x86_xpmap.c revision 1.61
1/* $NetBSD: x86_xpmap.c,v 1.61 2016/08/25 17:03:57 bouyer 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 1.61 2016/08/25 17:03:57 bouyer 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/mutex.h> 81#include <sys/cpu.h> 82 83#include <uvm/uvm.h> 84 85#include <x86/pmap.h> 86#include <machine/gdt.h> 87#include <xen/xenfunc.h> 88 89#include <dev/isa/isareg.h> 90#include <machine/isa_machdep.h> 91 92#undef XENDEBUG 93/* #define XENDEBUG_SYNC */ 94/* #define XENDEBUG_LOW */ 95 96#ifdef XENDEBUG 97#define XENPRINTF(x) printf x 98#define XENPRINTK(x) printk x 99#define XENPRINTK2(x) /* printk x */ 100 101static char XBUF[256]; 102#else 103#define XENPRINTF(x) 104#define XENPRINTK(x) 105#define XENPRINTK2(x) 106#endif 107#define PRINTF(x) printf x 108#define PRINTK(x) printk x 109 110volatile shared_info_t *HYPERVISOR_shared_info; 111/* Xen requires the start_info struct to be page aligned */ 112union start_info_union start_info_union __aligned(PAGE_SIZE); 113unsigned long *xpmap_phys_to_machine_mapping; 114kmutex_t pte_lock; 115 116void xen_failsafe_handler(void); 117 118#define HYPERVISOR_mmu_update_self(req, count, success_count) \ 119 HYPERVISOR_mmu_update((req), (count), (success_count), DOMID_SELF) 120 121/* 122 * kcpuset internally uses an array of uint32_t while xen uses an array of 123 * u_long. As we're little-endian we can cast one to the other. 124 */ 125typedef union { 126#ifdef _LP64 127 uint32_t xcpum_km[2]; 128#else 129 uint32_t xcpum_km[1]; 130#endif 131 u_long xcpum_xm; 132} xcpumask_t; 133 134void 135xen_failsafe_handler(void) 136{ 137 138 panic("xen_failsafe_handler called!\n"); 139} 140 141 142void 143xen_set_ldt(vaddr_t base, uint32_t entries) 144{ 145 vaddr_t va; 146 vaddr_t end; 147 pt_entry_t *ptp; 148 int s; 149 150#ifdef __x86_64__ 151 end = base + (entries << 3); 152#else 153 end = base + entries * sizeof(union descriptor); 154#endif 155 156 for (va = base; va < end; va += PAGE_SIZE) { 157 KASSERT(va >= VM_MIN_KERNEL_ADDRESS); 158 ptp = kvtopte(va); 159 XENPRINTF(("xen_set_ldt %#" PRIxVADDR " %d %p\n", 160 base, entries, ptp)); 161 pmap_pte_clearbits(ptp, PG_RW); 162 } 163 s = splvm(); 164 xpq_queue_set_ldt(base, entries); 165 splx(s); 166} 167 168#ifdef XENDEBUG 169void xpq_debug_dump(void); 170#endif 171 172#define XPQUEUE_SIZE 2048 173static mmu_update_t xpq_queue_array[MAXCPUS][XPQUEUE_SIZE]; 174static int xpq_idx_array[MAXCPUS]; 175 176#ifdef i386 177extern union descriptor tmpgdt[]; 178#endif /* i386 */ 179void 180xpq_flush_queue(void) 181{ 182 int i, ok = 0, ret; 183 184 mmu_update_t *xpq_queue = xpq_queue_array[curcpu()->ci_cpuid]; 185 int xpq_idx = xpq_idx_array[curcpu()->ci_cpuid]; 186 187 XENPRINTK2(("flush queue %p entries %d\n", xpq_queue, xpq_idx)); 188 for (i = 0; i < xpq_idx; i++) 189 XENPRINTK2(("%d: 0x%08" PRIx64 " 0x%08" PRIx64 "\n", i, 190 xpq_queue[i].ptr, xpq_queue[i].val)); 191 192retry: 193 ret = HYPERVISOR_mmu_update_self(xpq_queue, xpq_idx, &ok); 194 195 if (xpq_idx != 0 && ret < 0) { 196 struct cpu_info *ci; 197 CPU_INFO_ITERATOR cii; 198 199 printf("xpq_flush_queue: %d entries (%d successful) on " 200 "cpu%d (%ld)\n", 201 xpq_idx, ok, curcpu()->ci_index, curcpu()->ci_cpuid); 202 203 if (ok != 0) { 204 xpq_queue += ok; 205 xpq_idx -= ok; 206 ok = 0; 207 goto retry; 208 } 209 210 for (CPU_INFO_FOREACH(cii, ci)) { 211 xpq_queue = xpq_queue_array[ci->ci_cpuid]; 212 xpq_idx = xpq_idx_array[ci->ci_cpuid]; 213 printf("cpu%d (%ld):\n", ci->ci_index, ci->ci_cpuid); 214 for (i = 0; i < xpq_idx; i++) { 215 printf(" 0x%016" PRIx64 ": 0x%016" PRIx64 "\n", 216 xpq_queue[i].ptr, xpq_queue[i].val); 217 } 218#ifdef __x86_64__ 219 for (i = 0; i < PDIR_SLOT_PTE; i++) { 220 if (ci->ci_kpm_pdir[i] == 0) 221 continue; 222 printf(" kpm_pdir[%d]: 0x%" PRIx64 "\n", 223 i, ci->ci_kpm_pdir[i]); 224 } 225#endif 226 } 227 panic("HYPERVISOR_mmu_update failed, ret: %d\n", ret); 228 } 229 xpq_idx_array[curcpu()->ci_cpuid] = 0; 230} 231 232static inline void 233xpq_increment_idx(void) 234{ 235 236 if (__predict_false(++xpq_idx_array[curcpu()->ci_cpuid] == XPQUEUE_SIZE)) 237 xpq_flush_queue(); 238} 239 240void 241xpq_queue_machphys_update(paddr_t ma, paddr_t pa) 242{ 243 244 mmu_update_t *xpq_queue = xpq_queue_array[curcpu()->ci_cpuid]; 245 int xpq_idx = xpq_idx_array[curcpu()->ci_cpuid]; 246 247 XENPRINTK2(("xpq_queue_machphys_update ma=0x%" PRIx64 " pa=0x%" PRIx64 248 "\n", (int64_t)ma, (int64_t)pa)); 249 250 xpq_queue[xpq_idx].ptr = ma | MMU_MACHPHYS_UPDATE; 251 xpq_queue[xpq_idx].val = pa >> PAGE_SHIFT; 252 xpq_increment_idx(); 253#ifdef XENDEBUG_SYNC 254 xpq_flush_queue(); 255#endif 256} 257 258void 259xpq_queue_pte_update(paddr_t ptr, pt_entry_t val) 260{ 261 262 mmu_update_t *xpq_queue = xpq_queue_array[curcpu()->ci_cpuid]; 263 int xpq_idx = xpq_idx_array[curcpu()->ci_cpuid]; 264 265 KASSERT((ptr & 3) == 0); 266 xpq_queue[xpq_idx].ptr = (paddr_t)ptr | MMU_NORMAL_PT_UPDATE; 267 xpq_queue[xpq_idx].val = val; 268 xpq_increment_idx(); 269#ifdef XENDEBUG_SYNC 270 xpq_flush_queue(); 271#endif 272} 273 274void 275xpq_queue_pt_switch(paddr_t pa) 276{ 277 struct mmuext_op op; 278 xpq_flush_queue(); 279 280 XENPRINTK2(("xpq_queue_pt_switch: 0x%" PRIx64 " 0x%" PRIx64 "\n", 281 (int64_t)pa, (int64_t)pa)); 282 op.cmd = MMUEXT_NEW_BASEPTR; 283 op.arg1.mfn = pa >> PAGE_SHIFT; 284 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 285 panic("xpq_queue_pt_switch"); 286} 287 288void 289xpq_queue_pin_table(paddr_t pa, int lvl) 290{ 291 struct mmuext_op op; 292 293 xpq_flush_queue(); 294 295 XENPRINTK2(("xpq_queue_pin_l%d_table: %#" PRIxPADDR "\n", 296 lvl + 1, pa)); 297 298 op.arg1.mfn = pa >> PAGE_SHIFT; 299 op.cmd = lvl; 300 301 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 302 panic("xpq_queue_pin_table"); 303} 304 305void 306xpq_queue_unpin_table(paddr_t pa) 307{ 308 struct mmuext_op op; 309 310 xpq_flush_queue(); 311 312 XENPRINTK2(("xpq_queue_unpin_table: %#" PRIxPADDR "\n", pa)); 313 op.arg1.mfn = pa >> PAGE_SHIFT; 314 op.cmd = MMUEXT_UNPIN_TABLE; 315 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 316 panic("xpq_queue_unpin_table"); 317} 318 319void 320xpq_queue_set_ldt(vaddr_t va, uint32_t entries) 321{ 322 struct mmuext_op op; 323 324 xpq_flush_queue(); 325 326 XENPRINTK2(("xpq_queue_set_ldt\n")); 327 KASSERT(va == (va & ~PAGE_MASK)); 328 op.cmd = MMUEXT_SET_LDT; 329 op.arg1.linear_addr = va; 330 op.arg2.nr_ents = entries; 331 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 332 panic("xpq_queue_set_ldt"); 333} 334 335void 336xpq_queue_tlb_flush(void) 337{ 338 struct mmuext_op op; 339 340 xpq_flush_queue(); 341 342 XENPRINTK2(("xpq_queue_tlb_flush\n")); 343 op.cmd = MMUEXT_TLB_FLUSH_LOCAL; 344 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 345 panic("xpq_queue_tlb_flush"); 346} 347 348void 349xpq_flush_cache(void) 350{ 351 int s = splvm(); 352 353 xpq_flush_queue(); 354 355 XENPRINTK2(("xpq_queue_flush_cache\n")); 356 asm("wbinvd":::"memory"); 357 splx(s); /* XXX: removeme */ 358} 359 360void 361xpq_queue_invlpg(vaddr_t va) 362{ 363 struct mmuext_op op; 364 xpq_flush_queue(); 365 366 XENPRINTK2(("xpq_queue_invlpg %#" PRIxVADDR "\n", va)); 367 op.cmd = MMUEXT_INVLPG_LOCAL; 368 op.arg1.linear_addr = (va & ~PAGE_MASK); 369 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 370 panic("xpq_queue_invlpg"); 371} 372 373void 374xen_mcast_invlpg(vaddr_t va, kcpuset_t *kc) 375{ 376 xcpumask_t xcpumask; 377 mmuext_op_t op; 378 379 kcpuset_export_u32(kc, &xcpumask.xcpum_km[0], sizeof(xcpumask)); 380 381 /* Flush pending page updates */ 382 xpq_flush_queue(); 383 384 op.cmd = MMUEXT_INVLPG_MULTI; 385 op.arg1.linear_addr = va; 386 op.arg2.vcpumask = &xcpumask.xcpum_xm; 387 388 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 389 panic("xpq_queue_invlpg_all"); 390 } 391 392 return; 393} 394 395void 396xen_bcast_invlpg(vaddr_t va) 397{ 398 mmuext_op_t op; 399 400 /* Flush pending page updates */ 401 xpq_flush_queue(); 402 403 op.cmd = MMUEXT_INVLPG_ALL; 404 op.arg1.linear_addr = va; 405 406 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 407 panic("xpq_queue_invlpg_all"); 408 } 409 410 return; 411} 412 413/* This is a synchronous call. */ 414void 415xen_mcast_tlbflush(kcpuset_t *kc) 416{ 417 xcpumask_t xcpumask; 418 mmuext_op_t op; 419 420 kcpuset_export_u32(kc, &xcpumask.xcpum_km[0], sizeof(xcpumask)); 421 422 /* Flush pending page updates */ 423 xpq_flush_queue(); 424 425 op.cmd = MMUEXT_TLB_FLUSH_MULTI; 426 op.arg2.vcpumask = &xcpumask.xcpum_xm; 427 428 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 429 panic("xpq_queue_invlpg_all"); 430 } 431 432 return; 433} 434 435/* This is a synchronous call. */ 436void 437xen_bcast_tlbflush(void) 438{ 439 mmuext_op_t op; 440 441 /* Flush pending page updates */ 442 xpq_flush_queue(); 443 444 op.cmd = MMUEXT_TLB_FLUSH_ALL; 445 446 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 447 panic("xpq_queue_invlpg_all"); 448 } 449 450 return; 451} 452 453/* This is a synchronous call. */ 454void 455xen_vcpu_mcast_invlpg(vaddr_t sva, vaddr_t eva, kcpuset_t *kc) 456{ 457 KASSERT(eva > sva); 458 459 /* Flush pending page updates */ 460 xpq_flush_queue(); 461 462 /* Align to nearest page boundary */ 463 sva &= ~PAGE_MASK; 464 eva &= ~PAGE_MASK; 465 466 for ( ; sva <= eva; sva += PAGE_SIZE) { 467 xen_mcast_invlpg(sva, kc); 468 } 469 470 return; 471} 472 473/* This is a synchronous call. */ 474void 475xen_vcpu_bcast_invlpg(vaddr_t sva, vaddr_t eva) 476{ 477 KASSERT(eva > sva); 478 479 /* Flush pending page updates */ 480 xpq_flush_queue(); 481 482 /* Align to nearest page boundary */ 483 sva &= ~PAGE_MASK; 484 eva &= ~PAGE_MASK; 485 486 for ( ; sva <= eva; sva += PAGE_SIZE) { 487 xen_bcast_invlpg(sva); 488 } 489 490 return; 491} 492 493/* Copy a page */ 494void 495xen_copy_page(paddr_t srcpa, paddr_t dstpa) 496{ 497 mmuext_op_t op; 498 499 op.cmd = MMUEXT_COPY_PAGE; 500 op.arg1.mfn = xpmap_ptom(dstpa) >> PAGE_SHIFT; 501 op.arg2.src_mfn = xpmap_ptom(srcpa) >> PAGE_SHIFT; 502 503 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 504 panic(__func__); 505 } 506} 507 508/* Zero a physical page */ 509void 510xen_pagezero(paddr_t pa) 511{ 512 mmuext_op_t op; 513 514 op.cmd = MMUEXT_CLEAR_PAGE; 515 op.arg1.mfn = xpmap_ptom(pa) >> PAGE_SHIFT; 516 517 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) { 518 panic(__func__); 519 } 520} 521 522int 523xpq_update_foreign(paddr_t ptr, pt_entry_t val, int dom) 524{ 525 mmu_update_t op; 526 int ok; 527 528 xpq_flush_queue(); 529 530 op.ptr = ptr; 531 op.val = val; 532 if (HYPERVISOR_mmu_update(&op, 1, &ok, dom) < 0) 533 return EFAULT; 534 return (0); 535} 536 537#ifdef XENDEBUG 538void 539xpq_debug_dump(void) 540{ 541 int i; 542 543 mmu_update_t *xpq_queue = xpq_queue_array[curcpu()->ci_cpuid]; 544 int xpq_idx = xpq_idx_array[curcpu()->ci_cpuid]; 545 546 XENPRINTK2(("idx: %d\n", xpq_idx)); 547 for (i = 0; i < xpq_idx; i++) { 548 snprintf(XBUF, sizeof(XBUF), "%" PRIx64 " %08" PRIx64, 549 xpq_queue[i].ptr, xpq_queue[i].val); 550 if (++i < xpq_idx) 551 snprintf(XBUF + strlen(XBUF), 552 sizeof(XBUF) - strlen(XBUF), 553 "%" PRIx64 " %08" PRIx64, 554 xpq_queue[i].ptr, xpq_queue[i].val); 555 if (++i < xpq_idx) 556 snprintf(XBUF + strlen(XBUF), 557 sizeof(XBUF) - strlen(XBUF), 558 "%" PRIx64 " %08" PRIx64, 559 xpq_queue[i].ptr, xpq_queue[i].val); 560 if (++i < xpq_idx) 561 snprintf(XBUF + strlen(XBUF), 562 sizeof(XBUF) - strlen(XBUF), 563 "%" PRIx64 " %08" PRIx64, 564 xpq_queue[i].ptr, xpq_queue[i].val); 565 XENPRINTK2(("%d: %s\n", xpq_idx, XBUF)); 566 } 567} 568#endif 569 570 571extern volatile struct xencons_interface *xencons_interface; /* XXX */ 572extern struct xenstore_domain_interface *xenstore_interface; /* XXX */ 573 574static void xen_bt_set_readonly(vaddr_t); 575static void xen_bootstrap_tables(vaddr_t, vaddr_t, int, int, int); 576 577/* How many PDEs ? */ 578#if L2_SLOT_KERNBASE > 0 579#define TABLE_L2_ENTRIES (2 * (NKL2_KIMG_ENTRIES + 1)) 580#else 581#define TABLE_L2_ENTRIES (NKL2_KIMG_ENTRIES + 1) 582#endif 583 584/* 585 * Construct and switch to new pagetables 586 * first_avail is the first vaddr we can use after 587 * we get rid of Xen pagetables 588 */ 589 590vaddr_t xen_pmap_bootstrap(void); 591 592/* 593 * Function to get rid of Xen bootstrap tables 594 */ 595 596/* How many PDP do we need: */ 597#ifdef PAE 598/* 599 * For PAE, we consider a single contigous L2 "superpage" of 4 pages, 600 * all of them mapped by the L3 page. We also need a shadow page 601 * for L3[3]. 602 */ 603static const int l2_4_count = 6; 604#elif defined(__x86_64__) 605static const int l2_4_count = PTP_LEVELS; 606#else 607static const int l2_4_count = PTP_LEVELS - 1; 608#endif 609 610vaddr_t 611xen_pmap_bootstrap(void) 612{ 613 int count, oldcount; 614 long mapsize; 615 vaddr_t bootstrap_tables, init_tables; 616 617 xen_init_features(); 618 619 memset(xpq_idx_array, 0, sizeof(xpq_idx_array)); 620 621 xpmap_phys_to_machine_mapping = 622 (unsigned long *)xen_start_info.mfn_list; 623 init_tables = xen_start_info.pt_base; 624 __PRINTK(("xen_arch_pmap_bootstrap init_tables=0x%lx\n", init_tables)); 625 626 /* Space after Xen boostrap tables should be free */ 627 bootstrap_tables = xen_start_info.pt_base + 628 (xen_start_info.nr_pt_frames * PAGE_SIZE); 629 630 /* 631 * Calculate how many space we need 632 * first everything mapped before the Xen bootstrap tables 633 */ 634 mapsize = init_tables - KERNTEXTOFF; 635 /* after the tables we'll have: 636 * - UAREA 637 * - dummy user PGD (x86_64) 638 * - HYPERVISOR_shared_info 639 * - early_zerop 640 * - ISA I/O mem (if needed) 641 */ 642 mapsize += UPAGES * PAGE_SIZE; 643#ifdef __x86_64__ 644 mapsize += PAGE_SIZE; 645#endif 646 mapsize += PAGE_SIZE; 647 mapsize += PAGE_SIZE; 648 649#ifdef DOM0OPS 650 if (xendomain_is_dom0()) { 651 /* space for ISA I/O mem */ 652 mapsize += IOM_SIZE; 653 } 654#endif 655 /* at this point mapsize doesn't include the table size */ 656 657#ifdef __x86_64__ 658 count = TABLE_L2_ENTRIES; 659#else 660 count = (mapsize + (NBPD_L2 -1)) >> L2_SHIFT; 661#endif /* __x86_64__ */ 662 663 /* now compute how many L2 pages we need exactly */ 664 XENPRINTK(("bootstrap_final mapsize 0x%lx count %d\n", mapsize, count)); 665 while (mapsize + (count + l2_4_count) * PAGE_SIZE + KERNTEXTOFF > 666 ((long)count << L2_SHIFT) + KERNBASE) { 667 count++; 668 } 669#ifndef __x86_64__ 670 /* 671 * one more L2 page: we'll alocate several pages after kva_start 672 * in pmap_bootstrap() before pmap_growkernel(), which have not been 673 * counted here. It's not a big issue to allocate one more L2 as 674 * pmap_growkernel() will be called anyway. 675 */ 676 count++; 677 nkptp[1] = count; 678#endif 679 680 /* 681 * install bootstrap pages. We may need more L2 pages than will 682 * have the final table here, as it's installed after the final table 683 */ 684 oldcount = count; 685 686bootstrap_again: 687 XENPRINTK(("bootstrap_again oldcount %d\n", oldcount)); 688 /* 689 * Xen space we'll reclaim may not be enough for our new page tables, 690 * move bootstrap tables if necessary 691 */ 692 if (bootstrap_tables < init_tables + ((count + l2_4_count) * PAGE_SIZE)) 693 bootstrap_tables = init_tables + 694 ((count + l2_4_count) * PAGE_SIZE); 695 /* make sure we have enough to map the bootstrap_tables */ 696 if (bootstrap_tables + ((oldcount + l2_4_count) * PAGE_SIZE) > 697 ((long)oldcount << L2_SHIFT) + KERNBASE) { 698 oldcount++; 699 goto bootstrap_again; 700 } 701 702 /* Create temporary tables */ 703 xen_bootstrap_tables(xen_start_info.pt_base, bootstrap_tables, 704 xen_start_info.nr_pt_frames, oldcount, 0); 705 706 /* Create final tables */ 707 xen_bootstrap_tables(bootstrap_tables, init_tables, 708 oldcount + l2_4_count, count, 1); 709 710 /* zero out free space after tables */ 711 memset((void *)(init_tables + ((count + l2_4_count) * PAGE_SIZE)), 0, 712 (UPAGES + 1) * PAGE_SIZE); 713 714 /* Finally, flush TLB. */ 715 xpq_queue_tlb_flush(); 716 717 return (init_tables + ((count + l2_4_count) * PAGE_SIZE)); 718} 719 720/* 721 * Build a new table and switch to it. 722 * old_count is # of old tables (including PGD, PDTPE and PDE). 723 * new_count is # of new tables (PTE only). 724 * We assume the areas don't overlap. 725 */ 726static void 727xen_bootstrap_tables(vaddr_t old_pgd, vaddr_t new_pgd, 728 int old_count, int new_count, int final) 729{ 730 pd_entry_t *pdtpe, *pde, *pte; 731 pd_entry_t *bt_pgd; 732 paddr_t addr; 733 vaddr_t page, avail, map_end; 734 int i; 735 extern char __rodata_start; 736 extern char __data_start; 737 extern char __kernel_end; 738 extern char *early_zerop; /* from pmap.c */ 739 pt_entry_t pg_nx; 740 u_int descs[4]; 741 742 __PRINTK(("xen_bootstrap_tables(%#" PRIxVADDR ", %#" PRIxVADDR "," 743 " %d, %d)\n", 744 old_pgd, new_pgd, old_count, new_count)); 745 746 /* 747 * Set the NX/XD bit, if available. descs[3] = %edx. 748 */ 749 x86_cpuid(0x80000001, descs); 750 pg_nx = (descs[3] & CPUID_NOX) ? PG_NX : 0; 751 752 /* 753 * size of R/W area after kernel text: 754 * xencons_interface (if present) 755 * xenstore_interface (if present) 756 * table pages (new_count + l2_4_count entries) 757 * extra mappings (only when final is true): 758 * UAREA 759 * dummy user PGD (x86_64 only)/gdt page (i386 only) 760 * HYPERVISOR_shared_info 761 * early_zerop 762 * ISA I/O mem (if needed) 763 */ 764 map_end = new_pgd + ((new_count + l2_4_count) * PAGE_SIZE); 765 if (final) { 766 map_end += (UPAGES + 1) * PAGE_SIZE; 767 HYPERVISOR_shared_info = (shared_info_t *)map_end; 768 map_end += PAGE_SIZE; 769 early_zerop = (char *)map_end; 770 map_end += PAGE_SIZE; 771 } 772 773 /* 774 * we always set atdevbase, as it's used by init386 to find the first 775 * available VA. map_end is updated only if we are dom0, so 776 * atdevbase -> atdevbase + IOM_SIZE will be mapped only in 777 * this case. 778 */ 779 if (final) 780 atdevbase = map_end; 781#ifdef DOM0OPS 782 if (final && xendomain_is_dom0()) { 783 /* ISA I/O mem */ 784 map_end += IOM_SIZE; 785 } 786#endif /* DOM0OPS */ 787 788 __PRINTK(("xen_bootstrap_tables map_end 0x%lx\n", map_end)); 789 __PRINTK(("console %#lx ", xen_start_info.console_mfn)); 790 __PRINTK(("xenstore %#" PRIx32 "\n", xen_start_info.store_mfn)); 791 792 /* 793 * Create bootstrap page tables. What we need: 794 * - a PGD (level 4) 795 * - a PDTPE (level 3) 796 * - a PDE (level 2) 797 * - some PTEs (level 1) 798 */ 799 800 bt_pgd = (pd_entry_t *)new_pgd; 801 memset(bt_pgd, 0, PAGE_SIZE); 802 avail = new_pgd + PAGE_SIZE; 803 804#if PTP_LEVELS > 3 805 /* per-cpu L4 PD */ 806 pd_entry_t *bt_cpu_pgd = bt_pgd; 807 /* pmap_kernel() "shadow" L4 PD */ 808 bt_pgd = (pd_entry_t *)avail; 809 memset(bt_pgd, 0, PAGE_SIZE); 810 avail += PAGE_SIZE; 811 812 /* Install level 3 */ 813 pdtpe = (pd_entry_t *)avail; 814 memset(pdtpe, 0, PAGE_SIZE); 815 avail += PAGE_SIZE; 816 817 addr = ((u_long)pdtpe) - KERNBASE; 818 bt_pgd[pl4_pi(KERNTEXTOFF)] = bt_cpu_pgd[pl4_pi(KERNTEXTOFF)] = 819 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 820 821 __PRINTK(("L3 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 822 " -> L4[%#x]\n", 823 pdtpe, addr, bt_pgd[pl4_pi(KERNTEXTOFF)], pl4_pi(KERNTEXTOFF))); 824#else 825 pdtpe = bt_pgd; 826#endif /* PTP_LEVELS > 3 */ 827 828#if PTP_LEVELS > 2 829 /* Level 2 */ 830 pde = (pd_entry_t *)avail; 831 memset(pde, 0, PAGE_SIZE); 832 avail += PAGE_SIZE; 833 834 addr = ((u_long)pde) - KERNBASE; 835 pdtpe[pl3_pi(KERNTEXTOFF)] = 836 xpmap_ptom_masked(addr) | PG_k | PG_V | PG_RW; 837 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 838 " -> L3[%#x]\n", 839 pde, addr, pdtpe[pl3_pi(KERNTEXTOFF)], pl3_pi(KERNTEXTOFF))); 840#elif defined(PAE) 841 /* our PAE-style level 2: 5 contigous pages (4 L2 + 1 shadow) */ 842 pde = (pd_entry_t *)avail; 843 memset(pde, 0, PAGE_SIZE * 5); 844 avail += PAGE_SIZE * 5; 845 addr = ((u_long)pde) - KERNBASE; 846 /* 847 * enter L2 pages in the L3. 848 * The real L2 kernel PD will be the last one (so that 849 * pde[L2_SLOT_KERN] always point to the shadow). 850 */ 851 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 852 /* 853 * Xen doesn't want R/W mappings in L3 entries, it'll add it 854 * itself. 855 */ 856 pdtpe[i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 857 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 858 " -> L3[%#x]\n", 859 (vaddr_t)pde + PAGE_SIZE * i, addr, pdtpe[i], i)); 860 } 861 addr += PAGE_SIZE; 862 pdtpe[3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 863 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 864 " -> L3[%#x]\n", 865 (vaddr_t)pde + PAGE_SIZE * 4, addr, pdtpe[3], 3)); 866#else /* PAE */ 867 pde = bt_pgd; 868#endif /* PTP_LEVELS > 2 */ 869 870 /* Level 1 */ 871 page = KERNTEXTOFF; 872 for (i = 0; i < new_count; i ++) { 873 vaddr_t cur_page = page; 874 875 pte = (pd_entry_t *)avail; 876 avail += PAGE_SIZE; 877 878 memset(pte, 0, PAGE_SIZE); 879 while (pl2_pi(page) == pl2_pi(cur_page)) { 880 if (page >= map_end) { 881 /* not mapped at all */ 882 pte[pl1_pi(page)] = 0; 883 page += PAGE_SIZE; 884 continue; 885 } 886 pte[pl1_pi(page)] = xpmap_ptom_masked(page - KERNBASE); 887 if (page == (vaddr_t)HYPERVISOR_shared_info) { 888 pte[pl1_pi(page)] = xen_start_info.shared_info; 889 __PRINTK(("HYPERVISOR_shared_info " 890 "va %#lx pte %#" PRIxPADDR "\n", 891 HYPERVISOR_shared_info, pte[pl1_pi(page)])); 892 } 893 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 894 == xen_start_info.console.domU.mfn) { 895 xencons_interface = (void *)page; 896 pte[pl1_pi(page)] = xen_start_info.console_mfn; 897 pte[pl1_pi(page)] <<= PAGE_SHIFT; 898 __PRINTK(("xencons_interface " 899 "va %#lx pte %#" PRIxPADDR "\n", 900 xencons_interface, pte[pl1_pi(page)])); 901 } 902 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 903 == xen_start_info.store_mfn) { 904 xenstore_interface = (void *)page; 905 pte[pl1_pi(page)] = xen_start_info.store_mfn; 906 pte[pl1_pi(page)] <<= PAGE_SHIFT; 907 __PRINTK(("xenstore_interface " 908 "va %#lx pte %#" PRIxPADDR "\n", 909 xenstore_interface, pte[pl1_pi(page)])); 910 } 911#ifdef DOM0OPS 912 if (page >= (vaddr_t)atdevbase && 913 page < (vaddr_t)atdevbase + IOM_SIZE) { 914 pte[pl1_pi(page)] = 915 IOM_BEGIN + (page - (vaddr_t)atdevbase); 916 pte[pl1_pi(page)] |= pg_nx; 917 } 918#endif 919 920 pte[pl1_pi(page)] |= PG_k | PG_V; 921 if (page < (vaddr_t)&__rodata_start) { 922 /* Map the kernel text RX. */ 923 pte[pl1_pi(page)] |= PG_RO; 924 } else if (page >= (vaddr_t)&__rodata_start && 925 page < (vaddr_t)&__data_start) { 926 /* Map the kernel rodata R. */ 927 pte[pl1_pi(page)] |= PG_RO | pg_nx; 928 } else if (page >= old_pgd && 929 page < old_pgd + (old_count * PAGE_SIZE)) { 930 /* Map the old page tables R. */ 931 pte[pl1_pi(page)] |= PG_RO | pg_nx; 932 } else if (page >= new_pgd && 933 page < new_pgd + ((new_count + l2_4_count) * PAGE_SIZE)) { 934 /* Map the new page tables R. */ 935 pte[pl1_pi(page)] |= PG_RO | pg_nx; 936#ifdef i386 937 } else if (page == (vaddr_t)tmpgdt) { 938 /* 939 * Map bootstrap gdt R/O. Later, we 940 * will re-add this to page to uvm 941 * after making it writable. 942 */ 943 944 pte[pl1_pi(page)] = 0; 945 page += PAGE_SIZE; 946 continue; 947#endif /* i386 */ 948 } else if (page >= (vaddr_t)&__data_start && 949 page < (vaddr_t)&__kernel_end) { 950 /* Map the kernel data+bss RW. */ 951 pte[pl1_pi(page)] |= PG_RW | pg_nx; 952 } else { 953 /* map page RW */ 954 pte[pl1_pi(page)] |= PG_RW; 955 } 956 957 if ((page >= old_pgd && page < old_pgd + (old_count * PAGE_SIZE)) 958 || page >= new_pgd) { 959 __PRINTK(("va %#lx pa %#lx " 960 "entry 0x%" PRIxPADDR " -> L1[%#x]\n", 961 page, page - KERNBASE, 962 pte[pl1_pi(page)], pl1_pi(page))); 963 } 964 page += PAGE_SIZE; 965 } 966 967 addr = ((u_long) pte) - KERNBASE; 968 pde[pl2_pi(cur_page)] = 969 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 970 __PRINTK(("L1 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 971 " -> L2[%#x]\n", 972 pte, addr, pde[pl2_pi(cur_page)], pl2_pi(cur_page))); 973 /* Mark readonly */ 974 xen_bt_set_readonly((vaddr_t) pte); 975 } 976 977 /* Install recursive page tables mapping */ 978#ifdef PAE 979 /* 980 * we need a shadow page for the kernel's L2 page 981 * The real L2 kernel PD will be the last one (so that 982 * pde[L2_SLOT_KERN] always point to the shadow. 983 */ 984 memcpy(&pde[L2_SLOT_KERN + NPDPG], &pde[L2_SLOT_KERN], PAGE_SIZE); 985 cpu_info_primary.ci_kpm_pdir = &pde[L2_SLOT_KERN + NPDPG]; 986 cpu_info_primary.ci_kpm_pdirpa = 987 (vaddr_t) cpu_info_primary.ci_kpm_pdir - KERNBASE; 988 989 /* 990 * We don't enter a recursive entry from the L3 PD. Instead, 991 * we enter the first 4 L2 pages, which includes the kernel's L2 992 * shadow. But we have to entrer the shadow after switching 993 * %cr3, or Xen will refcount some PTE with the wrong type. 994 */ 995 addr = (u_long)pde - KERNBASE; 996 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 997 pde[PDIR_SLOT_PTE + i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 998 __PRINTK(("pde[%d] va %#" PRIxVADDR " pa %#" PRIxPADDR 999 " entry %#" PRIxPADDR "\n", 1000 (int)(PDIR_SLOT_PTE + i), pde + PAGE_SIZE * i, 1001 addr, pde[PDIR_SLOT_PTE + i])); 1002 } 1003#if 0 1004 addr += PAGE_SIZE; /* point to shadow L2 */ 1005 pde[PDIR_SLOT_PTE + 3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 1006 __PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n", 1007 (int)(PDIR_SLOT_PTE + 3), pde + PAGE_SIZE * 4, (long)addr, 1008 (int64_t)pde[PDIR_SLOT_PTE + 3])); 1009#endif 1010 /* Mark tables RO, and pin the kernel's shadow as L2 */ 1011 addr = (u_long)pde - KERNBASE; 1012 for (i = 0; i < 5; i++, addr += PAGE_SIZE) { 1013 xen_bt_set_readonly(((vaddr_t)pde) + PAGE_SIZE * i); 1014 if (i == 2 || i == 3) 1015 continue; 1016#if 0 1017 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", i, (int64_t)addr)); 1018 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 1019#endif 1020 } 1021 if (final) { 1022 addr = (u_long)pde - KERNBASE + 3 * PAGE_SIZE; 1023 __PRINTK(("pin L2 %d addr %#" PRIxPADDR "\n", 2, addr)); 1024 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 1025 } 1026#if 0 1027 addr = (u_long)pde - KERNBASE + 2 * PAGE_SIZE; 1028 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr)); 1029 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 1030#endif 1031#else /* PAE */ 1032 1033 /* Recursive entry in pmap_kernel(). */ 1034 bt_pgd[PDIR_SLOT_PTE] = xpmap_ptom_masked((paddr_t)bt_pgd - KERNBASE) 1035 | PG_k | PG_RO | PG_V | pg_nx; 1036#ifdef __x86_64__ 1037 /* Recursive entry in higher-level per-cpu PD. */ 1038 bt_cpu_pgd[PDIR_SLOT_PTE] = xpmap_ptom_masked((paddr_t)bt_cpu_pgd - KERNBASE) 1039 | PG_k | PG_RO | PG_V | pg_nx; 1040#endif 1041 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] va %#" PRIxVADDR " pa %#" PRIxPADDR 1042 " entry %#" PRIxPADDR "\n", new_pgd, (paddr_t)new_pgd - KERNBASE, 1043 bt_pgd[PDIR_SLOT_PTE])); 1044 1045 /* Mark tables RO */ 1046 xen_bt_set_readonly((vaddr_t) pde); 1047#endif 1048#if PTP_LEVELS > 2 || defined(PAE) 1049 xen_bt_set_readonly((vaddr_t) pdtpe); 1050#endif 1051#if PTP_LEVELS > 3 1052 xen_bt_set_readonly(new_pgd); 1053#endif 1054 1055 /* Pin the PGD */ 1056 __PRINTK(("pin PGD: %"PRIxVADDR"\n", new_pgd - KERNBASE)); 1057#ifdef __x86_64__ 1058 xpq_queue_pin_l4_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 1059#elif PAE 1060 xpq_queue_pin_l3_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 1061#else 1062 xpq_queue_pin_l2_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 1063#endif 1064 1065 /* Save phys. addr of PDP, for libkvm. */ 1066#ifdef PAE 1067 PDPpaddr = (u_long)pde - KERNBASE; /* PDP is the L2 with PAE */ 1068#else 1069 PDPpaddr = (u_long)bt_pgd - KERNBASE; 1070#endif 1071 1072 /* Switch to new tables */ 1073 __PRINTK(("switch to PGD\n")); 1074 xpq_queue_pt_switch(xpmap_ptom_masked(new_pgd - KERNBASE)); 1075 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] now entry %#" PRIxPADDR "\n", 1076 bt_pgd[PDIR_SLOT_PTE])); 1077 1078#ifdef PAE 1079 if (final) { 1080 /* save the address of the L3 page */ 1081 cpu_info_primary.ci_pae_l3_pdir = pdtpe; 1082 cpu_info_primary.ci_pae_l3_pdirpa = (new_pgd - KERNBASE); 1083 1084 /* now enter kernel's PTE mappings */ 1085 addr = (u_long)pde - KERNBASE + PAGE_SIZE * 3; 1086 xpq_queue_pte_update( 1087 xpmap_ptom(((vaddr_t)&pde[PDIR_SLOT_PTE + 3]) - KERNBASE), 1088 xpmap_ptom_masked(addr) | PG_k | PG_V); 1089 xpq_flush_queue(); 1090 } 1091#elif defined(__x86_64__) 1092 if (final) { 1093 /* save the address of the real per-cpu L4 pgd page */ 1094 cpu_info_primary.ci_kpm_pdir = bt_cpu_pgd; 1095 cpu_info_primary.ci_kpm_pdirpa = ((paddr_t) bt_cpu_pgd - KERNBASE); 1096 } 1097#endif 1098 __USE(pdtpe); 1099 1100 /* Now we can safely reclaim space taken by old tables */ 1101 1102 __PRINTK(("unpin old PGD\n")); 1103 /* Unpin old PGD */ 1104 xpq_queue_unpin_table(xpmap_ptom_masked(old_pgd - KERNBASE)); 1105 /* Mark old tables RW */ 1106 page = old_pgd; 1107 addr = (paddr_t) pde[pl2_pi(page)] & PG_FRAME; 1108 addr = xpmap_mtop(addr); 1109 pte = (pd_entry_t *) ((u_long)addr + KERNBASE); 1110 pte += pl1_pi(page); 1111 __PRINTK(("*pde %#" PRIxPADDR " addr %#" PRIxPADDR " pte %#lx\n", 1112 pde[pl2_pi(page)], addr, (long)pte)); 1113 while (page < old_pgd + (old_count * PAGE_SIZE) && page < map_end) { 1114 addr = xpmap_ptom(((u_long) pte) - KERNBASE); 1115 XENPRINTK(("addr %#" PRIxPADDR " pte %#lx " 1116 "*pte %#" PRIxPADDR "\n", 1117 addr, (long)pte, *pte)); 1118 xpq_queue_pte_update(addr, *pte | PG_RW); 1119 page += PAGE_SIZE; 1120 /* 1121 * Our PTEs are contiguous so it's safe to just "++" here. 1122 */ 1123 pte++; 1124 } 1125 xpq_flush_queue(); 1126} 1127 1128 1129/* 1130 * Bootstrap helper functions 1131 */ 1132 1133/* 1134 * Mark a page readonly 1135 * XXX: assuming vaddr = paddr + KERNBASE 1136 */ 1137 1138static void 1139xen_bt_set_readonly (vaddr_t page) 1140{ 1141 pt_entry_t entry; 1142 1143 entry = xpmap_ptom_masked(page - KERNBASE); 1144 entry |= PG_k | PG_V; 1145 1146 HYPERVISOR_update_va_mapping (page, entry, UVMF_INVLPG); 1147} 1148 1149#ifdef __x86_64__ 1150void 1151xen_set_user_pgd(paddr_t page) 1152{ 1153 struct mmuext_op op; 1154 int s = splvm(); 1155 1156 xpq_flush_queue(); 1157 op.cmd = MMUEXT_NEW_USER_BASEPTR; 1158 op.arg1.mfn = xpmap_ptom_masked(page) >> PAGE_SHIFT; 1159 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 1160 panic("xen_set_user_pgd: failed to install new user page" 1161 " directory %#" PRIxPADDR, page); 1162 splx(s); 1163} 1164#endif /* __x86_64__ */ 1165