osfmk/i386/pmap_common.c

   1 /*
   2  * Copyright (c) 2000-2010 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 unsigned int pmap_cache_attributes(ppnum_t pn) {
  65         if (pmap_get_cache_attributes(pn) & INTEL_PTE_NCACHE)
  66                 return (VM_WIMG_IO);
  67         else
  68                 return (VM_WIMG_COPYBACK);
  69 }
  70
  71 void    pmap_set_cache_attributes(ppnum_t pn, unsigned int cacheattr) {
  72         unsigned int current, template = 0;
  73         int pai;
  74
  75         if (cacheattr & VM_MEM_NOT_CACHEABLE) {
  76                 if(!(cacheattr & VM_MEM_GUARDED))
  77                         template |= PHYS_PTA;
  78                 template |= PHYS_NCACHE;
  79         }
  80
  81         pmap_intr_assert();
  82
  83         assert((pn != vm_page_fictitious_addr) && (pn != vm_page_guard_addr));
  84
  85         pai = ppn_to_pai(pn);
  86
  87         if (!IS_MANAGED_PAGE(pai)) {
  88                 return;
  89         }
  90
  91         /* override cache attributes for this phys page
  92          * Does not walk through existing mappings to adjust,
  93          * assumes page is disconnected
  94          */
  95
  96         LOCK_PVH(pai);
  97
  98         pmap_update_cache_attributes_locked(pn, template);
  99
 100         current = pmap_phys_attributes[pai] & PHYS_CACHEABILITY_MASK;
 101         pmap_phys_attributes[pai] &= ~PHYS_CACHEABILITY_MASK;
 102         pmap_phys_attributes[pai] |= template;
 103
 104         UNLOCK_PVH(pai);
 105
 106         if ((template & PHYS_NCACHE) && !(current & PHYS_NCACHE)) {
 107                 pmap_sync_page_attributes_phys(pn);
 108         }
 109 }
 110
 111 unsigned        pmap_get_cache_attributes(ppnum_t pn) {
 112         if (last_managed_page == 0)
 113                 return 0;
 114
 115         if (!IS_MANAGED_PAGE(ppn_to_pai(pn))) {
 116             return INTEL_PTE_NCACHE;
 117         }
 118
 119         /*
 120          * The cache attributes are read locklessly for efficiency.
 121          */
 122         unsigned int attr = pmap_phys_attributes[ppn_to_pai(pn)];
 123         unsigned int template = 0;
 124
 125         if (attr & PHYS_PTA)
 126                 template |= INTEL_PTE_PTA;
 127         if (attr & PHYS_NCACHE)
 128                 template |= INTEL_PTE_NCACHE;
 129         return template;
 130 }
 131
 132
 133
 134 boolean_t
 135 pmap_is_noencrypt(ppnum_t pn)
 136 {
 137         int             pai;
 138
 139         pai = ppn_to_pai(pn);
 140
 141         if (!IS_MANAGED_PAGE(pai))
 142                 return (FALSE);
 143
 144         if (pmap_phys_attributes[pai] & PHYS_NOENCRYPT)
 145                 return (TRUE);
 146
 147         return (FALSE);
 148 }
 149
 150
 151 void
 152 pmap_set_noencrypt(ppnum_t pn)
 153 {
 154         int             pai;
 155
 156         pai = ppn_to_pai(pn);
 157
 158         if (IS_MANAGED_PAGE(pai)) {
 159                 LOCK_PVH(pai);
 160
 161                 pmap_phys_attributes[pai] |= PHYS_NOENCRYPT;
 162
 163                 UNLOCK_PVH(pai);
 164         }
 165 }
 166
 167
 168 void
 169 pmap_clear_noencrypt(ppnum_t pn)
 170 {
 171         int             pai;
 172
 173         pai = ppn_to_pai(pn);
 174
 175         if (IS_MANAGED_PAGE(pai)) {
 176                 /*
 177                  * synchronization at VM layer prevents PHYS_NOENCRYPT
 178                  * from changing state, so we don't need the lock to inspect
 179                  */
 180                 if (pmap_phys_attributes[pai] & PHYS_NOENCRYPT) {
 181                         LOCK_PVH(pai);
 182
 183                         pmap_phys_attributes[pai] &= ~PHYS_NOENCRYPT;
 184
 185                         UNLOCK_PVH(pai);
 186                 }
 187         }
 188 }
 189
 190 void
 191 compute_pmap_gc_throttle(void *arg __unused)
 192 {
 193
 194 }
 195
 196
 197 void
 198 pmap_lock_phys_page(ppnum_t pn)
 199 {
 200         int             pai;
 201
 202         pai = ppn_to_pai(pn);
 203
 204         if (IS_MANAGED_PAGE(pai)) {
 205                 LOCK_PVH(pai);
 206         } else
 207                 simple_lock(&phys_backup_lock);
 208 }
 209
 210
 211 void
 212 pmap_unlock_phys_page(ppnum_t pn)
 213 {
 214         int             pai;
 215
 216         pai = ppn_to_pai(pn);
 217
 218         if (IS_MANAGED_PAGE(pai)) {
 219                 UNLOCK_PVH(pai);
 220         } else
 221                 simple_unlock(&phys_backup_lock);
 222 }
 223
 224
 225
 226 __private_extern__ void
 227 pmap_pagetable_corruption_msg_log(int (*log_func)(const char * fmt, ...)__printflike(1,2)) {
 228         if (pmap_pagetable_corruption_incidents > 0) {
 229                 int i, e = MIN(pmap_pagetable_corruption_incidents, PMAP_PAGETABLE_CORRUPTION_MAX_LOG);
 230                 (*log_func)("%u pagetable corruption incident(s) detected, timeout: %u\n", pmap_pagetable_corruption_incidents, pmap_pagetable_corruption_timeout);
 231                 for (i = 0; i < e; i++) {
 232                         (*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);
 233                 }
 234         }
 235 }
 236
 237 static inline void
 238 pmap_pagetable_corruption_log_setup(void) {
 239         if (pmap_pagetable_corruption_log_call == NULL) {
 240                 nanotime_to_absolutetime(PMAP_PAGETABLE_CORRUPTION_INTERVAL, 0, &pmap_pagetable_corruption_interval_abstime);
 241                 thread_call_setup(&pmap_pagetable_corruption_log_call_data,
 242                     (thread_call_func_t) pmap_pagetable_corruption_msg_log,
 243                     (thread_call_param_t) &printf);
 244                 pmap_pagetable_corruption_log_call = &pmap_pagetable_corruption_log_call_data;
 245         }
 246 }
 247
 248 void
 249 mapping_free_prime(void)
 250 {
 251         unsigned                i;
 252         pv_hashed_entry_t       pvh_e;
 253         pv_hashed_entry_t       pvh_eh;
 254         pv_hashed_entry_t       pvh_et;
 255         int                     pv_cnt;
 256
 257         /* Scale based on DRAM size */
 258         pv_hashed_low_water_mark = MAX(PV_HASHED_LOW_WATER_MARK_DEFAULT, ((uint32_t)(sane_size >> 30)) * 2000);
 259         pv_hashed_low_water_mark = MIN(pv_hashed_low_water_mark, 16000);
 260         /* Alterable via sysctl */
 261         pv_hashed_kern_low_water_mark = MAX(PV_HASHED_KERN_LOW_WATER_MARK_DEFAULT, ((uint32_t)(sane_size >> 30)) * 1000);
 262         pv_hashed_kern_low_water_mark = MIN(pv_hashed_kern_low_water_mark, 16000);
 263         pv_hashed_kern_alloc_chunk = PV_HASHED_KERN_ALLOC_CHUNK_INITIAL;
 264         pv_hashed_alloc_chunk = PV_HASHED_ALLOC_CHUNK_INITIAL;
 265
 266         pv_cnt = 0;
 267         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 268
 269         for (i = 0; i < (5 * PV_HASHED_ALLOC_CHUNK_INITIAL); i++) {
 270                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 271
 272                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 273                 pvh_eh = pvh_e;
 274
 275                 if (pvh_et == PV_HASHED_ENTRY_NULL)
 276                         pvh_et = pvh_e;
 277                 pv_cnt++;
 278         }
 279         PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 280
 281         pv_cnt = 0;
 282         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 283         for (i = 0; i < PV_HASHED_KERN_ALLOC_CHUNK_INITIAL; i++) {
 284                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 285
 286                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 287                 pvh_eh = pvh_e;
 288
 289                 if (pvh_et == PV_HASHED_ENTRY_NULL)
 290                         pvh_et = pvh_e;
 291                 pv_cnt++;
 292         }
 293         PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 294 }
 295
 296 void mapping_replenish(void);
 297
 298 void mapping_adjust(void) {
 299         kern_return_t mres;
 300
 301         pmap_pagetable_corruption_log_setup();
 302
 303         mres = kernel_thread_start_priority((thread_continue_t)mapping_replenish, NULL, MAXPRI_KERNEL, &mapping_replenish_thread);
 304         if (mres != KERN_SUCCESS) {
 305                 panic("pmap: mapping_replenish_thread creation failed");
 306         }
 307         thread_deallocate(mapping_replenish_thread);
 308 }
 309
 310 unsigned pmap_mapping_thread_wakeups;
 311 unsigned pmap_kernel_reserve_replenish_stat;
 312 unsigned pmap_user_reserve_replenish_stat;
 313 unsigned pmap_kern_reserve_alloc_stat;
 314
 315 void mapping_replenish(void)
 316 {
 317         pv_hashed_entry_t       pvh_e;
 318         pv_hashed_entry_t       pvh_eh;
 319         pv_hashed_entry_t       pvh_et;
 320         int                     pv_cnt;
 321         unsigned                i;
 322
 323         /* We qualify for VM privileges...*/
 324         current_thread()->options |= TH_OPT_VMPRIV;
 325
 326         for (;;) {
 327
 328                 while (pv_hashed_kern_free_count < pv_hashed_kern_low_water_mark) {
 329                         pv_cnt = 0;
 330                         pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 331
 332                         for (i = 0; i < pv_hashed_kern_alloc_chunk; i++) {
 333                                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 334                                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 335                                 pvh_eh = pvh_e;
 336
 337                                 if (pvh_et == PV_HASHED_ENTRY_NULL)
 338                                         pvh_et = pvh_e;
 339                                 pv_cnt++;
 340                         }
 341                         pmap_kernel_reserve_replenish_stat += pv_cnt;
 342                         PV_HASHED_KERN_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 343                 }
 344
 345                 pv_cnt = 0;
 346                 pvh_eh = pvh_et = PV_HASHED_ENTRY_NULL;
 347
 348                 if (pv_hashed_free_count < pv_hashed_low_water_mark) {
 349                         for (i = 0; i < pv_hashed_alloc_chunk; i++) {
 350                                 pvh_e = (pv_hashed_entry_t) zalloc(pv_hashed_list_zone);
 351
 352                                 pvh_e->qlink.next = (queue_entry_t)pvh_eh;
 353                                 pvh_eh = pvh_e;
 354
 355                                 if (pvh_et == PV_HASHED_ENTRY_NULL)
 356                                         pvh_et = pvh_e;
 357                                 pv_cnt++;
 358                         }
 359                         pmap_user_reserve_replenish_stat += pv_cnt;
 360                         PV_HASHED_FREE_LIST(pvh_eh, pvh_et, pv_cnt);
 361                 }
 362 /* Wake threads throttled while the kernel reserve was being replenished.
 363  */
 364                 if (pmap_pv_throttled_waiters) {
 365                         pmap_pv_throttled_waiters = 0;
 366                         thread_wakeup(&pmap_user_pv_throttle_event);
 367                 }
 368                 /* Check if the kernel pool has been depleted since the
 369                  * first pass, to reduce refill latency.
 370                  */
 371                 if (pv_hashed_kern_free_count < pv_hashed_kern_low_water_mark)
 372                         continue;
 373                 /* Block sans continuation to avoid yielding kernel stack */
 374                 assert_wait(&mapping_replenish_event, THREAD_UNINT);
 375                 mappingrecurse = 0;
 376                 thread_block(THREAD_CONTINUE_NULL);
 377                 pmap_mapping_thread_wakeups++;
 378         }
 379 }
 380
 381 /*
 382  *      Set specified attribute bits.
 383  */
 384
 385 void
 386 phys_attribute_set(
 387         ppnum_t         pn,
 388         int             bits)
 389 {
 390         int             pai;
 391
 392         pmap_intr_assert();
 393         assert(pn != vm_page_fictitious_addr);
 394         if (pn == vm_page_guard_addr)
 395                 return;
 396
 397         pai = ppn_to_pai(pn);
 398
 399         if (!IS_MANAGED_PAGE(pai)) {
 400                 /* Not a managed page.  */
 401                 return;
 402         }
 403
 404         LOCK_PVH(pai);
 405         pmap_phys_attributes[pai] |= bits;
 406         UNLOCK_PVH(pai);
 407 }
 408
 409 /*
 410  *      Set the modify bit on the specified physical page.
 411  */
 412
 413 void
 414 pmap_set_modify(ppnum_t pn)
 415 {
 416         phys_attribute_set(pn, PHYS_MODIFIED);
 417 }
 418
 419 /*
 420  *      Clear the modify bits on the specified physical page.
 421  */
 422
 423 void
 424 pmap_clear_modify(ppnum_t pn)
 425 {
 426         phys_attribute_clear(pn, PHYS_MODIFIED, 0, NULL);
 427 }
 428
 429 /*
 430  *      pmap_is_modified:
 431  *
 432  *      Return whether or not the specified physical page is modified
 433  *      by any physical maps.
 434  */
 435
 436 boolean_t
 437 pmap_is_modified(ppnum_t pn)
 438 {
 439         if (phys_attribute_test(pn, PHYS_MODIFIED))
 440                 return TRUE;
 441         return FALSE;
 442 }
 443
 444
 445 /*
 446  *      pmap_clear_reference:
 447  *
 448  *      Clear the reference bit on the specified physical page.
 449  */
 450
 451 void
 452 pmap_clear_reference(ppnum_t pn)
 453 {
 454         phys_attribute_clear(pn, PHYS_REFERENCED, 0, NULL);
 455 }
 456
 457 void
 458 pmap_set_reference(ppnum_t pn)
 459 {
 460         phys_attribute_set(pn, PHYS_REFERENCED);
 461 }
 462
 463 /*
 464  *      pmap_is_referenced:
 465  *
 466  *      Return whether or not the specified physical page is referenced
 467  *      by any physical maps.
 468  */
 469
 470 boolean_t
 471 pmap_is_referenced(ppnum_t pn)
 472 {
 473         if (phys_attribute_test(pn, PHYS_REFERENCED))
 474                 return TRUE;
 475         return FALSE;
 476 }
 477
 478
 479 /*
 480  * pmap_get_refmod(phys)
 481  *  returns the referenced and modified bits of the specified
 482  *  physical page.
 483  */
 484 unsigned int
 485 pmap_get_refmod(ppnum_t pn)
 486 {
 487         int             refmod;
 488         unsigned int    retval = 0;
 489
 490         refmod = phys_attribute_test(pn, PHYS_MODIFIED | PHYS_REFERENCED);
 491
 492         if (refmod & PHYS_MODIFIED)
 493                 retval |= VM_MEM_MODIFIED;
 494         if (refmod & PHYS_REFERENCED)
 495                 retval |= VM_MEM_REFERENCED;
 496
 497         return (retval);
 498 }
 499
 500
 501 void
 502 pmap_clear_refmod_options(ppnum_t pn, unsigned int mask, unsigned int options, void *arg)
 503 {
 504         unsigned int  x86Mask;
 505
 506         x86Mask = (   ((mask &   VM_MEM_MODIFIED)?   PHYS_MODIFIED : 0)
 507                       | ((mask & VM_MEM_REFERENCED)? PHYS_REFERENCED : 0));
 508
 509         phys_attribute_clear(pn, x86Mask, options, arg);
 510 }
 511
 512 /*
 513  * pmap_clear_refmod(phys, mask)
 514  *  clears the referenced and modified bits as specified by the mask
 515  *  of the specified physical page.
 516  */
 517 void
 518 pmap_clear_refmod(ppnum_t pn, unsigned int mask)
 519 {
 520         unsigned int  x86Mask;
 521
 522         x86Mask = (   ((mask &   VM_MEM_MODIFIED)?   PHYS_MODIFIED : 0)
 523                     | ((mask & VM_MEM_REFERENCED)? PHYS_REFERENCED : 0));
 524
 525         phys_attribute_clear(pn, x86Mask, 0, NULL);
 526 }
 527
 528 unsigned int
 529 pmap_disconnect(ppnum_t pa)
 530 {
 531         return (pmap_disconnect_options(pa, 0, NULL));
 532 }
 533
 534 /*
 535  *      Routine:
 536  *              pmap_disconnect_options
 537  *
 538  *      Function:
 539  *              Disconnect all mappings for this page and return reference and change status
 540  *              in generic format.
 541  *
 542  */
 543 unsigned int
 544 pmap_disconnect_options(ppnum_t pa, unsigned int options, void *arg)
 545 {
 546         unsigned refmod, vmrefmod = 0;
 547
 548         pmap_page_protect_options(pa, 0, options, arg);         /* disconnect the page */
 549
 550         pmap_assert(pa != vm_page_fictitious_addr);
 551         if ((pa == vm_page_guard_addr) || !IS_MANAGED_PAGE(pa) || (options & PMAP_OPTIONS_NOREFMOD))
 552                 return 0;
 553         refmod = pmap_phys_attributes[pa] & (PHYS_MODIFIED | PHYS_REFERENCED);
 554
 555         if (refmod & PHYS_MODIFIED)
 556                 vmrefmod |= VM_MEM_MODIFIED;
 557         if (refmod & PHYS_REFERENCED)
 558                 vmrefmod |= VM_MEM_REFERENCED;
 559
 560         return vmrefmod;
 561 }