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 pmap_cache_attributes(ppnum_t pn) {
  68         if (pmap_get_cache_attributes(pn, FALSE) & INTEL_PTE_NCACHE)
  69                 return (VM_WIMG_IO);
  70         else
  71                 return (VM_WIMG_COPYBACK);
  72 }
  73
  74 void    pmap_set_cache_attributes(ppnum_t pn, unsigned int cacheattr) {
  75         unsigned int current, template = 0;
  76         int pai;
  77
  78         if (cacheattr & VM_MEM_NOT_CACHEABLE) {
  79                 if(!(cacheattr & VM_MEM_GUARDED))
  80                         template |= PHYS_PTA;
  81                 template |= PHYS_NCACHE;
  82         }
  83
  84         pmap_intr_assert();
  85
  86         assert((pn != vm_page_fictitious_addr) && (pn != vm_page_guard_addr));
  87
  88         pai = ppn_to_pai(pn);
  89
  90         if (!IS_MANAGED_PAGE(pai)) {
  91                 return;
  92         }
  93
  94         /* override cache attributes for this phys page
  95          * Does not walk through existing mappings to adjust,
  96          * assumes page is disconnected
  97          */
  98
  99         LOCK_PVH(pai);
 100
 101         pmap_update_cache_attributes_locked(pn, template);
 102
 103         current = pmap_phys_attributes[pai] & PHYS_CACHEABILITY_MASK;
 104         pmap_phys_attributes[pai] &= ~PHYS_CACHEABILITY_MASK;
 105         pmap_phys_attributes[pai] |= template;
 106
 107         UNLOCK_PVH(pai);
 108
 109         if ((template & PHYS_NCACHE) && !(current & PHYS_NCACHE)) {
 110                 pmap_sync_page_attributes_phys(pn);
 111         }
 112 }
 113
 114 unsigned        pmap_get_cache_attributes(ppnum_t pn, boolean_t is_ept) {
 115         if (last_managed_page == 0)
 116                 return 0;
 117
 118         if (!IS_MANAGED_PAGE(ppn_to_pai(pn)))
 119             return PTE_NCACHE(is_ept);
 120
 121         /*
 122          * The cache attributes are read locklessly for efficiency.
 123          */
 124         unsigned int attr = pmap_phys_attributes[ppn_to_pai(pn)];
 125         unsigned int template = 0;
 126
 127         /*
 128          * The PTA bit is currently unsupported for EPT PTEs.
 129          */
 130         if ((attr & PHYS_PTA) && !is_ept)
 131                 template |= INTEL_PTE_PTA;
 132
 133         /*
 134          * If the page isn't marked as NCACHE, the default for EPT entries
 135          * is WB.
 136          */
 137         if (attr & PHYS_NCACHE)
 138                 template |= PTE_NCACHE(is_ept);
 139         else if (is_ept)
 140                 template |= INTEL_EPT_WB;
 141
 142         return template;
 143 }
 144
 145 boolean_t
 146 pmap_has_managed_page(ppnum_t first, ppnum_t last)
 147 {
 148         ppnum_t   pn;
 149     boolean_t result;
 150
 151     assert(last_managed_page);
 152     assert(first <= last);
 153
 154     for (result = FALSE, pn = first;
 155         !result
 156           && (pn <= last)
 157           && (pn <= last_managed_page);
 158          pn++)
 159     {
 160         result = (0 != (pmap_phys_attributes[pn] & PHYS_MANAGED));
 161     }
 162
 163         return (result);
 164 }
 165
 166 boolean_t
 167 pmap_is_noencrypt(ppnum_t pn)
 168 {
 169         int             pai;
 170
 171         pai = ppn_to_pai(pn);
 172
 173         if (!IS_MANAGED_PAGE(pai))
 174                 return (FALSE);
 175
 176         if (pmap_phys_attributes[pai] & PHYS_NOENCRYPT)
 177                 return (TRUE);
 178
 179         return (FALSE);
 180 }
 181
 182
 183 void
 184 pmap_set_noencrypt(ppnum_t pn)
 185 {
 186         int             pai;
 187
 188         pai = ppn_to_pai(pn);
 189
 190         if (IS_MANAGED_PAGE(pai)) {
 191                 LOCK_PVH(pai);
 192
 193                 pmap_phys_attributes[pai] |= PHYS_NOENCRYPT;
 194
 195                 UNLOCK_PVH(pai);
 196         }
 197 }
 198
 199
 200 void
 201 pmap_clear_noencrypt(ppnum_t pn)
 202 {
 203         int             pai;
 204
 205         pai = ppn_to_pai(pn);
 206
 207         if (IS_MANAGED_PAGE(pai)) {
 208                 /*
 209                  * synchronization at VM layer prevents PHYS_NOENCRYPT
 210                  * from changing state, so we don't need the lock to inspect
 211                  */
 212                 if (pmap_phys_attributes[pai] & PHYS_NOENCRYPT) {
 213                         LOCK_PVH(pai);
 214
 215                         pmap_phys_attributes[pai] &= ~PHYS_NOENCRYPT;
 216
 217                         UNLOCK_PVH(pai);
 218                 }
 219         }
 220 }
 221
 222 void
 223 compute_pmap_gc_throttle(void *arg __unused)
 224 {
 225
 226 }
 227
 228
 229 void
 230 pmap_lock_phys_page(ppnum_t pn)
 231 {
 232         int             pai;
 233
 234         pai = ppn_to_pai(pn);
 235
 236         if (IS_MANAGED_PAGE(pai)) {
 237                 LOCK_PVH(pai);
 238         } else
 239                 simple_lock(&phys_backup_lock);
 240 }
 241
 242
 243 void
 244 pmap_unlock_phys_page(ppnum_t pn)
 245 {
 246         int             pai;
 247
 248         pai = ppn_to_pai(pn);
 249
 250         if (IS_MANAGED_PAGE(pai)) {
 251                 UNLOCK_PVH(pai);
 252         } else
 253                 simple_unlock(&phys_backup_lock);
 254 }
 255
 256
 257
 258 __private_extern__ void
 259 pmap_pagetable_corruption_msg_log(int (*log_func)(const char * fmt, ...)__printflike(1,2)) {
 260         if (pmap_pagetable_corruption_incidents > 0) {
 261                 int i, e = MIN(pmap_pagetable_corruption_incidents, PMAP_PAGETABLE_CORRUPTION_MAX_LOG);
 262                 (*log_func)("%u pagetable corruption incident(s) detected, timeout: %u\n", pmap_pagetable_corruption_incidents, pmap_pagetable_corruption_timeout);
 263                 for (i = 0; i < e; i++) {
 264                         (*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);
 265                 }
 266         }
 267 }
 268
 269 static inline void
 270 pmap_pagetable_corruption_log_setup(void) {
 271         if (pmap_pagetable_corruption_log_call == NULL) {
 272                 nanotime_to_absolutetime(PMAP_PAGETABLE_CORRUPTION_INTERVAL, 0, &pmap_pagetable_corruption_interval_abstime);
 273                 thread_call_setup(&pmap_pagetable_corruption_log_call_data,
 274                     (thread_call_func_t) pmap_pagetable_corruption_msg_log,
 275                     (thread_call_param_t) &printf);
 276                 pmap_pagetable_corruption_log_call = &pmap_pagetable_corruption_log_call_data;
 277         }
 278 }
 279
 280 void
 281 mapping_free_prime(void)
 282 {
 283         unsigned                i;
 284         pv_hashed_entry_t       pvh_e;
 285         pv_hashed_entry_t       pvh_eh;
 286         pv_hashed_entry_t       pvh_et;
 287         int                     pv_cnt;
 288
 289         /* Scale based on DRAM size */
 290         pv_hashed_low_water_mark = MAX(PV_HASHED_LOW_WATER_MARK_DEFAULT, ((uint32_t)(sane_size >> 30)) * 2000);
 291         pv_hashed_low_water_mark = MIN(pv_hashed_low_water_mark, 16000);
 292         /* Alterable via sysctl */
 293         pv_hashed_kern_low_water_mark = MAX(PV_HASHED_KERN_LOW_WATER_MARK_DEFAULT, ((uint32_t)(sane_size >> 30)) * 1000);
 294         pv_hashed_kern_low_water_mark = MIN(pv_hashed_kern_low_water_mark, 16000);
 295         pv_hashed_kern_alloc_chunk = PV_HASHED_KERN_ALLOC_CHUNK_INITIAL;
 296         pv_hashed_alloc_chunk = PV_HASHED_ALLOC_CHUNK_INITIAL;
 297
 298         pv_cnt = 0;
 299         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 300
 301         for (i = 0; i < (5 * PV_HASHED_ALLOC_CHUNK_INITIAL); i++) {
 302                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 303
 304                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 305                 pvh_eh = pvh_e;
 306
 307                 if (pvh_et == PV_HASHED_ENTRY_NULL)
 308                         pvh_et = pvh_e;
 309                 pv_cnt++;
 310         }
 311         PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 312
 313         pv_cnt = 0;
 314         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 315         for (i = 0; i < PV_HASHED_KERN_ALLOC_CHUNK_INITIAL; i++) {
 316                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 317
 318                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 319                 pvh_eh = pvh_e;
 320
 321                 if (pvh_et == PV_HASHED_ENTRY_NULL)
 322                         pvh_et = pvh_e;
 323                 pv_cnt++;
 324         }
 325         PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 326 }
 327
 328 void mapping_replenish(void);
 329
 330 void mapping_adjust(void) {
 331         kern_return_t mres;
 332
 333         pmap_pagetable_corruption_log_setup();
 334
 335         mres = kernel_thread_start_priority((thread_continue_t)mapping_replenish, NULL, MAXPRI_KERNEL, &mapping_replenish_thread);
 336         if (mres != KERN_SUCCESS) {
 337                 panic("pmap: mapping_replenish_thread creation failed");
 338         }
 339         thread_deallocate(mapping_replenish_thread);
 340 }
 341
 342 unsigned pmap_mapping_thread_wakeups;
 343 unsigned pmap_kernel_reserve_replenish_stat;
 344 unsigned pmap_user_reserve_replenish_stat;
 345 unsigned pmap_kern_reserve_alloc_stat;
 346
 347 __attribute__((noreturn))
 348 void
 349 mapping_replenish(void)
 350 {
 351         pv_hashed_entry_t       pvh_e;
 352         pv_hashed_entry_t       pvh_eh;
 353         pv_hashed_entry_t       pvh_et;
 354         int                     pv_cnt;
 355         unsigned                i;
 356
 357         /* We qualify for VM privileges...*/
 358         current_thread()->options |= TH_OPT_VMPRIV;
 359
 360         for (;;) {
 361
 362                 while (pv_hashed_kern_free_count < pv_hashed_kern_low_water_mark) {
 363                         pv_cnt = 0;
 364                         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 365
 366                         for (i = 0; i < pv_hashed_kern_alloc_chunk; i++) {
 367                                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 368                                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 369                                 pvh_eh = pvh_e;
 370
 371                                 if (pvh_et == PV_HASHED_ENTRY_NULL)
 372                                         pvh_et = pvh_e;
 373                                 pv_cnt++;
 374                         }
 375                         pmap_kernel_reserve_replenish_stat += pv_cnt;
 376                         PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 377                 }
 378
 379                 pv_cnt = 0;
 380                 pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 381
 382                 if (pv_hashed_free_count < pv_hashed_low_water_mark) {
 383                         for (i = 0; i < pv_hashed_alloc_chunk; i++) {
 384                                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 385
 386                                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 387                                 pvh_eh = pvh_e;
 388
 389                                 if (pvh_et == PV_HASHED_ENTRY_NULL)
 390                                         pvh_et = pvh_e;
 391                                 pv_cnt++;
 392                         }
 393                         pmap_user_reserve_replenish_stat += pv_cnt;
 394                         PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 395                 }
 396 /* Wake threads throttled while the kernel reserve was being replenished.
 397  */
 398                 if (pmap_pv_throttled_waiters) {
 399                         pmap_pv_throttled_waiters = 0;
 400                         thread_wakeup(&pmap_user_pv_throttle_event);
 401                 }
 402                 /* Check if the kernel pool has been depleted since the
 403                  * first pass, to reduce refill latency.
 404                  */
 405                 if (pv_hashed_kern_free_count < pv_hashed_kern_low_water_mark)
 406                         continue;
 407                 /* Block sans continuation to avoid yielding kernel stack */
 408                 assert_wait(&mapping_replenish_event, THREAD_UNINT);
 409                 mappingrecurse = 0;
 410                 thread_block(THREAD_CONTINUE_NULL);
 411                 pmap_mapping_thread_wakeups++;
 412         }
 413 }
 414
 415 /*
 416  *      Set specified attribute bits.
 417  */
 418
 419 void
 420 phys_attribute_set(
 421         ppnum_t         pn,
 422         int             bits)
 423 {
 424         int             pai;
 425
 426         pmap_intr_assert();
 427         assert(pn != vm_page_fictitious_addr);
 428         if (pn == vm_page_guard_addr)
 429                 return;
 430
 431         pai = ppn_to_pai(pn);
 432
 433         if (!IS_MANAGED_PAGE(pai)) {
 434                 /* Not a managed page.  */
 435                 return;
 436         }
 437
 438         LOCK_PVH(pai);
 439         pmap_phys_attributes[pai] |= bits;
 440         UNLOCK_PVH(pai);
 441 }
 442
 443 /*
 444  *      Set the modify bit on the specified physical page.
 445  */
 446
 447 void
 448 pmap_set_modify(ppnum_t pn)
 449 {
 450         phys_attribute_set(pn, PHYS_MODIFIED);
 451 }
 452
 453 /*
 454  *      Clear the modify bits on the specified physical page.
 455  */
 456
 457 void
 458 pmap_clear_modify(ppnum_t pn)
 459 {
 460         phys_attribute_clear(pn, PHYS_MODIFIED, 0, NULL);
 461 }
 462
 463 /*
 464  *      pmap_is_modified:
 465  *
 466  *      Return whether or not the specified physical page is modified
 467  *      by any physical maps.
 468  */
 469
 470 boolean_t
 471 pmap_is_modified(ppnum_t pn)
 472 {
 473         if (phys_attribute_test(pn, PHYS_MODIFIED))
 474                 return TRUE;
 475         return FALSE;
 476 }
 477
 478
 479 /*
 480  *      pmap_clear_reference:
 481  *
 482  *      Clear the reference bit on the specified physical page.
 483  */
 484
 485 void
 486 pmap_clear_reference(ppnum_t pn)
 487 {
 488         phys_attribute_clear(pn, PHYS_REFERENCED, 0, NULL);
 489 }
 490
 491 void
 492 pmap_set_reference(ppnum_t pn)
 493 {
 494         phys_attribute_set(pn, PHYS_REFERENCED);
 495 }
 496
 497 /*
 498  *      pmap_is_referenced:
 499  *
 500  *      Return whether or not the specified physical page is referenced
 501  *      by any physical maps.
 502  */
 503
 504 boolean_t
 505 pmap_is_referenced(ppnum_t pn)
 506 {
 507         if (phys_attribute_test(pn, PHYS_REFERENCED))
 508                 return TRUE;
 509         return FALSE;
 510 }
 511
 512
 513 /*
 514  * pmap_get_refmod(phys)
 515  *  returns the referenced and modified bits of the specified
 516  *  physical page.
 517  */
 518 unsigned int
 519 pmap_get_refmod(ppnum_t pn)
 520 {
 521         int             refmod;
 522         unsigned int    retval = 0;
 523
 524         refmod = phys_attribute_test(pn, PHYS_MODIFIED | PHYS_REFERENCED);
 525
 526         if (refmod & PHYS_MODIFIED)
 527                 retval |= VM_MEM_MODIFIED;
 528         if (refmod & PHYS_REFERENCED)
 529                 retval |= VM_MEM_REFERENCED;
 530
 531         return (retval);
 532 }
 533
 534
 535 void
 536 pmap_clear_refmod_options(ppnum_t pn, unsigned int mask, unsigned int options, void *arg)
 537 {
 538         unsigned int  x86Mask;
 539
 540         x86Mask = (   ((mask &   VM_MEM_MODIFIED)?   PHYS_MODIFIED : 0)
 541                       | ((mask & VM_MEM_REFERENCED)? PHYS_REFERENCED : 0));
 542
 543         phys_attribute_clear(pn, x86Mask, options, arg);
 544 }
 545
 546 /*
 547  * pmap_clear_refmod(phys, mask)
 548  *  clears the referenced and modified bits as specified by the mask
 549  *  of the specified physical page.
 550  */
 551 void
 552 pmap_clear_refmod(ppnum_t pn, unsigned int mask)
 553 {
 554         unsigned int  x86Mask;
 555
 556         x86Mask = (   ((mask &   VM_MEM_MODIFIED)?   PHYS_MODIFIED : 0)
 557                     | ((mask & VM_MEM_REFERENCED)? PHYS_REFERENCED : 0));
 558
 559         phys_attribute_clear(pn, x86Mask, 0, NULL);
 560 }
 561
 562 unsigned int
 563 pmap_disconnect(ppnum_t pa)
 564 {
 565         return (pmap_disconnect_options(pa, 0, NULL));
 566 }
 567
 568 /*
 569  *      Routine:
 570  *              pmap_disconnect_options
 571  *
 572  *      Function:
 573  *              Disconnect all mappings for this page and return reference and change status
 574  *              in generic format.
 575  *
 576  */
 577 unsigned int
 578 pmap_disconnect_options(ppnum_t pa, unsigned int options, void *arg)
 579 {
 580         unsigned refmod, vmrefmod = 0;
 581
 582         pmap_page_protect_options(pa, 0, options, arg);         /* disconnect the page */
 583
 584         pmap_assert(pa != vm_page_fictitious_addr);
 585         if ((pa == vm_page_guard_addr) || !IS_MANAGED_PAGE(pa) || (options & PMAP_OPTIONS_NOREFMOD))
 586                 return 0;
 587         refmod = pmap_phys_attributes[pa] & (PHYS_MODIFIED | PHYS_REFERENCED);
 588
 589         if (refmod & PHYS_MODIFIED)
 590                 vmrefmod |= VM_MEM_MODIFIED;
 591         if (refmod & PHYS_REFERENCED)
 592                 vmrefmod |= VM_MEM_REFERENCED;
 593
 594         return vmrefmod;
 595 }