osfmk/i386/pmap_common.c

   1 /*
   2  * Copyright (c) 2000-2016 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 #include <vm/pmap.h>
  29 #include <kern/ledger.h>
  30 #include <i386/pmap_internal.h>
  31
  32
  33 /*
  34  *      Each entry in the pv_head_table is locked by a bit in the
  35  *      pv_lock_table.  The lock bits are accessed by the physical
  36  *      address of the page they lock.
  37  */
  38
  39 char    *pv_lock_table;         /* pointer to array of bits */
  40 char    *pv_hash_lock_table;
  41
  42 pv_rooted_entry_t       pv_head_table;          /* array of entries, one per
  43                                                  * page */
  44 uint32_t                        pv_hashed_free_count = 0;
  45 uint32_t                        pv_hashed_kern_free_count = 0;
  46
  47 pmap_pagetable_corruption_record_t pmap_pagetable_corruption_records[PMAP_PAGETABLE_CORRUPTION_MAX_LOG];
  48 uint32_t pmap_pagetable_corruption_incidents;
  49 uint64_t pmap_pagetable_corruption_last_abstime = (~(0ULL) >> 1);
  50 uint64_t pmap_pagetable_corruption_interval_abstime;
  51 thread_call_t   pmap_pagetable_corruption_log_call;
  52 static thread_call_data_t       pmap_pagetable_corruption_log_call_data;
  53 boolean_t pmap_pagetable_corruption_timeout = FALSE;
  54
  55 volatile uint32_t       mappingrecurse = 0;
  56
  57 uint32_t  pv_hashed_low_water_mark, pv_hashed_kern_low_water_mark, pv_hashed_alloc_chunk, pv_hashed_kern_alloc_chunk;
  58
  59 thread_t mapping_replenish_thread;
  60 event_t mapping_replenish_event, pmap_user_pv_throttle_event;
  61
  62 uint64_t pmap_pv_throttle_stat, pmap_pv_throttled_waiters;
  63
  64 int pmap_asserts_enabled = (DEBUG);
  65 int pmap_asserts_traced = 0;
  66
  67 unsigned int
  68 pmap_cache_attributes(ppnum_t pn)
  69 {
  70         int cacheattr = pmap_get_cache_attributes(pn, FALSE);
  71
  72         if (cacheattr & INTEL_PTE_NCACHE) {
  73                 if (cacheattr & INTEL_PTE_PAT) {
  74                         /* WC */
  75                         return VM_WIMG_WCOMB;
  76                 }
  77                 return VM_WIMG_IO;
  78         } else {
  79                 return VM_WIMG_COPYBACK;
  80         }
  81 }
  82
  83 void
  84 pmap_set_cache_attributes(ppnum_t pn, unsigned int cacheattr)
  85 {
  86         unsigned int current, template = 0;
  87         int pai;
  88
  89         if (cacheattr & VM_MEM_NOT_CACHEABLE) {
  90                 if (!(cacheattr & VM_MEM_GUARDED)) {
  91                         template |= PHYS_PAT;
  92                 }
  93                 template |= PHYS_NCACHE;
  94         }
  95
  96         pmap_intr_assert();
  97
  98         assert((pn != vm_page_fictitious_addr) && (pn != vm_page_guard_addr));
  99
 100         pai = ppn_to_pai(pn);
 101
 102         if (!IS_MANAGED_PAGE(pai)) {
 103                 return;
 104         }
 105
 106         /* override cache attributes for this phys page
 107          * Does not walk through existing mappings to adjust,
 108          * assumes page is disconnected
 109          */
 110
 111         LOCK_PVH(pai);
 112
 113         pmap_update_cache_attributes_locked(pn, template);
 114
 115         current = pmap_phys_attributes[pai] & PHYS_CACHEABILITY_MASK;
 116         pmap_phys_attributes[pai] &= ~PHYS_CACHEABILITY_MASK;
 117         pmap_phys_attributes[pai] |= template;
 118
 119         UNLOCK_PVH(pai);
 120
 121         if ((template & PHYS_NCACHE) && !(current & PHYS_NCACHE)) {
 122                 pmap_sync_page_attributes_phys(pn);
 123         }
 124 }
 125
 126 unsigned
 127 pmap_get_cache_attributes(ppnum_t pn, boolean_t is_ept)
 128 {
 129         if (last_managed_page == 0) {
 130                 return 0;
 131         }
 132
 133         if (!IS_MANAGED_PAGE(ppn_to_pai(pn))) {
 134                 return PTE_NCACHE(is_ept);
 135         }
 136
 137         /*
 138          * The cache attributes are read locklessly for efficiency.
 139          */
 140         unsigned int attr = pmap_phys_attributes[ppn_to_pai(pn)];
 141         unsigned int template = 0;
 142
 143         /*
 144          * The PTA bit is currently unsupported for EPT PTEs.
 145          */
 146         if ((attr & PHYS_PAT) && !is_ept) {
 147                 template |= INTEL_PTE_PAT;
 148         }
 149
 150         /*
 151          * If the page isn't marked as NCACHE, the default for EPT entries
 152          * is WB.
 153          */
 154         if (attr & PHYS_NCACHE) {
 155                 template |= PTE_NCACHE(is_ept);
 156         } else if (is_ept) {
 157                 template |= INTEL_EPT_WB;
 158         }
 159
 160         return template;
 161 }
 162
 163 boolean_t
 164 pmap_has_managed_page(ppnum_t first, ppnum_t last)
 165 {
 166         ppnum_t     pn, kdata_start, kdata_end;
 167         boolean_t   result;
 168         boot_args * args;
 169
 170         args        = (boot_args *) PE_state.bootArgs;
 171
 172         // Allow pages that the booter added to the end of the kernel.
 173         // We may miss reporting some pages in this range that were freed
 174         // with ml_static_free()
 175         kdata_start = atop_32(args->kaddr);
 176         kdata_end   = atop_32(args->kaddr + args->ksize);
 177
 178         assert(last_managed_page);
 179         assert(first <= last);
 180
 181         for (result = FALSE, pn = first;
 182             !result
 183             && (pn <= last)
 184             && (pn <= last_managed_page);
 185             pn++) {
 186                 if ((pn >= kdata_start) && (pn < kdata_end)) {
 187                         continue;
 188                 }
 189                 result = (0 != (pmap_phys_attributes[pn] & PHYS_MANAGED));
 190         }
 191
 192         return result;
 193 }
 194
 195 boolean_t
 196 pmap_is_noencrypt(ppnum_t pn)
 197 {
 198         int             pai;
 199
 200         pai = ppn_to_pai(pn);
 201
 202         if (!IS_MANAGED_PAGE(pai)) {
 203                 return FALSE;
 204         }
 205
 206         if (pmap_phys_attributes[pai] & PHYS_NOENCRYPT) {
 207                 return TRUE;
 208         }
 209
 210         return FALSE;
 211 }
 212
 213
 214 void
 215 pmap_set_noencrypt(ppnum_t pn)
 216 {
 217         int             pai;
 218
 219         pai = ppn_to_pai(pn);
 220
 221         if (IS_MANAGED_PAGE(pai)) {
 222                 LOCK_PVH(pai);
 223
 224                 pmap_phys_attributes[pai] |= PHYS_NOENCRYPT;
 225
 226                 UNLOCK_PVH(pai);
 227         }
 228 }
 229
 230
 231 void
 232 pmap_clear_noencrypt(ppnum_t pn)
 233 {
 234         int             pai;
 235
 236         pai = ppn_to_pai(pn);
 237
 238         if (IS_MANAGED_PAGE(pai)) {
 239                 /*
 240                  * synchronization at VM layer prevents PHYS_NOENCRYPT
 241                  * from changing state, so we don't need the lock to inspect
 242                  */
 243                 if (pmap_phys_attributes[pai] & PHYS_NOENCRYPT) {
 244                         LOCK_PVH(pai);
 245
 246                         pmap_phys_attributes[pai] &= ~PHYS_NOENCRYPT;
 247
 248                         UNLOCK_PVH(pai);
 249                 }
 250         }
 251 }
 252
 253 void
 254 compute_pmap_gc_throttle(void *arg __unused)
 255 {
 256 }
 257
 258
 259 void
 260 pmap_lock_phys_page(ppnum_t pn)
 261 {
 262         int             pai;
 263
 264         pai = ppn_to_pai(pn);
 265
 266         if (IS_MANAGED_PAGE(pai)) {
 267                 LOCK_PVH(pai);
 268         } else {
 269                 simple_lock(&phys_backup_lock, LCK_GRP_NULL);
 270         }
 271 }
 272
 273
 274 void
 275 pmap_unlock_phys_page(ppnum_t pn)
 276 {
 277         int             pai;
 278
 279         pai = ppn_to_pai(pn);
 280
 281         if (IS_MANAGED_PAGE(pai)) {
 282                 UNLOCK_PVH(pai);
 283         } else {
 284                 simple_unlock(&phys_backup_lock);
 285         }
 286 }
 287
 288
 289
 290 __private_extern__ void
 291 pmap_pagetable_corruption_msg_log(int (*log_func)(const char * fmt, ...)__printflike(1, 2))
 292 {
 293         if (pmap_pagetable_corruption_incidents > 0) {
 294                 int i, e = MIN(pmap_pagetable_corruption_incidents, PMAP_PAGETABLE_CORRUPTION_MAX_LOG);
 295                 (*log_func)("%u pagetable corruption incident(s) detected, timeout: %u\n", pmap_pagetable_corruption_incidents, pmap_pagetable_corruption_timeout);
 296                 for (i = 0; i < e; i++) {
 297                         (*log_func)("Incident 0x%x, reason: 0x%x, action: 0x%x, time: 0x%llx\n", pmap_pagetable_corruption_records[i].incident, pmap_pagetable_corruption_records[i].reason, pmap_pagetable_corruption_records[i].action, pmap_pagetable_corruption_records[i].abstime);
 298                 }
 299         }
 300 }
 301
 302 static inline void
 303 pmap_pagetable_corruption_log_setup(void)
 304 {
 305         if (pmap_pagetable_corruption_log_call == NULL) {
 306                 nanotime_to_absolutetime(PMAP_PAGETABLE_CORRUPTION_INTERVAL, 0, &pmap_pagetable_corruption_interval_abstime);
 307                 thread_call_setup(&pmap_pagetable_corruption_log_call_data,
 308                     (thread_call_func_t) pmap_pagetable_corruption_msg_log,
 309                     (thread_call_param_t) &printf);
 310                 pmap_pagetable_corruption_log_call = &pmap_pagetable_corruption_log_call_data;
 311         }
 312 }
 313
 314 void
 315 mapping_free_prime(void)
 316 {
 317         unsigned                i;
 318         pv_hashed_entry_t       pvh_e;
 319         pv_hashed_entry_t       pvh_eh;
 320         pv_hashed_entry_t       pvh_et;
 321         int                     pv_cnt;
 322
 323         /* Scale based on DRAM size */
 324         pv_hashed_low_water_mark = MAX(PV_HASHED_LOW_WATER_MARK_DEFAULT, ((uint32_t)(sane_size >> 30)) * 2000);
 325         pv_hashed_low_water_mark = MIN(pv_hashed_low_water_mark, 16000);
 326         /* Alterable via sysctl */
 327         pv_hashed_kern_low_water_mark = MAX(PV_HASHED_KERN_LOW_WATER_MARK_DEFAULT, ((uint32_t)(sane_size >> 30)) * 1000);
 328         pv_hashed_kern_low_water_mark = MIN(pv_hashed_kern_low_water_mark, 16000);
 329         pv_hashed_kern_alloc_chunk = PV_HASHED_KERN_ALLOC_CHUNK_INITIAL;
 330         pv_hashed_alloc_chunk = PV_HASHED_ALLOC_CHUNK_INITIAL;
 331
 332         pv_cnt = 0;
 333         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 334
 335         for (i = 0; i < (5 * PV_HASHED_ALLOC_CHUNK_INITIAL); i++) {
 336                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 337
 338                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 339                 pvh_eh = pvh_e;
 340
 341                 if (pvh_et == PV_HASHED_ENTRY_NULL) {
 342                         pvh_et = pvh_e;
 343                 }
 344                 pv_cnt++;
 345         }
 346         PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 347
 348         pv_cnt = 0;
 349         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 350         for (i = 0; i < PV_HASHED_KERN_ALLOC_CHUNK_INITIAL; i++) {
 351                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 352
 353                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 354                 pvh_eh = pvh_e;
 355
 356                 if (pvh_et == PV_HASHED_ENTRY_NULL) {
 357                         pvh_et = pvh_e;
 358                 }
 359                 pv_cnt++;
 360         }
 361         PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 362 }
 363
 364 void mapping_replenish(void);
 365
 366 void
 367 mapping_adjust(void)
 368 {
 369         kern_return_t mres;
 370
 371         pmap_pagetable_corruption_log_setup();
 372
 373         mres = kernel_thread_start_priority((thread_continue_t)mapping_replenish, NULL, MAXPRI_KERNEL, &mapping_replenish_thread);
 374         if (mres != KERN_SUCCESS) {
 375                 panic("pmap: mapping_replenish_thread creation failed");
 376         }
 377         thread_deallocate(mapping_replenish_thread);
 378 }
 379
 380 unsigned pmap_mapping_thread_wakeups;
 381 unsigned pmap_kernel_reserve_replenish_stat;
 382 unsigned pmap_user_reserve_replenish_stat;
 383 unsigned pmap_kern_reserve_alloc_stat;
 384
 385 __attribute__((noreturn))
 386 void
 387 mapping_replenish(void)
 388 {
 389         pv_hashed_entry_t       pvh_e;
 390         pv_hashed_entry_t       pvh_eh;
 391         pv_hashed_entry_t       pvh_et;
 392         int                     pv_cnt;
 393         unsigned                i;
 394
 395         /* We qualify for VM privileges...*/
 396         current_thread()->options |= TH_OPT_VMPRIV;
 397
 398         for (;;) {
 399                 while (pv_hashed_kern_free_count < pv_hashed_kern_low_water_mark) {
 400                         pv_cnt = 0;
 401                         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 402
 403                         for (i = 0; i < pv_hashed_kern_alloc_chunk; i++) {
 404                                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 405                                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 406                                 pvh_eh = pvh_e;
 407
 408                                 if (pvh_et == PV_HASHED_ENTRY_NULL) {
 409                                         pvh_et = pvh_e;
 410                                 }
 411                                 pv_cnt++;
 412                         }
 413                         pmap_kernel_reserve_replenish_stat += pv_cnt;
 414                         PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 415                 }
 416
 417                 pv_cnt = 0;
 418                 pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 419
 420                 if (pv_hashed_free_count < pv_hashed_low_water_mark) {
 421                         for (i = 0; i < pv_hashed_alloc_chunk; i++) {
 422                                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 423
 424                                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 425                                 pvh_eh = pvh_e;
 426
 427                                 if (pvh_et == PV_HASHED_ENTRY_NULL) {
 428                                         pvh_et = pvh_e;
 429                                 }
 430                                 pv_cnt++;
 431                         }
 432                         pmap_user_reserve_replenish_stat += pv_cnt;
 433                         PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 434                 }
 435 /* Wake threads throttled while the kernel reserve was being replenished.
 436  */
 437                 if (pmap_pv_throttled_waiters) {
 438                         pmap_pv_throttled_waiters = 0;
 439                         thread_wakeup(&pmap_user_pv_throttle_event);
 440                 }
 441                 /* Check if the kernel pool has been depleted since the
 442                  * first pass, to reduce refill latency.
 443                  */
 444                 if (pv_hashed_kern_free_count < pv_hashed_kern_low_water_mark) {
 445                         continue;
 446                 }
 447                 /* Block sans continuation to avoid yielding kernel stack */
 448                 assert_wait(&mapping_replenish_event, THREAD_UNINT);
 449                 mappingrecurse = 0;
 450                 thread_block(THREAD_CONTINUE_NULL);
 451                 pmap_mapping_thread_wakeups++;
 452         }
 453 }
 454
 455 /*
 456  *      Set specified attribute bits.
 457  */
 458
 459 void
 460 phys_attribute_set(
 461         ppnum_t         pn,
 462         int             bits)
 463 {
 464         int             pai;
 465
 466         pmap_intr_assert();
 467         assert(pn != vm_page_fictitious_addr);
 468         if (pn == vm_page_guard_addr) {
 469                 return;
 470         }
 471
 472         pai = ppn_to_pai(pn);
 473
 474         if (!IS_MANAGED_PAGE(pai)) {
 475                 /* Not a managed page.  */
 476                 return;
 477         }
 478
 479         LOCK_PVH(pai);
 480         pmap_phys_attributes[pai] |= bits;
 481         UNLOCK_PVH(pai);
 482 }
 483
 484 /*
 485  *      Set the modify bit on the specified physical page.
 486  */
 487
 488 void
 489 pmap_set_modify(ppnum_t pn)
 490 {
 491         phys_attribute_set(pn, PHYS_MODIFIED);
 492 }
 493
 494 /*
 495  *      Clear the modify bits on the specified physical page.
 496  */
 497
 498 void
 499 pmap_clear_modify(ppnum_t pn)
 500 {
 501         phys_attribute_clear(pn, PHYS_MODIFIED, 0, NULL);
 502 }
 503
 504 /*
 505  *      pmap_is_modified:
 506  *
 507  *      Return whether or not the specified physical page is modified
 508  *      by any physical maps.
 509  */
 510
 511 boolean_t
 512 pmap_is_modified(ppnum_t pn)
 513 {
 514         if (phys_attribute_test(pn, PHYS_MODIFIED)) {
 515                 return TRUE;
 516         }
 517         return FALSE;
 518 }
 519
 520
 521 /*
 522  *      pmap_clear_reference:
 523  *
 524  *      Clear the reference bit on the specified physical page.
 525  */
 526
 527 void
 528 pmap_clear_reference(ppnum_t pn)
 529 {
 530         phys_attribute_clear(pn, PHYS_REFERENCED, 0, NULL);
 531 }
 532
 533 void
 534 pmap_set_reference(ppnum_t pn)
 535 {
 536         phys_attribute_set(pn, PHYS_REFERENCED);
 537 }
 538
 539 /*
 540  *      pmap_is_referenced:
 541  *
 542  *      Return whether or not the specified physical page is referenced
 543  *      by any physical maps.
 544  */
 545
 546 boolean_t
 547 pmap_is_referenced(ppnum_t pn)
 548 {
 549         if (phys_attribute_test(pn, PHYS_REFERENCED)) {
 550                 return TRUE;
 551         }
 552         return FALSE;
 553 }
 554
 555
 556 /*
 557  * pmap_get_refmod(phys)
 558  *  returns the referenced and modified bits of the specified
 559  *  physical page.
 560  */
 561 unsigned int
 562 pmap_get_refmod(ppnum_t pn)
 563 {
 564         int             refmod;
 565         unsigned int    retval = 0;
 566
 567         refmod = phys_attribute_test(pn, PHYS_MODIFIED | PHYS_REFERENCED);
 568
 569         if (refmod & PHYS_MODIFIED) {
 570                 retval |= VM_MEM_MODIFIED;
 571         }
 572         if (refmod & PHYS_REFERENCED) {
 573                 retval |= VM_MEM_REFERENCED;
 574         }
 575
 576         return retval;
 577 }
 578
 579
 580 void
 581 pmap_clear_refmod_options(ppnum_t pn, unsigned int mask, unsigned int options, void *arg)
 582 {
 583         unsigned int  x86Mask;
 584
 585         x86Mask = (((mask &   VM_MEM_MODIFIED)?   PHYS_MODIFIED : 0)
 586             | ((mask & VM_MEM_REFERENCED)? PHYS_REFERENCED : 0));
 587
 588         phys_attribute_clear(pn, x86Mask, options, arg);
 589 }
 590
 591 /*
 592  * pmap_clear_refmod(phys, mask)
 593  *  clears the referenced and modified bits as specified by the mask
 594  *  of the specified physical page.
 595  */
 596 void
 597 pmap_clear_refmod(ppnum_t pn, unsigned int mask)
 598 {
 599         unsigned int  x86Mask;
 600
 601         x86Mask = (((mask &   VM_MEM_MODIFIED)?   PHYS_MODIFIED : 0)
 602             | ((mask & VM_MEM_REFERENCED)? PHYS_REFERENCED : 0));
 603
 604         phys_attribute_clear(pn, x86Mask, 0, NULL);
 605 }
 606
 607 unsigned int
 608 pmap_disconnect(ppnum_t pa)
 609 {
 610         return pmap_disconnect_options(pa, 0, NULL);
 611 }
 612
 613 /*
 614  *      Routine:
 615  *              pmap_disconnect_options
 616  *
 617  *      Function:
 618  *              Disconnect all mappings for this page and return reference and change status
 619  *              in generic format.
 620  *
 621  */
 622 unsigned int
 623 pmap_disconnect_options(ppnum_t pa, unsigned int options, void *arg)
 624 {
 625         unsigned refmod, vmrefmod = 0;
 626
 627         pmap_page_protect_options(pa, 0, options, arg);         /* disconnect the page */
 628
 629         pmap_assert(pa != vm_page_fictitious_addr);
 630         if ((pa == vm_page_guard_addr) || !IS_MANAGED_PAGE(pa) || (options & PMAP_OPTIONS_NOREFMOD)) {
 631                 return 0;
 632         }
 633         refmod = pmap_phys_attributes[pa] & (PHYS_MODIFIED | PHYS_REFERENCED);
 634
 635         if (refmod & PHYS_MODIFIED) {
 636                 vmrefmod |= VM_MEM_MODIFIED;
 637         }
 638         if (refmod & PHYS_REFERENCED) {
 639                 vmrefmod |= VM_MEM_REFERENCED;
 640         }
 641
 642         return vmrefmod;
 643 }