]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/pmap.h
xnu-792.24.17.tar.gz
[apple/xnu.git] / osfmk / i386 / pmap.h
index dc6f40bceb3f823a774280be6a81255bcc81f054..7faa4124cb69df304ebcf04918181f5574973b3b 100644 (file)
@@ -1,6 +1,5 @@
-
 /*
- * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
  *
  * @APPLE_LICENSE_HEADER_START@
  * 
@@ -59,7 +58,7 @@
  *
  *     Machine-dependent structures for the physical map module.
  */
-#ifdef KERNEL_PRIVATE
+
 #ifndef        _PMAP_MACHINE_
 #define _PMAP_MACHINE_ 1
 
 #include <kern/kern_types.h>
 #include <kern/thread.h>
 #include <kern/lock.h>
-
-#include <i386/mp.h>
-#include <i386/proc_reg.h>
+#define PMAP_QUEUE 1
+#ifdef PMAP_QUEUE
+#include <kern/queue.h>
+#endif
 
 /*
  *     Define the generic in terms of the specific
  *     i386/i486/i860 Page Table Entry
  */
 
+#ifdef PAE
+typedef uint64_t        pdpt_entry_t;
+typedef uint64_t       pt_entry_t;
+typedef uint64_t        pd_entry_t;
+typedef uint64_t        pmap_paddr_t;
+#else
+typedef uint32_t       pt_entry_t;
+typedef uint32_t        pd_entry_t;
+typedef uint32_t        pmap_paddr_t;
+#endif
+
+#define PT_ENTRY_NULL  ((pt_entry_t *) 0)
+#define PD_ENTRY_NULL  ((pt_entry_t *) 0)
+
 #endif /* ASSEMBLER */
 
+#ifdef PAE
 #define NPGPTD          4
 #define PDESHIFT        21
 #define PTEMASK         0x1ff
 #define PTEINDX         3
-
+#else
+#define NPGPTD          1
+#define PDESHIFT        22
+#define PTEMASK         0x3ff
+#define PTEINDX         2
+#endif
 #define PTESHIFT        12
 
 #define PDESIZE                sizeof(pd_entry_t) /* for assembly files */
 #define PTESIZE                sizeof(pt_entry_t) /* for assembly files */
 
 #define INTEL_OFFMASK  (I386_PGBYTES - 1)
-#define PG_FRAME        0x000FFFFFFFFFF000ULL
+#define PG_FRAME        (~((pmap_paddr_t)PAGE_MASK))
 #define NPTEPG          (PAGE_SIZE/(sizeof (pt_entry_t)))
-#define NPTDPG          (PAGE_SIZE/(sizeof (pd_entry_t)))
 
 #define NBPTD           (NPGPTD << PAGE_SHIFT)
 #define NPDEPTD         (NBPTD / (sizeof (pd_entry_t)))
 #define NBPDE           (1 << PDESHIFT)
 #define PDEMASK         (NBPDE - 1)
 
- /* cleanly define parameters for all the page table levels */
-typedef uint64_t        pml4_entry_t;
-#define NPML4PG         (PAGE_SIZE/(sizeof (pml4_entry_t)))
-#define PML4SHIFT       39
-#define PML4PGSHIFT     9
-#define NBPML4          (1ULL << PML4SHIFT)
-#define PML4MASK        (NBPML4-1)
-#define PML4_ENTRY_NULL ((pml4_entry_t *) 0)
-
-typedef uint64_t        pdpt_entry_t;
-#define NPDPTPG         (PAGE_SIZE/(sizeof (pdpt_entry_t)))
-#define PDPTSHIFT       30
-#define PDPTPGSHIFT     9
-#define NBPDPT          (1 << PDPTSHIFT)
-#define PDPTMASK        (NBPDPT-1)
-#define PDPT_ENTRY_NULL ((pdpt_entry_t *) 0)
-
-typedef uint64_t        pd_entry_t;
-#define NPDPG           (PAGE_SIZE/(sizeof (pd_entry_t)))
-#define PDSHIFT         21
-#define PDPGSHIFT       9
-#define NBPD            (1 << PDSHIFT)
-#define PDMASK          (NBPD-1)
-#define PD_ENTRY_NULL   ((pd_entry_t *) 0)
-
-typedef uint64_t        pt_entry_t;
-#define NPTPG           (PAGE_SIZE/(sizeof (pt_entry_t)))
-#define PTSHIFT         12
-#define PTPGSHIFT       9
-#define NBPT            (1 << PTSHIFT)
-#define PTMASK          (NBPT-1)
-#define PT_ENTRY_NULL  ((pt_entry_t *) 0)
-
-typedef uint64_t  pmap_paddr_t;
-
-/*
- * Atomic 64-bit store of a page table entry.
- */
-static inline void
-pmap_store_pte(pt_entry_t *entryp, pt_entry_t value)
-{
-       /*
-        * Load the new value into %ecx:%ebx
-        * Load the old value into %edx:%eax
-        * Compare-exchange-8bytes at address entryp (loaded in %edi)
-        * If the compare succeeds, the new value will have been stored.
-        * Otherwise, the old value changed and reloaded, so try again.
-        */
-       asm volatile(
-               "       movl    (%0), %%eax     \n\t"
-               "       movl    4(%0), %%edx    \n\t"
-               "1:                             \n\t"
-               "       cmpxchg8b (%0)          \n\t"
-               "       jnz 1b"
-               :
-               : "D" (entryp),
-                 "b" ((uint32_t)value),
-                 "c" ((uint32_t)(value >> 32))
-               : "eax", "edx", "memory");
-}
-
-/* in 64 bit spaces, the number of each type of page in the page tables */
-#define NPML4PGS        (1ULL * (PAGE_SIZE/(sizeof (pml4_entry_t))))
-#define NPDPTPGS        (NPML4PGS * (PAGE_SIZE/(sizeof (pdpt_entry_t))))
-#define NPDEPGS         (NPDPTPGS * (PAGE_SIZE/(sizeof (pd_entry_t))))
-#define NPTEPGS         (NPDEPGS * (PAGE_SIZE/(sizeof (pt_entry_t))))
-
-/*
- * The 64-bit kernel is remapped in uber-space which is at the base
- * the highest 4th-level directory (KERNEL_UBER_PML4_INDEX). That is,
- * 512GB from the top of virtual space (or zero).
- */
-#define KERNEL_UBER_PML4_INDEX 511
-#define KERNEL_UBER_BASE       (0ULL - NBPML4)
-#define KERNEL_UBER_BASE_HI32  ((uint32_t)(KERNEL_UBER_BASE >> 32))
-
 #define        VM_WIMG_COPYBACK        VM_MEM_COHERENT
 #define        VM_WIMG_DEFAULT         VM_MEM_COHERENT
 /* ?? intel ?? */
@@ -205,83 +148,56 @@ pmap_store_pte(pt_entry_t *entryp, pt_entry_t value)
 /* write combining mode, aka store gather */
 #define VM_WIMG_WCOMB          (VM_MEM_NOT_CACHEABLE | VM_MEM_COHERENT) 
 
-/*
- * Pte related macros
- */
-#define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDESHIFT)|((pti)<<PTESHIFT)))
-#define VADDR64(pmi, pdi, pti) ((vm_offset_t)(((pmi)<<PLM4SHIFT))((pdi)<<PDESHIFT)|((pti)<<PTESHIFT))
-
 /*
  * Size of Kernel address space.  This is the number of page table pages
  * (4MB each) to use for the kernel.  256 pages == 1 Gigabyte.
  * This **MUST** be a multiple of 4 (eg: 252, 256, 260, etc).
  */
 #ifndef KVA_PAGES
-#define KVA_PAGES      1024
+#define KVA_PAGES      256
 #endif
 
+/*
+ * Pte related macros
+ */
+#define VADDR(pdi, pti) ((vm_offset_t)(((pdi)<<PDESHIFT)|((pti)<<PTESHIFT)))
+
 #ifndef NKPT
+#ifdef PAE
 #define        NKPT            500     /* actual number of kernel page tables */
+#else
+#define        NKPT            32      /* initial number of kernel page tables */
+#endif
 #endif
 #ifndef NKPDE
 #define NKPDE  (KVA_PAGES - 1) /* addressable number of page tables/pde's */
 #endif
 
-
-enum high_cpu_types {
-  HIGH_CPU_ISS0,
-  HIGH_CPU_ISS1,
-  HIGH_CPU_DESC,
-  HIGH_CPU_LDT_BEGIN,
-  HIGH_CPU_LDT_END = HIGH_CPU_LDT_BEGIN + (LDTSZ / 512) - 1,
-  HIGH_CPU_END
-};
-
-enum  high_fixed_addresses {
-  HIGH_FIXED_TRAMPS,  /* must be first */
-  HIGH_FIXED_TRAMPS_END,
-  HIGH_FIXED_GDT,
-  HIGH_FIXED_IDT,
-  HIGH_FIXED_LDT_BEGIN,
-  HIGH_FIXED_LDT_END = HIGH_FIXED_LDT_BEGIN + (LDTSZ / 512) - 1,
-  HIGH_FIXED_KTSS,
-  HIGH_FIXED_DFTSS,
-  HIGH_FIXED_DBTSS,
-  HIGH_FIXED_CPUS_BEGIN,
-  HIGH_FIXED_CPUS_END = HIGH_FIXED_CPUS_BEGIN + (HIGH_CPU_END * MAX_CPUS) - 1,
-};
-
-
-/* XXX64  below PTDI values need cleanup */
 /*
  * The *PTDI values control the layout of virtual memory
  *
  */
-#define        KPTDI           (0x000)/* start of kernel virtual pde's */
+#ifdef PAE
+#define        KPTDI           (0x600)/* start of kernel virtual pde's */
 #define        PTDPTDI         (0x7F4) /* ptd entry that points to ptd! */
 #define        APTDPTDI        (0x7F8) /* alt ptd entry that points to APTD */
-#define        UMAXPTDI        (0x7F8) /* ptd entry for user space end */
+#define        UMAXPTDI        (0x5FC) /* ptd entry for user space end */
 #define        UMAXPTEOFF      (NPTEPG)        /* pte entry for user space end */
+#else
+#define        KPTDI           (0x300)/* start of kernel virtual pde's */
+#define        PTDPTDI         (0x3FD) /* ptd entry that points to ptd! */
+#define        APTDPTDI        (0x3FE) /* alt ptd entry that points to APTD */
+#define        UMAXPTDI        (0x2FF) /* ptd entry for user space end */
+#define        UMAXPTEOFF      (NPTEPG)        /* pte entry for user space end */
+#endif
 
 #define KERNBASE       VADDR(KPTDI,0)
 
-/*
- *     Convert address offset to directory address
- *     containing the page table pointer - legacy
- */
-/*#define pmap_pde(m,v) (&((m)->dirbase[(vm_offset_t)(v) >> PDESHIFT]))*/
-
-#define HIGH_MEM_BASE  ((uint32_t)( -NBPDE) )  /* shared gdt etc seg addr */ /* XXX64 ?? */
-#define pmap_index_to_virt(x)  (HIGH_MEM_BASE | ((unsigned)(x) << PAGE_SHIFT))
-
 /*
  *     Convert address offset to page descriptor index
  */
-#define pdenum(pmap, a) (((vm_offset_t)(a) >> PDESHIFT) & PDEMASK)
+#define pdenum(pmap, a) (((a) >> PDESHIFT) & PDEMASK)
 
-#define pdeidx(pmap, a)    (((a) >> PDSHIFT)   & ((1ULL<<(48 - PDSHIFT)) -1))
-#define pdptidx(pmap, a)   (((a) >> PDPTSHIFT) & ((1ULL<<(48 - PDPTSHIFT)) -1))
-#define pml4idx(pmap, a)   (((a) >> PML4SHIFT) & ((1ULL<<(48 - PML4SHIFT)) -1))
 
 /*
  *     Convert page descriptor index to user virtual address
@@ -291,7 +207,7 @@ enum  high_fixed_addresses {
 /*
  *     Convert address offset to page table index
  */
-#define ptenum(a)      (((vm_offset_t)(a) >> PTESHIFT) & PTEMASK)
+#define ptenum(a)      (((a) >> PTESHIFT) & PTEMASK)
 
 /*
  *     Hardware pte bit definitions (to be used directly on the ptes
@@ -309,22 +225,13 @@ enum  high_fixed_addresses {
 #define INTEL_PTE_PS            0x00000080
 #define INTEL_PTE_GLOBAL        0x00000100
 #define INTEL_PTE_WIRED                0x00000200
-#define INTEL_PTE_PFN          PG_FRAME
+#define INTEL_PTE_PFN          /*0xFFFFF000*/ (~0xFFF)
 #define INTEL_PTE_PTA          0x00000080
 
-#define INTEL_PTE_NX           (1ULL << 63)
-
-#define INTEL_PTE_INVALID       0
-
 #define        pa_to_pte(a)            ((a) & INTEL_PTE_PFN) /* XXX */
 #define        pte_to_pa(p)            ((p) & INTEL_PTE_PFN) /* XXX */
 #define        pte_increment_pa(p)     ((p) += INTEL_OFFMASK+1)
 
-#define pte_kernel_rw(p)          ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_RW))
-#define pte_kernel_ro(p)          ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID))
-#define pte_user_rw(p)            ((pt_entry)t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_USER|INTEL_PTE_RW))
-#define pte_user_ro(p)            ((pt_entry_t)(pa_to_pte(p) | INTEL_PTE_VALID|INTEL_PTE_USER))
-
 #define PMAP_DEFAULT_CACHE     0
 #define PMAP_INHIBIT_CACHE     1
 #define PMAP_GUARDED_CACHE     2
@@ -345,14 +252,9 @@ extern pt_entry_t PTmap[], APTmap[], Upte;
 extern pd_entry_t PTD[], APTD[], PTDpde[], APTDpde[], Upde;
 
 extern pd_entry_t *IdlePTD;    /* physical address of "Idle" state directory */
+#ifdef PAE
 extern pdpt_entry_t *IdlePDPT;
-
-extern pmap_paddr_t lo_kernel_cr3;
-
-extern pml4_entry_t *IdlePML4;
-extern pdpt_entry_t *IdlePDPT64;
-extern addr64_t     kernel64_cr3;
-extern boolean_t    no_shared_cr3;
+#endif
 
 /*
  * virtual address to page table entry and
@@ -360,7 +262,7 @@ extern boolean_t    no_shared_cr3;
  * Note: these work recursively, thus vtopte of a pte will give
  * the corresponding pde that in turn maps it.
  */
-#define        vtopte(va)      (PTmap + i386_btop((vm_offset_t)va))
+#define        vtopte(va)      (PTmap + i386_btop(va))
 
 
 typedef        volatile long   cpu_set;        /* set of CPUs - must be <= 32 */
@@ -379,50 +281,48 @@ struct md_page {
  */
 
 struct pmap {
-        pd_entry_t      *dirbase;        /* page directory pointer */
-       pmap_paddr_t    pdirbase;        /* phys. address of dirbase */
-        vm_object_t     pm_obj;         /* object to hold pde's */
+#ifdef PMAP_QUEUE
+  queue_head_t     pmap_link;   /* unordered queue of in use pmaps */
+#endif
+       pd_entry_t      *dirbase;       /* page directory pointer register */
+       pd_entry_t      *pdirbase;      /* phys. address of dirbase */
+        vm_object_t     pm_obj;         /* object to hold pte's */
        int             ref_count;      /* reference count */
-        int            nx_enabled;
-        boolean_t       pm_64bit;
-        boolean_t       pm_kernel_cr3;
        decl_simple_lock_data(,lock)    /* lock on map */
        struct pmap_statistics  stats;  /* map statistics */
+       cpu_set         cpus_using;     /* bitmap of cpus using pmap */
+#ifdef PAE
        vm_offset_t     pm_hold;        /* true pdpt zalloc addr */
-       pmap_paddr_t    pm_cr3;         /* physical addr */
-        pdpt_entry_t    *pm_pdpt;       /* KVA of 3rd level page */
-       pml4_entry_t    *pm_pml4;       /* VKA of top level */
-       vm_object_t     pm_obj_pdpt;    /* holds pdpt pages */
-       vm_object_t     pm_obj_pml4;    /* holds pml4 pages */
-       vm_object_t     pm_obj_top;     /* holds single top level page */
+       pdpt_entry_t    *pm_pdpt;       /* KVA of pg dir ptr table */
+       vm_offset_t     pm_ppdpt;       /* phy addr pdpt
+                                          should really be 32/64 bit */
+#endif
 };
 
-
-#define PMAP_PDPT_FIRST_WINDOW 0
-#define PMAP_PDPT_NWINDOWS 4
-#define PMAP_PDE_FIRST_WINDOW (PMAP_PDPT_NWINDOWS)
-#define PMAP_PDE_NWINDOWS 4
-#define PMAP_PTE_FIRST_WINDOW (PMAP_PDE_FIRST_WINDOW + PMAP_PDE_NWINDOWS)
-#define PMAP_PTE_NWINDOWS 4
-
-#define PMAP_NWINDOWS_FIRSTFREE (PMAP_PTE_FIRST_WINDOW + PMAP_PTE_NWINDOWS)
-#define PMAP_WINDOW_SIZE 8
-#define PMAP_NWINDOWS (PMAP_NWINDOWS_FIRSTFREE + PMAP_WINDOW_SIZE)
-
+#define PMAP_NWINDOWS  4
 typedef struct {
        pt_entry_t      *prv_CMAP;
        caddr_t         prv_CADDR;
 } mapwindow_t;
 
 typedef struct cpu_pmap {
-        int                     pdpt_window_index;
-        int                     pde_window_index;
-        int                     pte_window_index;
        mapwindow_t             mapwindow[PMAP_NWINDOWS];
+       struct pmap             *real_pmap;
+       struct pmap_update_list *update_list;
+       volatile boolean_t      update_needed;
 } cpu_pmap_t;
 
-
-extern mapwindow_t *pmap_get_mapwindow(pt_entry_t pentry);
+/*
+ * Should be rewritten in asm anyway.
+ */
+#define CM1 (current_cpu_datap()->cpu_pmap->mapwindow[0].prv_CMAP)
+#define CM2 (current_cpu_datap()->cpu_pmap->mapwindow[1].prv_CMAP)
+#define CM3 (current_cpu_datap()->cpu_pmap->mapwindow[2].prv_CMAP)
+#define CM4 (current_cpu_datap()->cpu_pmap->mapwindow[3].prv_CMAP)
+#define CA1 (current_cpu_datap()->cpu_pmap->mapwindow[0].prv_CADDR)
+#define CA2 (current_cpu_datap()->cpu_pmap->mapwindow[1].prv_CADDR)
+#define CA3 (current_cpu_datap()->cpu_pmap->mapwindow[2].prv_CADDR)
+#define CA4 (current_cpu_datap()->cpu_pmap->mapwindow[3].prv_CADDR)
 
 typedef struct pmap_memory_regions {
   ppnum_t base;
@@ -434,62 +334,117 @@ typedef struct pmap_memory_regions {
 unsigned pmap_memory_region_count;
 unsigned pmap_memory_region_current;
 
-#define PMAP_MEMORY_REGIONS_SIZE 128
+#define PMAP_MEMORY_REGIONS_SIZE 32
 
 extern pmap_memory_region_t pmap_memory_regions[];
 
-static inline void set_dirbase(pmap_t tpmap, __unused int tcpu) {
-       current_cpu_datap()->cpu_task_cr3 = (pmap_paddr_t)((tpmap)->pm_cr3);
-       current_cpu_datap()->cpu_task_map = tpmap->pm_64bit ? TASK_MAP_64BIT : TASK_MAP_32BIT;
+/* 
+ * Optimization avoiding some TLB flushes when switching to
+ * kernel-loaded threads.  This is effective only for i386:
+ * Since user task, kernel task and kernel loaded tasks share the
+ * same virtual space (with appropriate protections), any pmap
+ * allows mapping kernel and kernel loaded tasks. 
+ *
+ * The idea is to avoid switching to another pmap unnecessarily when
+ * switching to a kernel-loaded task, or when switching to the kernel
+ * itself.
+ *
+ * We store the pmap we are really using (from which we fetched the
+ * dirbase value) in current_cpu_datap()->cpu_pmap.real_pmap.
+ *
+ * Invariant:
+ * current_pmap() == current_cpu_datap()->cpu_pmap.real_pmap ||
+ *                     current_pmap() == kernel_pmap.
+ */
+#define        PMAP_REAL(my_cpu)       (cpu_datap(my_cpu)->cpu_pmap->real_pmap)
+
+#include <i386/proc_reg.h>
+/*
+ * If switching to the kernel pmap, don't incur the TLB cost of switching
+ * to its page tables, since all maps include the kernel map as a subset.
+ * Simply record that this CPU is logically on the kernel pmap (see
+ * pmap_destroy).
+ * 
+ * Similarly, if switching to a pmap (other than kernel_pmap that is already
+ * in use, don't do anything to the hardware, to avoid a TLB flush.
+ */
+
+#define        PMAP_CPU_SET(pmap, my_cpu) i_bit_set(my_cpu, &((pmap)->cpus_using))
+#define        PMAP_CPU_CLR(pmap, my_cpu) i_bit_clear(my_cpu, &((pmap)->cpus_using))
+
+#ifdef PAE
+#define PDIRBASE pm_ppdpt
+#else
+#define PDIRBASE pdirbase
+#endif
+#define        set_dirbase(mypmap, my_cpu) {                                   \
+       struct pmap     **ppmap = &PMAP_REAL(my_cpu);                   \
+       pmap_paddr_t    pdirbase = (pmap_paddr_t)((mypmap)->PDIRBASE);          \
+                                                                       \
+       if (*ppmap == (pmap_paddr_t)NULL) {                             \
+               *ppmap = (mypmap);                                      \
+               PMAP_CPU_SET((mypmap), my_cpu);                         \
+               set_cr3(pdirbase);                                      \
+       } else if ((mypmap) != kernel_pmap && (mypmap) != *ppmap ) {    \
+               if (*ppmap != kernel_pmap)                              \
+                       PMAP_CPU_CLR(*ppmap, my_cpu);                   \
+               *ppmap = (mypmap);                                      \
+               PMAP_CPU_SET((mypmap), my_cpu);                         \
+               set_cr3(pdirbase);                                      \
+       }                                                               \
+       assert((mypmap) == *ppmap || (mypmap) == kernel_pmap);          \
 }
 
+/*
+ *     List of cpus that are actively using mapped memory.  Any
+ *     pmap update operation must wait for all cpus in this list.
+ *     Update operations must still be queued to cpus not in this
+ *     list.
+ */
+extern cpu_set         cpus_active;
+
+/*
+ *     List of cpus that are idle, but still operating, and will want
+ *     to see any kernel pmap updates when they become active.
+ */
+extern cpu_set         cpus_idle;
+
+
+#define cpu_update_needed(cpu) cpu_datap(cpu)->cpu_pmap->update_needed
+#define cpu_update_list(cpu)   cpu_datap(cpu)->cpu_pmap->update_list
+
 /*
  *     External declarations for PMAP_ACTIVATE.
  */
 
-extern void            process_pmap_updates(void);
+extern void            process_pmap_updates(struct pmap *pmap);
 extern void            pmap_update_interrupt(void);
 
 /*
  *     Machine dependent routines that are used only for i386/i486/i860.
  */
 
-extern addr64_t                (kvtophys)(
+extern vm_offset_t     (kvtophys)(
                                vm_offset_t     addr);
 
 extern pt_entry_t      *pmap_pte(
                                struct pmap     *pmap,
-                               vm_map_offset_t addr);
-
-extern pd_entry_t      *pmap_pde(
-                               struct pmap     *pmap,
-                               vm_map_offset_t addr);
-
-extern pd_entry_t      *pmap64_pde(
-                               struct pmap     *pmap,
-                               vm_map_offset_t addr);
-
-extern pdpt_entry_t    *pmap64_pdpt(
-                               struct pmap     *pmap,
-                               vm_map_offset_t addr);
+                               vm_offset_t     addr);
 
 extern vm_offset_t     pmap_map(
                                vm_offset_t     virt,
-                               vm_map_offset_t start,
-                               vm_map_offset_t end,
-                               vm_prot_t       prot,
-                               unsigned int    flags);
+                               vm_offset_t     start,
+                               vm_offset_t     end,
+                               vm_prot_t       prot);
 
 extern vm_offset_t     pmap_map_bd(
                                vm_offset_t     virt,
-                               vm_map_offset_t start,
-                               vm_map_offset_t end,
-                               vm_prot_t       prot,
-                               unsigned int    flags);
+                               vm_offset_t     start,
+                               vm_offset_t     end,
+                               vm_prot_t       prot);
 
 extern void            pmap_bootstrap(
-                               vm_offset_t     load_start,
-                               boolean_t       IA32e);
+                               vm_offset_t     load_start);
 
 extern boolean_t       pmap_valid_page(
                                ppnum_t pn);
@@ -499,28 +454,14 @@ extern int                pmap_list_resident_pages(
                                vm_offset_t     *listp,
                                int             space);
 
-extern void             pmap_commpage32_init(
+extern void             pmap_commpage_init(
                                           vm_offset_t kernel,
                                           vm_offset_t user,
                                           int count);
-extern void             pmap_commpage64_init(
-                                          vm_offset_t  kernel,
-                                          vm_map_offset_t user,
-                                          int count);
-
 extern struct cpu_pmap *pmap_cpu_alloc(
                                boolean_t       is_boot_cpu);
 extern void            pmap_cpu_free(
                                struct cpu_pmap *cp);
-
-extern 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);
                                
 extern void invalidate_icache(vm_offset_t addr, unsigned cnt, int phys);
 extern void flush_dcache(vm_offset_t addr, unsigned count, int phys);
@@ -528,24 +469,6 @@ extern ppnum_t          pmap_find_phys(pmap_t map, addr64_t va);
 extern void pmap_sync_page_data_phys(ppnum_t pa);
 extern void pmap_sync_page_attributes_phys(ppnum_t pa);
 
-extern kern_return_t pmap_nest(pmap_t grand, pmap_t subord, addr64_t vstart, addr64_t nstart, uint64_t size);
-extern kern_return_t pmap_unnest(pmap_t grand, addr64_t vaddr);
-extern void pmap_map_sharedpage(task_t task, pmap_t pmap);
-extern void pmap_unmap_sharedpage(pmap_t pmap);
-extern void pmap_disable_NX(pmap_t pmap);
-extern void pmap_set_4GB_pagezero(pmap_t pmap);
-extern void pmap_clear_4GB_pagezero(pmap_t pmap);
-extern void pmap_load_kernel_cr3(void);
-extern vm_offset_t pmap_cpu_high_map_vaddr(int, enum high_cpu_types);
-extern vm_offset_t pmap_high_map_vaddr(enum high_cpu_types);
-extern vm_offset_t pmap_high_map(pt_entry_t, enum high_cpu_types);
-extern vm_offset_t pmap_cpu_high_shared_remap(int, enum high_cpu_types, vm_offset_t, int);
-extern vm_offset_t pmap_high_shared_remap(enum high_fixed_addresses, vm_offset_t, int);
-
-extern void pt_fake_zone_info(int *, vm_size_t *, vm_size_t *, vm_size_t *, vm_size_t *, int *, int *);
-
-
-
 /*
  *     Macros for speed.
  */
@@ -560,109 +483,125 @@ extern void pt_fake_zone_info(int *, vm_size_t *, vm_size_t *, vm_size_t *, vm_s
 #undef PMAP_DEACTIVATE_USER
 #endif
 
+/*
+ *     For multiple CPUS, PMAP_ACTIVATE and PMAP_DEACTIVATE must manage
+ *     fields to control TLB invalidation on other CPUS.
+ */
 
-#define PMAP_ACTIVATE_KERNEL(my_cpu)  {                                        \
-        spl_t          spl;                                            \
-                                                                        \
-        spl = splhigh();                                               \
-       if (current_cpu_datap()->cpu_tlb_invalid)                       \
-               process_pmap_updates();                                 \
-        splx(spl);                                                     \
+#define        PMAP_ACTIVATE_KERNEL(my_cpu)    {                               \
+                                                                       \
+       /*                                                              \
+        *      Let pmap updates proceed while we wait for this pmap.   \
+        */                                                             \
+       i_bit_clear((my_cpu), &cpus_active);                            \
+                                                                       \
+       /*                                                              \
+        *      Lock the pmap to put this cpu in its active set.        \
+        *      Wait for updates here.                                  \
+        */                                                             \
+       simple_lock(&kernel_pmap->lock);                                \
+                                                                       \
+       /*                                                              \
+        *      Process invalidate requests for the kernel pmap.        \
+        */                                                             \
+       if (cpu_update_needed(my_cpu))                                  \
+           process_pmap_updates(kernel_pmap);                          \
+                                                                       \
+       /*                                                              \
+        *      Mark that this cpu is using the pmap.                   \
+        */                                                             \
+       i_bit_set((my_cpu), &kernel_pmap->cpus_using);                  \
+                                                                       \
+       /*                                                              \
+        *      Mark this cpu active - IPL will be lowered by           \
+        *      load_context().                                         \
+        */                                                             \
+       i_bit_set((my_cpu), &cpus_active);                              \
+                                                                       \
+       simple_unlock(&kernel_pmap->lock);                              \
 }
 
-#define PMAP_DEACTIVATE_KERNEL(my_cpu) {                               \
-        spl_t          spl;                                            \
-                                                                        \
-        spl = splhigh();                                               \
-       process_pmap_updates();                                         \
-        splx(spl);                                                     \
+#define        PMAP_DEACTIVATE_KERNEL(my_cpu)  {                               \
+       /*                                                              \
+        *      Mark pmap no longer in use by this cpu even if          \
+        *      pmap is locked against updates.                         \
+        */                                                             \
+       i_bit_clear((my_cpu), &kernel_pmap->cpus_using);                \
+       i_bit_clear((my_cpu), &cpus_active);                            \
+       PMAP_REAL(my_cpu) = NULL;                                       \
 }
 
-
 #define PMAP_ACTIVATE_MAP(map, my_cpu) {                               \
        register pmap_t         tpmap;                                  \
-                                                                        \
-        tpmap = vm_map_pmap(map);                                      \
-        set_dirbase(tpmap, my_cpu);                                    \
+                                                                       \
+       tpmap = vm_map_pmap(map);                                       \
+       if (tpmap == kernel_pmap) {                                     \
+           /*                                                          \
+            *  If this is the kernel pmap, switch to its page tables.  \
+            */                                                         \
+           set_dirbase(kernel_pmap, my_cpu);                           \
+       }                                                               \
+       else {                                                          \
+           /*                                                          \
+            *  Let pmap updates proceed while we wait for this pmap.   \
+            */                                                         \
+           i_bit_clear((my_cpu), &cpus_active);                        \
+                                                                       \
+           /*                                                          \
+            *  Lock the pmap to put this cpu in its active set.        \
+            *  Wait for updates here.                                  \
+            */                                                         \
+           simple_lock(&tpmap->lock);                                  \
+                                                                       \
+           /*                                                          \
+            *  No need to invalidate the TLB - the entire user pmap    \
+            *  will be invalidated by reloading dirbase.               \
+            */                                                         \
+           set_dirbase(tpmap, my_cpu);                                 \
+                                                                       \
+           /*                                                          \
+            *  Mark this cpu active - IPL will be lowered by           \
+            *  load_context().                                         \
+            */                                                         \
+           i_bit_set((my_cpu), &cpus_active);                          \
+                                                                       \
+           simple_unlock(&tpmap->lock);                                \
+       }                                                               \
 }
 
 #define PMAP_DEACTIVATE_MAP(map, my_cpu)
 
-#define PMAP_ACTIVATE_USER(th, my_cpu)  {                              \
-        spl_t          spl;                                            \
-                                                                        \
-        spl = splhigh();                                               \
-        PMAP_ACTIVATE_MAP(th->map, my_cpu)                             \
-        splx(spl);                                                     \
+#define PMAP_ACTIVATE_USER(th, my_cpu)                               \
+       spl_t           spl;                                            \
+                                                                       \
+       spl = splhigh();                                                        \
+       PMAP_ACTIVATE_MAP(th->map, my_cpu)                              \
+       splx(spl);                                                      \
 }
 
 #define PMAP_DEACTIVATE_USER(th, my_cpu)
 
-
 #define        PMAP_SWITCH_CONTEXT(old_th, new_th, my_cpu) {                   \
        spl_t           spl;                                            \
-       pt_entry_t      *kpdp;                                          \
-       pt_entry_t      *updp;                                          \
-        int            i;                                              \
-        int            need_flush;                                     \
-                                                                        \
-        need_flush = 0;                                                        \
-        spl = splhigh();                                               \
+                                                                       \
        if (old_th->map != new_th->map) {                               \
+               spl = splhigh();                                                \
                PMAP_DEACTIVATE_MAP(old_th->map, my_cpu);               \
                PMAP_ACTIVATE_MAP(new_th->map, my_cpu);                 \
+               splx(spl);                                              \
        }                                                               \
-        kpdp = current_cpu_datap()->cpu_copywindow_pdp;                        \
-        for (i = 0; i < NCOPY_WINDOWS; i++) {                          \
-                if (new_th->machine.copy_window[i].user_base != (user_addr_t)-1) {     \
-                       updp = pmap_pde(new_th->map->pmap,              \
-                              new_th->machine.copy_window[i].user_base);\
-                       *kpdp = updp ? *updp : 0;                       \
-                }                                                      \
-                kpdp++;                                                        \
-        }                                                              \
-       splx(spl);                                                      \
-        if (new_th->machine.copyio_state == WINDOWS_OPENED)            \
-                need_flush = 1;                                                \
-        else                                                           \
-                new_th->machine.copyio_state = WINDOWS_DIRTY;          \
-        if (new_th->machine.physwindow_pte) {                          \
-                *(current_cpu_datap()->cpu_physwindow_ptep) =          \
-                                new_th->machine.physwindow_pte;        \
-                if (need_flush == 0)                                   \
-                        invlpg((uintptr_t)current_cpu_datap()->cpu_physwindow_base);\
-        }                                                              \
-        if (need_flush)                                                        \
-                flush_tlb();                                           \
 }
 
 #define        PMAP_SWITCH_USER(th, new_map, my_cpu) {                         \
        spl_t           spl;                                            \
                                                                        \
-       spl = splhigh();                                                \
+       spl = splhigh();                                                        \
        PMAP_DEACTIVATE_MAP(th->map, my_cpu);                           \
        th->map = new_map;                                              \
        PMAP_ACTIVATE_MAP(th->map, my_cpu);                             \
        splx(spl);                                                      \
-        inval_copy_windows(th);                                                \
 }
 
-/*
- * Marking the current cpu's cr3 inactive is achieved by setting its lsb.
- * Marking the current cpu's cr3 active once more involves clearng this bit.
- * Note that valid page tables are page-aligned and so the bottom 12 bits
- * are noramlly zero.
- * We can only mark the current cpu active/inactive but we can test any cpu.
- */
-#define CPU_CR3_MARK_INACTIVE()                                                \
-       current_cpu_datap()->cpu_active_cr3 |= 1
-
-#define CPU_CR3_MARK_ACTIVE()                                          \
-       current_cpu_datap()->cpu_active_cr3 &= ~1
-
-#define CPU_CR3_IS_ACTIVE(cpu)                                         \
-       ((cpu_datap(cpu)->cpu_active_cr3 & 1) == 0)
-
 #define MARK_CPU_IDLE(my_cpu)  {                                       \
        /*                                                              \
         *      Mark this cpu idle, and remove it from the active set,  \
@@ -672,16 +611,13 @@ extern void pt_fake_zone_info(int *, vm_size_t *, vm_size_t *, vm_size_t *, vm_s
         *      becomes active.                                         \
         */                                                             \
        int     s = splhigh();                                          \
-       if (!cpu_mode_is64bit() || no_shared_cr3)                       \
-               process_pmap_updates();                                 \
-       else                                                            \
-               pmap_load_kernel_cr3();                                 \
-       CPU_CR3_MARK_INACTIVE();                                        \
-       __asm__ volatile("mfence");                                     \
+       i_bit_set((my_cpu), &cpus_idle);                                \
+       i_bit_clear((my_cpu), &cpus_active);                            \
        splx(s);                                                        \
+       set_led(my_cpu);                                                \
 }
 
-#define MARK_CPU_ACTIVE(my_cpu) {                                      \
+#define MARK_CPU_ACTIVE(my_cpu)        {                                       \
                                                                        \
        int     s = splhigh();                                          \
        /*                                                              \
@@ -694,20 +630,23 @@ extern void pt_fake_zone_info(int *, vm_size_t *, vm_size_t *, vm_size_t *, vm_s
         *      set assures that we will receive another update         \
         *      interrupt if this happens.                              \
         */                                                             \
-       CPU_CR3_MARK_ACTIVE();                                          \
-       __asm__ volatile("mfence");                                     \
+       i_bit_clear((my_cpu), &cpus_idle);                              \
+                                                                       \
+       if (cpu_update_needed(my_cpu))                                  \
+           pmap_update_interrupt();                                    \
                                                                        \
-       if (current_cpu_datap()->cpu_tlb_invalid)                       \
-           process_pmap_updates();                                     \
+       /*                                                              \
+        *      Mark that this cpu is now active.                       \
+        */                                                             \
+       i_bit_set((my_cpu), &cpus_active);                              \
        splx(s);                                                        \
+       clear_led(my_cpu);                                              \
 }
 
 #define PMAP_CONTEXT(pmap, thread)
 
 #define pmap_kernel_va(VA)     \
-       ((((vm_offset_t) (VA)) >= vm_min_kernel_address) &&     \
-        (((vm_offset_t) (VA)) <= vm_max_kernel_address))
-
+       (((VA) >= VM_MIN_KERNEL_ADDRESS) && ((VA) <= VM_MAX_KERNEL_ADDRESS))
 
 #define pmap_resident_count(pmap)      ((pmap)->stats.resident_count)
 #define        pmap_copy(dst_pmap,src_pmap,dst_addr,len,src_addr)
@@ -718,8 +657,4 @@ extern void pt_fake_zone_info(int *, vm_size_t *, vm_size_t *, vm_size_t *, vm_s
 
 #endif /* ASSEMBLER */
 
-
 #endif /* _PMAP_MACHINE_ */
-
-
-#endif  /* KERNEL_PRIVATE */