xref: /linux/arch/x86/kernel/cpu/intel.c (revision 98b8788ae91694499d1995035625bea16a4db0c4)
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/ds.h>
16 #include <asm/bugs.h>
17 #include <asm/cpu.h>
18 
19 #ifdef CONFIG_X86_64
20 #include <linux/topology.h>
21 #include <asm/numa_64.h>
22 #endif
23 
24 #include "cpu.h"
25 
26 #ifdef CONFIG_X86_LOCAL_APIC
27 #include <asm/mpspec.h>
28 #include <asm/apic.h>
29 #endif
30 
31 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
32 {
33 	/* Unmask CPUID levels if masked: */
34 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
35 		u64 misc_enable;
36 
37 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
38 
39 		if (misc_enable & MSR_IA32_MISC_ENABLE_LIMIT_CPUID) {
40 			misc_enable &= ~MSR_IA32_MISC_ENABLE_LIMIT_CPUID;
41 			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
42 			c->cpuid_level = cpuid_eax(0);
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 #ifdef CONFIG_X86_64
51 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
52 #else
53 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
54 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
55 		c->x86_cache_alignment = 128;
56 #endif
57 
58 	/* CPUID workaround for 0F33/0F34 CPU */
59 	if (c->x86 == 0xF && c->x86_model == 0x3
60 	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
61 		c->x86_phys_bits = 36;
62 
63 	/*
64 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
65 	 * with P/T states and does not stop in deep C-states.
66 	 *
67 	 * It is also reliable across cores and sockets. (but not across
68 	 * cabinets - we turn it off in that case explicitly.)
69 	 */
70 	if (c->x86_power & (1 << 8)) {
71 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
72 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
73 		set_cpu_cap(c, X86_FEATURE_TSC_RELIABLE);
74 		sched_clock_stable = 1;
75 	}
76 
77 	/*
78 	 * There is a known erratum on Pentium III and Core Solo
79 	 * and Core Duo CPUs.
80 	 * " Page with PAT set to WC while associated MTRR is UC
81 	 *   may consolidate to UC "
82 	 * Because of this erratum, it is better to stick with
83 	 * setting WC in MTRR rather than using PAT on these CPUs.
84 	 *
85 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
86 	 */
87 	if (c->x86 == 6 && c->x86_model < 15)
88 		clear_cpu_cap(c, X86_FEATURE_PAT);
89 
90 #ifdef CONFIG_KMEMCHECK
91 	/*
92 	 * P4s have a "fast strings" feature which causes single-
93 	 * stepping REP instructions to only generate a #DB on
94 	 * cache-line boundaries.
95 	 *
96 	 * Ingo Molnar reported a Pentium D (model 6) and a Xeon
97 	 * (model 2) with the same problem.
98 	 */
99 	if (c->x86 == 15) {
100 		u64 misc_enable;
101 
102 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
103 
104 		if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) {
105 			printk(KERN_INFO "kmemcheck: Disabling fast string operations\n");
106 
107 			misc_enable &= ~MSR_IA32_MISC_ENABLE_FAST_STRING;
108 			wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
109 		}
110 	}
111 #endif
112 }
113 
114 #ifdef CONFIG_X86_32
115 /*
116  *	Early probe support logic for ppro memory erratum #50
117  *
118  *	This is called before we do cpu ident work
119  */
120 
121 int __cpuinit ppro_with_ram_bug(void)
122 {
123 	/* Uses data from early_cpu_detect now */
124 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
125 	    boot_cpu_data.x86 == 6 &&
126 	    boot_cpu_data.x86_model == 1 &&
127 	    boot_cpu_data.x86_mask < 8) {
128 		printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n");
129 		return 1;
130 	}
131 	return 0;
132 }
133 
134 #ifdef CONFIG_X86_F00F_BUG
135 static void __cpuinit trap_init_f00f_bug(void)
136 {
137 	__set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
138 
139 	/*
140 	 * Update the IDT descriptor and reload the IDT so that
141 	 * it uses the read-only mapped virtual address.
142 	 */
143 	idt_descr.address = fix_to_virt(FIX_F00F_IDT);
144 	load_idt(&idt_descr);
145 }
146 #endif
147 
148 static void __cpuinit intel_smp_check(struct cpuinfo_x86 *c)
149 {
150 #ifdef CONFIG_SMP
151 	/* calling is from identify_secondary_cpu() ? */
152 	if (c->cpu_index == boot_cpu_id)
153 		return;
154 
155 	/*
156 	 * Mask B, Pentium, but not Pentium MMX
157 	 */
158 	if (c->x86 == 5 &&
159 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
160 	    c->x86_model <= 3) {
161 		/*
162 		 * Remember we have B step Pentia with bugs
163 		 */
164 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
165 				    "with B stepping processors.\n");
166 	}
167 #endif
168 }
169 
170 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
171 {
172 	unsigned long lo, hi;
173 
174 #ifdef CONFIG_X86_F00F_BUG
175 	/*
176 	 * All current models of Pentium and Pentium with MMX technology CPUs
177 	 * have the F0 0F bug, which lets nonprivileged users lock up the
178 	 * system.
179 	 * Note that the workaround only should be initialized once...
180 	 */
181 	c->f00f_bug = 0;
182 	if (!paravirt_enabled() && c->x86 == 5) {
183 		static int f00f_workaround_enabled;
184 
185 		c->f00f_bug = 1;
186 		if (!f00f_workaround_enabled) {
187 			trap_init_f00f_bug();
188 			printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
189 			f00f_workaround_enabled = 1;
190 		}
191 	}
192 #endif
193 
194 	/*
195 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
196 	 * model 3 mask 3
197 	 */
198 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
199 		clear_cpu_cap(c, X86_FEATURE_SEP);
200 
201 	/*
202 	 * P4 Xeon errata 037 workaround.
203 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
204 	 */
205 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
206 		rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
207 		if ((lo & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE) == 0) {
208 			printk (KERN_INFO "CPU: C0 stepping P4 Xeon detected.\n");
209 			printk (KERN_INFO "CPU: Disabling hardware prefetching (Errata 037)\n");
210 			lo |= MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE;
211 			wrmsr(MSR_IA32_MISC_ENABLE, lo, hi);
212 		}
213 	}
214 
215 	/*
216 	 * See if we have a good local APIC by checking for buggy Pentia,
217 	 * i.e. all B steppings and the C2 stepping of P54C when using their
218 	 * integrated APIC (see 11AP erratum in "Pentium Processor
219 	 * Specification Update").
220 	 */
221 	if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
222 	    (c->x86_mask < 0x6 || c->x86_mask == 0xb))
223 		set_cpu_cap(c, X86_FEATURE_11AP);
224 
225 
226 #ifdef CONFIG_X86_INTEL_USERCOPY
227 	/*
228 	 * Set up the preferred alignment for movsl bulk memory moves
229 	 */
230 	switch (c->x86) {
231 	case 4:		/* 486: untested */
232 		break;
233 	case 5:		/* Old Pentia: untested */
234 		break;
235 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
236 		movsl_mask.mask = 7;
237 		break;
238 	case 15:	/* P4 is OK down to 8-byte alignment */
239 		movsl_mask.mask = 7;
240 		break;
241 	}
242 #endif
243 
244 #ifdef CONFIG_X86_NUMAQ
245 	numaq_tsc_disable();
246 #endif
247 
248 	intel_smp_check(c);
249 }
250 #else
251 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
252 {
253 }
254 #endif
255 
256 static void __cpuinit srat_detect_node(struct cpuinfo_x86 *c)
257 {
258 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
259 	unsigned node;
260 	int cpu = smp_processor_id();
261 	int apicid = cpu_has_apic ? hard_smp_processor_id() : c->apicid;
262 
263 	/* Don't do the funky fallback heuristics the AMD version employs
264 	   for now. */
265 	node = apicid_to_node[apicid];
266 	if (node == NUMA_NO_NODE)
267 		node = first_node(node_online_map);
268 	else if (!node_online(node)) {
269 		/* reuse the value from init_cpu_to_node() */
270 		node = cpu_to_node(cpu);
271 	}
272 	numa_set_node(cpu, node);
273 
274 	printk(KERN_INFO "CPU %d/0x%x -> Node %d\n", cpu, apicid, node);
275 #endif
276 }
277 
278 /*
279  * find out the number of processor cores on the die
280  */
281 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
282 {
283 	unsigned int eax, ebx, ecx, edx;
284 
285 	if (c->cpuid_level < 4)
286 		return 1;
287 
288 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
289 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
290 	if (eax & 0x1f)
291 		return (eax >> 26) + 1;
292 	else
293 		return 1;
294 }
295 
296 static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
297 {
298 	/* Intel VMX MSR indicated features */
299 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
300 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
301 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
302 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
303 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
304 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
305 
306 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
307 
308 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
309 	clear_cpu_cap(c, X86_FEATURE_VNMI);
310 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
311 	clear_cpu_cap(c, X86_FEATURE_EPT);
312 	clear_cpu_cap(c, X86_FEATURE_VPID);
313 
314 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
315 	msr_ctl = vmx_msr_high | vmx_msr_low;
316 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
317 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
318 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
319 		set_cpu_cap(c, X86_FEATURE_VNMI);
320 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
321 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
322 		      vmx_msr_low, vmx_msr_high);
323 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
324 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
325 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
326 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
327 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
328 			set_cpu_cap(c, X86_FEATURE_EPT);
329 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
330 			set_cpu_cap(c, X86_FEATURE_VPID);
331 	}
332 }
333 
334 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
335 {
336 	unsigned int l2 = 0;
337 
338 	early_init_intel(c);
339 
340 	intel_workarounds(c);
341 
342 	/*
343 	 * Detect the extended topology information if available. This
344 	 * will reinitialise the initial_apicid which will be used
345 	 * in init_intel_cacheinfo()
346 	 */
347 	detect_extended_topology(c);
348 
349 	l2 = init_intel_cacheinfo(c);
350 	if (c->cpuid_level > 9) {
351 		unsigned eax = cpuid_eax(10);
352 		/* Check for version and the number of counters */
353 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
354 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
355 	}
356 
357 	if (c->cpuid_level > 6) {
358 		unsigned ecx = cpuid_ecx(6);
359 		if (ecx & 0x01)
360 			set_cpu_cap(c, X86_FEATURE_APERFMPERF);
361 	}
362 
363 	if (cpu_has_xmm2)
364 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
365 	if (cpu_has_ds) {
366 		unsigned int l1;
367 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
368 		if (!(l1 & (1<<11)))
369 			set_cpu_cap(c, X86_FEATURE_BTS);
370 		if (!(l1 & (1<<12)))
371 			set_cpu_cap(c, X86_FEATURE_PEBS);
372 		ds_init_intel(c);
373 	}
374 
375 	if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush)
376 		set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
377 
378 #ifdef CONFIG_X86_64
379 	if (c->x86 == 15)
380 		c->x86_cache_alignment = c->x86_clflush_size * 2;
381 	if (c->x86 == 6)
382 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
383 #else
384 	/*
385 	 * Names for the Pentium II/Celeron processors
386 	 * detectable only by also checking the cache size.
387 	 * Dixon is NOT a Celeron.
388 	 */
389 	if (c->x86 == 6) {
390 		char *p = NULL;
391 
392 		switch (c->x86_model) {
393 		case 5:
394 			if (c->x86_mask == 0) {
395 				if (l2 == 0)
396 					p = "Celeron (Covington)";
397 				else if (l2 == 256)
398 					p = "Mobile Pentium II (Dixon)";
399 			}
400 			break;
401 
402 		case 6:
403 			if (l2 == 128)
404 				p = "Celeron (Mendocino)";
405 			else if (c->x86_mask == 0 || c->x86_mask == 5)
406 				p = "Celeron-A";
407 			break;
408 
409 		case 8:
410 			if (l2 == 128)
411 				p = "Celeron (Coppermine)";
412 			break;
413 		}
414 
415 		if (p)
416 			strcpy(c->x86_model_id, p);
417 	}
418 
419 	if (c->x86 == 15)
420 		set_cpu_cap(c, X86_FEATURE_P4);
421 	if (c->x86 == 6)
422 		set_cpu_cap(c, X86_FEATURE_P3);
423 #endif
424 
425 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
426 		/*
427 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
428 		 * detection.
429 		 */
430 		c->x86_max_cores = intel_num_cpu_cores(c);
431 #ifdef CONFIG_X86_32
432 		detect_ht(c);
433 #endif
434 	}
435 
436 	/* Work around errata */
437 	srat_detect_node(c);
438 
439 	if (cpu_has(c, X86_FEATURE_VMX))
440 		detect_vmx_virtcap(c);
441 }
442 
443 #ifdef CONFIG_X86_32
444 static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
445 {
446 	/*
447 	 * Intel PIII Tualatin. This comes in two flavours.
448 	 * One has 256kb of cache, the other 512. We have no way
449 	 * to determine which, so we use a boottime override
450 	 * for the 512kb model, and assume 256 otherwise.
451 	 */
452 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
453 		size = 256;
454 	return size;
455 }
456 #endif
457 
458 static const struct cpu_dev __cpuinitconst intel_cpu_dev = {
459 	.c_vendor	= "Intel",
460 	.c_ident	= { "GenuineIntel" },
461 #ifdef CONFIG_X86_32
462 	.c_models = {
463 		{ .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
464 		  {
465 			  [0] = "486 DX-25/33",
466 			  [1] = "486 DX-50",
467 			  [2] = "486 SX",
468 			  [3] = "486 DX/2",
469 			  [4] = "486 SL",
470 			  [5] = "486 SX/2",
471 			  [7] = "486 DX/2-WB",
472 			  [8] = "486 DX/4",
473 			  [9] = "486 DX/4-WB"
474 		  }
475 		},
476 		{ .vendor = X86_VENDOR_INTEL, .family = 5, .model_names =
477 		  {
478 			  [0] = "Pentium 60/66 A-step",
479 			  [1] = "Pentium 60/66",
480 			  [2] = "Pentium 75 - 200",
481 			  [3] = "OverDrive PODP5V83",
482 			  [4] = "Pentium MMX",
483 			  [7] = "Mobile Pentium 75 - 200",
484 			  [8] = "Mobile Pentium MMX"
485 		  }
486 		},
487 		{ .vendor = X86_VENDOR_INTEL, .family = 6, .model_names =
488 		  {
489 			  [0] = "Pentium Pro A-step",
490 			  [1] = "Pentium Pro",
491 			  [3] = "Pentium II (Klamath)",
492 			  [4] = "Pentium II (Deschutes)",
493 			  [5] = "Pentium II (Deschutes)",
494 			  [6] = "Mobile Pentium II",
495 			  [7] = "Pentium III (Katmai)",
496 			  [8] = "Pentium III (Coppermine)",
497 			  [10] = "Pentium III (Cascades)",
498 			  [11] = "Pentium III (Tualatin)",
499 		  }
500 		},
501 		{ .vendor = X86_VENDOR_INTEL, .family = 15, .model_names =
502 		  {
503 			  [0] = "Pentium 4 (Unknown)",
504 			  [1] = "Pentium 4 (Willamette)",
505 			  [2] = "Pentium 4 (Northwood)",
506 			  [4] = "Pentium 4 (Foster)",
507 			  [5] = "Pentium 4 (Foster)",
508 		  }
509 		},
510 	},
511 	.c_size_cache	= intel_size_cache,
512 #endif
513 	.c_early_init   = early_init_intel,
514 	.c_init		= init_intel,
515 	.c_x86_vendor	= X86_VENDOR_INTEL,
516 };
517 
518 cpu_dev_register(intel_cpu_dev);
519 
520