2 * Copyright (c) 2003-2010 Apple Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
30 * Here's what to do if you want to add a new routine to the comm page:
32 * 1. Add a definition for it's address in osfmk/i386/cpu_capabilities.h,
33 * being careful to reserve room for future expansion.
35 * 2. Write one or more versions of the routine, each with it's own
36 * commpage_descriptor. The tricky part is getting the "special",
37 * "musthave", and "canthave" fields right, so that exactly one
38 * version of the routine is selected for every machine.
39 * The source files should be in osfmk/i386/commpage/.
41 * 3. Add a ptr to your new commpage_descriptor(s) in the "routines"
42 * array in osfmk/i386/commpage/commpage_asm.s. There are two
43 * arrays, one for the 32-bit and one for the 64-bit commpage.
45 * 4. Write the code in Libc to use the new routine.
48 #include <mach/mach_types.h>
49 #include <mach/machine.h>
50 #include <mach/vm_map.h>
51 #include <mach/mach_vm.h>
52 #include <mach/machine.h>
53 #include <i386/cpuid.h>
55 #include <i386/rtclock_protos.h>
56 #include <i386/cpu_data.h>
57 #include <i386/machine_routines.h>
58 #include <i386/misc_protos.h>
59 #include <i386/cpuid.h>
60 #include <machine/cpu_capabilities.h>
61 #include <machine/commpage.h>
62 #include <machine/pmap.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_map.h>
66 #include <ipc/ipc_port.h>
68 #include <kern/page_decrypt.h>
69 #include <kern/processor.h>
71 /* the lists of commpage routines are in commpage_asm.s */
72 extern commpage_descriptor
* commpage_32_routines
[];
73 extern commpage_descriptor
* commpage_64_routines
[];
75 extern vm_map_t commpage32_map
; // the shared submap, set up in vm init
76 extern vm_map_t commpage64_map
; // the shared submap, set up in vm init
77 extern vm_map_t commpage_text32_map
; // the shared submap, set up in vm init
78 extern vm_map_t commpage_text64_map
; // the shared submap, set up in vm init
81 char *commPagePtr32
= NULL
; // virtual addr in kernel map of 32-bit commpage
82 char *commPagePtr64
= NULL
; // ...and of 64-bit commpage
83 char *commPageTextPtr32
= NULL
; // virtual addr in kernel map of 32-bit commpage
84 char *commPageTextPtr64
= NULL
; // ...and of 64-bit commpage
85 uint32_t _cpu_capabilities
= 0; // define the capability vector
87 int noVMX
= 0; /* if true, do not set kHasAltivec in ppc _cpu_capabilities */
89 typedef uint32_t commpage_address_t
;
91 static commpage_address_t next
; // next available address in comm page
92 static commpage_address_t cur_routine
; // comm page address of "current" routine
93 static boolean_t matched
; // true if we've found a match for "current" routine
95 static char *commPagePtr
; // virtual addr in kernel map of commpage we are working on
96 static commpage_address_t commPageBaseOffset
; // subtract from 32-bit runtime address to get offset in virtual commpage in kernel map
98 static commpage_time_data
*time_data32
= NULL
;
99 static commpage_time_data
*time_data64
= NULL
;
101 decl_simple_lock_data(static,commpage_active_cpus_lock
);
103 /* Allocate the commpage and add to the shared submap created by vm:
104 * 1. allocate a page in the kernel map (RW)
106 * 3. make a memory entry out of it
107 * 4. map that entry into the shared comm region map (R-only)
112 vm_map_t submap
, // commpage32_map or commpage_map64
113 size_t area_used
, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
116 vm_offset_t kernel_addr
= 0; // address of commpage in kernel map
117 vm_offset_t zero
= 0;
118 vm_size_t size
= area_used
; // size actually populated
119 vm_map_entry_t entry
;
124 panic("commpage submap is null");
126 if ((kr
= vm_map(kernel_map
,&kernel_addr
,area_used
,0,VM_FLAGS_ANYWHERE
,NULL
,0,FALSE
,VM_PROT_ALL
,VM_PROT_ALL
,VM_INHERIT_NONE
)))
127 panic("cannot allocate commpage %d", kr
);
129 if ((kr
= vm_map_wire(kernel_map
,kernel_addr
,kernel_addr
+area_used
,VM_PROT_DEFAULT
,FALSE
)))
130 panic("cannot wire commpage: %d", kr
);
133 * Now that the object is created and wired into the kernel map, mark it so that no delay
134 * copy-on-write will ever be performed on it as a result of mapping it into user-space.
135 * If such a delayed copy ever occurred, we could remove the kernel's wired mapping - and
136 * that would be a real disaster.
138 * JMM - What we really need is a way to create it like this in the first place.
140 if (!(kr
= vm_map_lookup_entry( kernel_map
, vm_map_trunc_page(kernel_addr
), &entry
) || entry
->is_sub_map
))
141 panic("cannot find commpage entry %d", kr
);
142 entry
->object
.vm_object
->copy_strategy
= MEMORY_OBJECT_COPY_NONE
;
144 if ((kr
= mach_make_memory_entry( kernel_map
, // target map
146 kernel_addr
, // offset (address in kernel map)
147 uperm
, // protections as specified
148 &handle
, // this is the object handle we get
149 NULL
))) // parent_entry (what is this?)
150 panic("cannot make entry for commpage %d", kr
);
152 if ((kr
= vm_map_64( submap
, // target map (shared submap)
153 &zero
, // address (map into 1st page in submap)
156 VM_FLAGS_FIXED
, // flags (it must be 1st page in submap)
157 handle
, // port is the memory entry we just made
158 0, // offset (map 1st page in memory entry)
160 uperm
, // cur_protection (R-only in user map)
161 uperm
, // max_protection
162 VM_INHERIT_SHARE
))) // inheritance
163 panic("cannot map commpage %d", kr
);
165 ipc_port_release(handle
);
166 /* Make the kernel mapping non-executable. This cannot be done
167 * at the time of map entry creation as mach_make_memory_entry
168 * cannot handle disjoint permissions at this time.
170 kr
= vm_protect(kernel_map
, kernel_addr
, area_used
, FALSE
, VM_PROT_READ
| VM_PROT_WRITE
);
171 assert (kr
== KERN_SUCCESS
);
173 return (void*)(intptr_t)kernel_addr
; // return address in kernel map
176 /* Get address (in kernel map) of a commpage field. */
180 commpage_address_t addr_at_runtime
)
182 return (void*) ((uintptr_t)commPagePtr
+ (addr_at_runtime
- commPageBaseOffset
));
185 /* Determine number of CPUs on this system. We cannot rely on
186 * machine_info.max_cpus this early in the boot.
189 commpage_cpus( void )
193 cpus
= ml_get_max_cpus(); // NB: this call can block
196 panic("commpage cpus==0");
203 /* Initialize kernel version of _cpu_capabilities vector (used by KEXTs.) */
206 commpage_init_cpu_capabilities( void )
210 ml_cpu_info_t cpu_info
;
213 ml_cpu_get_info(&cpu_info
);
215 switch (cpu_info
.vector_unit
) {
226 bits
|= kHasSupplementalSSE3
;
242 switch (cpu_info
.cache_line_size
) {
255 cpus
= commpage_cpus(); // how many CPUs do we have
260 bits
|= (cpus
<< kNumCPUsShift
);
262 bits
|= kFastThreadLocalStorage
; // we use %gs for TLS
264 if (cpu_mode_is64bit()) // k64Bit means processor is 64-bit capable
267 if (tscFreq
<= SLOW_TSC_THRESHOLD
) /* is TSC too slow for _commpage_nanotime? */
270 bits
|= (cpuid_features() & CPUID_FEATURE_AES
) ? kHasAES
: 0;
272 bits
|= (cpuid_features() & CPUID_FEATURE_F16C
) ? kHasF16C
: 0;
273 bits
|= (cpuid_features() & CPUID_FEATURE_RDRAND
) ? kHasRDRAND
: 0;
274 bits
|= ((cpuid_leaf7_features() & CPUID_LEAF7_FEATURE_ENFSTRG
) &&
275 (rdmsr64(MSR_IA32_MISC_ENABLE
) & 1ULL )) ? kHasENFSTRG
: 0;
277 _cpu_capabilities
= bits
; // set kernel version for use by drivers etc
281 _get_cpu_capabilities(void)
283 return _cpu_capabilities
;
286 /* Copy data into commpage. */
290 commpage_address_t address
,
294 void *dest
= commpage_addr_of(address
);
297 panic("commpage overlap at address 0x%p, 0x%x < 0x%x", dest
, address
, next
);
299 bcopy(source
,dest
,length
);
301 next
= address
+ length
;
304 /* Copy a routine into comm page if it matches running machine.
307 commpage_stuff_routine(
308 commpage_descriptor
*rd
)
312 if (rd
->commpage_address
!= cur_routine
) {
313 if ((cur_routine
!=0) && (matched
==0))
314 panic("commpage no match for last, next address %08x", rd
->commpage_address
);
315 cur_routine
= rd
->commpage_address
;
319 must
= _cpu_capabilities
& rd
->musthave
;
320 cant
= _cpu_capabilities
& rd
->canthave
;
322 if ((must
== rd
->musthave
) && (cant
== 0)) {
324 panic("commpage multiple matches for address %08x", rd
->commpage_address
);
327 commpage_stuff(rd
->commpage_address
,rd
->code_address
,rd
->code_length
);
331 /* Fill in the 32- or 64-bit commpage. Called once for each.
335 commpage_populate_one(
336 vm_map_t submap
, // commpage32_map or compage64_map
337 char ** kernAddressPtr
, // &commPagePtr32 or &commPagePtr64
338 size_t area_used
, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
339 commpage_address_t base_offset
, // will become commPageBaseOffset
340 commpage_time_data
** time_data
, // &time_data32 or &time_data64
341 const char* signature
, // "commpage 32-bit" or "commpage 64-bit"
349 short version
= _COMM_PAGE_THIS_VERSION
;
353 commPagePtr
= (char *)commpage_allocate( submap
, (vm_size_t
) area_used
, uperm
);
354 *kernAddressPtr
= commPagePtr
; // save address either in commPagePtr32 or 64
355 commPageBaseOffset
= base_offset
;
357 *time_data
= commpage_addr_of( _COMM_PAGE_TIME_DATA_START
);
359 /* Stuff in the constants. We move things into the comm page in strictly
360 * ascending order, so we can check for overlap and panic if so.
362 commpage_stuff(_COMM_PAGE_SIGNATURE
,signature
,(int)strlen(signature
));
363 commpage_stuff(_COMM_PAGE_VERSION
,&version
,sizeof(short));
364 commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES
,&_cpu_capabilities
,sizeof(int));
367 if (_cpu_capabilities
& kCache64
)
369 else if (_cpu_capabilities
& kCache128
)
371 commpage_stuff(_COMM_PAGE_CACHE_LINESIZE
,&c2
,2);
374 commpage_stuff(_COMM_PAGE_SPIN_COUNT
,&c4
,4);
376 /* machine_info valid after ml_get_max_cpus() */
377 c1
= machine_info
.physical_cpu_max
;
378 commpage_stuff(_COMM_PAGE_PHYSICAL_CPUS
,&c1
,1);
379 c1
= machine_info
.logical_cpu_max
;
380 commpage_stuff(_COMM_PAGE_LOGICAL_CPUS
,&c1
,1);
382 c8
= ml_cpu_cache_size(0);
383 commpage_stuff(_COMM_PAGE_MEMORY_SIZE
, &c8
, 8);
385 cfamily
= cpuid_info()->cpuid_cpufamily
;
386 commpage_stuff(_COMM_PAGE_CPUFAMILY
, &cfamily
, 4);
388 if (next
> _COMM_PAGE_END
)
389 panic("commpage overflow: next = 0x%08x, commPagePtr = 0x%p", next
, commPagePtr
);
394 /* Fill in commpages: called once, during kernel initialization, from the
395 * startup thread before user-mode code is running.
397 * See the top of this file for a list of what you have to do to add
398 * a new routine to the commpage.
402 commpage_populate( void )
404 commpage_init_cpu_capabilities();
406 commpage_populate_one( commpage32_map
,
408 _COMM_PAGE32_AREA_USED
,
409 _COMM_PAGE32_BASE_ADDRESS
,
414 pmap_commpage32_init((vm_offset_t
) commPagePtr32
, _COMM_PAGE32_BASE_ADDRESS
,
415 _COMM_PAGE32_AREA_USED
/INTEL_PGBYTES
);
417 time_data64
= time_data32
; /* if no 64-bit commpage, point to 32-bit */
419 if (_cpu_capabilities
& k64Bit
) {
420 commpage_populate_one( commpage64_map
,
422 _COMM_PAGE64_AREA_USED
,
423 _COMM_PAGE32_START_ADDRESS
, /* commpage address are relative to 32-bit commpage placement */
428 pmap_commpage64_init((vm_offset_t
) commPagePtr64
, _COMM_PAGE64_BASE_ADDRESS
,
429 _COMM_PAGE64_AREA_USED
/INTEL_PGBYTES
);
433 simple_lock_init(&commpage_active_cpus_lock
, 0);
435 commpage_update_active_cpus();
436 rtc_nanotime_init_commpage();
439 /* Fill in the common routines during kernel initialization.
440 * This is called before user-mode code is running.
442 void commpage_text_populate( void ){
443 commpage_descriptor
**rd
;
447 commPagePtr
= (char *) commpage_allocate(commpage_text32_map
, (vm_size_t
) _COMM_PAGE_TEXT_AREA_USED
, VM_PROT_READ
| VM_PROT_EXECUTE
);
448 commPageTextPtr32
= commPagePtr
;
450 char *cptr
= commPagePtr
;
452 for(; i
< _COMM_PAGE_TEXT_AREA_USED
; i
++){
456 commPageBaseOffset
= _COMM_PAGE_TEXT_START
;
457 for (rd
= commpage_32_routines
; *rd
!= NULL
; rd
++) {
458 commpage_stuff_routine(*rd
);
461 panic(" commpage_text no match for last routine ");
464 pmap_commpage32_init((vm_offset_t
) commPageTextPtr32
, _COMM_PAGE_TEXT_START
,
465 _COMM_PAGE_TEXT_AREA_USED
/INTEL_PGBYTES
);
468 if (_cpu_capabilities
& k64Bit
) {
471 commPagePtr
= (char *) commpage_allocate(commpage_text64_map
, (vm_size_t
) _COMM_PAGE_TEXT_AREA_USED
, VM_PROT_READ
| VM_PROT_EXECUTE
);
472 commPageTextPtr64
= commPagePtr
;
475 for(i
=0; i
<_COMM_PAGE_TEXT_AREA_USED
; i
++){
479 for (rd
= commpage_64_routines
; *rd
!=NULL
; rd
++) {
480 commpage_stuff_routine(*rd
);
484 pmap_commpage64_init((vm_offset_t
) commPageTextPtr64
, _COMM_PAGE_TEXT_START
,
485 _COMM_PAGE_TEXT_AREA_USED
/INTEL_PGBYTES
);
490 panic(" commpage_text no match for last routine ");
492 if (next
> _COMM_PAGE_TEXT_END
)
493 panic("commpage text overflow: next=0x%08x, commPagePtr=%p", next
, commPagePtr
);
497 /* Update commpage nanotime information. Note that we interleave
498 * setting the 32- and 64-bit commpages, in order to keep nanotime more
499 * nearly in sync between the two environments.
501 * This routine must be serialized by some external means, ie a lock.
505 commpage_set_nanotime(
511 commpage_time_data
*p32
= time_data32
;
512 commpage_time_data
*p64
= time_data64
;
513 static uint32_t generation
= 0;
516 if (p32
== NULL
) /* have commpages been allocated yet? */
519 if ( generation
!= p32
->nt_generation
)
520 panic("nanotime trouble 1"); /* possibly not serialized */
521 if ( ns_base
< p32
->nt_ns_base
)
522 panic("nanotime trouble 2");
523 if ((shift
!= 32) && ((_cpu_capabilities
& kSlow
)==0) )
524 panic("nanotime trouble 3");
526 next_gen
= ++generation
;
528 next_gen
= ++generation
;
530 p32
->nt_generation
= 0; /* mark invalid, so commpage won't try to use it */
531 p64
->nt_generation
= 0;
533 p32
->nt_tsc_base
= tsc_base
;
534 p64
->nt_tsc_base
= tsc_base
;
536 p32
->nt_ns_base
= ns_base
;
537 p64
->nt_ns_base
= ns_base
;
539 p32
->nt_scale
= scale
;
540 p64
->nt_scale
= scale
;
542 p32
->nt_shift
= shift
;
543 p64
->nt_shift
= shift
;
545 p32
->nt_generation
= next_gen
; /* mark data as valid */
546 p64
->nt_generation
= next_gen
;
550 /* Disable commpage gettimeofday(), forcing commpage to call through to the kernel. */
553 commpage_disable_timestamp( void )
555 time_data32
->gtod_generation
= 0;
556 time_data64
->gtod_generation
= 0;
560 /* Update commpage gettimeofday() information. As with nanotime(), we interleave
561 * updates to the 32- and 64-bit commpage, in order to keep time more nearly in sync
562 * between the two environments.
564 * This routine must be serializeed by some external means, ie a lock.
568 commpage_set_timestamp(
572 commpage_time_data
*p32
= time_data32
;
573 commpage_time_data
*p64
= time_data64
;
574 static uint32_t generation
= 0;
577 next_gen
= ++generation
;
579 next_gen
= ++generation
;
581 p32
->gtod_generation
= 0; /* mark invalid, so commpage won't try to use it */
582 p64
->gtod_generation
= 0;
584 p32
->gtod_ns_base
= abstime
;
585 p64
->gtod_ns_base
= abstime
;
587 p32
->gtod_sec_base
= secs
;
588 p64
->gtod_sec_base
= secs
;
590 p32
->gtod_generation
= next_gen
; /* mark data as valid */
591 p64
->gtod_generation
= next_gen
;
595 /* Update _COMM_PAGE_MEMORY_PRESSURE. Called periodically from vm's compute_memory_pressure() */
598 commpage_set_memory_pressure(
599 unsigned int pressure
)
606 cp
+= (_COMM_PAGE_MEMORY_PRESSURE
- _COMM_PAGE32_BASE_ADDRESS
);
608 *ip
= (uint32_t) pressure
;
613 cp
+= (_COMM_PAGE_MEMORY_PRESSURE
- _COMM_PAGE32_START_ADDRESS
);
615 *ip
= (uint32_t) pressure
;
621 /* Update _COMM_PAGE_SPIN_COUNT. We might want to reduce when running on a battery, etc. */
624 commpage_set_spin_count(
630 if (count
== 0) /* we test for 0 after decrement, not before */
635 cp
+= (_COMM_PAGE_SPIN_COUNT
- _COMM_PAGE32_BASE_ADDRESS
);
637 *ip
= (uint32_t) count
;
642 cp
+= (_COMM_PAGE_SPIN_COUNT
- _COMM_PAGE32_START_ADDRESS
);
644 *ip
= (uint32_t) count
;
649 /* Updated every time a logical CPU goes offline/online */
651 commpage_update_active_cpus(void)
654 volatile uint8_t *ip
;
656 /* At least 32-bit commpage must be initialized */
660 simple_lock(&commpage_active_cpus_lock
);
663 cp
+= (_COMM_PAGE_ACTIVE_CPUS
- _COMM_PAGE32_BASE_ADDRESS
);
664 ip
= (volatile uint8_t*) cp
;
665 *ip
= (uint8_t) processor_avail_count
;
669 cp
+= (_COMM_PAGE_ACTIVE_CPUS
- _COMM_PAGE32_START_ADDRESS
);
670 ip
= (volatile uint8_t*) cp
;
671 *ip
= (uint8_t) processor_avail_count
;
674 simple_unlock(&commpage_active_cpus_lock
);
677 extern user32_addr_t commpage_text32_location
;
678 extern user64_addr_t commpage_text64_location
;
680 /* Check to see if a given address is in the Preemption Free Zone (PFZ) */
683 commpage_is_in_pfz32(uint32_t addr32
)
685 if ( (addr32
>= (commpage_text32_location
+ _COMM_TEXT_PFZ_START_OFFSET
))
686 && (addr32
< (commpage_text32_location
+_COMM_TEXT_PFZ_END_OFFSET
))) {
694 commpage_is_in_pfz64(addr64_t addr64
)
696 if ( (addr64
>= (commpage_text64_location
+ _COMM_TEXT_PFZ_START_OFFSET
))
697 && (addr64
< (commpage_text64_location
+ _COMM_TEXT_PFZ_END_OFFSET
))) {