]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/i386_vm_init.c
xnu-2782.10.72.tar.gz
[apple/xnu.git] / osfmk / i386 / i386_vm_init.c
index 1ed0d11516ac0fb2a7619824db1a6fe7e2c10304..8a1d753b588ff8fbf8d5474f1c4d5a806f680133 100644 (file)
@@ -1,31 +1,29 @@
 /*
 /*
- * Copyright (c) 2003 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2003-2012 Apple Inc. All rights reserved.
  *
  *
- * @APPLE_LICENSE_OSREFERENCE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  * 
- * This file contains Original Code and/or Modifications of Original Code 
- * as defined in and that are subject to the Apple Public Source License 
- * Version 2.0 (the 'License'). You may not use this file except in 
- * compliance with the License.  The rights granted to you under the 
- * License may not be used to create, or enable the creation or 
- * redistribution of, unlawful or unlicensed copies of an Apple operating 
- * system, or to circumvent, violate, or enable the circumvention or 
- * violation of, any terms of an Apple operating system software license 
- * agreement.
- *
- * Please obtain a copy of the License at 
- * http://www.opensource.apple.com/apsl/ and read it before using this 
- * file.
- *
- * The Original Code and all software distributed under the License are 
- * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 
- * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 
- * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 
- * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 
- * Please see the License for the specific language governing rights and 
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ * 
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
  * limitations under the License.
  * limitations under the License.
- *
- * @APPLE_LICENSE_OSREFERENCE_HEADER_END@
+ * 
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_COPYRIGHT@
  */
 /*
  * @OSF_COPYRIGHT@
@@ -56,9 +54,6 @@
  * the rights to redistribute these changes.
  */
 
  * the rights to redistribute these changes.
  */
 
-#include <platforms.h>
-#include <mach_kdb.h>
-#include <himem.h>
 
 #include <mach/i386/vm_param.h>
 
 
 #include <mach/i386/vm_param.h>
 
 #include <vm/pmap.h>
 #include <vm/vm_kern.h>
 #include <i386/pmap.h>
 #include <vm/pmap.h>
 #include <vm/vm_kern.h>
 #include <i386/pmap.h>
-#include <i386/ipl.h>
-#include <i386/pio.h>
 #include <i386/misc_protos.h>
 #include <i386/misc_protos.h>
-#include <i386/mp_slave_boot.h>
 #include <i386/cpuid.h>
 #include <i386/cpuid.h>
-#ifdef __MACHO__
 #include <mach/thread_status.h>
 #include <mach/thread_status.h>
-#endif
+#include <pexpert/i386/efi.h>
+#include <i386/i386_lowmem.h>
+#include <x86_64/lowglobals.h>
+#include <i386/pal_routines.h>
+
+#include <mach-o/loader.h>
+#include <libkern/kernel_mach_header.h>
+
 
 vm_size_t      mem_size = 0; 
 
 vm_size_t      mem_size = 0; 
-vm_offset_t    first_avail = 0;/* first after page tables */
-vm_offset_t    last_addr;
+pmap_paddr_t   first_avail = 0;/* first after page tables */
 
 
-uint64_t        max_mem;
-uint64_t        sane_size = 0; /* we are going to use the booter memory
-                                 table info to construct this */
+uint64_t       max_mem;        /* Size of physical memory (bytes), adjusted by maxmem */
+uint64_t        mem_actual;
+uint64_t       sane_size = 0;  /* Memory size for defaults calculations */
+
+/*
+ * KASLR parameters
+ */
+ppnum_t                vm_kernel_base_page;
+vm_offset_t    vm_kernel_base;
+vm_offset_t    vm_kernel_top;
+vm_offset_t    vm_kernel_stext;
+vm_offset_t    vm_kernel_etext;
+vm_offset_t    vm_kernel_slide;
+vm_offset_t vm_hib_base;
+vm_offset_t    vm_kext_base = VM_MIN_KERNEL_AND_KEXT_ADDRESS;
+vm_offset_t    vm_kext_top = VM_MIN_KERNEL_ADDRESS;
+
+vm_offset_t vm_prelink_stext;
+vm_offset_t vm_prelink_etext;
+vm_offset_t vm_prelink_sinfo;
+vm_offset_t vm_prelink_einfo;
+vm_offset_t vm_slinkedit;
+vm_offset_t vm_elinkedit;
+
+#define MAXLORESERVE   (32 * 1024 * 1024)
+
+ppnum_t                max_ppnum = 0;
+ppnum_t                lowest_lo = 0;
+ppnum_t                lowest_hi = 0;
+ppnum_t                highest_hi = 0;
+
+enum {PMAP_MAX_RESERVED_RANGES = 32};
+uint32_t pmap_reserved_pages_allocated = 0;
+uint32_t pmap_reserved_range_indices[PMAP_MAX_RESERVED_RANGES];
+uint32_t pmap_last_reserved_range_index = 0;
+uint32_t pmap_reserved_ranges = 0;
+
+extern unsigned int bsd_mbuf_cluster_reserve(boolean_t *);
 
 pmap_paddr_t     avail_start, avail_end;
 vm_offset_t    virtual_avail, virtual_end;
 
 pmap_paddr_t     avail_start, avail_end;
 vm_offset_t    virtual_avail, virtual_end;
-pmap_paddr_t   avail_remaining;
+static pmap_paddr_t    avail_remaining;
 vm_offset_t     static_memory_end = 0;
 
 vm_offset_t     static_memory_end = 0;
 
-#ifndef __MACHO__
-extern char    edata, end;
-#endif
+vm_offset_t    sHIB, eHIB, stext, etext, sdata, edata, sconstdata, econstdata, end;
 
 
-#ifdef __MACHO__
-#include       <mach-o/loader.h>
-vm_offset_t    edata, etext, end;
+/*
+ * _mh_execute_header is the mach_header for the currently executing kernel
+ */
+vm_offset_t segTEXTB; unsigned long segSizeTEXT;
+vm_offset_t segDATAB; unsigned long segSizeDATA;
+vm_offset_t segLINKB; unsigned long segSizeLINK;
+vm_offset_t segPRELINKB; unsigned long segSizePRELINK;
+vm_offset_t segPRELINKINFOB; unsigned long segSizePRELINKINFO;
+vm_offset_t segHIBB; unsigned long segSizeHIB;
+vm_offset_t sectCONSTB; unsigned long sectSizeConst;
+
+boolean_t doconstro_override = FALSE;
+
+static kernel_segment_command_t *segTEXT, *segDATA;
+static kernel_section_t *cursectTEXT, *lastsectTEXT;
+static kernel_section_t *sectDCONST;
+
+extern uint64_t firmware_Conventional_bytes;
+extern uint64_t firmware_RuntimeServices_bytes;
+extern uint64_t firmware_ACPIReclaim_bytes;
+extern uint64_t firmware_ACPINVS_bytes;
+extern uint64_t firmware_PalCode_bytes;
+extern uint64_t firmware_Reserved_bytes;
+extern uint64_t firmware_Unusable_bytes;
+extern uint64_t firmware_other_bytes;
+uint64_t firmware_MMIO_bytes;
 
 /*
 
 /*
- * _mh_execute_header is the mach_header for the currently executing
- * 32 bit kernel
+ * Linker magic to establish the highest address in the kernel.
  */
  */
-extern struct mach_header _mh_execute_header;
-void *sectTEXTB; int sectSizeTEXT;
-void *sectDATAB; int sectSizeDATA;
-void *sectOBJCB; int sectSizeOBJC;
-void *sectLINKB; int sectSizeLINK;
-void *sectPRELINKB; int sectSizePRELINK;
-void *sectHIBB; int sectSizeHIB;
-
-extern void *getsegdatafromheader(struct mach_header *, const char *, int *);
-#endif
+extern void    *last_kernel_symbol;
 
 
+#if    DEBUG
+#define        PRINT_PMAP_MEMORY_TABLE
+#define DBG(x...)       kprintf(x)
+#else
+#define DBG(x...)
+#endif /* DEBUG */
 /*
  * Basic VM initialization.
  */
 void
 /*
  * Basic VM initialization.
  */
 void
-i386_vm_init(unsigned int maxmem, KernelBootArgs_t *args)
+i386_vm_init(uint64_t  maxmem,
+            boolean_t  IA32e,
+            boot_args  *args)
 {
        pmap_memory_region_t *pmptr;
 {
        pmap_memory_region_t *pmptr;
-       MemoryRange *mptr;
+        pmap_memory_region_t *prev_pmptr;
+       EfiMemoryRange *mptr;
+        unsigned int mcount;
+        unsigned int msize;
        ppnum_t fap;
        unsigned int i;
        ppnum_t fap;
        unsigned int i;
-       ppnum_t maxpg = (maxmem >> I386_PGSHIFT);
+       ppnum_t maxpg = 0;
+        uint32_t pmap_type;
+       uint32_t maxloreserve;
+       uint32_t maxdmaaddr;
+       uint32_t  mbuf_reserve = 0;
+       boolean_t mbuf_override = FALSE;
+       boolean_t coalescing_permitted;
+       vm_kernel_base_page = i386_btop(args->kaddr);
+       vm_offset_t base_address;
+       vm_offset_t static_base_address;
 
 
-#ifdef __MACHO__
-       /* Now retrieve addresses for end, edata, and etext 
-        * from MACH-O headers.
+       /*
+        * Establish the KASLR parameters.
         */
         */
+       static_base_address = ml_static_ptovirt(KERNEL_BASE_OFFSET);
+       base_address        = ml_static_ptovirt(args->kaddr);
+       vm_kernel_slide     = base_address - static_base_address;
+       if (args->kslide) {
+               kprintf("KASLR slide: 0x%016lx dynamic\n", vm_kernel_slide);
+               if (vm_kernel_slide != ((vm_offset_t)args->kslide))
+                       panic("Kernel base inconsistent with slide - rebased?");
+       } else {
+               /* No slide relative to on-disk symbols */
+               kprintf("KASLR slide: 0x%016lx static and ignored\n",
+                       vm_kernel_slide);
+               vm_kernel_slide = 0;
+       }
 
 
-       sectTEXTB = (void *) getsegdatafromheader(
-               &_mh_execute_header, "__TEXT", &sectSizeTEXT);
-       sectDATAB = (void *) getsegdatafromheader(
-               &_mh_execute_header, "__DATA", &sectSizeDATA);
-       sectOBJCB = (void *) getsegdatafromheader(
-               &_mh_execute_header, "__OBJC", &sectSizeOBJC);
-       sectLINKB = (void *) getsegdatafromheader(
-               &_mh_execute_header, "__LINKEDIT", &sectSizeLINK);
-       sectHIBB = (void *)getsegdatafromheader(
-               &_mh_execute_header, "__HIB", &sectSizeHIB);
-       sectPRELINKB = (void *) getsegdatafromheader(
-               &_mh_execute_header, "__PRELINK", &sectSizePRELINK);
-
-       etext = (vm_offset_t) sectTEXTB + sectSizeTEXT;
-       edata = (vm_offset_t) sectDATAB + sectSizeDATA;
-#endif
-#ifndef        __MACHO__
        /*
        /*
-        * Zero the BSS.
+        * Zero out local relocations to avoid confusing kxld.
+        * TODO: might be better to move this code to OSKext::initialize
         */
         */
-
-       bzero((char *)&edata,(unsigned)(&end - &edata));
-#endif
+       if (_mh_execute_header.flags & MH_PIE) {
+               struct load_command *loadcmd;
+               uint32_t cmd;
+
+               loadcmd = (struct load_command *)((uintptr_t)&_mh_execute_header +
+                                                 sizeof (_mh_execute_header));
+
+               for (cmd = 0; cmd < _mh_execute_header.ncmds; cmd++) {
+                       if (loadcmd->cmd == LC_DYSYMTAB) {
+                               struct dysymtab_command *dysymtab;
+
+                               dysymtab = (struct dysymtab_command *)loadcmd;
+                               dysymtab->nlocrel = 0;
+                               dysymtab->locreloff = 0;
+                               kprintf("Hiding local relocations\n");
+                               break;
+                       }
+                       loadcmd = (struct load_command *)((uintptr_t)loadcmd + loadcmd->cmdsize);
+               }
+       }
 
        /*
 
        /*
-        * Initialize the pic prior to any possible call to an spl.
+        * Now retrieve addresses for end, edata, and etext 
+        * from MACH-O headers.
         */
         */
+       segTEXTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
+                                       "__TEXT", &segSizeTEXT);
+       segDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
+                                       "__DATA", &segSizeDATA);
+       segLINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
+                                       "__LINKEDIT", &segSizeLINK);
+       segHIBB  = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
+                                       "__HIB", &segSizeHIB);
+       segPRELINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
+                                       "__PRELINK_TEXT", &segSizePRELINK);
+    segPRELINKINFOB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header,
+                    "__PRELINK_INFO", &segSizePRELINKINFO);
+       segTEXT = getsegbynamefromheader(&_mh_execute_header,
+                                       "__TEXT");
+       segDATA = getsegbynamefromheader(&_mh_execute_header,
+                                       "__DATA");
+       sectDCONST = getsectbynamefromheader(&_mh_execute_header,
+                                       "__DATA", "__const");
+       cursectTEXT = lastsectTEXT = firstsect(segTEXT);
+       /* Discover the last TEXT section within the TEXT segment */
+       while ((cursectTEXT = nextsect(segTEXT, cursectTEXT)) != NULL) {
+               lastsectTEXT = cursectTEXT;
+       }
+
+       sHIB  = segHIBB;
+       eHIB  = segHIBB + segSizeHIB;
+       vm_hib_base = sHIB;
+       /* Zero-padded from ehib to stext if text is 2M-aligned */
+       stext = segTEXTB;
+       lowGlo.lgStext = stext;
+       etext = (vm_offset_t) round_page_64(lastsectTEXT->addr + lastsectTEXT->size);
+       /* Zero-padded from etext to sdata if text is 2M-aligned */
+       sdata = segDATAB;
+       edata = segDATAB + segSizeDATA;
+
+       sectCONSTB = (vm_offset_t) sectDCONST->addr;
+       sectSizeConst = sectDCONST->size;
+       sconstdata = sectCONSTB;
+       econstdata = sectCONSTB + sectSizeConst;
+
+       if (sectSizeConst & PAGE_MASK) {
+               kernel_section_t *ns = nextsect(segDATA, sectDCONST);
+               if (ns && !(ns->addr & PAGE_MASK))
+                       doconstro_override = TRUE;
+       } else
+               doconstro_override = TRUE;
+
+       DBG("segTEXTB    = %p\n", (void *) segTEXTB);
+       DBG("segDATAB    = %p\n", (void *) segDATAB);
+       DBG("segLINKB    = %p\n", (void *) segLINKB);
+       DBG("segHIBB     = %p\n", (void *) segHIBB);
+       DBG("segPRELINKB = %p\n", (void *) segPRELINKB);
+    DBG("segPRELINKINFOB = %p\n", (void *) segPRELINKINFOB);
+       DBG("sHIB        = %p\n", (void *) sHIB);
+       DBG("eHIB        = %p\n", (void *) eHIB);
+       DBG("stext       = %p\n", (void *) stext);
+       DBG("etext       = %p\n", (void *) etext);
+       DBG("sdata       = %p\n", (void *) sdata);
+       DBG("edata       = %p\n", (void *) edata);
+       DBG("sconstdata  = %p\n", (void *) sconstdata);
+       DBG("econstdata  = %p\n", (void *) econstdata);
+       DBG("kernel_top  = %p\n", (void *) &last_kernel_symbol);
+
+       vm_kernel_base  = sHIB;
+       vm_kernel_top   = (vm_offset_t) &last_kernel_symbol;
+       vm_kernel_stext = stext;
+       vm_kernel_etext = etext;
+    vm_prelink_stext = segPRELINKB;
+    vm_prelink_etext = segPRELINKB + segSizePRELINK;
+    vm_prelink_sinfo = segPRELINKINFOB;
+    vm_prelink_einfo = segPRELINKINFOB + segSizePRELINKINFO;
+    vm_slinkedit = segLINKB;
+    vm_elinkedit = segLINKB + segSizePRELINK;
 
 
-       set_cpu_model();
        vm_set_page_size();
 
        /*
        vm_set_page_size();
 
        /*
@@ -177,256 +326,534 @@ i386_vm_init(unsigned int maxmem, KernelBootArgs_t *args)
        avail_remaining = 0;
        avail_end = 0;
        pmptr = pmap_memory_regions;
        avail_remaining = 0;
        avail_end = 0;
        pmptr = pmap_memory_regions;
+        prev_pmptr = 0;
        pmap_memory_region_count = pmap_memory_region_current = 0;
        fap = (ppnum_t) i386_btop(first_avail);
        pmap_memory_region_count = pmap_memory_region_current = 0;
        fap = (ppnum_t) i386_btop(first_avail);
-       mptr = args->memoryMap;
 
 
-#ifdef PAE
+       mptr = (EfiMemoryRange *)ml_static_ptovirt((vm_offset_t)args->MemoryMap);
+        if (args->MemoryMapDescriptorSize == 0)
+               panic("Invalid memory map descriptor size");
+        msize = args->MemoryMapDescriptorSize;
+        mcount = args->MemoryMapSize / msize;
+
 #define FOURGIG 0x0000000100000000ULL
 #define FOURGIG 0x0000000100000000ULL
-       for (i=0; i < args->memoryMapCount; i++,mptr++) {
-         ppnum_t base, top;
-
-         base = (ppnum_t) (mptr->base >> I386_PGSHIFT);
-         top = (ppnum_t) ((mptr->base + mptr->length) >> I386_PGSHIFT) - 1;
-
-         if (maxmem) {
-           if (base >= maxpg) break;
-           top = (top > maxpg)? maxpg : top;
-         }
-
-         if (kMemoryRangeUsable != mptr->type) continue;
-         sane_size += (uint64_t)(mptr->length);
-#ifdef DEVICES_HANDLE_64BIT_IO  /* XXX enable else clause  when I/O to high memory works */
-         if (top < fap) {
-           /* entire range below first_avail */
-           continue;
-         } else if (mptr->base >= FOURGIG) {
-           /* entire range above 4GB (pre PAE) */
-           continue;
-         } else if ( (base < fap) &&
-                     (top > fap)) {
-           /* spans first_avail */
-           /*  put mem below first avail in table but
-               mark already allocated */
-           pmptr->base = base;
-           pmptr->alloc = pmptr->end = (fap - 1);
-            pmptr->type = mptr->type;
-           /* we bump these here inline so the accounting below works
-              correctly */
-           pmptr++;
-           pmap_memory_region_count++;
-           pmptr->alloc = pmptr->base = fap;
-            pmptr->type = mptr->type;
-           pmptr->end = top;
-         } else if ( (mptr->base < FOURGIG) &&
-                     ((mptr->base+mptr->length) > FOURGIG) ) {
-           /* spans across 4GB (pre PAE) */
-           pmptr->alloc = pmptr->base = base;
-            pmptr->type = mptr->type;
-           pmptr->end = (FOURGIG >> I386_PGSHIFT) - 1;
-         } else {
-           /* entire range useable */
-           pmptr->alloc = pmptr->base = base;
-            pmptr->type = mptr->type;
-           pmptr->end = top;
-         }
-#else
-         if (top < fap) {
-           /* entire range below first_avail */
-           continue;
-         } else if ( (base < fap) &&
-                     (top > fap)) {
-           /* spans first_avail */
-           pmptr->alloc = pmptr->base = fap;
-            pmptr->type = mptr->type;
-           pmptr->end = top;
-         } else {
-           /* entire range useable */
-           pmptr->alloc = pmptr->base = base;
-            pmptr->type = mptr->type;
-           pmptr->end = top;
-         }
+#define ONEGIG  0x0000000040000000ULL
+
+       for (i = 0; i < mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
+               ppnum_t base, top;
+               uint64_t region_bytes = 0;
+
+               if (pmap_memory_region_count >= PMAP_MEMORY_REGIONS_SIZE) {
+                       kprintf("WARNING: truncating memory region count at %d\n", pmap_memory_region_count);
+                       break;
+               }
+               base = (ppnum_t) (mptr->PhysicalStart >> I386_PGSHIFT);
+               top = (ppnum_t) (((mptr->PhysicalStart) >> I386_PGSHIFT) + mptr->NumberOfPages - 1);
+
+               if (base == 0) {
+                       /*
+                        * Avoid having to deal with the edge case of the 
+                        * very first possible physical page and the roll-over
+                        * to -1; just ignore that page.
+                        */
+                       kprintf("WARNING: ignoring first page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top);
+                       base++;
+               }
+               if (top + 1 == 0) {
+                       /*
+                        * Avoid having to deal with the edge case of the 
+                        * very last possible physical page and the roll-over
+                        * to 0; just ignore that page.
+                        */
+                       kprintf("WARNING: ignoring last page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top);
+                       top--;
+               }
+               if (top < base) {
+                       /*
+                        * That was the only page in that region, so
+                        * ignore the whole region.
+                        */
+                       continue;
+               }
+
+#if    MR_RSV_TEST
+               static uint32_t nmr = 0;
+               if ((base > 0x20000) && (nmr++ < 4))
+                       mptr->Attribute |= EFI_MEMORY_KERN_RESERVED;
 #endif
 #endif
-         if (i386_ptob(pmptr->end) > avail_end ) {
-           avail_end = i386_ptob(pmptr->end);
-         }
-         avail_remaining += (pmptr->end - pmptr->base);
-         pmap_memory_region_count++;
-         pmptr++;
+               region_bytes = (uint64_t)(mptr->NumberOfPages << I386_PGSHIFT);
+               pmap_type = mptr->Type;
+
+               switch (mptr->Type) {
+               case kEfiLoaderCode:
+               case kEfiLoaderData:
+               case kEfiBootServicesCode:
+               case kEfiBootServicesData:
+               case kEfiConventionalMemory:
+                       /*
+                        * Consolidate usable memory types into one.
+                        */
+                       pmap_type = kEfiConventionalMemory;
+                       sane_size += region_bytes;
+                       firmware_Conventional_bytes += region_bytes;
+                       break;
+                       /*
+                        * sane_size should reflect the total amount of physical
+                        * RAM in the system, not just the amount that is
+                        * available for the OS to use.
+                        * FIXME:Consider deriving this value from SMBIOS tables
+                        * rather than reverse engineering the memory map.
+                        * Alternatively, see
+                        * <rdar://problem/4642773> Memory map should
+                        * describe all memory
+                        * Firmware on some systems guarantees that the memory
+                        * map is complete via the "RomReservedMemoryTracked"
+                        * feature field--consult that where possible to
+                        * avoid the "round up to 128M" workaround below.
+                        */
+
+               case kEfiRuntimeServicesCode:
+               case kEfiRuntimeServicesData:
+                       firmware_RuntimeServices_bytes += region_bytes;
+                       sane_size += region_bytes;
+                       break;
+               case kEfiACPIReclaimMemory:
+                       firmware_ACPIReclaim_bytes += region_bytes;
+                       sane_size += region_bytes;
+                       break;
+               case kEfiACPIMemoryNVS:
+                       firmware_ACPINVS_bytes += region_bytes;
+                       sane_size += region_bytes;
+                       break;
+               case kEfiPalCode:
+                       firmware_PalCode_bytes += region_bytes;
+                       sane_size += region_bytes;
+                       break;
+
+               case kEfiReservedMemoryType:
+                       firmware_Reserved_bytes += region_bytes;
+                       break;
+               case kEfiUnusableMemory:
+                       firmware_Unusable_bytes += region_bytes;
+                       break;
+               case kEfiMemoryMappedIO:
+               case kEfiMemoryMappedIOPortSpace:
+                       firmware_MMIO_bytes += region_bytes;
+                       break;
+               default:
+                       firmware_other_bytes += region_bytes;
+                       break;
+               }
+
+               DBG("EFI region %d: type %u/%d, base 0x%x, top 0x%x %s\n",
+                   i, mptr->Type, pmap_type, base, top,
+                   (mptr->Attribute&EFI_MEMORY_KERN_RESERVED)? "RESERVED" :
+                   (mptr->Attribute&EFI_MEMORY_RUNTIME)? "RUNTIME" : "");
+
+               if (maxpg) {
+                       if (base >= maxpg)
+                               break;
+                       top = (top > maxpg) ? maxpg : top;
+               }
+
+               /*
+                * handle each region
+                */
+               if ((mptr->Attribute & EFI_MEMORY_RUNTIME) == EFI_MEMORY_RUNTIME ||
+                   pmap_type != kEfiConventionalMemory) {
+                       prev_pmptr = 0;
+                       continue;
+               } else {
+                       /*
+                        * Usable memory region
+                        */
+                       if (top < I386_LOWMEM_RESERVED ||
+                           !pal_is_usable_memory(base, top)) {
+                               prev_pmptr = 0;
+                               continue;
+                       }
+                       /*
+                        * A range may be marked with with the
+                        * EFI_MEMORY_KERN_RESERVED attribute
+                        * on some systems, to indicate that the range
+                        * must not be made available to devices.
+                        */
+
+                       if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED) {
+                               if (++pmap_reserved_ranges > PMAP_MAX_RESERVED_RANGES) {
+                                       panic("Too many reserved ranges %u\n", pmap_reserved_ranges);
+                               }
+                       }
+
+                       if (top < fap) {
+                               /*
+                                * entire range below first_avail
+                                * salvage some low memory pages
+                                * we use some very low memory at startup
+                                * mark as already allocated here
+                                */
+                               if (base >= I386_LOWMEM_RESERVED)
+                                       pmptr->base = base;
+                               else
+                                       pmptr->base = I386_LOWMEM_RESERVED;
+
+                               pmptr->end = top;
+
+
+                               if ((mptr->Attribute & EFI_MEMORY_KERN_RESERVED) &&
+                                   (top < vm_kernel_base_page)) {
+                                       pmptr->alloc_up = pmptr->base;
+                                       pmptr->alloc_down = pmptr->end;
+                                       pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
+                               }
+                               else {
+                                       /*
+                                        * mark as already mapped
+                                        */
+                                       pmptr->alloc_up = top + 1;
+                                       pmptr->alloc_down = top;
+                               }
+                               pmptr->type = pmap_type;
+                               pmptr->attribute = mptr->Attribute;
+                       }
+                       else if ( (base < fap) && (top > fap) ) {
+                               /*
+                                * spans first_avail
+                                * put mem below first avail in table but
+                                * mark already allocated
+                                */
+                               pmptr->base = base;
+                               pmptr->end = (fap - 1);
+                               pmptr->alloc_up = pmptr->end + 1;
+                               pmptr->alloc_down = pmptr->end;
+                               pmptr->type = pmap_type;
+                               pmptr->attribute = mptr->Attribute;
+                               /*
+                                * we bump these here inline so the accounting
+                                * below works correctly
+                                */
+                               pmptr++;
+                               pmap_memory_region_count++;
+
+                               pmptr->alloc_up = pmptr->base = fap;
+                               pmptr->type = pmap_type;
+                               pmptr->attribute = mptr->Attribute;
+                               pmptr->alloc_down = pmptr->end = top;
+
+                               if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
+                                       pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
+                       } else {
+                               /*
+                                * entire range useable
+                                */
+                               pmptr->alloc_up = pmptr->base = base;
+                               pmptr->type = pmap_type;
+                               pmptr->attribute = mptr->Attribute;
+                               pmptr->alloc_down = pmptr->end = top;
+                               if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED)
+                                       pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
+                       }
+
+                       if (i386_ptob(pmptr->end) > avail_end )
+                               avail_end = i386_ptob(pmptr->end);
+
+                       avail_remaining += (pmptr->end - pmptr->base);
+                       coalescing_permitted = (prev_pmptr && (pmptr->attribute == prev_pmptr->attribute) && ((pmptr->attribute & EFI_MEMORY_KERN_RESERVED) == 0));
+                       /*
+                        * Consolidate contiguous memory regions, if possible
+                        */
+                       if (prev_pmptr &&
+                           (pmptr->type == prev_pmptr->type) &&
+                           (coalescing_permitted) &&
+                           (pmptr->base == pmptr->alloc_up) &&
+                           (prev_pmptr->end == prev_pmptr->alloc_down) &&
+                           (pmptr->base == (prev_pmptr->end + 1)))
+                       {
+                               prev_pmptr->end = pmptr->end;
+                               prev_pmptr->alloc_down = pmptr->alloc_down;
+                       } else {
+                               pmap_memory_region_count++;
+                               prev_pmptr = pmptr;
+                               pmptr++;
+                       }
+               }
        }
        }
-#else  /* non PAE follows */
-#define FOURGIG 0x0000000100000000ULL
-       for (i=0; i < args->memoryMapCount; i++,mptr++) {
-         ppnum_t base, top;
-
-         base = (ppnum_t) (mptr->base >> I386_PGSHIFT);
-         top = (ppnum_t) ((mptr->base + mptr->length) >> I386_PGSHIFT) - 1;
-
-         if (maxmem) {
-           if (base >= maxpg) break;
-           top = (top > maxpg)? maxpg : top;
-         }
-
-         if (kMemoryRangeUsable != mptr->type) continue;
-
-          // save other regions
-          if (kMemoryRangeNVS == mptr->type) {
-              // Mark this as a memory range (for hibernation),
-              // but don't count as usable memory
-              pmptr->base = base;
-              pmptr->end = ((mptr->base + mptr->length + I386_PGBYTES - 1) >> I386_PGSHIFT) - 1;
-              pmptr->alloc = pmptr->end;
-              pmptr->type = mptr->type;
-              kprintf("NVS region: 0x%x ->0x%x\n", pmptr->base, pmptr->end);
-          } else if (kMemoryRangeUsable != mptr->type) {
-              continue;
-          } else {
-              // Usable memory region
-         sane_size += (uint64_t)(mptr->length);
-         if (top < fap) {
-           /* entire range below first_avail */
-           /* salvage some low memory pages */
-           /* we use some very low memory at startup */
-           /* mark as already allocated here */
-           pmptr->base = 0x18; /* PAE and HIB use below this */
-           pmptr->alloc = pmptr->end = top;  /* mark as already mapped */
-           pmptr->type = mptr->type;
-         } else if (mptr->base >= FOURGIG) {
-           /* entire range above 4GB (pre PAE) */
-           continue;
-         } else if ( (base < fap) &&
-                     (top > fap)) {
-           /* spans first_avail */
-           /*  put mem below first avail in table but
-               mark already allocated */
-           pmptr->base = base;
-           pmptr->alloc = pmptr->end = (fap - 1);
-            pmptr->type = mptr->type;
-           /* we bump these here inline so the accounting below works
-              correctly */
-           pmptr++;
-           pmap_memory_region_count++;
-           pmptr->alloc = pmptr->base = fap;
-            pmptr->type = mptr->type;
-           pmptr->end = top;
-         } else if ( (mptr->base < FOURGIG) &&
-                     ((mptr->base+mptr->length) > FOURGIG) ) {
-           /* spans across 4GB (pre PAE) */
-           pmptr->alloc = pmptr->base = base;
-            pmptr->type = mptr->type;
-           pmptr->end = (FOURGIG >> I386_PGSHIFT) - 1;
-         } else {
-           /* entire range useable */
-           pmptr->alloc = pmptr->base = base;
-            pmptr->type = mptr->type;
-           pmptr->end = top;
-         }
-
-         if (i386_ptob(pmptr->end) > avail_end ) {
-           avail_end = i386_ptob(pmptr->end);
-         }
-
-         avail_remaining += (pmptr->end - pmptr->base);
-         pmap_memory_region_count++;
-         pmptr++;
-         }
-       }
-#endif
 
 #ifdef PRINT_PMAP_MEMORY_TABLE
 
 #ifdef PRINT_PMAP_MEMORY_TABLE
- {
-   unsigned int j;
-  pmap_memory_region_t *p = pmap_memory_regions;
-   for (j=0;j<pmap_memory_region_count;j++, p++) {
-     kprintf("%d base 0x%x alloc 0x%x top 0x%x\n",j,
-            p->base, p->alloc, p->end);
-   }
- }
+       {
+        unsigned int j;
+        pmap_memory_region_t *p = pmap_memory_regions;
+        addr64_t region_start, region_end;
+        addr64_t efi_start, efi_end;
+        for (j=0;j<pmap_memory_region_count;j++, p++) {
+            kprintf("pmap region %d type %d base 0x%llx alloc_up 0x%llx alloc_down 0x%llx top 0x%llx\n",
+                   j, p->type,
+                    (addr64_t) p->base  << I386_PGSHIFT,
+                   (addr64_t) p->alloc_up << I386_PGSHIFT,
+                   (addr64_t) p->alloc_down << I386_PGSHIFT,
+                   (addr64_t) p->end   << I386_PGSHIFT);
+            region_start = (addr64_t) p->base << I386_PGSHIFT;
+            region_end = ((addr64_t) p->end << I386_PGSHIFT) - 1;
+           mptr = (EfiMemoryRange *) ml_static_ptovirt((vm_offset_t)args->MemoryMap);
+            for (i=0; i<mcount; i++, mptr = (EfiMemoryRange *)(((vm_offset_t)mptr) + msize)) {
+                if (mptr->Type != kEfiLoaderCode &&
+                    mptr->Type != kEfiLoaderData &&
+                    mptr->Type != kEfiBootServicesCode &&
+                    mptr->Type != kEfiBootServicesData &&
+                    mptr->Type != kEfiConventionalMemory) {
+                efi_start = (addr64_t)mptr->PhysicalStart;
+                efi_end = efi_start + ((vm_offset_t)mptr->NumberOfPages << I386_PGSHIFT) - 1;
+                if ((efi_start >= region_start && efi_start <= region_end) ||
+                    (efi_end >= region_start && efi_end <= region_end)) {
+                    kprintf(" *** Overlapping region with EFI runtime region %d\n", i);
+                }
+              }
+            }
+          }
+       }
 #endif
 
        avail_start = first_avail;
 #endif
 
        avail_start = first_avail;
+       mem_actual = sane_size;
 
 
-       if (maxmem) {  /* if user set maxmem try to use it */
-         uint64_t  tmp = (uint64_t)maxmem;
-         /* can't set below first_avail or above actual memory */
-         if ( (maxmem > first_avail) && (tmp < sane_size) ) {
-           sane_size = tmp;
-           avail_end = maxmem;
-         }
+       /*
+        * For user visible memory size, round up to 128 Mb - accounting for the various stolen memory
+        * not reported by EFI.
+        */
+
+       sane_size = (sane_size + 128 * MB - 1) & ~((uint64_t)(128 * MB - 1));
+
+       /*
+        * We cap at KERNEL_MAXMEM bytes (currently 32GB for K32, 96GB for K64).
+        * Unless overriden by the maxmem= boot-arg
+        * -- which is a non-zero maxmem argument to this function.
+        */
+       if (maxmem == 0 && sane_size > KERNEL_MAXMEM) {
+               maxmem = KERNEL_MAXMEM;
+               printf("Physical memory %lld bytes capped at %dGB\n",
+                       sane_size, (uint32_t) (KERNEL_MAXMEM/GB));
+       }
+
+       /*
+        * if user set maxmem, reduce memory sizes
+        */
+       if ( (maxmem > (uint64_t)first_avail) && (maxmem < sane_size)) {
+               ppnum_t discarded_pages  = (ppnum_t)((sane_size - maxmem) >> I386_PGSHIFT);
+               ppnum_t highest_pn = 0;
+               ppnum_t cur_end  = 0;
+               uint64_t        pages_to_use;
+               unsigned        cur_region = 0;
+
+               sane_size = maxmem;
+
+               if (avail_remaining > discarded_pages)
+                       avail_remaining -= discarded_pages;
+               else
+                       avail_remaining = 0;
+               
+               pages_to_use = avail_remaining;
+
+               while (cur_region < pmap_memory_region_count && pages_to_use) {
+                       for (cur_end = pmap_memory_regions[cur_region].base;
+                            cur_end < pmap_memory_regions[cur_region].end && pages_to_use;
+                            cur_end++) {
+                               if (cur_end > highest_pn)
+                                       highest_pn = cur_end;
+                               pages_to_use--;
+                       }
+                       if (pages_to_use == 0) {
+                               pmap_memory_regions[cur_region].end = cur_end;
+                               pmap_memory_regions[cur_region].alloc_down = cur_end;
+                       }
+
+                       cur_region++;
+               }
+               pmap_memory_region_count = cur_region;
+
+               avail_end = i386_ptob(highest_pn + 1);
        }
        }
-       // round up to a megabyte - mostly accounting for the
-       // low mem madness
-       sane_size += ( 0x100000ULL - 1);
-       sane_size &=  ~0xFFFFFULL;
-
-#ifndef PAE
-       if (sane_size < FOURGIG)
-         mem_size = (unsigned long) sane_size;
-       else
-         mem_size = (unsigned long) (FOURGIG >> 1);
-#else
-         mem_size = (unsigned long) sane_size;
-#endif
 
 
+       /*
+        * mem_size is only a 32 bit container... follow the PPC route
+        * and pin it to a 2 Gbyte maximum
+        */
+       if (sane_size > (FOURGIG >> 1))
+               mem_size = (vm_size_t)(FOURGIG >> 1);
+       else
+               mem_size = (vm_size_t)sane_size;
        max_mem = sane_size;
 
        max_mem = sane_size;
 
-       /* now make sane size sane */
-#define MIN(a,b)       (((a)<(b))?(a):(b))
-#define MEG            (1024*1024)
-       sane_size = MIN(sane_size, 256*MEG);
+       kprintf("Physical memory %llu MB\n", sane_size/MB);
+
+       max_valid_low_ppnum = (2 * GB) / PAGE_SIZE;
+
+       if (!PE_parse_boot_argn("max_valid_dma_addr", &maxdmaaddr, sizeof (maxdmaaddr))) {
+               max_valid_dma_address = (uint64_t)4 * (uint64_t)GB;
+       } else {
+               max_valid_dma_address = ((uint64_t) maxdmaaddr) * MB;
+
+               if ((max_valid_dma_address / PAGE_SIZE) < max_valid_low_ppnum)
+                       max_valid_low_ppnum = (ppnum_t)(max_valid_dma_address / PAGE_SIZE);
+       }
+       if (avail_end >= max_valid_dma_address) {
+
+               if (!PE_parse_boot_argn("maxloreserve", &maxloreserve, sizeof (maxloreserve))) {
 
 
-       kprintf("Physical memory %d MB\n",
-               mem_size/MEG);
+                       if (sane_size >= (ONEGIG * 15))
+                               maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 4;
+                       else if (sane_size >= (ONEGIG * 7))
+                               maxloreserve = (MAXLORESERVE / PAGE_SIZE) * 2;
+                       else
+                               maxloreserve = MAXLORESERVE / PAGE_SIZE;
 
 
+#if SOCKETS
+                       mbuf_reserve = bsd_mbuf_cluster_reserve(&mbuf_override) / PAGE_SIZE;
+#endif
+               } else
+                       maxloreserve = (maxloreserve * (1024 * 1024)) / PAGE_SIZE;
+
+               if (maxloreserve) {
+                       vm_lopage_free_limit = maxloreserve;
+                       
+                       if (mbuf_override == TRUE) {
+                               vm_lopage_free_limit += mbuf_reserve;
+                               vm_lopage_lowater = 0;
+                       } else
+                               vm_lopage_lowater = vm_lopage_free_limit / 16;
+
+                       vm_lopage_refill = TRUE;
+                       vm_lopage_needed = TRUE;
+               }
+       }
+       
        /*
         *      Initialize kernel physical map.
         *      Kernel virtual address starts at VM_KERNEL_MIN_ADDRESS.
         */
        /*
         *      Initialize kernel physical map.
         *      Kernel virtual address starts at VM_KERNEL_MIN_ADDRESS.
         */
-       pmap_bootstrap(0);
-
-
+       kprintf("avail_remaining = 0x%lx\n", (unsigned long)avail_remaining);
+       pmap_bootstrap(0, IA32e);
 }
 
 }
 
+
 unsigned int
 pmap_free_pages(void)
 {
 unsigned int
 pmap_free_pages(void)
 {
-       return avail_remaining;
+       return (unsigned int)avail_remaining;
 }
 
 }
 
+
+boolean_t pmap_next_page_reserved(ppnum_t *);
+
+/*
+ * Pick a page from a "kernel private" reserved range; works around
+ * errata on some hardware.
+ */
 boolean_t
 boolean_t
-pmap_next_page(
-              ppnum_t *pn)
+pmap_next_page_reserved(ppnum_t *pn) {
+       if (pmap_reserved_ranges) {
+               uint32_t n;
+               pmap_memory_region_t *region;
+               for (n = 0; n < pmap_last_reserved_range_index; n++) {
+                       uint32_t reserved_index = pmap_reserved_range_indices[n];
+                       region = &pmap_memory_regions[reserved_index];
+                       if (region->alloc_up <= region->alloc_down) {
+                               *pn = region->alloc_up++;
+                               avail_remaining--;
+
+                               if (*pn > max_ppnum)
+                                       max_ppnum = *pn;
+
+                               if (lowest_lo == 0 || *pn < lowest_lo)
+                                       lowest_lo = *pn;
+
+                               pmap_reserved_pages_allocated++;
+#if DEBUG
+                               if (region->alloc_up > region->alloc_down) {
+                                       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);
+                               }
+#endif
+                               return TRUE;
+                       }
+               }
+       }
+       return FALSE;
+}
+
+
+boolean_t
+pmap_next_page_hi(
+                 ppnum_t *pn)
 {
 {
+       pmap_memory_region_t *region;
+       int     n;
+
+       if (pmap_next_page_reserved(pn))
+               return TRUE;
+
+       if (avail_remaining) {
+               for (n = pmap_memory_region_count - 1; n >= 0; n--) {
+                       region = &pmap_memory_regions[n];
+
+                       if (region->alloc_down >= region->alloc_up) {
+                               *pn = region->alloc_down--;
+                               avail_remaining--;
 
 
-       while (pmap_memory_region_current < pmap_memory_region_count) {
-         if (pmap_memory_regions[pmap_memory_region_current].alloc ==
-             pmap_memory_regions[pmap_memory_region_current].end) {
-           pmap_memory_region_current++;
-           continue;
-         }
-         *pn = pmap_memory_regions[pmap_memory_region_current].alloc++;
-         avail_remaining--;
+                               if (*pn > max_ppnum)
+                                       max_ppnum = *pn;
 
 
-         return TRUE;
+                                if (lowest_lo == 0 || *pn < lowest_lo)
+                                        lowest_lo = *pn;
+
+                                if (lowest_hi == 0 || *pn < lowest_hi)
+                                        lowest_hi = *pn;
+
+                                if (*pn > highest_hi)
+                                        highest_hi = *pn;
+
+                               return TRUE;
+                       }
+               }
        }
        return FALSE;
 }
 
        }
        return FALSE;
 }
 
+
+boolean_t
+pmap_next_page(
+              ppnum_t *pn)
+{
+       if (avail_remaining) while (pmap_memory_region_current < pmap_memory_region_count) {
+               if (pmap_memory_regions[pmap_memory_region_current].alloc_up >
+                   pmap_memory_regions[pmap_memory_region_current].alloc_down) {
+                       pmap_memory_region_current++;
+                       continue;
+               }
+               *pn = pmap_memory_regions[pmap_memory_region_current].alloc_up++;
+               avail_remaining--;
+
+               if (*pn > max_ppnum)
+                       max_ppnum = *pn;
+
+               if (lowest_lo == 0 || *pn < lowest_lo)
+                       lowest_lo = *pn;
+
+               return TRUE;
+       }
+       return FALSE;
+}
+
+
 boolean_t
 pmap_valid_page(
        ppnum_t pn)
 {
 boolean_t
 pmap_valid_page(
        ppnum_t pn)
 {
-  unsigned int i;
-  pmap_memory_region_t *pmptr = pmap_memory_regions;
-
-  assert(pn);
-  for (i=0; i<pmap_memory_region_count; i++, pmptr++) {
-    if ( (pn >= pmptr->base) && (pn <= pmptr->end) ) {
-        if (pmptr->type == kMemoryRangeUsable)
-            return TRUE;
-        else
-            return FALSE;
-    }
-  }
-  return FALSE;
+        unsigned int i;
+       pmap_memory_region_t *pmptr = pmap_memory_regions;
+
+       for (i = 0; i < pmap_memory_region_count; i++, pmptr++) {
+               if ( (pn >= pmptr->base) && (pn <= pmptr->end) )
+                       return TRUE;
+       }
+       return FALSE;
 }
 }
+