1 /* 2 * Machine specific setup for xen 3 * 4 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 5 */ 6 7 #include <linux/init.h> 8 #include <linux/sched.h> 9 #include <linux/mm.h> 10 #include <linux/pm.h> 11 #include <linux/memblock.h> 12 #include <linux/cpuidle.h> 13 #include <linux/cpufreq.h> 14 15 #include <asm/elf.h> 16 #include <asm/vdso.h> 17 #include <asm/e820/api.h> 18 #include <asm/setup.h> 19 #include <asm/acpi.h> 20 #include <asm/numa.h> 21 #include <asm/xen/hypervisor.h> 22 #include <asm/xen/hypercall.h> 23 24 #include <xen/xen.h> 25 #include <xen/page.h> 26 #include <xen/interface/callback.h> 27 #include <xen/interface/memory.h> 28 #include <xen/interface/physdev.h> 29 #include <xen/features.h> 30 #include <xen/hvc-console.h> 31 #include "xen-ops.h" 32 #include "vdso.h" 33 #include "mmu.h" 34 35 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024) 36 37 /* Amount of extra memory space we add to the e820 ranges */ 38 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; 39 40 /* Number of pages released from the initial allocation. */ 41 unsigned long xen_released_pages; 42 43 /* E820 map used during setting up memory. */ 44 static struct e820_table xen_e820_table __initdata; 45 46 /* 47 * Buffer used to remap identity mapped pages. We only need the virtual space. 48 * The physical page behind this address is remapped as needed to different 49 * buffer pages. 50 */ 51 #define REMAP_SIZE (P2M_PER_PAGE - 3) 52 static struct { 53 unsigned long next_area_mfn; 54 unsigned long target_pfn; 55 unsigned long size; 56 unsigned long mfns[REMAP_SIZE]; 57 } xen_remap_buf __initdata __aligned(PAGE_SIZE); 58 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY; 59 60 /* 61 * The maximum amount of extra memory compared to the base size. The 62 * main scaling factor is the size of struct page. At extreme ratios 63 * of base:extra, all the base memory can be filled with page 64 * structures for the extra memory, leaving no space for anything 65 * else. 66 * 67 * 10x seems like a reasonable balance between scaling flexibility and 68 * leaving a practically usable system. 69 */ 70 #define EXTRA_MEM_RATIO (10) 71 72 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB); 73 74 static void __init xen_parse_512gb(void) 75 { 76 bool val = false; 77 char *arg; 78 79 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit"); 80 if (!arg) 81 return; 82 83 arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit="); 84 if (!arg) 85 val = true; 86 else if (strtobool(arg + strlen("xen_512gb_limit="), &val)) 87 return; 88 89 xen_512gb_limit = val; 90 } 91 92 static void __init xen_add_extra_mem(unsigned long start_pfn, 93 unsigned long n_pfns) 94 { 95 int i; 96 97 /* 98 * No need to check for zero size, should happen rarely and will only 99 * write a new entry regarded to be unused due to zero size. 100 */ 101 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { 102 /* Add new region. */ 103 if (xen_extra_mem[i].n_pfns == 0) { 104 xen_extra_mem[i].start_pfn = start_pfn; 105 xen_extra_mem[i].n_pfns = n_pfns; 106 break; 107 } 108 /* Append to existing region. */ 109 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns == 110 start_pfn) { 111 xen_extra_mem[i].n_pfns += n_pfns; 112 break; 113 } 114 } 115 if (i == XEN_EXTRA_MEM_MAX_REGIONS) 116 printk(KERN_WARNING "Warning: not enough extra memory regions\n"); 117 118 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns)); 119 } 120 121 static void __init xen_del_extra_mem(unsigned long start_pfn, 122 unsigned long n_pfns) 123 { 124 int i; 125 unsigned long start_r, size_r; 126 127 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { 128 start_r = xen_extra_mem[i].start_pfn; 129 size_r = xen_extra_mem[i].n_pfns; 130 131 /* Start of region. */ 132 if (start_r == start_pfn) { 133 BUG_ON(n_pfns > size_r); 134 xen_extra_mem[i].start_pfn += n_pfns; 135 xen_extra_mem[i].n_pfns -= n_pfns; 136 break; 137 } 138 /* End of region. */ 139 if (start_r + size_r == start_pfn + n_pfns) { 140 BUG_ON(n_pfns > size_r); 141 xen_extra_mem[i].n_pfns -= n_pfns; 142 break; 143 } 144 /* Mid of region. */ 145 if (start_pfn > start_r && start_pfn < start_r + size_r) { 146 BUG_ON(start_pfn + n_pfns > start_r + size_r); 147 xen_extra_mem[i].n_pfns = start_pfn - start_r; 148 /* Calling memblock_reserve() again is okay. */ 149 xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r - 150 (start_pfn + n_pfns)); 151 break; 152 } 153 } 154 memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns)); 155 } 156 157 /* 158 * Called during boot before the p2m list can take entries beyond the 159 * hypervisor supplied p2m list. Entries in extra mem are to be regarded as 160 * invalid. 161 */ 162 unsigned long __ref xen_chk_extra_mem(unsigned long pfn) 163 { 164 int i; 165 166 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { 167 if (pfn >= xen_extra_mem[i].start_pfn && 168 pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns) 169 return INVALID_P2M_ENTRY; 170 } 171 172 return IDENTITY_FRAME(pfn); 173 } 174 175 /* 176 * Mark all pfns of extra mem as invalid in p2m list. 177 */ 178 void __init xen_inv_extra_mem(void) 179 { 180 unsigned long pfn, pfn_s, pfn_e; 181 int i; 182 183 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { 184 if (!xen_extra_mem[i].n_pfns) 185 continue; 186 pfn_s = xen_extra_mem[i].start_pfn; 187 pfn_e = pfn_s + xen_extra_mem[i].n_pfns; 188 for (pfn = pfn_s; pfn < pfn_e; pfn++) 189 set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 190 } 191 } 192 193 /* 194 * Finds the next RAM pfn available in the E820 map after min_pfn. 195 * This function updates min_pfn with the pfn found and returns 196 * the size of that range or zero if not found. 197 */ 198 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn) 199 { 200 const struct e820_entry *entry = xen_e820_table.entries; 201 unsigned int i; 202 unsigned long done = 0; 203 204 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) { 205 unsigned long s_pfn; 206 unsigned long e_pfn; 207 208 if (entry->type != E820_TYPE_RAM) 209 continue; 210 211 e_pfn = PFN_DOWN(entry->addr + entry->size); 212 213 /* We only care about E820 after this */ 214 if (e_pfn <= *min_pfn) 215 continue; 216 217 s_pfn = PFN_UP(entry->addr); 218 219 /* If min_pfn falls within the E820 entry, we want to start 220 * at the min_pfn PFN. 221 */ 222 if (s_pfn <= *min_pfn) { 223 done = e_pfn - *min_pfn; 224 } else { 225 done = e_pfn - s_pfn; 226 *min_pfn = s_pfn; 227 } 228 break; 229 } 230 231 return done; 232 } 233 234 static int __init xen_free_mfn(unsigned long mfn) 235 { 236 struct xen_memory_reservation reservation = { 237 .address_bits = 0, 238 .extent_order = 0, 239 .domid = DOMID_SELF 240 }; 241 242 set_xen_guest_handle(reservation.extent_start, &mfn); 243 reservation.nr_extents = 1; 244 245 return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); 246 } 247 248 /* 249 * This releases a chunk of memory and then does the identity map. It's used 250 * as a fallback if the remapping fails. 251 */ 252 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn, 253 unsigned long end_pfn, unsigned long nr_pages) 254 { 255 unsigned long pfn, end; 256 int ret; 257 258 WARN_ON(start_pfn > end_pfn); 259 260 /* Release pages first. */ 261 end = min(end_pfn, nr_pages); 262 for (pfn = start_pfn; pfn < end; pfn++) { 263 unsigned long mfn = pfn_to_mfn(pfn); 264 265 /* Make sure pfn exists to start with */ 266 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn) 267 continue; 268 269 ret = xen_free_mfn(mfn); 270 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret); 271 272 if (ret == 1) { 273 xen_released_pages++; 274 if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY)) 275 break; 276 } else 277 break; 278 } 279 280 set_phys_range_identity(start_pfn, end_pfn); 281 } 282 283 /* 284 * Helper function to update the p2m and m2p tables and kernel mapping. 285 */ 286 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn) 287 { 288 struct mmu_update update = { 289 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE, 290 .val = pfn 291 }; 292 293 /* Update p2m */ 294 if (!set_phys_to_machine(pfn, mfn)) { 295 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n", 296 pfn, mfn); 297 BUG(); 298 } 299 300 /* Update m2p */ 301 if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) { 302 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n", 303 mfn, pfn); 304 BUG(); 305 } 306 307 /* Update kernel mapping, but not for highmem. */ 308 if (pfn >= PFN_UP(__pa(high_memory - 1))) 309 return; 310 311 if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT), 312 mfn_pte(mfn, PAGE_KERNEL), 0)) { 313 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n", 314 mfn, pfn); 315 BUG(); 316 } 317 } 318 319 /* 320 * This function updates the p2m and m2p tables with an identity map from 321 * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the 322 * original allocation at remap_pfn. The information needed for remapping is 323 * saved in the memory itself to avoid the need for allocating buffers. The 324 * complete remap information is contained in a list of MFNs each containing 325 * up to REMAP_SIZE MFNs and the start target PFN for doing the remap. 326 * This enables us to preserve the original mfn sequence while doing the 327 * remapping at a time when the memory management is capable of allocating 328 * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and 329 * its callers. 330 */ 331 static void __init xen_do_set_identity_and_remap_chunk( 332 unsigned long start_pfn, unsigned long size, unsigned long remap_pfn) 333 { 334 unsigned long buf = (unsigned long)&xen_remap_buf; 335 unsigned long mfn_save, mfn; 336 unsigned long ident_pfn_iter, remap_pfn_iter; 337 unsigned long ident_end_pfn = start_pfn + size; 338 unsigned long left = size; 339 unsigned int i, chunk; 340 341 WARN_ON(size == 0); 342 343 mfn_save = virt_to_mfn(buf); 344 345 for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn; 346 ident_pfn_iter < ident_end_pfn; 347 ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) { 348 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE; 349 350 /* Map first pfn to xen_remap_buf */ 351 mfn = pfn_to_mfn(ident_pfn_iter); 352 set_pte_mfn(buf, mfn, PAGE_KERNEL); 353 354 /* Save mapping information in page */ 355 xen_remap_buf.next_area_mfn = xen_remap_mfn; 356 xen_remap_buf.target_pfn = remap_pfn_iter; 357 xen_remap_buf.size = chunk; 358 for (i = 0; i < chunk; i++) 359 xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i); 360 361 /* Put remap buf into list. */ 362 xen_remap_mfn = mfn; 363 364 /* Set identity map */ 365 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk); 366 367 left -= chunk; 368 } 369 370 /* Restore old xen_remap_buf mapping */ 371 set_pte_mfn(buf, mfn_save, PAGE_KERNEL); 372 } 373 374 /* 375 * This function takes a contiguous pfn range that needs to be identity mapped 376 * and: 377 * 378 * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn. 379 * 2) Calls the do_ function to actually do the mapping/remapping work. 380 * 381 * The goal is to not allocate additional memory but to remap the existing 382 * pages. In the case of an error the underlying memory is simply released back 383 * to Xen and not remapped. 384 */ 385 static unsigned long __init xen_set_identity_and_remap_chunk( 386 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages, 387 unsigned long remap_pfn) 388 { 389 unsigned long pfn; 390 unsigned long i = 0; 391 unsigned long n = end_pfn - start_pfn; 392 393 if (remap_pfn == 0) 394 remap_pfn = nr_pages; 395 396 while (i < n) { 397 unsigned long cur_pfn = start_pfn + i; 398 unsigned long left = n - i; 399 unsigned long size = left; 400 unsigned long remap_range_size; 401 402 /* Do not remap pages beyond the current allocation */ 403 if (cur_pfn >= nr_pages) { 404 /* Identity map remaining pages */ 405 set_phys_range_identity(cur_pfn, cur_pfn + size); 406 break; 407 } 408 if (cur_pfn + size > nr_pages) 409 size = nr_pages - cur_pfn; 410 411 remap_range_size = xen_find_pfn_range(&remap_pfn); 412 if (!remap_range_size) { 413 pr_warning("Unable to find available pfn range, not remapping identity pages\n"); 414 xen_set_identity_and_release_chunk(cur_pfn, 415 cur_pfn + left, nr_pages); 416 break; 417 } 418 /* Adjust size to fit in current e820 RAM region */ 419 if (size > remap_range_size) 420 size = remap_range_size; 421 422 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn); 423 424 /* Update variables to reflect new mappings. */ 425 i += size; 426 remap_pfn += size; 427 } 428 429 /* 430 * If the PFNs are currently mapped, the VA mapping also needs 431 * to be updated to be 1:1. 432 */ 433 for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++) 434 (void)HYPERVISOR_update_va_mapping( 435 (unsigned long)__va(pfn << PAGE_SHIFT), 436 mfn_pte(pfn, PAGE_KERNEL_IO), 0); 437 438 return remap_pfn; 439 } 440 441 static unsigned long __init xen_count_remap_pages( 442 unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages, 443 unsigned long remap_pages) 444 { 445 if (start_pfn >= nr_pages) 446 return remap_pages; 447 448 return remap_pages + min(end_pfn, nr_pages) - start_pfn; 449 } 450 451 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages, 452 unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn, 453 unsigned long nr_pages, unsigned long last_val)) 454 { 455 phys_addr_t start = 0; 456 unsigned long ret_val = 0; 457 const struct e820_entry *entry = xen_e820_table.entries; 458 int i; 459 460 /* 461 * Combine non-RAM regions and gaps until a RAM region (or the 462 * end of the map) is reached, then call the provided function 463 * to perform its duty on the non-RAM region. 464 * 465 * The combined non-RAM regions are rounded to a whole number 466 * of pages so any partial pages are accessible via the 1:1 467 * mapping. This is needed for some BIOSes that put (for 468 * example) the DMI tables in a reserved region that begins on 469 * a non-page boundary. 470 */ 471 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) { 472 phys_addr_t end = entry->addr + entry->size; 473 if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) { 474 unsigned long start_pfn = PFN_DOWN(start); 475 unsigned long end_pfn = PFN_UP(end); 476 477 if (entry->type == E820_TYPE_RAM) 478 end_pfn = PFN_UP(entry->addr); 479 480 if (start_pfn < end_pfn) 481 ret_val = func(start_pfn, end_pfn, nr_pages, 482 ret_val); 483 start = end; 484 } 485 } 486 487 return ret_val; 488 } 489 490 /* 491 * Remap the memory prepared in xen_do_set_identity_and_remap_chunk(). 492 * The remap information (which mfn remap to which pfn) is contained in the 493 * to be remapped memory itself in a linked list anchored at xen_remap_mfn. 494 * This scheme allows to remap the different chunks in arbitrary order while 495 * the resulting mapping will be independant from the order. 496 */ 497 void __init xen_remap_memory(void) 498 { 499 unsigned long buf = (unsigned long)&xen_remap_buf; 500 unsigned long mfn_save, pfn; 501 unsigned long remapped = 0; 502 unsigned int i; 503 unsigned long pfn_s = ~0UL; 504 unsigned long len = 0; 505 506 mfn_save = virt_to_mfn(buf); 507 508 while (xen_remap_mfn != INVALID_P2M_ENTRY) { 509 /* Map the remap information */ 510 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL); 511 512 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]); 513 514 pfn = xen_remap_buf.target_pfn; 515 for (i = 0; i < xen_remap_buf.size; i++) { 516 xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]); 517 remapped++; 518 pfn++; 519 } 520 if (pfn_s == ~0UL || pfn == pfn_s) { 521 pfn_s = xen_remap_buf.target_pfn; 522 len += xen_remap_buf.size; 523 } else if (pfn_s + len == xen_remap_buf.target_pfn) { 524 len += xen_remap_buf.size; 525 } else { 526 xen_del_extra_mem(pfn_s, len); 527 pfn_s = xen_remap_buf.target_pfn; 528 len = xen_remap_buf.size; 529 } 530 xen_remap_mfn = xen_remap_buf.next_area_mfn; 531 } 532 533 if (pfn_s != ~0UL && len) 534 xen_del_extra_mem(pfn_s, len); 535 536 set_pte_mfn(buf, mfn_save, PAGE_KERNEL); 537 538 pr_info("Remapped %ld page(s)\n", remapped); 539 } 540 541 static unsigned long __init xen_get_pages_limit(void) 542 { 543 unsigned long limit; 544 545 #ifdef CONFIG_X86_32 546 limit = GB(64) / PAGE_SIZE; 547 #else 548 limit = MAXMEM / PAGE_SIZE; 549 if (!xen_initial_domain() && xen_512gb_limit) 550 limit = GB(512) / PAGE_SIZE; 551 #endif 552 return limit; 553 } 554 555 static unsigned long __init xen_get_max_pages(void) 556 { 557 unsigned long max_pages, limit; 558 domid_t domid = DOMID_SELF; 559 long ret; 560 561 limit = xen_get_pages_limit(); 562 max_pages = limit; 563 564 /* 565 * For the initial domain we use the maximum reservation as 566 * the maximum page. 567 * 568 * For guest domains the current maximum reservation reflects 569 * the current maximum rather than the static maximum. In this 570 * case the e820 map provided to us will cover the static 571 * maximum region. 572 */ 573 if (xen_initial_domain()) { 574 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid); 575 if (ret > 0) 576 max_pages = ret; 577 } 578 579 return min(max_pages, limit); 580 } 581 582 static void __init xen_align_and_add_e820_region(phys_addr_t start, 583 phys_addr_t size, int type) 584 { 585 phys_addr_t end = start + size; 586 587 /* Align RAM regions to page boundaries. */ 588 if (type == E820_TYPE_RAM) { 589 start = PAGE_ALIGN(start); 590 end &= ~((phys_addr_t)PAGE_SIZE - 1); 591 } 592 593 e820__range_add(start, end - start, type); 594 } 595 596 static void __init xen_ignore_unusable(void) 597 { 598 struct e820_entry *entry = xen_e820_table.entries; 599 unsigned int i; 600 601 for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) { 602 if (entry->type == E820_TYPE_UNUSABLE) 603 entry->type = E820_TYPE_RAM; 604 } 605 } 606 607 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size) 608 { 609 struct e820_entry *entry; 610 unsigned mapcnt; 611 phys_addr_t end; 612 613 if (!size) 614 return false; 615 616 end = start + size; 617 entry = xen_e820_table.entries; 618 619 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) { 620 if (entry->type == E820_TYPE_RAM && entry->addr <= start && 621 (entry->addr + entry->size) >= end) 622 return false; 623 624 entry++; 625 } 626 627 return true; 628 } 629 630 /* 631 * Find a free area in physical memory not yet reserved and compliant with 632 * E820 map. 633 * Used to relocate pre-allocated areas like initrd or p2m list which are in 634 * conflict with the to be used E820 map. 635 * In case no area is found, return 0. Otherwise return the physical address 636 * of the area which is already reserved for convenience. 637 */ 638 phys_addr_t __init xen_find_free_area(phys_addr_t size) 639 { 640 unsigned mapcnt; 641 phys_addr_t addr, start; 642 struct e820_entry *entry = xen_e820_table.entries; 643 644 for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) { 645 if (entry->type != E820_TYPE_RAM || entry->size < size) 646 continue; 647 start = entry->addr; 648 for (addr = start; addr < start + size; addr += PAGE_SIZE) { 649 if (!memblock_is_reserved(addr)) 650 continue; 651 start = addr + PAGE_SIZE; 652 if (start + size > entry->addr + entry->size) 653 break; 654 } 655 if (addr >= start + size) { 656 memblock_reserve(start, size); 657 return start; 658 } 659 } 660 661 return 0; 662 } 663 664 /* 665 * Like memcpy, but with physical addresses for dest and src. 666 */ 667 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src, 668 phys_addr_t n) 669 { 670 phys_addr_t dest_off, src_off, dest_len, src_len, len; 671 void *from, *to; 672 673 while (n) { 674 dest_off = dest & ~PAGE_MASK; 675 src_off = src & ~PAGE_MASK; 676 dest_len = n; 677 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off) 678 dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off; 679 src_len = n; 680 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off) 681 src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off; 682 len = min(dest_len, src_len); 683 to = early_memremap(dest - dest_off, dest_len + dest_off); 684 from = early_memremap(src - src_off, src_len + src_off); 685 memcpy(to, from, len); 686 early_memunmap(to, dest_len + dest_off); 687 early_memunmap(from, src_len + src_off); 688 n -= len; 689 dest += len; 690 src += len; 691 } 692 } 693 694 /* 695 * Reserve Xen mfn_list. 696 */ 697 static void __init xen_reserve_xen_mfnlist(void) 698 { 699 phys_addr_t start, size; 700 701 if (xen_start_info->mfn_list >= __START_KERNEL_map) { 702 start = __pa(xen_start_info->mfn_list); 703 size = PFN_ALIGN(xen_start_info->nr_pages * 704 sizeof(unsigned long)); 705 } else { 706 start = PFN_PHYS(xen_start_info->first_p2m_pfn); 707 size = PFN_PHYS(xen_start_info->nr_p2m_frames); 708 } 709 710 memblock_reserve(start, size); 711 if (!xen_is_e820_reserved(start, size)) 712 return; 713 714 #ifdef CONFIG_X86_32 715 /* 716 * Relocating the p2m on 32 bit system to an arbitrary virtual address 717 * is not supported, so just give up. 718 */ 719 xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n"); 720 BUG(); 721 #else 722 xen_relocate_p2m(); 723 memblock_free(start, size); 724 #endif 725 } 726 727 /** 728 * machine_specific_memory_setup - Hook for machine specific memory setup. 729 **/ 730 char * __init xen_memory_setup(void) 731 { 732 unsigned long max_pfn, pfn_s, n_pfns; 733 phys_addr_t mem_end, addr, size, chunk_size; 734 u32 type; 735 int rc; 736 struct xen_memory_map memmap; 737 unsigned long max_pages; 738 unsigned long extra_pages = 0; 739 int i; 740 int op; 741 742 xen_parse_512gb(); 743 max_pfn = xen_get_pages_limit(); 744 max_pfn = min(max_pfn, xen_start_info->nr_pages); 745 mem_end = PFN_PHYS(max_pfn); 746 747 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries); 748 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries); 749 750 op = xen_initial_domain() ? 751 XENMEM_machine_memory_map : 752 XENMEM_memory_map; 753 rc = HYPERVISOR_memory_op(op, &memmap); 754 if (rc == -ENOSYS) { 755 BUG_ON(xen_initial_domain()); 756 memmap.nr_entries = 1; 757 xen_e820_table.entries[0].addr = 0ULL; 758 xen_e820_table.entries[0].size = mem_end; 759 /* 8MB slack (to balance backend allocations). */ 760 xen_e820_table.entries[0].size += 8ULL << 20; 761 xen_e820_table.entries[0].type = E820_TYPE_RAM; 762 rc = 0; 763 } 764 BUG_ON(rc); 765 BUG_ON(memmap.nr_entries == 0); 766 xen_e820_table.nr_entries = memmap.nr_entries; 767 768 /* 769 * Xen won't allow a 1:1 mapping to be created to UNUSABLE 770 * regions, so if we're using the machine memory map leave the 771 * region as RAM as it is in the pseudo-physical map. 772 * 773 * UNUSABLE regions in domUs are not handled and will need 774 * a patch in the future. 775 */ 776 if (xen_initial_domain()) 777 xen_ignore_unusable(); 778 779 /* Make sure the Xen-supplied memory map is well-ordered. */ 780 e820__update_table(&xen_e820_table); 781 782 max_pages = xen_get_max_pages(); 783 784 /* How many extra pages do we need due to remapping? */ 785 max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages); 786 787 if (max_pages > max_pfn) 788 extra_pages += max_pages - max_pfn; 789 790 /* 791 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO 792 * factor the base size. On non-highmem systems, the base 793 * size is the full initial memory allocation; on highmem it 794 * is limited to the max size of lowmem, so that it doesn't 795 * get completely filled. 796 * 797 * Make sure we have no memory above max_pages, as this area 798 * isn't handled by the p2m management. 799 * 800 * In principle there could be a problem in lowmem systems if 801 * the initial memory is also very large with respect to 802 * lowmem, but we won't try to deal with that here. 803 */ 804 extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)), 805 extra_pages, max_pages - max_pfn); 806 i = 0; 807 addr = xen_e820_table.entries[0].addr; 808 size = xen_e820_table.entries[0].size; 809 while (i < xen_e820_table.nr_entries) { 810 bool discard = false; 811 812 chunk_size = size; 813 type = xen_e820_table.entries[i].type; 814 815 if (type == E820_TYPE_RAM) { 816 if (addr < mem_end) { 817 chunk_size = min(size, mem_end - addr); 818 } else if (extra_pages) { 819 chunk_size = min(size, PFN_PHYS(extra_pages)); 820 pfn_s = PFN_UP(addr); 821 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s; 822 extra_pages -= n_pfns; 823 xen_add_extra_mem(pfn_s, n_pfns); 824 xen_max_p2m_pfn = pfn_s + n_pfns; 825 } else 826 discard = true; 827 } 828 829 if (!discard) 830 xen_align_and_add_e820_region(addr, chunk_size, type); 831 832 addr += chunk_size; 833 size -= chunk_size; 834 if (size == 0) { 835 i++; 836 if (i < xen_e820_table.nr_entries) { 837 addr = xen_e820_table.entries[i].addr; 838 size = xen_e820_table.entries[i].size; 839 } 840 } 841 } 842 843 /* 844 * Set the rest as identity mapped, in case PCI BARs are 845 * located here. 846 */ 847 set_phys_range_identity(addr / PAGE_SIZE, ~0ul); 848 849 /* 850 * In domU, the ISA region is normal, usable memory, but we 851 * reserve ISA memory anyway because too many things poke 852 * about in there. 853 */ 854 e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED); 855 856 e820__update_table(e820_table); 857 858 /* 859 * Check whether the kernel itself conflicts with the target E820 map. 860 * Failing now is better than running into weird problems later due 861 * to relocating (and even reusing) pages with kernel text or data. 862 */ 863 if (xen_is_e820_reserved(__pa_symbol(_text), 864 __pa_symbol(__bss_stop) - __pa_symbol(_text))) { 865 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n"); 866 BUG(); 867 } 868 869 /* 870 * Check for a conflict of the hypervisor supplied page tables with 871 * the target E820 map. 872 */ 873 xen_pt_check_e820(); 874 875 xen_reserve_xen_mfnlist(); 876 877 /* Check for a conflict of the initrd with the target E820 map. */ 878 if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image, 879 boot_params.hdr.ramdisk_size)) { 880 phys_addr_t new_area, start, size; 881 882 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size); 883 if (!new_area) { 884 xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n"); 885 BUG(); 886 } 887 888 start = boot_params.hdr.ramdisk_image; 889 size = boot_params.hdr.ramdisk_size; 890 xen_phys_memcpy(new_area, start, size); 891 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n", 892 start, start + size, new_area, new_area + size); 893 memblock_free(start, size); 894 boot_params.hdr.ramdisk_image = new_area; 895 boot_params.ext_ramdisk_image = new_area >> 32; 896 } 897 898 /* 899 * Set identity map on non-RAM pages and prepare remapping the 900 * underlying RAM. 901 */ 902 xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk); 903 904 pr_info("Released %ld page(s)\n", xen_released_pages); 905 906 return "Xen"; 907 } 908 909 /* 910 * Machine specific memory setup for auto-translated guests. 911 */ 912 char * __init xen_auto_xlated_memory_setup(void) 913 { 914 struct xen_memory_map memmap; 915 int i; 916 int rc; 917 918 memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries); 919 set_xen_guest_handle(memmap.buffer, xen_e820_table.entries); 920 921 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap); 922 if (rc < 0) 923 panic("No memory map (%d)\n", rc); 924 925 xen_e820_table.nr_entries = memmap.nr_entries; 926 927 e820__update_table(&xen_e820_table); 928 929 for (i = 0; i < xen_e820_table.nr_entries; i++) 930 e820__range_add(xen_e820_table.entries[i].addr, xen_e820_table.entries[i].size, xen_e820_table.entries[i].type); 931 932 /* Remove p2m info, it is not needed. */ 933 xen_start_info->mfn_list = 0; 934 xen_start_info->first_p2m_pfn = 0; 935 xen_start_info->nr_p2m_frames = 0; 936 937 return "Xen"; 938 } 939 940 /* 941 * Set the bit indicating "nosegneg" library variants should be used. 942 * We only need to bother in pure 32-bit mode; compat 32-bit processes 943 * can have un-truncated segments, so wrapping around is allowed. 944 */ 945 static void __init fiddle_vdso(void) 946 { 947 #ifdef CONFIG_X86_32 948 u32 *mask = vdso_image_32.data + 949 vdso_image_32.sym_VDSO32_NOTE_MASK; 950 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; 951 #endif 952 } 953 954 static int register_callback(unsigned type, const void *func) 955 { 956 struct callback_register callback = { 957 .type = type, 958 .address = XEN_CALLBACK(__KERNEL_CS, func), 959 .flags = CALLBACKF_mask_events, 960 }; 961 962 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback); 963 } 964 965 void xen_enable_sysenter(void) 966 { 967 int ret; 968 unsigned sysenter_feature; 969 970 #ifdef CONFIG_X86_32 971 sysenter_feature = X86_FEATURE_SEP; 972 #else 973 sysenter_feature = X86_FEATURE_SYSENTER32; 974 #endif 975 976 if (!boot_cpu_has(sysenter_feature)) 977 return; 978 979 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target); 980 if(ret != 0) 981 setup_clear_cpu_cap(sysenter_feature); 982 } 983 984 void xen_enable_syscall(void) 985 { 986 #ifdef CONFIG_X86_64 987 int ret; 988 989 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target); 990 if (ret != 0) { 991 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret); 992 /* Pretty fatal; 64-bit userspace has no other 993 mechanism for syscalls. */ 994 } 995 996 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) { 997 ret = register_callback(CALLBACKTYPE_syscall32, 998 xen_syscall32_target); 999 if (ret != 0) 1000 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32); 1001 } 1002 #endif /* CONFIG_X86_64 */ 1003 } 1004 1005 void __init xen_pvmmu_arch_setup(void) 1006 { 1007 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments); 1008 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables); 1009 1010 HYPERVISOR_vm_assist(VMASST_CMD_enable, 1011 VMASST_TYPE_pae_extended_cr3); 1012 1013 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) || 1014 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback)) 1015 BUG(); 1016 1017 xen_enable_sysenter(); 1018 xen_enable_syscall(); 1019 } 1020 1021 /* This function is not called for HVM domains */ 1022 void __init xen_arch_setup(void) 1023 { 1024 xen_panic_handler_init(); 1025 xen_pvmmu_arch_setup(); 1026 1027 #ifdef CONFIG_ACPI 1028 if (!(xen_start_info->flags & SIF_INITDOMAIN)) { 1029 printk(KERN_INFO "ACPI in unprivileged domain disabled\n"); 1030 disable_acpi(); 1031 } 1032 #endif 1033 1034 memcpy(boot_command_line, xen_start_info->cmd_line, 1035 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ? 1036 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE); 1037 1038 /* Set up idle, making sure it calls safe_halt() pvop */ 1039 disable_cpuidle(); 1040 disable_cpufreq(); 1041 WARN_ON(xen_set_default_idle()); 1042 fiddle_vdso(); 1043 #ifdef CONFIG_NUMA 1044 numa_off = 1; 1045 #endif 1046 } 1047