]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - arch/x86/kernel/cpu/intel.c
Merge branch 'x86/core' into tracing/ftrace
[linux-2.6-omap-h63xx.git] / arch / x86 / kernel / cpu / intel.c
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         /*
59          * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
60          * with P/T states and does not stop in deep C-states.
61          *
62          * It is also reliable across cores and sockets. (but not across
63          * cabinets - we turn it off in that case explicitly.)
64          */
65         if (c->x86_power & (1 << 8)) {
66                 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
67                 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
68                 set_cpu_cap(c, X86_FEATURE_TSC_RELIABLE);
69                 sched_clock_stable = 1;
70         }
71
72         /*
73          * There is a known erratum on Pentium III and Core Solo
74          * and Core Duo CPUs.
75          * " Page with PAT set to WC while associated MTRR is UC
76          *   may consolidate to UC "
77          * Because of this erratum, it is better to stick with
78          * setting WC in MTRR rather than using PAT on these CPUs.
79          *
80          * Enable PAT WC only on P4, Core 2 or later CPUs.
81          */
82         if (c->x86 == 6 && c->x86_model < 15)
83                 clear_cpu_cap(c, X86_FEATURE_PAT);
84 }
85
86 #ifdef CONFIG_X86_32
87 /*
88  *      Early probe support logic for ppro memory erratum #50
89  *
90  *      This is called before we do cpu ident work
91  */
92
93 int __cpuinit ppro_with_ram_bug(void)
94 {
95         /* Uses data from early_cpu_detect now */
96         if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
97             boot_cpu_data.x86 == 6 &&
98             boot_cpu_data.x86_model == 1 &&
99             boot_cpu_data.x86_mask < 8) {
100                 printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n");
101                 return 1;
102         }
103         return 0;
104 }
105
106 #ifdef CONFIG_X86_F00F_BUG
107 static void __cpuinit trap_init_f00f_bug(void)
108 {
109         __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
110
111         /*
112          * Update the IDT descriptor and reload the IDT so that
113          * it uses the read-only mapped virtual address.
114          */
115         idt_descr.address = fix_to_virt(FIX_F00F_IDT);
116         load_idt(&idt_descr);
117 }
118 #endif
119
120 static void __cpuinit intel_smp_check(struct cpuinfo_x86 *c)
121 {
122 #ifdef CONFIG_SMP
123         /* calling is from identify_secondary_cpu() ? */
124         if (c->cpu_index == boot_cpu_id)
125                 return;
126
127         /*
128          * Mask B, Pentium, but not Pentium MMX
129          */
130         if (c->x86 == 5 &&
131             c->x86_mask >= 1 && c->x86_mask <= 4 &&
132             c->x86_model <= 3) {
133                 /*
134                  * Remember we have B step Pentia with bugs
135                  */
136                 WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
137                                     "with B stepping processors.\n");
138         }
139 #endif
140 }
141
142 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
143 {
144         unsigned long lo, hi;
145
146 #ifdef CONFIG_X86_F00F_BUG
147         /*
148          * All current models of Pentium and Pentium with MMX technology CPUs
149          * have the F0 0F bug, which lets nonprivileged users lock up the system.
150          * Note that the workaround only should be initialized once...
151          */
152         c->f00f_bug = 0;
153         if (!paravirt_enabled() && c->x86 == 5) {
154                 static int f00f_workaround_enabled;
155
156                 c->f00f_bug = 1;
157                 if (!f00f_workaround_enabled) {
158                         trap_init_f00f_bug();
159                         printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
160                         f00f_workaround_enabled = 1;
161                 }
162         }
163 #endif
164
165         /*
166          * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
167          * model 3 mask 3
168          */
169         if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
170                 clear_cpu_cap(c, X86_FEATURE_SEP);
171
172         /*
173          * P4 Xeon errata 037 workaround.
174          * Hardware prefetcher may cause stale data to be loaded into the cache.
175          */
176         if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
177                 rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
178                 if ((lo & MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE) == 0) {
179                         printk (KERN_INFO "CPU: C0 stepping P4 Xeon detected.\n");
180                         printk (KERN_INFO "CPU: Disabling hardware prefetching (Errata 037)\n");
181                         lo |= MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE;
182                         wrmsr (MSR_IA32_MISC_ENABLE, lo, hi);
183                 }
184         }
185
186         /*
187          * See if we have a good local APIC by checking for buggy Pentia,
188          * i.e. all B steppings and the C2 stepping of P54C when using their
189          * integrated APIC (see 11AP erratum in "Pentium Processor
190          * Specification Update").
191          */
192         if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
193             (c->x86_mask < 0x6 || c->x86_mask == 0xb))
194                 set_cpu_cap(c, X86_FEATURE_11AP);
195
196
197 #ifdef CONFIG_X86_INTEL_USERCOPY
198         /*
199          * Set up the preferred alignment for movsl bulk memory moves
200          */
201         switch (c->x86) {
202         case 4:         /* 486: untested */
203                 break;
204         case 5:         /* Old Pentia: untested */
205                 break;
206         case 6:         /* PII/PIII only like movsl with 8-byte alignment */
207                 movsl_mask.mask = 7;
208                 break;
209         case 15:        /* P4 is OK down to 8-byte alignment */
210                 movsl_mask.mask = 7;
211                 break;
212         }
213 #endif
214
215 #ifdef CONFIG_X86_NUMAQ
216         numaq_tsc_disable();
217 #endif
218
219         intel_smp_check(c);
220 }
221 #else
222 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
223 {
224 }
225 #endif
226
227 static void __cpuinit srat_detect_node(void)
228 {
229 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
230         unsigned node;
231         int cpu = smp_processor_id();
232         int apicid = hard_smp_processor_id();
233
234         /* Don't do the funky fallback heuristics the AMD version employs
235            for now. */
236         node = apicid_to_node[apicid];
237         if (node == NUMA_NO_NODE || !node_online(node))
238                 node = first_node(node_online_map);
239         numa_set_node(cpu, node);
240
241         printk(KERN_INFO "CPU %d/0x%x -> Node %d\n", cpu, apicid, node);
242 #endif
243 }
244
245 /*
246  * find out the number of processor cores on the die
247  */
248 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
249 {
250         unsigned int eax, ebx, ecx, edx;
251
252         if (c->cpuid_level < 4)
253                 return 1;
254
255         /* Intel has a non-standard dependency on %ecx for this CPUID level. */
256         cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
257         if (eax & 0x1f)
258                 return ((eax >> 26) + 1);
259         else
260                 return 1;
261 }
262
263 static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
264 {
265         /* Intel VMX MSR indicated features */
266 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW    0x00200000
267 #define X86_VMX_FEATURE_PROC_CTLS_VNMI          0x00400000
268 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS      0x80000000
269 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC    0x00000001
270 #define X86_VMX_FEATURE_PROC_CTLS2_EPT          0x00000002
271 #define X86_VMX_FEATURE_PROC_CTLS2_VPID         0x00000020
272
273         u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
274
275         clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
276         clear_cpu_cap(c, X86_FEATURE_VNMI);
277         clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
278         clear_cpu_cap(c, X86_FEATURE_EPT);
279         clear_cpu_cap(c, X86_FEATURE_VPID);
280
281         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
282         msr_ctl = vmx_msr_high | vmx_msr_low;
283         if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
284                 set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
285         if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
286                 set_cpu_cap(c, X86_FEATURE_VNMI);
287         if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
288                 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
289                       vmx_msr_low, vmx_msr_high);
290                 msr_ctl2 = vmx_msr_high | vmx_msr_low;
291                 if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
292                     (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
293                         set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
294                 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
295                         set_cpu_cap(c, X86_FEATURE_EPT);
296                 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
297                         set_cpu_cap(c, X86_FEATURE_VPID);
298         }
299 }
300
301 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
302 {
303         unsigned int l2 = 0;
304
305         early_init_intel(c);
306
307         intel_workarounds(c);
308
309         /*
310          * Detect the extended topology information if available. This
311          * will reinitialise the initial_apicid which will be used
312          * in init_intel_cacheinfo()
313          */
314         detect_extended_topology(c);
315
316         l2 = init_intel_cacheinfo(c);
317         if (c->cpuid_level > 9) {
318                 unsigned eax = cpuid_eax(10);
319                 /* Check for version and the number of counters */
320                 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
321                         set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
322         }
323
324         if (cpu_has_xmm2)
325                 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
326         if (cpu_has_ds) {
327                 unsigned int l1;
328                 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
329                 if (!(l1 & (1<<11)))
330                         set_cpu_cap(c, X86_FEATURE_BTS);
331                 if (!(l1 & (1<<12)))
332                         set_cpu_cap(c, X86_FEATURE_PEBS);
333                 ds_init_intel(c);
334         }
335
336         if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush)
337                 set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
338
339 #ifdef CONFIG_X86_64
340         if (c->x86 == 15)
341                 c->x86_cache_alignment = c->x86_clflush_size * 2;
342         if (c->x86 == 6)
343                 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
344 #else
345         /*
346          * Names for the Pentium II/Celeron processors
347          * detectable only by also checking the cache size.
348          * Dixon is NOT a Celeron.
349          */
350         if (c->x86 == 6) {
351                 char *p = NULL;
352
353                 switch (c->x86_model) {
354                 case 5:
355                         if (c->x86_mask == 0) {
356                                 if (l2 == 0)
357                                         p = "Celeron (Covington)";
358                                 else if (l2 == 256)
359                                         p = "Mobile Pentium II (Dixon)";
360                         }
361                         break;
362
363                 case 6:
364                         if (l2 == 128)
365                                 p = "Celeron (Mendocino)";
366                         else if (c->x86_mask == 0 || c->x86_mask == 5)
367                                 p = "Celeron-A";
368                         break;
369
370                 case 8:
371                         if (l2 == 128)
372                                 p = "Celeron (Coppermine)";
373                         break;
374                 }
375
376                 if (p)
377                         strcpy(c->x86_model_id, p);
378         }
379
380         if (c->x86 == 15)
381                 set_cpu_cap(c, X86_FEATURE_P4);
382         if (c->x86 == 6)
383                 set_cpu_cap(c, X86_FEATURE_P3);
384 #endif
385
386         if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
387                 /*
388                  * let's use the legacy cpuid vector 0x1 and 0x4 for topology
389                  * detection.
390                  */
391                 c->x86_max_cores = intel_num_cpu_cores(c);
392 #ifdef CONFIG_X86_32
393                 detect_ht(c);
394 #endif
395         }
396
397         /* Work around errata */
398         srat_detect_node();
399
400         if (cpu_has(c, X86_FEATURE_VMX))
401                 detect_vmx_virtcap(c);
402 }
403
404 #ifdef CONFIG_X86_32
405 static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
406 {
407         /*
408          * Intel PIII Tualatin. This comes in two flavours.
409          * One has 256kb of cache, the other 512. We have no way
410          * to determine which, so we use a boottime override
411          * for the 512kb model, and assume 256 otherwise.
412          */
413         if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
414                 size = 256;
415         return size;
416 }
417 #endif
418
419 static struct cpu_dev intel_cpu_dev __cpuinitdata = {
420         .c_vendor       = "Intel",
421         .c_ident        = { "GenuineIntel" },
422 #ifdef CONFIG_X86_32
423         .c_models = {
424                 { .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
425                   {
426                           [0] = "486 DX-25/33",
427                           [1] = "486 DX-50",
428                           [2] = "486 SX",
429                           [3] = "486 DX/2",
430                           [4] = "486 SL",
431                           [5] = "486 SX/2",
432                           [7] = "486 DX/2-WB",
433                           [8] = "486 DX/4",
434                           [9] = "486 DX/4-WB"
435                   }
436                 },
437                 { .vendor = X86_VENDOR_INTEL, .family = 5, .model_names =
438                   {
439                           [0] = "Pentium 60/66 A-step",
440                           [1] = "Pentium 60/66",
441                           [2] = "Pentium 75 - 200",
442                           [3] = "OverDrive PODP5V83",
443                           [4] = "Pentium MMX",
444                           [7] = "Mobile Pentium 75 - 200",
445                           [8] = "Mobile Pentium MMX"
446                   }
447                 },
448                 { .vendor = X86_VENDOR_INTEL, .family = 6, .model_names =
449                   {
450                           [0] = "Pentium Pro A-step",
451                           [1] = "Pentium Pro",
452                           [3] = "Pentium II (Klamath)",
453                           [4] = "Pentium II (Deschutes)",
454                           [5] = "Pentium II (Deschutes)",
455                           [6] = "Mobile Pentium II",
456                           [7] = "Pentium III (Katmai)",
457                           [8] = "Pentium III (Coppermine)",
458                           [10] = "Pentium III (Cascades)",
459                           [11] = "Pentium III (Tualatin)",
460                   }
461                 },
462                 { .vendor = X86_VENDOR_INTEL, .family = 15, .model_names =
463                   {
464                           [0] = "Pentium 4 (Unknown)",
465                           [1] = "Pentium 4 (Willamette)",
466                           [2] = "Pentium 4 (Northwood)",
467                           [4] = "Pentium 4 (Foster)",
468                           [5] = "Pentium 4 (Foster)",
469                   }
470                 },
471         },
472         .c_size_cache   = intel_size_cache,
473 #endif
474         .c_early_init   = early_init_intel,
475         .c_init         = init_intel,
476         .c_x86_vendor   = X86_VENDOR_INTEL,
477 };
478
479 cpu_dev_register(intel_cpu_dev);
480