xref: /linux/arch/x86/kernel/cpu/intel.c (revision 00a6d7b6762c27d441e9ac8faff36384bc0fc180)
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/processor.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 static void early_init_intel(struct cpuinfo_x86 *c)
29 {
30 	u64 misc_enable;
31 
32 	/* Unmask CPUID levels if masked: */
33 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
34 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
35 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
36 			c->cpuid_level = cpuid_eax(0);
37 			get_cpu_cap(c);
38 		}
39 	}
40 
41 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
42 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
43 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
44 
45 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) {
46 		unsigned lower_word;
47 
48 		wrmsr(MSR_IA32_UCODE_REV, 0, 0);
49 		/* Required by the SDM */
50 		sync_core();
51 		rdmsr(MSR_IA32_UCODE_REV, lower_word, c->microcode);
52 	}
53 
54 	/*
55 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
56 	 *
57 	 * A race condition between speculative fetches and invalidating
58 	 * a large page.  This is worked around in microcode, but we
59 	 * need the microcode to have already been loaded... so if it is
60 	 * not, recommend a BIOS update and disable large pages.
61 	 */
62 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 &&
63 	    c->microcode < 0x20e) {
64 		printk(KERN_WARNING "Atom PSE erratum detected, BIOS microcode update recommended\n");
65 		clear_cpu_cap(c, X86_FEATURE_PSE);
66 	}
67 
68 #ifdef CONFIG_X86_64
69 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
70 #else
71 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
72 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
73 		c->x86_cache_alignment = 128;
74 #endif
75 
76 	/* CPUID workaround for 0F33/0F34 CPU */
77 	if (c->x86 == 0xF && c->x86_model == 0x3
78 	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
79 		c->x86_phys_bits = 36;
80 
81 	/*
82 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
83 	 * with P/T states and does not stop in deep C-states.
84 	 *
85 	 * It is also reliable across cores and sockets. (but not across
86 	 * cabinets - we turn it off in that case explicitly.)
87 	 */
88 	if (c->x86_power & (1 << 8)) {
89 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
90 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
91 		if (!check_tsc_unstable())
92 			set_sched_clock_stable();
93 	}
94 
95 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
96 	if (c->x86 == 6) {
97 		switch (c->x86_model) {
98 		case 0x27:	/* Penwell */
99 		case 0x35:	/* Cloverview */
100 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
101 			break;
102 		default:
103 			break;
104 		}
105 	}
106 
107 	/*
108 	 * There is a known erratum on Pentium III and Core Solo
109 	 * and Core Duo CPUs.
110 	 * " Page with PAT set to WC while associated MTRR is UC
111 	 *   may consolidate to UC "
112 	 * Because of this erratum, it is better to stick with
113 	 * setting WC in MTRR rather than using PAT on these CPUs.
114 	 *
115 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
116 	 */
117 	if (c->x86 == 6 && c->x86_model < 15)
118 		clear_cpu_cap(c, X86_FEATURE_PAT);
119 
120 #ifdef CONFIG_KMEMCHECK
121 	/*
122 	 * P4s have a "fast strings" feature which causes single-
123 	 * stepping REP instructions to only generate a #DB on
124 	 * cache-line boundaries.
125 	 *
126 	 * Ingo Molnar reported a Pentium D (model 6) and a Xeon
127 	 * (model 2) with the same problem.
128 	 */
129 	if (c->x86 == 15)
130 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
131 				  MSR_IA32_MISC_ENABLE_FAST_STRING_BIT) > 0)
132 			pr_info("kmemcheck: Disabling fast string operations\n");
133 #endif
134 
135 	/*
136 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
137 	 * clear the fast string and enhanced fast string CPU capabilities.
138 	 */
139 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
140 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
141 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
142 			printk(KERN_INFO "Disabled fast string operations\n");
143 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
144 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
145 		}
146 	}
147 }
148 
149 #ifdef CONFIG_X86_32
150 /*
151  *	Early probe support logic for ppro memory erratum #50
152  *
153  *	This is called before we do cpu ident work
154  */
155 
156 int ppro_with_ram_bug(void)
157 {
158 	/* Uses data from early_cpu_detect now */
159 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
160 	    boot_cpu_data.x86 == 6 &&
161 	    boot_cpu_data.x86_model == 1 &&
162 	    boot_cpu_data.x86_mask < 8) {
163 		printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n");
164 		return 1;
165 	}
166 	return 0;
167 }
168 
169 static void intel_smp_check(struct cpuinfo_x86 *c)
170 {
171 	/* calling is from identify_secondary_cpu() ? */
172 	if (!c->cpu_index)
173 		return;
174 
175 	/*
176 	 * Mask B, Pentium, but not Pentium MMX
177 	 */
178 	if (c->x86 == 5 &&
179 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
180 	    c->x86_model <= 3) {
181 		/*
182 		 * Remember we have B step Pentia with bugs
183 		 */
184 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
185 				    "with B stepping processors.\n");
186 	}
187 }
188 
189 static int forcepae;
190 static int __init forcepae_setup(char *__unused)
191 {
192 	forcepae = 1;
193 	return 1;
194 }
195 __setup("forcepae", forcepae_setup);
196 
197 static void intel_workarounds(struct cpuinfo_x86 *c)
198 {
199 #ifdef CONFIG_X86_F00F_BUG
200 	/*
201 	 * All current models of Pentium and Pentium with MMX technology CPUs
202 	 * have the F0 0F bug, which lets nonprivileged users lock up the
203 	 * system. Announce that the fault handler will be checking for it.
204 	 */
205 	clear_cpu_bug(c, X86_BUG_F00F);
206 	if (!paravirt_enabled() && c->x86 == 5) {
207 		static int f00f_workaround_enabled;
208 
209 		set_cpu_bug(c, X86_BUG_F00F);
210 		if (!f00f_workaround_enabled) {
211 			printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
212 			f00f_workaround_enabled = 1;
213 		}
214 	}
215 #endif
216 
217 	/*
218 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
219 	 * model 3 mask 3
220 	 */
221 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
222 		clear_cpu_cap(c, X86_FEATURE_SEP);
223 
224 	/*
225 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
226 	 * functionally usable PAE implementation.
227 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
228 	 */
229 	if (forcepae) {
230 		printk(KERN_WARNING "PAE forced!\n");
231 		set_cpu_cap(c, X86_FEATURE_PAE);
232 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
233 	}
234 
235 	/*
236 	 * P4 Xeon errata 037 workaround.
237 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
238 	 */
239 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
240 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
241 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT)
242 		    > 0) {
243 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
244 			pr_info("CPU: Disabling hardware prefetching (Errata 037)\n");
245 		}
246 	}
247 
248 	/*
249 	 * See if we have a good local APIC by checking for buggy Pentia,
250 	 * i.e. all B steppings and the C2 stepping of P54C when using their
251 	 * integrated APIC (see 11AP erratum in "Pentium Processor
252 	 * Specification Update").
253 	 */
254 	if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
255 	    (c->x86_mask < 0x6 || c->x86_mask == 0xb))
256 		set_cpu_cap(c, X86_FEATURE_11AP);
257 
258 
259 #ifdef CONFIG_X86_INTEL_USERCOPY
260 	/*
261 	 * Set up the preferred alignment for movsl bulk memory moves
262 	 */
263 	switch (c->x86) {
264 	case 4:		/* 486: untested */
265 		break;
266 	case 5:		/* Old Pentia: untested */
267 		break;
268 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
269 		movsl_mask.mask = 7;
270 		break;
271 	case 15:	/* P4 is OK down to 8-byte alignment */
272 		movsl_mask.mask = 7;
273 		break;
274 	}
275 #endif
276 
277 	intel_smp_check(c);
278 }
279 #else
280 static void intel_workarounds(struct cpuinfo_x86 *c)
281 {
282 }
283 #endif
284 
285 static void srat_detect_node(struct cpuinfo_x86 *c)
286 {
287 #ifdef CONFIG_NUMA
288 	unsigned node;
289 	int cpu = smp_processor_id();
290 
291 	/* Don't do the funky fallback heuristics the AMD version employs
292 	   for now. */
293 	node = numa_cpu_node(cpu);
294 	if (node == NUMA_NO_NODE || !node_online(node)) {
295 		/* reuse the value from init_cpu_to_node() */
296 		node = cpu_to_node(cpu);
297 	}
298 	numa_set_node(cpu, node);
299 #endif
300 }
301 
302 /*
303  * find out the number of processor cores on the die
304  */
305 static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
306 {
307 	unsigned int eax, ebx, ecx, edx;
308 
309 	if (c->cpuid_level < 4)
310 		return 1;
311 
312 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
313 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
314 	if (eax & 0x1f)
315 		return (eax >> 26) + 1;
316 	else
317 		return 1;
318 }
319 
320 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
321 {
322 	/* Intel VMX MSR indicated features */
323 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
324 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
325 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
326 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
327 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
328 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
329 
330 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
331 
332 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
333 	clear_cpu_cap(c, X86_FEATURE_VNMI);
334 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
335 	clear_cpu_cap(c, X86_FEATURE_EPT);
336 	clear_cpu_cap(c, X86_FEATURE_VPID);
337 
338 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
339 	msr_ctl = vmx_msr_high | vmx_msr_low;
340 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
341 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
342 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
343 		set_cpu_cap(c, X86_FEATURE_VNMI);
344 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
345 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
346 		      vmx_msr_low, vmx_msr_high);
347 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
348 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
349 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
350 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
351 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
352 			set_cpu_cap(c, X86_FEATURE_EPT);
353 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
354 			set_cpu_cap(c, X86_FEATURE_VPID);
355 	}
356 }
357 
358 static void init_intel(struct cpuinfo_x86 *c)
359 {
360 	unsigned int l2 = 0;
361 
362 	early_init_intel(c);
363 
364 	intel_workarounds(c);
365 
366 	/*
367 	 * Detect the extended topology information if available. This
368 	 * will reinitialise the initial_apicid which will be used
369 	 * in init_intel_cacheinfo()
370 	 */
371 	detect_extended_topology(c);
372 
373 	l2 = init_intel_cacheinfo(c);
374 	if (c->cpuid_level > 9) {
375 		unsigned eax = cpuid_eax(10);
376 		/* Check for version and the number of counters */
377 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
378 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
379 	}
380 
381 	if (cpu_has_xmm2)
382 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
383 	if (cpu_has_ds) {
384 		unsigned int l1;
385 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
386 		if (!(l1 & (1<<11)))
387 			set_cpu_cap(c, X86_FEATURE_BTS);
388 		if (!(l1 & (1<<12)))
389 			set_cpu_cap(c, X86_FEATURE_PEBS);
390 	}
391 
392 	if (c->x86 == 6 && cpu_has_clflush &&
393 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
394 		set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
395 
396 #ifdef CONFIG_X86_64
397 	if (c->x86 == 15)
398 		c->x86_cache_alignment = c->x86_clflush_size * 2;
399 	if (c->x86 == 6)
400 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
401 #else
402 	/*
403 	 * Names for the Pentium II/Celeron processors
404 	 * detectable only by also checking the cache size.
405 	 * Dixon is NOT a Celeron.
406 	 */
407 	if (c->x86 == 6) {
408 		char *p = NULL;
409 
410 		switch (c->x86_model) {
411 		case 5:
412 			if (l2 == 0)
413 				p = "Celeron (Covington)";
414 			else if (l2 == 256)
415 				p = "Mobile Pentium II (Dixon)";
416 			break;
417 
418 		case 6:
419 			if (l2 == 128)
420 				p = "Celeron (Mendocino)";
421 			else if (c->x86_mask == 0 || c->x86_mask == 5)
422 				p = "Celeron-A";
423 			break;
424 
425 		case 8:
426 			if (l2 == 128)
427 				p = "Celeron (Coppermine)";
428 			break;
429 		}
430 
431 		if (p)
432 			strcpy(c->x86_model_id, p);
433 	}
434 
435 	if (c->x86 == 15)
436 		set_cpu_cap(c, X86_FEATURE_P4);
437 	if (c->x86 == 6)
438 		set_cpu_cap(c, X86_FEATURE_P3);
439 #endif
440 
441 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
442 		/*
443 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
444 		 * detection.
445 		 */
446 		c->x86_max_cores = intel_num_cpu_cores(c);
447 #ifdef CONFIG_X86_32
448 		detect_ht(c);
449 #endif
450 	}
451 
452 	/* Work around errata */
453 	srat_detect_node(c);
454 
455 	if (cpu_has(c, X86_FEATURE_VMX))
456 		detect_vmx_virtcap(c);
457 
458 	/*
459 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if BIOS did not.
460 	 * x86_energy_perf_policy(8) is available to change it at run-time
461 	 */
462 	if (cpu_has(c, X86_FEATURE_EPB)) {
463 		u64 epb;
464 
465 		rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
466 		if ((epb & 0xF) == ENERGY_PERF_BIAS_PERFORMANCE) {
467 			printk_once(KERN_WARNING "ENERGY_PERF_BIAS:"
468 				" Set to 'normal', was 'performance'\n"
469 				"ENERGY_PERF_BIAS: View and update with"
470 				" x86_energy_perf_policy(8)\n");
471 			epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
472 			wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
473 		}
474 	}
475 }
476 
477 #ifdef CONFIG_X86_32
478 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
479 {
480 	/*
481 	 * Intel PIII Tualatin. This comes in two flavours.
482 	 * One has 256kb of cache, the other 512. We have no way
483 	 * to determine which, so we use a boottime override
484 	 * for the 512kb model, and assume 256 otherwise.
485 	 */
486 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
487 		size = 256;
488 	return size;
489 }
490 #endif
491 
492 #define TLB_INST_4K	0x01
493 #define TLB_INST_4M	0x02
494 #define TLB_INST_2M_4M	0x03
495 
496 #define TLB_INST_ALL	0x05
497 #define TLB_INST_1G	0x06
498 
499 #define TLB_DATA_4K	0x11
500 #define TLB_DATA_4M	0x12
501 #define TLB_DATA_2M_4M	0x13
502 #define TLB_DATA_4K_4M	0x14
503 
504 #define TLB_DATA_1G	0x16
505 
506 #define TLB_DATA0_4K	0x21
507 #define TLB_DATA0_4M	0x22
508 #define TLB_DATA0_2M_4M	0x23
509 
510 #define STLB_4K		0x41
511 #define STLB_4K_2M	0x42
512 
513 static const struct _tlb_table intel_tlb_table[] = {
514 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
515 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
516 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
517 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
518 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
519 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
520 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
521 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
522 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
523 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
524 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
525 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
526 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
527 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
528 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
529 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
530 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
531 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
532 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
533 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
534 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
535 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
536 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
537 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
538 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
539 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
540 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set ssociative" },
541 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set ssociative" },
542 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
543 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
544 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
545 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
546 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
547 	{ 0x00, 0, 0 }
548 };
549 
550 static void intel_tlb_lookup(const unsigned char desc)
551 {
552 	unsigned char k;
553 	if (desc == 0)
554 		return;
555 
556 	/* look up this descriptor in the table */
557 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
558 			intel_tlb_table[k].descriptor != 0; k++)
559 		;
560 
561 	if (intel_tlb_table[k].tlb_type == 0)
562 		return;
563 
564 	switch (intel_tlb_table[k].tlb_type) {
565 	case STLB_4K:
566 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
567 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
568 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
569 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
570 		break;
571 	case STLB_4K_2M:
572 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
573 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
574 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
575 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
576 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
577 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
578 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
579 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
580 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
581 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
582 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
583 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
584 		break;
585 	case TLB_INST_ALL:
586 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
587 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
588 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
589 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
590 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
591 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
592 		break;
593 	case TLB_INST_4K:
594 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
595 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
596 		break;
597 	case TLB_INST_4M:
598 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
599 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
600 		break;
601 	case TLB_INST_2M_4M:
602 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
603 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
604 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
605 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
606 		break;
607 	case TLB_DATA_4K:
608 	case TLB_DATA0_4K:
609 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
610 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
611 		break;
612 	case TLB_DATA_4M:
613 	case TLB_DATA0_4M:
614 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
615 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
616 		break;
617 	case TLB_DATA_2M_4M:
618 	case TLB_DATA0_2M_4M:
619 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
620 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
621 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
622 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
623 		break;
624 	case TLB_DATA_4K_4M:
625 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
626 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
627 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
628 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
629 		break;
630 	case TLB_DATA_1G:
631 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
632 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
633 		break;
634 	}
635 }
636 
637 static void intel_tlb_flushall_shift_set(struct cpuinfo_x86 *c)
638 {
639 	switch ((c->x86 << 8) + c->x86_model) {
640 	case 0x60f: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */
641 	case 0x616: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */
642 	case 0x617: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */
643 	case 0x61d: /* six-core 45 nm xeon "Dunnington" */
644 		tlb_flushall_shift = -1;
645 		break;
646 	case 0x63a: /* Ivybridge */
647 		tlb_flushall_shift = 2;
648 		break;
649 	case 0x61a: /* 45 nm nehalem, "Bloomfield" */
650 	case 0x61e: /* 45 nm nehalem, "Lynnfield" */
651 	case 0x625: /* 32 nm nehalem, "Clarkdale" */
652 	case 0x62c: /* 32 nm nehalem, "Gulftown" */
653 	case 0x62e: /* 45 nm nehalem-ex, "Beckton" */
654 	case 0x62f: /* 32 nm Xeon E7 */
655 	case 0x62a: /* SandyBridge */
656 	case 0x62d: /* SandyBridge, "Romely-EP" */
657 	default:
658 		tlb_flushall_shift = 6;
659 	}
660 }
661 
662 static void intel_detect_tlb(struct cpuinfo_x86 *c)
663 {
664 	int i, j, n;
665 	unsigned int regs[4];
666 	unsigned char *desc = (unsigned char *)regs;
667 
668 	if (c->cpuid_level < 2)
669 		return;
670 
671 	/* Number of times to iterate */
672 	n = cpuid_eax(2) & 0xFF;
673 
674 	for (i = 0 ; i < n ; i++) {
675 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
676 
677 		/* If bit 31 is set, this is an unknown format */
678 		for (j = 0 ; j < 3 ; j++)
679 			if (regs[j] & (1 << 31))
680 				regs[j] = 0;
681 
682 		/* Byte 0 is level count, not a descriptor */
683 		for (j = 1 ; j < 16 ; j++)
684 			intel_tlb_lookup(desc[j]);
685 	}
686 	intel_tlb_flushall_shift_set(c);
687 }
688 
689 static const struct cpu_dev intel_cpu_dev = {
690 	.c_vendor	= "Intel",
691 	.c_ident	= { "GenuineIntel" },
692 #ifdef CONFIG_X86_32
693 	.legacy_models = {
694 		{ .family = 4, .model_names =
695 		  {
696 			  [0] = "486 DX-25/33",
697 			  [1] = "486 DX-50",
698 			  [2] = "486 SX",
699 			  [3] = "486 DX/2",
700 			  [4] = "486 SL",
701 			  [5] = "486 SX/2",
702 			  [7] = "486 DX/2-WB",
703 			  [8] = "486 DX/4",
704 			  [9] = "486 DX/4-WB"
705 		  }
706 		},
707 		{ .family = 5, .model_names =
708 		  {
709 			  [0] = "Pentium 60/66 A-step",
710 			  [1] = "Pentium 60/66",
711 			  [2] = "Pentium 75 - 200",
712 			  [3] = "OverDrive PODP5V83",
713 			  [4] = "Pentium MMX",
714 			  [7] = "Mobile Pentium 75 - 200",
715 			  [8] = "Mobile Pentium MMX"
716 		  }
717 		},
718 		{ .family = 6, .model_names =
719 		  {
720 			  [0] = "Pentium Pro A-step",
721 			  [1] = "Pentium Pro",
722 			  [3] = "Pentium II (Klamath)",
723 			  [4] = "Pentium II (Deschutes)",
724 			  [5] = "Pentium II (Deschutes)",
725 			  [6] = "Mobile Pentium II",
726 			  [7] = "Pentium III (Katmai)",
727 			  [8] = "Pentium III (Coppermine)",
728 			  [10] = "Pentium III (Cascades)",
729 			  [11] = "Pentium III (Tualatin)",
730 		  }
731 		},
732 		{ .family = 15, .model_names =
733 		  {
734 			  [0] = "Pentium 4 (Unknown)",
735 			  [1] = "Pentium 4 (Willamette)",
736 			  [2] = "Pentium 4 (Northwood)",
737 			  [4] = "Pentium 4 (Foster)",
738 			  [5] = "Pentium 4 (Foster)",
739 		  }
740 		},
741 	},
742 	.legacy_cache_size = intel_size_cache,
743 #endif
744 	.c_detect_tlb	= intel_detect_tlb,
745 	.c_early_init   = early_init_intel,
746 	.c_init		= init_intel,
747 	.c_x86_vendor	= X86_VENDOR_INTEL,
748 };
749 
750 cpu_dev_register(intel_cpu_dev);
751 
752