kernel/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 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG_LIMIT
  95 #define P2M_LIMIT CONFIG_XEN_BALLOON_MEMORY_HOTPLUG_LIMIT
  96 #else
  97 #define P2M_LIMIT 0
  98 #endif
  99
 100 static DEFINE_SPINLOCK(p2m_update_lock);
 101
 102 static unsigned long *p2m_mid_missing_mfn;
 103 static unsigned long *p2m_top_mfn;
 104 static unsigned long **p2m_top_mfn_p;
 105 static unsigned long *p2m_missing;
 106 static unsigned long *p2m_identity;
 107 static pte_t *p2m_missing_pte;
 108 static pte_t *p2m_identity_pte;
 109
 110 static inline unsigned p2m_top_index(unsigned long pfn)
 111 {
 112         BUG_ON(pfn >= MAX_P2M_PFN);
 113         return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
 114 }
 115
 116 static inline unsigned p2m_mid_index(unsigned long pfn)
 117 {
 118         return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
 119 }
 120
 121 static inline unsigned p2m_index(unsigned long pfn)
 122 {
 123         return pfn % P2M_PER_PAGE;
 124 }
 125
 126 static void p2m_top_mfn_init(unsigned long *top)
 127 {
 128         unsigned i;
 129
 130         for (i = 0; i < P2M_TOP_PER_PAGE; i++)
 131                 top[i] = virt_to_mfn(p2m_mid_missing_mfn);
 132 }
 133
 134 static void p2m_top_mfn_p_init(unsigned long **top)
 135 {
 136         unsigned i;
 137
 138         for (i = 0; i < P2M_TOP_PER_PAGE; i++)
 139                 top[i] = p2m_mid_missing_mfn;
 140 }
 141
 142 static void p2m_mid_mfn_init(unsigned long *mid, unsigned long *leaf)
 143 {
 144         unsigned i;
 145
 146         for (i = 0; i < P2M_MID_PER_PAGE; i++)
 147                 mid[i] = virt_to_mfn(leaf);
 148 }
 149
 150 static void p2m_init(unsigned long *p2m)
 151 {
 152         unsigned i;
 153
 154         for (i = 0; i < P2M_PER_PAGE; i++)
 155                 p2m[i] = INVALID_P2M_ENTRY;
 156 }
 157
 158 static void p2m_init_identity(unsigned long *p2m, unsigned long pfn)
 159 {
 160         unsigned i;
 161
 162         for (i = 0; i < P2M_PER_PAGE; i++)
 163                 p2m[i] = IDENTITY_FRAME(pfn + i);
 164 }
 165
 166 static void * __ref alloc_p2m_page(void)
 167 {
 168         if (unlikely(!slab_is_available()))
 169                 return alloc_bootmem_align(PAGE_SIZE, PAGE_SIZE);
 170
 171         return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
 172 }
 173
 174 static void __ref free_p2m_page(void *p)
 175 {
 176         if (unlikely(!slab_is_available())) {
 177                 free_bootmem((unsigned long)p, PAGE_SIZE);
 178                 return;
 179         }
 180
 181         free_page((unsigned long)p);
 182 }
 183
 184 /*
 185  * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
 186  *
 187  * This is called both at boot time, and after resuming from suspend:
 188  * - At boot time we're called rather early, and must use alloc_bootmem*()
 189  *   to allocate memory.
 190  *
 191  * - After resume we're called from within stop_machine, but the mfn
 192  *   tree should already be completely allocated.
 193  */
 194 void __ref xen_build_mfn_list_list(void)
 195 {
 196         unsigned long pfn, mfn;
 197         pte_t *ptep;
 198         unsigned int level, topidx, mididx;
 199         unsigned long *mid_mfn_p;
 200
 201         if (xen_feature(XENFEAT_auto_translated_physmap))
 202                 return;
 203
 204         /* Pre-initialize p2m_top_mfn to be completely missing */
 205         if (p2m_top_mfn == NULL) {
 206                 p2m_mid_missing_mfn = alloc_p2m_page();
 207                 p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
 208
 209                 p2m_top_mfn_p = alloc_p2m_page();
 210                 p2m_top_mfn_p_init(p2m_top_mfn_p);
 211
 212                 p2m_top_mfn = alloc_p2m_page();
 213                 p2m_top_mfn_init(p2m_top_mfn);
 214         } else {
 215                 /* Reinitialise, mfn's all change after migration */
 216                 p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
 217         }
 218
 219         for (pfn = 0; pfn < xen_max_p2m_pfn && pfn < MAX_P2M_PFN;
 220              pfn += P2M_PER_PAGE) {
 221                 topidx = p2m_top_index(pfn);
 222                 mididx = p2m_mid_index(pfn);
 223
 224                 mid_mfn_p = p2m_top_mfn_p[topidx];
 225                 ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn),
 226                                       &level);
 227                 BUG_ON(!ptep || level != PG_LEVEL_4K);
 228                 mfn = pte_mfn(*ptep);
 229                 ptep = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
 230
 231                 /* Don't bother allocating any mfn mid levels if
 232                  * they're just missing, just update the stored mfn,
 233                  * since all could have changed over a migrate.
 234                  */
 235                 if (ptep == p2m_missing_pte || ptep == p2m_identity_pte) {
 236                         BUG_ON(mididx);
 237                         BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
 238                         p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
 239                         pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
 240                         continue;
 241                 }
 242
 243                 if (mid_mfn_p == p2m_mid_missing_mfn) {
 244                         mid_mfn_p = alloc_p2m_page();
 245                         p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
 246
 247                         p2m_top_mfn_p[topidx] = mid_mfn_p;
 248                 }
 249
 250                 p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
 251                 mid_mfn_p[mididx] = mfn;
 252         }
 253 }
 254
 255 void xen_setup_mfn_list_list(void)
 256 {
 257         if (xen_feature(XENFEAT_auto_translated_physmap))
 258                 return;
 259
 260         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
 261
 262         HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
 263                 virt_to_mfn(p2m_top_mfn);
 264         HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn;
 265 }
 266
 267 /* Set up p2m_top to point to the domain-builder provided p2m pages */
 268 void __init xen_build_dynamic_phys_to_machine(void)
 269 {
 270         unsigned long pfn;
 271
 272          if (xen_feature(XENFEAT_auto_translated_physmap))
 273                 return;
 274
 275         xen_p2m_addr = (unsigned long *)xen_start_info->mfn_list;
 276         xen_p2m_size = ALIGN(xen_start_info->nr_pages, P2M_PER_PAGE);
 277
 278         for (pfn = xen_start_info->nr_pages; pfn < xen_p2m_size; pfn++)
 279                 xen_p2m_addr[pfn] = INVALID_P2M_ENTRY;
 280
 281         xen_max_p2m_pfn = xen_p2m_size;
 282 }
 283
 284 #define P2M_TYPE_IDENTITY       0
 285 #define P2M_TYPE_MISSING        1
 286 #define P2M_TYPE_PFN            2
 287 #define P2M_TYPE_UNKNOWN        3
 288
 289 static int xen_p2m_elem_type(unsigned long pfn)
 290 {
 291         unsigned long mfn;
 292
 293         if (pfn >= xen_p2m_size)
 294                 return P2M_TYPE_IDENTITY;
 295
 296         mfn = xen_p2m_addr[pfn];
 297
 298         if (mfn == INVALID_P2M_ENTRY)
 299                 return P2M_TYPE_MISSING;
 300
 301         if (mfn & IDENTITY_FRAME_BIT)
 302                 return P2M_TYPE_IDENTITY;
 303
 304         return P2M_TYPE_PFN;
 305 }
 306
 307 static void __init xen_rebuild_p2m_list(unsigned long *p2m)
 308 {
 309         unsigned int i, chunk;
 310         unsigned long pfn;
 311         unsigned long *mfns;
 312         pte_t *ptep;
 313         pmd_t *pmdp;
 314         int type;
 315
 316         p2m_missing = alloc_p2m_page();
 317         p2m_init(p2m_missing);
 318         p2m_identity = alloc_p2m_page();
 319         p2m_init(p2m_identity);
 320
 321         p2m_missing_pte = alloc_p2m_page();
 322         paravirt_alloc_pte(&init_mm, __pa(p2m_missing_pte) >> PAGE_SHIFT);
 323         p2m_identity_pte = alloc_p2m_page();
 324         paravirt_alloc_pte(&init_mm, __pa(p2m_identity_pte) >> PAGE_SHIFT);
 325         for (i = 0; i < PTRS_PER_PTE; i++) {
 326                 set_pte(p2m_missing_pte + i,
 327                         pfn_pte(PFN_DOWN(__pa(p2m_missing)), PAGE_KERNEL_RO));
 328                 set_pte(p2m_identity_pte + i,
 329                         pfn_pte(PFN_DOWN(__pa(p2m_identity)), PAGE_KERNEL_RO));
 330         }
 331
 332         for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += chunk) {
 333                 /*
 334                  * Try to map missing/identity PMDs or p2m-pages if possible.
 335                  * We have to respect the structure of the mfn_list_list
 336                  * which will be built just afterwards.
 337                  * Chunk size to test is one p2m page if we are in the middle
 338                  * of a mfn_list_list mid page and the complete mid page area
 339                  * if we are at index 0 of the mid page. Please note that a
 340                  * mid page might cover more than one PMD, e.g. on 32 bit PAE
 341                  * kernels.
 342                  */
 343                 chunk = (pfn & (P2M_PER_PAGE * P2M_MID_PER_PAGE - 1)) ?
 344                         P2M_PER_PAGE : P2M_PER_PAGE * P2M_MID_PER_PAGE;
 345
 346                 type = xen_p2m_elem_type(pfn);
 347                 i = 0;
 348                 if (type != P2M_TYPE_PFN)
 349                         for (i = 1; i < chunk; i++)
 350                                 if (xen_p2m_elem_type(pfn + i) != type)
 351                                         break;
 352                 if (i < chunk)
 353                         /* Reset to minimal chunk size. */
 354                         chunk = P2M_PER_PAGE;
 355
 356                 if (type == P2M_TYPE_PFN || i < chunk) {
 357                         /* Use initial p2m page contents. */
 358 #ifdef CONFIG_X86_64
 359                         mfns = alloc_p2m_page();
 360                         copy_page(mfns, xen_p2m_addr + pfn);
 361 #else
 362                         mfns = xen_p2m_addr + pfn;
 363 #endif
 364                         ptep = populate_extra_pte((unsigned long)(p2m + pfn));
 365                         set_pte(ptep,
 366                                 pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL));
 367                         continue;
 368                 }
 369
 370                 if (chunk == P2M_PER_PAGE) {
 371                         /* Map complete missing or identity p2m-page. */
 372                         mfns = (type == P2M_TYPE_MISSING) ?
 373                                 p2m_missing : p2m_identity;
 374                         ptep = populate_extra_pte((unsigned long)(p2m + pfn));
 375                         set_pte(ptep,
 376                                 pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL_RO));
 377                         continue;
 378                 }
 379
 380                 /* Complete missing or identity PMD(s) can be mapped. */
 381                 ptep = (type == P2M_TYPE_MISSING) ?
 382                         p2m_missing_pte : p2m_identity_pte;
 383                 for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
 384                         pmdp = populate_extra_pmd(
 385                                 (unsigned long)(p2m + pfn) + i * PMD_SIZE);
 386                         set_pmd(pmdp, __pmd(__pa(ptep) | _KERNPG_TABLE));
 387                 }
 388         }
 389 }
 390
 391 void __init xen_vmalloc_p2m_tree(void)
 392 {
 393         static struct vm_struct vm;
 394         unsigned long p2m_limit;
 395
 396         p2m_limit = (phys_addr_t)P2M_LIMIT * 1024 * 1024 * 1024 / PAGE_SIZE;
 397         vm.flags = VM_ALLOC;
 398         vm.size = ALIGN(sizeof(unsigned long) * max(xen_max_p2m_pfn, p2m_limit),
 399                         PMD_SIZE * PMDS_PER_MID_PAGE);
 400         vm_area_register_early(&vm, PMD_SIZE * PMDS_PER_MID_PAGE);
 401         pr_notice("p2m virtual area at %p, size is %lx\n", vm.addr, vm.size);
 402
 403         xen_max_p2m_pfn = vm.size / sizeof(unsigned long);
 404
 405         xen_rebuild_p2m_list(vm.addr);
 406
 407         xen_p2m_addr = vm.addr;
 408         xen_p2m_size = xen_max_p2m_pfn;
 409
 410         xen_inv_extra_mem();
 411 }
 412
 413 unsigned long get_phys_to_machine(unsigned long pfn)
 414 {
 415         pte_t *ptep;
 416         unsigned int level;
 417
 418         if (unlikely(pfn >= xen_p2m_size)) {
 419                 if (pfn < xen_max_p2m_pfn)
 420                         return xen_chk_extra_mem(pfn);
 421
 422                 return IDENTITY_FRAME(pfn);
 423         }
 424
 425         ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
 426         BUG_ON(!ptep || level != PG_LEVEL_4K);
 427
 428         /*
 429          * The INVALID_P2M_ENTRY is filled in both p2m_*identity
 430          * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
 431          * would be wrong.
 432          */
 433         if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
 434                 return IDENTITY_FRAME(pfn);
 435
 436         return xen_p2m_addr[pfn];
 437 }
 438 EXPORT_SYMBOL_GPL(get_phys_to_machine);
 439
 440 /*
 441  * Allocate new pmd(s). It is checked whether the old pmd is still in place.
 442  * If not, nothing is changed. This is okay as the only reason for allocating
 443  * a new pmd is to replace p2m_missing_pte or p2m_identity_pte by a individual
 444  * pmd. In case of PAE/x86-32 there are multiple pmds to allocate!
 445  */
 446 static pte_t *alloc_p2m_pmd(unsigned long addr, pte_t *pte_pg)
 447 {
 448         pte_t *ptechk;
 449         pte_t *pte_newpg[PMDS_PER_MID_PAGE];
 450         pmd_t *pmdp;
 451         unsigned int level;
 452         unsigned long flags;
 453         unsigned long vaddr;
 454         int i;
 455
 456         /* Do all allocations first to bail out in error case. */
 457         for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
 458                 pte_newpg[i] = alloc_p2m_page();
 459                 if (!pte_newpg[i]) {
 460                         for (i--; i >= 0; i--)
 461                                 free_p2m_page(pte_newpg[i]);
 462
 463                         return NULL;
 464                 }
 465         }
 466
 467         vaddr = addr & ~(PMD_SIZE * PMDS_PER_MID_PAGE - 1);
 468
 469         for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
 470                 copy_page(pte_newpg[i], pte_pg);
 471                 paravirt_alloc_pte(&init_mm, __pa(pte_newpg[i]) >> PAGE_SHIFT);
 472
 473                 pmdp = lookup_pmd_address(vaddr);
 474                 BUG_ON(!pmdp);
 475
 476                 spin_lock_irqsave(&p2m_update_lock, flags);
 477
 478                 ptechk = lookup_address(vaddr, &level);
 479                 if (ptechk == pte_pg) {
 480                         set_pmd(pmdp,
 481                                 __pmd(__pa(pte_newpg[i]) | _KERNPG_TABLE));
 482                         pte_newpg[i] = NULL;
 483                 }
 484
 485                 spin_unlock_irqrestore(&p2m_update_lock, flags);
 486
 487                 if (pte_newpg[i]) {
 488                         paravirt_release_pte(__pa(pte_newpg[i]) >> PAGE_SHIFT);
 489                         free_p2m_page(pte_newpg[i]);
 490                 }
 491
 492                 vaddr += PMD_SIZE;
 493         }
 494
 495         return lookup_address(addr, &level);
 496 }
 497
 498 /*
 499  * Fully allocate the p2m structure for a given pfn.  We need to check
 500  * that both the top and mid levels are allocated, and make sure the
 501  * parallel mfn tree is kept in sync.  We may race with other cpus, so
 502  * the new pages are installed with cmpxchg; if we lose the race then
 503  * simply free the page we allocated and use the one that's there.
 504  */
 505 static bool alloc_p2m(unsigned long pfn)
 506 {
 507         unsigned topidx, mididx;
 508         unsigned long *top_mfn_p, *mid_mfn;
 509         pte_t *ptep, *pte_pg;
 510         unsigned int level;
 511         unsigned long flags;
 512         unsigned long addr = (unsigned long)(xen_p2m_addr + pfn);
 513         unsigned long p2m_pfn;
 514
 515         topidx = p2m_top_index(pfn);
 516         mididx = p2m_mid_index(pfn);
 517
 518         ptep = lookup_address(addr, &level);
 519         BUG_ON(!ptep || level != PG_LEVEL_4K);
 520         pte_pg = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
 521
 522         if (pte_pg == p2m_missing_pte || pte_pg == p2m_identity_pte) {
 523                 /* PMD level is missing, allocate a new one */
 524                 ptep = alloc_p2m_pmd(addr, pte_pg);
 525                 if (!ptep)
 526                         return false;
 527         }
 528
 529         if (p2m_top_mfn) {
 530                 top_mfn_p = &p2m_top_mfn[topidx];
 531                 mid_mfn = ACCESS_ONCE(p2m_top_mfn_p[topidx]);
 532
 533                 BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
 534
 535                 if (mid_mfn == p2m_mid_missing_mfn) {
 536                         /* Separately check the mid mfn level */
 537                         unsigned long missing_mfn;
 538                         unsigned long mid_mfn_mfn;
 539                         unsigned long old_mfn;
 540
 541                         mid_mfn = alloc_p2m_page();
 542                         if (!mid_mfn)
 543                                 return false;
 544
 545                         p2m_mid_mfn_init(mid_mfn, p2m_missing);
 546
 547                         missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
 548                         mid_mfn_mfn = virt_to_mfn(mid_mfn);
 549                         old_mfn = cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn);
 550                         if (old_mfn != missing_mfn) {
 551                                 free_p2m_page(mid_mfn);
 552                                 mid_mfn = mfn_to_virt(old_mfn);
 553                         } else {
 554                                 p2m_top_mfn_p[topidx] = mid_mfn;
 555                         }
 556                 }
 557         } else {
 558                 mid_mfn = NULL;
 559         }
 560
 561         p2m_pfn = pte_pfn(READ_ONCE(*ptep));
 562         if (p2m_pfn == PFN_DOWN(__pa(p2m_identity)) ||
 563             p2m_pfn == PFN_DOWN(__pa(p2m_missing))) {
 564                 /* p2m leaf page is missing */
 565                 unsigned long *p2m;
 566
 567                 p2m = alloc_p2m_page();
 568                 if (!p2m)
 569                         return false;
 570
 571                 if (p2m_pfn == PFN_DOWN(__pa(p2m_missing)))
 572                         p2m_init(p2m);
 573                 else
 574                         p2m_init_identity(p2m, pfn & ~(P2M_PER_PAGE - 1));
 575
 576                 spin_lock_irqsave(&p2m_update_lock, flags);
 577
 578                 if (pte_pfn(*ptep) == p2m_pfn) {
 579                         set_pte(ptep,
 580                                 pfn_pte(PFN_DOWN(__pa(p2m)), PAGE_KERNEL));
 581                         if (mid_mfn)
 582                                 mid_mfn[mididx] = virt_to_mfn(p2m);
 583                         p2m = NULL;
 584                 }
 585
 586                 spin_unlock_irqrestore(&p2m_update_lock, flags);
 587
 588                 if (p2m)
 589                         free_p2m_page(p2m);
 590         }
 591
 592         return true;
 593 }
 594
 595 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
 596                                       unsigned long pfn_e)
 597 {
 598         unsigned long pfn;
 599
 600         if (unlikely(pfn_s >= xen_p2m_size))
 601                 return 0;
 602
 603         if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
 604                 return pfn_e - pfn_s;
 605
 606         if (pfn_s > pfn_e)
 607                 return 0;
 608
 609         if (pfn_e > xen_p2m_size)
 610                 pfn_e = xen_p2m_size;
 611
 612         for (pfn = pfn_s; pfn < pfn_e; pfn++)
 613                 xen_p2m_addr[pfn] = IDENTITY_FRAME(pfn);
 614
 615         return pfn - pfn_s;
 616 }
 617
 618 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
 619 {
 620         pte_t *ptep;
 621         unsigned int level;
 622
 623         /* don't track P2M changes in autotranslate guests */
 624         if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
 625                 return true;
 626
 627         if (unlikely(pfn >= xen_p2m_size)) {
 628                 BUG_ON(mfn != INVALID_P2M_ENTRY);
 629                 return true;
 630         }
 631
 632         if (likely(!xen_safe_write_ulong(xen_p2m_addr + pfn, mfn)))
 633                 return true;
 634
 635         ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
 636         BUG_ON(!ptep || level != PG_LEVEL_4K);
 637
 638         if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_missing)))
 639                 return mfn == INVALID_P2M_ENTRY;
 640
 641         if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
 642                 return mfn == IDENTITY_FRAME(pfn);
 643
 644         return false;
 645 }
 646
 647 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
 648 {
 649         if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
 650                 if (!alloc_p2m(pfn))
 651                         return false;
 652
 653                 return __set_phys_to_machine(pfn, mfn);
 654         }
 655
 656         return true;
 657 }
 658
 659 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
 660                             struct gnttab_map_grant_ref *kmap_ops,
 661                             struct page **pages, unsigned int count)
 662 {
 663         int i, ret = 0;
 664         pte_t *pte;
 665
 666         if (xen_feature(XENFEAT_auto_translated_physmap))
 667                 return 0;
 668
 669         if (kmap_ops) {
 670                 ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref,
 671                                                 kmap_ops, count);
 672                 if (ret)
 673                         goto out;
 674         }
 675
 676         for (i = 0; i < count; i++) {
 677                 unsigned long mfn, pfn;
 678
 679                 /* Do not add to override if the map failed. */
 680                 if (map_ops[i].status)
 681                         continue;
 682
 683                 if (map_ops[i].flags & GNTMAP_contains_pte) {
 684                         pte = (pte_t *)(mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
 685                                 (map_ops[i].host_addr & ~PAGE_MASK));
 686                         mfn = pte_mfn(*pte);
 687                 } else {
 688                         mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
 689                 }
 690                 pfn = page_to_pfn(pages[i]);
 691
 692                 WARN(pfn_to_mfn(pfn) != INVALID_P2M_ENTRY, "page must be ballooned");
 693
 694                 if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) {
 695                         ret = -ENOMEM;
 696                         goto out;
 697                 }
 698         }
 699
 700 out:
 701         return ret;
 702 }
 703 EXPORT_SYMBOL_GPL(set_foreign_p2m_mapping);
 704
 705 int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
 706                               struct gnttab_unmap_grant_ref *kunmap_ops,
 707                               struct page **pages, unsigned int count)
 708 {
 709         int i, ret = 0;
 710
 711         if (xen_feature(XENFEAT_auto_translated_physmap))
 712                 return 0;
 713
 714         for (i = 0; i < count; i++) {
 715                 unsigned long mfn = __pfn_to_mfn(page_to_pfn(pages[i]));
 716                 unsigned long pfn = page_to_pfn(pages[i]);
 717
 718                 if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) {
 719                         ret = -EINVAL;
 720                         goto out;
 721                 }
 722
 723                 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 724         }
 725         if (kunmap_ops)
 726                 ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
 727                                                 kunmap_ops, count);
 728 out:
 729         return ret;
 730 }
 731 EXPORT_SYMBOL_GPL(clear_foreign_p2m_mapping);
 732
 733 #ifdef CONFIG_XEN_DEBUG_FS
 734 #include <linux/debugfs.h>
 735 #include "debugfs.h"
 736 static int p2m_dump_show(struct seq_file *m, void *v)
 737 {
 738         static const char * const type_name[] = {
 739                                 [P2M_TYPE_IDENTITY] = "identity",
 740                                 [P2M_TYPE_MISSING] = "missing",
 741                                 [P2M_TYPE_PFN] = "pfn",
 742                                 [P2M_TYPE_UNKNOWN] = "abnormal"};
 743         unsigned long pfn, first_pfn;
 744         int type, prev_type;
 745
 746         prev_type = xen_p2m_elem_type(0);
 747         first_pfn = 0;
 748
 749         for (pfn = 0; pfn < xen_p2m_size; pfn++) {
 750                 type = xen_p2m_elem_type(pfn);
 751                 if (type != prev_type) {
 752                         seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
 753                                    type_name[prev_type]);
 754                         prev_type = type;
 755                         first_pfn = pfn;
 756                 }
 757         }
 758         seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
 759                    type_name[prev_type]);
 760         return 0;
 761 }
 762
 763 static int p2m_dump_open(struct inode *inode, struct file *filp)
 764 {
 765         return single_open(filp, p2m_dump_show, NULL);
 766 }
 767
 768 static const struct file_operations p2m_dump_fops = {
 769         .open           = p2m_dump_open,
 770         .read           = seq_read,
 771         .llseek         = seq_lseek,
 772         .release        = single_release,
 773 };
 774
 775 static struct dentry *d_mmu_debug;
 776
 777 static int __init xen_p2m_debugfs(void)
 778 {
 779         struct dentry *d_xen = xen_init_debugfs();
 780
 781         if (d_xen == NULL)
 782                 return -ENOMEM;
 783
 784         d_mmu_debug = debugfs_create_dir("mmu", d_xen);
 785
 786         debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
 787         return 0;
 788 }
 789 fs_initcall(xen_p2m_debugfs);
 790 #endif /* CONFIG_XEN_DEBUG_FS */