xref: /haiku/src/system/kernel/arch/x86/arch_cpu.cpp (revision 56ce1249b2b02f9222de845a52d60b0c8207519d)
1 /*
2  * Copyright 2018, Jérôme Duval, jerome.duval@gmail.com.
3  * Copyright 2002-2010, Axel Dörfler, axeld@pinc-software.de.
4  * Copyright 2013, Paweł Dziepak, pdziepak@quarnos.org.
5  * Copyright 2012, Alex Smith, alex@alex-smith.me.uk.
6  * Distributed under the terms of the MIT License.
7  *
8  * Copyright 2001-2002, Travis Geiselbrecht. All rights reserved.
9  * Distributed under the terms of the NewOS License.
10  */
11 
12 
13 #include <cpu.h>
14 
15 #include <string.h>
16 #include <stdlib.h>
17 #include <stdio.h>
18 
19 #include <algorithm>
20 
21 #include <ACPI.h>
22 
23 #include <boot_device.h>
24 #include <commpage.h>
25 #include <debug.h>
26 #include <elf.h>
27 #include <safemode.h>
28 #include <smp.h>
29 #include <util/BitUtils.h>
30 #include <vm/vm.h>
31 #include <vm/vm_types.h>
32 #include <vm/VMAddressSpace.h>
33 
34 #include <arch_system_info.h>
35 #include <arch/x86/apic.h>
36 #include <boot/kernel_args.h>
37 
38 #include "paging/X86PagingStructures.h"
39 #include "paging/X86VMTranslationMap.h"
40 
41 
42 #define DUMP_FEATURE_STRING	1
43 #define DUMP_CPU_TOPOLOGY	1
44 #define DUMP_CPU_PATCHLEVEL	1
45 
46 
47 /* cpu vendor info */
48 struct cpu_vendor_info {
49 	const char *vendor;
50 	const char *ident_string[2];
51 };
52 
53 static const struct cpu_vendor_info vendor_info[VENDOR_NUM] = {
54 	{ "Intel", { "GenuineIntel" } },
55 	{ "AMD", { "AuthenticAMD" } },
56 	{ "Cyrix", { "CyrixInstead" } },
57 	{ "UMC", { "UMC UMC UMC" } },
58 	{ "NexGen", { "NexGenDriven" } },
59 	{ "Centaur", { "CentaurHauls" } },
60 	{ "Rise", { "RiseRiseRise" } },
61 	{ "Transmeta", { "GenuineTMx86", "TransmetaCPU" } },
62 	{ "NSC", { "Geode by NSC" } },
63 	{ "Hygon", { "HygonGenuine" } },
64 };
65 
66 #define K8_SMIONCMPHALT			(1ULL << 27)
67 #define K8_C1EONCMPHALT			(1ULL << 28)
68 
69 #define K8_CMPHALT				(K8_SMIONCMPHALT | K8_C1EONCMPHALT)
70 
71 struct set_mtrr_parameter {
72 	int32	index;
73 	uint64	base;
74 	uint64	length;
75 	uint8	type;
76 };
77 
78 struct set_mtrrs_parameter {
79 	const x86_mtrr_info*	infos;
80 	uint32					count;
81 	uint8					defaultType;
82 };
83 
84 
85 #ifdef __x86_64__
86 extern addr_t _stac;
87 extern addr_t _clac;
88 extern addr_t _xsave;
89 extern addr_t _xsavec;
90 extern addr_t _xrstor;
91 uint64 gXsaveMask;
92 uint64 gFPUSaveLength = 512;
93 bool gHasXsave = false;
94 bool gHasXsavec = false;
95 #endif
96 
97 extern "C" void x86_reboot(void);
98 	// from arch.S
99 
100 void (*gCpuIdleFunc)(void);
101 #ifndef __x86_64__
102 void (*gX86SwapFPUFunc)(void* oldState, const void* newState) = x86_noop_swap;
103 bool gHasSSE = false;
104 #endif
105 
106 static uint32 sCpuRendezvous;
107 static uint32 sCpuRendezvous2;
108 static uint32 sCpuRendezvous3;
109 static vint32 sTSCSyncRendezvous;
110 
111 /* Some specials for the double fault handler */
112 static uint8* sDoubleFaultStacks;
113 static const size_t kDoubleFaultStackSize = 4096;	// size per CPU
114 
115 static x86_cpu_module_info* sCpuModule;
116 
117 
118 /* CPU topology information */
119 static uint32 (*sGetCPUTopologyID)(int currentCPU);
120 static uint32 sHierarchyMask[CPU_TOPOLOGY_LEVELS];
121 static uint32 sHierarchyShift[CPU_TOPOLOGY_LEVELS];
122 
123 /* Cache topology information */
124 static uint32 sCacheSharingMask[CPU_MAX_CACHE_LEVEL];
125 
126 static void* sUcodeData = NULL;
127 static size_t sUcodeDataSize = 0;
128 static struct intel_microcode_header* sLoadedUcodeUpdate;
129 static spinlock sUcodeUpdateLock = B_SPINLOCK_INITIALIZER;
130 
131 
132 static status_t
133 acpi_shutdown(bool rebootSystem)
134 {
135 	if (debug_debugger_running() || !are_interrupts_enabled())
136 		return B_ERROR;
137 
138 	acpi_module_info* acpi;
139 	if (get_module(B_ACPI_MODULE_NAME, (module_info**)&acpi) != B_OK)
140 		return B_NOT_SUPPORTED;
141 
142 	status_t status;
143 	if (rebootSystem) {
144 		status = acpi->reboot();
145 	} else {
146 		status = acpi->prepare_sleep_state(ACPI_POWER_STATE_OFF, NULL, 0);
147 		if (status == B_OK) {
148 			//cpu_status state = disable_interrupts();
149 			status = acpi->enter_sleep_state(ACPI_POWER_STATE_OFF);
150 			//restore_interrupts(state);
151 		}
152 	}
153 
154 	put_module(B_ACPI_MODULE_NAME);
155 	return status;
156 }
157 
158 
159 /*!	Disable CPU caches, and invalidate them. */
160 static void
161 disable_caches()
162 {
163 	x86_write_cr0((x86_read_cr0() | CR0_CACHE_DISABLE)
164 		& ~CR0_NOT_WRITE_THROUGH);
165 	wbinvd();
166 	arch_cpu_global_TLB_invalidate();
167 }
168 
169 
170 /*!	Invalidate CPU caches, and enable them. */
171 static void
172 enable_caches()
173 {
174 	wbinvd();
175 	arch_cpu_global_TLB_invalidate();
176 	x86_write_cr0(x86_read_cr0()
177 		& ~(CR0_CACHE_DISABLE | CR0_NOT_WRITE_THROUGH));
178 }
179 
180 
181 static void
182 set_mtrr(void* _parameter, int cpu)
183 {
184 	struct set_mtrr_parameter* parameter
185 		= (struct set_mtrr_parameter*)_parameter;
186 
187 	// wait until all CPUs have arrived here
188 	smp_cpu_rendezvous(&sCpuRendezvous);
189 
190 	// One CPU has to reset sCpuRendezvous3 -- it is needed to prevent the CPU
191 	// that initiated the call_all_cpus() from doing that again and clearing
192 	// sCpuRendezvous2 before the last CPU has actually left the loop in
193 	// smp_cpu_rendezvous();
194 	if (cpu == 0)
195 		atomic_set((int32*)&sCpuRendezvous3, 0);
196 
197 	disable_caches();
198 
199 	sCpuModule->set_mtrr(parameter->index, parameter->base, parameter->length,
200 		parameter->type);
201 
202 	enable_caches();
203 
204 	// wait until all CPUs have arrived here
205 	smp_cpu_rendezvous(&sCpuRendezvous2);
206 	smp_cpu_rendezvous(&sCpuRendezvous3);
207 }
208 
209 
210 static void
211 set_mtrrs(void* _parameter, int cpu)
212 {
213 	set_mtrrs_parameter* parameter = (set_mtrrs_parameter*)_parameter;
214 
215 	// wait until all CPUs have arrived here
216 	smp_cpu_rendezvous(&sCpuRendezvous);
217 
218 	// One CPU has to reset sCpuRendezvous3 -- it is needed to prevent the CPU
219 	// that initiated the call_all_cpus() from doing that again and clearing
220 	// sCpuRendezvous2 before the last CPU has actually left the loop in
221 	// smp_cpu_rendezvous();
222 	if (cpu == 0)
223 		atomic_set((int32*)&sCpuRendezvous3, 0);
224 
225 	disable_caches();
226 
227 	sCpuModule->set_mtrrs(parameter->defaultType, parameter->infos,
228 		parameter->count);
229 
230 	enable_caches();
231 
232 	// wait until all CPUs have arrived here
233 	smp_cpu_rendezvous(&sCpuRendezvous2);
234 	smp_cpu_rendezvous(&sCpuRendezvous3);
235 }
236 
237 
238 static void
239 init_mtrrs(void* _unused, int cpu)
240 {
241 	// wait until all CPUs have arrived here
242 	smp_cpu_rendezvous(&sCpuRendezvous);
243 
244 	// One CPU has to reset sCpuRendezvous3 -- it is needed to prevent the CPU
245 	// that initiated the call_all_cpus() from doing that again and clearing
246 	// sCpuRendezvous2 before the last CPU has actually left the loop in
247 	// smp_cpu_rendezvous();
248 	if (cpu == 0)
249 		atomic_set((int32*)&sCpuRendezvous3, 0);
250 
251 	disable_caches();
252 
253 	sCpuModule->init_mtrrs();
254 
255 	enable_caches();
256 
257 	// wait until all CPUs have arrived here
258 	smp_cpu_rendezvous(&sCpuRendezvous2);
259 	smp_cpu_rendezvous(&sCpuRendezvous3);
260 }
261 
262 
263 uint32
264 x86_count_mtrrs(void)
265 {
266 	if (sCpuModule == NULL)
267 		return 0;
268 
269 	return sCpuModule->count_mtrrs();
270 }
271 
272 
273 void
274 x86_set_mtrr(uint32 index, uint64 base, uint64 length, uint8 type)
275 {
276 	struct set_mtrr_parameter parameter;
277 	parameter.index = index;
278 	parameter.base = base;
279 	parameter.length = length;
280 	parameter.type = type;
281 
282 	sCpuRendezvous = sCpuRendezvous2 = 0;
283 	call_all_cpus(&set_mtrr, &parameter);
284 }
285 
286 
287 status_t
288 x86_get_mtrr(uint32 index, uint64* _base, uint64* _length, uint8* _type)
289 {
290 	// the MTRRs are identical on all CPUs, so it doesn't matter
291 	// on which CPU this runs
292 	return sCpuModule->get_mtrr(index, _base, _length, _type);
293 }
294 
295 
296 void
297 x86_set_mtrrs(uint8 defaultType, const x86_mtrr_info* infos, uint32 count)
298 {
299 	if (sCpuModule == NULL)
300 		return;
301 
302 	struct set_mtrrs_parameter parameter;
303 	parameter.defaultType = defaultType;
304 	parameter.infos = infos;
305 	parameter.count = count;
306 
307 	sCpuRendezvous = sCpuRendezvous2 = 0;
308 	call_all_cpus(&set_mtrrs, &parameter);
309 }
310 
311 
312 void
313 x86_init_fpu(void)
314 {
315 	// All x86_64 CPUs support SSE, don't need to bother checking for it.
316 #ifndef __x86_64__
317 	if (!x86_check_feature(IA32_FEATURE_FPU, FEATURE_COMMON)) {
318 		// No FPU... time to install one in your 386?
319 		dprintf("%s: Warning: CPU has no reported FPU.\n", __func__);
320 		gX86SwapFPUFunc = x86_noop_swap;
321 		return;
322 	}
323 
324 	if (!x86_check_feature(IA32_FEATURE_SSE, FEATURE_COMMON)
325 		|| !x86_check_feature(IA32_FEATURE_FXSR, FEATURE_COMMON)) {
326 		dprintf("%s: CPU has no SSE... just enabling FPU.\n", __func__);
327 		// we don't have proper SSE support, just enable FPU
328 		x86_write_cr0(x86_read_cr0() & ~(CR0_FPU_EMULATION | CR0_MONITOR_FPU));
329 		gX86SwapFPUFunc = x86_fnsave_swap;
330 		return;
331 	}
332 #endif
333 
334 	dprintf("%s: CPU has SSE... enabling FXSR and XMM.\n", __func__);
335 #ifndef __x86_64__
336 	// enable OS support for SSE
337 	x86_write_cr4(x86_read_cr4() | CR4_OS_FXSR | CR4_OS_XMM_EXCEPTION);
338 	x86_write_cr0(x86_read_cr0() & ~(CR0_FPU_EMULATION | CR0_MONITOR_FPU));
339 
340 	gX86SwapFPUFunc = x86_fxsave_swap;
341 	gHasSSE = true;
342 #endif
343 }
344 
345 
346 #if DUMP_FEATURE_STRING
347 static void
348 dump_feature_string(int currentCPU, cpu_ent* cpu)
349 {
350 	char features[512];
351 	features[0] = 0;
352 
353 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_FPU)
354 		strlcat(features, "fpu ", sizeof(features));
355 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_VME)
356 		strlcat(features, "vme ", sizeof(features));
357 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_DE)
358 		strlcat(features, "de ", sizeof(features));
359 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_PSE)
360 		strlcat(features, "pse ", sizeof(features));
361 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_TSC)
362 		strlcat(features, "tsc ", sizeof(features));
363 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_MSR)
364 		strlcat(features, "msr ", sizeof(features));
365 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_PAE)
366 		strlcat(features, "pae ", sizeof(features));
367 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_MCE)
368 		strlcat(features, "mce ", sizeof(features));
369 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_CX8)
370 		strlcat(features, "cx8 ", sizeof(features));
371 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_APIC)
372 		strlcat(features, "apic ", sizeof(features));
373 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_SEP)
374 		strlcat(features, "sep ", sizeof(features));
375 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_MTRR)
376 		strlcat(features, "mtrr ", sizeof(features));
377 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_PGE)
378 		strlcat(features, "pge ", sizeof(features));
379 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_MCA)
380 		strlcat(features, "mca ", sizeof(features));
381 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_CMOV)
382 		strlcat(features, "cmov ", sizeof(features));
383 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_PAT)
384 		strlcat(features, "pat ", sizeof(features));
385 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_PSE36)
386 		strlcat(features, "pse36 ", sizeof(features));
387 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_PSN)
388 		strlcat(features, "psn ", sizeof(features));
389 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_CLFSH)
390 		strlcat(features, "clfsh ", sizeof(features));
391 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_DS)
392 		strlcat(features, "ds ", sizeof(features));
393 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_ACPI)
394 		strlcat(features, "acpi ", sizeof(features));
395 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_MMX)
396 		strlcat(features, "mmx ", sizeof(features));
397 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_FXSR)
398 		strlcat(features, "fxsr ", sizeof(features));
399 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_SSE)
400 		strlcat(features, "sse ", sizeof(features));
401 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_SSE2)
402 		strlcat(features, "sse2 ", sizeof(features));
403 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_SS)
404 		strlcat(features, "ss ", sizeof(features));
405 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_HTT)
406 		strlcat(features, "htt ", sizeof(features));
407 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_TM)
408 		strlcat(features, "tm ", sizeof(features));
409 	if (cpu->arch.feature[FEATURE_COMMON] & IA32_FEATURE_PBE)
410 		strlcat(features, "pbe ", sizeof(features));
411 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_SSE3)
412 		strlcat(features, "sse3 ", sizeof(features));
413 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_PCLMULQDQ)
414 		strlcat(features, "pclmulqdq ", sizeof(features));
415 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_DTES64)
416 		strlcat(features, "dtes64 ", sizeof(features));
417 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_MONITOR)
418 		strlcat(features, "monitor ", sizeof(features));
419 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_DSCPL)
420 		strlcat(features, "dscpl ", sizeof(features));
421 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_VMX)
422 		strlcat(features, "vmx ", sizeof(features));
423 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_SMX)
424 		strlcat(features, "smx ", sizeof(features));
425 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_EST)
426 		strlcat(features, "est ", sizeof(features));
427 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_TM2)
428 		strlcat(features, "tm2 ", sizeof(features));
429 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_SSSE3)
430 		strlcat(features, "ssse3 ", sizeof(features));
431 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_CNXTID)
432 		strlcat(features, "cnxtid ", sizeof(features));
433 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_FMA)
434 		strlcat(features, "fma ", sizeof(features));
435 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_CX16)
436 		strlcat(features, "cx16 ", sizeof(features));
437 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_XTPR)
438 		strlcat(features, "xtpr ", sizeof(features));
439 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_PDCM)
440 		strlcat(features, "pdcm ", sizeof(features));
441 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_PCID)
442 		strlcat(features, "pcid ", sizeof(features));
443 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_DCA)
444 		strlcat(features, "dca ", sizeof(features));
445 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_SSE4_1)
446 		strlcat(features, "sse4_1 ", sizeof(features));
447 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_SSE4_2)
448 		strlcat(features, "sse4_2 ", sizeof(features));
449 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_X2APIC)
450 		strlcat(features, "x2apic ", sizeof(features));
451 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_MOVBE)
452 		strlcat(features, "movbe ", sizeof(features));
453 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_POPCNT)
454 		strlcat(features, "popcnt ", sizeof(features));
455 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_TSCDEADLINE)
456 		strlcat(features, "tscdeadline ", sizeof(features));
457 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_AES)
458 		strlcat(features, "aes ", sizeof(features));
459 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_XSAVE)
460 		strlcat(features, "xsave ", sizeof(features));
461 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_OSXSAVE)
462 		strlcat(features, "osxsave ", sizeof(features));
463 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_AVX)
464 		strlcat(features, "avx ", sizeof(features));
465 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_F16C)
466 		strlcat(features, "f16c ", sizeof(features));
467 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_RDRND)
468 		strlcat(features, "rdrnd ", sizeof(features));
469 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_HYPERVISOR)
470 		strlcat(features, "hypervisor ", sizeof(features));
471 	if (cpu->arch.feature[FEATURE_EXT_AMD] & IA32_FEATURE_AMD_EXT_SYSCALL)
472 		strlcat(features, "syscall ", sizeof(features));
473 	if (cpu->arch.feature[FEATURE_EXT_AMD] & IA32_FEATURE_AMD_EXT_NX)
474 		strlcat(features, "nx ", sizeof(features));
475 	if (cpu->arch.feature[FEATURE_EXT_AMD] & IA32_FEATURE_AMD_EXT_MMXEXT)
476 		strlcat(features, "mmxext ", sizeof(features));
477 	if (cpu->arch.feature[FEATURE_EXT_AMD] & IA32_FEATURE_AMD_EXT_FFXSR)
478 		strlcat(features, "ffxsr ", sizeof(features));
479 	if (cpu->arch.feature[FEATURE_EXT_AMD] & IA32_FEATURE_AMD_EXT_PDPE1GB)
480 		strlcat(features, "pdpe1gb ", sizeof(features));
481 	if (cpu->arch.feature[FEATURE_EXT_AMD] & IA32_FEATURE_AMD_EXT_LONG)
482 		strlcat(features, "long ", sizeof(features));
483 	if (cpu->arch.feature[FEATURE_EXT_AMD] & IA32_FEATURE_AMD_EXT_3DNOWEXT)
484 		strlcat(features, "3dnowext ", sizeof(features));
485 	if (cpu->arch.feature[FEATURE_EXT_AMD] & IA32_FEATURE_AMD_EXT_3DNOW)
486 		strlcat(features, "3dnow ", sizeof(features));
487 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_DTS)
488 		strlcat(features, "dts ", sizeof(features));
489 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_ITB)
490 		strlcat(features, "itb ", sizeof(features));
491 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_ARAT)
492 		strlcat(features, "arat ", sizeof(features));
493 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_PLN)
494 		strlcat(features, "pln ", sizeof(features));
495 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_ECMD)
496 		strlcat(features, "ecmd ", sizeof(features));
497 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_PTM)
498 		strlcat(features, "ptm ", sizeof(features));
499 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP)
500 		strlcat(features, "hwp ", sizeof(features));
501 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_NOTIFY)
502 		strlcat(features, "hwp_notify ", sizeof(features));
503 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_ACTWIN)
504 		strlcat(features, "hwp_actwin ", sizeof(features));
505 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_EPP)
506 		strlcat(features, "hwp_epp ", sizeof(features));
507 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_PLR)
508 		strlcat(features, "hwp_plr ", sizeof(features));
509 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HDC)
510 		strlcat(features, "hdc ", sizeof(features));
511 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_TBMT3)
512 		strlcat(features, "tbmt3 ", sizeof(features));
513 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_CAP)
514 		strlcat(features, "hwp_cap ", sizeof(features));
515 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_PECI)
516 		strlcat(features, "hwp_peci ", sizeof(features));
517 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_FLEX)
518 		strlcat(features, "hwp_flex ", sizeof(features));
519 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_FAST)
520 		strlcat(features, "hwp_fast ", sizeof(features));
521 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HW_FEEDBACK)
522 		strlcat(features, "hw_feedback ", sizeof(features));
523 	if (cpu->arch.feature[FEATURE_6_EAX] & IA32_FEATURE_HWP_IGNIDL)
524 		strlcat(features, "hwp_ignidl ", sizeof(features));
525 	if (cpu->arch.feature[FEATURE_6_ECX] & IA32_FEATURE_APERFMPERF)
526 		strlcat(features, "aperfmperf ", sizeof(features));
527 	if (cpu->arch.feature[FEATURE_6_ECX] & IA32_FEATURE_EPB)
528 		strlcat(features, "epb ", sizeof(features));
529 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_TSC_ADJUST)
530 		strlcat(features, "tsc_adjust ", sizeof(features));
531 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_SGX)
532 		strlcat(features, "sgx ", sizeof(features));
533 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_BMI1)
534 		strlcat(features, "bmi1 ", sizeof(features));
535 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_HLE)
536 		strlcat(features, "hle ", sizeof(features));
537 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX2)
538 		strlcat(features, "avx2 ", sizeof(features));
539 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_SMEP)
540 		strlcat(features, "smep ", sizeof(features));
541 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_BMI2)
542 		strlcat(features, "bmi2 ", sizeof(features));
543 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_ERMS)
544 		strlcat(features, "erms ", sizeof(features));
545 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_INVPCID)
546 		strlcat(features, "invpcid ", sizeof(features));
547 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_RTM)
548 		strlcat(features, "rtm ", sizeof(features));
549 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_CQM)
550 		strlcat(features, "cqm ", sizeof(features));
551 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_MPX)
552 		strlcat(features, "mpx ", sizeof(features));
553 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_RDT_A)
554 		strlcat(features, "rdt_a ", sizeof(features));
555 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX512F)
556 		strlcat(features, "avx512f ", sizeof(features));
557 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX512DQ)
558 		strlcat(features, "avx512dq ", sizeof(features));
559 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_RDSEED)
560 		strlcat(features, "rdseed ", sizeof(features));
561 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_ADX)
562 		strlcat(features, "adx ", sizeof(features));
563 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_SMAP)
564 		strlcat(features, "smap ", sizeof(features));
565 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX512IFMA)
566 		strlcat(features, "avx512ifma ", sizeof(features));
567 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_PCOMMIT)
568 		strlcat(features, "pcommit ", sizeof(features));
569 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_CLFLUSHOPT)
570 		strlcat(features, "cflushopt ", sizeof(features));
571 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_CLWB)
572 		strlcat(features, "clwb ", sizeof(features));
573 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_INTEL_PT)
574 		strlcat(features, "intel_pt ", sizeof(features));
575 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX512PF)
576 		strlcat(features, "avx512pf ", sizeof(features));
577 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX512ER)
578 		strlcat(features, "avx512er ", sizeof(features));
579 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX512CD)
580 		strlcat(features, "avx512cd ", sizeof(features));
581 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_SHA_NI)
582 		strlcat(features, "sha_ni ", sizeof(features));
583 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX512BW)
584 		strlcat(features, "avx512bw ", sizeof(features));
585 	if (cpu->arch.feature[FEATURE_7_EBX] & IA32_FEATURE_AVX512VI)
586 		strlcat(features, "avx512vi ", sizeof(features));
587 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_IBRS)
588 		strlcat(features, "ibrs ", sizeof(features));
589 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_STIBP)
590 		strlcat(features, "stibp ", sizeof(features));
591 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_L1D_FLUSH)
592 		strlcat(features, "l1d_flush ", sizeof(features));
593 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_ARCH_CAPABILITIES)
594 		strlcat(features, "msr_arch ", sizeof(features));
595 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_SSBD)
596 		strlcat(features, "ssbd ", sizeof(features));
597 	if (cpu->arch.feature[FEATURE_D_1_EAX] & IA32_FEATURE_XSAVEOPT)
598 		strlcat(features, "xsaveopt ", sizeof(features));
599 	if (cpu->arch.feature[FEATURE_D_1_EAX] & IA32_FEATURE_XSAVEC)
600 		strlcat(features, "xsavec ", sizeof(features));
601 	if (cpu->arch.feature[FEATURE_D_1_EAX] & IA32_FEATURE_XGETBV1)
602 		strlcat(features, "xgetbv1 ", sizeof(features));
603 	if (cpu->arch.feature[FEATURE_D_1_EAX] & IA32_FEATURE_XSAVES)
604 		strlcat(features, "xsaves ", sizeof(features));
605 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_CLZERO)
606 		strlcat(features, "clzero ", sizeof(features));
607 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_IBPB)
608 		strlcat(features, "ibpb ", sizeof(features));
609 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_AMD_SSBD)
610 		strlcat(features, "amd_ssbd ", sizeof(features));
611 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_VIRT_SSBD)
612 		strlcat(features, "virt_ssbd ", sizeof(features));
613 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_AMD_SSB_NO)
614 		strlcat(features, "amd_ssb_no ", sizeof(features));
615 	dprintf("CPU %d: features: %s\n", currentCPU, features);
616 }
617 #endif	// DUMP_FEATURE_STRING
618 
619 
620 static void
621 compute_cpu_hierarchy_masks(int maxLogicalID, int maxCoreID)
622 {
623 	ASSERT(maxLogicalID >= maxCoreID);
624 	const int kMaxSMTID = maxLogicalID / maxCoreID;
625 
626 	sHierarchyMask[CPU_TOPOLOGY_SMT] = kMaxSMTID - 1;
627 	sHierarchyShift[CPU_TOPOLOGY_SMT] = 0;
628 
629 	sHierarchyMask[CPU_TOPOLOGY_CORE] = (maxCoreID - 1) * kMaxSMTID;
630 	sHierarchyShift[CPU_TOPOLOGY_CORE]
631 		= count_set_bits(sHierarchyMask[CPU_TOPOLOGY_SMT]);
632 
633 	const uint32 kSinglePackageMask = sHierarchyMask[CPU_TOPOLOGY_SMT]
634 		| sHierarchyMask[CPU_TOPOLOGY_CORE];
635 	sHierarchyMask[CPU_TOPOLOGY_PACKAGE] = ~kSinglePackageMask;
636 	sHierarchyShift[CPU_TOPOLOGY_PACKAGE] = count_set_bits(kSinglePackageMask);
637 }
638 
639 
640 static uint32
641 get_cpu_legacy_initial_apic_id(int /* currentCPU */)
642 {
643 	cpuid_info cpuid;
644 	get_current_cpuid(&cpuid, 1, 0);
645 	return cpuid.regs.ebx >> 24;
646 }
647 
648 
649 static inline status_t
650 detect_amd_cpu_topology(uint32 maxBasicLeaf, uint32 maxExtendedLeaf)
651 {
652 	sGetCPUTopologyID = get_cpu_legacy_initial_apic_id;
653 
654 	cpuid_info cpuid;
655 	get_current_cpuid(&cpuid, 1, 0);
656 	int maxLogicalID = next_power_of_2((cpuid.regs.ebx >> 16) & 0xff);
657 
658 	int maxCoreID = 1;
659 	if (maxExtendedLeaf >= 0x80000008) {
660 		get_current_cpuid(&cpuid, 0x80000008, 0);
661 		maxCoreID = (cpuid.regs.ecx >> 12) & 0xf;
662 		if (maxCoreID != 0)
663 			maxCoreID = 1 << maxCoreID;
664 		else
665 			maxCoreID = next_power_of_2((cpuid.regs.edx & 0xf) + 1);
666 	}
667 
668 	if (maxExtendedLeaf >= 0x80000001) {
669 		get_current_cpuid(&cpuid, 0x80000001, 0);
670 		if (x86_check_feature(IA32_FEATURE_AMD_EXT_CMPLEGACY,
671 				FEATURE_EXT_AMD_ECX))
672 			maxCoreID = maxLogicalID;
673 	}
674 
675 	compute_cpu_hierarchy_masks(maxLogicalID, maxCoreID);
676 
677 	return B_OK;
678 }
679 
680 
681 static void
682 detect_amd_cache_topology(uint32 maxExtendedLeaf)
683 {
684 	if (!x86_check_feature(IA32_FEATURE_AMD_EXT_TOPOLOGY, FEATURE_EXT_AMD_ECX))
685 		return;
686 
687 	if (maxExtendedLeaf < 0x8000001d)
688 		return;
689 
690 	uint8 hierarchyLevels[CPU_MAX_CACHE_LEVEL];
691 	int maxCacheLevel = 0;
692 
693 	int currentLevel = 0;
694 	int cacheType;
695 	do {
696 		cpuid_info cpuid;
697 		get_current_cpuid(&cpuid, 0x8000001d, currentLevel);
698 
699 		cacheType = cpuid.regs.eax & 0x1f;
700 		if (cacheType == 0)
701 			break;
702 
703 		int cacheLevel = (cpuid.regs.eax >> 5) & 0x7;
704 		int coresCount = next_power_of_2(((cpuid.regs.eax >> 14) & 0x3f) + 1);
705 		hierarchyLevels[cacheLevel - 1]
706 			= coresCount * (sHierarchyMask[CPU_TOPOLOGY_SMT] + 1);
707 		maxCacheLevel = std::max(maxCacheLevel, cacheLevel);
708 
709 		currentLevel++;
710 	} while (true);
711 
712 	for (int i = 0; i < maxCacheLevel; i++)
713 		sCacheSharingMask[i] = ~uint32(hierarchyLevels[i] - 1);
714 	gCPUCacheLevelCount = maxCacheLevel;
715 }
716 
717 
718 static uint32
719 get_intel_cpu_initial_x2apic_id(int /* currentCPU */)
720 {
721 	cpuid_info cpuid;
722 	get_current_cpuid(&cpuid, 11, 0);
723 	return cpuid.regs.edx;
724 }
725 
726 
727 static inline status_t
728 detect_intel_cpu_topology_x2apic(uint32 maxBasicLeaf)
729 {
730 	if (maxBasicLeaf < 11)
731 		return B_UNSUPPORTED;
732 
733 	uint8 hierarchyLevels[CPU_TOPOLOGY_LEVELS] = { 0 };
734 
735 	int currentLevel = 0;
736 	int levelType;
737 	unsigned int levelsSet = 0;
738 
739 	do {
740 		cpuid_info cpuid;
741 		get_current_cpuid(&cpuid, 11, currentLevel);
742 		if (currentLevel == 0 && cpuid.regs.ebx == 0)
743 			return B_UNSUPPORTED;
744 
745 		levelType = (cpuid.regs.ecx >> 8) & 0xff;
746 		int levelValue = cpuid.regs.eax & 0x1f;
747 
748 		switch (levelType) {
749 			case 1:	// SMT
750 				hierarchyLevels[CPU_TOPOLOGY_SMT] = levelValue;
751 				levelsSet |= 1;
752 				break;
753 			case 2:	// core
754 				hierarchyLevels[CPU_TOPOLOGY_CORE] = levelValue;
755 				levelsSet |= 2;
756 				break;
757 		}
758 
759 		currentLevel++;
760 	} while (levelType != 0 && levelsSet != 3);
761 
762 	sGetCPUTopologyID = get_intel_cpu_initial_x2apic_id;
763 
764 	for (int i = 1; i < CPU_TOPOLOGY_LEVELS; i++) {
765 		if ((levelsSet & (1u << i)) != 0)
766 			continue;
767 		hierarchyLevels[i] = hierarchyLevels[i - 1];
768 	}
769 
770 	for (int i = 0; i < CPU_TOPOLOGY_LEVELS; i++) {
771 		uint32 mask = ~uint32(0);
772 		if (i < CPU_TOPOLOGY_LEVELS - 1)
773 			mask = (1u << hierarchyLevels[i]) - 1;
774 		if (i > 0)
775 			mask &= ~sHierarchyMask[i - 1];
776 		sHierarchyMask[i] = mask;
777 		sHierarchyShift[i] = i > 0 ? hierarchyLevels[i - 1] : 0;
778 	}
779 
780 	return B_OK;
781 }
782 
783 
784 static inline status_t
785 detect_intel_cpu_topology_legacy(uint32 maxBasicLeaf)
786 {
787 	sGetCPUTopologyID = get_cpu_legacy_initial_apic_id;
788 
789 	cpuid_info cpuid;
790 
791 	get_current_cpuid(&cpuid, 1, 0);
792 	int maxLogicalID = next_power_of_2((cpuid.regs.ebx >> 16) & 0xff);
793 
794 	int maxCoreID = 1;
795 	if (maxBasicLeaf >= 4) {
796 		get_current_cpuid(&cpuid, 4, 0);
797 		maxCoreID = next_power_of_2((cpuid.regs.eax >> 26) + 1);
798 	}
799 
800 	compute_cpu_hierarchy_masks(maxLogicalID, maxCoreID);
801 
802 	return B_OK;
803 }
804 
805 
806 static void
807 detect_intel_cache_topology(uint32 maxBasicLeaf)
808 {
809 	if (maxBasicLeaf < 4)
810 		return;
811 
812 	uint8 hierarchyLevels[CPU_MAX_CACHE_LEVEL];
813 	int maxCacheLevel = 0;
814 
815 	int currentLevel = 0;
816 	int cacheType;
817 	do {
818 		cpuid_info cpuid;
819 		get_current_cpuid(&cpuid, 4, currentLevel);
820 
821 		cacheType = cpuid.regs.eax & 0x1f;
822 		if (cacheType == 0)
823 			break;
824 
825 		int cacheLevel = (cpuid.regs.eax >> 5) & 0x7;
826 		hierarchyLevels[cacheLevel - 1]
827 			= next_power_of_2(((cpuid.regs.eax >> 14) & 0x3f) + 1);
828 		maxCacheLevel = std::max(maxCacheLevel, cacheLevel);
829 
830 		currentLevel++;
831 	} while (true);
832 
833 	for (int i = 0; i < maxCacheLevel; i++)
834 		sCacheSharingMask[i] = ~uint32(hierarchyLevels[i] - 1);
835 
836 	gCPUCacheLevelCount = maxCacheLevel;
837 }
838 
839 
840 static uint32
841 get_simple_cpu_topology_id(int currentCPU)
842 {
843 	return currentCPU;
844 }
845 
846 
847 static inline int
848 get_topology_level_id(uint32 id, cpu_topology_level level)
849 {
850 	ASSERT(level < CPU_TOPOLOGY_LEVELS);
851 	return (id & sHierarchyMask[level]) >> sHierarchyShift[level];
852 }
853 
854 
855 static void
856 detect_cpu_topology(int currentCPU, cpu_ent* cpu, uint32 maxBasicLeaf,
857 	uint32 maxExtendedLeaf)
858 {
859 	if (currentCPU == 0) {
860 		memset(sCacheSharingMask, 0xff, sizeof(sCacheSharingMask));
861 
862 		status_t result = B_UNSUPPORTED;
863 		if (x86_check_feature(IA32_FEATURE_HTT, FEATURE_COMMON)) {
864 			if (cpu->arch.vendor == VENDOR_AMD
865 				|| cpu->arch.vendor == VENDOR_HYGON) {
866 				result = detect_amd_cpu_topology(maxBasicLeaf, maxExtendedLeaf);
867 
868 				if (result == B_OK)
869 					detect_amd_cache_topology(maxExtendedLeaf);
870 			}
871 
872 			if (cpu->arch.vendor == VENDOR_INTEL) {
873 				result = detect_intel_cpu_topology_x2apic(maxBasicLeaf);
874 				if (result != B_OK)
875 					result = detect_intel_cpu_topology_legacy(maxBasicLeaf);
876 
877 				if (result == B_OK)
878 					detect_intel_cache_topology(maxBasicLeaf);
879 			}
880 		}
881 
882 		if (result != B_OK) {
883 			dprintf("No CPU topology information available.\n");
884 
885 			sGetCPUTopologyID = get_simple_cpu_topology_id;
886 
887 			sHierarchyMask[CPU_TOPOLOGY_PACKAGE] = ~uint32(0);
888 		}
889 	}
890 
891 	ASSERT(sGetCPUTopologyID != NULL);
892 	int topologyID = sGetCPUTopologyID(currentCPU);
893 	cpu->topology_id[CPU_TOPOLOGY_SMT]
894 		= get_topology_level_id(topologyID, CPU_TOPOLOGY_SMT);
895 	cpu->topology_id[CPU_TOPOLOGY_CORE]
896 		= get_topology_level_id(topologyID, CPU_TOPOLOGY_CORE);
897 	cpu->topology_id[CPU_TOPOLOGY_PACKAGE]
898 		= get_topology_level_id(topologyID, CPU_TOPOLOGY_PACKAGE);
899 
900 	unsigned int i;
901 	for (i = 0; i < gCPUCacheLevelCount; i++)
902 		cpu->cache_id[i] = topologyID & sCacheSharingMask[i];
903 	for (; i < CPU_MAX_CACHE_LEVEL; i++)
904 		cpu->cache_id[i] = -1;
905 
906 #if DUMP_CPU_TOPOLOGY
907 	dprintf("CPU %d: apic id %d, package %d, core %d, smt %d\n", currentCPU,
908 		topologyID, cpu->topology_id[CPU_TOPOLOGY_PACKAGE],
909 		cpu->topology_id[CPU_TOPOLOGY_CORE],
910 		cpu->topology_id[CPU_TOPOLOGY_SMT]);
911 
912 	if (gCPUCacheLevelCount > 0) {
913 		char cacheLevels[256];
914 		unsigned int offset = 0;
915 		for (i = 0; i < gCPUCacheLevelCount; i++) {
916 			offset += snprintf(cacheLevels + offset,
917 					sizeof(cacheLevels) - offset,
918 					" L%d id %d%s", i + 1, cpu->cache_id[i],
919 					i < gCPUCacheLevelCount - 1 ? "," : "");
920 
921 			if (offset >= sizeof(cacheLevels))
922 				break;
923 		}
924 
925 		dprintf("CPU %d: cache sharing:%s\n", currentCPU, cacheLevels);
926 	}
927 #endif
928 }
929 
930 
931 static void
932 detect_intel_patch_level(cpu_ent* cpu)
933 {
934 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_HYPERVISOR) {
935 		cpu->arch.patch_level = 0;
936 		return;
937 	}
938 
939 	x86_write_msr(IA32_MSR_UCODE_REV, 0);
940 	cpuid_info cpuid;
941 	get_current_cpuid(&cpuid, 1, 0);
942 
943 	uint64 value = x86_read_msr(IA32_MSR_UCODE_REV);
944 	cpu->arch.patch_level = value >> 32;
945 }
946 
947 
948 static void
949 detect_amd_patch_level(cpu_ent* cpu)
950 {
951 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_HYPERVISOR) {
952 		cpu->arch.patch_level = 0;
953 		return;
954 	}
955 
956 	uint64 value = x86_read_msr(IA32_MSR_UCODE_REV);
957 	cpu->arch.patch_level = value >> 32;
958 }
959 
960 
961 static struct intel_microcode_header*
962 find_microcode_intel(addr_t data, size_t size, uint32 patchLevel)
963 {
964 	// 9.11.3 Processor Identification
965 	cpuid_info cpuid;
966 	get_current_cpuid(&cpuid, 1, 0);
967 	uint32 signature = cpuid.regs.eax;
968 	// 9.11.4 Platform Identification
969 	uint64 platformBits = (x86_read_msr(IA32_MSR_PLATFORM_ID) >> 50) & 0x7;
970 	uint64 mask = 1 << platformBits;
971 
972 	while (size > 0) {
973 		if (size < sizeof(struct intel_microcode_header)) {
974 			dprintf("find_microcode_intel update is too small for header\n");
975 			break;
976 		}
977 		struct intel_microcode_header* header =
978 			(struct intel_microcode_header*)data;
979 
980 		uint32 totalSize = header->total_size;
981 		uint32 dataSize = header->data_size;
982 		if (dataSize == 0) {
983 			dataSize = 2000;
984 			totalSize = sizeof(struct intel_microcode_header)
985 				+ dataSize;
986 		}
987 		if (totalSize > size) {
988 			dprintf("find_microcode_intel update is too small for data\n");
989 			break;
990 		}
991 
992 		uint32* dwords = (uint32*)data;
993 		// prepare the next update
994 		size -= totalSize;
995 		data += totalSize;
996 
997 		if (header->loader_revision != 1) {
998 			dprintf("find_microcode_intel incorrect loader version\n");
999 			continue;
1000 		}
1001 		// 9.11.6 The microcode update data requires a 16-byte boundary
1002 		// alignment.
1003 		if (((addr_t)header % 16) != 0) {
1004 			dprintf("find_microcode_intel incorrect alignment\n");
1005 			continue;
1006 		}
1007 		uint32 sum = 0;
1008 		for (uint32 i = 0; i < totalSize / 4; i++) {
1009 			sum += dwords[i];
1010 		}
1011 		if (sum != 0) {
1012 			dprintf("find_microcode_intel incorrect checksum\n");
1013 			continue;
1014 		}
1015 		if (patchLevel > header->update_revision) {
1016 			dprintf("find_microcode_intel update_revision is lower\n");
1017 			continue;
1018 		}
1019 		if (signature == header->processor_signature
1020 			&& (mask & header->processor_flags) != 0) {
1021 			return header;
1022 		}
1023 		if (totalSize <= (sizeof(struct intel_microcode_header) + dataSize
1024 			+ sizeof(struct intel_microcode_extended_signature_header))) {
1025 			continue;
1026 		}
1027 		struct intel_microcode_extended_signature_header* extSigHeader =
1028 			(struct intel_microcode_extended_signature_header*)((addr_t)header
1029 				+ sizeof(struct intel_microcode_header) + dataSize);
1030 		struct intel_microcode_extended_signature* extended_signature =
1031 			(struct intel_microcode_extended_signature*)((addr_t)extSigHeader
1032 				+ sizeof(struct intel_microcode_extended_signature_header));
1033 		for (uint32 i = 0; i < extSigHeader->extended_signature_count; i++) {
1034 			if (signature == extended_signature[i].processor_signature
1035 				&& (mask & extended_signature[i].processor_flags) != 0)
1036 				return header;
1037 		}
1038 	}
1039 	return NULL;
1040 }
1041 
1042 
1043 static void
1044 load_microcode_intel(int currentCPU, cpu_ent* cpu)
1045 {
1046 	// serialize for HT cores
1047 	if (currentCPU != 0)
1048 		acquire_spinlock(&sUcodeUpdateLock);
1049 	detect_intel_patch_level(cpu);
1050 	uint32 revision = cpu->arch.patch_level;
1051 	struct intel_microcode_header* update = sLoadedUcodeUpdate;
1052 	if (update == NULL) {
1053 		update = find_microcode_intel((addr_t)sUcodeData, sUcodeDataSize,
1054 			revision);
1055 	}
1056 	if (update != NULL) {
1057 		addr_t data = (addr_t)update + sizeof(struct intel_microcode_header);
1058 		wbinvd();
1059 		x86_write_msr(IA32_MSR_UCODE_WRITE, data);
1060 		detect_intel_patch_level(cpu);
1061 		if (revision == cpu->arch.patch_level) {
1062 			dprintf("CPU %d: update failed\n", currentCPU);
1063 		} else {
1064 			if (sLoadedUcodeUpdate == NULL)
1065 				sLoadedUcodeUpdate = update;
1066 			dprintf("CPU %d: updated from revision %" B_PRIu32 " to %" B_PRIu32
1067 				"\n", currentCPU, revision, cpu->arch.patch_level);
1068 		}
1069 	} else {
1070 		dprintf("CPU %d: no update found\n", currentCPU);
1071 	}
1072 	if (currentCPU != 0)
1073 		release_spinlock(&sUcodeUpdateLock);
1074 }
1075 
1076 
1077 static void
1078 load_microcode_amd(int currentCPU, cpu_ent* cpu)
1079 {
1080 	dprintf("CPU %d: no update found\n", currentCPU);
1081 }
1082 
1083 
1084 static void
1085 load_microcode(int currentCPU)
1086 {
1087 	if (sUcodeData == NULL)
1088 		return;
1089 	cpu_ent* cpu = get_cpu_struct();
1090 	if ((cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_HYPERVISOR) != 0)
1091 		return;
1092 	if (cpu->arch.vendor == VENDOR_INTEL)
1093 		load_microcode_intel(currentCPU, cpu);
1094 	else if (cpu->arch.vendor == VENDOR_AMD)
1095 		load_microcode_amd(currentCPU, cpu);
1096 }
1097 
1098 
1099 static void
1100 detect_cpu(int currentCPU)
1101 {
1102 	cpu_ent* cpu = get_cpu_struct();
1103 	char vendorString[17];
1104 	cpuid_info cpuid;
1105 
1106 	// clear out the cpu info data
1107 	cpu->arch.vendor = VENDOR_UNKNOWN;
1108 	cpu->arch.vendor_name = "UNKNOWN VENDOR";
1109 	cpu->arch.feature[FEATURE_COMMON] = 0;
1110 	cpu->arch.feature[FEATURE_EXT] = 0;
1111 	cpu->arch.feature[FEATURE_EXT_AMD] = 0;
1112 	cpu->arch.feature[FEATURE_7_EBX] = 0;
1113 	cpu->arch.feature[FEATURE_7_ECX] = 0;
1114 	cpu->arch.feature[FEATURE_7_EDX] = 0;
1115 	cpu->arch.feature[FEATURE_D_1_EAX] = 0;
1116 	cpu->arch.model_name[0] = 0;
1117 
1118 	// print some fun data
1119 	get_current_cpuid(&cpuid, 0, 0);
1120 	uint32 maxBasicLeaf = cpuid.eax_0.max_eax;
1121 
1122 	// build the vendor string
1123 	memset(vendorString, 0, sizeof(vendorString));
1124 	memcpy(vendorString, cpuid.eax_0.vendor_id, sizeof(cpuid.eax_0.vendor_id));
1125 
1126 	// get the family, model, stepping
1127 	get_current_cpuid(&cpuid, 1, 0);
1128 	cpu->arch.type = cpuid.eax_1.type;
1129 	cpu->arch.family = cpuid.eax_1.family;
1130 	cpu->arch.extended_family = cpuid.eax_1.extended_family;
1131 	cpu->arch.model = cpuid.eax_1.model;
1132 	cpu->arch.extended_model = cpuid.eax_1.extended_model;
1133 	cpu->arch.stepping = cpuid.eax_1.stepping;
1134 	dprintf("CPU %d: type %d family %d extended_family %d model %d "
1135 		"extended_model %d stepping %d, string '%s'\n",
1136 		currentCPU, cpu->arch.type, cpu->arch.family,
1137 		cpu->arch.extended_family, cpu->arch.model,
1138 		cpu->arch.extended_model, cpu->arch.stepping, vendorString);
1139 
1140 	// figure out what vendor we have here
1141 
1142 	for (int32 i = 0; i < VENDOR_NUM; i++) {
1143 		if (vendor_info[i].ident_string[0]
1144 			&& !strcmp(vendorString, vendor_info[i].ident_string[0])) {
1145 			cpu->arch.vendor = (x86_vendors)i;
1146 			cpu->arch.vendor_name = vendor_info[i].vendor;
1147 			break;
1148 		}
1149 		if (vendor_info[i].ident_string[1]
1150 			&& !strcmp(vendorString, vendor_info[i].ident_string[1])) {
1151 			cpu->arch.vendor = (x86_vendors)i;
1152 			cpu->arch.vendor_name = vendor_info[i].vendor;
1153 			break;
1154 		}
1155 	}
1156 
1157 	// see if we can get the model name
1158 	get_current_cpuid(&cpuid, 0x80000000, 0);
1159 	uint32 maxExtendedLeaf = cpuid.eax_0.max_eax;
1160 	if (maxExtendedLeaf >= 0x80000004) {
1161 		// build the model string (need to swap ecx/edx data before copying)
1162 		unsigned int temp;
1163 		memset(cpu->arch.model_name, 0, sizeof(cpu->arch.model_name));
1164 
1165 		get_current_cpuid(&cpuid, 0x80000002, 0);
1166 		temp = cpuid.regs.edx;
1167 		cpuid.regs.edx = cpuid.regs.ecx;
1168 		cpuid.regs.ecx = temp;
1169 		memcpy(cpu->arch.model_name, cpuid.as_chars, sizeof(cpuid.as_chars));
1170 
1171 		get_current_cpuid(&cpuid, 0x80000003, 0);
1172 		temp = cpuid.regs.edx;
1173 		cpuid.regs.edx = cpuid.regs.ecx;
1174 		cpuid.regs.ecx = temp;
1175 		memcpy(cpu->arch.model_name + 16, cpuid.as_chars,
1176 			sizeof(cpuid.as_chars));
1177 
1178 		get_current_cpuid(&cpuid, 0x80000004, 0);
1179 		temp = cpuid.regs.edx;
1180 		cpuid.regs.edx = cpuid.regs.ecx;
1181 		cpuid.regs.ecx = temp;
1182 		memcpy(cpu->arch.model_name + 32, cpuid.as_chars,
1183 			sizeof(cpuid.as_chars));
1184 
1185 		// some cpus return a right-justified string
1186 		int32 i = 0;
1187 		while (cpu->arch.model_name[i] == ' ')
1188 			i++;
1189 		if (i > 0) {
1190 			memmove(cpu->arch.model_name, &cpu->arch.model_name[i],
1191 				strlen(&cpu->arch.model_name[i]) + 1);
1192 		}
1193 
1194 		dprintf("CPU %d: vendor '%s' model name '%s'\n",
1195 			currentCPU, cpu->arch.vendor_name, cpu->arch.model_name);
1196 	} else {
1197 		strlcpy(cpu->arch.model_name, "unknown", sizeof(cpu->arch.model_name));
1198 	}
1199 
1200 	// load feature bits
1201 	get_current_cpuid(&cpuid, 1, 0);
1202 	cpu->arch.feature[FEATURE_COMMON] = cpuid.eax_1.features; // edx
1203 	cpu->arch.feature[FEATURE_EXT] = cpuid.eax_1.extended_features; // ecx
1204 
1205 	if (maxExtendedLeaf >= 0x80000001) {
1206 		get_current_cpuid(&cpuid, 0x80000001, 0);
1207 		if (cpu->arch.vendor == VENDOR_AMD)
1208 			cpu->arch.feature[FEATURE_EXT_AMD_ECX] = cpuid.regs.ecx; // ecx
1209 		cpu->arch.feature[FEATURE_EXT_AMD] = cpuid.regs.edx; // edx
1210 		if (cpu->arch.vendor != VENDOR_AMD)
1211 			cpu->arch.feature[FEATURE_EXT_AMD] &= IA32_FEATURES_INTEL_EXT;
1212 	}
1213 
1214 	if (maxBasicLeaf >= 5) {
1215 		get_current_cpuid(&cpuid, 5, 0);
1216 		cpu->arch.feature[FEATURE_5_ECX] = cpuid.regs.ecx;
1217 	}
1218 
1219 	if (maxBasicLeaf >= 6) {
1220 		get_current_cpuid(&cpuid, 6, 0);
1221 		cpu->arch.feature[FEATURE_6_EAX] = cpuid.regs.eax;
1222 		cpu->arch.feature[FEATURE_6_ECX] = cpuid.regs.ecx;
1223 	}
1224 
1225 	if (maxBasicLeaf >= 7) {
1226 		get_current_cpuid(&cpuid, 7, 0);
1227 		cpu->arch.feature[FEATURE_7_EBX] = cpuid.regs.ebx;
1228 		cpu->arch.feature[FEATURE_7_ECX] = cpuid.regs.ecx;
1229 		cpu->arch.feature[FEATURE_7_EDX] = cpuid.regs.edx;
1230 	}
1231 
1232 	if (maxBasicLeaf >= 0xd) {
1233 		get_current_cpuid(&cpuid, 0xd, 1);
1234 		cpu->arch.feature[FEATURE_D_1_EAX] = cpuid.regs.eax;
1235 	}
1236 
1237 	if (maxExtendedLeaf >= 0x80000007) {
1238 		get_current_cpuid(&cpuid, 0x80000007, 0);
1239 		cpu->arch.feature[FEATURE_EXT_7_EDX] = cpuid.regs.edx;
1240 	}
1241 
1242 	if (maxExtendedLeaf >= 0x80000008) {
1243 		get_current_cpuid(&cpuid, 0x80000008, 0);
1244 			cpu->arch.feature[FEATURE_EXT_8_EBX] = cpuid.regs.ebx;
1245 	}
1246 
1247 	detect_cpu_topology(currentCPU, cpu, maxBasicLeaf, maxExtendedLeaf);
1248 
1249 	if (cpu->arch.vendor == VENDOR_INTEL)
1250 		detect_intel_patch_level(cpu);
1251 	else if (cpu->arch.vendor == VENDOR_AMD)
1252 		detect_amd_patch_level(cpu);
1253 
1254 #if DUMP_FEATURE_STRING
1255 	dump_feature_string(currentCPU, cpu);
1256 #endif
1257 #if DUMP_CPU_PATCHLEVEL
1258 	dprintf("CPU %d: patch_level %" B_PRIu32 "\n", currentCPU,
1259 		cpu->arch.patch_level);
1260 #endif
1261 }
1262 
1263 
1264 bool
1265 x86_check_feature(uint32 feature, enum x86_feature_type type)
1266 {
1267 	cpu_ent* cpu = get_cpu_struct();
1268 
1269 #if 0
1270 	int i;
1271 	dprintf("x86_check_feature: feature 0x%x, type %d\n", feature, type);
1272 	for (i = 0; i < FEATURE_NUM; i++) {
1273 		dprintf("features %d: 0x%x\n", i, cpu->arch.feature[i]);
1274 	}
1275 #endif
1276 
1277 	return (cpu->arch.feature[type] & feature) != 0;
1278 }
1279 
1280 
1281 void*
1282 x86_get_double_fault_stack(int32 cpu, size_t* _size)
1283 {
1284 	*_size = kDoubleFaultStackSize;
1285 	return sDoubleFaultStacks + kDoubleFaultStackSize * cpu;
1286 }
1287 
1288 
1289 /*!	Returns the index of the current CPU. Can only be called from the double
1290 	fault handler.
1291 */
1292 int32
1293 x86_double_fault_get_cpu(void)
1294 {
1295 	addr_t stack = x86_get_stack_frame();
1296 	return (stack - (addr_t)sDoubleFaultStacks) / kDoubleFaultStackSize;
1297 }
1298 
1299 
1300 //	#pragma mark -
1301 
1302 
1303 status_t
1304 arch_cpu_preboot_init_percpu(kernel_args* args, int cpu)
1305 {
1306 	// On SMP system we want to synchronize the CPUs' TSCs, so system_time()
1307 	// will return consistent values.
1308 	if (smp_get_num_cpus() > 1) {
1309 		// let the first CPU prepare the rendezvous point
1310 		if (cpu == 0)
1311 			sTSCSyncRendezvous = smp_get_num_cpus() - 1;
1312 
1313 		// One CPU after the other will drop out of this loop and be caught by
1314 		// the loop below, until the last CPU (0) gets there. Save for +/- a few
1315 		// cycles the CPUs should pass the second loop at the same time.
1316 		while (sTSCSyncRendezvous != cpu) {
1317 		}
1318 
1319 		sTSCSyncRendezvous = cpu - 1;
1320 
1321 		while (sTSCSyncRendezvous != -1) {
1322 		}
1323 
1324 		// reset TSC to 0
1325 		x86_write_msr(IA32_MSR_TSC, 0);
1326 	}
1327 
1328 	x86_descriptors_preboot_init_percpu(args, cpu);
1329 
1330 	return B_OK;
1331 }
1332 
1333 
1334 static void
1335 halt_idle(void)
1336 {
1337 	asm("hlt");
1338 }
1339 
1340 
1341 static void
1342 amdc1e_noarat_idle(void)
1343 {
1344 	uint64 msr = x86_read_msr(K8_MSR_IPM);
1345 	if (msr & K8_CMPHALT)
1346 		x86_write_msr(K8_MSR_IPM, msr & ~K8_CMPHALT);
1347 	halt_idle();
1348 }
1349 
1350 
1351 static bool
1352 detect_amdc1e_noarat()
1353 {
1354 	cpu_ent* cpu = get_cpu_struct();
1355 
1356 	if (cpu->arch.vendor != VENDOR_AMD)
1357 		return false;
1358 
1359 	// Family 0x12 and higher processors support ARAT
1360 	// Family lower than 0xf processors doesn't support C1E
1361 	// Family 0xf with model <= 0x40 procssors doesn't support C1E
1362 	uint32 family = cpu->arch.family + cpu->arch.extended_family;
1363 	uint32 model = (cpu->arch.extended_model << 4) | cpu->arch.model;
1364 	return (family < 0x12 && family > 0xf) || (family == 0xf && model > 0x40);
1365 }
1366 
1367 
1368 status_t
1369 arch_cpu_init_percpu(kernel_args* args, int cpu)
1370 {
1371 	load_microcode(cpu);
1372 	detect_cpu(cpu);
1373 
1374 	if (!gCpuIdleFunc) {
1375 		if (detect_amdc1e_noarat())
1376 			gCpuIdleFunc = amdc1e_noarat_idle;
1377 		else
1378 			gCpuIdleFunc = halt_idle;
1379 	}
1380 
1381 	if (x86_check_feature(IA32_FEATURE_MCE, FEATURE_COMMON))
1382 		x86_write_cr4(x86_read_cr4() | IA32_CR4_MCE);
1383 
1384 #ifdef __x86_64__
1385 	// if RDTSCP is available write cpu number in TSC_AUX
1386 	if (x86_check_feature(IA32_FEATURE_AMD_EXT_RDTSCP, FEATURE_EXT_AMD))
1387 		x86_write_msr(IA32_MSR_TSC_AUX, cpu);
1388 #endif
1389 
1390 	return __x86_patch_errata_percpu(cpu);
1391 }
1392 
1393 
1394 status_t
1395 arch_cpu_init(kernel_args* args)
1396 {
1397 	if (args->ucode_data != NULL
1398 		&& args->ucode_data_size > 0) {
1399 		sUcodeData = args->ucode_data;
1400 		sUcodeDataSize = args->ucode_data_size;
1401 	} else {
1402 		dprintf("CPU: no microcode provided\n");
1403 	}
1404 
1405 	// init the TSC -> system_time() conversion factors
1406 
1407 	uint32 conversionFactor = args->arch_args.system_time_cv_factor;
1408 	uint64 conversionFactorNsecs = (uint64)conversionFactor * 1000;
1409 
1410 #ifdef __x86_64__
1411 	// The x86_64 system_time() implementation uses 64-bit multiplication and
1412 	// therefore shifting is not necessary for low frequencies (it's also not
1413 	// too likely that there'll be any x86_64 CPUs clocked under 1GHz).
1414 	__x86_setup_system_time((uint64)conversionFactor << 32,
1415 		conversionFactorNsecs);
1416 #else
1417 	if (conversionFactorNsecs >> 32 != 0) {
1418 		// the TSC frequency is < 1 GHz, which forces us to shift the factor
1419 		__x86_setup_system_time(conversionFactor, conversionFactorNsecs >> 16,
1420 			true);
1421 	} else {
1422 		// the TSC frequency is >= 1 GHz
1423 		__x86_setup_system_time(conversionFactor, conversionFactorNsecs, false);
1424 	}
1425 #endif
1426 
1427 	// Initialize descriptor tables.
1428 	x86_descriptors_init(args);
1429 
1430 	return B_OK;
1431 }
1432 
1433 
1434 #ifdef __x86_64__
1435 static void
1436 enable_smap(void* dummy, int cpu)
1437 {
1438 	x86_write_cr4(x86_read_cr4() | IA32_CR4_SMAP);
1439 }
1440 
1441 
1442 static void
1443 enable_smep(void* dummy, int cpu)
1444 {
1445 	x86_write_cr4(x86_read_cr4() | IA32_CR4_SMEP);
1446 }
1447 
1448 
1449 static void
1450 enable_osxsave(void* dummy, int cpu)
1451 {
1452 	x86_write_cr4(x86_read_cr4() | IA32_CR4_OSXSAVE);
1453 }
1454 
1455 
1456 static void
1457 enable_xsavemask(void* dummy, int cpu)
1458 {
1459 	xsetbv(0, gXsaveMask);
1460 }
1461 #endif
1462 
1463 
1464 status_t
1465 arch_cpu_init_post_vm(kernel_args* args)
1466 {
1467 	uint32 i;
1468 
1469 	// allocate an area for the double fault stacks
1470 	virtual_address_restrictions virtualRestrictions = {};
1471 	virtualRestrictions.address_specification = B_ANY_KERNEL_ADDRESS;
1472 	physical_address_restrictions physicalRestrictions = {};
1473 	create_area_etc(B_SYSTEM_TEAM, "double fault stacks",
1474 		kDoubleFaultStackSize * smp_get_num_cpus(), B_FULL_LOCK,
1475 		B_KERNEL_READ_AREA | B_KERNEL_WRITE_AREA, CREATE_AREA_DONT_WAIT, 0,
1476 		&virtualRestrictions, &physicalRestrictions,
1477 		(void**)&sDoubleFaultStacks);
1478 
1479 	X86PagingStructures* kernelPagingStructures
1480 		= static_cast<X86VMTranslationMap*>(
1481 			VMAddressSpace::Kernel()->TranslationMap())->PagingStructures();
1482 
1483 	// Set active translation map on each CPU.
1484 	for (i = 0; i < args->num_cpus; i++) {
1485 		gCPU[i].arch.active_paging_structures = kernelPagingStructures;
1486 		kernelPagingStructures->AddReference();
1487 	}
1488 
1489 	if (!apic_available())
1490 		x86_init_fpu();
1491 	// else fpu gets set up in smp code
1492 
1493 #ifdef __x86_64__
1494 	// if available enable SMEP (Supervisor Memory Execution Protection)
1495 	if (x86_check_feature(IA32_FEATURE_SMEP, FEATURE_7_EBX)) {
1496 		if (!get_safemode_boolean(B_SAFEMODE_DISABLE_SMEP_SMAP, false)) {
1497 			dprintf("enable SMEP\n");
1498 			call_all_cpus_sync(&enable_smep, NULL);
1499 		} else
1500 			dprintf("SMEP disabled per safemode setting\n");
1501 	}
1502 
1503 	// if available enable SMAP (Supervisor Memory Access Protection)
1504 	if (x86_check_feature(IA32_FEATURE_SMAP, FEATURE_7_EBX)) {
1505 		if (!get_safemode_boolean(B_SAFEMODE_DISABLE_SMEP_SMAP, false)) {
1506 			dprintf("enable SMAP\n");
1507 			call_all_cpus_sync(&enable_smap, NULL);
1508 
1509 			arch_altcodepatch_replace(ALTCODEPATCH_TAG_STAC, &_stac, 3);
1510 			arch_altcodepatch_replace(ALTCODEPATCH_TAG_CLAC, &_clac, 3);
1511 		} else
1512 			dprintf("SMAP disabled per safemode setting\n");
1513 	}
1514 
1515 	// if available enable XSAVE (XSAVE and extended states)
1516 	gHasXsave = x86_check_feature(IA32_FEATURE_EXT_XSAVE, FEATURE_EXT);
1517 	if (gHasXsave) {
1518 		gHasXsavec = x86_check_feature(IA32_FEATURE_XSAVEC,
1519 			FEATURE_D_1_EAX);
1520 
1521 		call_all_cpus_sync(&enable_osxsave, NULL);
1522 		gXsaveMask = IA32_XCR0_X87 | IA32_XCR0_SSE;
1523 		cpuid_info cpuid;
1524 		get_current_cpuid(&cpuid, 0xd, 0);
1525 		gXsaveMask |= (cpuid.regs.eax & IA32_XCR0_AVX);
1526 		call_all_cpus_sync(&enable_xsavemask, NULL);
1527 		get_current_cpuid(&cpuid, 0xd, 0);
1528 		gFPUSaveLength = cpuid.regs.ebx;
1529 
1530 		arch_altcodepatch_replace(ALTCODEPATCH_TAG_XSAVE,
1531 			gHasXsavec ? &_xsavec : &_xsave, 4);
1532 		arch_altcodepatch_replace(ALTCODEPATCH_TAG_XRSTOR,
1533 			&_xrstor, 4);
1534 
1535 		dprintf("enable %s 0x%" B_PRIx64 " %" B_PRId64 "\n",
1536 			gHasXsavec ? "XSAVEC" : "XSAVE", gXsaveMask, gFPUSaveLength);
1537 	}
1538 
1539 #endif
1540 
1541 	return B_OK;
1542 }
1543 
1544 
1545 status_t
1546 arch_cpu_init_post_modules(kernel_args* args)
1547 {
1548 	// initialize CPU module
1549 
1550 	void* cookie = open_module_list("cpu");
1551 
1552 	while (true) {
1553 		char name[B_FILE_NAME_LENGTH];
1554 		size_t nameLength = sizeof(name);
1555 
1556 		if (read_next_module_name(cookie, name, &nameLength) != B_OK
1557 			|| get_module(name, (module_info**)&sCpuModule) == B_OK)
1558 			break;
1559 	}
1560 
1561 	close_module_list(cookie);
1562 
1563 	// initialize MTRRs if available
1564 	if (x86_count_mtrrs() > 0) {
1565 		sCpuRendezvous = sCpuRendezvous2 = 0;
1566 		call_all_cpus(&init_mtrrs, NULL);
1567 	}
1568 
1569 	size_t threadExitLen = (addr_t)x86_end_userspace_thread_exit
1570 		- (addr_t)x86_userspace_thread_exit;
1571 	addr_t threadExitPosition = fill_commpage_entry(
1572 		COMMPAGE_ENTRY_X86_THREAD_EXIT, (const void*)x86_userspace_thread_exit,
1573 		threadExitLen);
1574 
1575 	// add the functions to the commpage image
1576 	image_id image = get_commpage_image();
1577 
1578 	elf_add_memory_image_symbol(image, "commpage_thread_exit",
1579 		threadExitPosition, threadExitLen, B_SYMBOL_TYPE_TEXT);
1580 
1581 	return B_OK;
1582 }
1583 
1584 
1585 void
1586 arch_cpu_user_TLB_invalidate(void)
1587 {
1588 	x86_write_cr3(x86_read_cr3());
1589 }
1590 
1591 
1592 void
1593 arch_cpu_global_TLB_invalidate(void)
1594 {
1595 	uint32 flags = x86_read_cr4();
1596 
1597 	if (flags & IA32_CR4_GLOBAL_PAGES) {
1598 		// disable and reenable the global pages to flush all TLBs regardless
1599 		// of the global page bit
1600 		x86_write_cr4(flags & ~IA32_CR4_GLOBAL_PAGES);
1601 		x86_write_cr4(flags | IA32_CR4_GLOBAL_PAGES);
1602 	} else {
1603 		cpu_status state = disable_interrupts();
1604 		arch_cpu_user_TLB_invalidate();
1605 		restore_interrupts(state);
1606 	}
1607 }
1608 
1609 
1610 void
1611 arch_cpu_invalidate_TLB_range(addr_t start, addr_t end)
1612 {
1613 	int32 num_pages = end / B_PAGE_SIZE - start / B_PAGE_SIZE;
1614 	while (num_pages-- >= 0) {
1615 		invalidate_TLB(start);
1616 		start += B_PAGE_SIZE;
1617 	}
1618 }
1619 
1620 
1621 void
1622 arch_cpu_invalidate_TLB_list(addr_t pages[], int num_pages)
1623 {
1624 	int i;
1625 	for (i = 0; i < num_pages; i++) {
1626 		invalidate_TLB(pages[i]);
1627 	}
1628 }
1629 
1630 
1631 status_t
1632 arch_cpu_shutdown(bool rebootSystem)
1633 {
1634 	if (acpi_shutdown(rebootSystem) == B_OK)
1635 		return B_OK;
1636 
1637 	if (!rebootSystem) {
1638 #ifndef __x86_64__
1639 		return apm_shutdown();
1640 #else
1641 		return B_NOT_SUPPORTED;
1642 #endif
1643 	}
1644 
1645 	cpu_status state = disable_interrupts();
1646 
1647 	// try to reset the system using the keyboard controller
1648 	out8(0xfe, 0x64);
1649 
1650 	// Give some time to the controller to do its job (0.5s)
1651 	snooze(500000);
1652 
1653 	// if that didn't help, try it this way
1654 	x86_reboot();
1655 
1656 	restore_interrupts(state);
1657 	return B_ERROR;
1658 }
1659 
1660 
1661 void
1662 arch_cpu_sync_icache(void* address, size_t length)
1663 {
1664 	// instruction cache is always consistent on x86
1665 }
1666 
1667