xref: /linux/arch/x86/kernel/cpu/intel.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
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 #include <asm/numa_64.h>
21 #endif
22 
23 #include "cpu.h"
24 
25 #ifdef CONFIG_X86_LOCAL_APIC
26 #include <asm/mpspec.h>
27 #include <asm/apic.h>
28 #endif
29 
30 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
31 {
32 	u64 misc_enable;
33 
34 	/* Unmask CPUID levels if masked: */
35 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
36 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
37 
38 		if (misc_enable & MSR_IA32_MISC_ENABLE_LIMIT_CPUID) {
39 			misc_enable &= ~MSR_IA32_MISC_ENABLE_LIMIT_CPUID;
40 			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
41 			c->cpuid_level = cpuid_eax(0);
42 			get_cpu_cap(c);
43 		}
44 	}
45 
46 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
47 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
48 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
49 
50 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) {
51 		unsigned lower_word;
52 
53 		wrmsr(MSR_IA32_UCODE_REV, 0, 0);
54 		/* Required by the SDM */
55 		sync_core();
56 		rdmsr(MSR_IA32_UCODE_REV, lower_word, c->microcode);
57 	}
58 
59 	/*
60 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
61 	 *
62 	 * A race condition between speculative fetches and invalidating
63 	 * a large page.  This is worked around in microcode, but we
64 	 * need the microcode to have already been loaded... so if it is
65 	 * not, recommend a BIOS update and disable large pages.
66 	 */
67 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 &&
68 	    c->microcode < 0x20e) {
69 		printk(KERN_WARNING "Atom PSE erratum detected, BIOS microcode update recommended\n");
70 		clear_cpu_cap(c, X86_FEATURE_PSE);
71 	}
72 
73 #ifdef CONFIG_X86_64
74 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
75 #else
76 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
77 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
78 		c->x86_cache_alignment = 128;
79 #endif
80 
81 	/* CPUID workaround for 0F33/0F34 CPU */
82 	if (c->x86 == 0xF && c->x86_model == 0x3
83 	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
84 		c->x86_phys_bits = 36;
85 
86 	/*
87 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
88 	 * with P/T states and does not stop in deep C-states.
89 	 *
90 	 * It is also reliable across cores and sockets. (but not across
91 	 * cabinets - we turn it off in that case explicitly.)
92 	 */
93 	if (c->x86_power & (1 << 8)) {
94 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
95 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
96 		if (!check_tsc_unstable())
97 			sched_clock_stable = 1;
98 	}
99 
100 	/*
101 	 * There is a known erratum on Pentium III and Core Solo
102 	 * and Core Duo CPUs.
103 	 * " Page with PAT set to WC while associated MTRR is UC
104 	 *   may consolidate to UC "
105 	 * Because of this erratum, it is better to stick with
106 	 * setting WC in MTRR rather than using PAT on these CPUs.
107 	 *
108 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
109 	 */
110 	if (c->x86 == 6 && c->x86_model < 15)
111 		clear_cpu_cap(c, X86_FEATURE_PAT);
112 
113 #ifdef CONFIG_KMEMCHECK
114 	/*
115 	 * P4s have a "fast strings" feature which causes single-
116 	 * stepping REP instructions to only generate a #DB on
117 	 * cache-line boundaries.
118 	 *
119 	 * Ingo Molnar reported a Pentium D (model 6) and a Xeon
120 	 * (model 2) with the same problem.
121 	 */
122 	if (c->x86 == 15) {
123 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
124 
125 		if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) {
126 			printk(KERN_INFO "kmemcheck: Disabling fast string operations\n");
127 
128 			misc_enable &= ~MSR_IA32_MISC_ENABLE_FAST_STRING;
129 			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
130 		}
131 	}
132 #endif
133 
134 	/*
135 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
136 	 * clear the fast string and enhanced fast string CPU capabilities.
137 	 */
138 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
139 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
140 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
141 			printk(KERN_INFO "Disabled fast string operations\n");
142 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
143 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
144 		}
145 	}
146 }
147 
148 #ifdef CONFIG_X86_32
149 /*
150  *	Early probe support logic for ppro memory erratum #50
151  *
152  *	This is called before we do cpu ident work
153  */
154 
155 int __cpuinit ppro_with_ram_bug(void)
156 {
157 	/* Uses data from early_cpu_detect now */
158 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
159 	    boot_cpu_data.x86 == 6 &&
160 	    boot_cpu_data.x86_model == 1 &&
161 	    boot_cpu_data.x86_mask < 8) {
162 		printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n");
163 		return 1;
164 	}
165 	return 0;
166 }
167 
168 #ifdef CONFIG_X86_F00F_BUG
169 static void __cpuinit trap_init_f00f_bug(void)
170 {
171 	__set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
172 
173 	/*
174 	 * Update the IDT descriptor and reload the IDT so that
175 	 * it uses the read-only mapped virtual address.
176 	 */
177 	idt_descr.address = fix_to_virt(FIX_F00F_IDT);
178 	load_idt(&idt_descr);
179 }
180 #endif
181 
182 static void __cpuinit intel_smp_check(struct cpuinfo_x86 *c)
183 {
184 #ifdef CONFIG_SMP
185 	/* calling is from identify_secondary_cpu() ? */
186 	if (!c->cpu_index)
187 		return;
188 
189 	/*
190 	 * Mask B, Pentium, but not Pentium MMX
191 	 */
192 	if (c->x86 == 5 &&
193 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
194 	    c->x86_model <= 3) {
195 		/*
196 		 * Remember we have B step Pentia with bugs
197 		 */
198 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
199 				    "with B stepping processors.\n");
200 	}
201 #endif
202 }
203 
204 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
205 {
206 	unsigned long lo, hi;
207 
208 #ifdef CONFIG_X86_F00F_BUG
209 	/*
210 	 * All current models of Pentium and Pentium with MMX technology CPUs
211 	 * have the F0 0F bug, which lets nonprivileged users lock up the
212 	 * system.
213 	 * Note that the workaround only should be initialized once...
214 	 */
215 	c->f00f_bug = 0;
216 	if (!paravirt_enabled() && c->x86 == 5) {
217 		static int f00f_workaround_enabled;
218 
219 		c->f00f_bug = 1;
220 		if (!f00f_workaround_enabled) {
221 			trap_init_f00f_bug();
222 			printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
223 			f00f_workaround_enabled = 1;
224 		}
225 	}
226 #endif
227 
228 	/*
229 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
230 	 * model 3 mask 3
231 	 */
232 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
233 		clear_cpu_cap(c, X86_FEATURE_SEP);
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 		rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
241 		if ((lo & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE) == 0) {
242 			printk (KERN_INFO "CPU: C0 stepping P4 Xeon detected.\n");
243 			printk (KERN_INFO "CPU: Disabling hardware prefetching (Errata 037)\n");
244 			lo |= MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE;
245 			wrmsr(MSR_IA32_MISC_ENABLE, lo, hi);
246 		}
247 	}
248 
249 	/*
250 	 * See if we have a good local APIC by checking for buggy Pentia,
251 	 * i.e. all B steppings and the C2 stepping of P54C when using their
252 	 * integrated APIC (see 11AP erratum in "Pentium Processor
253 	 * Specification Update").
254 	 */
255 	if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
256 	    (c->x86_mask < 0x6 || c->x86_mask == 0xb))
257 		set_cpu_cap(c, X86_FEATURE_11AP);
258 
259 
260 #ifdef CONFIG_X86_INTEL_USERCOPY
261 	/*
262 	 * Set up the preferred alignment for movsl bulk memory moves
263 	 */
264 	switch (c->x86) {
265 	case 4:		/* 486: untested */
266 		break;
267 	case 5:		/* Old Pentia: untested */
268 		break;
269 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
270 		movsl_mask.mask = 7;
271 		break;
272 	case 15:	/* P4 is OK down to 8-byte alignment */
273 		movsl_mask.mask = 7;
274 		break;
275 	}
276 #endif
277 
278 #ifdef CONFIG_X86_NUMAQ
279 	numaq_tsc_disable();
280 #endif
281 
282 	intel_smp_check(c);
283 }
284 #else
285 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
286 {
287 }
288 #endif
289 
290 static void __cpuinit srat_detect_node(struct cpuinfo_x86 *c)
291 {
292 #ifdef CONFIG_NUMA
293 	unsigned node;
294 	int cpu = smp_processor_id();
295 
296 	/* Don't do the funky fallback heuristics the AMD version employs
297 	   for now. */
298 	node = numa_cpu_node(cpu);
299 	if (node == NUMA_NO_NODE || !node_online(node)) {
300 		/* reuse the value from init_cpu_to_node() */
301 		node = cpu_to_node(cpu);
302 	}
303 	numa_set_node(cpu, node);
304 #endif
305 }
306 
307 /*
308  * find out the number of processor cores on the die
309  */
310 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
311 {
312 	unsigned int eax, ebx, ecx, edx;
313 
314 	if (c->cpuid_level < 4)
315 		return 1;
316 
317 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
318 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
319 	if (eax & 0x1f)
320 		return (eax >> 26) + 1;
321 	else
322 		return 1;
323 }
324 
325 static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
326 {
327 	/* Intel VMX MSR indicated features */
328 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
329 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
330 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
331 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
332 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
333 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
334 
335 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
336 
337 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
338 	clear_cpu_cap(c, X86_FEATURE_VNMI);
339 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
340 	clear_cpu_cap(c, X86_FEATURE_EPT);
341 	clear_cpu_cap(c, X86_FEATURE_VPID);
342 
343 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
344 	msr_ctl = vmx_msr_high | vmx_msr_low;
345 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
346 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
347 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
348 		set_cpu_cap(c, X86_FEATURE_VNMI);
349 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
350 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
351 		      vmx_msr_low, vmx_msr_high);
352 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
353 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
354 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
355 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
356 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
357 			set_cpu_cap(c, X86_FEATURE_EPT);
358 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
359 			set_cpu_cap(c, X86_FEATURE_VPID);
360 	}
361 }
362 
363 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
364 {
365 	unsigned int l2 = 0;
366 
367 	early_init_intel(c);
368 
369 	intel_workarounds(c);
370 
371 	/*
372 	 * Detect the extended topology information if available. This
373 	 * will reinitialise the initial_apicid which will be used
374 	 * in init_intel_cacheinfo()
375 	 */
376 	detect_extended_topology(c);
377 
378 	l2 = init_intel_cacheinfo(c);
379 	if (c->cpuid_level > 9) {
380 		unsigned eax = cpuid_eax(10);
381 		/* Check for version and the number of counters */
382 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
383 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
384 	}
385 
386 	if (cpu_has_xmm2)
387 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
388 	if (cpu_has_ds) {
389 		unsigned int l1;
390 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
391 		if (!(l1 & (1<<11)))
392 			set_cpu_cap(c, X86_FEATURE_BTS);
393 		if (!(l1 & (1<<12)))
394 			set_cpu_cap(c, X86_FEATURE_PEBS);
395 	}
396 
397 	if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush)
398 		set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
399 
400 #ifdef CONFIG_X86_64
401 	if (c->x86 == 15)
402 		c->x86_cache_alignment = c->x86_clflush_size * 2;
403 	if (c->x86 == 6)
404 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
405 #else
406 	/*
407 	 * Names for the Pentium II/Celeron processors
408 	 * detectable only by also checking the cache size.
409 	 * Dixon is NOT a Celeron.
410 	 */
411 	if (c->x86 == 6) {
412 		char *p = NULL;
413 
414 		switch (c->x86_model) {
415 		case 5:
416 			if (l2 == 0)
417 				p = "Celeron (Covington)";
418 			else if (l2 == 256)
419 				p = "Mobile Pentium II (Dixon)";
420 			break;
421 
422 		case 6:
423 			if (l2 == 128)
424 				p = "Celeron (Mendocino)";
425 			else if (c->x86_mask == 0 || c->x86_mask == 5)
426 				p = "Celeron-A";
427 			break;
428 
429 		case 8:
430 			if (l2 == 128)
431 				p = "Celeron (Coppermine)";
432 			break;
433 		}
434 
435 		if (p)
436 			strcpy(c->x86_model_id, p);
437 	}
438 
439 	if (c->x86 == 15)
440 		set_cpu_cap(c, X86_FEATURE_P4);
441 	if (c->x86 == 6)
442 		set_cpu_cap(c, X86_FEATURE_P3);
443 #endif
444 
445 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
446 		/*
447 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
448 		 * detection.
449 		 */
450 		c->x86_max_cores = intel_num_cpu_cores(c);
451 #ifdef CONFIG_X86_32
452 		detect_ht(c);
453 #endif
454 	}
455 
456 	/* Work around errata */
457 	srat_detect_node(c);
458 
459 	if (cpu_has(c, X86_FEATURE_VMX))
460 		detect_vmx_virtcap(c);
461 
462 	/*
463 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if BIOS did not.
464 	 * x86_energy_perf_policy(8) is available to change it at run-time
465 	 */
466 	if (cpu_has(c, X86_FEATURE_EPB)) {
467 		u64 epb;
468 
469 		rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
470 		if ((epb & 0xF) == ENERGY_PERF_BIAS_PERFORMANCE) {
471 			printk_once(KERN_WARNING "ENERGY_PERF_BIAS:"
472 				" Set to 'normal', was 'performance'\n"
473 				"ENERGY_PERF_BIAS: View and update with"
474 				" x86_energy_perf_policy(8)\n");
475 			epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
476 			wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
477 		}
478 	}
479 }
480 
481 #ifdef CONFIG_X86_32
482 static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
483 {
484 	/*
485 	 * Intel PIII Tualatin. This comes in two flavours.
486 	 * One has 256kb of cache, the other 512. We have no way
487 	 * to determine which, so we use a boottime override
488 	 * for the 512kb model, and assume 256 otherwise.
489 	 */
490 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
491 		size = 256;
492 	return size;
493 }
494 #endif
495 
496 static const struct cpu_dev __cpuinitconst intel_cpu_dev = {
497 	.c_vendor	= "Intel",
498 	.c_ident	= { "GenuineIntel" },
499 #ifdef CONFIG_X86_32
500 	.c_models = {
501 		{ .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
502 		  {
503 			  [0] = "486 DX-25/33",
504 			  [1] = "486 DX-50",
505 			  [2] = "486 SX",
506 			  [3] = "486 DX/2",
507 			  [4] = "486 SL",
508 			  [5] = "486 SX/2",
509 			  [7] = "486 DX/2-WB",
510 			  [8] = "486 DX/4",
511 			  [9] = "486 DX/4-WB"
512 		  }
513 		},
514 		{ .vendor = X86_VENDOR_INTEL, .family = 5, .model_names =
515 		  {
516 			  [0] = "Pentium 60/66 A-step",
517 			  [1] = "Pentium 60/66",
518 			  [2] = "Pentium 75 - 200",
519 			  [3] = "OverDrive PODP5V83",
520 			  [4] = "Pentium MMX",
521 			  [7] = "Mobile Pentium 75 - 200",
522 			  [8] = "Mobile Pentium MMX"
523 		  }
524 		},
525 		{ .vendor = X86_VENDOR_INTEL, .family = 6, .model_names =
526 		  {
527 			  [0] = "Pentium Pro A-step",
528 			  [1] = "Pentium Pro",
529 			  [3] = "Pentium II (Klamath)",
530 			  [4] = "Pentium II (Deschutes)",
531 			  [5] = "Pentium II (Deschutes)",
532 			  [6] = "Mobile Pentium II",
533 			  [7] = "Pentium III (Katmai)",
534 			  [8] = "Pentium III (Coppermine)",
535 			  [10] = "Pentium III (Cascades)",
536 			  [11] = "Pentium III (Tualatin)",
537 		  }
538 		},
539 		{ .vendor = X86_VENDOR_INTEL, .family = 15, .model_names =
540 		  {
541 			  [0] = "Pentium 4 (Unknown)",
542 			  [1] = "Pentium 4 (Willamette)",
543 			  [2] = "Pentium 4 (Northwood)",
544 			  [4] = "Pentium 4 (Foster)",
545 			  [5] = "Pentium 4 (Foster)",
546 		  }
547 		},
548 	},
549 	.c_size_cache	= intel_size_cache,
550 #endif
551 	.c_early_init   = early_init_intel,
552 	.c_init		= init_intel,
553 	.c_x86_vendor	= X86_VENDOR_INTEL,
554 };
555 
556 cpu_dev_register(intel_cpu_dev);
557 
558