arch/x86/xen/setup.c

   1 /*
   2  * Machine specific setup for xen
   3  *
   4  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
   5  */
   6
   7 #include <linux/module.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.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-ops.h"
  31 #include "vdso.h"
  32
  33 /* These are code, but not functions.  Defined in entry.S */
  34 extern const char xen_hypervisor_callback[];
  35 extern const char xen_failsafe_callback[];
  36 #ifdef CONFIG_X86_64
  37 extern const char nmi[];
  38 #endif
  39 extern void xen_sysenter_target(void);
  40 extern void xen_syscall_target(void);
  41 extern void xen_syscall32_target(void);
  42
  43 /* Amount of extra memory space we add to the e820 ranges */
  44 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
  45
  46 /* Number of pages released from the initial allocation. */
  47 unsigned long xen_released_pages;
  48
  49 /*
  50  * The maximum amount of extra memory compared to the base size.  The
  51  * main scaling factor is the size of struct page.  At extreme ratios
  52  * of base:extra, all the base memory can be filled with page
  53  * structures for the extra memory, leaving no space for anything
  54  * else.
  55  *
  56  * 10x seems like a reasonable balance between scaling flexibility and
  57  * leaving a practically usable system.
  58  */
  59 #define EXTRA_MEM_RATIO         (10)
  60
  61 static void __init xen_add_extra_mem(u64 start, u64 size)
  62 {
  63         unsigned long pfn;
  64         int i;
  65
  66         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
  67                 /* Add new region. */
  68                 if (xen_extra_mem[i].size == 0) {
  69                         xen_extra_mem[i].start = start;
  70                         xen_extra_mem[i].size  = size;
  71                         break;
  72                 }
  73                 /* Append to existing region. */
  74                 if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
  75                         xen_extra_mem[i].size += size;
  76                         break;
  77                 }
  78         }
  79         if (i == XEN_EXTRA_MEM_MAX_REGIONS)
  80                 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
  81
  82         memblock_reserve(start, size);
  83
  84         xen_max_p2m_pfn = PFN_DOWN(start + size);
  85         for (pfn = PFN_DOWN(start); pfn < xen_max_p2m_pfn; pfn++) {
  86                 unsigned long mfn = pfn_to_mfn(pfn);
  87
  88                 if (WARN(mfn == pfn, "Trying to over-write 1-1 mapping (pfn: %lx)\n", pfn))
  89                         continue;
  90                 WARN(mfn != INVALID_P2M_ENTRY, "Trying to remove %lx which has %lx mfn!\n",
  91                         pfn, mfn);
  92
  93                 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
  94         }
  95 }
  96
  97 static unsigned long __init xen_do_chunk(unsigned long start,
  98                                          unsigned long end, bool release)
  99 {
 100         struct xen_memory_reservation reservation = {
 101                 .address_bits = 0,
 102                 .extent_order = 0,
 103                 .domid        = DOMID_SELF
 104         };
 105         unsigned long len = 0;
 106         unsigned long pfn;
 107         int ret;
 108
 109         for (pfn = start; pfn < end; pfn++) {
 110                 unsigned long frame;
 111                 unsigned long mfn = pfn_to_mfn(pfn);
 112
 113                 if (release) {
 114                         /* Make sure pfn exists to start with */
 115                         if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
 116                                 continue;
 117                         frame = mfn;
 118                 } else {
 119                         if (mfn != INVALID_P2M_ENTRY)
 120                                 continue;
 121                         frame = pfn;
 122                 }
 123                 set_xen_guest_handle(reservation.extent_start, &frame);
 124                 reservation.nr_extents = 1;
 125
 126                 ret = HYPERVISOR_memory_op(release ? XENMEM_decrease_reservation : XENMEM_populate_physmap,
 127                                            &reservation);
 128                 WARN(ret != 1, "Failed to %s pfn %lx err=%d\n",
 129                      release ? "release" : "populate", pfn, ret);
 130
 131                 if (ret == 1) {
 132                         if (!early_set_phys_to_machine(pfn, release ? INVALID_P2M_ENTRY : frame)) {
 133                                 if (release)
 134                                         break;
 135                                 set_xen_guest_handle(reservation.extent_start, &frame);
 136                                 reservation.nr_extents = 1;
 137                                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
 138                                                            &reservation);
 139                                 break;
 140                         }
 141                         len++;
 142                 } else
 143                         break;
 144         }
 145         if (len)
 146                 printk(KERN_INFO "%s %lx-%lx pfn range: %lu pages %s\n",
 147                        release ? "Freeing" : "Populating",
 148                        start, end, len,
 149                        release ? "freed" : "added");
 150
 151         return len;
 152 }
 153
 154 static unsigned long __init xen_release_chunk(unsigned long start,
 155                                               unsigned long end)
 156 {
 157         return xen_do_chunk(start, end, true);
 158 }
 159
 160 static unsigned long __init xen_populate_chunk(
 161         const struct e820entry *list, size_t map_size,
 162         unsigned long max_pfn, unsigned long *last_pfn,
 163         unsigned long credits_left)
 164 {
 165         const struct e820entry *entry;
 166         unsigned int i;
 167         unsigned long done = 0;
 168         unsigned long dest_pfn;
 169
 170         for (i = 0, entry = list; i < map_size; i++, entry++) {
 171                 unsigned long s_pfn;
 172                 unsigned long e_pfn;
 173                 unsigned long pfns;
 174                 long capacity;
 175
 176                 if (credits_left <= 0)
 177                         break;
 178
 179                 if (entry->type != E820_RAM)
 180                         continue;
 181
 182                 e_pfn = PFN_DOWN(entry->addr + entry->size);
 183
 184                 /* We only care about E820 after the xen_start_info->nr_pages */
 185                 if (e_pfn <= max_pfn)
 186                         continue;
 187
 188                 s_pfn = PFN_UP(entry->addr);
 189                 /* If the E820 falls within the nr_pages, we want to start
 190                  * at the nr_pages PFN.
 191                  * If that would mean going past the E820 entry, skip it
 192                  */
 193                 if (s_pfn <= max_pfn) {
 194                         capacity = e_pfn - max_pfn;
 195                         dest_pfn = max_pfn;
 196                 } else {
 197                         capacity = e_pfn - s_pfn;
 198                         dest_pfn = s_pfn;
 199                 }
 200
 201                 if (credits_left < capacity)
 202                         capacity = credits_left;
 203
 204                 pfns = xen_do_chunk(dest_pfn, dest_pfn + capacity, false);
 205                 done += pfns;
 206                 *last_pfn = (dest_pfn + pfns);
 207                 if (pfns < capacity)
 208                         break;
 209                 credits_left -= pfns;
 210         }
 211         return done;
 212 }
 213
 214 static void __init xen_set_identity_and_release_chunk(
 215         unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
 216         unsigned long *released, unsigned long *identity)
 217 {
 218         unsigned long pfn;
 219
 220         /*
 221          * If the PFNs are currently mapped, clear the mappings
 222          * (except for the ISA region which must be 1:1 mapped) to
 223          * release the refcounts (in Xen) on the original frames.
 224          */
 225         for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++) {
 226                 pte_t pte = __pte_ma(0);
 227
 228                 if (pfn < PFN_UP(ISA_END_ADDRESS))
 229                         pte = mfn_pte(pfn, PAGE_KERNEL_IO);
 230
 231                 (void)HYPERVISOR_update_va_mapping(
 232                         (unsigned long)__va(pfn << PAGE_SHIFT), pte, 0);
 233         }
 234
 235         if (start_pfn < nr_pages)
 236                 *released += xen_release_chunk(
 237                         start_pfn, min(end_pfn, nr_pages));
 238
 239         *identity += set_phys_range_identity(start_pfn, end_pfn);
 240 }
 241
 242 static unsigned long __init xen_set_identity_and_release(
 243         const struct e820entry *list, size_t map_size, unsigned long nr_pages)
 244 {
 245         phys_addr_t start = 0;
 246         unsigned long released = 0;
 247         unsigned long identity = 0;
 248         const struct e820entry *entry;
 249         int i;
 250
 251         /*
 252          * Combine non-RAM regions and gaps until a RAM region (or the
 253          * end of the map) is reached, then set the 1:1 map and
 254          * release the pages (if available) in those non-RAM regions.
 255          *
 256          * The combined non-RAM regions are rounded to a whole number
 257          * of pages so any partial pages are accessible via the 1:1
 258          * mapping.  This is needed for some BIOSes that put (for
 259          * example) the DMI tables in a reserved region that begins on
 260          * a non-page boundary.
 261          */
 262         for (i = 0, entry = list; i < map_size; i++, entry++) {
 263                 phys_addr_t end = entry->addr + entry->size;
 264                 if (entry->type == E820_RAM || i == map_size - 1) {
 265                         unsigned long start_pfn = PFN_DOWN(start);
 266                         unsigned long end_pfn = PFN_UP(end);
 267
 268                         if (entry->type == E820_RAM)
 269                                 end_pfn = PFN_UP(entry->addr);
 270
 271                         if (start_pfn < end_pfn)
 272                                 xen_set_identity_and_release_chunk(
 273                                         start_pfn, end_pfn, nr_pages,
 274                                         &released, &identity);
 275
 276                         start = end;
 277                 }
 278         }
 279
 280         if (released)
 281                 printk(KERN_INFO "Released %lu pages of unused memory\n", released);
 282         if (identity)
 283                 printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity);
 284
 285         return released;
 286 }
 287
 288 static unsigned long __init xen_get_max_pages(void)
 289 {
 290         unsigned long max_pages = MAX_DOMAIN_PAGES;
 291         domid_t domid = DOMID_SELF;
 292         int ret;
 293
 294         /*
 295          * For the initial domain we use the maximum reservation as
 296          * the maximum page.
 297          *
 298          * For guest domains the current maximum reservation reflects
 299          * the current maximum rather than the static maximum. In this
 300          * case the e820 map provided to us will cover the static
 301          * maximum region.
 302          */
 303         if (xen_initial_domain()) {
 304                 ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
 305                 if (ret > 0)
 306                         max_pages = ret;
 307         }
 308
 309         return min(max_pages, MAX_DOMAIN_PAGES);
 310 }
 311
 312 static void xen_align_and_add_e820_region(u64 start, u64 size, int type)
 313 {
 314         u64 end = start + size;
 315
 316         /* Align RAM regions to page boundaries. */
 317         if (type == E820_RAM) {
 318                 start = PAGE_ALIGN(start);
 319                 end &= ~((u64)PAGE_SIZE - 1);
 320         }
 321
 322         e820_add_region(start, end - start, type);
 323 }
 324
 325 void xen_ignore_unusable(struct e820entry *list, size_t map_size)
 326 {
 327         struct e820entry *entry;
 328         unsigned int i;
 329
 330         for (i = 0, entry = list; i < map_size; i++, entry++) {
 331                 if (entry->type == E820_UNUSABLE)
 332                         entry->type = E820_RAM;
 333         }
 334 }
 335
 336 /**
 337  * machine_specific_memory_setup - Hook for machine specific memory setup.
 338  **/
 339 char * __init xen_memory_setup(void)
 340 {
 341         static struct e820entry map[E820MAX] __initdata;
 342
 343         unsigned long max_pfn = xen_start_info->nr_pages;
 344         unsigned long long mem_end;
 345         int rc;
 346         struct xen_memory_map memmap;
 347         unsigned long max_pages;
 348         unsigned long last_pfn = 0;
 349         unsigned long extra_pages = 0;
 350         unsigned long populated;
 351         int i;
 352         int op;
 353
 354         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
 355         mem_end = PFN_PHYS(max_pfn);
 356
 357         memmap.nr_entries = E820MAX;
 358         set_xen_guest_handle(memmap.buffer, map);
 359
 360         op = xen_initial_domain() ?
 361                 XENMEM_machine_memory_map :
 362                 XENMEM_memory_map;
 363         rc = HYPERVISOR_memory_op(op, &memmap);
 364         if (rc == -ENOSYS) {
 365                 BUG_ON(xen_initial_domain());
 366                 memmap.nr_entries = 1;
 367                 map[0].addr = 0ULL;
 368                 map[0].size = mem_end;
 369                 /* 8MB slack (to balance backend allocations). */
 370                 map[0].size += 8ULL << 20;
 371                 map[0].type = E820_RAM;
 372                 rc = 0;
 373         }
 374         BUG_ON(rc);
 375
 376         /*
 377          * Xen won't allow a 1:1 mapping to be created to UNUSABLE
 378          * regions, so if we're using the machine memory map leave the
 379          * region as RAM as it is in the pseudo-physical map.
 380          *
 381          * UNUSABLE regions in domUs are not handled and will need
 382          * a patch in the future.
 383          */
 384         if (xen_initial_domain())
 385                 xen_ignore_unusable(map, memmap.nr_entries);
 386
 387         /* Make sure the Xen-supplied memory map is well-ordered. */
 388         sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
 389
 390         max_pages = xen_get_max_pages();
 391         if (max_pages > max_pfn)
 392                 extra_pages += max_pages - max_pfn;
 393
 394         /*
 395          * Set P2M for all non-RAM pages and E820 gaps to be identity
 396          * type PFNs.  Any RAM pages that would be made inaccesible by
 397          * this are first released.
 398          */
 399         xen_released_pages = xen_set_identity_and_release(
 400                 map, memmap.nr_entries, max_pfn);
 401
 402         /*
 403          * Populate back the non-RAM pages and E820 gaps that had been
 404          * released. */
 405         populated = xen_populate_chunk(map, memmap.nr_entries,
 406                         max_pfn, &last_pfn, xen_released_pages);
 407
 408         xen_released_pages -= populated;
 409         extra_pages += xen_released_pages;
 410
 411         if (last_pfn > max_pfn) {
 412                 max_pfn = min(MAX_DOMAIN_PAGES, last_pfn);
 413                 mem_end = PFN_PHYS(max_pfn);
 414         }
 415         /*
 416          * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
 417          * factor the base size.  On non-highmem systems, the base
 418          * size is the full initial memory allocation; on highmem it
 419          * is limited to the max size of lowmem, so that it doesn't
 420          * get completely filled.
 421          *
 422          * In principle there could be a problem in lowmem systems if
 423          * the initial memory is also very large with respect to
 424          * lowmem, but we won't try to deal with that here.
 425          */
 426         extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
 427                           extra_pages);
 428         i = 0;
 429         while (i < memmap.nr_entries) {
 430                 u64 addr = map[i].addr;
 431                 u64 size = map[i].size;
 432                 u32 type = map[i].type;
 433
 434                 if (type == E820_RAM) {
 435                         if (addr < mem_end) {
 436                                 size = min(size, mem_end - addr);
 437                         } else if (extra_pages) {
 438                                 size = min(size, (u64)extra_pages * PAGE_SIZE);
 439                                 extra_pages -= size / PAGE_SIZE;
 440                                 xen_add_extra_mem(addr, size);
 441                         } else
 442                                 type = E820_UNUSABLE;
 443                 }
 444
 445                 xen_align_and_add_e820_region(addr, size, type);
 446
 447                 map[i].addr += size;
 448                 map[i].size -= size;
 449                 if (map[i].size == 0)
 450                         i++;
 451         }
 452
 453         /*
 454          * In domU, the ISA region is normal, usable memory, but we
 455          * reserve ISA memory anyway because too many things poke
 456          * about in there.
 457          */
 458         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
 459                         E820_RESERVED);
 460
 461         /*
 462          * Reserve Xen bits:
 463          *  - mfn_list
 464          *  - xen_start_info
 465          * See comment above "struct start_info" in <xen/interface/xen.h>
 466          * We tried to make the the memblock_reserve more selective so
 467          * that it would be clear what region is reserved. Sadly we ran
 468          * in the problem wherein on a 64-bit hypervisor with a 32-bit
 469          * initial domain, the pt_base has the cr3 value which is not
 470          * neccessarily where the pagetable starts! As Jan put it: "
 471          * Actually, the adjustment turns out to be correct: The page
 472          * tables for a 32-on-64 dom0 get allocated in the order "first L1",
 473          * "first L2", "first L3", so the offset to the page table base is
 474          * indeed 2. When reading xen/include/public/xen.h's comment
 475          * very strictly, this is not a violation (since there nothing is said
 476          * that the first thing in the page table space is pointed to by
 477          * pt_base; I admit that this seems to be implied though, namely
 478          * do I think that it is implied that the page table space is the
 479          * range [pt_base, pt_base + nt_pt_frames), whereas that
 480          * range here indeed is [pt_base - 2, pt_base - 2 + nt_pt_frames),
 481          * which - without a priori knowledge - the kernel would have
 482          * difficulty to figure out)." - so lets just fall back to the
 483          * easy way and reserve the whole region.
 484          */
 485         memblock_reserve(__pa(xen_start_info->mfn_list),
 486                          xen_start_info->pt_base - xen_start_info->mfn_list);
 487
 488         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 489
 490         return "Xen";
 491 }
 492
 493 /*
 494  * Set the bit indicating "nosegneg" library variants should be used.
 495  * We only need to bother in pure 32-bit mode; compat 32-bit processes
 496  * can have un-truncated segments, so wrapping around is allowed.
 497  */
 498 static void __init fiddle_vdso(void)
 499 {
 500 #ifdef CONFIG_X86_32
 501         u32 *mask;
 502         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
 503         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 504         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
 505         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
 506 #endif
 507 }
 508
 509 static int register_callback(unsigned type, const void *func)
 510 {
 511         struct callback_register callback = {
 512                 .type = type,
 513                 .address = XEN_CALLBACK(__KERNEL_CS, func),
 514                 .flags = CALLBACKF_mask_events,
 515         };
 516
 517         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
 518 }
 519
 520 void xen_enable_sysenter(void)
 521 {
 522         int ret;
 523         unsigned sysenter_feature;
 524
 525 #ifdef CONFIG_X86_32
 526         sysenter_feature = X86_FEATURE_SEP;
 527 #else
 528         sysenter_feature = X86_FEATURE_SYSENTER32;
 529 #endif
 530
 531         if (!boot_cpu_has(sysenter_feature))
 532                 return;
 533
 534         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
 535         if(ret != 0)
 536                 setup_clear_cpu_cap(sysenter_feature);
 537 }
 538
 539 void xen_enable_syscall(void)
 540 {
 541 #ifdef CONFIG_X86_64
 542         int ret;
 543
 544         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
 545         if (ret != 0) {
 546                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
 547                 /* Pretty fatal; 64-bit userspace has no other
 548                    mechanism for syscalls. */
 549         }
 550
 551         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
 552                 ret = register_callback(CALLBACKTYPE_syscall32,
 553                                         xen_syscall32_target);
 554                 if (ret != 0)
 555                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 556         }
 557 #endif /* CONFIG_X86_64 */
 558 }
 559 void xen_enable_nmi(void)
 560 {
 561 #ifdef CONFIG_X86_64
 562         if (register_callback(CALLBACKTYPE_nmi, nmi))
 563                 BUG();
 564 #endif
 565 }
 566 void __init xen_arch_setup(void)
 567 {
 568         xen_panic_handler_init();
 569
 570         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 571         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
 572
 573         if (!xen_feature(XENFEAT_auto_translated_physmap))
 574                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
 575                                      VMASST_TYPE_pae_extended_cr3);
 576
 577         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
 578             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
 579                 BUG();
 580
 581         xen_enable_sysenter();
 582         xen_enable_syscall();
 583         xen_enable_nmi();
 584 #ifdef CONFIG_ACPI
 585         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
 586                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
 587                 disable_acpi();
 588         }
 589 #endif
 590
 591         memcpy(boot_command_line, xen_start_info->cmd_line,
 592                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
 593                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
 594
 595         /* Set up idle, making sure it calls safe_halt() pvop */
 596         disable_cpuidle();
 597         disable_cpufreq();
 598         WARN_ON(xen_set_default_idle());
 599         fiddle_vdso();
 600 #ifdef CONFIG_NUMA
 601         numa_off = 1;
 602 #endif
 603 }