]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/i386_vm_init.c
xnu-2050.48.11.tar.gz
[apple/xnu.git] / osfmk / i386 / i386_vm_init.c
index 40086ffd1087a3b3ee3c854a43ea6957b335a041..9a9735c5c5b0a4ae8f95189286285f131da50b8e 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003-2006 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2003-2008 Apple Computer, Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
@@ -55,7 +55,6 @@
  */
 
 #include <platforms.h>
-#include <mach_kdb.h>
 
 #include <mach/i386/vm_param.h>
 
 #include <vm/pmap.h>
 #include <vm/vm_kern.h>
 #include <i386/pmap.h>
-#include <i386/ipl.h>
 #include <i386/misc_protos.h>
-#include <i386/mp_slave_boot.h>
 #include <i386/cpuid.h>
 #include <mach/thread_status.h>
 #include <pexpert/i386/efi.h>
-#include "i386_lowmem.h"
+#include <i386/i386_lowmem.h>
+#ifdef __x86_64__
+#include <x86_64/lowglobals.h>
+#else
+#include <i386/lowglobals.h>
+#endif
+#include <i386/pal_routines.h>
+
+#include <mach-o/loader.h>
+#include <libkern/kernel_mach_header.h>
+
 
 vm_size_t      mem_size = 0; 
-vm_offset_t    first_avail = 0;/* first after page tables */
+pmap_paddr_t   first_avail = 0;/* first after page tables */
 
 uint64_t       max_mem;        /* Size of physical memory (bytes), adjusted by maxmem */
 uint64_t        mem_actual;
-uint64_t       sane_size = 0;  /* Memory size to use for defaults calculations */
-
-#define MAXBOUNCEPOOL  (128 * 1024 * 1024)
-#define MAXLORESERVE   ( 32 * 1024 * 1024)
+uint64_t       sane_size = 0;  /* Memory size for defaults calculations */
 
-extern int bsd_mbuf_cluster_reserve(void);
+/*
+ * 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;
 
+#define MAXLORESERVE   (32 * 1024 * 1024)
 
-uint32_t       bounce_pool_base = 0;
-uint32_t       bounce_pool_size = 0;
+ppnum_t                max_ppnum = 0;
+ppnum_t                lowest_lo = 0;
+ppnum_t                lowest_hi = 0;
+ppnum_t                highest_hi = 0;
 
-static void    reserve_bouncepool(uint32_t);
+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;
 static pmap_paddr_t    avail_remaining;
 vm_offset_t     static_memory_end = 0;
 
-#include       <mach-o/loader.h>
-vm_offset_t    edata, etext, end;
+vm_offset_t    sHIB, eHIB, stext, etext, sdata, edata, sconstdata, econstdata, end;
 
 /*
- * _mh_execute_header is the mach_header for the currently executing
- * 32 bit kernel
+ * _mh_execute_header is the mach_header for the currently executing 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 *);
-extern struct segment_command *getsegbyname(const char *);
-extern struct section *firstsect(struct segment_command *);
-extern struct section *nextsect(struct segment_command *, struct section *);
-
-
-void
-i386_macho_zerofill(void)
-{
-       struct segment_command  *sgp;
-       struct section          *sp;
-
-       sgp = getsegbyname("__DATA");
-       if (sgp) {
-               sp = firstsect(sgp);
-               if (sp) {
-                       do {
-                               if ((sp->flags & S_ZEROFILL))
-                                       bzero((char *) sp->addr, sp->size);
-                       } while ((sp = nextsect(sgp, sp)));
-               }
-       }
-
-       return;
-}
+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 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;
 
+/*
+ * Linker magic to establish the highest address in the kernel.
+ * This is replicated from libsa which marks last_kernel_symbol
+ * but that's not visible from here in osfmk.
+ */
+__asm__(".zerofill __LAST, __last, _kernel_top, 0");
+extern void    *kernel_top;
+
+#if    DEBUG
+#define        PRINT_PMAP_MEMORY_TABLE
+#define DBG(x...)       kprintf(x)
+#else
+#define DBG(x...)
+#endif /* DEBUG */
 /*
  * Basic VM initialization.
  */
@@ -165,30 +186,129 @@ i386_vm_init(uint64_t    maxmem,
        unsigned int safeboot;
        ppnum_t maxpg = 0;
         uint32_t pmap_type;
-       uint32_t maxbouncepoolsize;
        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);
+#ifdef __x86_64__
+       vm_offset_t base_address;
+       vm_offset_t static_base_address;
+
+       /*
+        * 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;
+       }
+
+       /*
+        * Zero out local relocations to avoid confusing kxld.
+        * TODO: might be better to move this code to OSKext::initialize
+        */
+       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);
+               }
+       }
 
+#endif // __x86_64__
+        
        /*
         * 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);
+       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;
+       }
 
-       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;
+       sHIB  = segHIBB;
+       eHIB  = segHIBB + segSizeHIB;
+       /* Zero-padded from ehib to stext if text is 2M-aligned */
+       stext = segTEXTB;
+#ifdef __x86_64__
+       lowGlo.lgStext = stext;
+#endif
+       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("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 *) &kernel_top);
+
+       vm_kernel_base  = sHIB;
+       vm_kernel_top   = (vm_offset_t) &kernel_top;
+       vm_kernel_stext = stext;
+       vm_kernel_etext = etext;
 
        vm_set_page_size();
 
@@ -206,23 +326,33 @@ i386_vm_init(uint64_t     maxmem,
        pmap_memory_region_count = pmap_memory_region_current = 0;
        fap = (ppnum_t) i386_btop(first_avail);
 
-       mptr = (EfiMemoryRange *)args->MemoryMap;
+       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 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;
+               top = (ppnum_t) (((mptr->PhysicalStart) >> I386_PGSHIFT) + mptr->NumberOfPages - 1);
+
+#if    MR_RSV_TEST
+               static uint32_t nmr = 0;
+               if ((base > 0x20000) && (nmr++ < 4))
+                       mptr->Attribute |= EFI_MEMORY_KERN_RESERVED;
+#endif
+               region_bytes = (uint64_t)(mptr->NumberOfPages << I386_PGSHIFT);
+               pmap_type = mptr->Type;
 
                switch (mptr->Type) {
                case kEfiLoaderCode:
@@ -234,31 +364,61 @@ i386_vm_init(uint64_t     maxmem,
                         * Consolidate usable memory types into one.
                         */
                        pmap_type = kEfiConventionalMemory;
-                       sane_size += (uint64_t)(mptr->NumberOfPages << I386_PGSHIFT);
+                       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:
-                       /*
-                        * 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
-                        */
-                       sane_size += (uint64_t)(mptr->NumberOfPages << I386_PGSHIFT);
-                       /* fall thru */
+                       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:
-               case kEfiReservedMemoryType:
+                       firmware_MMIO_bytes += region_bytes;
+                       break;
                default:
-                       pmap_type = mptr->Type;
+                       firmware_other_bytes += region_bytes;
+                       break;
                }
 
-               kprintf("EFI region: type = %u/%d,  base = 0x%x,  top = 0x%x\n", mptr->Type, pmap_type, base, top);
+               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)
@@ -277,10 +437,24 @@ i386_vm_init(uint64_t     maxmem,
                        /*
                         * Usable memory region
                         */
-                       if (top < I386_LOWMEM_RESERVED) {
+                       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
@@ -292,11 +466,23 @@ i386_vm_init(uint64_t     maxmem,
                                        pmptr->base = base;
                                else
                                        pmptr->base = I386_LOWMEM_RESERVED;
-                               /*
-                                * mark as already mapped
-                                */
-                               pmptr->alloc = pmptr->end = top;
+
+                               pmptr->end = top;
+
+
+                               if ((mptr->Attribute & EFI_MEMORY_KERN_RESERVED) &&
+                                   (top < vm_kernel_base_page)) {
+                                       pmptr->alloc = pmptr->base;
+                                       pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count;
+                               }
+                               else {
+                                       /*
+                                        * mark as already mapped
+                                        */
+                                       pmptr->alloc = top;
+                               }
                                pmptr->type = pmap_type;
+                               pmptr->attribute = mptr->Attribute;
                        }
                        else if ( (base < fap) && (top > fap) ) {
                                /*
@@ -307,38 +493,50 @@ i386_vm_init(uint64_t     maxmem,
                                pmptr->base = base;
                                pmptr->alloc = pmptr->end = (fap - 1);
                                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 = pmptr->base = fap;
                                pmptr->type = pmap_type;
+                               pmptr->attribute = mptr->Attribute;
                                pmptr->end = top;
-                       }
-                       else {
+
+                               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 = pmptr->base = base;
                                pmptr->type = pmap_type;
+                               pmptr->attribute = mptr->Attribute;
                                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 &&
-                           pmptr->base == pmptr->alloc &&
-                           pmptr->base == (prev_pmptr->end + 1)) {
-                               prev_pmptr->end = pmptr->end;
+                           (pmptr->type == prev_pmptr->type) &&
+                           (coalescing_permitted) &&
+                           (pmptr->base == pmptr->alloc) &&
+                           (pmptr->base == (prev_pmptr->end + 1)))
+                       {
+                               if (prev_pmptr->end == prev_pmptr->alloc)
+                                       prev_pmptr->alloc = pmptr->base;
+                               prev_pmptr->end = pmptr->end;
                        } else {
                                pmap_memory_region_count++;
                                prev_pmptr = pmptr;
@@ -347,71 +545,67 @@ i386_vm_init(uint64_t     maxmem,
                }
        }
 
-
 #ifdef PRINT_PMAP_MEMORY_TABLE
        {
         unsigned int j;
         pmap_memory_region_t *p = pmap_memory_regions;
-        vm_offset_t region_start, region_end;
-        vm_offset_t efi_start, efi_end;
+        addr64_t region_start, region_end;
+        addr64_t efi_start, efi_end;
         for (j=0;j<pmap_memory_region_count;j++, p++) {
-            kprintf("type %d base 0x%x alloc 0x%x top 0x%x\n", p->type,
-                    p->base << I386_PGSHIFT, p->alloc << I386_PGSHIFT, p->end << I386_PGSHIFT);
-            region_start = p->base << I386_PGSHIFT;
-            region_end = (p->end << I386_PGSHIFT) - 1;
-            mptr = args->MemoryMap;
+            kprintf("pmap region %d type %d base 0x%llx alloc 0x%llx top 0x%llx\n",
+                   j, p->type,
+                    (addr64_t) p->base  << I386_PGSHIFT,
+                   (addr64_t) p->alloc << 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 = (vm_offset_t)mptr->PhysicalStart;
+                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;
        mem_actual = sane_size;
 
-#define MEG            (1024*1024ULL)
-#define GIG            (1024*MEG)
-
        /*
         * 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 * MEG - 1) & ~((uint64_t)(128 * MEG - 1));
+       sane_size = (sane_size + 128 * MB - 1) & ~((uint64_t)(128 * MB - 1));
 
-#if defined(__i386__)
-#define K32_MAXMEM     (32*GIG)
        /*
-        * For K32 we cap at K32_MAXMEM GB (currently 32GB).
+        * 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 > K32_MAXMEM) {
-               maxmem = K32_MAXMEM;
-               printf("Physical memory %lld bytes capped at %dGB for 32-bit kernel\n",
-                       sane_size, (uint32_t) (K32_MAXMEM/GIG));
+       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));
        }
-#endif
+
        /*
         * if user set maxmem, reduce memory sizes
         */
        if ( (maxmem > (uint64_t)first_avail) && (maxmem < sane_size)) {
-               ppnum_t discarded_pages  = (sane_size - maxmem) >> I386_PGSHIFT;
+               ppnum_t discarded_pages  = (ppnum_t)((sane_size - maxmem) >> I386_PGSHIFT);
                ppnum_t highest_pn = 0;
-               ppnum_t cur_alloc  = 0;
+               ppnum_t cur_end  = 0;
                uint64_t        pages_to_use;
                unsigned        cur_region = 0;
 
@@ -425,15 +619,15 @@ i386_vm_init(uint64_t     maxmem,
                pages_to_use = avail_remaining;
 
                while (cur_region < pmap_memory_region_count && pages_to_use) {
-                       for (cur_alloc = pmap_memory_regions[cur_region].alloc;
-                            cur_alloc < pmap_memory_regions[cur_region].end && pages_to_use;
-                            cur_alloc++) {
-                               if (cur_alloc > highest_pn)
-                                       highest_pn = cur_alloc;
+                       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_alloc;
+                               pmap_memory_regions[cur_region].end = cur_end;
 
                        cur_region++;
                }
@@ -452,41 +646,52 @@ i386_vm_init(uint64_t     maxmem,
                mem_size = (vm_size_t)sane_size;
        max_mem = sane_size;
 
-       kprintf("Physical memory %llu MB\n", sane_size/MEG);
+       kprintf("Physical memory %llu MB\n", sane_size/MB);
 
-       if (!PE_parse_boot_argn("max_valid_dma_addr", &maxdmaaddr, sizeof (maxdmaaddr)))
-               max_valid_dma_address = 1024ULL * 1024ULL * 4096ULL;
-       else
-               max_valid_dma_address = ((uint64_t) maxdmaaddr) * 1024ULL * 1024ULL;
-
-       if (!PE_parse_boot_argn("maxbouncepool", &maxbouncepoolsize, sizeof (maxbouncepoolsize)))
-               maxbouncepoolsize = MAXBOUNCEPOOL;
-       else
-               maxbouncepoolsize = maxbouncepoolsize * (1024 * 1024);
-
-       /*
-        * bsd_mbuf_cluster_reserve depends on sane_size being set
-        * in order to correctly determine the size of the mbuf pool
-        * that will be reserved
-        */
-       if (!PE_parse_boot_argn("maxloreserve", &maxloreserve, sizeof (maxloreserve)))
-               maxloreserve = MAXLORESERVE + bsd_mbuf_cluster_reserve();
-       else
-               maxloreserve = maxloreserve * (1024 * 1024);
+       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 (maxbouncepoolsize)
-                       reserve_bouncepool(maxbouncepoolsize);
 
-               if (maxloreserve)
-                       vm_lopage_poolsize = maxloreserve / PAGE_SIZE;
+               if (!PE_parse_boot_argn("maxloreserve", &maxloreserve, sizeof (maxloreserve))) {
+
+                       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;
+
+                       mbuf_reserve = bsd_mbuf_cluster_reserve(&mbuf_override) / PAGE_SIZE;
+               } 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.
         */
+       kprintf("avail_remaining = 0x%lx\n", (unsigned long)avail_remaining);
        pmap_bootstrap(0, IA32e);
 }
 
@@ -494,7 +699,83 @@ i386_vm_init(uint64_t      maxmem,
 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
+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 < region->end) {
+                               *pn = region->alloc++;
+                               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 == region->end) {
+                                       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 != region->end) {
+                               *pn = region->alloc++;
+                               avail_remaining--;
+
+                               if (*pn > max_ppnum)
+                                       max_ppnum = *pn;
+
+                                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;
 }
 
 
@@ -502,16 +783,21 @@ 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 ==
-                   pmap_memory_regions[pmap_memory_region_current].end) {
-                       pmap_memory_region_current++;
+               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;
+
+               if (lowest_lo == 0 || *pn < lowest_lo)
+                       lowest_lo = *pn;
+
                return TRUE;
        }
        return FALSE;
@@ -525,7 +811,6 @@ pmap_valid_page(
         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) )
                        return TRUE;
@@ -533,28 +818,3 @@ pmap_valid_page(
        return FALSE;
 }
 
-
-static void
-reserve_bouncepool(uint32_t bounce_pool_wanted)
-{
-       pmap_memory_region_t *pmptr  = pmap_memory_regions;
-       pmap_memory_region_t *lowest = NULL;
-        unsigned int i;
-       unsigned int pages_needed;
-
-       pages_needed = bounce_pool_wanted / PAGE_SIZE;
-
-       for (i = 0; i < pmap_memory_region_count; i++, pmptr++) {
-               if ( (pmptr->end - pmptr->alloc) >= pages_needed ) {
-                       if ( (lowest == NULL) || (pmptr->alloc < lowest->alloc) )
-                               lowest = pmptr;
-               }
-       }
-       if ( (lowest != NULL) ) {
-               bounce_pool_base = lowest->alloc * PAGE_SIZE;
-               bounce_pool_size = bounce_pool_wanted;
-
-               lowest->alloc += pages_needed;
-               avail_remaining -= pages_needed;
-       }
-}