xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 78c99ba1)
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 
11 #include <asm/processor.h>
12 #include <asm/pgtable.h>
13 #include <asm/msr.h>
14 #include <asm/uaccess.h>
15 #include <asm/ds.h>
16 #include <asm/bugs.h>
17 #include <asm/cpu.h>
18 
19 #ifdef CONFIG_X86_64
20 #include <asm/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 
91 #ifdef CONFIG_X86_32
92 /*
93  *	Early probe support logic for ppro memory erratum #50
94  *
95  *	This is called before we do cpu ident work
96  */
97 
98 int __cpuinit ppro_with_ram_bug(void)
99 {
100 	/* Uses data from early_cpu_detect now */
101 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
102 	    boot_cpu_data.x86 == 6 &&
103 	    boot_cpu_data.x86_model == 1 &&
104 	    boot_cpu_data.x86_mask < 8) {
105 		printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n");
106 		return 1;
107 	}
108 	return 0;
109 }
110 
111 #ifdef CONFIG_X86_F00F_BUG
112 static void __cpuinit trap_init_f00f_bug(void)
113 {
114 	__set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
115 
116 	/*
117 	 * Update the IDT descriptor and reload the IDT so that
118 	 * it uses the read-only mapped virtual address.
119 	 */
120 	idt_descr.address = fix_to_virt(FIX_F00F_IDT);
121 	load_idt(&idt_descr);
122 }
123 #endif
124 
125 static void __cpuinit intel_smp_check(struct cpuinfo_x86 *c)
126 {
127 #ifdef CONFIG_SMP
128 	/* calling is from identify_secondary_cpu() ? */
129 	if (c->cpu_index == boot_cpu_id)
130 		return;
131 
132 	/*
133 	 * Mask B, Pentium, but not Pentium MMX
134 	 */
135 	if (c->x86 == 5 &&
136 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
137 	    c->x86_model <= 3) {
138 		/*
139 		 * Remember we have B step Pentia with bugs
140 		 */
141 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
142 				    "with B stepping processors.\n");
143 	}
144 #endif
145 }
146 
147 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
148 {
149 	unsigned long lo, hi;
150 
151 #ifdef CONFIG_X86_F00F_BUG
152 	/*
153 	 * All current models of Pentium and Pentium with MMX technology CPUs
154 	 * have the F0 0F bug, which lets nonprivileged users lock up the system.
155 	 * Note that the workaround only should be initialized once...
156 	 */
157 	c->f00f_bug = 0;
158 	if (!paravirt_enabled() && c->x86 == 5) {
159 		static int f00f_workaround_enabled;
160 
161 		c->f00f_bug = 1;
162 		if (!f00f_workaround_enabled) {
163 			trap_init_f00f_bug();
164 			printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
165 			f00f_workaround_enabled = 1;
166 		}
167 	}
168 #endif
169 
170 	/*
171 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
172 	 * model 3 mask 3
173 	 */
174 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
175 		clear_cpu_cap(c, X86_FEATURE_SEP);
176 
177 	/*
178 	 * P4 Xeon errata 037 workaround.
179 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
180 	 */
181 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
182 		rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
183 		if ((lo & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE) == 0) {
184 			printk (KERN_INFO "CPU: C0 stepping P4 Xeon detected.\n");
185 			printk (KERN_INFO "CPU: Disabling hardware prefetching (Errata 037)\n");
186 			lo |= MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE;
187 			wrmsr (MSR_IA32_MISC_ENABLE, lo, hi);
188 		}
189 	}
190 
191 	/*
192 	 * See if we have a good local APIC by checking for buggy Pentia,
193 	 * i.e. all B steppings and the C2 stepping of P54C when using their
194 	 * integrated APIC (see 11AP erratum in "Pentium Processor
195 	 * Specification Update").
196 	 */
197 	if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
198 	    (c->x86_mask < 0x6 || c->x86_mask == 0xb))
199 		set_cpu_cap(c, X86_FEATURE_11AP);
200 
201 
202 #ifdef CONFIG_X86_INTEL_USERCOPY
203 	/*
204 	 * Set up the preferred alignment for movsl bulk memory moves
205 	 */
206 	switch (c->x86) {
207 	case 4:		/* 486: untested */
208 		break;
209 	case 5:		/* Old Pentia: untested */
210 		break;
211 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
212 		movsl_mask.mask = 7;
213 		break;
214 	case 15:	/* P4 is OK down to 8-byte alignment */
215 		movsl_mask.mask = 7;
216 		break;
217 	}
218 #endif
219 
220 #ifdef CONFIG_X86_NUMAQ
221 	numaq_tsc_disable();
222 #endif
223 
224 	intel_smp_check(c);
225 }
226 #else
227 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
228 {
229 }
230 #endif
231 
232 static void __cpuinit srat_detect_node(struct cpuinfo_x86 *c)
233 {
234 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
235 	unsigned node;
236 	int cpu = smp_processor_id();
237 	int apicid = cpu_has_apic ? hard_smp_processor_id() : c->apicid;
238 
239 	/* Don't do the funky fallback heuristics the AMD version employs
240 	   for now. */
241 	node = apicid_to_node[apicid];
242 	if (node == NUMA_NO_NODE || !node_online(node))
243 		node = first_node(node_online_map);
244 	numa_set_node(cpu, node);
245 
246 	printk(KERN_INFO "CPU %d/0x%x -> Node %d\n", cpu, apicid, node);
247 #endif
248 }
249 
250 /*
251  * find out the number of processor cores on the die
252  */
253 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
254 {
255 	unsigned int eax, ebx, ecx, edx;
256 
257 	if (c->cpuid_level < 4)
258 		return 1;
259 
260 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
261 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
262 	if (eax & 0x1f)
263 		return ((eax >> 26) + 1);
264 	else
265 		return 1;
266 }
267 
268 static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
269 {
270 	/* Intel VMX MSR indicated features */
271 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
272 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
273 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
274 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
275 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
276 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
277 
278 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
279 
280 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
281 	clear_cpu_cap(c, X86_FEATURE_VNMI);
282 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
283 	clear_cpu_cap(c, X86_FEATURE_EPT);
284 	clear_cpu_cap(c, X86_FEATURE_VPID);
285 
286 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
287 	msr_ctl = vmx_msr_high | vmx_msr_low;
288 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
289 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
290 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
291 		set_cpu_cap(c, X86_FEATURE_VNMI);
292 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
293 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
294 		      vmx_msr_low, vmx_msr_high);
295 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
296 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
297 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
298 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
299 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
300 			set_cpu_cap(c, X86_FEATURE_EPT);
301 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
302 			set_cpu_cap(c, X86_FEATURE_VPID);
303 	}
304 }
305 
306 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
307 {
308 	unsigned int l2 = 0;
309 
310 	early_init_intel(c);
311 
312 	intel_workarounds(c);
313 
314 	/*
315 	 * Detect the extended topology information if available. This
316 	 * will reinitialise the initial_apicid which will be used
317 	 * in init_intel_cacheinfo()
318 	 */
319 	detect_extended_topology(c);
320 
321 	l2 = init_intel_cacheinfo(c);
322 	if (c->cpuid_level > 9) {
323 		unsigned eax = cpuid_eax(10);
324 		/* Check for version and the number of counters */
325 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
326 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
327 	}
328 
329 	if (cpu_has_xmm2)
330 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
331 	if (cpu_has_ds) {
332 		unsigned int l1;
333 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
334 		if (!(l1 & (1<<11)))
335 			set_cpu_cap(c, X86_FEATURE_BTS);
336 		if (!(l1 & (1<<12)))
337 			set_cpu_cap(c, X86_FEATURE_PEBS);
338 		ds_init_intel(c);
339 	}
340 
341 	if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush)
342 		set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
343 
344 #ifdef CONFIG_X86_64
345 	if (c->x86 == 15)
346 		c->x86_cache_alignment = c->x86_clflush_size * 2;
347 	if (c->x86 == 6)
348 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
349 #else
350 	/*
351 	 * Names for the Pentium II/Celeron processors
352 	 * detectable only by also checking the cache size.
353 	 * Dixon is NOT a Celeron.
354 	 */
355 	if (c->x86 == 6) {
356 		char *p = NULL;
357 
358 		switch (c->x86_model) {
359 		case 5:
360 			if (c->x86_mask == 0) {
361 				if (l2 == 0)
362 					p = "Celeron (Covington)";
363 				else if (l2 == 256)
364 					p = "Mobile Pentium II (Dixon)";
365 			}
366 			break;
367 
368 		case 6:
369 			if (l2 == 128)
370 				p = "Celeron (Mendocino)";
371 			else if (c->x86_mask == 0 || c->x86_mask == 5)
372 				p = "Celeron-A";
373 			break;
374 
375 		case 8:
376 			if (l2 == 128)
377 				p = "Celeron (Coppermine)";
378 			break;
379 		}
380 
381 		if (p)
382 			strcpy(c->x86_model_id, p);
383 	}
384 
385 	if (c->x86 == 15)
386 		set_cpu_cap(c, X86_FEATURE_P4);
387 	if (c->x86 == 6)
388 		set_cpu_cap(c, X86_FEATURE_P3);
389 #endif
390 
391 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
392 		/*
393 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
394 		 * detection.
395 		 */
396 		c->x86_max_cores = intel_num_cpu_cores(c);
397 #ifdef CONFIG_X86_32
398 		detect_ht(c);
399 #endif
400 	}
401 
402 	/* Work around errata */
403 	srat_detect_node(c);
404 
405 	if (cpu_has(c, X86_FEATURE_VMX))
406 		detect_vmx_virtcap(c);
407 }
408 
409 #ifdef CONFIG_X86_32
410 static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
411 {
412 	/*
413 	 * Intel PIII Tualatin. This comes in two flavours.
414 	 * One has 256kb of cache, the other 512. We have no way
415 	 * to determine which, so we use a boottime override
416 	 * for the 512kb model, and assume 256 otherwise.
417 	 */
418 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
419 		size = 256;
420 	return size;
421 }
422 #endif
423 
424 static const struct cpu_dev __cpuinitconst intel_cpu_dev = {
425 	.c_vendor	= "Intel",
426 	.c_ident	= { "GenuineIntel" },
427 #ifdef CONFIG_X86_32
428 	.c_models = {
429 		{ .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
430 		  {
431 			  [0] = "486 DX-25/33",
432 			  [1] = "486 DX-50",
433 			  [2] = "486 SX",
434 			  [3] = "486 DX/2",
435 			  [4] = "486 SL",
436 			  [5] = "486 SX/2",
437 			  [7] = "486 DX/2-WB",
438 			  [8] = "486 DX/4",
439 			  [9] = "486 DX/4-WB"
440 		  }
441 		},
442 		{ .vendor = X86_VENDOR_INTEL, .family = 5, .model_names =
443 		  {
444 			  [0] = "Pentium 60/66 A-step",
445 			  [1] = "Pentium 60/66",
446 			  [2] = "Pentium 75 - 200",
447 			  [3] = "OverDrive PODP5V83",
448 			  [4] = "Pentium MMX",
449 			  [7] = "Mobile Pentium 75 - 200",
450 			  [8] = "Mobile Pentium MMX"
451 		  }
452 		},
453 		{ .vendor = X86_VENDOR_INTEL, .family = 6, .model_names =
454 		  {
455 			  [0] = "Pentium Pro A-step",
456 			  [1] = "Pentium Pro",
457 			  [3] = "Pentium II (Klamath)",
458 			  [4] = "Pentium II (Deschutes)",
459 			  [5] = "Pentium II (Deschutes)",
460 			  [6] = "Mobile Pentium II",
461 			  [7] = "Pentium III (Katmai)",
462 			  [8] = "Pentium III (Coppermine)",
463 			  [10] = "Pentium III (Cascades)",
464 			  [11] = "Pentium III (Tualatin)",
465 		  }
466 		},
467 		{ .vendor = X86_VENDOR_INTEL, .family = 15, .model_names =
468 		  {
469 			  [0] = "Pentium 4 (Unknown)",
470 			  [1] = "Pentium 4 (Willamette)",
471 			  [2] = "Pentium 4 (Northwood)",
472 			  [4] = "Pentium 4 (Foster)",
473 			  [5] = "Pentium 4 (Foster)",
474 		  }
475 		},
476 	},
477 	.c_size_cache	= intel_size_cache,
478 #endif
479 	.c_early_init   = early_init_intel,
480 	.c_init		= init_intel,
481 	.c_x86_vendor	= X86_VENDOR_INTEL,
482 };
483 
484 cpu_dev_register(intel_cpu_dev);
485 
486