osfmk/i386/i386_vm_init.c

   1 /*
   2  * Copyright (c) 2003-2008 Apple Computer, 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 /*
  29  * @OSF_COPYRIGHT@
  30  */
  31 /*
  32  * Mach Operating System
  33  * Copyright (c) 1991,1990,1989, 1988 Carnegie Mellon University
  34  * All Rights Reserved.
  35  *
  36  * Permission to use, copy, modify and distribute this software and its
  37  * documentation is hereby granted, provided that both the copyright
  38  * notice and this permission notice appear in all copies of the
  39  * software, derivative works or modified versions, and any portions
  40  * thereof, and that both notices appear in supporting documentation.
  41  *
  42  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  43  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  44  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  45  *
  46  * Carnegie Mellon requests users of this software to return to
  47  *
  48  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  49  *  School of Computer Science
  50  *  Carnegie Mellon University
  51  *  Pittsburgh PA 15213-3890
  52  *
  53  * any improvements or extensions that they make and grant Carnegie Mellon
  54  * the rights to redistribute these changes.
  55  */
  56
  57 #include <platforms.h>
  58
  59 #include <mach/i386/vm_param.h>
  60
  61 #include <string.h>
  62 #include <mach/vm_param.h>
  63 #include <mach/vm_prot.h>
  64 #include <mach/machine.h>
  65 #include <mach/time_value.h>
  66 #include <kern/spl.h>
  67 #include <kern/assert.h>
  68 #include <kern/debug.h>
  69 #include <kern/misc_protos.h>
  70 #include <kern/cpu_data.h>
  71 #include <kern/processor.h>
  72 #include <vm/vm_page.h>
  73 #include <vm/pmap.h>
  74 #include <vm/vm_kern.h>
  75 #include <i386/pmap.h>
  76 #include <i386/misc_protos.h>
  77 #include <i386/cpuid.h>
  78 #include <mach/thread_status.h>
  79 #include <pexpert/i386/efi.h>
  80 #include <i386/i386_lowmem.h>
  81 #ifdef __x86_64__
  82 #include <x86_64/lowglobals.h>
  83 #else
  84 #include <i386/lowglobals.h>
  85 #endif
  86 #include <i386/pal_routines.h>
  87
  88 #include <mach-o/loader.h>
  89 #include <libkern/kernel_mach_header.h>
  90
  91
  92 vm_size_t       mem_size = 0;
  93 pmap_paddr_t    first_avail = 0;/* first after page tables */
  94
  95 uint64_t        max_mem;        /* Size of physical memory (bytes), adjusted by maxmem */
  96 uint64_t        mem_actual;
  97 uint64_t        sane_size = 0;  /* Memory size for defaults calculations */
  98
  99 /*
 100  * KASLR parameters
 101  */
 102 ppnum_t         vm_kernel_base_page;
 103 vm_offset_t     vm_kernel_base;
 104 vm_offset_t     vm_kernel_top;
 105 vm_offset_t     vm_kernel_stext;
 106 vm_offset_t     vm_kernel_etext;
 107 vm_offset_t     vm_kernel_slide;
 108
 109 #define MAXLORESERVE    (32 * 1024 * 1024)
 110
 111 ppnum_t         max_ppnum = 0;
 112 ppnum_t         lowest_lo = 0;
 113 ppnum_t         lowest_hi = 0;
 114 ppnum_t         highest_hi = 0;
 115
 116 enum {PMAP_MAX_RESERVED_RANGES = 32};
 117 uint32_t pmap_reserved_pages_allocated = 0;
 118 uint32_t pmap_reserved_range_indices[PMAP_MAX_RESERVED_RANGES];
 119 uint32_t pmap_last_reserved_range_index = 0;
 120 uint32_t pmap_reserved_ranges = 0;
 121
 122 extern unsigned int bsd_mbuf_cluster_reserve(boolean_t *);
 123
 124 pmap_paddr_t     avail_start, avail_end;
 125 vm_offset_t     virtual_avail, virtual_end;
 126 static pmap_paddr_t     avail_remaining;
 127 vm_offset_t     static_memory_end = 0;
 128
 129 vm_offset_t     sHIB, eHIB, stext, etext, sdata, edata, sconstdata, econstdata, end;
 130
 131 /*
 132  * _mh_execute_header is the mach_header for the currently executing kernel
 133  */
 134 vm_offset_t segTEXTB; unsigned long segSizeTEXT;
 135 vm_offset_t segDATAB; unsigned long segSizeDATA;
 136 vm_offset_t segLINKB; unsigned long segSizeLINK;
 137 vm_offset_t segPRELINKB; unsigned long segSizePRELINK;
 138 vm_offset_t segHIBB; unsigned long segSizeHIB;
 139 vm_offset_t sectCONSTB; unsigned long sectSizeConst;
 140
 141 boolean_t doconstro_override = FALSE;
 142
 143 static kernel_segment_command_t *segTEXT, *segDATA;
 144 static kernel_section_t *cursectTEXT, *lastsectTEXT;
 145 static kernel_section_t *sectDCONST;
 146
 147 extern uint64_t firmware_Conventional_bytes;
 148 extern uint64_t firmware_RuntimeServices_bytes;
 149 extern uint64_t firmware_ACPIReclaim_bytes;
 150 extern uint64_t firmware_ACPINVS_bytes;
 151 extern uint64_t firmware_PalCode_bytes;
 152 extern uint64_t firmware_Reserved_bytes;
 153 extern uint64_t firmware_Unusable_bytes;
 154 extern uint64_t firmware_other_bytes;
 155 uint64_t firmware_MMIO_bytes;
 156
 157 /*
 158  * Linker magic to establish the highest address in the kernel.
 159  * This is replicated from libsa which marks last_kernel_symbol
 160  * but that's not visible from here in osfmk.
 161  */
 162 __asm__(".zerofill __LAST, __last, _kernel_top, 0");
 163 extern void     *kernel_top;
 164
 165 #if     DEBUG
 166 #define PRINT_PMAP_MEMORY_TABLE
 167 #define DBG(x...)       kprintf(x)
 168 #else
 169 #define DBG(x...)
 170 #endif /* DEBUG */
 171 /*
 172  * Basic VM initialization.
 173  */
 174 void
 175 i386_vm_init(uint64_t   maxmem,
 176              boolean_t  IA32e,
 177              boot_args  *args)
 178 {
 179         pmap_memory_region_t *pmptr;
 180         pmap_memory_region_t *prev_pmptr;
 181         EfiMemoryRange *mptr;
 182         unsigned int mcount;
 183         unsigned int msize;
 184         ppnum_t fap;
 185         unsigned int i;
 186         unsigned int safeboot;
 187         ppnum_t maxpg = 0;
 188         uint32_t pmap_type;
 189         uint32_t maxloreserve;
 190         uint32_t maxdmaaddr;
 191         uint32_t  mbuf_reserve = 0;
 192         boolean_t mbuf_override = FALSE;
 193         boolean_t coalescing_permitted;
 194         vm_kernel_base_page = i386_btop(args->kaddr);
 195 #ifdef __x86_64__
 196         vm_offset_t base_address;
 197         vm_offset_t static_base_address;
 198
 199         /*
 200          * Establish the KASLR parameters.
 201          */
 202         static_base_address = ml_static_ptovirt(KERNEL_BASE_OFFSET);
 203         base_address        = ml_static_ptovirt(args->kaddr);
 204         vm_kernel_slide     = base_address - static_base_address;
 205         if (args->kslide) {
 206                 kprintf("KASLR slide: 0x%016lx dynamic\n", vm_kernel_slide);
 207                 if (vm_kernel_slide != ((vm_offset_t)args->kslide))
 208                         panic("Kernel base inconsistent with slide - rebased?");
 209         } else {
 210                 /* No slide relative to on-disk symbols */
 211                 kprintf("KASLR slide: 0x%016lx static and ignored\n",
 212                         vm_kernel_slide);
 213                 vm_kernel_slide = 0;
 214         }
 215
 216         /*
 217          * Zero out local relocations to avoid confusing kxld.
 218          * TODO: might be better to move this code to OSKext::initialize
 219          */
 220         if (_mh_execute_header.flags & MH_PIE) {
 221                 struct load_command *loadcmd;
 222                 uint32_t cmd;
 223
 224                 loadcmd = (struct load_command *)((uintptr_t)&_mh_execute_header +
 225                                                   sizeof (_mh_execute_header));
 226
 227                 for (cmd = 0; cmd < _mh_execute_header.ncmds; cmd++) {
 228                         if (loadcmd->cmd == LC_DYSYMTAB) {
 229                                 struct dysymtab_command *dysymtab;
 230
 231                                 dysymtab = (struct dysymtab_command *)loadcmd;
 232                                 dysymtab->nlocrel = 0;
 233                                 dysymtab->locreloff = 0;
 234                                 kprintf("Hiding local relocations\n");
 235                                 break;
 236                         }
 237                         loadcmd = (struct load_command *)((uintptr_t)loadcmd + loadcmd->cmdsize);
 238                 }
 239         }
 240
 241 #endif // __x86_64__
 242
 243         /*
 244          * Now retrieve addresses for end, edata, and etext
 245          * from MACH-O headers.
 246          */
 247         segTEXTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 248                                         "__TEXT", &segSizeTEXT);
 249         segDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 250                                         "__DATA", &segSizeDATA);
 251         segLINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 252                                         "__LINKEDIT", &segSizeLINK);
 253         segHIBB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 254                                         "__HIB", &segSizeHIB);
 255         segPRELINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 256                                         "__PRELINK_TEXT", &segSizePRELINK);
 257         segTEXT = getsegbynamefromheader(&_mh_execute_header,
 258                                         "__TEXT");
 259         segDATA = getsegbynamefromheader(&_mh_execute_header,
 260                                         "__DATA");
 261         sectDCONST = getsectbynamefromheader(&_mh_execute_header,
 262                                         "__DATA", "__const");
 263         cursectTEXT = lastsectTEXT = firstsect(segTEXT);
 264         /* Discover the last TEXT section within the TEXT segment */
 265         while ((cursectTEXT = nextsect(segTEXT, cursectTEXT)) != NULL) {
 266                 lastsectTEXT = cursectTEXT;
 267         }
 268
 269         sHIB  = segHIBB;
 270         eHIB  = segHIBB + segSizeHIB;
 271         /* Zero-padded from ehib to stext if text is 2M-aligned */
 272         stext = segTEXTB;
 273 #ifdef __x86_64__
 274         lowGlo.lgStext = stext;
 275 #endif
 276         etext = (vm_offset_t) round_page_64(lastsectTEXT->addr + lastsectTEXT->size);
 277         /* Zero-padded from etext to sdata if text is 2M-aligned */
 278         sdata = segDATAB;
 279         edata = segDATAB + segSizeDATA;
 280
 281         sectCONSTB = (vm_offset_t) sectDCONST->addr;
 282         sectSizeConst = sectDCONST->size;
 283         sconstdata = sectCONSTB;
 284         econstdata = sectCONSTB + sectSizeConst;
 285
 286         if (sectSizeConst & PAGE_MASK) {
 287                 kernel_section_t *ns = nextsect(segDATA, sectDCONST);
 288                 if (ns && !(ns->addr & PAGE_MASK))
 289                         doconstro_override = TRUE;
 290         } else
 291                 doconstro_override = TRUE;
 292
 293         DBG("segTEXTB    = %p\n", (void *) segTEXTB);
 294         DBG("segDATAB    = %p\n", (void *) segDATAB);
 295         DBG("segLINKB    = %p\n", (void *) segLINKB);
 296         DBG("segHIBB     = %p\n", (void *) segHIBB);
 297         DBG("segPRELINKB = %p\n", (void *) segPRELINKB);
 298         DBG("sHIB        = %p\n", (void *) sHIB);
 299         DBG("eHIB        = %p\n", (void *) eHIB);
 300         DBG("stext       = %p\n", (void *) stext);
 301         DBG("etext       = %p\n", (void *) etext);
 302         DBG("sdata       = %p\n", (void *) sdata);
 303         DBG("edata       = %p\n", (void *) edata);
 304         DBG("sconstdata  = %p\n", (void *) sconstdata);
 305         DBG("econstdata  = %p\n", (void *) econstdata);
 306         DBG("kernel_top  = %p\n", (void *) &kernel_top);
 307
 308         vm_kernel_base  = sHIB;
 309         vm_kernel_top   = (vm_offset_t) &kernel_top;
 310         vm_kernel_stext = stext;
 311         vm_kernel_etext = etext;
 312
 313         vm_set_page_size();
 314
 315         /*
 316          * Compute the memory size.
 317          */
 318
 319         if ((1 == vm_himemory_mode) || PE_parse_boot_argn("-x", &safeboot, sizeof (safeboot))) {
 320                 maxpg = 1 << (32 - I386_PGSHIFT);
 321         }
 322         avail_remaining = 0;
 323         avail_end = 0;
 324         pmptr = pmap_memory_regions;
 325         prev_pmptr = 0;
 326         pmap_memory_region_count = pmap_memory_region_current = 0;
 327         fap = (ppnum_t) i386_btop(first_avail);
 328
 329         mptr = (EfiMemoryRange *)ml_static_ptovirt((vm_offset_t)args->MemoryMap);
 330         if (args->MemoryMapDescriptorSize == 0)
 331                 panic("Invalid memory map descriptor size");
 332         msize = args->MemoryMapDescriptorSize;
 333         mcount = args->MemoryMapSize / msize;
 334
 335 #define FOURGIG 0x0000000100000000ULL
 336 #define ONEGIG  0x0000000040000000ULL
 337
 338         for (i = 0; i < mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
 339                 ppnum_t base, top;
 340                 uint64_t region_bytes = 0;
 341
 342                 if (pmap_memory_region_count >= PMAP_MEMORY_REGIONS_SIZE) {
 343                         kprintf("WARNING: truncating memory region count at %d\n", pmap_memory_region_count);
 344                         break;
 345                 }
 346                 base = (ppnum_t) (mptr->PhysicalStart >> I386_PGSHIFT);
 347                 top = (ppnum_t) (((mptr->PhysicalStart) >> I386_PGSHIFT) + mptr->NumberOfPages - 1);
 348
 349 #if     MR_RSV_TEST
 350                 static uint32_t nmr = 0;
 351                 if ((base > 0x20000) && (nmr++ < 4))
 352                         mptr->Attribute |= EFI_MEMORY_KERN_RESERVED;
 353 #endif
 354                 region_bytes = (uint64_t)(mptr->NumberOfPages << I386_PGSHIFT);
 355                 pmap_type = mptr->Type;
 356
 357                 switch (mptr->Type) {
 358                 case kEfiLoaderCode:
 359                 case kEfiLoaderData:
 360                 case kEfiBootServicesCode:
 361                 case kEfiBootServicesData:
 362                 case kEfiConventionalMemory:
 363                         /*
 364                          * Consolidate usable memory types into one.
 365                          */
 366                         pmap_type = kEfiConventionalMemory;
 367                         sane_size += region_bytes;
 368                         firmware_Conventional_bytes += region_bytes;
 369                         break;
 370                         /*
 371                          * sane_size should reflect the total amount of physical
 372                          * RAM in the system, not just the amount that is
 373                          * available for the OS to use.
 374                          * FIXME:Consider deriving this value from SMBIOS tables
 375                          * rather than reverse engineering the memory map.
 376                          * Alternatively, see
 377                          * <rdar://problem/4642773> Memory map should
 378                          * describe all memory
 379                          * Firmware on some systems guarantees that the memory
 380                          * map is complete via the "RomReservedMemoryTracked"
 381                          * feature field--consult that where possible to
 382                          * avoid the "round up to 128M" workaround below.
 383                          */
 384
 385                 case kEfiRuntimeServicesCode:
 386                 case kEfiRuntimeServicesData:
 387                         firmware_RuntimeServices_bytes += region_bytes;
 388                         sane_size += region_bytes;
 389                         break;
 390                 case kEfiACPIReclaimMemory:
 391                         firmware_ACPIReclaim_bytes += region_bytes;
 392                         sane_size += region_bytes;
 393                         break;
 394                 case kEfiACPIMemoryNVS:
 395                         firmware_ACPINVS_bytes += region_bytes;
 396                         sane_size += region_bytes;
 397                         break;
 398                 case kEfiPalCode:
 399                         firmware_PalCode_bytes += region_bytes;
 400                         sane_size += region_bytes;
 401                         break;
 402
 403                 case kEfiReservedMemoryType:
 404                         firmware_Reserved_bytes += region_bytes;
 405                         break;
 406                 case kEfiUnusableMemory:
 407                         firmware_Unusable_bytes += region_bytes;
 408                         break;
 409                 case kEfiMemoryMappedIO:
 410                 case kEfiMemoryMappedIOPortSpace:
 411                         firmware_MMIO_bytes += region_bytes;
 412                         break;
 413                 default:
 414                         firmware_other_bytes += region_bytes;
 415                         break;
 416                 }
 417
 418                 DBG("EFI region %d: type %u/%d, base 0x%x, top 0x%x %s\n",
 419                     i, mptr->Type, pmap_type, base, top,
 420                     (mptr->Attribute&EFI_MEMORY_KERN_RESERVED)? "RESERVED" :
 421                     (mptr->Attribute&EFI_MEMORY_RUNTIME)? "RUNTIME" : "");
 422
 423                 if (maxpg) {
 424                         if (base >= maxpg)
 425                                 break;
 426                         top = (top > maxpg) ? maxpg : top;
 427                 }
 428
 429                 /*
 430                  * handle each region
 431                  */
 432                 if ((mptr->Attribute & EFI_MEMORY_RUNTIME) == EFI_MEMORY_RUNTIME ||
 433                     pmap_type != kEfiConventionalMemory) {
 434                         prev_pmptr = 0;
 435                         continue;
 436                 } else {
 437                         /*
 438                          * Usable memory region
 439                          */
 440                         if (top < I386_LOWMEM_RESERVED ||
 441                             !pal_is_usable_memory(base, top)) {
 442                                 prev_pmptr = 0;
 443                                 continue;
 444                         }
 445                         /*
 446                          * A range may be marked with with the
 447                          * EFI_MEMORY_KERN_RESERVED attribute
 448                          * on some systems, to indicate that the range
 449                          * must not be made available to devices.
 450                          */
 451
 452                         if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED) {
 453                                 if (++pmap_reserved_ranges > PMAP_MAX_RESERVED_RANGES) {
 454                                         panic("Too many reserved ranges %u\n", pmap_reserved_ranges);
 455                                 }
 456                         }
 457
 458                         if (top < fap) {
 459                                 /*
 460                                  * entire range below first_avail
 461                                  * salvage some low memory pages
 462                                  * we use some very low memory at startup
 463                                  * mark as already allocated here
 464                                  */
 465                                 if (base >= I386_LOWMEM_RESERVED)
 466                                         pmptr->base = base;
 467                                 else
 468                                         pmptr->base = I386_LOWMEM_RESERVED;
 469
 470                                 pmptr->end = top;
 471
 472
 473                                 if ((mptr->Attribute & EFI_MEMORY_KERN_RESERVED) &&
 474                                     (top < vm_kernel_base_page)) {
 475                                         pmptr->alloc = pmptr->base;
 476                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 477                                 }
 478                                 else {
 479                                         /*
 480                                          * mark as already mapped
 481                                          */
 482                                         pmptr->alloc = top;
 483                                 }
 484                                 pmptr->type = pmap_type;
 485                                 pmptr->attribute = mptr->Attribute;
 486                         }
 487                         else if ( (base < fap) && (top > fap) ) {
 488                                 /*
 489                                  * spans first_avail
 490                                  * put mem below first avail in table but
 491                                  * mark already allocated
 492                                  */
 493                                 pmptr->base = base;
 494                                 pmptr->alloc = pmptr->end = (fap - 1);
 495                                 pmptr->type = pmap_type;
 496                                 pmptr->attribute = mptr->Attribute;
 497                                 /*
 498                                  * we bump these here inline so the accounting
 499                                  * below works correctly
 500                                  */
 501                                 pmptr++;
 502                                 pmap_memory_region_count++;
 503
 504                                 pmptr->alloc = pmptr->base = fap;
 505                                 pmptr->type = pmap_type;
 506                                 pmptr->attribute = mptr->Attribute;
 507                                 pmptr->end = top;
 508
 509                                 if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
 510                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 511                         } else {
 512                                 /*
 513                                  * entire range useable
 514                                  */
 515                                 pmptr->alloc = pmptr->base = base;
 516                                 pmptr->type = pmap_type;
 517                                 pmptr->attribute = mptr->Attribute;
 518                                 pmptr->end = top;
 519                                 if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
 520                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 521                         }
 522
 523                         if (i386_ptob(pmptr->end) > avail_end )
 524                                 avail_end = i386_ptob(pmptr->end);
 525
 526                         avail_remaining += (pmptr->end - pmptr->base);
 527                         coalescing_permitted = (prev_pmptr && (pmptr->attribute == prev_pmptr->attribute) && ((pmptr->attribute & EFI_MEMORY_KERN_RESERVED) == 0));
 528                         /*
 529                          * Consolidate contiguous memory regions, if possible
 530                          */
 531                         if (prev_pmptr &&
 532                             (pmptr->type == prev_pmptr->type) &&
 533                             (coalescing_permitted) &&
 534                             (pmptr->base == pmptr->alloc) &&
 535                             (pmptr->base == (prev_pmptr->end + 1)))
 536                         {
 537                                 if (prev_pmptr->end == prev_pmptr->alloc)
 538                                         prev_pmptr->alloc = pmptr->base;
 539                                 prev_pmptr->end = pmptr->end;
 540                         } else {
 541                                 pmap_memory_region_count++;
 542                                 prev_pmptr = pmptr;
 543                                 pmptr++;
 544                         }
 545                 }
 546         }
 547
 548 #ifdef PRINT_PMAP_MEMORY_TABLE
 549         {
 550         unsigned int j;
 551         pmap_memory_region_t *p = pmap_memory_regions;
 552         addr64_t region_start, region_end;
 553         addr64_t efi_start, efi_end;
 554         for (j=0;j<pmap_memory_region_count;j++, p++) {
 555             kprintf("pmap region %d type %d base 0x%llx alloc 0x%llx top 0x%llx\n",
 556                     j, p->type,
 557                     (addr64_t) p->base  << I386_PGSHIFT,
 558                     (addr64_t) p->alloc << I386_PGSHIFT,
 559                     (addr64_t) p->end   << I386_PGSHIFT);
 560             region_start = (addr64_t) p->base << I386_PGSHIFT;
 561             region_end = ((addr64_t) p->end << I386_PGSHIFT) - 1;
 562             mptr = (EfiMemoryRange *) ml_static_ptovirt((vm_offset_t)args->MemoryMap);
 563             for (i=0; i<mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
 564                 if (mptr->Type != kEfiLoaderCode &&
 565                     mptr->Type != kEfiLoaderData &&
 566                     mptr->Type != kEfiBootServicesCode &&
 567                     mptr->Type != kEfiBootServicesData &&
 568                     mptr->Type != kEfiConventionalMemory) {
 569                 efi_start = (addr64_t)mptr->PhysicalStart;
 570                 efi_end = efi_start + ((vm_offset_t)mptr->NumberOfPages << I386_PGSHIFT) - 1;
 571                 if ((efi_start >= region_start && efi_start <= region_end) ||
 572                     (efi_end >= region_start && efi_end <= region_end)) {
 573                     kprintf(" *** Overlapping region with EFI runtime region %d\n", i);
 574                 }
 575               }
 576             }
 577           }
 578         }
 579 #endif
 580
 581         avail_start = first_avail;
 582         mem_actual = sane_size;
 583
 584         /*
 585          * For user visible memory size, round up to 128 Mb - accounting for the various stolen memory
 586          * not reported by EFI.
 587          */
 588
 589         sane_size = (sane_size + 128 * MB - 1) & ~((uint64_t)(128 * MB - 1));
 590
 591         /*
 592          * We cap at KERNEL_MAXMEM bytes (currently 32GB for K32, 96GB for K64).
 593          * Unless overriden by the maxmem= boot-arg
 594          * -- which is a non-zero maxmem argument to this function.
 595          */
 596         if (maxmem == 0 && sane_size > KERNEL_MAXMEM) {
 597                 maxmem = KERNEL_MAXMEM;
 598                 printf("Physical memory %lld bytes capped at %dGB\n",
 599                         sane_size, (uint32_t) (KERNEL_MAXMEM/GB));
 600         }
 601
 602         /*
 603          * if user set maxmem, reduce memory sizes
 604          */
 605         if ( (maxmem > (uint64_t)first_avail) && (maxmem < sane_size)) {
 606                 ppnum_t discarded_pages  = (ppnum_t)((sane_size - maxmem) >> I386_PGSHIFT);
 607                 ppnum_t highest_pn = 0;
 608                 ppnum_t cur_end  = 0;
 609                 uint64_t        pages_to_use;
 610                 unsigned        cur_region = 0;
 611
 612                 sane_size = maxmem;
 613
 614                 if (avail_remaining > discarded_pages)
 615                         avail_remaining -= discarded_pages;
 616                 else
 617                         avail_remaining = 0;
 618
 619                 pages_to_use = avail_remaining;
 620
 621                 while (cur_region < pmap_memory_region_count && pages_to_use) {
 622                         for (cur_end = pmap_memory_regions[cur_region].base;
 623                              cur_end < pmap_memory_regions[cur_region].end && pages_to_use;
 624                              cur_end++) {
 625                                 if (cur_end > highest_pn)
 626                                         highest_pn = cur_end;
 627                                 pages_to_use--;
 628                         }
 629                         if (pages_to_use == 0)
 630                                 pmap_memory_regions[cur_region].end = cur_end;
 631
 632                         cur_region++;
 633                 }
 634                 pmap_memory_region_count = cur_region;
 635
 636                 avail_end = i386_ptob(highest_pn + 1);
 637         }
 638
 639         /*
 640          * mem_size is only a 32 bit container... follow the PPC route
 641          * and pin it to a 2 Gbyte maximum
 642          */
 643         if (sane_size > (FOURGIG >> 1))
 644                 mem_size = (vm_size_t)(FOURGIG >> 1);
 645         else
 646                 mem_size = (vm_size_t)sane_size;
 647         max_mem = sane_size;
 648
 649         kprintf("Physical memory %llu MB\n", sane_size/MB);
 650
 651         max_valid_low_ppnum = (2 * GB) / PAGE_SIZE;
 652
 653         if (!PE_parse_boot_argn("max_valid_dma_addr", &maxdmaaddr, sizeof (maxdmaaddr))) {
 654                 max_valid_dma_address = (uint64_t)4 * (uint64_t)GB;
 655         } else {
 656                 max_valid_dma_address = ((uint64_t) maxdmaaddr) * MB;
 657
 658                 if ((max_valid_dma_address / PAGE_SIZE) < max_valid_low_ppnum)
 659                         max_valid_low_ppnum = (ppnum_t)(max_valid_dma_address / PAGE_SIZE);
 660         }
 661         if (avail_end >= max_valid_dma_address) {
 662
 663                 if (!PE_parse_boot_argn("maxloreserve", &maxloreserve, sizeof (maxloreserve))) {
 664
 665                         if (sane_size >= (ONEGIG * 15))
 666                                 maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 4;
 667                         else if (sane_size >= (ONEGIG * 7))
 668                                 maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 2;
 669                         else
 670                                 maxloreserve = MAXLORESERVE / PAGE_SIZE;
 671
 672                         mbuf_reserve = bsd_mbuf_cluster_reserve(&mbuf_override) / PAGE_SIZE;
 673                 } else
 674                         maxloreserve = (maxloreserve * (1024 * 1024)) / PAGE_SIZE;
 675
 676                 if (maxloreserve) {
 677                         vm_lopage_free_limit = maxloreserve;
 678
 679                         if (mbuf_override == TRUE) {
 680                                 vm_lopage_free_limit += mbuf_reserve;
 681                                 vm_lopage_lowater = 0;
 682                         } else
 683                                 vm_lopage_lowater = vm_lopage_free_limit / 16;
 684
 685                         vm_lopage_refill = TRUE;
 686                         vm_lopage_needed = TRUE;
 687                 }
 688         }
 689
 690         /*
 691          *      Initialize kernel physical map.
 692          *      Kernel virtual address starts at VM_KERNEL_MIN_ADDRESS.
 693          */
 694         kprintf("avail_remaining = 0x%lx\n", (unsigned long)avail_remaining);
 695         pmap_bootstrap(0, IA32e);
 696 }
 697
 698
 699 unsigned int
 700 pmap_free_pages(void)
 701 {
 702         return (unsigned int)avail_remaining;
 703 }
 704
 705
 706 boolean_t pmap_next_page_reserved(ppnum_t *);
 707
 708 /*
 709  * Pick a page from a "kernel private" reserved range; works around
 710  * errata on some hardware.
 711  */
 712 boolean_t
 713 pmap_next_page_reserved(ppnum_t *pn) {
 714         if (pmap_reserved_ranges) {
 715                 uint32_t n;
 716                 pmap_memory_region_t *region;
 717                 for (n = 0; n < pmap_last_reserved_range_index; n++) {
 718                         uint32_t reserved_index = pmap_reserved_range_indices[n];
 719                         region = &pmap_memory_regions[reserved_index];
 720                         if (region->alloc < region->end) {
 721                                 *pn = region->alloc++;
 722                                 avail_remaining--;
 723
 724                                 if (*pn > max_ppnum)
 725                                         max_ppnum = *pn;
 726
 727                                 if (lowest_lo == 0 || *pn < lowest_lo)
 728                                         lowest_lo = *pn;
 729
 730                                 pmap_reserved_pages_allocated++;
 731 #if DEBUG
 732                                 if (region->alloc == region->end) {
 733                                         kprintf("Exhausted reserved range index: %u, base: 0x%x end: 0x%x, type: 0x%x, attribute: 0x%llx\n", reserved_index, region->base, region->end, region->type, region->attribute);
 734                                 }
 735 #endif
 736                                 return TRUE;
 737                         }
 738                 }
 739         }
 740         return FALSE;
 741 }
 742
 743
 744 boolean_t
 745 pmap_next_page_hi(
 746                   ppnum_t *pn)
 747 {
 748         pmap_memory_region_t *region;
 749         int     n;
 750
 751         if (pmap_next_page_reserved(pn))
 752                 return TRUE;
 753
 754         if (avail_remaining) {
 755                 for (n = pmap_memory_region_count - 1; n >= 0; n--) {
 756                         region = &pmap_memory_regions[n];
 757
 758                         if (region->alloc != region->end) {
 759                                 *pn = region->alloc++;
 760                                 avail_remaining--;
 761
 762                                 if (*pn > max_ppnum)
 763                                         max_ppnum = *pn;
 764
 765                                 if (lowest_lo == 0 || *pn < lowest_lo)
 766                                         lowest_lo = *pn;
 767
 768                                 if (lowest_hi == 0 || *pn < lowest_hi)
 769                                         lowest_hi = *pn;
 770
 771                                 if (*pn > highest_hi)
 772                                         highest_hi = *pn;
 773
 774                                 return TRUE;
 775                         }
 776                 }
 777         }
 778         return FALSE;
 779 }
 780
 781
 782 boolean_t
 783 pmap_next_page(
 784                ppnum_t *pn)
 785 {
 786         if (avail_remaining) while (pmap_memory_region_current < pmap_memory_region_count) {
 787                 if (pmap_memory_regions[pmap_memory_region_current].alloc ==
 788                                 pmap_memory_regions[pmap_memory_region_current].end) {
 789                         pmap_memory_region_current++;
 790                         continue;
 791                 }
 792                 *pn = pmap_memory_regions[pmap_memory_region_current].alloc++;
 793                 avail_remaining--;
 794
 795                 if (*pn > max_ppnum)
 796                         max_ppnum = *pn;
 797
 798                 if (lowest_lo == 0 || *pn < lowest_lo)
 799                         lowest_lo = *pn;
 800
 801                 return TRUE;
 802         }
 803         return FALSE;
 804 }
 805
 806
 807 boolean_t
 808 pmap_valid_page(
 809         ppnum_t pn)
 810 {
 811         unsigned int i;
 812         pmap_memory_region_t *pmptr = pmap_memory_regions;
 813
 814         for (i = 0; i < pmap_memory_region_count; i++, pmptr++) {
 815                 if ( (pn >= pmptr->base) && (pn <= pmptr->end) )
 816                         return TRUE;
 817         }
 818         return FALSE;
 819 }
 820