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_xnu.h>
64 #include <vm/vm_map_internal.h>
65 #include <vm/vm_map_xnu.h>
66 #include <stdatomic.h>
67
68 #include <ipc/ipc_port.h>
69
70 #include <kern/page_decrypt.h>
71 #include <kern/processor.h>
72
73 #include <sys/kdebug.h>
74 #include <sys/random.h>
75
76 #if CONFIG_ATM
77 #include <atm/atm_internal.h>
78 #endif
79
80 /* the lists of commpage routines are in commpage_asm.s */
81 extern commpage_descriptor* commpage_32_routines[];
82 extern commpage_descriptor* commpage_64_routines[];
83
84 extern vm_map_t commpage32_map; // the shared submap, set up in vm init
85 extern vm_map_t commpage64_map; // the shared submap, set up in vm init
86 extern vm_map_t commpage_text32_map; // the shared submap, set up in vm init
87 extern vm_map_t commpage_text64_map; // the shared submap, set up in vm init
88
89
90 char *commPagePtr32 = NULL; // virtual addr in kernel map of 32-bit commpage
91 char *commPagePtr64 = NULL; // ...and of 64-bit commpage
92 char *commPageTextPtr32 = NULL; // virtual addr in kernel map of 32-bit commpage
93 char *commPageTextPtr64 = NULL; // ...and of 64-bit commpage
94
95 uint64_t _cpu_capabilities = 0; // define the capability vector
96
97 typedef uint32_t commpage_address_t;
98
99 static commpage_address_t next; // next available address in comm page
100
101 static char *commPagePtr; // virtual addr in kernel map of commpage we are working on
102 static commpage_address_t commPageBaseOffset; // subtract from 32-bit runtime address to get offset in virtual commpage in kernel map
103
104 static commpage_time_data *time_data32 = NULL;
105 static commpage_time_data *time_data64 = NULL;
106 static new_commpage_timeofday_data_t *gtod_time_data32 = NULL;
107 static new_commpage_timeofday_data_t *gtod_time_data64 = NULL;
108
109
110 decl_simple_lock_data(static, commpage_active_cpus_lock);
111
112 /* Allocate the commpage and add to the shared submap created by vm:
113 * 1. allocate a page in the kernel map (RW)
114 * 2. wire it down
115 * 3. make a memory entry out of it
116 * 4. map that entry into the shared comm region map (R-only)
117 */
118
119 static void*
commpage_allocate(vm_map_t submap,size_t area_used,vm_prot_t uperm)120 commpage_allocate(
121 vm_map_t submap, // commpage32_map or commpage_map64
122 size_t area_used, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
123 vm_prot_t uperm)
124 {
125 mach_vm_offset_t kernel_addr = 0; // address of commpage in kernel map
126 mach_vm_offset_t zero = 0;
127 vm_size_t size = area_used; // size actually populated
128 vm_map_entry_t entry;
129 ipc_port_t handle;
130 kern_return_t kr;
131 vm_map_kernel_flags_t vmk_flags;
132
133 if (submap == NULL) {
134 panic("commpage submap is null");
135 }
136
137 kr = mach_vm_map_kernel(kernel_map,
138 &kernel_addr,
139 area_used,
140 0,
141 VM_MAP_KERNEL_FLAGS_ANYWHERE(.vm_tag = 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_FIXED();
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 = mach_vm_map_kernel(
193 submap, // target map (shared submap)
194 &zero, // address (map into 1st page in submap)
195 area_used, // size
196 0, // mask
197 vmk_flags,
198 handle, // port is the memory entry we just made
199 0, // offset (map 1st page in memory entry)
200 FALSE, // copy
201 uperm, // cur_protection (R-only in user map)
202 uperm, // max_protection
203 VM_INHERIT_SHARE); // inheritance
204 if (kr != KERN_SUCCESS) {
205 panic("cannot map commpage %d", kr);
206 }
207
208 ipc_port_release(handle);
209 /* Make the kernel mapping non-executable. This cannot be done
210 * at the time of map entry creation as mach_make_memory_entry
211 * cannot handle disjoint permissions at this time.
212 */
213 kr = vm_protect(kernel_map, kernel_addr, area_used, FALSE, VM_PROT_READ | VM_PROT_WRITE);
214 assert(kr == KERN_SUCCESS);
215
216 return (void*)(intptr_t)kernel_addr; // return address in kernel map
217 }
218
219 /* Get address (in kernel map) of a commpage field. */
220
221 static void*
commpage_addr_of(commpage_address_t addr_at_runtime)222 commpage_addr_of(
223 commpage_address_t addr_at_runtime )
224 {
225 return (void*) ((uintptr_t)commPagePtr + (addr_at_runtime - commPageBaseOffset));
226 }
227
228 /*
229 * Calculate address of data within 32- and 64-bit commpages (not to be used with commpage
230 * text).
231 */
232 static void*
commpage_specific_addr_of(char * commPageBase,commpage_address_t addr_at_runtime)233 commpage_specific_addr_of(char *commPageBase, commpage_address_t addr_at_runtime)
234 {
235 /*
236 * Note that the base address (_COMM_PAGE32_BASE_ADDRESS) is the same for
237 * 32- and 64-bit commpages
238 */
239 return (void*) ((uintptr_t)commPageBase + (addr_at_runtime - _COMM_PAGE32_BASE_ADDRESS));
240 }
241
242 /* Determine number of CPUs on this system. We cannot rely on
243 * machine_info.max_cpus this early in the boot.
244 */
245 static int
commpage_cpus(void)246 commpage_cpus( void )
247 {
248 unsigned int cpus;
249
250 cpus = ml_wait_max_cpus(); // NB: this call can block
251
252 if (cpus == 0) {
253 panic("commpage cpus==0");
254 }
255 if (cpus > 0xFF) {
256 cpus = 0xFF;
257 }
258
259 return cpus;
260 }
261
262 /* Initialize kernel version of _cpu_capabilities vector (used by KEXTs.) */
263
264 static void
commpage_init_cpu_capabilities(void)265 commpage_init_cpu_capabilities( void )
266 {
267 uint64_t bits;
268 int cpus;
269 ml_cpu_info_t cpu_info;
270
271 bits = 0;
272 ml_cpu_get_info(&cpu_info);
273
274 switch (cpu_info.vector_unit) {
275 case 9:
276 bits |= kHasAVX1_0;
277 OS_FALLTHROUGH;
278 case 8:
279 bits |= kHasSSE4_2;
280 OS_FALLTHROUGH;
281 case 7:
282 bits |= kHasSSE4_1;
283 OS_FALLTHROUGH;
284 case 6:
285 bits |= kHasSupplementalSSE3;
286 OS_FALLTHROUGH;
287 case 5:
288 bits |= kHasSSE3;
289 OS_FALLTHROUGH;
290 case 4:
291 bits |= kHasSSE2;
292 OS_FALLTHROUGH;
293 case 3:
294 bits |= kHasSSE;
295 OS_FALLTHROUGH;
296 case 2:
297 bits |= kHasMMX;
298 OS_FALLTHROUGH;
299 default:
300 break;
301 }
302 switch (cpu_info.cache_line_size) {
303 case 128:
304 bits |= kCache128;
305 break;
306 case 64:
307 bits |= kCache64;
308 break;
309 case 32:
310 bits |= kCache32;
311 break;
312 default:
313 break;
314 }
315 cpus = commpage_cpus(); // how many CPUs do we have
316
317 bits |= (cpus << kNumCPUsShift);
318
319 bits |= kFastThreadLocalStorage; // we use %gs for TLS
320
321 #define setif(_bits, _bit, _condition) \
322 if (_condition) _bits |= _bit
323
324 setif(bits, kUP, cpus == 1);
325 setif(bits, k64Bit, cpu_mode_is64bit());
326 setif(bits, kSlow, tscFreq <= SLOW_TSC_THRESHOLD);
327
328 setif(bits, kHasAES, cpuid_features() &
329 CPUID_FEATURE_AES);
330 setif(bits, kHasF16C, cpuid_features() &
331 CPUID_FEATURE_F16C);
332 setif(bits, kHasRDRAND, cpuid_features() &
333 CPUID_FEATURE_RDRAND);
334 setif(bits, kHasFMA, cpuid_features() &
335 CPUID_FEATURE_FMA);
336
337 setif(bits, kHasBMI1, cpuid_leaf7_features() &
338 CPUID_LEAF7_FEATURE_BMI1);
339 setif(bits, kHasBMI2, cpuid_leaf7_features() &
340 CPUID_LEAF7_FEATURE_BMI2);
341 /* Do not advertise RTM and HLE if the TSX FORCE ABORT WA is required */
342 if (cpuid_wa_required(CPU_INTEL_TSXFA) & CWA_OFF) {
343 setif(bits, kHasRTM, cpuid_leaf7_features() &
344 CPUID_LEAF7_FEATURE_RTM);
345 setif(bits, kHasHLE, cpuid_leaf7_features() &
346 CPUID_LEAF7_FEATURE_HLE);
347 }
348 setif(bits, kHasAVX2_0, cpuid_leaf7_features() &
349 CPUID_LEAF7_FEATURE_AVX2);
350 setif(bits, kHasRDSEED, cpuid_leaf7_features() &
351 CPUID_LEAF7_FEATURE_RDSEED);
352 setif(bits, kHasADX, cpuid_leaf7_features() &
353 CPUID_LEAF7_FEATURE_ADX);
354
355 #if 0 /* The kernel doesn't support MPX or SGX */
356 setif(bits, kHasMPX, cpuid_leaf7_features() &
357 CPUID_LEAF7_FEATURE_MPX);
358 setif(bits, kHasSGX, cpuid_leaf7_features() &
359 CPUID_LEAF7_FEATURE_SGX);
360 #endif
361
362 if (ml_fpu_avx512_enabled()) {
363 setif(bits, kHasAVX512F, cpuid_leaf7_features() &
364 CPUID_LEAF7_FEATURE_AVX512F);
365 setif(bits, kHasAVX512CD, cpuid_leaf7_features() &
366 CPUID_LEAF7_FEATURE_AVX512CD);
367 setif(bits, kHasAVX512DQ, cpuid_leaf7_features() &
368 CPUID_LEAF7_FEATURE_AVX512DQ);
369 setif(bits, kHasAVX512BW, cpuid_leaf7_features() &
370 CPUID_LEAF7_FEATURE_AVX512BW);
371 setif(bits, kHasAVX512VL, cpuid_leaf7_features() &
372 CPUID_LEAF7_FEATURE_AVX512VL);
373 setif(bits, kHasAVX512IFMA, cpuid_leaf7_features() &
374 CPUID_LEAF7_FEATURE_AVX512IFMA);
375 setif(bits, kHasAVX512VBMI, cpuid_leaf7_features() &
376 CPUID_LEAF7_FEATURE_AVX512VBMI);
377 setif(bits, kHasVAES, cpuid_leaf7_features() &
378 CPUID_LEAF7_FEATURE_VAES);
379 setif(bits, kHasVPCLMULQDQ, cpuid_leaf7_features() &
380 CPUID_LEAF7_FEATURE_VPCLMULQDQ);
381 setif(bits, kHasAVX512VNNI, cpuid_leaf7_features() &
382 CPUID_LEAF7_FEATURE_AVX512VNNI);
383 setif(bits, kHasAVX512BITALG, cpuid_leaf7_features() &
384 CPUID_LEAF7_FEATURE_AVX512BITALG);
385 setif(bits, kHasAVX512VPOPCNTDQ, cpuid_leaf7_features() &
386 CPUID_LEAF7_FEATURE_AVX512VPCDQ);
387 }
388
389 if (cpuid_leaf7_features() & CPUID_LEAF7_FEATURE_ERMS) {
390 uint64_t misc_enable = rdmsr64(MSR_IA32_MISC_ENABLE);
391 setif(bits, kHasENFSTRG, (misc_enable & 1ULL));
392 }
393
394 _cpu_capabilities = bits; // set kernel version for use by drivers etc
395 }
396
397 /* initialize the approx_time_supported flag and set the approx time to 0.
398 * Called during initial commpage population.
399 */
400 static void
commpage_mach_approximate_time_init(void)401 commpage_mach_approximate_time_init(void)
402 {
403 char *cp = commPagePtr32;
404 uint8_t supported;
405
406 #ifdef CONFIG_MACH_APPROXIMATE_TIME
407 supported = 1;
408 #else
409 supported = 0;
410 #endif
411 if (cp) {
412 cp += (_COMM_PAGE_APPROX_TIME_SUPPORTED - _COMM_PAGE32_BASE_ADDRESS);
413 *(boolean_t *)cp = supported;
414 }
415
416 cp = commPagePtr64;
417 if (cp) {
418 cp += (_COMM_PAGE_APPROX_TIME_SUPPORTED - _COMM_PAGE32_START_ADDRESS);
419 *(boolean_t *)cp = supported;
420 }
421 commpage_update_mach_approximate_time(0);
422 }
423
424 static void
commpage_mach_continuous_time_init(void)425 commpage_mach_continuous_time_init(void)
426 {
427 commpage_update_mach_continuous_time(0);
428 }
429
430 static void
commpage_boottime_init(void)431 commpage_boottime_init(void)
432 {
433 clock_sec_t secs;
434 clock_usec_t microsecs;
435 clock_get_boottime_microtime(&secs, µsecs);
436 commpage_update_boottime(secs * USEC_PER_SEC + microsecs);
437 }
438
439 uint64_t
_get_cpu_capabilities(void)440 _get_cpu_capabilities(void)
441 {
442 return _cpu_capabilities;
443 }
444
445 /* Copy data into commpage. */
446
447 static void
commpage_stuff(commpage_address_t address,const void * source,int length)448 commpage_stuff(
449 commpage_address_t address,
450 const void *source,
451 int length )
452 {
453 void *dest = commpage_addr_of(address);
454
455 if (address < next) {
456 panic("commpage overlap at address 0x%p, 0x%x < 0x%x", dest, address, next);
457 }
458
459 bcopy(source, dest, length);
460
461 next = address + length;
462 }
463
464 /*
465 * Updates both the 32-bit and 64-bit commpages with the new data.
466 */
467 static void
commpage_update(commpage_address_t address,const void * source,int length)468 commpage_update(commpage_address_t address, const void *source, int length)
469 {
470 void *dest = commpage_specific_addr_of(commPagePtr32, address);
471 bcopy(source, dest, length);
472
473 dest = commpage_specific_addr_of(commPagePtr64, address);
474 bcopy(source, dest, length);
475 }
476
477 void
commpage_post_ucode_update(void)478 commpage_post_ucode_update(void)
479 {
480 commpage_init_cpu_capabilities();
481 commpage_update(_COMM_PAGE_CPU_CAPABILITIES64, &_cpu_capabilities, sizeof(_cpu_capabilities));
482 commpage_update(_COMM_PAGE_CPU_CAPABILITIES, &_cpu_capabilities, sizeof(uint32_t));
483 }
484
485 /* Copy a routine into comm page if it matches running machine.
486 */
487 static void
commpage_stuff_routine(commpage_descriptor * rd)488 commpage_stuff_routine(
489 commpage_descriptor *rd )
490 {
491 commpage_stuff(rd->commpage_address, rd->code_address, rd->code_length);
492 }
493
494
495 /* Fill in the 32- or 64-bit commpage. Called once for each.
496 */
497
498 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)499 commpage_populate_one(
500 vm_map_t submap, // commpage32_map or compage64_map
501 char ** kernAddressPtr, // &commPagePtr32 or &commPagePtr64
502 size_t area_used, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
503 commpage_address_t base_offset, // will become commPageBaseOffset
504 commpage_time_data** time_data, // &time_data32 or &time_data64
505 new_commpage_timeofday_data_t** gtod_time_data, // >od_time_data32 or >od_time_data64
506 const char* signature, // "commpage 32-bit" or "commpage 64-bit"
507 vm_prot_t uperm)
508 {
509 uint8_t c1;
510 uint16_t c2;
511 uint64_t c8;
512 uint8_t c256[256] = {0};
513 uint32_t cfamily;
514 short version = _COMM_PAGE_THIS_VERSION;
515
516 next = 0;
517 commPagePtr = (char *)commpage_allocate( submap, (vm_size_t) area_used, uperm );
518 *kernAddressPtr = commPagePtr; // save address either in commPagePtr32 or 64
519 commPageBaseOffset = base_offset;
520
521 *time_data = commpage_addr_of( _COMM_PAGE_TIME_DATA_START );
522 *gtod_time_data = commpage_addr_of( _COMM_PAGE_NEWTIMEOFDAY_DATA );
523
524 /* Stuff in the constants. We move things into the comm page in strictly
525 * ascending order, so we can check for overlap and panic if so.
526 * Note: the 32-bit cpu_capabilities vector is retained in addition to
527 * the expanded 64-bit vector.
528 */
529 commpage_stuff(_COMM_PAGE_SIGNATURE, signature, (int)MIN(_COMM_PAGE_SIGNATURELEN, strlen(signature)));
530 commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES64, &_cpu_capabilities, sizeof(_cpu_capabilities));
531 commpage_stuff(_COMM_PAGE_VERSION, &version, sizeof(short));
532 commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES, &_cpu_capabilities, sizeof(uint32_t));
533
534 c2 = 32; // default
535 if (_cpu_capabilities & kCache64) {
536 c2 = 64;
537 } else if (_cpu_capabilities & kCache128) {
538 c2 = 128;
539 }
540 commpage_stuff(_COMM_PAGE_CACHE_LINESIZE, &c2, 2);
541
542 /* machine_info valid after ml_wait_max_cpus() */
543 c1 = machine_info.physical_cpu_max;
544 commpage_stuff(_COMM_PAGE_PHYSICAL_CPUS, &c1, 1);
545 c1 = machine_info.logical_cpu_max;
546 commpage_stuff(_COMM_PAGE_LOGICAL_CPUS, &c1, 1);
547 c1 = ml_get_cluster_count();
548 commpage_stuff(_COMM_PAGE_CPU_CLUSTERS, &c1, 1);
549
550 c8 = ml_cpu_cache_size(0);
551 commpage_stuff(_COMM_PAGE_MEMORY_SIZE, &c8, 8);
552
553 cfamily = cpuid_info()->cpuid_cpufamily;
554 commpage_stuff(_COMM_PAGE_CPUFAMILY, &cfamily, 4);
555 c1 = PAGE_SHIFT;
556 commpage_stuff(_COMM_PAGE_KERNEL_PAGE_SHIFT, &c1, 1);
557 commpage_stuff(_COMM_PAGE_USER_PAGE_SHIFT_64, &c1, 1);
558
559 ml_map_cpus_to_clusters(c256);
560 commpage_stuff(_COMM_PAGE_CPU_TO_CLUSTER, c256, 256);
561
562 if (next > _COMM_PAGE_END) {
563 panic("commpage overflow: next = 0x%08x, commPagePtr = 0x%p", next, commPagePtr);
564 }
565 }
566
567
568 /* Fill in commpages: called once, during kernel initialization, from the
569 * startup thread before user-mode code is running.
570 *
571 * See the top of this file for a list of what you have to do to add
572 * a new routine to the commpage.
573 */
574
575 void
commpage_populate(void)576 commpage_populate( void )
577 {
578 commpage_init_cpu_capabilities();
579
580 commpage_populate_one( commpage32_map,
581 &commPagePtr32,
582 _COMM_PAGE32_AREA_USED,
583 _COMM_PAGE32_BASE_ADDRESS,
584 &time_data32,
585 >od_time_data32,
586 _COMM_PAGE32_SIGNATURE_STRING,
587 VM_PROT_READ);
588 #ifndef __LP64__
589 pmap_commpage32_init((vm_offset_t) commPagePtr32, _COMM_PAGE32_BASE_ADDRESS,
590 _COMM_PAGE32_AREA_USED / INTEL_PGBYTES);
591 #endif
592 time_data64 = time_data32; /* if no 64-bit commpage, point to 32-bit */
593 gtod_time_data64 = gtod_time_data32;
594
595 if (_cpu_capabilities & k64Bit) {
596 commpage_populate_one( commpage64_map,
597 &commPagePtr64,
598 _COMM_PAGE64_AREA_USED,
599 _COMM_PAGE32_START_ADDRESS, /* commpage address are relative to 32-bit commpage placement */
600 &time_data64,
601 >od_time_data64,
602 _COMM_PAGE64_SIGNATURE_STRING,
603 VM_PROT_READ);
604 #ifndef __LP64__
605 pmap_commpage64_init((vm_offset_t) commPagePtr64, _COMM_PAGE64_BASE_ADDRESS,
606 _COMM_PAGE64_AREA_USED / INTEL_PGBYTES);
607 #endif
608 }
609
610 simple_lock_init(&commpage_active_cpus_lock, 0);
611
612 commpage_update_active_cpus();
613 commpage_mach_approximate_time_init();
614 commpage_mach_continuous_time_init();
615 commpage_boottime_init();
616 rtc_nanotime_init_commpage();
617 commpage_update_kdebug_state();
618 #if CONFIG_ATM
619 commpage_update_atm_diagnostic_config(atm_get_diagnostic_config());
620 #endif
621
622 /*
623 * Set random values for targets in Apple Security Bounty
624 * addr should be unmapped for userland processes
625 * kaddr should be unmapped for kernel
626 */
627 uint64_t asb_value, asb_addr, asb_kvalue, asb_kaddr;
628 uint64_t asb_rand_vals[] = {
629 0x93e78adcded4d3d5, 0xd16c5b76ad99bccf, 0x67dfbbd12c4a594e, 0x7365636e6f6f544f,
630 0x239a974c9811e04b, 0xbf60e7fa45741446, 0x8acf5210b466b05, 0x67dfbbd12c4a594e
631 };
632 const int nrandval = sizeof(asb_rand_vals) / sizeof(asb_rand_vals[0]);
633 uint8_t randidx;
634 read_random(&randidx, sizeof(uint8_t));
635
636
637 asb_value = asb_rand_vals[randidx++ % nrandval];
638 commpage_update(_COMM_PAGE_ASB_TARGET_VALUE, &asb_value, sizeof(asb_value));
639
640 asb_addr = asb_rand_vals[randidx++ % nrandval];
641 uint64_t user_min = MACH_VM_MAX_ADDRESS;
642 uint64_t user_max = UINT64_MAX;
643 asb_addr %= (user_max - user_min);
644 asb_addr += user_min;
645 commpage_update(_COMM_PAGE_ASB_TARGET_ADDRESS, &asb_addr, sizeof(asb_addr));
646
647 asb_kvalue = asb_rand_vals[randidx++ % nrandval];
648 commpage_update(_COMM_PAGE_ASB_TARGET_KERN_VALUE, &asb_kvalue, sizeof(asb_kvalue));
649
650 asb_kaddr = asb_rand_vals[randidx++ % nrandval];
651 uint64_t kernel_min = 0x0LL;
652 uint64_t kernel_max = VM_MIN_KERNEL_ADDRESS;
653 asb_kaddr %= (kernel_max - kernel_min);
654 asb_kaddr += kernel_min;
655 commpage_update(_COMM_PAGE_ASB_TARGET_KERN_ADDRESS, &asb_kaddr, sizeof(asb_kaddr));
656
657 vm_map_seal(commpage32_map, true /* nested_pmap */);
658 vm_map_seal(commpage64_map, true /* nested_pmap */);
659 }
660
661 /* Fill in the common routines during kernel initialization.
662 * This is called before user-mode code is running.
663 */
664 void
commpage_text_populate(void)665 commpage_text_populate( void )
666 {
667 commpage_descriptor **rd;
668
669 next = 0;
670 commPagePtr = (char *) commpage_allocate(commpage_text32_map, (vm_size_t) _COMM_PAGE_TEXT_AREA_USED, VM_PROT_READ | VM_PROT_EXECUTE);
671 commPageTextPtr32 = commPagePtr;
672
673 char *cptr = commPagePtr;
674 int i = 0;
675 for (; i < _COMM_PAGE_TEXT_AREA_USED; i++) {
676 cptr[i] = 0xCC;
677 }
678
679 commPageBaseOffset = _COMM_PAGE_TEXT_START;
680 for (rd = commpage_32_routines; *rd != NULL; rd++) {
681 commpage_stuff_routine(*rd);
682 }
683
684 #ifndef __LP64__
685 pmap_commpage32_init((vm_offset_t) commPageTextPtr32, _COMM_PAGE_TEXT_START,
686 _COMM_PAGE_TEXT_AREA_USED / INTEL_PGBYTES);
687 #endif
688
689 if (_cpu_capabilities & k64Bit) {
690 next = 0;
691 commPagePtr = (char *) commpage_allocate(commpage_text64_map, (vm_size_t) _COMM_PAGE_TEXT_AREA_USED, VM_PROT_READ | VM_PROT_EXECUTE);
692 commPageTextPtr64 = commPagePtr;
693
694 cptr = commPagePtr;
695 for (i = 0; i < _COMM_PAGE_TEXT_AREA_USED; i++) {
696 cptr[i] = 0xCC;
697 }
698
699 for (rd = commpage_64_routines; *rd != NULL; rd++) {
700 commpage_stuff_routine(*rd);
701 }
702
703 #ifndef __LP64__
704 pmap_commpage64_init((vm_offset_t) commPageTextPtr64, _COMM_PAGE_TEXT_START,
705 _COMM_PAGE_TEXT_AREA_USED / INTEL_PGBYTES);
706 #endif
707 }
708
709 if (next > _COMM_PAGE_TEXT_END) {
710 panic("commpage text overflow: next=0x%08x, commPagePtr=%p", next, commPagePtr);
711 }
712
713 vm_map_seal(commpage_text32_map, true /* nested_pmap */);
714 vm_map_seal(commpage_text64_map, true /* nested_pmap */);
715 }
716
717 /* Update commpage nanotime information.
718 *
719 * This routine must be serialized by some external means, ie a lock.
720 */
721
722 void
commpage_set_nanotime(uint64_t tsc_base,uint64_t ns_base,uint32_t scale,uint32_t shift)723 commpage_set_nanotime(
724 uint64_t tsc_base,
725 uint64_t ns_base,
726 uint32_t scale,
727 uint32_t shift )
728 {
729 commpage_time_data *p32 = time_data32;
730 commpage_time_data *p64 = time_data64;
731 static uint32_t generation = 0;
732 uint32_t next_gen;
733
734 if (p32 == NULL) { /* have commpages been allocated yet? */
735 return;
736 }
737
738 if (generation != p32->nt_generation) {
739 panic("nanotime trouble 1"); /* possibly not serialized */
740 }
741 if (ns_base < p32->nt_ns_base) {
742 panic("nanotime trouble 2");
743 }
744 if ((shift != 0) && ((_cpu_capabilities & kSlow) == 0)) {
745 panic("nanotime trouble 3");
746 }
747
748 next_gen = ++generation;
749 if (next_gen == 0) {
750 next_gen = ++generation;
751 }
752
753 p32->nt_generation = 0; /* mark invalid, so commpage won't try to use it */
754 p64->nt_generation = 0;
755
756 p32->nt_tsc_base = tsc_base;
757 p64->nt_tsc_base = tsc_base;
758
759 p32->nt_ns_base = ns_base;
760 p64->nt_ns_base = ns_base;
761
762 p32->nt_scale = scale;
763 p64->nt_scale = scale;
764
765 p32->nt_shift = shift;
766 p64->nt_shift = shift;
767
768 p32->nt_generation = next_gen; /* mark data as valid */
769 p64->nt_generation = next_gen;
770 }
771
772 /* Update commpage gettimeofday() information. As with nanotime(), we interleave
773 * updates to the 32- and 64-bit commpage, in order to keep time more nearly in sync
774 * between the two environments.
775 *
776 * This routine must be serializeed by some external means, ie a lock.
777 */
778
779 void
commpage_set_timestamp(uint64_t abstime,uint64_t sec,uint64_t frac,uint64_t scale,uint64_t tick_per_sec)780 commpage_set_timestamp(
781 uint64_t abstime,
782 uint64_t sec,
783 uint64_t frac,
784 uint64_t scale,
785 uint64_t tick_per_sec)
786 {
787 new_commpage_timeofday_data_t *p32 = gtod_time_data32;
788 new_commpage_timeofday_data_t *p64 = gtod_time_data64;
789
790 p32->TimeStamp_tick = 0x0ULL;
791 p64->TimeStamp_tick = 0x0ULL;
792
793 p32->TimeStamp_sec = sec;
794 p64->TimeStamp_sec = sec;
795
796 p32->TimeStamp_frac = frac;
797 p64->TimeStamp_frac = frac;
798
799 p32->Ticks_scale = scale;
800 p64->Ticks_scale = scale;
801
802 p32->Ticks_per_sec = tick_per_sec;
803 p64->Ticks_per_sec = tick_per_sec;
804
805 p32->TimeStamp_tick = abstime;
806 p64->TimeStamp_tick = abstime;
807 }
808
809 /* Update _COMM_PAGE_MEMORY_PRESSURE. Called periodically from vm's compute_memory_pressure() */
810
811 void
commpage_set_memory_pressure(unsigned int pressure)812 commpage_set_memory_pressure(
813 unsigned int pressure )
814 {
815 char *cp;
816 uint32_t *ip;
817
818 cp = commPagePtr32;
819 if (cp) {
820 cp += (_COMM_PAGE_MEMORY_PRESSURE - _COMM_PAGE32_BASE_ADDRESS);
821 ip = (uint32_t*) (void *) cp;
822 *ip = (uint32_t) pressure;
823 }
824
825 cp = commPagePtr64;
826 if (cp) {
827 cp += (_COMM_PAGE_MEMORY_PRESSURE - _COMM_PAGE32_START_ADDRESS);
828 ip = (uint32_t*) (void *) cp;
829 *ip = (uint32_t) pressure;
830 }
831 }
832
833 /* Updated every time a logical CPU goes offline/online */
834 void
commpage_update_active_cpus(void)835 commpage_update_active_cpus(void)
836 {
837 char *cp;
838 volatile uint8_t *ip;
839
840 /* At least 32-bit commpage must be initialized */
841 if (!commPagePtr32) {
842 return;
843 }
844
845 simple_lock(&commpage_active_cpus_lock, LCK_GRP_NULL);
846
847 cp = commPagePtr32;
848 cp += (_COMM_PAGE_ACTIVE_CPUS - _COMM_PAGE32_BASE_ADDRESS);
849 ip = (volatile uint8_t*) cp;
850 *ip = (uint8_t) processor_avail_count_user;
851
852 cp = commPagePtr64;
853 if (cp) {
854 cp += (_COMM_PAGE_ACTIVE_CPUS - _COMM_PAGE32_START_ADDRESS);
855 ip = (volatile uint8_t*) cp;
856 *ip = (uint8_t) processor_avail_count_user;
857 }
858
859 simple_unlock(&commpage_active_cpus_lock);
860 }
861
862 /*
863 * Update the commpage with current kdebug state. This currently has bits for
864 * global trace state, and typefilter enablement. It is likely additional state
865 * will be tracked in the future.
866 *
867 * INVARIANT: This value will always be 0 if global tracing is disabled. This
868 * allows simple guard tests of "if (*_COMM_PAGE_KDEBUG_ENABLE) { ... }"
869 */
870 void
commpage_update_kdebug_state(void)871 commpage_update_kdebug_state(void)
872 {
873 volatile uint32_t *saved_data_ptr;
874 char *cp;
875
876 cp = commPagePtr32;
877 if (cp) {
878 cp += (_COMM_PAGE_KDEBUG_ENABLE - _COMM_PAGE32_BASE_ADDRESS);
879 saved_data_ptr = (volatile uint32_t *)cp;
880 *saved_data_ptr = kdebug_commpage_state();
881 }
882
883 cp = commPagePtr64;
884 if (cp) {
885 cp += (_COMM_PAGE_KDEBUG_ENABLE - _COMM_PAGE32_START_ADDRESS);
886 saved_data_ptr = (volatile uint32_t *)cp;
887 *saved_data_ptr = kdebug_commpage_state();
888 }
889 }
890
891 /* Ditto for atm_diagnostic_config */
892 void
commpage_update_atm_diagnostic_config(uint32_t diagnostic_config)893 commpage_update_atm_diagnostic_config(uint32_t diagnostic_config)
894 {
895 volatile uint32_t *saved_data_ptr;
896 char *cp;
897
898 cp = commPagePtr32;
899 if (cp) {
900 cp += (_COMM_PAGE_ATM_DIAGNOSTIC_CONFIG - _COMM_PAGE32_BASE_ADDRESS);
901 saved_data_ptr = (volatile uint32_t *)cp;
902 *saved_data_ptr = diagnostic_config;
903 }
904
905 cp = commPagePtr64;
906 if (cp) {
907 cp += (_COMM_PAGE_ATM_DIAGNOSTIC_CONFIG - _COMM_PAGE32_START_ADDRESS);
908 saved_data_ptr = (volatile uint32_t *)cp;
909 *saved_data_ptr = diagnostic_config;
910 }
911 }
912
913 /*
914 * update the commpage with if dtrace user land probes are enabled
915 */
916 void
commpage_update_dof(boolean_t enabled)917 commpage_update_dof(boolean_t enabled)
918 {
919 #if CONFIG_DTRACE
920 char *cp;
921
922 cp = commPagePtr32;
923 if (cp) {
924 cp += (_COMM_PAGE_DTRACE_DOF_ENABLED - _COMM_PAGE32_BASE_ADDRESS);
925 *cp = (enabled ? 1 : 0);
926 }
927
928 cp = commPagePtr64;
929 if (cp) {
930 cp += (_COMM_PAGE_DTRACE_DOF_ENABLED - _COMM_PAGE32_START_ADDRESS);
931 *cp = (enabled ? 1 : 0);
932 }
933 #else
934 (void)enabled;
935 #endif
936 }
937
938
939 /*
940 * update the dyld global config flags
941 */
942 void
commpage_update_dyld_flags(uint64_t value)943 commpage_update_dyld_flags(uint64_t value)
944 {
945 char *cp;
946
947 cp = commPagePtr32;
948 if (cp) {
949 cp += (_COMM_PAGE_DYLD_FLAGS - _COMM_PAGE32_BASE_ADDRESS);
950 *(uint64_t *)cp = value;
951 }
952
953 cp = commPagePtr64;
954 if (cp) {
955 cp += (_COMM_PAGE_DYLD_FLAGS - _COMM_PAGE32_BASE_ADDRESS);
956 *(uint64_t *)cp = value;
957 }
958 }
959
960
961 /*
962 * update the commpage data for last known value of mach_absolute_time()
963 */
964
965 void
commpage_update_mach_approximate_time(uint64_t abstime)966 commpage_update_mach_approximate_time(uint64_t abstime)
967 {
968 #ifdef CONFIG_MACH_APPROXIMATE_TIME
969 uint64_t saved_data;
970 char *cp;
971
972 cp = commPagePtr32;
973 if (cp) {
974 cp += (_COMM_PAGE_APPROX_TIME - _COMM_PAGE32_BASE_ADDRESS);
975 saved_data = atomic_load_explicit((_Atomic uint64_t *)(uintptr_t)cp, memory_order_relaxed);
976 if (saved_data < abstime) {
977 /* ignoring the success/fail return value assuming that
978 * if the value has been updated since we last read it,
979 * "someone" has a newer timestamp than us and ours is
980 * now invalid. */
981 atomic_compare_exchange_strong_explicit((_Atomic uint64_t *)(uintptr_t)cp,
982 &saved_data, abstime, memory_order_relaxed, memory_order_relaxed);
983 }
984 }
985 cp = commPagePtr64;
986 if (cp) {
987 cp += (_COMM_PAGE_APPROX_TIME - _COMM_PAGE32_START_ADDRESS);
988 saved_data = atomic_load_explicit((_Atomic uint64_t *)(uintptr_t)cp, memory_order_relaxed);
989 if (saved_data < abstime) {
990 /* ignoring the success/fail return value assuming that
991 * if the value has been updated since we last read it,
992 * "someone" has a newer timestamp than us and ours is
993 * now invalid. */
994 atomic_compare_exchange_strong_explicit((_Atomic uint64_t *)(uintptr_t)cp,
995 &saved_data, abstime, memory_order_relaxed, memory_order_relaxed);
996 }
997 }
998 #else
999 #pragma unused (abstime)
1000 #endif
1001 }
1002
1003 void
commpage_update_mach_continuous_time(uint64_t sleeptime)1004 commpage_update_mach_continuous_time(uint64_t sleeptime)
1005 {
1006 char *cp;
1007 cp = commPagePtr32;
1008 if (cp) {
1009 cp += (_COMM_PAGE_CONT_TIMEBASE - _COMM_PAGE32_START_ADDRESS);
1010 *(uint64_t *)cp = sleeptime;
1011 }
1012
1013 cp = commPagePtr64;
1014 if (cp) {
1015 cp += (_COMM_PAGE_CONT_TIMEBASE - _COMM_PAGE32_START_ADDRESS);
1016 *(uint64_t *)cp = sleeptime;
1017 }
1018 }
1019
1020 void
commpage_update_boottime(uint64_t boottime)1021 commpage_update_boottime(uint64_t boottime)
1022 {
1023 char *cp;
1024 cp = commPagePtr32;
1025 if (cp) {
1026 cp += (_COMM_PAGE_BOOTTIME_USEC - _COMM_PAGE32_START_ADDRESS);
1027 *(uint64_t *)cp = boottime;
1028 }
1029
1030 cp = commPagePtr64;
1031 if (cp) {
1032 cp += (_COMM_PAGE_BOOTTIME_USEC - _COMM_PAGE32_START_ADDRESS);
1033 *(uint64_t *)cp = boottime;
1034 }
1035 }
1036
1037
1038 extern user32_addr_t commpage_text32_location;
1039 extern user64_addr_t commpage_text64_location;
1040
1041 /* Check to see if a given address is in the Preemption Free Zone (PFZ) */
1042
1043 uint32_t
commpage_is_in_pfz32(uint32_t addr32)1044 commpage_is_in_pfz32(uint32_t addr32)
1045 {
1046 if ((addr32 >= (commpage_text32_location + _COMM_TEXT_PFZ_START_OFFSET))
1047 && (addr32 < (commpage_text32_location + _COMM_TEXT_PFZ_END_OFFSET))) {
1048 return 1;
1049 } else {
1050 return 0;
1051 }
1052 }
1053
1054 uint32_t
commpage_is_in_pfz64(addr64_t addr64)1055 commpage_is_in_pfz64(addr64_t addr64)
1056 {
1057 if ((addr64 >= (commpage_text64_location + _COMM_TEXT_PFZ_START_OFFSET))
1058 && (addr64 < (commpage_text64_location + _COMM_TEXT_PFZ_END_OFFSET))) {
1059 return 1;
1060 } else {
1061 return 0;
1062 }
1063 }
1064