]>
Commit | Line | Data |
---|---|---|
43866e37 | 1 | /* |
6d2010ae | 2 | * Copyright (c) 2003-2010 Apple Inc. All rights reserved. |
43866e37 | 3 | * |
2d21ac55 | 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
43866e37 | 5 | * |
2d21ac55 A |
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. | |
8f6c56a5 | 14 | * |
2d21ac55 A |
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 | |
8f6c56a5 A |
20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, | |
2d21ac55 A |
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. | |
8f6c56a5 | 25 | * |
2d21ac55 | 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
43866e37 A |
27 | */ |
28 | ||
55e303ae A |
29 | /* |
30 | * Here's what to do if you want to add a new routine to the comm page: | |
31 | * | |
0c530ab8 | 32 | * 1. Add a definition for it's address in osfmk/i386/cpu_capabilities.h, |
55e303ae A |
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. | |
0c530ab8 | 39 | * The source files should be in osfmk/i386/commpage/. |
55e303ae A |
40 | * |
41 | * 3. Add a ptr to your new commpage_descriptor(s) in the "routines" | |
0c530ab8 A |
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. | |
55e303ae A |
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> | |
91447636 | 50 | #include <mach/vm_map.h> |
b0d623f7 | 51 | #include <mach/mach_vm.h> |
7e4a7d39 A |
52 | #include <mach/machine.h> |
53 | #include <i386/cpuid.h> | |
2d21ac55 | 54 | #include <i386/tsc.h> |
6d2010ae | 55 | #include <i386/rtclock_protos.h> |
2d21ac55 | 56 | #include <i386/cpu_data.h> |
b0d623f7 A |
57 | #include <i386/machine_routines.h> |
58 | #include <i386/misc_protos.h> | |
7e4a7d39 | 59 | #include <i386/cpuid.h> |
43866e37 A |
60 | #include <machine/cpu_capabilities.h> |
61 | #include <machine/commpage.h> | |
55e303ae A |
62 | #include <machine/pmap.h> |
63 | #include <vm/vm_kern.h> | |
91447636 | 64 | #include <vm/vm_map.h> |
5ba3f43e | 65 | #include <stdatomic.h> |
b0d623f7 | 66 | |
91447636 A |
67 | #include <ipc/ipc_port.h> |
68 | ||
0c530ab8 | 69 | #include <kern/page_decrypt.h> |
6d2010ae | 70 | #include <kern/processor.h> |
4452a7af | 71 | |
a1c7dba1 A |
72 | #include <sys/kdebug.h> |
73 | ||
3e170ce0 A |
74 | #if CONFIG_ATM |
75 | #include <atm/atm_internal.h> | |
76 | #endif | |
77 | ||
0c530ab8 A |
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[]; | |
4452a7af | 81 | |
2d21ac55 A |
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 | |
316670eb A |
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 | ||
4452a7af | 87 | |
0c530ab8 A |
88 | char *commPagePtr32 = NULL; // virtual addr in kernel map of 32-bit commpage |
89 | char *commPagePtr64 = NULL; // ...and of 64-bit commpage | |
bd504ef0 A |
90 | char *commPageTextPtr32 = NULL; // virtual addr in kernel map of 32-bit commpage |
91 | char *commPageTextPtr64 = NULL; // ...and of 64-bit commpage | |
6601e61a | 92 | |
bd504ef0 | 93 | uint64_t _cpu_capabilities = 0; // define the capability vector |
0c530ab8 | 94 | |
b0d623f7 A |
95 | typedef uint32_t commpage_address_t; |
96 | ||
bd504ef0 | 97 | static commpage_address_t next; // next available address in comm page |
0c530ab8 A |
98 | |
99 | static char *commPagePtr; // virtual addr in kernel map of commpage we are working on | |
b0d623f7 | 100 | static commpage_address_t commPageBaseOffset; // subtract from 32-bit runtime address to get offset in virtual commpage in kernel map |
55e303ae | 101 | |
2d21ac55 A |
102 | static commpage_time_data *time_data32 = NULL; |
103 | static commpage_time_data *time_data64 = NULL; | |
5ba3f43e A |
104 | static new_commpage_timeofday_data_t *gtod_time_data32 = NULL; |
105 | static new_commpage_timeofday_data_t *gtod_time_data64 = NULL; | |
106 | ||
2d21ac55 | 107 | |
6d2010ae A |
108 | decl_simple_lock_data(static,commpage_active_cpus_lock); |
109 | ||
55e303ae A |
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* | |
0c530ab8 | 118 | commpage_allocate( |
2d21ac55 | 119 | vm_map_t submap, // commpage32_map or commpage_map64 |
316670eb A |
120 | size_t area_used, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED |
121 | vm_prot_t uperm) | |
55e303ae | 122 | { |
2d21ac55 | 123 | vm_offset_t kernel_addr = 0; // address of commpage in kernel map |
0c530ab8 A |
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; | |
316670eb | 128 | kern_return_t kr; |
0c530ab8 A |
129 | |
130 | if (submap == NULL) | |
131 | panic("commpage submap is null"); | |
132 | ||
5ba3f43e | 133 | if ((kr = vm_map_kernel(kernel_map, |
3e170ce0 A |
134 | &kernel_addr, |
135 | area_used, | |
136 | 0, | |
5ba3f43e A |
137 | VM_FLAGS_ANYWHERE, |
138 | VM_KERN_MEMORY_OSFMK, | |
3e170ce0 A |
139 | NULL, |
140 | 0, | |
141 | FALSE, | |
142 | VM_PROT_ALL, | |
143 | VM_PROT_ALL, | |
144 | VM_INHERIT_NONE))) | |
316670eb | 145 | panic("cannot allocate commpage %d", kr); |
0c530ab8 | 146 | |
5ba3f43e | 147 | if ((kr = vm_map_wire_kernel(kernel_map, |
3e170ce0 A |
148 | kernel_addr, |
149 | kernel_addr+area_used, | |
5ba3f43e | 150 | VM_PROT_DEFAULT, VM_KERN_MEMORY_OSFMK, |
3e170ce0 | 151 | FALSE))) |
316670eb | 152 | panic("cannot wire commpage: %d", kr); |
0c530ab8 A |
153 | |
154 | /* | |
155 | * Now that the object is created and wired into the kernel map, mark it so that no delay | |
156 | * copy-on-write will ever be performed on it as a result of mapping it into user-space. | |
157 | * If such a delayed copy ever occurred, we could remove the kernel's wired mapping - and | |
158 | * that would be a real disaster. | |
159 | * | |
160 | * JMM - What we really need is a way to create it like this in the first place. | |
161 | */ | |
39236c6e | 162 | 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)) |
316670eb | 163 | panic("cannot find commpage entry %d", kr); |
3e170ce0 | 164 | VME_OBJECT(entry)->copy_strategy = MEMORY_OBJECT_COPY_NONE; |
0c530ab8 | 165 | |
316670eb | 166 | if ((kr = mach_make_memory_entry( kernel_map, // target map |
0c530ab8 A |
167 | &size, // size |
168 | kernel_addr, // offset (address in kernel map) | |
316670eb | 169 | uperm, // protections as specified |
0c530ab8 | 170 | &handle, // this is the object handle we get |
316670eb A |
171 | NULL ))) // parent_entry (what is this?) |
172 | panic("cannot make entry for commpage %d", kr); | |
0c530ab8 | 173 | |
5ba3f43e | 174 | if ((kr = vm_map_64_kernel( submap, // target map (shared submap) |
0c530ab8 A |
175 | &zero, // address (map into 1st page in submap) |
176 | area_used, // size | |
177 | 0, // mask | |
178 | VM_FLAGS_FIXED, // flags (it must be 1st page in submap) | |
5ba3f43e | 179 | VM_KERN_MEMORY_NONE, |
0c530ab8 A |
180 | handle, // port is the memory entry we just made |
181 | 0, // offset (map 1st page in memory entry) | |
182 | FALSE, // copy | |
316670eb A |
183 | uperm, // cur_protection (R-only in user map) |
184 | uperm, // max_protection | |
185 | VM_INHERIT_SHARE ))) // inheritance | |
186 | panic("cannot map commpage %d", kr); | |
0c530ab8 A |
187 | |
188 | ipc_port_release(handle); | |
316670eb A |
189 | /* Make the kernel mapping non-executable. This cannot be done |
190 | * at the time of map entry creation as mach_make_memory_entry | |
191 | * cannot handle disjoint permissions at this time. | |
192 | */ | |
193 | kr = vm_protect(kernel_map, kernel_addr, area_used, FALSE, VM_PROT_READ | VM_PROT_WRITE); | |
194 | assert (kr == KERN_SUCCESS); | |
0c530ab8 | 195 | |
b0d623f7 | 196 | return (void*)(intptr_t)kernel_addr; // return address in kernel map |
55e303ae A |
197 | } |
198 | ||
199 | /* Get address (in kernel map) of a commpage field. */ | |
200 | ||
91447636 | 201 | static void* |
55e303ae | 202 | commpage_addr_of( |
b0d623f7 | 203 | commpage_address_t addr_at_runtime ) |
55e303ae | 204 | { |
b0d623f7 | 205 | return (void*) ((uintptr_t)commPagePtr + (addr_at_runtime - commPageBaseOffset)); |
55e303ae A |
206 | } |
207 | ||
208 | /* Determine number of CPUs on this system. We cannot rely on | |
209 | * machine_info.max_cpus this early in the boot. | |
210 | */ | |
211 | static int | |
212 | commpage_cpus( void ) | |
213 | { | |
214 | int cpus; | |
215 | ||
216 | cpus = ml_get_max_cpus(); // NB: this call can block | |
217 | ||
218 | if (cpus == 0) | |
219 | panic("commpage cpus==0"); | |
220 | if (cpus > 0xFF) | |
221 | cpus = 0xFF; | |
222 | ||
223 | return cpus; | |
224 | } | |
43866e37 | 225 | |
55e303ae | 226 | /* Initialize kernel version of _cpu_capabilities vector (used by KEXTs.) */ |
43866e37 | 227 | |
55e303ae A |
228 | static void |
229 | commpage_init_cpu_capabilities( void ) | |
230 | { | |
bd504ef0 | 231 | uint64_t bits; |
55e303ae A |
232 | int cpus; |
233 | ml_cpu_info_t cpu_info; | |
43866e37 | 234 | |
55e303ae A |
235 | bits = 0; |
236 | ml_cpu_get_info(&cpu_info); | |
237 | ||
238 | switch (cpu_info.vector_unit) { | |
6d2010ae A |
239 | case 9: |
240 | bits |= kHasAVX1_0; | |
241 | /* fall thru */ | |
2d21ac55 A |
242 | case 8: |
243 | bits |= kHasSSE4_2; | |
244 | /* fall thru */ | |
245 | case 7: | |
246 | bits |= kHasSSE4_1; | |
247 | /* fall thru */ | |
0c530ab8 A |
248 | case 6: |
249 | bits |= kHasSupplementalSSE3; | |
250 | /* fall thru */ | |
55e303ae | 251 | case 5: |
91447636 | 252 | bits |= kHasSSE3; |
55e303ae A |
253 | /* fall thru */ |
254 | case 4: | |
255 | bits |= kHasSSE2; | |
256 | /* fall thru */ | |
257 | case 3: | |
258 | bits |= kHasSSE; | |
259 | /* fall thru */ | |
260 | case 2: | |
261 | bits |= kHasMMX; | |
262 | default: | |
263 | break; | |
264 | } | |
265 | switch (cpu_info.cache_line_size) { | |
266 | case 128: | |
267 | bits |= kCache128; | |
268 | break; | |
269 | case 64: | |
270 | bits |= kCache64; | |
271 | break; | |
272 | case 32: | |
273 | bits |= kCache32; | |
274 | break; | |
275 | default: | |
276 | break; | |
277 | } | |
278 | cpus = commpage_cpus(); // how many CPUs do we have | |
279 | ||
55e303ae A |
280 | bits |= (cpus << kNumCPUsShift); |
281 | ||
91447636 A |
282 | bits |= kFastThreadLocalStorage; // we use %gs for TLS |
283 | ||
bd504ef0 A |
284 | #define setif(_bits, _bit, _condition) \ |
285 | if (_condition) _bits |= _bit | |
286 | ||
287 | setif(bits, kUP, cpus == 1); | |
288 | setif(bits, k64Bit, cpu_mode_is64bit()); | |
289 | setif(bits, kSlow, tscFreq <= SLOW_TSC_THRESHOLD); | |
290 | ||
291 | setif(bits, kHasAES, cpuid_features() & | |
292 | CPUID_FEATURE_AES); | |
293 | setif(bits, kHasF16C, cpuid_features() & | |
294 | CPUID_FEATURE_F16C); | |
295 | setif(bits, kHasRDRAND, cpuid_features() & | |
296 | CPUID_FEATURE_RDRAND); | |
297 | setif(bits, kHasFMA, cpuid_features() & | |
298 | CPUID_FEATURE_FMA); | |
299 | ||
300 | setif(bits, kHasBMI1, cpuid_leaf7_features() & | |
301 | CPUID_LEAF7_FEATURE_BMI1); | |
302 | setif(bits, kHasBMI2, cpuid_leaf7_features() & | |
303 | CPUID_LEAF7_FEATURE_BMI2); | |
304 | setif(bits, kHasRTM, cpuid_leaf7_features() & | |
305 | CPUID_LEAF7_FEATURE_RTM); | |
306 | setif(bits, kHasHLE, cpuid_leaf7_features() & | |
307 | CPUID_LEAF7_FEATURE_HLE); | |
308 | setif(bits, kHasAVX2_0, cpuid_leaf7_features() & | |
309 | CPUID_LEAF7_FEATURE_AVX2); | |
a1c7dba1 A |
310 | setif(bits, kHasRDSEED, cpuid_features() & |
311 | CPUID_LEAF7_FEATURE_RDSEED); | |
312 | setif(bits, kHasADX, cpuid_features() & | |
313 | CPUID_LEAF7_FEATURE_ADX); | |
bd504ef0 | 314 | |
5ba3f43e | 315 | #if 0 /* The kernel doesn't support MPX or SGX */ |
2dced7af A |
316 | setif(bits, kHasMPX, cpuid_leaf7_features() & |
317 | CPUID_LEAF7_FEATURE_MPX); | |
318 | setif(bits, kHasSGX, cpuid_leaf7_features() & | |
319 | CPUID_LEAF7_FEATURE_SGX); | |
5ba3f43e A |
320 | #endif |
321 | ||
322 | #if !defined(RC_HIDE_XNU_J137) | |
323 | if (ml_fpu_avx512_enabled()) { | |
324 | setif(bits, kHasAVX512F, cpuid_leaf7_features() & | |
325 | CPUID_LEAF7_FEATURE_AVX512F); | |
326 | setif(bits, kHasAVX512CD, cpuid_leaf7_features() & | |
327 | CPUID_LEAF7_FEATURE_AVX512CD); | |
328 | setif(bits, kHasAVX512DQ, cpuid_leaf7_features() & | |
329 | CPUID_LEAF7_FEATURE_AVX512DQ); | |
330 | setif(bits, kHasAVX512BW, cpuid_leaf7_features() & | |
331 | CPUID_LEAF7_FEATURE_AVX512BW); | |
332 | setif(bits, kHasAVX512VL, cpuid_leaf7_features() & | |
333 | CPUID_LEAF7_FEATURE_AVX512VL); | |
334 | setif(bits, kHasAVX512IFMA, cpuid_leaf7_features() & | |
335 | CPUID_LEAF7_FEATURE_AVX512IFMA); | |
336 | setif(bits, kHasAVX512VBMI, cpuid_leaf7_features() & | |
337 | CPUID_LEAF7_FEATURE_AVX512VBMI); | |
338 | } | |
339 | ||
340 | #endif /* not RC_HIDE_XNU_J137 */ | |
bd504ef0 A |
341 | uint64_t misc_enable = rdmsr64(MSR_IA32_MISC_ENABLE); |
342 | setif(bits, kHasENFSTRG, (misc_enable & 1ULL) && | |
343 | (cpuid_leaf7_features() & | |
fe8ab488 | 344 | CPUID_LEAF7_FEATURE_ERMS)); |
bd504ef0 | 345 | |
55e303ae A |
346 | _cpu_capabilities = bits; // set kernel version for use by drivers etc |
347 | } | |
348 | ||
fe8ab488 A |
349 | /* initialize the approx_time_supported flag and set the approx time to 0. |
350 | * Called during initial commpage population. | |
351 | */ | |
352 | static void | |
353 | commpage_mach_approximate_time_init(void) | |
354 | { | |
39037602 | 355 | char *cp = commPagePtr32; |
fe8ab488 A |
356 | uint8_t supported; |
357 | ||
358 | #ifdef CONFIG_MACH_APPROXIMATE_TIME | |
359 | supported = 1; | |
360 | #else | |
361 | supported = 0; | |
362 | #endif | |
363 | if ( cp ) { | |
39037602 | 364 | cp += (_COMM_PAGE_APPROX_TIME_SUPPORTED - _COMM_PAGE32_BASE_ADDRESS); |
fe8ab488 A |
365 | *(boolean_t *)cp = supported; |
366 | } | |
39037602 A |
367 | |
368 | cp = commPagePtr64; | |
fe8ab488 | 369 | if ( cp ) { |
39037602 | 370 | cp += (_COMM_PAGE_APPROX_TIME_SUPPORTED - _COMM_PAGE32_START_ADDRESS); |
fe8ab488 A |
371 | *(boolean_t *)cp = supported; |
372 | } | |
373 | commpage_update_mach_approximate_time(0); | |
374 | } | |
375 | ||
39037602 A |
376 | static void |
377 | commpage_mach_continuous_time_init(void) | |
378 | { | |
379 | commpage_update_mach_continuous_time(0); | |
380 | } | |
381 | ||
382 | static void | |
383 | commpage_boottime_init(void) | |
384 | { | |
385 | clock_sec_t secs; | |
386 | clock_usec_t microsecs; | |
387 | clock_get_boottime_microtime(&secs, µsecs); | |
388 | commpage_update_boottime(secs * USEC_PER_SEC + microsecs); | |
389 | } | |
fe8ab488 | 390 | |
bd504ef0 | 391 | uint64_t |
2d21ac55 | 392 | _get_cpu_capabilities(void) |
0c530ab8 A |
393 | { |
394 | return _cpu_capabilities; | |
395 | } | |
396 | ||
55e303ae A |
397 | /* Copy data into commpage. */ |
398 | ||
399 | static void | |
400 | commpage_stuff( | |
b0d623f7 | 401 | commpage_address_t address, |
0c530ab8 | 402 | const void *source, |
55e303ae A |
403 | int length ) |
404 | { | |
405 | void *dest = commpage_addr_of(address); | |
406 | ||
b0d623f7 | 407 | if (address < next) |
6d2010ae | 408 | panic("commpage overlap at address 0x%p, 0x%x < 0x%x", dest, address, next); |
55e303ae A |
409 | |
410 | bcopy(source,dest,length); | |
43866e37 | 411 | |
b0d623f7 | 412 | next = address + length; |
55e303ae A |
413 | } |
414 | ||
415 | /* Copy a routine into comm page if it matches running machine. | |
416 | */ | |
417 | static void | |
418 | commpage_stuff_routine( | |
bd504ef0 | 419 | commpage_descriptor *rd ) |
55e303ae | 420 | { |
bd504ef0 | 421 | commpage_stuff(rd->commpage_address,rd->code_address,rd->code_length); |
55e303ae A |
422 | } |
423 | ||
0c530ab8 | 424 | /* Fill in the 32- or 64-bit commpage. Called once for each. |
55e303ae A |
425 | */ |
426 | ||
0c530ab8 A |
427 | static void |
428 | commpage_populate_one( | |
2d21ac55 | 429 | vm_map_t submap, // commpage32_map or compage64_map |
0c530ab8 A |
430 | char ** kernAddressPtr, // &commPagePtr32 or &commPagePtr64 |
431 | size_t area_used, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED | |
b0d623f7 | 432 | commpage_address_t base_offset, // will become commPageBaseOffset |
2d21ac55 | 433 | commpage_time_data** time_data, // &time_data32 or &time_data64 |
5ba3f43e | 434 | new_commpage_timeofday_data_t** gtod_time_data, // >od_time_data32 or >od_time_data64 |
316670eb A |
435 | const char* signature, // "commpage 32-bit" or "commpage 64-bit" |
436 | vm_prot_t uperm) | |
55e303ae | 437 | { |
bd504ef0 A |
438 | uint8_t c1; |
439 | uint16_t c2; | |
440 | int c4; | |
441 | uint64_t c8; | |
6d2010ae | 442 | uint32_t cfamily; |
55e303ae | 443 | short version = _COMM_PAGE_THIS_VERSION; |
55e303ae | 444 | |
b0d623f7 | 445 | next = 0; |
316670eb | 446 | commPagePtr = (char *)commpage_allocate( submap, (vm_size_t) area_used, uperm ); |
0c530ab8 A |
447 | *kernAddressPtr = commPagePtr; // save address either in commPagePtr32 or 64 |
448 | commPageBaseOffset = base_offset; | |
b0d623f7 | 449 | |
2d21ac55 | 450 | *time_data = commpage_addr_of( _COMM_PAGE_TIME_DATA_START ); |
5ba3f43e | 451 | *gtod_time_data = commpage_addr_of( _COMM_PAGE_NEWTIMEOFDAY_DATA ); |
55e303ae A |
452 | |
453 | /* Stuff in the constants. We move things into the comm page in strictly | |
454 | * ascending order, so we can check for overlap and panic if so. | |
bd504ef0 A |
455 | * Note: the 32-bit cpu_capabilities vector is retained in addition to |
456 | * the expanded 64-bit vector. | |
55e303ae | 457 | */ |
bd504ef0 A |
458 | commpage_stuff(_COMM_PAGE_SIGNATURE,signature,(int)MIN(_COMM_PAGE_SIGNATURELEN, strlen(signature))); |
459 | commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES64,&_cpu_capabilities,sizeof(_cpu_capabilities)); | |
6d2010ae | 460 | commpage_stuff(_COMM_PAGE_VERSION,&version,sizeof(short)); |
bd504ef0 | 461 | commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES,&_cpu_capabilities,sizeof(uint32_t)); |
0c530ab8 | 462 | |
6d2010ae A |
463 | c2 = 32; // default |
464 | if (_cpu_capabilities & kCache64) | |
91447636 A |
465 | c2 = 64; |
466 | else if (_cpu_capabilities & kCache128) | |
467 | c2 = 128; | |
468 | commpage_stuff(_COMM_PAGE_CACHE_LINESIZE,&c2,2); | |
bd504ef0 | 469 | |
b0d623f7 A |
470 | c4 = MP_SPIN_TRIES; |
471 | commpage_stuff(_COMM_PAGE_SPIN_COUNT,&c4,4); | |
91447636 | 472 | |
6d2010ae A |
473 | /* machine_info valid after ml_get_max_cpus() */ |
474 | c1 = machine_info.physical_cpu_max; | |
475 | commpage_stuff(_COMM_PAGE_PHYSICAL_CPUS,&c1,1); | |
476 | c1 = machine_info.logical_cpu_max; | |
477 | commpage_stuff(_COMM_PAGE_LOGICAL_CPUS,&c1,1); | |
478 | ||
479 | c8 = ml_cpu_cache_size(0); | |
480 | commpage_stuff(_COMM_PAGE_MEMORY_SIZE, &c8, 8); | |
481 | ||
482 | cfamily = cpuid_info()->cpuid_cpufamily; | |
483 | commpage_stuff(_COMM_PAGE_CPUFAMILY, &cfamily, 4); | |
6601e61a | 484 | |
b0d623f7 A |
485 | if (next > _COMM_PAGE_END) |
486 | panic("commpage overflow: next = 0x%08x, commPagePtr = 0x%p", next, commPagePtr); | |
91447636 | 487 | |
43866e37 | 488 | } |
91447636 | 489 | |
0c530ab8 A |
490 | |
491 | /* Fill in commpages: called once, during kernel initialization, from the | |
492 | * startup thread before user-mode code is running. | |
493 | * | |
494 | * See the top of this file for a list of what you have to do to add | |
495 | * a new routine to the commpage. | |
496 | */ | |
91447636 A |
497 | |
498 | void | |
0c530ab8 | 499 | commpage_populate( void ) |
91447636 | 500 | { |
0c530ab8 A |
501 | commpage_init_cpu_capabilities(); |
502 | ||
2d21ac55 | 503 | commpage_populate_one( commpage32_map, |
0c530ab8 A |
504 | &commPagePtr32, |
505 | _COMM_PAGE32_AREA_USED, | |
506 | _COMM_PAGE32_BASE_ADDRESS, | |
2d21ac55 | 507 | &time_data32, |
5ba3f43e | 508 | >od_time_data32, |
316670eb A |
509 | "commpage 32-bit", |
510 | VM_PROT_READ); | |
b0d623f7 | 511 | #ifndef __LP64__ |
0c530ab8 A |
512 | pmap_commpage32_init((vm_offset_t) commPagePtr32, _COMM_PAGE32_BASE_ADDRESS, |
513 | _COMM_PAGE32_AREA_USED/INTEL_PGBYTES); | |
b0d623f7 | 514 | #endif |
2d21ac55 | 515 | time_data64 = time_data32; /* if no 64-bit commpage, point to 32-bit */ |
5ba3f43e | 516 | gtod_time_data64 = gtod_time_data32; |
0c530ab8 A |
517 | |
518 | if (_cpu_capabilities & k64Bit) { | |
2d21ac55 | 519 | commpage_populate_one( commpage64_map, |
0c530ab8 A |
520 | &commPagePtr64, |
521 | _COMM_PAGE64_AREA_USED, | |
b0d623f7 | 522 | _COMM_PAGE32_START_ADDRESS, /* commpage address are relative to 32-bit commpage placement */ |
2d21ac55 | 523 | &time_data64, |
5ba3f43e | 524 | >od_time_data64, |
316670eb A |
525 | "commpage 64-bit", |
526 | VM_PROT_READ); | |
b0d623f7 | 527 | #ifndef __LP64__ |
0c530ab8 A |
528 | pmap_commpage64_init((vm_offset_t) commPagePtr64, _COMM_PAGE64_BASE_ADDRESS, |
529 | _COMM_PAGE64_AREA_USED/INTEL_PGBYTES); | |
b0d623f7 | 530 | #endif |
0c530ab8 | 531 | } |
6601e61a | 532 | |
6d2010ae A |
533 | simple_lock_init(&commpage_active_cpus_lock, 0); |
534 | ||
535 | commpage_update_active_cpus(); | |
a1c7dba1 | 536 | commpage_mach_approximate_time_init(); |
39037602 A |
537 | commpage_mach_continuous_time_init(); |
538 | commpage_boottime_init(); | |
0c530ab8 | 539 | rtc_nanotime_init_commpage(); |
39037602 | 540 | commpage_update_kdebug_state(); |
3e170ce0 A |
541 | #if CONFIG_ATM |
542 | commpage_update_atm_diagnostic_config(atm_get_diagnostic_config()); | |
543 | #endif | |
91447636 | 544 | } |
2d21ac55 | 545 | |
316670eb A |
546 | /* Fill in the common routines during kernel initialization. |
547 | * This is called before user-mode code is running. | |
548 | */ | |
549 | void commpage_text_populate( void ){ | |
550 | commpage_descriptor **rd; | |
551 | ||
bd504ef0 | 552 | next = 0; |
316670eb A |
553 | commPagePtr = (char *) commpage_allocate(commpage_text32_map, (vm_size_t) _COMM_PAGE_TEXT_AREA_USED, VM_PROT_READ | VM_PROT_EXECUTE); |
554 | commPageTextPtr32 = commPagePtr; | |
555 | ||
556 | char *cptr = commPagePtr; | |
557 | int i=0; | |
558 | for(; i< _COMM_PAGE_TEXT_AREA_USED; i++){ | |
559 | cptr[i]=0xCC; | |
560 | } | |
561 | ||
562 | commPageBaseOffset = _COMM_PAGE_TEXT_START; | |
563 | for (rd = commpage_32_routines; *rd != NULL; rd++) { | |
564 | commpage_stuff_routine(*rd); | |
565 | } | |
316670eb A |
566 | |
567 | #ifndef __LP64__ | |
568 | pmap_commpage32_init((vm_offset_t) commPageTextPtr32, _COMM_PAGE_TEXT_START, | |
569 | _COMM_PAGE_TEXT_AREA_USED/INTEL_PGBYTES); | |
570 | #endif | |
571 | ||
572 | if (_cpu_capabilities & k64Bit) { | |
bd504ef0 | 573 | next = 0; |
316670eb A |
574 | commPagePtr = (char *) commpage_allocate(commpage_text64_map, (vm_size_t) _COMM_PAGE_TEXT_AREA_USED, VM_PROT_READ | VM_PROT_EXECUTE); |
575 | commPageTextPtr64 = commPagePtr; | |
576 | ||
577 | cptr=commPagePtr; | |
578 | for(i=0; i<_COMM_PAGE_TEXT_AREA_USED; i++){ | |
579 | cptr[i]=0xCC; | |
580 | } | |
581 | ||
582 | for (rd = commpage_64_routines; *rd !=NULL; rd++) { | |
583 | commpage_stuff_routine(*rd); | |
584 | } | |
585 | ||
586 | #ifndef __LP64__ | |
587 | pmap_commpage64_init((vm_offset_t) commPageTextPtr64, _COMM_PAGE_TEXT_START, | |
588 | _COMM_PAGE_TEXT_AREA_USED/INTEL_PGBYTES); | |
589 | #endif | |
590 | } | |
591 | ||
316670eb A |
592 | if (next > _COMM_PAGE_TEXT_END) |
593 | panic("commpage text overflow: next=0x%08x, commPagePtr=%p", next, commPagePtr); | |
594 | ||
595 | } | |
2d21ac55 | 596 | |
bd504ef0 | 597 | /* Update commpage nanotime information. |
2d21ac55 A |
598 | * |
599 | * This routine must be serialized by some external means, ie a lock. | |
600 | */ | |
601 | ||
602 | void | |
603 | commpage_set_nanotime( | |
604 | uint64_t tsc_base, | |
605 | uint64_t ns_base, | |
606 | uint32_t scale, | |
607 | uint32_t shift ) | |
608 | { | |
609 | commpage_time_data *p32 = time_data32; | |
610 | commpage_time_data *p64 = time_data64; | |
611 | static uint32_t generation = 0; | |
612 | uint32_t next_gen; | |
613 | ||
614 | if (p32 == NULL) /* have commpages been allocated yet? */ | |
615 | return; | |
616 | ||
617 | if ( generation != p32->nt_generation ) | |
618 | panic("nanotime trouble 1"); /* possibly not serialized */ | |
619 | if ( ns_base < p32->nt_ns_base ) | |
620 | panic("nanotime trouble 2"); | |
bd504ef0 | 621 | if ((shift != 0) && ((_cpu_capabilities & kSlow)==0) ) |
2d21ac55 A |
622 | panic("nanotime trouble 3"); |
623 | ||
624 | next_gen = ++generation; | |
625 | if (next_gen == 0) | |
626 | next_gen = ++generation; | |
627 | ||
628 | p32->nt_generation = 0; /* mark invalid, so commpage won't try to use it */ | |
629 | p64->nt_generation = 0; | |
630 | ||
631 | p32->nt_tsc_base = tsc_base; | |
632 | p64->nt_tsc_base = tsc_base; | |
633 | ||
634 | p32->nt_ns_base = ns_base; | |
635 | p64->nt_ns_base = ns_base; | |
636 | ||
637 | p32->nt_scale = scale; | |
638 | p64->nt_scale = scale; | |
639 | ||
640 | p32->nt_shift = shift; | |
641 | p64->nt_shift = shift; | |
642 | ||
643 | p32->nt_generation = next_gen; /* mark data as valid */ | |
644 | p64->nt_generation = next_gen; | |
645 | } | |
646 | ||
2d21ac55 A |
647 | /* Update commpage gettimeofday() information. As with nanotime(), we interleave |
648 | * updates to the 32- and 64-bit commpage, in order to keep time more nearly in sync | |
649 | * between the two environments. | |
650 | * | |
651 | * This routine must be serializeed by some external means, ie a lock. | |
652 | */ | |
5ba3f43e A |
653 | |
654 | void | |
655 | commpage_set_timestamp( | |
656 | uint64_t abstime, | |
657 | uint64_t sec, | |
658 | uint64_t frac, | |
659 | uint64_t scale, | |
660 | uint64_t tick_per_sec) | |
2d21ac55 | 661 | { |
5ba3f43e A |
662 | new_commpage_timeofday_data_t *p32 = gtod_time_data32; |
663 | new_commpage_timeofday_data_t *p64 = gtod_time_data64; | |
2d21ac55 | 664 | |
5ba3f43e A |
665 | p32->TimeStamp_tick = 0x0ULL; |
666 | p64->TimeStamp_tick = 0x0ULL; | |
667 | ||
668 | p32->TimeStamp_sec = sec; | |
669 | p64->TimeStamp_sec = sec; | |
670 | ||
671 | p32->TimeStamp_frac = frac; | |
672 | p64->TimeStamp_frac = frac; | |
b0d623f7 | 673 | |
5ba3f43e A |
674 | p32->Ticks_scale = scale; |
675 | p64->Ticks_scale = scale; | |
676 | ||
677 | p32->Ticks_per_sec = tick_per_sec; | |
678 | p64->Ticks_per_sec = tick_per_sec; | |
679 | ||
680 | p32->TimeStamp_tick = abstime; | |
681 | p64->TimeStamp_tick = abstime; | |
682 | } | |
b0d623f7 A |
683 | |
684 | /* Update _COMM_PAGE_MEMORY_PRESSURE. Called periodically from vm's compute_memory_pressure() */ | |
685 | ||
686 | void | |
687 | commpage_set_memory_pressure( | |
688 | unsigned int pressure ) | |
689 | { | |
690 | char *cp; | |
691 | uint32_t *ip; | |
692 | ||
693 | cp = commPagePtr32; | |
694 | if ( cp ) { | |
695 | cp += (_COMM_PAGE_MEMORY_PRESSURE - _COMM_PAGE32_BASE_ADDRESS); | |
bd504ef0 | 696 | ip = (uint32_t*) (void *) cp; |
b0d623f7 A |
697 | *ip = (uint32_t) pressure; |
698 | } | |
699 | ||
700 | cp = commPagePtr64; | |
701 | if ( cp ) { | |
702 | cp += (_COMM_PAGE_MEMORY_PRESSURE - _COMM_PAGE32_START_ADDRESS); | |
bd504ef0 | 703 | ip = (uint32_t*) (void *) cp; |
b0d623f7 A |
704 | *ip = (uint32_t) pressure; |
705 | } | |
706 | ||
707 | } | |
708 | ||
709 | ||
710 | /* Update _COMM_PAGE_SPIN_COUNT. We might want to reduce when running on a battery, etc. */ | |
711 | ||
712 | void | |
713 | commpage_set_spin_count( | |
714 | unsigned int count ) | |
715 | { | |
716 | char *cp; | |
717 | uint32_t *ip; | |
718 | ||
719 | if (count == 0) /* we test for 0 after decrement, not before */ | |
720 | count = 1; | |
721 | ||
722 | cp = commPagePtr32; | |
723 | if ( cp ) { | |
724 | cp += (_COMM_PAGE_SPIN_COUNT - _COMM_PAGE32_BASE_ADDRESS); | |
bd504ef0 | 725 | ip = (uint32_t*) (void *) cp; |
b0d623f7 A |
726 | *ip = (uint32_t) count; |
727 | } | |
728 | ||
729 | cp = commPagePtr64; | |
730 | if ( cp ) { | |
731 | cp += (_COMM_PAGE_SPIN_COUNT - _COMM_PAGE32_START_ADDRESS); | |
bd504ef0 | 732 | ip = (uint32_t*) (void *) cp; |
b0d623f7 A |
733 | *ip = (uint32_t) count; |
734 | } | |
735 | ||
736 | } | |
737 | ||
6d2010ae A |
738 | /* Updated every time a logical CPU goes offline/online */ |
739 | void | |
740 | commpage_update_active_cpus(void) | |
741 | { | |
742 | char *cp; | |
743 | volatile uint8_t *ip; | |
744 | ||
745 | /* At least 32-bit commpage must be initialized */ | |
746 | if (!commPagePtr32) | |
747 | return; | |
748 | ||
749 | simple_lock(&commpage_active_cpus_lock); | |
750 | ||
751 | cp = commPagePtr32; | |
752 | cp += (_COMM_PAGE_ACTIVE_CPUS - _COMM_PAGE32_BASE_ADDRESS); | |
753 | ip = (volatile uint8_t*) cp; | |
754 | *ip = (uint8_t) processor_avail_count; | |
755 | ||
756 | cp = commPagePtr64; | |
757 | if ( cp ) { | |
758 | cp += (_COMM_PAGE_ACTIVE_CPUS - _COMM_PAGE32_START_ADDRESS); | |
759 | ip = (volatile uint8_t*) cp; | |
760 | *ip = (uint8_t) processor_avail_count; | |
761 | } | |
762 | ||
763 | simple_unlock(&commpage_active_cpus_lock); | |
764 | } | |
765 | ||
a1c7dba1 | 766 | /* |
39037602 A |
767 | * Update the commpage with current kdebug state. This currently has bits for |
768 | * global trace state, and typefilter enablement. It is likely additional state | |
769 | * will be tracked in the future. | |
770 | * | |
771 | * INVARIANT: This value will always be 0 if global tracing is disabled. This | |
772 | * allows simple guard tests of "if (*_COMM_PAGE_KDEBUG_ENABLE) { ... }" | |
a1c7dba1 A |
773 | */ |
774 | void | |
39037602 | 775 | commpage_update_kdebug_state(void) |
a1c7dba1 A |
776 | { |
777 | volatile uint32_t *saved_data_ptr; | |
778 | char *cp; | |
779 | ||
780 | cp = commPagePtr32; | |
781 | if (cp) { | |
782 | cp += (_COMM_PAGE_KDEBUG_ENABLE - _COMM_PAGE32_BASE_ADDRESS); | |
783 | saved_data_ptr = (volatile uint32_t *)cp; | |
39037602 | 784 | *saved_data_ptr = kdebug_commpage_state(); |
a1c7dba1 A |
785 | } |
786 | ||
787 | cp = commPagePtr64; | |
39037602 | 788 | if (cp) { |
a1c7dba1 A |
789 | cp += (_COMM_PAGE_KDEBUG_ENABLE - _COMM_PAGE32_START_ADDRESS); |
790 | saved_data_ptr = (volatile uint32_t *)cp; | |
39037602 | 791 | *saved_data_ptr = kdebug_commpage_state(); |
a1c7dba1 A |
792 | } |
793 | } | |
794 | ||
3e170ce0 A |
795 | /* Ditto for atm_diagnostic_config */ |
796 | void | |
797 | commpage_update_atm_diagnostic_config(uint32_t diagnostic_config) | |
798 | { | |
799 | volatile uint32_t *saved_data_ptr; | |
800 | char *cp; | |
801 | ||
802 | cp = commPagePtr32; | |
803 | if (cp) { | |
804 | cp += (_COMM_PAGE_ATM_DIAGNOSTIC_CONFIG - _COMM_PAGE32_BASE_ADDRESS); | |
805 | saved_data_ptr = (volatile uint32_t *)cp; | |
806 | *saved_data_ptr = diagnostic_config; | |
807 | } | |
808 | ||
809 | cp = commPagePtr64; | |
810 | if ( cp ) { | |
811 | cp += (_COMM_PAGE_ATM_DIAGNOSTIC_CONFIG - _COMM_PAGE32_START_ADDRESS); | |
812 | saved_data_ptr = (volatile uint32_t *)cp; | |
813 | *saved_data_ptr = diagnostic_config; | |
814 | } | |
815 | } | |
a1c7dba1 | 816 | |
fe8ab488 A |
817 | /* |
818 | * update the commpage data for last known value of mach_absolute_time() | |
819 | */ | |
820 | ||
821 | void | |
822 | commpage_update_mach_approximate_time(uint64_t abstime) | |
823 | { | |
824 | #ifdef CONFIG_MACH_APPROXIMATE_TIME | |
825 | uint64_t saved_data; | |
a1c7dba1 A |
826 | char *cp; |
827 | ||
828 | cp = commPagePtr32; | |
fe8ab488 | 829 | if ( cp ) { |
a1c7dba1 | 830 | cp += (_COMM_PAGE_APPROX_TIME - _COMM_PAGE32_BASE_ADDRESS); |
5ba3f43e | 831 | saved_data = atomic_load_explicit((_Atomic uint64_t *)(uintptr_t)cp, memory_order_relaxed); |
fe8ab488 A |
832 | if (saved_data < abstime) { |
833 | /* ignoring the success/fail return value assuming that | |
834 | * if the value has been updated since we last read it, | |
835 | * "someone" has a newer timestamp than us and ours is | |
836 | * now invalid. */ | |
5ba3f43e A |
837 | atomic_compare_exchange_strong_explicit((_Atomic uint64_t *)(uintptr_t)cp, |
838 | &saved_data, abstime, memory_order_relaxed, memory_order_relaxed); | |
fe8ab488 A |
839 | } |
840 | } | |
a1c7dba1 | 841 | cp = commPagePtr64; |
fe8ab488 | 842 | if ( cp ) { |
a1c7dba1 | 843 | cp += (_COMM_PAGE_APPROX_TIME - _COMM_PAGE32_START_ADDRESS); |
5ba3f43e | 844 | saved_data = atomic_load_explicit((_Atomic uint64_t *)(uintptr_t)cp, memory_order_relaxed); |
fe8ab488 A |
845 | if (saved_data < abstime) { |
846 | /* ignoring the success/fail return value assuming that | |
847 | * if the value has been updated since we last read it, | |
848 | * "someone" has a newer timestamp than us and ours is | |
849 | * now invalid. */ | |
5ba3f43e A |
850 | atomic_compare_exchange_strong_explicit((_Atomic uint64_t *)(uintptr_t)cp, |
851 | &saved_data, abstime, memory_order_relaxed, memory_order_relaxed); | |
fe8ab488 A |
852 | } |
853 | } | |
854 | #else | |
855 | #pragma unused (abstime) | |
856 | #endif | |
857 | } | |
858 | ||
39037602 A |
859 | void |
860 | commpage_update_mach_continuous_time(uint64_t sleeptime) | |
861 | { | |
862 | char *cp; | |
863 | cp = commPagePtr32; | |
864 | if (cp) { | |
865 | cp += (_COMM_PAGE_CONT_TIMEBASE - _COMM_PAGE32_START_ADDRESS); | |
866 | *(uint64_t *)cp = sleeptime; | |
867 | } | |
868 | ||
869 | cp = commPagePtr64; | |
870 | if (cp) { | |
871 | cp += (_COMM_PAGE_CONT_TIMEBASE - _COMM_PAGE32_START_ADDRESS); | |
872 | *(uint64_t *)cp = sleeptime; | |
873 | } | |
874 | } | |
875 | ||
876 | void | |
877 | commpage_update_boottime(uint64_t boottime) | |
878 | { | |
879 | char *cp; | |
880 | cp = commPagePtr32; | |
881 | if (cp) { | |
882 | cp += (_COMM_PAGE_BOOTTIME_USEC - _COMM_PAGE32_START_ADDRESS); | |
883 | *(uint64_t *)cp = boottime; | |
884 | } | |
885 | ||
886 | cp = commPagePtr64; | |
887 | if (cp) { | |
888 | cp += (_COMM_PAGE_BOOTTIME_USEC - _COMM_PAGE32_START_ADDRESS); | |
889 | *(uint64_t *)cp = boottime; | |
890 | } | |
891 | } | |
892 | ||
fe8ab488 | 893 | |
316670eb A |
894 | extern user32_addr_t commpage_text32_location; |
895 | extern user64_addr_t commpage_text64_location; | |
b0d623f7 A |
896 | |
897 | /* Check to see if a given address is in the Preemption Free Zone (PFZ) */ | |
898 | ||
899 | uint32_t | |
900 | commpage_is_in_pfz32(uint32_t addr32) | |
901 | { | |
316670eb A |
902 | if ( (addr32 >= (commpage_text32_location + _COMM_TEXT_PFZ_START_OFFSET)) |
903 | && (addr32 < (commpage_text32_location+_COMM_TEXT_PFZ_END_OFFSET))) { | |
b0d623f7 A |
904 | return 1; |
905 | } | |
906 | else | |
907 | return 0; | |
908 | } | |
909 | ||
910 | uint32_t | |
911 | commpage_is_in_pfz64(addr64_t addr64) | |
912 | { | |
316670eb A |
913 | if ( (addr64 >= (commpage_text64_location + _COMM_TEXT_PFZ_START_OFFSET)) |
914 | && (addr64 < (commpage_text64_location + _COMM_TEXT_PFZ_END_OFFSET))) { | |
b0d623f7 A |
915 | return 1; |
916 | } | |
917 | else | |
918 | return 0; | |
919 | } | |
920 |