osfmk/i386/i386_vm_init.c

   1 /*
   2  * Copyright (c) 2003-2012 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 /*
  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
  58 #include <mach/i386/vm_param.h>
  59
  60 #include <string.h>
  61 #include <mach/vm_param.h>
  62 #include <mach/vm_prot.h>
  63 #include <mach/machine.h>
  64 #include <mach/time_value.h>
  65 #include <kern/spl.h>
  66 #include <kern/assert.h>
  67 #include <kern/debug.h>
  68 #include <kern/misc_protos.h>
  69 #include <kern/cpu_data.h>
  70 #include <kern/processor.h>
  71 #include <vm/vm_page.h>
  72 #include <vm/pmap.h>
  73 #include <vm/vm_kern.h>
  74 #include <i386/pmap.h>
  75 #include <i386/misc_protos.h>
  76 #include <i386/cpuid.h>
  77 #include <mach/thread_status.h>
  78 #include <pexpert/i386/efi.h>
  79 #include <i386/i386_lowmem.h>
  80 #include <x86_64/lowglobals.h>
  81 #include <i386/pal_routines.h>
  82
  83 #include <mach-o/loader.h>
  84 #include <libkern/kernel_mach_header.h>
  85
  86
  87 vm_size_t       mem_size = 0;
  88 pmap_paddr_t    first_avail = 0;/* first after page tables */
  89
  90 uint64_t        max_mem;        /* Size of physical memory (bytes), adjusted by maxmem */
  91 uint64_t        mem_actual;
  92 uint64_t        sane_size = 0;  /* Memory size for defaults calculations */
  93
  94 /*
  95  * KASLR parameters
  96  */
  97 ppnum_t         vm_kernel_base_page;
  98 vm_offset_t     vm_kernel_base;
  99 vm_offset_t     vm_kernel_top;
 100 vm_offset_t     vm_kernel_stext;
 101 vm_offset_t     vm_kernel_etext;
 102 vm_offset_t     vm_kernel_slide;
 103 vm_offset_t     vm_kernel_slid_base;
 104 vm_offset_t     vm_kernel_slid_top;
 105 vm_offset_t vm_hib_base;
 106 vm_offset_t     vm_kext_base = VM_MIN_KERNEL_AND_KEXT_ADDRESS;
 107 vm_offset_t     vm_kext_top = VM_MIN_KERNEL_ADDRESS;
 108
 109 vm_offset_t vm_prelink_stext;
 110 vm_offset_t vm_prelink_etext;
 111 vm_offset_t vm_prelink_sinfo;
 112 vm_offset_t vm_prelink_einfo;
 113 vm_offset_t vm_slinkedit;
 114 vm_offset_t vm_elinkedit;
 115
 116 #define MAXLORESERVE    (32 * 1024 * 1024)
 117
 118 ppnum_t         max_ppnum = 0;
 119 ppnum_t         lowest_lo = 0;
 120 ppnum_t         lowest_hi = 0;
 121 ppnum_t         highest_hi = 0;
 122
 123 enum {PMAP_MAX_RESERVED_RANGES = 32};
 124 uint32_t pmap_reserved_pages_allocated = 0;
 125 uint32_t pmap_reserved_range_indices[PMAP_MAX_RESERVED_RANGES];
 126 uint32_t pmap_last_reserved_range_index = 0;
 127 uint32_t pmap_reserved_ranges = 0;
 128
 129 extern unsigned int bsd_mbuf_cluster_reserve(boolean_t *);
 130
 131 pmap_paddr_t     avail_start, avail_end;
 132 vm_offset_t     virtual_avail, virtual_end;
 133 static pmap_paddr_t     avail_remaining;
 134 vm_offset_t     static_memory_end = 0;
 135
 136 vm_offset_t     sHIB, eHIB, stext, etext, sdata, edata, end, sconst, econst;
 137
 138 /*
 139  * _mh_execute_header is the mach_header for the currently executing kernel
 140  */
 141 vm_offset_t segTEXTB; unsigned long segSizeTEXT;
 142 vm_offset_t segDATAB; unsigned long segSizeDATA;
 143 vm_offset_t segLINKB; unsigned long segSizeLINK;
 144 vm_offset_t segPRELINKTEXTB; unsigned long segSizePRELINKTEXT;
 145 vm_offset_t segPRELINKINFOB; unsigned long segSizePRELINKINFO;
 146 vm_offset_t segHIBB; unsigned long segSizeHIB;
 147 unsigned long segSizeConst;
 148
 149 static kernel_segment_command_t *segTEXT, *segDATA;
 150 static kernel_section_t *cursectTEXT, *lastsectTEXT;
 151 static kernel_segment_command_t *segCONST;
 152
 153 extern uint64_t firmware_Conventional_bytes;
 154 extern uint64_t firmware_RuntimeServices_bytes;
 155 extern uint64_t firmware_ACPIReclaim_bytes;
 156 extern uint64_t firmware_ACPINVS_bytes;
 157 extern uint64_t firmware_PalCode_bytes;
 158 extern uint64_t firmware_Reserved_bytes;
 159 extern uint64_t firmware_Unusable_bytes;
 160 extern uint64_t firmware_other_bytes;
 161 uint64_t firmware_MMIO_bytes;
 162
 163 /*
 164  * Linker magic to establish the highest address in the kernel.
 165  */
 166 extern void     *last_kernel_symbol;
 167
 168 #if     DEBUG
 169 #define PRINT_PMAP_MEMORY_TABLE
 170 #define DBG(x...)       kprintf(x)
 171 #else
 172 #define DBG(x...)
 173 #endif /* DEBUG */
 174 /*
 175  * Basic VM initialization.
 176  */
 177 void
 178 i386_vm_init(uint64_t   maxmem,
 179              boolean_t  IA32e,
 180              boot_args  *args)
 181 {
 182         pmap_memory_region_t *pmptr;
 183         pmap_memory_region_t *prev_pmptr;
 184         EfiMemoryRange *mptr;
 185         unsigned int mcount;
 186         unsigned int msize;
 187         ppnum_t fap;
 188         unsigned int i;
 189         ppnum_t maxpg = 0;
 190         uint32_t pmap_type;
 191         uint32_t maxloreserve;
 192         uint32_t maxdmaaddr;
 193         uint32_t  mbuf_reserve = 0;
 194         boolean_t mbuf_override = FALSE;
 195         boolean_t coalescing_permitted;
 196         vm_kernel_base_page = i386_btop(args->kaddr);
 197         vm_offset_t base_address;
 198         vm_offset_t static_base_address;
 199
 200         /*
 201          * Establish the KASLR parameters.
 202          */
 203         static_base_address = ml_static_ptovirt(KERNEL_BASE_OFFSET);
 204         base_address        = ml_static_ptovirt(args->kaddr);
 205         vm_kernel_slide     = base_address - static_base_address;
 206         if (args->kslide) {
 207                 kprintf("KASLR slide: 0x%016lx dynamic\n", vm_kernel_slide);
 208                 if (vm_kernel_slide != ((vm_offset_t)args->kslide))
 209                         panic("Kernel base inconsistent with slide - rebased?");
 210         } else {
 211                 /* No slide relative to on-disk symbols */
 212                 kprintf("KASLR slide: 0x%016lx static and ignored\n",
 213                         vm_kernel_slide);
 214                 vm_kernel_slide = 0;
 215         }
 216
 217         /*
 218          * Zero out local relocations to avoid confusing kxld.
 219          * TODO: might be better to move this code to OSKext::initialize
 220          */
 221         if (_mh_execute_header.flags & MH_PIE) {
 222                 struct load_command *loadcmd;
 223                 uint32_t cmd;
 224
 225                 loadcmd = (struct load_command *)((uintptr_t)&_mh_execute_header +
 226                                                   sizeof (_mh_execute_header));
 227
 228                 for (cmd = 0; cmd < _mh_execute_header.ncmds; cmd++) {
 229                         if (loadcmd->cmd == LC_DYSYMTAB) {
 230                                 struct dysymtab_command *dysymtab;
 231
 232                                 dysymtab = (struct dysymtab_command *)loadcmd;
 233                                 dysymtab->nlocrel = 0;
 234                                 dysymtab->locreloff = 0;
 235                                 kprintf("Hiding local relocations\n");
 236                                 break;
 237                         }
 238                         loadcmd = (struct load_command *)((uintptr_t)loadcmd + loadcmd->cmdsize);
 239                 }
 240         }
 241
 242         /*
 243          * Now retrieve addresses for end, edata, and etext
 244          * from MACH-O headers.
 245          */
 246         segTEXTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 247                                         "__TEXT", &segSizeTEXT);
 248         segDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 249                                         "__DATA", &segSizeDATA);
 250         segLINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 251                                         "__LINKEDIT", &segSizeLINK);
 252         segHIBB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 253                                         "__HIB", &segSizeHIB);
 254         segPRELINKTEXTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 255                                                 "__PRELINK_TEXT", &segSizePRELINKTEXT);
 256         segPRELINKINFOB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 257                                                 "__PRELINK_INFO", &segSizePRELINKINFO);
 258         segTEXT = getsegbynamefromheader(&_mh_execute_header,
 259                                         "__TEXT");
 260         segDATA = getsegbynamefromheader(&_mh_execute_header,
 261                                         "__DATA");
 262         segCONST = getsegbynamefromheader(&_mh_execute_header,
 263                                         "__CONST");
 264         cursectTEXT = lastsectTEXT = firstsect(segTEXT);
 265         /* Discover the last TEXT section within the TEXT segment */
 266         while ((cursectTEXT = nextsect(segTEXT, cursectTEXT)) != NULL) {
 267                 lastsectTEXT = cursectTEXT;
 268         }
 269
 270         sHIB  = segHIBB;
 271         eHIB  = segHIBB + segSizeHIB;
 272         vm_hib_base = sHIB;
 273         /* Zero-padded from ehib to stext if text is 2M-aligned */
 274         stext = segTEXTB;
 275         lowGlo.lgStext = stext;
 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         sconst = segCONST->vmaddr;
 282         segSizeConst = segCONST->vmsize;
 283         econst = sconst + segSizeConst;
 284
 285         assert(((sconst|econst) & PAGE_MASK) == 0);
 286
 287         DBG("segTEXTB    = %p\n", (void *) segTEXTB);
 288         DBG("segDATAB    = %p\n", (void *) segDATAB);
 289         DBG("segLINKB    = %p\n", (void *) segLINKB);
 290         DBG("segHIBB     = %p\n", (void *) segHIBB);
 291         DBG("segPRELINKTEXTB = %p\n", (void *) segPRELINKTEXTB);
 292         DBG("segPRELINKINFOB = %p\n", (void *) segPRELINKINFOB);
 293         DBG("sHIB        = %p\n", (void *) sHIB);
 294         DBG("eHIB        = %p\n", (void *) eHIB);
 295         DBG("stext       = %p\n", (void *) stext);
 296         DBG("etext       = %p\n", (void *) etext);
 297         DBG("sdata       = %p\n", (void *) sdata);
 298         DBG("edata       = %p\n", (void *) edata);
 299         DBG("sconst      = %p\n", (void *) sconst);
 300         DBG("econst      = %p\n", (void *) econst);
 301         DBG("kernel_top  = %p\n", (void *) &last_kernel_symbol);
 302
 303         vm_kernel_base  = sHIB;
 304         vm_kernel_top   = (vm_offset_t) &last_kernel_symbol;
 305         vm_kernel_stext = stext;
 306         vm_kernel_etext = etext;
 307         vm_prelink_stext = segPRELINKTEXTB;
 308         vm_prelink_etext = segPRELINKTEXTB + segSizePRELINKTEXT;
 309         vm_prelink_sinfo = segPRELINKINFOB;
 310         vm_prelink_einfo = segPRELINKINFOB + segSizePRELINKINFO;
 311         vm_slinkedit = segLINKB;
 312         vm_elinkedit = segLINKB + segSizePRELINKTEXT;
 313         vm_kernel_slid_base = vm_kext_base;
 314         vm_kernel_slid_top = vm_elinkedit;
 315
 316         vm_set_page_size();
 317
 318         /*
 319          * Compute the memory size.
 320          */
 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 (base == 0) {
 350                         /*
 351                          * Avoid having to deal with the edge case of the
 352                          * very first possible physical page and the roll-over
 353                          * to -1; just ignore that page.
 354                          */
 355                         kprintf("WARNING: ignoring first page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top);
 356                         base++;
 357                 }
 358                 if (top + 1 == 0) {
 359                         /*
 360                          * Avoid having to deal with the edge case of the
 361                          * very last possible physical page and the roll-over
 362                          * to 0; just ignore that page.
 363                          */
 364                         kprintf("WARNING: ignoring last page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top);
 365                         top--;
 366                 }
 367                 if (top < base) {
 368                         /*
 369                          * That was the only page in that region, so
 370                          * ignore the whole region.
 371                          */
 372                         continue;
 373                 }
 374
 375 #if     MR_RSV_TEST
 376                 static uint32_t nmr = 0;
 377                 if ((base > 0x20000) && (nmr++ < 4))
 378                         mptr->Attribute |= EFI_MEMORY_KERN_RESERVED;
 379 #endif
 380                 region_bytes = (uint64_t)(mptr->NumberOfPages << I386_PGSHIFT);
 381                 pmap_type = mptr->Type;
 382
 383                 switch (mptr->Type) {
 384                 case kEfiLoaderCode:
 385                 case kEfiLoaderData:
 386                 case kEfiBootServicesCode:
 387                 case kEfiBootServicesData:
 388                 case kEfiConventionalMemory:
 389                         /*
 390                          * Consolidate usable memory types into one.
 391                          */
 392                         pmap_type = kEfiConventionalMemory;
 393                         sane_size += region_bytes;
 394                         firmware_Conventional_bytes += region_bytes;
 395                         break;
 396                         /*
 397                          * sane_size should reflect the total amount of physical
 398                          * RAM in the system, not just the amount that is
 399                          * available for the OS to use.
 400                          * We now get this value from SMBIOS tables
 401                          * rather than reverse engineering the memory map.
 402                          * But the legacy computation of "sane_size" is kept
 403                          * for diagnostic information.
 404                          */
 405
 406                 case kEfiRuntimeServicesCode:
 407                 case kEfiRuntimeServicesData:
 408                         firmware_RuntimeServices_bytes += region_bytes;
 409                         sane_size += region_bytes;
 410                         break;
 411                 case kEfiACPIReclaimMemory:
 412                         firmware_ACPIReclaim_bytes += region_bytes;
 413                         sane_size += region_bytes;
 414                         break;
 415                 case kEfiACPIMemoryNVS:
 416                         firmware_ACPINVS_bytes += region_bytes;
 417                         sane_size += region_bytes;
 418                         break;
 419                 case kEfiPalCode:
 420                         firmware_PalCode_bytes += region_bytes;
 421                         sane_size += region_bytes;
 422                         break;
 423
 424                 case kEfiReservedMemoryType:
 425                         firmware_Reserved_bytes += region_bytes;
 426                         break;
 427                 case kEfiUnusableMemory:
 428                         firmware_Unusable_bytes += region_bytes;
 429                         break;
 430                 case kEfiMemoryMappedIO:
 431                 case kEfiMemoryMappedIOPortSpace:
 432                         firmware_MMIO_bytes += region_bytes;
 433                         break;
 434                 default:
 435                         firmware_other_bytes += region_bytes;
 436                         break;
 437                 }
 438
 439                 DBG("EFI region %d: type %u/%d, base 0x%x, top 0x%x %s\n",
 440                     i, mptr->Type, pmap_type, base, top,
 441                     (mptr->Attribute&EFI_MEMORY_KERN_RESERVED)? "RESERVED" :
 442                     (mptr->Attribute&EFI_MEMORY_RUNTIME)? "RUNTIME" : "");
 443
 444                 if (maxpg) {
 445                         if (base >= maxpg)
 446                                 break;
 447                         top = (top > maxpg) ? maxpg : top;
 448                 }
 449
 450                 /*
 451                  * handle each region
 452                  */
 453                 if ((mptr->Attribute & EFI_MEMORY_RUNTIME) == EFI_MEMORY_RUNTIME ||
 454                     pmap_type != kEfiConventionalMemory) {
 455                         prev_pmptr = 0;
 456                         continue;
 457                 } else {
 458                         /*
 459                          * Usable memory region
 460                          */
 461                         if (top < I386_LOWMEM_RESERVED ||
 462                             !pal_is_usable_memory(base, top)) {
 463                                 prev_pmptr = 0;
 464                                 continue;
 465                         }
 466                         /*
 467                          * A range may be marked with with the
 468                          * EFI_MEMORY_KERN_RESERVED attribute
 469                          * on some systems, to indicate that the range
 470                          * must not be made available to devices.
 471                          */
 472
 473                         if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED) {
 474                                 if (++pmap_reserved_ranges > PMAP_MAX_RESERVED_RANGES) {
 475                                         panic("Too many reserved ranges %u\n", pmap_reserved_ranges);
 476                                 }
 477                         }
 478
 479                         if (top < fap) {
 480                                 /*
 481                                  * entire range below first_avail
 482                                  * salvage some low memory pages
 483                                  * we use some very low memory at startup
 484                                  * mark as already allocated here
 485                                  */
 486                                 if (base >= I386_LOWMEM_RESERVED)
 487                                         pmptr->base = base;
 488                                 else
 489                                         pmptr->base = I386_LOWMEM_RESERVED;
 490
 491                                 pmptr->end = top;
 492
 493
 494                                 if ((mptr->Attribute & EFI_MEMORY_KERN_RESERVED) &&
 495                                     (top < vm_kernel_base_page)) {
 496                                         pmptr->alloc_up = pmptr->base;
 497                                         pmptr->alloc_down = pmptr->end;
 498                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 499                                 }
 500                                 else {
 501                                         /*
 502                                          * mark as already mapped
 503                                          */
 504                                         pmptr->alloc_up = top + 1;
 505                                         pmptr->alloc_down = top;
 506                                 }
 507                                 pmptr->type = pmap_type;
 508                                 pmptr->attribute = mptr->Attribute;
 509                         }
 510                         else if ( (base < fap) && (top > fap) ) {
 511                                 /*
 512                                  * spans first_avail
 513                                  * put mem below first avail in table but
 514                                  * mark already allocated
 515                                  */
 516                                 pmptr->base = base;
 517                                 pmptr->end = (fap - 1);
 518                                 pmptr->alloc_up = pmptr->end + 1;
 519                                 pmptr->alloc_down = pmptr->end;
 520                                 pmptr->type = pmap_type;
 521                                 pmptr->attribute = mptr->Attribute;
 522                                 /*
 523                                  * we bump these here inline so the accounting
 524                                  * below works correctly
 525                                  */
 526                                 pmptr++;
 527                                 pmap_memory_region_count++;
 528
 529                                 pmptr->alloc_up = pmptr->base = fap;
 530                                 pmptr->type = pmap_type;
 531                                 pmptr->attribute = mptr->Attribute;
 532                                 pmptr->alloc_down = pmptr->end = top;
 533
 534                                 if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
 535                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 536                         } else {
 537                                 /*
 538                                  * entire range useable
 539                                  */
 540                                 pmptr->alloc_up = pmptr->base = base;
 541                                 pmptr->type = pmap_type;
 542                                 pmptr->attribute = mptr->Attribute;
 543                                 pmptr->alloc_down = pmptr->end = top;
 544                                 if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
 545                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 546                         }
 547
 548                         if (i386_ptob(pmptr->end) > avail_end )
 549                                 avail_end = i386_ptob(pmptr->end);
 550
 551                         avail_remaining += (pmptr->end - pmptr->base);
 552                         coalescing_permitted = (prev_pmptr && (pmptr->attribute == prev_pmptr->attribute) && ((pmptr->attribute & EFI_MEMORY_KERN_RESERVED) == 0));
 553                         /*
 554                          * Consolidate contiguous memory regions, if possible
 555                          */
 556                         if (prev_pmptr &&
 557                             (pmptr->type == prev_pmptr->type) &&
 558                             (coalescing_permitted) &&
 559                             (pmptr->base == pmptr->alloc_up) &&
 560                             (prev_pmptr->end == prev_pmptr->alloc_down) &&
 561                             (pmptr->base == (prev_pmptr->end + 1)))
 562                         {
 563                                 prev_pmptr->end = pmptr->end;
 564                                 prev_pmptr->alloc_down = pmptr->alloc_down;
 565                         } else {
 566                                 pmap_memory_region_count++;
 567                                 prev_pmptr = pmptr;
 568                                 pmptr++;
 569                         }
 570                 }
 571         }
 572
 573 #ifdef PRINT_PMAP_MEMORY_TABLE
 574         {
 575         unsigned int j;
 576         pmap_memory_region_t *p = pmap_memory_regions;
 577         addr64_t region_start, region_end;
 578         addr64_t efi_start, efi_end;
 579         for (j=0;j<pmap_memory_region_count;j++, p++) {
 580             kprintf("pmap region %d type %d base 0x%llx alloc_up 0x%llx alloc_down 0x%llx top 0x%llx\n",
 581                     j, p->type,
 582                     (addr64_t) p->base  << I386_PGSHIFT,
 583                     (addr64_t) p->alloc_up << I386_PGSHIFT,
 584                     (addr64_t) p->alloc_down << I386_PGSHIFT,
 585                     (addr64_t) p->end   << I386_PGSHIFT);
 586             region_start = (addr64_t) p->base << I386_PGSHIFT;
 587             region_end = ((addr64_t) p->end << I386_PGSHIFT) - 1;
 588             mptr = (EfiMemoryRange *) ml_static_ptovirt((vm_offset_t)args->MemoryMap);
 589             for (i=0; i<mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
 590                 if (mptr->Type != kEfiLoaderCode &&
 591                     mptr->Type != kEfiLoaderData &&
 592                     mptr->Type != kEfiBootServicesCode &&
 593                     mptr->Type != kEfiBootServicesData &&
 594                     mptr->Type != kEfiConventionalMemory) {
 595                 efi_start = (addr64_t)mptr->PhysicalStart;
 596                 efi_end = efi_start + ((vm_offset_t)mptr->NumberOfPages << I386_PGSHIFT) - 1;
 597                 if ((efi_start >= region_start && efi_start <= region_end) ||
 598                     (efi_end >= region_start && efi_end <= region_end)) {
 599                     kprintf(" *** Overlapping region with EFI runtime region %d\n", i);
 600                 }
 601               }
 602             }
 603           }
 604         }
 605 #endif
 606
 607         avail_start = first_avail;
 608         mem_actual = args->PhysicalMemorySize;
 609
 610         /*
 611          * For user visible memory size, round up to 128 Mb
 612          * - accounting for the various stolen memory not reported by EFI.
 613          * This is maintained for historical, comparison purposes but
 614          * we now use the memory size reported by EFI/Booter.
 615          */
 616         sane_size = (sane_size + 128 * MB - 1) & ~((uint64_t)(128 * MB - 1));
 617         if (sane_size != mem_actual)
 618                 printf("mem_actual: 0x%llx\n legacy sane_size: 0x%llx\n",
 619                         mem_actual, sane_size);
 620         sane_size = mem_actual;
 621
 622         /*
 623          * We cap at KERNEL_MAXMEM bytes (currently 32GB for K32, 96GB for K64).
 624          * Unless overriden by the maxmem= boot-arg
 625          * -- which is a non-zero maxmem argument to this function.
 626          */
 627         if (maxmem == 0 && sane_size > KERNEL_MAXMEM) {
 628                 maxmem = KERNEL_MAXMEM;
 629                 printf("Physical memory %lld bytes capped at %dGB\n",
 630                         sane_size, (uint32_t) (KERNEL_MAXMEM/GB));
 631         }
 632
 633         /*
 634          * if user set maxmem, reduce memory sizes
 635          */
 636         if ( (maxmem > (uint64_t)first_avail) && (maxmem < sane_size)) {
 637                 ppnum_t discarded_pages  = (ppnum_t)((sane_size - maxmem) >> I386_PGSHIFT);
 638                 ppnum_t highest_pn = 0;
 639                 ppnum_t cur_end  = 0;
 640                 uint64_t        pages_to_use;
 641                 unsigned        cur_region = 0;
 642
 643                 sane_size = maxmem;
 644
 645                 if (avail_remaining > discarded_pages)
 646                         avail_remaining -= discarded_pages;
 647                 else
 648                         avail_remaining = 0;
 649
 650                 pages_to_use = avail_remaining;
 651
 652                 while (cur_region < pmap_memory_region_count && pages_to_use) {
 653                         for (cur_end = pmap_memory_regions[cur_region].base;
 654                              cur_end < pmap_memory_regions[cur_region].end && pages_to_use;
 655                              cur_end++) {
 656                                 if (cur_end > highest_pn)
 657                                         highest_pn = cur_end;
 658                                 pages_to_use--;
 659                         }
 660                         if (pages_to_use == 0) {
 661                                 pmap_memory_regions[cur_region].end = cur_end;
 662                                 pmap_memory_regions[cur_region].alloc_down = cur_end;
 663                         }
 664
 665                         cur_region++;
 666                 }
 667                 pmap_memory_region_count = cur_region;
 668
 669                 avail_end = i386_ptob(highest_pn + 1);
 670         }
 671
 672         /*
 673          * mem_size is only a 32 bit container... follow the PPC route
 674          * and pin it to a 2 Gbyte maximum
 675          */
 676         if (sane_size > (FOURGIG >> 1))
 677                 mem_size = (vm_size_t)(FOURGIG >> 1);
 678         else
 679                 mem_size = (vm_size_t)sane_size;
 680         max_mem = sane_size;
 681
 682         kprintf("Physical memory %llu MB\n", sane_size/MB);
 683
 684         max_valid_low_ppnum = (2 * GB) / PAGE_SIZE;
 685
 686         if (!PE_parse_boot_argn("max_valid_dma_addr", &maxdmaaddr, sizeof (maxdmaaddr))) {
 687                 max_valid_dma_address = (uint64_t)4 * (uint64_t)GB;
 688         } else {
 689                 max_valid_dma_address = ((uint64_t) maxdmaaddr) * MB;
 690
 691                 if ((max_valid_dma_address / PAGE_SIZE) < max_valid_low_ppnum)
 692                         max_valid_low_ppnum = (ppnum_t)(max_valid_dma_address / PAGE_SIZE);
 693         }
 694         if (avail_end >= max_valid_dma_address) {
 695
 696                 if (!PE_parse_boot_argn("maxloreserve", &maxloreserve, sizeof (maxloreserve))) {
 697
 698                         if (sane_size >= (ONEGIG * 15))
 699                                 maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 4;
 700                         else if (sane_size >= (ONEGIG * 7))
 701                                 maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 2;
 702                         else
 703                                 maxloreserve = MAXLORESERVE / PAGE_SIZE;
 704
 705 #if SOCKETS
 706                         mbuf_reserve = bsd_mbuf_cluster_reserve(&mbuf_override) / PAGE_SIZE;
 707 #endif
 708                 } else
 709                         maxloreserve = (maxloreserve * (1024 * 1024)) / PAGE_SIZE;
 710
 711                 if (maxloreserve) {
 712                         vm_lopage_free_limit = maxloreserve;
 713
 714                         if (mbuf_override == TRUE) {
 715                                 vm_lopage_free_limit += mbuf_reserve;
 716                                 vm_lopage_lowater = 0;
 717                         } else
 718                                 vm_lopage_lowater = vm_lopage_free_limit / 16;
 719
 720                         vm_lopage_refill = TRUE;
 721                         vm_lopage_needed = TRUE;
 722                 }
 723         }
 724
 725         /*
 726          *      Initialize kernel physical map.
 727          *      Kernel virtual address starts at VM_KERNEL_MIN_ADDRESS.
 728          */
 729         kprintf("avail_remaining = 0x%lx\n", (unsigned long)avail_remaining);
 730         pmap_bootstrap(0, IA32e);
 731 }
 732
 733
 734 unsigned int
 735 pmap_free_pages(void)
 736 {
 737         return (unsigned int)avail_remaining;
 738 }
 739
 740
 741 boolean_t pmap_next_page_reserved(ppnum_t *);
 742
 743 /*
 744  * Pick a page from a "kernel private" reserved range; works around
 745  * errata on some hardware.
 746  */
 747 boolean_t
 748 pmap_next_page_reserved(ppnum_t *pn) {
 749         if (pmap_reserved_ranges) {
 750                 uint32_t n;
 751                 pmap_memory_region_t *region;
 752                 for (n = 0; n < pmap_last_reserved_range_index; n++) {
 753                         uint32_t reserved_index = pmap_reserved_range_indices[n];
 754                         region = &pmap_memory_regions[reserved_index];
 755                         if (region->alloc_up <= region->alloc_down) {
 756                                 *pn = region->alloc_up++;
 757                                 avail_remaining--;
 758
 759                                 if (*pn > max_ppnum)
 760                                         max_ppnum = *pn;
 761
 762                                 if (lowest_lo == 0 || *pn < lowest_lo)
 763                                         lowest_lo = *pn;
 764
 765                                 pmap_reserved_pages_allocated++;
 766 #if DEBUG
 767                                 if (region->alloc_up > region->alloc_down) {
 768                                         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);
 769                                 }
 770 #endif
 771                                 return TRUE;
 772                         }
 773                 }
 774         }
 775         return FALSE;
 776 }
 777
 778
 779 boolean_t
 780 pmap_next_page_hi(
 781                   ppnum_t *pn)
 782 {
 783         pmap_memory_region_t *region;
 784         int     n;
 785
 786         if (pmap_next_page_reserved(pn))
 787                 return TRUE;
 788
 789         if (avail_remaining) {
 790                 for (n = pmap_memory_region_count - 1; n >= 0; n--) {
 791                         region = &pmap_memory_regions[n];
 792
 793                         if (region->alloc_down >= region->alloc_up) {
 794                                 *pn = region->alloc_down--;
 795                                 avail_remaining--;
 796
 797                                 if (*pn > max_ppnum)
 798                                         max_ppnum = *pn;
 799
 800                                 if (lowest_lo == 0 || *pn < lowest_lo)
 801                                         lowest_lo = *pn;
 802
 803                                 if (lowest_hi == 0 || *pn < lowest_hi)
 804                                         lowest_hi = *pn;
 805
 806                                 if (*pn > highest_hi)
 807                                         highest_hi = *pn;
 808
 809                                 return TRUE;
 810                         }
 811                 }
 812         }
 813         return FALSE;
 814 }
 815
 816
 817 boolean_t
 818 pmap_next_page(
 819                ppnum_t *pn)
 820 {
 821         if (avail_remaining) while (pmap_memory_region_current < pmap_memory_region_count) {
 822                 if (pmap_memory_regions[pmap_memory_region_current].alloc_up >
 823                     pmap_memory_regions[pmap_memory_region_current].alloc_down) {
 824                         pmap_memory_region_current++;
 825                         continue;
 826                 }
 827                 *pn = pmap_memory_regions[pmap_memory_region_current].alloc_up++;
 828                 avail_remaining--;
 829
 830                 if (*pn > max_ppnum)
 831                         max_ppnum = *pn;
 832
 833                 if (lowest_lo == 0 || *pn < lowest_lo)
 834                         lowest_lo = *pn;
 835
 836                 return TRUE;
 837         }
 838         return FALSE;
 839 }
 840
 841
 842 boolean_t
 843 pmap_valid_page(
 844         ppnum_t pn)
 845 {
 846         unsigned int i;
 847         pmap_memory_region_t *pmptr = pmap_memory_regions;
 848
 849         for (i = 0; i < pmap_memory_region_count; i++, pmptr++) {
 850                 if ( (pn >= pmptr->base) && (pn <= pmptr->end) )
 851                         return TRUE;
 852         }
 853         return FALSE;
 854 }
 855