x86: set X86_FEATURE_TSC_RELIABLE
[linux-flexiantxendom0-natty.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
18 #ifdef CONFIG_X86_64
19 #include <asm/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 #include <mach_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
74 #ifdef CONFIG_X86_32
75 /*
76  *      Early probe support logic for ppro memory erratum #50
77  *
78  *      This is called before we do cpu ident work
79  */
80
81 int __cpuinit ppro_with_ram_bug(void)
82 {
83         /* Uses data from early_cpu_detect now */
84         if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
85             boot_cpu_data.x86 == 6 &&
86             boot_cpu_data.x86_model == 1 &&
87             boot_cpu_data.x86_mask < 8) {
88                 printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n");
89                 return 1;
90         }
91         return 0;
92 }
93
94 #ifdef CONFIG_X86_F00F_BUG
95 static void __cpuinit trap_init_f00f_bug(void)
96 {
97         __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
98
99         /*
100          * Update the IDT descriptor and reload the IDT so that
101          * it uses the read-only mapped virtual address.
102          */
103         idt_descr.address = fix_to_virt(FIX_F00F_IDT);
104         load_idt(&idt_descr);
105 }
106 #endif
107
108 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
109 {
110         unsigned long lo, hi;
111
112 #ifdef CONFIG_X86_F00F_BUG
113         /*
114          * All current models of Pentium and Pentium with MMX technology CPUs
115          * have the F0 0F bug, which lets nonprivileged users lock up the system.
116          * Note that the workaround only should be initialized once...
117          */
118         c->f00f_bug = 0;
119         if (!paravirt_enabled() && c->x86 == 5) {
120                 static int f00f_workaround_enabled;
121
122                 c->f00f_bug = 1;
123                 if (!f00f_workaround_enabled) {
124                         trap_init_f00f_bug();
125                         printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
126                         f00f_workaround_enabled = 1;
127                 }
128         }
129 #endif
130
131         /*
132          * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
133          * model 3 mask 3
134          */
135         if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
136                 clear_cpu_cap(c, X86_FEATURE_SEP);
137
138         /*
139          * P4 Xeon errata 037 workaround.
140          * Hardware prefetcher may cause stale data to be loaded into the cache.
141          */
142         if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
143                 rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
144                 if ((lo & (1<<9)) == 0) {
145                         printk (KERN_INFO "CPU: C0 stepping P4 Xeon detected.\n");
146                         printk (KERN_INFO "CPU: Disabling hardware prefetching (Errata 037)\n");
147                         lo |= (1<<9);   /* Disable hw prefetching */
148                         wrmsr (MSR_IA32_MISC_ENABLE, lo, hi);
149                 }
150         }
151
152         /*
153          * See if we have a good local APIC by checking for buggy Pentia,
154          * i.e. all B steppings and the C2 stepping of P54C when using their
155          * integrated APIC (see 11AP erratum in "Pentium Processor
156          * Specification Update").
157          */
158         if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
159             (c->x86_mask < 0x6 || c->x86_mask == 0xb))
160                 set_cpu_cap(c, X86_FEATURE_11AP);
161
162
163 #ifdef CONFIG_X86_INTEL_USERCOPY
164         /*
165          * Set up the preferred alignment for movsl bulk memory moves
166          */
167         switch (c->x86) {
168         case 4:         /* 486: untested */
169                 break;
170         case 5:         /* Old Pentia: untested */
171                 break;
172         case 6:         /* PII/PIII only like movsl with 8-byte alignment */
173                 movsl_mask.mask = 7;
174                 break;
175         case 15:        /* P4 is OK down to 8-byte alignment */
176                 movsl_mask.mask = 7;
177                 break;
178         }
179 #endif
180
181 #ifdef CONFIG_X86_NUMAQ
182         numaq_tsc_disable();
183 #endif
184 }
185 #else
186 static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
187 {
188 }
189 #endif
190
191 static void __cpuinit srat_detect_node(void)
192 {
193 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
194         unsigned node;
195         int cpu = smp_processor_id();
196         int apicid = hard_smp_processor_id();
197
198         /* Don't do the funky fallback heuristics the AMD version employs
199            for now. */
200         node = apicid_to_node[apicid];
201         if (node == NUMA_NO_NODE || !node_online(node))
202                 node = first_node(node_online_map);
203         numa_set_node(cpu, node);
204
205         printk(KERN_INFO "CPU %d/0x%x -> Node %d\n", cpu, apicid, node);
206 #endif
207 }
208
209 /*
210  * find out the number of processor cores on the die
211  */
212 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
213 {
214         unsigned int eax, ebx, ecx, edx;
215
216         if (c->cpuid_level < 4)
217                 return 1;
218
219         /* Intel has a non-standard dependency on %ecx for this CPUID level. */
220         cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
221         if (eax & 0x1f)
222                 return ((eax >> 26) + 1);
223         else
224                 return 1;
225 }
226
227 static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
228 {
229         /* Intel VMX MSR indicated features */
230 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW    0x00200000
231 #define X86_VMX_FEATURE_PROC_CTLS_VNMI          0x00400000
232 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS      0x80000000
233 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC    0x00000001
234 #define X86_VMX_FEATURE_PROC_CTLS2_EPT          0x00000002
235 #define X86_VMX_FEATURE_PROC_CTLS2_VPID         0x00000020
236
237         u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
238
239         clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
240         clear_cpu_cap(c, X86_FEATURE_VNMI);
241         clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
242         clear_cpu_cap(c, X86_FEATURE_EPT);
243         clear_cpu_cap(c, X86_FEATURE_VPID);
244
245         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
246         msr_ctl = vmx_msr_high | vmx_msr_low;
247         if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
248                 set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
249         if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
250                 set_cpu_cap(c, X86_FEATURE_VNMI);
251         if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
252                 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
253                       vmx_msr_low, vmx_msr_high);
254                 msr_ctl2 = vmx_msr_high | vmx_msr_low;
255                 if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
256                     (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
257                         set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
258                 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
259                         set_cpu_cap(c, X86_FEATURE_EPT);
260                 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
261                         set_cpu_cap(c, X86_FEATURE_VPID);
262         }
263 }
264
265 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
266 {
267         unsigned int l2 = 0;
268
269         early_init_intel(c);
270
271         intel_workarounds(c);
272
273         /*
274          * Detect the extended topology information if available. This
275          * will reinitialise the initial_apicid which will be used
276          * in init_intel_cacheinfo()
277          */
278         detect_extended_topology(c);
279
280         l2 = init_intel_cacheinfo(c);
281         if (c->cpuid_level > 9) {
282                 unsigned eax = cpuid_eax(10);
283                 /* Check for version and the number of counters */
284                 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
285                         set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
286         }
287
288         if (cpu_has_xmm2)
289                 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
290         if (cpu_has_ds) {
291                 unsigned int l1;
292                 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
293                 if (!(l1 & (1<<11)))
294                         set_cpu_cap(c, X86_FEATURE_BTS);
295                 if (!(l1 & (1<<12)))
296                         set_cpu_cap(c, X86_FEATURE_PEBS);
297                 ds_init_intel(c);
298         }
299
300         if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush)
301                 set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
302
303 #ifdef CONFIG_X86_64
304         if (c->x86 == 15)
305                 c->x86_cache_alignment = c->x86_clflush_size * 2;
306         if (c->x86 == 6)
307                 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
308 #else
309         /*
310          * Names for the Pentium II/Celeron processors
311          * detectable only by also checking the cache size.
312          * Dixon is NOT a Celeron.
313          */
314         if (c->x86 == 6) {
315                 char *p = NULL;
316
317                 switch (c->x86_model) {
318                 case 5:
319                         if (c->x86_mask == 0) {
320                                 if (l2 == 0)
321                                         p = "Celeron (Covington)";
322                                 else if (l2 == 256)
323                                         p = "Mobile Pentium II (Dixon)";
324                         }
325                         break;
326
327                 case 6:
328                         if (l2 == 128)
329                                 p = "Celeron (Mendocino)";
330                         else if (c->x86_mask == 0 || c->x86_mask == 5)
331                                 p = "Celeron-A";
332                         break;
333
334                 case 8:
335                         if (l2 == 128)
336                                 p = "Celeron (Coppermine)";
337                         break;
338                 }
339
340                 if (p)
341                         strcpy(c->x86_model_id, p);
342         }
343
344         if (c->x86 == 15)
345                 set_cpu_cap(c, X86_FEATURE_P4);
346         if (c->x86 == 6)
347                 set_cpu_cap(c, X86_FEATURE_P3);
348 #endif
349
350         if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
351                 /*
352                  * let's use the legacy cpuid vector 0x1 and 0x4 for topology
353                  * detection.
354                  */
355                 c->x86_max_cores = intel_num_cpu_cores(c);
356 #ifdef CONFIG_X86_32
357                 detect_ht(c);
358 #endif
359         }
360
361         /* Work around errata */
362         srat_detect_node();
363
364         if (cpu_has(c, X86_FEATURE_VMX))
365                 detect_vmx_virtcap(c);
366 }
367
368 #ifdef CONFIG_X86_32
369 static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
370 {
371         /*
372          * Intel PIII Tualatin. This comes in two flavours.
373          * One has 256kb of cache, the other 512. We have no way
374          * to determine which, so we use a boottime override
375          * for the 512kb model, and assume 256 otherwise.
376          */
377         if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
378                 size = 256;
379         return size;
380 }
381 #endif
382
383 static struct cpu_dev intel_cpu_dev __cpuinitdata = {
384         .c_vendor       = "Intel",
385         .c_ident        = { "GenuineIntel" },
386 #ifdef CONFIG_X86_32
387         .c_models = {
388                 { .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
389                   {
390                           [0] = "486 DX-25/33",
391                           [1] = "486 DX-50",
392                           [2] = "486 SX",
393                           [3] = "486 DX/2",
394                           [4] = "486 SL",
395                           [5] = "486 SX/2",
396                           [7] = "486 DX/2-WB",
397                           [8] = "486 DX/4",
398                           [9] = "486 DX/4-WB"
399                   }
400                 },
401                 { .vendor = X86_VENDOR_INTEL, .family = 5, .model_names =
402                   {
403                           [0] = "Pentium 60/66 A-step",
404                           [1] = "Pentium 60/66",
405                           [2] = "Pentium 75 - 200",
406                           [3] = "OverDrive PODP5V83",
407                           [4] = "Pentium MMX",
408                           [7] = "Mobile Pentium 75 - 200",
409                           [8] = "Mobile Pentium MMX"
410                   }
411                 },
412                 { .vendor = X86_VENDOR_INTEL, .family = 6, .model_names =
413                   {
414                           [0] = "Pentium Pro A-step",
415                           [1] = "Pentium Pro",
416                           [3] = "Pentium II (Klamath)",
417                           [4] = "Pentium II (Deschutes)",
418                           [5] = "Pentium II (Deschutes)",
419                           [6] = "Mobile Pentium II",
420                           [7] = "Pentium III (Katmai)",
421                           [8] = "Pentium III (Coppermine)",
422                           [10] = "Pentium III (Cascades)",
423                           [11] = "Pentium III (Tualatin)",
424                   }
425                 },
426                 { .vendor = X86_VENDOR_INTEL, .family = 15, .model_names =
427                   {
428                           [0] = "Pentium 4 (Unknown)",
429                           [1] = "Pentium 4 (Willamette)",
430                           [2] = "Pentium 4 (Northwood)",
431                           [4] = "Pentium 4 (Foster)",
432                           [5] = "Pentium 4 (Foster)",
433                   }
434                 },
435         },
436         .c_size_cache   = intel_size_cache,
437 #endif
438         .c_early_init   = early_init_intel,
439         .c_init         = init_intel,
440         .c_x86_vendor   = X86_VENDOR_INTEL,
441 };
442
443 cpu_dev_register(intel_cpu_dev);
444