/*
- * Copyright (c) 2003-2006 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2003-2010 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
#include <mach/mach_types.h>
#include <mach/machine.h>
#include <mach/vm_map.h>
+#include <mach/mach_vm.h>
+#include <mach/machine.h>
+#include <i386/cpuid.h>
+#include <i386/tsc.h>
+#include <i386/rtclock_protos.h>
+#include <i386/cpu_data.h>
#include <i386/machine_routines.h>
#include <i386/misc_protos.h>
+#include <i386/cpuid.h>
#include <machine/cpu_capabilities.h>
#include <machine/commpage.h>
#include <machine/pmap.h>
#include <vm/vm_kern.h>
#include <vm/vm_map.h>
+
#include <ipc/ipc_port.h>
#include <kern/page_decrypt.h>
+#include <kern/processor.h>
/* the lists of commpage routines are in commpage_asm.s */
extern commpage_descriptor* commpage_32_routines[];
extern commpage_descriptor* commpage_64_routines[];
-/* translated commpage descriptors from commpage_sigs.c */
-extern commpage_descriptor sigdata_descriptor;
-extern commpage_descriptor *ba_descriptors[];
-
-extern vm_map_t com_region_map32; // the shared submap, set up in vm init
-extern vm_map_t com_region_map64; // the shared submap, set up in vm init
+extern vm_map_t commpage32_map; // the shared submap, set up in vm init
+extern vm_map_t commpage64_map; // the shared submap, set up in vm init
char *commPagePtr32 = NULL; // virtual addr in kernel map of 32-bit commpage
char *commPagePtr64 = NULL; // ...and of 64-bit commpage
-int _cpu_capabilities = 0; // define the capability vector
+uint32_t _cpu_capabilities = 0; // define the capability vector
int noVMX = 0; /* if true, do not set kHasAltivec in ppc _cpu_capabilities */
-void* dsmos_blobs[3]; /* ptrs to the system integrity data in each commpage */
-int dsmos_blob_count = 0;
+typedef uint32_t commpage_address_t;
-static uintptr_t next; // next available byte in comm page
-static int cur_routine; // comm page address of "current" routine
-static int matched; // true if we've found a match for "current" routine
+static commpage_address_t next; // next available address in comm page
+static commpage_address_t cur_routine; // comm page address of "current" routine
+static boolean_t matched; // true if we've found a match for "current" routine
static char *commPagePtr; // virtual addr in kernel map of commpage we are working on
-static size_t commPageBaseOffset; // add to 32-bit runtime address to get offset in commpage
+static commpage_address_t commPageBaseOffset; // subtract from 32-bit runtime address to get offset in virtual commpage in kernel map
+
+static commpage_time_data *time_data32 = NULL;
+static commpage_time_data *time_data64 = NULL;
+
+decl_simple_lock_data(static,commpage_active_cpus_lock);
/* Allocate the commpage and add to the shared submap created by vm:
* 1. allocate a page in the kernel map (RW)
static void*
commpage_allocate(
- vm_map_t submap, // com_region_map32 or com_region_map64
+ vm_map_t submap, // commpage32_map or commpage_map64
size_t area_used ) // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
{
- vm_offset_t kernel_addr; // address of commpage in kernel map
+ vm_offset_t kernel_addr = 0; // address of commpage in kernel map
vm_offset_t zero = 0;
vm_size_t size = area_used; // size actually populated
vm_map_entry_t entry;
if (submap == NULL)
panic("commpage submap is null");
- if (vm_allocate(kernel_map,&kernel_addr,area_used,VM_FLAGS_ANYWHERE))
+ if (vm_map(kernel_map,&kernel_addr,area_used,0,VM_FLAGS_ANYWHERE,NULL,0,FALSE,VM_PROT_ALL,VM_PROT_ALL,VM_INHERIT_NONE))
panic("cannot allocate commpage");
if (vm_map_wire(kernel_map,kernel_addr,kernel_addr+area_used,VM_PROT_DEFAULT,FALSE))
if (mach_make_memory_entry( kernel_map, // target map
&size, // size
kernel_addr, // offset (address in kernel map)
- VM_PROT_DEFAULT, // map it RW
+ VM_PROT_ALL, // map it RWX
&handle, // this is the object handle we get
NULL )) // parent_entry (what is this?)
panic("cannot make entry for commpage");
handle, // port is the memory entry we just made
0, // offset (map 1st page in memory entry)
FALSE, // copy
- VM_PROT_READ, // cur_protection (R-only in user map)
- VM_PROT_READ, // max_protection
+ VM_PROT_READ|VM_PROT_EXECUTE, // cur_protection (R-only in user map)
+ VM_PROT_READ|VM_PROT_EXECUTE, // max_protection
VM_INHERIT_SHARE )) // inheritance
panic("cannot map commpage");
ipc_port_release(handle);
-
- return (void*) kernel_addr; // return address in kernel map
+
+ // Initialize the text section of the commpage with INT3
+ char *commpage_ptr = (char*)(intptr_t)kernel_addr;
+ vm_size_t i;
+ for( i = _COMM_PAGE_TEXT_START - _COMM_PAGE_START_ADDRESS; i < size; i++ )
+ // This is the hex for the X86 opcode INT3
+ commpage_ptr[i] = 0xCC;
+
+ return (void*)(intptr_t)kernel_addr; // return address in kernel map
}
/* Get address (in kernel map) of a commpage field. */
static void*
commpage_addr_of(
- int addr_at_runtime )
+ commpage_address_t addr_at_runtime )
{
- return (void*) ((uintptr_t)commPagePtr + addr_at_runtime - commPageBaseOffset);
+ return (void*) ((uintptr_t)commPagePtr + (addr_at_runtime - commPageBaseOffset));
}
/* Determine number of CPUs on this system. We cannot rely on
static void
commpage_init_cpu_capabilities( void )
{
- int bits;
+ uint32_t bits;
int cpus;
ml_cpu_info_t cpu_info;
ml_cpu_get_info(&cpu_info);
switch (cpu_info.vector_unit) {
+ case 9:
+ bits |= kHasAVX1_0;
+ /* fall thru */
+ case 8:
+ bits |= kHasSSE4_2;
+ /* fall thru */
+ case 7:
+ bits |= kHasSSE4_1;
+ /* fall thru */
case 6:
bits |= kHasSupplementalSSE3;
/* fall thru */
if (cpu_mode_is64bit()) // k64Bit means processor is 64-bit capable
bits |= k64Bit;
+ if (tscFreq <= SLOW_TSC_THRESHOLD) /* is TSC too slow for _commpage_nanotime? */
+ bits |= kSlow;
+
+ if (cpuid_features() & CPUID_FEATURE_AES)
+ bits |= kHasAES;
+
_cpu_capabilities = bits; // set kernel version for use by drivers etc
}
int
-_get_cpu_capabilities()
+_get_cpu_capabilities(void)
{
return _cpu_capabilities;
}
static void
commpage_stuff(
- int address,
+ commpage_address_t address,
const void *source,
int length )
{
void *dest = commpage_addr_of(address);
- if ((uintptr_t)dest < next)
- panic("commpage overlap at address 0x%x, 0x%x < 0x%x", address, dest, next);
+ if (address < next)
+ panic("commpage overlap at address 0x%p, 0x%x < 0x%x", dest, address, next);
bcopy(source,dest,length);
- next = ((uintptr_t)dest + length);
-}
-
-static void
-commpage_stuff_swap(
- int address,
- void *source,
- int length,
- int legacy )
-{
- if ( legacy ) {
- void *dest = commpage_addr_of(address);
- dest = (void *)((uintptr_t) dest + _COMM_PAGE_SIGS_OFFSET);
- switch (length) {
- case 2:
- OSWriteSwapInt16(dest, 0, *(uint16_t *)source);
- break;
- case 4:
- OSWriteSwapInt32(dest, 0, *(uint32_t *)source);
- break;
- case 8:
- OSWriteSwapInt64(dest, 0, *(uint64_t *)source);
- break;
- }
- }
-}
-
-static void
-commpage_stuff2(
- int address,
- void *source,
- int length,
- int legacy )
-{
- commpage_stuff_swap(address, source, length, legacy);
- commpage_stuff(address, source, length);
+ next = address + length;
}
/* Copy a routine into comm page if it matches running machine.
commpage_stuff_routine(
commpage_descriptor *rd )
{
- int must,cant;
+ uint32_t must,cant;
if (rd->commpage_address != cur_routine) {
if ((cur_routine!=0) && (matched==0))
}
/* Fill in the 32- or 64-bit commpage. Called once for each.
- * The 32-bit ("legacy") commpage has a bunch of stuff added to it
- * for translated processes, some of which is byte-swapped.
*/
static void
commpage_populate_one(
- vm_map_t submap, // com_region_map32 or com_region_map64
+ vm_map_t submap, // commpage32_map or compage64_map
char ** kernAddressPtr, // &commPagePtr32 or &commPagePtr64
size_t area_used, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
- size_t base_offset, // will become commPageBaseOffset
+ commpage_address_t base_offset, // will become commPageBaseOffset
commpage_descriptor** commpage_routines, // list of routine ptrs for this commpage
- boolean_t legacy, // true if 32-bit commpage
+ commpage_time_data** time_data, // &time_data32 or &time_data64
const char* signature ) // "commpage 32-bit" or "commpage 64-bit"
{
+ uint8_t c1;
short c2;
- static double two52 = 1048576.0 * 1048576.0 * 4096.0; // 2**52
- static double ten6 = 1000000.0; // 10**6
+ int c4;
+ uint64_t c8;
+ uint32_t cfamily;
commpage_descriptor **rd;
short version = _COMM_PAGE_THIS_VERSION;
- int swapcaps;
- next = (uintptr_t) NULL;
+ next = 0;
cur_routine = 0;
commPagePtr = (char *)commpage_allocate( submap, (vm_size_t) area_used );
*kernAddressPtr = commPagePtr; // save address either in commPagePtr32 or 64
commPageBaseOffset = base_offset;
+ *time_data = commpage_addr_of( _COMM_PAGE_TIME_DATA_START );
+
/* Stuff in the constants. We move things into the comm page in strictly
* ascending order, so we can check for overlap and panic if so.
*/
- commpage_stuff(_COMM_PAGE_SIGNATURE,signature,strlen(signature));
- commpage_stuff2(_COMM_PAGE_VERSION,&version,sizeof(short),legacy);
+ commpage_stuff(_COMM_PAGE_SIGNATURE,signature,(int)strlen(signature));
+ commpage_stuff(_COMM_PAGE_VERSION,&version,sizeof(short));
commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES,&_cpu_capabilities,sizeof(int));
- /* excuse our magic constants, we cannot include ppc/cpu_capabilities.h */
- /* always set kCache32 and kDcbaAvailable */
- swapcaps = 0x44;
- if ( _cpu_capabilities & kUP )
- swapcaps |= (kUP + (1 << kNumCPUsShift));
- else
- swapcaps |= 2 << kNumCPUsShift; /* limit #cpus to 2 */
- if ( ! noVMX ) /* if rosetta will be emulating altivec... */
- swapcaps |= 0x101; /* ...then set kHasAltivec and kDataStreamsAvailable too */
- commpage_stuff_swap(_COMM_PAGE_CPU_CAPABILITIES, &swapcaps, sizeof(int), legacy);
- c2 = 32;
- commpage_stuff_swap(_COMM_PAGE_CACHE_LINESIZE,&c2,2,legacy);
-
- if (_cpu_capabilities & kCache32)
- c2 = 32;
- else if (_cpu_capabilities & kCache64)
+ c2 = 32; // default
+ if (_cpu_capabilities & kCache64)
c2 = 64;
else if (_cpu_capabilities & kCache128)
c2 = 128;
commpage_stuff(_COMM_PAGE_CACHE_LINESIZE,&c2,2);
+
+ c4 = MP_SPIN_TRIES;
+ commpage_stuff(_COMM_PAGE_SPIN_COUNT,&c4,4);
- if ( legacy ) {
- commpage_stuff2(_COMM_PAGE_2_TO_52,&two52,8,legacy);
- commpage_stuff2(_COMM_PAGE_10_TO_6,&ten6,8,legacy);
- }
+ /* machine_info valid after ml_get_max_cpus() */
+ c1 = machine_info.physical_cpu_max;
+ commpage_stuff(_COMM_PAGE_PHYSICAL_CPUS,&c1,1);
+ c1 = machine_info.logical_cpu_max;
+ commpage_stuff(_COMM_PAGE_LOGICAL_CPUS,&c1,1);
+
+ c8 = ml_cpu_cache_size(0);
+ commpage_stuff(_COMM_PAGE_MEMORY_SIZE, &c8, 8);
+
+ cfamily = cpuid_info()->cpuid_cpufamily;
+ commpage_stuff(_COMM_PAGE_CPUFAMILY, &cfamily, 4);
for( rd = commpage_routines; *rd != NULL ; rd++ )
commpage_stuff_routine(*rd);
if (!matched)
panic("commpage no match on last routine");
- if (next > (uintptr_t)_COMM_PAGE_END)
- panic("commpage overflow: next = 0x%08x, commPagePtr = 0x%08x", next, (uintptr_t)commPagePtr);
+ if (next > _COMM_PAGE_END)
+ panic("commpage overflow: next = 0x%08x, commPagePtr = 0x%p", next, commPagePtr);
- if ( legacy ) {
- next = (uintptr_t) NULL;
- for( rd = ba_descriptors; *rd != NULL ; rd++ )
- commpage_stuff_routine(*rd);
-
- next = (uintptr_t) NULL;
- commpage_stuff_routine(&sigdata_descriptor);
- }
-
- /* salt away a ptr to the system integrity data in this commpage */
- dsmos_blobs[dsmos_blob_count++] =
- commpage_addr_of( _COMM_PAGE_SYSTEM_INTEGRITY );
}
{
commpage_init_cpu_capabilities();
- commpage_populate_one( com_region_map32,
+ commpage_populate_one( commpage32_map,
&commPagePtr32,
_COMM_PAGE32_AREA_USED,
_COMM_PAGE32_BASE_ADDRESS,
commpage_32_routines,
- TRUE, /* legacy (32-bit) commpage */
+ &time_data32,
"commpage 32-bit");
+#ifndef __LP64__
pmap_commpage32_init((vm_offset_t) commPagePtr32, _COMM_PAGE32_BASE_ADDRESS,
_COMM_PAGE32_AREA_USED/INTEL_PGBYTES);
+#endif
+ time_data64 = time_data32; /* if no 64-bit commpage, point to 32-bit */
if (_cpu_capabilities & k64Bit) {
- commpage_populate_one( com_region_map64,
+ commpage_populate_one( commpage64_map,
&commPagePtr64,
_COMM_PAGE64_AREA_USED,
- _COMM_PAGE32_START_ADDRESS, /* because kernel is built 32-bit */
+ _COMM_PAGE32_START_ADDRESS, /* commpage address are relative to 32-bit commpage placement */
commpage_64_routines,
- FALSE, /* not a legacy commpage */
+ &time_data64,
"commpage 64-bit");
+#ifndef __LP64__
pmap_commpage64_init((vm_offset_t) commPagePtr64, _COMM_PAGE64_BASE_ADDRESS,
_COMM_PAGE64_AREA_USED/INTEL_PGBYTES);
+#endif
}
+ simple_lock_init(&commpage_active_cpus_lock, 0);
+
+ commpage_update_active_cpus();
rtc_nanotime_init_commpage();
}
+
+
+/* Update commpage nanotime information. Note that we interleave
+ * setting the 32- and 64-bit commpages, in order to keep nanotime more
+ * nearly in sync between the two environments.
+ *
+ * This routine must be serialized by some external means, ie a lock.
+ */
+
+void
+commpage_set_nanotime(
+ uint64_t tsc_base,
+ uint64_t ns_base,
+ uint32_t scale,
+ uint32_t shift )
+{
+ commpage_time_data *p32 = time_data32;
+ commpage_time_data *p64 = time_data64;
+ static uint32_t generation = 0;
+ uint32_t next_gen;
+
+ if (p32 == NULL) /* have commpages been allocated yet? */
+ return;
+
+ if ( generation != p32->nt_generation )
+ panic("nanotime trouble 1"); /* possibly not serialized */
+ if ( ns_base < p32->nt_ns_base )
+ panic("nanotime trouble 2");
+ if ((shift != 32) && ((_cpu_capabilities & kSlow)==0) )
+ panic("nanotime trouble 3");
+
+ next_gen = ++generation;
+ if (next_gen == 0)
+ next_gen = ++generation;
+
+ p32->nt_generation = 0; /* mark invalid, so commpage won't try to use it */
+ p64->nt_generation = 0;
+
+ p32->nt_tsc_base = tsc_base;
+ p64->nt_tsc_base = tsc_base;
+
+ p32->nt_ns_base = ns_base;
+ p64->nt_ns_base = ns_base;
+
+ p32->nt_scale = scale;
+ p64->nt_scale = scale;
+
+ p32->nt_shift = shift;
+ p64->nt_shift = shift;
+
+ p32->nt_generation = next_gen; /* mark data as valid */
+ p64->nt_generation = next_gen;
+}
+
+
+/* Disable commpage gettimeofday(), forcing commpage to call through to the kernel. */
+
+void
+commpage_disable_timestamp( void )
+{
+ time_data32->gtod_generation = 0;
+ time_data64->gtod_generation = 0;
+}
+
+
+/* Update commpage gettimeofday() information. As with nanotime(), we interleave
+ * updates to the 32- and 64-bit commpage, in order to keep time more nearly in sync
+ * between the two environments.
+ *
+ * This routine must be serializeed by some external means, ie a lock.
+ */
+
+ void
+ commpage_set_timestamp(
+ uint64_t abstime,
+ uint64_t secs )
+{
+ commpage_time_data *p32 = time_data32;
+ commpage_time_data *p64 = time_data64;
+ static uint32_t generation = 0;
+ uint32_t next_gen;
+
+ next_gen = ++generation;
+ if (next_gen == 0)
+ next_gen = ++generation;
+
+ p32->gtod_generation = 0; /* mark invalid, so commpage won't try to use it */
+ p64->gtod_generation = 0;
+
+ p32->gtod_ns_base = abstime;
+ p64->gtod_ns_base = abstime;
+
+ p32->gtod_sec_base = secs;
+ p64->gtod_sec_base = secs;
+
+ p32->gtod_generation = next_gen; /* mark data as valid */
+ p64->gtod_generation = next_gen;
+}
+
+
+/* Update _COMM_PAGE_MEMORY_PRESSURE. Called periodically from vm's compute_memory_pressure() */
+
+void
+commpage_set_memory_pressure(
+ unsigned int pressure )
+{
+ char *cp;
+ uint32_t *ip;
+
+ cp = commPagePtr32;
+ if ( cp ) {
+ cp += (_COMM_PAGE_MEMORY_PRESSURE - _COMM_PAGE32_BASE_ADDRESS);
+ ip = (uint32_t*) cp;
+ *ip = (uint32_t) pressure;
+ }
+
+ cp = commPagePtr64;
+ if ( cp ) {
+ cp += (_COMM_PAGE_MEMORY_PRESSURE - _COMM_PAGE32_START_ADDRESS);
+ ip = (uint32_t*) cp;
+ *ip = (uint32_t) pressure;
+ }
+
+}
+
+
+/* Update _COMM_PAGE_SPIN_COUNT. We might want to reduce when running on a battery, etc. */
+
+void
+commpage_set_spin_count(
+ unsigned int count )
+{
+ char *cp;
+ uint32_t *ip;
+
+ if (count == 0) /* we test for 0 after decrement, not before */
+ count = 1;
+
+ cp = commPagePtr32;
+ if ( cp ) {
+ cp += (_COMM_PAGE_SPIN_COUNT - _COMM_PAGE32_BASE_ADDRESS);
+ ip = (uint32_t*) cp;
+ *ip = (uint32_t) count;
+ }
+
+ cp = commPagePtr64;
+ if ( cp ) {
+ cp += (_COMM_PAGE_SPIN_COUNT - _COMM_PAGE32_START_ADDRESS);
+ ip = (uint32_t*) cp;
+ *ip = (uint32_t) count;
+ }
+
+}
+
+/* Updated every time a logical CPU goes offline/online */
+void
+commpage_update_active_cpus(void)
+{
+ char *cp;
+ volatile uint8_t *ip;
+
+ /* At least 32-bit commpage must be initialized */
+ if (!commPagePtr32)
+ return;
+
+ simple_lock(&commpage_active_cpus_lock);
+
+ cp = commPagePtr32;
+ cp += (_COMM_PAGE_ACTIVE_CPUS - _COMM_PAGE32_BASE_ADDRESS);
+ ip = (volatile uint8_t*) cp;
+ *ip = (uint8_t) processor_avail_count;
+
+ cp = commPagePtr64;
+ if ( cp ) {
+ cp += (_COMM_PAGE_ACTIVE_CPUS - _COMM_PAGE32_START_ADDRESS);
+ ip = (volatile uint8_t*) cp;
+ *ip = (uint8_t) processor_avail_count;
+ }
+
+ simple_unlock(&commpage_active_cpus_lock);
+}
+
+
+/* Check to see if a given address is in the Preemption Free Zone (PFZ) */
+
+uint32_t
+commpage_is_in_pfz32(uint32_t addr32)
+{
+ if ( (addr32 >= _COMM_PAGE_PFZ_START) && (addr32 < _COMM_PAGE_PFZ_END)) {
+ return 1;
+ }
+ else
+ return 0;
+}
+
+uint32_t
+commpage_is_in_pfz64(addr64_t addr64)
+{
+ if ( (addr64 >= _COMM_PAGE_32_TO_64(_COMM_PAGE_PFZ_START))
+ && (addr64 < _COMM_PAGE_32_TO_64(_COMM_PAGE_PFZ_END))) {
+ return 1;
+ }
+ else
+ return 0;
+}
+
projects / apple / xnu.git / blobdiff