arch/x86/xen/setup.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
xen_parse_512gb
xen_add_extra_mem
xen_del_extra_mem
xen_chk_extra_mem
xen_inv_extra_mem
xen_find_pfn_range
xen_free_mfn
xen_set_identity_and_release_chunk
xen_update_mem_tables
xen_do_set_identity_and_remap_chunk
xen_set_identity_and_remap_chunk
xen_count_remap_pages
xen_foreach_remap_area
xen_remap_memory
xen_get_pages_limit
xen_get_max_pages
xen_align_and_add_e820_region
xen_ignore_unusable
xen_is_e820_reserved
xen_find_free_area
xen_phys_memcpy
xen_reserve_xen_mfnlist
xen_memory_setup
fiddle_vdso
register_callback
xen_enable_sysenter
xen_enable_syscall
xen_pvmmu_arch_setup
xen_arch_setup
   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Machine specific setup for xen
   4  *
   5  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
   6  */
   7 
   8 #include <linux/init.h>
   9 #include <linux/sched.h>
  10 #include <linux/mm.h>
  11 #include <linux/pm.h>
  12 #include <linux/memblock.h>
  13 #include <linux/cpuidle.h>
  14 #include <linux/cpufreq.h>
  15 #include <linux/memory_hotplug.h>
  16 
  17 #include <asm/elf.h>
  18 #include <asm/vdso.h>
  19 #include <asm/e820/api.h>
  20 #include <asm/setup.h>
  21 #include <asm/acpi.h>
  22 #include <asm/numa.h>
  23 #include <asm/xen/hypervisor.h>
  24 #include <asm/xen/hypercall.h>
  25 
  26 #include <xen/xen.h>
  27 #include <xen/page.h>
  28 #include <xen/interface/callback.h>
  29 #include <xen/interface/memory.h>
  30 #include <xen/interface/physdev.h>
  31 #include <xen/features.h>
  32 #include <xen/hvc-console.h>
  33 #include "xen-ops.h"
  34 #include "vdso.h"
  35 #include "mmu.h"
  36 
  37 #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024)
  38 
  39 /* Amount of extra memory space we add to the e820 ranges */
  40 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
  41 
  42 /* Number of pages released from the initial allocation. */
  43 unsigned long xen_released_pages;
  44 
  45 /* E820 map used during setting up memory. */
  46 static struct e820_table xen_e820_table __initdata;
  47 
  48 /*
  49  * Buffer used to remap identity mapped pages. We only need the virtual space.
  50  * The physical page behind this address is remapped as needed to different
  51  * buffer pages.
  52  */
  53 #define REMAP_SIZE      (P2M_PER_PAGE - 3)
  54 static struct {
  55         unsigned long   next_area_mfn;
  56         unsigned long   target_pfn;
  57         unsigned long   size;
  58         unsigned long   mfns[REMAP_SIZE];
  59 } xen_remap_buf __initdata __aligned(PAGE_SIZE);
  60 static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY;
  61 
  62 /* 
  63  * The maximum amount of extra memory compared to the base size.  The
  64  * main scaling factor is the size of struct page.  At extreme ratios
  65  * of base:extra, all the base memory can be filled with page
  66  * structures for the extra memory, leaving no space for anything
  67  * else.
  68  * 
  69  * 10x seems like a reasonable balance between scaling flexibility and
  70  * leaving a practically usable system.
  71  */
  72 #define EXTRA_MEM_RATIO         (10)
  73 
  74 static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB);
  75 
  76 static void __init xen_parse_512gb(void)
  77 {
  78         bool val = false;
  79         char *arg;
  80 
  81         arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit");
  82         if (!arg)
  83                 return;
  84 
  85         arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit=");
  86         if (!arg)
  87                 val = true;
  88         else if (strtobool(arg + strlen("xen_512gb_limit="), &val))
  89                 return;
  90 
  91         xen_512gb_limit = val;
  92 }
  93 
  94 static void __init xen_add_extra_mem(unsigned long start_pfn,
  95                                      unsigned long n_pfns)
  96 {
  97         int i;
  98 
  99         /*
 100          * No need to check for zero size, should happen rarely and will only
 101          * write a new entry regarded to be unused due to zero size.
 102          */
 103         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 104                 /* Add new region. */
 105                 if (xen_extra_mem[i].n_pfns == 0) {
 106                         xen_extra_mem[i].start_pfn = start_pfn;
 107                         xen_extra_mem[i].n_pfns = n_pfns;
 108                         break;
 109                 }
 110                 /* Append to existing region. */
 111                 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
 112                     start_pfn) {
 113                         xen_extra_mem[i].n_pfns += n_pfns;
 114                         break;
 115                 }
 116         }
 117         if (i == XEN_EXTRA_MEM_MAX_REGIONS)
 118                 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
 119 
 120         memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 121 }
 122 
 123 static void __init xen_del_extra_mem(unsigned long start_pfn,
 124                                      unsigned long n_pfns)
 125 {
 126         int i;
 127         unsigned long start_r, size_r;
 128 
 129         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 130                 start_r = xen_extra_mem[i].start_pfn;
 131                 size_r = xen_extra_mem[i].n_pfns;
 132 
 133                 /* Start of region. */
 134                 if (start_r == start_pfn) {
 135                         BUG_ON(n_pfns > size_r);
 136                         xen_extra_mem[i].start_pfn += n_pfns;
 137                         xen_extra_mem[i].n_pfns -= n_pfns;
 138                         break;
 139                 }
 140                 /* End of region. */
 141                 if (start_r + size_r == start_pfn + n_pfns) {
 142                         BUG_ON(n_pfns > size_r);
 143                         xen_extra_mem[i].n_pfns -= n_pfns;
 144                         break;
 145                 }
 146                 /* Mid of region. */
 147                 if (start_pfn > start_r && start_pfn < start_r + size_r) {
 148                         BUG_ON(start_pfn + n_pfns > start_r + size_r);
 149                         xen_extra_mem[i].n_pfns = start_pfn - start_r;
 150                         /* Calling memblock_reserve() again is okay. */
 151                         xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r -
 152                                           (start_pfn + n_pfns));
 153                         break;
 154                 }
 155         }
 156         memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
 157 }
 158 
 159 /*
 160  * Called during boot before the p2m list can take entries beyond the
 161  * hypervisor supplied p2m list. Entries in extra mem are to be regarded as
 162  * invalid.
 163  */
 164 unsigned long __ref xen_chk_extra_mem(unsigned long pfn)
 165 {
 166         int i;
 167 
 168         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 169                 if (pfn >= xen_extra_mem[i].start_pfn &&
 170                     pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns)
 171                         return INVALID_P2M_ENTRY;
 172         }
 173 
 174         return IDENTITY_FRAME(pfn);
 175 }
 176 
 177 /*
 178  * Mark all pfns of extra mem as invalid in p2m list.
 179  */
 180 void __init xen_inv_extra_mem(void)
 181 {
 182         unsigned long pfn, pfn_s, pfn_e;
 183         int i;
 184 
 185         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 186                 if (!xen_extra_mem[i].n_pfns)
 187                         continue;
 188                 pfn_s = xen_extra_mem[i].start_pfn;
 189                 pfn_e = pfn_s + xen_extra_mem[i].n_pfns;
 190                 for (pfn = pfn_s; pfn < pfn_e; pfn++)
 191                         set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 192         }
 193 }
 194 
 195 /*
 196  * Finds the next RAM pfn available in the E820 map after min_pfn.
 197  * This function updates min_pfn with the pfn found and returns
 198  * the size of that range or zero if not found.
 199  */
 200 static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn)
 201 {
 202         const struct e820_entry *entry = xen_e820_table.entries;
 203         unsigned int i;
 204         unsigned long done = 0;
 205 
 206         for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
 207                 unsigned long s_pfn;
 208                 unsigned long e_pfn;
 209 
 210                 if (entry->type != E820_TYPE_RAM)
 211                         continue;
 212 
 213                 e_pfn = PFN_DOWN(entry->addr + entry->size);
 214 
 215                 /* We only care about E820 after this */
 216                 if (e_pfn <= *min_pfn)
 217                         continue;
 218 
 219                 s_pfn = PFN_UP(entry->addr);
 220 
 221                 /* If min_pfn falls within the E820 entry, we want to start
 222                  * at the min_pfn PFN.
 223                  */
 224                 if (s_pfn <= *min_pfn) {
 225                         done = e_pfn - *min_pfn;
 226                 } else {
 227                         done = e_pfn - s_pfn;
 228                         *min_pfn = s_pfn;
 229                 }
 230                 break;
 231         }
 232 
 233         return done;
 234 }
 235 
 236 static int __init xen_free_mfn(unsigned long mfn)
 237 {
 238         struct xen_memory_reservation reservation = {
 239                 .address_bits = 0,
 240                 .extent_order = 0,
 241                 .domid        = DOMID_SELF
 242         };
 243 
 244         set_xen_guest_handle(reservation.extent_start, &mfn);
 245         reservation.nr_extents = 1;
 246 
 247         return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
 248 }
 249 
 250 /*
 251  * This releases a chunk of memory and then does the identity map. It's used
 252  * as a fallback if the remapping fails.
 253  */
 254 static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn,
 255                         unsigned long end_pfn, unsigned long nr_pages)
 256 {
 257         unsigned long pfn, end;
 258         int ret;
 259 
 260         WARN_ON(start_pfn > end_pfn);
 261 
 262         /* Release pages first. */
 263         end = min(end_pfn, nr_pages);
 264         for (pfn = start_pfn; pfn < end; pfn++) {
 265                 unsigned long mfn = pfn_to_mfn(pfn);
 266 
 267                 /* Make sure pfn exists to start with */
 268                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
 269                         continue;
 270 
 271                 ret = xen_free_mfn(mfn);
 272                 WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
 273 
 274                 if (ret == 1) {
 275                         xen_released_pages++;
 276                         if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY))
 277                                 break;
 278                 } else
 279                         break;
 280         }
 281 
 282         set_phys_range_identity(start_pfn, end_pfn);
 283 }
 284 
 285 /*
 286  * Helper function to update the p2m and m2p tables and kernel mapping.
 287  */
 288 static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn)
 289 {
 290         struct mmu_update update = {
 291                 .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE,
 292                 .val = pfn
 293         };
 294 
 295         /* Update p2m */
 296         if (!set_phys_to_machine(pfn, mfn)) {
 297                 WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n",
 298                      pfn, mfn);
 299                 BUG();
 300         }
 301 
 302         /* Update m2p */
 303         if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) {
 304                 WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n",
 305                      mfn, pfn);
 306                 BUG();
 307         }
 308 
 309         /* Update kernel mapping, but not for highmem. */
 310         if (pfn >= PFN_UP(__pa(high_memory - 1)))
 311                 return;
 312 
 313         if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT),
 314                                          mfn_pte(mfn, PAGE_KERNEL), 0)) {
 315                 WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n",
 316                       mfn, pfn);
 317                 BUG();
 318         }
 319 }
 320 
 321 /*
 322  * This function updates the p2m and m2p tables with an identity map from
 323  * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the
 324  * original allocation at remap_pfn. The information needed for remapping is
 325  * saved in the memory itself to avoid the need for allocating buffers. The
 326  * complete remap information is contained in a list of MFNs each containing
 327  * up to REMAP_SIZE MFNs and the start target PFN for doing the remap.
 328  * This enables us to preserve the original mfn sequence while doing the
 329  * remapping at a time when the memory management is capable of allocating
 330  * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and
 331  * its callers.
 332  */
 333 static void __init xen_do_set_identity_and_remap_chunk(
 334         unsigned long start_pfn, unsigned long size, unsigned long remap_pfn)
 335 {
 336         unsigned long buf = (unsigned long)&xen_remap_buf;
 337         unsigned long mfn_save, mfn;
 338         unsigned long ident_pfn_iter, remap_pfn_iter;
 339         unsigned long ident_end_pfn = start_pfn + size;
 340         unsigned long left = size;
 341         unsigned int i, chunk;
 342 
 343         WARN_ON(size == 0);
 344 
 345         mfn_save = virt_to_mfn(buf);
 346 
 347         for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn;
 348              ident_pfn_iter < ident_end_pfn;
 349              ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) {
 350                 chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE;
 351 
 352                 /* Map first pfn to xen_remap_buf */
 353                 mfn = pfn_to_mfn(ident_pfn_iter);
 354                 set_pte_mfn(buf, mfn, PAGE_KERNEL);
 355 
 356                 /* Save mapping information in page */
 357                 xen_remap_buf.next_area_mfn = xen_remap_mfn;
 358                 xen_remap_buf.target_pfn = remap_pfn_iter;
 359                 xen_remap_buf.size = chunk;
 360                 for (i = 0; i < chunk; i++)
 361                         xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i);
 362 
 363                 /* Put remap buf into list. */
 364                 xen_remap_mfn = mfn;
 365 
 366                 /* Set identity map */
 367                 set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk);
 368 
 369                 left -= chunk;
 370         }
 371 
 372         /* Restore old xen_remap_buf mapping */
 373         set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
 374 }
 375 
 376 /*
 377  * This function takes a contiguous pfn range that needs to be identity mapped
 378  * and:
 379  *
 380  *  1) Finds a new range of pfns to use to remap based on E820 and remap_pfn.
 381  *  2) Calls the do_ function to actually do the mapping/remapping work.
 382  *
 383  * The goal is to not allocate additional memory but to remap the existing
 384  * pages. In the case of an error the underlying memory is simply released back
 385  * to Xen and not remapped.
 386  */
 387 static unsigned long __init xen_set_identity_and_remap_chunk(
 388         unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
 389         unsigned long remap_pfn)
 390 {
 391         unsigned long pfn;
 392         unsigned long i = 0;
 393         unsigned long n = end_pfn - start_pfn;
 394 
 395         if (remap_pfn == 0)
 396                 remap_pfn = nr_pages;
 397 
 398         while (i < n) {
 399                 unsigned long cur_pfn = start_pfn + i;
 400                 unsigned long left = n - i;
 401                 unsigned long size = left;
 402                 unsigned long remap_range_size;
 403 
 404                 /* Do not remap pages beyond the current allocation */
 405                 if (cur_pfn >= nr_pages) {
 406                         /* Identity map remaining pages */
 407                         set_phys_range_identity(cur_pfn, cur_pfn + size);
 408                         break;
 409                 }
 410                 if (cur_pfn + size > nr_pages)
 411                         size = nr_pages - cur_pfn;
 412 
 413                 remap_range_size = xen_find_pfn_range(&remap_pfn);
 414                 if (!remap_range_size) {
 415                         pr_warning("Unable to find available pfn range, not remapping identity pages\n");
 416                         xen_set_identity_and_release_chunk(cur_pfn,
 417                                                 cur_pfn + left, nr_pages);
 418                         break;
 419                 }
 420                 /* Adjust size to fit in current e820 RAM region */
 421                 if (size > remap_range_size)
 422                         size = remap_range_size;
 423 
 424                 xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn);
 425 
 426                 /* Update variables to reflect new mappings. */
 427                 i += size;
 428                 remap_pfn += size;
 429         }
 430 
 431         /*
 432          * If the PFNs are currently mapped, the VA mapping also needs
 433          * to be updated to be 1:1.
 434          */
 435         for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
 436                 (void)HYPERVISOR_update_va_mapping(
 437                         (unsigned long)__va(pfn << PAGE_SHIFT),
 438                         mfn_pte(pfn, PAGE_KERNEL_IO), 0);
 439 
 440         return remap_pfn;
 441 }
 442 
 443 static unsigned long __init xen_count_remap_pages(
 444         unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
 445         unsigned long remap_pages)
 446 {
 447         if (start_pfn >= nr_pages)
 448                 return remap_pages;
 449 
 450         return remap_pages + min(end_pfn, nr_pages) - start_pfn;
 451 }
 452 
 453 static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages,
 454         unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn,
 455                               unsigned long nr_pages, unsigned long last_val))
 456 {
 457         phys_addr_t start = 0;
 458         unsigned long ret_val = 0;
 459         const struct e820_entry *entry = xen_e820_table.entries;
 460         int i;
 461 
 462         /*
 463          * Combine non-RAM regions and gaps until a RAM region (or the
 464          * end of the map) is reached, then call the provided function
 465          * to perform its duty on the non-RAM region.
 466          *
 467          * The combined non-RAM regions are rounded to a whole number
 468          * of pages so any partial pages are accessible via the 1:1
 469          * mapping.  This is needed for some BIOSes that put (for
 470          * example) the DMI tables in a reserved region that begins on
 471          * a non-page boundary.
 472          */
 473         for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
 474                 phys_addr_t end = entry->addr + entry->size;
 475                 if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) {
 476                         unsigned long start_pfn = PFN_DOWN(start);
 477                         unsigned long end_pfn = PFN_UP(end);
 478 
 479                         if (entry->type == E820_TYPE_RAM)
 480                                 end_pfn = PFN_UP(entry->addr);
 481 
 482                         if (start_pfn < end_pfn)
 483                                 ret_val = func(start_pfn, end_pfn, nr_pages,
 484                                                ret_val);
 485                         start = end;
 486                 }
 487         }
 488 
 489         return ret_val;
 490 }
 491 
 492 /*
 493  * Remap the memory prepared in xen_do_set_identity_and_remap_chunk().
 494  * The remap information (which mfn remap to which pfn) is contained in the
 495  * to be remapped memory itself in a linked list anchored at xen_remap_mfn.
 496  * This scheme allows to remap the different chunks in arbitrary order while
 497  * the resulting mapping will be independent from the order.
 498  */
 499 void __init xen_remap_memory(void)
 500 {
 501         unsigned long buf = (unsigned long)&xen_remap_buf;
 502         unsigned long mfn_save, pfn;
 503         unsigned long remapped = 0;
 504         unsigned int i;
 505         unsigned long pfn_s = ~0UL;
 506         unsigned long len = 0;
 507 
 508         mfn_save = virt_to_mfn(buf);
 509 
 510         while (xen_remap_mfn != INVALID_P2M_ENTRY) {
 511                 /* Map the remap information */
 512                 set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL);
 513 
 514                 BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]);
 515 
 516                 pfn = xen_remap_buf.target_pfn;
 517                 for (i = 0; i < xen_remap_buf.size; i++) {
 518                         xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]);
 519                         remapped++;
 520                         pfn++;
 521                 }
 522                 if (pfn_s == ~0UL || pfn == pfn_s) {
 523                         pfn_s = xen_remap_buf.target_pfn;
 524                         len += xen_remap_buf.size;
 525                 } else if (pfn_s + len == xen_remap_buf.target_pfn) {
 526                         len += xen_remap_buf.size;
 527                 } else {
 528                         xen_del_extra_mem(pfn_s, len);
 529                         pfn_s = xen_remap_buf.target_pfn;
 530                         len = xen_remap_buf.size;
 531                 }
 532                 xen_remap_mfn = xen_remap_buf.next_area_mfn;
 533         }
 534 
 535         if (pfn_s != ~0UL && len)
 536                 xen_del_extra_mem(pfn_s, len);
 537 
 538         set_pte_mfn(buf, mfn_save, PAGE_KERNEL);
 539 
 540         pr_info("Remapped %ld page(s)\n", remapped);
 541 }
 542 
 543 static unsigned long __init xen_get_pages_limit(void)
 544 {
 545         unsigned long limit;
 546 
 547 #ifdef CONFIG_X86_32
 548         limit = GB(64) / PAGE_SIZE;
 549 #else
 550         limit = MAXMEM / PAGE_SIZE;
 551         if (!xen_initial_domain() && xen_512gb_limit)
 552                 limit = GB(512) / PAGE_SIZE;
 553 #endif
 554         return limit;
 555 }
 556 
 557 static unsigned long __init xen_get_max_pages(void)
 558 {
 559         unsigned long max_pages, limit;
 560         domid_t domid = DOMID_SELF;
 561         long ret;
 562 
 563         limit = xen_get_pages_limit();
 564         max_pages = limit;
 565 
 566         /*
 567          * For the initial domain we use the maximum reservation as
 568          * the maximum page.
 569          *
 570          * For guest domains the current maximum reservation reflects
 571          * the current maximum rather than the static maximum. In this
 572          * case the e820 map provided to us will cover the static
 573          * maximum region.
 574          */
 575         if (xen_initial_domain()) {
 576                 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
 577                 if (ret > 0)
 578                         max_pages = ret;
 579         }
 580 
 581         return min(max_pages, limit);
 582 }
 583 
 584 static void __init xen_align_and_add_e820_region(phys_addr_t start,
 585                                                  phys_addr_t size, int type)
 586 {
 587         phys_addr_t end = start + size;
 588 
 589         /* Align RAM regions to page boundaries. */
 590         if (type == E820_TYPE_RAM) {
 591                 start = PAGE_ALIGN(start);
 592                 end &= ~((phys_addr_t)PAGE_SIZE - 1);
 593 #ifdef CONFIG_MEMORY_HOTPLUG
 594                 /*
 595                  * Don't allow adding memory not in E820 map while booting the
 596                  * system. Once the balloon driver is up it will remove that
 597                  * restriction again.
 598                  */
 599                 max_mem_size = end;
 600 #endif
 601         }
 602 
 603         e820__range_add(start, end - start, type);
 604 }
 605 
 606 static void __init xen_ignore_unusable(void)
 607 {
 608         struct e820_entry *entry = xen_e820_table.entries;
 609         unsigned int i;
 610 
 611         for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) {
 612                 if (entry->type == E820_TYPE_UNUSABLE)
 613                         entry->type = E820_TYPE_RAM;
 614         }
 615 }
 616 
 617 bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size)
 618 {
 619         struct e820_entry *entry;
 620         unsigned mapcnt;
 621         phys_addr_t end;
 622 
 623         if (!size)
 624                 return false;
 625 
 626         end = start + size;
 627         entry = xen_e820_table.entries;
 628 
 629         for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) {
 630                 if (entry->type == E820_TYPE_RAM && entry->addr <= start &&
 631                     (entry->addr + entry->size) >= end)
 632                         return false;
 633 
 634                 entry++;
 635         }
 636 
 637         return true;
 638 }
 639 
 640 /*
 641  * Find a free area in physical memory not yet reserved and compliant with
 642  * E820 map.
 643  * Used to relocate pre-allocated areas like initrd or p2m list which are in
 644  * conflict with the to be used E820 map.
 645  * In case no area is found, return 0. Otherwise return the physical address
 646  * of the area which is already reserved for convenience.
 647  */
 648 phys_addr_t __init xen_find_free_area(phys_addr_t size)
 649 {
 650         unsigned mapcnt;
 651         phys_addr_t addr, start;
 652         struct e820_entry *entry = xen_e820_table.entries;
 653 
 654         for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) {
 655                 if (entry->type != E820_TYPE_RAM || entry->size < size)
 656                         continue;
 657                 start = entry->addr;
 658                 for (addr = start; addr < start + size; addr += PAGE_SIZE) {
 659                         if (!memblock_is_reserved(addr))
 660                                 continue;
 661                         start = addr + PAGE_SIZE;
 662                         if (start + size > entry->addr + entry->size)
 663                                 break;
 664                 }
 665                 if (addr >= start + size) {
 666                         memblock_reserve(start, size);
 667                         return start;
 668                 }
 669         }
 670 
 671         return 0;
 672 }
 673 
 674 /*
 675  * Like memcpy, but with physical addresses for dest and src.
 676  */
 677 static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src,
 678                                    phys_addr_t n)
 679 {
 680         phys_addr_t dest_off, src_off, dest_len, src_len, len;
 681         void *from, *to;
 682 
 683         while (n) {
 684                 dest_off = dest & ~PAGE_MASK;
 685                 src_off = src & ~PAGE_MASK;
 686                 dest_len = n;
 687                 if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off)
 688                         dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off;
 689                 src_len = n;
 690                 if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off)
 691                         src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off;
 692                 len = min(dest_len, src_len);
 693                 to = early_memremap(dest - dest_off, dest_len + dest_off);
 694                 from = early_memremap(src - src_off, src_len + src_off);
 695                 memcpy(to, from, len);
 696                 early_memunmap(to, dest_len + dest_off);
 697                 early_memunmap(from, src_len + src_off);
 698                 n -= len;
 699                 dest += len;
 700                 src += len;
 701         }
 702 }
 703 
 704 /*
 705  * Reserve Xen mfn_list.
 706  */
 707 static void __init xen_reserve_xen_mfnlist(void)
 708 {
 709         phys_addr_t start, size;
 710 
 711         if (xen_start_info->mfn_list >= __START_KERNEL_map) {
 712                 start = __pa(xen_start_info->mfn_list);
 713                 size = PFN_ALIGN(xen_start_info->nr_pages *
 714                                  sizeof(unsigned long));
 715         } else {
 716                 start = PFN_PHYS(xen_start_info->first_p2m_pfn);
 717                 size = PFN_PHYS(xen_start_info->nr_p2m_frames);
 718         }
 719 
 720         memblock_reserve(start, size);
 721         if (!xen_is_e820_reserved(start, size))
 722                 return;
 723 
 724 #ifdef CONFIG_X86_32
 725         /*
 726          * Relocating the p2m on 32 bit system to an arbitrary virtual address
 727          * is not supported, so just give up.
 728          */
 729         xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n");
 730         BUG();
 731 #else
 732         xen_relocate_p2m();
 733         memblock_free(start, size);
 734 #endif
 735 }
 736 
 737 /**
 738  * machine_specific_memory_setup - Hook for machine specific memory setup.
 739  **/
 740 char * __init xen_memory_setup(void)
 741 {
 742         unsigned long max_pfn, pfn_s, n_pfns;
 743         phys_addr_t mem_end, addr, size, chunk_size;
 744         u32 type;
 745         int rc;
 746         struct xen_memory_map memmap;
 747         unsigned long max_pages;
 748         unsigned long extra_pages = 0;
 749         int i;
 750         int op;
 751 
 752         xen_parse_512gb();
 753         max_pfn = xen_get_pages_limit();
 754         max_pfn = min(max_pfn, xen_start_info->nr_pages);
 755         mem_end = PFN_PHYS(max_pfn);
 756 
 757         memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries);
 758         set_xen_guest_handle(memmap.buffer, xen_e820_table.entries);
 759 
 760 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_XEN_BALLOON)
 761         xen_saved_max_mem_size = max_mem_size;
 762 #endif
 763 
 764         op = xen_initial_domain() ?
 765                 XENMEM_machine_memory_map :
 766                 XENMEM_memory_map;
 767         rc = HYPERVISOR_memory_op(op, &memmap);
 768         if (rc == -ENOSYS) {
 769                 BUG_ON(xen_initial_domain());
 770                 memmap.nr_entries = 1;
 771                 xen_e820_table.entries[0].addr = 0ULL;
 772                 xen_e820_table.entries[0].size = mem_end;
 773                 /* 8MB slack (to balance backend allocations). */
 774                 xen_e820_table.entries[0].size += 8ULL << 20;
 775                 xen_e820_table.entries[0].type = E820_TYPE_RAM;
 776                 rc = 0;
 777         }
 778         BUG_ON(rc);
 779         BUG_ON(memmap.nr_entries == 0);
 780         xen_e820_table.nr_entries = memmap.nr_entries;
 781 
 782         /*
 783          * Xen won't allow a 1:1 mapping to be created to UNUSABLE
 784          * regions, so if we're using the machine memory map leave the
 785          * region as RAM as it is in the pseudo-physical map.
 786          *
 787          * UNUSABLE regions in domUs are not handled and will need
 788          * a patch in the future.
 789          */
 790         if (xen_initial_domain())
 791                 xen_ignore_unusable();
 792 
 793         /* Make sure the Xen-supplied memory map is well-ordered. */
 794         e820__update_table(&xen_e820_table);
 795 
 796         max_pages = xen_get_max_pages();
 797 
 798         /* How many extra pages do we need due to remapping? */
 799         max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages);
 800 
 801         if (max_pages > max_pfn)
 802                 extra_pages += max_pages - max_pfn;
 803 
 804         /*
 805          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
 806          * factor the base size.  On non-highmem systems, the base
 807          * size is the full initial memory allocation; on highmem it
 808          * is limited to the max size of lowmem, so that it doesn't
 809          * get completely filled.
 810          *
 811          * Make sure we have no memory above max_pages, as this area
 812          * isn't handled by the p2m management.
 813          *
 814          * In principle there could be a problem in lowmem systems if
 815          * the initial memory is also very large with respect to
 816          * lowmem, but we won't try to deal with that here.
 817          */
 818         extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
 819                            extra_pages, max_pages - max_pfn);
 820         i = 0;
 821         addr = xen_e820_table.entries[0].addr;
 822         size = xen_e820_table.entries[0].size;
 823         while (i < xen_e820_table.nr_entries) {
 824                 bool discard = false;
 825 
 826                 chunk_size = size;
 827                 type = xen_e820_table.entries[i].type;
 828 
 829                 if (type == E820_TYPE_RAM) {
 830                         if (addr < mem_end) {
 831                                 chunk_size = min(size, mem_end - addr);
 832                         } else if (extra_pages) {
 833                                 chunk_size = min(size, PFN_PHYS(extra_pages));
 834                                 pfn_s = PFN_UP(addr);
 835                                 n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s;
 836                                 extra_pages -= n_pfns;
 837                                 xen_add_extra_mem(pfn_s, n_pfns);
 838                                 xen_max_p2m_pfn = pfn_s + n_pfns;
 839                         } else
 840                                 discard = true;
 841                 }
 842 
 843                 if (!discard)
 844                         xen_align_and_add_e820_region(addr, chunk_size, type);
 845 
 846                 addr += chunk_size;
 847                 size -= chunk_size;
 848                 if (size == 0) {
 849                         i++;
 850                         if (i < xen_e820_table.nr_entries) {
 851                                 addr = xen_e820_table.entries[i].addr;
 852                                 size = xen_e820_table.entries[i].size;
 853                         }
 854                 }
 855         }
 856 
 857         /*
 858          * Set the rest as identity mapped, in case PCI BARs are
 859          * located here.
 860          */
 861         set_phys_range_identity(addr / PAGE_SIZE, ~0ul);
 862 
 863         /*
 864          * In domU, the ISA region is normal, usable memory, but we
 865          * reserve ISA memory anyway because too many things poke
 866          * about in there.
 867          */
 868         e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED);
 869 
 870         e820__update_table(e820_table);
 871 
 872         /*
 873          * Check whether the kernel itself conflicts with the target E820 map.
 874          * Failing now is better than running into weird problems later due
 875          * to relocating (and even reusing) pages with kernel text or data.
 876          */
 877         if (xen_is_e820_reserved(__pa_symbol(_text),
 878                         __pa_symbol(__bss_stop) - __pa_symbol(_text))) {
 879                 xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n");
 880                 BUG();
 881         }
 882 
 883         /*
 884          * Check for a conflict of the hypervisor supplied page tables with
 885          * the target E820 map.
 886          */
 887         xen_pt_check_e820();
 888 
 889         xen_reserve_xen_mfnlist();
 890 
 891         /* Check for a conflict of the initrd with the target E820 map. */
 892         if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image,
 893                                  boot_params.hdr.ramdisk_size)) {
 894                 phys_addr_t new_area, start, size;
 895 
 896                 new_area = xen_find_free_area(boot_params.hdr.ramdisk_size);
 897                 if (!new_area) {
 898                         xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n");
 899                         BUG();
 900                 }
 901 
 902                 start = boot_params.hdr.ramdisk_image;
 903                 size = boot_params.hdr.ramdisk_size;
 904                 xen_phys_memcpy(new_area, start, size);
 905                 pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n",
 906                         start, start + size, new_area, new_area + size);
 907                 memblock_free(start, size);
 908                 boot_params.hdr.ramdisk_image = new_area;
 909                 boot_params.ext_ramdisk_image = new_area >> 32;
 910         }
 911 
 912         /*
 913          * Set identity map on non-RAM pages and prepare remapping the
 914          * underlying RAM.
 915          */
 916         xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk);
 917 
 918         pr_info("Released %ld page(s)\n", xen_released_pages);
 919 
 920         return "Xen";
 921 }
 922 
 923 /*
 924  * Set the bit indicating "nosegneg" library variants should be used.
 925  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 926  * can have un-truncated segments, so wrapping around is allowed.
 927  */
 928 static void __init fiddle_vdso(void)
 929 {
 930 #ifdef CONFIG_X86_32
 931         u32 *mask = vdso_image_32.data +
 932                 vdso_image_32.sym_VDSO32_NOTE_MASK;
 933         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 934 #endif
 935 }
 936 
 937 static int register_callback(unsigned type, const void *func)
 938 {
 939         struct callback_register callback = {
 940                 .type = type,
 941                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 942                 .flags = CALLBACKF_mask_events,
 943         };
 944 
 945         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 946 }
 947 
 948 void xen_enable_sysenter(void)
 949 {
 950         int ret;
 951         unsigned sysenter_feature;
 952 
 953 #ifdef CONFIG_X86_32
 954         sysenter_feature = X86_FEATURE_SEP;
 955 #else
 956         sysenter_feature = X86_FEATURE_SYSENTER32;
 957 #endif
 958 
 959         if (!boot_cpu_has(sysenter_feature))
 960                 return;
 961 
 962         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 963         if(ret != 0)
 964                 setup_clear_cpu_cap(sysenter_feature);
 965 }
 966 
 967 void xen_enable_syscall(void)
 968 {
 969 #ifdef CONFIG_X86_64
 970         int ret;
 971 
 972         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 973         if (ret != 0) {
 974                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 975                 /* Pretty fatal; 64-bit userspace has no other
 976                    mechanism for syscalls. */
 977         }
 978 
 979         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 980                 ret = register_callback(CALLBACKTYPE_syscall32,
 981                                         xen_syscall32_target);
 982                 if (ret != 0)
 983                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 984         }
 985 #endif /* CONFIG_X86_64 */
 986 }
 987 
 988 void __init xen_pvmmu_arch_setup(void)
 989 {
 990         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 991         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 992 
 993         HYPERVISOR_vm_assist(VMASST_CMD_enable,
 994                              VMASST_TYPE_pae_extended_cr3);
 995 
 996         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 997             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 998                 BUG();
 999 
1000         xen_enable_sysenter();
1001         xen_enable_syscall();
1002 }
1003 
1004 /* This function is not called for HVM domains */
1005 void __init xen_arch_setup(void)
1006 {
1007         xen_panic_handler_init();
1008         xen_pvmmu_arch_setup();
1009 
1010 #ifdef CONFIG_ACPI
1011         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
1012                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
1013                 disable_acpi();
1014         }
1015 #endif
1016 
1017         memcpy(boot_command_line, xen_start_info->cmd_line,
1018                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
1019                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
1020 
1021         /* Set up idle, making sure it calls safe_halt() pvop */
1022         disable_cpuidle();
1023         disable_cpufreq();
1024         WARN_ON(xen_set_default_idle());
1025         fiddle_vdso();
1026 #ifdef CONFIG_NUMA
1027         numa_off = 1;
1028 #endif
1029 }
/* [<][>][^][v][top][bottom][index][help] */
root/arch/x86/xen/setup.c

DEFINITIONS
