1/*
2 * Copyright (c) 2003-2007 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
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.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
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.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28
29/*
30 *	Here's what to do if you want to add a new routine to the comm page:
31 *
32 *		1. Add a definition for it's address in osfmk/i386/cpu_capabilities.h,
33 *		   being careful to reserve room for future expansion.
34 *
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/.
40 *
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.
44 *
45 *		4. Write the code in Libc to use the new routine.
46 */
47
48#include <mach/mach_types.h>
49#include <mach/machine.h>
50#include <mach/vm_map.h>
51#include <i386/machine_routines.h>
52#include <i386/misc_protos.h>
53#include <i386/tsc.h>
54#include <i386/cpu_data.h>
55#include <machine/cpu_capabilities.h>
56#include <machine/commpage.h>
57#include <machine/pmap.h>
58#include <vm/vm_kern.h>
59#include <vm/vm_map.h>
60#include <ipc/ipc_port.h>
61
62#include <kern/page_decrypt.h>
63
64/* the lists of commpage routines are in commpage_asm.s  */
65extern	commpage_descriptor*	commpage_32_routines[];
66extern	commpage_descriptor*	commpage_64_routines[];
67
68/* translated commpage descriptors from commpage_sigs.c  */
69extern	commpage_descriptor sigdata_descriptor;
70extern	commpage_descriptor *ba_descriptors[];
71
72extern vm_map_t	commpage32_map;	// the shared submap, set up in vm init
73extern vm_map_t	commpage64_map;	// the shared submap, set up in vm init
74
75char	*commPagePtr32 = NULL;		// virtual addr in kernel map of 32-bit commpage
76char	*commPagePtr64 = NULL;		// ...and of 64-bit commpage
77int     _cpu_capabilities = 0;          // define the capability vector
78
79int	noVMX = 0;		/* if true, do not set kHasAltivec in ppc _cpu_capabilities */
80
81static uintptr_t next;			// next available byte in comm page
82static int     	cur_routine;		// comm page address of "current" routine
83static int     	matched;		// true if we've found a match for "current" routine
84
85static char    *commPagePtr;		// virtual addr in kernel map of commpage we are working on
86static size_t	commPageBaseOffset;	// add to 32-bit runtime address to get offset in commpage
87
88static	commpage_time_data	*time_data32 = NULL;
89static	commpage_time_data	*time_data64 = NULL;
90
91/* Allocate the commpage and add to the shared submap created by vm:
92 * 	1. allocate a page in the kernel map (RW)
93 *	2. wire it down
94 *	3. make a memory entry out of it
95 *	4. map that entry into the shared comm region map (R-only)
96 */
97
98static  void*
99commpage_allocate(
100	vm_map_t	submap,			// commpage32_map or commpage_map64
101	size_t		area_used )		// _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
102{
103	vm_offset_t	kernel_addr = 0;	// address of commpage in kernel map
104	vm_offset_t	zero = 0;
105	vm_size_t	size = area_used;	// size actually populated
106	vm_map_entry_t	entry;
107	ipc_port_t	handle;
108
109	if (submap == NULL)
110		panic("commpage submap is null");
111
112	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))
113		panic("cannot allocate commpage");
114
115	if (vm_map_wire(kernel_map,kernel_addr,kernel_addr+area_used,VM_PROT_DEFAULT,FALSE))
116		panic("cannot wire commpage");
117
118	/*
119	 * Now that the object is created and wired into the kernel map, mark it so that no delay
120	 * copy-on-write will ever be performed on it as a result of mapping it into user-space.
121	 * If such a delayed copy ever occurred, we could remove the kernel's wired mapping - and
122	 * that would be a real disaster.
123	 *
124	 * JMM - What we really need is a way to create it like this in the first place.
125	 */
126	if (!vm_map_lookup_entry( kernel_map, vm_map_trunc_page(kernel_addr), &entry) || entry->is_sub_map)
127		panic("cannot find commpage entry");
128	entry->object.vm_object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
129
130	if (mach_make_memory_entry( kernel_map,		// target map
131				    &size,		// size
132				    kernel_addr,	// offset (address in kernel map)
133				    VM_PROT_ALL,	// map it RWX
134				    &handle,		// this is the object handle we get
135				    NULL ))		// parent_entry (what is this?)
136		panic("cannot make entry for commpage");
137
138	if (vm_map_64(	submap,				// target map (shared submap)
139			&zero,				// address (map into 1st page in submap)
140			area_used,			// size
141			0,				// mask
142			VM_FLAGS_FIXED,			// flags (it must be 1st page in submap)
143			handle,				// port is the memory entry we just made
144			0,                              // offset (map 1st page in memory entry)
145			FALSE,                          // copy
146			VM_PROT_READ|VM_PROT_EXECUTE,   // cur_protection (R-only in user map)
147			VM_PROT_READ|VM_PROT_EXECUTE,   // max_protection
148			VM_INHERIT_SHARE ))             // inheritance
149		panic("cannot map commpage");
150
151	ipc_port_release(handle);
152
153	return (void*) kernel_addr;                     // return address in kernel map
154}
155
156/* Get address (in kernel map) of a commpage field. */
157
158static void*
159commpage_addr_of(
160    int     addr_at_runtime )
161{
162    return  (void*) ((uintptr_t)commPagePtr + addr_at_runtime - commPageBaseOffset);
163}
164
165/* Determine number of CPUs on this system.  We cannot rely on
166 * machine_info.max_cpus this early in the boot.
167 */
168static int
169commpage_cpus( void )
170{
171	int cpus;
172
173	cpus = ml_get_max_cpus();                   // NB: this call can block
174
175	if (cpus == 0)
176		panic("commpage cpus==0");
177	if (cpus > 0xFF)
178		cpus = 0xFF;
179
180	return cpus;
181}
182
183/* Initialize kernel version of _cpu_capabilities vector (used by KEXTs.) */
184
185static void
186commpage_init_cpu_capabilities( void )
187{
188	int bits;
189	int cpus;
190	ml_cpu_info_t cpu_info;
191
192	bits = 0;
193	ml_cpu_get_info(&cpu_info);
194
195	switch (cpu_info.vector_unit) {
196		case 8:
197			bits |= kHasSSE4_2;
198			/* fall thru */
199		case 7:
200			bits |= kHasSSE4_1;
201			/* fall thru */
202		case 6:
203			bits |= kHasSupplementalSSE3;
204			/* fall thru */
205		case 5:
206			bits |= kHasSSE3;
207			/* fall thru */
208		case 4:
209			bits |= kHasSSE2;
210			/* fall thru */
211		case 3:
212			bits |= kHasSSE;
213			/* fall thru */
214		case 2:
215			bits |= kHasMMX;
216		default:
217			break;
218	}
219	switch (cpu_info.cache_line_size) {
220		case 128:
221			bits |= kCache128;
222			break;
223		case 64:
224			bits |= kCache64;
225			break;
226		case 32:
227			bits |= kCache32;
228			break;
229		default:
230			break;
231	}
232	cpus = commpage_cpus();			// how many CPUs do we have
233
234	if (cpus == 1)
235		bits |= kUP;
236
237	bits |= (cpus << kNumCPUsShift);
238
239	bits |= kFastThreadLocalStorage;	// we use %gs for TLS
240
241	if (cpu_mode_is64bit())			// k64Bit means processor is 64-bit capable
242		bits |= k64Bit;
243
244	if (tscFreq <= SLOW_TSC_THRESHOLD)	/* is TSC too slow for _commpage_nanotime?  */
245		bits |= kSlow;
246
247	_cpu_capabilities = bits;		// set kernel version for use by drivers etc
248}
249
250int
251_get_cpu_capabilities(void)
252{
253	return _cpu_capabilities;
254}
255
256/* Copy data into commpage. */
257
258static void
259commpage_stuff(
260    int 	address,
261    const void 	*source,
262    int 	length	)
263{
264    void	*dest = commpage_addr_of(address);
265
266    if ((uintptr_t)dest < next)
267        panic("commpage overlap at address 0x%x, %p < 0x%lx", address, dest, next);
268
269    bcopy(source,dest,length);
270
271    next = ((uintptr_t)dest + length);
272}
273
274static void
275commpage_stuff_swap(
276	int	address,
277	void	*source,
278	int	length,
279	int	legacy )
280{
281	if ( legacy ) {
282		void *dest = commpage_addr_of(address);
283		dest = (void *)((uintptr_t) dest + _COMM_PAGE_SIGS_OFFSET);
284		switch (length) {
285			case 2:
286				OSWriteSwapInt16(dest, 0, *(uint16_t *)source);
287				break;
288			case 4:
289				OSWriteSwapInt32(dest, 0, *(uint32_t *)source);
290				break;
291			case 8:
292				OSWriteSwapInt64(dest, 0, *(uint64_t *)source);
293				break;
294		}
295	}
296}
297
298static void
299commpage_stuff2(
300	int	address,
301	void	*source,
302	int	length,
303	int	legacy )
304{
305	commpage_stuff_swap(address, source, length, legacy);
306	commpage_stuff(address, source, length);
307}
308
309/* Copy a routine into comm page if it matches running machine.
310 */
311static void
312commpage_stuff_routine(
313    commpage_descriptor	*rd	)
314{
315    int		must,cant;
316
317    if (rd->commpage_address != cur_routine) {
318        if ((cur_routine!=0) && (matched==0))
319            panic("commpage no match for last, next address %08lx", rd->commpage_address);
320        cur_routine = rd->commpage_address;
321        matched = 0;
322    }
323
324    must = _cpu_capabilities & rd->musthave;
325    cant = _cpu_capabilities & rd->canthave;
326
327    if ((must == rd->musthave) && (cant == 0)) {
328        if (matched)
329            panic("commpage multiple matches for address %08lx", rd->commpage_address);
330        matched = 1;
331
332        commpage_stuff(rd->commpage_address,rd->code_address,rd->code_length);
333	}
334}
335
336/* Fill in the 32- or 64-bit commpage.  Called once for each.
337 * The 32-bit ("legacy") commpage has a bunch of stuff added to it
338 * for translated processes, some of which is byte-swapped.
339 */
340
341static void
342commpage_populate_one(
343	vm_map_t	submap,		// commpage32_map or compage64_map
344	char **		kernAddressPtr,	// &commPagePtr32 or &commPagePtr64
345	size_t		area_used,	// _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
346	size_t		base_offset,	// will become commPageBaseOffset
347	commpage_descriptor** commpage_routines, // list of routine ptrs for this commpage
348	boolean_t	legacy,		// true if 32-bit commpage
349	commpage_time_data** time_data,	// &time_data32 or &time_data64
350	const char*	signature )	// "commpage 32-bit" or "commpage 64-bit"
351{
352   	short   c2;
353	static double   two52 = 1048576.0 * 1048576.0 * 4096.0; // 2**52
354	static double   ten6 = 1000000.0;                       // 10**6
355	commpage_descriptor **rd;
356	short   version = _COMM_PAGE_THIS_VERSION;
357	int		swapcaps;
358
359	next = (uintptr_t) NULL;
360	cur_routine = 0;
361	commPagePtr = (char *)commpage_allocate( submap, (vm_size_t) area_used );
362	*kernAddressPtr = commPagePtr;				// save address either in commPagePtr32 or 64
363	commPageBaseOffset = base_offset;
364
365	*time_data = commpage_addr_of( _COMM_PAGE_TIME_DATA_START );
366
367	/* Stuff in the constants.  We move things into the comm page in strictly
368	* ascending order, so we can check for overlap and panic if so.
369	*/
370	commpage_stuff(_COMM_PAGE_SIGNATURE,signature,strlen(signature));
371	commpage_stuff2(_COMM_PAGE_VERSION,&version,sizeof(short),legacy);
372	commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES,&_cpu_capabilities,sizeof(int));
373
374	/* excuse our magic constants, we cannot include ppc/cpu_capabilities.h */
375	/* always set kCache32 and kDcbaAvailable */
376	swapcaps =  0x44;
377	if ( _cpu_capabilities & kUP )
378		swapcaps |= (kUP + (1 << kNumCPUsShift));
379	else
380		swapcaps |= 2 << kNumCPUsShift;	/* limit #cpus to 2 */
381	if ( ! noVMX )		/* if rosetta will be emulating altivec... */
382		swapcaps |= 0x101;	/* ...then set kHasAltivec and kDataStreamsAvailable too */
383	commpage_stuff_swap(_COMM_PAGE_CPU_CAPABILITIES, &swapcaps, sizeof(int), legacy);
384	c2 = 32;
385	commpage_stuff_swap(_COMM_PAGE_CACHE_LINESIZE,&c2,2,legacy);
386
387	if (_cpu_capabilities & kCache32)
388		c2 = 32;
389	else if (_cpu_capabilities & kCache64)
390		c2 = 64;
391	else if (_cpu_capabilities & kCache128)
392		c2 = 128;
393	commpage_stuff(_COMM_PAGE_CACHE_LINESIZE,&c2,2);
394
395	if ( legacy ) {
396		commpage_stuff2(_COMM_PAGE_2_TO_52,&two52,8,legacy);
397		commpage_stuff2(_COMM_PAGE_10_TO_6,&ten6,8,legacy);
398	}
399
400	for( rd = commpage_routines; *rd != NULL ; rd++ )
401		commpage_stuff_routine(*rd);
402
403	if (!matched)
404		panic("commpage no match on last routine");
405
406	if (next > (uintptr_t)_COMM_PAGE_END)
407		panic("commpage overflow: next = 0x%08lx, commPagePtr = 0x%08lx", next, (uintptr_t)commPagePtr);
408
409	if ( legacy ) {
410		next = (uintptr_t) NULL;
411		for( rd = ba_descriptors; *rd != NULL ; rd++ )
412			commpage_stuff_routine(*rd);
413
414		next = (uintptr_t) NULL;
415		commpage_stuff_routine(&sigdata_descriptor);
416	}
417}
418
419
420/* Fill in commpages: called once, during kernel initialization, from the
421 * startup thread before user-mode code is running.
422 *
423 * See the top of this file for a list of what you have to do to add
424 * a new routine to the commpage.
425 */
426
427void
428commpage_populate( void )
429{
430	commpage_init_cpu_capabilities();
431
432	commpage_populate_one(	commpage32_map,
433				&commPagePtr32,
434				_COMM_PAGE32_AREA_USED,
435				_COMM_PAGE32_BASE_ADDRESS,
436				commpage_32_routines,
437				TRUE,			/* legacy (32-bit) commpage */
438				&time_data32,
439				"commpage 32-bit");
440	pmap_commpage32_init((vm_offset_t) commPagePtr32, _COMM_PAGE32_BASE_ADDRESS,
441			   _COMM_PAGE32_AREA_USED/INTEL_PGBYTES);
442
443	time_data64 = time_data32;			/* if no 64-bit commpage, point to 32-bit */
444
445	if (_cpu_capabilities & k64Bit) {
446		commpage_populate_one(	commpage64_map,
447					&commPagePtr64,
448					_COMM_PAGE64_AREA_USED,
449					_COMM_PAGE32_START_ADDRESS, /* because kernel is built 32-bit */
450					commpage_64_routines,
451					FALSE,		/* not a legacy commpage */
452					&time_data64,
453					"commpage 64-bit");
454		pmap_commpage64_init((vm_offset_t) commPagePtr64, _COMM_PAGE64_BASE_ADDRESS,
455				   _COMM_PAGE64_AREA_USED/INTEL_PGBYTES);
456	}
457
458	rtc_nanotime_init_commpage();
459}
460
461
462/* Update commpage nanotime information.  Note that we interleave
463 * setting the 32- and 64-bit commpages, in order to keep nanotime more
464 * nearly in sync between the two environments.
465 *
466 * This routine must be serialized by some external means, ie a lock.
467 */
468
469void
470commpage_set_nanotime(
471	uint64_t	tsc_base,
472	uint64_t	ns_base,
473	uint32_t	scale,
474	uint32_t	shift )
475{
476	commpage_time_data	*p32 = time_data32;
477	commpage_time_data	*p64 = time_data64;
478	static uint32_t	generation = 0;
479	uint32_t	next_gen;
480
481	if (p32 == NULL)		/* have commpages been allocated yet? */
482		return;
483
484	if ( generation != p32->nt_generation )
485		panic("nanotime trouble 1");	/* possibly not serialized */
486	if ( ns_base < p32->nt_ns_base )
487		panic("nanotime trouble 2");
488	if ((shift != 32) && ((_cpu_capabilities & kSlow)==0) )
489		panic("nanotime trouble 3");
490
491	next_gen = ++generation;
492	if (next_gen == 0)
493		next_gen = ++generation;
494
495	p32->nt_generation = 0;		/* mark invalid, so commpage won't try to use it */
496	p64->nt_generation = 0;
497
498	p32->nt_tsc_base = tsc_base;
499	p64->nt_tsc_base = tsc_base;
500
501	p32->nt_ns_base = ns_base;
502	p64->nt_ns_base = ns_base;
503
504	p32->nt_scale = scale;
505	p64->nt_scale = scale;
506
507	p32->nt_shift = shift;
508	p64->nt_shift = shift;
509
510	p32->nt_generation = next_gen;	/* mark data as valid */
511	p64->nt_generation = next_gen;
512}
513
514
515/* Disable commpage gettimeofday(), forcing commpage to call through to the kernel.  */
516
517void
518commpage_disable_timestamp( void )
519{
520	time_data32->gtod_generation = 0;
521	time_data64->gtod_generation = 0;
522}
523
524
525/* Update commpage gettimeofday() information.  As with nanotime(), we interleave
526 * updates to the 32- and 64-bit commpage, in order to keep time more nearly in sync
527 * between the two environments.
528 *
529 * This routine must be serializeed by some external means, ie a lock.
530 */
531
532 void
533 commpage_set_timestamp(
534	uint64_t	abstime,
535	uint64_t	secs )
536{
537	commpage_time_data	*p32 = time_data32;
538	commpage_time_data	*p64 = time_data64;
539	static uint32_t	generation = 0;
540	uint32_t	next_gen;
541
542	next_gen = ++generation;
543	if (next_gen == 0)
544		next_gen = ++generation;
545
546	p32->gtod_generation = 0;		/* mark invalid, so commpage won't try to use it */
547	p64->gtod_generation = 0;
548
549	p32->gtod_ns_base = abstime;
550	p64->gtod_ns_base = abstime;
551
552	p32->gtod_sec_base = secs;
553	p64->gtod_sec_base = secs;
554
555	p32->gtod_generation = next_gen;	/* mark data as valid */
556	p64->gtod_generation = next_gen;
557}
558