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     vm_prelink_stext = segPRELINKB;
 314     vm_prelink_etext = segPRELINKB + segSizePRELINK;
 315     vm_prelink_sinfo = segPRELINKINFOB;
 316     vm_prelink_einfo = segPRELINKINFOB + segSizePRELINKINFO;
 317     vm_slinkedit = segLINKB;
 318     vm_elinkedit = segLINKB + segSizePRELINK;
 319
 320         vm_set_page_size();
 321
 322         /*
 323          * Compute the memory size.
 324          */
 325
 326         avail_remaining = 0;
 327         avail_end = 0;
 328         pmptr = pmap_memory_regions;
 329         prev_pmptr = 0;
 330         pmap_memory_region_count = pmap_memory_region_current = 0;
 331         fap = (ppnum_t) i386_btop(first_avail);
 332
 333         mptr = (EfiMemoryRange *)ml_static_ptovirt((vm_offset_t)args->MemoryMap);
 334         if (args->MemoryMapDescriptorSize == 0)
 335                 panic("Invalid memory map descriptor size");
 336         msize = args->MemoryMapDescriptorSize;
 337         mcount = args->MemoryMapSize / msize;
 338
 339 #define FOURGIG 0x0000000100000000ULL
 340 #define ONEGIG  0x0000000040000000ULL
 341
 342         for (i = 0; i < mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
 343                 ppnum_t base, top;
 344                 uint64_t region_bytes = 0;
 345
 346                 if (pmap_memory_region_count >= PMAP_MEMORY_REGIONS_SIZE) {
 347                         kprintf("WARNING: truncating memory region count at %d\n", pmap_memory_region_count);
 348                         break;
 349                 }
 350                 base = (ppnum_t) (mptr->PhysicalStart >> I386_PGSHIFT);
 351                 top = (ppnum_t) (((mptr->PhysicalStart) >> I386_PGSHIFT) + mptr->NumberOfPages - 1);
 352
 353                 if (base == 0) {
 354                         /*
 355                          * Avoid having to deal with the edge case of the
 356                          * very first possible physical page and the roll-over
 357                          * to -1; just ignore that page.
 358                          */
 359                         kprintf("WARNING: ignoring first page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top);
 360                         base++;
 361                 }
 362                 if (top + 1 == 0) {
 363                         /*
 364                          * Avoid having to deal with the edge case of the
 365                          * very last possible physical page and the roll-over
 366                          * to 0; just ignore that page.
 367                          */
 368                         kprintf("WARNING: ignoring last page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top);
 369                         top--;
 370                 }
 371                 if (top < base) {
 372                         /*
 373                          * That was the only page in that region, so
 374                          * ignore the whole region.
 375                          */
 376                         continue;
 377                 }
 378
 379 #if     MR_RSV_TEST
 380                 static uint32_t nmr = 0;
 381                 if ((base > 0x20000) && (nmr++ < 4))
 382                         mptr->Attribute |= EFI_MEMORY_KERN_RESERVED;
 383 #endif
 384                 region_bytes = (uint64_t)(mptr->NumberOfPages << I386_PGSHIFT);
 385                 pmap_type = mptr->Type;
 386
 387                 switch (mptr->Type) {
 388                 case kEfiLoaderCode:
 389                 case kEfiLoaderData:
 390                 case kEfiBootServicesCode:
 391                 case kEfiBootServicesData:
 392                 case kEfiConventionalMemory:
 393                         /*
 394                          * Consolidate usable memory types into one.
 395                          */
 396                         pmap_type = kEfiConventionalMemory;
 397                         sane_size += region_bytes;
 398                         firmware_Conventional_bytes += region_bytes;
 399                         break;
 400                         /*
 401                          * sane_size should reflect the total amount of physical
 402                          * RAM in the system, not just the amount that is
 403                          * available for the OS to use.
 404                          * FIXME:Consider deriving this value from SMBIOS tables
 405                          * rather than reverse engineering the memory map.
 406                          * Alternatively, see
 407                          * <rdar://problem/4642773> Memory map should
 408                          * describe all memory
 409                          * Firmware on some systems guarantees that the memory
 410                          * map is complete via the "RomReservedMemoryTracked"
 411                          * feature field--consult that where possible to
 412                          * avoid the "round up to 128M" workaround below.
 413                          */
 414
 415                 case kEfiRuntimeServicesCode:
 416                 case kEfiRuntimeServicesData:
 417                         firmware_RuntimeServices_bytes += region_bytes;
 418                         sane_size += region_bytes;
 419                         break;
 420                 case kEfiACPIReclaimMemory:
 421                         firmware_ACPIReclaim_bytes += region_bytes;
 422                         sane_size += region_bytes;
 423                         break;
 424                 case kEfiACPIMemoryNVS:
 425                         firmware_ACPINVS_bytes += region_bytes;
 426                         sane_size += region_bytes;
 427                         break;
 428                 case kEfiPalCode:
 429                         firmware_PalCode_bytes += region_bytes;
 430                         sane_size += region_bytes;
 431                         break;
 432
 433                 case kEfiReservedMemoryType:
 434                         firmware_Reserved_bytes += region_bytes;
 435                         break;
 436                 case kEfiUnusableMemory:
 437                         firmware_Unusable_bytes += region_bytes;
 438                         break;
 439                 case kEfiMemoryMappedIO:
 440                 case kEfiMemoryMappedIOPortSpace:
 441                         firmware_MMIO_bytes += region_bytes;
 442                         break;
 443                 default:
 444                         firmware_other_bytes += region_bytes;
 445                         break;
 446                 }
 447
 448                 DBG("EFI region %d: type %u/%d, base 0x%x, top 0x%x %s\n",
 449                     i, mptr->Type, pmap_type, base, top,
 450                     (mptr->Attribute&EFI_MEMORY_KERN_RESERVED)? "RESERVED" :
 451                     (mptr->Attribute&EFI_MEMORY_RUNTIME)? "RUNTIME" : "");
 452
 453                 if (maxpg) {
 454                         if (base >= maxpg)
 455                                 break;
 456                         top = (top > maxpg) ? maxpg : top;
 457                 }
 458
 459                 /*
 460                  * handle each region
 461                  */
 462                 if ((mptr->Attribute & EFI_MEMORY_RUNTIME) == EFI_MEMORY_RUNTIME ||
 463                     pmap_type != kEfiConventionalMemory) {
 464                         prev_pmptr = 0;
 465                         continue;
 466                 } else {
 467                         /*
 468                          * Usable memory region
 469                          */
 470                         if (top < I386_LOWMEM_RESERVED ||
 471                             !pal_is_usable_memory(base, top)) {
 472                                 prev_pmptr = 0;
 473                                 continue;
 474                         }
 475                         /*
 476                          * A range may be marked with with the
 477                          * EFI_MEMORY_KERN_RESERVED attribute
 478                          * on some systems, to indicate that the range
 479                          * must not be made available to devices.
 480                          */
 481
 482                         if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED) {
 483                                 if (++pmap_reserved_ranges > PMAP_MAX_RESERVED_RANGES) {
 484                                         panic("Too many reserved ranges %u\n", pmap_reserved_ranges);
 485                                 }
 486                         }
 487
 488                         if (top < fap) {
 489                                 /*
 490                                  * entire range below first_avail
 491                                  * salvage some low memory pages
 492                                  * we use some very low memory at startup
 493                                  * mark as already allocated here
 494                                  */
 495                                 if (base >= I386_LOWMEM_RESERVED)
 496                                         pmptr->base = base;
 497                                 else
 498                                         pmptr->base = I386_LOWMEM_RESERVED;
 499
 500                                 pmptr->end = top;
 501
 502
 503                                 if ((mptr->Attribute & EFI_MEMORY_KERN_RESERVED) &&
 504                                     (top < vm_kernel_base_page)) {
 505                                         pmptr->alloc_up = pmptr->base;
 506                                         pmptr->alloc_down = pmptr->end;
 507                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 508                                 }
 509                                 else {
 510                                         /*
 511                                          * mark as already mapped
 512                                          */
 513                                         pmptr->alloc_up = top + 1;
 514                                         pmptr->alloc_down = top;
 515                                 }
 516                                 pmptr->type = pmap_type;
 517                                 pmptr->attribute = mptr->Attribute;
 518                         }
 519                         else if ( (base < fap) && (top > fap) ) {
 520                                 /*
 521                                  * spans first_avail
 522                                  * put mem below first avail in table but
 523                                  * mark already allocated
 524                                  */
 525                                 pmptr->base = base;
 526                                 pmptr->end = (fap - 1);
 527                                 pmptr->alloc_up = pmptr->end + 1;
 528                                 pmptr->alloc_down = pmptr->end;
 529                                 pmptr->type = pmap_type;
 530                                 pmptr->attribute = mptr->Attribute;
 531                                 /*
 532                                  * we bump these here inline so the accounting
 533                                  * below works correctly
 534                                  */
 535                                 pmptr++;
 536                                 pmap_memory_region_count++;
 537
 538                                 pmptr->alloc_up = pmptr->base = fap;
 539                                 pmptr->type = pmap_type;
 540                                 pmptr->attribute = mptr->Attribute;
 541                                 pmptr->alloc_down = pmptr->end = top;
 542
 543                                 if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
 544                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 545                         } else {
 546                                 /*
 547                                  * entire range useable
 548                                  */
 549                                 pmptr->alloc_up = pmptr->base = base;
 550                                 pmptr->type = pmap_type;
 551                                 pmptr->attribute = mptr->Attribute;
 552                                 pmptr->alloc_down = pmptr->end = top;
 553                                 if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
 554                                         pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
 555                         }
 556
 557                         if (i386_ptob(pmptr->end) > avail_end )
 558                                 avail_end = i386_ptob(pmptr->end);
 559
 560                         avail_remaining += (pmptr->end - pmptr->base);
 561                         coalescing_permitted = (prev_pmptr && (pmptr->attribute == prev_pmptr->attribute) && ((pmptr->attribute & EFI_MEMORY_KERN_RESERVED) == 0));
 562                         /*
 563                          * Consolidate contiguous memory regions, if possible
 564                          */
 565                         if (prev_pmptr &&
 566                             (pmptr->type == prev_pmptr->type) &&
 567                             (coalescing_permitted) &&
 568                             (pmptr->base == pmptr->alloc_up) &&
 569                             (prev_pmptr->end == prev_pmptr->alloc_down) &&
 570                             (pmptr->base == (prev_pmptr->end + 1)))
 571                         {
 572                                 prev_pmptr->end = pmptr->end;
 573                                 prev_pmptr->alloc_down = pmptr->alloc_down;
 574                         } else {
 575                                 pmap_memory_region_count++;
 576                                 prev_pmptr = pmptr;
 577                                 pmptr++;
 578                         }
 579                 }
 580         }
 581
 582 #ifdef PRINT_PMAP_MEMORY_TABLE
 583         {
 584         unsigned int j;
 585         pmap_memory_region_t *p = pmap_memory_regions;
 586         addr64_t region_start, region_end;
 587         addr64_t efi_start, efi_end;
 588         for (j=0;j<pmap_memory_region_count;j++, p++) {
 589             kprintf("pmap region %d type %d base 0x%llx alloc_up 0x%llx alloc_down 0x%llx top 0x%llx\n",
 590                     j, p->type,
 591                     (addr64_t) p->base  << I386_PGSHIFT,
 592                     (addr64_t) p->alloc_up << I386_PGSHIFT,
 593                     (addr64_t) p->alloc_down << I386_PGSHIFT,
 594                     (addr64_t) p->end   << I386_PGSHIFT);
 595             region_start = (addr64_t) p->base << I386_PGSHIFT;
 596             region_end = ((addr64_t) p->end << I386_PGSHIFT) - 1;
 597             mptr = (EfiMemoryRange *) ml_static_ptovirt((vm_offset_t)args->MemoryMap);
 598             for (i=0; i<mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
 599                 if (mptr->Type != kEfiLoaderCode &&
 600                     mptr->Type != kEfiLoaderData &&
 601                     mptr->Type != kEfiBootServicesCode &&
 602                     mptr->Type != kEfiBootServicesData &&
 603                     mptr->Type != kEfiConventionalMemory) {
 604                 efi_start = (addr64_t)mptr->PhysicalStart;
 605                 efi_end = efi_start + ((vm_offset_t)mptr->NumberOfPages << I386_PGSHIFT) - 1;
 606                 if ((efi_start >= region_start && efi_start <= region_end) ||
 607                     (efi_end >= region_start && efi_end <= region_end)) {
 608                     kprintf(" *** Overlapping region with EFI runtime region %d\n", i);
 609                 }
 610               }
 611             }
 612           }
 613         }
 614 #endif
 615
 616         avail_start = first_avail;
 617         mem_actual = sane_size;
 618
 619         /*
 620          * For user visible memory size, round up to 128 Mb - accounting for the various stolen memory
 621          * not reported by EFI.
 622          */
 623
 624         sane_size = (sane_size + 128 * MB - 1) & ~((uint64_t)(128 * MB - 1));
 625
 626         /*
 627          * We cap at KERNEL_MAXMEM bytes (currently 32GB for K32, 96GB for K64).
 628          * Unless overriden by the maxmem= boot-arg
 629          * -- which is a non-zero maxmem argument to this function.
 630          */
 631         if (maxmem == 0 && sane_size > KERNEL_MAXMEM) {
 632                 maxmem = KERNEL_MAXMEM;
 633                 printf("Physical memory %lld bytes capped at %dGB\n",
 634                         sane_size, (uint32_t) (KERNEL_MAXMEM/GB));
 635         }
 636
 637         /*
 638          * if user set maxmem, reduce memory sizes
 639          */
 640         if ( (maxmem > (uint64_t)first_avail) && (maxmem < sane_size)) {
 641                 ppnum_t discarded_pages  = (ppnum_t)((sane_size - maxmem) >> I386_PGSHIFT);
 642                 ppnum_t highest_pn = 0;
 643                 ppnum_t cur_end  = 0;
 644                 uint64_t        pages_to_use;
 645                 unsigned        cur_region = 0;
 646
 647                 sane_size = maxmem;
 648
 649                 if (avail_remaining > discarded_pages)
 650                         avail_remaining -= discarded_pages;
 651                 else
 652                         avail_remaining = 0;
 653
 654                 pages_to_use = avail_remaining;
 655
 656                 while (cur_region < pmap_memory_region_count && pages_to_use) {
 657                         for (cur_end = pmap_memory_regions[cur_region].base;
 658                              cur_end < pmap_memory_regions[cur_region].end && pages_to_use;
 659                              cur_end++) {
 660                                 if (cur_end > highest_pn)
 661                                         highest_pn = cur_end;
 662                                 pages_to_use--;
 663                         }
 664                         if (pages_to_use == 0) {
 665                                 pmap_memory_regions[cur_region].end = cur_end;
 666                                 pmap_memory_regions[cur_region].alloc_down = cur_end;
 667                         }
 668
 669                         cur_region++;
 670                 }
 671                 pmap_memory_region_count = cur_region;
 672
 673                 avail_end = i386_ptob(highest_pn + 1);
 674         }
 675
 676         /*
 677          * mem_size is only a 32 bit container... follow the PPC route
 678          * and pin it to a 2 Gbyte maximum
 679          */
 680         if (sane_size > (FOURGIG >> 1))
 681                 mem_size = (vm_size_t)(FOURGIG >> 1);
 682         else
 683                 mem_size = (vm_size_t)sane_size;
 684         max_mem = sane_size;
 685
 686         kprintf("Physical memory %llu MB\n", sane_size/MB);
 687
 688         max_valid_low_ppnum = (2 * GB) / PAGE_SIZE;
 689
 690         if (!PE_parse_boot_argn("max_valid_dma_addr", &maxdmaaddr, sizeof (maxdmaaddr))) {
 691                 max_valid_dma_address = (uint64_t)4 * (uint64_t)GB;
 692         } else {
 693                 max_valid_dma_address = ((uint64_t) maxdmaaddr) * MB;
 694
 695                 if ((max_valid_dma_address / PAGE_SIZE) < max_valid_low_ppnum)
 696                         max_valid_low_ppnum = (ppnum_t)(max_valid_dma_address / PAGE_SIZE);
 697         }
 698         if (avail_end >= max_valid_dma_address) {
 699
 700                 if (!PE_parse_boot_argn("maxloreserve", &maxloreserve, sizeof (maxloreserve))) {
 701
 702                         if (sane_size >= (ONEGIG * 15))
 703                                 maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 4;
 704                         else if (sane_size >= (ONEGIG * 7))
 705                                 maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 2;
 706                         else
 707                                 maxloreserve = MAXLORESERVE / PAGE_SIZE;
 708
 709 #if SOCKETS
 710                         mbuf_reserve = bsd_mbuf_cluster_reserve(&mbuf_override) / PAGE_SIZE;
 711 #endif
 712                 } else
 713                         maxloreserve = (maxloreserve * (1024 * 1024)) / PAGE_SIZE;
 714
 715                 if (maxloreserve) {
 716                         vm_lopage_free_limit = maxloreserve;
 717
 718                         if (mbuf_override == TRUE) {
 719                                 vm_lopage_free_limit += mbuf_reserve;
 720                                 vm_lopage_lowater = 0;
 721                         } else
 722                                 vm_lopage_lowater = vm_lopage_free_limit / 16;
 723
 724                         vm_lopage_refill = TRUE;
 725                         vm_lopage_needed = TRUE;
 726                 }
 727         }
 728
 729         /*
 730          *      Initialize kernel physical map.
 731          *      Kernel virtual address starts at VM_KERNEL_MIN_ADDRESS.
 732          */
 733         kprintf("avail_remaining = 0x%lx\n", (unsigned long)avail_remaining);
 734         pmap_bootstrap(0, IA32e);
 735 }
 736
 737
 738 unsigned int
 739 pmap_free_pages(void)
 740 {
 741         return (unsigned int)avail_remaining;
 742 }
 743
 744
 745 boolean_t pmap_next_page_reserved(ppnum_t *);
 746
 747 /*
 748  * Pick a page from a "kernel private" reserved range; works around
 749  * errata on some hardware.
 750  */
 751 boolean_t
 752 pmap_next_page_reserved(ppnum_t *pn) {
 753         if (pmap_reserved_ranges) {
 754                 uint32_t n;
 755                 pmap_memory_region_t *region;
 756                 for (n = 0; n < pmap_last_reserved_range_index; n++) {
 757                         uint32_t reserved_index = pmap_reserved_range_indices[n];
 758                         region = &pmap_memory_regions[reserved_index];
 759                         if (region->alloc_up <= region->alloc_down) {
 760                                 *pn = region->alloc_up++;
 761                                 avail_remaining--;
 762
 763                                 if (*pn > max_ppnum)
 764                                         max_ppnum = *pn;
 765
 766                                 if (lowest_lo == 0 || *pn < lowest_lo)
 767                                         lowest_lo = *pn;
 768
 769                                 pmap_reserved_pages_allocated++;
 770 #if DEBUG
 771                                 if (region->alloc_up > region->alloc_down) {
 772                                         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);
 773                                 }
 774 #endif
 775                                 return TRUE;
 776                         }
 777                 }
 778         }
 779         return FALSE;
 780 }
 781
 782
 783 boolean_t
 784 pmap_next_page_hi(
 785                   ppnum_t *pn)
 786 {
 787         pmap_memory_region_t *region;
 788         int     n;
 789
 790         if (pmap_next_page_reserved(pn))
 791                 return TRUE;
 792
 793         if (avail_remaining) {
 794                 for (n = pmap_memory_region_count - 1; n >= 0; n--) {
 795                         region = &pmap_memory_regions[n];
 796
 797                         if (region->alloc_down >= region->alloc_up) {
 798                                 *pn = region->alloc_down--;
 799                                 avail_remaining--;
 800
 801                                 if (*pn > max_ppnum)
 802                                         max_ppnum = *pn;
 803
 804                                 if (lowest_lo == 0 || *pn < lowest_lo)
 805                                         lowest_lo = *pn;
 806
 807                                 if (lowest_hi == 0 || *pn < lowest_hi)
 808                                         lowest_hi = *pn;
 809
 810                                 if (*pn > highest_hi)
 811                                         highest_hi = *pn;
 812
 813                                 return TRUE;
 814                         }
 815                 }
 816         }
 817         return FALSE;
 818 }
 819
 820
 821 boolean_t
 822 pmap_next_page(
 823                ppnum_t *pn)
 824 {
 825         if (avail_remaining) while (pmap_memory_region_current < pmap_memory_region_count) {
 826                 if (pmap_memory_regions[pmap_memory_region_current].alloc_up >
 827                     pmap_memory_regions[pmap_memory_region_current].alloc_down) {
 828                         pmap_memory_region_current++;
 829                         continue;
 830                 }
 831                 *pn = pmap_memory_regions[pmap_memory_region_current].alloc_up++;
 832                 avail_remaining--;
 833
 834                 if (*pn > max_ppnum)
 835                         max_ppnum = *pn;
 836
 837                 if (lowest_lo == 0 || *pn < lowest_lo)
 838                         lowest_lo = *pn;
 839
 840                 return TRUE;
 841         }
 842         return FALSE;
 843 }
 844
 845
 846 boolean_t
 847 pmap_valid_page(
 848         ppnum_t pn)
 849 {
 850         unsigned int i;
 851         pmap_memory_region_t *pmptr = pmap_memory_regions;
 852
 853         for (i = 0; i < pmap_memory_region_count; i++, pmptr++) {
 854                 if ( (pn >= pmptr->base) && (pn <= pmptr->end) )
 855                         return TRUE;
 856         }
 857         return FALSE;
 858 }
 859