]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/ppc/pmap.c
xnu-792.6.22.tar.gz
[apple/xnu.git] / osfmk / ppc / pmap.c
index 5b7bc5b8736369934e5c64c63b6253fd4740cf61..875ee6912b1da84204222abbafdc76679b5ad17e 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved.
  *
  * @APPLE_LICENSE_HEADER_START@
  * 
  */
 
 #include <zone_debug.h>
-#include <cpus.h>
 #include <debug.h>
 #include <mach_kgdb.h>
 #include <mach_vm_debug.h>
 #include <db_machine_commands.h>
 
 #include <kern/thread.h>
+#include <kern/simple_lock.h>
 #include <mach/vm_attributes.h>
 #include <mach/vm_param.h>
+#include <vm/vm_kern.h>
 #include <kern/spl.h>
 
 #include <kern/misc_protos.h>
 #include <vm/vm_page.h>
 
 #include <ppc/pmap.h>
-#include <ppc/pmap_internals.h>
 #include <ppc/mem.h>
 #include <ppc/mappings.h>
 
 #include <ppc/new_screen.h>
 #include <ppc/Firmware.h>
 #include <ppc/savearea.h>
+#include <ppc/cpu_internal.h>
+#include <ppc/exception.h>
+#include <ppc/low_trace.h>
+#include <ppc/lowglobals.h>
 #include <ddb/db_output.h>
+#include <machine/cpu_capabilities.h>
 
-#if    DB_MACHINE_COMMANDS
-/* optionally enable traces of pmap operations in post-mortem trace table */
-/* #define PMAP_LOWTRACE 1 */
-#define PMAP_LOWTRACE 0
-#else  /* DB_MACHINE_COMMANDS */
-/* Can not trace even if we wanted to */
-#define PMAP_LOWTRACE 0
-#endif /* DB_MACHINE_COMMANDS */
+#include <vm/vm_protos.h> /* must be last */
 
-#define PERFTIMES 0
-
-#if PERFTIMES && DEBUG
-#define debugLog2(a, b, c) dbgLog2(a, b, c)
-#else
-#define debugLog2(a, b, c)
-#endif
 
 extern unsigned int    avail_remaining;
-extern unsigned int    mappingdeb0;
-extern struct  Saveanchor saveanchor;                                          /* Aliged savearea anchor */
-extern int             real_ncpus;                                                                     /* Number of actual CPUs */
-unsigned int debugbackpocket;                                                          /* (TEST/DEBUG) */
+unsigned int   debugbackpocket;                                                        /* (TEST/DEBUG) */
 
-vm_offset_t            avail_next;
 vm_offset_t            first_free_virt;
 int            current_free_region;                                            /* Used in pmap_next_page */
 
+pmapTransTab *pmapTrans;                                                                       /* Point to the hash to pmap translations */
+struct phys_entry *phys_table;
+
 /* forward */
+static void pmap_map_physical(void);
+static void pmap_map_iohole(addr64_t paddr, addr64_t size);
 void pmap_activate(pmap_t pmap, thread_t th, int which_cpu);
 void pmap_deactivate(pmap_t pmap, thread_t th, int which_cpu);
-void copy_to_phys(vm_offset_t sva, vm_offset_t dpa, int bytecount);
 
-#if MACH_VM_DEBUG
-int pmap_list_resident_pages(pmap_t pmap, vm_offset_t *listp, int space);
-#endif
-
-#if DEBUG
-#define PDB_USER       0x01    /* exported functions */
-#define PDB_MAPPING    0x02    /* low-level mapping routines */
-#define PDB_ENTER      0x04    /* pmap_enter specifics */
-#define PDB_COPY       0x08    /* copy page debugging */
-#define PDB_ZERO       0x10    /* zero page debugging */
-#define PDB_WIRED      0x20    /* things concerning wired entries */
-#define PDB_PTEG       0x40    /* PTEG overflows */
-#define PDB_LOCK       0x100   /* locks */
-#define PDB_IO         0x200   /* Improper use of WIMG_IO checks - PCI machines */
-
-int pmdebug=0;
-#endif
+extern void hw_hash_init(void);
 
 /*  NOTE:  kernel_pmap_store must be in V=R storage and aligned!!!!!!!!!!!!!! */
 
 extern struct pmap     kernel_pmap_store;
 pmap_t         kernel_pmap;                    /* Pointer to kernel pmap and anchor for in-use pmaps */                
+addr64_t       kernel_pmap_phys;               /* Pointer to kernel pmap and anchor for in-use pmaps, physical address */              
 pmap_t         cursor_pmap;                    /* Pointer to last pmap allocated or previous if removed from in-use list */
+pmap_t         sharedPmap;                             /* Pointer to common pmap for 64-bit address spaces */
 struct zone    *pmap_zone;                             /* zone of pmap structures */
 boolean_t      pmap_initialized = FALSE;
 
+int ppc_max_pmaps;                                     /* Maximum number of concurrent address spaces allowed. This is machine dependent */    
+addr64_t vm_max_address;                       /* Maximum effective address supported */
+addr64_t vm_max_physical;                      /* Maximum physical address supported */
+
 /*
  * Physical-to-virtual translations are handled by inverted page table
  * structures, phys_tables.  Multiple mappings of a single page are handled
@@ -180,11 +161,6 @@ boolean_t  pmap_initialized = FALSE;
  * for phys_tables of the physical memory we know about, but more may be
  * added as it is discovered (eg. by drivers).
  */
-struct phys_entry *phys_table;         /* For debugging */
-
-lock_t pmap_system_lock;
-
-decl_simple_lock_data(,tlb_system_lock)
 
 /*
  *     free pmap list. caches the first free_pmap_max pmaps that are freed up
@@ -198,121 +174,38 @@ decl_simple_lock_data(,free_pmap_lock)
  * Function to get index into phys_table for a given physical address
  */
 
-struct phys_entry *pmap_find_physentry(vm_offset_t pa)
+struct phys_entry *pmap_find_physentry(ppnum_t pa)
 {
        int i;
-       struct phys_entry *entry;
+       unsigned int entry;
 
-       for (i = pmap_mem_regions_count-1; i >= 0; i--) {
-               if (pa < pmap_mem_regions[i].start)
-                       continue;
-               if (pa >= pmap_mem_regions[i].end)
-                       return PHYS_NULL;
+       for (i = pmap_mem_regions_count - 1; i >= 0; i--) {
+               if (pa < pmap_mem_regions[i].mrStart) continue; /* See if we fit in this region */
+               if (pa > pmap_mem_regions[i].mrEnd) continue;   /* Check the end too */
                
-               entry = &pmap_mem_regions[i].phys_table[(pa - pmap_mem_regions[i].start) >> PPC_PGSHIFT];
-               __asm__ volatile("dcbt 0,%0" : : "r" (entry));  /* We will use this in a little bit */
-               return entry;
+               entry = (unsigned int)pmap_mem_regions[i].mrPhysTab + ((pa - pmap_mem_regions[i].mrStart) * sizeof(phys_entry_t));
+               return (struct phys_entry *)entry;
        }
-       kprintf("DEBUG : pmap_find_physentry 0x%08x out of range\n",pa);
-       return PHYS_NULL;
+//     kprintf("DEBUG - pmap_find_physentry: page 0x%08X not found\n", pa);
+       return 0;
 }
 
 /*
  * kern_return_t
  * pmap_add_physical_memory(vm_offset_t spa, vm_offset_t epa,
  *                          boolean_t available, unsigned int attr)
- *     Allocate some extra physentries for the physical addresses given,
- *     specifying some default attribute that on the powerpc specifies
- *      the default cachability for any mappings using these addresses
- *     If the memory is marked as available, it is added to the general
- *     VM pool, otherwise it is not (it is reserved for card IO etc).
+ *
+ *     THIS IS NOT SUPPORTED
  */
-kern_return_t pmap_add_physical_memory(vm_offset_t spa, vm_offset_t epa,
-                                      boolean_t available, unsigned int attr)
+kern_return_t
+pmap_add_physical_memory(
+       __unused vm_offset_t spa, 
+       __unused vm_offset_t epa,
+       __unused boolean_t available,
+       __unused unsigned int attr)
 {
-       int i,j;
-       spl_t s;
-
-       /* Only map whole pages */
        
        panic("Forget it! You can't map no more memory, you greedy puke!\n");
-
-       spa = trunc_page(spa);
-       epa = round_page(epa);
-
-       /* First check that the region doesn't already exist */
-
-       assert (epa >= spa);
-       for (i = 0; i < pmap_mem_regions_count; i++) {
-               /* If we're below the next region, then no conflict */
-               if (epa < pmap_mem_regions[i].start)
-                       break;
-               if (spa < pmap_mem_regions[i].end) {
-#if DEBUG
-                       kprintf("pmap_add_physical_memory(0x%08x,0x%08x,0x%08x) - memory already present\n",spa,epa,attr);
-#endif /* DEBUG */
-                       return KERN_NO_SPACE;
-               }
-       }
-
-#if DEBUG
-       kprintf("pmap_add_physical_memory; region insert spot: %d out of %d\n", i, pmap_mem_regions_count);     /* (TEST/DEBUG) */
-#endif
-
-       /* Check that we've got enough space for another region */
-       if (pmap_mem_regions_count == PMAP_MEM_REGION_MAX)
-               return KERN_RESOURCE_SHORTAGE;
-
-       /* Once here, i points to the mem_region above ours in physical mem */
-
-       /* allocate a new phys_table for this new region */
-#if DEBUG
-       kprintf("pmap_add_physical_memory; kalloc\n");  /* (TEST/DEBUG) */
-#endif
-
-       phys_table =  (struct phys_entry *)
-               kalloc(sizeof(struct phys_entry) * atop(epa-spa));
-#if DEBUG
-       kprintf("pmap_add_physical_memory; new phys_table: %08X\n", phys_table);        /* (TEST/DEBUG) */
-#endif
-
-       /* Initialise the new phys_table entries */
-       for (j = 0; j < atop(epa-spa); j++) {
-               
-               phys_table[j].phys_link = MAPPING_NULL;
-               
-               mapping_phys_init(&phys_table[j], spa+(j*PAGE_SIZE), attr);     /* Initialize the hardware specific portions */
-
-       }
-       s = splhigh();
-       
-       /* Move all the phys_table entries up some to make room in
-        * the ordered list.
-        */
-       for (j = pmap_mem_regions_count; j > i ; j--)
-               pmap_mem_regions[j] = pmap_mem_regions[j-1];
-
-       /* Insert a new entry with some memory to back it */
-
-       pmap_mem_regions[i].start            = spa;
-       pmap_mem_regions[i].end           = epa;
-       pmap_mem_regions[i].phys_table    = phys_table;
-
-       pmap_mem_regions_count++;
-       splx(s);
-       
-#if DEBUG
-       for(i=0; i<pmap_mem_regions_count; i++) {                       /* (TEST/DEBUG) */
-               kprintf("region %d: %08X %08X %08X\n", i, pmap_mem_regions[i].start,
-                       pmap_mem_regions[i].end, pmap_mem_regions[i].phys_table);       /* (TEST/DEBUG) */
-       }
-#endif
-
-       if (available) {
-               kprintf("warning : pmap_add_physical_mem() "
-                      "available not yet supported\n");
-       }
-
        return KERN_SUCCESS;
 }
 
@@ -329,7 +222,10 @@ kern_return_t pmap_add_physical_memory(vm_offset_t spa, vm_offset_t epa,
  *     Pages with a contiguous virtual address range, the same protection, and attributes.
  *     therefore, we map it with a single block.
  *
+ *     Note that this call will only map into 32-bit space
+ *
  */
+
 vm_offset_t
 pmap_map(
        vm_offset_t va,
@@ -338,72 +234,91 @@ pmap_map(
        vm_prot_t prot)
 {
 
-
-       if (spa == epa)
-               return(va);
+       addr64_t colladr;
+       
+       if (spa == epa) return(va);
 
        assert(epa > spa);
-       debugLog2(40, va, spa);                                                         /* Log pmap_map call */
-
-       pmap_map_block(kernel_pmap, va, spa, epa - spa, prot, PTE_WIMG_DEFAULT, blkPerm);       /* Set up a permanent block mapped area */
 
-       debugLog2(41, epa, prot);                                                       /* Log pmap_map call */
+       colladr = mapping_make(kernel_pmap, (addr64_t)va, (ppnum_t)(spa >> 12), (mmFlgBlock | mmFlgPerm), (epa - spa) >> 12, prot & VM_PROT_ALL);
 
+       if(colladr) {                                                                                   /* Was something already mapped in the range? */
+               panic("pmap_map: attempt to map previously mapped range - va = %08X, pa = %08X, epa = %08X, collision = %016llX\n",
+                       va, spa, epa, colladr);
+       }                               
        return(va);
 }
 
 /*
- * pmap_map_bd(va, spa, epa, prot)
- *     Back-door routine for mapping kernel VM at initialisation.
- *     Used for mapping memory outside the known physical memory
- *      space, with caching disabled. Designed for use by device probes.
- * 
- *     A virtual address range starting at "va" is mapped to the physical
- *     address range "spa" to "epa" with machine independent protection
- *     "prot".
- *
- *     "va", "spa", and "epa" are byte addresses and must be on machine
- *     independent page boundaries.
- *
- * WARNING: The current version of memcpy() can use the dcbz instruction
- * on the destination addresses.  This will cause an alignment exception
- * and consequent overhead if the destination is caching-disabled.  So
- * avoid memcpy()ing into the memory mapped by this function.
- *
- * also, many other pmap_ routines will misbehave if you try and change
- * protections or remove these mappings, they are designed to be permanent.
- *
- * These areas will be added to the autogen list, if possible.  Existing translations
- * are overridden and their mapping stuctures are released.  This takes place in
- * the autogen_map function.
+ * pmap_map_physical()
+ *     Maps physical memory into the kernel's address map beginning at lgPMWvaddr, the
+ *  physical memory window.
  *
- * Locking:
- *     this routine is called only during system initialization when only
- *     one processor is active, so no need to take locks...
  */
-vm_offset_t
-pmap_map_bd(
-       vm_offset_t va,
-       vm_offset_t spa,
-       vm_offset_t epa,
-       vm_prot_t prot)
+void
+pmap_map_physical()
 {
-       register struct mapping *mp;
-       register struct phys_entry      *pp;
-       
+       unsigned region;
+       uint64_t msize, size;
+       addr64_t paddr, vaddr, colladdr;
 
-       if (spa == epa)
-               return(va);
+       /* Iterate over physical memory regions, block mapping each into the kernel's address map */    
+       for (region = 0; region < (unsigned)pmap_mem_regions_count; region++) {
+               paddr = ((addr64_t)pmap_mem_regions[region].mrStart << 12);     /* Get starting physical address */
+               size  = (((addr64_t)pmap_mem_regions[region].mrEnd + 1) << 12) - paddr;
 
-       assert(epa > spa);
+               vaddr = paddr + lowGlo.lgPMWvaddr;                                      /* Get starting virtual address */
+
+               while (size > 0) {
+                       
+                       msize = ((size > 0x0000020000000000ULL) ? 0x0000020000000000ULL : size);        /* Get size, but no more than 2TBs */
+                       
+                       colladdr = mapping_make(kernel_pmap, vaddr, (paddr >> 12),
+                               (mmFlgBlock | mmFlgPerm), (msize >> 12),
+                               (VM_PROT_READ | VM_PROT_WRITE));
+                       if (colladdr) {
+                               panic ("pmap_map_physical: mapping failure - va = %016llX, pa = %08X, size = %08X, collision = %016llX\n",
+                                          vaddr, (paddr >> 12), (msize >> 12), colladdr);
+                       }
+
+                       vaddr = vaddr + (uint64_t)msize;                                /* Point to the next virtual addr */
+                       paddr = paddr + (uint64_t)msize;                                /* Point to the next physical addr */
+                       size  -= msize;
+               }
+       }
+}
+
+/*
+ * pmap_map_iohole(addr64_t paddr, addr64_t size)
+ *     Maps an I/O hole into the kernel's address map at its proper offset in
+ *     the physical memory window.
+ *
+ */
+void
+pmap_map_iohole(addr64_t paddr, addr64_t size)
+{
 
-       debugLog2(42, va, epa);                                                         /* Log pmap_map_bd call */
+       addr64_t vaddr, colladdr, msize;
+       uint32_t psize;
 
-       pmap_map_block(kernel_pmap, va, spa, epa - spa, prot, PTE_WIMG_IO, blkPerm);    /* Set up autogen area */
+       vaddr = paddr + lowGlo.lgPMWvaddr;                                              /* Get starting virtual address */              
 
-       debugLog2(43, epa, prot);                                                       /* Log pmap_map_bd exit */
+       while (size > 0) {
 
-       return(va);
+               msize = ((size > 0x0000020000000000ULL) ? 0x0000020000000000ULL : size);        /* Get size, but no more than 2TBs */
+               
+               colladdr = mapping_make(kernel_pmap, vaddr, (paddr >> 12),
+                       (mmFlgBlock | mmFlgPerm | mmFlgGuarded | mmFlgCInhib), (msize >> 12),
+                       (VM_PROT_READ | VM_PROT_WRITE));
+               if (colladdr) {
+                       panic ("pmap_map_iohole: mapping failed - va = %016llX, pa = %08X, size = %08X, collision = %016llX\n",
+                                  vaddr, (paddr >> 12), (msize >> 12), colladdr);
+               }
+
+               vaddr = vaddr + (uint64_t)msize;                                        /* Point to the next virtual addr */
+               paddr = paddr + (uint64_t)msize;                                        /* Point to the next physical addr */
+               size  -= msize;
+       }       
 }
 
 /*
@@ -412,362 +327,231 @@ pmap_map_bd(
  *     Called with mapping done by BATs. Page_size must already be set.
  *
  *     Parameters:
- *     mem_size:       Total memory present
+ *     msize:  Total memory present
  *     first_avail:    First virtual address available
- *     first_phys_avail:       First physical address available
+ *     kmapsize:       Size of kernel text and data
  */
 void
-pmap_bootstrap(unsigned int mem_size, vm_offset_t *first_avail, vm_offset_t *first_phys_avail, unsigned int kmapsize)
+pmap_bootstrap(uint64_t msize, vm_offset_t *first_avail, unsigned int kmapsize)
 {
-       register struct mapping *mp;
        vm_offset_t     addr;
        vm_size_t               size;
-       int                     i, num, j, rsize, mapsize, vmpagesz, vmmapsz;
-       unsigned int     mask;
-       vm_offset_t             first_used_addr;
-       PCA                             *pcaptr;
-       savectl                 *savec, *savec2;
-       vm_offset_t             save, save2;
+       unsigned int    i, num, mapsize, vmpagesz, vmmapsz, nbits;
+       signed                  bank;
+       uint64_t                tmemsize;
+       uint_t                  htslop;
+       vm_offset_t             first_used_addr, PCAsize;
+       struct phys_entry *phys_entry;
 
-       *first_avail = round_page(*first_avail);
-       
-#if DEBUG
-       kprintf("first_avail=%08X; first_phys_avail=%08X; avail_remaining=%d\n", 
-               *first_avail, *first_phys_avail, avail_remaining);
-#endif
-
-       assert(PAGE_SIZE == PPC_PGBYTES);
+       *first_avail = round_page(*first_avail);                                /* Make sure we start out on a page boundary */
+       vm_last_addr = VM_MAX_KERNEL_ADDRESS;                                   /* Set the highest address know to VM */
 
        /*
         * Initialize kernel pmap
         */
        kernel_pmap = &kernel_pmap_store;
+       kernel_pmap_phys = (addr64_t)&kernel_pmap_store;
        cursor_pmap = &kernel_pmap_store;
 
-       lock_init(&pmap_system_lock,
-                 FALSE,                /* NOT a sleep lock */
-                 ETAP_VM_PMAP_SYS,
-                 ETAP_VM_PMAP_SYS_I);
-
-       simple_lock_init(&kernel_pmap->lock, ETAP_VM_PMAP_KERNEL);
-
        kernel_pmap->pmap_link.next = (queue_t)kernel_pmap;             /* Set up anchor forward */
        kernel_pmap->pmap_link.prev = (queue_t)kernel_pmap;             /* Set up anchor reverse */
        kernel_pmap->ref_count = 1;
+       kernel_pmap->pmapFlags = pmapKeyDef;                                    /* Set the default keys */
+       kernel_pmap->pmapCCtl = pmapCCtlVal;                                    /* Initialize cache control */
        kernel_pmap->space = PPC_SID_KERNEL;
-       kernel_pmap->pmapvr = 0;                                                /* Virtual = Real  */
-       kernel_pmap->bmaps = 0;                                                 /* No block pages just yet */
-       for(i=0; i < 128; i++) {                                                /* Clear usage slots */
-               kernel_pmap->pmapUsage[i] = 0;
-       }
-       for(i=0; i < 16; i++) {                                                 /* Initialize for laughs */
-               kernel_pmap->pmapSegs[i] = SEG_REG_PROT | (i << 20) | PPC_SID_KERNEL;
-       }
-
-       /*
-        * Allocate: (from first_avail up)
-        *      Aligned to its own size:
-        *       hash table (for mem size 2**x, allocate 2**(x-10) entries)
-        *       mapping table (same size and immediatly following hash table)
-        */
-       /* hash_table_size must be a power of 2, recommended sizes are
-        * taken from PPC601 User Manual, table 6-19. We take the next
-        * highest size if mem_size is not a power of two.
-        * TODO NMGS make this configurable at boot time.
-        */
-
-       num = sizeof(pte_t) * (mem_size >> 10);
-
-       for (hash_table_size = 64 * 1024;       /* minimum size = 64Kbytes */
-            hash_table_size < num; 
-            hash_table_size *= 2)
-               continue;
-
-       /* Scale to within any physical memory layout constraints */
-       do {
-               num = atop(mem_size);   /* num now holds mem_size in pages */
-
-               /* size of all structures that we're going to allocate */
+       kernel_pmap->pmapvr = 0;                                                                /* Virtual = Real  */
 
-               size = (vm_size_t) (
-                       (InitialSaveBloks * PAGE_SIZE) +        /* Allow space for the initial context saveareas */
-                       (8 * PAGE_SIZE) +                                       /* For backpocket saveareas */
-                       hash_table_size +                                       /* For hash table */
-                       hash_table_size +                                       /* For PTEG allocation table */
-                       (num * sizeof(struct phys_entry))       /* For the physical entries */
-                       );
-
-               mapsize = size = round_page(size);              /* Get size of area to map that we just calculated */
-               mapsize = mapsize + kmapsize;                   /* Account for the kernel text size */
+/*
+ *  IBM's recommended hash table size is one PTEG for every 2 physical pages.
+ *  However, we have found that OSX rarely uses more than 4 PTEs in a PTEG
+ *  with this size table.  Therefore, by default we allocate a hash table
+ *  one half IBM's recommended size, ie one PTEG per 4 pages.  The "ht_shift" boot-arg
+ *  can be used to override the default hash table size.
+ *     We will allocate the hash table in physical RAM, outside of kernel virtual memory,
+ *     at the top of the highest bank that will contain it.
+ *     Note that "bank" doesn't refer to a physical memory slot here, it is a range of
+ *     physically contiguous memory.
+ *
+ *     The PCA will go there as well, immediately before the hash table.
+ */
+       nbits = cntlzw(((msize << 1) - 1) >> 32);                               /* Get first bit in upper half */
+       if (nbits == 32)                                        /* If upper half was empty, find bit in bottom half */
+        nbits = nbits + cntlzw((uint_t)((msize << 1) - 1));
+       tmemsize = 0x8000000000000000ULL >> nbits;              /* Get memory size rounded up to power of 2 */
+       
+    /* Calculate hash table size:  First, make sure we don't overflow 32-bit arithmetic. */
+       if (tmemsize > 0x0000002000000000ULL)
+        tmemsize = 0x0000002000000000ULL;
+
+    /* Second, calculate IBM recommended hash table size, ie one PTEG per 2 physical pages */
+       hash_table_size = (uint_t)(tmemsize >> 13) * PerProcTable[0].ppe_vaddr->pf.pfPTEG;
+    
+    /* Third, cut this in half to produce the OSX default, ie one PTEG per 4 physical pages */
+    hash_table_size >>= 1;
+    
+    /* Fourth, adjust default size per "ht_shift" boot arg */
+    if (hash_table_shift >= 0)                              /* if positive, make size bigger */
+        hash_table_size <<= hash_table_shift;
+    else                                                    /* if "ht_shift" is negative, make smaller */
+        hash_table_size >>= (-hash_table_shift);
+    
+    /* Fifth, make sure we are at least minimum size */
+       if (hash_table_size < (256 * 1024))
+        hash_table_size = (256 * 1024);
+
+       while(1) {                                                                                              /* Try to fit hash table in PCA into contiguous memory */
+
+               if(hash_table_size < (256 * 1024)) {                            /* Have we dropped too short? This should never, ever happen */
+                       panic("pmap_bootstrap: Can't find space for hash table\n");     /* This will never print, system isn't up far enough... */
+               }
 
-               vmpagesz = round_page(num * sizeof(struct vm_page));    /* Allow for all vm_pages needed to map physical mem */
-               vmmapsz = round_page((num / 8) * sizeof(struct vm_map_entry));  /* Allow for vm_maps */
+               PCAsize = (hash_table_size / PerProcTable[0].ppe_vaddr->pf.pfPTEG) * sizeof(PCA_t);     /* Get total size of PCA table */
+               PCAsize = round_page(PCAsize);                                  /* Make sure it is at least a page long */
+       
+               for(bank = pmap_mem_regions_count - 1; bank >= 0; bank--) {     /* Search backwards through banks */
+                       
+                       hash_table_base = ((addr64_t)pmap_mem_regions[bank].mrEnd << 12) - hash_table_size + PAGE_SIZE; /* Get tenative address */
+                       
+                       htslop = hash_table_base & (hash_table_size - 1);       /* Get the extra that we will round down when we align */
+                       hash_table_base = hash_table_base & -(addr64_t)hash_table_size; /* Round down to correct boundary */
+                       
+                       if((hash_table_base - round_page(PCAsize)) >= ((addr64_t)pmap_mem_regions[bank].mrStart << 12)) break;  /* Leave if we fit */
+               }
                
-               mapsize = mapsize + vmpagesz + vmmapsz; /* Add the VM system estimates into the grand total */
-
-               mapsize = mapsize + (4 * 1024 * 1024);  /* Allow for 4 meg of extra mappings */
-               mapsize = ((mapsize / PAGE_SIZE) + MAPPERBLOK - 1) / MAPPERBLOK;        /* Get number of blocks of mappings we need */
-               mapsize = mapsize + ((mapsize  + MAPPERBLOK - 1) / MAPPERBLOK); /* Account for the mappings themselves */
-
-#if DEBUG
-               kprintf("pmap_bootstrap: initial vm_pages     = %08X\n", vmpagesz);
-               kprintf("pmap_bootstrap: initial vm_maps      = %08X\n", vmmapsz);
-               kprintf("pmap_bootstrap: size before mappings = %08X\n", size);
-               kprintf("pmap_bootstrap: kernel map size      = %08X\n", kmapsize);
-               kprintf("pmap_bootstrap: mapping blocks rqrd  = %08X\n", mapsize);
-#endif
+               if(bank >= 0) break;                                                            /* We are done if we found a suitable bank */
                
-               size = size + (mapsize * PAGE_SIZE);    /* Get the true size we need */
-
-               /* hash table must be aligned to its size */
-
-               addr = (*first_avail +
-                       (hash_table_size-1)) & ~(hash_table_size-1);
+               hash_table_size = hash_table_size >> 1;                         /* Try the next size down */
+       }
 
-               if (addr + size > pmap_mem_regions[0].end) {
-                       hash_table_size /= 2;
-               } else {
-                       break;
+       if(htslop) {                                                                                    /* If there was slop (i.e., wasted pages for alignment) add a new region */
+               for(i = pmap_mem_regions_count - 1; i >= (unsigned)bank; i--) { /* Copy from end to our bank, including our bank */
+                       pmap_mem_regions[i + 1].mrStart  = pmap_mem_regions[i].mrStart; /* Set the start of the bank */
+                       pmap_mem_regions[i + 1].mrAStart = pmap_mem_regions[i].mrAStart;        /* Set the start of allocatable area */
+                       pmap_mem_regions[i + 1].mrEnd    = pmap_mem_regions[i].mrEnd;   /* Set the end address of bank */
+                       pmap_mem_regions[i + 1].mrAEnd   = pmap_mem_regions[i].mrAEnd;  /* Set the end address of allocatable area */
                }
-               /* If we have had to shrink hash table to too small, panic */
-               if (hash_table_size == 32 * 1024)
-                       panic("cannot lay out pmap memory map correctly");
-       } while (1);
+               
+               pmap_mem_regions[i + 1].mrStart  = (hash_table_base + hash_table_size) >> 12;   /* Set the start of the next bank to the start of the slop area */
+               pmap_mem_regions[i + 1].mrAStart = (hash_table_base + hash_table_size) >> 12;   /* Set the start of allocatable area to the start of the slop area */
+               pmap_mem_regions[i].mrEnd        = (hash_table_base + hash_table_size - 4096) >> 12;    /* Set the end of our bank to the end of the hash table */
+               
+       }               
        
-#if DEBUG
-       kprintf("hash table size=%08X, total size of area=%08X, addr=%08X\n", 
-               hash_table_size, size, addr);
-#endif
-       if (round_page(*first_phys_avail) < trunc_page(addr)) {
-               /* We are stepping over at least one page here, so
-                * add this region to the free regions so that it can
-                * be allocated by pmap_steal
-                */
-               free_regions[free_regions_count].start = round_page(*first_phys_avail);
-               free_regions[free_regions_count].end = trunc_page(addr);
-
-               avail_remaining += (free_regions[free_regions_count].end -
-                                   free_regions[free_regions_count].start) /
-                                           PPC_PGBYTES;
-#if DEBUG
-               kprintf("ADDED FREE REGION from 0x%08x to 0x%08x, avail_remaining = %d\n",
-                       free_regions[free_regions_count].start,free_regions[free_regions_count].end, 
-                       avail_remaining);
-#endif /* DEBUG */
-               free_regions_count++;
-       }
+       pmap_mem_regions[bank].mrAEnd = (hash_table_base - PCAsize - 4096) >> 12;       /* Set the maximum allocatable in this bank */
+       
+       hw_hash_init();                                                                                 /* Initiaize the hash table and PCA */
+       hw_setup_trans();                                                                               /* Set up hardware registers needed for translation */
+       
+/*
+ *     The hash table is now all initialized and so is the PCA.  Go on to do the rest of it.
+ *     This allocation is from the bottom up.
+ */    
+       
+       num = atop_64(msize);                                                                           /* Get number of pages in all of memory */
+
+/* Figure out how much we need to allocate */
 
-       /* Zero everything - this also invalidates the hash table entries */
-       bzero((char *)addr, size);
+       size = (vm_size_t) (
+               (InitialSaveBloks * PAGE_SIZE) +                                        /* Allow space for the initial context saveareas */
+               (BackPocketSaveBloks * PAGE_SIZE) +                                     /* For backpocket saveareas */
+               trcWork.traceSize +                                                             /* Size of trace table */
+               ((((1 << maxAdrSpb) * sizeof(pmapTransTab)) + 4095) & -4096) +  /* Size of pmap translate table */
+               (((num * sizeof(struct phys_entry)) + 4095) & -4096)    /* For the physical entries */
+       );
 
-       /* Set up some pointers to our new structures */
+       mapsize = size = round_page(size);                                              /* Get size of area to map that we just calculated */
+       mapsize = mapsize + kmapsize;                                                   /* Account for the kernel text size */
 
-       /* from here,  addr points to the next free address */
+       vmpagesz = round_page(num * sizeof(struct vm_page));    /* Allow for all vm_pages needed to map physical mem */
+       vmmapsz = round_page((num / 8) * sizeof(struct vm_map_entry));  /* Allow for vm_maps */
        
-       first_used_addr = addr; /* remember where we started */
+       mapsize = mapsize + vmpagesz + vmmapsz;                                 /* Add the VM system estimates into the grand total */
 
-       /* Set up hash table address and dma buffer address, keeping
-        * alignment. These mappings are all 1-1,  so dma_r == dma_v
-        * 
-        * If hash_table_size == dma_buffer_alignment, then put hash_table
-        * first, since dma_buffer_size may be smaller than alignment, but
-        * hash table alignment==hash_table_size.
-        */
-       hash_table_base = addr;
-               
-       addr += hash_table_size;
-       addr += hash_table_size;                                                        /* Add another for the PTEG Control Area */
-       assert((hash_table_base & (hash_table_size-1)) == 0);
+       mapsize = mapsize + (4 * 1024 * 1024);                                  /* Allow for 4 meg of extra mappings */
+       mapsize = ((mapsize / PAGE_SIZE) + MAPPERBLOK - 1) / MAPPERBLOK;        /* Get number of blocks of mappings we need */
+       mapsize = mapsize + ((mapsize  + MAPPERBLOK - 1) / MAPPERBLOK); /* Account for the mappings themselves */
 
-       pcaptr = (PCA *)(hash_table_base+hash_table_size);      /* Point to the PCA table */
-       
-       for(i=0; i < (hash_table_size/64) ; i++) {                      /* For all of PTEG control areas: */
-               pcaptr[i].flgs.PCAalflgs.PCAfree=0xFF;                  /* Mark all slots free */
-               pcaptr[i].flgs.PCAalflgs.PCAsteal=0x01;                 /* Initialize steal position */
-       }
-       
-/*
- *             Allocate our initial context save areas.  As soon as we do this,
- *             we can take an interrupt. We do the saveareas here, 'cause they're guaranteed
- *             to be at least page aligned.
- */
-       save2 = addr;                                                                           /* Remember first page */       
-       save =  addr;                                                                           /* Point to the whole block of blocks */        
-       savec2 = (savectl *)(addr + PAGE_SIZE - sizeof(savectl));       /* Point to the first's control area */
+       size = size + (mapsize * PAGE_SIZE);                                    /* Get the true size we need */
 
-       for(i=0; i < InitialSaveBloks; i++) {                           /* Initialize the saveareas */
+       /* hash table must be aligned to its size */
 
-               savec = (savectl *)(save + PAGE_SIZE - sizeof(savectl));        /* Get the control area for this one */
+       addr = *first_avail;                                                                    /* Set the address to start allocations */
+       first_used_addr = addr;                                                                 /* Remember where we started */
 
-               savec->sac_alloc = sac_empty;                                   /* Mark both free */
-               savec->sac_vrswap = 0;                                                  /* V=R, so the translation factor is 0 */
-               savec->sac_flags = sac_perm;                                    /* Mark it permanent */
+       bzero((char *)addr, size);                                                              /* Clear everything that we are allocating */
 
-               savec->sac_flags |= 0x0000EE00;                                 /* (TEST/DEBUG) */
+       savearea_init(addr);                                                                    /* Initialize the savearea chains and data */
 
-               save += PAGE_SIZE;                                                              /* Jump up to the next one now */
-               
-               savec->sac_next = (unsigned int *)save;                 /* Link these two */
-       
-       }
-       
-       savec->sac_next = (unsigned int *)0;                            /* Clear the forward pointer for the last */
-       savec2->sac_alloc &= 0x7FFFFFFF;                                        /* Mark the first one in use */
-
-       saveanchor.savefree             = (unsigned int)save2;          /* Point to the first one */
-       saveanchor.savecount    = InitialSaveBloks * sac_cnt;   /* The total number of save areas allocated */
-       saveanchor.saveinuse    = 1;                                            /* Number of areas in use */
-       saveanchor.savemin              = InitialSaveMin;                       /* We abend if lower than this */
-       saveanchor.saveneghyst  = InitialNegHysteresis;         /* The minimum number to keep free (must be a multiple of sac_cnt) */
-       saveanchor.savetarget   = InitialSaveTarget;            /* The target point for free save areas (must be a multiple of sac_cnt) */
-       saveanchor.saveposhyst  = InitialPosHysteresis;         /* The high water mark for free save areas (must be a multiple of sac_cnt) */
-       __asm__ volatile ("mtsprg 1, %0" : : "r" (save2));      /* Tell the exception handler about it */
-               
-       addr += InitialSaveBloks * PAGE_SIZE;                           /* Move up the next free address */
+       addr = (vm_offset_t)((unsigned int)addr + ((InitialSaveBloks + BackPocketSaveBloks) * PAGE_SIZE));      /* Point past saveareas */
 
-       save2 = addr;                                                                           
-       save =  addr;                                                                           
-       savec2 = (savectl *)(addr + PAGE_SIZE - sizeof(savectl));       
+       trcWork.traceCurr = (unsigned int)addr;                                 /* Set first trace slot to use */
+       trcWork.traceStart = (unsigned int)addr;                                /* Set start of trace table */
+       trcWork.traceEnd = (unsigned int)addr + trcWork.traceSize;              /* Set end of trace table */
 
-       for(i=0; i < 8; i++) {                                                          /* Allocate backpocket saveareas */
-       
-               savec = (savectl *)(save + PAGE_SIZE - sizeof(savectl));        
-       
-               savec->sac_alloc = sac_empty;                                   
-               savec->sac_vrswap = 0;                                                  
-               savec->sac_flags = sac_perm;                                    
-               savec->sac_flags |= 0x0000EE00;                                 
+       addr = (vm_offset_t)trcWork.traceEnd;                                   /* Set next allocatable location */
                
-               save += PAGE_SIZE;                                                              
+       pmapTrans = (pmapTransTab *)addr;                                               /* Point to the pmap to hash translation table */
                
-               savec->sac_next = (unsigned int *)save;                 
-       
-       }
-       
-       savec->sac_next = (unsigned int *)0;                            
-       savec2->sac_alloc &= 0x7FFFFFFF;                                        
-       debugbackpocket = save2;                                                        
-       addr += 8 * PAGE_SIZE;                                                          
-                       
-       /* phys_table is static to help debugging,
-        * this variable is no longer actually used
-        * outside of this scope
-        */
+       pmapTrans[PPC_SID_KERNEL].pmapPAddr = (addr64_t)((uintptr_t)kernel_pmap);       /* Initialize the kernel pmap in the translate table */
+       pmapTrans[PPC_SID_KERNEL].pmapVAddr = CAST_DOWN(unsigned int, kernel_pmap);  /* Initialize the kernel pmap in the translate table */
+               
+       addr += ((((1 << maxAdrSpb) * sizeof(pmapTransTab)) + 4095) & -4096);   /* Point past pmap translate table */
 
-       phys_table = (struct phys_entry *) addr;
+/*     NOTE: the phys_table must be within the first 2GB of physical RAM. This makes sure we only need to do 32-bit arithmetic */
 
-#if DEBUG
-       kprintf("hash_table_base                 =%08X\n", hash_table_base);
-       kprintf("phys_table                      =%08X\n", phys_table);
-       kprintf("pmap_mem_regions_count          =%08X\n", pmap_mem_regions_count);
-#endif
+       phys_entry = (struct phys_entry *) addr;                                /* Get pointer to physical table */
 
-       for (i = 0; i < pmap_mem_regions_count; i++) {
+       for (bank = 0; bank < pmap_mem_regions_count; bank++) { /* Set pointer and initialize all banks of ram */
                
-               pmap_mem_regions[i].phys_table = phys_table;
-               rsize = (pmap_mem_regions[i].end - (unsigned int)pmap_mem_regions[i].start)/PAGE_SIZE;
+               pmap_mem_regions[bank].mrPhysTab = phys_entry;          /* Set pointer to the physical table for this bank */
                
-#if DEBUG
-               kprintf("Initializing physical table for region %d\n", i);
-               kprintf("   table=%08X, size=%08X, start=%08X, end=%08X\n",
-                       phys_table, rsize, pmap_mem_regions[i].start, 
-                       (unsigned int)pmap_mem_regions[i].end);
-#endif         
-               
-               for (j = 0; j < rsize; j++) {
-                       phys_table[j].phys_link = MAPPING_NULL;
-                       mapping_phys_init(&phys_table[j], (unsigned int)pmap_mem_regions[i].start+(j*PAGE_SIZE), 
-                               PTE_WIMG_DEFAULT);                                              /* Initializes hw specific storage attributes */
-               }
-               phys_table = phys_table +
-                       atop(pmap_mem_regions[i].end - pmap_mem_regions[i].start);
+               phys_entry = phys_entry + (pmap_mem_regions[bank].mrEnd - pmap_mem_regions[bank].mrStart + 1);  /* Point to the next */
        }
 
-       /* restore phys_table for debug */
-       phys_table = (struct phys_entry *) addr;
-
-       addr += sizeof(struct phys_entry) * num;
+       addr += (((num * sizeof(struct phys_entry)) + 4095) & -4096);   /* Step on past the physical entries */
        
-       simple_lock_init(&tlb_system_lock, ETAP_VM_PMAP_TLB);
-
-       /* Initialise the registers necessary for supporting the hashtable */
-#if DEBUG
-       kprintf("*** hash_table_init: base=%08X, size=%08X\n", hash_table_base, hash_table_size);
-#endif
-
-       hash_table_init(hash_table_base, hash_table_size);
-                       
 /*
  *             Remaining space is for mapping entries.  Tell the initializer routine that
  *             the mapping system can't release this block because it's permanently assigned
  */
 
-       mapping_init();                                                                 /* Initialize the mapping tables */
+       mapping_init();                                                                                 /* Initialize the mapping tables */
 
        for(i = addr; i < first_used_addr + size; i += PAGE_SIZE) {     /* Add initial mapping blocks */
-               mapping_free_init(i, 1, 0);                                     /* Pass block address and say that this one is not releasable */
+               mapping_free_init(i, 1, 0);                                                     /* Pass block address and say that this one is not releasable */
        }
-       mapCtl.mapcmin = MAPPERBLOK;                                    /* Make sure we only adjust one at a time */
-
-#if DEBUG
-
-       kprintf("mapping kernel memory from 0x%08x to 0x%08x, to address 0x%08x\n",
-                first_used_addr, round_page(first_used_addr+size),
-                first_used_addr);
-#endif /* DEBUG */
+       mapCtl.mapcmin = MAPPERBLOK;                                                    /* Make sure we only adjust one at a time */
 
        /* Map V=R the page tables */
        pmap_map(first_used_addr, first_used_addr,
-                round_page(first_used_addr+size), VM_PROT_READ | VM_PROT_WRITE);
+                round_page(first_used_addr + size), VM_PROT_READ | VM_PROT_WRITE);
 
-#if DEBUG
+       *first_avail = round_page(first_used_addr + size);              /* Set next available page */
+       first_free_virt = *first_avail;                                                 /* Ditto */
+       
+       /* For 64-bit machines, block map physical memory and the I/O hole into kernel space */
+       if(BootProcInfo.pf.Available & pf64Bit) {                               /* Are we on a 64-bit machine? */
+               lowGlo.lgPMWvaddr = PHYS_MEM_WINDOW_VADDR;                      /* Initialize the physical memory window's virtual address */
 
-       for(i=first_used_addr; i < round_page(first_used_addr+size); i+=PAGE_SIZE) {    /* Step through all these mappings */
-               if(i != (j = kvtophys(i))) {                                                    /* Verify that the mapping was made V=R */
-                       kprintf("*** V=R mapping failed to verify: V=%08X; R=%08X\n", i, j);
-               }
+               pmap_map_physical();                                                            /* Block map physical memory into the window */
+               
+               pmap_map_iohole(IO_MEM_WINDOW_VADDR, IO_MEM_WINDOW_SIZE);
+                                                                                                                       /* Block map the I/O hole */
        }
-#endif
-
-       *first_avail = round_page(first_used_addr + size);
-       first_free_virt = round_page(first_used_addr + size);
 
        /* All the rest of memory is free - add it to the free
         * regions so that it can be allocated by pmap_steal
         */
-       free_regions[free_regions_count].start = *first_avail;
-       free_regions[free_regions_count].end = pmap_mem_regions[0].end;
-
-       avail_remaining += (free_regions[free_regions_count].end -
-                           free_regions[free_regions_count].start) /
-                                   PPC_PGBYTES;
-
-#if DEBUG
-       kprintf("ADDED FREE REGION from 0x%08x to 0x%08x, avail_remaining = %d\n",
-               free_regions[free_regions_count].start,free_regions[free_regions_count].end, 
-               avail_remaining);
-#endif /* DEBUG */
-
-       free_regions_count++;
 
-       current_free_region = 0;
+       pmap_mem_regions[0].mrAStart = (*first_avail >> 12);    /* Set up the free area to start allocations (always in the first bank) */
 
-       avail_next = free_regions[current_free_region].start;
-       
-#if DEBUG
-       kprintf("Number of free regions=%d\n",free_regions_count);      /* (TEST/DEBUG) */
-       kprintf("Current free region=%d\n",current_free_region);        /* (TEST/DEBUG) */
-       for(i=0;i<free_regions_count; i++) {                                    /* (TEST/DEBUG) */
-               kprintf("Free region %3d - from %08X to %08X\n", i, free_regions[i].start,
-                       free_regions[i].end);                                                   /* (TEST/DEBUG) */
+       current_free_region = 0;                                                                /* Set that we will start allocating in bank 0 */
+       avail_remaining = 0;                                                                    /* Clear free page count */
+       for(bank = 0; bank < pmap_mem_regions_count; bank++) {  /* Total up all of the pages in the system that are available */
+               avail_remaining += (pmap_mem_regions[bank].mrAEnd - pmap_mem_regions[bank].mrAStart) + 1;       /* Add in allocatable pages in this bank */
        }
-       for (i = 0; i < pmap_mem_regions_count; i++) {                  /* (TEST/DEBUG) */
-               kprintf("PMAP region %3d - from %08X to %08X; phys=%08X\n", i,  /* (TEST/DEBUG) */
-                       pmap_mem_regions[i].start,                                              /* (TEST/DEBUG) */
-                       pmap_mem_regions[i].end,                                                /* (TEST/DEBUG) */
-                       pmap_mem_regions[i].phys_table);                                /* (TEST/DEBUG) */
-       }
-#endif
+
 
 }
 
@@ -787,7 +571,6 @@ void
 pmap_init(void)
 {
 
-
        pmap_zone = zinit(pmapSize, 400 * pmapSize, 4096, "pmap");
 #if    ZONE_DEBUG
        zone_debug_disable(pmap_zone);          /* Can't debug this one 'cause it messes with size and alignment */
@@ -800,7 +583,8 @@ pmap_init(void)
         */
        free_pmap_list = 0;                                     /* Set that there are no free pmaps */
        free_pmap_count = 0;
-       simple_lock_init(&free_pmap_lock, ETAP_VM_PMAP_CACHE);
+       simple_lock_init(&free_pmap_lock, 0);
+       
 }
 
 unsigned int pmap_free_pages(void)
@@ -808,63 +592,32 @@ unsigned int pmap_free_pages(void)
        return avail_remaining;
 }
 
-boolean_t pmap_next_page(vm_offset_t *addrp)
+/*
+ *     This function allocates physical pages.
+ */
+
+/* Non-optimal, but only used for virtual memory startup.
+ * Allocate memory from a table of free physical addresses
+ * If there are no more free entries, too bad. 
+ */
+
+boolean_t pmap_next_page(ppnum_t *addrp)
 {
-       /* Non optimal, but only used for virtual memory startup.
-     * Allocate memory from a table of free physical addresses
-        * If there are no more free entries, too bad. We have two
-        * tables to look through, free_regions[] which holds free
-        * regions from inside pmap_mem_regions[0], and the others...
-        * pmap_mem_regions[1..]
-     */
-        
-       /* current_free_region indicates the next free entry,
-        * if it's less than free_regions_count, then we're still
-        * in free_regions, otherwise we're in pmap_mem_regions
-        */
+               int i;
 
-       if (current_free_region >= free_regions_count) {
-               /* We're into the pmap_mem_regions, handle this
-                * separately to free_regions
-                */
-
-               int current_pmap_mem_region = current_free_region -
-                                        free_regions_count + 1;
-               if (current_pmap_mem_region > pmap_mem_regions_count)
-                       return FALSE;
-               *addrp = avail_next;
-               avail_next += PAGE_SIZE;
-               avail_remaining--;
-               if (avail_next >= pmap_mem_regions[current_pmap_mem_region].end) {
-                       current_free_region++;
-                       current_pmap_mem_region++;
-                       avail_next = pmap_mem_regions[current_pmap_mem_region].start;
-#if DEBUG
-                       kprintf("pmap_next_page : next region start=0x%08x\n",avail_next);
-#endif /* DEBUG */
-               }
-               return TRUE;
-       }
+       if(current_free_region >= pmap_mem_regions_count) return FALSE; /* Return failure if we have used everything... */
        
-       /* We're in the free_regions, allocate next page and increment
-        * counters
-        */
-       *addrp = avail_next;
-
-       avail_next += PAGE_SIZE;
-       avail_remaining--;
-
-       if (avail_next >= free_regions[current_free_region].end) {
-               current_free_region++;
-               if (current_free_region < free_regions_count)
-                       avail_next = free_regions[current_free_region].start;
-               else
-                       avail_next = pmap_mem_regions[current_free_region -
-                                                free_regions_count + 1].start;
-#if DEBUG
-               kprintf("pmap_next_page : next region start=0x%08x\n",avail_next);
-#endif 
+       for(i = current_free_region; i < pmap_mem_regions_count; i++) { /* Find the next bank with free pages */
+               if(pmap_mem_regions[i].mrAStart <= pmap_mem_regions[i].mrAEnd) break;   /* Found one */
        }
+       
+       current_free_region = i;                                                                                /* Set our current bank */
+       if(i >= pmap_mem_regions_count) return FALSE;                                   /* Couldn't find a free page */
+
+       *addrp = pmap_mem_regions[i].mrAStart;                                  /* Allocate the page */
+       pmap_mem_regions[i].mrAStart = pmap_mem_regions[i].mrAStart + 1;        /* Set the next one to go */
+       avail_remaining--;                                                                                              /* Drop free count */
+
        return TRUE;
 }
 
@@ -873,7 +626,7 @@ void pmap_virtual_space(
        vm_offset_t *endp)
 {
        *startp = round_page(first_free_virt);
-       *endp   = VM_MAX_KERNEL_ADDRESS;
+       *endp   = vm_last_addr;
 }
 
 /*
@@ -895,21 +648,12 @@ void pmap_virtual_space(
  * only, and is bounded by that size.
  */
 pmap_t
-pmap_create(vm_size_t size)
+pmap_create(vm_map_size_t size)
 {
-       pmap_t pmap, ckpmap, fore, aft;
-       int s, i;
-       space_t sid;
+       pmap_t pmap, ckpmap, fore;
+       int s;
        unsigned int currSID;
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00001, size, 0);                  /* (TEST/DEBUG) */
-#endif
-
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_create(size=%x)%c", size, size ? '\n' : ' ');
-#endif
+       addr64_t physpmap;
 
        /*
         * A software use-only map doesn't even need a pmap structure.
@@ -925,76 +669,69 @@ pmap_create(vm_size_t size)
        s = splhigh();
        simple_lock(&free_pmap_lock);
        
-       if(free_pmap_list) {                                    /* Any free? */
-               pmap = free_pmap_list;                          /* Yes, allocate it */
-               free_pmap_list = (pmap_t)pmap->bmaps;   /* Dequeue this one (we chain free ones through bmaps) */
+       if(free_pmap_list) {                                                    /* Any free? */
+               pmap = free_pmap_list;                                          /* Yes, allocate it */
+               free_pmap_list = (pmap_t)pmap->freepmap;        /* Dequeue this one (we chain free ones through freepmap) */
                free_pmap_count--;
        }
        else {
-               simple_unlock(&free_pmap_lock);         /* Unlock just in case */
+               simple_unlock(&free_pmap_lock);                         /* Unlock just in case */
                splx(s);
 
-               pmap = (pmap_t) zalloc(pmap_zone);      /* Get one */
+               pmap = (pmap_t) zalloc(pmap_zone);                      /* Get one */
                if (pmap == PMAP_NULL) return(PMAP_NULL);       /* Handle out-of-memory condition */
                
-               bzero((char *)pmap, pmapSize);          /* Clean up the pmap */
+               bzero((char *)pmap, pmapSize);                          /* Clean up the pmap */
                
                s = splhigh();
-               simple_lock(&free_pmap_lock);           /* Lock it back up      */
+               simple_lock(&free_pmap_lock);                           /* Lock it back up      */
                
-               ckpmap = cursor_pmap;                           /* Get starting point for free ID search */
-               currSID = ckpmap->spaceNum;                     /* Get the actual space ID number */
+               ckpmap = cursor_pmap;                                           /* Get starting point for free ID search */
+               currSID = ckpmap->spaceNum;                                     /* Get the actual space ID number */
 
-               while(1) {                                                      /* Keep trying until something happens */
+               while(1) {                                                                      /* Keep trying until something happens */
                
-                       currSID = (currSID + 1) & SID_MAX;      /* Get the next in the sequence */
+                       currSID = (currSID + 1) & (maxAdrSp - 1);       /* Get the next in the sequence */
+                       if(((currSID * incrVSID) & (maxAdrSp - 1)) == invalSpace) continue;     /* Skip the space we have reserved */
                        ckpmap = (pmap_t)ckpmap->pmap_link.next;        /* On to the next in-use pmap */
        
                        if(ckpmap->spaceNum != currSID) break;  /* If we are out of sequence, this is free */
                        
-                       if(ckpmap == cursor_pmap) {             /* See if we have 2^20 already allocated */
-                               panic("pmap_create: Maximum number (2^20) active address spaces reached\n");    /* Die pig dog */
+                       if(ckpmap == cursor_pmap) {                             /* See if we have 2^20 already allocated */
+                               panic("pmap_create: Maximum number (%d) active address spaces reached\n", maxAdrSp);    /* Die pig dog */
                        }
                }
 
-               pmap->space = (currSID * incrVSID) & SID_MAX;   /* Calculate the actual VSID */
-               pmap->spaceNum = currSID;                       /* Set the space ID number */
-
+               pmap->space = (currSID * incrVSID) & (maxAdrSp - 1);    /* Calculate the actual VSID */
+               pmap->spaceNum = currSID;                                       /* Set the space ID number */
 /*
  *             Now we link into the chain just before the out of sequence guy.
  */
 
-               fore = (pmap_t)ckpmap->pmap_link.prev;  /* Get the current's previous */
-               pmap->pmap_link.next = (queue_t)ckpmap; /* My next points to the current */
-               fore->pmap_link.next = (queue_t)pmap;   /* Current's previous's next points to me */
-               pmap->pmap_link.prev = (queue_t)fore;   /* My prev points to what the current pointed to */
-               ckpmap->pmap_link.prev = (queue_t)pmap; /* Current's prev points to me */
-
-               simple_lock_init(&pmap->lock, ETAP_VM_PMAP);
-               pmap->pmapvr = (unsigned int)pmap ^ (unsigned int)pmap_extract(kernel_pmap, (vm_offset_t)pmap); /* Get physical pointer to the pmap and make mask */
+               fore = (pmap_t)ckpmap->pmap_link.prev;          /* Get the current's previous */
+               pmap->pmap_link.next = (queue_t)ckpmap;         /* My next points to the current */
+               fore->pmap_link.next = (queue_t)pmap;           /* Current's previous's next points to me */
+               pmap->pmap_link.prev = (queue_t)fore;           /* My prev points to what the current pointed to */
+               ckpmap->pmap_link.prev = (queue_t)pmap;         /* Current's prev points to me */
+               
+               physpmap = ((addr64_t)pmap_find_phys(kernel_pmap, (addr64_t)((uintptr_t)pmap)) << 12) | (addr64_t)((unsigned int)pmap & 0xFFF); /* Get the physical address of the pmap */
+               
+               pmap->pmapvr = (addr64_t)((uintptr_t)pmap) ^ physpmap;  /* Make V to R translation mask */
+               
+               pmapTrans[pmap->space].pmapPAddr = physpmap;    /* Set translate table physical to point to us */
+               pmapTrans[pmap->space].pmapVAddr = CAST_DOWN(unsigned int, pmap);       /* Set translate table virtual to point to us */
        }
+
+       pmap->pmapVmmExt = 0;                                                   /* Clear VMM extension block vaddr */
+       pmap->pmapVmmExtPhys = 0;                                               /*  and the paddr, too */
+       pmap->pmapFlags = pmapKeyDef;                                   /* Set default key */
+       pmap->pmapCCtl = pmapCCtlVal;                                   /* Initialize cache control */
        pmap->ref_count = 1;
        pmap->stats.resident_count = 0;
        pmap->stats.wired_count = 0;
-       pmap->bmaps = 0;                                                /* Clear block map pointer to 0 */
-       pmap->vflags = 0;                                               /* Mark all alternates invalid for now */
-       for(i=0; i < 128; i++) {                                /* Clean out usage slots */
-               pmap->pmapUsage[i] = 0;
-       }
-       for(i=0; i < 16; i++) {                                 /* Initialize for laughs */
-               pmap->pmapSegs[i] = SEG_REG_PROT | (i << 20) | pmap->space;
-       }
-       
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00002, (unsigned int)pmap, (unsigned int)pmap->space);    /* (TEST/DEBUG) */
-#endif
-
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("-> %x, space id = %d\n", pmap, pmap->space);
-#endif
-
+       pmap->pmapSCSubTag = 0x0000000000000000ULL;             /* Make sure this is clean an tidy */
        simple_unlock(&free_pmap_lock);
+
        splx(s);
        return(pmap);
 }
@@ -1014,15 +751,6 @@ pmap_destroy(pmap_t pmap)
        spl_t s;
        pmap_t fore, aft;
 
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00003, (unsigned int)pmap, 0);                    /* (TEST/DEBUG) */
-#endif
-
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_destroy(pmap=%x)\n", pmap);
-#endif
-
        if (pmap == PMAP_NULL)
                return;
 
@@ -1031,13 +759,17 @@ pmap_destroy(pmap_t pmap)
 
        if(ref_count < 0)                                                                               /* Did we go too far? */
                panic("pmap_destroy(): ref_count < 0");
+
+       if (!(pmap->pmapFlags & pmapVMgsaa)) {                                  /* Don't try this for a shadow assist guest */
+               pmap_unmap_sharedpage(pmap);                                            /* Remove any mapping of page -1 */
+       }
        
 #ifdef notdef
        if(pmap->stats.resident_count != 0)
                panic("PMAP_DESTROY: pmap not empty");
 #else
        if(pmap->stats.resident_count != 0) {
-               pmap_remove(pmap, 0, 0xFFFFF000);
+               pmap_remove(pmap, 0, 0xFFFFFFFFFFFFF000ULL);
        }
 #endif
 
@@ -1051,9 +783,9 @@ pmap_destroy(pmap_t pmap)
         */
        simple_lock(&free_pmap_lock);
        
-       if (free_pmap_count <= free_pmap_max) {                                 /* Do we have enough spares? */
+       if (free_pmap_count <= free_pmap_max) {         /* Do we have enough spares? */
                
-               pmap->bmaps = (struct blokmap *)free_pmap_list;         /* Queue in front */
+               pmap->freepmap = free_pmap_list;                /* Queue in front */
                free_pmap_list = pmap;
                free_pmap_count++;
                simple_unlock(&free_pmap_lock);
@@ -1065,7 +797,9 @@ pmap_destroy(pmap_t pmap)
                fore->pmap_link.next = pmap->pmap_link.next;    /* My previous's next is my next */
                aft->pmap_link.prev = pmap->pmap_link.prev;             /* My next's previous is my previous */ 
                simple_unlock(&free_pmap_lock);
-               zfree(pmap_zone, (vm_offset_t) pmap);
+               pmapTrans[pmap->space].pmapPAddr = -1;                  /* Invalidate the translate table physical */
+               pmapTrans[pmap->space].pmapVAddr = -1;                  /* Invalidate the translate table virtual */
+               zfree(pmap_zone, pmap);
        }
        splx(s);
 }
@@ -1077,17 +811,6 @@ pmap_destroy(pmap_t pmap)
 void
 pmap_reference(pmap_t pmap)
 {
-       spl_t s;
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00004, (unsigned int)pmap, 0);                    /* (TEST/DEBUG) */
-#endif
-
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_reference(pmap=%x)\n", pmap);
-#endif
-
        if (pmap != PMAP_NULL) hw_atomic_add(&pmap->ref_count, 1);      /* Bump the count */
 }
 
@@ -1102,17 +825,56 @@ void pmap_remove_some_phys(
             vm_offset_t pa)
 {
        register struct phys_entry      *pp;
-       register struct mapping         *mp, *mpv;
-
+       register struct mapping         *mp;
+       unsigned int pindex;
 
-       if (pmap == PMAP_NULL) return;                          /* Do nothing if no pmap */
+       if (pmap == PMAP_NULL) {                                        /* This should never be called with a null pmap */
+               panic("pmap_remove_some_phys: null pmap\n");
+       }
 
-       pp = pmap_find_physentry(pa);                           /* Get the physent for this page */
-       if (pp == PHYS_NULL) return;                            /* Leave if not in physical RAM */
+       pp = mapping_phys_lookup(pa, &pindex);          /* Get physical entry */
+       if (pp == 0) return;                                            /* Leave if not in physical RAM */
+
+       do {                                                                            /* Keep going until we toss all pages from this pmap */
+               if (pmap->pmapFlags & pmapVMhost) {
+                       mp = hw_purge_phys(pp);                         /* Toss a map */
+                       switch ((unsigned int)mp & mapRetCode) {
+                               case mapRtOK:
+                                       mapping_free(mp);                       /* Return mapping to free inventory */
+                                       break;
+                               case mapRtGuest:
+                                       break;                                          /* Don't try to return a guest mapping */
+                               case mapRtEmpty:
+                                       break;                                          /* Physent chain empty, we're done */
+                               case mapRtNotFnd:                               
+                                       break;                                          /* Mapping disappeared on us, retry */  
+                               default:
+                                       panic("pmap_remove_some_phys: hw_purge_phys failed - pp = %08X, pmap = %08X, code = %08X\n",
+                                                       pp, pmap, mp);          /* Handle failure with our usual lack of tact */
+                       }
+               } else { 
+                       mp = hw_purge_space(pp, pmap);          /* Toss a map */
+                       switch ((unsigned int)mp & mapRetCode) {
+                               case mapRtOK:
+                                       mapping_free(mp);                       /* Return mapping to free inventory */
+                                       break;
+                               case mapRtEmpty:
+                                       break;                                          /* Physent chain empty, we're done */
+                               case mapRtNotFnd:                               
+                                       break;                                          /* Mapping disappeared on us, retry */  
+                               default:
+                                       panic("pmap_remove_some_phys: hw_purge_phys failed - pp = %08X, pmap = %08X, code = %08X\n",
+                                                       pp, pmap, mp);          /* Handle failure with our usual lack of tact */
+                       }
+               }
+       } while (mapRtEmpty != ((unsigned int)mp & mapRetCode));
 
-       mapping_purge_pmap(pp, pmap);   
+#if DEBUG      
+       if ((pmap->pmapFlags & pmapVMhost) && !pmap_verify_free(pa)) 
+               panic("pmap_remove_some_phys: cruft left behind - pa = %08X, pmap = %08X\n", pa, pmap);
+#endif
 
-       return;                                                 /* Leave... */
+       return;                                                                         /* Leave... */
 }
 
 /*
@@ -1128,25 +890,13 @@ void pmap_remove_some_phys(
 void
 pmap_remove(
            pmap_t pmap,
-           vm_offset_t sva,
-           vm_offset_t eva)
+           addr64_t sva,
+           addr64_t eva)
 {
-       spl_t                   spl;
-       struct mapping  *mp, *blm;
-       vm_offset_t             lpage;
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00005, (unsigned int)pmap, sva|((eva-sva)>>12));  /* (TEST/DEBUG) */
-#endif
+       addr64_t                va, endva;
 
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_remove(pmap=%x, sva=%x, eva=%x)\n",
-                      pmap, sva, eva);
-#endif
+       if (pmap == PMAP_NULL) return;                                  /* Leave if software pmap */
 
-       if (pmap == PMAP_NULL)
-               return;
 
        /* It is just possible that eva might have wrapped around to zero,
         * and sometimes we get asked to liberate something of size zero
@@ -1155,49 +905,17 @@ pmap_remove(
        assert(eva >= sva);
 
        /* If these are not page aligned the loop might not terminate */
-       assert((sva == trunc_page(sva)) && (eva == trunc_page(eva)));
+       assert((sva == trunc_page_64(sva)) && (eva == trunc_page_64(eva)));
 
-       /* We liberate addresses from high to low, since the stack grows
-        * down. This means that we won't need to test addresses below
-        * the limit of stack growth
-        */
+       va = sva & -4096LL;                                                     /* Round start down to a page */
+       endva = eva & -4096LL;                                          /* Round end down to a page */
 
-       debugLog2(44, sva, eva);                                        /* Log pmap_map call */
-       
-       sva = trunc_page(sva);                                          /* Make it clean */
-       lpage = trunc_page(eva) - PAGE_SIZE;            /* Point to the last page contained in the range */
-
-/*
- *     Here we will remove all of the block mappings that overlap this range.
- *     hw_rem_blk removes one mapping in the range and returns.  If it returns
- *     0, there are no blocks in the range.
- */
-
-       while(mp = (mapping *)hw_rem_blk(pmap, sva, lpage)) {   /* Keep going until no more */
-               if((unsigned int)mp & 1) {                                                      /* Make sure we don't unmap a permanent one */
-                       blm = (blokmap *)hw_cpv((mapping *)((unsigned int)mp & 0xFFFFFFFE));            /* Get virtual address */
-                       panic("mapping_remove: attempt to unmap a permanent mapping - pmap = %08X, va = %08X, mapping = %08X\n",
-                               pmap, sva, blm);
-               }
-               mapping_free(hw_cpv(mp));                                                       /* Release it */
-       }
-
-       while (pmap->stats.resident_count && (eva > sva)) {
-
-               eva -= PAGE_SIZE;                                               /* Back up a page */
-
-#if 1
-               if((0x00008000 >> (sva >> 28)) & pmap->vflags)
-                  panic("pmap_remove: attempt to remove nested vaddr; pmap = %08X, vaddr = %08X\n", pmap, sva);        /* (TEST/DEBUG) panic */
-#endif
-               if(!(pmap->pmapUsage[(eva >> pmapUsageShft) & pmapUsageMask])) {        /* See if this chunk has anything in it */
-                       eva = eva & (-pmapUsageSize);           /* Back up into the previous slot */
-                       continue;                                                       /* Check the next... */
-               }
-               mapping_remove(pmap, eva);                              /* Remove the mapping for this address */
+       while(1) {                                                                      /* Go until we finish the range */
+               va = mapping_remove(pmap, va);                  /* Remove the mapping and see what's next */
+               va = va & -4096LL;                                              /* Make sure the "not found" indication is clear */
+               if((va == 0) || (va >= endva)) break;   /* End loop if we finish range or run off the end */
        }
 
-       debugLog2(45, 0, 0);                                            /* Log pmap_map call */
 }
 
 /*
@@ -1209,24 +927,15 @@ pmap_remove(
  */
 void
 pmap_page_protect(
-       vm_offset_t pa,
+       ppnum_t pa,
        vm_prot_t prot)
 {
        register struct phys_entry      *pp;
        boolean_t                       remove;
+       unsigned int            pindex;
+       mapping_t                       *mp;
 
 
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00006, (unsigned int)pa, (unsigned int)prot);     /* (TEST/DEBUG) */
-#endif
-
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_page_protect(pa=%x, prot=%x)\n", pa, prot);
-#endif
-
-       debugLog2(46, pa, prot);                                        /* Log pmap_page_protect call */
-
        switch (prot) {
                case VM_PROT_READ:
                case VM_PROT_READ|VM_PROT_EXECUTE:
@@ -1239,23 +948,87 @@ pmap_page_protect(
                        break;
        }
 
-       pp = pmap_find_physentry(pa);                           /* Get the physent for this page */
-       if (pp == PHYS_NULL) return;                            /* Leave if not in physical RAM */
+
+       pp = mapping_phys_lookup(pa, &pindex);          /* Get physical entry */
+       if (pp == 0) return;                                            /* Leave if not in physical RAM */
 
        if (remove) {                                                           /* If the protection was set to none, we'll remove all mappings */
-               mapping_purge(pp);                                              /* Get rid of them all */
+               
+               do {                                                                    /* Keep going until we toss all pages from this physical page */
+                       mp = hw_purge_phys(pp);                         /* Toss a map */
+                       switch ((unsigned int)mp & mapRetCode) {
+                               case mapRtOK:
+                                                       mapping_free(mp);       /* Return mapping to free inventory */
+                                                       break;
+                               case mapRtGuest:
+                                                       break;                          /* Don't try to return a guest mapping */
+                               case mapRtNotFnd:
+                                                       break;                          /* Mapping disappeared on us, retry */
+                               case mapRtEmpty:
+                                                       break;                          /* Physent chain empty, we're done */
+                               default:        panic("pmap_page_protect: hw_purge_phys failed - pp = %08X, code = %08X\n",
+                                                                 pp, mp);              /* Handle failure with our usual lack of tact */
+                       }
+               } while (mapRtEmpty != ((unsigned int)mp & mapRetCode));
+
+#if DEBUG
+               if (!pmap_verify_free(pa)) 
+                       panic("pmap_page_protect: cruft left behind - pa = %08X\n", pa);
+#endif
 
-               debugLog2(47, 0, 0);                                    /* Log pmap_map call */
                return;                                                                 /* Leave... */
        }
+
+/*     When we get here, it means that we are to change the protection for a 
+ *     physical page.  
+ */
+       mapping_protect_phys(pa, prot & VM_PROT_ALL);   /* Change protection of all mappings to page. */
+
+}
+
+/*
+ *     Routine:
+ *             pmap_disconnect
+ *
+ *     Function:
+ *             Disconnect all mappings for this page and return reference and change status
+ *             in generic format.
+ *
+ */
+unsigned int pmap_disconnect(
+       ppnum_t pa)
+{
+       register struct phys_entry *pp;
+       unsigned int                            pindex;
+       mapping_t                                  *mp;
        
-       /*      When we get here, it means that we are to change the protection for a 
-        *      physical page.  
-        */
-        
-       mapping_protect_phys(pp, prot, 0);                      /* Change protection of all mappings to page. */
+       pp = mapping_phys_lookup(pa, &pindex);          /* Get physical entry */
+       if (pp == 0) return (0);                                        /* Return null ref and chg if not in physical RAM */
+       do {                                                                            /* Iterate until all mappings are dead and gone */
+               mp = hw_purge_phys(pp);                                 /* Disconnect a mapping */
+               if (!mp) break;                                                 /* All mappings are gone, leave the loop */
+               switch ((unsigned int)mp & mapRetCode) {
+                       case mapRtOK:
+                                               mapping_free(mp);               /* Return mapping to free inventory */
+                                               break;
+                       case mapRtGuest:
+                                               break;                                  /* Don't try to return a guest mapping */
+                       case mapRtNotFnd:
+                                               break;                                  /* Mapping disappeared on us, retry */
+                       case mapRtEmpty:
+                                               break;                                  /* Physent chain empty, we're done */
+                       default:        panic("hw_purge_phys: hw_purge_phys failed - pp = %08X, code = %08X\n",
+                                                         pp, mp);                      /* Handle failure with our usual lack of tact */
+               }
+       } while (mapRtEmpty != ((unsigned int)mp & mapRetCode));
 
-       debugLog2(47, 1, 0);                                            /* Log pmap_map call */
+#if DEBUG
+       if (!pmap_verify_free(pa)) 
+               panic("pmap_disconnect: cruft left behind - pa = %08X\n", pa);
+#endif
+
+       return (mapping_tst_refmod(pa));                        /* Return page ref and chg in generic format */
 }
 
 /*
@@ -1270,60 +1043,32 @@ pmap_page_protect(
  */
 void pmap_protect(
             pmap_t pmap,
-            vm_offset_t sva, 
-            vm_offset_t eva,
+            vm_map_offset_t sva, 
+            vm_map_offset_t eva,
             vm_prot_t prot)
 {
-       spl_t                                           spl;
-       register struct phys_entry      *pp;
-       register struct mapping         *mp, *mpv;
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00008, (unsigned int)pmap, (unsigned int)(sva|((eva-sva)>>12)));  /* (TEST/DEBUG) */
-#endif
 
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_protect(pmap=%x, sva=%x, eva=%x, prot=%x)\n", pmap, sva, eva, prot);
-       
-       assert(sva < eva);
-#endif
+       addr64_t va, endva;
 
        if (pmap == PMAP_NULL) return;                          /* Do nothing if no pmap */
 
-       debugLog2(48, sva, eva);                                        /* Log pmap_map call */
-
        if (prot == VM_PROT_NONE) {                                     /* Should we kill the address range?? */
-               pmap_remove(pmap, sva, eva);                    /* Yeah, dump 'em */
-
-               debugLog2(49, prot, 0);                                                         /* Log pmap_map call */
-
+               pmap_remove(pmap, (addr64_t)sva, (addr64_t)eva);        /* Yeah, dump 'em */
                return;                                                                 /* Leave... */
        }
 
-       sva = trunc_page(sva);                                          /* Start up a page boundary */
-       
-       while(sva < eva) {                                                      /* Step through */
-
-               if(!(pmap->pmapUsage[(sva >> pmapUsageShft) & pmapUsageMask])) {        /* See if this chunk has anything in it */
-                       sva = (sva + pmapUsageSize) &(-pmapUsageSize);  /* Jump up into the next slot if nothing here */
-                       if(!sva) break;                                         /* We tried to wrap, kill loop... */
-                       continue;                                                       /* Check the next... */
-               }
-
-#if 1
-               if((0x00008000 >> (sva >> 28)) & pmap->vflags)
-                  panic("pmap_protect: attempt to protect nested vaddr; pmap = %08X, vaddr = %08X\n", pmap, sva);      /* (TEST/DEBUG) panic */
-#endif
+       va = sva & -4096LL;                                                     /* Round start down to a page */
+       endva = eva & -4096LL;                                          /* Round end down to a page */
 
-               mapping_protect(pmap, sva, prot);               /* Change the protection on the page */
-               sva += PAGE_SIZE;                                               /* On to the next page */
+       while(1) {                                                                      /* Go until we finish the range */
+               mapping_protect(pmap, va, prot & VM_PROT_ALL, &va);     /* Change the protection and see what's next */
+               if((va == 0) || (va >= endva)) break;   /* End loop if we finish range or run off the end */
        }
 
-       debugLog2(49, prot, 1);                                         /* Log pmap_map call */
-       return;                                                                         /* Leave... */
 }
 
+
+
 /*
  * pmap_enter
  *
@@ -1338,51 +1083,96 @@ void pmap_protect(
  *     insert this page into the given map NOW.
  */
 void
-pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot,
-          boolean_t wired)
+pmap_enter(pmap_t pmap, vm_map_offset_t va, ppnum_t pa, vm_prot_t prot, 
+               unsigned int flags, __unused boolean_t wired)
 {
-       spl_t                           spl;
-       struct mapping          *mp;
-       struct phys_entry       *pp;
-       int                                     memattr;
+       unsigned int            mflags;
+       addr64_t                        colva;
+       
+       if (pmap == PMAP_NULL) return;                                  /* Leave if software pmap */
+
+       mflags = 0;                                                                             /* Make sure this is initialized to nothing special */
+       if(!(flags & VM_WIMG_USE_DEFAULT)) {                    /* Are they supplying the attributes? */
+               mflags = mmFlgUseAttr | (flags & VM_MEM_GUARDED) | ((flags & VM_MEM_NOT_CACHEABLE) >> 1);       /* Convert to our mapping_make flags */
+       }
        
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00009, (unsigned int)pmap, (unsigned int)va);     /* (TEST/DEBUG) */
-       dbgTrace(0xF1D04009, (unsigned int)pa, (unsigned int)prot);     /* (TEST/DEBUG) */
-#endif         
+/*
+ *     It is possible to hang here if another processor is remapping any pages we collide with and are removing
+ */ 
+
+       while(1) {                                                                              /* Keep trying the enter until it goes in */
        
-       if (pmap == PMAP_NULL) return;                                  /* If they gave us no pmap, just leave... */
+               colva = mapping_make(pmap, va, pa, mflags, 1, prot & VM_PROT_ALL);      /* Enter the mapping into the pmap */
+               
+               if(!colva) break;                                                       /* If there were no collisions, we are done... */
+               
+               mapping_remove(pmap, colva);                            /* Remove the mapping that collided */
+       }
+}
 
-       debugLog2(50, va, pa);                                                  /* Log pmap_map call */
+/*
+ *             Enters translations for odd-sized V=F blocks.
+ *
+ *             The higher level VM map should be locked to insure that we don't have a
+ *             double diddle here.
+ *
+ *             We panic if we get a block that overlaps with another. We do not merge adjacent
+ *             blocks because removing any address within a block removes the entire block and if
+ *             would really mess things up if we trashed too much.
+ *
+ *             Once a block is mapped, it is unmutable, that is, protection, catch mode, etc. can
+ *             not be changed.  The block must be unmapped and then remapped with the new stuff.
+ *             We also do not keep track of reference or change flags.
+ *
+ *             Any block that is larger than 256MB must be a multiple of 32MB.  We panic if it is not.
+ *
+ *             Note that pmap_map_block_rc is the same but doesn't panic if collision.
+ *
+ */
+void pmap_map_block(pmap_t pmap, addr64_t va, ppnum_t pa, uint32_t size, vm_prot_t prot, int attr, unsigned int flags) {       /* Map an autogenned block */
 
-       pp = pmap_find_physentry(pa);                                   /* Get the physent for this physical page */
+       unsigned int            mflags;
+       addr64_t                        colva;
 
-       if((0x00008000 >> (va >> 28)) & pmap->vflags)
-          panic("pmap_enter: attempt to map into nested vaddr; pmap = %08X, vaddr = %08X\n", pmap, va);        /* (TEST/DEBUG) panic */
+       
+       if (pmap == PMAP_NULL) {                                                /* Did they give us a pmap? */
+               panic("pmap_map_block: null pmap\n");           /* No, like that's dumb... */
+       }
 
-       spl=splhigh();                                                                  /* Have to disallow interrupts between the
-                                                                                                          time we possibly clear a mapping and the time
-                                                                                                          we get it remapped again.  An I/O SLIH could
-                                                                                                          try to drive an IOR using the page before
-                                                                                                          we get it mapped (Dude! This was a tough 
-                                                                                                          bug!!!!) */
+//     kprintf("pmap_map_block: (%08X) va = %016llX, pa = %08X, size = %08X, prot = %08X, attr = %08X, flags = %08X\n",        /* (BRINGUP) */
+//             current_thread(), va, pa, size, prot, attr, flags);     /* (BRINGUP) */
 
-       mapping_remove(pmap, va);                                               /* Remove any other mapping at this address */
+       mflags = mmFlgBlock | mmFlgUseAttr | (attr & VM_MEM_GUARDED) | ((attr & VM_MEM_NOT_CACHEABLE) >> 1);    /* Convert to our mapping_make flags */
+       if(flags) mflags |= mmFlgPerm;                                  /* Mark permanent if requested */
+       
+       colva = mapping_make(pmap, va, pa, mflags, size, prot); /* Enter the mapping into the pmap */
        
-       memattr = PTE_WIMG_IO;                                                  /* Assume I/O mapping for a moment */
-       if(pp) memattr = ((pp->pte1&0x00000078) >> 3);  /* Set the attribute to the physical default */
+       if(colva) {                                                                             /* If there was a collision, panic */
+               panic("pmap_map_block: mapping error %d, pmap = %08X, va = %016llX\n", (uint32_t)(colva & mapRetCode), pmap, va);
+       }
+       
+       return;                                                                                 /* Return */
+}
 
-       mp=mapping_make(pmap, pp, va, pa, prot, memattr, 0);    /* Make the address mapping */
+int pmap_map_block_rc(pmap_t pmap, addr64_t va, ppnum_t pa, uint32_t size, vm_prot_t prot, int attr, unsigned int flags) {     /* Map an autogenned block */
 
-       splx(spl);                                                                              /* I'm not busy no more - come what may */
+       unsigned int            mflags;
+       addr64_t                        colva;
 
-       debugLog2(51, prot, 0);                                                 /* Log pmap_map call */
+       
+       if (pmap == PMAP_NULL) {                                                /* Did they give us a pmap? */
+               panic("pmap_map_block_rc: null pmap\n");        /* No, like that's dumb... */
+       }
 
-#if    DEBUG
-       if (pmdebug & (PDB_USER|PDB_ENTER))
-               kprintf("leaving pmap_enter\n");
-#endif
+       mflags = mmFlgBlock | mmFlgUseAttr | (attr & VM_MEM_GUARDED) | ((attr & VM_MEM_NOT_CACHEABLE) >> 1);    /* Convert to our mapping_make flags */
+       if(flags) mflags |= mmFlgPerm;                                  /* Mark permanent if requested */
 
+       colva = mapping_make(pmap, va, pa, mflags, size, prot); /* Enter the mapping into the pmap */
+       
+       if(colva) return 0;                                                             /* If there was a collision, fail */
+       
+       return 1;                                                                               /* Return true of we worked */
 }
 
 /*
@@ -1390,62 +1180,90 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_offset_t pa, vm_prot_t prot,
  *     returns the physical address corrsponding to the 
  *     virtual address specified by pmap and va if the
  *     virtual address is mapped and 0 if it is not.
+ *     Note: we assume nothing is ever mapped to phys 0.
+ *
+ *     NOTE: This call always will fail for physical addresses greater than 0xFFFFF000.
  */
-vm_offset_t pmap_extract(pmap_t pmap, vm_offset_t va) {
+vm_offset_t pmap_extract(pmap_t pmap, vm_map_offset_t va) {
 
        spl_t                                   spl;
-       register struct mapping *mp, *mpv;
+       register struct mapping *mp;
        register vm_offset_t    pa;
-       unsigned int                    seg;
-       pmap_t                                  actpmap;
-
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D0000B, (unsigned int)pmap, (unsigned int)va);     /* (TEST/DEBUG) */
-#endif
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_extract(pmap=%x, va=%x)\n", pmap, va);
-#endif
-
-       seg = va >> 28;                                                                 /* Isolate segment */
-       if((0x00008000 >> seg) & pmap->vflags) actpmap = pmap->pmapPmaps[seg];  /* Set nested pmap if there is one */
-       else actpmap = pmap;                                                    /* Otherwise use the one passed in */
+       addr64_t                                nextva;
+       ppnum_t                                 ppoffset;
+       unsigned int                    gva;
 
-       pa = (vm_offset_t) 0;                                                   /* Clear this to 0 */ 
+#ifdef BOGUSCOMPAT
+       panic("pmap_extract: THIS CALL IS BOGUS. NEVER USE IT EVER. So there...\n");    /* Don't use this */
+#else
 
-       debugLog2(52, actpmap->space, va);                                      /* Log pmap_map call */
+       gva = (unsigned int)va;                                                 /* Make sure we don't have a sign */
 
        spl = splhigh();                                                                /* We can't allow any loss of control here */
+       
+       mp = mapping_find(pmap, (addr64_t)gva, &nextva,1);      /* Find the mapping for this address */
+       
+       if(!mp) {                                                                               /* Is the page mapped? */
+               splx(spl);                                                                      /* Enable interrupts */
+               return 0;                                                                       /* Pass back 0 if not found */
+       }
 
-       if(mp=hw_lock_phys_vir(actpmap->space, va)) {   /* Find the mapping for this vaddr and lock physent */  
-               if((unsigned int)mp&1) {                                        /* Did the lock on the phys entry time out? */
-                       panic("pmap_extract: timeout obtaining lock on physical entry\n");      /* Scream bloody murder! */
-                       splx(spl);                                                              /* Interruptions are cool now */
-                       return 0;
-               }
+       ppoffset = (ppnum_t)(((gva & -4096LL) - (mp->mpVAddr & -4096LL)) >> 12);        /* Get offset from va to base va */
+       
+       
+       pa = mp->mpPAddr + ppoffset;                                    /* Remember ppage because mapping may vanish after drop call */
+                       
+       mapping_drop_busy(mp);                                                  /* We have everything we need from the mapping */
+       splx(spl);                                                                              /* Restore 'rupts */
 
-               mpv = hw_cpv(mp);                                                       /* Get virtual address of mapping */
-               pa = (vm_offset_t)((mpv->PTEr & -PAGE_SIZE) | ((unsigned int)va & (PAGE_SIZE-1)));      /* Build the physical address */
-               if(mpv->physent) hw_unlock_bit((unsigned int *)&mpv->physent->phys_link, PHYS_LOCK);    /* Unlock the physical entry */
-               splx(spl);                                                                      /* Interruptions are cool now */
+       if(pa > maxPPage32) return 0;                                   /* Force large addresses to fail */
+       
+       pa = (pa << 12) | (va & 0xFFF);                                 /* Convert physical page number to address */
+       
+#endif
+       return pa;                                                                              /* Return physical address or 0 */
+}
+
+/*
+ * ppnum_t pmap_find_phys(pmap, addr64_t va)
+ *     returns the physical page corrsponding to the 
+ *     virtual address specified by pmap and va if the
+ *     virtual address is mapped and 0 if it is not.
+ *     Note: we assume nothing is ever mapped to phys 0.
+ *
+ */
+ppnum_t pmap_find_phys(pmap_t pmap, addr64_t va) {
 
-               debugLog2(53, pa, 0);                                           /* Log pmap_map call */
+       spl_t                                   spl;
+       register struct mapping *mp;
+       ppnum_t                                 pa, ppoffset;
+       addr64_t                                nextva;
 
-               return pa;                                                                      /* Return the physical address... */
+       spl = splhigh();                                                                /* We can't allow any loss of control here */
+       
+       mp = mapping_find(pmap, va, &nextva, 1);                /* Find the mapping for this address */
+       
+       if(!mp) {                                                                               /* Is the page mapped? */
+               splx(spl);                                                                      /* Enable interrupts */
+               return 0;                                                                       /* Pass back 0 if not found */
        }
+               
+       
+       ppoffset = (ppnum_t)(((va & -4096LL) - (mp->mpVAddr & -4096LL)) >> 12); /* Get offset from va to base va */
+       
+       pa = mp->mpPAddr + ppoffset;                                    /* Get the actual physical address */
+
+       mapping_drop_busy(mp);                                                  /* We have everything we need from the mapping */
 
-       pa = hw_cvp_blk(pmap, va);                                              /* Try to convert odd-sized page (returns 0 if not found) */
-                                                                                                       /* Note no nested pmaps here */
        splx(spl);                                                                              /* Restore 'rupts */
-       debugLog2(53, pa, 0);                                                   /* Log pmap_map call */
        return pa;                                                                              /* Return physical address or 0 */
 }
 
+
 /*
  *     pmap_attributes:
  *
- *     Set/Get special memory attributes; Set is not implemented.
+ *     Set/Get special memory attributes; not implemented.
  *
  *     Note: 'VAL_GET_INFO' is used to return info about a page.
  *       If less than 1 page is specified, return the physical page
@@ -1454,174 +1272,67 @@ vm_offset_t pmap_extract(pmap_t pmap, vm_offset_t va) {
  *       of resident pages and the number of shared (more than
  *       one mapping) pages in the range;
  *
+ *
  */
 kern_return_t
-pmap_attribute(pmap, address, size, attribute, value)
-       pmap_t                  pmap;
-       vm_offset_t             address;
-       vm_size_t               size;
-       vm_machine_attribute_t  attribute;
-       vm_machine_attribute_val_t* value;      
+pmap_attribute(
+       __unused pmap_t                         pmap,
+       __unused vm_map_offset_t                address,
+       __unused vm_map_size_t                  size,
+       __unused vm_machine_attribute_t         attribute,
+       __unused vm_machine_attribute_val_t*    value)  
 {
-       spl_t                   s;
-       vm_offset_t     sva, eva;
-       vm_offset_t             pa;
-       kern_return_t   ret;
-       register struct mapping *mp, *mpv;
-       register struct phys_entry *pp;
-       int                     total, seg;
-       pmap_t                  actpmap;
-
-       if (attribute != MATTR_CACHE)
-               return KERN_INVALID_ARGUMENT;
+       
+       return KERN_INVALID_ARGUMENT;
 
-       /* We can't get the caching attribute for more than one page
-        * at a time
-        */
-       if ((*value == MATTR_VAL_GET) &&
-           (trunc_page(address) != trunc_page(address+size-1)))
-               return KERN_INVALID_ARGUMENT;
+}
 
-       if (pmap == PMAP_NULL)
-               return KERN_SUCCESS;
-
-       sva = trunc_page(address);
-       eva = round_page(address + size);
-       ret = KERN_SUCCESS;
-
-       debugLog2(54, address, attribute);                                              /* Log pmap_map call */
-
-       switch (*value) {
-               case MATTR_VAL_CACHE_SYNC:                                                      /* sync I+D caches */
-               case MATTR_VAL_CACHE_FLUSH:                                                     /* flush from all caches */
-               case MATTR_VAL_DCACHE_FLUSH:                                            /* flush from data cache(s) */
-               case MATTR_VAL_ICACHE_FLUSH:                                            /* flush from instr cache(s) */
-                       sva = trunc_page(sva);
-                       s = splhigh();
-
-                       while (sva < eva) {     
-                               seg = sva >> 28;                                                        /* Isolate segment */
-                               if((0x00008000 >> seg) & pmap->vflags) actpmap = pmap->pmapPmaps[seg];  /* Set nested pmap if there is one */
-                               else actpmap = pmap;                                            /* Otherwise use the one passed in */
-                       
 /*
- *                             Note: the following should work ok with nested pmaps because there are not overlayed mappings 
+ * pmap_attribute_cache_sync(vm_offset_t pa)
+ * 
+ * Invalidates all of the instruction cache on a physical page and
+ * pushes any dirty data from the data cache for the same physical page
  */
-                               if(!(actpmap->pmapUsage[(sva >> pmapUsageShft) & pmapUsageMask])) {     /* See if this chunk has anything in it */
-                                       sva = (sva + pmapUsageSize) & (-pmapUsageSize); /* Jump up into the next slot if nothing here */
-                                       if(!sva) break;                                                 /* We tried to wrap, kill loop... */
-                                       continue;                                                               /* Check the next... */
-                               }
-
-                               if(!(mp = hw_lock_phys_vir(actpmap->space, sva))) {     /* Find the mapping for this vaddr and lock physent */
-                                       sva += PAGE_SIZE;                                               /* Point to the next page */
-                                       continue;                                                               /* Skip if the page is not mapped... */
-                               }
-
-                               if((unsigned int)mp&1) {                                        /* Did the lock on the phys entry time out? */
-                                       panic("pmap_attribute: timeout obtaining lock on physical entry\n");    /* Scream bloody murder! */
-                                       continue;
-                               }
-                               
-                               mpv = hw_cpv(mp);                                                       /* Get virtual address of mapping */
-                               if((unsigned int)mpv->physent) {                        /* Is there a physical entry? */
-                                       pa = (vm_offset_t)mpv->physent->pte1 & -PAGE_SIZE;      /* Yes, get the physical address from there */
-                               } 
-                               else {
-                                       pa = (vm_offset_t)(mpv->PTEr & PAGE_SIZE);      /* Otherwise from the mapping */
-                               }
-
-                               switch (*value) {                                                       /* What type was that again? */
-                                       case MATTR_VAL_CACHE_SYNC:                              /* It is sync I+D caches */
-                                               sync_cache(pa, PAGE_SIZE);                      /* Sync up dem caches */
-                                               break;                                                          /* Done with this one here... */
-                                       
-                                       case MATTR_VAL_CACHE_FLUSH:                             /* It is flush from all caches */
-                                               flush_dcache(pa, PAGE_SIZE, TRUE);      /* Flush out the data cache */
-                                               invalidate_icache(pa, PAGE_SIZE, TRUE); /* Flush out the instruction cache */
-                                               break;                                                          /* Done with this one here... */
-                                       
-                                       case MATTR_VAL_DCACHE_FLUSH:                    /* It is flush from data cache(s) */
-                                               flush_dcache(pa, PAGE_SIZE, TRUE);      /* Flush out the data cache */
-                                               break;                                                          /* Done with this one here... */
-
-                                       case MATTR_VAL_ICACHE_FLUSH:                    /* It is flush from instr cache(s) */
-                                               invalidate_icache(pa, PAGE_SIZE, TRUE); /* Flush out the instruction cache */
-                                               break;                                                          /* Done with this one here... */
-                               }
-                               if(mpv->physent) hw_unlock_bit((unsigned int *)&mpv->physent->phys_link, PHYS_LOCK);    /* Unlock the physical entry if it exists*/
-                               
-                               sva += PAGE_SIZE;                                                       /* Point to the next page */
-                       }
-                       splx(s);
-                       break;
-
-               case MATTR_VAL_GET_INFO:                                                        /* Get info */
-                       total = 0;
-                       s = splhigh();                                                                  /* Lock 'em out */
-               
-                       if (size <= PAGE_SIZE) {                                                /* Do they want just one page */
-                               seg = sva >> 28;                                                        /* Isolate segment */
-                               if((0x00008000 >> seg) & pmap->vflags) actpmap = pmap->pmapPmaps[seg];  /* Set nested pmap if there is one */
-                               else actpmap = pmap;                                            /* Otherwise use the one passed in */
-                               if(!(mp = hw_lock_phys_vir(actpmap->space, sva))) {     /* Find the mapping for this vaddr and lock physent */
-                                       *value = 0;                                                             /* Return nothing if no mapping */
-                               }
-                               else {
-                                       if((unsigned int)mp&1) {                                /* Did the lock on the phys entry time out? */
-                                               panic("pmap_attribute: timeout obtaining lock on physical entry\n");    /* Scream bloody murder! */
-                                       }
-                                       mpv = hw_cpv(mp);                                               /* Get virtual address of mapping */
-                                       if(pp = mpv->physent) {                                 /* Check for a physical entry */
-                                               total = 0;                                                      /* Clear the count */
-                                               for (mpv = (mapping *)hw_cpv((mapping *)((unsigned int)pp->phys_link & ~PHYS_FLAGS)); mpv != NULL; mpv = hw_cpv(mp->next)) total++;     /* Count the mapping */
-                                               *value = (vm_machine_attribute_val_t) ((pp->pte1 & -PAGE_SIZE) | total);        /* Pass back the physical address and the count of mappings */
-                                               hw_unlock_bit((unsigned int *)&pp->phys_link, PHYS_LOCK);       /* Clear the physical entry lock */
-                                       }
-                                       else {                                                                  /* This is the case for an I/O mapped area */
-                                               *value = (vm_machine_attribute_val_t) ((mpv->PTEr & -PAGE_SIZE) | 1);   /* Pass back the physical address and the count of mappings */
-                                       }
-                               }
-                       }
-                       else {
-                               total = 0;
-                               while (sva < eva) {
-                                       seg = sva >> 28;                                                /* Isolate segment */
-                                       if((0x00008000 >> seg) & pmap->vflags) actpmap = pmap->pmapPmaps[seg];  /* Set nested pmap if there is one */
-                                       else actpmap = pmap;                                    /* Otherwise use the one passed in */
-                                       
-                                       if(!(actpmap->pmapUsage[(sva >> pmapUsageShft) & pmapUsageMask])) {     /* See if this chunk has anything in it */
-                                               sva = (sva + pmapUsageSize) & (-pmapUsageSize); /* Jump up into the next slot if nothing here */
-                                               if(!sva) break;                                         /* We tried to wrap, kill loop... */
-                                               continue;                                                       /* Check the next... */
-                                       }
-                                       if(mp = hw_lock_phys_vir(actpmap->space, sva)) {        /* Find the mapping for this vaddr and lock physent */
-                                               if((unsigned int)mp&1) {                        /* Did the lock on the phys entry time out? */
-                                                       panic("pmap_attribute: timeout obtaining lock on physical entry\n");    /* Scream bloody murder! */
-                                                       continue;
-                                               }
-                                               mpv = hw_cpv(mp);                                       /* Get virtual address of mapping */
-                                               total += 65536 + (mpv->physent && ((mapping *)((unsigned int)mpv->physent->phys_link & -32))->next);    /* Count the "resident" and shared pages */
-                                               hw_unlock_bit((unsigned int *)&mpv->physent->phys_link, PHYS_LOCK);     /* Clear the physical entry lock */
-                                       }
-                                       sva += PAGE_SIZE;
-                               }
-                               *value = total;
-                       }
-                       splx(s);
-                       break;
+kern_return_t pmap_attribute_cache_sync(ppnum_t pp, vm_size_t size,
+                               __unused vm_machine_attribute_t  attribute,
+                               __unused vm_machine_attribute_val_t* value) {
        
-               case MATTR_VAL_GET:             /* return current value */
-               case MATTR_VAL_OFF:             /* turn attribute off */
-               case MATTR_VAL_ON:              /* turn attribute on */
-               default:
-                       ret = KERN_INVALID_ARGUMENT;
-                       break;
+       spl_t s;
+       unsigned int i, npages;
+       
+       npages = round_page(size) >> 12;                        /* Get the number of pages to do */
+       
+       for(i = 0; i < npages; i++) {                           /* Do all requested pages */
+               s = splhigh();                                                  /* No interruptions here */
+               sync_ppage(pp + i);                                             /* Go flush data cache and invalidate icache */
+               splx(s);                                                                /* Allow interruptions */
        }
+       
+       return KERN_SUCCESS;
+}
 
-       debugLog2(55, 0, 0);                                    /* Log pmap_map call */
+/*
+ * pmap_sync_page_data_phys(ppnum_t pa)
+ * 
+ * Invalidates all of the instruction cache on a physical page and
+ * pushes any dirty data from the data cache for the same physical page
+ */
+void pmap_sync_page_data_phys(ppnum_t pa) {
+       
+       spl_t s;
+       
+       s = splhigh();                                                          /* No interruptions here */
+       sync_ppage(pa);                                                         /* Sync up dem caches */
+       splx(s);                                                                        /* Allow interruptions */
+       return;
+}
 
-       return ret;
+void
+pmap_sync_page_attributes_phys(ppnum_t pa)
+{
+       pmap_sync_page_data_phys(pa);
 }
 
 /*
@@ -1631,7 +1342,7 @@ pmap_attribute(pmap, address, size, attribute, value)
  * It isn't implemented or needed or wanted.
  */
 void
-pmap_collect(pmap_t pmap)
+pmap_collect(__unused pmap_t pmap)
 {
        return;
 }
@@ -1645,9 +1356,9 @@ pmap_collect(pmap_t pmap)
  */
 void
 pmap_activate(
-       pmap_t pmap,
-       thread_t th,
-       int which_cpu)
+       __unused pmap_t pmap,
+       __unused thread_t th,
+       __unused int which_cpu)
 {
        return;
 }
@@ -1657,76 +1368,13 @@ pmap_activate(
  */
 void
 pmap_deactivate(
-       pmap_t pmap,
-       thread_t th,
-       int which_cpu)
+       __unused pmap_t pmap,
+       __unused thread_t th,
+       __unused int which_cpu)
 {
        return;
 }
 
-#if DEBUG
-
-/*
- * pmap_zero_page
- * pmap_copy page
- * 
- * are implemented in movc.s, these
- * are just wrappers to help debugging
- */
-
-extern void pmap_zero_page_assembler(vm_offset_t p);
-extern void pmap_copy_page_assembler(vm_offset_t src, vm_offset_t dst);
-
-/*
- * pmap_zero_page(pa)
- *
- * pmap_zero_page zeros the specified (machine independent) page pa.
- */
-void
-pmap_zero_page(
-       vm_offset_t p)
-{
-       register struct mapping *mp;
-       register struct phys_entry *pp;
-
-       if (pmdebug & (PDB_USER|PDB_ZERO))
-               kprintf("pmap_zero_page(pa=%x)\n", p);
-
-       /*
-        * XXX can these happen?
-        */
-       if (pmap_find_physentry(p) == PHYS_NULL)
-               panic("zero_page: physaddr out of range");
-
-       pmap_zero_page_assembler(p);
-}
-
-/*
- * pmap_copy_page(src, dst)
- *
- * pmap_copy_page copies the specified (machine independent)
- * page from physical address src to physical address dst.
- *
- * We need to invalidate the cache for address dst before
- * we do the copy. Apparently there won't be any mappings
- * to the dst address normally.
- */
-void
-pmap_copy_page(
-       vm_offset_t src,
-       vm_offset_t dst)
-{
-       register struct phys_entry *pp;
-
-       if (pmdebug & (PDB_USER|PDB_COPY))
-               kprintf("pmap_copy_page(spa=%x, dpa=%x)\n", src, dst);
-       if (pmdebug & PDB_COPY)
-               kprintf("pmap_copy_page: phys_copy(%x, %x, %x)\n",
-                      src, dst, PAGE_SIZE);
-
-       pmap_copy_page_assembler(src, dst);
-}
-#endif /* DEBUG */
 
 /*
  * pmap_pageable(pmap, s, e, pageable)
@@ -1745,10 +1393,10 @@ pmap_copy_page(
  */
 void
 pmap_pageable(
-       pmap_t          pmap,
-       vm_offset_t     start,
-       vm_offset_t     end,
-       boolean_t       pageable)
+       __unused pmap_t                         pmap,
+       __unused vm_map_offset_t        start,
+       __unused vm_map_offset_t        end,
+       __unused boolean_t                      pageable)
 {
 
        return;                                                                                         /* This is not used... */
@@ -1756,13 +1404,13 @@ pmap_pageable(
 }
 /*
  *     Routine:        pmap_change_wiring
- *     NOTE USED ANYMORE.
+ *     NOT USED ANYMORE.
  */
 void
 pmap_change_wiring(
-       register pmap_t pmap,
-       vm_offset_t     va,
-       boolean_t       wired)
+       __unused pmap_t                         pmap,
+       __unused vm_map_offset_t        va,
+       __unused boolean_t                      wired)
 {
        return;                                                                                         /* This is not used... */
 }
@@ -1773,47 +1421,52 @@ pmap_change_wiring(
  *     virtual address range determined by [s, e] and pmap,
  *     s and e must be on machine independent page boundaries and
  *     s must be less than or equal to e.
+ *
+ *  Note that this function will not descend nested pmaps.
  */
 void
 pmap_modify_pages(
             pmap_t pmap,
-            vm_offset_t sva, 
-            vm_offset_t eva)
+            vm_map_offset_t sva, 
+            vm_map_offset_t eva)
 {
        spl_t           spl;
-       mapping         *mp;
+       mapping_t       *mp;
+       ppnum_t         pa;
+       addr64_t                va, endva;
+       unsigned int    savetype;
 
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00010, (unsigned int)pmap, (unsigned int)(sva|((eva-sva)>>12)));  /* (TEST/DEBUG) */
-#endif
-
-#if DEBUG
-       if (pmdebug & PDB_USER) kprintf("pmap_modify_pages(pmap=%x, sva=%x, eva=%x)\n", pmap, sva, eva);
-#endif
-
-       if (pmap == PMAP_NULL) return;                                          /* If no pmap, can't do it... */
-
-       debugLog2(56, sva, eva);                                                        /* Log pmap_map call */
+       if (pmap == PMAP_NULL) return;                                  /* If no pmap, can't do it... */
+       
+       va = sva & -4096;                                                               /* Round to page */
+       endva = eva & -4096;                                                    /* Round to page */
 
-       spl=splhigh();                                                                          /* Don't bother me */
+       while (va < endva) {                                                    /* Walk through all pages */
 
-       for ( ; sva < eva; sva += PAGE_SIZE) {                          /* Cycle through the whole range */     
-               mp = hw_lock_phys_vir(pmap->space, sva);                /* Lock the physical entry for this mapping */
-               if(mp) {                                                                                /* Did we find one? */
-                       if((unsigned int)mp&1) {                                        /* Did the lock on the phys entry time out? */
-                               panic("pmap_modify_pages: timeout obtaining lock on physical entry\n"); /* Scream bloody murder! */
-                               continue;
-                       }
-                       mp = hw_cpv(mp);                                                        /* Convert to virtual addressing */                             
-                       if(!mp->physent) continue;                                      /* No physical entry means an I/O page, we can't set attributes */
-                       mapping_set_mod(mp->physent);                           /* Set the modfied bit for this page */
-                       hw_unlock_bit((unsigned int *)&mp->physent->phys_link, PHYS_LOCK);      /* Unlock the physical entry */
+               spl = splhigh();                                                        /* We can't allow any loss of control here */
+       
+               mp = mapping_find(pmap, (addr64_t)va, &va, 0);  /* Find the mapping for this address */
+               
+               if(!mp) {                                                                       /* Is the page mapped? */
+                       splx(spl);                                                              /* Page not mapped, restore interruptions */
+                       if((va == 0) || (va >= endva)) break;   /* We are done if there are no more or we hit the end... */
+                       continue;                                                               /* We are not done and there is more to check... */
                }
+               
+               savetype = mp->mpFlags & mpType;                        /* Remember the type */
+               pa = mp->mpPAddr;                                                       /* Remember ppage because mapping may vanish after drop call */
+       
+               mapping_drop_busy(mp);                                          /* We have everything we need from the mapping */
+       
+               splx(spl);                                                                      /* Restore 'rupts */
+       
+               if(savetype != mpNormal) continue;                      /* Can't mess around with these guys... */      
+               
+               mapping_set_mod(pa);                                            /* Set the modfied bit for this page */
+               
+               if(va == 0) break;                                                      /* We hit the end of the pmap, might as well leave now... */
        }
-       splx(spl);                                                                                      /* Restore the interrupt level */
-
-       debugLog2(57, 0, 0);                                                            /* Log pmap_map call */
-       return;                                                                                         /* Leave... */
+       return;                                                                                 /* Leave... */
 }
 
 /*
@@ -1824,38 +1477,11 @@ pmap_modify_pages(
  *     independant page boundary.
  */
 void
-pmap_clear_modify(vm_offset_t pa)
+pmap_clear_modify(ppnum_t pa)
 {
-       register struct phys_entry      *pp;
-       spl_t           spl;
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00011, (unsigned int)pa, 0);                      /* (TEST/DEBUG) */
-#endif
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_clear_modify(pa=%x)\n", pa);
-#endif
-
-       pp = pmap_find_physentry(pa);                                           /* Find the physent for this page */
-       if (pp == PHYS_NULL) return;                                            /* If there isn't one, just leave... */
-
-       debugLog2(58, pa, 0);                                   /* Log pmap_map call */
 
-       spl=splhigh();                                                                          /* Don't bother me */
+       mapping_clr_mod(pa);                            /* Clear all change bits for physical page */
 
-       if(!hw_lock_bit((unsigned int *)&pp->phys_link, PHYS_LOCK, LockTimeOut)) {      /* Try to get the lock on the physical entry */
-               panic("pmap_clear_modify: Timeout getting lock on physent at %08X\n", pp);      /* Arrrgghhhh! */
-               splx(spl);                                                                              /* Restore 'rupts */
-               return;                                                                                 /* Should die before here */
-       }
-
-       mapping_clr_mod(pp);                                                            /* Clear all change bits for physical page */
-
-       hw_unlock_bit((unsigned int *)&pp->phys_link, PHYS_LOCK);       /* Unlock the physical entry */
-       splx(spl);                                                                                      /* Restore the interrupt level */
-
-       debugLog2(59, 0, 0);                                    /* Log pmap_map call */
 }
 
 /*
@@ -1864,42 +1490,10 @@ pmap_clear_modify(vm_offset_t pa)
  *     since the last call to pmap_clear_modify().
  */
 boolean_t
-pmap_is_modified(register vm_offset_t pa)
+pmap_is_modified(register ppnum_t pa)
 {
-       register struct phys_entry      *pp;
-       spl_t           spl;
-       boolean_t       ret;
-
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00012, (unsigned int)pa, 0);                      /* (TEST/DEBUG) */
-#endif
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_is_modified(pa=%x)\n", pa);
-#endif
-
-       pp = pmap_find_physentry(pa);                                           /* Find the physent for this page */
-       if (pp == PHYS_NULL) return(FALSE);                                     /* Just indicate not set... */
+       return mapping_tst_mod(pa);     /* Check for modified */
        
-       debugLog2(60, pa, 0);                                   /* Log pmap_map call */
-
-       spl=splhigh();                                                                          /* Don't bother me */
-
-       if(!hw_lock_bit((unsigned int *)&pp->phys_link, PHYS_LOCK, LockTimeOut)) {      /* Try to get the lock on the physical entry */
-               panic("pmap_is_modified: Timeout getting lock on physent at %08X\n", pp);       /* Arrrgghhhh! */
-               splx(spl);                                                                              /* Restore 'rupts */
-               return 0;                                                                               /* Should die before here */
-       }
-       
-       ret = mapping_tst_mod(pp);                                                      /* Check for modified */
-       
-       hw_unlock_bit((unsigned int *)&pp->phys_link, PHYS_LOCK);       /* Unlock the physical entry */
-       splx(spl);                                                                                      /* Restore the interrupt level */
-       
-       debugLog2(61, ret, 0);                                  /* Log pmap_map call */
-
-       return ret;                                                                             
 }
 
 /*
@@ -1909,31 +1503,9 @@ pmap_is_modified(register vm_offset_t pa)
  *
  */
 void
-pmap_clear_reference(vm_offset_t pa)
+pmap_clear_reference(ppnum_t pa)
 {
-       register struct phys_entry      *pp;
-       spl_t           spl;
-
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00013, (unsigned int)pa, 0);                              /* (TEST/DEBUG) */
-#endif
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_clear_reference(pa=%x)\n", pa);
-#endif
-
-       pp = pmap_find_physentry(pa);                                                   /* Find the physent for this page */
-       if (pp == PHYS_NULL) return;                                                    /* If there isn't one, just leave... */
-
-       debugLog2(62, pa, 0);                                   /* Log pmap_map call */
-
-       spl=splhigh();                                                                                  /* Don't bother me */
-       mapping_clr_ref(pp);                                                                    /* Clear all reference bits for physical page */
-       splx(spl);                                                                                              /* Restore the interrupt level */
-
-       debugLog2(63, 0, 0);                                    /* Log pmap_map call */
-
+       mapping_clr_ref(pa);                    /* Check for modified */
 }
 
 /*
@@ -1942,50 +1514,59 @@ pmap_clear_reference(vm_offset_t pa)
  *     since the last call to pmap_clear_reference().
  */
 boolean_t
-pmap_is_referenced(vm_offset_t pa)
+pmap_is_referenced(ppnum_t pa)
 {
-       register struct phys_entry      *pp;
-       spl_t           spl;
-       boolean_t       ret;
+       return mapping_tst_ref(pa);     /* Check for referenced */
+}
 
+/*
+ * pmap_get_refmod(phys)
+ *  returns the referenced and modified bits of the specified
+ *  physical page.
+ */
+unsigned int
+pmap_get_refmod(ppnum_t pa)
+{
+       return (mapping_tst_refmod(pa));
+}
 
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00014, (unsigned int)pa, 0);                      /* (TEST/DEBUG) */
-#endif
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_is_referenced(pa=%x)\n", pa);
-#endif
+/*
+ * pmap_clear_refmod(phys, mask)
+ *  clears the referenced and modified bits as specified by the mask
+ *  of the specified physical page.
+ */
+void
+pmap_clear_refmod(ppnum_t pa, unsigned int mask)
+{
+       mapping_clr_refmod(pa, mask);
+}
 
-       pp = pmap_find_physentry(pa);                                           /* Find the physent for this page */
-       if (pp == PHYS_NULL) return(FALSE);                                     /* Just indicate not set... */
-       
-       debugLog2(64, pa, 0);                                   /* Log pmap_map call */
+/*
+ * pmap_eligible_for_execute(ppnum_t pa)
+ *     return true if physical address is eligible to contain executable code;
+ *  otherwise, return false
+ */
+boolean_t
+pmap_eligible_for_execute(ppnum_t pa)
+{
+       phys_entry_t *physent;
+       unsigned int  pindex;
 
-       spl=splhigh();                                                                          /* Don't bother me */
+       physent = mapping_phys_lookup(pa, &pindex);                             /* Get physical entry */
 
-       if(!hw_lock_bit((unsigned int *)&pp->phys_link, PHYS_LOCK, LockTimeOut)) {      /* Try to get the lock on the physical entry */
-               panic("pmap_is_referenced: Timeout getting lock on physent at %08X\n", pp);     /* Arrrgghhhh! */
-               splx(spl);                                                                              /* Restore 'rupts */
-               return 0;                                                                               /* Should die before here */
-       }
-       
-       ret = mapping_tst_ref(pp);                                                      /* Check for referenced */
-       
-       hw_unlock_bit((unsigned int *)&pp->phys_link, PHYS_LOCK);       /* Unlock the physical entry */
-       splx(spl);                                                                                      /* Restore the interrupt level */
-       
-       debugLog2(65, ret, 0);                                  /* Log pmap_map call */
+       if((!physent) || (physent->ppLink & ppG))
+               return 0;                                                                                       /* If there is no physical entry or marked guarded,
+                                                                      the entry is not eligible for execute */
 
-       return ret;                                                                             
+       return 1;                                                                                               /* Otherwise, entry is eligible for execute */
 }
 
 #if    MACH_VM_DEBUG
 int
 pmap_list_resident_pages(
-       register pmap_t         pmap,
-       register vm_offset_t    *listp,
-       register int            space)
+       __unused pmap_t         pmap,
+       __unused vm_offset_t    *listp,
+       __unused int            space)
 {
        return 0;
 }
@@ -2003,56 +1584,36 @@ pmap_copy_part_page(
        vm_offset_t     dst_offset,
        vm_size_t       len)
 {
-       register struct phys_entry *pp_src, *pp_dst;
-       spl_t   s;
-
-
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00019, (unsigned int)src+src_offset, (unsigned int)dst+dst_offset);       /* (TEST/DEBUG) */
-       dbgTrace(0xF1D04019, (unsigned int)len, 0);                     /* (TEST/DEBUG) */
-#endif
-       s = splhigh();
+       addr64_t fsrc, fdst;
 
-        assert(((dst & PAGE_MASK)+dst_offset+len) <= PAGE_SIZE);
-        assert(((src & PAGE_MASK)+src_offset+len) <= PAGE_SIZE);
+       assert(((dst <<12) & PAGE_MASK+dst_offset+len) <= PAGE_SIZE);
+       assert(((src <<12) & PAGE_MASK+src_offset+len) <= PAGE_SIZE);
 
-       /*
-        * Since the source and destination are physical addresses, 
-        * turn off data translation to perform a  bcopy() in bcopy_phys().
-        */
-       phys_copy((vm_offset_t) src+src_offset,
-                 (vm_offset_t) dst+dst_offset, len);
+       fsrc = ((addr64_t)src << 12) + src_offset;
+       fdst = ((addr64_t)dst << 12) + dst_offset;
 
-       splx(s);
+       phys_copy(fsrc, fdst, len);                                                             /* Copy the stuff physically */
 }
 
 void
 pmap_zero_part_page(
-       vm_offset_t     p,
-       vm_offset_t     offset,
-       vm_size_t       len)
+       __unused vm_offset_t            p,
+       __unused vm_offset_t    offset,
+       __unused vm_size_t      len)
 {
     panic("pmap_zero_part_page");
 }
 
-boolean_t pmap_verify_free(vm_offset_t pa) {
+boolean_t pmap_verify_free(ppnum_t pa) {
 
        struct phys_entry       *pp;
+       unsigned int pindex;
 
-#if PMAP_LOWTRACE
-       dbgTrace(0xF1D00007, (unsigned int)pa, 0);                              /* (TEST/DEBUG) */
-#endif
-
-#if DEBUG
-       if (pmdebug & PDB_USER)
-               kprintf("pmap_verify_free(pa=%x)\n", pa);
-#endif
+       pp = mapping_phys_lookup(pa, &pindex);  /* Get physical entry */
+       if (pp == 0) return FALSE;                                      /* If there isn't one, show no mapping... */
 
-       if (!pmap_initialized) return(TRUE);
-
-       pp = pmap_find_physentry(pa);                                                   /* Look up the physical entry */
-       if (pp == PHYS_NULL) return FALSE;                                              /* If there isn't one, show no mapping... */
-       return ((mapping *)((unsigned int)pp->phys_link & ~PHYS_FLAGS) == MAPPING_NULL);        /* Otherwise, return TRUE if mapping exists... */
+       if(pp->ppLink & ~(ppLock | ppFlags)) return FALSE;      /* We have at least one mapping */
+       return TRUE;                                                            /* No mappings */
 }
 
 
@@ -2060,15 +1621,8 @@ boolean_t pmap_verify_free(vm_offset_t pa) {
 
 void pmap_switch(pmap_t map)
 {
-       unsigned int i;
-
-#if DEBUG
-       if (watchacts & WA_PCB) {
-               kprintf("Switching to map at 0x%08x, space=%d\n",
-                      map,map->space);
-       }
-#endif /* DEBUG */
-
+       hw_blow_seg(lowGlo.lgUMWvaddr);                                 /* Blow off the first segment */
+       hw_blow_seg(lowGlo.lgUMWvaddr + 0x10000000ULL); /* Blow off the second segment */
 
 /* when changing to kernel space, don't bother
  * doing anything, the kernel is mapped from here already.
@@ -2082,113 +1636,112 @@ void pmap_switch(pmap_t map)
 }
 
 /*
- *     kern_return_t pmap_nest(grand, subord, vaddr, size)
+ *     kern_return_t pmap_nest(grand, subord, vstart, size)
  *
  *     grand  = the pmap that we will nest subord into
  *     subord = the pmap that goes into the grand
- *     vaddr  = start of range in pmap to be inserted
- *     size   = size of range in pmap to be inserted
+ *     vstart  = start of range in pmap to be inserted
+ *     nstart  = start of range in pmap nested pmap
+ *     size   = Size of nest area (up to 2TB)
  *
  *     Inserts a pmap into another.  This is used to implement shared segments.
  *     On the current PPC processors, this is limited to segment (256MB) aligned
  *     segment sized ranges.
+ *
+ *     We actually kinda allow recursive nests.  The gating factor is that we do not allow 
+ *     nesting on top of something that is already mapped, i.e., the range must be empty.
+ *
+ *     Note that we depend upon higher level VM locks to insure that things don't change while
+ *     we are doing this.  For example, VM should not be doing any pmap enters while it is nesting
+ *     or do 2 nests at once.
  */
 
-kern_return_t pmap_nest(pmap_t grand, pmap_t subord, vm_offset_t vaddr, vm_size_t size) {
-                       
-       unsigned int oflags, seg, grandr;
-       int i;
+kern_return_t pmap_nest(pmap_t grand, pmap_t subord, addr64_t vstart, addr64_t nstart, uint64_t size) {
+               
+       addr64_t vend, colladdr;
+       unsigned int msize;
+       int nlists;
+       mapping_t *mp;
        
-       if(size != 0x10000000) return KERN_INVALID_VALUE;       /* We can only do this for 256MB for now */
-       if(vaddr & 0x0FFFFFFF) return KERN_INVALID_VALUE;       /* We can only do this aligned to 256MB */
-
-       while(1) {                                                                      /* Test and set the subordinate flag */
-               oflags = subord->vflags & ~pmapAltSeg;  /* Get old unset value */
-               if(subord->vflags & pmapAltSeg) {               /* Are trying to nest one already nested? */
-                       panic("pmap_nest: Attempt to nest an already nested pmap\n");
-               }
-               if(hw_compare_and_store(oflags, oflags | pmapSubord, &subord->vflags)) break;   /* Done if we got it set */
-       }
+       if(size & 0x0FFFFFFFULL) return KERN_INVALID_VALUE;     /* We can only do this for multiples of 256MB */
+       if((size >> 25) > 65536)  return KERN_INVALID_VALUE;    /* Max size we can nest is 2TB */
+       if(vstart & 0x0FFFFFFFULL) return KERN_INVALID_VALUE;   /* We can only do this aligned to 256MB */
+       if(nstart & 0x0FFFFFFFULL) return KERN_INVALID_VALUE;   /* We can only do this aligned to 256MB */
        
-       simple_lock(&grand->lock);                                      /* Lock the superior pmap */
-
-       if(grand->vflags & pmapSubord) {                        /* Are we only one level deep? */
-               simple_unlock(&grand->lock);                    /* Unlock the superior pmap */
-               panic("pmap_nest: Attempt to nest into subordinate pmap\n");
-               return KERN_FAILURE;                                    /* Shame on you */
-       }
-
-       seg = vaddr >> 28;                                                      /* Isolate the segment number */
-       if((0x00008000 >> seg) & grand->vflags) {       /* See if it is already in use */
-               simple_unlock(&grand->lock);                    /* Unlock the superior pmap */
-               panic("pmap_nest: Attempt to nest into already nested segment\n");
-               return KERN_FAILURE;                                    /* Shame on you */
+       if(size == 0) {                                                         /*      Is the size valid? */
+               panic("pmap_nest: size is invalid - %016llX\n", size);
        }
        
-       grand->pmapPmaps[seg] = subord;                         /* Set the pointer to the subordinate */
-       grand->pmapSegs[seg] = SEG_REG_PROT | (seg << 20) | subord->space;              /* Set the vsid to the subordinate's vsid */
-       grand->vflags |= (0x00008000 >> seg);           /* Set in-use bit */
+       msize = (size >> 25) - 1;                                                       /* Change size to blocks of 32MB */
+       
+       nlists = mapSetLists(grand);                                            /* Set number of lists this will be on */
 
-       grandr = (unsigned int)grand ^ grand->pmapvr;   /* Get real address of the grand pmap */
+       mp = mapping_alloc(nlists);                                                     /* Get a spare mapping block */
        
-       simple_unlock(&grand->lock);                            /* Unlock the grand pmap */
+       mp->mpFlags = 0x01000000 | mpNest | mpPerm | mpBSu | nlists;    /* Make this a permanent nested pmap with a 32MB basic size unit */
+                                                                                                               /* Set the flags. Make sure busy count is 1 */
+       mp->mpSpace = subord->space;                                            /* Set the address space/pmap lookup ID */
+       mp->u.mpBSize = msize;                                                          /* Set the size */
+       mp->mpPte = 0;                                                                          /* Set the PTE invalid */
+       mp->mpPAddr = 0;                                                                        /* Set the physical page number */
+       mp->mpVAddr = vstart;                                                           /* Set the address */
+       mp->mpNestReloc = nstart - vstart;                                      /* Set grand to nested vaddr relocation value */
        
-
-/*
- *     Note that the following will force the segment registers to be reloaded following
- *     the next interrupt on all processors if they are using the pmap we just changed.
- *
- */
-
-
-       for(i=0; i < real_ncpus; i++) {                                                 /* Cycle through processors */
-               (void)hw_compare_and_store((unsigned int)grandr, 0, &per_proc_info[i].Lastpmap);        /* Clear if ours */
+       colladdr = hw_add_map(grand, mp);                                       /* Go add the mapping to the pmap */
+       
+       if(colladdr) {                                                                          /* Did it collide? */
+               vend = vstart + size - 4096;                                    /* Point to the last page we would cover in nest */     
+               panic("pmap_nest: attempt to nest into a non-empty range - pmap = %08X, start = %016llX, end = %016llX\n",
+                       grand, vstart, vend);
        }
-               
-       return KERN_SUCCESS;                                            /* Bye, bye, butterfly... */
+       
+       return KERN_SUCCESS;
 }
 
-
 /*
- *     kern_return_t pmap_unnest(grand, vaddr, size)
+ *     kern_return_t pmap_unnest(grand, vaddr)
  *
  *     grand  = the pmap that we will nest subord into
- *     vaddr  = start of range in pmap to be inserted
- *     size   = size of range in pmap to be inserted
+ *     vaddr  = start of range in pmap to be unnested
  *
  *     Removes a pmap from another.  This is used to implement shared segments.
  *     On the current PPC processors, this is limited to segment (256MB) aligned
  *     segment sized ranges.
  */
 
-kern_return_t pmap_unnest(pmap_t grand, vm_offset_t vaddr, vm_size_t size) {
+kern_return_t pmap_unnest(pmap_t grand, addr64_t vaddr) {
                        
-       unsigned int oflags, seg, grandr, tstamp;
-       int i, tcpu, mycpu;
+       unsigned int tstamp, i, mycpu;
+       addr64_t nextva;
+       spl_t s;
+       mapping_t *mp;
                
-       if(size != 0x10000000) return KERN_INVALID_VALUE;       /* We can only do this for 256MB for now */
-       if(vaddr & 0x0FFFFFFF) return KERN_INVALID_VALUE;       /* We can only do this aligned to 256MB */
-
-       simple_lock(&grand->lock);                                      /* Lock the superior pmap */
-       disable_preemption();                                           /* It's all for me! */
-
-       seg = vaddr >> 28;                                                      /* Isolate the segment number */
-       if(!((0x00008000 >> seg) & grand->vflags)) {    /* See if it is already in use */
-               enable_preemption();                                    /* Ok, your turn */
-               simple_unlock(&grand->lock);                    /* Unlock the superior pmap */
-               panic("pmap_unnest: Attempt to unnest an unnested segment\n");
-               return KERN_FAILURE;                                    /* Shame on you */
+       s = splhigh();                                                                          /* Make sure interruptions are disabled */
+
+       mp = mapping_find(grand, vaddr, &nextva, 0);            /* Find the nested map */
+
+       if(((unsigned int)mp & mapRetCode) != mapRtOK) {        /* See if it was even nested */
+               panic("pmap_unnest: Attempt to unnest an unnested segment - va = %016llX\n", vaddr);
+       }
+
+       if((mp->mpFlags & mpType) != mpNest) {                          /* Did we find something other than a nest? */
+               panic("pmap_unnest: Attempt to unnest something that is not a nest - va = %016llX\n", vaddr);
        }
        
-       grand->pmapPmaps[seg] = (pmap_t)0;                      /* Clear the pointer to the subordinate */
-       grand->pmapSegs[seg] = grand->space;            /* Set the pointer to the subordinate's vsid */
-       grand->pmapSegs[seg] = SEG_REG_PROT | (seg << 20) | grand->space;               /* Set the vsid to the grand's vsid */
-       grand->vflags &= ~(0x00008000 >> seg);          /* Clear in-use bit */
+       if(mp->mpVAddr != vaddr) {                                                      /* Make sure the address is the same */
+               panic("pmap_unnest: Attempt to unnest something that is not at start of nest - va = %016llX\n", vaddr);
+       }
 
-       grandr = (unsigned int)grand ^ grand->pmapvr;   /* Get real address of the grand pmap */
+       (void)hw_atomic_and(&mp->mpFlags, ~mpPerm);                     /* Show that this mapping is now removable */
        
-       simple_unlock(&grand->lock);                            /* Unlock the superior pmap */
+       mapping_drop_busy(mp);                                                          /* Go ahead and release the mapping now */
 
+       splx(s);                                                                                        /* Restore 'rupts */
+               
+       (void)mapping_remove(grand, vaddr);                                     /* Toss the nested pmap mapping */
+       
+       invalidateSegs(grand);                                                          /* Invalidate the pmap segment cache */
+       
 /*
  *     Note that the following will force the segment registers to be reloaded 
  *     on all processors (if they are using the pmap we just changed) before returning.
@@ -2204,69 +1757,302 @@ kern_return_t pmap_unnest(pmap_t grand, vm_offset_t vaddr, vm_size_t size) {
  */
 
 
-       mycpu = cpu_number();                                           /* Who am I? Am I just a dream? */
-       for(i=0; i < real_ncpus; i++) {                         /* Cycle through processors */
-               if(hw_compare_and_store((unsigned int)grandr, 0, &per_proc_info[i].Lastpmap)) { /* Clear if ours and kick the other guy if he was using it */
-                       if(i == mycpu) continue;                                /* Don't diddle ourselves */
-                       tstamp = per_proc_info[i].ruptStamp[1]; /* Save the processor's last interrupt time stamp */
-                       if(cpu_signal(i, SIGPwake, 0, 0) != KERN_SUCCESS) {     /* Make sure we see the pmap change */
-                               continue;
-                       }
-                       if(!hw_cpu_wcng(&per_proc_info[i].ruptStamp[1], tstamp, LockTimeOut)) { /* Wait for the other processors to enter debug */
-                               panic("pmap_unnest: Other processor (%d) did not see interruption request\n", i);
+       for(i=0; i < real_ncpus; i++) {                                         /* Cycle through processors */
+               disable_preemption();
+               mycpu = cpu_number();                                                           /* Who am I? Am I just a dream? */
+               if((unsigned int)grand == PerProcTable[i].ppe_vaddr->ppUserPmapVirt) {  /* Is this guy using the changed pmap? */
+                       
+                       PerProcTable[i].ppe_vaddr->ppInvSeg = 1;        /* Show that we need to invalidate the segments */
+                       
+                       if(i != mycpu) {
+               
+                               tstamp = PerProcTable[i].ppe_vaddr->ruptStamp[1];               /* Save the processor's last interrupt time stamp */
+                               if(cpu_signal(i, SIGPcpureq, CPRQsegload, 0) == KERN_SUCCESS) { /* Make sure we see the pmap change */
+                                       if(!hw_cpu_wcng(&PerProcTable[i].ppe_vaddr->ruptStamp[1], tstamp, LockTimeOut)) {       /* Wait for the other processors to enter debug */
+                                               panic("pmap_unnest: Other processor (%d) did not see interruption request\n", i);
+                                       }
+                               }
                        }
                }
+               enable_preemption();
        }
 
-       enable_preemption();                                            /* Others can run now */
-       return KERN_SUCCESS;                                            /* Bye, bye, butterfly... */
+       return KERN_SUCCESS;                                                            /* Bye, bye, butterfly... */
 }
 
 
-void pmap_ver(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) {
+/*
+ *     void MapUserMemoryWindowInit(void)
+ *
+ *     Initialize anything we need to in order to map user address space slices into
+ *     the kernel.  Primarily used for copy in/out.
+ *
+ *     Currently we only support one 512MB slot for this purpose.  There are two special
+ *     mappings defined for the purpose: the special pmap nest, and linkage mapping.
+ *
+ *     The special pmap nest (which is allocated in this function) is used as a place holder
+ *     in the kernel's pmap search list. It is 512MB long and covers the address range
+ *     starting at lgUMWvaddr.  It points to no actual memory and when the fault handler 
+ *     hits in it, it knows to look in the per_proc and start using the linkage
+ *     mapping contained therin.
+ *
+ *     The linkage mapping is used to glue the user address space slice into the 
+ *     kernel.  It contains the relocation information used to transform the faulting
+ *     kernel address into the user address space.  It also provides the link to the
+ *     user's pmap.  This is pointed to by the per_proc and is switched in and out
+ *     whenever there is a context switch.
+ *
+ */
 
-       int cnt, i, j, k;
-       vm_offset_t xx;
+void MapUserMemoryWindowInit(void) {
+               
+       addr64_t colladdr;
+       int nlists;
+       mapping_t *mp;
+       
+       nlists = mapSetLists(kernel_pmap);                                      /* Set number of lists this will be on */
+       
+       mp = mapping_alloc(nlists);                                                     /* Get a spare mapping block */
+
+       mp->mpFlags = 0x01000000 | mpLinkage | mpPerm | mpBSu | nlists; /* Make this a permanent nested pmap with a 32MB basic size unit */
+                                                                                                               /* Set the flags. Make sure busy count is 1 */
+       mp->mpSpace = kernel_pmap->space;                                       /* Set the address space/pmap lookup ID */
+       mp->u.mpBSize = 15;                                                                     /* Set the size to 2 segments in 32MB chunks - 1 */
+       mp->mpPte = 0;                                                                          /* Means nothing */
+       mp->mpPAddr = 0;                                                                        /* Means nothing */
+       mp->mpVAddr = lowGlo.lgUMWvaddr;                                        /* Set the address range we cover */
+       mp->mpNestReloc = 0;                                                            /* Means nothing */
        
-       if(!pmap) return;
+       colladdr = hw_add_map(kernel_pmap, mp);                         /* Go add the mapping to the pmap */
+       
+       if(colladdr) {                                                                          /* Did it collide? */
+               panic("MapUserMemoryWindowInit: MapUserMemoryWindow range already mapped\n");
+       }
+       
+       return;
+}
 
-       sva = trunc_page(sva);
-       eva = trunc_page(eva);
+/*
+ *     addr64_t MapUserMemoryWindow(vm_map_t map, vm_offset_t va, size)
+ *
+ *     map  = the vm_map that we are mapping into the kernel
+ *     va = start of the address range we are mapping
+ *     Note that we do not test validty, we chose to trust our fellows...
+ *
+ *     Maps a 512M slice of a user address space into a predefined kernel range
+ *     on a per-thread basis. We map only the first 256M segment, allowing the
+ *  second 256M segment to fault in as needed. This allows our clients to access
+ *  an arbitrarily aligned operand up to 256M in size.
+ *
+ *  In the future, the restriction of a predefined range may be loosened.
+ *
+ *     Builds the proper linkage map to map the user range
+ *  We will round this down to the previous segment boundary and calculate
+ *     the relocation to the kernel slot
+ *
+ *     We always make a segment table entry here if we need to.  This is mainly because of
+ *     copyin/out and if we don't, there will be multiple segment faults for
+ *     each system call.  I have seen upwards of 30000 per second.
+ *
+ *     We do check, however, to see if the slice is already mapped and if so,
+ *     we just exit.  This is done for performance reasons.  It was found that 
+ *     there was a considerable boost in copyin/out performance if we did not
+ *     invalidate the segment at ReleaseUserAddressSpace time, so we dumped the
+ *     restriction that you had to bracket MapUserMemoryWindow.  Further, there 
+ *     is a yet further boost if you didn't need to map it each time.  The theory
+ *     behind this is that many times copies are to or from the same segment and
+ *     done multiple times within the same system call.  To take advantage of that,
+ *     we check umwSpace and umwRelo to see if we've already got it.  
+ *
+ *     We also need to half-invalidate the slice when we context switch or go
+ *     back to user state.  A half-invalidate does not clear the actual mapping,
+ *     but it does force the MapUserMemoryWindow function to reload the segment
+ *     register/SLBE.  If this is not done, we can end up some pretty severe
+ *     performance penalties. If we map a slice, and the cached space/relocation is
+ *     the same, we won't reload the segment registers.  Howver, since we ran someone else,
+ *     our SR is cleared and we will take a fault.  This is reasonable if we block
+ *     while copying (e.g., we took a page fault), but it is not reasonable when we 
+ *     just start.  For this reason, we half-invalidate to make sure that the SR is
+ *     explicitly reloaded.
+ *      
+ *     Note that we do not go to the trouble of making a pmap segment cache
+ *     entry for these guys because they are very short term -- 99.99% of the time
+ *     they will be unmapped before the next context switch.
+ *
+ */
+
+addr64_t MapUserMemoryWindow(
+       vm_map_t map,
+       addr64_t va) {
+               
+       addr64_t baddrs, reladd;
+       thread_t thread;
+       mapping_t *mp;
        
-       for(i = 0; i < (pmapUsageMask + 1); i++) {      /* Step through them all */
-               if((pmap->pmapUsage[i]) > 8192) {               /* See if this is a sane number */
-                       panic("pmap_ver: pmapUsage slot for %08X has invalid count (%d) for pmap %08X\n",
-                               i * pmapUsageSize, pmap->pmapUsage[i], pmap);
-               }
+       baddrs = va & 0xFFFFFFFFF0000000ULL;                            /* Isolate the segment */
+       thread = current_thread();                                                      /* Remember our activation */
+
+       reladd = baddrs - lowGlo.lgUMWvaddr;                            /* Get the relocation from user to kernel */
+       
+       if((thread->machine.umwSpace == map->pmap->space) && (thread->machine.umwRelo == reladd)) {     /* Already mapped? */
+               return ((va & 0x0FFFFFFFULL) | lowGlo.lgUMWvaddr);      /* Pass back the kernel address we are to use */
        }
-       j = 0;
-       while(1) {                                                                      /* Try multiple times */
-               cnt = 0;
-               for(i = 0; i < (pmapUsageMask + 1); i++) {      /* Step through them all */
-                       cnt = cnt + pmap->pmapUsage[i];                 /* Sum all slots */
+
+       disable_preemption();                                                           /* Don't move... */     
+       
+       mp = (mapping_t *)&(getPerProc()->ppUMWmp);                     /* Make up for C */
+       thread->machine.umwRelo = reladd;                                       /* Relocation from user to kernel */
+       mp->mpNestReloc = reladd;                                                       /* Relocation from user to kernel */
+       
+       thread->machine.umwSpace = map->pmap->space;            /* Set the address space/pmap lookup ID */
+       mp->mpSpace = map->pmap->space;                                         /* Set the address space/pmap lookup ID */
+       
+/*
+ *     Here we make an assumption that we are going to be using the base pmap's address space.
+ *     If we are wrong, and that would be very, very, very rare, the fault handler will fix us up.
+ */ 
+
+       hw_map_seg(map->pmap,  lowGlo.lgUMWvaddr, baddrs);      /* Make the entry for the first segment */
+
+       enable_preemption();                                                            /* Let's move */
+       return ((va & 0x0FFFFFFFULL) | lowGlo.lgUMWvaddr);      /* Pass back the kernel address we are to use */
+}
+
+
+/*
+ *     kern_return_t pmap_boot_map(size)
+ *
+ *     size   = size of virtual address range to be mapped
+ *
+ *     This function is used to assign a range of virtual addresses before VM in 
+ *     initialized.  It starts at VM_MAX_KERNEL_ADDRESS and works downward.
+ *     The variable vm_last_addr contains the current highest possible VM
+ *     assignable address.  It is a panic to attempt to call this after VM has
+ *     started up.  The only problem is, is that we may not have the serial or
+ *     framebuffer mapped, so we'll never know we died.........
+ */
+
+vm_offset_t pmap_boot_map(vm_size_t size) {
+                       
+       if(kernel_map != VM_MAP_NULL) {                         /* Has VM already started? */
+               panic("pmap_boot_map: VM started\n");
+       }
+       
+       size = round_page(size);                                        /* Make sure this is in pages */
+       vm_last_addr = vm_last_addr - size;                     /* Allocate the memory */
+       return (vm_last_addr + 1);                                      /* Return the vaddr we just allocated */
+
+}
+
+
+/*
+ *     void pmap_init_sharedpage(void);
+ *
+ *     Hack map for the 64-bit commpage
+ */
+
+void pmap_init_sharedpage(vm_offset_t cpg){
+       
+       addr64_t cva, cpoff;
+       ppnum_t cpphys;
+       
+       sharedPmap = pmap_create(0);                            /* Get a pmap to hold the common segment */
+       if(!sharedPmap) {                                                       /* Check for errors */
+               panic("pmap_init_sharedpage: couldn't make sharedPmap\n");
+       }
+
+       for(cpoff = 0; cpoff < _COMM_PAGE_AREA_USED; cpoff += 4096) {   /* Step along now */
+       
+               cpphys = pmap_find_phys(kernel_pmap, (addr64_t)cpg + cpoff);
+               if(!cpphys) {
+                       panic("pmap_init_sharedpage: compage %08X not mapped in kernel\n", cpg + cpoff);
                }
-               if(cnt == pmap->stats.resident_count) break;    /* We're ok if we match... */
                
-               j++;
-               for(i = 0; i < 100000; i++) {
-                       k = j + i;
-               }
-               if(j >= 10) {
-                       panic("pmap_ver: pmapUsage total (%d) does not match resident count (%d) for pmap %08X\n",
-                               cnt, pmap->stats.resident_count, pmap);
+               cva = mapping_make(sharedPmap, (addr64_t)((uint32_t)_COMM_PAGE_BASE_ADDRESS) + cpoff,
+                       cpphys, mmFlgPerm, 1, VM_PROT_READ);    /* Map the page read only */
+               if(cva) {                                                               /* Check for errors */
+                       panic("pmap_init_sharedpage: couldn't map commpage page - cva = %016llX\n", cva);
                }
+       
        }
+               
+       return;
+}
+
+
+/*
+ *     void pmap_map_sharedpage(pmap_t pmap);
+ *
+ *     Maps the last segment in a 64-bit address space
+ *
+ *     
+ */
+
+void pmap_map_sharedpage(task_t task, pmap_t pmap){
        
-       for(xx = sva; xx < eva; xx += PAGE_SIZE) {      /* See if any slots not clear */
-               if(pmap_extract(pmap, xx)) {
-                       panic("pmap_ver: range (%08X to %08X) not empty at %08X for pmap %08X\n",
-                               sva, eva, xx, pmap);
+       kern_return_t ret;
+
+       if(task_has_64BitAddr(task) || _cpu_capabilities & k64Bit) {    /* Should we map the 64-bit page -1? */
+               ret = pmap_nest(pmap, sharedPmap, 0xFFFFFFFFF0000000ULL, 0x00000000F0000000ULL,
+                       0x0000000010000000ULL);                         /* Nest the highest possible segment to map comm page */
+               if(ret != KERN_SUCCESS) {                               /* Did it work? */
+                       panic("pmap_map_sharedpage: couldn't nest shared page - ret = %08X\n", ret);
                }
        }
+
+       return;
 }
 
 
+/*
+ *     void pmap_unmap_sharedpage(pmap_t pmap);
+ *
+ *     Unmaps the last segment in a 64-bit address space
+ *
+ */
 
+void pmap_unmap_sharedpage(pmap_t pmap){
+       
+       kern_return_t ret;
+       mapping_t *mp;
+       boolean_t inter;
+       int gotnest;
+       addr64_t nextva;
 
+       if(BootProcInfo.pf.Available & pf64Bit) {               /* Are we on a 64-bit machine? */
+               
+               inter  = ml_set_interrupts_enabled(FALSE);      /* Disable interruptions for now */
+               mp = hw_find_map(pmap, 0xFFFFFFFFF0000000ULL, &nextva); /* Find the mapping for this address */
+               if((unsigned int)mp == mapRtBadLk) {            /* Did we lock up ok? */
+                       panic("pmap_unmap_sharedpage: mapping lock failure - rc = %08X, pmap = %08X\n", mp, pmap);      /* Die... */
+               }
+               
+               gotnest = 0;                                                            /* Assume nothing here */
+               if(mp) {
+                       gotnest = ((mp->mpFlags & mpType) == mpNest);
+                                                                                                       /* Remember if we have a nest here */
+                       mapping_drop_busy(mp);                                  /* We have everything we need from the mapping */
+               }
+               ml_set_interrupts_enabled(inter);                       /* Put interrupts back to what they were */
+               
+               if(!gotnest) return;                                            /* Leave if there isn't any nesting here */
+               
+               ret = pmap_unnest(pmap, 0xFFFFFFFFF0000000ULL); /* Unnest the max 64-bit page */
+               
+               if(ret != KERN_SUCCESS) {                                       /* Did it work? */
+                       panic("pmap_unmap_sharedpage: couldn't unnest shared page - ret = %08X\n", ret);
+               }
+       }
+       
+       return;
+}
 
 
+/* temporary workaround */
+boolean_t
+coredumpok(
+       __unused vm_map_t map,
+       __unused vm_offset_t va)
+{
+       return TRUE;
+}