xref: /haiku/src/system/kernel/arch/x86/arch_cpu.cpp (revision 52c4471a3024d2eb81fe88e2c3982b9f8daa5e56)
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 void* 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_ECX] & IA32_FEATURE_AVX512VMBI)
588 		strlcat(features, "avx512vmbi ", sizeof(features));
589 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_UMIP)
590 		strlcat(features, "umip ", sizeof(features));
591 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_PKU)
592 		strlcat(features, "pku ", sizeof(features));
593 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_OSPKE)
594 		strlcat(features, "ospke ", sizeof(features));
595 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_AVX512VMBI2)
596 		strlcat(features, "avx512vmbi2 ", sizeof(features));
597 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_GFNI)
598 		strlcat(features, "gfni ", sizeof(features));
599 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_VAES)
600 		strlcat(features, "vaes ", sizeof(features));
601 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_VPCLMULQDQ)
602 		strlcat(features, "vpclmulqdq ", sizeof(features));
603 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_AVX512_VNNI)
604 		strlcat(features, "avx512vnni ", sizeof(features));
605 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_AVX512_BITALG)
606 		strlcat(features, "avx512bitalg ", sizeof(features));
607 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_AVX512_VPOPCNTDQ)
608 		strlcat(features, "avx512vpopcntdq ", sizeof(features));
609 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_LA57)
610 		strlcat(features, "la57 ", sizeof(features));
611 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_RDPID)
612 		strlcat(features, "rdpid ", sizeof(features));
613 	if (cpu->arch.feature[FEATURE_7_ECX] & IA32_FEATURE_SGX_LC)
614 		strlcat(features, "sgx_lc ", sizeof(features));
615 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_IBRS)
616 		strlcat(features, "ibrs ", sizeof(features));
617 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_STIBP)
618 		strlcat(features, "stibp ", sizeof(features));
619 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_L1D_FLUSH)
620 		strlcat(features, "l1d_flush ", sizeof(features));
621 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_ARCH_CAPABILITIES)
622 		strlcat(features, "msr_arch ", sizeof(features));
623 	if (cpu->arch.feature[FEATURE_7_EDX] & IA32_FEATURE_SSBD)
624 		strlcat(features, "ssbd ", sizeof(features));
625 	if (cpu->arch.feature[FEATURE_D_1_EAX] & IA32_FEATURE_XSAVEOPT)
626 		strlcat(features, "xsaveopt ", sizeof(features));
627 	if (cpu->arch.feature[FEATURE_D_1_EAX] & IA32_FEATURE_XSAVEC)
628 		strlcat(features, "xsavec ", sizeof(features));
629 	if (cpu->arch.feature[FEATURE_D_1_EAX] & IA32_FEATURE_XGETBV1)
630 		strlcat(features, "xgetbv1 ", sizeof(features));
631 	if (cpu->arch.feature[FEATURE_D_1_EAX] & IA32_FEATURE_XSAVES)
632 		strlcat(features, "xsaves ", sizeof(features));
633 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_CLZERO)
634 		strlcat(features, "clzero ", sizeof(features));
635 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_IBPB)
636 		strlcat(features, "ibpb ", sizeof(features));
637 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_AMD_SSBD)
638 		strlcat(features, "amd_ssbd ", sizeof(features));
639 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_VIRT_SSBD)
640 		strlcat(features, "virt_ssbd ", sizeof(features));
641 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_AMD_SSB_NO)
642 		strlcat(features, "amd_ssb_no ", sizeof(features));
643 	if (cpu->arch.feature[FEATURE_EXT_8_EBX] & IA32_FEATURE_CPPC)
644 		strlcat(features, "cppc ", sizeof(features));
645 	dprintf("CPU %d: features: %s\n", currentCPU, features);
646 }
647 #endif	// DUMP_FEATURE_STRING
648 
649 
650 static void
651 compute_cpu_hierarchy_masks(int maxLogicalID, int maxCoreID)
652 {
653 	ASSERT(maxLogicalID >= maxCoreID);
654 	const int kMaxSMTID = maxLogicalID / maxCoreID;
655 
656 	sHierarchyMask[CPU_TOPOLOGY_SMT] = kMaxSMTID - 1;
657 	sHierarchyShift[CPU_TOPOLOGY_SMT] = 0;
658 
659 	sHierarchyMask[CPU_TOPOLOGY_CORE] = (maxCoreID - 1) * kMaxSMTID;
660 	sHierarchyShift[CPU_TOPOLOGY_CORE]
661 		= count_set_bits(sHierarchyMask[CPU_TOPOLOGY_SMT]);
662 
663 	const uint32 kSinglePackageMask = sHierarchyMask[CPU_TOPOLOGY_SMT]
664 		| sHierarchyMask[CPU_TOPOLOGY_CORE];
665 	sHierarchyMask[CPU_TOPOLOGY_PACKAGE] = ~kSinglePackageMask;
666 	sHierarchyShift[CPU_TOPOLOGY_PACKAGE] = count_set_bits(kSinglePackageMask);
667 }
668 
669 
670 static uint32
671 get_cpu_legacy_initial_apic_id(int /* currentCPU */)
672 {
673 	cpuid_info cpuid;
674 	get_current_cpuid(&cpuid, 1, 0);
675 	return cpuid.regs.ebx >> 24;
676 }
677 
678 
679 static inline status_t
680 detect_amd_cpu_topology(uint32 maxBasicLeaf, uint32 maxExtendedLeaf)
681 {
682 	sGetCPUTopologyID = get_cpu_legacy_initial_apic_id;
683 
684 	cpuid_info cpuid;
685 	get_current_cpuid(&cpuid, 1, 0);
686 	int maxLogicalID = next_power_of_2((cpuid.regs.ebx >> 16) & 0xff);
687 
688 	int maxCoreID = 1;
689 	if (maxExtendedLeaf >= 0x80000008) {
690 		get_current_cpuid(&cpuid, 0x80000008, 0);
691 		maxCoreID = (cpuid.regs.ecx >> 12) & 0xf;
692 		if (maxCoreID != 0)
693 			maxCoreID = 1 << maxCoreID;
694 		else
695 			maxCoreID = next_power_of_2((cpuid.regs.edx & 0xf) + 1);
696 	}
697 
698 	if (maxExtendedLeaf >= 0x80000001) {
699 		get_current_cpuid(&cpuid, 0x80000001, 0);
700 		if (x86_check_feature(IA32_FEATURE_AMD_EXT_CMPLEGACY,
701 				FEATURE_EXT_AMD_ECX))
702 			maxCoreID = maxLogicalID;
703 	}
704 
705 	compute_cpu_hierarchy_masks(maxLogicalID, maxCoreID);
706 
707 	return B_OK;
708 }
709 
710 
711 static void
712 detect_amd_cache_topology(uint32 maxExtendedLeaf)
713 {
714 	if (!x86_check_feature(IA32_FEATURE_AMD_EXT_TOPOLOGY, FEATURE_EXT_AMD_ECX))
715 		return;
716 
717 	if (maxExtendedLeaf < 0x8000001d)
718 		return;
719 
720 	uint8 hierarchyLevels[CPU_MAX_CACHE_LEVEL];
721 	int maxCacheLevel = 0;
722 
723 	int currentLevel = 0;
724 	int cacheType;
725 	do {
726 		cpuid_info cpuid;
727 		get_current_cpuid(&cpuid, 0x8000001d, currentLevel);
728 
729 		cacheType = cpuid.regs.eax & 0x1f;
730 		if (cacheType == 0)
731 			break;
732 
733 		int cacheLevel = (cpuid.regs.eax >> 5) & 0x7;
734 		int coresCount = next_power_of_2(((cpuid.regs.eax >> 14) & 0x3f) + 1);
735 		hierarchyLevels[cacheLevel - 1]
736 			= coresCount * (sHierarchyMask[CPU_TOPOLOGY_SMT] + 1);
737 		maxCacheLevel = std::max(maxCacheLevel, cacheLevel);
738 
739 		currentLevel++;
740 	} while (true);
741 
742 	for (int i = 0; i < maxCacheLevel; i++)
743 		sCacheSharingMask[i] = ~uint32(hierarchyLevels[i] - 1);
744 	gCPUCacheLevelCount = maxCacheLevel;
745 }
746 
747 
748 static uint32
749 get_intel_cpu_initial_x2apic_id(int /* currentCPU */)
750 {
751 	cpuid_info cpuid;
752 	get_current_cpuid(&cpuid, 11, 0);
753 	return cpuid.regs.edx;
754 }
755 
756 
757 static inline status_t
758 detect_intel_cpu_topology_x2apic(uint32 maxBasicLeaf)
759 {
760 
761 	uint32 leaf = 0;
762 	cpuid_info cpuid;
763 	if (maxBasicLeaf >= 0x1f) {
764 		get_current_cpuid(&cpuid, 0x1f, 0);
765 		if (cpuid.regs.ebx != 0)
766 			leaf = 0x1f;
767 	}
768 	if (maxBasicLeaf >= 0xb && leaf == 0) {
769 		get_current_cpuid(&cpuid, 0xb, 0);
770 		if (cpuid.regs.ebx != 0)
771 			leaf = 0xb;
772 	}
773 	if (leaf == 0)
774 		return B_UNSUPPORTED;
775 
776 	uint8 hierarchyLevels[CPU_TOPOLOGY_LEVELS] = { 0 };
777 
778 	int currentLevel = 0;
779 	unsigned int levelsSet = 0;
780 	do {
781 		cpuid_info cpuid;
782 		get_current_cpuid(&cpuid, leaf, currentLevel++);
783 		int levelType = (cpuid.regs.ecx >> 8) & 0xff;
784 		int levelValue = cpuid.regs.eax & 0x1f;
785 
786 		if (levelType == 0)
787 			break;
788 
789 		switch (levelType) {
790 			case 1:	// SMT
791 				hierarchyLevels[CPU_TOPOLOGY_SMT] = levelValue;
792 				levelsSet |= 1;
793 				break;
794 			case 2:	// core
795 				hierarchyLevels[CPU_TOPOLOGY_CORE] = levelValue;
796 				levelsSet |= 2;
797 				break;
798 		}
799 
800 	} while (levelsSet != 3);
801 
802 	sGetCPUTopologyID = get_intel_cpu_initial_x2apic_id;
803 
804 	for (int i = 1; i < CPU_TOPOLOGY_LEVELS; i++) {
805 		if ((levelsSet & (1u << i)) != 0)
806 			continue;
807 		hierarchyLevels[i] = hierarchyLevels[i - 1];
808 	}
809 
810 	for (int i = 0; i < CPU_TOPOLOGY_LEVELS; i++) {
811 		uint32 mask = ~uint32(0);
812 		if (i < CPU_TOPOLOGY_LEVELS - 1)
813 			mask = (1u << hierarchyLevels[i]) - 1;
814 		if (i > 0)
815 			mask &= ~sHierarchyMask[i - 1];
816 		sHierarchyMask[i] = mask;
817 		sHierarchyShift[i] = i > 0 ? hierarchyLevels[i - 1] : 0;
818 	}
819 
820 	return B_OK;
821 }
822 
823 
824 static inline status_t
825 detect_intel_cpu_topology_legacy(uint32 maxBasicLeaf)
826 {
827 	sGetCPUTopologyID = get_cpu_legacy_initial_apic_id;
828 
829 	cpuid_info cpuid;
830 
831 	get_current_cpuid(&cpuid, 1, 0);
832 	int maxLogicalID = next_power_of_2((cpuid.regs.ebx >> 16) & 0xff);
833 
834 	int maxCoreID = 1;
835 	if (maxBasicLeaf >= 4) {
836 		get_current_cpuid(&cpuid, 4, 0);
837 		maxCoreID = next_power_of_2((cpuid.regs.eax >> 26) + 1);
838 	}
839 
840 	compute_cpu_hierarchy_masks(maxLogicalID, maxCoreID);
841 
842 	return B_OK;
843 }
844 
845 
846 static void
847 detect_intel_cache_topology(uint32 maxBasicLeaf)
848 {
849 	if (maxBasicLeaf < 4)
850 		return;
851 
852 	uint8 hierarchyLevels[CPU_MAX_CACHE_LEVEL];
853 	int maxCacheLevel = 0;
854 
855 	int currentLevel = 0;
856 	int cacheType;
857 	do {
858 		cpuid_info cpuid;
859 		get_current_cpuid(&cpuid, 4, currentLevel);
860 
861 		cacheType = cpuid.regs.eax & 0x1f;
862 		if (cacheType == 0)
863 			break;
864 
865 		int cacheLevel = (cpuid.regs.eax >> 5) & 0x7;
866 		hierarchyLevels[cacheLevel - 1]
867 			= next_power_of_2(((cpuid.regs.eax >> 14) & 0x3f) + 1);
868 		maxCacheLevel = std::max(maxCacheLevel, cacheLevel);
869 
870 		currentLevel++;
871 	} while (true);
872 
873 	for (int i = 0; i < maxCacheLevel; i++)
874 		sCacheSharingMask[i] = ~uint32(hierarchyLevels[i] - 1);
875 
876 	gCPUCacheLevelCount = maxCacheLevel;
877 }
878 
879 
880 static uint32
881 get_simple_cpu_topology_id(int currentCPU)
882 {
883 	return currentCPU;
884 }
885 
886 
887 static inline int
888 get_topology_level_id(uint32 id, cpu_topology_level level)
889 {
890 	ASSERT(level < CPU_TOPOLOGY_LEVELS);
891 	return (id & sHierarchyMask[level]) >> sHierarchyShift[level];
892 }
893 
894 
895 static void
896 detect_cpu_topology(int currentCPU, cpu_ent* cpu, uint32 maxBasicLeaf,
897 	uint32 maxExtendedLeaf)
898 {
899 	if (currentCPU == 0) {
900 		memset(sCacheSharingMask, 0xff, sizeof(sCacheSharingMask));
901 
902 		status_t result = B_UNSUPPORTED;
903 		if (x86_check_feature(IA32_FEATURE_HTT, FEATURE_COMMON)) {
904 			if (cpu->arch.vendor == VENDOR_AMD
905 				|| cpu->arch.vendor == VENDOR_HYGON) {
906 				result = detect_amd_cpu_topology(maxBasicLeaf, maxExtendedLeaf);
907 
908 				if (result == B_OK)
909 					detect_amd_cache_topology(maxExtendedLeaf);
910 			}
911 
912 			if (cpu->arch.vendor == VENDOR_INTEL) {
913 				result = detect_intel_cpu_topology_x2apic(maxBasicLeaf);
914 				if (result != B_OK)
915 					result = detect_intel_cpu_topology_legacy(maxBasicLeaf);
916 
917 				if (result == B_OK)
918 					detect_intel_cache_topology(maxBasicLeaf);
919 			}
920 		}
921 
922 		if (result != B_OK) {
923 			dprintf("No CPU topology information available.\n");
924 
925 			sGetCPUTopologyID = get_simple_cpu_topology_id;
926 
927 			sHierarchyMask[CPU_TOPOLOGY_PACKAGE] = ~uint32(0);
928 		}
929 	}
930 
931 	ASSERT(sGetCPUTopologyID != NULL);
932 	int topologyID = sGetCPUTopologyID(currentCPU);
933 	cpu->topology_id[CPU_TOPOLOGY_SMT]
934 		= get_topology_level_id(topologyID, CPU_TOPOLOGY_SMT);
935 	cpu->topology_id[CPU_TOPOLOGY_CORE]
936 		= get_topology_level_id(topologyID, CPU_TOPOLOGY_CORE);
937 	cpu->topology_id[CPU_TOPOLOGY_PACKAGE]
938 		= get_topology_level_id(topologyID, CPU_TOPOLOGY_PACKAGE);
939 
940 	unsigned int i;
941 	for (i = 0; i < gCPUCacheLevelCount; i++)
942 		cpu->cache_id[i] = topologyID & sCacheSharingMask[i];
943 	for (; i < CPU_MAX_CACHE_LEVEL; i++)
944 		cpu->cache_id[i] = -1;
945 
946 #if DUMP_CPU_TOPOLOGY
947 	dprintf("CPU %d: apic id %d, package %d, core %d, smt %d\n", currentCPU,
948 		topologyID, cpu->topology_id[CPU_TOPOLOGY_PACKAGE],
949 		cpu->topology_id[CPU_TOPOLOGY_CORE],
950 		cpu->topology_id[CPU_TOPOLOGY_SMT]);
951 
952 	if (gCPUCacheLevelCount > 0) {
953 		char cacheLevels[256];
954 		unsigned int offset = 0;
955 		for (i = 0; i < gCPUCacheLevelCount; i++) {
956 			offset += snprintf(cacheLevels + offset,
957 					sizeof(cacheLevels) - offset,
958 					" L%d id %d%s", i + 1, cpu->cache_id[i],
959 					i < gCPUCacheLevelCount - 1 ? "," : "");
960 
961 			if (offset >= sizeof(cacheLevels))
962 				break;
963 		}
964 
965 		dprintf("CPU %d: cache sharing:%s\n", currentCPU, cacheLevels);
966 	}
967 #endif
968 }
969 
970 
971 static void
972 detect_intel_patch_level(cpu_ent* cpu)
973 {
974 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_HYPERVISOR) {
975 		cpu->arch.patch_level = 0;
976 		return;
977 	}
978 
979 	x86_write_msr(IA32_MSR_UCODE_REV, 0);
980 	cpuid_info cpuid;
981 	get_current_cpuid(&cpuid, 1, 0);
982 
983 	uint64 value = x86_read_msr(IA32_MSR_UCODE_REV);
984 	cpu->arch.patch_level = value >> 32;
985 }
986 
987 
988 static void
989 detect_amd_patch_level(cpu_ent* cpu)
990 {
991 	if (cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_HYPERVISOR) {
992 		cpu->arch.patch_level = 0;
993 		return;
994 	}
995 
996 	uint64 value = x86_read_msr(IA32_MSR_UCODE_REV);
997 	cpu->arch.patch_level = (uint32)value;
998 }
999 
1000 
1001 static struct intel_microcode_header*
1002 find_microcode_intel(addr_t data, size_t size, uint32 patchLevel)
1003 {
1004 	// 9.11.3 Processor Identification
1005 	cpuid_info cpuid;
1006 	get_current_cpuid(&cpuid, 1, 0);
1007 	uint32 signature = cpuid.regs.eax;
1008 	// 9.11.4 Platform Identification
1009 	uint64 platformBits = (x86_read_msr(IA32_MSR_PLATFORM_ID) >> 50) & 0x7;
1010 	uint64 mask = 1 << platformBits;
1011 
1012 	while (size > 0) {
1013 		if (size < sizeof(struct intel_microcode_header)) {
1014 			dprintf("find_microcode_intel update is too small for header\n");
1015 			break;
1016 		}
1017 		struct intel_microcode_header* header =
1018 			(struct intel_microcode_header*)data;
1019 
1020 		uint32 totalSize = header->total_size;
1021 		uint32 dataSize = header->data_size;
1022 		if (dataSize == 0) {
1023 			dataSize = 2000;
1024 			totalSize = sizeof(struct intel_microcode_header)
1025 				+ dataSize;
1026 		}
1027 		if (totalSize > size) {
1028 			dprintf("find_microcode_intel update is too small for data\n");
1029 			break;
1030 		}
1031 
1032 		uint32* dwords = (uint32*)data;
1033 		// prepare the next update
1034 		size -= totalSize;
1035 		data += totalSize;
1036 
1037 		if (header->loader_revision != 1) {
1038 			dprintf("find_microcode_intel incorrect loader version\n");
1039 			continue;
1040 		}
1041 		// 9.11.6 The microcode update data requires a 16-byte boundary
1042 		// alignment.
1043 		if (((addr_t)header % 16) != 0) {
1044 			dprintf("find_microcode_intel incorrect alignment\n");
1045 			continue;
1046 		}
1047 		uint32 sum = 0;
1048 		for (uint32 i = 0; i < totalSize / 4; i++) {
1049 			sum += dwords[i];
1050 		}
1051 		if (sum != 0) {
1052 			dprintf("find_microcode_intel incorrect checksum\n");
1053 			continue;
1054 		}
1055 		if (patchLevel > header->update_revision) {
1056 			dprintf("find_microcode_intel update_revision is lower\n");
1057 			continue;
1058 		}
1059 		if (signature == header->processor_signature
1060 			&& (mask & header->processor_flags) != 0) {
1061 			return header;
1062 		}
1063 		if (totalSize <= (sizeof(struct intel_microcode_header) + dataSize
1064 			+ sizeof(struct intel_microcode_extended_signature_header))) {
1065 			continue;
1066 		}
1067 		struct intel_microcode_extended_signature_header* extSigHeader =
1068 			(struct intel_microcode_extended_signature_header*)((addr_t)header
1069 				+ sizeof(struct intel_microcode_header) + dataSize);
1070 		struct intel_microcode_extended_signature* extended_signature =
1071 			(struct intel_microcode_extended_signature*)((addr_t)extSigHeader
1072 				+ sizeof(struct intel_microcode_extended_signature_header));
1073 		for (uint32 i = 0; i < extSigHeader->extended_signature_count; i++) {
1074 			if (signature == extended_signature[i].processor_signature
1075 				&& (mask & extended_signature[i].processor_flags) != 0)
1076 				return header;
1077 		}
1078 	}
1079 	return NULL;
1080 }
1081 
1082 
1083 static void
1084 load_microcode_intel(int currentCPU, cpu_ent* cpu)
1085 {
1086 	// serialize for HT cores
1087 	if (currentCPU != 0)
1088 		acquire_spinlock(&sUcodeUpdateLock);
1089 	detect_intel_patch_level(cpu);
1090 	uint32 revision = cpu->arch.patch_level;
1091 	struct intel_microcode_header* update = (struct intel_microcode_header*)sLoadedUcodeUpdate;
1092 	if (update == NULL) {
1093 		update = find_microcode_intel((addr_t)sUcodeData, sUcodeDataSize,
1094 			revision);
1095 	}
1096 	if (update != NULL) {
1097 		addr_t data = (addr_t)update + sizeof(struct intel_microcode_header);
1098 		wbinvd();
1099 		x86_write_msr(IA32_MSR_UCODE_WRITE, data);
1100 		detect_intel_patch_level(cpu);
1101 		if (revision == cpu->arch.patch_level) {
1102 			dprintf("CPU %d: update failed\n", currentCPU);
1103 		} else {
1104 			if (sLoadedUcodeUpdate == NULL)
1105 				sLoadedUcodeUpdate = update;
1106 			dprintf("CPU %d: updated from revision %" B_PRIu32 " to %" B_PRIu32
1107 				"\n", currentCPU, revision, cpu->arch.patch_level);
1108 		}
1109 	} else {
1110 		dprintf("CPU %d: no update found\n", currentCPU);
1111 	}
1112 	if (currentCPU != 0)
1113 		release_spinlock(&sUcodeUpdateLock);
1114 }
1115 
1116 
1117 static struct amd_microcode_header*
1118 find_microcode_amd(addr_t data, size_t size, uint32 patchLevel)
1119 {
1120 	// 9.11.3 Processor Identification
1121 	cpuid_info cpuid;
1122 	get_current_cpuid(&cpuid, 1, 0);
1123 	uint32 signature = cpuid.regs.eax;
1124 
1125 	if (size < sizeof(struct amd_container_header)) {
1126 		dprintf("find_microcode_amd update is too small for header\n");
1127 		return NULL;
1128 	}
1129 	struct amd_container_header* container = (struct amd_container_header*)data;
1130 	if (container->magic != 0x414d44) {
1131 		dprintf("find_microcode_amd update invalid magic\n");
1132 		return NULL;
1133 	}
1134 
1135 	size -= sizeof(*container);
1136 	data += sizeof(*container);
1137 
1138 	struct amd_section_header* section =
1139 		(struct amd_section_header*)data;
1140 	if (section->type != 0 || section->size == 0) {
1141 		dprintf("find_microcode_amd update first section invalid\n");
1142 		return NULL;
1143 	}
1144 
1145 	size -= sizeof(*section);
1146 	data += sizeof(*section);
1147 
1148 	amd_equiv_cpu_entry* table = (amd_equiv_cpu_entry*)data;
1149 	size -= section->size;
1150 	data += section->size;
1151 
1152 	uint16 equiv_id = 0;
1153 	for (uint32 i = 0; table[i].installed_cpu != 0; i++) {
1154 		if (signature == table[i].equiv_cpu) {
1155 			equiv_id = table[i].equiv_cpu;
1156 			dprintf("find_microcode_amd found equiv cpu: %x\n", equiv_id);
1157 			break;
1158 		}
1159 	}
1160 	if (equiv_id == 0) {
1161 		dprintf("find_microcode_amd update cpu not found in equiv table\n");
1162 		return NULL;
1163 	}
1164 
1165 	while (size > sizeof(amd_section_header)) {
1166 		struct amd_section_header* section = (struct amd_section_header*)data;
1167 		size -= sizeof(*section);
1168 		data += sizeof(*section);
1169 
1170 		if (section->type != 1 || section->size > size
1171 			|| section->size < sizeof(amd_microcode_header)) {
1172 			dprintf("find_microcode_amd update firmware section invalid\n");
1173 			return NULL;
1174 		}
1175 		struct amd_microcode_header* header = (struct amd_microcode_header*)data;
1176 		size -= section->size;
1177 		data += section->size;
1178 
1179 		if (header->processor_rev_id != equiv_id) {
1180 			dprintf("find_microcode_amd update found rev_id %x\n", header->processor_rev_id);
1181 			continue;
1182 		}
1183 		if (patchLevel >= header->patch_id) {
1184 			dprintf("find_microcode_intel update_revision is lower\n");
1185 			continue;
1186 		}
1187 		if (header->nb_dev_id != 0 || header->sb_dev_id != 0) {
1188 			dprintf("find_microcode_amd update chipset specific firmware\n");
1189 			continue;
1190 		}
1191 		if (((addr_t)header % 16) != 0) {
1192 			dprintf("find_microcode_amd incorrect alignment\n");
1193 			continue;
1194 		}
1195 
1196 		return header;
1197 	}
1198 	dprintf("find_microcode_amd no fw update found for this cpu\n");
1199 	return NULL;
1200 }
1201 
1202 
1203 static void
1204 load_microcode_amd(int currentCPU, cpu_ent* cpu)
1205 {
1206 	// serialize for HT cores
1207 	if (currentCPU != 0)
1208 		acquire_spinlock(&sUcodeUpdateLock);
1209 	detect_amd_patch_level(cpu);
1210 	uint32 revision = cpu->arch.patch_level;
1211 	struct amd_microcode_header* update = (struct amd_microcode_header*)sLoadedUcodeUpdate;
1212 	if (update == NULL) {
1213 		update = find_microcode_amd((addr_t)sUcodeData, sUcodeDataSize,
1214 			revision);
1215 	}
1216 	if (update != NULL) {
1217 		addr_t data = (addr_t)update;
1218 		wbinvd();
1219 
1220 		x86_write_msr(MSR_K8_UCODE_UPDATE, data);
1221 
1222 		detect_amd_patch_level(cpu);
1223 		if (revision == cpu->arch.patch_level) {
1224 			dprintf("CPU %d: update failed\n", currentCPU);
1225 		} else {
1226 			if (sLoadedUcodeUpdate == NULL)
1227 				sLoadedUcodeUpdate = update;
1228 			dprintf("CPU %d: updated from revision 0x%" B_PRIx32 " to 0x%" B_PRIx32
1229 				"\n", currentCPU, revision, cpu->arch.patch_level);
1230 		}
1231 
1232 	} else {
1233 		dprintf("CPU %d: no update found\n", currentCPU);
1234 	}
1235 
1236 	if (currentCPU != 0)
1237 		release_spinlock(&sUcodeUpdateLock);
1238 }
1239 
1240 
1241 static void
1242 load_microcode(int currentCPU)
1243 {
1244 	if (sUcodeData == NULL)
1245 		return;
1246 	cpu_ent* cpu = get_cpu_struct();
1247 	if ((cpu->arch.feature[FEATURE_EXT] & IA32_FEATURE_EXT_HYPERVISOR) != 0)
1248 		return;
1249 	if (cpu->arch.vendor == VENDOR_INTEL)
1250 		load_microcode_intel(currentCPU, cpu);
1251 	else if (cpu->arch.vendor == VENDOR_AMD)
1252 		load_microcode_amd(currentCPU, cpu);
1253 }
1254 
1255 
1256 static void
1257 detect_cpu(int currentCPU, bool full = true)
1258 {
1259 	cpu_ent* cpu = get_cpu_struct();
1260 	char vendorString[17];
1261 	cpuid_info cpuid;
1262 
1263 	// clear out the cpu info data
1264 	cpu->arch.vendor = VENDOR_UNKNOWN;
1265 	cpu->arch.vendor_name = "UNKNOWN VENDOR";
1266 	cpu->arch.feature[FEATURE_COMMON] = 0;
1267 	cpu->arch.feature[FEATURE_EXT] = 0;
1268 	cpu->arch.feature[FEATURE_EXT_AMD] = 0;
1269 	cpu->arch.feature[FEATURE_7_EBX] = 0;
1270 	cpu->arch.feature[FEATURE_7_ECX] = 0;
1271 	cpu->arch.feature[FEATURE_7_EDX] = 0;
1272 	cpu->arch.feature[FEATURE_D_1_EAX] = 0;
1273 	cpu->arch.model_name[0] = 0;
1274 
1275 	// print some fun data
1276 	get_current_cpuid(&cpuid, 0, 0);
1277 	uint32 maxBasicLeaf = cpuid.eax_0.max_eax;
1278 
1279 	// build the vendor string
1280 	memset(vendorString, 0, sizeof(vendorString));
1281 	memcpy(vendorString, cpuid.eax_0.vendor_id, sizeof(cpuid.eax_0.vendor_id));
1282 
1283 	// get the family, model, stepping
1284 	get_current_cpuid(&cpuid, 1, 0);
1285 	cpu->arch.type = cpuid.eax_1.type;
1286 	cpu->arch.family = cpuid.eax_1.family;
1287 	cpu->arch.extended_family = cpuid.eax_1.extended_family;
1288 	cpu->arch.model = cpuid.eax_1.model;
1289 	cpu->arch.extended_model = cpuid.eax_1.extended_model;
1290 	cpu->arch.stepping = cpuid.eax_1.stepping;
1291 	if (full) {
1292 		dprintf("CPU %d: type %d family %d extended_family %d model %d "
1293 			"extended_model %d stepping %d, string '%s'\n",
1294 			currentCPU, cpu->arch.type, cpu->arch.family,
1295 			cpu->arch.extended_family, cpu->arch.model,
1296 			cpu->arch.extended_model, cpu->arch.stepping, vendorString);
1297 	}
1298 
1299 	// figure out what vendor we have here
1300 
1301 	for (int32 i = 0; i < VENDOR_NUM; i++) {
1302 		if (vendor_info[i].ident_string[0]
1303 			&& !strcmp(vendorString, vendor_info[i].ident_string[0])) {
1304 			cpu->arch.vendor = (x86_vendors)i;
1305 			cpu->arch.vendor_name = vendor_info[i].vendor;
1306 			break;
1307 		}
1308 		if (vendor_info[i].ident_string[1]
1309 			&& !strcmp(vendorString, vendor_info[i].ident_string[1])) {
1310 			cpu->arch.vendor = (x86_vendors)i;
1311 			cpu->arch.vendor_name = vendor_info[i].vendor;
1312 			break;
1313 		}
1314 	}
1315 
1316 	// see if we can get the model name
1317 	get_current_cpuid(&cpuid, 0x80000000, 0);
1318 	uint32 maxExtendedLeaf = cpuid.eax_0.max_eax;
1319 	if (maxExtendedLeaf >= 0x80000004) {
1320 		// build the model string (need to swap ecx/edx data before copying)
1321 		unsigned int temp;
1322 		memset(cpu->arch.model_name, 0, sizeof(cpu->arch.model_name));
1323 
1324 		get_current_cpuid(&cpuid, 0x80000002, 0);
1325 		temp = cpuid.regs.edx;
1326 		cpuid.regs.edx = cpuid.regs.ecx;
1327 		cpuid.regs.ecx = temp;
1328 		memcpy(cpu->arch.model_name, cpuid.as_chars, sizeof(cpuid.as_chars));
1329 
1330 		get_current_cpuid(&cpuid, 0x80000003, 0);
1331 		temp = cpuid.regs.edx;
1332 		cpuid.regs.edx = cpuid.regs.ecx;
1333 		cpuid.regs.ecx = temp;
1334 		memcpy(cpu->arch.model_name + 16, cpuid.as_chars,
1335 			sizeof(cpuid.as_chars));
1336 
1337 		get_current_cpuid(&cpuid, 0x80000004, 0);
1338 		temp = cpuid.regs.edx;
1339 		cpuid.regs.edx = cpuid.regs.ecx;
1340 		cpuid.regs.ecx = temp;
1341 		memcpy(cpu->arch.model_name + 32, cpuid.as_chars,
1342 			sizeof(cpuid.as_chars));
1343 
1344 		// some cpus return a right-justified string
1345 		int32 i = 0;
1346 		while (cpu->arch.model_name[i] == ' ')
1347 			i++;
1348 		if (i > 0) {
1349 			memmove(cpu->arch.model_name, &cpu->arch.model_name[i],
1350 				strlen(&cpu->arch.model_name[i]) + 1);
1351 		}
1352 
1353 		if (full) {
1354 			dprintf("CPU %d: vendor '%s' model name '%s'\n",
1355 				currentCPU, cpu->arch.vendor_name, cpu->arch.model_name);
1356 		}
1357 	} else {
1358 		strlcpy(cpu->arch.model_name, "unknown", sizeof(cpu->arch.model_name));
1359 	}
1360 
1361 	// load feature bits
1362 	get_current_cpuid(&cpuid, 1, 0);
1363 	cpu->arch.feature[FEATURE_COMMON] = cpuid.eax_1.features; // edx
1364 	cpu->arch.feature[FEATURE_EXT] = cpuid.eax_1.extended_features; // ecx
1365 
1366 	if (!full)
1367 		return;
1368 
1369 	if (maxExtendedLeaf >= 0x80000001) {
1370 		get_current_cpuid(&cpuid, 0x80000001, 0);
1371 		if (cpu->arch.vendor == VENDOR_AMD)
1372 			cpu->arch.feature[FEATURE_EXT_AMD_ECX] = cpuid.regs.ecx; // ecx
1373 		cpu->arch.feature[FEATURE_EXT_AMD] = cpuid.regs.edx; // edx
1374 		if (cpu->arch.vendor != VENDOR_AMD)
1375 			cpu->arch.feature[FEATURE_EXT_AMD] &= IA32_FEATURES_INTEL_EXT;
1376 	}
1377 
1378 	if (maxBasicLeaf >= 5) {
1379 		get_current_cpuid(&cpuid, 5, 0);
1380 		cpu->arch.feature[FEATURE_5_ECX] = cpuid.regs.ecx;
1381 	}
1382 
1383 	if (maxBasicLeaf >= 6) {
1384 		get_current_cpuid(&cpuid, 6, 0);
1385 		cpu->arch.feature[FEATURE_6_EAX] = cpuid.regs.eax;
1386 		cpu->arch.feature[FEATURE_6_ECX] = cpuid.regs.ecx;
1387 	}
1388 
1389 	if (maxBasicLeaf >= 7) {
1390 		get_current_cpuid(&cpuid, 7, 0);
1391 		cpu->arch.feature[FEATURE_7_EBX] = cpuid.regs.ebx;
1392 		cpu->arch.feature[FEATURE_7_ECX] = cpuid.regs.ecx;
1393 		cpu->arch.feature[FEATURE_7_EDX] = cpuid.regs.edx;
1394 	}
1395 
1396 	if (maxBasicLeaf >= 0xd) {
1397 		get_current_cpuid(&cpuid, 0xd, 1);
1398 		cpu->arch.feature[FEATURE_D_1_EAX] = cpuid.regs.eax;
1399 	}
1400 
1401 	if (maxExtendedLeaf >= 0x80000007) {
1402 		get_current_cpuid(&cpuid, 0x80000007, 0);
1403 		cpu->arch.feature[FEATURE_EXT_7_EDX] = cpuid.regs.edx;
1404 	}
1405 
1406 	if (maxExtendedLeaf >= 0x80000008) {
1407 		get_current_cpuid(&cpuid, 0x80000008, 0);
1408 			cpu->arch.feature[FEATURE_EXT_8_EBX] = cpuid.regs.ebx;
1409 	}
1410 
1411 	detect_cpu_topology(currentCPU, cpu, maxBasicLeaf, maxExtendedLeaf);
1412 
1413 	if (cpu->arch.vendor == VENDOR_INTEL)
1414 		detect_intel_patch_level(cpu);
1415 	else if (cpu->arch.vendor == VENDOR_AMD)
1416 		detect_amd_patch_level(cpu);
1417 
1418 #if DUMP_FEATURE_STRING
1419 	dump_feature_string(currentCPU, cpu);
1420 #endif
1421 #if DUMP_CPU_PATCHLEVEL
1422 	dprintf("CPU %d: patch_level %" B_PRIx32 "\n", currentCPU,
1423 		cpu->arch.patch_level);
1424 #endif
1425 }
1426 
1427 
1428 bool
1429 x86_check_feature(uint32 feature, enum x86_feature_type type)
1430 {
1431 	cpu_ent* cpu = get_cpu_struct();
1432 
1433 #if 0
1434 	int i;
1435 	dprintf("x86_check_feature: feature 0x%x, type %d\n", feature, type);
1436 	for (i = 0; i < FEATURE_NUM; i++) {
1437 		dprintf("features %d: 0x%x\n", i, cpu->arch.feature[i]);
1438 	}
1439 #endif
1440 
1441 	return (cpu->arch.feature[type] & feature) != 0;
1442 }
1443 
1444 
1445 void*
1446 x86_get_double_fault_stack(int32 cpu, size_t* _size)
1447 {
1448 	*_size = kDoubleFaultStackSize;
1449 	return sDoubleFaultStacks + kDoubleFaultStackSize * cpu;
1450 }
1451 
1452 
1453 /*!	Returns the index of the current CPU. Can only be called from the double
1454 	fault handler.
1455 */
1456 int32
1457 x86_double_fault_get_cpu(void)
1458 {
1459 	addr_t stack = x86_get_stack_frame();
1460 	return (stack - (addr_t)sDoubleFaultStacks) / kDoubleFaultStackSize;
1461 }
1462 
1463 
1464 //	#pragma mark -
1465 
1466 
1467 status_t
1468 arch_cpu_preboot_init_percpu(kernel_args* args, int cpu)
1469 {
1470 	// On SMP system we want to synchronize the CPUs' TSCs, so system_time()
1471 	// will return consistent values.
1472 	if (smp_get_num_cpus() > 1) {
1473 		// let the first CPU prepare the rendezvous point
1474 		if (cpu == 0)
1475 			sTSCSyncRendezvous = smp_get_num_cpus() - 1;
1476 
1477 		// One CPU after the other will drop out of this loop and be caught by
1478 		// the loop below, until the last CPU (0) gets there. Save for +/- a few
1479 		// cycles the CPUs should pass the second loop at the same time.
1480 		while (sTSCSyncRendezvous != cpu) {
1481 		}
1482 
1483 		sTSCSyncRendezvous = cpu - 1;
1484 
1485 		while (sTSCSyncRendezvous != -1) {
1486 		}
1487 
1488 		// reset TSC to 0
1489 		x86_write_msr(IA32_MSR_TSC, 0);
1490 	}
1491 
1492 	x86_descriptors_preboot_init_percpu(args, cpu);
1493 
1494 	return B_OK;
1495 }
1496 
1497 
1498 static void
1499 halt_idle(void)
1500 {
1501 	asm("hlt");
1502 }
1503 
1504 
1505 static void
1506 amdc1e_noarat_idle(void)
1507 {
1508 	uint64 msr = x86_read_msr(K8_MSR_IPM);
1509 	if (msr & K8_CMPHALT)
1510 		x86_write_msr(K8_MSR_IPM, msr & ~K8_CMPHALT);
1511 	halt_idle();
1512 }
1513 
1514 
1515 static bool
1516 detect_amdc1e_noarat()
1517 {
1518 	cpu_ent* cpu = get_cpu_struct();
1519 
1520 	if (cpu->arch.vendor != VENDOR_AMD)
1521 		return false;
1522 
1523 	// Family 0x12 and higher processors support ARAT
1524 	// Family lower than 0xf processors doesn't support C1E
1525 	// Family 0xf with model <= 0x40 procssors doesn't support C1E
1526 	uint32 family = cpu->arch.family + cpu->arch.extended_family;
1527 	uint32 model = (cpu->arch.extended_model << 4) | cpu->arch.model;
1528 	return (family < 0x12 && family > 0xf) || (family == 0xf && model > 0x40);
1529 }
1530 
1531 
1532 static void
1533 init_tsc_with_cpuid(kernel_args* args, uint32* conversionFactor)
1534 {
1535 	cpu_ent* cpu = get_cpu_struct();
1536 	if (cpu->arch.vendor != VENDOR_INTEL)
1537 		return;
1538 	uint32 model = (cpu->arch.extended_model << 4) | cpu->arch.model;
1539 	cpuid_info cpuid;
1540 	get_current_cpuid(&cpuid, 0, 0);
1541 	uint32 maxBasicLeaf = cpuid.eax_0.max_eax;
1542 	if (maxBasicLeaf < 0x15)
1543 		return;
1544 
1545 	get_current_cpuid(&cpuid, 0x15, 0);
1546 	if (cpuid.regs.eax == 0 || cpuid.regs.ebx == 0)
1547 		return;
1548 	uint32 khz = cpuid.regs.ecx / 1000;
1549 	uint32 denominator = cpuid.regs.eax;
1550 	uint32 numerator = cpuid.regs.ebx;
1551 	if (khz == 0 && model == 0x5f) {
1552 		// CPUID 0x16 isn't supported, hardcoding
1553 		khz = 25000;
1554 	}
1555 
1556 	if (khz == 0 && maxBasicLeaf >= 0x16) {
1557 		// for these CPUs the base frequency is also the tsc frequency
1558 		get_current_cpuid(&cpuid, 0x16, 0);
1559 		khz = cpuid.regs.eax * 1000 * denominator / numerator;
1560 	}
1561 	if (khz == 0)
1562 		return;
1563 	dprintf("CPU: using TSC frequency from CPUID\n");
1564 	// compute for microseconds as follows (1000000 << 32) / (tsc freq in Hz),
1565 	// or (1000 << 32) / (tsc freq in kHz)
1566 	*conversionFactor = (1000ULL << 32) / (khz * numerator / denominator);
1567 	// overwrite the bootloader value
1568 	args->arch_args.system_time_cv_factor = *conversionFactor;
1569 }
1570 
1571 
1572 static void
1573 init_tsc_with_msr(kernel_args* args, uint32* conversionFactor)
1574 {
1575 	cpu_ent* cpuEnt = get_cpu_struct();
1576 	if (cpuEnt->arch.vendor != VENDOR_AMD)
1577 		return;
1578 	uint32 family = cpuEnt->arch.family + cpuEnt->arch.extended_family;
1579 	if (family < 0x10)
1580 		return;
1581 	uint64 value = x86_read_msr(MSR_F10H_HWCR);
1582 	if ((value & HWCR_TSCFREQSEL) == 0)
1583 		return;
1584 
1585 	value = x86_read_msr(MSR_F10H_PSTATEDEF(0));
1586 	if ((value & PSTATEDEF_EN) == 0)
1587 		return;
1588 	if (family != 0x17 && family != 0x19)
1589 		return;
1590 
1591 	uint64 khz = 200 * 1000;
1592 	uint32 denominator = (value >> 8) & 0x3f;
1593 	if (denominator < 0x8 || denominator > 0x2c)
1594 		return;
1595 	if (denominator > 0x1a && (denominator % 2) == 1)
1596 		return;
1597 	uint32 numerator = value & 0xff;
1598 	if (numerator < 0x10)
1599 		return;
1600 
1601 	dprintf("CPU: using TSC frequency from MSR %" B_PRIu64 "\n", khz * numerator / denominator);
1602 	// compute for microseconds as follows (1000000 << 32) / (tsc freq in Hz),
1603 	// or (1000 << 32) / (tsc freq in kHz)
1604 	*conversionFactor = (1000ULL << 32) / (khz * numerator / denominator);
1605 	// overwrite the bootloader value
1606 	args->arch_args.system_time_cv_factor = *conversionFactor;
1607 }
1608 
1609 
1610 static void
1611 init_tsc(kernel_args* args)
1612 {
1613 	// init the TSC -> system_time() conversion factors
1614 
1615 	// try to find the TSC frequency with CPUID
1616 	uint32 conversionFactor = args->arch_args.system_time_cv_factor;
1617 	init_tsc_with_cpuid(args, &conversionFactor);
1618 	init_tsc_with_msr(args, &conversionFactor);
1619 	uint64 conversionFactorNsecs = (uint64)conversionFactor * 1000;
1620 
1621 
1622 #ifdef __x86_64__
1623 	// The x86_64 system_time() implementation uses 64-bit multiplication and
1624 	// therefore shifting is not necessary for low frequencies (it's also not
1625 	// too likely that there'll be any x86_64 CPUs clocked under 1GHz).
1626 	__x86_setup_system_time((uint64)conversionFactor << 32,
1627 		conversionFactorNsecs);
1628 #else
1629 	if (conversionFactorNsecs >> 32 != 0) {
1630 		// the TSC frequency is < 1 GHz, which forces us to shift the factor
1631 		__x86_setup_system_time(conversionFactor, conversionFactorNsecs >> 16,
1632 			true);
1633 	} else {
1634 		// the TSC frequency is >= 1 GHz
1635 		__x86_setup_system_time(conversionFactor, conversionFactorNsecs, false);
1636 	}
1637 #endif
1638 }
1639 
1640 
1641 status_t
1642 arch_cpu_init_percpu(kernel_args* args, int cpu)
1643 {
1644 	detect_cpu(cpu, false);
1645 	load_microcode(cpu);
1646 	detect_cpu(cpu);
1647 
1648 	if (cpu == 0)
1649 		init_tsc(args);
1650 
1651 	if (!gCpuIdleFunc) {
1652 		if (detect_amdc1e_noarat())
1653 			gCpuIdleFunc = amdc1e_noarat_idle;
1654 		else
1655 			gCpuIdleFunc = halt_idle;
1656 	}
1657 
1658 	if (x86_check_feature(IA32_FEATURE_MCE, FEATURE_COMMON))
1659 		x86_write_cr4(x86_read_cr4() | IA32_CR4_MCE);
1660 
1661 #ifdef __x86_64__
1662 	// if RDTSCP is available write cpu number in TSC_AUX
1663 	if (x86_check_feature(IA32_FEATURE_AMD_EXT_RDTSCP, FEATURE_EXT_AMD))
1664 		x86_write_msr(IA32_MSR_TSC_AUX, cpu);
1665 
1666 	// make LFENCE a dispatch serializing instruction on AMD 64bit
1667 	cpu_ent* cpuEnt = get_cpu_struct();
1668 	if (cpuEnt->arch.vendor == VENDOR_AMD) {
1669 		uint32 family = cpuEnt->arch.family + cpuEnt->arch.extended_family;
1670 		if (family >= 0x10 && family != 0x11) {
1671 			uint64 value = x86_read_msr(MSR_F10H_DE_CFG);
1672 			if ((value & DE_CFG_SERIALIZE_LFENCE) == 0)
1673 				x86_write_msr(MSR_F10H_DE_CFG, value | DE_CFG_SERIALIZE_LFENCE);
1674 		}
1675 	}
1676 #endif
1677 
1678 	if (x86_check_feature(IA32_FEATURE_APERFMPERF, FEATURE_6_ECX)) {
1679 		gCPU[cpu].arch.mperf_prev = x86_read_msr(IA32_MSR_MPERF);
1680 		gCPU[cpu].arch.aperf_prev = x86_read_msr(IA32_MSR_APERF);
1681 		gCPU[cpu].arch.frequency = 0;
1682 		gCPU[cpu].arch.perf_timestamp = 0;
1683 	}
1684 	return __x86_patch_errata_percpu(cpu);
1685 }
1686 
1687 
1688 status_t
1689 arch_cpu_init(kernel_args* args)
1690 {
1691 	if (args->ucode_data != NULL
1692 		&& args->ucode_data_size > 0) {
1693 		sUcodeData = args->ucode_data;
1694 		sUcodeDataSize = args->ucode_data_size;
1695 	} else {
1696 		dprintf("CPU: no microcode provided\n");
1697 	}
1698 
1699 	// Initialize descriptor tables.
1700 	x86_descriptors_init(args);
1701 
1702 	return B_OK;
1703 }
1704 
1705 
1706 #ifdef __x86_64__
1707 static void
1708 enable_smap(void* dummy, int cpu)
1709 {
1710 	x86_write_cr4(x86_read_cr4() | IA32_CR4_SMAP);
1711 }
1712 
1713 
1714 static void
1715 enable_smep(void* dummy, int cpu)
1716 {
1717 	x86_write_cr4(x86_read_cr4() | IA32_CR4_SMEP);
1718 }
1719 
1720 
1721 static void
1722 enable_osxsave(void* dummy, int cpu)
1723 {
1724 	x86_write_cr4(x86_read_cr4() | IA32_CR4_OSXSAVE);
1725 }
1726 
1727 
1728 static void
1729 enable_xsavemask(void* dummy, int cpu)
1730 {
1731 	xsetbv(0, gXsaveMask);
1732 }
1733 #endif
1734 
1735 
1736 status_t
1737 arch_cpu_init_post_vm(kernel_args* args)
1738 {
1739 	uint32 i;
1740 
1741 	// allocate an area for the double fault stacks
1742 	virtual_address_restrictions virtualRestrictions = {};
1743 	virtualRestrictions.address_specification = B_ANY_KERNEL_ADDRESS;
1744 	physical_address_restrictions physicalRestrictions = {};
1745 	create_area_etc(B_SYSTEM_TEAM, "double fault stacks",
1746 		kDoubleFaultStackSize * smp_get_num_cpus(), B_FULL_LOCK,
1747 		B_KERNEL_READ_AREA | B_KERNEL_WRITE_AREA, CREATE_AREA_DONT_WAIT, 0,
1748 		&virtualRestrictions, &physicalRestrictions,
1749 		(void**)&sDoubleFaultStacks);
1750 
1751 	X86PagingStructures* kernelPagingStructures
1752 		= static_cast<X86VMTranslationMap*>(
1753 			VMAddressSpace::Kernel()->TranslationMap())->PagingStructures();
1754 
1755 	// Set active translation map on each CPU.
1756 	for (i = 0; i < args->num_cpus; i++) {
1757 		gCPU[i].arch.active_paging_structures = kernelPagingStructures;
1758 		kernelPagingStructures->AddReference();
1759 	}
1760 
1761 	if (!apic_available())
1762 		x86_init_fpu();
1763 	// else fpu gets set up in smp code
1764 
1765 #ifdef __x86_64__
1766 	// if available enable SMEP (Supervisor Memory Execution Protection)
1767 	if (x86_check_feature(IA32_FEATURE_SMEP, FEATURE_7_EBX)) {
1768 		if (!get_safemode_boolean(B_SAFEMODE_DISABLE_SMEP_SMAP, false)) {
1769 			dprintf("enable SMEP\n");
1770 			call_all_cpus_sync(&enable_smep, NULL);
1771 		} else
1772 			dprintf("SMEP disabled per safemode setting\n");
1773 	}
1774 
1775 	// if available enable SMAP (Supervisor Memory Access Protection)
1776 	if (x86_check_feature(IA32_FEATURE_SMAP, FEATURE_7_EBX)) {
1777 		if (!get_safemode_boolean(B_SAFEMODE_DISABLE_SMEP_SMAP, false)) {
1778 			dprintf("enable SMAP\n");
1779 			call_all_cpus_sync(&enable_smap, NULL);
1780 
1781 			arch_altcodepatch_replace(ALTCODEPATCH_TAG_STAC, &_stac, 3);
1782 			arch_altcodepatch_replace(ALTCODEPATCH_TAG_CLAC, &_clac, 3);
1783 		} else
1784 			dprintf("SMAP disabled per safemode setting\n");
1785 	}
1786 
1787 	// if available enable XSAVE (XSAVE and extended states)
1788 	gHasXsave = x86_check_feature(IA32_FEATURE_EXT_XSAVE, FEATURE_EXT);
1789 	if (gHasXsave) {
1790 		gHasXsavec = x86_check_feature(IA32_FEATURE_XSAVEC,
1791 			FEATURE_D_1_EAX);
1792 
1793 		call_all_cpus_sync(&enable_osxsave, NULL);
1794 		gXsaveMask = IA32_XCR0_X87 | IA32_XCR0_SSE;
1795 		cpuid_info cpuid;
1796 		get_current_cpuid(&cpuid, 0xd, 0);
1797 		gXsaveMask |= (cpuid.regs.eax & IA32_XCR0_AVX);
1798 		call_all_cpus_sync(&enable_xsavemask, NULL);
1799 		get_current_cpuid(&cpuid, 0xd, 0);
1800 		gFPUSaveLength = cpuid.regs.ebx;
1801 		if (gFPUSaveLength > sizeof(((struct arch_thread *)0)->fpu_state))
1802 			gFPUSaveLength = 832;
1803 
1804 		arch_altcodepatch_replace(ALTCODEPATCH_TAG_XSAVE,
1805 			gHasXsavec ? &_xsavec : &_xsave, 4);
1806 		arch_altcodepatch_replace(ALTCODEPATCH_TAG_XRSTOR,
1807 			&_xrstor, 4);
1808 
1809 		dprintf("enable %s 0x%" B_PRIx64 " %" B_PRId64 "\n",
1810 			gHasXsavec ? "XSAVEC" : "XSAVE", gXsaveMask, gFPUSaveLength);
1811 	}
1812 
1813 #endif
1814 
1815 	return B_OK;
1816 }
1817 
1818 
1819 status_t
1820 arch_cpu_init_post_modules(kernel_args* args)
1821 {
1822 	// initialize CPU module
1823 
1824 	void* cookie = open_module_list("cpu");
1825 
1826 	while (true) {
1827 		char name[B_FILE_NAME_LENGTH];
1828 		size_t nameLength = sizeof(name);
1829 
1830 		if (read_next_module_name(cookie, name, &nameLength) != B_OK
1831 			|| get_module(name, (module_info**)&sCpuModule) == B_OK)
1832 			break;
1833 	}
1834 
1835 	close_module_list(cookie);
1836 
1837 	// initialize MTRRs if available
1838 	if (x86_count_mtrrs() > 0) {
1839 		sCpuRendezvous = sCpuRendezvous2 = 0;
1840 		call_all_cpus(&init_mtrrs, NULL);
1841 	}
1842 
1843 	size_t threadExitLen = (addr_t)x86_end_userspace_thread_exit
1844 		- (addr_t)x86_userspace_thread_exit;
1845 	addr_t threadExitPosition = fill_commpage_entry(
1846 		COMMPAGE_ENTRY_X86_THREAD_EXIT, (const void*)x86_userspace_thread_exit,
1847 		threadExitLen);
1848 
1849 	// add the functions to the commpage image
1850 	image_id image = get_commpage_image();
1851 
1852 	elf_add_memory_image_symbol(image, "commpage_thread_exit",
1853 		threadExitPosition, threadExitLen, B_SYMBOL_TYPE_TEXT);
1854 
1855 	return B_OK;
1856 }
1857 
1858 
1859 void
1860 arch_cpu_user_TLB_invalidate(void)
1861 {
1862 	x86_write_cr3(x86_read_cr3());
1863 }
1864 
1865 
1866 void
1867 arch_cpu_global_TLB_invalidate(void)
1868 {
1869 	uint32 flags = x86_read_cr4();
1870 
1871 	if (flags & IA32_CR4_GLOBAL_PAGES) {
1872 		// disable and reenable the global pages to flush all TLBs regardless
1873 		// of the global page bit
1874 		x86_write_cr4(flags & ~IA32_CR4_GLOBAL_PAGES);
1875 		x86_write_cr4(flags | IA32_CR4_GLOBAL_PAGES);
1876 	} else {
1877 		cpu_status state = disable_interrupts();
1878 		arch_cpu_user_TLB_invalidate();
1879 		restore_interrupts(state);
1880 	}
1881 }
1882 
1883 
1884 void
1885 arch_cpu_invalidate_TLB_range(addr_t start, addr_t end)
1886 {
1887 	int32 num_pages = end / B_PAGE_SIZE - start / B_PAGE_SIZE;
1888 	while (num_pages-- >= 0) {
1889 		invalidate_TLB(start);
1890 		start += B_PAGE_SIZE;
1891 	}
1892 }
1893 
1894 
1895 void
1896 arch_cpu_invalidate_TLB_list(addr_t pages[], int num_pages)
1897 {
1898 	int i;
1899 	for (i = 0; i < num_pages; i++) {
1900 		invalidate_TLB(pages[i]);
1901 	}
1902 }
1903 
1904 
1905 status_t
1906 arch_cpu_shutdown(bool rebootSystem)
1907 {
1908 	if (acpi_shutdown(rebootSystem) == B_OK)
1909 		return B_OK;
1910 
1911 	if (!rebootSystem) {
1912 #ifndef __x86_64__
1913 		return apm_shutdown();
1914 #else
1915 		return B_NOT_SUPPORTED;
1916 #endif
1917 	}
1918 
1919 	cpu_status state = disable_interrupts();
1920 
1921 	// try to reset the system using the keyboard controller
1922 	out8(0xfe, 0x64);
1923 
1924 	// Give some time to the controller to do its job (0.5s)
1925 	snooze(500000);
1926 
1927 	// if that didn't help, try it this way
1928 	x86_reboot();
1929 
1930 	restore_interrupts(state);
1931 	return B_ERROR;
1932 }
1933 
1934 
1935 void
1936 arch_cpu_sync_icache(void* address, size_t length)
1937 {
1938 	// instruction cache is always consistent on x86
1939 }
1940 
1941