xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 23c2b932)
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/module.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 
17 #ifdef CONFIG_X86_64
18 #include <linux/topology.h>
19 #endif
20 
21 #include "cpu.h"
22 
23 #ifdef CONFIG_X86_LOCAL_APIC
24 #include <asm/mpspec.h>
25 #include <asm/apic.h>
26 #endif
27 
28 /*
29  * Just in case our CPU detection goes bad, or you have a weird system,
30  * allow a way to override the automatic disabling of MPX.
31  */
32 static int forcempx;
33 
34 static int __init forcempx_setup(char *__unused)
35 {
36 	forcempx = 1;
37 
38 	return 1;
39 }
40 __setup("intel-skd-046-workaround=disable", forcempx_setup);
41 
42 void check_mpx_erratum(struct cpuinfo_x86 *c)
43 {
44 	if (forcempx)
45 		return;
46 	/*
47 	 * Turn off the MPX feature on CPUs where SMEP is not
48 	 * available or disabled.
49 	 *
50 	 * Works around Intel Erratum SKD046: "Branch Instructions
51 	 * May Initialize MPX Bound Registers Incorrectly".
52 	 *
53 	 * This might falsely disable MPX on systems without
54 	 * SMEP, like Atom processors without SMEP.  But there
55 	 * is no such hardware known at the moment.
56 	 */
57 	if (cpu_has(c, X86_FEATURE_MPX) && !cpu_has(c, X86_FEATURE_SMEP)) {
58 		setup_clear_cpu_cap(X86_FEATURE_MPX);
59 		pr_warn("x86/mpx: Disabling MPX since SMEP not present\n");
60 	}
61 }
62 
63 static void early_init_intel(struct cpuinfo_x86 *c)
64 {
65 	u64 misc_enable;
66 
67 	/* Unmask CPUID levels if masked: */
68 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
69 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
70 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
71 			c->cpuid_level = cpuid_eax(0);
72 			get_cpu_cap(c);
73 		}
74 	}
75 
76 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
77 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
78 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
79 
80 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) {
81 		unsigned lower_word;
82 
83 		wrmsr(MSR_IA32_UCODE_REV, 0, 0);
84 		/* Required by the SDM */
85 		sync_core();
86 		rdmsr(MSR_IA32_UCODE_REV, lower_word, c->microcode);
87 	}
88 
89 	/*
90 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
91 	 *
92 	 * A race condition between speculative fetches and invalidating
93 	 * a large page.  This is worked around in microcode, but we
94 	 * need the microcode to have already been loaded... so if it is
95 	 * not, recommend a BIOS update and disable large pages.
96 	 */
97 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 &&
98 	    c->microcode < 0x20e) {
99 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
100 		clear_cpu_cap(c, X86_FEATURE_PSE);
101 	}
102 
103 #ifdef CONFIG_X86_64
104 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
105 #else
106 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
107 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
108 		c->x86_cache_alignment = 128;
109 #endif
110 
111 	/* CPUID workaround for 0F33/0F34 CPU */
112 	if (c->x86 == 0xF && c->x86_model == 0x3
113 	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
114 		c->x86_phys_bits = 36;
115 
116 	/*
117 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
118 	 * with P/T states and does not stop in deep C-states.
119 	 *
120 	 * It is also reliable across cores and sockets. (but not across
121 	 * cabinets - we turn it off in that case explicitly.)
122 	 */
123 	if (c->x86_power & (1 << 8)) {
124 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
125 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
126 		if (!check_tsc_unstable())
127 			set_sched_clock_stable();
128 	}
129 
130 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
131 	if (c->x86 == 6) {
132 		switch (c->x86_model) {
133 		case 0x27:	/* Penwell */
134 		case 0x35:	/* Cloverview */
135 		case 0x4a:	/* Merrifield */
136 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
137 			break;
138 		default:
139 			break;
140 		}
141 	}
142 
143 	/*
144 	 * There is a known erratum on Pentium III and Core Solo
145 	 * and Core Duo CPUs.
146 	 * " Page with PAT set to WC while associated MTRR is UC
147 	 *   may consolidate to UC "
148 	 * Because of this erratum, it is better to stick with
149 	 * setting WC in MTRR rather than using PAT on these CPUs.
150 	 *
151 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
152 	 */
153 	if (c->x86 == 6 && c->x86_model < 15)
154 		clear_cpu_cap(c, X86_FEATURE_PAT);
155 
156 #ifdef CONFIG_KMEMCHECK
157 	/*
158 	 * P4s have a "fast strings" feature which causes single-
159 	 * stepping REP instructions to only generate a #DB on
160 	 * cache-line boundaries.
161 	 *
162 	 * Ingo Molnar reported a Pentium D (model 6) and a Xeon
163 	 * (model 2) with the same problem.
164 	 */
165 	if (c->x86 == 15)
166 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
167 				  MSR_IA32_MISC_ENABLE_FAST_STRING_BIT) > 0)
168 			pr_info("kmemcheck: Disabling fast string operations\n");
169 #endif
170 
171 	/*
172 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
173 	 * clear the fast string and enhanced fast string CPU capabilities.
174 	 */
175 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
176 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
177 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
178 			pr_info("Disabled fast string operations\n");
179 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
180 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
181 		}
182 	}
183 
184 	/*
185 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
186 	 * "The operating system also is required to invalidate (i.e., flush)
187 	 *  the TLB when any changes are made to any of the page table entries.
188 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
189 	 *
190 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
191 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
192 	 * to be modified.
193 	 */
194 	if (c->x86 == 5 && c->x86_model == 9) {
195 		pr_info("Disabling PGE capability bit\n");
196 		setup_clear_cpu_cap(X86_FEATURE_PGE);
197 	}
198 
199 	if (c->cpuid_level >= 0x00000001) {
200 		u32 eax, ebx, ecx, edx;
201 
202 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
203 		/*
204 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
205 		 * apicids which are reserved per package. Store the resulting
206 		 * shift value for the package management code.
207 		 */
208 		if (edx & (1U << 28))
209 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
210 	}
211 
212 	check_mpx_erratum(c);
213 }
214 
215 #ifdef CONFIG_X86_32
216 /*
217  *	Early probe support logic for ppro memory erratum #50
218  *
219  *	This is called before we do cpu ident work
220  */
221 
222 int ppro_with_ram_bug(void)
223 {
224 	/* Uses data from early_cpu_detect now */
225 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
226 	    boot_cpu_data.x86 == 6 &&
227 	    boot_cpu_data.x86_model == 1 &&
228 	    boot_cpu_data.x86_mask < 8) {
229 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
230 		return 1;
231 	}
232 	return 0;
233 }
234 
235 static void intel_smp_check(struct cpuinfo_x86 *c)
236 {
237 	/* calling is from identify_secondary_cpu() ? */
238 	if (!c->cpu_index)
239 		return;
240 
241 	/*
242 	 * Mask B, Pentium, but not Pentium MMX
243 	 */
244 	if (c->x86 == 5 &&
245 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
246 	    c->x86_model <= 3) {
247 		/*
248 		 * Remember we have B step Pentia with bugs
249 		 */
250 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
251 				    "with B stepping processors.\n");
252 	}
253 }
254 
255 static int forcepae;
256 static int __init forcepae_setup(char *__unused)
257 {
258 	forcepae = 1;
259 	return 1;
260 }
261 __setup("forcepae", forcepae_setup);
262 
263 static void intel_workarounds(struct cpuinfo_x86 *c)
264 {
265 #ifdef CONFIG_X86_F00F_BUG
266 	/*
267 	 * All models of Pentium and Pentium with MMX technology CPUs
268 	 * have the F0 0F bug, which lets nonprivileged users lock up the
269 	 * system. Announce that the fault handler will be checking for it.
270 	 * The Quark is also family 5, but does not have the same bug.
271 	 */
272 	clear_cpu_bug(c, X86_BUG_F00F);
273 	if (c->x86 == 5 && c->x86_model < 9) {
274 		static int f00f_workaround_enabled;
275 
276 		set_cpu_bug(c, X86_BUG_F00F);
277 		if (!f00f_workaround_enabled) {
278 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
279 			f00f_workaround_enabled = 1;
280 		}
281 	}
282 #endif
283 
284 	/*
285 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
286 	 * model 3 mask 3
287 	 */
288 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
289 		clear_cpu_cap(c, X86_FEATURE_SEP);
290 
291 	/*
292 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
293 	 * functionally usable PAE implementation.
294 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
295 	 */
296 	if (forcepae) {
297 		pr_warn("PAE forced!\n");
298 		set_cpu_cap(c, X86_FEATURE_PAE);
299 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
300 	}
301 
302 	/*
303 	 * P4 Xeon errata 037 workaround.
304 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
305 	 */
306 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
307 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
308 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT)
309 		    > 0) {
310 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
311 			pr_info("CPU: Disabling hardware prefetching (Errata 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 #ifdef CONFIG_X86_64
513 	if (c->x86 == 15)
514 		c->x86_cache_alignment = c->x86_clflush_size * 2;
515 	if (c->x86 == 6)
516 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
517 #else
518 	/*
519 	 * Names for the Pentium II/Celeron processors
520 	 * detectable only by also checking the cache size.
521 	 * Dixon is NOT a Celeron.
522 	 */
523 	if (c->x86 == 6) {
524 		char *p = NULL;
525 
526 		switch (c->x86_model) {
527 		case 5:
528 			if (l2 == 0)
529 				p = "Celeron (Covington)";
530 			else if (l2 == 256)
531 				p = "Mobile Pentium II (Dixon)";
532 			break;
533 
534 		case 6:
535 			if (l2 == 128)
536 				p = "Celeron (Mendocino)";
537 			else if (c->x86_mask == 0 || c->x86_mask == 5)
538 				p = "Celeron-A";
539 			break;
540 
541 		case 8:
542 			if (l2 == 128)
543 				p = "Celeron (Coppermine)";
544 			break;
545 		}
546 
547 		if (p)
548 			strcpy(c->x86_model_id, p);
549 	}
550 
551 	if (c->x86 == 15)
552 		set_cpu_cap(c, X86_FEATURE_P4);
553 	if (c->x86 == 6)
554 		set_cpu_cap(c, X86_FEATURE_P3);
555 #endif
556 
557 	/* Work around errata */
558 	srat_detect_node(c);
559 
560 	if (cpu_has(c, X86_FEATURE_VMX))
561 		detect_vmx_virtcap(c);
562 
563 	init_intel_energy_perf(c);
564 }
565 
566 #ifdef CONFIG_X86_32
567 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
568 {
569 	/*
570 	 * Intel PIII Tualatin. This comes in two flavours.
571 	 * One has 256kb of cache, the other 512. We have no way
572 	 * to determine which, so we use a boottime override
573 	 * for the 512kb model, and assume 256 otherwise.
574 	 */
575 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
576 		size = 256;
577 
578 	/*
579 	 * Intel Quark SoC X1000 contains a 4-way set associative
580 	 * 16K cache with a 16 byte cache line and 256 lines per tag
581 	 */
582 	if ((c->x86 == 5) && (c->x86_model == 9))
583 		size = 16;
584 	return size;
585 }
586 #endif
587 
588 #define TLB_INST_4K	0x01
589 #define TLB_INST_4M	0x02
590 #define TLB_INST_2M_4M	0x03
591 
592 #define TLB_INST_ALL	0x05
593 #define TLB_INST_1G	0x06
594 
595 #define TLB_DATA_4K	0x11
596 #define TLB_DATA_4M	0x12
597 #define TLB_DATA_2M_4M	0x13
598 #define TLB_DATA_4K_4M	0x14
599 
600 #define TLB_DATA_1G	0x16
601 
602 #define TLB_DATA0_4K	0x21
603 #define TLB_DATA0_4M	0x22
604 #define TLB_DATA0_2M_4M	0x23
605 
606 #define STLB_4K		0x41
607 #define STLB_4K_2M	0x42
608 
609 static const struct _tlb_table intel_tlb_table[] = {
610 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
611 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
612 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
613 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
614 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
615 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
616 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
617 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
618 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
619 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
620 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
621 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
622 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
623 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
624 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
625 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
626 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
627 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
628 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
629 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
630 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
631 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
632 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
633 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
634 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
635 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
636 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
637 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
638 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
639 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
640 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
641 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
642 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
643 	{ 0x00, 0, 0 }
644 };
645 
646 static void intel_tlb_lookup(const unsigned char desc)
647 {
648 	unsigned char k;
649 	if (desc == 0)
650 		return;
651 
652 	/* look up this descriptor in the table */
653 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
654 			intel_tlb_table[k].descriptor != 0; k++)
655 		;
656 
657 	if (intel_tlb_table[k].tlb_type == 0)
658 		return;
659 
660 	switch (intel_tlb_table[k].tlb_type) {
661 	case STLB_4K:
662 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
663 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
664 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
665 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
666 		break;
667 	case STLB_4K_2M:
668 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
669 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
670 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
671 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
672 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
673 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
674 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
675 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
676 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
677 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
678 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
679 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
680 		break;
681 	case TLB_INST_ALL:
682 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
683 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
684 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
685 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
686 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
687 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
688 		break;
689 	case TLB_INST_4K:
690 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
691 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
692 		break;
693 	case TLB_INST_4M:
694 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
695 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
696 		break;
697 	case TLB_INST_2M_4M:
698 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
699 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
700 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
701 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
702 		break;
703 	case TLB_DATA_4K:
704 	case TLB_DATA0_4K:
705 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
706 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
707 		break;
708 	case TLB_DATA_4M:
709 	case TLB_DATA0_4M:
710 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
711 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
712 		break;
713 	case TLB_DATA_2M_4M:
714 	case TLB_DATA0_2M_4M:
715 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
716 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
717 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
718 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
719 		break;
720 	case TLB_DATA_4K_4M:
721 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
722 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
723 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
724 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
725 		break;
726 	case TLB_DATA_1G:
727 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
728 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
729 		break;
730 	}
731 }
732 
733 static void intel_detect_tlb(struct cpuinfo_x86 *c)
734 {
735 	int i, j, n;
736 	unsigned int regs[4];
737 	unsigned char *desc = (unsigned char *)regs;
738 
739 	if (c->cpuid_level < 2)
740 		return;
741 
742 	/* Number of times to iterate */
743 	n = cpuid_eax(2) & 0xFF;
744 
745 	for (i = 0 ; i < n ; i++) {
746 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
747 
748 		/* If bit 31 is set, this is an unknown format */
749 		for (j = 0 ; j < 3 ; j++)
750 			if (regs[j] & (1 << 31))
751 				regs[j] = 0;
752 
753 		/* Byte 0 is level count, not a descriptor */
754 		for (j = 1 ; j < 16 ; j++)
755 			intel_tlb_lookup(desc[j]);
756 	}
757 }
758 
759 static const struct cpu_dev intel_cpu_dev = {
760 	.c_vendor	= "Intel",
761 	.c_ident	= { "GenuineIntel" },
762 #ifdef CONFIG_X86_32
763 	.legacy_models = {
764 		{ .family = 4, .model_names =
765 		  {
766 			  [0] = "486 DX-25/33",
767 			  [1] = "486 DX-50",
768 			  [2] = "486 SX",
769 			  [3] = "486 DX/2",
770 			  [4] = "486 SL",
771 			  [5] = "486 SX/2",
772 			  [7] = "486 DX/2-WB",
773 			  [8] = "486 DX/4",
774 			  [9] = "486 DX/4-WB"
775 		  }
776 		},
777 		{ .family = 5, .model_names =
778 		  {
779 			  [0] = "Pentium 60/66 A-step",
780 			  [1] = "Pentium 60/66",
781 			  [2] = "Pentium 75 - 200",
782 			  [3] = "OverDrive PODP5V83",
783 			  [4] = "Pentium MMX",
784 			  [7] = "Mobile Pentium 75 - 200",
785 			  [8] = "Mobile Pentium MMX",
786 			  [9] = "Quark SoC X1000",
787 		  }
788 		},
789 		{ .family = 6, .model_names =
790 		  {
791 			  [0] = "Pentium Pro A-step",
792 			  [1] = "Pentium Pro",
793 			  [3] = "Pentium II (Klamath)",
794 			  [4] = "Pentium II (Deschutes)",
795 			  [5] = "Pentium II (Deschutes)",
796 			  [6] = "Mobile Pentium II",
797 			  [7] = "Pentium III (Katmai)",
798 			  [8] = "Pentium III (Coppermine)",
799 			  [10] = "Pentium III (Cascades)",
800 			  [11] = "Pentium III (Tualatin)",
801 		  }
802 		},
803 		{ .family = 15, .model_names =
804 		  {
805 			  [0] = "Pentium 4 (Unknown)",
806 			  [1] = "Pentium 4 (Willamette)",
807 			  [2] = "Pentium 4 (Northwood)",
808 			  [4] = "Pentium 4 (Foster)",
809 			  [5] = "Pentium 4 (Foster)",
810 		  }
811 		},
812 	},
813 	.legacy_cache_size = intel_size_cache,
814 #endif
815 	.c_detect_tlb	= intel_detect_tlb,
816 	.c_early_init   = early_init_intel,
817 	.c_init		= init_intel,
818 	.c_bsp_resume	= intel_bsp_resume,
819 	.c_x86_vendor	= X86_VENDOR_INTEL,
820 };
821 
822 cpu_dev_register(intel_cpu_dev);
823 
824