xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision f3539c12)
1 #include <linux/kernel.h>
2 
3 #include <linux/string.h>
4 #include <linux/bitops.h>
5 #include <linux/smp.h>
6 #include <linux/sched.h>
7 #include <linux/thread_info.h>
8 #include <linux/init.h>
9 #include <linux/uaccess.h>
10 
11 #include <asm/cpufeature.h>
12 #include <asm/pgtable.h>
13 #include <asm/msr.h>
14 #include <asm/bugs.h>
15 #include <asm/cpu.h>
16 #include <asm/intel-family.h>
17 
18 #ifdef CONFIG_X86_64
19 #include <linux/topology.h>
20 #endif
21 
22 #include "cpu.h"
23 
24 #ifdef CONFIG_X86_LOCAL_APIC
25 #include <asm/mpspec.h>
26 #include <asm/apic.h>
27 #endif
28 
29 /*
30  * Just in case our CPU detection goes bad, or you have a weird system,
31  * allow a way to override the automatic disabling of MPX.
32  */
33 static int forcempx;
34 
35 static int __init forcempx_setup(char *__unused)
36 {
37 	forcempx = 1;
38 
39 	return 1;
40 }
41 __setup("intel-skd-046-workaround=disable", forcempx_setup);
42 
43 void check_mpx_erratum(struct cpuinfo_x86 *c)
44 {
45 	if (forcempx)
46 		return;
47 	/*
48 	 * Turn off the MPX feature on CPUs where SMEP is not
49 	 * available or disabled.
50 	 *
51 	 * Works around Intel Erratum SKD046: "Branch Instructions
52 	 * May Initialize MPX Bound Registers Incorrectly".
53 	 *
54 	 * This might falsely disable MPX on systems without
55 	 * SMEP, like Atom processors without SMEP.  But there
56 	 * is no such hardware known at the moment.
57 	 */
58 	if (cpu_has(c, X86_FEATURE_MPX) && !cpu_has(c, X86_FEATURE_SMEP)) {
59 		setup_clear_cpu_cap(X86_FEATURE_MPX);
60 		pr_warn("x86/mpx: Disabling MPX since SMEP not present\n");
61 	}
62 }
63 
64 static void early_init_intel(struct cpuinfo_x86 *c)
65 {
66 	u64 misc_enable;
67 
68 	/* Unmask CPUID levels if masked: */
69 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
70 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
71 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
72 			c->cpuid_level = cpuid_eax(0);
73 			get_cpu_cap(c);
74 		}
75 	}
76 
77 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
78 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
79 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
80 
81 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) {
82 		unsigned lower_word;
83 
84 		wrmsr(MSR_IA32_UCODE_REV, 0, 0);
85 		/* Required by the SDM */
86 		sync_core();
87 		rdmsr(MSR_IA32_UCODE_REV, lower_word, c->microcode);
88 	}
89 
90 	/*
91 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
92 	 *
93 	 * A race condition between speculative fetches and invalidating
94 	 * a large page.  This is worked around in microcode, but we
95 	 * need the microcode to have already been loaded... so if it is
96 	 * not, recommend a BIOS update and disable large pages.
97 	 */
98 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 &&
99 	    c->microcode < 0x20e) {
100 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
101 		clear_cpu_cap(c, X86_FEATURE_PSE);
102 	}
103 
104 #ifdef CONFIG_X86_64
105 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
106 #else
107 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
108 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
109 		c->x86_cache_alignment = 128;
110 #endif
111 
112 	/* CPUID workaround for 0F33/0F34 CPU */
113 	if (c->x86 == 0xF && c->x86_model == 0x3
114 	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
115 		c->x86_phys_bits = 36;
116 
117 	/*
118 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
119 	 * with P/T states and does not stop in deep C-states.
120 	 *
121 	 * It is also reliable across cores and sockets. (but not across
122 	 * cabinets - we turn it off in that case explicitly.)
123 	 */
124 	if (c->x86_power & (1 << 8)) {
125 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
126 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
127 		if (!check_tsc_unstable())
128 			set_sched_clock_stable();
129 	}
130 
131 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
132 	if (c->x86 == 6) {
133 		switch (c->x86_model) {
134 		case 0x27:	/* Penwell */
135 		case 0x35:	/* Cloverview */
136 		case 0x4a:	/* Merrifield */
137 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
138 			break;
139 		default:
140 			break;
141 		}
142 	}
143 
144 	/*
145 	 * There is a known erratum on Pentium III and Core Solo
146 	 * and Core Duo CPUs.
147 	 * " Page with PAT set to WC while associated MTRR is UC
148 	 *   may consolidate to UC "
149 	 * Because of this erratum, it is better to stick with
150 	 * setting WC in MTRR rather than using PAT on these CPUs.
151 	 *
152 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
153 	 */
154 	if (c->x86 == 6 && c->x86_model < 15)
155 		clear_cpu_cap(c, X86_FEATURE_PAT);
156 
157 #ifdef CONFIG_KMEMCHECK
158 	/*
159 	 * P4s have a "fast strings" feature which causes single-
160 	 * stepping REP instructions to only generate a #DB on
161 	 * cache-line boundaries.
162 	 *
163 	 * Ingo Molnar reported a Pentium D (model 6) and a Xeon
164 	 * (model 2) with the same problem.
165 	 */
166 	if (c->x86 == 15)
167 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
168 				  MSR_IA32_MISC_ENABLE_FAST_STRING_BIT) > 0)
169 			pr_info("kmemcheck: Disabling fast string operations\n");
170 #endif
171 
172 	/*
173 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
174 	 * clear the fast string and enhanced fast string CPU capabilities.
175 	 */
176 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
177 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
178 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
179 			pr_info("Disabled fast string operations\n");
180 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
181 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
182 		}
183 	}
184 
185 	/*
186 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
187 	 * "The operating system also is required to invalidate (i.e., flush)
188 	 *  the TLB when any changes are made to any of the page table entries.
189 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
190 	 *
191 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
192 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
193 	 * to be modified.
194 	 */
195 	if (c->x86 == 5 && c->x86_model == 9) {
196 		pr_info("Disabling PGE capability bit\n");
197 		setup_clear_cpu_cap(X86_FEATURE_PGE);
198 	}
199 
200 	if (c->cpuid_level >= 0x00000001) {
201 		u32 eax, ebx, ecx, edx;
202 
203 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
204 		/*
205 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
206 		 * apicids which are reserved per package. Store the resulting
207 		 * shift value for the package management code.
208 		 */
209 		if (edx & (1U << 28))
210 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
211 	}
212 
213 	check_mpx_erratum(c);
214 }
215 
216 #ifdef CONFIG_X86_32
217 /*
218  *	Early probe support logic for ppro memory erratum #50
219  *
220  *	This is called before we do cpu ident work
221  */
222 
223 int ppro_with_ram_bug(void)
224 {
225 	/* Uses data from early_cpu_detect now */
226 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
227 	    boot_cpu_data.x86 == 6 &&
228 	    boot_cpu_data.x86_model == 1 &&
229 	    boot_cpu_data.x86_mask < 8) {
230 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
231 		return 1;
232 	}
233 	return 0;
234 }
235 
236 static void intel_smp_check(struct cpuinfo_x86 *c)
237 {
238 	/* calling is from identify_secondary_cpu() ? */
239 	if (!c->cpu_index)
240 		return;
241 
242 	/*
243 	 * Mask B, Pentium, but not Pentium MMX
244 	 */
245 	if (c->x86 == 5 &&
246 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
247 	    c->x86_model <= 3) {
248 		/*
249 		 * Remember we have B step Pentia with bugs
250 		 */
251 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
252 				    "with B stepping processors.\n");
253 	}
254 }
255 
256 static int forcepae;
257 static int __init forcepae_setup(char *__unused)
258 {
259 	forcepae = 1;
260 	return 1;
261 }
262 __setup("forcepae", forcepae_setup);
263 
264 static void intel_workarounds(struct cpuinfo_x86 *c)
265 {
266 #ifdef CONFIG_X86_F00F_BUG
267 	/*
268 	 * All models of Pentium and Pentium with MMX technology CPUs
269 	 * have the F0 0F bug, which lets nonprivileged users lock up the
270 	 * system. Announce that the fault handler will be checking for it.
271 	 * The Quark is also family 5, but does not have the same bug.
272 	 */
273 	clear_cpu_bug(c, X86_BUG_F00F);
274 	if (c->x86 == 5 && c->x86_model < 9) {
275 		static int f00f_workaround_enabled;
276 
277 		set_cpu_bug(c, X86_BUG_F00F);
278 		if (!f00f_workaround_enabled) {
279 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
280 			f00f_workaround_enabled = 1;
281 		}
282 	}
283 #endif
284 
285 	/*
286 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
287 	 * model 3 mask 3
288 	 */
289 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
290 		clear_cpu_cap(c, X86_FEATURE_SEP);
291 
292 	/*
293 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
294 	 * functionally usable PAE implementation.
295 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
296 	 */
297 	if (forcepae) {
298 		pr_warn("PAE forced!\n");
299 		set_cpu_cap(c, X86_FEATURE_PAE);
300 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
301 	}
302 
303 	/*
304 	 * P4 Xeon erratum 037 workaround.
305 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
306 	 */
307 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
308 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
309 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
310 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
311 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
312 		}
313 	}
314 
315 	/*
316 	 * See if we have a good local APIC by checking for buggy Pentia,
317 	 * i.e. all B steppings and the C2 stepping of P54C when using their
318 	 * integrated APIC (see 11AP erratum in "Pentium Processor
319 	 * Specification Update").
320 	 */
321 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
322 	    (c->x86_mask < 0x6 || c->x86_mask == 0xb))
323 		set_cpu_bug(c, X86_BUG_11AP);
324 
325 
326 #ifdef CONFIG_X86_INTEL_USERCOPY
327 	/*
328 	 * Set up the preferred alignment for movsl bulk memory moves
329 	 */
330 	switch (c->x86) {
331 	case 4:		/* 486: untested */
332 		break;
333 	case 5:		/* Old Pentia: untested */
334 		break;
335 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
336 		movsl_mask.mask = 7;
337 		break;
338 	case 15:	/* P4 is OK down to 8-byte alignment */
339 		movsl_mask.mask = 7;
340 		break;
341 	}
342 #endif
343 
344 	intel_smp_check(c);
345 }
346 #else
347 static void intel_workarounds(struct cpuinfo_x86 *c)
348 {
349 }
350 #endif
351 
352 static void srat_detect_node(struct cpuinfo_x86 *c)
353 {
354 #ifdef CONFIG_NUMA
355 	unsigned node;
356 	int cpu = smp_processor_id();
357 
358 	/* Don't do the funky fallback heuristics the AMD version employs
359 	   for now. */
360 	node = numa_cpu_node(cpu);
361 	if (node == NUMA_NO_NODE || !node_online(node)) {
362 		/* reuse the value from init_cpu_to_node() */
363 		node = cpu_to_node(cpu);
364 	}
365 	numa_set_node(cpu, node);
366 #endif
367 }
368 
369 /*
370  * find out the number of processor cores on the die
371  */
372 static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
373 {
374 	unsigned int eax, ebx, ecx, edx;
375 
376 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
377 		return 1;
378 
379 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
380 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
381 	if (eax & 0x1f)
382 		return (eax >> 26) + 1;
383 	else
384 		return 1;
385 }
386 
387 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
388 {
389 	/* Intel VMX MSR indicated features */
390 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
391 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
392 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
393 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
394 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
395 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
396 
397 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
398 
399 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
400 	clear_cpu_cap(c, X86_FEATURE_VNMI);
401 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
402 	clear_cpu_cap(c, X86_FEATURE_EPT);
403 	clear_cpu_cap(c, X86_FEATURE_VPID);
404 
405 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
406 	msr_ctl = vmx_msr_high | vmx_msr_low;
407 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
408 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
409 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
410 		set_cpu_cap(c, X86_FEATURE_VNMI);
411 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
412 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
413 		      vmx_msr_low, vmx_msr_high);
414 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
415 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
416 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
417 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
418 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
419 			set_cpu_cap(c, X86_FEATURE_EPT);
420 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
421 			set_cpu_cap(c, X86_FEATURE_VPID);
422 	}
423 }
424 
425 static void init_intel_energy_perf(struct cpuinfo_x86 *c)
426 {
427 	u64 epb;
428 
429 	/*
430 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
431 	 * (x86_energy_perf_policy(8) is available to change it at run-time.)
432 	 */
433 	if (!cpu_has(c, X86_FEATURE_EPB))
434 		return;
435 
436 	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
437 	if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
438 		return;
439 
440 	pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
441 	pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
442 	epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
443 	wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
444 }
445 
446 static void intel_bsp_resume(struct cpuinfo_x86 *c)
447 {
448 	/*
449 	 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
450 	 * so reinitialize it properly like during bootup:
451 	 */
452 	init_intel_energy_perf(c);
453 }
454 
455 static void init_intel(struct cpuinfo_x86 *c)
456 {
457 	unsigned int l2 = 0;
458 
459 	early_init_intel(c);
460 
461 	intel_workarounds(c);
462 
463 	/*
464 	 * Detect the extended topology information if available. This
465 	 * will reinitialise the initial_apicid which will be used
466 	 * in init_intel_cacheinfo()
467 	 */
468 	detect_extended_topology(c);
469 
470 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
471 		/*
472 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
473 		 * detection.
474 		 */
475 		c->x86_max_cores = intel_num_cpu_cores(c);
476 #ifdef CONFIG_X86_32
477 		detect_ht(c);
478 #endif
479 	}
480 
481 	l2 = init_intel_cacheinfo(c);
482 
483 	/* Detect legacy cache sizes if init_intel_cacheinfo did not */
484 	if (l2 == 0) {
485 		cpu_detect_cache_sizes(c);
486 		l2 = c->x86_cache_size;
487 	}
488 
489 	if (c->cpuid_level > 9) {
490 		unsigned eax = cpuid_eax(10);
491 		/* Check for version and the number of counters */
492 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
493 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
494 	}
495 
496 	if (cpu_has(c, X86_FEATURE_XMM2))
497 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
498 
499 	if (boot_cpu_has(X86_FEATURE_DS)) {
500 		unsigned int l1;
501 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
502 		if (!(l1 & (1<<11)))
503 			set_cpu_cap(c, X86_FEATURE_BTS);
504 		if (!(l1 & (1<<12)))
505 			set_cpu_cap(c, X86_FEATURE_PEBS);
506 	}
507 
508 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
509 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
510 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
511 
512 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
513 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
514 		set_cpu_bug(c, X86_BUG_MONITOR);
515 
516 #ifdef CONFIG_X86_64
517 	if (c->x86 == 15)
518 		c->x86_cache_alignment = c->x86_clflush_size * 2;
519 	if (c->x86 == 6)
520 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
521 #else
522 	/*
523 	 * Names for the Pentium II/Celeron processors
524 	 * detectable only by also checking the cache size.
525 	 * Dixon is NOT a Celeron.
526 	 */
527 	if (c->x86 == 6) {
528 		char *p = NULL;
529 
530 		switch (c->x86_model) {
531 		case 5:
532 			if (l2 == 0)
533 				p = "Celeron (Covington)";
534 			else if (l2 == 256)
535 				p = "Mobile Pentium II (Dixon)";
536 			break;
537 
538 		case 6:
539 			if (l2 == 128)
540 				p = "Celeron (Mendocino)";
541 			else if (c->x86_mask == 0 || c->x86_mask == 5)
542 				p = "Celeron-A";
543 			break;
544 
545 		case 8:
546 			if (l2 == 128)
547 				p = "Celeron (Coppermine)";
548 			break;
549 		}
550 
551 		if (p)
552 			strcpy(c->x86_model_id, p);
553 	}
554 
555 	if (c->x86 == 15)
556 		set_cpu_cap(c, X86_FEATURE_P4);
557 	if (c->x86 == 6)
558 		set_cpu_cap(c, X86_FEATURE_P3);
559 #endif
560 
561 	/* Work around errata */
562 	srat_detect_node(c);
563 
564 	if (cpu_has(c, X86_FEATURE_VMX))
565 		detect_vmx_virtcap(c);
566 
567 	init_intel_energy_perf(c);
568 }
569 
570 #ifdef CONFIG_X86_32
571 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
572 {
573 	/*
574 	 * Intel PIII Tualatin. This comes in two flavours.
575 	 * One has 256kb of cache, the other 512. We have no way
576 	 * to determine which, so we use a boottime override
577 	 * for the 512kb model, and assume 256 otherwise.
578 	 */
579 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
580 		size = 256;
581 
582 	/*
583 	 * Intel Quark SoC X1000 contains a 4-way set associative
584 	 * 16K cache with a 16 byte cache line and 256 lines per tag
585 	 */
586 	if ((c->x86 == 5) && (c->x86_model == 9))
587 		size = 16;
588 	return size;
589 }
590 #endif
591 
592 #define TLB_INST_4K	0x01
593 #define TLB_INST_4M	0x02
594 #define TLB_INST_2M_4M	0x03
595 
596 #define TLB_INST_ALL	0x05
597 #define TLB_INST_1G	0x06
598 
599 #define TLB_DATA_4K	0x11
600 #define TLB_DATA_4M	0x12
601 #define TLB_DATA_2M_4M	0x13
602 #define TLB_DATA_4K_4M	0x14
603 
604 #define TLB_DATA_1G	0x16
605 
606 #define TLB_DATA0_4K	0x21
607 #define TLB_DATA0_4M	0x22
608 #define TLB_DATA0_2M_4M	0x23
609 
610 #define STLB_4K		0x41
611 #define STLB_4K_2M	0x42
612 
613 static const struct _tlb_table intel_tlb_table[] = {
614 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
615 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
616 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
617 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
618 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
619 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
620 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
621 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
622 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
623 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
624 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
625 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
626 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
627 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
628 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
629 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
630 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
631 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
632 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
633 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
634 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
635 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
636 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
637 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
638 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
639 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
640 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
641 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
642 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
643 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
644 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
645 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
646 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
647 	{ 0x00, 0, 0 }
648 };
649 
650 static void intel_tlb_lookup(const unsigned char desc)
651 {
652 	unsigned char k;
653 	if (desc == 0)
654 		return;
655 
656 	/* look up this descriptor in the table */
657 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
658 			intel_tlb_table[k].descriptor != 0; k++)
659 		;
660 
661 	if (intel_tlb_table[k].tlb_type == 0)
662 		return;
663 
664 	switch (intel_tlb_table[k].tlb_type) {
665 	case STLB_4K:
666 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
667 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
668 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
669 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
670 		break;
671 	case STLB_4K_2M:
672 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
673 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
674 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
675 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
676 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
677 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
678 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
679 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
680 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
681 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
682 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
683 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
684 		break;
685 	case TLB_INST_ALL:
686 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
687 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
688 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
689 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
690 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
691 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
692 		break;
693 	case TLB_INST_4K:
694 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
695 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
696 		break;
697 	case TLB_INST_4M:
698 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
699 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
700 		break;
701 	case TLB_INST_2M_4M:
702 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
703 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
704 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
705 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
706 		break;
707 	case TLB_DATA_4K:
708 	case TLB_DATA0_4K:
709 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
710 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
711 		break;
712 	case TLB_DATA_4M:
713 	case TLB_DATA0_4M:
714 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
715 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
716 		break;
717 	case TLB_DATA_2M_4M:
718 	case TLB_DATA0_2M_4M:
719 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
720 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
721 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
722 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
723 		break;
724 	case TLB_DATA_4K_4M:
725 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
726 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
727 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
728 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
729 		break;
730 	case TLB_DATA_1G:
731 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
732 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
733 		break;
734 	}
735 }
736 
737 static void intel_detect_tlb(struct cpuinfo_x86 *c)
738 {
739 	int i, j, n;
740 	unsigned int regs[4];
741 	unsigned char *desc = (unsigned char *)regs;
742 
743 	if (c->cpuid_level < 2)
744 		return;
745 
746 	/* Number of times to iterate */
747 	n = cpuid_eax(2) & 0xFF;
748 
749 	for (i = 0 ; i < n ; i++) {
750 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
751 
752 		/* If bit 31 is set, this is an unknown format */
753 		for (j = 0 ; j < 3 ; j++)
754 			if (regs[j] & (1 << 31))
755 				regs[j] = 0;
756 
757 		/* Byte 0 is level count, not a descriptor */
758 		for (j = 1 ; j < 16 ; j++)
759 			intel_tlb_lookup(desc[j]);
760 	}
761 }
762 
763 static const struct cpu_dev intel_cpu_dev = {
764 	.c_vendor	= "Intel",
765 	.c_ident	= { "GenuineIntel" },
766 #ifdef CONFIG_X86_32
767 	.legacy_models = {
768 		{ .family = 4, .model_names =
769 		  {
770 			  [0] = "486 DX-25/33",
771 			  [1] = "486 DX-50",
772 			  [2] = "486 SX",
773 			  [3] = "486 DX/2",
774 			  [4] = "486 SL",
775 			  [5] = "486 SX/2",
776 			  [7] = "486 DX/2-WB",
777 			  [8] = "486 DX/4",
778 			  [9] = "486 DX/4-WB"
779 		  }
780 		},
781 		{ .family = 5, .model_names =
782 		  {
783 			  [0] = "Pentium 60/66 A-step",
784 			  [1] = "Pentium 60/66",
785 			  [2] = "Pentium 75 - 200",
786 			  [3] = "OverDrive PODP5V83",
787 			  [4] = "Pentium MMX",
788 			  [7] = "Mobile Pentium 75 - 200",
789 			  [8] = "Mobile Pentium MMX",
790 			  [9] = "Quark SoC X1000",
791 		  }
792 		},
793 		{ .family = 6, .model_names =
794 		  {
795 			  [0] = "Pentium Pro A-step",
796 			  [1] = "Pentium Pro",
797 			  [3] = "Pentium II (Klamath)",
798 			  [4] = "Pentium II (Deschutes)",
799 			  [5] = "Pentium II (Deschutes)",
800 			  [6] = "Mobile Pentium II",
801 			  [7] = "Pentium III (Katmai)",
802 			  [8] = "Pentium III (Coppermine)",
803 			  [10] = "Pentium III (Cascades)",
804 			  [11] = "Pentium III (Tualatin)",
805 		  }
806 		},
807 		{ .family = 15, .model_names =
808 		  {
809 			  [0] = "Pentium 4 (Unknown)",
810 			  [1] = "Pentium 4 (Willamette)",
811 			  [2] = "Pentium 4 (Northwood)",
812 			  [4] = "Pentium 4 (Foster)",
813 			  [5] = "Pentium 4 (Foster)",
814 		  }
815 		},
816 	},
817 	.legacy_cache_size = intel_size_cache,
818 #endif
819 	.c_detect_tlb	= intel_detect_tlb,
820 	.c_early_init   = early_init_intel,
821 	.c_init		= init_intel,
822 	.c_bsp_resume	= intel_bsp_resume,
823 	.c_x86_vendor	= X86_VENDOR_INTEL,
824 };
825 
826 cpu_dev_register(intel_cpu_dev);
827 
828