arch/x86/xen/p2m.c

   1 /*
   2  * Xen leaves the responsibility for maintaining p2m mappings to the
   3  * guests themselves, but it must also access and update the p2m array
   4  * during suspend/resume when all the pages are reallocated.
   5  *
   6  * The logical flat p2m table is mapped to a linear kernel memory area.
   7  * For accesses by Xen a three-level tree linked via mfns only is set up to
   8  * allow the address space to be sparse.
   9  *
  10  *               Xen
  11  *                |
  12  *          p2m_top_mfn
  13  *              /   \
  14  * p2m_mid_mfn p2m_mid_mfn
  15  *         /           /
  16  *  p2m p2m p2m ...
  17  *
  18  * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p.
  19  *
  20  * The p2m_top_mfn level is limited to 1 page, so the maximum representable
  21  * pseudo-physical address space is:
  22  *  P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages
  23  *
  24  * P2M_PER_PAGE depends on the architecture, as a mfn is always
  25  * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to
  26  * 512 and 1024 entries respectively.
  27  *
  28  * In short, these structures contain the Machine Frame Number (MFN) of the PFN.
  29  *
  30  * However not all entries are filled with MFNs. Specifically for all other
  31  * leaf entries, or for the top  root, or middle one, for which there is a void
  32  * entry, we assume it is  "missing". So (for example)
  33  *  pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY.
  34  * We have a dedicated page p2m_missing with all entries being
  35  * INVALID_P2M_ENTRY. This page may be referenced multiple times in the p2m
  36  * list/tree in case there are multiple areas with P2M_PER_PAGE invalid pfns.
  37  *
  38  * We also have the possibility of setting 1-1 mappings on certain regions, so
  39  * that:
  40  *  pfn_to_mfn(0xc0000)=0xc0000
  41  *
  42  * The benefit of this is, that we can assume for non-RAM regions (think
  43  * PCI BARs, or ACPI spaces), we can create mappings easily because we
  44  * get the PFN value to match the MFN.
  45  *
  46  * For this to work efficiently we have one new page p2m_identity. All entries
  47  * in p2m_identity are set to INVALID_P2M_ENTRY type (Xen toolstack only
  48  * recognizes that and MFNs, no other fancy value).
  49  *
  50  * On lookup we spot that the entry points to p2m_identity and return the
  51  * identity value instead of dereferencing and returning INVALID_P2M_ENTRY.
  52  * If the entry points to an allocated page, we just proceed as before and
  53  * return the PFN. If the PFN has IDENTITY_FRAME_BIT set we unmask that in
  54  * appropriate functions (pfn_to_mfn).
  55  *
  56  * The reason for having the IDENTITY_FRAME_BIT instead of just returning the
  57  * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a
  58  * non-identity pfn. To protect ourselves against we elect to set (and get) the
  59  * IDENTITY_FRAME_BIT on all identity mapped PFNs.
  60  */
  61
  62 #include <linux/init.h>
  63 #include <linux/module.h>
  64 #include <linux/list.h>
  65 #include <linux/hash.h>
  66 #include <linux/sched.h>
  67 #include <linux/seq_file.h>
  68 #include <linux/bootmem.h>
  69 #include <linux/slab.h>
  70
  71 #include <asm/cache.h>
  72 #include <asm/setup.h>
  73 #include <asm/uaccess.h>
  74
  75 #include <asm/xen/page.h>
  76 #include <asm/xen/hypercall.h>
  77 #include <asm/xen/hypervisor.h>
  78 #include <xen/balloon.h>
  79 #include <xen/grant_table.h>
  80
  81 #include "p2m.h"
  82 #include "multicalls.h"
  83 #include "xen-ops.h"
  84
  85 #define PMDS_PER_MID_PAGE       (P2M_MID_PER_PAGE / PTRS_PER_PTE)
  86
  87 unsigned long *xen_p2m_addr __read_mostly;
  88 EXPORT_SYMBOL_GPL(xen_p2m_addr);
  89 unsigned long xen_p2m_size __read_mostly;
  90 EXPORT_SYMBOL_GPL(xen_p2m_size);
  91 unsigned long xen_max_p2m_pfn __read_mostly;
  92 EXPORT_SYMBOL_GPL(xen_max_p2m_pfn);
  93
  94 static DEFINE_SPINLOCK(p2m_update_lock);
  95
  96 static unsigned long *p2m_mid_missing_mfn;
  97 static unsigned long *p2m_top_mfn;
  98 static unsigned long **p2m_top_mfn_p;
  99 static unsigned long *p2m_missing;
 100 static unsigned long *p2m_identity;
 101 static pte_t *p2m_missing_pte;
 102 static pte_t *p2m_identity_pte;
 103
 104 static inline unsigned p2m_top_index(unsigned long pfn)
 105 {
 106         BUG_ON(pfn >= MAX_P2M_PFN);
 107         return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
 108 }
 109
 110 static inline unsigned p2m_mid_index(unsigned long pfn)
 111 {
 112         return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
 113 }
 114
 115 static inline unsigned p2m_index(unsigned long pfn)
 116 {
 117         return pfn % P2M_PER_PAGE;
 118 }
 119
 120 static void p2m_top_mfn_init(unsigned long *top)
 121 {
 122         unsigned i;
 123
 124         for (i = 0; i < P2M_TOP_PER_PAGE; i++)
 125                 top[i] = virt_to_mfn(p2m_mid_missing_mfn);
 126 }
 127
 128 static void p2m_top_mfn_p_init(unsigned long **top)
 129 {
 130         unsigned i;
 131
 132         for (i = 0; i < P2M_TOP_PER_PAGE; i++)
 133                 top[i] = p2m_mid_missing_mfn;
 134 }
 135
 136 static void p2m_mid_mfn_init(unsigned long *mid, unsigned long *leaf)
 137 {
 138         unsigned i;
 139
 140         for (i = 0; i < P2M_MID_PER_PAGE; i++)
 141                 mid[i] = virt_to_mfn(leaf);
 142 }
 143
 144 static void p2m_init(unsigned long *p2m)
 145 {
 146         unsigned i;
 147
 148         for (i = 0; i < P2M_PER_PAGE; i++)
 149                 p2m[i] = INVALID_P2M_ENTRY;
 150 }
 151
 152 static void p2m_init_identity(unsigned long *p2m, unsigned long pfn)
 153 {
 154         unsigned i;
 155
 156         for (i = 0; i < P2M_PER_PAGE; i++)
 157                 p2m[i] = IDENTITY_FRAME(pfn + i);
 158 }
 159
 160 static void * __ref alloc_p2m_page(void)
 161 {
 162         if (unlikely(!slab_is_available()))
 163                 return alloc_bootmem_align(PAGE_SIZE, PAGE_SIZE);
 164
 165         return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
 166 }
 167
 168 static void __ref free_p2m_page(void *p)
 169 {
 170         if (unlikely(!slab_is_available())) {
 171                 free_bootmem((unsigned long)p, PAGE_SIZE);
 172                 return;
 173         }
 174
 175         free_page((unsigned long)p);
 176 }
 177
 178 /*
 179  * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
 180  *
 181  * This is called both at boot time, and after resuming from suspend:
 182  * - At boot time we're called rather early, and must use alloc_bootmem*()
 183  *   to allocate memory.
 184  *
 185  * - After resume we're called from within stop_machine, but the mfn
 186  *   tree should already be completely allocated.
 187  */
 188 void __ref xen_build_mfn_list_list(void)
 189 {
 190         unsigned long pfn, mfn;
 191         pte_t *ptep;
 192         unsigned int level, topidx, mididx;
 193         unsigned long *mid_mfn_p;
 194
 195         if (xen_feature(XENFEAT_auto_translated_physmap))
 196                 return;
 197
 198         /* Pre-initialize p2m_top_mfn to be completely missing */
 199         if (p2m_top_mfn == NULL) {
 200                 p2m_mid_missing_mfn = alloc_p2m_page();
 201                 p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
 202
 203                 p2m_top_mfn_p = alloc_p2m_page();
 204                 p2m_top_mfn_p_init(p2m_top_mfn_p);
 205
 206                 p2m_top_mfn = alloc_p2m_page();
 207                 p2m_top_mfn_init(p2m_top_mfn);
 208         } else {
 209                 /* Reinitialise, mfn's all change after migration */
 210                 p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
 211         }
 212
 213         for (pfn = 0; pfn < xen_max_p2m_pfn && pfn < MAX_P2M_PFN;
 214              pfn += P2M_PER_PAGE) {
 215                 topidx = p2m_top_index(pfn);
 216                 mididx = p2m_mid_index(pfn);
 217
 218                 mid_mfn_p = p2m_top_mfn_p[topidx];
 219                 ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn),
 220                                       &level);
 221                 BUG_ON(!ptep || level != PG_LEVEL_4K);
 222                 mfn = pte_mfn(*ptep);
 223                 ptep = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
 224
 225                 /* Don't bother allocating any mfn mid levels if
 226                  * they're just missing, just update the stored mfn,
 227                  * since all could have changed over a migrate.
 228                  */
 229                 if (ptep == p2m_missing_pte || ptep == p2m_identity_pte) {
 230                         BUG_ON(mididx);
 231                         BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
 232                         p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
 233                         pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
 234                         continue;
 235                 }
 236
 237                 if (mid_mfn_p == p2m_mid_missing_mfn) {
 238                         mid_mfn_p = alloc_p2m_page();
 239                         p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
 240
 241                         p2m_top_mfn_p[topidx] = mid_mfn_p;
 242                 }
 243
 244                 p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
 245                 mid_mfn_p[mididx] = mfn;
 246         }
 247 }
 248
 249 void xen_setup_mfn_list_list(void)
 250 {
 251         if (xen_feature(XENFEAT_auto_translated_physmap))
 252                 return;
 253
 254         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
 255
 256         HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
 257                 virt_to_mfn(p2m_top_mfn);
 258         HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn;
 259 }
 260
 261 /* Set up p2m_top to point to the domain-builder provided p2m pages */
 262 void __init xen_build_dynamic_phys_to_machine(void)
 263 {
 264         unsigned long pfn;
 265
 266          if (xen_feature(XENFEAT_auto_translated_physmap))
 267                 return;
 268
 269         xen_p2m_addr = (unsigned long *)xen_start_info->mfn_list;
 270         xen_p2m_size = ALIGN(xen_start_info->nr_pages, P2M_PER_PAGE);
 271
 272         for (pfn = xen_start_info->nr_pages; pfn < xen_p2m_size; pfn++)
 273                 xen_p2m_addr[pfn] = INVALID_P2M_ENTRY;
 274
 275         xen_max_p2m_pfn = xen_p2m_size;
 276 }
 277
 278 #define P2M_TYPE_IDENTITY       0
 279 #define P2M_TYPE_MISSING        1
 280 #define P2M_TYPE_PFN            2
 281 #define P2M_TYPE_UNKNOWN        3
 282
 283 static int xen_p2m_elem_type(unsigned long pfn)
 284 {
 285         unsigned long mfn;
 286
 287         if (pfn >= xen_p2m_size)
 288                 return P2M_TYPE_IDENTITY;
 289
 290         mfn = xen_p2m_addr[pfn];
 291
 292         if (mfn == INVALID_P2M_ENTRY)
 293                 return P2M_TYPE_MISSING;
 294
 295         if (mfn & IDENTITY_FRAME_BIT)
 296                 return P2M_TYPE_IDENTITY;
 297
 298         return P2M_TYPE_PFN;
 299 }
 300
 301 static void __init xen_rebuild_p2m_list(unsigned long *p2m)
 302 {
 303         unsigned int i, chunk;
 304         unsigned long pfn;
 305         unsigned long *mfns;
 306         pte_t *ptep;
 307         pmd_t *pmdp;
 308         int type;
 309
 310         p2m_missing = alloc_p2m_page();
 311         p2m_init(p2m_missing);
 312         p2m_identity = alloc_p2m_page();
 313         p2m_init(p2m_identity);
 314
 315         p2m_missing_pte = alloc_p2m_page();
 316         paravirt_alloc_pte(&init_mm, __pa(p2m_missing_pte) >> PAGE_SHIFT);
 317         p2m_identity_pte = alloc_p2m_page();
 318         paravirt_alloc_pte(&init_mm, __pa(p2m_identity_pte) >> PAGE_SHIFT);
 319         for (i = 0; i < PTRS_PER_PTE; i++) {
 320                 set_pte(p2m_missing_pte + i,
 321                         pfn_pte(PFN_DOWN(__pa(p2m_missing)), PAGE_KERNEL_RO));
 322                 set_pte(p2m_identity_pte + i,
 323                         pfn_pte(PFN_DOWN(__pa(p2m_identity)), PAGE_KERNEL_RO));
 324         }
 325
 326         for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += chunk) {
 327                 /*
 328                  * Try to map missing/identity PMDs or p2m-pages if possible.
 329                  * We have to respect the structure of the mfn_list_list
 330                  * which will be built just afterwards.
 331                  * Chunk size to test is one p2m page if we are in the middle
 332                  * of a mfn_list_list mid page and the complete mid page area
 333                  * if we are at index 0 of the mid page. Please note that a
 334                  * mid page might cover more than one PMD, e.g. on 32 bit PAE
 335                  * kernels.
 336                  */
 337                 chunk = (pfn & (P2M_PER_PAGE * P2M_MID_PER_PAGE - 1)) ?
 338                         P2M_PER_PAGE : P2M_PER_PAGE * P2M_MID_PER_PAGE;
 339
 340                 type = xen_p2m_elem_type(pfn);
 341                 i = 0;
 342                 if (type != P2M_TYPE_PFN)
 343                         for (i = 1; i < chunk; i++)
 344                                 if (xen_p2m_elem_type(pfn + i) != type)
 345                                         break;
 346                 if (i < chunk)
 347                         /* Reset to minimal chunk size. */
 348                         chunk = P2M_PER_PAGE;
 349
 350                 if (type == P2M_TYPE_PFN || i < chunk) {
 351                         /* Use initial p2m page contents. */
 352 #ifdef CONFIG_X86_64
 353                         mfns = alloc_p2m_page();
 354                         copy_page(mfns, xen_p2m_addr + pfn);
 355 #else
 356                         mfns = xen_p2m_addr + pfn;
 357 #endif
 358                         ptep = populate_extra_pte((unsigned long)(p2m + pfn));
 359                         set_pte(ptep,
 360                                 pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL));
 361                         continue;
 362                 }
 363
 364                 if (chunk == P2M_PER_PAGE) {
 365                         /* Map complete missing or identity p2m-page. */
 366                         mfns = (type == P2M_TYPE_MISSING) ?
 367                                 p2m_missing : p2m_identity;
 368                         ptep = populate_extra_pte((unsigned long)(p2m + pfn));
 369                         set_pte(ptep,
 370                                 pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL_RO));
 371                         continue;
 372                 }
 373
 374                 /* Complete missing or identity PMD(s) can be mapped. */
 375                 ptep = (type == P2M_TYPE_MISSING) ?
 376                         p2m_missing_pte : p2m_identity_pte;
 377                 for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
 378                         pmdp = populate_extra_pmd(
 379                                 (unsigned long)(p2m + pfn) + i * PMD_SIZE);
 380                         set_pmd(pmdp, __pmd(__pa(ptep) | _KERNPG_TABLE));
 381                 }
 382         }
 383 }
 384
 385 void __init xen_vmalloc_p2m_tree(void)
 386 {
 387         static struct vm_struct vm;
 388
 389         vm.flags = VM_ALLOC;
 390         vm.size = ALIGN(sizeof(unsigned long) * xen_max_p2m_pfn,
 391                         PMD_SIZE * PMDS_PER_MID_PAGE);
 392         vm_area_register_early(&vm, PMD_SIZE * PMDS_PER_MID_PAGE);
 393         pr_notice("p2m virtual area at %p, size is %lx\n", vm.addr, vm.size);
 394
 395         xen_max_p2m_pfn = vm.size / sizeof(unsigned long);
 396
 397         xen_rebuild_p2m_list(vm.addr);
 398
 399         xen_p2m_addr = vm.addr;
 400         xen_p2m_size = xen_max_p2m_pfn;
 401
 402         xen_inv_extra_mem();
 403 }
 404
 405 unsigned long get_phys_to_machine(unsigned long pfn)
 406 {
 407         pte_t *ptep;
 408         unsigned int level;
 409
 410         if (unlikely(pfn >= xen_p2m_size)) {
 411                 if (pfn < xen_max_p2m_pfn)
 412                         return xen_chk_extra_mem(pfn);
 413
 414                 return IDENTITY_FRAME(pfn);
 415         }
 416
 417         ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
 418         BUG_ON(!ptep || level != PG_LEVEL_4K);
 419
 420         /*
 421          * The INVALID_P2M_ENTRY is filled in both p2m_*identity
 422          * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
 423          * would be wrong.
 424          */
 425         if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
 426                 return IDENTITY_FRAME(pfn);
 427
 428         return xen_p2m_addr[pfn];
 429 }
 430 EXPORT_SYMBOL_GPL(get_phys_to_machine);
 431
 432 /*
 433  * Allocate new pmd(s). It is checked whether the old pmd is still in place.
 434  * If not, nothing is changed. This is okay as the only reason for allocating
 435  * a new pmd is to replace p2m_missing_pte or p2m_identity_pte by a individual
 436  * pmd. In case of PAE/x86-32 there are multiple pmds to allocate!
 437  */
 438 static pte_t *alloc_p2m_pmd(unsigned long addr, pte_t *pte_pg)
 439 {
 440         pte_t *ptechk;
 441         pte_t *pte_newpg[PMDS_PER_MID_PAGE];
 442         pmd_t *pmdp;
 443         unsigned int level;
 444         unsigned long flags;
 445         unsigned long vaddr;
 446         int i;
 447
 448         /* Do all allocations first to bail out in error case. */
 449         for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
 450                 pte_newpg[i] = alloc_p2m_page();
 451                 if (!pte_newpg[i]) {
 452                         for (i--; i >= 0; i--)
 453                                 free_p2m_page(pte_newpg[i]);
 454
 455                         return NULL;
 456                 }
 457         }
 458
 459         vaddr = addr & ~(PMD_SIZE * PMDS_PER_MID_PAGE - 1);
 460
 461         for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
 462                 copy_page(pte_newpg[i], pte_pg);
 463                 paravirt_alloc_pte(&init_mm, __pa(pte_newpg[i]) >> PAGE_SHIFT);
 464
 465                 pmdp = lookup_pmd_address(vaddr);
 466                 BUG_ON(!pmdp);
 467
 468                 spin_lock_irqsave(&p2m_update_lock, flags);
 469
 470                 ptechk = lookup_address(vaddr, &level);
 471                 if (ptechk == pte_pg) {
 472                         set_pmd(pmdp,
 473                                 __pmd(__pa(pte_newpg[i]) | _KERNPG_TABLE));
 474                         pte_newpg[i] = NULL;
 475                 }
 476
 477                 spin_unlock_irqrestore(&p2m_update_lock, flags);
 478
 479                 if (pte_newpg[i]) {
 480                         paravirt_release_pte(__pa(pte_newpg[i]) >> PAGE_SHIFT);
 481                         free_p2m_page(pte_newpg[i]);
 482                 }
 483
 484                 vaddr += PMD_SIZE;
 485         }
 486
 487         return lookup_address(addr, &level);
 488 }
 489
 490 /*
 491  * Fully allocate the p2m structure for a given pfn.  We need to check
 492  * that both the top and mid levels are allocated, and make sure the
 493  * parallel mfn tree is kept in sync.  We may race with other cpus, so
 494  * the new pages are installed with cmpxchg; if we lose the race then
 495  * simply free the page we allocated and use the one that's there.
 496  */
 497 static bool alloc_p2m(unsigned long pfn)
 498 {
 499         unsigned topidx, mididx;
 500         unsigned long *top_mfn_p, *mid_mfn;
 501         pte_t *ptep, *pte_pg;
 502         unsigned int level;
 503         unsigned long flags;
 504         unsigned long addr = (unsigned long)(xen_p2m_addr + pfn);
 505         unsigned long p2m_pfn;
 506
 507         topidx = p2m_top_index(pfn);
 508         mididx = p2m_mid_index(pfn);
 509
 510         ptep = lookup_address(addr, &level);
 511         BUG_ON(!ptep || level != PG_LEVEL_4K);
 512         pte_pg = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
 513
 514         if (pte_pg == p2m_missing_pte || pte_pg == p2m_identity_pte) {
 515                 /* PMD level is missing, allocate a new one */
 516                 ptep = alloc_p2m_pmd(addr, pte_pg);
 517                 if (!ptep)
 518                         return false;
 519         }
 520
 521         if (p2m_top_mfn) {
 522                 top_mfn_p = &p2m_top_mfn[topidx];
 523                 mid_mfn = ACCESS_ONCE(p2m_top_mfn_p[topidx]);
 524
 525                 BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
 526
 527                 if (mid_mfn == p2m_mid_missing_mfn) {
 528                         /* Separately check the mid mfn level */
 529                         unsigned long missing_mfn;
 530                         unsigned long mid_mfn_mfn;
 531                         unsigned long old_mfn;
 532
 533                         mid_mfn = alloc_p2m_page();
 534                         if (!mid_mfn)
 535                                 return false;
 536
 537                         p2m_mid_mfn_init(mid_mfn, p2m_missing);
 538
 539                         missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
 540                         mid_mfn_mfn = virt_to_mfn(mid_mfn);
 541                         old_mfn = cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn);
 542                         if (old_mfn != missing_mfn) {
 543                                 free_p2m_page(mid_mfn);
 544                                 mid_mfn = mfn_to_virt(old_mfn);
 545                         } else {
 546                                 p2m_top_mfn_p[topidx] = mid_mfn;
 547                         }
 548                 }
 549         } else {
 550                 mid_mfn = NULL;
 551         }
 552
 553         p2m_pfn = pte_pfn(ACCESS_ONCE(*ptep));
 554         if (p2m_pfn == PFN_DOWN(__pa(p2m_identity)) ||
 555             p2m_pfn == PFN_DOWN(__pa(p2m_missing))) {
 556                 /* p2m leaf page is missing */
 557                 unsigned long *p2m;
 558
 559                 p2m = alloc_p2m_page();
 560                 if (!p2m)
 561                         return false;
 562
 563                 if (p2m_pfn == PFN_DOWN(__pa(p2m_missing)))
 564                         p2m_init(p2m);
 565                 else
 566                         p2m_init_identity(p2m, pfn);
 567
 568                 spin_lock_irqsave(&p2m_update_lock, flags);
 569
 570                 if (pte_pfn(*ptep) == p2m_pfn) {
 571                         set_pte(ptep,
 572                                 pfn_pte(PFN_DOWN(__pa(p2m)), PAGE_KERNEL));
 573                         if (mid_mfn)
 574                                 mid_mfn[mididx] = virt_to_mfn(p2m);
 575                         p2m = NULL;
 576                 }
 577
 578                 spin_unlock_irqrestore(&p2m_update_lock, flags);
 579
 580                 if (p2m)
 581                         free_p2m_page(p2m);
 582         }
 583
 584         return true;
 585 }
 586
 587 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
 588                                       unsigned long pfn_e)
 589 {
 590         unsigned long pfn;
 591
 592         if (unlikely(pfn_s >= xen_p2m_size))
 593                 return 0;
 594
 595         if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
 596                 return pfn_e - pfn_s;
 597
 598         if (pfn_s > pfn_e)
 599                 return 0;
 600
 601         if (pfn_e > xen_p2m_size)
 602                 pfn_e = xen_p2m_size;
 603
 604         for (pfn = pfn_s; pfn < pfn_e; pfn++)
 605                 xen_p2m_addr[pfn] = IDENTITY_FRAME(pfn);
 606
 607         return pfn - pfn_s;
 608 }
 609
 610 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
 611 {
 612         pte_t *ptep;
 613         unsigned int level;
 614
 615         /* don't track P2M changes in autotranslate guests */
 616         if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
 617                 return true;
 618
 619         if (unlikely(pfn >= xen_p2m_size)) {
 620                 BUG_ON(mfn != INVALID_P2M_ENTRY);
 621                 return true;
 622         }
 623
 624         if (likely(!xen_safe_write_ulong(xen_p2m_addr + pfn, mfn)))
 625                 return true;
 626
 627         ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
 628         BUG_ON(!ptep || level != PG_LEVEL_4K);
 629
 630         if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_missing)))
 631                 return mfn == INVALID_P2M_ENTRY;
 632
 633         if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
 634                 return mfn == IDENTITY_FRAME(pfn);
 635
 636         return false;
 637 }
 638
 639 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
 640 {
 641         if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
 642                 if (!alloc_p2m(pfn))
 643                         return false;
 644
 645                 return __set_phys_to_machine(pfn, mfn);
 646         }
 647
 648         return true;
 649 }
 650
 651 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
 652                             struct gnttab_map_grant_ref *kmap_ops,
 653                             struct page **pages, unsigned int count)
 654 {
 655         int i, ret = 0;
 656         pte_t *pte;
 657
 658         if (xen_feature(XENFEAT_auto_translated_physmap))
 659                 return 0;
 660
 661         if (kmap_ops) {
 662                 ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref,
 663                                                 kmap_ops, count);
 664                 if (ret)
 665                         goto out;
 666         }
 667
 668         for (i = 0; i < count; i++) {
 669                 unsigned long mfn, pfn;
 670
 671                 /* Do not add to override if the map failed. */
 672                 if (map_ops[i].status)
 673                         continue;
 674
 675                 if (map_ops[i].flags & GNTMAP_contains_pte) {
 676                         pte = (pte_t *)(mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
 677                                 (map_ops[i].host_addr & ~PAGE_MASK));
 678                         mfn = pte_mfn(*pte);
 679                 } else {
 680                         mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
 681                 }
 682                 pfn = page_to_pfn(pages[i]);
 683
 684                 WARN(pfn_to_mfn(pfn) != INVALID_P2M_ENTRY, "page must be ballooned");
 685
 686                 if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) {
 687                         ret = -ENOMEM;
 688                         goto out;
 689                 }
 690         }
 691
 692 out:
 693         return ret;
 694 }
 695 EXPORT_SYMBOL_GPL(set_foreign_p2m_mapping);
 696
 697 int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
 698                               struct gnttab_unmap_grant_ref *kunmap_ops,
 699                               struct page **pages, unsigned int count)
 700 {
 701         int i, ret = 0;
 702
 703         if (xen_feature(XENFEAT_auto_translated_physmap))
 704                 return 0;
 705
 706         for (i = 0; i < count; i++) {
 707                 unsigned long mfn = __pfn_to_mfn(page_to_pfn(pages[i]));
 708                 unsigned long pfn = page_to_pfn(pages[i]);
 709
 710                 if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) {
 711                         ret = -EINVAL;
 712                         goto out;
 713                 }
 714
 715                 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 716         }
 717         if (kunmap_ops)
 718                 ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
 719                                                 kunmap_ops, count);
 720 out:
 721         return ret;
 722 }
 723 EXPORT_SYMBOL_GPL(clear_foreign_p2m_mapping);
 724
 725 #ifdef CONFIG_XEN_DEBUG_FS
 726 #include <linux/debugfs.h>
 727 #include "debugfs.h"
 728 static int p2m_dump_show(struct seq_file *m, void *v)
 729 {
 730         static const char * const type_name[] = {
 731                                 [P2M_TYPE_IDENTITY] = "identity",
 732                                 [P2M_TYPE_MISSING] = "missing",
 733                                 [P2M_TYPE_PFN] = "pfn",
 734                                 [P2M_TYPE_UNKNOWN] = "abnormal"};
 735         unsigned long pfn, first_pfn;
 736         int type, prev_type;
 737
 738         prev_type = xen_p2m_elem_type(0);
 739         first_pfn = 0;
 740
 741         for (pfn = 0; pfn < xen_p2m_size; pfn++) {
 742                 type = xen_p2m_elem_type(pfn);
 743                 if (type != prev_type) {
 744                         seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
 745                                    type_name[prev_type]);
 746                         prev_type = type;
 747                         first_pfn = pfn;
 748                 }
 749         }
 750         seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
 751                    type_name[prev_type]);
 752         return 0;
 753 }
 754
 755 static int p2m_dump_open(struct inode *inode, struct file *filp)
 756 {
 757         return single_open(filp, p2m_dump_show, NULL);
 758 }
 759
 760 static const struct file_operations p2m_dump_fops = {
 761         .open           = p2m_dump_open,
 762         .read           = seq_read,
 763         .llseek         = seq_lseek,
 764         .release        = single_release,
 765 };
 766
 767 static struct dentry *d_mmu_debug;
 768
 769 static int __init xen_p2m_debugfs(void)
 770 {
 771         struct dentry *d_xen = xen_init_debugfs();
 772
 773         if (d_xen == NULL)
 774                 return -ENOMEM;
 775
 776         d_mmu_debug = debugfs_create_dir("mmu", d_xen);
 777
 778         debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
 779         return 0;
 780 }
 781 fs_initcall(xen_p2m_debugfs);
 782 #endif /* CONFIG_XEN_DEBUG_FS */