xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 840ef8b7cc584a23c4f9d05352f4dbaf8e56e5ab)
1 #include <linux/init.h>
2 #include <linux/kernel.h>
3 
4 #include <linux/string.h>
5 #include <linux/bitops.h>
6 #include <linux/smp.h>
7 #include <linux/sched.h>
8 #include <linux/thread_info.h>
9 #include <linux/module.h>
10 #include <linux/uaccess.h>
11 
12 #include <asm/processor.h>
13 #include <asm/pgtable.h>
14 #include <asm/msr.h>
15 #include <asm/bugs.h>
16 #include <asm/cpu.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 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
30 {
31 	u64 misc_enable;
32 
33 	/* Unmask CPUID levels if masked: */
34 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
35 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
36 
37 		if (misc_enable & MSR_IA32_MISC_ENABLE_LIMIT_CPUID) {
38 			misc_enable &= ~MSR_IA32_MISC_ENABLE_LIMIT_CPUID;
39 			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
40 			c->cpuid_level = cpuid_eax(0);
41 			get_cpu_cap(c);
42 		}
43 	}
44 
45 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
46 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
47 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
48 
49 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) {
50 		unsigned lower_word;
51 
52 		wrmsr(MSR_IA32_UCODE_REV, 0, 0);
53 		/* Required by the SDM */
54 		sync_core();
55 		rdmsr(MSR_IA32_UCODE_REV, lower_word, c->microcode);
56 	}
57 
58 	/*
59 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
60 	 *
61 	 * A race condition between speculative fetches and invalidating
62 	 * a large page.  This is worked around in microcode, but we
63 	 * need the microcode to have already been loaded... so if it is
64 	 * not, recommend a BIOS update and disable large pages.
65 	 */
66 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 &&
67 	    c->microcode < 0x20e) {
68 		printk(KERN_WARNING "Atom PSE erratum detected, BIOS microcode update recommended\n");
69 		clear_cpu_cap(c, X86_FEATURE_PSE);
70 	}
71 
72 #ifdef CONFIG_X86_64
73 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
74 #else
75 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
76 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
77 		c->x86_cache_alignment = 128;
78 #endif
79 
80 	/* CPUID workaround for 0F33/0F34 CPU */
81 	if (c->x86 == 0xF && c->x86_model == 0x3
82 	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
83 		c->x86_phys_bits = 36;
84 
85 	/*
86 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
87 	 * with P/T states and does not stop in deep C-states.
88 	 *
89 	 * It is also reliable across cores and sockets. (but not across
90 	 * cabinets - we turn it off in that case explicitly.)
91 	 */
92 	if (c->x86_power & (1 << 8)) {
93 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
94 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
95 		if (!check_tsc_unstable())
96 			sched_clock_stable = 1;
97 	}
98 
99 	/*
100 	 * There is a known erratum on Pentium III and Core Solo
101 	 * and Core Duo CPUs.
102 	 * " Page with PAT set to WC while associated MTRR is UC
103 	 *   may consolidate to UC "
104 	 * Because of this erratum, it is better to stick with
105 	 * setting WC in MTRR rather than using PAT on these CPUs.
106 	 *
107 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
108 	 */
109 	if (c->x86 == 6 && c->x86_model < 15)
110 		clear_cpu_cap(c, X86_FEATURE_PAT);
111 
112 #ifdef CONFIG_KMEMCHECK
113 	/*
114 	 * P4s have a "fast strings" feature which causes single-
115 	 * stepping REP instructions to only generate a #DB on
116 	 * cache-line boundaries.
117 	 *
118 	 * Ingo Molnar reported a Pentium D (model 6) and a Xeon
119 	 * (model 2) with the same problem.
120 	 */
121 	if (c->x86 == 15) {
122 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
123 
124 		if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) {
125 			printk(KERN_INFO "kmemcheck: Disabling fast string operations\n");
126 
127 			misc_enable &= ~MSR_IA32_MISC_ENABLE_FAST_STRING;
128 			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
129 		}
130 	}
131 #endif
132 
133 	/*
134 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
135 	 * clear the fast string and enhanced fast string CPU capabilities.
136 	 */
137 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
138 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
139 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
140 			printk(KERN_INFO "Disabled fast string operations\n");
141 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
142 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
143 		}
144 	}
145 }
146 
147 #ifdef CONFIG_X86_32
148 /*
149  *	Early probe support logic for ppro memory erratum #50
150  *
151  *	This is called before we do cpu ident work
152  */
153 
154 int __cpuinit ppro_with_ram_bug(void)
155 {
156 	/* Uses data from early_cpu_detect now */
157 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
158 	    boot_cpu_data.x86 == 6 &&
159 	    boot_cpu_data.x86_model == 1 &&
160 	    boot_cpu_data.x86_mask < 8) {
161 		printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n");
162 		return 1;
163 	}
164 	return 0;
165 }
166 
167 #ifdef CONFIG_X86_F00F_BUG
168 static void __cpuinit trap_init_f00f_bug(void)
169 {
170 	__set_fixmap(FIX_F00F_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
171 
172 	/*
173 	 * Update the IDT descriptor and reload the IDT so that
174 	 * it uses the read-only mapped virtual address.
175 	 */
176 	idt_descr.address = fix_to_virt(FIX_F00F_IDT);
177 	load_idt(&idt_descr);
178 }
179 #endif
180 
181 static void __cpuinit intel_smp_check(struct cpuinfo_x86 *c)
182 {
183 	/* calling is from identify_secondary_cpu() ? */
184 	if (!c->cpu_index)
185 		return;
186 
187 	/*
188 	 * Mask B, Pentium, but not Pentium MMX
189 	 */
190 	if (c->x86 == 5 &&
191 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
192 	    c->x86_model <= 3) {
193 		/*
194 		 * Remember we have B step Pentia with bugs
195 		 */
196 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
197 				    "with B stepping processors.\n");
198 	}
199 }
200 
201 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
202 {
203 	unsigned long lo, hi;
204 
205 #ifdef CONFIG_X86_F00F_BUG
206 	/*
207 	 * All current models of Pentium and Pentium with MMX technology CPUs
208 	 * have the F0 0F bug, which lets nonprivileged users lock up the
209 	 * system.
210 	 * Note that the workaround only should be initialized once...
211 	 */
212 	c->f00f_bug = 0;
213 	if (!paravirt_enabled() && c->x86 == 5) {
214 		static int f00f_workaround_enabled;
215 
216 		c->f00f_bug = 1;
217 		if (!f00f_workaround_enabled) {
218 			trap_init_f00f_bug();
219 			printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
220 			f00f_workaround_enabled = 1;
221 		}
222 	}
223 #endif
224 
225 	/*
226 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
227 	 * model 3 mask 3
228 	 */
229 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
230 		clear_cpu_cap(c, X86_FEATURE_SEP);
231 
232 	/*
233 	 * P4 Xeon errata 037 workaround.
234 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
235 	 */
236 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
237 		rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
238 		if ((lo & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE) == 0) {
239 			printk (KERN_INFO "CPU: C0 stepping P4 Xeon detected.\n");
240 			printk (KERN_INFO "CPU: Disabling hardware prefetching (Errata 037)\n");
241 			lo |= MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE;
242 			wrmsr(MSR_IA32_MISC_ENABLE, lo, hi);
243 		}
244 	}
245 
246 	/*
247 	 * See if we have a good local APIC by checking for buggy Pentia,
248 	 * i.e. all B steppings and the C2 stepping of P54C when using their
249 	 * integrated APIC (see 11AP erratum in "Pentium Processor
250 	 * Specification Update").
251 	 */
252 	if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
253 	    (c->x86_mask < 0x6 || c->x86_mask == 0xb))
254 		set_cpu_cap(c, X86_FEATURE_11AP);
255 
256 
257 #ifdef CONFIG_X86_INTEL_USERCOPY
258 	/*
259 	 * Set up the preferred alignment for movsl bulk memory moves
260 	 */
261 	switch (c->x86) {
262 	case 4:		/* 486: untested */
263 		break;
264 	case 5:		/* Old Pentia: untested */
265 		break;
266 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
267 		movsl_mask.mask = 7;
268 		break;
269 	case 15:	/* P4 is OK down to 8-byte alignment */
270 		movsl_mask.mask = 7;
271 		break;
272 	}
273 #endif
274 
275 #ifdef CONFIG_X86_NUMAQ
276 	numaq_tsc_disable();
277 #endif
278 
279 	intel_smp_check(c);
280 }
281 #else
282 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
283 {
284 }
285 #endif
286 
287 static void __cpuinit srat_detect_node(struct cpuinfo_x86 *c)
288 {
289 #ifdef CONFIG_NUMA
290 	unsigned node;
291 	int cpu = smp_processor_id();
292 
293 	/* Don't do the funky fallback heuristics the AMD version employs
294 	   for now. */
295 	node = numa_cpu_node(cpu);
296 	if (node == NUMA_NO_NODE || !node_online(node)) {
297 		/* reuse the value from init_cpu_to_node() */
298 		node = cpu_to_node(cpu);
299 	}
300 	numa_set_node(cpu, node);
301 #endif
302 }
303 
304 /*
305  * find out the number of processor cores on the die
306  */
307 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
308 {
309 	unsigned int eax, ebx, ecx, edx;
310 
311 	if (c->cpuid_level < 4)
312 		return 1;
313 
314 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
315 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
316 	if (eax & 0x1f)
317 		return (eax >> 26) + 1;
318 	else
319 		return 1;
320 }
321 
322 static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
323 {
324 	/* Intel VMX MSR indicated features */
325 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
326 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
327 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
328 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
329 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
330 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
331 
332 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
333 
334 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
335 	clear_cpu_cap(c, X86_FEATURE_VNMI);
336 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
337 	clear_cpu_cap(c, X86_FEATURE_EPT);
338 	clear_cpu_cap(c, X86_FEATURE_VPID);
339 
340 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
341 	msr_ctl = vmx_msr_high | vmx_msr_low;
342 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
343 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
344 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
345 		set_cpu_cap(c, X86_FEATURE_VNMI);
346 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
347 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
348 		      vmx_msr_low, vmx_msr_high);
349 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
350 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
351 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
352 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
353 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
354 			set_cpu_cap(c, X86_FEATURE_EPT);
355 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
356 			set_cpu_cap(c, X86_FEATURE_VPID);
357 	}
358 }
359 
360 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
361 {
362 	unsigned int l2 = 0;
363 
364 	early_init_intel(c);
365 
366 	intel_workarounds(c);
367 
368 	/*
369 	 * Detect the extended topology information if available. This
370 	 * will reinitialise the initial_apicid which will be used
371 	 * in init_intel_cacheinfo()
372 	 */
373 	detect_extended_topology(c);
374 
375 	l2 = init_intel_cacheinfo(c);
376 	if (c->cpuid_level > 9) {
377 		unsigned eax = cpuid_eax(10);
378 		/* Check for version and the number of counters */
379 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
380 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
381 	}
382 
383 	if (cpu_has_xmm2)
384 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
385 	if (cpu_has_ds) {
386 		unsigned int l1;
387 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
388 		if (!(l1 & (1<<11)))
389 			set_cpu_cap(c, X86_FEATURE_BTS);
390 		if (!(l1 & (1<<12)))
391 			set_cpu_cap(c, X86_FEATURE_PEBS);
392 	}
393 
394 	if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush)
395 		set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
396 
397 #ifdef CONFIG_X86_64
398 	if (c->x86 == 15)
399 		c->x86_cache_alignment = c->x86_clflush_size * 2;
400 	if (c->x86 == 6)
401 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
402 #else
403 	/*
404 	 * Names for the Pentium II/Celeron processors
405 	 * detectable only by also checking the cache size.
406 	 * Dixon is NOT a Celeron.
407 	 */
408 	if (c->x86 == 6) {
409 		char *p = NULL;
410 
411 		switch (c->x86_model) {
412 		case 5:
413 			if (l2 == 0)
414 				p = "Celeron (Covington)";
415 			else if (l2 == 256)
416 				p = "Mobile Pentium II (Dixon)";
417 			break;
418 
419 		case 6:
420 			if (l2 == 128)
421 				p = "Celeron (Mendocino)";
422 			else if (c->x86_mask == 0 || c->x86_mask == 5)
423 				p = "Celeron-A";
424 			break;
425 
426 		case 8:
427 			if (l2 == 128)
428 				p = "Celeron (Coppermine)";
429 			break;
430 		}
431 
432 		if (p)
433 			strcpy(c->x86_model_id, p);
434 	}
435 
436 	if (c->x86 == 15)
437 		set_cpu_cap(c, X86_FEATURE_P4);
438 	if (c->x86 == 6)
439 		set_cpu_cap(c, X86_FEATURE_P3);
440 #endif
441 
442 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
443 		/*
444 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
445 		 * detection.
446 		 */
447 		c->x86_max_cores = intel_num_cpu_cores(c);
448 #ifdef CONFIG_X86_32
449 		detect_ht(c);
450 #endif
451 	}
452 
453 	/* Work around errata */
454 	srat_detect_node(c);
455 
456 	if (cpu_has(c, X86_FEATURE_VMX))
457 		detect_vmx_virtcap(c);
458 
459 	/*
460 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if BIOS did not.
461 	 * x86_energy_perf_policy(8) is available to change it at run-time
462 	 */
463 	if (cpu_has(c, X86_FEATURE_EPB)) {
464 		u64 epb;
465 
466 		rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
467 		if ((epb & 0xF) == ENERGY_PERF_BIAS_PERFORMANCE) {
468 			printk_once(KERN_WARNING "ENERGY_PERF_BIAS:"
469 				" Set to 'normal', was 'performance'\n"
470 				"ENERGY_PERF_BIAS: View and update with"
471 				" x86_energy_perf_policy(8)\n");
472 			epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
473 			wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
474 		}
475 	}
476 }
477 
478 #ifdef CONFIG_X86_32
479 static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
480 {
481 	/*
482 	 * Intel PIII Tualatin. This comes in two flavours.
483 	 * One has 256kb of cache, the other 512. We have no way
484 	 * to determine which, so we use a boottime override
485 	 * for the 512kb model, and assume 256 otherwise.
486 	 */
487 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
488 		size = 256;
489 	return size;
490 }
491 #endif
492 
493 #define TLB_INST_4K	0x01
494 #define TLB_INST_4M	0x02
495 #define TLB_INST_2M_4M	0x03
496 
497 #define TLB_INST_ALL	0x05
498 #define TLB_INST_1G	0x06
499 
500 #define TLB_DATA_4K	0x11
501 #define TLB_DATA_4M	0x12
502 #define TLB_DATA_2M_4M	0x13
503 #define TLB_DATA_4K_4M	0x14
504 
505 #define TLB_DATA_1G	0x16
506 
507 #define TLB_DATA0_4K	0x21
508 #define TLB_DATA0_4M	0x22
509 #define TLB_DATA0_2M_4M	0x23
510 
511 #define STLB_4K		0x41
512 
513 static const struct _tlb_table intel_tlb_table[] __cpuinitconst = {
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 	{ 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 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
538 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
539 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
540 	{ 0x00, 0, 0 }
541 };
542 
543 static void __cpuinit intel_tlb_lookup(const unsigned char desc)
544 {
545 	unsigned char k;
546 	if (desc == 0)
547 		return;
548 
549 	/* look up this descriptor in the table */
550 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
551 			intel_tlb_table[k].descriptor != 0; k++)
552 		;
553 
554 	if (intel_tlb_table[k].tlb_type == 0)
555 		return;
556 
557 	switch (intel_tlb_table[k].tlb_type) {
558 	case STLB_4K:
559 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
560 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
561 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
562 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
563 		break;
564 	case TLB_INST_ALL:
565 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
566 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
567 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
568 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
569 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
570 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
571 		break;
572 	case TLB_INST_4K:
573 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
574 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
575 		break;
576 	case TLB_INST_4M:
577 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
578 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
579 		break;
580 	case TLB_INST_2M_4M:
581 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
582 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
583 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
584 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
585 		break;
586 	case TLB_DATA_4K:
587 	case TLB_DATA0_4K:
588 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
589 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
590 		break;
591 	case TLB_DATA_4M:
592 	case TLB_DATA0_4M:
593 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
594 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
595 		break;
596 	case TLB_DATA_2M_4M:
597 	case TLB_DATA0_2M_4M:
598 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
599 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
600 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
601 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
602 		break;
603 	case TLB_DATA_4K_4M:
604 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
605 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
606 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
607 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
608 		break;
609 	}
610 }
611 
612 static void __cpuinit intel_tlb_flushall_shift_set(struct cpuinfo_x86 *c)
613 {
614 	switch ((c->x86 << 8) + c->x86_model) {
615 	case 0x60f: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */
616 	case 0x616: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */
617 	case 0x617: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */
618 	case 0x61d: /* six-core 45 nm xeon "Dunnington" */
619 		tlb_flushall_shift = -1;
620 		break;
621 	case 0x61a: /* 45 nm nehalem, "Bloomfield" */
622 	case 0x61e: /* 45 nm nehalem, "Lynnfield" */
623 	case 0x625: /* 32 nm nehalem, "Clarkdale" */
624 	case 0x62c: /* 32 nm nehalem, "Gulftown" */
625 	case 0x62e: /* 45 nm nehalem-ex, "Beckton" */
626 	case 0x62f: /* 32 nm Xeon E7 */
627 		tlb_flushall_shift = 6;
628 		break;
629 	case 0x62a: /* SandyBridge */
630 	case 0x62d: /* SandyBridge, "Romely-EP" */
631 		tlb_flushall_shift = 5;
632 		break;
633 	case 0x63a: /* Ivybridge */
634 		tlb_flushall_shift = 1;
635 		break;
636 	default:
637 		tlb_flushall_shift = 6;
638 	}
639 }
640 
641 static void __cpuinit intel_detect_tlb(struct cpuinfo_x86 *c)
642 {
643 	int i, j, n;
644 	unsigned int regs[4];
645 	unsigned char *desc = (unsigned char *)regs;
646 
647 	if (c->cpuid_level < 2)
648 		return;
649 
650 	/* Number of times to iterate */
651 	n = cpuid_eax(2) & 0xFF;
652 
653 	for (i = 0 ; i < n ; i++) {
654 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
655 
656 		/* If bit 31 is set, this is an unknown format */
657 		for (j = 0 ; j < 3 ; j++)
658 			if (regs[j] & (1 << 31))
659 				regs[j] = 0;
660 
661 		/* Byte 0 is level count, not a descriptor */
662 		for (j = 1 ; j < 16 ; j++)
663 			intel_tlb_lookup(desc[j]);
664 	}
665 	intel_tlb_flushall_shift_set(c);
666 }
667 
668 static const struct cpu_dev __cpuinitconst intel_cpu_dev = {
669 	.c_vendor	= "Intel",
670 	.c_ident	= { "GenuineIntel" },
671 #ifdef CONFIG_X86_32
672 	.c_models = {
673 		{ .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
674 		  {
675 			  [0] = "486 DX-25/33",
676 			  [1] = "486 DX-50",
677 			  [2] = "486 SX",
678 			  [3] = "486 DX/2",
679 			  [4] = "486 SL",
680 			  [5] = "486 SX/2",
681 			  [7] = "486 DX/2-WB",
682 			  [8] = "486 DX/4",
683 			  [9] = "486 DX/4-WB"
684 		  }
685 		},
686 		{ .vendor = X86_VENDOR_INTEL, .family = 5, .model_names =
687 		  {
688 			  [0] = "Pentium 60/66 A-step",
689 			  [1] = "Pentium 60/66",
690 			  [2] = "Pentium 75 - 200",
691 			  [3] = "OverDrive PODP5V83",
692 			  [4] = "Pentium MMX",
693 			  [7] = "Mobile Pentium 75 - 200",
694 			  [8] = "Mobile Pentium MMX"
695 		  }
696 		},
697 		{ .vendor = X86_VENDOR_INTEL, .family = 6, .model_names =
698 		  {
699 			  [0] = "Pentium Pro A-step",
700 			  [1] = "Pentium Pro",
701 			  [3] = "Pentium II (Klamath)",
702 			  [4] = "Pentium II (Deschutes)",
703 			  [5] = "Pentium II (Deschutes)",
704 			  [6] = "Mobile Pentium II",
705 			  [7] = "Pentium III (Katmai)",
706 			  [8] = "Pentium III (Coppermine)",
707 			  [10] = "Pentium III (Cascades)",
708 			  [11] = "Pentium III (Tualatin)",
709 		  }
710 		},
711 		{ .vendor = X86_VENDOR_INTEL, .family = 15, .model_names =
712 		  {
713 			  [0] = "Pentium 4 (Unknown)",
714 			  [1] = "Pentium 4 (Willamette)",
715 			  [2] = "Pentium 4 (Northwood)",
716 			  [4] = "Pentium 4 (Foster)",
717 			  [5] = "Pentium 4 (Foster)",
718 		  }
719 		},
720 	},
721 	.c_size_cache	= intel_size_cache,
722 #endif
723 	.c_detect_tlb	= intel_detect_tlb,
724 	.c_early_init   = early_init_intel,
725 	.c_init		= init_intel,
726 	.c_x86_vendor	= X86_VENDOR_INTEL,
727 };
728 
729 cpu_dev_register(intel_cpu_dev);
730 
731