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 p2m table is logically a flat array, but we implement it as a 7 * three-level tree to allow the address space to be sparse. 8 * 9 * Xen 10 * | 11 * p2m_top p2m_top_mfn 12 * / \ / \ 13 * p2m_mid p2m_mid p2m_mid_mfn p2m_mid_mfn 14 * / \ / \ / / 15 * p2m p2m p2m p2m p2m p2m p2m ... 16 * 17 * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p. 18 * 19 * The p2m_top and p2m_top_mfn levels are limited to 1 page, so the 20 * maximum representable pseudo-physical address space is: 21 * P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages 22 * 23 * P2M_PER_PAGE depends on the architecture, as a mfn is always 24 * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to 25 * 512 and 1024 entries respectively. 26 * 27 * In short, these structures contain the Machine Frame Number (MFN) of the PFN. 28 * 29 * However not all entries are filled with MFNs. Specifically for all other 30 * leaf entries, or for the top root, or middle one, for which there is a void 31 * entry, we assume it is "missing". So (for example) 32 * pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY. 33 * 34 * We also have the possibility of setting 1-1 mappings on certain regions, so 35 * that: 36 * pfn_to_mfn(0xc0000)=0xc0000 37 * 38 * The benefit of this is, that we can assume for non-RAM regions (think 39 * PCI BARs, or ACPI spaces), we can create mappings easily b/c we 40 * get the PFN value to match the MFN. 41 * 42 * For this to work efficiently we have one new page p2m_identity and 43 * allocate (via reserved_brk) any other pages we need to cover the sides 44 * (1GB or 4MB boundary violations). All entries in p2m_identity are set to 45 * INVALID_P2M_ENTRY type (Xen toolstack only recognizes that and MFNs, 46 * no other fancy value). 47 * 48 * On lookup we spot that the entry points to p2m_identity and return the 49 * identity value instead of dereferencing and returning INVALID_P2M_ENTRY. 50 * If the entry points to an allocated page, we just proceed as before and 51 * return the PFN. If the PFN has IDENTITY_FRAME_BIT set we unmask that in 52 * appropriate functions (pfn_to_mfn). 53 * 54 * The reason for having the IDENTITY_FRAME_BIT instead of just returning the 55 * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a 56 * non-identity pfn. To protect ourselves against we elect to set (and get) the 57 * IDENTITY_FRAME_BIT on all identity mapped PFNs. 58 * 59 * This simplistic diagram is used to explain the more subtle piece of code. 60 * There is also a digram of the P2M at the end that can help. 61 * Imagine your E820 looking as so: 62 * 63 * 1GB 2GB 64 * /-------------------+---------\/----\ /----------\ /---+-----\ 65 * | System RAM | Sys RAM ||ACPI| | reserved | | Sys RAM | 66 * \-------------------+---------/\----/ \----------/ \---+-----/ 67 * ^- 1029MB ^- 2001MB 68 * 69 * [1029MB = 263424 (0x40500), 2001MB = 512256 (0x7D100), 70 * 2048MB = 524288 (0x80000)] 71 * 72 * And dom0_mem=max:3GB,1GB is passed in to the guest, meaning memory past 1GB 73 * is actually not present (would have to kick the balloon driver to put it in). 74 * 75 * When we are told to set the PFNs for identity mapping (see patch: "xen/setup: 76 * Set identity mapping for non-RAM E820 and E820 gaps.") we pass in the start 77 * of the PFN and the end PFN (263424 and 512256 respectively). The first step 78 * is to reserve_brk a top leaf page if the p2m[1] is missing. The top leaf page 79 * covers 512^2 of page estate (1GB) and in case the start or end PFN is not 80 * aligned on 512^2*PAGE_SIZE (1GB) we loop on aligned 1GB PFNs from start pfn 81 * to end pfn. We reserve_brk top leaf pages if they are missing (means they 82 * point to p2m_mid_missing). 83 * 84 * With the E820 example above, 263424 is not 1GB aligned so we allocate a 85 * reserve_brk page which will cover the PFNs estate from 0x40000 to 0x80000. 86 * Each entry in the allocate page is "missing" (points to p2m_missing). 87 * 88 * Next stage is to determine if we need to do a more granular boundary check 89 * on the 4MB (or 2MB depending on architecture) off the start and end pfn's. 90 * We check if the start pfn and end pfn violate that boundary check, and if 91 * so reserve_brk a middle (p2m[x][y]) leaf page. This way we have a much finer 92 * granularity of setting which PFNs are missing and which ones are identity. 93 * In our example 263424 and 512256 both fail the check so we reserve_brk two 94 * pages. Populate them with INVALID_P2M_ENTRY (so they both have "missing" 95 * values) and assign them to p2m[1][2] and p2m[1][488] respectively. 96 * 97 * At this point we would at minimum reserve_brk one page, but could be up to 98 * three. Each call to set_phys_range_identity has at maximum a three page 99 * cost. If we were to query the P2M at this stage, all those entries from 100 * start PFN through end PFN (so 1029MB -> 2001MB) would return 101 * INVALID_P2M_ENTRY ("missing"). 102 * 103 * The next step is to walk from the start pfn to the end pfn setting 104 * the IDENTITY_FRAME_BIT on each PFN. This is done in set_phys_range_identity. 105 * If we find that the middle leaf is pointing to p2m_missing we can swap it 106 * over to p2m_identity - this way covering 4MB (or 2MB) PFN space. At this 107 * point we do not need to worry about boundary aligment (so no need to 108 * reserve_brk a middle page, figure out which PFNs are "missing" and which 109 * ones are identity), as that has been done earlier. If we find that the 110 * middle leaf is not occupied by p2m_identity or p2m_missing, we dereference 111 * that page (which covers 512 PFNs) and set the appropriate PFN with 112 * IDENTITY_FRAME_BIT. In our example 263424 and 512256 end up there, and we 113 * set from p2m[1][2][256->511] and p2m[1][488][0->256] with 114 * IDENTITY_FRAME_BIT set. 115 * 116 * All other regions that are void (or not filled) either point to p2m_missing 117 * (considered missing) or have the default value of INVALID_P2M_ENTRY (also 118 * considered missing). In our case, p2m[1][2][0->255] and p2m[1][488][257->511] 119 * contain the INVALID_P2M_ENTRY value and are considered "missing." 120 * 121 * This is what the p2m ends up looking (for the E820 above) with this 122 * fabulous drawing: 123 * 124 * p2m /--------------\ 125 * /-----\ | &mfn_list[0],| /-----------------\ 126 * | 0 |------>| &mfn_list[1],| /---------------\ | ~0, ~0, .. | 127 * |-----| | ..., ~0, ~0 | | ~0, ~0, [x]---+----->| IDENTITY [@256] | 128 * | 1 |---\ \--------------/ | [p2m_identity]+\ | IDENTITY [@257] | 129 * |-----| \ | [p2m_identity]+\\ | .... | 130 * | 2 |--\ \-------------------->| ... | \\ \----------------/ 131 * |-----| \ \---------------/ \\ 132 * | 3 |\ \ \\ p2m_identity 133 * |-----| \ \-------------------->/---------------\ /-----------------\ 134 * | .. +->+ | [p2m_identity]+-->| ~0, ~0, ~0, ... | 135 * \-----/ / | [p2m_identity]+-->| ..., ~0 | 136 * / /---------------\ | .... | \-----------------/ 137 * / | IDENTITY[@0] | /-+-[x], ~0, ~0.. | 138 * / | IDENTITY[@256]|<----/ \---------------/ 139 * / | ~0, ~0, .... | 140 * | \---------------/ 141 * | 142 * p2m_mid_missing p2m_missing 143 * /-----------------\ /------------\ 144 * | [p2m_missing] +---->| ~0, ~0, ~0 | 145 * | [p2m_missing] +---->| ..., ~0 | 146 * \-----------------/ \------------/ 147 * 148 * where ~0 is INVALID_P2M_ENTRY. IDENTITY is (PFN | IDENTITY_BIT) 149 */ 150 151 #include <linux/init.h> 152 #include <linux/module.h> 153 #include <linux/list.h> 154 #include <linux/hash.h> 155 #include <linux/sched.h> 156 #include <linux/seq_file.h> 157 158 #include <asm/cache.h> 159 #include <asm/setup.h> 160 161 #include <asm/xen/page.h> 162 #include <asm/xen/hypercall.h> 163 #include <asm/xen/hypervisor.h> 164 #include <xen/grant_table.h> 165 166 #include "multicalls.h" 167 #include "xen-ops.h" 168 169 static void __init m2p_override_init(void); 170 171 unsigned long xen_max_p2m_pfn __read_mostly; 172 173 #define P2M_PER_PAGE (PAGE_SIZE / sizeof(unsigned long)) 174 #define P2M_MID_PER_PAGE (PAGE_SIZE / sizeof(unsigned long *)) 175 #define P2M_TOP_PER_PAGE (PAGE_SIZE / sizeof(unsigned long **)) 176 177 #define MAX_P2M_PFN (P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE) 178 179 /* Placeholders for holes in the address space */ 180 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_PER_PAGE); 181 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing, P2M_MID_PER_PAGE); 182 static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_missing_mfn, P2M_MID_PER_PAGE); 183 184 static RESERVE_BRK_ARRAY(unsigned long **, p2m_top, P2M_TOP_PER_PAGE); 185 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn, P2M_TOP_PER_PAGE); 186 static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p, P2M_TOP_PER_PAGE); 187 188 static RESERVE_BRK_ARRAY(unsigned long, p2m_identity, P2M_PER_PAGE); 189 190 RESERVE_BRK(p2m_mid, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE))); 191 RESERVE_BRK(p2m_mid_mfn, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE))); 192 193 /* We might hit two boundary violations at the start and end, at max each 194 * boundary violation will require three middle nodes. */ 195 RESERVE_BRK(p2m_mid_identity, PAGE_SIZE * 2 * 3); 196 197 /* When we populate back during bootup, the amount of pages can vary. The 198 * max we have is seen is 395979, but that does not mean it can't be more. 199 * Some machines can have 3GB I/O holes even. With early_can_reuse_p2m_middle 200 * it can re-use Xen provided mfn_list array, so we only need to allocate at 201 * most three P2M top nodes. */ 202 RESERVE_BRK(p2m_populated, PAGE_SIZE * 3); 203 204 static inline unsigned p2m_top_index(unsigned long pfn) 205 { 206 BUG_ON(pfn >= MAX_P2M_PFN); 207 return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE); 208 } 209 210 static inline unsigned p2m_mid_index(unsigned long pfn) 211 { 212 return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE; 213 } 214 215 static inline unsigned p2m_index(unsigned long pfn) 216 { 217 return pfn % P2M_PER_PAGE; 218 } 219 220 static void p2m_top_init(unsigned long ***top) 221 { 222 unsigned i; 223 224 for (i = 0; i < P2M_TOP_PER_PAGE; i++) 225 top[i] = p2m_mid_missing; 226 } 227 228 static void p2m_top_mfn_init(unsigned long *top) 229 { 230 unsigned i; 231 232 for (i = 0; i < P2M_TOP_PER_PAGE; i++) 233 top[i] = virt_to_mfn(p2m_mid_missing_mfn); 234 } 235 236 static void p2m_top_mfn_p_init(unsigned long **top) 237 { 238 unsigned i; 239 240 for (i = 0; i < P2M_TOP_PER_PAGE; i++) 241 top[i] = p2m_mid_missing_mfn; 242 } 243 244 static void p2m_mid_init(unsigned long **mid) 245 { 246 unsigned i; 247 248 for (i = 0; i < P2M_MID_PER_PAGE; i++) 249 mid[i] = p2m_missing; 250 } 251 252 static void p2m_mid_mfn_init(unsigned long *mid) 253 { 254 unsigned i; 255 256 for (i = 0; i < P2M_MID_PER_PAGE; i++) 257 mid[i] = virt_to_mfn(p2m_missing); 258 } 259 260 static void p2m_init(unsigned long *p2m) 261 { 262 unsigned i; 263 264 for (i = 0; i < P2M_MID_PER_PAGE; i++) 265 p2m[i] = INVALID_P2M_ENTRY; 266 } 267 268 /* 269 * Build the parallel p2m_top_mfn and p2m_mid_mfn structures 270 * 271 * This is called both at boot time, and after resuming from suspend: 272 * - At boot time we're called very early, and must use extend_brk() 273 * to allocate memory. 274 * 275 * - After resume we're called from within stop_machine, but the mfn 276 * tree should alreay be completely allocated. 277 */ 278 void __ref xen_build_mfn_list_list(void) 279 { 280 unsigned long pfn; 281 282 /* Pre-initialize p2m_top_mfn to be completely missing */ 283 if (p2m_top_mfn == NULL) { 284 p2m_mid_missing_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE); 285 p2m_mid_mfn_init(p2m_mid_missing_mfn); 286 287 p2m_top_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE); 288 p2m_top_mfn_p_init(p2m_top_mfn_p); 289 290 p2m_top_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE); 291 p2m_top_mfn_init(p2m_top_mfn); 292 } else { 293 /* Reinitialise, mfn's all change after migration */ 294 p2m_mid_mfn_init(p2m_mid_missing_mfn); 295 } 296 297 for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += P2M_PER_PAGE) { 298 unsigned topidx = p2m_top_index(pfn); 299 unsigned mididx = p2m_mid_index(pfn); 300 unsigned long **mid; 301 unsigned long *mid_mfn_p; 302 303 mid = p2m_top[topidx]; 304 mid_mfn_p = p2m_top_mfn_p[topidx]; 305 306 /* Don't bother allocating any mfn mid levels if 307 * they're just missing, just update the stored mfn, 308 * since all could have changed over a migrate. 309 */ 310 if (mid == p2m_mid_missing) { 311 BUG_ON(mididx); 312 BUG_ON(mid_mfn_p != p2m_mid_missing_mfn); 313 p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn); 314 pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE; 315 continue; 316 } 317 318 if (mid_mfn_p == p2m_mid_missing_mfn) { 319 /* 320 * XXX boot-time only! We should never find 321 * missing parts of the mfn tree after 322 * runtime. extend_brk() will BUG if we call 323 * it too late. 324 */ 325 mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE); 326 p2m_mid_mfn_init(mid_mfn_p); 327 328 p2m_top_mfn_p[topidx] = mid_mfn_p; 329 } 330 331 p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p); 332 mid_mfn_p[mididx] = virt_to_mfn(mid[mididx]); 333 } 334 } 335 336 void xen_setup_mfn_list_list(void) 337 { 338 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); 339 340 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = 341 virt_to_mfn(p2m_top_mfn); 342 HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn; 343 } 344 345 /* Set up p2m_top to point to the domain-builder provided p2m pages */ 346 void __init xen_build_dynamic_phys_to_machine(void) 347 { 348 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list; 349 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages); 350 unsigned long pfn; 351 352 xen_max_p2m_pfn = max_pfn; 353 354 p2m_missing = extend_brk(PAGE_SIZE, PAGE_SIZE); 355 p2m_init(p2m_missing); 356 357 p2m_mid_missing = extend_brk(PAGE_SIZE, PAGE_SIZE); 358 p2m_mid_init(p2m_mid_missing); 359 360 p2m_top = extend_brk(PAGE_SIZE, PAGE_SIZE); 361 p2m_top_init(p2m_top); 362 363 p2m_identity = extend_brk(PAGE_SIZE, PAGE_SIZE); 364 p2m_init(p2m_identity); 365 366 /* 367 * The domain builder gives us a pre-constructed p2m array in 368 * mfn_list for all the pages initially given to us, so we just 369 * need to graft that into our tree structure. 370 */ 371 for (pfn = 0; pfn < max_pfn; pfn += P2M_PER_PAGE) { 372 unsigned topidx = p2m_top_index(pfn); 373 unsigned mididx = p2m_mid_index(pfn); 374 375 if (p2m_top[topidx] == p2m_mid_missing) { 376 unsigned long **mid = extend_brk(PAGE_SIZE, PAGE_SIZE); 377 p2m_mid_init(mid); 378 379 p2m_top[topidx] = mid; 380 } 381 382 /* 383 * As long as the mfn_list has enough entries to completely 384 * fill a p2m page, pointing into the array is ok. But if 385 * not the entries beyond the last pfn will be undefined. 386 */ 387 if (unlikely(pfn + P2M_PER_PAGE > max_pfn)) { 388 unsigned long p2midx; 389 390 p2midx = max_pfn % P2M_PER_PAGE; 391 for ( ; p2midx < P2M_PER_PAGE; p2midx++) 392 mfn_list[pfn + p2midx] = INVALID_P2M_ENTRY; 393 } 394 p2m_top[topidx][mididx] = &mfn_list[pfn]; 395 } 396 397 m2p_override_init(); 398 } 399 #ifdef CONFIG_X86_64 400 #include <linux/bootmem.h> 401 unsigned long __init xen_revector_p2m_tree(void) 402 { 403 unsigned long va_start; 404 unsigned long va_end; 405 unsigned long pfn; 406 unsigned long pfn_free = 0; 407 unsigned long *mfn_list = NULL; 408 unsigned long size; 409 410 va_start = xen_start_info->mfn_list; 411 /*We copy in increments of P2M_PER_PAGE * sizeof(unsigned long), 412 * so make sure it is rounded up to that */ 413 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); 414 va_end = va_start + size; 415 416 /* If we were revectored already, don't do it again. */ 417 if (va_start <= __START_KERNEL_map && va_start >= __PAGE_OFFSET) 418 return 0; 419 420 mfn_list = alloc_bootmem_align(size, PAGE_SIZE); 421 if (!mfn_list) { 422 pr_warn("Could not allocate space for a new P2M tree!\n"); 423 return xen_start_info->mfn_list; 424 } 425 /* Fill it out with INVALID_P2M_ENTRY value */ 426 memset(mfn_list, 0xFF, size); 427 428 for (pfn = 0; pfn < ALIGN(MAX_DOMAIN_PAGES, P2M_PER_PAGE); pfn += P2M_PER_PAGE) { 429 unsigned topidx = p2m_top_index(pfn); 430 unsigned mididx; 431 unsigned long *mid_p; 432 433 if (!p2m_top[topidx]) 434 continue; 435 436 if (p2m_top[topidx] == p2m_mid_missing) 437 continue; 438 439 mididx = p2m_mid_index(pfn); 440 mid_p = p2m_top[topidx][mididx]; 441 if (!mid_p) 442 continue; 443 if ((mid_p == p2m_missing) || (mid_p == p2m_identity)) 444 continue; 445 446 if ((unsigned long)mid_p == INVALID_P2M_ENTRY) 447 continue; 448 449 /* The old va. Rebase it on mfn_list */ 450 if (mid_p >= (unsigned long *)va_start && mid_p <= (unsigned long *)va_end) { 451 unsigned long *new; 452 453 if (pfn_free > (size / sizeof(unsigned long))) { 454 WARN(1, "Only allocated for %ld pages, but we want %ld!\n", 455 size / sizeof(unsigned long), pfn_free); 456 return 0; 457 } 458 new = &mfn_list[pfn_free]; 459 460 copy_page(new, mid_p); 461 p2m_top[topidx][mididx] = &mfn_list[pfn_free]; 462 p2m_top_mfn_p[topidx][mididx] = virt_to_mfn(&mfn_list[pfn_free]); 463 464 pfn_free += P2M_PER_PAGE; 465 466 } 467 /* This should be the leafs allocated for identity from _brk. */ 468 } 469 return (unsigned long)mfn_list; 470 471 } 472 #else 473 unsigned long __init xen_revector_p2m_tree(void) 474 { 475 return 0; 476 } 477 #endif 478 unsigned long get_phys_to_machine(unsigned long pfn) 479 { 480 unsigned topidx, mididx, idx; 481 482 if (unlikely(pfn >= MAX_P2M_PFN)) 483 return INVALID_P2M_ENTRY; 484 485 topidx = p2m_top_index(pfn); 486 mididx = p2m_mid_index(pfn); 487 idx = p2m_index(pfn); 488 489 /* 490 * The INVALID_P2M_ENTRY is filled in both p2m_*identity 491 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY 492 * would be wrong. 493 */ 494 if (p2m_top[topidx][mididx] == p2m_identity) 495 return IDENTITY_FRAME(pfn); 496 497 return p2m_top[topidx][mididx][idx]; 498 } 499 EXPORT_SYMBOL_GPL(get_phys_to_machine); 500 501 static void *alloc_p2m_page(void) 502 { 503 return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT); 504 } 505 506 static void free_p2m_page(void *p) 507 { 508 free_page((unsigned long)p); 509 } 510 511 /* 512 * Fully allocate the p2m structure for a given pfn. We need to check 513 * that both the top and mid levels are allocated, and make sure the 514 * parallel mfn tree is kept in sync. We may race with other cpus, so 515 * the new pages are installed with cmpxchg; if we lose the race then 516 * simply free the page we allocated and use the one that's there. 517 */ 518 static bool alloc_p2m(unsigned long pfn) 519 { 520 unsigned topidx, mididx; 521 unsigned long ***top_p, **mid; 522 unsigned long *top_mfn_p, *mid_mfn; 523 524 topidx = p2m_top_index(pfn); 525 mididx = p2m_mid_index(pfn); 526 527 top_p = &p2m_top[topidx]; 528 mid = *top_p; 529 530 if (mid == p2m_mid_missing) { 531 /* Mid level is missing, allocate a new one */ 532 mid = alloc_p2m_page(); 533 if (!mid) 534 return false; 535 536 p2m_mid_init(mid); 537 538 if (cmpxchg(top_p, p2m_mid_missing, mid) != p2m_mid_missing) 539 free_p2m_page(mid); 540 } 541 542 top_mfn_p = &p2m_top_mfn[topidx]; 543 mid_mfn = p2m_top_mfn_p[topidx]; 544 545 BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p); 546 547 if (mid_mfn == p2m_mid_missing_mfn) { 548 /* Separately check the mid mfn level */ 549 unsigned long missing_mfn; 550 unsigned long mid_mfn_mfn; 551 552 mid_mfn = alloc_p2m_page(); 553 if (!mid_mfn) 554 return false; 555 556 p2m_mid_mfn_init(mid_mfn); 557 558 missing_mfn = virt_to_mfn(p2m_mid_missing_mfn); 559 mid_mfn_mfn = virt_to_mfn(mid_mfn); 560 if (cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn) != missing_mfn) 561 free_p2m_page(mid_mfn); 562 else 563 p2m_top_mfn_p[topidx] = mid_mfn; 564 } 565 566 if (p2m_top[topidx][mididx] == p2m_identity || 567 p2m_top[topidx][mididx] == p2m_missing) { 568 /* p2m leaf page is missing */ 569 unsigned long *p2m; 570 unsigned long *p2m_orig = p2m_top[topidx][mididx]; 571 572 p2m = alloc_p2m_page(); 573 if (!p2m) 574 return false; 575 576 p2m_init(p2m); 577 578 if (cmpxchg(&mid[mididx], p2m_orig, p2m) != p2m_orig) 579 free_p2m_page(p2m); 580 else 581 mid_mfn[mididx] = virt_to_mfn(p2m); 582 } 583 584 return true; 585 } 586 587 static bool __init early_alloc_p2m_middle(unsigned long pfn, bool check_boundary) 588 { 589 unsigned topidx, mididx, idx; 590 unsigned long *p2m; 591 unsigned long *mid_mfn_p; 592 593 topidx = p2m_top_index(pfn); 594 mididx = p2m_mid_index(pfn); 595 idx = p2m_index(pfn); 596 597 /* Pfff.. No boundary cross-over, lets get out. */ 598 if (!idx && check_boundary) 599 return false; 600 601 WARN(p2m_top[topidx][mididx] == p2m_identity, 602 "P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n", 603 topidx, mididx); 604 605 /* 606 * Could be done by xen_build_dynamic_phys_to_machine.. 607 */ 608 if (p2m_top[topidx][mididx] != p2m_missing) 609 return false; 610 611 /* Boundary cross-over for the edges: */ 612 p2m = extend_brk(PAGE_SIZE, PAGE_SIZE); 613 614 p2m_init(p2m); 615 616 p2m_top[topidx][mididx] = p2m; 617 618 /* For save/restore we need to MFN of the P2M saved */ 619 620 mid_mfn_p = p2m_top_mfn_p[topidx]; 621 WARN(mid_mfn_p[mididx] != virt_to_mfn(p2m_missing), 622 "P2M_TOP_P[%d][%d] != MFN of p2m_missing!\n", 623 topidx, mididx); 624 mid_mfn_p[mididx] = virt_to_mfn(p2m); 625 626 return true; 627 } 628 629 static bool __init early_alloc_p2m(unsigned long pfn) 630 { 631 unsigned topidx = p2m_top_index(pfn); 632 unsigned long *mid_mfn_p; 633 unsigned long **mid; 634 635 mid = p2m_top[topidx]; 636 mid_mfn_p = p2m_top_mfn_p[topidx]; 637 if (mid == p2m_mid_missing) { 638 mid = extend_brk(PAGE_SIZE, PAGE_SIZE); 639 640 p2m_mid_init(mid); 641 642 p2m_top[topidx] = mid; 643 644 BUG_ON(mid_mfn_p != p2m_mid_missing_mfn); 645 } 646 /* And the save/restore P2M tables.. */ 647 if (mid_mfn_p == p2m_mid_missing_mfn) { 648 mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE); 649 p2m_mid_mfn_init(mid_mfn_p); 650 651 p2m_top_mfn_p[topidx] = mid_mfn_p; 652 p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p); 653 /* Note: we don't set mid_mfn_p[midix] here, 654 * look in early_alloc_p2m_middle */ 655 } 656 return true; 657 } 658 659 /* 660 * Skim over the P2M tree looking at pages that are either filled with 661 * INVALID_P2M_ENTRY or with 1:1 PFNs. If found, re-use that page and 662 * replace the P2M leaf with a p2m_missing or p2m_identity. 663 * Stick the old page in the new P2M tree location. 664 */ 665 bool __init early_can_reuse_p2m_middle(unsigned long set_pfn, unsigned long set_mfn) 666 { 667 unsigned topidx; 668 unsigned mididx; 669 unsigned ident_pfns; 670 unsigned inv_pfns; 671 unsigned long *p2m; 672 unsigned long *mid_mfn_p; 673 unsigned idx; 674 unsigned long pfn; 675 676 /* We only look when this entails a P2M middle layer */ 677 if (p2m_index(set_pfn)) 678 return false; 679 680 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_PER_PAGE) { 681 topidx = p2m_top_index(pfn); 682 683 if (!p2m_top[topidx]) 684 continue; 685 686 if (p2m_top[topidx] == p2m_mid_missing) 687 continue; 688 689 mididx = p2m_mid_index(pfn); 690 p2m = p2m_top[topidx][mididx]; 691 if (!p2m) 692 continue; 693 694 if ((p2m == p2m_missing) || (p2m == p2m_identity)) 695 continue; 696 697 if ((unsigned long)p2m == INVALID_P2M_ENTRY) 698 continue; 699 700 ident_pfns = 0; 701 inv_pfns = 0; 702 for (idx = 0; idx < P2M_PER_PAGE; idx++) { 703 /* IDENTITY_PFNs are 1:1 */ 704 if (p2m[idx] == IDENTITY_FRAME(pfn + idx)) 705 ident_pfns++; 706 else if (p2m[idx] == INVALID_P2M_ENTRY) 707 inv_pfns++; 708 else 709 break; 710 } 711 if ((ident_pfns == P2M_PER_PAGE) || (inv_pfns == P2M_PER_PAGE)) 712 goto found; 713 } 714 return false; 715 found: 716 /* Found one, replace old with p2m_identity or p2m_missing */ 717 p2m_top[topidx][mididx] = (ident_pfns ? p2m_identity : p2m_missing); 718 /* And the other for save/restore.. */ 719 mid_mfn_p = p2m_top_mfn_p[topidx]; 720 /* NOTE: Even if it is a p2m_identity it should still be point to 721 * a page filled with INVALID_P2M_ENTRY entries. */ 722 mid_mfn_p[mididx] = virt_to_mfn(p2m_missing); 723 724 /* Reset where we want to stick the old page in. */ 725 topidx = p2m_top_index(set_pfn); 726 mididx = p2m_mid_index(set_pfn); 727 728 /* This shouldn't happen */ 729 if (WARN_ON(p2m_top[topidx] == p2m_mid_missing)) 730 early_alloc_p2m(set_pfn); 731 732 if (WARN_ON(p2m_top[topidx][mididx] != p2m_missing)) 733 return false; 734 735 p2m_init(p2m); 736 p2m_top[topidx][mididx] = p2m; 737 mid_mfn_p = p2m_top_mfn_p[topidx]; 738 mid_mfn_p[mididx] = virt_to_mfn(p2m); 739 740 return true; 741 } 742 bool __init early_set_phys_to_machine(unsigned long pfn, unsigned long mfn) 743 { 744 if (unlikely(!__set_phys_to_machine(pfn, mfn))) { 745 if (!early_alloc_p2m(pfn)) 746 return false; 747 748 if (early_can_reuse_p2m_middle(pfn, mfn)) 749 return __set_phys_to_machine(pfn, mfn); 750 751 if (!early_alloc_p2m_middle(pfn, false /* boundary crossover OK!*/)) 752 return false; 753 754 if (!__set_phys_to_machine(pfn, mfn)) 755 return false; 756 } 757 758 return true; 759 } 760 unsigned long __init set_phys_range_identity(unsigned long pfn_s, 761 unsigned long pfn_e) 762 { 763 unsigned long pfn; 764 765 if (unlikely(pfn_s >= MAX_P2M_PFN || pfn_e >= MAX_P2M_PFN)) 766 return 0; 767 768 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) 769 return pfn_e - pfn_s; 770 771 if (pfn_s > pfn_e) 772 return 0; 773 774 for (pfn = (pfn_s & ~(P2M_MID_PER_PAGE * P2M_PER_PAGE - 1)); 775 pfn < ALIGN(pfn_e, (P2M_MID_PER_PAGE * P2M_PER_PAGE)); 776 pfn += P2M_MID_PER_PAGE * P2M_PER_PAGE) 777 { 778 WARN_ON(!early_alloc_p2m(pfn)); 779 } 780 781 early_alloc_p2m_middle(pfn_s, true); 782 early_alloc_p2m_middle(pfn_e, true); 783 784 for (pfn = pfn_s; pfn < pfn_e; pfn++) 785 if (!__set_phys_to_machine(pfn, IDENTITY_FRAME(pfn))) 786 break; 787 788 if (!WARN((pfn - pfn_s) != (pfn_e - pfn_s), 789 "Identity mapping failed. We are %ld short of 1-1 mappings!\n", 790 (pfn_e - pfn_s) - (pfn - pfn_s))) 791 printk(KERN_DEBUG "1-1 mapping on %lx->%lx\n", pfn_s, pfn); 792 793 return pfn - pfn_s; 794 } 795 796 /* Try to install p2m mapping; fail if intermediate bits missing */ 797 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn) 798 { 799 unsigned topidx, mididx, idx; 800 801 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) { 802 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY); 803 return true; 804 } 805 if (unlikely(pfn >= MAX_P2M_PFN)) { 806 BUG_ON(mfn != INVALID_P2M_ENTRY); 807 return true; 808 } 809 810 topidx = p2m_top_index(pfn); 811 mididx = p2m_mid_index(pfn); 812 idx = p2m_index(pfn); 813 814 /* For sparse holes were the p2m leaf has real PFN along with 815 * PCI holes, stick in the PFN as the MFN value. 816 */ 817 if (mfn != INVALID_P2M_ENTRY && (mfn & IDENTITY_FRAME_BIT)) { 818 if (p2m_top[topidx][mididx] == p2m_identity) 819 return true; 820 821 /* Swap over from MISSING to IDENTITY if needed. */ 822 if (p2m_top[topidx][mididx] == p2m_missing) { 823 WARN_ON(cmpxchg(&p2m_top[topidx][mididx], p2m_missing, 824 p2m_identity) != p2m_missing); 825 return true; 826 } 827 } 828 829 if (p2m_top[topidx][mididx] == p2m_missing) 830 return mfn == INVALID_P2M_ENTRY; 831 832 p2m_top[topidx][mididx][idx] = mfn; 833 834 return true; 835 } 836 837 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn) 838 { 839 if (unlikely(!__set_phys_to_machine(pfn, mfn))) { 840 if (!alloc_p2m(pfn)) 841 return false; 842 843 if (!__set_phys_to_machine(pfn, mfn)) 844 return false; 845 } 846 847 return true; 848 } 849 850 #define M2P_OVERRIDE_HASH_SHIFT 10 851 #define M2P_OVERRIDE_HASH (1 << M2P_OVERRIDE_HASH_SHIFT) 852 853 static RESERVE_BRK_ARRAY(struct list_head, m2p_overrides, M2P_OVERRIDE_HASH); 854 static DEFINE_SPINLOCK(m2p_override_lock); 855 856 static void __init m2p_override_init(void) 857 { 858 unsigned i; 859 860 m2p_overrides = extend_brk(sizeof(*m2p_overrides) * M2P_OVERRIDE_HASH, 861 sizeof(unsigned long)); 862 863 for (i = 0; i < M2P_OVERRIDE_HASH; i++) 864 INIT_LIST_HEAD(&m2p_overrides[i]); 865 } 866 867 static unsigned long mfn_hash(unsigned long mfn) 868 { 869 return hash_long(mfn, M2P_OVERRIDE_HASH_SHIFT); 870 } 871 872 /* Add an MFN override for a particular page */ 873 int m2p_add_override(unsigned long mfn, struct page *page, 874 struct gnttab_map_grant_ref *kmap_op) 875 { 876 unsigned long flags; 877 unsigned long pfn; 878 unsigned long uninitialized_var(address); 879 unsigned level; 880 pte_t *ptep = NULL; 881 int ret = 0; 882 883 pfn = page_to_pfn(page); 884 if (!PageHighMem(page)) { 885 address = (unsigned long)__va(pfn << PAGE_SHIFT); 886 ptep = lookup_address(address, &level); 887 if (WARN(ptep == NULL || level != PG_LEVEL_4K, 888 "m2p_add_override: pfn %lx not mapped", pfn)) 889 return -EINVAL; 890 } 891 WARN_ON(PagePrivate(page)); 892 SetPagePrivate(page); 893 set_page_private(page, mfn); 894 page->index = pfn_to_mfn(pfn); 895 896 if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) 897 return -ENOMEM; 898 899 if (kmap_op != NULL) { 900 if (!PageHighMem(page)) { 901 struct multicall_space mcs = 902 xen_mc_entry(sizeof(*kmap_op)); 903 904 MULTI_grant_table_op(mcs.mc, 905 GNTTABOP_map_grant_ref, kmap_op, 1); 906 907 xen_mc_issue(PARAVIRT_LAZY_MMU); 908 } 909 } 910 spin_lock_irqsave(&m2p_override_lock, flags); 911 list_add(&page->lru, &m2p_overrides[mfn_hash(mfn)]); 912 spin_unlock_irqrestore(&m2p_override_lock, flags); 913 914 /* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in 915 * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other 916 * pfn so that the following mfn_to_pfn(mfn) calls will return the 917 * pfn from the m2p_override (the backend pfn) instead. 918 * We need to do this because the pages shared by the frontend 919 * (xen-blkfront) can be already locked (lock_page, called by 920 * do_read_cache_page); when the userspace backend tries to use them 921 * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so 922 * do_blockdev_direct_IO is going to try to lock the same pages 923 * again resulting in a deadlock. 924 * As a side effect get_user_pages_fast might not be safe on the 925 * frontend pages while they are being shared with the backend, 926 * because mfn_to_pfn (that ends up being called by GUPF) will 927 * return the backend pfn rather than the frontend pfn. */ 928 ret = __get_user(pfn, &machine_to_phys_mapping[mfn]); 929 if (ret == 0 && get_phys_to_machine(pfn) == mfn) 930 set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)); 931 932 return 0; 933 } 934 EXPORT_SYMBOL_GPL(m2p_add_override); 935 int m2p_remove_override(struct page *page, 936 struct gnttab_map_grant_ref *kmap_op) 937 { 938 unsigned long flags; 939 unsigned long mfn; 940 unsigned long pfn; 941 unsigned long uninitialized_var(address); 942 unsigned level; 943 pte_t *ptep = NULL; 944 int ret = 0; 945 946 pfn = page_to_pfn(page); 947 mfn = get_phys_to_machine(pfn); 948 if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) 949 return -EINVAL; 950 951 if (!PageHighMem(page)) { 952 address = (unsigned long)__va(pfn << PAGE_SHIFT); 953 ptep = lookup_address(address, &level); 954 955 if (WARN(ptep == NULL || level != PG_LEVEL_4K, 956 "m2p_remove_override: pfn %lx not mapped", pfn)) 957 return -EINVAL; 958 } 959 960 spin_lock_irqsave(&m2p_override_lock, flags); 961 list_del(&page->lru); 962 spin_unlock_irqrestore(&m2p_override_lock, flags); 963 WARN_ON(!PagePrivate(page)); 964 ClearPagePrivate(page); 965 966 set_phys_to_machine(pfn, page->index); 967 if (kmap_op != NULL) { 968 if (!PageHighMem(page)) { 969 struct multicall_space mcs; 970 struct gnttab_unmap_grant_ref *unmap_op; 971 972 /* 973 * It might be that we queued all the m2p grant table 974 * hypercalls in a multicall, then m2p_remove_override 975 * get called before the multicall has actually been 976 * issued. In this case handle is going to -1 because 977 * it hasn't been modified yet. 978 */ 979 if (kmap_op->handle == -1) 980 xen_mc_flush(); 981 /* 982 * Now if kmap_op->handle is negative it means that the 983 * hypercall actually returned an error. 984 */ 985 if (kmap_op->handle == GNTST_general_error) { 986 printk(KERN_WARNING "m2p_remove_override: " 987 "pfn %lx mfn %lx, failed to modify kernel mappings", 988 pfn, mfn); 989 return -1; 990 } 991 992 mcs = xen_mc_entry( 993 sizeof(struct gnttab_unmap_grant_ref)); 994 unmap_op = mcs.args; 995 unmap_op->host_addr = kmap_op->host_addr; 996 unmap_op->handle = kmap_op->handle; 997 unmap_op->dev_bus_addr = 0; 998 999 MULTI_grant_table_op(mcs.mc, 1000 GNTTABOP_unmap_grant_ref, unmap_op, 1); 1001 1002 xen_mc_issue(PARAVIRT_LAZY_MMU); 1003 1004 set_pte_at(&init_mm, address, ptep, 1005 pfn_pte(pfn, PAGE_KERNEL)); 1006 __flush_tlb_single(address); 1007 kmap_op->host_addr = 0; 1008 } 1009 } 1010 1011 /* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present 1012 * somewhere in this domain, even before being added to the 1013 * m2p_override (see comment above in m2p_add_override). 1014 * If there are no other entries in the m2p_override corresponding 1015 * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for 1016 * the original pfn (the one shared by the frontend): the backend 1017 * cannot do any IO on this page anymore because it has been 1018 * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of 1019 * the original pfn causes mfn_to_pfn(mfn) to return the frontend 1020 * pfn again. */ 1021 mfn &= ~FOREIGN_FRAME_BIT; 1022 ret = __get_user(pfn, &machine_to_phys_mapping[mfn]); 1023 if (ret == 0 && get_phys_to_machine(pfn) == FOREIGN_FRAME(mfn) && 1024 m2p_find_override(mfn) == NULL) 1025 set_phys_to_machine(pfn, mfn); 1026 1027 return 0; 1028 } 1029 EXPORT_SYMBOL_GPL(m2p_remove_override); 1030 1031 struct page *m2p_find_override(unsigned long mfn) 1032 { 1033 unsigned long flags; 1034 struct list_head *bucket = &m2p_overrides[mfn_hash(mfn)]; 1035 struct page *p, *ret; 1036 1037 ret = NULL; 1038 1039 spin_lock_irqsave(&m2p_override_lock, flags); 1040 1041 list_for_each_entry(p, bucket, lru) { 1042 if (page_private(p) == mfn) { 1043 ret = p; 1044 break; 1045 } 1046 } 1047 1048 spin_unlock_irqrestore(&m2p_override_lock, flags); 1049 1050 return ret; 1051 } 1052 1053 unsigned long m2p_find_override_pfn(unsigned long mfn, unsigned long pfn) 1054 { 1055 struct page *p = m2p_find_override(mfn); 1056 unsigned long ret = pfn; 1057 1058 if (p) 1059 ret = page_to_pfn(p); 1060 1061 return ret; 1062 } 1063 EXPORT_SYMBOL_GPL(m2p_find_override_pfn); 1064 1065 #ifdef CONFIG_XEN_DEBUG_FS 1066 #include <linux/debugfs.h> 1067 #include "debugfs.h" 1068 static int p2m_dump_show(struct seq_file *m, void *v) 1069 { 1070 static const char * const level_name[] = { "top", "middle", 1071 "entry", "abnormal", "error"}; 1072 #define TYPE_IDENTITY 0 1073 #define TYPE_MISSING 1 1074 #define TYPE_PFN 2 1075 #define TYPE_UNKNOWN 3 1076 static const char * const type_name[] = { 1077 [TYPE_IDENTITY] = "identity", 1078 [TYPE_MISSING] = "missing", 1079 [TYPE_PFN] = "pfn", 1080 [TYPE_UNKNOWN] = "abnormal"}; 1081 unsigned long pfn, prev_pfn_type = 0, prev_pfn_level = 0; 1082 unsigned int uninitialized_var(prev_level); 1083 unsigned int uninitialized_var(prev_type); 1084 1085 if (!p2m_top) 1086 return 0; 1087 1088 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn++) { 1089 unsigned topidx = p2m_top_index(pfn); 1090 unsigned mididx = p2m_mid_index(pfn); 1091 unsigned idx = p2m_index(pfn); 1092 unsigned lvl, type; 1093 1094 lvl = 4; 1095 type = TYPE_UNKNOWN; 1096 if (p2m_top[topidx] == p2m_mid_missing) { 1097 lvl = 0; type = TYPE_MISSING; 1098 } else if (p2m_top[topidx] == NULL) { 1099 lvl = 0; type = TYPE_UNKNOWN; 1100 } else if (p2m_top[topidx][mididx] == NULL) { 1101 lvl = 1; type = TYPE_UNKNOWN; 1102 } else if (p2m_top[topidx][mididx] == p2m_identity) { 1103 lvl = 1; type = TYPE_IDENTITY; 1104 } else if (p2m_top[topidx][mididx] == p2m_missing) { 1105 lvl = 1; type = TYPE_MISSING; 1106 } else if (p2m_top[topidx][mididx][idx] == 0) { 1107 lvl = 2; type = TYPE_UNKNOWN; 1108 } else if (p2m_top[topidx][mididx][idx] == IDENTITY_FRAME(pfn)) { 1109 lvl = 2; type = TYPE_IDENTITY; 1110 } else if (p2m_top[topidx][mididx][idx] == INVALID_P2M_ENTRY) { 1111 lvl = 2; type = TYPE_MISSING; 1112 } else if (p2m_top[topidx][mididx][idx] == pfn) { 1113 lvl = 2; type = TYPE_PFN; 1114 } else if (p2m_top[topidx][mididx][idx] != pfn) { 1115 lvl = 2; type = TYPE_PFN; 1116 } 1117 if (pfn == 0) { 1118 prev_level = lvl; 1119 prev_type = type; 1120 } 1121 if (pfn == MAX_DOMAIN_PAGES-1) { 1122 lvl = 3; 1123 type = TYPE_UNKNOWN; 1124 } 1125 if (prev_type != type) { 1126 seq_printf(m, " [0x%lx->0x%lx] %s\n", 1127 prev_pfn_type, pfn, type_name[prev_type]); 1128 prev_pfn_type = pfn; 1129 prev_type = type; 1130 } 1131 if (prev_level != lvl) { 1132 seq_printf(m, " [0x%lx->0x%lx] level %s\n", 1133 prev_pfn_level, pfn, level_name[prev_level]); 1134 prev_pfn_level = pfn; 1135 prev_level = lvl; 1136 } 1137 } 1138 return 0; 1139 #undef TYPE_IDENTITY 1140 #undef TYPE_MISSING 1141 #undef TYPE_PFN 1142 #undef TYPE_UNKNOWN 1143 } 1144 1145 static int p2m_dump_open(struct inode *inode, struct file *filp) 1146 { 1147 return single_open(filp, p2m_dump_show, NULL); 1148 } 1149 1150 static const struct file_operations p2m_dump_fops = { 1151 .open = p2m_dump_open, 1152 .read = seq_read, 1153 .llseek = seq_lseek, 1154 .release = single_release, 1155 }; 1156 1157 static struct dentry *d_mmu_debug; 1158 1159 static int __init xen_p2m_debugfs(void) 1160 { 1161 struct dentry *d_xen = xen_init_debugfs(); 1162 1163 if (d_xen == NULL) 1164 return -ENOMEM; 1165 1166 d_mmu_debug = debugfs_create_dir("mmu", d_xen); 1167 1168 debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops); 1169 return 0; 1170 } 1171 fs_initcall(xen_p2m_debugfs); 1172 #endif /* CONFIG_XEN_DEBUG_FS */ 1173