?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[PATCH] create and destroy cache sysfs entries based on cpu notifiers
[linux-2.6/verdex.git] / arch / i386 / kernel / cpu / intel_cacheinfo.c
blobf0839334881c76a49fbf7cc191f1c20a82e556ce
1 /*
2 * Routines to indentify caches on Intel CPU.
3 *
4 * Changes:
5 * Venkatesh Pallipadi : Adding cache identification through cpuid(4)
6 * Ashok Raj <ashok.raj@intel.com>: Work with CPU hotplug infrastructure.
7 */
8
9 #include <linux/init.h>
10 #include <linux/slab.h>
11 #include <linux/device.h>
12 #include <linux/compiler.h>
13 #include <linux/cpu.h>
14
15 #include <asm/processor.h>
16 #include <asm/smp.h>
17
18 #define LVL_1_INST 1
19 #define LVL_1_DATA 2
20 #define LVL_2 3
21 #define LVL_3 4
22 #define LVL_TRACE 5
23
24 struct _cache_table
25 {
26 unsigned char descriptor;
27 char cache_type;
28 short size;
29 };
30
31 /* all the cache descriptor types we care about (no TLB or trace cache entries) */
32 static struct _cache_table cache_table[] __cpuinitdata =
33 {
34 { 0x06, LVL_1_INST, 8 }, /* 4-way set assoc, 32 byte line size */
35 { 0x08, LVL_1_INST, 16 }, /* 4-way set assoc, 32 byte line size */
36 { 0x0a, LVL_1_DATA, 8 }, /* 2 way set assoc, 32 byte line size */
37 { 0x0c, LVL_1_DATA, 16 }, /* 4-way set assoc, 32 byte line size */
38 { 0x22, LVL_3, 512 }, /* 4-way set assoc, sectored cache, 64 byte line size */
39 { 0x23, LVL_3, 1024 }, /* 8-way set assoc, sectored cache, 64 byte line size */
40 { 0x25, LVL_3, 2048 }, /* 8-way set assoc, sectored cache, 64 byte line size */
41 { 0x29, LVL_3, 4096 }, /* 8-way set assoc, sectored cache, 64 byte line size */
42 { 0x2c, LVL_1_DATA, 32 }, /* 8-way set assoc, 64 byte line size */
43 { 0x30, LVL_1_INST, 32 }, /* 8-way set assoc, 64 byte line size */
44 { 0x39, LVL_2, 128 }, /* 4-way set assoc, sectored cache, 64 byte line size */
45 { 0x3b, LVL_2, 128 }, /* 2-way set assoc, sectored cache, 64 byte line size */
46 { 0x3c, LVL_2, 256 }, /* 4-way set assoc, sectored cache, 64 byte line size */
47 { 0x41, LVL_2, 128 }, /* 4-way set assoc, 32 byte line size */
48 { 0x42, LVL_2, 256 }, /* 4-way set assoc, 32 byte line size */
49 { 0x43, LVL_2, 512 }, /* 4-way set assoc, 32 byte line size */
50 { 0x44, LVL_2, 1024 }, /* 4-way set assoc, 32 byte line size */
51 { 0x45, LVL_2, 2048 }, /* 4-way set assoc, 32 byte line size */
52 { 0x60, LVL_1_DATA, 16 }, /* 8-way set assoc, sectored cache, 64 byte line size */
53 { 0x66, LVL_1_DATA, 8 }, /* 4-way set assoc, sectored cache, 64 byte line size */
54 { 0x67, LVL_1_DATA, 16 }, /* 4-way set assoc, sectored cache, 64 byte line size */
55 { 0x68, LVL_1_DATA, 32 }, /* 4-way set assoc, sectored cache, 64 byte line size */
56 { 0x70, LVL_TRACE, 12 }, /* 8-way set assoc */
57 { 0x71, LVL_TRACE, 16 }, /* 8-way set assoc */
58 { 0x72, LVL_TRACE, 32 }, /* 8-way set assoc */
59 { 0x78, LVL_2, 1024 }, /* 4-way set assoc, 64 byte line size */
60 { 0x79, LVL_2, 128 }, /* 8-way set assoc, sectored cache, 64 byte line size */
61 { 0x7a, LVL_2, 256 }, /* 8-way set assoc, sectored cache, 64 byte line size */
62 { 0x7b, LVL_2, 512 }, /* 8-way set assoc, sectored cache, 64 byte line size */
63 { 0x7c, LVL_2, 1024 }, /* 8-way set assoc, sectored cache, 64 byte line size */
64 { 0x7d, LVL_2, 2048 }, /* 8-way set assoc, 64 byte line size */
65 { 0x7f, LVL_2, 512 }, /* 2-way set assoc, 64 byte line size */
66 { 0x82, LVL_2, 256 }, /* 8-way set assoc, 32 byte line size */
67 { 0x83, LVL_2, 512 }, /* 8-way set assoc, 32 byte line size */
68 { 0x84, LVL_2, 1024 }, /* 8-way set assoc, 32 byte line size */
69 { 0x85, LVL_2, 2048 }, /* 8-way set assoc, 32 byte line size */
70 { 0x86, LVL_2, 512 }, /* 4-way set assoc, 64 byte line size */
71 { 0x87, LVL_2, 1024 }, /* 8-way set assoc, 64 byte line size */
72 { 0x00, 0, 0}
73 };
74
75
76 enum _cache_type
77 {
78 CACHE_TYPE_NULL = 0,
79 CACHE_TYPE_DATA = 1,
80 CACHE_TYPE_INST = 2,
81 CACHE_TYPE_UNIFIED = 3
82 };
83
84 union _cpuid4_leaf_eax {
85 struct {
86 enum _cache_type type:5;
87 unsigned int level:3;
88 unsigned int is_self_initializing:1;
89 unsigned int is_fully_associative:1;
90 unsigned int reserved:4;
91 unsigned int num_threads_sharing:12;
92 unsigned int num_cores_on_die:6;
93 } split;
94 u32 full;
95 };
96
97 union _cpuid4_leaf_ebx {
98 struct {
99 unsigned int coherency_line_size:12;
100 unsigned int physical_line_partition:10;
101 unsigned int ways_of_associativity:10;
102 } split;
103 u32 full;
104 };
105
106 union _cpuid4_leaf_ecx {
107 struct {
108 unsigned int number_of_sets:32;
109 } split;
110 u32 full;
111 };
112
113 struct _cpuid4_info {
114 union _cpuid4_leaf_eax eax;
115 union _cpuid4_leaf_ebx ebx;
116 union _cpuid4_leaf_ecx ecx;
117 unsigned long size;
118 cpumask_t shared_cpu_map;
119 };
120
121 static unsigned short num_cache_leaves;
122
123 static int __cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf)
124 {
125 unsigned int eax, ebx, ecx, edx;
126 union _cpuid4_leaf_eax cache_eax;
127
128 cpuid_count(4, index, &eax, &ebx, &ecx, &edx);
129 cache_eax.full = eax;
130 if (cache_eax.split.type == CACHE_TYPE_NULL)
131 return -EIO; /* better error ? */
132
133 this_leaf->eax.full = eax;
134 this_leaf->ebx.full = ebx;
135 this_leaf->ecx.full = ecx;
136 this_leaf->size = (this_leaf->ecx.split.number_of_sets + 1) *
137 (this_leaf->ebx.split.coherency_line_size + 1) *
138 (this_leaf->ebx.split.physical_line_partition + 1) *
139 (this_leaf->ebx.split.ways_of_associativity + 1);
140 return 0;
141 }
142
143 static int __init find_num_cache_leaves(void)
144 {
145 unsigned int eax, ebx, ecx, edx;
146 union _cpuid4_leaf_eax cache_eax;
147 int i = -1;
148
149 do {
150 ++i;
151 /* Do cpuid(4) loop to find out num_cache_leaves */
152 cpuid_count(4, i, &eax, &ebx, &ecx, &edx);
153 cache_eax.full = eax;
154 } while (cache_eax.split.type != CACHE_TYPE_NULL);
155 return i;
156 }
157
158 unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
159 {
160 unsigned int trace = 0, l1i = 0, l1d = 0, l2 = 0, l3 = 0; /* Cache sizes */
161 unsigned int new_l1d = 0, new_l1i = 0; /* Cache sizes from cpuid(4) */
162 unsigned int new_l2 = 0, new_l3 = 0, i; /* Cache sizes from cpuid(4) */
163
164 if (c->cpuid_level > 4) {
165 static int is_initialized;
166
167 if (is_initialized == 0) {
168 /* Init num_cache_leaves from boot CPU */
169 num_cache_leaves = find_num_cache_leaves();
170 is_initialized++;
171 }
172
173 /*
174 * Whenever possible use cpuid(4), deterministic cache
175 * parameters cpuid leaf to find the cache details
176 */
177 for (i = 0; i < num_cache_leaves; i++) {
178 struct _cpuid4_info this_leaf;
179
180 int retval;
181
182 retval = cpuid4_cache_lookup(i, &this_leaf);
183 if (retval >= 0) {
184 switch(this_leaf.eax.split.level) {
185 case 1:
186 if (this_leaf.eax.split.type ==
187 CACHE_TYPE_DATA)
188 new_l1d = this_leaf.size/1024;
189 else if (this_leaf.eax.split.type ==
190 CACHE_TYPE_INST)
191 new_l1i = this_leaf.size/1024;
192 break;
193 case 2:
194 new_l2 = this_leaf.size/1024;
195 break;
196 case 3:
197 new_l3 = this_leaf.size/1024;
198 break;
199 default:
200 break;
201 }
202 }
203 }
204 }
205 if (c->cpuid_level > 1) {
206 /* supports eax=2 call */
207 int i, j, n;
208 int regs[4];
209 unsigned char *dp = (unsigned char *)regs;
210
211 /* Number of times to iterate */
212 n = cpuid_eax(2) & 0xFF;
213
214 for ( i = 0 ; i < n ; i++ ) {
215 cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
216
217 /* If bit 31 is set, this is an unknown format */
218 for ( j = 0 ; j < 3 ; j++ ) {
219 if ( regs[j] < 0 ) regs[j] = 0;
220 }
221
222 /* Byte 0 is level count, not a descriptor */
223 for ( j = 1 ; j < 16 ; j++ ) {
224 unsigned char des = dp[j];
225 unsigned char k = 0;
226
227 /* look up this descriptor in the table */
228 while (cache_table[k].descriptor != 0)
229 {
230 if (cache_table[k].descriptor == des) {
231 switch (cache_table[k].cache_type) {
232 case LVL_1_INST:
233 l1i += cache_table[k].size;
234 break;
235 case LVL_1_DATA:
236 l1d += cache_table[k].size;
237 break;
238 case LVL_2:
239 l2 += cache_table[k].size;
240 break;
241 case LVL_3:
242 l3 += cache_table[k].size;
243 break;
244 case LVL_TRACE:
245 trace += cache_table[k].size;
246 break;
247 }
248
249 break;
250 }
251
252 k++;
253 }
254 }
255 }
256
257 if (new_l1d)
258 l1d = new_l1d;
259
260 if (new_l1i)
261 l1i = new_l1i;
262
263 if (new_l2)
264 l2 = new_l2;
265
266 if (new_l3)
267 l3 = new_l3;
268
269 if ( trace )
270 printk (KERN_INFO "CPU: Trace cache: %dK uops", trace);
271 else if ( l1i )
272 printk (KERN_INFO "CPU: L1 I cache: %dK", l1i);
273 if ( l1d )
274 printk(", L1 D cache: %dK\n", l1d);
275 else
276 printk("\n");
277 if ( l2 )
278 printk(KERN_INFO "CPU: L2 cache: %dK\n", l2);
279 if ( l3 )
280 printk(KERN_INFO "CPU: L3 cache: %dK\n", l3);
281
282 c->x86_cache_size = l3 ? l3 : (l2 ? l2 : (l1i+l1d));
283 }
284
285 return l2;
286 }
287
288 /* pointer to _cpuid4_info array (for each cache leaf) */
289 static struct _cpuid4_info *cpuid4_info[NR_CPUS];
290 #define CPUID4_INFO_IDX(x,y) (&((cpuid4_info[x])[y]))
291
292 #ifdef CONFIG_SMP
293 static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index)
294 {
295 struct _cpuid4_info *this_leaf;
296 unsigned long num_threads_sharing;
297 #ifdef CONFIG_X86_HT
298 struct cpuinfo_x86 *c = cpu_data + cpu;
299 #endif
300
301 this_leaf = CPUID4_INFO_IDX(cpu, index);
302 num_threads_sharing = 1 + this_leaf->eax.split.num_threads_sharing;
303
304 if (num_threads_sharing == 1)
305 cpu_set(cpu, this_leaf->shared_cpu_map);
306 #ifdef CONFIG_X86_HT
307 else if (num_threads_sharing == smp_num_siblings)
308 this_leaf->shared_cpu_map = cpu_sibling_map[cpu];
309 else if (num_threads_sharing == (c->x86_num_cores * smp_num_siblings))
310 this_leaf->shared_cpu_map = cpu_core_map[cpu];
311 else
312 printk(KERN_DEBUG "Number of CPUs sharing cache didn't match "
313 "any known set of CPUs\n");
314 #endif
315 }
316 #else
317 static void __init cache_shared_cpu_map_setup(unsigned int cpu, int index) {}
318 #endif
319
320 static void free_cache_attributes(unsigned int cpu)
321 {
322 kfree(cpuid4_info[cpu]);
323 cpuid4_info[cpu] = NULL;
324 }
325
326 static int __cpuinit detect_cache_attributes(unsigned int cpu)
327 {
328 struct _cpuid4_info *this_leaf;
329 unsigned long j;
330 int retval;
331 cpumask_t oldmask;
332
333 if (num_cache_leaves == 0)
334 return -ENOENT;
335
336 cpuid4_info[cpu] = kmalloc(
337 sizeof(struct _cpuid4_info) * num_cache_leaves, GFP_KERNEL);
338 if (unlikely(cpuid4_info[cpu] == NULL))
339 return -ENOMEM;
340 memset(cpuid4_info[cpu], 0,
341 sizeof(struct _cpuid4_info) * num_cache_leaves);
342
343 oldmask = current->cpus_allowed;
344 retval = set_cpus_allowed(current, cpumask_of_cpu(cpu));
345 if (retval)
346 goto out;
347
348 /* Do cpuid and store the results */
349 retval = 0;
350 for (j = 0; j < num_cache_leaves; j++) {
351 this_leaf = CPUID4_INFO_IDX(cpu, j);
352 retval = cpuid4_cache_lookup(j, this_leaf);
353 if (unlikely(retval < 0))
354 break;
355 cache_shared_cpu_map_setup(cpu, j);
356 }
357 set_cpus_allowed(current, oldmask);
358
359 out:
360 if (retval)
361 free_cache_attributes(cpu);
362 return retval;
363 }
364
365 #ifdef CONFIG_SYSFS
366
367 #include <linux/kobject.h>
368 #include <linux/sysfs.h>
369
370 extern struct sysdev_class cpu_sysdev_class; /* from drivers/base/cpu.c */
371
372 /* pointer to kobject for cpuX/cache */
373 static struct kobject * cache_kobject[NR_CPUS];
374
375 struct _index_kobject {
376 struct kobject kobj;
377 unsigned int cpu;
378 unsigned short index;
379 };
380
381 /* pointer to array of kobjects for cpuX/cache/indexY */
382 static struct _index_kobject *index_kobject[NR_CPUS];
383 #define INDEX_KOBJECT_PTR(x,y) (&((index_kobject[x])[y]))
384
385 #define show_one_plus(file_name, object, val) \
386 static ssize_t show_##file_name \
387 (struct _cpuid4_info *this_leaf, char *buf) \
388 { \
389 return sprintf (buf, "%lu\n", (unsigned long)this_leaf->object + val); \
390 }
391
392 show_one_plus(level, eax.split.level, 0);
393 show_one_plus(coherency_line_size, ebx.split.coherency_line_size, 1);
394 show_one_plus(physical_line_partition, ebx.split.physical_line_partition, 1);
395 show_one_plus(ways_of_associativity, ebx.split.ways_of_associativity, 1);
396 show_one_plus(number_of_sets, ecx.split.number_of_sets, 1);
397
398 static ssize_t show_size(struct _cpuid4_info *this_leaf, char *buf)
399 {
400 return sprintf (buf, "%luK\n", this_leaf->size / 1024);
401 }
402
403 static ssize_t show_shared_cpu_map(struct _cpuid4_info *this_leaf, char *buf)
404 {
405 char mask_str[NR_CPUS];
406 cpumask_scnprintf(mask_str, NR_CPUS, this_leaf->shared_cpu_map);
407 return sprintf(buf, "%s\n", mask_str);
408 }
409
410 static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf) {
411 switch(this_leaf->eax.split.type) {
412 case CACHE_TYPE_DATA:
413 return sprintf(buf, "Data\n");
414 break;
415 case CACHE_TYPE_INST:
416 return sprintf(buf, "Instruction\n");
417 break;
418 case CACHE_TYPE_UNIFIED:
419 return sprintf(buf, "Unified\n");
420 break;
421 default:
422 return sprintf(buf, "Unknown\n");
423 break;
424 }
425 }
426
427 struct _cache_attr {
428 struct attribute attr;
429 ssize_t (*show)(struct _cpuid4_info *, char *);
430 ssize_t (*store)(struct _cpuid4_info *, const char *, size_t count);
431 };
432
433 #define define_one_ro(_name) \
434 static struct _cache_attr _name = \
435 __ATTR(_name, 0444, show_##_name, NULL)
436
437 define_one_ro(level);
438 define_one_ro(type);
439 define_one_ro(coherency_line_size);
440 define_one_ro(physical_line_partition);
441 define_one_ro(ways_of_associativity);
442 define_one_ro(number_of_sets);
443 define_one_ro(size);
444 define_one_ro(shared_cpu_map);
445
446 static struct attribute * default_attrs[] = {
447 &type.attr,
448 &level.attr,
449 &coherency_line_size.attr,
450 &physical_line_partition.attr,
451 &ways_of_associativity.attr,
452 &number_of_sets.attr,
453 &size.attr,
454 &shared_cpu_map.attr,
455 NULL
456 };
457
458 #define to_object(k) container_of(k, struct _index_kobject, kobj)
459 #define to_attr(a) container_of(a, struct _cache_attr, attr)
460
461 static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
462 {
463 struct _cache_attr *fattr = to_attr(attr);
464 struct _index_kobject *this_leaf = to_object(kobj);
465 ssize_t ret;
466
467 ret = fattr->show ?
468 fattr->show(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
469 buf) :
470 0;
471 return ret;
472 }
473
474 static ssize_t store(struct kobject * kobj, struct attribute * attr,
475 const char * buf, size_t count)
476 {
477 return 0;
478 }
479
480 static struct sysfs_ops sysfs_ops = {
481 .show = show,
482 .store = store,
483 };
484
485 static struct kobj_type ktype_cache = {
486 .sysfs_ops = &sysfs_ops,
487 .default_attrs = default_attrs,
488 };
489
490 static struct kobj_type ktype_percpu_entry = {
491 .sysfs_ops = &sysfs_ops,
492 };
493
494 static void cpuid4_cache_sysfs_exit(unsigned int cpu)
495 {
496 kfree(cache_kobject[cpu]);
497 kfree(index_kobject[cpu]);
498 cache_kobject[cpu] = NULL;
499 index_kobject[cpu] = NULL;
500 free_cache_attributes(cpu);
501 }
502
503 static int __cpuinit cpuid4_cache_sysfs_init(unsigned int cpu)
504 {
505
506 if (num_cache_leaves == 0)
507 return -ENOENT;
508
509 detect_cache_attributes(cpu);
510 if (cpuid4_info[cpu] == NULL)
511 return -ENOENT;
512
513 /* Allocate all required memory */
514 cache_kobject[cpu] = kmalloc(sizeof(struct kobject), GFP_KERNEL);
515 if (unlikely(cache_kobject[cpu] == NULL))
516 goto err_out;
517 memset(cache_kobject[cpu], 0, sizeof(struct kobject));
518
519 index_kobject[cpu] = kmalloc(
520 sizeof(struct _index_kobject ) * num_cache_leaves, GFP_KERNEL);
521 if (unlikely(index_kobject[cpu] == NULL))
522 goto err_out;
523 memset(index_kobject[cpu], 0,
524 sizeof(struct _index_kobject) * num_cache_leaves);
525
526 return 0;
527
528 err_out:
529 cpuid4_cache_sysfs_exit(cpu);
530 return -ENOMEM;
531 }
532
533 /* Add/Remove cache interface for CPU device */
534 static int __cpuinit cache_add_dev(struct sys_device * sys_dev)
535 {
536 unsigned int cpu = sys_dev->id;
537 unsigned long i, j;
538 struct _index_kobject *this_object;
539 int retval = 0;
540
541 retval = cpuid4_cache_sysfs_init(cpu);
542 if (unlikely(retval < 0))
543 return retval;
544
545 cache_kobject[cpu]->parent = &sys_dev->kobj;
546 kobject_set_name(cache_kobject[cpu], "%s", "cache");
547 cache_kobject[cpu]->ktype = &ktype_percpu_entry;
548 retval = kobject_register(cache_kobject[cpu]);
549
550 for (i = 0; i < num_cache_leaves; i++) {
551 this_object = INDEX_KOBJECT_PTR(cpu,i);
552 this_object->cpu = cpu;
553 this_object->index = i;
554 this_object->kobj.parent = cache_kobject[cpu];
555 kobject_set_name(&(this_object->kobj), "index%1lu", i);
556 this_object->kobj.ktype = &ktype_cache;
557 retval = kobject_register(&(this_object->kobj));
558 if (unlikely(retval)) {
559 for (j = 0; j < i; j++) {
560 kobject_unregister(
561 &(INDEX_KOBJECT_PTR(cpu,j)->kobj));
562 }
563 kobject_unregister(cache_kobject[cpu]);
564 cpuid4_cache_sysfs_exit(cpu);
565 break;
566 }
567 }
568 return retval;
569 }
570
571 static void __cpuexit cache_remove_dev(struct sys_device * sys_dev)
572 {
573 unsigned int cpu = sys_dev->id;
574 unsigned long i;
575
576 for (i = 0; i < num_cache_leaves; i++)
577 kobject_unregister(&(INDEX_KOBJECT_PTR(cpu,i)->kobj));
578 kobject_unregister(cache_kobject[cpu]);
579 cpuid4_cache_sysfs_exit(cpu);
580 return;
581 }
582
583 static int __cpuinit cacheinfo_cpu_callback(struct notifier_block *nfb,
584 unsigned long action, void *hcpu)
585 {
586 unsigned int cpu = (unsigned long)hcpu;
587 struct sys_device *sys_dev;
588
589 sys_dev = get_cpu_sysdev(cpu);
590 switch (action) {
591 case CPU_ONLINE:
592 cache_add_dev(sys_dev);
593 break;
594 case CPU_DEAD:
595 cache_remove_dev(sys_dev);
596 break;
597 }
598 return NOTIFY_OK;
599 }
600
601 static struct notifier_block cacheinfo_cpu_notifier =
602 {
603 .notifier_call = cacheinfo_cpu_callback,
604 };
605
606 static int __cpuinit cache_sysfs_init(void)
607 {
608 int i;
609
610 if (num_cache_leaves == 0)
611 return 0;
612
613 register_cpu_notifier(&cacheinfo_cpu_notifier);
614
615 for_each_online_cpu(i) {
616 cacheinfo_cpu_callback(&cacheinfo_cpu_notifier, CPU_ONLINE,
617 (void *)(long)i);
618 }
619
620 return 0;
621 }
622
623 device_initcall(cache_sysfs_init);
624
625 #endif
Verdex Pro support