1 /* $NetBSD: x86_xpmap.c,v 1.53 2014/05/06 04:26:24 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 1.53 2014/05/06 04:26:24 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/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 */ 100 101 static 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 109 110 volatile shared_info_t *HYPERVISOR_shared_info; 111 /* Xen requires the start_info struct to be page aligned */ 112 union start_info_union start_info_union __aligned(PAGE_SIZE); 113 unsigned long *xpmap_phys_to_machine_mapping; 114 kmutex_t pte_lock; 115 116 void 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 */ 125 typedef 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; 133 134 void 135 xen_failsafe_handler(void) 136 { 137 138 panic("xen_failsafe_handler called!\n"); 139 } 140 141 142 void 143 xen_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 169 void xpq_debug_dump(void); 170 #endif 171 172 #define XPQUEUE_SIZE 2048 173 static mmu_update_t xpq_queue_array[MAXCPUS][XPQUEUE_SIZE]; 174 static int xpq_idx_array[MAXCPUS]; 175 176 #ifdef i386 177 extern union descriptor tmpgdt[]; 178 #endif /* i386 */ 179 void 180 xpq_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)); 191 192 retry: 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 } 231 232 static inline void 233 xpq_increment_idx(void) 234 { 235 236 if (__predict_false(++xpq_idx_array[curcpu()->ci_cpuid] == XPQUEUE_SIZE)) 237 xpq_flush_queue(); 238 } 239 240 void 241 xpq_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 } 257 258 void 259 xpq_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 } 273 274 void 275 xpq_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 } 287 288 void 289 xpq_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 } 304 305 void 306 xpq_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 } 318 319 void 320 xpq_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 } 334 335 void 336 xpq_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 } 347 348 void 349 xpq_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 } 359 360 void 361 xpq_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 } 372 373 void 374 xen_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 } 394 395 void 396 xen_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. */ 414 void 415 xen_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. */ 436 void 437 xen_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. */ 454 void 455 xen_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. */ 474 void 475 xen_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 */ 494 void 495 xen_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 */ 509 void 510 xen_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 } 521 522 int 523 xpq_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 538 void 539 xpq_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 569 570 571 extern volatile struct xencons_interface *xencons_interface; /* XXX */ 572 extern struct xenstore_domain_interface *xenstore_interface; /* XXX */ 573 574 static void xen_bt_set_readonly (vaddr_t); 575 static 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 */ 589 590 vaddr_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 */ 603 static const int l2_4_count = 6; 604 #elif defined(__x86_64__) 605 static const int l2_4_count = PTP_LEVELS; 606 #else 607 static const int l2_4_count = PTP_LEVELS - 1; 608 #endif 609 610 vaddr_t 611 xen_pmap_bootstrap(void) 612 { 613 int count, oldcount; 614 long mapsize; 615 vaddr_t bootstrap_tables, init_tables; 616 617 memset(xpq_idx_array, 0, sizeof xpq_idx_array); 618 619 xpmap_phys_to_machine_mapping = 620 (unsigned long *)xen_start_info.mfn_list; 621 init_tables = xen_start_info.pt_base; 622 __PRINTK(("xen_arch_pmap_bootstrap init_tables=0x%lx\n", init_tables)); 623 624 /* Space after Xen boostrap tables should be free */ 625 bootstrap_tables = xen_start_info.pt_base + 626 (xen_start_info.nr_pt_frames * PAGE_SIZE); 627 628 /* 629 * Calculate how many space we need 630 * first everything mapped before the Xen bootstrap tables 631 */ 632 mapsize = init_tables - KERNTEXTOFF; 633 /* after the tables we'll have: 634 * - UAREA 635 * - dummy user PGD (x86_64) 636 * - HYPERVISOR_shared_info 637 * - early_zerop 638 * - ISA I/O mem (if needed) 639 */ 640 mapsize += UPAGES * NBPG; 641 #ifdef __x86_64__ 642 mapsize += NBPG; 643 #endif 644 mapsize += NBPG; 645 mapsize += NBPG; 646 647 #ifdef DOM0OPS 648 if (xendomain_is_dom0()) { 649 /* space for ISA I/O mem */ 650 mapsize += IOM_SIZE; 651 } 652 #endif 653 /* at this point mapsize doens't include the table size */ 654 655 #ifdef __x86_64__ 656 count = TABLE_L2_ENTRIES; 657 #else 658 count = (mapsize + (NBPD_L2 -1)) >> L2_SHIFT; 659 #endif /* __x86_64__ */ 660 661 /* now compute how many L2 pages we need exactly */ 662 XENPRINTK(("bootstrap_final mapsize 0x%lx count %d\n", mapsize, count)); 663 while (mapsize + (count + l2_4_count) * PAGE_SIZE + KERNTEXTOFF > 664 ((long)count << L2_SHIFT) + KERNBASE) { 665 count++; 666 } 667 #ifndef __x86_64__ 668 /* 669 * one more L2 page: we'll alocate several pages after kva_start 670 * in pmap_bootstrap() before pmap_growkernel(), which have not been 671 * counted here. It's not a big issue to allocate one more L2 as 672 * pmap_growkernel() will be called anyway. 673 */ 674 count++; 675 nkptp[1] = count; 676 #endif 677 678 /* 679 * install bootstrap pages. We may need more L2 pages than will 680 * have the final table here, as it's installed after the final table 681 */ 682 oldcount = count; 683 684 bootstrap_again: 685 XENPRINTK(("bootstrap_again oldcount %d\n", oldcount)); 686 /* 687 * Xen space we'll reclaim may not be enough for our new page tables, 688 * move bootstrap tables if necessary 689 */ 690 if (bootstrap_tables < init_tables + ((count + l2_4_count) * PAGE_SIZE)) 691 bootstrap_tables = init_tables + 692 ((count + l2_4_count) * PAGE_SIZE); 693 /* make sure we have enough to map the bootstrap_tables */ 694 if (bootstrap_tables + ((oldcount + l2_4_count) * PAGE_SIZE) > 695 ((long)oldcount << L2_SHIFT) + KERNBASE) { 696 oldcount++; 697 goto bootstrap_again; 698 } 699 700 /* Create temporary tables */ 701 xen_bootstrap_tables(xen_start_info.pt_base, bootstrap_tables, 702 xen_start_info.nr_pt_frames, oldcount, 0); 703 704 /* Create final tables */ 705 xen_bootstrap_tables(bootstrap_tables, init_tables, 706 oldcount + l2_4_count, count, 1); 707 708 /* zero out free space after tables */ 709 memset((void *)(init_tables + ((count + l2_4_count) * PAGE_SIZE)), 0, 710 (UPAGES + 1) * NBPG); 711 712 /* Finally, flush TLB. */ 713 xpq_queue_tlb_flush(); 714 715 return (init_tables + ((count + l2_4_count) * PAGE_SIZE)); 716 } 717 718 /* 719 * Build a new table and switch to it 720 * old_count is # of old tables (including PGD, PDTPE and PDE) 721 * new_count is # of new tables (PTE only) 722 * we assume areas don't overlap 723 */ 724 static void 725 xen_bootstrap_tables (vaddr_t old_pgd, vaddr_t new_pgd, 726 int old_count, int new_count, int final) 727 { 728 pd_entry_t *pdtpe, *pde, *pte; 729 pd_entry_t *bt_pgd; 730 paddr_t addr; 731 vaddr_t page, avail, text_end, map_end; 732 int i; 733 extern char __data_start; 734 extern char *early_zerop; /* from pmap.c */ 735 736 __PRINTK(("xen_bootstrap_tables(%#" PRIxVADDR ", %#" PRIxVADDR "," 737 " %d, %d)\n", 738 old_pgd, new_pgd, old_count, new_count)); 739 text_end = ((vaddr_t)&__data_start) & ~PAGE_MASK; 740 /* 741 * size of R/W area after kernel text: 742 * xencons_interface (if present) 743 * xenstore_interface (if present) 744 * table pages (new_count + l2_4_count entries) 745 * extra mappings (only when final is true): 746 * UAREA 747 * dummy user PGD (x86_64 only)/gdt page (i386 only) 748 * HYPERVISOR_shared_info 749 * early_zerop 750 * ISA I/O mem (if needed) 751 */ 752 map_end = new_pgd + ((new_count + l2_4_count) * NBPG); 753 if (final) { 754 map_end += (UPAGES + 1) * NBPG; 755 HYPERVISOR_shared_info = (shared_info_t *)map_end; 756 map_end += NBPG; 757 early_zerop = (char *)map_end; 758 map_end += NBPG; 759 } 760 /* 761 * we always set atdevbase, as it's used by init386 to find the first 762 * available VA. map_end is updated only if we are dom0, so 763 * atdevbase -> atdevbase + IOM_SIZE will be mapped only in 764 * this case. 765 */ 766 if (final) 767 atdevbase = map_end; 768 #ifdef DOM0OPS 769 if (final && xendomain_is_dom0()) { 770 /* ISA I/O mem */ 771 map_end += IOM_SIZE; 772 } 773 #endif /* DOM0OPS */ 774 775 __PRINTK(("xen_bootstrap_tables text_end 0x%lx map_end 0x%lx\n", 776 text_end, map_end)); 777 __PRINTK(("console %#lx ", xen_start_info.console_mfn)); 778 __PRINTK(("xenstore %#" PRIx32 "\n", xen_start_info.store_mfn)); 779 780 /* 781 * Create bootstrap page tables 782 * What we need: 783 * - a PGD (level 4) 784 * - a PDTPE (level 3) 785 * - a PDE (level2) 786 * - some PTEs (level 1) 787 */ 788 789 bt_pgd = (pd_entry_t *) new_pgd; 790 memset (bt_pgd, 0, PAGE_SIZE); 791 avail = new_pgd + PAGE_SIZE; 792 #if PTP_LEVELS > 3 793 /* per-cpu L4 PD */ 794 pd_entry_t *bt_cpu_pgd = bt_pgd; 795 /* pmap_kernel() "shadow" L4 PD */ 796 bt_pgd = (pd_entry_t *) avail; 797 memset(bt_pgd, 0, PAGE_SIZE); 798 avail += PAGE_SIZE; 799 800 /* Install level 3 */ 801 pdtpe = (pd_entry_t *) avail; 802 memset (pdtpe, 0, PAGE_SIZE); 803 avail += PAGE_SIZE; 804 805 addr = ((u_long) pdtpe) - KERNBASE; 806 bt_pgd[pl4_pi(KERNTEXTOFF)] = bt_cpu_pgd[pl4_pi(KERNTEXTOFF)] = 807 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 808 809 __PRINTK(("L3 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 810 " -> L4[%#x]\n", 811 pdtpe, addr, bt_pgd[pl4_pi(KERNTEXTOFF)], pl4_pi(KERNTEXTOFF))); 812 #else 813 pdtpe = bt_pgd; 814 #endif /* PTP_LEVELS > 3 */ 815 816 #if PTP_LEVELS > 2 817 /* Level 2 */ 818 pde = (pd_entry_t *) avail; 819 memset(pde, 0, PAGE_SIZE); 820 avail += PAGE_SIZE; 821 822 addr = ((u_long) pde) - KERNBASE; 823 pdtpe[pl3_pi(KERNTEXTOFF)] = 824 xpmap_ptom_masked(addr) | PG_k | PG_V | PG_RW; 825 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 826 " -> L3[%#x]\n", 827 pde, addr, pdtpe[pl3_pi(KERNTEXTOFF)], pl3_pi(KERNTEXTOFF))); 828 #elif defined(PAE) 829 /* our PAE-style level 2: 5 contigous pages (4 L2 + 1 shadow) */ 830 pde = (pd_entry_t *) avail; 831 memset(pde, 0, PAGE_SIZE * 5); 832 avail += PAGE_SIZE * 5; 833 addr = ((u_long) pde) - KERNBASE; 834 /* 835 * enter L2 pages in the L3. 836 * The real L2 kernel PD will be the last one (so that 837 * pde[L2_SLOT_KERN] always point to the shadow). 838 */ 839 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 840 /* 841 * Xen doesn't want R/W mappings in L3 entries, it'll add it 842 * itself. 843 */ 844 pdtpe[i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 845 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 846 " -> L3[%#x]\n", 847 (vaddr_t)pde + PAGE_SIZE * i, addr, pdtpe[i], i)); 848 } 849 addr += PAGE_SIZE; 850 pdtpe[3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 851 __PRINTK(("L2 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 852 " -> L3[%#x]\n", 853 (vaddr_t)pde + PAGE_SIZE * 4, addr, pdtpe[3], 3)); 854 855 #else /* PAE */ 856 pde = bt_pgd; 857 #endif /* PTP_LEVELS > 2 */ 858 859 /* Level 1 */ 860 page = KERNTEXTOFF; 861 for (i = 0; i < new_count; i ++) { 862 vaddr_t cur_page = page; 863 864 pte = (pd_entry_t *) avail; 865 avail += PAGE_SIZE; 866 867 memset(pte, 0, PAGE_SIZE); 868 while (pl2_pi(page) == pl2_pi (cur_page)) { 869 if (page >= map_end) { 870 /* not mapped at all */ 871 pte[pl1_pi(page)] = 0; 872 page += PAGE_SIZE; 873 continue; 874 } 875 pte[pl1_pi(page)] = xpmap_ptom_masked(page - KERNBASE); 876 if (page == (vaddr_t)HYPERVISOR_shared_info) { 877 pte[pl1_pi(page)] = xen_start_info.shared_info; 878 __PRINTK(("HYPERVISOR_shared_info " 879 "va %#lx pte %#" PRIxPADDR "\n", 880 HYPERVISOR_shared_info, pte[pl1_pi(page)])); 881 } 882 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 883 == xen_start_info.console.domU.mfn) { 884 xencons_interface = (void *)page; 885 pte[pl1_pi(page)] = xen_start_info.console_mfn; 886 pte[pl1_pi(page)] <<= PAGE_SHIFT; 887 __PRINTK(("xencons_interface " 888 "va %#lx pte %#" PRIxPADDR "\n", 889 xencons_interface, pte[pl1_pi(page)])); 890 } 891 if ((xpmap_ptom_masked(page - KERNBASE) >> PAGE_SHIFT) 892 == xen_start_info.store_mfn) { 893 xenstore_interface = (void *)page; 894 pte[pl1_pi(page)] = xen_start_info.store_mfn; 895 pte[pl1_pi(page)] <<= PAGE_SHIFT; 896 __PRINTK(("xenstore_interface " 897 "va %#lx pte %#" PRIxPADDR "\n", 898 xenstore_interface, pte[pl1_pi(page)])); 899 } 900 #ifdef DOM0OPS 901 if (page >= (vaddr_t)atdevbase && 902 page < (vaddr_t)atdevbase + IOM_SIZE) { 903 pte[pl1_pi(page)] = 904 IOM_BEGIN + (page - (vaddr_t)atdevbase); 905 } 906 #endif 907 pte[pl1_pi(page)] |= PG_k | PG_V; 908 if (page < text_end) { 909 /* map kernel text RO */ 910 pte[pl1_pi(page)] |= 0; 911 } else if (page >= old_pgd 912 && page < old_pgd + (old_count * PAGE_SIZE)) { 913 /* map old page tables RO */ 914 pte[pl1_pi(page)] |= 0; 915 } else if (page >= new_pgd && 916 page < new_pgd + ((new_count + l2_4_count) * PAGE_SIZE)) { 917 /* map new page tables RO */ 918 pte[pl1_pi(page)] |= 0; 919 #ifdef i386 920 } else if (page == (vaddr_t)tmpgdt) { 921 /* 922 * Map bootstrap gdt R/O. Later, we 923 * will re-add this to page to uvm 924 * after making it writable. 925 */ 926 927 pte[pl1_pi(page)] = 0; 928 page += PAGE_SIZE; 929 continue; 930 #endif /* i386 */ 931 } else { 932 /* map page RW */ 933 pte[pl1_pi(page)] |= PG_RW; 934 } 935 936 if ((page >= old_pgd && page < old_pgd + (old_count * PAGE_SIZE)) 937 || page >= new_pgd) { 938 __PRINTK(("va %#lx pa %#lx " 939 "entry 0x%" PRIxPADDR " -> L1[%#x]\n", 940 page, page - KERNBASE, 941 pte[pl1_pi(page)], pl1_pi(page))); 942 } 943 page += PAGE_SIZE; 944 } 945 946 addr = ((u_long) pte) - KERNBASE; 947 pde[pl2_pi(cur_page)] = 948 xpmap_ptom_masked(addr) | PG_k | PG_RW | PG_V; 949 __PRINTK(("L1 va %#lx pa %#" PRIxPADDR " entry %#" PRIxPADDR 950 " -> L2[%#x]\n", 951 pte, addr, pde[pl2_pi(cur_page)], pl2_pi(cur_page))); 952 /* Mark readonly */ 953 xen_bt_set_readonly((vaddr_t) pte); 954 } 955 956 /* Install recursive page tables mapping */ 957 #ifdef PAE 958 /* 959 * we need a shadow page for the kernel's L2 page 960 * The real L2 kernel PD will be the last one (so that 961 * pde[L2_SLOT_KERN] always point to the shadow. 962 */ 963 memcpy(&pde[L2_SLOT_KERN + NPDPG], &pde[L2_SLOT_KERN], PAGE_SIZE); 964 cpu_info_primary.ci_kpm_pdir = &pde[L2_SLOT_KERN + NPDPG]; 965 cpu_info_primary.ci_kpm_pdirpa = 966 (vaddr_t) cpu_info_primary.ci_kpm_pdir - KERNBASE; 967 968 /* 969 * We don't enter a recursive entry from the L3 PD. Instead, 970 * we enter the first 4 L2 pages, which includes the kernel's L2 971 * shadow. But we have to entrer the shadow after switching 972 * %cr3, or Xen will refcount some PTE with the wrong type. 973 */ 974 addr = (u_long)pde - KERNBASE; 975 for (i = 0; i < 3; i++, addr += PAGE_SIZE) { 976 pde[PDIR_SLOT_PTE + i] = xpmap_ptom_masked(addr) | PG_k | PG_V; 977 __PRINTK(("pde[%d] va %#" PRIxVADDR " pa %#" PRIxPADDR 978 " entry %#" PRIxPADDR "\n", 979 (int)(PDIR_SLOT_PTE + i), pde + PAGE_SIZE * i, 980 addr, pde[PDIR_SLOT_PTE + i])); 981 } 982 #if 0 983 addr += PAGE_SIZE; /* point to shadow L2 */ 984 pde[PDIR_SLOT_PTE + 3] = xpmap_ptom_masked(addr) | PG_k | PG_V; 985 __PRINTK(("pde[%d] va 0x%lx pa 0x%lx entry 0x%" PRIx64 "\n", 986 (int)(PDIR_SLOT_PTE + 3), pde + PAGE_SIZE * 4, (long)addr, 987 (int64_t)pde[PDIR_SLOT_PTE + 3])); 988 #endif 989 /* Mark tables RO, and pin the kernel's shadow as L2 */ 990 addr = (u_long)pde - KERNBASE; 991 for (i = 0; i < 5; i++, addr += PAGE_SIZE) { 992 xen_bt_set_readonly(((vaddr_t)pde) + PAGE_SIZE * i); 993 if (i == 2 || i == 3) 994 continue; 995 #if 0 996 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", i, (int64_t)addr)); 997 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 998 #endif 999 } 1000 if (final) { 1001 addr = (u_long)pde - KERNBASE + 3 * PAGE_SIZE; 1002 __PRINTK(("pin L2 %d addr %#" PRIxPADDR "\n", 2, addr)); 1003 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 1004 } 1005 #if 0 1006 addr = (u_long)pde - KERNBASE + 2 * PAGE_SIZE; 1007 __PRINTK(("pin L2 %d addr 0x%" PRIx64 "\n", 2, (int64_t)addr)); 1008 xpq_queue_pin_l2_table(xpmap_ptom_masked(addr)); 1009 #endif 1010 #else /* PAE */ 1011 /* recursive entry in higher-level per-cpu PD and pmap_kernel() */ 1012 bt_pgd[PDIR_SLOT_PTE] = xpmap_ptom_masked((paddr_t)bt_pgd - KERNBASE) | PG_k | PG_V; 1013 #ifdef __x86_64__ 1014 bt_cpu_pgd[PDIR_SLOT_PTE] = 1015 xpmap_ptom_masked((paddr_t)bt_cpu_pgd - KERNBASE) | PG_k | PG_V; 1016 #endif /* __x86_64__ */ 1017 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] va %#" PRIxVADDR " pa %#" PRIxPADDR 1018 " entry %#" PRIxPADDR "\n", new_pgd, (paddr_t)new_pgd - KERNBASE, 1019 bt_pgd[PDIR_SLOT_PTE])); 1020 /* Mark tables RO */ 1021 xen_bt_set_readonly((vaddr_t) pde); 1022 #endif 1023 #if PTP_LEVELS > 2 || defined(PAE) 1024 xen_bt_set_readonly((vaddr_t) pdtpe); 1025 #endif 1026 #if PTP_LEVELS > 3 1027 xen_bt_set_readonly(new_pgd); 1028 #endif 1029 /* Pin the PGD */ 1030 __PRINTK(("pin PGD: %"PRIxVADDR"\n", new_pgd - KERNBASE)); 1031 #ifdef __x86_64__ 1032 xpq_queue_pin_l4_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 1033 #elif PAE 1034 xpq_queue_pin_l3_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 1035 #else 1036 xpq_queue_pin_l2_table(xpmap_ptom_masked(new_pgd - KERNBASE)); 1037 #endif 1038 1039 /* Save phys. addr of PDP, for libkvm. */ 1040 #ifdef PAE 1041 PDPpaddr = (u_long)pde - KERNBASE; /* PDP is the L2 with PAE */ 1042 #else 1043 PDPpaddr = (u_long)bt_pgd - KERNBASE; 1044 #endif 1045 1046 /* Switch to new tables */ 1047 __PRINTK(("switch to PGD\n")); 1048 xpq_queue_pt_switch(xpmap_ptom_masked(new_pgd - KERNBASE)); 1049 __PRINTK(("bt_pgd[PDIR_SLOT_PTE] now entry %#" PRIxPADDR "\n", 1050 bt_pgd[PDIR_SLOT_PTE])); 1051 1052 #ifdef PAE 1053 if (final) { 1054 /* save the address of the L3 page */ 1055 cpu_info_primary.ci_pae_l3_pdir = pdtpe; 1056 cpu_info_primary.ci_pae_l3_pdirpa = (new_pgd - KERNBASE); 1057 1058 /* now enter kernel's PTE mappings */ 1059 addr = (u_long)pde - KERNBASE + PAGE_SIZE * 3; 1060 xpq_queue_pte_update( 1061 xpmap_ptom(((vaddr_t)&pde[PDIR_SLOT_PTE + 3]) - KERNBASE), 1062 xpmap_ptom_masked(addr) | PG_k | PG_V); 1063 xpq_flush_queue(); 1064 } 1065 #elif defined(__x86_64__) 1066 if (final) { 1067 /* save the address of the real per-cpu L4 pgd page */ 1068 cpu_info_primary.ci_kpm_pdir = bt_cpu_pgd; 1069 cpu_info_primary.ci_kpm_pdirpa = ((paddr_t) bt_cpu_pgd - KERNBASE); 1070 } 1071 #endif 1072 __USE(pdtpe); 1073 1074 /* Now we can safely reclaim space taken by old tables */ 1075 1076 __PRINTK(("unpin old PGD\n")); 1077 /* Unpin old PGD */ 1078 xpq_queue_unpin_table(xpmap_ptom_masked(old_pgd - KERNBASE)); 1079 /* Mark old tables RW */ 1080 page = old_pgd; 1081 addr = (paddr_t) pde[pl2_pi(page)] & PG_FRAME; 1082 addr = xpmap_mtop(addr); 1083 pte = (pd_entry_t *) ((u_long)addr + KERNBASE); 1084 pte += pl1_pi(page); 1085 __PRINTK(("*pde %#" PRIxPADDR " addr %#" PRIxPADDR " pte %#lx\n", 1086 pde[pl2_pi(page)], addr, (long)pte)); 1087 while (page < old_pgd + (old_count * PAGE_SIZE) && page < map_end) { 1088 addr = xpmap_ptom(((u_long) pte) - KERNBASE); 1089 XENPRINTK(("addr %#" PRIxPADDR " pte %#lx " 1090 "*pte %#" PRIxPADDR "\n", 1091 addr, (long)pte, *pte)); 1092 xpq_queue_pte_update(addr, *pte | PG_RW); 1093 page += PAGE_SIZE; 1094 /* 1095 * Our ptes are contiguous 1096 * so it's safe to just "++" here 1097 */ 1098 pte++; 1099 } 1100 xpq_flush_queue(); 1101 } 1102 1103 1104 /* 1105 * Bootstrap helper functions 1106 */ 1107 1108 /* 1109 * Mark a page readonly 1110 * XXX: assuming vaddr = paddr + KERNBASE 1111 */ 1112 1113 static void 1114 xen_bt_set_readonly (vaddr_t page) 1115 { 1116 pt_entry_t entry; 1117 1118 entry = xpmap_ptom_masked(page - KERNBASE); 1119 entry |= PG_k | PG_V; 1120 1121 HYPERVISOR_update_va_mapping (page, entry, UVMF_INVLPG); 1122 } 1123 1124 #ifdef __x86_64__ 1125 void 1126 xen_set_user_pgd(paddr_t page) 1127 { 1128 struct mmuext_op op; 1129 int s = splvm(); 1130 1131 xpq_flush_queue(); 1132 op.cmd = MMUEXT_NEW_USER_BASEPTR; 1133 op.arg1.mfn = xpmap_ptom_masked(page) >> PAGE_SHIFT; 1134 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0) 1135 panic("xen_set_user_pgd: failed to install new user page" 1136 " directory %#" PRIxPADDR, page); 1137 splx(s); 1138 } 1139 #endif /* __x86_64__ */ 1140