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 * Andi Kleen / Andreas Herrmann : CPUID4 emulation on AMD.
8 */
9
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/device.h>
13 #include <linux/compiler.h>
14 #include <linux/cpu.h>
15 #include <linux/sched.h>
16 #include <linux/pci.h>
17
18 #include <asm/processor.h>
19 #include <asm/smp.h>
20
21 #define LVL_1_INST 1
22 #define LVL_1_DATA 2
23 #define LVL_2 3
24 #define LVL_3 4
25 #define LVL_TRACE 5
26
27 struct _cache_table
28 {
29 unsigned char descriptor;
30 char cache_type;
31 short size;
32 };
33
34 /* all the cache descriptor types we care about (no TLB or trace cache entries) */
35 static struct _cache_table cache_table[] __cpuinitdata =
36 {
37 { 0x06, LVL_1_INST, 8 }, /* 4-way set assoc, 32 byte line size */
38 { 0x08, LVL_1_INST, 16 }, /* 4-way set assoc, 32 byte line size */
39 { 0x09, LVL_1_INST, 32 }, /* 4-way set assoc, 64 byte line size */
40 { 0x0a, LVL_1_DATA, 8 }, /* 2 way set assoc, 32 byte line size */
41 { 0x0c, LVL_1_DATA, 16 }, /* 4-way set assoc, 32 byte line size */
42 { 0x0d, LVL_1_DATA, 16 }, /* 4-way set assoc, 64 byte line size */
43 { 0x21, LVL_2, 256 }, /* 8-way set assoc, 64 byte line size */
44 { 0x22, LVL_3, 512 }, /* 4-way set assoc, sectored cache, 64 byte line size */
45 { 0x23, LVL_3, 1024 }, /* 8-way set assoc, sectored cache, 64 byte line size */
46 { 0x25, LVL_3, 2048 }, /* 8-way set assoc, sectored cache, 64 byte line size */
47 { 0x29, LVL_3, 4096 }, /* 8-way set assoc, sectored cache, 64 byte line size */
48 { 0x2c, LVL_1_DATA, 32 }, /* 8-way set assoc, 64 byte line size */
49 { 0x30, LVL_1_INST, 32 }, /* 8-way set assoc, 64 byte line size */
50 { 0x39, LVL_2, 128 }, /* 4-way set assoc, sectored cache, 64 byte line size */
51 { 0x3a, LVL_2, 192 }, /* 6-way set assoc, sectored cache, 64 byte line size */
52 { 0x3b, LVL_2, 128 }, /* 2-way set assoc, sectored cache, 64 byte line size */
53 { 0x3c, LVL_2, 256 }, /* 4-way set assoc, sectored cache, 64 byte line size */
54 { 0x3d, LVL_2, 384 }, /* 6-way set assoc, sectored cache, 64 byte line size */
55 { 0x3e, LVL_2, 512 }, /* 4-way set assoc, sectored cache, 64 byte line size */
56 { 0x3f, LVL_2, 256 }, /* 2-way set assoc, 64 byte line size */
57 { 0x41, LVL_2, 128 }, /* 4-way set assoc, 32 byte line size */
58 { 0x42, LVL_2, 256 }, /* 4-way set assoc, 32 byte line size */
59 { 0x43, LVL_2, 512 }, /* 4-way set assoc, 32 byte line size */
60 { 0x44, LVL_2, 1024 }, /* 4-way set assoc, 32 byte line size */
61 { 0x45, LVL_2, 2048 }, /* 4-way set assoc, 32 byte line size */
62 { 0x46, LVL_3, 4096 }, /* 4-way set assoc, 64 byte line size */
63 { 0x47, LVL_3, 8192 }, /* 8-way set assoc, 64 byte line size */
64 { 0x49, LVL_3, 4096 }, /* 16-way set assoc, 64 byte line size */
65 { 0x4a, LVL_3, 6144 }, /* 12-way set assoc, 64 byte line size */
66 { 0x4b, LVL_3, 8192 }, /* 16-way set assoc, 64 byte line size */
67 { 0x4c, LVL_3, 12288 }, /* 12-way set assoc, 64 byte line size */
68 { 0x4d, LVL_3, 16384 }, /* 16-way set assoc, 64 byte line size */
69 { 0x4e, LVL_2, 6144 }, /* 24-way set assoc, 64 byte line size */
70 { 0x60, LVL_1_DATA, 16 }, /* 8-way set assoc, sectored cache, 64 byte line size */
71 { 0x66, LVL_1_DATA, 8 }, /* 4-way set assoc, sectored cache, 64 byte line size */
72 { 0x67, LVL_1_DATA, 16 }, /* 4-way set assoc, sectored cache, 64 byte line size */
73 { 0x68, LVL_1_DATA, 32 }, /* 4-way set assoc, sectored cache, 64 byte line size */
74 { 0x70, LVL_TRACE, 12 }, /* 8-way set assoc */
75 { 0x71, LVL_TRACE, 16 }, /* 8-way set assoc */
76 { 0x72, LVL_TRACE, 32 }, /* 8-way set assoc */
77 { 0x73, LVL_TRACE, 64 }, /* 8-way set assoc */
78 { 0x78, LVL_2, 1024 }, /* 4-way set assoc, 64 byte line size */
79 { 0x79, LVL_2, 128 }, /* 8-way set assoc, sectored cache, 64 byte line size */
80 { 0x7a, LVL_2, 256 }, /* 8-way set assoc, sectored cache, 64 byte line size */
81 { 0x7b, LVL_2, 512 }, /* 8-way set assoc, sectored cache, 64 byte line size */
82 { 0x7c, LVL_2, 1024 }, /* 8-way set assoc, sectored cache, 64 byte line size */
83 { 0x7d, LVL_2, 2048 }, /* 8-way set assoc, 64 byte line size */
84 { 0x7f, LVL_2, 512 }, /* 2-way set assoc, 64 byte line size */
85 { 0x82, LVL_2, 256 }, /* 8-way set assoc, 32 byte line size */
86 { 0x83, LVL_2, 512 }, /* 8-way set assoc, 32 byte line size */
87 { 0x84, LVL_2, 1024 }, /* 8-way set assoc, 32 byte line size */
88 { 0x85, LVL_2, 2048 }, /* 8-way set assoc, 32 byte line size */
89 { 0x86, LVL_2, 512 }, /* 4-way set assoc, 64 byte line size */
90 { 0x87, LVL_2, 1024 }, /* 8-way set assoc, 64 byte line size */
91 { 0xd0, LVL_3, 512 }, /* 4-way set assoc, 64 byte line size */
92 { 0xd1, LVL_3, 1024 }, /* 4-way set assoc, 64 byte line size */
93 { 0xd2, LVL_3, 2048 }, /* 4-way set assoc, 64 byte line size */
94 { 0xd6, LVL_3, 1024 }, /* 8-way set assoc, 64 byte line size */
95 { 0xd7, LVL_3, 2038 }, /* 8-way set assoc, 64 byte line size */
96 { 0xd8, LVL_3, 4096 }, /* 12-way set assoc, 64 byte line size */
97 { 0xdc, LVL_3, 2048 }, /* 12-way set assoc, 64 byte line size */
98 { 0xdd, LVL_3, 4096 }, /* 12-way set assoc, 64 byte line size */
99 { 0xde, LVL_3, 8192 }, /* 12-way set assoc, 64 byte line size */
100 { 0xe2, LVL_3, 2048 }, /* 16-way set assoc, 64 byte line size */
101 { 0xe3, LVL_3, 4096 }, /* 16-way set assoc, 64 byte line size */
102 { 0xe4, LVL_3, 8192 }, /* 16-way set assoc, 64 byte line size */
103 { 0x00, 0, 0}
104 };
105
106
107 enum _cache_type
108 {
109 CACHE_TYPE_NULL = 0,
110 CACHE_TYPE_DATA = 1,
111 CACHE_TYPE_INST = 2,
112 CACHE_TYPE_UNIFIED = 3
113 };
114
115 union _cpuid4_leaf_eax {
116 struct {
117 enum _cache_type type:5;
118 unsigned int level:3;
119 unsigned int is_self_initializing:1;
120 unsigned int is_fully_associative:1;
121 unsigned int reserved:4;
122 unsigned int num_threads_sharing:12;
123 unsigned int num_cores_on_die:6;
124 } split;
125 u32 full;
126 };
127
128 union _cpuid4_leaf_ebx {
129 struct {
130 unsigned int coherency_line_size:12;
131 unsigned int physical_line_partition:10;
132 unsigned int ways_of_associativity:10;
133 } split;
134 u32 full;
135 };
136
137 union _cpuid4_leaf_ecx {
138 struct {
139 unsigned int number_of_sets:32;
140 } split;
141 u32 full;
142 };
143
144 struct _cpuid4_info {
145 union _cpuid4_leaf_eax eax;
146 union _cpuid4_leaf_ebx ebx;
147 union _cpuid4_leaf_ecx ecx;
148 unsigned long size;
149 unsigned long can_disable;
150 cpumask_t shared_cpu_map; /* future?: only cpus/node is needed */
151 };
152
153 #ifdef CONFIG_PCI
154 static struct pci_device_id k8_nb_id[] = {
155 { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1103) },
156 { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1203) },
157 {}
158 };
159 #endif
160
161 unsigned short num_cache_leaves;
162
163 /* AMD doesn't have CPUID4. Emulate it here to report the same
164 information to the user. This makes some assumptions about the machine:
165 L2 not shared, no SMT etc. that is currently true on AMD CPUs.
166
167 In theory the TLBs could be reported as fake type (they are in "dummy").
168 Maybe later */
169 union l1_cache {
170 struct {
171 unsigned line_size : 8;
172 unsigned lines_per_tag : 8;
173 unsigned assoc : 8;
174 unsigned size_in_kb : 8;
175 };
176 unsigned val;
177 };
178
179 union l2_cache {
180 struct {
181 unsigned line_size : 8;
182 unsigned lines_per_tag : 4;
183 unsigned assoc : 4;
184 unsigned size_in_kb : 16;
185 };
186 unsigned val;
187 };
188
189 union l3_cache {
190 struct {
191 unsigned line_size : 8;
192 unsigned lines_per_tag : 4;
193 unsigned assoc : 4;
194 unsigned res : 2;
195 unsigned size_encoded : 14;
196 };
197 unsigned val;
198 };
199
200 static unsigned short assocs[] __cpuinitdata = {
201 [1] = 1, [2] = 2, [4] = 4, [6] = 8,
202 [8] = 16, [0xa] = 32, [0xb] = 48,
203 [0xc] = 64,
204 [0xf] = 0xffff // ??
205 };
206
207 static unsigned char levels[] __cpuinitdata = { 1, 1, 2, 3 };
208 static unsigned char types[] __cpuinitdata = { 1, 2, 3, 3 };
209
210 static void __cpuinit
amd_cpuid4(int leaf,union _cpuid4_leaf_eax * eax,union _cpuid4_leaf_ebx * ebx,union _cpuid4_leaf_ecx * ecx)211 amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax,
212 union _cpuid4_leaf_ebx *ebx,
213 union _cpuid4_leaf_ecx *ecx)
214 {
215 unsigned dummy;
216 unsigned line_size, lines_per_tag, assoc, size_in_kb;
217 union l1_cache l1i, l1d;
218 union l2_cache l2;
219 union l3_cache l3;
220 union l1_cache *l1 = &l1d;
221
222 eax->full = 0;
223 ebx->full = 0;
224 ecx->full = 0;
225
226 cpuid(0x80000005, &dummy, &dummy, &l1d.val, &l1i.val);
227 cpuid(0x80000006, &dummy, &dummy, &l2.val, &l3.val);
228
229 switch (leaf) {
230 case 1:
231 l1 = &l1i;
232 case 0:
233 if (!l1->val)
234 return;
235 assoc = l1->assoc;
236 line_size = l1->line_size;
237 lines_per_tag = l1->lines_per_tag;
238 size_in_kb = l1->size_in_kb;
239 break;
240 case 2:
241 if (!l2.val)
242 return;
243 assoc = l2.assoc;
244 line_size = l2.line_size;
245 lines_per_tag = l2.lines_per_tag;
246 /* cpu_data has errata corrections for K7 applied */
247 size_in_kb = current_cpu_data.x86_cache_size;
248 break;
249 case 3:
250 if (!l3.val)
251 return;
252 assoc = l3.assoc;
253 line_size = l3.line_size;
254 lines_per_tag = l3.lines_per_tag;
255 size_in_kb = l3.size_encoded * 512;
256 break;
257 default:
258 return;
259 }
260
261 eax->split.is_self_initializing = 1;
262 eax->split.type = types[leaf];
263 eax->split.level = levels[leaf];
264 if (leaf == 3)
265 eax->split.num_threads_sharing = current_cpu_data.x86_max_cores - 1;
266 else
267 eax->split.num_threads_sharing = 0;
268 eax->split.num_cores_on_die = current_cpu_data.x86_max_cores - 1;
269
270
271 if (assoc == 0xf)
272 eax->split.is_fully_associative = 1;
273 ebx->split.coherency_line_size = line_size - 1;
274 ebx->split.ways_of_associativity = assocs[assoc] - 1;
275 ebx->split.physical_line_partition = lines_per_tag - 1;
276 ecx->split.number_of_sets = (size_in_kb * 1024) / line_size /
277 (ebx->split.ways_of_associativity + 1) - 1;
278 }
279
280 static void __cpuinit
amd_check_l3_disable(int index,struct _cpuid4_info * this_leaf)281 amd_check_l3_disable(int index, struct _cpuid4_info *this_leaf)
282 {
283 if (index < 3)
284 return;
285 this_leaf->can_disable = 1;
286 }
287
288 static int
cpuid4_cache_lookup(int index,struct _cpuid4_info * this_leaf)289 __cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf)
290 {
291 union _cpuid4_leaf_eax eax;
292 union _cpuid4_leaf_ebx ebx;
293 union _cpuid4_leaf_ecx ecx;
294 unsigned edx;
295
296 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
297 amd_cpuid4(index, &eax, &ebx, &ecx);
298 if (boot_cpu_data.x86 >= 0x10)
299 amd_check_l3_disable(index, this_leaf);
300 } else {
301 cpuid_count(4, index, &eax.full, &ebx.full, &ecx.full, &edx);
302 }
303
304 if (eax.split.type == CACHE_TYPE_NULL)
305 return -EIO; /* better error ? */
306
307 this_leaf->eax = eax;
308 this_leaf->ebx = ebx;
309 this_leaf->ecx = ecx;
310 this_leaf->size = (ecx.split.number_of_sets + 1) *
311 (ebx.split.coherency_line_size + 1) *
312 (ebx.split.physical_line_partition + 1) *
313 (ebx.split.ways_of_associativity + 1);
314 return 0;
315 }
316
find_num_cache_leaves(void)317 static int __cpuinit find_num_cache_leaves(void)
318 {
319 unsigned int eax, ebx, ecx, edx;
320 union _cpuid4_leaf_eax cache_eax;
321 int i = -1;
322
323 do {
324 ++i;
325 /* Do cpuid(4) loop to find out num_cache_leaves */
326 cpuid_count(4, i, &eax, &ebx, &ecx, &edx);
327 cache_eax.full = eax;
328 } while (cache_eax.split.type != CACHE_TYPE_NULL);
329 return i;
330 }
331
init_intel_cacheinfo(struct cpuinfo_x86 * c)332 unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
333 {
334 unsigned int trace = 0, l1i = 0, l1d = 0, l2 = 0, l3 = 0; /* Cache sizes */
335 unsigned int new_l1d = 0, new_l1i = 0; /* Cache sizes from cpuid(4) */
336 unsigned int new_l2 = 0, new_l3 = 0, i; /* Cache sizes from cpuid(4) */
337 unsigned int l2_id = 0, l3_id = 0, num_threads_sharing, index_msb;
338 #ifdef CONFIG_X86_HT
339 unsigned int cpu = c->cpu_index;
340 #endif
341
342 if (c->cpuid_level > 3) {
343 static int is_initialized;
344
345 if (is_initialized == 0) {
346 /* Init num_cache_leaves from boot CPU */
347 num_cache_leaves = find_num_cache_leaves();
348 is_initialized++;
349 }
350
351 /*
352 * Whenever possible use cpuid(4), deterministic cache
353 * parameters cpuid leaf to find the cache details
354 */
355 for (i = 0; i < num_cache_leaves; i++) {
356 struct _cpuid4_info this_leaf;
357
358 int retval;
359
360 retval = cpuid4_cache_lookup(i, &this_leaf);
361 if (retval >= 0) {
362 switch(this_leaf.eax.split.level) {
363 case 1:
364 if (this_leaf.eax.split.type ==
365 CACHE_TYPE_DATA)
366 new_l1d = this_leaf.size/1024;
367 else if (this_leaf.eax.split.type ==
368 CACHE_TYPE_INST)
369 new_l1i = this_leaf.size/1024;
370 break;
371 case 2:
372 new_l2 = this_leaf.size/1024;
373 num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
374 index_msb = get_count_order(num_threads_sharing);
375 l2_id = c->apicid >> index_msb;
376 break;
377 case 3:
378 new_l3 = this_leaf.size/1024;
379 num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
380 index_msb = get_count_order(num_threads_sharing);
381 l3_id = c->apicid >> index_msb;
382 break;
383 default:
384 break;
385 }
386 }
387 }
388 }
389 /*
390 * Don't use cpuid2 if cpuid4 is supported. For P4, we use cpuid2 for
391 * trace cache
392 */
393 if ((num_cache_leaves == 0 || c->x86 == 15) && c->cpuid_level > 1) {
394 /* supports eax=2 call */
395 int j, n;
396 unsigned int regs[4];
397 unsigned char *dp = (unsigned char *)regs;
398 int only_trace = 0;
399
400 if (num_cache_leaves != 0 && c->x86 == 15)
401 only_trace = 1;
402
403 /* Number of times to iterate */
404 n = cpuid_eax(2) & 0xFF;
405
406 for ( i = 0 ; i < n ; i++ ) {
407 cpuid(2, ®s[0], ®s[1], ®s[2], ®s[3]);
408
409 /* If bit 31 is set, this is an unknown format */
410 for ( j = 0 ; j < 3 ; j++ ) {
411 if (regs[j] & (1 << 31)) regs[j] = 0;
412 }
413
414 /* Byte 0 is level count, not a descriptor */
415 for ( j = 1 ; j < 16 ; j++ ) {
416 unsigned char des = dp[j];
417 unsigned char k = 0;
418
419 /* look up this descriptor in the table */
420 while (cache_table[k].descriptor != 0)
421 {
422 if (cache_table[k].descriptor == des) {
423 if (only_trace && cache_table[k].cache_type != LVL_TRACE)
424 break;
425 switch (cache_table[k].cache_type) {
426 case LVL_1_INST:
427 l1i += cache_table[k].size;
428 break;
429 case LVL_1_DATA:
430 l1d += cache_table[k].size;
431 break;
432 case LVL_2:
433 l2 += cache_table[k].size;
434 break;
435 case LVL_3:
436 l3 += cache_table[k].size;
437 break;
438 case LVL_TRACE:
439 trace += cache_table[k].size;
440 break;
441 }
442
443 break;
444 }
445
446 k++;
447 }
448 }
449 }
450 }
451
452 if (new_l1d)
453 l1d = new_l1d;
454
455 if (new_l1i)
456 l1i = new_l1i;
457
458 if (new_l2) {
459 l2 = new_l2;
460 #ifdef CONFIG_X86_HT
461 per_cpu(cpu_llc_id, cpu) = l2_id;
462 #endif
463 }
464
465 if (new_l3) {
466 l3 = new_l3;
467 #ifdef CONFIG_X86_HT
468 per_cpu(cpu_llc_id, cpu) = l3_id;
469 #endif
470 }
471
472 if (trace)
473 printk (KERN_INFO "CPU: Trace cache: %dK uops", trace);
474 else if ( l1i )
475 printk (KERN_INFO "CPU: L1 I cache: %dK", l1i);
476
477 if (l1d)
478 printk(", L1 D cache: %dK\n", l1d);
479 else
480 printk("\n");
481
482 if (l2)
483 printk(KERN_INFO "CPU: L2 cache: %dK\n", l2);
484
485 if (l3)
486 printk(KERN_INFO "CPU: L3 cache: %dK\n", l3);
487
488 c->x86_cache_size = l3 ? l3 : (l2 ? l2 : (l1i+l1d));
489
490 return l2;
491 }
492
493 /* pointer to _cpuid4_info array (for each cache leaf) */
494 static DEFINE_PER_CPU(struct _cpuid4_info *, cpuid4_info);
495 #define CPUID4_INFO_IDX(x, y) (&((per_cpu(cpuid4_info, x))[y]))
496
497 #ifdef CONFIG_SMP
cache_shared_cpu_map_setup(unsigned int cpu,int index)498 static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index)
499 {
500 struct _cpuid4_info *this_leaf, *sibling_leaf;
501 unsigned long num_threads_sharing;
502 int index_msb, i;
503 struct cpuinfo_x86 *c = &cpu_data(cpu);
504
505 this_leaf = CPUID4_INFO_IDX(cpu, index);
506 num_threads_sharing = 1 + this_leaf->eax.split.num_threads_sharing;
507
508 if (num_threads_sharing == 1)
509 cpu_set(cpu, this_leaf->shared_cpu_map);
510 else {
511 index_msb = get_count_order(num_threads_sharing);
512
513 for_each_online_cpu(i) {
514 if (cpu_data(i).apicid >> index_msb ==
515 c->apicid >> index_msb) {
516 cpu_set(i, this_leaf->shared_cpu_map);
517 if (i != cpu && per_cpu(cpuid4_info, i)) {
518 sibling_leaf = CPUID4_INFO_IDX(i, index);
519 cpu_set(cpu, sibling_leaf->shared_cpu_map);
520 }
521 }
522 }
523 }
524 }
cache_remove_shared_cpu_map(unsigned int cpu,int index)525 static void __cpuinit cache_remove_shared_cpu_map(unsigned int cpu, int index)
526 {
527 struct _cpuid4_info *this_leaf, *sibling_leaf;
528 int sibling;
529
530 this_leaf = CPUID4_INFO_IDX(cpu, index);
531 for_each_cpu_mask_nr(sibling, this_leaf->shared_cpu_map) {
532 sibling_leaf = CPUID4_INFO_IDX(sibling, index);
533 cpu_clear(cpu, sibling_leaf->shared_cpu_map);
534 }
535 }
536 #else
cache_shared_cpu_map_setup(unsigned int cpu,int index)537 static void __cpuinit cache_shared_cpu_map_setup(unsigned int cpu, int index) {}
cache_remove_shared_cpu_map(unsigned int cpu,int index)538 static void __cpuinit cache_remove_shared_cpu_map(unsigned int cpu, int index) {}
539 #endif
540
free_cache_attributes(unsigned int cpu)541 static void __cpuinit free_cache_attributes(unsigned int cpu)
542 {
543 int i;
544
545 for (i = 0; i < num_cache_leaves; i++)
546 cache_remove_shared_cpu_map(cpu, i);
547
548 kfree(per_cpu(cpuid4_info, cpu));
549 per_cpu(cpuid4_info, cpu) = NULL;
550 }
551
get_cpu_leaves(void * _retval)552 static void __cpuinit get_cpu_leaves(void *_retval)
553 {
554 int j, *retval = _retval, cpu = smp_processor_id();
555
556 /* Do cpuid and store the results */
557 for (j = 0; j < num_cache_leaves; j++) {
558 struct _cpuid4_info *this_leaf;
559 this_leaf = CPUID4_INFO_IDX(cpu, j);
560 *retval = cpuid4_cache_lookup(j, this_leaf);
561 if (unlikely(*retval < 0)) {
562 int i;
563
564 for (i = 0; i < j; i++)
565 cache_remove_shared_cpu_map(cpu, i);
566 break;
567 }
568 cache_shared_cpu_map_setup(cpu, j);
569 }
570 }
571
detect_cache_attributes(unsigned int cpu)572 static int __cpuinit detect_cache_attributes(unsigned int cpu)
573 {
574 int retval;
575
576 if (num_cache_leaves == 0)
577 return -ENOENT;
578
579 per_cpu(cpuid4_info, cpu) = kzalloc(
580 sizeof(struct _cpuid4_info) * num_cache_leaves, GFP_KERNEL);
581 if (per_cpu(cpuid4_info, cpu) == NULL)
582 return -ENOMEM;
583
584 smp_call_function_single(cpu, get_cpu_leaves, &retval, true);
585 if (retval) {
586 kfree(per_cpu(cpuid4_info, cpu));
587 per_cpu(cpuid4_info, cpu) = NULL;
588 }
589
590 return retval;
591 }
592
593 #ifdef CONFIG_SYSFS
594
595 #include <linux/kobject.h>
596 #include <linux/sysfs.h>
597
598 extern struct sysdev_class cpu_sysdev_class; /* from drivers/base/cpu.c */
599
600 /* pointer to kobject for cpuX/cache */
601 static DEFINE_PER_CPU(struct kobject *, cache_kobject);
602
603 struct _index_kobject {
604 struct kobject kobj;
605 unsigned int cpu;
606 unsigned short index;
607 };
608
609 /* pointer to array of kobjects for cpuX/cache/indexY */
610 static DEFINE_PER_CPU(struct _index_kobject *, index_kobject);
611 #define INDEX_KOBJECT_PTR(x, y) (&((per_cpu(index_kobject, x))[y]))
612
613 #define show_one_plus(file_name, object, val) \
614 static ssize_t show_##file_name \
615 (struct _cpuid4_info *this_leaf, char *buf) \
616 { \
617 return sprintf (buf, "%lu\n", (unsigned long)this_leaf->object + val); \
618 }
619
620 show_one_plus(level, eax.split.level, 0);
621 show_one_plus(coherency_line_size, ebx.split.coherency_line_size, 1);
622 show_one_plus(physical_line_partition, ebx.split.physical_line_partition, 1);
623 show_one_plus(ways_of_associativity, ebx.split.ways_of_associativity, 1);
624 show_one_plus(number_of_sets, ecx.split.number_of_sets, 1);
625
show_size(struct _cpuid4_info * this_leaf,char * buf)626 static ssize_t show_size(struct _cpuid4_info *this_leaf, char *buf)
627 {
628 return sprintf (buf, "%luK\n", this_leaf->size / 1024);
629 }
630
show_shared_cpu_map_func(struct _cpuid4_info * this_leaf,int type,char * buf)631 static ssize_t show_shared_cpu_map_func(struct _cpuid4_info *this_leaf,
632 int type, char *buf)
633 {
634 ptrdiff_t len = PTR_ALIGN(buf + PAGE_SIZE - 1, PAGE_SIZE) - buf;
635 int n = 0;
636
637 if (len > 1) {
638 cpumask_t *mask = &this_leaf->shared_cpu_map;
639
640 n = type?
641 cpulist_scnprintf(buf, len-2, mask) :
642 cpumask_scnprintf(buf, len-2, mask);
643 buf[n++] = '\n';
644 buf[n] = '\0';
645 }
646 return n;
647 }
648
show_shared_cpu_map(struct _cpuid4_info * leaf,char * buf)649 static inline ssize_t show_shared_cpu_map(struct _cpuid4_info *leaf, char *buf)
650 {
651 return show_shared_cpu_map_func(leaf, 0, buf);
652 }
653
show_shared_cpu_list(struct _cpuid4_info * leaf,char * buf)654 static inline ssize_t show_shared_cpu_list(struct _cpuid4_info *leaf, char *buf)
655 {
656 return show_shared_cpu_map_func(leaf, 1, buf);
657 }
658
show_type(struct _cpuid4_info * this_leaf,char * buf)659 static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf)
660 {
661 switch (this_leaf->eax.split.type) {
662 case CACHE_TYPE_DATA:
663 return sprintf(buf, "Data\n");
664 case CACHE_TYPE_INST:
665 return sprintf(buf, "Instruction\n");
666 case CACHE_TYPE_UNIFIED:
667 return sprintf(buf, "Unified\n");
668 default:
669 return sprintf(buf, "Unknown\n");
670 }
671 }
672
673 #define to_object(k) container_of(k, struct _index_kobject, kobj)
674 #define to_attr(a) container_of(a, struct _cache_attr, attr)
675
676 #ifdef CONFIG_PCI
get_k8_northbridge(int node)677 static struct pci_dev *get_k8_northbridge(int node)
678 {
679 struct pci_dev *dev = NULL;
680 int i;
681
682 for (i = 0; i <= node; i++) {
683 do {
684 dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev);
685 if (!dev)
686 break;
687 } while (!pci_match_id(&k8_nb_id[0], dev));
688 if (!dev)
689 break;
690 }
691 return dev;
692 }
693 #else
get_k8_northbridge(int node)694 static struct pci_dev *get_k8_northbridge(int node)
695 {
696 return NULL;
697 }
698 #endif
699
show_cache_disable(struct _cpuid4_info * this_leaf,char * buf)700 static ssize_t show_cache_disable(struct _cpuid4_info *this_leaf, char *buf)
701 {
702 int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map));
703 struct pci_dev *dev = NULL;
704 ssize_t ret = 0;
705 int i;
706
707 if (!this_leaf->can_disable)
708 return sprintf(buf, "Feature not enabled\n");
709
710 dev = get_k8_northbridge(node);
711 if (!dev) {
712 printk(KERN_ERR "Attempting AMD northbridge operation on a system with no northbridge\n");
713 return -EINVAL;
714 }
715
716 for (i = 0; i < 2; i++) {
717 unsigned int reg;
718
719 pci_read_config_dword(dev, 0x1BC + i * 4, ®);
720
721 ret += sprintf(buf, "%sEntry: %d\n", buf, i);
722 ret += sprintf(buf, "%sReads: %s\tNew Entries: %s\n",
723 buf,
724 reg & 0x80000000 ? "Disabled" : "Allowed",
725 reg & 0x40000000 ? "Disabled" : "Allowed");
726 ret += sprintf(buf, "%sSubCache: %x\tIndex: %x\n",
727 buf, (reg & 0x30000) >> 16, reg & 0xfff);
728 }
729 return ret;
730 }
731
732 static ssize_t
store_cache_disable(struct _cpuid4_info * this_leaf,const char * buf,size_t count)733 store_cache_disable(struct _cpuid4_info *this_leaf, const char *buf,
734 size_t count)
735 {
736 int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map));
737 struct pci_dev *dev = NULL;
738 unsigned int ret, index, val;
739
740 if (!this_leaf->can_disable)
741 return 0;
742
743 if (strlen(buf) > 15)
744 return -EINVAL;
745
746 ret = sscanf(buf, "%x %x", &index, &val);
747 if (ret != 2)
748 return -EINVAL;
749 if (index > 1)
750 return -EINVAL;
751
752 val |= 0xc0000000;
753 dev = get_k8_northbridge(node);
754 if (!dev) {
755 printk(KERN_ERR "Attempting AMD northbridge operation on a system with no northbridge\n");
756 return -EINVAL;
757 }
758
759 pci_write_config_dword(dev, 0x1BC + index * 4, val & ~0x40000000);
760 wbinvd();
761 pci_write_config_dword(dev, 0x1BC + index * 4, val);
762
763 return 1;
764 }
765
766 struct _cache_attr {
767 struct attribute attr;
768 ssize_t (*show)(struct _cpuid4_info *, char *);
769 ssize_t (*store)(struct _cpuid4_info *, const char *, size_t count);
770 };
771
772 #define define_one_ro(_name) \
773 static struct _cache_attr _name = \
774 __ATTR(_name, 0444, show_##_name, NULL)
775
776 define_one_ro(level);
777 define_one_ro(type);
778 define_one_ro(coherency_line_size);
779 define_one_ro(physical_line_partition);
780 define_one_ro(ways_of_associativity);
781 define_one_ro(number_of_sets);
782 define_one_ro(size);
783 define_one_ro(shared_cpu_map);
784 define_one_ro(shared_cpu_list);
785
786 static struct _cache_attr cache_disable = __ATTR(cache_disable, 0644, show_cache_disable, store_cache_disable);
787
788 static struct attribute * default_attrs[] = {
789 &type.attr,
790 &level.attr,
791 &coherency_line_size.attr,
792 &physical_line_partition.attr,
793 &ways_of_associativity.attr,
794 &number_of_sets.attr,
795 &size.attr,
796 &shared_cpu_map.attr,
797 &shared_cpu_list.attr,
798 &cache_disable.attr,
799 NULL
800 };
801
show(struct kobject * kobj,struct attribute * attr,char * buf)802 static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
803 {
804 struct _cache_attr *fattr = to_attr(attr);
805 struct _index_kobject *this_leaf = to_object(kobj);
806 ssize_t ret;
807
808 ret = fattr->show ?
809 fattr->show(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
810 buf) :
811 0;
812 return ret;
813 }
814
store(struct kobject * kobj,struct attribute * attr,const char * buf,size_t count)815 static ssize_t store(struct kobject * kobj, struct attribute * attr,
816 const char * buf, size_t count)
817 {
818 struct _cache_attr *fattr = to_attr(attr);
819 struct _index_kobject *this_leaf = to_object(kobj);
820 ssize_t ret;
821
822 ret = fattr->store ?
823 fattr->store(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
824 buf, count) :
825 0;
826 return ret;
827 }
828
829 static struct sysfs_ops sysfs_ops = {
830 .show = show,
831 .store = store,
832 };
833
834 static struct kobj_type ktype_cache = {
835 .sysfs_ops = &sysfs_ops,
836 .default_attrs = default_attrs,
837 };
838
839 static struct kobj_type ktype_percpu_entry = {
840 .sysfs_ops = &sysfs_ops,
841 };
842
cpuid4_cache_sysfs_exit(unsigned int cpu)843 static void __cpuinit cpuid4_cache_sysfs_exit(unsigned int cpu)
844 {
845 kfree(per_cpu(cache_kobject, cpu));
846 kfree(per_cpu(index_kobject, cpu));
847 per_cpu(cache_kobject, cpu) = NULL;
848 per_cpu(index_kobject, cpu) = NULL;
849 free_cache_attributes(cpu);
850 }
851
cpuid4_cache_sysfs_init(unsigned int cpu)852 static int __cpuinit cpuid4_cache_sysfs_init(unsigned int cpu)
853 {
854 int err;
855
856 if (num_cache_leaves == 0)
857 return -ENOENT;
858
859 err = detect_cache_attributes(cpu);
860 if (err)
861 return err;
862
863 /* Allocate all required memory */
864 per_cpu(cache_kobject, cpu) =
865 kzalloc(sizeof(struct kobject), GFP_KERNEL);
866 if (unlikely(per_cpu(cache_kobject, cpu) == NULL))
867 goto err_out;
868
869 per_cpu(index_kobject, cpu) = kzalloc(
870 sizeof(struct _index_kobject ) * num_cache_leaves, GFP_KERNEL);
871 if (unlikely(per_cpu(index_kobject, cpu) == NULL))
872 goto err_out;
873
874 return 0;
875
876 err_out:
877 cpuid4_cache_sysfs_exit(cpu);
878 return -ENOMEM;
879 }
880
881 static cpumask_t cache_dev_map = CPU_MASK_NONE;
882
883 /* Add/Remove cache interface for CPU device */
cache_add_dev(struct sys_device * sys_dev)884 static int __cpuinit cache_add_dev(struct sys_device * sys_dev)
885 {
886 unsigned int cpu = sys_dev->id;
887 unsigned long i, j;
888 struct _index_kobject *this_object;
889 int retval;
890
891 retval = cpuid4_cache_sysfs_init(cpu);
892 if (unlikely(retval < 0))
893 return retval;
894
895 retval = kobject_init_and_add(per_cpu(cache_kobject, cpu),
896 &ktype_percpu_entry,
897 &sys_dev->kobj, "%s", "cache");
898 if (retval < 0) {
899 cpuid4_cache_sysfs_exit(cpu);
900 return retval;
901 }
902
903 for (i = 0; i < num_cache_leaves; i++) {
904 this_object = INDEX_KOBJECT_PTR(cpu,i);
905 this_object->cpu = cpu;
906 this_object->index = i;
907 retval = kobject_init_and_add(&(this_object->kobj),
908 &ktype_cache,
909 per_cpu(cache_kobject, cpu),
910 "index%1lu", i);
911 if (unlikely(retval)) {
912 for (j = 0; j < i; j++) {
913 kobject_put(&(INDEX_KOBJECT_PTR(cpu,j)->kobj));
914 }
915 kobject_put(per_cpu(cache_kobject, cpu));
916 cpuid4_cache_sysfs_exit(cpu);
917 return retval;
918 }
919 kobject_uevent(&(this_object->kobj), KOBJ_ADD);
920 }
921 cpu_set(cpu, cache_dev_map);
922
923 kobject_uevent(per_cpu(cache_kobject, cpu), KOBJ_ADD);
924 return 0;
925 }
926
cache_remove_dev(struct sys_device * sys_dev)927 static void __cpuinit cache_remove_dev(struct sys_device * sys_dev)
928 {
929 unsigned int cpu = sys_dev->id;
930 unsigned long i;
931
932 if (per_cpu(cpuid4_info, cpu) == NULL)
933 return;
934 if (!cpu_isset(cpu, cache_dev_map))
935 return;
936 cpu_clear(cpu, cache_dev_map);
937
938 for (i = 0; i < num_cache_leaves; i++)
939 kobject_put(&(INDEX_KOBJECT_PTR(cpu,i)->kobj));
940 kobject_put(per_cpu(cache_kobject, cpu));
941 cpuid4_cache_sysfs_exit(cpu);
942 }
943
cacheinfo_cpu_callback(struct notifier_block * nfb,unsigned long action,void * hcpu)944 static int __cpuinit cacheinfo_cpu_callback(struct notifier_block *nfb,
945 unsigned long action, void *hcpu)
946 {
947 unsigned int cpu = (unsigned long)hcpu;
948 struct sys_device *sys_dev;
949
950 sys_dev = get_cpu_sysdev(cpu);
951 switch (action) {
952 case CPU_ONLINE:
953 case CPU_ONLINE_FROZEN:
954 cache_add_dev(sys_dev);
955 break;
956 case CPU_DEAD:
957 case CPU_DEAD_FROZEN:
958 cache_remove_dev(sys_dev);
959 break;
960 }
961 return NOTIFY_OK;
962 }
963
964 static struct notifier_block __cpuinitdata cacheinfo_cpu_notifier =
965 {
966 .notifier_call = cacheinfo_cpu_callback,
967 };
968
cache_sysfs_init(void)969 static int __cpuinit cache_sysfs_init(void)
970 {
971 int i;
972
973 if (num_cache_leaves == 0)
974 return 0;
975
976 for_each_online_cpu(i) {
977 int err;
978 struct sys_device *sys_dev = get_cpu_sysdev(i);
979
980 err = cache_add_dev(sys_dev);
981 if (err)
982 return err;
983 }
984 register_hotcpu_notifier(&cacheinfo_cpu_notifier);
985 return 0;
986 }
987
988 device_initcall(cache_sysfs_init);
989
990 #endif
991