]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/loose_ends.c
xnu-1228.5.18.tar.gz
[apple/xnu.git] / osfmk / i386 / loose_ends.c
index b0cd27fffdea95a0cd901f63905aa54f6ab5336c..d5ed99e5115978138a8a8edd81c58dcd1615f019 100644 (file)
@@ -1,23 +1,29 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License").  You may not use this file except in compliance with the
- * License.  Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
  * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ * 
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_COPYRIGHT@
 #include <mach/i386/vm_param.h>
 #include <kern/kern_types.h>
 #include <kern/misc_protos.h>
+#include <sys/errno.h>
+#include <i386/param.h>
 #include <i386/misc_protos.h>
+#include <i386/cpu_data.h>
+#include <i386/machine_routines.h>
+#include <i386/cpuid.h>
+#include <vm/pmap.h>
+#include <vm/vm_map.h>
+#include <vm/vm_kern.h>
+#include <vm/vm_fault.h>
 
-       /*
-        * Should be rewritten in asm anyway.
-        */
+#include <libkern/OSAtomic.h>
+#include <sys/kdebug.h>
 
+#if 0
 
-/*
- * copy 'size' bytes from physical to physical address
- * the caller must validate the physical ranges 
- *
- * if flush_action == 0, no cache flush necessary
- * if flush_action == 1, flush the source
- * if flush_action == 2, flush the dest
- * if flush_action == 3, flush both source and dest
- */
+#undef KERNEL_DEBUG
+#define KERNEL_DEBUG KERNEL_DEBUG_CONSTANT
+#define KDEBUG 1
 
-kern_return_t copyp2p(vm_offset_t source, vm_offset_t dest, unsigned int size, unsigned int flush_action) {
+#endif
 
-        switch(flush_action) {
-       case 1:
-               flush_dcache(source, size, 1);
-               break;
-       case 2:
-               flush_dcache(dest, size, 1);
-               break;
-       case 3:
-               flush_dcache(source, size, 1);
-               flush_dcache(dest, size, 1);
-               break;
+/* XXX - should be gone from here */
+extern void            invalidate_icache64(addr64_t addr, unsigned cnt, int phys);
+extern void            flush_dcache64(addr64_t addr, unsigned count, int phys);
+extern boolean_t       phys_page_exists(ppnum_t);
+extern void            bcopy_no_overwrite(const char *from, char *to,vm_size_t bytes);
+extern void            pmap_set_reference(ppnum_t pn);
+extern void            mapping_set_mod(ppnum_t pa); 
+extern void            mapping_set_ref(ppnum_t pn);
 
-       }
-        bcopy_phys((char *)source, (char *)dest, size);        /* Do a physical copy */
-
-        switch(flush_action) {
-       case 1:
-               flush_dcache(source, size, 1);
-               break;
-       case 2:
-               flush_dcache(dest, size, 1);
-               break;
-       case 3:
-               flush_dcache(source, size, 1);
-               flush_dcache(dest, size, 1);
-               break;
+extern void            ovbcopy(const char      *from,
+                               char            *to,
+                               vm_size_t       nbytes);
+void machine_callstack(natural_t *buf, vm_size_t callstack_max);
 
-       }
+
+#define value_64bit(value)  ((value) & 0xFFFFFFFF00000000LL)
+#define low32(x)  ((unsigned int)((x) & 0x00000000FFFFFFFFLL))
+
+void
+bzero_phys_nc(
+             addr64_t src64,
+             vm_size_t bytes)
+{
+  bzero_phys(src64,bytes);
 }
 
+void
+bzero_phys(
+          addr64_t src64,
+          vm_size_t bytes)
+{
+        mapwindow_t *map;
+
+        mp_disable_preemption();
 
+       map = pmap_get_mapwindow((pt_entry_t)(INTEL_PTE_VALID | INTEL_PTE_RW | ((pmap_paddr_t)src64 & PG_FRAME) | INTEL_PTE_REF | INTEL_PTE_MOD));
 
-/*
- *              Copies data from a physical page to a virtual page.  This is used to
- *              move data from the kernel to user state.
- *
- */
+       bzero((void *)((uintptr_t)map->prv_CADDR | ((uint32_t)src64 & INTEL_OFFMASK)), bytes);
 
-kern_return_t
-copyp2v(char *from, char *to, unsigned int size) {
+       pmap_put_mapwindow(map);
 
-  return(copyout(phystokv(from), to, size));
+       mp_enable_preemption();
 }
 
+
 /*
  * bcopy_phys - like bcopy but copies from/to physical addresses.
- *              this is trivial since all phys mem is mapped into 
- *              kernel virtual space
  */
 
 void
-bcopy_phys(const char *from, char *to, vm_size_t bytes)
+bcopy_phys(
+          addr64_t src64,
+          addr64_t dst64,
+          vm_size_t bytes)
 {
-  bcopy((char *)phystokv(from), (char *)phystokv(to), bytes);
-}
+        mapwindow_t *src_map, *dst_map;
+
+       /* ensure we stay within a page */
+       if ( ((((uint32_t)src64 & (NBPG-1)) + bytes) > NBPG) || ((((uint32_t)dst64 & (NBPG-1)) + bytes) > NBPG) ) {
+               panic("bcopy_phys alignment");
+       }
+       mp_disable_preemption();
+
+       src_map = pmap_get_mapwindow((pt_entry_t)(INTEL_PTE_VALID | ((pmap_paddr_t)src64 & PG_FRAME) | INTEL_PTE_REF));
+       dst_map = pmap_get_mapwindow((pt_entry_t)(INTEL_PTE_VALID | INTEL_PTE_RW | ((pmap_paddr_t)dst64 & PG_FRAME) |
+                                                 INTEL_PTE_REF | INTEL_PTE_MOD));
+
+       bcopy((void *) ((uintptr_t)src_map->prv_CADDR | ((uint32_t)src64 & INTEL_OFFMASK)),
+             (void *) ((uintptr_t)dst_map->prv_CADDR | ((uint32_t)dst64 & INTEL_OFFMASK)), bytes);
 
+       pmap_put_mapwindow(src_map);
+       pmap_put_mapwindow(dst_map);
+
+       mp_enable_preemption();
+}
 
 /* 
  * ovbcopy - like bcopy, but recognizes overlapping ranges and handles 
@@ -158,20 +186,264 @@ ovbcopy(
        }
 }
 
-void
-bcopy(
-       const char      *from,
-       char            *to,
-       vm_size_t       bytes)          /* num bytes to copy */
+
+/*
+ *  Read data from a physical address. Memory should not be cache inhibited.
+ */
+
+
+static unsigned int
+ml_phys_read_data(pmap_paddr_t paddr, int size )
+{
+        mapwindow_t *map;
+       unsigned int result;
+
+       mp_disable_preemption();
+
+       map = pmap_get_mapwindow((pt_entry_t)(INTEL_PTE_VALID | (paddr & PG_FRAME) | INTEL_PTE_REF));
+
+        switch (size) {
+            unsigned char s1;
+            unsigned short s2;
+        case 1:
+            s1 = *(unsigned char *)((uintptr_t)map->prv_CADDR | ((uint32_t)paddr & INTEL_OFFMASK));
+            result = s1;
+            break;
+        case 2:
+            s2 = *(unsigned short *)((uintptr_t)map->prv_CADDR | ((uint32_t)paddr & INTEL_OFFMASK));
+            result = s2;
+            break;
+        case 4:
+        default:
+            result = *(unsigned int *)((uintptr_t)map->prv_CADDR | ((uint32_t)paddr & INTEL_OFFMASK));
+            break;
+        }
+        pmap_put_mapwindow(map);
+
+       mp_enable_preemption();
+
+        return result;
+}
+
+static unsigned long long
+ml_phys_read_long_long(pmap_paddr_t paddr )
+{
+        mapwindow_t *map;
+       unsigned long long result;
+
+       mp_disable_preemption();
+
+       map = pmap_get_mapwindow((pt_entry_t)(INTEL_PTE_VALID | (paddr & PG_FRAME) | INTEL_PTE_REF));
+
+       result = *(unsigned long long *)((uintptr_t)map->prv_CADDR | ((uint32_t)paddr & INTEL_OFFMASK));
+
+        pmap_put_mapwindow(map);
+
+       mp_enable_preemption();
+
+        return result;
+}
+
+
+
+unsigned int ml_phys_read( vm_offset_t paddr)
+{
+        return ml_phys_read_data((pmap_paddr_t)paddr, 4);
+}
+
+unsigned int ml_phys_read_word(vm_offset_t paddr) {
+
+        return ml_phys_read_data((pmap_paddr_t)paddr, 4);
+}
+
+unsigned int ml_phys_read_64(addr64_t paddr64)
+{
+        return ml_phys_read_data((pmap_paddr_t)paddr64, 4);
+}
+
+unsigned int ml_phys_read_word_64(addr64_t paddr64)
+{
+        return ml_phys_read_data((pmap_paddr_t)paddr64, 4);
+}
+
+unsigned int ml_phys_read_half(vm_offset_t paddr)
+{
+        return ml_phys_read_data((pmap_paddr_t)paddr, 2);
+}
+
+unsigned int ml_phys_read_half_64(addr64_t paddr64)
 {
-       ovbcopy(from, to, bytes);
+        return ml_phys_read_data((pmap_paddr_t)paddr64, 2);
 }
 
+unsigned int ml_phys_read_byte(vm_offset_t paddr)
+{
+        return ml_phys_read_data((pmap_paddr_t)paddr, 1);
+}
+
+unsigned int ml_phys_read_byte_64(addr64_t paddr64)
+{
+        return ml_phys_read_data((pmap_paddr_t)paddr64, 1);
+}
+
+unsigned long long ml_phys_read_double(vm_offset_t paddr)
+{
+        return ml_phys_read_long_long((pmap_paddr_t)paddr);
+}
+
+unsigned long long ml_phys_read_double_64(addr64_t paddr64)
+{
+        return ml_phys_read_long_long((pmap_paddr_t)paddr64);
+}
+
+
+
+/*
+ *  Write data to a physical address. Memory should not be cache inhibited.
+ */
+
+static void
+ml_phys_write_data(pmap_paddr_t paddr, unsigned long data, int size)
+{
+        mapwindow_t *map;
+
+       mp_disable_preemption();
+
+       map = pmap_get_mapwindow((pt_entry_t)(INTEL_PTE_VALID | INTEL_PTE_RW | (paddr & PG_FRAME) | 
+                                         INTEL_PTE_REF | INTEL_PTE_MOD));
+
+        switch (size) {
+        case 1:
+           *(unsigned char *)((uintptr_t)map->prv_CADDR | ((uint32_t)paddr & INTEL_OFFMASK)) = (unsigned char)data;
+            break;
+        case 2:
+           *(unsigned short *)((uintptr_t)map->prv_CADDR | ((uint32_t)paddr & INTEL_OFFMASK)) = (unsigned short)data;
+            break;
+        case 4:
+        default:
+           *(unsigned int *)((uintptr_t)map->prv_CADDR | ((uint32_t)paddr & INTEL_OFFMASK)) = data;
+            break;
+        }
+        pmap_put_mapwindow(map);
+
+       mp_enable_preemption();
+}
+
+static void
+ml_phys_write_long_long(pmap_paddr_t paddr, unsigned long long data)
+{
+        mapwindow_t *map;
+
+       mp_disable_preemption();
+
+       map = pmap_get_mapwindow((pt_entry_t)(INTEL_PTE_VALID | INTEL_PTE_RW | (paddr & PG_FRAME) | 
+                                             INTEL_PTE_REF | INTEL_PTE_MOD));
+
+       *(unsigned long long *)((uintptr_t)map->prv_CADDR | ((uint32_t)paddr & INTEL_OFFMASK)) = data;
+
+        pmap_put_mapwindow(map);
+
+       mp_enable_preemption();
+}
+
+
+
+void ml_phys_write_byte(vm_offset_t paddr, unsigned int data)
+{
+        ml_phys_write_data((pmap_paddr_t)paddr, data, 1);
+}
+
+void ml_phys_write_byte_64(addr64_t paddr64, unsigned int data)
+{
+        ml_phys_write_data((pmap_paddr_t)paddr64, data, 1);
+}
+
+void ml_phys_write_half(vm_offset_t paddr, unsigned int data)
+{
+        ml_phys_write_data((pmap_paddr_t)paddr, data, 2);
+}
+
+void ml_phys_write_half_64(addr64_t paddr64, unsigned int data)
+{
+        ml_phys_write_data((pmap_paddr_t)paddr64, data, 2);
+}
+
+void ml_phys_write(vm_offset_t paddr, unsigned int data)
+{
+        ml_phys_write_data((pmap_paddr_t)paddr, data, 4);
+}
+
+void ml_phys_write_64(addr64_t paddr64, unsigned int data)
+{
+        ml_phys_write_data((pmap_paddr_t)paddr64, data, 4);
+}
+
+void ml_phys_write_word(vm_offset_t paddr, unsigned int data)
+{
+        ml_phys_write_data((pmap_paddr_t)paddr, data, 4);
+}
+
+void ml_phys_write_word_64(addr64_t paddr64, unsigned int data)
+{
+        ml_phys_write_data((pmap_paddr_t)paddr64, data, 4);
+}
+
+void ml_phys_write_double(vm_offset_t paddr, unsigned long long data)
+{
+        ml_phys_write_long_long((pmap_paddr_t)paddr, data);
+}
+
+void ml_phys_write_double_64(addr64_t paddr64, unsigned long long data)
+{
+        ml_phys_write_long_long((pmap_paddr_t)paddr64, data);
+}
+
+
+/* PCI config cycle probing
+ *
+ *
+ *      Read the memory location at physical address paddr.
+ *  This is a part of a device probe, so there is a good chance we will
+ *  have a machine check here. So we have to be able to handle that.
+ *  We assume that machine checks are enabled both in MSR and HIDs
+ */
+
+boolean_t
+ml_probe_read(vm_offset_t paddr, unsigned int *val)
+{
+    if ((PAGE_SIZE - (paddr & PAGE_MASK)) < 4)
+        return FALSE;
+
+    *val = ml_phys_read((pmap_paddr_t)paddr);
+
+    return TRUE;
+}
+
+/*
+ *  Read the memory location at physical address paddr.
+ *  This is a part of a device probe, so there is a good chance we will
+ *  have a machine check here. So we have to be able to handle that.
+ *  We assume that machine checks are enabled both in MSR and HIDs
+ */
+boolean_t 
+ml_probe_read_64(addr64_t paddr64, unsigned int *val)
+{
+    if ((PAGE_SIZE - (paddr64 & PAGE_MASK)) < 4)
+        return FALSE;
+
+    *val = ml_phys_read_64((pmap_paddr_t)paddr64);
+    return TRUE;
+}
+
+
 int bcmp(
-       const char      *a,
-       const char      *b,
-       vm_size_t       len)
+       const void      *pa,
+       const void      *pb,
+       size_t  len)
 {
+       const char *a = (const char *)pa;
+       const char *b = (const char *)pb;
+
        if (len == 0)
                return 0;
 
@@ -184,13 +456,16 @@ int bcmp(
 }
 
 int
-memcmp(s1, s2, n)
-       register char *s1, *s2;
-       register n;
+memcmp(const void *s1, const void *s2, size_t n)
 {
-       while (--n >= 0)
-               if (*s1++ != *s2++)
-                       return (*--s1 - *--s2);
+       if (n != 0) {
+               const unsigned char *p1 = s1, *p2 = s2;
+
+               do {
+                       if (*p1++ != *p2++)
+                               return (*--p1 - *--p2);
+               } while (--n != 0);
+       }
        return (0);
 }
 
@@ -211,95 +486,725 @@ strlen(
        return string - 1 - ret;
 }
 
-#include <libkern/OSAtomic.h>
-
 uint32_t
-hw_atomic_add(
-       uint32_t        *dest,
-       uint32_t        delt)
+hw_compare_and_store(uint32_t oldval, uint32_t newval, volatile uint32_t *dest)
 {
-       uint32_t        oldValue;
-       uint32_t        newValue;
-       
-       do {
-               oldValue = *dest;
-               newValue = (oldValue + delt);
-       } while (!OSCompareAndSwap((UInt32)oldValue,
-                                                                       (UInt32)newValue, (UInt32 *)dest));
-       
-       return newValue;
+       return OSCompareAndSwap((UInt32)oldval,
+                               (UInt32)newval,
+                               (volatile UInt32 *)dest);
 }
 
-uint32_t
-hw_atomic_sub(
-       uint32_t        *dest,
-       uint32_t        delt)
+#if    MACH_ASSERT
+
+/*
+ * Machine-dependent routine to fill in an array with up to callstack_max
+ * levels of return pc information.
+ */
+void machine_callstack(
+       __unused natural_t      *buf,
+       __unused vm_size_t      callstack_max)
 {
-       uint32_t        oldValue;
-       uint32_t        newValue;
-       
-       do {
-               oldValue = *dest;
-               newValue = (oldValue - delt);
-       } while (!OSCompareAndSwap((UInt32)oldValue,
-                                                                       (UInt32)newValue, (UInt32 *)dest));
-       
-       return newValue;
 }
 
-uint32_t
-hw_atomic_or(
-       uint32_t        *dest,
-       uint32_t        mask)
+#endif /* MACH_ASSERT */
+
+void fillPage(ppnum_t pa, unsigned int fill)
 {
-       uint32_t        oldValue;
-       uint32_t        newValue;
-       
-       do {
-               oldValue = *dest;
-               newValue = (oldValue | mask);
-       } while (!OSCompareAndSwap((UInt32)oldValue,
-                                                                       (UInt32)newValue, (UInt32 *)dest));
-       
-       return newValue;
+        mapwindow_t *map;
+       pmap_paddr_t src;
+       int i;
+       int cnt = PAGE_SIZE/sizeof(unsigned int);
+       unsigned int *addr;
+
+       mp_disable_preemption();
+
+       src = i386_ptob(pa);
+       map = pmap_get_mapwindow((pt_entry_t)(INTEL_PTE_VALID | INTEL_PTE_RW | (src & PG_FRAME) | 
+                                             INTEL_PTE_REF | INTEL_PTE_MOD));
+
+       for (i = 0, addr = (unsigned int *)map->prv_CADDR; i < cnt ; i++ )
+               *addr++ = fill;
+
+       pmap_put_mapwindow(map);
+
+       mp_enable_preemption();
 }
 
-uint32_t
-hw_atomic_and(
-       uint32_t        *dest,
-       uint32_t        mask)
+static inline void __sfence(void)
 {
-       uint32_t        oldValue;
-       uint32_t        newValue;
-       
-       do {
-               oldValue = *dest;
-               newValue = (oldValue & mask);
-       } while (!OSCompareAndSwap((UInt32)oldValue,
-                                                                       (UInt32)newValue, (UInt32 *)dest));
-       
-       return newValue;
+    __asm__ volatile("sfence");
+}
+static inline void __mfence(void)
+{
+    __asm__ volatile("mfence");
+}
+static inline void __wbinvd(void)
+{
+    __asm__ volatile("wbinvd");
+}
+static inline void __clflush(void *ptr)
+{
+       __asm__ volatile("clflush (%0)" : : "r" (ptr));
 }
 
-uint32_t
-hw_compare_and_store(
-       uint32_t        oldval,
-       uint32_t        newval,
-       uint32_t        *dest)
+void dcache_incoherent_io_store64(addr64_t pa, unsigned int count)
 {
-       return OSCompareAndSwap((UInt32)oldval, (UInt32)newval, (UInt32 *)dest);
+        mapwindow_t *map;
+        uint32_t  linesize = cpuid_info()->cache_linesize;
+        addr64_t  addr;
+        uint32_t  offset, chunk;
+        boolean_t istate;
+
+       __mfence();
+
+        istate = ml_set_interrupts_enabled(FALSE);
+
+        offset = pa & (linesize - 1);
+        addr   = pa - offset;
+
+        map = pmap_get_mapwindow((pt_entry_t)(i386_ptob(atop_64(addr)) | INTEL_PTE_VALID));
+
+        count += offset;
+        offset = addr & ((addr64_t) (page_size - 1));
+        chunk  = page_size - offset;
+
+        do
+        {
+            if (chunk > count)
+                chunk = count;
+    
+            for (; offset < chunk; offset += linesize)
+                __clflush((void *)(((uintptr_t)map->prv_CADDR) + offset));
+
+            count -= chunk;
+            addr  += chunk;
+            chunk  = page_size;
+            offset = 0;
+
+           if (count) {
+               pmap_store_pte(map->prv_CMAP, (pt_entry_t)(i386_ptob(atop_64(addr)) | INTEL_PTE_VALID));
+               invlpg((uintptr_t)map->prv_CADDR);
+           }
+        }
+        while (count);
+
+        pmap_put_mapwindow(map);
+
+        (void) ml_set_interrupts_enabled(istate);
+
+       __mfence();
+}
+
+void dcache_incoherent_io_flush64(addr64_t pa, unsigned int count)
+{
+    return(dcache_incoherent_io_store64(pa,count));
+}
+
+void
+flush_dcache64(__unused addr64_t addr,
+              __unused unsigned count,
+              __unused int phys)
+{
+}
+
+void
+invalidate_icache64(__unused addr64_t addr,
+                   __unused unsigned count,
+                   __unused int phys)
+{
+}
+
+
+addr64_t         vm_last_addr;
+
+void
+mapping_set_mod(ppnum_t pn)
+{
+  pmap_set_modify(pn);
+}
+
+void
+mapping_set_ref(ppnum_t pn)
+{
+  pmap_set_reference(pn);
+}
+
+void
+cache_flush_page_phys(ppnum_t pa)
+{
+        mapwindow_t     *map;
+       boolean_t       istate;
+       int             i;
+       unsigned char   *cacheline_addr;
+       int             cacheline_size = cpuid_info()->cache_linesize;
+       int             cachelines_in_page = PAGE_SIZE/cacheline_size;
+
+       __mfence();
+
+       istate = ml_set_interrupts_enabled(FALSE);
+
+        map = pmap_get_mapwindow((pt_entry_t)(i386_ptob(pa) | INTEL_PTE_VALID));
+
+       for (i = 0, cacheline_addr = (unsigned char *)map->prv_CADDR;
+            i < cachelines_in_page;
+            i++, cacheline_addr += cacheline_size) {
+               __clflush((void *) cacheline_addr);
+       }
+        pmap_put_mapwindow(map);
+
+       (void) ml_set_interrupts_enabled(istate);
+
+       __mfence();
 }
 
-#if    MACH_ASSERT
 
 /*
- * Machine-dependent routine to fill in an array with up to callstack_max
- * levels of return pc information.
+ * the copy engine has the following characteristics
+ *   - copyio handles copies to/from user or kernel space
+ *   - copypv deals with physical or virtual addresses
+ *
+ * implementation details as follows
+ *   - a cache of up to NCOPY_WINDOWS is maintained per thread for
+ *     access of user virutal space
+ *   - the window size is determined by the amount of virtual space
+ *     that can be mapped by a single page table
+ *   - the mapping is done by copying the page table pointer from
+ *     the user's directory entry corresponding to the window's
+ *     address in user space to the directory entry corresponding
+ *     to the window slot in the kernel's address space
+ *   - the set of mappings is preserved across context switches,
+ *     so the copy can run with pre-emption enabled
+ *   - there is a gdt entry set up to anchor the kernel window on
+ *     each processor
+ *   - the copies are done using the selector corresponding to the
+ *     gdt entry
+ *   - the addresses corresponding to the user virtual address are
+ *     relative to the beginning of the window being used to map
+ *     that region... thus the thread can be pre-empted and switched
+ *     to a different processor while in the midst of a copy
+ *   - the window caches must be invalidated if the pmap changes out
+ *     from under the thread... this can happen during vfork/exec...
+ *     inval_copy_windows is the invalidation routine to be used
+ *   - the copyio engine has 4 different states associated with it
+ *     that allows for lazy tlb flushes and the ability to avoid
+ *     a flush all together if we've just come from user space
+ *     the 4 states are as follows...
+ *
+ *     WINDOWS_OPENED - set by copyio to indicate to the context
+ *       switch code that it is necessary to do a tlbflush after
+ *       switching the windows since we're in the middle of a copy
+ *
+ *     WINDOWS_CLOSED - set by copyio to indicate that it's done
+ *       using the windows, so that the context switch code need
+ *       not do the tlbflush... instead it will set the state to...
+ *
+ *     WINDOWS_DIRTY - set by the context switch code to indicate
+ *       to the copy engine that it is responsible for doing a 
+ *       tlbflush before using the windows again... it's also
+ *       set by the inval_copy_windows routine to indicate the
+ *       same responsibility.
+ *
+ *     WINDOWS_CLEAN - set by the return to user path to indicate
+ *       that a tlbflush has happened and that there is no need
+ *       for copyio to do another when it is entered next...
+ *
+ *   - a window for mapping single physical pages is provided for copypv
+ *   - this window is maintained across context switches and has the
+ *     same characteristics as the user space windows w/r to pre-emption
  */
-void machine_callstack(
-       natural_t       *buf,
-       vm_size_t       callstack_max)
+
+extern int copyout_user(const char *, vm_offset_t, vm_size_t);
+extern int copyout_kern(const char *, vm_offset_t, vm_size_t);
+extern int copyin_user(const vm_offset_t, char *, vm_size_t);
+extern int copyin_kern(const vm_offset_t, char *, vm_size_t);
+extern int copyoutphys_user(const char *, vm_offset_t, vm_size_t);
+extern int copyoutphys_kern(const char *, vm_offset_t, vm_size_t);
+extern int copyinphys_user(const vm_offset_t, char *, vm_size_t);
+extern int copyinphys_kern(const vm_offset_t, char *, vm_size_t);
+extern int copyinstr_user(const vm_offset_t, char *, vm_size_t, vm_size_t *);
+extern int copyinstr_kern(const vm_offset_t, char *, vm_size_t, vm_size_t *);
+
+static int copyio(int, user_addr_t, char *, vm_size_t, vm_size_t *, int);
+static int copyio_phys(addr64_t, addr64_t, vm_size_t, int);
+
+
+#define COPYIN         0
+#define COPYOUT                1
+#define COPYINSTR      2
+#define COPYINPHYS     3
+#define COPYOUTPHYS    4
+
+
+
+void inval_copy_windows(thread_t thread)
 {
+        int    i;
+       
+       for (i = 0; i < NCOPY_WINDOWS; i++) {
+                thread->machine.copy_window[i].user_base = -1;
+       }
+       thread->machine.nxt_window = 0;
+       thread->machine.copyio_state = WINDOWS_DIRTY;
+
+       KERNEL_DEBUG(0xeff70058 | DBG_FUNC_NONE, (int)thread, (int)thread->map, 0, 0, 0);
 }
 
-#endif /* MACH_ASSERT */
+
+static int
+copyio(int copy_type, user_addr_t user_addr, char *kernel_addr,
+       vm_size_t nbytes, vm_size_t *lencopied, int use_kernel_map)
+{
+        thread_t       thread;
+       pmap_t          pmap;
+       pt_entry_t      *updp;
+       pt_entry_t      *kpdp;
+       user_addr_t     user_base;
+       vm_offset_t     user_offset;
+       vm_offset_t     kern_vaddr;
+       vm_size_t       cnt;
+       vm_size_t       bytes_copied;
+       int             error = 0;
+       int             window_index;
+       int             copyio_state;
+        boolean_t      istate;
+#if KDEBUG
+       int             debug_type = 0xeff70010;
+       debug_type += (copy_type << 2);
+#endif
+
+       thread = current_thread();
+
+       KERNEL_DEBUG(debug_type | DBG_FUNC_START, (int)(user_addr >> 32), (int)user_addr,
+                    (int)nbytes, thread->machine.copyio_state, 0);
+
+       if (nbytes == 0) {
+               KERNEL_DEBUG(debug_type | DBG_FUNC_END, (unsigned)user_addr,
+                            (unsigned)kernel_addr, (unsigned)nbytes, 0, 0);
+               return (0);
+       }
+        pmap = thread->map->pmap;
+
+#if CONFIG_DTRACE
+       thread->machine.specFlags |= CopyIOActive;
+#endif /* CONFIG_DTRACE */
+
+        if (pmap == kernel_pmap || use_kernel_map) {
+
+               kern_vaddr = (vm_offset_t)user_addr;
+         
+               switch (copy_type) {
+
+               case COPYIN:
+                       error = copyin_kern(kern_vaddr, kernel_addr, nbytes);
+                       break;
+
+               case COPYOUT:
+                       error = copyout_kern(kernel_addr, kern_vaddr, nbytes);
+                       break;
+
+               case COPYINSTR:
+                       error = copyinstr_kern(kern_vaddr, kernel_addr, nbytes, lencopied);
+                       break;
+
+               case COPYINPHYS:
+                       error = copyinphys_kern(kern_vaddr, kernel_addr, nbytes);
+                       break;
+
+               case COPYOUTPHYS:
+                       error = copyoutphys_kern(kernel_addr, kern_vaddr, nbytes);
+                       break;
+               }
+               KERNEL_DEBUG(debug_type | DBG_FUNC_END, (unsigned)kern_vaddr,
+                            (unsigned)kernel_addr, (unsigned)nbytes,
+                            error | 0x80000000, 0);
+
+#if CONFIG_DTRACE
+       thread->machine.specFlags &= ~CopyIOActive;
+#endif /* CONFIG_DTRACE */
+
+               return (error);
+       }
+       user_base = user_addr & ~((user_addr_t)(NBPDE - 1));
+       user_offset = user_addr & (NBPDE - 1);
+
+       KERNEL_DEBUG(debug_type | DBG_FUNC_NONE, (int)(user_base >> 32), (int)user_base,
+                    (int)user_offset, 0, 0);
+
+       cnt = NBPDE - user_offset;
+
+       if (cnt > nbytes)
+               cnt = nbytes;
+
+       istate = ml_set_interrupts_enabled(FALSE);
+
+       copyio_state = thread->machine.copyio_state;
+       thread->machine.copyio_state = WINDOWS_OPENED;
+
+       (void) ml_set_interrupts_enabled(istate);
+
+
+       for (;;) {
+
+               for (window_index = 0; window_index < NCOPY_WINDOWS; window_index++) {
+                       if (thread->machine.copy_window[window_index].user_base == user_base)
+                               break;
+               }
+               if (window_index >= NCOPY_WINDOWS) {
+
+                       window_index = thread->machine.nxt_window;
+                       thread->machine.nxt_window++;
+
+                       if (thread->machine.nxt_window >= NCOPY_WINDOWS)
+                               thread->machine.nxt_window = 0;
+                       thread->machine.copy_window[window_index].user_base = user_base;
+
+                       /*
+                        * it's necessary to disable pre-emption
+                        * since I have to compute the kernel descriptor pointer
+                        * for the new window
+                        */
+                       istate = ml_set_interrupts_enabled(FALSE);
+
+                       updp = pmap_pde(pmap, user_base);
+
+                       kpdp = current_cpu_datap()->cpu_copywindow_pdp;
+                       kpdp += window_index;
+
+                       pmap_store_pte(kpdp, updp ? *updp : 0);
+
+                       (void) ml_set_interrupts_enabled(istate);
+
+                       copyio_state = WINDOWS_DIRTY;
+
+                       KERNEL_DEBUG(0xeff70040 | DBG_FUNC_NONE, window_index,
+                                    (unsigned)user_base, (unsigned)updp,
+                                    (unsigned)kpdp, 0);
+
+               }
+#if JOE_DEBUG
+               else {
+                       istate = ml_set_interrupts_enabled(FALSE);
+
+                       updp = pmap_pde(pmap, user_base);
+
+                       kpdp = current_cpu_datap()->cpu_copywindow_pdp;
+
+                       kpdp += window_index;
+
+                       if ((*kpdp & PG_FRAME) != (*updp & PG_FRAME)) {
+                               panic("copyio: user pdp mismatch - kpdp = 0x%x,  updp = 0x%x\n", kpdp, updp);
+                       }
+                       (void) ml_set_interrupts_enabled(istate);
+               }
+#endif
+               if (copyio_state == WINDOWS_DIRTY) {
+                       flush_tlb();
+
+                       copyio_state = WINDOWS_CLEAN;
+
+                       KERNEL_DEBUG(0xeff70054 | DBG_FUNC_NONE, window_index, 0, 0, 0, 0);
+               }
+               user_offset += (window_index * NBPDE);
+
+               KERNEL_DEBUG(0xeff70044 | DBG_FUNC_NONE, (unsigned)user_offset,
+                            (unsigned)kernel_addr, cnt, 0, 0);
+
+               switch (copy_type) {
+
+               case COPYIN:
+                       error = copyin_user(user_offset, kernel_addr, cnt);
+                       break;
+                       
+               case COPYOUT:
+                       error = copyout_user(kernel_addr, user_offset, cnt);
+                       break;
+
+               case COPYINPHYS:
+                       error = copyinphys_user(user_offset, kernel_addr, cnt);
+                       break;
+                       
+               case COPYOUTPHYS:
+                       error = copyoutphys_user(kernel_addr, user_offset, cnt);
+                       break;
+
+               case COPYINSTR:
+                       error = copyinstr_user(user_offset, kernel_addr, cnt, &bytes_copied);
+
+                       /*
+                        * lencopied should be updated on success
+                        * or ENAMETOOLONG...  but not EFAULT
+                        */
+                       if (error != EFAULT)
+                               *lencopied += bytes_copied;
+
+                       /*
+                        * if we still have room, then the ENAMETOOLONG
+                        * is just an artifact of the buffer straddling
+                        * a window boundary and we should continue
+                        */
+                       if (error == ENAMETOOLONG && nbytes > cnt)
+                               error = 0;
+
+                       if (error) {
+#if KDEBUG
+                               nbytes = *lencopied;
+#endif
+                               break;
+                       }
+                       if (*(kernel_addr + bytes_copied - 1) == 0) {
+                               /*
+                                * we found a NULL terminator... we're done
+                                */
+#if KDEBUG
+                               nbytes = *lencopied;
+#endif
+                               goto done;
+                       }
+                       if (cnt == nbytes) {
+                               /*
+                                * no more room in the buffer and we haven't
+                                * yet come across a NULL terminator
+                                */
+#if KDEBUG
+                               nbytes = *lencopied;
+#endif
+                               error = ENAMETOOLONG;
+                               break;
+                       }
+                       assert(cnt == bytes_copied);
+
+                       break;
+               }
+               if (error)
+                       break;
+               if ((nbytes -= cnt) == 0)
+                       break;
+
+               kernel_addr += cnt;
+               user_base += NBPDE;
+               user_offset = 0;
+
+               if (nbytes > NBPDE)
+                       cnt = NBPDE;
+               else
+                       cnt = nbytes;
+       }
+done:
+       thread->machine.copyio_state = WINDOWS_CLOSED;
+
+       KERNEL_DEBUG(debug_type | DBG_FUNC_END, (unsigned)user_addr,
+                    (unsigned)kernel_addr, (unsigned)nbytes, error, 0);
+
+#if CONFIG_DTRACE
+       thread->machine.specFlags &= ~CopyIOActive;
+#endif /* CONFIG_DTRACE */
+
+       return (error);
+}
+
+
+static int
+copyio_phys(addr64_t source, addr64_t sink, vm_size_t csize, int which)
+{
+        pmap_paddr_t paddr;
+       user_addr_t vaddr;
+       char        *window_offset;
+       pt_entry_t  pentry;
+       int         ctype;
+       int         retval;
+       boolean_t   istate;
+
+       if (which & cppvPsnk) {
+               paddr  = (pmap_paddr_t)sink;
+               vaddr  = (user_addr_t)source;
+               ctype  = COPYINPHYS;
+               pentry = (pt_entry_t)(INTEL_PTE_VALID | (paddr & PG_FRAME) | INTEL_PTE_RW);
+       } else {
+               paddr  = (pmap_paddr_t)source;
+               vaddr  = (user_addr_t)sink;
+               ctype  = COPYOUTPHYS;
+               pentry = (pt_entry_t)(INTEL_PTE_VALID | (paddr & PG_FRAME));
+       }
+       window_offset = (char *)((uint32_t)paddr & (PAGE_SIZE - 1));
+
+       if (current_thread()->machine.physwindow_busy) {
+               pt_entry_t      old_pentry;
+
+               KERNEL_DEBUG(0xeff70048 | DBG_FUNC_NONE, paddr, csize, 0, -1, 0);
+               /*
+                * we had better be targeting wired memory at this point
+                * we will not be able to handle a fault with interrupts
+                * disabled... we disable them because we can't tolerate
+                * being preempted during this nested use of the window
+                */
+               istate = ml_set_interrupts_enabled(FALSE);
+
+               old_pentry = *(current_cpu_datap()->cpu_physwindow_ptep);
+               pmap_store_pte((current_cpu_datap()->cpu_physwindow_ptep), pentry);
+
+               invlpg((uintptr_t)current_cpu_datap()->cpu_physwindow_base);
+
+               retval = copyio(ctype, vaddr, window_offset, csize, NULL, which & cppvKmap);
+
+               pmap_store_pte((current_cpu_datap()->cpu_physwindow_ptep), old_pentry);
+
+               invlpg((uintptr_t)current_cpu_datap()->cpu_physwindow_base);
+
+               (void) ml_set_interrupts_enabled(istate);
+       } else {
+               /*
+                * mark the window as in use... if an interrupt hits while we're
+                * busy, or we trigger another coyppv from the fault path into
+                * the driver on a user address space page fault due to a copyin/out
+                * then we need to save and restore the current window state instead
+                * of caching the window preserving it across context switches
+                */
+               current_thread()->machine.physwindow_busy = 1;
+
+               if (current_thread()->machine.physwindow_pte != pentry) {
+                       KERNEL_DEBUG(0xeff70048 | DBG_FUNC_NONE, paddr, csize, 0, 0, 0);
+
+                       current_thread()->machine.physwindow_pte = pentry;
+                       
+                       /*
+                        * preemption at this point would be bad since we
+                        * could end up on the other processor after we grabbed the
+                        * pointer to the current cpu data area, but before we finished
+                        * using it to stuff the page table entry since we would
+                        * be modifying a window that no longer belonged to us
+                        * the invlpg can be done unprotected since it only flushes
+                        * this page address from the tlb... if it flushes the wrong
+                        * one, no harm is done, and the context switch that moved us
+                        * to the other processor will have already take care of 
+                        * flushing the tlb after it reloaded the page table from machine.physwindow_pte
+                        */
+                       istate = ml_set_interrupts_enabled(FALSE);
+                       pmap_store_pte((current_cpu_datap()->cpu_physwindow_ptep), pentry);
+                       (void) ml_set_interrupts_enabled(istate);
+
+                       invlpg((uintptr_t)current_cpu_datap()->cpu_physwindow_base);
+               }
+#if JOE_DEBUG
+               else {
+                       if (pentry !=
+                           (*(current_cpu_datap()->cpu_physwindow_ptep) & (INTEL_PTE_VALID | PG_FRAME | INTEL_PTE_RW)))
+                               panic("copyio_phys: pentry != *physwindow_ptep");
+               }
+#endif
+               retval = copyio(ctype, vaddr, window_offset, csize, NULL, which & cppvKmap);
+
+               current_thread()->machine.physwindow_busy = 0;
+       }
+       return (retval);
+}
+
+int
+copyinmsg(const user_addr_t user_addr, char *kernel_addr, vm_size_t nbytes)
+{
+        return (copyio(COPYIN, user_addr, kernel_addr, nbytes, NULL, 0));
+}    
+
+int
+copyin(const user_addr_t user_addr, char *kernel_addr, vm_size_t nbytes)
+{
+        return (copyio(COPYIN, user_addr, kernel_addr, nbytes, NULL, 0));
+}
+
+int
+copyinstr(const user_addr_t user_addr,  char *kernel_addr, vm_size_t nbytes, vm_size_t *lencopied)
+{
+       *lencopied = 0;
+
+        return (copyio(COPYINSTR, user_addr, kernel_addr, nbytes, lencopied, 0));
+}
+
+int
+copyoutmsg(const char *kernel_addr, user_addr_t user_addr, vm_size_t nbytes)
+{
+        return (copyio(COPYOUT, user_addr, kernel_addr, nbytes, NULL, 0));
+}
+
+int
+copyout(const void *kernel_addr, user_addr_t user_addr, vm_size_t nbytes)
+{
+        return (copyio(COPYOUT, user_addr, kernel_addr, nbytes, NULL, 0));
+}
+
+
+kern_return_t
+copypv(addr64_t src64, addr64_t snk64, unsigned int size, int which)
+{
+       unsigned int lop, csize;
+       int bothphys = 0;
+       
+       KERNEL_DEBUG(0xeff7004c | DBG_FUNC_START, (unsigned)src64,
+                    (unsigned)snk64, size, which, 0);
+
+       if ((which & (cppvPsrc | cppvPsnk)) == 0 )                              /* Make sure that only one is virtual */
+               panic("copypv: no more than 1 parameter may be virtual\n");     /* Not allowed */
+
+       if ((which & (cppvPsrc | cppvPsnk)) == (cppvPsrc | cppvPsnk))
+               bothphys = 1;                                                   /* both are physical */
+
+       while (size) {
+         
+               if (bothphys) {
+                       lop = (unsigned int)(PAGE_SIZE - (snk64 & (PAGE_SIZE - 1)));            /* Assume sink smallest */
+
+                       if (lop > (unsigned int)(PAGE_SIZE - (src64 & (PAGE_SIZE - 1))))
+                               lop = (unsigned int)(PAGE_SIZE - (src64 & (PAGE_SIZE - 1)));    /* No, source is smaller */
+               } else {
+                       /*
+                        * only need to compute the resid for the physical page
+                        * address... we don't care about where we start/finish in
+                        * the virtual since we just call the normal copyin/copyout
+                        */
+                       if (which & cppvPsrc)
+                               lop = (unsigned int)(PAGE_SIZE - (src64 & (PAGE_SIZE - 1)));
+                       else
+                               lop = (unsigned int)(PAGE_SIZE - (snk64 & (PAGE_SIZE - 1)));
+               }
+               csize = size;                                           /* Assume we can copy it all */
+               if (lop < size)
+                       csize = lop;                                    /* Nope, we can't do it all */
+#if 0          
+               /*
+                * flush_dcache64 is currently a nop on the i386... 
+                * it's used when copying to non-system memory such
+                * as video capture cards... on PPC there was a need
+                * to flush due to how we mapped this memory... not
+                * sure if it's needed on i386.
+                */
+               if (which & cppvFsrc)
+                       flush_dcache64(src64, csize, 1);                /* If requested, flush source before move */
+               if (which & cppvFsnk)
+                       flush_dcache64(snk64, csize, 1);                /* If requested, flush sink before move */
+#endif
+               if (bothphys)
+                       bcopy_phys(src64, snk64, csize);                /* Do a physical copy, virtually */
+               else {
+                       if (copyio_phys(src64, snk64, csize, which))
+                               return (KERN_FAILURE);
+               }
+#if 0
+               if (which & cppvFsrc)
+                       flush_dcache64(src64, csize, 1);        /* If requested, flush source after move */
+               if (which & cppvFsnk)
+                       flush_dcache64(snk64, csize, 1);        /* If requested, flush sink after move */
+#endif
+               size   -= csize;                                        /* Calculate what is left */
+               snk64 += csize;                                 /* Bump sink to next physical address */
+               src64 += csize;                                 /* Bump source to next physical address */
+       }
+       KERNEL_DEBUG(0xeff7004c | DBG_FUNC_END, (unsigned)src64,
+                    (unsigned)snk64, size, which, 0);
+
+       return KERN_SUCCESS;
+}
+
+#if !MACH_KDP
+void
+kdp_register_callout(void)
+{
+}
+#endif