947#endif 948#if PTP_LEVELS > 2 || defined(PAE) 949 xen_bt_set_readonly((vaddr_t) pdtpe); 950#endif 951#if PTP_LEVELS > 3 952 xen_bt_set_readonly(new_pgd); 953#endif 954 /* Pin the PGD */ 955 __PRINTK(("pin PGD: %"PRIxVADDR"\n", new_pgd - KERNBASE)); 956#ifdef __x86_64__ 957 xpq_queue_pin_l4_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 958#elif PAE 959 xpq_queue_pin_l3_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 960#else 961 xpq_queue_pin_l2_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 962#endif 963 964 /* Save phys. addr of PDP, for libkvm. */ 965#ifdef PAE 966 PDPpaddr = (u_long)pde - KERNBASE; /* PDP is the L2 with PAE */ 967#else 968 PDPpaddr = (u_long)bt_pgd - KERNBASE; 969#endif 970 971 /* Switch to new tables */ 972 __PRINTK(("switch to PGD\n")); 973 xpq_queue_pt_switch(xpmap_ptom_masked(new_pgd - KERNBASE)); 974 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] now entry %#" PRIxPADDR "\n", 975 bt_pgd[PDIR_SLOT_PTE])); 976 977#ifdef PAE 978 if (final) { 979 /* save the address of the L3 page */ 980 cpu_info_primary.ci_pae_l3_pdir = pdtpe; 981 cpu_info_primary.ci_pae_l3_pdirpa = (new_pgd - KERNBASE); 982 983 /* now enter kernel's PTE mappings */ 984 addr = (u_long)pde - KERNBASE + PAGE_SIZE * 3; 985 xpq_queue_pte_update( 986 xpmap_ptom(((vaddr_t)&pde[PDIR_SLOT_PTE + 3]) - KERNBASE), 987 xpmap_ptom_masked(addr) | PG_k | PG_V); 988 xpq_flush_queue(); 989 } 990#elif defined(__x86_64__) 991 if (final) { 992 /* save the address of the real per-cpu L4 pgd page */ 993 cpu_info_primary.ci_kpm_pdir = bt_cpu_pgd; 994 cpu_info_primary.ci_kpm_pdirpa = ((paddr_t) bt_cpu_pgd - KERNBASE); 995 } 996#endif 997 998 /* Now we can safely reclaim space taken by old tables */ 999 1000 __PRINTK(("unpin old PGD\n")); 1001 /* Unpin old PGD */ 1002 xpq_queue_unpin_table(xpmap_ptom_masked(old_pgd - KERNBASE)); 1003 /* Mark old tables RW */ 1004 page = old_pgd; 1005 addr = (paddr_t) pde[pl2_pi(page)] & PG_FRAME; 1006 addr = xpmap_mtop(addr); 1007 pte = (pd_entry_t *) ((u_long)addr + KERNBASE); 1008 pte += pl1_pi(page); 1009 __PRINTK(("*pde %#" PRIxPADDR " addr %#" PRIxPADDR " pte %#lx\n", 1010 pde[pl2_pi(page)], addr, (long)pte)); 1011 while (page < old_pgd + (old_count * PAGE_SIZE) && page < map_end) { 1012 addr = xpmap_ptom(((u_long) pte) - KERNBASE); 1013 XENPRINTK(("addr %#" PRIxPADDR " pte %#lx " 1014 "*pte %#" PRIxPADDR "\n", 1015 addr, (long)pte, *pte)); 1016 xpq_queue_pte_update(addr, *pte | PG_RW); 1017 page += PAGE_SIZE; 1018 /* 1019 * Our ptes are contiguous 1020 * so it's safe to just "++" here 1021 */ 1022 pte++; 1023 } 1024 xpq_flush_queue(); 1025} 1026 1027 1028/* 1029 * Bootstrap helper functions 1030 */ 1031 1032/* 1033 * Mark a page readonly 1034 * XXX: assuming vaddr = paddr + KERNBASE 1035 */ 1036 1037static void 1038xen_bt_set_readonly (vaddr_t page) 1039{ 1040 pt_entry_t entry; 1041 1042 entry = xpmap_ptom_masked(page - KERNBASE); 1043 entry |= PG_k | PG_V; 1044 1045 HYPERVISOR_update_va_mapping (page, entry, UVMF_INVLPG); 1046} 1047 1048#ifdef __x86_64__ 1049void 1050xen_set_user_pgd(paddr_t page) 1051{ 1052 struct mmuext_op op; 1053 int s = splvm(); 1054 1055 xpq_flush_queue(); 1056 op.cmd = MMUEXT_NEW_USER_BASEPTR; 1057 op.arg1.mfn = pfn_to_mfn(page >> PAGE_SHIFT); 1058 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 1059 panic("xen_set_user_pgd: failed to install new user page" 1060 " directory %#" PRIxPADDR, page); 1061 splx(s); 1062} 1063#endif /* __x86_64__ */ 1064