1 /*
2 * Copyright (c) 2003-2020 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 <mach/mach_vm.h>
52 #include <mach/machine.h>
53 #include <i386/cpuid.h>
54 #include <i386/tsc.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>
65 #include <stdatomic.h>
66
67 #include <ipc/ipc_port.h>
68
69 #include <kern/page_decrypt.h>
70 #include <kern/processor.h>
71
72 #include <sys/kdebug.h>
73
74 #if CONFIG_ATM
75 #include <atm/atm_internal.h>
76 #endif
77
78 /* the lists of commpage routines are in commpage_asm.s */
79 extern commpage_descriptor* commpage_32_routines[];
80 extern commpage_descriptor* commpage_64_routines[];
81
82 extern vm_map_t commpage32_map; // the shared submap, set up in vm init
83 extern vm_map_t commpage64_map; // the shared submap, set up in vm init
84 extern vm_map_t commpage_text32_map; // the shared submap, set up in vm init
85 extern vm_map_t commpage_text64_map; // the shared submap, set up in vm init
86
87
88 char *commPagePtr32 = NULL; // virtual addr in kernel map of 32-bit commpage
89 char *commPagePtr64 = NULL; // ...and of 64-bit commpage
90 char *commPageTextPtr32 = NULL; // virtual addr in kernel map of 32-bit commpage
91 char *commPageTextPtr64 = NULL; // ...and of 64-bit commpage
92
93 uint64_t _cpu_capabilities = 0; // define the capability vector
94
95 typedef uint32_t commpage_address_t;
96
97 static commpage_address_t next; // next available address in comm page
98
99 static char *commPagePtr; // virtual addr in kernel map of commpage we are working on
100 static commpage_address_t commPageBaseOffset; // subtract from 32-bit runtime address to get offset in virtual commpage in kernel map
101
102 static commpage_time_data *time_data32 = NULL;
103 static commpage_time_data *time_data64 = NULL;
104 static new_commpage_timeofday_data_t *gtod_time_data32 = NULL;
105 static new_commpage_timeofday_data_t *gtod_time_data64 = NULL;
106
107
108 decl_simple_lock_data(static, commpage_active_cpus_lock);
109
110 /* Allocate the commpage and add to the shared submap created by vm:
111 * 1. allocate a page in the kernel map (RW)
112 * 2. wire it down
113 * 3. make a memory entry out of it
114 * 4. map that entry into the shared comm region map (R-only)
115 */
116
117 static void*
commpage_allocate(vm_map_t submap,size_t area_used,vm_prot_t uperm)118 commpage_allocate(
119 vm_map_t submap, // commpage32_map or commpage_map64
120 size_t area_used, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
121 vm_prot_t uperm)
122 {
123 vm_offset_t kernel_addr = 0; // address of commpage in kernel map
124 vm_offset_t zero = 0;
125 vm_size_t size = area_used; // size actually populated
126 vm_map_entry_t entry;
127 ipc_port_t handle;
128 kern_return_t kr;
129 vm_map_kernel_flags_t vmk_flags;
130
131 if (submap == NULL) {
132 panic("commpage submap is null");
133 }
134
135 kr = vm_map_kernel(kernel_map,
136 &kernel_addr,
137 area_used,
138 0,
139 VM_FLAGS_ANYWHERE,
140 VM_MAP_KERNEL_FLAGS_NONE,
141 VM_KERN_MEMORY_OSFMK,
142 NULL,
143 0,
144 FALSE,
145 VM_PROT_ALL,
146 VM_PROT_ALL,
147 VM_INHERIT_NONE);
148 if (kr != KERN_SUCCESS) {
149 panic("cannot allocate commpage %d", kr);
150 }
151
152 if ((kr = vm_map_wire_kernel(kernel_map,
153 kernel_addr,
154 kernel_addr + area_used,
155 VM_PROT_DEFAULT, VM_KERN_MEMORY_OSFMK,
156 FALSE))) {
157 panic("cannot wire commpage: %d", kr);
158 }
159
160 /*
161 * Now that the object is created and wired into the kernel map, mark it so that no delay
162 * copy-on-write will ever be performed on it as a result of mapping it into user-space.
163 * If such a delayed copy ever occurred, we could remove the kernel's wired mapping - and
164 * that would be a real disaster.
165 *
166 * JMM - What we really need is a way to create it like this in the first place.
167 */
168 if (!(kr = vm_map_lookup_entry( kernel_map, vm_map_trunc_page(kernel_addr, VM_MAP_PAGE_MASK(kernel_map)), &entry) || entry->is_sub_map)) {
169 panic("cannot find commpage entry %d", kr);
170 }
171 VME_OBJECT(entry)->copy_strategy = MEMORY_OBJECT_COPY_NONE;
172
173 if ((kr = mach_make_memory_entry( kernel_map, // target map
174 &size, // size
175 kernel_addr, // offset (address in kernel map)
176 uperm, // protections as specified
177 &handle, // this is the object handle we get
178 NULL ))) { // parent_entry (what is this?)
179 panic("cannot make entry for commpage %d", kr);
180 }
181
182 vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
183 if (uperm == (VM_PROT_READ | VM_PROT_EXECUTE)) {
184 /*
185 * Mark this unsigned executable mapping as "jit" to avoid
186 * code-signing violations when attempting to execute unsigned
187 * code.
188 */
189 vmk_flags.vmkf_map_jit = TRUE;
190 }
191
192 kr = vm_map_64_kernel(
193 submap, // target map (shared submap)
194 &zero, // address (map into 1st page in submap)
195 area_used, // size
196 0, // mask
197 VM_FLAGS_FIXED, // flags (it must be 1st page in submap)
198 vmk_flags,
199 VM_KERN_MEMORY_NONE,
200 handle, // port is the memory entry we just made
201 0, // offset (map 1st page in memory entry)
202 FALSE, // copy
203 uperm, // cur_protection (R-only in user map)
204 uperm, // max_protection
205 VM_INHERIT_SHARE); // inheritance
206 if (kr != KERN_SUCCESS) {
207 panic("cannot map commpage %d", kr);
208 }
209
210 ipc_port_release(handle);
211 /* Make the kernel mapping non-executable. This cannot be done
212 * at the time of map entry creation as mach_make_memory_entry
213 * cannot handle disjoint permissions at this time.
214 */
215 kr = vm_protect(kernel_map, kernel_addr, area_used, FALSE, VM_PROT_READ | VM_PROT_WRITE);
216 assert(kr == KERN_SUCCESS);
217
218 return (void*)(intptr_t)kernel_addr; // return address in kernel map
219 }
220
221 /* Get address (in kernel map) of a commpage field. */
222
223 static void*
commpage_addr_of(commpage_address_t addr_at_runtime)224 commpage_addr_of(
225 commpage_address_t addr_at_runtime )
226 {
227 return (void*) ((uintptr_t)commPagePtr + (addr_at_runtime - commPageBaseOffset));
228 }
229
230 /*
231 * Calculate address of data within 32- and 64-bit commpages (not to be used with commpage
232 * text).
233 */
234 static void*
commpage_specific_addr_of(char * commPageBase,commpage_address_t addr_at_runtime)235 commpage_specific_addr_of(char *commPageBase, commpage_address_t addr_at_runtime)
236 {
237 /*
238 * Note that the base address (_COMM_PAGE32_BASE_ADDRESS) is the same for
239 * 32- and 64-bit commpages
240 */
241 return (void*) ((uintptr_t)commPageBase + (addr_at_runtime - _COMM_PAGE32_BASE_ADDRESS));
242 }
243
244 /* Determine number of CPUs on this system. We cannot rely on
245 * machine_info.max_cpus this early in the boot.
246 */
247 static int
commpage_cpus(void)248 commpage_cpus( void )
249 {
250 unsigned int cpus;
251
252 cpus = ml_wait_max_cpus(); // NB: this call can block
253
254 if (cpus == 0) {
255 panic("commpage cpus==0");
256 }
257 if (cpus > 0xFF) {
258 cpus = 0xFF;
259 }
260
261 return cpus;
262 }
263
264 /* Initialize kernel version of _cpu_capabilities vector (used by KEXTs.) */
265
266 static void
commpage_init_cpu_capabilities(void)267 commpage_init_cpu_capabilities( void )
268 {
269 uint64_t bits;
270 int cpus;
271 ml_cpu_info_t cpu_info;
272
273 bits = 0;
274 ml_cpu_get_info(&cpu_info);
275
276 switch (cpu_info.vector_unit) {
277 case 9:
278 bits |= kHasAVX1_0;
279 OS_FALLTHROUGH;
280 case 8:
281 bits |= kHasSSE4_2;
282 OS_FALLTHROUGH;
283 case 7:
284 bits |= kHasSSE4_1;
285 OS_FALLTHROUGH;
286 case 6:
287 bits |= kHasSupplementalSSE3;
288 OS_FALLTHROUGH;
289 case 5:
290 bits |= kHasSSE3;
291 OS_FALLTHROUGH;
292 case 4:
293 bits |= kHasSSE2;
294 OS_FALLTHROUGH;
295 case 3:
296 bits |= kHasSSE;
297 OS_FALLTHROUGH;
298 case 2:
299 bits |= kHasMMX;
300 OS_FALLTHROUGH;
301 default:
302 break;
303 }
304 switch (cpu_info.cache_line_size) {
305 case 128:
306 bits |= kCache128;
307 break;
308 case 64:
309 bits |= kCache64;
310 break;
311 case 32:
312 bits |= kCache32;
313 break;
314 default:
315 break;
316 }
317 cpus = commpage_cpus(); // how many CPUs do we have
318
319 bits |= (cpus << kNumCPUsShift);
320
321 bits |= kFastThreadLocalStorage; // we use %gs for TLS
322
323 #define setif(_bits, _bit, _condition) \
324 if (_condition) _bits |= _bit
325
326 setif(bits, kUP, cpus == 1);
327 setif(bits, k64Bit, cpu_mode_is64bit());
328 setif(bits, kSlow, tscFreq <= SLOW_TSC_THRESHOLD);
329
330 setif(bits, kHasAES, cpuid_features() &
331 CPUID_FEATURE_AES);
332 setif(bits, kHasF16C, cpuid_features() &
333 CPUID_FEATURE_F16C);
334 setif(bits, kHasRDRAND, cpuid_features() &
335 CPUID_FEATURE_RDRAND);
336 setif(bits, kHasFMA, cpuid_features() &
337 CPUID_FEATURE_FMA);
338
339 setif(bits, kHasBMI1, cpuid_leaf7_features() &
340 CPUID_LEAF7_FEATURE_BMI1);
341 setif(bits, kHasBMI2, cpuid_leaf7_features() &
342 CPUID_LEAF7_FEATURE_BMI2);
343 /* Do not advertise RTM and HLE if the TSX FORCE ABORT WA is required */
344 if (cpuid_wa_required(CPU_INTEL_TSXFA) & CWA_OFF) {
345 setif(bits, kHasRTM, cpuid_leaf7_features() &
346 CPUID_LEAF7_FEATURE_RTM);
347 setif(bits, kHasHLE, cpuid_leaf7_features() &
348 CPUID_LEAF7_FEATURE_HLE);
349 }
350 setif(bits, kHasAVX2_0, cpuid_leaf7_features() &
351 CPUID_LEAF7_FEATURE_AVX2);
352 setif(bits, kHasRDSEED, cpuid_leaf7_features() &
353 CPUID_LEAF7_FEATURE_RDSEED);
354 setif(bits, kHasADX, cpuid_leaf7_features() &
355 CPUID_LEAF7_FEATURE_ADX);
356
357 #if 0 /* The kernel doesn't support MPX or SGX */
358 setif(bits, kHasMPX, cpuid_leaf7_features() &
359 CPUID_LEAF7_FEATURE_MPX);
360 setif(bits, kHasSGX, cpuid_leaf7_features() &
361 CPUID_LEAF7_FEATURE_SGX);
362 #endif
363
364 if (ml_fpu_avx512_enabled()) {
365 setif(bits, kHasAVX512F, cpuid_leaf7_features() &
366 CPUID_LEAF7_FEATURE_AVX512F);
367 setif(bits, kHasAVX512CD, cpuid_leaf7_features() &
368 CPUID_LEAF7_FEATURE_AVX512CD);
369 setif(bits, kHasAVX512DQ, cpuid_leaf7_features() &
370 CPUID_LEAF7_FEATURE_AVX512DQ);
371 setif(bits, kHasAVX512BW, cpuid_leaf7_features() &
372 CPUID_LEAF7_FEATURE_AVX512BW);
373 setif(bits, kHasAVX512VL, cpuid_leaf7_features() &
374 CPUID_LEAF7_FEATURE_AVX512VL);
375 setif(bits, kHasAVX512IFMA, cpuid_leaf7_features() &
376 CPUID_LEAF7_FEATURE_AVX512IFMA);
377 setif(bits, kHasAVX512VBMI, cpuid_leaf7_features() &
378 CPUID_LEAF7_FEATURE_AVX512VBMI);
379 setif(bits, kHasVAES, cpuid_leaf7_features() &
380 CPUID_LEAF7_FEATURE_VAES);
381 setif(bits, kHasVPCLMULQDQ, cpuid_leaf7_features() &
382 CPUID_LEAF7_FEATURE_VPCLMULQDQ);
383 setif(bits, kHasAVX512VNNI, cpuid_leaf7_features() &
384 CPUID_LEAF7_FEATURE_AVX512VNNI);
385 setif(bits, kHasAVX512BITALG, cpuid_leaf7_features() &
386 CPUID_LEAF7_FEATURE_AVX512BITALG);
387 setif(bits, kHasAVX512VPOPCNTDQ, cpuid_leaf7_features() &
388 CPUID_LEAF7_FEATURE_AVX512VPCDQ);
389 }
390
391 uint64_t misc_enable = rdmsr64(MSR_IA32_MISC_ENABLE);
392 setif(bits, kHasENFSTRG, (misc_enable & 1ULL) &&
393 (cpuid_leaf7_features() &
394 CPUID_LEAF7_FEATURE_ERMS));
395
396 _cpu_capabilities = bits; // set kernel version for use by drivers etc
397 }
398
399 /* initialize the approx_time_supported flag and set the approx time to 0.
400 * Called during initial commpage population.
401 */
402 static void
commpage_mach_approximate_time_init(void)403 commpage_mach_approximate_time_init(void)
404 {
405 char *cp = commPagePtr32;
406 uint8_t supported;
407
408 #ifdef CONFIG_MACH_APPROXIMATE_TIME
409 supported = 1;
410 #else
411 supported = 0;
412 #endif
413 if (cp) {
414 cp += (_COMM_PAGE_APPROX_TIME_SUPPORTED - _COMM_PAGE32_BASE_ADDRESS);
415 *(boolean_t *)cp = supported;
416 }
417
418 cp = commPagePtr64;
419 if (cp) {
420 cp += (_COMM_PAGE_APPROX_TIME_SUPPORTED - _COMM_PAGE32_START_ADDRESS);
421 *(boolean_t *)cp = supported;
422 }
423 commpage_update_mach_approximate_time(0);
424 }
425
426 static void
commpage_mach_continuous_time_init(void)427 commpage_mach_continuous_time_init(void)
428 {
429 commpage_update_mach_continuous_time(0);
430 }
431
432 static void
commpage_boottime_init(void)433 commpage_boottime_init(void)
434 {
435 clock_sec_t secs;
436 clock_usec_t microsecs;
437 clock_get_boottime_microtime(&secs, µsecs);
438 commpage_update_boottime(secs * USEC_PER_SEC + microsecs);
439 }
440
441 uint64_t
_get_cpu_capabilities(void)442 _get_cpu_capabilities(void)
443 {
444 return _cpu_capabilities;
445 }
446
447 /* Copy data into commpage. */
448
449 static void
commpage_stuff(commpage_address_t address,const void * source,int length)450 commpage_stuff(
451 commpage_address_t address,
452 const void *source,
453 int length )
454 {
455 void *dest = commpage_addr_of(address);
456
457 if (address < next) {
458 panic("commpage overlap at address 0x%p, 0x%x < 0x%x", dest, address, next);
459 }
460
461 bcopy(source, dest, length);
462
463 next = address + length;
464 }
465
466 /*
467 * Updates both the 32-bit and 64-bit commpages with the new data.
468 */
469 static void
commpage_update(commpage_address_t address,const void * source,int length)470 commpage_update(commpage_address_t address, const void *source, int length)
471 {
472 void *dest = commpage_specific_addr_of(commPagePtr32, address);
473 bcopy(source, dest, length);
474
475 dest = commpage_specific_addr_of(commPagePtr64, address);
476 bcopy(source, dest, length);
477 }
478
479 void
commpage_post_ucode_update(void)480 commpage_post_ucode_update(void)
481 {
482 commpage_init_cpu_capabilities();
483 commpage_update(_COMM_PAGE_CPU_CAPABILITIES64, &_cpu_capabilities, sizeof(_cpu_capabilities));
484 commpage_update(_COMM_PAGE_CPU_CAPABILITIES, &_cpu_capabilities, sizeof(uint32_t));
485 }
486
487 /* Copy a routine into comm page if it matches running machine.
488 */
489 static void
commpage_stuff_routine(commpage_descriptor * rd)490 commpage_stuff_routine(
491 commpage_descriptor *rd )
492 {
493 commpage_stuff(rd->commpage_address, rd->code_address, rd->code_length);
494 }
495
496
497 /* Fill in the 32- or 64-bit commpage. Called once for each.
498 */
499
500 static void
commpage_populate_one(vm_map_t submap,char ** kernAddressPtr,size_t area_used,commpage_address_t base_offset,commpage_time_data ** time_data,new_commpage_timeofday_data_t ** gtod_time_data,const char * signature,vm_prot_t uperm)501 commpage_populate_one(
502 vm_map_t submap, // commpage32_map or compage64_map
503 char ** kernAddressPtr, // &commPagePtr32 or &commPagePtr64
504 size_t area_used, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
505 commpage_address_t base_offset, // will become commPageBaseOffset
506 commpage_time_data** time_data, // &time_data32 or &time_data64
507 new_commpage_timeofday_data_t** gtod_time_data, // >od_time_data32 or >od_time_data64
508 const char* signature, // "commpage 32-bit" or "commpage 64-bit"
509 vm_prot_t uperm)
510 {
511 uint8_t c1;
512 uint16_t c2;
513 uint64_t c8;
514 uint32_t cfamily;
515 short version = _COMM_PAGE_THIS_VERSION;
516
517 next = 0;
518 commPagePtr = (char *)commpage_allocate( submap, (vm_size_t) area_used, uperm );
519 *kernAddressPtr = commPagePtr; // save address either in commPagePtr32 or 64
520 commPageBaseOffset = base_offset;
521
522 *time_data = commpage_addr_of( _COMM_PAGE_TIME_DATA_START );
523 *gtod_time_data = commpage_addr_of( _COMM_PAGE_NEWTIMEOFDAY_DATA );
524
525 /* Stuff in the constants. We move things into the comm page in strictly
526 * ascending order, so we can check for overlap and panic if so.
527 * Note: the 32-bit cpu_capabilities vector is retained in addition to
528 * the expanded 64-bit vector.
529 */
530 commpage_stuff(_COMM_PAGE_SIGNATURE, signature, (int)MIN(_COMM_PAGE_SIGNATURELEN, strlen(signature)));
531 commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES64, &_cpu_capabilities, sizeof(_cpu_capabilities));
532 commpage_stuff(_COMM_PAGE_VERSION, &version, sizeof(short));
533 commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES, &_cpu_capabilities, sizeof(uint32_t));
534
535 c2 = 32; // default
536 if (_cpu_capabilities & kCache64) {
537 c2 = 64;
538 } else if (_cpu_capabilities & kCache128) {
539 c2 = 128;
540 }
541 commpage_stuff(_COMM_PAGE_CACHE_LINESIZE, &c2, 2);
542
543 /* machine_info valid after ml_wait_max_cpus() */
544 c1 = machine_info.physical_cpu_max;
545 commpage_stuff(_COMM_PAGE_PHYSICAL_CPUS, &c1, 1);
546 c1 = machine_info.logical_cpu_max;
547 commpage_stuff(_COMM_PAGE_LOGICAL_CPUS, &c1, 1);
548
549 c8 = ml_cpu_cache_size(0);
550 commpage_stuff(_COMM_PAGE_MEMORY_SIZE, &c8, 8);
551
552 cfamily = cpuid_info()->cpuid_cpufamily;
553 commpage_stuff(_COMM_PAGE_CPUFAMILY, &cfamily, 4);
554 c1 = PAGE_SHIFT;
555 commpage_stuff(_COMM_PAGE_KERNEL_PAGE_SHIFT, &c1, 1);
556 commpage_stuff(_COMM_PAGE_USER_PAGE_SHIFT_64, &c1, 1);
557
558 if (next > _COMM_PAGE_END) {
559 panic("commpage overflow: next = 0x%08x, commPagePtr = 0x%p", next, commPagePtr);
560 }
561 }
562
563
564 /* Fill in commpages: called once, during kernel initialization, from the
565 * startup thread before user-mode code is running.
566 *
567 * See the top of this file for a list of what you have to do to add
568 * a new routine to the commpage.
569 */
570
571 void
commpage_populate(void)572 commpage_populate( void )
573 {
574 commpage_init_cpu_capabilities();
575
576 commpage_populate_one( commpage32_map,
577 &commPagePtr32,
578 _COMM_PAGE32_AREA_USED,
579 _COMM_PAGE32_BASE_ADDRESS,
580 &time_data32,
581 >od_time_data32,
582 _COMM_PAGE32_SIGNATURE_STRING,
583 VM_PROT_READ);
584 #ifndef __LP64__
585 pmap_commpage32_init((vm_offset_t) commPagePtr32, _COMM_PAGE32_BASE_ADDRESS,
586 _COMM_PAGE32_AREA_USED / INTEL_PGBYTES);
587 #endif
588 time_data64 = time_data32; /* if no 64-bit commpage, point to 32-bit */
589 gtod_time_data64 = gtod_time_data32;
590
591 if (_cpu_capabilities & k64Bit) {
592 commpage_populate_one( commpage64_map,
593 &commPagePtr64,
594 _COMM_PAGE64_AREA_USED,
595 _COMM_PAGE32_START_ADDRESS, /* commpage address are relative to 32-bit commpage placement */
596 &time_data64,
597 >od_time_data64,
598 _COMM_PAGE64_SIGNATURE_STRING,
599 VM_PROT_READ);
600 #ifndef __LP64__
601 pmap_commpage64_init((vm_offset_t) commPagePtr64, _COMM_PAGE64_BASE_ADDRESS,
602 _COMM_PAGE64_AREA_USED / INTEL_PGBYTES);
603 #endif
604 }
605
606 simple_lock_init(&commpage_active_cpus_lock, 0);
607
608 commpage_update_active_cpus();
609 commpage_mach_approximate_time_init();
610 commpage_mach_continuous_time_init();
611 commpage_boottime_init();
612 rtc_nanotime_init_commpage();
613 commpage_update_kdebug_state();
614 #if CONFIG_ATM
615 commpage_update_atm_diagnostic_config(atm_get_diagnostic_config());
616 #endif
617 }
618
619 /* Fill in the common routines during kernel initialization.
620 * This is called before user-mode code is running.
621 */
622 void
commpage_text_populate(void)623 commpage_text_populate( void )
624 {
625 commpage_descriptor **rd;
626
627 next = 0;
628 commPagePtr = (char *) commpage_allocate(commpage_text32_map, (vm_size_t) _COMM_PAGE_TEXT_AREA_USED, VM_PROT_READ | VM_PROT_EXECUTE);
629 commPageTextPtr32 = commPagePtr;
630
631 char *cptr = commPagePtr;
632 int i = 0;
633 for (; i < _COMM_PAGE_TEXT_AREA_USED; i++) {
634 cptr[i] = 0xCC;
635 }
636
637 commPageBaseOffset = _COMM_PAGE_TEXT_START;
638 for (rd = commpage_32_routines; *rd != NULL; rd++) {
639 commpage_stuff_routine(*rd);
640 }
641
642 #ifndef __LP64__
643 pmap_commpage32_init((vm_offset_t) commPageTextPtr32, _COMM_PAGE_TEXT_START,
644 _COMM_PAGE_TEXT_AREA_USED / INTEL_PGBYTES);
645 #endif
646
647 if (_cpu_capabilities & k64Bit) {
648 next = 0;
649 commPagePtr = (char *) commpage_allocate(commpage_text64_map, (vm_size_t) _COMM_PAGE_TEXT_AREA_USED, VM_PROT_READ | VM_PROT_EXECUTE);
650 commPageTextPtr64 = commPagePtr;
651
652 cptr = commPagePtr;
653 for (i = 0; i < _COMM_PAGE_TEXT_AREA_USED; i++) {
654 cptr[i] = 0xCC;
655 }
656
657 for (rd = commpage_64_routines; *rd != NULL; rd++) {
658 commpage_stuff_routine(*rd);
659 }
660
661 #ifndef __LP64__
662 pmap_commpage64_init((vm_offset_t) commPageTextPtr64, _COMM_PAGE_TEXT_START,
663 _COMM_PAGE_TEXT_AREA_USED / INTEL_PGBYTES);
664 #endif
665 }
666
667 if (next > _COMM_PAGE_TEXT_END) {
668 panic("commpage text overflow: next=0x%08x, commPagePtr=%p", next, commPagePtr);
669 }
670 }
671
672 /* Update commpage nanotime information.
673 *
674 * This routine must be serialized by some external means, ie a lock.
675 */
676
677 void
commpage_set_nanotime(uint64_t tsc_base,uint64_t ns_base,uint32_t scale,uint32_t shift)678 commpage_set_nanotime(
679 uint64_t tsc_base,
680 uint64_t ns_base,
681 uint32_t scale,
682 uint32_t shift )
683 {
684 commpage_time_data *p32 = time_data32;
685 commpage_time_data *p64 = time_data64;
686 static uint32_t generation = 0;
687 uint32_t next_gen;
688
689 if (p32 == NULL) { /* have commpages been allocated yet? */
690 return;
691 }
692
693 if (generation != p32->nt_generation) {
694 panic("nanotime trouble 1"); /* possibly not serialized */
695 }
696 if (ns_base < p32->nt_ns_base) {
697 panic("nanotime trouble 2");
698 }
699 if ((shift != 0) && ((_cpu_capabilities & kSlow) == 0)) {
700 panic("nanotime trouble 3");
701 }
702
703 next_gen = ++generation;
704 if (next_gen == 0) {
705 next_gen = ++generation;
706 }
707
708 p32->nt_generation = 0; /* mark invalid, so commpage won't try to use it */
709 p64->nt_generation = 0;
710
711 p32->nt_tsc_base = tsc_base;
712 p64->nt_tsc_base = tsc_base;
713
714 p32->nt_ns_base = ns_base;
715 p64->nt_ns_base = ns_base;
716
717 p32->nt_scale = scale;
718 p64->nt_scale = scale;
719
720 p32->nt_shift = shift;
721 p64->nt_shift = shift;
722
723 p32->nt_generation = next_gen; /* mark data as valid */
724 p64->nt_generation = next_gen;
725 }
726
727 /* Update commpage gettimeofday() information. As with nanotime(), we interleave
728 * updates to the 32- and 64-bit commpage, in order to keep time more nearly in sync
729 * between the two environments.
730 *
731 * This routine must be serializeed by some external means, ie a lock.
732 */
733
734 void
commpage_set_timestamp(uint64_t abstime,uint64_t sec,uint64_t frac,uint64_t scale,uint64_t tick_per_sec)735 commpage_set_timestamp(
736 uint64_t abstime,
737 uint64_t sec,
738 uint64_t frac,
739 uint64_t scale,
740 uint64_t tick_per_sec)
741 {
742 new_commpage_timeofday_data_t *p32 = gtod_time_data32;
743 new_commpage_timeofday_data_t *p64 = gtod_time_data64;
744
745 p32->TimeStamp_tick = 0x0ULL;
746 p64->TimeStamp_tick = 0x0ULL;
747
748 p32->TimeStamp_sec = sec;
749 p64->TimeStamp_sec = sec;
750
751 p32->TimeStamp_frac = frac;
752 p64->TimeStamp_frac = frac;
753
754 p32->Ticks_scale = scale;
755 p64->Ticks_scale = scale;
756
757 p32->Ticks_per_sec = tick_per_sec;
758 p64->Ticks_per_sec = tick_per_sec;
759
760 p32->TimeStamp_tick = abstime;
761 p64->TimeStamp_tick = abstime;
762 }
763
764 /* Update _COMM_PAGE_MEMORY_PRESSURE. Called periodically from vm's compute_memory_pressure() */
765
766 void
commpage_set_memory_pressure(unsigned int pressure)767 commpage_set_memory_pressure(
768 unsigned int pressure )
769 {
770 char *cp;
771 uint32_t *ip;
772
773 cp = commPagePtr32;
774 if (cp) {
775 cp += (_COMM_PAGE_MEMORY_PRESSURE - _COMM_PAGE32_BASE_ADDRESS);
776 ip = (uint32_t*) (void *) cp;
777 *ip = (uint32_t) pressure;
778 }
779
780 cp = commPagePtr64;
781 if (cp) {
782 cp += (_COMM_PAGE_MEMORY_PRESSURE - _COMM_PAGE32_START_ADDRESS);
783 ip = (uint32_t*) (void *) cp;
784 *ip = (uint32_t) pressure;
785 }
786 }
787
788 /* Updated every time a logical CPU goes offline/online */
789 void
commpage_update_active_cpus(void)790 commpage_update_active_cpus(void)
791 {
792 char *cp;
793 volatile uint8_t *ip;
794
795 /* At least 32-bit commpage must be initialized */
796 if (!commPagePtr32) {
797 return;
798 }
799
800 simple_lock(&commpage_active_cpus_lock, LCK_GRP_NULL);
801
802 cp = commPagePtr32;
803 cp += (_COMM_PAGE_ACTIVE_CPUS - _COMM_PAGE32_BASE_ADDRESS);
804 ip = (volatile uint8_t*) cp;
805 *ip = (uint8_t) processor_avail_count_user;
806
807 cp = commPagePtr64;
808 if (cp) {
809 cp += (_COMM_PAGE_ACTIVE_CPUS - _COMM_PAGE32_START_ADDRESS);
810 ip = (volatile uint8_t*) cp;
811 *ip = (uint8_t) processor_avail_count_user;
812 }
813
814 simple_unlock(&commpage_active_cpus_lock);
815 }
816
817 /*
818 * Update the commpage with current kdebug state. This currently has bits for
819 * global trace state, and typefilter enablement. It is likely additional state
820 * will be tracked in the future.
821 *
822 * INVARIANT: This value will always be 0 if global tracing is disabled. This
823 * allows simple guard tests of "if (*_COMM_PAGE_KDEBUG_ENABLE) { ... }"
824 */
825 void
commpage_update_kdebug_state(void)826 commpage_update_kdebug_state(void)
827 {
828 volatile uint32_t *saved_data_ptr;
829 char *cp;
830
831 cp = commPagePtr32;
832 if (cp) {
833 cp += (_COMM_PAGE_KDEBUG_ENABLE - _COMM_PAGE32_BASE_ADDRESS);
834 saved_data_ptr = (volatile uint32_t *)cp;
835 *saved_data_ptr = kdebug_commpage_state();
836 }
837
838 cp = commPagePtr64;
839 if (cp) {
840 cp += (_COMM_PAGE_KDEBUG_ENABLE - _COMM_PAGE32_START_ADDRESS);
841 saved_data_ptr = (volatile uint32_t *)cp;
842 *saved_data_ptr = kdebug_commpage_state();
843 }
844 }
845
846 /* Ditto for atm_diagnostic_config */
847 void
commpage_update_atm_diagnostic_config(uint32_t diagnostic_config)848 commpage_update_atm_diagnostic_config(uint32_t diagnostic_config)
849 {
850 volatile uint32_t *saved_data_ptr;
851 char *cp;
852
853 cp = commPagePtr32;
854 if (cp) {
855 cp += (_COMM_PAGE_ATM_DIAGNOSTIC_CONFIG - _COMM_PAGE32_BASE_ADDRESS);
856 saved_data_ptr = (volatile uint32_t *)cp;
857 *saved_data_ptr = diagnostic_config;
858 }
859
860 cp = commPagePtr64;
861 if (cp) {
862 cp += (_COMM_PAGE_ATM_DIAGNOSTIC_CONFIG - _COMM_PAGE32_START_ADDRESS);
863 saved_data_ptr = (volatile uint32_t *)cp;
864 *saved_data_ptr = diagnostic_config;
865 }
866 }
867
868 /*
869 * update the commpage with if dtrace user land probes are enabled
870 */
871 void
commpage_update_dof(boolean_t enabled)872 commpage_update_dof(boolean_t enabled)
873 {
874 #if CONFIG_DTRACE
875 char *cp;
876
877 cp = commPagePtr32;
878 if (cp) {
879 cp += (_COMM_PAGE_DTRACE_DOF_ENABLED - _COMM_PAGE32_BASE_ADDRESS);
880 *cp = (enabled ? 1 : 0);
881 }
882
883 cp = commPagePtr64;
884 if (cp) {
885 cp += (_COMM_PAGE_DTRACE_DOF_ENABLED - _COMM_PAGE32_START_ADDRESS);
886 *cp = (enabled ? 1 : 0);
887 }
888 #else
889 (void)enabled;
890 #endif
891 }
892
893
894 /*
895 * update the dyld global config flags
896 */
897 void
commpage_update_dyld_flags(uint64_t value)898 commpage_update_dyld_flags(uint64_t value)
899 {
900 char *cp;
901
902 cp = commPagePtr32;
903 if (cp) {
904 cp += (_COMM_PAGE_DYLD_FLAGS - _COMM_PAGE32_BASE_ADDRESS);
905 *(uint64_t *)cp = value;
906 }
907
908 cp = commPagePtr64;
909 if (cp) {
910 cp += (_COMM_PAGE_DYLD_FLAGS - _COMM_PAGE32_BASE_ADDRESS);
911 *(uint64_t *)cp = value;
912 }
913 }
914
915
916 /*
917 * update the commpage data for last known value of mach_absolute_time()
918 */
919
920 void
commpage_update_mach_approximate_time(uint64_t abstime)921 commpage_update_mach_approximate_time(uint64_t abstime)
922 {
923 #ifdef CONFIG_MACH_APPROXIMATE_TIME
924 uint64_t saved_data;
925 char *cp;
926
927 cp = commPagePtr32;
928 if (cp) {
929 cp += (_COMM_PAGE_APPROX_TIME - _COMM_PAGE32_BASE_ADDRESS);
930 saved_data = atomic_load_explicit((_Atomic uint64_t *)(uintptr_t)cp, memory_order_relaxed);
931 if (saved_data < abstime) {
932 /* ignoring the success/fail return value assuming that
933 * if the value has been updated since we last read it,
934 * "someone" has a newer timestamp than us and ours is
935 * now invalid. */
936 atomic_compare_exchange_strong_explicit((_Atomic uint64_t *)(uintptr_t)cp,
937 &saved_data, abstime, memory_order_relaxed, memory_order_relaxed);
938 }
939 }
940 cp = commPagePtr64;
941 if (cp) {
942 cp += (_COMM_PAGE_APPROX_TIME - _COMM_PAGE32_START_ADDRESS);
943 saved_data = atomic_load_explicit((_Atomic uint64_t *)(uintptr_t)cp, memory_order_relaxed);
944 if (saved_data < abstime) {
945 /* ignoring the success/fail return value assuming that
946 * if the value has been updated since we last read it,
947 * "someone" has a newer timestamp than us and ours is
948 * now invalid. */
949 atomic_compare_exchange_strong_explicit((_Atomic uint64_t *)(uintptr_t)cp,
950 &saved_data, abstime, memory_order_relaxed, memory_order_relaxed);
951 }
952 }
953 #else
954 #pragma unused (abstime)
955 #endif
956 }
957
958 void
commpage_update_mach_continuous_time(uint64_t sleeptime)959 commpage_update_mach_continuous_time(uint64_t sleeptime)
960 {
961 char *cp;
962 cp = commPagePtr32;
963 if (cp) {
964 cp += (_COMM_PAGE_CONT_TIMEBASE - _COMM_PAGE32_START_ADDRESS);
965 *(uint64_t *)cp = sleeptime;
966 }
967
968 cp = commPagePtr64;
969 if (cp) {
970 cp += (_COMM_PAGE_CONT_TIMEBASE - _COMM_PAGE32_START_ADDRESS);
971 *(uint64_t *)cp = sleeptime;
972 }
973 }
974
975 void
commpage_update_boottime(uint64_t boottime)976 commpage_update_boottime(uint64_t boottime)
977 {
978 char *cp;
979 cp = commPagePtr32;
980 if (cp) {
981 cp += (_COMM_PAGE_BOOTTIME_USEC - _COMM_PAGE32_START_ADDRESS);
982 *(uint64_t *)cp = boottime;
983 }
984
985 cp = commPagePtr64;
986 if (cp) {
987 cp += (_COMM_PAGE_BOOTTIME_USEC - _COMM_PAGE32_START_ADDRESS);
988 *(uint64_t *)cp = boottime;
989 }
990 }
991
992
993 extern user32_addr_t commpage_text32_location;
994 extern user64_addr_t commpage_text64_location;
995
996 /* Check to see if a given address is in the Preemption Free Zone (PFZ) */
997
998 uint32_t
commpage_is_in_pfz32(uint32_t addr32)999 commpage_is_in_pfz32(uint32_t addr32)
1000 {
1001 if ((addr32 >= (commpage_text32_location + _COMM_TEXT_PFZ_START_OFFSET))
1002 && (addr32 < (commpage_text32_location + _COMM_TEXT_PFZ_END_OFFSET))) {
1003 return 1;
1004 } else {
1005 return 0;
1006 }
1007 }
1008
1009 uint32_t
commpage_is_in_pfz64(addr64_t addr64)1010 commpage_is_in_pfz64(addr64_t addr64)
1011 {
1012 if ((addr64 >= (commpage_text64_location + _COMM_TEXT_PFZ_START_OFFSET))
1013 && (addr64 < (commpage_text64_location + _COMM_TEXT_PFZ_END_OFFSET))) {
1014 return 1;
1015 } else {
1016 return 0;
1017 }
1018 }
1019