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_hib_base;
 104 vm_offset_t     vm_kext_base = VM_MIN_KERNEL_AND_KEXT_ADDRESS;
 105 vm_offset_t     vm_kext_top = VM_MIN_KERNEL_ADDRESS;
 106
 107 vm_offset_t vm_prelink_stext;
 108 vm_offset_t vm_prelink_etext;
 109 vm_offset_t vm_prelink_sinfo;
 110 vm_offset_t vm_prelink_einfo;
 111 vm_offset_t vm_slinkedit;
 112 vm_offset_t vm_elinkedit;
 113
 114 #define MAXLORESERVE    (32 * 1024 * 1024)
 115
 116 ppnum_t         max_ppnum = 0;
 117 ppnum_t         lowest_lo = 0;
 118 ppnum_t         lowest_hi = 0;
 119 ppnum_t         highest_hi = 0;
 120
 121 enum {PMAP_MAX_RESERVED_RANGES = 32};
 122 uint32_t pmap_reserved_pages_allocated = 0;
 123 uint32_t pmap_reserved_range_indices[PMAP_MAX_RESERVED_RANGES];
 124 uint32_t pmap_last_reserved_range_index = 0;
 125 uint32_t pmap_reserved_ranges = 0;
 126
 127 extern unsigned int bsd_mbuf_cluster_reserve(boolean_t *);
 128
 129 pmap_paddr_t     avail_start, avail_end;
 130 vm_offset_t     virtual_avail, virtual_end;
 131 static pmap_paddr_t     avail_remaining;
 132 vm_offset_t     static_memory_end = 0;
 133
 134 vm_offset_t     sHIB, eHIB, stext, etext, sdata, edata, sconstdata, econstdata, end;
 135
 136 /*
 137  * _mh_execute_header is the mach_header for the currently executing kernel
 138  */
 139 vm_offset_t segTEXTB; unsigned long segSizeTEXT;
 140 vm_offset_t segDATAB; unsigned long segSizeDATA;
 141 vm_offset_t segLINKB; unsigned long segSizeLINK;
 142 vm_offset_t segPRELINKB; unsigned long segSizePRELINK;
 143 vm_offset_t segPRELINKINFOB; unsigned long segSizePRELINKINFO;
 144 vm_offset_t segHIBB; unsigned long segSizeHIB;
 145 vm_offset_t sectCONSTB; unsigned long sectSizeConst;
 146
 147 boolean_t doconstro_override = FALSE;
 148
 149 static kernel_segment_command_t *segTEXT, *segDATA;
 150 static kernel_section_t *cursectTEXT, *lastsectTEXT;
 151 static kernel_section_t *sectDCONST;
 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     segPRELINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
 255                                                      "__PRELINK_TEXT", &segSizePRELINK);
 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         sectDCONST = getsectbynamefromheader(&_mh_execute_header,
 263                                         "__DATA", "__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         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("segPRELINKINFOB = %p\n", (void *) segPRELINKINFOB);
 299         DBG("sHIB        = %p\n", (void *) sHIB);
 300         DBG("eHIB        = %p\n", (void *) eHIB);
 301         DBG("stext       = %p\n", (void *) stext);
 302         DBG("etext       = %p\n", (void *) etext);
 303         DBG("sdata       = %p\n", (void *) sdata);
 304         DBG("edata       = %p\n", (void *) edata);
 305         DBG("sconstdata  = %p\n", (void *) sconstdata);
 306         DBG("econstdata  = %p\n", (void *) econstdata);
 307         DBG("kernel_top  = %p\n", (void *) &last_kernel_symbol);
 308
 309         vm_kernel_base  = sHIB;
 310         vm_kernel_top   = (vm_offset_t) &last_kernel_symbol;
 311         vm_kernel_stext = stext;
 312         vm_kernel_etext = etext;
 313
 314     vm_prelink_stext = segPRELINKB;
 315     vm_prelink_etext = segPRELINKB + segSizePRELINK;
 316     vm_prelink_sinfo = segPRELINKINFOB;
 317     vm_prelink_einfo = segPRELINKINFOB + segSizePRELINKINFO;
 318     vm_slinkedit = segLINKB;
 319     vm_elinkedit = segLINKB + segSizePRELINK;
 320
 321         vm_set_page_size();
 322
 323         /*
 324          * Compute the memory size.
 325          */
 326
 327         avail_remaining = 0;
 328         avail_end = 0;
 329         pmptr = pmap_memory_regions;
 330         prev_pmptr = 0;
 331         pmap_memory_region_count = pmap_memory_region_current = 0;
 332         fap = (ppnum_t) i386_btop(first_avail);
 333
 334         mptr = (EfiMemoryRange *)ml_static_ptovirt((vm_offset_t)args->MemoryMap);
 335         if (args->MemoryMapDescriptorSize == 0)
 336                 panic("Invalid memory map descriptor size");
 337         msize = args->MemoryMapDescriptorSize;
 338         mcount = args->MemoryMapSize / msize;
 339
 340 #define FOURGIG 0x0000000100000000ULL
 341 #define ONEGIG  0x0000000040000000ULL
 342
 343         for (i = 0; i < mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
 344                 ppnum_t base, top;
 345                 uint64_t region_bytes = 0;
 346
 347                 if (pmap_memory_region_count >= PMAP_MEMORY_REGIONS_SIZE) {
 348                         kprintf("WARNING: truncating memory region count at %d\n", pmap_memory_region_count);
 349                         break;
 350                 }
 351                 base = (ppnum_t) (mptr->PhysicalStart >> I386_PGSHIFT);
 352                 top = (ppnum_t) (((mptr->PhysicalStart) >> I386_PGSHIFT) + mptr->NumberOfPages - 1);
 353
 354                 if (base == 0) {
 355                         /*
 356                          * Avoid having to deal with the edge case of the
 357                          * very first possible physical page and the roll-over
 358                          * to -1; just ignore that page.
 359                          */
 360                         kprintf("WARNING: ignoring first page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top);
 361                         base++;
 362                 }
 363                 if (top + 1 == 0) {
 364                         /*
 365                          * Avoid having to deal with the edge case of the
 366                          * very last possible physical page and the roll-over
 367                          * to 0; just ignore that page.
 368                          */
 369                         kprintf("WARNING: ignoring last page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top);
 370                         top--;
 371                 }
 372                 if (top < base) {
 373                         /*
 374                          * That was the only page in that region, so
 375                          * ignore the whole region.
 376                          */
 377                         continue;
 378                 }
 379
 380 #if     MR_RSV_TEST
 381                 static uint32_t nmr = 0;
 382                 if ((base > 0x20000) && (nmr++ < 4))
 383                         mptr->Attribute |= EFI_MEMORY_KERN_RESERVED;
 384 #endif
 385                 region_bytes = (uint64_t)(mptr->NumberOfPages << I386_PGSHIFT);
 386                 pmap_type = mptr->Type;
 387
 388                 switch (mptr->Type) {
 389                 case kEfiLoaderCode:
 390                 case kEfiLoaderData:
 391                 case kEfiBootServicesCode:
 392                 case kEfiBootServicesData:
 393                 case kEfiConventionalMemory:
 394                         /*
 395                          * Consolidate usable memory types into one.
 396                          */
 397                         pmap_type = kEfiConventionalMemory;
 398                         sane_size += region_bytes;
 399                         firmware_Conventional_bytes += region_bytes;
 400                         break;
 401                         /*
 402                          * sane_size should reflect the total amount of physical
 403                          * RAM in the system, not just the amount that is
 404                          * available for the OS to use.
 405                          * We now get this value from SMBIOS tables
 406                          * rather than reverse engineering the memory map.
 407                          * But the legacy computation of "sane_size" is kept
 408                          * for diagnostic information.
 409                          */
 410
 411                 case kEfiRuntimeServicesCode:
 412                 case kEfiRuntimeServicesData:
 413                         firmware_RuntimeServices_bytes += region_bytes;
 414                         sane_size += region_bytes;
 415                         break;
 416                 case kEfiACPIReclaimMemory:
 417                         firmware_ACPIReclaim_bytes += region_bytes;
 418                         sane_size += region_bytes;
 419                         break;
 420                 case kEfiACPIMemoryNVS:
 421                         firmware_ACPINVS_bytes += region_bytes;
 422                         sane_size += region_bytes;
 423                         break;
 424                 case kEfiPalCode:
 425                         firmware_PalCode_bytes += region_bytes;
 426                         sane_size += region_bytes;
 427                         break;
 428
 429                 case kEfiReservedMemoryType:
 430                         firmware_Reserved_bytes += region_bytes;
 431                         break;
 432                 case kEfiUnusableMemory:
 433                         firmware_Unusable_bytes += region_bytes;
 434                         break;
 435                 case kEfiMemoryMappedIO:
 436                 case kEfiMemoryMappedIOPortSpace:
 437                         firmware_MMIO_bytes += region_bytes;
 438                         break;
 439                 default:
 440                         firmware_other_bytes += region_bytes;
 441                         break;
 442                 }
 443
 444                 DBG("EFI region %d: type %u/%d, base 0x%x, top 0x%x %s\n",
 445                     i, mptr->Type, pmap_type, base, top,
 446                     (mptr->Attribute&EFI_MEMORY_KERN_RESERVED)? "RESERVED" :
 447                     (mptr->Attribute&EFI_MEMORY_RUNTIME)? "RUNTIME" : "");
 448
 449                 if (maxpg) {
 450                         if (base >= maxpg)
 451                                 break;
 452                         top = (top > maxpg) ? maxpg : top;
 453                 }
 454
 455                 /*
 456                  * handle each region
 457                  */
 458                 if ((mptr->Attribute & EFI_MEMORY_RUNTIME) == EFI_MEMORY_RUNTIME ||
 459                     pmap_type != kEfiConventionalMemory) {
 460                         prev_pmptr = 0;
 461                         continue;
 462                 } else {
 463                         /*
 464                          * Usable memory region
 465                          */
 466                         if (top < I386_LOWMEM_RESERVED ||
 467                             !pal_is_usable_memory(base, top)) {
 468                                 prev_pmptr = 0;
 469                                 continue;
 470                         }
 471                         /*
 472                          * A range may be marked with with the
 473                          * EFI_MEMORY_KERN_RESERVED attribute
 474                          * on some systems, to indicate that the range
 475                          * must not be made available to devices.
 476                          */
 477
 478                         if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED) {
 479                                 if (++pmap_reserved_ranges > PMAP_MAX_RESERVED_RANGES) {
 480                                         panic("Too many reserved ranges %u\n", pmap_reserved_ranges);
 481                                 }
 482                         }
 483
 484                         if (top < fap) {
 485                                 /*
 486                                  * entire range below first_avail
 487                                  * salvage some low memory pages
 488                                  * we use some very low memory at startup
 489                                  * mark as already allocated here
 490                                  */
 491                                 if (base >= I386_LOWMEM_RESERVED)
 492                                         pmptr->base = base;
 493                                 else
 494                                         pmptr->base = I386_LOWMEM_RESERVED;
 495
 496                                 pmptr->end = top;
 497
 498
 499                                 if ((mptr->Attribute & EFI_MEMORY_KERN_RESERVED) &&
 500                                     (top < vm_kernel_base_page)) {
 501                                         pmptr->alloc_up = pmptr->base;
 502                                         pmptr->alloc_down = pmptr->end;
 503                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 504                                 }
 505                                 else {
 506                                         /*
 507                                          * mark as already mapped
 508                                          */
 509                                         pmptr->alloc_up = top + 1;
 510                                         pmptr->alloc_down = top;
 511                                 }
 512                                 pmptr->type = pmap_type;
 513                                 pmptr->attribute = mptr->Attribute;
 514                         }
 515                         else if ( (base < fap) && (top > fap) ) {
 516                                 /*
 517                                  * spans first_avail
 518                                  * put mem below first avail in table but
 519                                  * mark already allocated
 520                                  */
 521                                 pmptr->base = base;
 522                                 pmptr->end = (fap - 1);
 523                                 pmptr->alloc_up = pmptr->end + 1;
 524                                 pmptr->alloc_down = pmptr->end;
 525                                 pmptr->type = pmap_type;
 526                                 pmptr->attribute = mptr->Attribute;
 527                                 /*
 528                                  * we bump these here inline so the accounting
 529                                  * below works correctly
 530                                  */
 531                                 pmptr++;
 532                                 pmap_memory_region_count++;
 533
 534                                 pmptr->alloc_up = pmptr->base = fap;
 535                                 pmptr->type = pmap_type;
 536                                 pmptr->attribute = mptr->Attribute;
 537                                 pmptr->alloc_down = pmptr->end = top;
 538
 539                                 if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
 540                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 541                         } else {
 542                                 /*
 543                                  * entire range useable
 544                                  */
 545                                 pmptr->alloc_up = pmptr->base = base;
 546                                 pmptr->type = pmap_type;
 547                                 pmptr->attribute = mptr->Attribute;
 548                                 pmptr->alloc_down = pmptr->end = top;
 549                                 if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
 550                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 551                         }
 552
 553                         if (i386_ptob(pmptr->end) > avail_end )
 554                                 avail_end = i386_ptob(pmptr->end);
 555
 556                         avail_remaining += (pmptr->end - pmptr->base);
 557                         coalescing_permitted = (prev_pmptr && (pmptr->attribute == prev_pmptr->attribute) && ((pmptr->attribute & EFI_MEMORY_KERN_RESERVED) == 0));
 558                         /*
 559                          * Consolidate contiguous memory regions, if possible
 560                          */
 561                         if (prev_pmptr &&
 562                             (pmptr->type == prev_pmptr->type) &&
 563                             (coalescing_permitted) &&
 564                             (pmptr->base == pmptr->alloc_up) &&
 565                             (prev_pmptr->end == prev_pmptr->alloc_down) &&
 566                             (pmptr->base == (prev_pmptr->end + 1)))
 567                         {
 568                                 prev_pmptr->end = pmptr->end;
 569                                 prev_pmptr->alloc_down = pmptr->alloc_down;
 570                         } else {
 571                                 pmap_memory_region_count++;
 572                                 prev_pmptr = pmptr;
 573                                 pmptr++;
 574                         }
 575                 }
 576         }
 577
 578 #ifdef PRINT_PMAP_MEMORY_TABLE
 579         {
 580         unsigned int j;
 581         pmap_memory_region_t *p = pmap_memory_regions;
 582         addr64_t region_start, region_end;
 583         addr64_t efi_start, efi_end;
 584         for (j=0;j<pmap_memory_region_count;j++, p++) {
 585             kprintf("pmap region %d type %d base 0x%llx alloc_up 0x%llx alloc_down 0x%llx top 0x%llx\n",
 586                     j, p->type,
 587                     (addr64_t) p->base  << I386_PGSHIFT,
 588                     (addr64_t) p->alloc_up << I386_PGSHIFT,
 589                     (addr64_t) p->alloc_down << I386_PGSHIFT,
 590                     (addr64_t) p->end   << I386_PGSHIFT);
 591             region_start = (addr64_t) p->base << I386_PGSHIFT;
 592             region_end = ((addr64_t) p->end << I386_PGSHIFT) - 1;
 593             mptr = (EfiMemoryRange *) ml_static_ptovirt((vm_offset_t)args->MemoryMap);
 594             for (i=0; i<mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
 595                 if (mptr->Type != kEfiLoaderCode &&
 596                     mptr->Type != kEfiLoaderData &&
 597                     mptr->Type != kEfiBootServicesCode &&
 598                     mptr->Type != kEfiBootServicesData &&
 599                     mptr->Type != kEfiConventionalMemory) {
 600                 efi_start = (addr64_t)mptr->PhysicalStart;
 601                 efi_end = efi_start + ((vm_offset_t)mptr->NumberOfPages << I386_PGSHIFT) - 1;
 602                 if ((efi_start >= region_start && efi_start <= region_end) ||
 603                     (efi_end >= region_start && efi_end <= region_end)) {
 604                     kprintf(" *** Overlapping region with EFI runtime region %d\n", i);
 605                 }
 606               }
 607             }
 608           }
 609         }
 610 #endif
 611
 612         avail_start = first_avail;
 613         mem_actual = args->PhysicalMemorySize;
 614
 615         /*
 616          * For user visible memory size, round up to 128 Mb
 617          * - accounting for the various stolen memory not reported by EFI.
 618          * This is maintained for historical, comparison purposes but
 619          * we now use the memory size reported by EFI/Booter.
 620          */
 621         sane_size = (sane_size + 128 * MB - 1) & ~((uint64_t)(128 * MB - 1));
 622         if (sane_size != mem_actual)
 623                 printf("mem_actual: 0x%llx\n legacy sane_size: 0x%llx\n",
 624                         mem_actual, sane_size);
 625         sane_size = mem_actual;
 626
 627         /*
 628          * We cap at KERNEL_MAXMEM bytes (currently 32GB for K32, 96GB for K64).
 629          * Unless overriden by the maxmem= boot-arg
 630          * -- which is a non-zero maxmem argument to this function.
 631          */
 632         if (maxmem == 0 && sane_size > KERNEL_MAXMEM) {
 633                 maxmem = KERNEL_MAXMEM;
 634                 printf("Physical memory %lld bytes capped at %dGB\n",
 635                         sane_size, (uint32_t) (KERNEL_MAXMEM/GB));
 636         }
 637
 638         /*
 639          * if user set maxmem, reduce memory sizes
 640          */
 641         if ( (maxmem > (uint64_t)first_avail) && (maxmem < sane_size)) {
 642                 ppnum_t discarded_pages  = (ppnum_t)((sane_size - maxmem) >> I386_PGSHIFT);
 643                 ppnum_t highest_pn = 0;
 644                 ppnum_t cur_end  = 0;
 645                 uint64_t        pages_to_use;
 646                 unsigned        cur_region = 0;
 647
 648                 sane_size = maxmem;
 649
 650                 if (avail_remaining > discarded_pages)
 651                         avail_remaining -= discarded_pages;
 652                 else
 653                         avail_remaining = 0;
 654
 655                 pages_to_use = avail_remaining;
 656
 657                 while (cur_region < pmap_memory_region_count && pages_to_use) {
 658                         for (cur_end = pmap_memory_regions[cur_region].base;
 659                              cur_end < pmap_memory_regions[cur_region].end && pages_to_use;
 660                              cur_end++) {
 661                                 if (cur_end > highest_pn)
 662                                         highest_pn = cur_end;
 663                                 pages_to_use--;
 664                         }
 665                         if (pages_to_use == 0) {
 666                                 pmap_memory_regions[cur_region].end = cur_end;
 667                                 pmap_memory_regions[cur_region].alloc_down = cur_end;
 668                         }
 669
 670                         cur_region++;
 671                 }
 672                 pmap_memory_region_count = cur_region;
 673
 674                 avail_end = i386_ptob(highest_pn + 1);
 675         }
 676
 677         /*
 678          * mem_size is only a 32 bit container... follow the PPC route
 679          * and pin it to a 2 Gbyte maximum
 680          */
 681         if (sane_size > (FOURGIG >> 1))
 682                 mem_size = (vm_size_t)(FOURGIG >> 1);
 683         else
 684                 mem_size = (vm_size_t)sane_size;
 685         max_mem = sane_size;
 686
 687         kprintf("Physical memory %llu MB\n", sane_size/MB);
 688
 689         max_valid_low_ppnum = (2 * GB) / PAGE_SIZE;
 690
 691         if (!PE_parse_boot_argn("max_valid_dma_addr", &maxdmaaddr, sizeof (maxdmaaddr))) {
 692                 max_valid_dma_address = (uint64_t)4 * (uint64_t)GB;
 693         } else {
 694                 max_valid_dma_address = ((uint64_t) maxdmaaddr) * MB;
 695
 696                 if ((max_valid_dma_address / PAGE_SIZE) < max_valid_low_ppnum)
 697                         max_valid_low_ppnum = (ppnum_t)(max_valid_dma_address / PAGE_SIZE);
 698         }
 699         if (avail_end >= max_valid_dma_address) {
 700
 701                 if (!PE_parse_boot_argn("maxloreserve", &maxloreserve, sizeof (maxloreserve))) {
 702
 703                         if (sane_size >= (ONEGIG * 15))
 704                                 maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 4;
 705                         else if (sane_size >= (ONEGIG * 7))
 706                                 maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 2;
 707                         else
 708                                 maxloreserve = MAXLORESERVE / PAGE_SIZE;
 709
 710 #if SOCKETS
 711                         mbuf_reserve = bsd_mbuf_cluster_reserve(&mbuf_override) / PAGE_SIZE;
 712 #endif
 713                 } else
 714                         maxloreserve = (maxloreserve * (1024 * 1024)) / PAGE_SIZE;
 715
 716                 if (maxloreserve) {
 717                         vm_lopage_free_limit = maxloreserve;
 718
 719                         if (mbuf_override == TRUE) {
 720                                 vm_lopage_free_limit += mbuf_reserve;
 721                                 vm_lopage_lowater = 0;
 722                         } else
 723                                 vm_lopage_lowater = vm_lopage_free_limit / 16;
 724
 725                         vm_lopage_refill = TRUE;
 726                         vm_lopage_needed = TRUE;
 727                 }
 728         }
 729
 730         /*
 731          *      Initialize kernel physical map.
 732          *      Kernel virtual address starts at VM_KERNEL_MIN_ADDRESS.
 733          */
 734         kprintf("avail_remaining = 0x%lx\n", (unsigned long)avail_remaining);
 735         pmap_bootstrap(0, IA32e);
 736 }
 737
 738
 739 unsigned int
 740 pmap_free_pages(void)
 741 {
 742         return (unsigned int)avail_remaining;
 743 }
 744
 745
 746 boolean_t pmap_next_page_reserved(ppnum_t *);
 747
 748 /*
 749  * Pick a page from a "kernel private" reserved range; works around
 750  * errata on some hardware.
 751  */
 752 boolean_t
 753 pmap_next_page_reserved(ppnum_t *pn) {
 754         if (pmap_reserved_ranges) {
 755                 uint32_t n;
 756                 pmap_memory_region_t *region;
 757                 for (n = 0; n < pmap_last_reserved_range_index; n++) {
 758                         uint32_t reserved_index = pmap_reserved_range_indices[n];
 759                         region = &pmap_memory_regions[reserved_index];
 760                         if (region->alloc_up <= region->alloc_down) {
 761                                 *pn = region->alloc_up++;
 762                                 avail_remaining--;
 763
 764                                 if (*pn > max_ppnum)
 765                                         max_ppnum = *pn;
 766
 767                                 if (lowest_lo == 0 || *pn < lowest_lo)
 768                                         lowest_lo = *pn;
 769
 770                                 pmap_reserved_pages_allocated++;
 771 #if DEBUG
 772                                 if (region->alloc_up > region->alloc_down) {
 773                                         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);
 774                                 }
 775 #endif
 776                                 return TRUE;
 777                         }
 778                 }
 779         }
 780         return FALSE;
 781 }
 782
 783
 784 boolean_t
 785 pmap_next_page_hi(
 786                   ppnum_t *pn)
 787 {
 788         pmap_memory_region_t *region;
 789         int     n;
 790
 791         if (pmap_next_page_reserved(pn))
 792                 return TRUE;
 793
 794         if (avail_remaining) {
 795                 for (n = pmap_memory_region_count - 1; n >= 0; n--) {
 796                         region = &pmap_memory_regions[n];
 797
 798                         if (region->alloc_down >= region->alloc_up) {
 799                                 *pn = region->alloc_down--;
 800                                 avail_remaining--;
 801
 802                                 if (*pn > max_ppnum)
 803                                         max_ppnum = *pn;
 804
 805                                 if (lowest_lo == 0 || *pn < lowest_lo)
 806                                         lowest_lo = *pn;
 807
 808                                 if (lowest_hi == 0 || *pn < lowest_hi)
 809                                         lowest_hi = *pn;
 810
 811                                 if (*pn > highest_hi)
 812                                         highest_hi = *pn;
 813
 814                                 return TRUE;
 815                         }
 816                 }
 817         }
 818         return FALSE;
 819 }
 820
 821
 822 boolean_t
 823 pmap_next_page(
 824                ppnum_t *pn)
 825 {
 826         if (avail_remaining) while (pmap_memory_region_current < pmap_memory_region_count) {
 827                 if (pmap_memory_regions[pmap_memory_region_current].alloc_up >
 828                     pmap_memory_regions[pmap_memory_region_current].alloc_down) {
 829                         pmap_memory_region_current++;
 830                         continue;
 831                 }
 832                 *pn = pmap_memory_regions[pmap_memory_region_current].alloc_up++;
 833                 avail_remaining--;
 834
 835                 if (*pn > max_ppnum)
 836                         max_ppnum = *pn;
 837
 838                 if (lowest_lo == 0 || *pn < lowest_lo)
 839                         lowest_lo = *pn;
 840
 841                 return TRUE;
 842         }
 843         return FALSE;
 844 }
 845
 846
 847 boolean_t
 848 pmap_valid_page(
 849         ppnum_t pn)
 850 {
 851         unsigned int i;
 852         pmap_memory_region_t *pmptr = pmap_memory_regions;
 853
 854         for (i = 0; i < pmap_memory_region_count; i++, pmptr++) {
 855                 if ( (pn >= pmptr->base) && (pn <= pmptr->end) )
 856                         return TRUE;
 857         }
 858         return FALSE;
 859 }
 860