x86_xpmap.c revision 1.8
1/* $NetBSD: x86_xpmap.c,v 1.8 2008/04/14 13:38:03 cegger 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 * 3. All advertising materials mentioning features or use of this software 31 * must display the following acknowledgement: 32 * This product includes software developed by Manuel Bouyer. 33 * 4. The name of the author may not be used to endorse or promote products 34 * derived from this software without specific prior written permission. 35 * 36 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 37 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 38 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 39 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 40 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 41 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 42 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 43 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 44 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 45 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 46 * 47 */ 48 49/* 50 * 51 * Copyright (c) 2004 Christian Limpach. 52 * All rights reserved. 53 * 54 * Redistribution and use in source and binary forms, with or without 55 * modification, are permitted provided that the following conditions 56 * are met: 57 * 1. Redistributions of source code must retain the above copyright 58 * notice, this list of conditions and the following disclaimer. 59 * 2. Redistributions in binary form must reproduce the above copyright 60 * notice, this list of conditions and the following disclaimer in the 61 * documentation and/or other materials provided with the distribution. 62 * 3. All advertising materials mentioning features or use of this software 63 * must display the following acknowledgement: 64 * This product includes software developed by Christian Limpach. 65 * 4. The name of the author may not be used to endorse or promote products 66 * derived from this software without specific prior written permission. 67 * 68 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 69 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 70 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 71 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 72 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 73 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 74 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 75 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 76 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 77 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 78 */ 79 80 81#include <sys/cdefs.h> 82__KERNEL_RCSID(0, "$NetBSD: x86_xpmap.c,v 1.8 2008/04/14 13:38:03 cegger Exp $"); 83 84#include "opt_xen.h" 85#include "opt_ddb.h" 86#include "ksyms.h" 87 88#include <sys/param.h> 89#include <sys/systm.h> 90 91#include <uvm/uvm.h> 92 93#include <machine/pmap.h> 94#include <machine/gdt.h> 95#include <xen/xenfunc.h> 96 97#include <dev/isa/isareg.h> 98#include <machine/isa_machdep.h> 99 100#undef XENDEBUG 101/* #define XENDEBUG_SYNC */ 102/* #define XENDEBUG_LOW */ 103 104#ifdef XENDEBUG 105#define XENPRINTF(x) printf x 106#define XENPRINTK(x) printk x 107#define XENPRINTK2(x) /* printk x */ 108 109static char XBUF[256]; 110#else 111#define XENPRINTF(x) 112#define XENPRINTK(x) 113#define XENPRINTK2(x) 114#endif 115#define PRINTF(x) printf x 116#define PRINTK(x) printk x 117 118/* on x86_64 kernel runs in ring 3 */ 119#ifdef __x86_64__ 120#define PG_k PG_u 121#else 122#define PG_k 0 123#endif 124 125volatile shared_info_t *HYPERVISOR_shared_info; 126union start_info_union start_info_union; 127unsigned long *xpmap_phys_to_machine_mapping; 128 129void xen_failsafe_handler(void); 130 131#ifdef XEN3 132#define HYPERVISOR_mmu_update_self(req, count, success_count) \ 133 HYPERVISOR_mmu_update((req), (count), (success_count), DOMID_SELF) 134#else 135#define HYPERVISOR_mmu_update_self(req, count, success_count) \ 136 HYPERVISOR_mmu_update((req), (count), (success_count)) 137#endif 138 139void 140xen_failsafe_handler(void) 141{ 142 143 panic("xen_failsafe_handler called!\n"); 144} 145 146 147void 148xen_set_ldt(vaddr_t base, uint32_t entries) 149{ 150 vaddr_t va; 151 vaddr_t end; 152 pt_entry_t *ptp; 153 int s; 154 155#ifdef __x86_64__ 156 end = base + (entries << 3); 157#else 158 end = base + entries * sizeof(union descriptor); 159#endif 160 161 for (va = base; va < end; va += PAGE_SIZE) { 162 KASSERT(va >= VM_MIN_KERNEL_ADDRESS); 163 ptp = kvtopte(va); 164 XENPRINTF(("xen_set_ldt %p %d %p\n", (void *)base, 165 entries, ptp)); 166 pmap_pte_clearbits(ptp, PG_RW); 167 } 168 s = splvm(); 169 xpq_queue_set_ldt(base, entries); 170 xpq_flush_queue(); 171 splx(s); 172} 173 174#ifdef XENDEBUG 175void xpq_debug_dump(void); 176#endif 177 178#define XPQUEUE_SIZE 2048 179static mmu_update_t xpq_queue[XPQUEUE_SIZE]; 180static int xpq_idx = 0; 181 182void 183xpq_flush_queue(void) 184{ 185 int i, ok; 186 187 XENPRINTK2(("flush queue %p entries %d\n", xpq_queue, xpq_idx)); 188 for (i = 0; i < xpq_idx; i++) 189 XENPRINTK2(("%d: %p %08" PRIx64 "\n", i, 190 (uint64_t)xpq_queue[i].ptr, (uint64_t)xpq_queue[i].val)); 191 if (xpq_idx != 0 && 192 HYPERVISOR_mmu_update_self(xpq_queue, xpq_idx, &ok) < 0) { 193 printf("xpq_flush_queue: %d entries \n", xpq_idx); 194 for (i = 0; i < xpq_idx; i++) 195 printf("0x%016" PRIx64 ": 0x%016" PRIx64 "\n", 196 (uint64_t)xpq_queue[i].ptr, 197 (uint64_t)xpq_queue[i].val); 198 panic("HYPERVISOR_mmu_update failed\n"); 199 } 200 xpq_idx = 0; 201} 202 203static inline void 204xpq_increment_idx(void) 205{ 206 207 xpq_idx++; 208 if (__predict_false(xpq_idx == XPQUEUE_SIZE)) 209 xpq_flush_queue(); 210} 211 212void 213xpq_queue_machphys_update(paddr_t ma, paddr_t pa) 214{ 215 XENPRINTK2(("xpq_queue_machphys_update ma=0x%" PRIx64 " pa=0x%" PRIx64 216 "\n", (int64_t)ma, (int64_t)pa)); 217 xpq_queue[xpq_idx].ptr = ma | MMU_MACHPHYS_UPDATE; 218 xpq_queue[xpq_idx].val = (pa - XPMAP_OFFSET) >> PAGE_SHIFT; 219 xpq_increment_idx(); 220#ifdef XENDEBUG_SYNC 221 xpq_flush_queue(); 222#endif 223} 224 225void 226xpq_queue_pte_update(paddr_t ptr, pt_entry_t val) 227{ 228 229 KASSERT((ptr & 3) == 0); 230 xpq_queue[xpq_idx].ptr = (paddr_t)ptr | MMU_NORMAL_PT_UPDATE; 231 xpq_queue[xpq_idx].val = val; 232 xpq_increment_idx(); 233#ifdef XENDEBUG_SYNC 234 xpq_flush_queue(); 235#endif 236} 237 238#ifdef XEN3 239void 240xpq_queue_pt_switch(paddr_t pa) 241{ 242 struct mmuext_op op; 243 xpq_flush_queue(); 244 245 XENPRINTK2(("xpq_queue_pt_switch: 0x%" PRIx64 " 0x%" PRIx64 "\n", 246 (int64_t)pa, (int64_t)pa)); 247 op.cmd = MMUEXT_NEW_BASEPTR; 248 op.arg1.mfn = pa >> PAGE_SHIFT; 249 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 250 panic("xpq_queue_pt_switch"); 251} 252 253void 254xpq_queue_pin_table(paddr_t pa) 255{ 256 struct mmuext_op op; 257 xpq_flush_queue(); 258 259 XENPRINTK2(("xpq_queue_pin_table: 0x%" PRIx64 " 0x%" PRIx64 "\n", 260 (int64_t)pa, (int64_t)pa)); 261 op.arg1.mfn = pa >> PAGE_SHIFT; 262 263#if defined(__x86_64__) 264 op.cmd = MMUEXT_PIN_L4_TABLE; 265#else 266 op.cmd = MMUEXT_PIN_L2_TABLE; 267#endif 268 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 269 panic("xpq_queue_pin_table"); 270} 271 272#ifdef PAE 273static void 274xpq_queue_pin_l3_table(paddr_t pa) 275{ 276 struct mmuext_op op; 277 xpq_flush_queue(); 278 279 XENPRINTK2(("xpq_queue_pin_l2_table: 0x%" PRIx64 " 0x%" PRIx64 "\n", 280 (int64_t)pa, (int64_t)pa)); 281 op.arg1.mfn = pa >> PAGE_SHIFT; 282 283 op.cmd = MMUEXT_PIN_L3_TABLE; 284 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 285 panic("xpq_queue_pin_table"); 286} 287#endif 288 289void 290xpq_queue_unpin_table(paddr_t pa) 291{ 292 struct mmuext_op op; 293 xpq_flush_queue(); 294 295 XENPRINTK2(("xpq_queue_unpin_table: 0x%" PRIx64 " 0x%" PRIx64 "\n", 296 (int64_t)pa, (int64_t)pa)); 297 op.arg1.mfn = pa >> PAGE_SHIFT; 298 op.cmd = MMUEXT_UNPIN_TABLE; 299 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 300 panic("xpq_queue_unpin_table"); 301} 302 303void 304xpq_queue_set_ldt(vaddr_t va, uint32_t entries) 305{ 306 struct mmuext_op op; 307 xpq_flush_queue(); 308 309 XENPRINTK2(("xpq_queue_set_ldt\n")); 310 KASSERT(va == (va & ~PAGE_MASK)); 311 op.cmd = MMUEXT_SET_LDT; 312 op.arg1.linear_addr = va; 313 op.arg2.nr_ents = entries; 314 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 315 panic("xpq_queue_set_ldt"); 316} 317 318void 319xpq_queue_tlb_flush(void) 320{ 321 struct mmuext_op op; 322 xpq_flush_queue(); 323 324 XENPRINTK2(("xpq_queue_tlb_flush\n")); 325 op.cmd = MMUEXT_TLB_FLUSH_LOCAL; 326 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 327 panic("xpq_queue_tlb_flush"); 328} 329 330void 331xpq_flush_cache(void) 332{ 333 struct mmuext_op op; 334 int s = splvm(); 335 xpq_flush_queue(); 336 337 XENPRINTK2(("xpq_queue_flush_cache\n")); 338 op.cmd = MMUEXT_FLUSH_CACHE; 339 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 340 panic("xpq_flush_cache"); 341 splx(s); 342} 343 344void 345xpq_queue_invlpg(vaddr_t va) 346{ 347 struct mmuext_op op; 348 xpq_flush_queue(); 349 350 XENPRINTK2(("xpq_queue_invlpg %p\n", (void *)va)); 351 op.cmd = MMUEXT_INVLPG_LOCAL; 352 op.arg1.linear_addr = (va & ~PAGE_MASK); 353 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 354 panic("xpq_queue_invlpg"); 355} 356 357int 358xpq_update_foreign(paddr_t ptr, pt_entry_t val, int dom) 359{ 360 mmu_update_t op; 361 int ok; 362 xpq_flush_queue(); 363 364 op.ptr = ptr; 365 op.val = val; 366 if (HYPERVISOR_mmu_update(&op, 1, &ok, dom) < 0) 367 return EFAULT; 368 return (0); 369} 370#else /* XEN3 */ 371void 372xpq_queue_pt_switch(paddr_t pa) 373{ 374 375 XENPRINTK2(("xpq_queue_pt_switch: %p %p\n", (void *)pa, (void *)pa)); 376 xpq_queue[xpq_idx].ptr = pa | MMU_EXTENDED_COMMAND; 377 xpq_queue[xpq_idx].val = MMUEXT_NEW_BASEPTR; 378 xpq_increment_idx(); 379} 380 381void 382xpq_queue_pin_table(paddr_t pa) 383{ 384 385 XENPRINTK2(("xpq_queue_pin_table: %p %p\n", (void *)pa, (void *)pa)); 386 xpq_queue[xpq_idx].ptr = pa | MMU_EXTENDED_COMMAND; 387 xpq_queue[xpq_idx].val = MMUEXT_PIN_L2_TABLE; 388 xpq_increment_idx(); 389} 390 391void 392xpq_queue_unpin_table(paddr_t pa) 393{ 394 395 XENPRINTK2(("xpq_queue_unpin_table: %p %p\n", (void *)pa, (void *)pa)); 396 xpq_queue[xpq_idx].ptr = pa | MMU_EXTENDED_COMMAND; 397 xpq_queue[xpq_idx].val = MMUEXT_UNPIN_TABLE; 398 xpq_increment_idx(); 399} 400 401void 402xpq_queue_set_ldt(vaddr_t va, uint32_t entries) 403{ 404 405 XENPRINTK2(("xpq_queue_set_ldt\n")); 406 KASSERT(va == (va & ~PAGE_MASK)); 407 xpq_queue[xpq_idx].ptr = MMU_EXTENDED_COMMAND | va; 408 xpq_queue[xpq_idx].val = MMUEXT_SET_LDT | (entries << MMUEXT_CMD_SHIFT); 409 xpq_increment_idx(); 410} 411 412void 413xpq_queue_tlb_flush(void) 414{ 415 416 XENPRINTK2(("xpq_queue_tlb_flush\n")); 417 xpq_queue[xpq_idx].ptr = MMU_EXTENDED_COMMAND; 418 xpq_queue[xpq_idx].val = MMUEXT_TLB_FLUSH; 419 xpq_increment_idx(); 420} 421 422void 423xpq_flush_cache(void) 424{ 425 int s = splvm(); 426 427 XENPRINTK2(("xpq_queue_flush_cache\n")); 428 xpq_queue[xpq_idx].ptr = MMU_EXTENDED_COMMAND; 429 xpq_queue[xpq_idx].val = MMUEXT_FLUSH_CACHE; 430 xpq_increment_idx(); 431 xpq_flush_queue(); 432 splx(s); 433} 434 435void 436xpq_queue_invlpg(vaddr_t va) 437{ 438 439 XENPRINTK2(("xpq_queue_invlpg %p\n", (void *)va)); 440 xpq_queue[xpq_idx].ptr = (va & ~PAGE_MASK) | MMU_EXTENDED_COMMAND; 441 xpq_queue[xpq_idx].val = MMUEXT_INVLPG; 442 xpq_increment_idx(); 443} 444 445int 446xpq_update_foreign(paddr_t ptr, pt_entry_t val, int dom) 447{ 448 mmu_update_t xpq_up[3]; 449 450 xpq_up[0].ptr = MMU_EXTENDED_COMMAND; 451 xpq_up[0].val = MMUEXT_SET_FOREIGNDOM | (dom << 16); 452 xpq_up[1].ptr = ptr; 453 xpq_up[1].val = val; 454 if (HYPERVISOR_mmu_update_self(xpq_up, 2, NULL) < 0) 455 return EFAULT; 456 return (0); 457} 458#endif /* XEN3 */ 459 460#ifdef XENDEBUG 461void 462xpq_debug_dump(void) 463{ 464 int i; 465 466 XENPRINTK2(("idx: %d\n", xpq_idx)); 467 for (i = 0; i < xpq_idx; i++) { 468 sprintf(XBUF, "%" PRIx64 " %08" PRIx64, 469 (uint64_t)xpq_queue[i].ptr, (uint64_t)xpq_queue[i].val); 470 if (++i < xpq_idx) 471 sprintf(XBUF + strlen(XBUF), "%" PRIx64 " %08" PRIx64, 472 (uint64_t)xpq_queue[i].ptr, (uint64_t)xpq_queue[i].val); 473 if (++i < xpq_idx) 474 sprintf(XBUF + strlen(XBUF), "%" PRIx64 " %08" PRIx64, 475 (uint64_t)xpq_queue[i].ptr, (uint64_t)xpq_queue[i].val); 476 if (++i < xpq_idx) 477 sprintf(XBUF + strlen(XBUF), "%" PRIx64 " %08" PRIx64, 478 (uint64_t)xpq_queue[i].ptr, (uint64_t)xpq_queue[i].val); 479 XENPRINTK2(("%d: %s\n", xpq_idx, XBUF)); 480 } 481} 482#endif 483 484 485extern volatile struct xencons_interface *xencons_interface; /* XXX */ 486extern struct xenstore_domain_interface *xenstore_interface; /* XXX */ 487 488static void xen_bt_set_readonly (vaddr_t); 489static void xen_bootstrap_tables (vaddr_t, vaddr_t, int, int, int); 490 491/* How many PDEs ? */ 492#if L2_SLOT_KERNBASE > 0 493#define TABLE_L2_ENTRIES (2 * (NKL2_KIMG_ENTRIES + 1)) 494#else 495#define TABLE_L2_ENTRIES (NKL2_KIMG_ENTRIES + 1) 496#endif 497 498/* 499 * Construct and switch to new pagetables 500 * first_avail is the first vaddr we can use after 501 * we get rid of Xen pagetables 502 */ 503 504vaddr_t xen_pmap_bootstrap (void); 505 506/* 507 * Function to get rid of Xen bootstrap tables 508 */ 509 510/* How many PDP do we need: */ 511#ifdef PAE 512/* 513 * For PAE, we consider a single contigous L2 "superpage" of 4 pages, 514 * all of them mapped by the L3 page. We also need a shadow page 515 * for L3[3]. 516 */ 517static const int l2_4_count = 6; 518#else 519static const int l2_4_count = PTP_LEVELS - 1; 520#endif 521 522vaddr_t 523xen_pmap_bootstrap(void) 524{ 525 int count, oldcount; 526 long mapsize; 527 vaddr_t bootstrap_tables, init_tables; 528 529 xpmap_phys_to_machine_mapping = 530 (unsigned long *)xen_start_info.mfn_list; 531 init_tables = xen_start_info.pt_base; 532 __PRINTK(("xen_arch_pmap_bootstrap init_tables=0x%lx\n", init_tables)); 533 534 /* Space after Xen boostrap tables should be free */ 535 bootstrap_tables = xen_start_info.pt_base + 536 (xen_start_info.nr_pt_frames * PAGE_SIZE); 537 538 /* 539 * Calculate how many space we need 540 * first everything mapped before the Xen bootstrap tables 541 */ 542 mapsize = init_tables - KERNTEXTOFF; 543 /* after the tables we'll have: 544 * - UAREA 545 * - dummy user PGD (x86_64) 546 * - HYPERVISOR_shared_info 547 * - ISA I/O mem (if needed) 548 */ 549 mapsize += UPAGES * NBPG; 550#ifdef __x86_64__ 551 mapsize += NBPG; 552#endif 553 mapsize += NBPG; 554 555#ifdef DOM0OPS 556 if (xen_start_info.flags & SIF_INITDOMAIN) { 557 /* space for ISA I/O mem */ 558 mapsize += IOM_SIZE; 559 } 560#endif 561 /* at this point mapsize doens't include the table size */ 562 563#ifdef __x86_64__ 564 count = TABLE_L2_ENTRIES; 565#else 566 count = (mapsize + (NBPD_L2 -1)) >> L2_SHIFT; 567#endif /* __x86_64__ */ 568 569 /* now compute how many L2 pages we need exactly */ 570 XENPRINTK(("bootstrap_final mapsize 0x%lx count %d\n", mapsize, count)); 571 while (mapsize + (count + l2_4_count) * PAGE_SIZE + KERNTEXTOFF > 572 ((long)count << L2_SHIFT) + KERNBASE) { 573 count++; 574 } 575#ifndef __x86_64__ 576 /* 577 * one more L2 page: we'll alocate several pages after kva_start 578 * in pmap_bootstrap() before pmap_growkernel(), which have not been 579 * counted here. It's not a big issue to allocate one more L2 as 580 * pmap_growkernel() will be called anyway. 581 */ 582 count++; 583 nkptp[1] = count; 584#endif 585 586 /* 587 * install bootstrap pages. We may need more L2 pages than will 588 * have the final table here, as it's installed after the final table 589 */ 590 oldcount = count; 591 592bootstrap_again: 593 XENPRINTK(("bootstrap_again oldcount %d\n", oldcount)); 594 /* 595 * Xen space we'll reclaim may not be enough for our new page tables, 596 * move bootstrap tables if necessary 597 */ 598 if (bootstrap_tables < init_tables + ((count + l2_4_count) * PAGE_SIZE)) 599 bootstrap_tables = init_tables + 600 ((count + l2_4_count) * PAGE_SIZE); 601 /* make sure we have enough to map the bootstrap_tables */ 602 if (bootstrap_tables + ((oldcount + l2_4_count) * PAGE_SIZE) > 603 ((long)oldcount << L2_SHIFT) + KERNBASE) { 604 oldcount++; 605 goto bootstrap_again; 606 } 607 608 /* Create temporary tables */ 609 xen_bootstrap_tables(xen_start_info.pt_base, bootstrap_tables, 610 xen_start_info.nr_pt_frames, oldcount, 0); 611 612 /* Create final tables */ 613 xen_bootstrap_tables(bootstrap_tables, init_tables, 614 oldcount + l2_4_count, count, 1); 615 616 /* zero out free space after tables */ 617 memset((void *)(init_tables + ((count + l2_4_count) * PAGE_SIZE)), 0, 618 (UPAGES + 1) * NBPG); 619 return (init_tables + ((count + l2_4_count) * PAGE_SIZE)); 620} 621 622 623/* 624 * Build a new table and switch to it 625 * old_count is # of old tables (including PGD, PDTPE and PDE) 626 * new_count is # of new tables (PTE only) 627 * we assume areas don't overlap 628 */ 629 630 631static void 632xen_bootstrap_tables (vaddr_t old_pgd, vaddr_t new_pgd, 633 int old_count, int new_count, int final) 634{ 635 pd_entry_t *pdtpe, *pde, *pte; 636 pd_entry_t *cur_pgd, *bt_pgd; 637 paddr_t addr; 638 vaddr_t page, avail, text_end, map_end; 639 int i; 640 extern char __data_start; 641 642 __PRINTK(("xen_bootstrap_tables(0x%lx, 0x%lx, %d, %d)\n", 643 old_pgd, new_pgd, old_count, new_count)); 644 text_end = ((vaddr_t)&__data_start) & ~PAGE_MASK; 645 /* 646 * size of R/W area after kernel text: 647 * xencons_interface (if present) 648 * xenstore_interface (if present) 649 * table pages (new_count + l2_4_count entries) 650 * extra mappings (only when final is true): 651 * UAREA 652 * dummy user PGD (x86_64 only)/gdt page (i386 only) 653 * HYPERVISOR_shared_info 654 * ISA I/O mem (if needed) 655 */ 656 map_end = new_pgd + ((new_count + l2_4_count) * NBPG); 657 if (final) { 658 map_end += (UPAGES + 1) * NBPG; 659 HYPERVISOR_shared_info = (shared_info_t *)map_end; 660 map_end += NBPG; 661 } 662 /* 663 * we always set atdevbase, as it's used by init386 to find the first 664 * available VA. map_end is updated only if we are dom0, so 665 * atdevbase -> atdevbase + IOM_SIZE will be mapped only in 666 * this case. 667 */ 668 if (final) 669 atdevbase = map_end; 670#ifdef DOM0OPS 671 if (final && (xen_start_info.flags & SIF_INITDOMAIN)) { 672 /* ISA I/O mem */ 673 map_end += IOM_SIZE; 674 } 675#endif /* DOM0OPS */ 676 677 __PRINTK(("xen_bootstrap_tables text_end 0x%lx map_end 0x%lx\n", 678 text_end, map_end)); 679 __PRINTK(("console 0x%lx ", xen_start_info.console_mfn)); 680 __PRINTK(("xenstore 0x%lx\n", xen_start_info.store_mfn)); 681 682 /* 683 * Create bootstrap page tables 684 * What we need: 685 * - a PGD (level 4) 686 * - a PDTPE (level 3) 687 * - a PDE (level2) 688 * - some PTEs (level 1) 689 */ 690 691 cur_pgd = (pd_entry_t *) old_pgd; 692 bt_pgd = (pd_entry_t *) new_pgd; 693 memset (bt_pgd, 0, PAGE_SIZE); 694 avail = new_pgd + PAGE_SIZE; 695#if PTP_LEVELS > 3 696 /* Install level 3 */ 697 pdtpe = (pd_entry_t *) avail; 698 memset (pdtpe, 0, PAGE_SIZE); 699 avail += PAGE_SIZE; 700 701 addr = ((u_long) pdtpe) - KERNBASE; 702 bt_pgd[pl4_pi(KERNTEXTOFF)] = 703 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 704 705 __PRINTK(("L3 va 0x%lx pa 0x%" PRIx64 " entry 0x%" PRIx64 " -> L4[0x%x]\n", 706 pdtpe, (uint64_t)addr, (uint64_t)bt_pgd[pl4_pi(KERNTEXTOFF)], 707 pl4_pi(KERNTEXTOFF))); 708#else 709 pdtpe = bt_pgd; 710#endif /* PTP_LEVELS > 3 */ 711 712#if PTP_LEVELS > 2 713 /* Level 2 */ 714 pde = (pd_entry_t *) avail; 715 memset(pde, 0, PAGE_SIZE); 716 avail += PAGE_SIZE; 717 718 addr = ((u_long) pde) - KERNBASE; 719 pdtpe[pl3_pi(KERNTEXTOFF)] = 720 xpmap_ptom_masked(addr) | PG_k | PG_V | PG_RW; 721 __PRINTK(("L2 va 0x%lx pa 0x%" PRIx64 " entry 0x%" PRIx64 " -> L3[0x%x]\n", 722 pde, (int64_t)addr, (int64_t)pdtpe[pl3_pi(KERNTEXTOFF)], 723 pl3_pi(KERNTEXTOFF))); 724#elif defined(PAE) 725 /* our PAE-style level 2: 5 contigous pages (4 L2 + 1 shadow) */ 726 pde = (pd_entry_t *) avail; 727 memset(pde, 0, PAGE_SIZE * 5); 728 avail += PAGE_SIZE * 5; 729 addr = ((u_long) pde) - KERNBASE; 730 /* 731 * enter L2 pages in the L3. 732 * The real L2 kernel PD will be the last one (so that 733 * pde[L2_SLOT_KERN] always point to the shadow). 734 */ 735 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 736 /* 737 * Xen doens't want R/W mappings in L3 entries, it'll add it 738 * itself. 739 */ 740 pdtpe[i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 741 __PRINTK(("L2 va 0x%lx pa 0x%" PRIx64 " entry 0x%" PRIx64 742 " -> L3[0x%x]\n", (vaddr_t)pde + PAGE_SIZE * i, 743 (int64_t)addr, (int64_t)pdtpe[i], i)); 744 } 745 addr += PAGE_SIZE; 746 pdtpe[3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 747 __PRINTK(("L2 va 0x%lx pa 0x%" PRIx64 " entry 0x%" PRIx64 748 " -> L3[0x%x]\n", (vaddr_t)pde + PAGE_SIZE * 4, 749 (int64_t)addr, (int64_t)pdtpe[3], 3)); 750 751#else /* PAE */ 752 pde = bt_pgd; 753#endif /* PTP_LEVELS > 2 */ 754 755 /* Level 1 */ 756 page = KERNTEXTOFF; 757 for (i = 0; i < new_count; i ++) { 758 vaddr_t cur_page = page; 759 760 pte = (pd_entry_t *) avail; 761 avail += PAGE_SIZE; 762 763 memset(pte, 0, PAGE_SIZE); 764 while (pl2_pi(page) == pl2_pi (cur_page)) { 765 if (page >= map_end) { 766 /* not mapped at all */ 767 pte[pl1_pi(page)] = 0; 768 page += PAGE_SIZE; 769 continue; 770 } 771 pte[pl1_pi(page)] = xpmap_ptom_masked(page - KERNBASE); 772 if (page == (vaddr_t)HYPERVISOR_shared_info) { 773 pte[pl1_pi(page)] = xen_start_info.shared_info; 774 __PRINTK(("HYPERVISOR_shared_info " 775 "va 0x%lx pte 0x%" PRIx64 "\n", 776 HYPERVISOR_shared_info, (int64_t)pte[pl1_pi(page)])); 777 } 778#ifdef XEN3 779 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 780 == xen_start_info.console_mfn) { 781 xencons_interface = (void *)page; 782 pte[pl1_pi(page)] = xen_start_info.console_mfn; 783 pte[pl1_pi(page)] <<= PAGE_SHIFT; 784 __PRINTK(("xencons_interface " 785 "va 0x%lx pte 0x%" PRIx64 "\n", 786 xencons_interface, (int64_t)pte[pl1_pi(page)])); 787 } 788 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 789 == xen_start_info.store_mfn) { 790 xenstore_interface = (void *)page; 791 pte[pl1_pi(page)] = xen_start_info.store_mfn; 792 pte[pl1_pi(page)] <<= PAGE_SHIFT; 793 __PRINTK(("xenstore_interface " 794 "va 0x%lx pte 0x%" PRIx64 "\n", 795 xenstore_interface, (int64_t)pte[pl1_pi(page)])); 796 } 797#endif /* XEN3 */ 798#ifdef DOM0OPS 799 if (page >= (vaddr_t)atdevbase && 800 page < (vaddr_t)atdevbase + IOM_SIZE) { 801 pte[pl1_pi(page)] = 802 IOM_BEGIN + (page - (vaddr_t)atdevbase); 803 } 804#endif 805 pte[pl1_pi(page)] |= PG_k | PG_V; 806 if (page < text_end) { 807 /* map kernel text RO */ 808 pte[pl1_pi(page)] |= 0; 809 } else if (page >= old_pgd 810 && page < old_pgd + (old_count * PAGE_SIZE)) { 811 /* map old page tables RO */ 812 pte[pl1_pi(page)] |= 0; 813 } else if (page >= new_pgd && 814 page < new_pgd + ((new_count + l2_4_count) * PAGE_SIZE)) { 815 /* map new page tables RO */ 816 pte[pl1_pi(page)] |= 0; 817 } else { 818 /* map page RW */ 819 pte[pl1_pi(page)] |= PG_RW; 820 } 821 822 if ((page >= old_pgd && page < old_pgd + (old_count * PAGE_SIZE)) || page >= new_pgd) 823 __PRINTK(("va 0x%lx pa 0x%lx " 824 "entry 0x%" PRIx64 " -> L1[0x%x]\n", 825 page, page - KERNBASE, 826 (int64_t)pte[pl1_pi(page)], pl1_pi(page))); 827 page += PAGE_SIZE; 828 } 829 830 addr = ((u_long) pte) - KERNBASE; 831 pde[pl2_pi(cur_page)] = 832 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 833 __PRINTK(("L1 va 0x%lx pa 0x%" PRIx64 " entry 0x%" PRIx64 834 " -> L2[0x%x]\n", pte, (int64_t)addr, 835 (int64_t)pde[pl2_pi(cur_page)], pl2_pi(cur_page))); 836 /* Mark readonly */ 837 xen_bt_set_readonly((vaddr_t) pte); 838 } 839 840 /* Install recursive page tables mapping */ 841#ifdef PAE 842 /* 843 * we need a shadow page for the kernel's L2 page 844 * The real L2 kernel PD will be the last one (so that 845 * pde[L2_SLOT_KERN] always point to the shadow. 846 */ 847 memcpy(&pde[L2_SLOT_KERN + NPDPG], &pde[L2_SLOT_KERN], PAGE_SIZE); 848 pmap_kl2pd = &pde[L2_SLOT_KERN + NPDPG]; 849 pmap_kl2paddr = (u_long)pmap_kl2pd - KERNBASE; 850 851 /* 852 * We don't enter a recursive entry from the L3 PD. Instead, 853 * we enter the first 4 L2 pages, which includes the kernel's L2 854 * shadow. But we have to entrer the shadow after switching 855 * %cr3, or Xen will refcount some PTE with the wrong type. 856 */ 857 addr = (u_long)pde - KERNBASE; 858 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 859 pde[PDIR_SLOT_PTE + i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 860 __PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n", 861 (int)(PDIR_SLOT_PTE + i), pde + PAGE_SIZE * i, (long)addr, 862 (int64_t)pde[PDIR_SLOT_PTE + i])); 863 } 864#if 0 865 addr += PAGE_SIZE; /* point to shadow L2 */ 866 pde[PDIR_SLOT_PTE + 3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 867 __PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n", 868 (int)(PDIR_SLOT_PTE + 3), pde + PAGE_SIZE * 4, (long)addr, 869 (int64_t)pde[PDIR_SLOT_PTE + 3])); 870#endif 871 /* Mark tables RO, and pin the kenrel's shadow as L2 */ 872 addr = (u_long)pde - KERNBASE; 873 for (i = 0; i < 5; i++, addr += PAGE_SIZE) { 874 xen_bt_set_readonly(((vaddr_t)pde) + PAGE_SIZE * i); 875 if (i == 2 || i == 3) 876 continue; 877#if 0 878 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", i, (int64_t)addr)); 879 xpq_queue_pin_table(xpmap_ptom_masked(addr)); 880#endif 881 } 882 if (final) { 883 addr = (u_long)pde - KERNBASE + 3 * PAGE_SIZE; 884 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr)); 885 xpq_queue_pin_table(xpmap_ptom_masked(addr)); 886 } 887#if 0 888 addr = (u_long)pde - KERNBASE + 2 * PAGE_SIZE; 889 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr)); 890 xpq_queue_pin_table(xpmap_ptom_masked(addr)); 891#endif 892#else /* PAE */ 893 /* recursive entry in higher-level PD */ 894 bt_pgd[PDIR_SLOT_PTE] = 895 xpmap_ptom_masked(new_pgd - KERNBASE) | PG_k | PG_V; 896 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] va 0x%lx pa 0x%" PRIx64 897 " entry 0x%" PRIx64 "\n", new_pgd, (int64_t)new_pgd - KERNBASE, 898 (int64_t)bt_pgd[PDIR_SLOT_PTE])); 899 /* Mark tables RO */ 900 xen_bt_set_readonly((vaddr_t) pde); 901#endif 902#if PTP_LEVELS > 2 || defined(PAE) 903 xen_bt_set_readonly((vaddr_t) pdtpe); 904#endif 905#if PTP_LEVELS > 3 906 xen_bt_set_readonly(new_pgd); 907#endif 908 /* Pin the PGD */ 909 __PRINTK(("pin PDG\n")); 910#ifdef PAE 911 xpq_queue_pin_l3_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 912#else 913 xpq_queue_pin_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 914#endif 915#ifdef __i386__ 916 /* Save phys. addr of PDP, for libkvm. */ 917 PDPpaddr = (long)pde; 918#ifdef PAE 919 /* also save the address of the L3 page */ 920 pmap_l3pd = pdtpe; 921 pmap_l3paddr = (new_pgd - KERNBASE); 922#endif /* PAE */ 923#endif /* i386 */ 924 /* Switch to new tables */ 925 __PRINTK(("switch to PDG\n")); 926 xpq_queue_pt_switch(xpmap_ptom_masked(new_pgd - KERNBASE)); 927 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] now entry 0x%" PRIx64 "\n", 928 (int64_t)bt_pgd[PDIR_SLOT_PTE])); 929#ifdef PAE 930 if (final) { 931 /* now enter kernel's PTE mappings */ 932 addr = (u_long)pde - KERNBASE + PAGE_SIZE * 3; 933 xpq_queue_pte_update( 934 xpmap_ptom(((vaddr_t)&pde[PDIR_SLOT_PTE + 3]) - KERNBASE), 935 xpmap_ptom_masked(addr) | PG_k | PG_V); 936 xpq_flush_queue(); 937 } 938#endif 939 940 941 942 /* Now we can safely reclaim space taken by old tables */ 943 944 __PRINTK(("unpin old PDG\n")); 945 /* Unpin old PGD */ 946 xpq_queue_unpin_table(xpmap_ptom_masked(old_pgd - KERNBASE)); 947 /* Mark old tables RW */ 948 page = old_pgd; 949 addr = (paddr_t) pde[pl2_pi(page)] & PG_FRAME; 950 addr = xpmap_mtop(addr); 951 pte = (pd_entry_t *) ((u_long)addr + KERNBASE); 952 pte += pl1_pi(page); 953 __PRINTK(("*pde 0x%" PRIx64 " addr 0x%" PRIx64 " pte 0x%lx\n", 954 (int64_t)pde[pl2_pi(page)], (int64_t)addr, (long)pte)); 955 while (page < old_pgd + (old_count * PAGE_SIZE) && page < map_end) { 956 addr = xpmap_ptom(((u_long) pte) - KERNBASE); 957 XENPRINTK(("addr 0x%" PRIx64 " pte 0x%lx *pte 0x%" PRIx64 "\n", 958 (int64_t)addr, (long)pte, (int64_t)*pte)); 959 xpq_queue_pte_update(addr, *pte | PG_RW); 960 page += PAGE_SIZE; 961 /* 962 * Our ptes are contiguous 963 * so it's safe to just "++" here 964 */ 965 pte++; 966 } 967 xpq_flush_queue(); 968} 969 970 971/* 972 * Bootstrap helper functions 973 */ 974 975/* 976 * Mark a page readonly 977 * XXX: assuming vaddr = paddr + KERNBASE 978 */ 979 980static void 981xen_bt_set_readonly (vaddr_t page) 982{ 983 pt_entry_t entry; 984 985 entry = xpmap_ptom_masked(page - KERNBASE); 986 entry |= PG_k | PG_V; 987 988 HYPERVISOR_update_va_mapping (page, entry, UVMF_INVLPG); 989} 990 991#ifdef __x86_64__ 992void 993xen_set_user_pgd(paddr_t page) 994{ 995 struct mmuext_op op; 996 int s = splvm(); 997 998 xpq_flush_queue(); 999 op.cmd = MMUEXT_NEW_USER_BASEPTR; 1000 op.arg1.mfn = xpmap_phys_to_machine_mapping[page >> PAGE_SHIFT]; 1001 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 1002 panic("xen_set_user_pgd: failed to install new user page" 1003 " directory %lx", page); 1004 splx(s); 1005} 1006#endif /* __x86_64__ */ 1007