2 * Routines to indentify caches on Intel CPU.
5 * Venkatesh Pallipadi : Adding cache identification through cpuid(4)
8 #include <linux/init.h>
9 #include <linux/slab.h>
10 #include <linux/device.h>
11 #include <linux/compiler.h>
12 #include <linux/cpu.h>
14 #include <asm/processor.h>
25 unsigned char descriptor;
30 /* all the cache descriptor types we care about (no TLB or trace cache entries) */
31 static struct _cache_table cache_table[] __devinitdata =
33 { 0x06, LVL_1_INST, 8 }, /* 4-way set assoc, 32 byte line size */
34 { 0x08, LVL_1_INST, 16 }, /* 4-way set assoc, 32 byte line size */
35 { 0x0a, LVL_1_DATA, 8 }, /* 2 way set assoc, 32 byte line size */
36 { 0x0c, LVL_1_DATA, 16 }, /* 4-way set assoc, 32 byte line size */
37 { 0x22, LVL_3, 512 }, /* 4-way set assoc, sectored cache, 64 byte line size */
38 { 0x23, LVL_3, 1024 }, /* 8-way set assoc, sectored cache, 64 byte line size */
39 { 0x25, LVL_3, 2048 }, /* 8-way set assoc, sectored cache, 64 byte line size */
40 { 0x29, LVL_3, 4096 }, /* 8-way set assoc, sectored cache, 64 byte line size */
41 { 0x2c, LVL_1_DATA, 32 }, /* 8-way set assoc, 64 byte line size */
42 { 0x30, LVL_1_INST, 32 }, /* 8-way set assoc, 64 byte line size */
43 { 0x39, LVL_2, 128 }, /* 4-way set assoc, sectored cache, 64 byte line size */
44 { 0x3b, LVL_2, 128 }, /* 2-way set assoc, sectored cache, 64 byte line size */
45 { 0x3c, LVL_2, 256 }, /* 4-way set assoc, sectored cache, 64 byte line size */
46 { 0x41, LVL_2, 128 }, /* 4-way set assoc, 32 byte line size */
47 { 0x42, LVL_2, 256 }, /* 4-way set assoc, 32 byte line size */
48 { 0x43, LVL_2, 512 }, /* 4-way set assoc, 32 byte line size */
49 { 0x44, LVL_2, 1024 }, /* 4-way set assoc, 32 byte line size */
50 { 0x45, LVL_2, 2048 }, /* 4-way set assoc, 32 byte line size */
51 { 0x60, LVL_1_DATA, 16 }, /* 8-way set assoc, sectored cache, 64 byte line size */
52 { 0x66, LVL_1_DATA, 8 }, /* 4-way set assoc, sectored cache, 64 byte line size */
53 { 0x67, LVL_1_DATA, 16 }, /* 4-way set assoc, sectored cache, 64 byte line size */
54 { 0x68, LVL_1_DATA, 32 }, /* 4-way set assoc, sectored cache, 64 byte line size */
55 { 0x70, LVL_TRACE, 12 }, /* 8-way set assoc */
56 { 0x71, LVL_TRACE, 16 }, /* 8-way set assoc */
57 { 0x72, LVL_TRACE, 32 }, /* 8-way set assoc */
58 { 0x78, LVL_2, 1024 }, /* 4-way set assoc, 64 byte line size */
59 { 0x79, LVL_2, 128 }, /* 8-way set assoc, sectored cache, 64 byte line size */
60 { 0x7a, LVL_2, 256 }, /* 8-way set assoc, sectored cache, 64 byte line size */
61 { 0x7b, LVL_2, 512 }, /* 8-way set assoc, sectored cache, 64 byte line size */
62 { 0x7c, LVL_2, 1024 }, /* 8-way set assoc, sectored cache, 64 byte line size */
63 { 0x7d, LVL_2, 2048 }, /* 8-way set assoc, 64 byte line size */
64 { 0x7f, LVL_2, 512 }, /* 2-way set assoc, 64 byte line size */
65 { 0x82, LVL_2, 256 }, /* 8-way set assoc, 32 byte line size */
66 { 0x83, LVL_2, 512 }, /* 8-way set assoc, 32 byte line size */
67 { 0x84, LVL_2, 1024 }, /* 8-way set assoc, 32 byte line size */
68 { 0x85, LVL_2, 2048 }, /* 8-way set assoc, 32 byte line size */
69 { 0x86, LVL_2, 512 }, /* 4-way set assoc, 64 byte line size */
70 { 0x87, LVL_2, 1024 }, /* 8-way set assoc, 64 byte line size */
80 CACHE_TYPE_UNIFIED = 3
83 union _cpuid4_leaf_eax {
85 enum _cache_type type:5;
87 unsigned int is_self_initializing:1;
88 unsigned int is_fully_associative:1;
89 unsigned int reserved:4;
90 unsigned int num_threads_sharing:12;
91 unsigned int num_cores_on_die:6;
96 union _cpuid4_leaf_ebx {
98 unsigned int coherency_line_size:12;
99 unsigned int physical_line_partition:10;
100 unsigned int ways_of_associativity:10;
105 union _cpuid4_leaf_ecx {
107 unsigned int number_of_sets:32;
112 struct _cpuid4_info {
113 union _cpuid4_leaf_eax eax;
114 union _cpuid4_leaf_ebx ebx;
115 union _cpuid4_leaf_ecx ecx;
117 cpumask_t shared_cpu_map;
120 #define MAX_CACHE_LEAVES 4
121 static unsigned short num_cache_leaves;
123 static int __devinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf)
125 unsigned int eax, ebx, ecx, edx;
126 union _cpuid4_leaf_eax cache_eax;
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 ? */
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);
143 static int __init find_num_cache_leaves(void)
145 unsigned int eax, ebx, ecx, edx;
146 union _cpuid4_leaf_eax cache_eax;
150 retval = MAX_CACHE_LEAVES;
151 /* Do cpuid(4) loop to find out num_cache_leaves */
152 for (i = 0; i < MAX_CACHE_LEAVES; i++) {
153 cpuid_count(4, i, &eax, &ebx, &ecx, &edx);
154 cache_eax.full = eax;
155 if (cache_eax.split.type == CACHE_TYPE_NULL) {
163 unsigned int __devinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
165 unsigned int trace = 0, l1i = 0, l1d = 0, l2 = 0, l3 = 0; /* Cache sizes */
166 unsigned int new_l1d = 0, new_l1i = 0; /* Cache sizes from cpuid(4) */
167 unsigned int new_l2 = 0, new_l3 = 0, i; /* Cache sizes from cpuid(4) */
169 if (c->cpuid_level > 4) {
170 static int is_initialized;
172 if (is_initialized == 0) {
173 /* Init num_cache_leaves from boot CPU */
174 num_cache_leaves = find_num_cache_leaves();
179 * Whenever possible use cpuid(4), deterministic cache
180 * parameters cpuid leaf to find the cache details
182 for (i = 0; i < num_cache_leaves; i++) {
183 struct _cpuid4_info this_leaf;
187 retval = cpuid4_cache_lookup(i, &this_leaf);
189 switch(this_leaf.eax.split.level) {
191 if (this_leaf.eax.split.type ==
193 new_l1d = this_leaf.size/1024;
194 else if (this_leaf.eax.split.type ==
196 new_l1i = this_leaf.size/1024;
199 new_l2 = this_leaf.size/1024;
202 new_l3 = this_leaf.size/1024;
210 if (c->cpuid_level > 1) {
211 /* supports eax=2 call */
214 unsigned char *dp = (unsigned char *)regs;
216 /* Number of times to iterate */
217 n = cpuid_eax(2) & 0xFF;
219 for ( i = 0 ; i < n ; i++ ) {
220 cpuid(2, ®s[0], ®s[1], ®s[2], ®s[3]);
222 /* If bit 31 is set, this is an unknown format */
223 for ( j = 0 ; j < 3 ; j++ ) {
224 if ( regs[j] < 0 ) regs[j] = 0;
227 /* Byte 0 is level count, not a descriptor */
228 for ( j = 1 ; j < 16 ; j++ ) {
229 unsigned char des = dp[j];
232 /* look up this descriptor in the table */
233 while (cache_table[k].descriptor != 0)
235 if (cache_table[k].descriptor == des) {
236 switch (cache_table[k].cache_type) {
238 l1i += cache_table[k].size;
241 l1d += cache_table[k].size;
244 l2 += cache_table[k].size;
247 l3 += cache_table[k].size;
250 trace += cache_table[k].size;
275 printk (KERN_INFO "CPU: Trace cache: %dK uops", trace);
277 printk (KERN_INFO "CPU: L1 I cache: %dK", l1i);
279 printk(", L1 D cache: %dK\n", l1d);
283 printk(KERN_INFO "CPU: L2 cache: %dK\n", l2);
285 printk(KERN_INFO "CPU: L3 cache: %dK\n", l3);
288 * This assumes the L3 cache is shared; it typically lives in
289 * the northbridge. The L1 caches are included by the L2
290 * cache, and so should not be included for the purpose of
291 * SMP switching weights.
293 c->x86_cache_size = l2 ? l2 : (l1i+l1d);
299 /* pointer to _cpuid4_info array (for each cache leaf) */
300 static struct _cpuid4_info *cpuid4_info[NR_CPUS];
301 #define CPUID4_INFO_IDX(x,y) (&((cpuid4_info[x])[y]))
304 static void __devinit cache_shared_cpu_map_setup(unsigned int cpu, int index)
306 struct _cpuid4_info *this_leaf;
307 unsigned long num_threads_sharing;
309 struct cpuinfo_x86 *c = cpu_data + cpu;
312 this_leaf = CPUID4_INFO_IDX(cpu, index);
313 num_threads_sharing = 1 + this_leaf->eax.split.num_threads_sharing;
315 if (num_threads_sharing == 1)
316 cpu_set(cpu, this_leaf->shared_cpu_map);
318 else if (num_threads_sharing == smp_num_siblings)
319 this_leaf->shared_cpu_map = cpu_sibling_map[cpu];
320 else if (num_threads_sharing == (c->x86_num_cores * smp_num_siblings))
321 this_leaf->shared_cpu_map = cpu_core_map[cpu];
323 printk(KERN_DEBUG "Number of CPUs sharing cache didn't match "
324 "any known set of CPUs\n");
328 static void __init cache_shared_cpu_map_setup(unsigned int cpu, int index) {}
331 static void free_cache_attributes(unsigned int cpu)
333 kfree(cpuid4_info[cpu]);
334 cpuid4_info[cpu] = NULL;
337 static int __devinit detect_cache_attributes(unsigned int cpu)
339 struct _cpuid4_info *this_leaf;
344 if (num_cache_leaves == 0)
347 cpuid4_info[cpu] = kmalloc(
348 sizeof(struct _cpuid4_info) * num_cache_leaves, GFP_KERNEL);
349 if (unlikely(cpuid4_info[cpu] == NULL))
351 memset(cpuid4_info[cpu], 0,
352 sizeof(struct _cpuid4_info) * num_cache_leaves);
354 oldmask = current->cpus_allowed;
355 retval = set_cpus_allowed(current, cpumask_of_cpu(cpu));
359 /* Do cpuid and store the results */
361 for (j = 0; j < num_cache_leaves; j++) {
362 this_leaf = CPUID4_INFO_IDX(cpu, j);
363 retval = cpuid4_cache_lookup(j, this_leaf);
364 if (unlikely(retval < 0))
366 cache_shared_cpu_map_setup(cpu, j);
368 set_cpus_allowed(current, oldmask);
372 free_cache_attributes(cpu);
378 #include <linux/kobject.h>
379 #include <linux/sysfs.h>
381 extern struct sysdev_class cpu_sysdev_class; /* from drivers/base/cpu.c */
383 /* pointer to kobject for cpuX/cache */
384 static struct kobject * cache_kobject[NR_CPUS];
386 struct _index_kobject {
389 unsigned short index;
392 /* pointer to array of kobjects for cpuX/cache/indexY */
393 static struct _index_kobject *index_kobject[NR_CPUS];
394 #define INDEX_KOBJECT_PTR(x,y) (&((index_kobject[x])[y]))
396 #define show_one_plus(file_name, object, val) \
397 static ssize_t show_##file_name \
398 (struct _cpuid4_info *this_leaf, char *buf) \
400 return sprintf (buf, "%lu\n", (unsigned long)this_leaf->object + val); \
403 show_one_plus(level, eax.split.level, 0);
404 show_one_plus(coherency_line_size, ebx.split.coherency_line_size, 1);
405 show_one_plus(physical_line_partition, ebx.split.physical_line_partition, 1);
406 show_one_plus(ways_of_associativity, ebx.split.ways_of_associativity, 1);
407 show_one_plus(number_of_sets, ecx.split.number_of_sets, 1);
409 static ssize_t show_size(struct _cpuid4_info *this_leaf, char *buf)
411 return sprintf (buf, "%luK\n", this_leaf->size / 1024);
414 static ssize_t show_shared_cpu_map(struct _cpuid4_info *this_leaf, char *buf)
416 char mask_str[NR_CPUS];
417 cpumask_scnprintf(mask_str, NR_CPUS, this_leaf->shared_cpu_map);
418 return sprintf(buf, "%s\n", mask_str);
421 static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf) {
422 switch(this_leaf->eax.split.type) {
423 case CACHE_TYPE_DATA:
424 return sprintf(buf, "Data\n");
426 case CACHE_TYPE_INST:
427 return sprintf(buf, "Instruction\n");
429 case CACHE_TYPE_UNIFIED:
430 return sprintf(buf, "Unified\n");
433 return sprintf(buf, "Unknown\n");
439 struct attribute attr;
440 ssize_t (*show)(struct _cpuid4_info *, char *);
441 ssize_t (*store)(struct _cpuid4_info *, const char *, size_t count);
444 #define define_one_ro(_name) \
445 static struct _cache_attr _name = \
446 __ATTR(_name, 0444, show_##_name, NULL)
448 define_one_ro(level);
450 define_one_ro(coherency_line_size);
451 define_one_ro(physical_line_partition);
452 define_one_ro(ways_of_associativity);
453 define_one_ro(number_of_sets);
455 define_one_ro(shared_cpu_map);
457 static struct attribute * default_attrs[] = {
460 &coherency_line_size.attr,
461 &physical_line_partition.attr,
462 &ways_of_associativity.attr,
463 &number_of_sets.attr,
465 &shared_cpu_map.attr,
469 #define to_object(k) container_of(k, struct _index_kobject, kobj)
470 #define to_attr(a) container_of(a, struct _cache_attr, attr)
472 static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
474 struct _cache_attr *fattr = to_attr(attr);
475 struct _index_kobject *this_leaf = to_object(kobj);
479 fattr->show(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
485 static ssize_t store(struct kobject * kobj, struct attribute * attr,
486 const char * buf, size_t count)
491 static struct sysfs_ops sysfs_ops = {
496 static struct kobj_type ktype_cache = {
497 .sysfs_ops = &sysfs_ops,
498 .default_attrs = default_attrs,
501 static struct kobj_type ktype_percpu_entry = {
502 .sysfs_ops = &sysfs_ops,
505 static void cpuid4_cache_sysfs_exit(unsigned int cpu)
507 kfree(cache_kobject[cpu]);
508 kfree(index_kobject[cpu]);
509 cache_kobject[cpu] = NULL;
510 index_kobject[cpu] = NULL;
511 free_cache_attributes(cpu);
514 static int __devinit cpuid4_cache_sysfs_init(unsigned int cpu)
517 if (num_cache_leaves == 0)
520 detect_cache_attributes(cpu);
521 if (cpuid4_info[cpu] == NULL)
524 /* Allocate all required memory */
525 cache_kobject[cpu] = kmalloc(sizeof(struct kobject), GFP_KERNEL);
526 if (unlikely(cache_kobject[cpu] == NULL))
528 memset(cache_kobject[cpu], 0, sizeof(struct kobject));
530 index_kobject[cpu] = kmalloc(
531 sizeof(struct _index_kobject ) * num_cache_leaves, GFP_KERNEL);
532 if (unlikely(index_kobject[cpu] == NULL))
534 memset(index_kobject[cpu], 0,
535 sizeof(struct _index_kobject) * num_cache_leaves);
540 cpuid4_cache_sysfs_exit(cpu);
544 /* Add/Remove cache interface for CPU device */
545 static int __devinit cache_add_dev(struct sys_device * sys_dev)
547 unsigned int cpu = sys_dev->id;
549 struct _index_kobject *this_object;
552 retval = cpuid4_cache_sysfs_init(cpu);
553 if (unlikely(retval < 0))
556 cache_kobject[cpu]->parent = &sys_dev->kobj;
557 kobject_set_name(cache_kobject[cpu], "%s", "cache");
558 cache_kobject[cpu]->ktype = &ktype_percpu_entry;
559 retval = kobject_register(cache_kobject[cpu]);
561 for (i = 0; i < num_cache_leaves; i++) {
562 this_object = INDEX_KOBJECT_PTR(cpu,i);
563 this_object->cpu = cpu;
564 this_object->index = i;
565 this_object->kobj.parent = cache_kobject[cpu];
566 kobject_set_name(&(this_object->kobj), "index%1lu", i);
567 this_object->kobj.ktype = &ktype_cache;
568 retval = kobject_register(&(this_object->kobj));
569 if (unlikely(retval)) {
570 for (j = 0; j < i; j++) {
572 &(INDEX_KOBJECT_PTR(cpu,j)->kobj));
574 kobject_unregister(cache_kobject[cpu]);
575 cpuid4_cache_sysfs_exit(cpu);
582 static int __devexit cache_remove_dev(struct sys_device * sys_dev)
584 unsigned int cpu = sys_dev->id;
587 for (i = 0; i < num_cache_leaves; i++)
588 kobject_unregister(&(INDEX_KOBJECT_PTR(cpu,i)->kobj));
589 kobject_unregister(cache_kobject[cpu]);
590 cpuid4_cache_sysfs_exit(cpu);
594 static struct sysdev_driver cache_sysdev_driver = {
595 .add = cache_add_dev,
596 .remove = __devexit_p(cache_remove_dev),
599 /* Register/Unregister the cpu_cache driver */
600 static int __devinit cache_register_driver(void)
602 if (num_cache_leaves == 0)
605 return sysdev_driver_register(&cpu_sysdev_class,&cache_sysdev_driver);
608 device_initcall(cache_register_driver);