xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision f6723b56)
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 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
35 
36 		if (misc_enable & MSR_IA32_MISC_ENABLE_LIMIT_CPUID) {
37 			misc_enable &= ~MSR_IA32_MISC_ENABLE_LIMIT_CPUID;
38 			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
39 			c->cpuid_level = cpuid_eax(0);
40 			get_cpu_cap(c);
41 		}
42 	}
43 
44 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
45 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
46 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
47 
48 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) {
49 		unsigned lower_word;
50 
51 		wrmsr(MSR_IA32_UCODE_REV, 0, 0);
52 		/* Required by the SDM */
53 		sync_core();
54 		rdmsr(MSR_IA32_UCODE_REV, lower_word, c->microcode);
55 	}
56 
57 	/*
58 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
59 	 *
60 	 * A race condition between speculative fetches and invalidating
61 	 * a large page.  This is worked around in microcode, but we
62 	 * need the microcode to have already been loaded... so if it is
63 	 * not, recommend a BIOS update and disable large pages.
64 	 */
65 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 &&
66 	    c->microcode < 0x20e) {
67 		printk(KERN_WARNING "Atom PSE erratum detected, BIOS microcode update recommended\n");
68 		clear_cpu_cap(c, X86_FEATURE_PSE);
69 	}
70 
71 #ifdef CONFIG_X86_64
72 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
73 #else
74 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
75 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
76 		c->x86_cache_alignment = 128;
77 #endif
78 
79 	/* CPUID workaround for 0F33/0F34 CPU */
80 	if (c->x86 == 0xF && c->x86_model == 0x3
81 	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
82 		c->x86_phys_bits = 36;
83 
84 	/*
85 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
86 	 * with P/T states and does not stop in deep C-states.
87 	 *
88 	 * It is also reliable across cores and sockets. (but not across
89 	 * cabinets - we turn it off in that case explicitly.)
90 	 */
91 	if (c->x86_power & (1 << 8)) {
92 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
93 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
94 		if (!check_tsc_unstable())
95 			set_sched_clock_stable();
96 	}
97 
98 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
99 	if (c->x86 == 6) {
100 		switch (c->x86_model) {
101 		case 0x27:	/* Penwell */
102 		case 0x35:	/* Cloverview */
103 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
104 			break;
105 		default:
106 			break;
107 		}
108 	}
109 
110 	/*
111 	 * There is a known erratum on Pentium III and Core Solo
112 	 * and Core Duo CPUs.
113 	 * " Page with PAT set to WC while associated MTRR is UC
114 	 *   may consolidate to UC "
115 	 * Because of this erratum, it is better to stick with
116 	 * setting WC in MTRR rather than using PAT on these CPUs.
117 	 *
118 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
119 	 */
120 	if (c->x86 == 6 && c->x86_model < 15)
121 		clear_cpu_cap(c, X86_FEATURE_PAT);
122 
123 #ifdef CONFIG_KMEMCHECK
124 	/*
125 	 * P4s have a "fast strings" feature which causes single-
126 	 * stepping REP instructions to only generate a #DB on
127 	 * cache-line boundaries.
128 	 *
129 	 * Ingo Molnar reported a Pentium D (model 6) and a Xeon
130 	 * (model 2) with the same problem.
131 	 */
132 	if (c->x86 == 15) {
133 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
134 
135 		if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) {
136 			printk(KERN_INFO "kmemcheck: Disabling fast string operations\n");
137 
138 			misc_enable &= ~MSR_IA32_MISC_ENABLE_FAST_STRING;
139 			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
140 		}
141 	}
142 #endif
143 
144 	/*
145 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
146 	 * clear the fast string and enhanced fast string CPU capabilities.
147 	 */
148 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
149 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
150 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
151 			printk(KERN_INFO "Disabled fast string operations\n");
152 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
153 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
154 		}
155 	}
156 }
157 
158 #ifdef CONFIG_X86_32
159 /*
160  *	Early probe support logic for ppro memory erratum #50
161  *
162  *	This is called before we do cpu ident work
163  */
164 
165 int ppro_with_ram_bug(void)
166 {
167 	/* Uses data from early_cpu_detect now */
168 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
169 	    boot_cpu_data.x86 == 6 &&
170 	    boot_cpu_data.x86_model == 1 &&
171 	    boot_cpu_data.x86_mask < 8) {
172 		printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n");
173 		return 1;
174 	}
175 	return 0;
176 }
177 
178 static void intel_smp_check(struct cpuinfo_x86 *c)
179 {
180 	/* calling is from identify_secondary_cpu() ? */
181 	if (!c->cpu_index)
182 		return;
183 
184 	/*
185 	 * Mask B, Pentium, but not Pentium MMX
186 	 */
187 	if (c->x86 == 5 &&
188 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
189 	    c->x86_model <= 3) {
190 		/*
191 		 * Remember we have B step Pentia with bugs
192 		 */
193 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
194 				    "with B stepping processors.\n");
195 	}
196 }
197 
198 static void intel_workarounds(struct cpuinfo_x86 *c)
199 {
200 	unsigned long lo, hi;
201 
202 #ifdef CONFIG_X86_F00F_BUG
203 	/*
204 	 * All current models of Pentium and Pentium with MMX technology CPUs
205 	 * have the F0 0F bug, which lets nonprivileged users lock up the
206 	 * system. Announce that the fault handler will be checking for it.
207 	 */
208 	clear_cpu_bug(c, X86_BUG_F00F);
209 	if (!paravirt_enabled() && c->x86 == 5) {
210 		static int f00f_workaround_enabled;
211 
212 		set_cpu_bug(c, X86_BUG_F00F);
213 		if (!f00f_workaround_enabled) {
214 			printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
215 			f00f_workaround_enabled = 1;
216 		}
217 	}
218 #endif
219 
220 	/*
221 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
222 	 * model 3 mask 3
223 	 */
224 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
225 		clear_cpu_cap(c, X86_FEATURE_SEP);
226 
227 	/*
228 	 * P4 Xeon errata 037 workaround.
229 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
230 	 */
231 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
232 		rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
233 		if ((lo & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE) == 0) {
234 			printk (KERN_INFO "CPU: C0 stepping P4 Xeon detected.\n");
235 			printk (KERN_INFO "CPU: Disabling hardware prefetching (Errata 037)\n");
236 			lo |= MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE;
237 			wrmsr(MSR_IA32_MISC_ENABLE, lo, hi);
238 		}
239 	}
240 
241 	/*
242 	 * See if we have a good local APIC by checking for buggy Pentia,
243 	 * i.e. all B steppings and the C2 stepping of P54C when using their
244 	 * integrated APIC (see 11AP erratum in "Pentium Processor
245 	 * Specification Update").
246 	 */
247 	if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
248 	    (c->x86_mask < 0x6 || c->x86_mask == 0xb))
249 		set_cpu_cap(c, X86_FEATURE_11AP);
250 
251 
252 #ifdef CONFIG_X86_INTEL_USERCOPY
253 	/*
254 	 * Set up the preferred alignment for movsl bulk memory moves
255 	 */
256 	switch (c->x86) {
257 	case 4:		/* 486: untested */
258 		break;
259 	case 5:		/* Old Pentia: untested */
260 		break;
261 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
262 		movsl_mask.mask = 7;
263 		break;
264 	case 15:	/* P4 is OK down to 8-byte alignment */
265 		movsl_mask.mask = 7;
266 		break;
267 	}
268 #endif
269 
270 #ifdef CONFIG_X86_NUMAQ
271 	numaq_tsc_disable();
272 #endif
273 
274 	intel_smp_check(c);
275 }
276 #else
277 static void intel_workarounds(struct cpuinfo_x86 *c)
278 {
279 }
280 #endif
281 
282 static void srat_detect_node(struct cpuinfo_x86 *c)
283 {
284 #ifdef CONFIG_NUMA
285 	unsigned node;
286 	int cpu = smp_processor_id();
287 
288 	/* Don't do the funky fallback heuristics the AMD version employs
289 	   for now. */
290 	node = numa_cpu_node(cpu);
291 	if (node == NUMA_NO_NODE || !node_online(node)) {
292 		/* reuse the value from init_cpu_to_node() */
293 		node = cpu_to_node(cpu);
294 	}
295 	numa_set_node(cpu, node);
296 #endif
297 }
298 
299 /*
300  * find out the number of processor cores on the die
301  */
302 static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
303 {
304 	unsigned int eax, ebx, ecx, edx;
305 
306 	if (c->cpuid_level < 4)
307 		return 1;
308 
309 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
310 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
311 	if (eax & 0x1f)
312 		return (eax >> 26) + 1;
313 	else
314 		return 1;
315 }
316 
317 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
318 {
319 	/* Intel VMX MSR indicated features */
320 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
321 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
322 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
323 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
324 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
325 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
326 
327 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
328 
329 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
330 	clear_cpu_cap(c, X86_FEATURE_VNMI);
331 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
332 	clear_cpu_cap(c, X86_FEATURE_EPT);
333 	clear_cpu_cap(c, X86_FEATURE_VPID);
334 
335 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
336 	msr_ctl = vmx_msr_high | vmx_msr_low;
337 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
338 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
339 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
340 		set_cpu_cap(c, X86_FEATURE_VNMI);
341 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
342 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
343 		      vmx_msr_low, vmx_msr_high);
344 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
345 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
346 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
347 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
348 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
349 			set_cpu_cap(c, X86_FEATURE_EPT);
350 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
351 			set_cpu_cap(c, X86_FEATURE_VPID);
352 	}
353 }
354 
355 static void init_intel(struct cpuinfo_x86 *c)
356 {
357 	unsigned int l2 = 0;
358 
359 	early_init_intel(c);
360 
361 	intel_workarounds(c);
362 
363 	/*
364 	 * Detect the extended topology information if available. This
365 	 * will reinitialise the initial_apicid which will be used
366 	 * in init_intel_cacheinfo()
367 	 */
368 	detect_extended_topology(c);
369 
370 	l2 = init_intel_cacheinfo(c);
371 	if (c->cpuid_level > 9) {
372 		unsigned eax = cpuid_eax(10);
373 		/* Check for version and the number of counters */
374 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
375 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
376 	}
377 
378 	if (cpu_has_xmm2)
379 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
380 	if (cpu_has_ds) {
381 		unsigned int l1;
382 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
383 		if (!(l1 & (1<<11)))
384 			set_cpu_cap(c, X86_FEATURE_BTS);
385 		if (!(l1 & (1<<12)))
386 			set_cpu_cap(c, X86_FEATURE_PEBS);
387 	}
388 
389 	if (c->x86 == 6 && cpu_has_clflush &&
390 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
391 		set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
392 
393 #ifdef CONFIG_X86_64
394 	if (c->x86 == 15)
395 		c->x86_cache_alignment = c->x86_clflush_size * 2;
396 	if (c->x86 == 6)
397 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
398 #else
399 	/*
400 	 * Names for the Pentium II/Celeron processors
401 	 * detectable only by also checking the cache size.
402 	 * Dixon is NOT a Celeron.
403 	 */
404 	if (c->x86 == 6) {
405 		char *p = NULL;
406 
407 		switch (c->x86_model) {
408 		case 5:
409 			if (l2 == 0)
410 				p = "Celeron (Covington)";
411 			else if (l2 == 256)
412 				p = "Mobile Pentium II (Dixon)";
413 			break;
414 
415 		case 6:
416 			if (l2 == 128)
417 				p = "Celeron (Mendocino)";
418 			else if (c->x86_mask == 0 || c->x86_mask == 5)
419 				p = "Celeron-A";
420 			break;
421 
422 		case 8:
423 			if (l2 == 128)
424 				p = "Celeron (Coppermine)";
425 			break;
426 		}
427 
428 		if (p)
429 			strcpy(c->x86_model_id, p);
430 	}
431 
432 	if (c->x86 == 15)
433 		set_cpu_cap(c, X86_FEATURE_P4);
434 	if (c->x86 == 6)
435 		set_cpu_cap(c, X86_FEATURE_P3);
436 #endif
437 
438 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
439 		/*
440 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
441 		 * detection.
442 		 */
443 		c->x86_max_cores = intel_num_cpu_cores(c);
444 #ifdef CONFIG_X86_32
445 		detect_ht(c);
446 #endif
447 	}
448 
449 	/* Work around errata */
450 	srat_detect_node(c);
451 
452 	if (cpu_has(c, X86_FEATURE_VMX))
453 		detect_vmx_virtcap(c);
454 
455 	/*
456 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if BIOS did not.
457 	 * x86_energy_perf_policy(8) is available to change it at run-time
458 	 */
459 	if (cpu_has(c, X86_FEATURE_EPB)) {
460 		u64 epb;
461 
462 		rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
463 		if ((epb & 0xF) == ENERGY_PERF_BIAS_PERFORMANCE) {
464 			printk_once(KERN_WARNING "ENERGY_PERF_BIAS:"
465 				" Set to 'normal', was 'performance'\n"
466 				"ENERGY_PERF_BIAS: View and update with"
467 				" x86_energy_perf_policy(8)\n");
468 			epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
469 			wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
470 		}
471 	}
472 }
473 
474 #ifdef CONFIG_X86_32
475 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
476 {
477 	/*
478 	 * Intel PIII Tualatin. This comes in two flavours.
479 	 * One has 256kb of cache, the other 512. We have no way
480 	 * to determine which, so we use a boottime override
481 	 * for the 512kb model, and assume 256 otherwise.
482 	 */
483 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
484 		size = 256;
485 	return size;
486 }
487 #endif
488 
489 #define TLB_INST_4K	0x01
490 #define TLB_INST_4M	0x02
491 #define TLB_INST_2M_4M	0x03
492 
493 #define TLB_INST_ALL	0x05
494 #define TLB_INST_1G	0x06
495 
496 #define TLB_DATA_4K	0x11
497 #define TLB_DATA_4M	0x12
498 #define TLB_DATA_2M_4M	0x13
499 #define TLB_DATA_4K_4M	0x14
500 
501 #define TLB_DATA_1G	0x16
502 
503 #define TLB_DATA0_4K	0x21
504 #define TLB_DATA0_4M	0x22
505 #define TLB_DATA0_2M_4M	0x23
506 
507 #define STLB_4K		0x41
508 #define STLB_4K_2M	0x42
509 
510 static const struct _tlb_table intel_tlb_table[] = {
511 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
512 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
513 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
514 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
515 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
516 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
517 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
518 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
519 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
520 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
521 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
522 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
523 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
524 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
525 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
526 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
527 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
528 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
529 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
530 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
531 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
532 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
533 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
534 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
535 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
536 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
537 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set ssociative" },
538 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set ssociative" },
539 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
540 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
541 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
542 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
543 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
544 	{ 0x00, 0, 0 }
545 };
546 
547 static void intel_tlb_lookup(const unsigned char desc)
548 {
549 	unsigned char k;
550 	if (desc == 0)
551 		return;
552 
553 	/* look up this descriptor in the table */
554 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
555 			intel_tlb_table[k].descriptor != 0; k++)
556 		;
557 
558 	if (intel_tlb_table[k].tlb_type == 0)
559 		return;
560 
561 	switch (intel_tlb_table[k].tlb_type) {
562 	case STLB_4K:
563 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
564 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
565 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
566 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
567 		break;
568 	case STLB_4K_2M:
569 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
570 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
571 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
572 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
573 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
574 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
575 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
576 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
577 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
578 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
579 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
580 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
581 		break;
582 	case TLB_INST_ALL:
583 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
584 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
585 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
586 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
587 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
588 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
589 		break;
590 	case TLB_INST_4K:
591 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
592 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
593 		break;
594 	case TLB_INST_4M:
595 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
596 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
597 		break;
598 	case TLB_INST_2M_4M:
599 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
600 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
601 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
602 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
603 		break;
604 	case TLB_DATA_4K:
605 	case TLB_DATA0_4K:
606 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
607 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
608 		break;
609 	case TLB_DATA_4M:
610 	case TLB_DATA0_4M:
611 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
612 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
613 		break;
614 	case TLB_DATA_2M_4M:
615 	case TLB_DATA0_2M_4M:
616 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
617 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
618 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
619 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
620 		break;
621 	case TLB_DATA_4K_4M:
622 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
623 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
624 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
625 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
626 		break;
627 	case TLB_DATA_1G:
628 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
629 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
630 		break;
631 	}
632 }
633 
634 static void intel_tlb_flushall_shift_set(struct cpuinfo_x86 *c)
635 {
636 	switch ((c->x86 << 8) + c->x86_model) {
637 	case 0x60f: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */
638 	case 0x616: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */
639 	case 0x617: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */
640 	case 0x61d: /* six-core 45 nm xeon "Dunnington" */
641 		tlb_flushall_shift = -1;
642 		break;
643 	case 0x63a: /* Ivybridge */
644 		tlb_flushall_shift = 2;
645 		break;
646 	case 0x61a: /* 45 nm nehalem, "Bloomfield" */
647 	case 0x61e: /* 45 nm nehalem, "Lynnfield" */
648 	case 0x625: /* 32 nm nehalem, "Clarkdale" */
649 	case 0x62c: /* 32 nm nehalem, "Gulftown" */
650 	case 0x62e: /* 45 nm nehalem-ex, "Beckton" */
651 	case 0x62f: /* 32 nm Xeon E7 */
652 	case 0x62a: /* SandyBridge */
653 	case 0x62d: /* SandyBridge, "Romely-EP" */
654 	default:
655 		tlb_flushall_shift = 6;
656 	}
657 }
658 
659 static void intel_detect_tlb(struct cpuinfo_x86 *c)
660 {
661 	int i, j, n;
662 	unsigned int regs[4];
663 	unsigned char *desc = (unsigned char *)regs;
664 
665 	if (c->cpuid_level < 2)
666 		return;
667 
668 	/* Number of times to iterate */
669 	n = cpuid_eax(2) & 0xFF;
670 
671 	for (i = 0 ; i < n ; i++) {
672 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
673 
674 		/* If bit 31 is set, this is an unknown format */
675 		for (j = 0 ; j < 3 ; j++)
676 			if (regs[j] & (1 << 31))
677 				regs[j] = 0;
678 
679 		/* Byte 0 is level count, not a descriptor */
680 		for (j = 1 ; j < 16 ; j++)
681 			intel_tlb_lookup(desc[j]);
682 	}
683 	intel_tlb_flushall_shift_set(c);
684 }
685 
686 static const struct cpu_dev intel_cpu_dev = {
687 	.c_vendor	= "Intel",
688 	.c_ident	= { "GenuineIntel" },
689 #ifdef CONFIG_X86_32
690 	.legacy_models = {
691 		{ .family = 4, .model_names =
692 		  {
693 			  [0] = "486 DX-25/33",
694 			  [1] = "486 DX-50",
695 			  [2] = "486 SX",
696 			  [3] = "486 DX/2",
697 			  [4] = "486 SL",
698 			  [5] = "486 SX/2",
699 			  [7] = "486 DX/2-WB",
700 			  [8] = "486 DX/4",
701 			  [9] = "486 DX/4-WB"
702 		  }
703 		},
704 		{ .family = 5, .model_names =
705 		  {
706 			  [0] = "Pentium 60/66 A-step",
707 			  [1] = "Pentium 60/66",
708 			  [2] = "Pentium 75 - 200",
709 			  [3] = "OverDrive PODP5V83",
710 			  [4] = "Pentium MMX",
711 			  [7] = "Mobile Pentium 75 - 200",
712 			  [8] = "Mobile Pentium MMX"
713 		  }
714 		},
715 		{ .family = 6, .model_names =
716 		  {
717 			  [0] = "Pentium Pro A-step",
718 			  [1] = "Pentium Pro",
719 			  [3] = "Pentium II (Klamath)",
720 			  [4] = "Pentium II (Deschutes)",
721 			  [5] = "Pentium II (Deschutes)",
722 			  [6] = "Mobile Pentium II",
723 			  [7] = "Pentium III (Katmai)",
724 			  [8] = "Pentium III (Coppermine)",
725 			  [10] = "Pentium III (Cascades)",
726 			  [11] = "Pentium III (Tualatin)",
727 		  }
728 		},
729 		{ .family = 15, .model_names =
730 		  {
731 			  [0] = "Pentium 4 (Unknown)",
732 			  [1] = "Pentium 4 (Willamette)",
733 			  [2] = "Pentium 4 (Northwood)",
734 			  [4] = "Pentium 4 (Foster)",
735 			  [5] = "Pentium 4 (Foster)",
736 		  }
737 		},
738 	},
739 	.legacy_cache_size = intel_size_cache,
740 #endif
741 	.c_detect_tlb	= intel_detect_tlb,
742 	.c_early_init   = early_init_intel,
743 	.c_init		= init_intel,
744 	.c_x86_vendor	= X86_VENDOR_INTEL,
745 };
746 
747 cpu_dev_register(intel_cpu_dev);
748 
749