[CPUFREQ] X86_GX_SUSPMOD must depend on PCI
[linux-2.6/linux-2.6-stable.git] / arch / i386 / kernel / cpu / intel_cacheinfo.c
blobfbfd374aa336aff467612054031a19356b41b1a7
1 /*
2 * Routines to indentify caches on Intel CPU.
4 * Changes:
5 * Venkatesh Pallipadi : Adding cache identification through cpuid(4)
6 * Ashok Raj <ashok.raj@intel.com>: Work with CPU hotplug infrastructure.
7 */
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 #include <linux/sched.h>
16 #include <asm/processor.h>
17 #include <asm/smp.h>
19 #define LVL_1_INST 1
20 #define LVL_1_DATA 2
21 #define LVL_2 3
22 #define LVL_3 4
23 #define LVL_TRACE 5
25 struct _cache_table
27 unsigned char descriptor;
28 char cache_type;
29 short size;
32 /* all the cache descriptor types we care about (no TLB or trace cache entries) */
33 static struct _cache_table cache_table[] __cpuinitdata =
35 { 0x06, LVL_1_INST, 8 }, /* 4-way set assoc, 32 byte line size */
36 { 0x08, LVL_1_INST, 16 }, /* 4-way set assoc, 32 byte line size */
37 { 0x0a, LVL_1_DATA, 8 }, /* 2 way set assoc, 32 byte line size */
38 { 0x0c, LVL_1_DATA, 16 }, /* 4-way set assoc, 32 byte line size */
39 { 0x22, LVL_3, 512 }, /* 4-way set assoc, sectored cache, 64 byte line size */
40 { 0x23, LVL_3, 1024 }, /* 8-way set assoc, sectored cache, 64 byte line size */
41 { 0x25, LVL_3, 2048 }, /* 8-way set assoc, sectored cache, 64 byte line size */
42 { 0x29, LVL_3, 4096 }, /* 8-way set assoc, sectored cache, 64 byte line size */
43 { 0x2c, LVL_1_DATA, 32 }, /* 8-way set assoc, 64 byte line size */
44 { 0x30, LVL_1_INST, 32 }, /* 8-way set assoc, 64 byte line size */
45 { 0x39, LVL_2, 128 }, /* 4-way set assoc, sectored cache, 64 byte line size */
46 { 0x3b, LVL_2, 128 }, /* 2-way set assoc, sectored cache, 64 byte line size */
47 { 0x3c, LVL_2, 256 }, /* 4-way set assoc, sectored cache, 64 byte line size */
48 { 0x41, LVL_2, 128 }, /* 4-way set assoc, 32 byte line size */
49 { 0x42, LVL_2, 256 }, /* 4-way set assoc, 32 byte line size */
50 { 0x43, LVL_2, 512 }, /* 4-way set assoc, 32 byte line size */
51 { 0x44, LVL_2, 1024 }, /* 4-way set assoc, 32 byte line size */
52 { 0x45, LVL_2, 2048 }, /* 4-way set assoc, 32 byte line size */
53 { 0x60, LVL_1_DATA, 16 }, /* 8-way set assoc, sectored cache, 64 byte line size */
54 { 0x66, LVL_1_DATA, 8 }, /* 4-way set assoc, sectored cache, 64 byte line size */
55 { 0x67, LVL_1_DATA, 16 }, /* 4-way set assoc, sectored cache, 64 byte line size */
56 { 0x68, LVL_1_DATA, 32 }, /* 4-way set assoc, sectored cache, 64 byte line size */
57 { 0x70, LVL_TRACE, 12 }, /* 8-way set assoc */
58 { 0x71, LVL_TRACE, 16 }, /* 8-way set assoc */
59 { 0x72, LVL_TRACE, 32 }, /* 8-way set assoc */
60 { 0x78, LVL_2, 1024 }, /* 4-way set assoc, 64 byte line size */
61 { 0x79, LVL_2, 128 }, /* 8-way set assoc, sectored cache, 64 byte line size */
62 { 0x7a, LVL_2, 256 }, /* 8-way set assoc, sectored cache, 64 byte line size */
63 { 0x7b, LVL_2, 512 }, /* 8-way set assoc, sectored cache, 64 byte line size */
64 { 0x7c, LVL_2, 1024 }, /* 8-way set assoc, sectored cache, 64 byte line size */
65 { 0x7d, LVL_2, 2048 }, /* 8-way set assoc, 64 byte line size */
66 { 0x7f, LVL_2, 512 }, /* 2-way set assoc, 64 byte line size */
67 { 0x82, LVL_2, 256 }, /* 8-way set assoc, 32 byte line size */
68 { 0x83, LVL_2, 512 }, /* 8-way set assoc, 32 byte line size */
69 { 0x84, LVL_2, 1024 }, /* 8-way set assoc, 32 byte line size */
70 { 0x85, LVL_2, 2048 }, /* 8-way set assoc, 32 byte line size */
71 { 0x86, LVL_2, 512 }, /* 4-way set assoc, 64 byte line size */
72 { 0x87, LVL_2, 1024 }, /* 8-way set assoc, 64 byte line size */
73 { 0x00, 0, 0}
77 enum _cache_type
79 CACHE_TYPE_NULL = 0,
80 CACHE_TYPE_DATA = 1,
81 CACHE_TYPE_INST = 2,
82 CACHE_TYPE_UNIFIED = 3
85 union _cpuid4_leaf_eax {
86 struct {
87 enum _cache_type type:5;
88 unsigned int level:3;
89 unsigned int is_self_initializing:1;
90 unsigned int is_fully_associative:1;
91 unsigned int reserved:4;
92 unsigned int num_threads_sharing:12;
93 unsigned int num_cores_on_die:6;
94 } split;
95 u32 full;
98 union _cpuid4_leaf_ebx {
99 struct {
100 unsigned int coherency_line_size:12;
101 unsigned int physical_line_partition:10;
102 unsigned int ways_of_associativity:10;
103 } split;
104 u32 full;
107 union _cpuid4_leaf_ecx {
108 struct {
109 unsigned int number_of_sets:32;
110 } split;
111 u32 full;
114 struct _cpuid4_info {
115 union _cpuid4_leaf_eax eax;
116 union _cpuid4_leaf_ebx ebx;
117 union _cpuid4_leaf_ecx ecx;
118 unsigned long size;
119 cpumask_t shared_cpu_map;
122 static unsigned short num_cache_leaves;
124 static int __cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf)
126 unsigned int eax, ebx, ecx, edx;
127 union _cpuid4_leaf_eax cache_eax;
129 cpuid_count(4, index, &eax, &ebx, &ecx, &edx);
130 cache_eax.full = eax;
131 if (cache_eax.split.type == CACHE_TYPE_NULL)
132 return -EIO; /* better error ? */
134 this_leaf->eax.full = eax;
135 this_leaf->ebx.full = ebx;
136 this_leaf->ecx.full = ecx;
137 this_leaf->size = (this_leaf->ecx.split.number_of_sets + 1) *
138 (this_leaf->ebx.split.coherency_line_size + 1) *
139 (this_leaf->ebx.split.physical_line_partition + 1) *
140 (this_leaf->ebx.split.ways_of_associativity + 1);
141 return 0;
144 static int __init find_num_cache_leaves(void)
146 unsigned int eax, ebx, ecx, edx;
147 union _cpuid4_leaf_eax cache_eax;
148 int i = -1;
150 do {
151 ++i;
152 /* Do cpuid(4) loop to find out num_cache_leaves */
153 cpuid_count(4, i, &eax, &ebx, &ecx, &edx);
154 cache_eax.full = eax;
155 } while (cache_eax.split.type != CACHE_TYPE_NULL);
156 return i;
159 unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
161 unsigned int trace = 0, l1i = 0, l1d = 0, l2 = 0, l3 = 0; /* Cache sizes */
162 unsigned int new_l1d = 0, new_l1i = 0; /* Cache sizes from cpuid(4) */
163 unsigned int new_l2 = 0, new_l3 = 0, i; /* Cache sizes from cpuid(4) */
165 if (c->cpuid_level > 4) {
166 static int is_initialized;
168 if (is_initialized == 0) {
169 /* Init num_cache_leaves from boot CPU */
170 num_cache_leaves = find_num_cache_leaves();
171 is_initialized++;
175 * Whenever possible use cpuid(4), deterministic cache
176 * parameters cpuid leaf to find the cache details
178 for (i = 0; i < num_cache_leaves; i++) {
179 struct _cpuid4_info this_leaf;
181 int retval;
183 retval = cpuid4_cache_lookup(i, &this_leaf);
184 if (retval >= 0) {
185 switch(this_leaf.eax.split.level) {
186 case 1:
187 if (this_leaf.eax.split.type ==
188 CACHE_TYPE_DATA)
189 new_l1d = this_leaf.size/1024;
190 else if (this_leaf.eax.split.type ==
191 CACHE_TYPE_INST)
192 new_l1i = this_leaf.size/1024;
193 break;
194 case 2:
195 new_l2 = this_leaf.size/1024;
196 break;
197 case 3:
198 new_l3 = this_leaf.size/1024;
199 break;
200 default:
201 break;
206 if (c->cpuid_level > 1) {
207 /* supports eax=2 call */
208 int i, j, n;
209 int regs[4];
210 unsigned char *dp = (unsigned char *)regs;
212 /* Number of times to iterate */
213 n = cpuid_eax(2) & 0xFF;
215 for ( i = 0 ; i < n ; i++ ) {
216 cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
218 /* If bit 31 is set, this is an unknown format */
219 for ( j = 0 ; j < 3 ; j++ ) {
220 if ( regs[j] < 0 ) regs[j] = 0;
223 /* Byte 0 is level count, not a descriptor */
224 for ( j = 1 ; j < 16 ; j++ ) {
225 unsigned char des = dp[j];
226 unsigned char k = 0;
228 /* look up this descriptor in the table */
229 while (cache_table[k].descriptor != 0)
231 if (cache_table[k].descriptor == des) {
232 switch (cache_table[k].cache_type) {
233 case LVL_1_INST:
234 l1i += cache_table[k].size;
235 break;
236 case LVL_1_DATA:
237 l1d += cache_table[k].size;
238 break;
239 case LVL_2:
240 l2 += cache_table[k].size;
241 break;
242 case LVL_3:
243 l3 += cache_table[k].size;
244 break;
245 case LVL_TRACE:
246 trace += cache_table[k].size;
247 break;
250 break;
253 k++;
258 if (new_l1d)
259 l1d = new_l1d;
261 if (new_l1i)
262 l1i = new_l1i;
264 if (new_l2)
265 l2 = new_l2;
267 if (new_l3)
268 l3 = new_l3;
270 if ( trace )
271 printk (KERN_INFO "CPU: Trace cache: %dK uops", trace);
272 else if ( l1i )
273 printk (KERN_INFO "CPU: L1 I cache: %dK", l1i);
274 if ( l1d )
275 printk(", L1 D cache: %dK\n", l1d);
276 else
277 printk("\n");
278 if ( l2 )
279 printk(KERN_INFO "CPU: L2 cache: %dK\n", l2);
280 if ( l3 )
281 printk(KERN_INFO "CPU: L3 cache: %dK\n", l3);
283 c->x86_cache_size = l3 ? l3 : (l2 ? l2 : (l1i+l1d));
286 return l2;
289 /* pointer to _cpuid4_info array (for each cache leaf) */
290 static struct _cpuid4_info *cpuid4_info[NR_CPUS];
291 #define CPUID4_INFO_IDX(x,y) (&((cpuid4_info[x])[y]))
293 #ifdef CONFIG_SMP
294 static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index)
296 struct _cpuid4_info *this_leaf, *sibling_leaf;
297 unsigned long num_threads_sharing;
298 int index_msb, i;
299 struct cpuinfo_x86 *c = cpu_data;
301 this_leaf = CPUID4_INFO_IDX(cpu, index);
302 num_threads_sharing = 1 + this_leaf->eax.split.num_threads_sharing;
304 if (num_threads_sharing == 1)
305 cpu_set(cpu, this_leaf->shared_cpu_map);
306 else {
307 index_msb = get_count_order(num_threads_sharing);
309 for_each_online_cpu(i) {
310 if (c[i].apicid >> index_msb ==
311 c[cpu].apicid >> index_msb) {
312 cpu_set(i, this_leaf->shared_cpu_map);
313 if (i != cpu && cpuid4_info[i]) {
314 sibling_leaf = CPUID4_INFO_IDX(i, index);
315 cpu_set(cpu, sibling_leaf->shared_cpu_map);
321 static void __devinit cache_remove_shared_cpu_map(unsigned int cpu, int index)
323 struct _cpuid4_info *this_leaf, *sibling_leaf;
324 int sibling;
326 this_leaf = CPUID4_INFO_IDX(cpu, index);
327 for_each_cpu_mask(sibling, this_leaf->shared_cpu_map) {
328 sibling_leaf = CPUID4_INFO_IDX(sibling, index);
329 cpu_clear(cpu, sibling_leaf->shared_cpu_map);
332 #else
333 static void __init cache_shared_cpu_map_setup(unsigned int cpu, int index) {}
334 static void __init cache_remove_shared_cpu_map(unsigned int cpu, int index) {}
335 #endif
337 static void free_cache_attributes(unsigned int cpu)
339 kfree(cpuid4_info[cpu]);
340 cpuid4_info[cpu] = NULL;
343 static int __cpuinit detect_cache_attributes(unsigned int cpu)
345 struct _cpuid4_info *this_leaf;
346 unsigned long j;
347 int retval;
348 cpumask_t oldmask;
350 if (num_cache_leaves == 0)
351 return -ENOENT;
353 cpuid4_info[cpu] = kmalloc(
354 sizeof(struct _cpuid4_info) * num_cache_leaves, GFP_KERNEL);
355 if (unlikely(cpuid4_info[cpu] == NULL))
356 return -ENOMEM;
357 memset(cpuid4_info[cpu], 0,
358 sizeof(struct _cpuid4_info) * num_cache_leaves);
360 oldmask = current->cpus_allowed;
361 retval = set_cpus_allowed(current, cpumask_of_cpu(cpu));
362 if (retval)
363 goto out;
365 /* Do cpuid and store the results */
366 retval = 0;
367 for (j = 0; j < num_cache_leaves; j++) {
368 this_leaf = CPUID4_INFO_IDX(cpu, j);
369 retval = cpuid4_cache_lookup(j, this_leaf);
370 if (unlikely(retval < 0))
371 break;
372 cache_shared_cpu_map_setup(cpu, j);
374 set_cpus_allowed(current, oldmask);
376 out:
377 if (retval)
378 free_cache_attributes(cpu);
379 return retval;
382 #ifdef CONFIG_SYSFS
384 #include <linux/kobject.h>
385 #include <linux/sysfs.h>
387 extern struct sysdev_class cpu_sysdev_class; /* from drivers/base/cpu.c */
389 /* pointer to kobject for cpuX/cache */
390 static struct kobject * cache_kobject[NR_CPUS];
392 struct _index_kobject {
393 struct kobject kobj;
394 unsigned int cpu;
395 unsigned short index;
398 /* pointer to array of kobjects for cpuX/cache/indexY */
399 static struct _index_kobject *index_kobject[NR_CPUS];
400 #define INDEX_KOBJECT_PTR(x,y) (&((index_kobject[x])[y]))
402 #define show_one_plus(file_name, object, val) \
403 static ssize_t show_##file_name \
404 (struct _cpuid4_info *this_leaf, char *buf) \
406 return sprintf (buf, "%lu\n", (unsigned long)this_leaf->object + val); \
409 show_one_plus(level, eax.split.level, 0);
410 show_one_plus(coherency_line_size, ebx.split.coherency_line_size, 1);
411 show_one_plus(physical_line_partition, ebx.split.physical_line_partition, 1);
412 show_one_plus(ways_of_associativity, ebx.split.ways_of_associativity, 1);
413 show_one_plus(number_of_sets, ecx.split.number_of_sets, 1);
415 static ssize_t show_size(struct _cpuid4_info *this_leaf, char *buf)
417 return sprintf (buf, "%luK\n", this_leaf->size / 1024);
420 static ssize_t show_shared_cpu_map(struct _cpuid4_info *this_leaf, char *buf)
422 char mask_str[NR_CPUS];
423 cpumask_scnprintf(mask_str, NR_CPUS, this_leaf->shared_cpu_map);
424 return sprintf(buf, "%s\n", mask_str);
427 static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf) {
428 switch(this_leaf->eax.split.type) {
429 case CACHE_TYPE_DATA:
430 return sprintf(buf, "Data\n");
431 break;
432 case CACHE_TYPE_INST:
433 return sprintf(buf, "Instruction\n");
434 break;
435 case CACHE_TYPE_UNIFIED:
436 return sprintf(buf, "Unified\n");
437 break;
438 default:
439 return sprintf(buf, "Unknown\n");
440 break;
444 struct _cache_attr {
445 struct attribute attr;
446 ssize_t (*show)(struct _cpuid4_info *, char *);
447 ssize_t (*store)(struct _cpuid4_info *, const char *, size_t count);
450 #define define_one_ro(_name) \
451 static struct _cache_attr _name = \
452 __ATTR(_name, 0444, show_##_name, NULL)
454 define_one_ro(level);
455 define_one_ro(type);
456 define_one_ro(coherency_line_size);
457 define_one_ro(physical_line_partition);
458 define_one_ro(ways_of_associativity);
459 define_one_ro(number_of_sets);
460 define_one_ro(size);
461 define_one_ro(shared_cpu_map);
463 static struct attribute * default_attrs[] = {
464 &type.attr,
465 &level.attr,
466 &coherency_line_size.attr,
467 &physical_line_partition.attr,
468 &ways_of_associativity.attr,
469 &number_of_sets.attr,
470 &size.attr,
471 &shared_cpu_map.attr,
472 NULL
475 #define to_object(k) container_of(k, struct _index_kobject, kobj)
476 #define to_attr(a) container_of(a, struct _cache_attr, attr)
478 static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
480 struct _cache_attr *fattr = to_attr(attr);
481 struct _index_kobject *this_leaf = to_object(kobj);
482 ssize_t ret;
484 ret = fattr->show ?
485 fattr->show(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
486 buf) :
488 return ret;
491 static ssize_t store(struct kobject * kobj, struct attribute * attr,
492 const char * buf, size_t count)
494 return 0;
497 static struct sysfs_ops sysfs_ops = {
498 .show = show,
499 .store = store,
502 static struct kobj_type ktype_cache = {
503 .sysfs_ops = &sysfs_ops,
504 .default_attrs = default_attrs,
507 static struct kobj_type ktype_percpu_entry = {
508 .sysfs_ops = &sysfs_ops,
511 static void cpuid4_cache_sysfs_exit(unsigned int cpu)
513 kfree(cache_kobject[cpu]);
514 kfree(index_kobject[cpu]);
515 cache_kobject[cpu] = NULL;
516 index_kobject[cpu] = NULL;
517 free_cache_attributes(cpu);
520 static int __cpuinit cpuid4_cache_sysfs_init(unsigned int cpu)
523 if (num_cache_leaves == 0)
524 return -ENOENT;
526 detect_cache_attributes(cpu);
527 if (cpuid4_info[cpu] == NULL)
528 return -ENOENT;
530 /* Allocate all required memory */
531 cache_kobject[cpu] = kmalloc(sizeof(struct kobject), GFP_KERNEL);
532 if (unlikely(cache_kobject[cpu] == NULL))
533 goto err_out;
534 memset(cache_kobject[cpu], 0, sizeof(struct kobject));
536 index_kobject[cpu] = kmalloc(
537 sizeof(struct _index_kobject ) * num_cache_leaves, GFP_KERNEL);
538 if (unlikely(index_kobject[cpu] == NULL))
539 goto err_out;
540 memset(index_kobject[cpu], 0,
541 sizeof(struct _index_kobject) * num_cache_leaves);
543 return 0;
545 err_out:
546 cpuid4_cache_sysfs_exit(cpu);
547 return -ENOMEM;
550 /* Add/Remove cache interface for CPU device */
551 static int __cpuinit cache_add_dev(struct sys_device * sys_dev)
553 unsigned int cpu = sys_dev->id;
554 unsigned long i, j;
555 struct _index_kobject *this_object;
556 int retval = 0;
558 retval = cpuid4_cache_sysfs_init(cpu);
559 if (unlikely(retval < 0))
560 return retval;
562 cache_kobject[cpu]->parent = &sys_dev->kobj;
563 kobject_set_name(cache_kobject[cpu], "%s", "cache");
564 cache_kobject[cpu]->ktype = &ktype_percpu_entry;
565 retval = kobject_register(cache_kobject[cpu]);
567 for (i = 0; i < num_cache_leaves; i++) {
568 this_object = INDEX_KOBJECT_PTR(cpu,i);
569 this_object->cpu = cpu;
570 this_object->index = i;
571 this_object->kobj.parent = cache_kobject[cpu];
572 kobject_set_name(&(this_object->kobj), "index%1lu", i);
573 this_object->kobj.ktype = &ktype_cache;
574 retval = kobject_register(&(this_object->kobj));
575 if (unlikely(retval)) {
576 for (j = 0; j < i; j++) {
577 kobject_unregister(
578 &(INDEX_KOBJECT_PTR(cpu,j)->kobj));
580 kobject_unregister(cache_kobject[cpu]);
581 cpuid4_cache_sysfs_exit(cpu);
582 break;
585 return retval;
588 static void __cpuexit cache_remove_dev(struct sys_device * sys_dev)
590 unsigned int cpu = sys_dev->id;
591 unsigned long i;
593 for (i = 0; i < num_cache_leaves; i++) {
594 cache_remove_shared_cpu_map(cpu, i);
595 kobject_unregister(&(INDEX_KOBJECT_PTR(cpu,i)->kobj));
597 kobject_unregister(cache_kobject[cpu]);
598 cpuid4_cache_sysfs_exit(cpu);
599 return;
602 static int __cpuinit cacheinfo_cpu_callback(struct notifier_block *nfb,
603 unsigned long action, void *hcpu)
605 unsigned int cpu = (unsigned long)hcpu;
606 struct sys_device *sys_dev;
608 sys_dev = get_cpu_sysdev(cpu);
609 switch (action) {
610 case CPU_ONLINE:
611 cache_add_dev(sys_dev);
612 break;
613 case CPU_DEAD:
614 cache_remove_dev(sys_dev);
615 break;
617 return NOTIFY_OK;
620 static struct notifier_block cacheinfo_cpu_notifier =
622 .notifier_call = cacheinfo_cpu_callback,
625 static int __cpuinit cache_sysfs_init(void)
627 int i;
629 if (num_cache_leaves == 0)
630 return 0;
632 register_cpu_notifier(&cacheinfo_cpu_notifier);
634 for_each_online_cpu(i) {
635 cacheinfo_cpu_callback(&cacheinfo_cpu_notifier, CPU_ONLINE,
636 (void *)(long)i);
639 return 0;
642 device_initcall(cache_sysfs_init);
644 #endif