?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / arch / x86 / kernel / cpu / intel_cacheinfo.c
blob54d04d5741481164fea1045eb9a96b752ae452c2
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Routines to identify caches on Intel CPU.
4 *
5 * Changes:
6 * Venkatesh Pallipadi : Adding cache identification through cpuid(4)
7 * Ashok Raj <ashok.raj@intel.com>: Work with CPU hotplug infrastructure.
8 * Andi Kleen / Andreas Herrmann : CPUID4 emulation on AMD.
9 */
10
11 #include <linux/slab.h>
12 #include <linux/cacheinfo.h>
13 #include <linux/cpu.h>
14 #include <linux/sched.h>
15 #include <linux/capability.h>
16 #include <linux/sysfs.h>
17 #include <linux/pci.h>
18
19 #include <asm/cpufeature.h>
20 #include <asm/amd_nb.h>
21 #include <asm/smp.h>
22
23 #define LVL_1_INST 1
24 #define LVL_1_DATA 2
25 #define LVL_2 3
26 #define LVL_3 4
27 #define LVL_TRACE 5
28
29 struct _cache_table {
30 unsigned char descriptor;
31 char cache_type;
32 short size;
33 };
34
35 #define MB(x) ((x) * 1024)
36
37 /* All the cache descriptor types we care about (no TLB or
38 trace cache entries) */
39
40 static const struct _cache_table cache_table[] =
41 {
42 { 0x06, LVL_1_INST, 8 }, /* 4-way set assoc, 32 byte line size */
43 { 0x08, LVL_1_INST, 16 }, /* 4-way set assoc, 32 byte line size */
44 { 0x09, LVL_1_INST, 32 }, /* 4-way set assoc, 64 byte line size */
45 { 0x0a, LVL_1_DATA, 8 }, /* 2 way set assoc, 32 byte line size */
46 { 0x0c, LVL_1_DATA, 16 }, /* 4-way set assoc, 32 byte line size */
47 { 0x0d, LVL_1_DATA, 16 }, /* 4-way set assoc, 64 byte line size */
48 { 0x0e, LVL_1_DATA, 24 }, /* 6-way set assoc, 64 byte line size */
49 { 0x21, LVL_2, 256 }, /* 8-way set assoc, 64 byte line size */
50 { 0x22, LVL_3, 512 }, /* 4-way set assoc, sectored cache, 64 byte line size */
51 { 0x23, LVL_3, MB(1) }, /* 8-way set assoc, sectored cache, 64 byte line size */
52 { 0x25, LVL_3, MB(2) }, /* 8-way set assoc, sectored cache, 64 byte line size */
53 { 0x29, LVL_3, MB(4) }, /* 8-way set assoc, sectored cache, 64 byte line size */
54 { 0x2c, LVL_1_DATA, 32 }, /* 8-way set assoc, 64 byte line size */
55 { 0x30, LVL_1_INST, 32 }, /* 8-way set assoc, 64 byte line size */
56 { 0x39, LVL_2, 128 }, /* 4-way set assoc, sectored cache, 64 byte line size */
57 { 0x3a, LVL_2, 192 }, /* 6-way set assoc, sectored cache, 64 byte line size */
58 { 0x3b, LVL_2, 128 }, /* 2-way set assoc, sectored cache, 64 byte line size */
59 { 0x3c, LVL_2, 256 }, /* 4-way set assoc, sectored cache, 64 byte line size */
60 { 0x3d, LVL_2, 384 }, /* 6-way set assoc, sectored cache, 64 byte line size */
61 { 0x3e, LVL_2, 512 }, /* 4-way set assoc, sectored cache, 64 byte line size */
62 { 0x3f, LVL_2, 256 }, /* 2-way set assoc, 64 byte line size */
63 { 0x41, LVL_2, 128 }, /* 4-way set assoc, 32 byte line size */
64 { 0x42, LVL_2, 256 }, /* 4-way set assoc, 32 byte line size */
65 { 0x43, LVL_2, 512 }, /* 4-way set assoc, 32 byte line size */
66 { 0x44, LVL_2, MB(1) }, /* 4-way set assoc, 32 byte line size */
67 { 0x45, LVL_2, MB(2) }, /* 4-way set assoc, 32 byte line size */
68 { 0x46, LVL_3, MB(4) }, /* 4-way set assoc, 64 byte line size */
69 { 0x47, LVL_3, MB(8) }, /* 8-way set assoc, 64 byte line size */
70 { 0x48, LVL_2, MB(3) }, /* 12-way set assoc, 64 byte line size */
71 { 0x49, LVL_3, MB(4) }, /* 16-way set assoc, 64 byte line size */
72 { 0x4a, LVL_3, MB(6) }, /* 12-way set assoc, 64 byte line size */
73 { 0x4b, LVL_3, MB(8) }, /* 16-way set assoc, 64 byte line size */
74 { 0x4c, LVL_3, MB(12) }, /* 12-way set assoc, 64 byte line size */
75 { 0x4d, LVL_3, MB(16) }, /* 16-way set assoc, 64 byte line size */
76 { 0x4e, LVL_2, MB(6) }, /* 24-way set assoc, 64 byte line size */
77 { 0x60, LVL_1_DATA, 16 }, /* 8-way set assoc, sectored cache, 64 byte line size */
78 { 0x66, LVL_1_DATA, 8 }, /* 4-way set assoc, sectored cache, 64 byte line size */
79 { 0x67, LVL_1_DATA, 16 }, /* 4-way set assoc, sectored cache, 64 byte line size */
80 { 0x68, LVL_1_DATA, 32 }, /* 4-way set assoc, sectored cache, 64 byte line size */
81 { 0x70, LVL_TRACE, 12 }, /* 8-way set assoc */
82 { 0x71, LVL_TRACE, 16 }, /* 8-way set assoc */
83 { 0x72, LVL_TRACE, 32 }, /* 8-way set assoc */
84 { 0x73, LVL_TRACE, 64 }, /* 8-way set assoc */
85 { 0x78, LVL_2, MB(1) }, /* 4-way set assoc, 64 byte line size */
86 { 0x79, LVL_2, 128 }, /* 8-way set assoc, sectored cache, 64 byte line size */
87 { 0x7a, LVL_2, 256 }, /* 8-way set assoc, sectored cache, 64 byte line size */
88 { 0x7b, LVL_2, 512 }, /* 8-way set assoc, sectored cache, 64 byte line size */
89 { 0x7c, LVL_2, MB(1) }, /* 8-way set assoc, sectored cache, 64 byte line size */
90 { 0x7d, LVL_2, MB(2) }, /* 8-way set assoc, 64 byte line size */
91 { 0x7f, LVL_2, 512 }, /* 2-way set assoc, 64 byte line size */
92 { 0x80, LVL_2, 512 }, /* 8-way set assoc, 64 byte line size */
93 { 0x82, LVL_2, 256 }, /* 8-way set assoc, 32 byte line size */
94 { 0x83, LVL_2, 512 }, /* 8-way set assoc, 32 byte line size */
95 { 0x84, LVL_2, MB(1) }, /* 8-way set assoc, 32 byte line size */
96 { 0x85, LVL_2, MB(2) }, /* 8-way set assoc, 32 byte line size */
97 { 0x86, LVL_2, 512 }, /* 4-way set assoc, 64 byte line size */
98 { 0x87, LVL_2, MB(1) }, /* 8-way set assoc, 64 byte line size */
99 { 0xd0, LVL_3, 512 }, /* 4-way set assoc, 64 byte line size */
100 { 0xd1, LVL_3, MB(1) }, /* 4-way set assoc, 64 byte line size */
101 { 0xd2, LVL_3, MB(2) }, /* 4-way set assoc, 64 byte line size */
102 { 0xd6, LVL_3, MB(1) }, /* 8-way set assoc, 64 byte line size */
103 { 0xd7, LVL_3, MB(2) }, /* 8-way set assoc, 64 byte line size */
104 { 0xd8, LVL_3, MB(4) }, /* 12-way set assoc, 64 byte line size */
105 { 0xdc, LVL_3, MB(2) }, /* 12-way set assoc, 64 byte line size */
106 { 0xdd, LVL_3, MB(4) }, /* 12-way set assoc, 64 byte line size */
107 { 0xde, LVL_3, MB(8) }, /* 12-way set assoc, 64 byte line size */
108 { 0xe2, LVL_3, MB(2) }, /* 16-way set assoc, 64 byte line size */
109 { 0xe3, LVL_3, MB(4) }, /* 16-way set assoc, 64 byte line size */
110 { 0xe4, LVL_3, MB(8) }, /* 16-way set assoc, 64 byte line size */
111 { 0xea, LVL_3, MB(12) }, /* 24-way set assoc, 64 byte line size */
112 { 0xeb, LVL_3, MB(18) }, /* 24-way set assoc, 64 byte line size */
113 { 0xec, LVL_3, MB(24) }, /* 24-way set assoc, 64 byte line size */
114 { 0x00, 0, 0}
115 };
116
117
118 enum _cache_type {
119 CTYPE_NULL = 0,
120 CTYPE_DATA = 1,
121 CTYPE_INST = 2,
122 CTYPE_UNIFIED = 3
123 };
124
125 union _cpuid4_leaf_eax {
126 struct {
127 enum _cache_type type:5;
128 unsigned int level:3;
129 unsigned int is_self_initializing:1;
130 unsigned int is_fully_associative:1;
131 unsigned int reserved:4;
132 unsigned int num_threads_sharing:12;
133 unsigned int num_cores_on_die:6;
134 } split;
135 u32 full;
136 };
137
138 union _cpuid4_leaf_ebx {
139 struct {
140 unsigned int coherency_line_size:12;
141 unsigned int physical_line_partition:10;
142 unsigned int ways_of_associativity:10;
143 } split;
144 u32 full;
145 };
146
147 union _cpuid4_leaf_ecx {
148 struct {
149 unsigned int number_of_sets:32;
150 } split;
151 u32 full;
152 };
153
154 struct _cpuid4_info_regs {
155 union _cpuid4_leaf_eax eax;
156 union _cpuid4_leaf_ebx ebx;
157 union _cpuid4_leaf_ecx ecx;
158 unsigned int id;
159 unsigned long size;
160 struct amd_northbridge *nb;
161 };
162
163 static unsigned short num_cache_leaves;
164
165 /* AMD doesn't have CPUID4. Emulate it here to report the same
166 information to the user. This makes some assumptions about the machine:
167 L2 not shared, no SMT etc. that is currently true on AMD CPUs.
168
169 In theory the TLBs could be reported as fake type (they are in "dummy").
170 Maybe later */
171 union l1_cache {
172 struct {
173 unsigned line_size:8;
174 unsigned lines_per_tag:8;
175 unsigned assoc:8;
176 unsigned size_in_kb:8;
177 };
178 unsigned val;
179 };
180
181 union l2_cache {
182 struct {
183 unsigned line_size:8;
184 unsigned lines_per_tag:4;
185 unsigned assoc:4;
186 unsigned size_in_kb:16;
187 };
188 unsigned val;
189 };
190
191 union l3_cache {
192 struct {
193 unsigned line_size:8;
194 unsigned lines_per_tag:4;
195 unsigned assoc:4;
196 unsigned res:2;
197 unsigned size_encoded:14;
198 };
199 unsigned val;
200 };
201
202 static const unsigned short assocs[] = {
203 [1] = 1,
204 [2] = 2,
205 [4] = 4,
206 [6] = 8,
207 [8] = 16,
208 [0xa] = 32,
209 [0xb] = 48,
210 [0xc] = 64,
211 [0xd] = 96,
212 [0xe] = 128,
213 [0xf] = 0xffff /* fully associative - no way to show this currently */
214 };
215
216 static const unsigned char levels[] = { 1, 1, 2, 3 };
217 static const unsigned char types[] = { 1, 2, 3, 3 };
218
219 static const enum cache_type cache_type_map[] = {
220 [CTYPE_NULL] = CACHE_TYPE_NOCACHE,
221 [CTYPE_DATA] = CACHE_TYPE_DATA,
222 [CTYPE_INST] = CACHE_TYPE_INST,
223 [CTYPE_UNIFIED] = CACHE_TYPE_UNIFIED,
224 };
225
226 static void
227 amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax,
228 union _cpuid4_leaf_ebx *ebx,
229 union _cpuid4_leaf_ecx *ecx)
230 {
231 unsigned dummy;
232 unsigned line_size, lines_per_tag, assoc, size_in_kb;
233 union l1_cache l1i, l1d;
234 union l2_cache l2;
235 union l3_cache l3;
236 union l1_cache *l1 = &l1d;
237
238 eax->full = 0;
239 ebx->full = 0;
240 ecx->full = 0;
241
242 cpuid(0x80000005, &dummy, &dummy, &l1d.val, &l1i.val);
243 cpuid(0x80000006, &dummy, &dummy, &l2.val, &l3.val);
244
245 switch (leaf) {
246 case 1:
247 l1 = &l1i;
248 case 0:
249 if (!l1->val)
250 return;
251 assoc = assocs[l1->assoc];
252 line_size = l1->line_size;
253 lines_per_tag = l1->lines_per_tag;
254 size_in_kb = l1->size_in_kb;
255 break;
256 case 2:
257 if (!l2.val)
258 return;
259 assoc = assocs[l2.assoc];
260 line_size = l2.line_size;
261 lines_per_tag = l2.lines_per_tag;
262 /* cpu_data has errata corrections for K7 applied */
263 size_in_kb = __this_cpu_read(cpu_info.x86_cache_size);
264 break;
265 case 3:
266 if (!l3.val)
267 return;
268 assoc = assocs[l3.assoc];
269 line_size = l3.line_size;
270 lines_per_tag = l3.lines_per_tag;
271 size_in_kb = l3.size_encoded * 512;
272 if (boot_cpu_has(X86_FEATURE_AMD_DCM)) {
273 size_in_kb = size_in_kb >> 1;
274 assoc = assoc >> 1;
275 }
276 break;
277 default:
278 return;
279 }
280
281 eax->split.is_self_initializing = 1;
282 eax->split.type = types[leaf];
283 eax->split.level = levels[leaf];
284 eax->split.num_threads_sharing = 0;
285 eax->split.num_cores_on_die = __this_cpu_read(cpu_info.x86_max_cores) - 1;
286
287
288 if (assoc == 0xffff)
289 eax->split.is_fully_associative = 1;
290 ebx->split.coherency_line_size = line_size - 1;
291 ebx->split.ways_of_associativity = assoc - 1;
292 ebx->split.physical_line_partition = lines_per_tag - 1;
293 ecx->split.number_of_sets = (size_in_kb * 1024) / line_size /
294 (ebx->split.ways_of_associativity + 1) - 1;
295 }
296
297 #if defined(CONFIG_AMD_NB) && defined(CONFIG_SYSFS)
298
299 /*
300 * L3 cache descriptors
301 */
302 static void amd_calc_l3_indices(struct amd_northbridge *nb)
303 {
304 struct amd_l3_cache *l3 = &nb->l3_cache;
305 unsigned int sc0, sc1, sc2, sc3;
306 u32 val = 0;
307
308 pci_read_config_dword(nb->misc, 0x1C4, &val);
309
310 /* calculate subcache sizes */
311 l3->subcaches[0] = sc0 = !(val & BIT(0));
312 l3->subcaches[1] = sc1 = !(val & BIT(4));
313
314 if (boot_cpu_data.x86 == 0x15) {
315 l3->subcaches[0] = sc0 += !(val & BIT(1));
316 l3->subcaches[1] = sc1 += !(val & BIT(5));
317 }
318
319 l3->subcaches[2] = sc2 = !(val & BIT(8)) + !(val & BIT(9));
320 l3->subcaches[3] = sc3 = !(val & BIT(12)) + !(val & BIT(13));
321
322 l3->indices = (max(max3(sc0, sc1, sc2), sc3) << 10) - 1;
323 }
324
325 /*
326 * check whether a slot used for disabling an L3 index is occupied.
327 * @l3: L3 cache descriptor
328 * @slot: slot number (0..1)
329 *
330 * @returns: the disabled index if used or negative value if slot free.
331 */
332 static int amd_get_l3_disable_slot(struct amd_northbridge *nb, unsigned slot)
333 {
334 unsigned int reg = 0;
335
336 pci_read_config_dword(nb->misc, 0x1BC + slot * 4, &reg);
337
338 /* check whether this slot is activated already */
339 if (reg & (3UL << 30))
340 return reg & 0xfff;
341
342 return -1;
343 }
344
345 static ssize_t show_cache_disable(struct cacheinfo *this_leaf, char *buf,
346 unsigned int slot)
347 {
348 int index;
349 struct amd_northbridge *nb = this_leaf->priv;
350
351 index = amd_get_l3_disable_slot(nb, slot);
352 if (index >= 0)
353 return sprintf(buf, "%d\n", index);
354
355 return sprintf(buf, "FREE\n");
356 }
357
358 #define SHOW_CACHE_DISABLE(slot) \
359 static ssize_t \
360 cache_disable_##slot##_show(struct device *dev, \
361 struct device_attribute *attr, char *buf) \
362 { \
363 struct cacheinfo *this_leaf = dev_get_drvdata(dev); \
364 return show_cache_disable(this_leaf, buf, slot); \
365 }
366 SHOW_CACHE_DISABLE(0)
367 SHOW_CACHE_DISABLE(1)
368
369 static void amd_l3_disable_index(struct amd_northbridge *nb, int cpu,
370 unsigned slot, unsigned long idx)
371 {
372 int i;
373
374 idx |= BIT(30);
375
376 /*
377 * disable index in all 4 subcaches
378 */
379 for (i = 0; i < 4; i++) {
380 u32 reg = idx | (i << 20);
381
382 if (!nb->l3_cache.subcaches[i])
383 continue;
384
385 pci_write_config_dword(nb->misc, 0x1BC + slot * 4, reg);
386
387 /*
388 * We need to WBINVD on a core on the node containing the L3
389 * cache which indices we disable therefore a simple wbinvd()
390 * is not sufficient.
391 */
392 wbinvd_on_cpu(cpu);
393
394 reg |= BIT(31);
395 pci_write_config_dword(nb->misc, 0x1BC + slot * 4, reg);
396 }
397 }
398
399 /*
400 * disable a L3 cache index by using a disable-slot
401 *
402 * @l3: L3 cache descriptor
403 * @cpu: A CPU on the node containing the L3 cache
404 * @slot: slot number (0..1)
405 * @index: index to disable
406 *
407 * @return: 0 on success, error status on failure
408 */
409 static int amd_set_l3_disable_slot(struct amd_northbridge *nb, int cpu,
410 unsigned slot, unsigned long index)
411 {
412 int ret = 0;
413
414 /* check if @slot is already used or the index is already disabled */
415 ret = amd_get_l3_disable_slot(nb, slot);
416 if (ret >= 0)
417 return -EEXIST;
418
419 if (index > nb->l3_cache.indices)
420 return -EINVAL;
421
422 /* check whether the other slot has disabled the same index already */
423 if (index == amd_get_l3_disable_slot(nb, !slot))
424 return -EEXIST;
425
426 amd_l3_disable_index(nb, cpu, slot, index);
427
428 return 0;
429 }
430
431 static ssize_t store_cache_disable(struct cacheinfo *this_leaf,
432 const char *buf, size_t count,
433 unsigned int slot)
434 {
435 unsigned long val = 0;
436 int cpu, err = 0;
437 struct amd_northbridge *nb = this_leaf->priv;
438
439 if (!capable(CAP_SYS_ADMIN))
440 return -EPERM;
441
442 cpu = cpumask_first(&this_leaf->shared_cpu_map);
443
444 if (kstrtoul(buf, 10, &val) < 0)
445 return -EINVAL;
446
447 err = amd_set_l3_disable_slot(nb, cpu, slot, val);
448 if (err) {
449 if (err == -EEXIST)
450 pr_warn("L3 slot %d in use/index already disabled!\n",
451 slot);
452 return err;
453 }
454 return count;
455 }
456
457 #define STORE_CACHE_DISABLE(slot) \
458 static ssize_t \
459 cache_disable_##slot##_store(struct device *dev, \
460 struct device_attribute *attr, \
461 const char *buf, size_t count) \
462 { \
463 struct cacheinfo *this_leaf = dev_get_drvdata(dev); \
464 return store_cache_disable(this_leaf, buf, count, slot); \
465 }
466 STORE_CACHE_DISABLE(0)
467 STORE_CACHE_DISABLE(1)
468
469 static ssize_t subcaches_show(struct device *dev,
470 struct device_attribute *attr, char *buf)
471 {
472 struct cacheinfo *this_leaf = dev_get_drvdata(dev);
473 int cpu = cpumask_first(&this_leaf->shared_cpu_map);
474
475 return sprintf(buf, "%x\n", amd_get_subcaches(cpu));
476 }
477
478 static ssize_t subcaches_store(struct device *dev,
479 struct device_attribute *attr,
480 const char *buf, size_t count)
481 {
482 struct cacheinfo *this_leaf = dev_get_drvdata(dev);
483 int cpu = cpumask_first(&this_leaf->shared_cpu_map);
484 unsigned long val;
485
486 if (!capable(CAP_SYS_ADMIN))
487 return -EPERM;
488
489 if (kstrtoul(buf, 16, &val) < 0)
490 return -EINVAL;
491
492 if (amd_set_subcaches(cpu, val))
493 return -EINVAL;
494
495 return count;
496 }
497
498 static DEVICE_ATTR_RW(cache_disable_0);
499 static DEVICE_ATTR_RW(cache_disable_1);
500 static DEVICE_ATTR_RW(subcaches);
501
502 static umode_t
503 cache_private_attrs_is_visible(struct kobject *kobj,
504 struct attribute *attr, int unused)
505 {
506 struct device *dev = kobj_to_dev(kobj);
507 struct cacheinfo *this_leaf = dev_get_drvdata(dev);
508 umode_t mode = attr->mode;
509
510 if (!this_leaf->priv)
511 return 0;
512
513 if ((attr == &dev_attr_subcaches.attr) &&
514 amd_nb_has_feature(AMD_NB_L3_PARTITIONING))
515 return mode;
516
517 if ((attr == &dev_attr_cache_disable_0.attr ||
518 attr == &dev_attr_cache_disable_1.attr) &&
519 amd_nb_has_feature(AMD_NB_L3_INDEX_DISABLE))
520 return mode;
521
522 return 0;
523 }
524
525 static struct attribute_group cache_private_group = {
526 .is_visible = cache_private_attrs_is_visible,
527 };
528
529 static void init_amd_l3_attrs(void)
530 {
531 int n = 1;
532 static struct attribute **amd_l3_attrs;
533
534 if (amd_l3_attrs) /* already initialized */
535 return;
536
537 if (amd_nb_has_feature(AMD_NB_L3_INDEX_DISABLE))
538 n += 2;
539 if (amd_nb_has_feature(AMD_NB_L3_PARTITIONING))
540 n += 1;
541
542 amd_l3_attrs = kcalloc(n, sizeof(*amd_l3_attrs), GFP_KERNEL);
543 if (!amd_l3_attrs)
544 return;
545
546 n = 0;
547 if (amd_nb_has_feature(AMD_NB_L3_INDEX_DISABLE)) {
548 amd_l3_attrs[n++] = &dev_attr_cache_disable_0.attr;
549 amd_l3_attrs[n++] = &dev_attr_cache_disable_1.attr;
550 }
551 if (amd_nb_has_feature(AMD_NB_L3_PARTITIONING))
552 amd_l3_attrs[n++] = &dev_attr_subcaches.attr;
553
554 cache_private_group.attrs = amd_l3_attrs;
555 }
556
557 const struct attribute_group *
558 cache_get_priv_group(struct cacheinfo *this_leaf)
559 {
560 struct amd_northbridge *nb = this_leaf->priv;
561
562 if (this_leaf->level < 3 || !nb)
563 return NULL;
564
565 if (nb && nb->l3_cache.indices)
566 init_amd_l3_attrs();
567
568 return &cache_private_group;
569 }
570
571 static void amd_init_l3_cache(struct _cpuid4_info_regs *this_leaf, int index)
572 {
573 int node;
574
575 /* only for L3, and not in virtualized environments */
576 if (index < 3)
577 return;
578
579 node = amd_get_nb_id(smp_processor_id());
580 this_leaf->nb = node_to_amd_nb(node);
581 if (this_leaf->nb && !this_leaf->nb->l3_cache.indices)
582 amd_calc_l3_indices(this_leaf->nb);
583 }
584 #else
585 #define amd_init_l3_cache(x, y)
586 #endif /* CONFIG_AMD_NB && CONFIG_SYSFS */
587
588 static int
589 cpuid4_cache_lookup_regs(int index, struct _cpuid4_info_regs *this_leaf)
590 {
591 union _cpuid4_leaf_eax eax;
592 union _cpuid4_leaf_ebx ebx;
593 union _cpuid4_leaf_ecx ecx;
594 unsigned edx;
595
596 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
597 if (boot_cpu_has(X86_FEATURE_TOPOEXT))
598 cpuid_count(0x8000001d, index, &eax.full,
599 &ebx.full, &ecx.full, &edx);
600 else
601 amd_cpuid4(index, &eax, &ebx, &ecx);
602 amd_init_l3_cache(this_leaf, index);
603 } else {
604 cpuid_count(4, index, &eax.full, &ebx.full, &ecx.full, &edx);
605 }
606
607 if (eax.split.type == CTYPE_NULL)
608 return -EIO; /* better error ? */
609
610 this_leaf->eax = eax;
611 this_leaf->ebx = ebx;
612 this_leaf->ecx = ecx;
613 this_leaf->size = (ecx.split.number_of_sets + 1) *
614 (ebx.split.coherency_line_size + 1) *
615 (ebx.split.physical_line_partition + 1) *
616 (ebx.split.ways_of_associativity + 1);
617 return 0;
618 }
619
620 static int find_num_cache_leaves(struct cpuinfo_x86 *c)
621 {
622 unsigned int eax, ebx, ecx, edx, op;
623 union _cpuid4_leaf_eax cache_eax;
624 int i = -1;
625
626 if (c->x86_vendor == X86_VENDOR_AMD)
627 op = 0x8000001d;
628 else
629 op = 4;
630
631 do {
632 ++i;
633 /* Do cpuid(op) loop to find out num_cache_leaves */
634 cpuid_count(op, i, &eax, &ebx, &ecx, &edx);
635 cache_eax.full = eax;
636 } while (cache_eax.split.type != CTYPE_NULL);
637 return i;
638 }
639
640 void init_amd_cacheinfo(struct cpuinfo_x86 *c)
641 {
642
643 if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
644 num_cache_leaves = find_num_cache_leaves(c);
645 } else if (c->extended_cpuid_level >= 0x80000006) {
646 if (cpuid_edx(0x80000006) & 0xf000)
647 num_cache_leaves = 4;
648 else
649 num_cache_leaves = 3;
650 }
651 }
652
653 unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c)
654 {
655 /* Cache sizes */
656 unsigned int trace = 0, l1i = 0, l1d = 0, l2 = 0, l3 = 0;
657 unsigned int new_l1d = 0, new_l1i = 0; /* Cache sizes from cpuid(4) */
658 unsigned int new_l2 = 0, new_l3 = 0, i; /* Cache sizes from cpuid(4) */
659 unsigned int l2_id = 0, l3_id = 0, num_threads_sharing, index_msb;
660 #ifdef CONFIG_SMP
661 unsigned int cpu = c->cpu_index;
662 #endif
663
664 if (c->cpuid_level > 3) {
665 static int is_initialized;
666
667 if (is_initialized == 0) {
668 /* Init num_cache_leaves from boot CPU */
669 num_cache_leaves = find_num_cache_leaves(c);
670 is_initialized++;
671 }
672
673 /*
674 * Whenever possible use cpuid(4), deterministic cache
675 * parameters cpuid leaf to find the cache details
676 */
677 for (i = 0; i < num_cache_leaves; i++) {
678 struct _cpuid4_info_regs this_leaf = {};
679 int retval;
680
681 retval = cpuid4_cache_lookup_regs(i, &this_leaf);
682 if (retval < 0)
683 continue;
684
685 switch (this_leaf.eax.split.level) {
686 case 1:
687 if (this_leaf.eax.split.type == CTYPE_DATA)
688 new_l1d = this_leaf.size/1024;
689 else if (this_leaf.eax.split.type == CTYPE_INST)
690 new_l1i = this_leaf.size/1024;
691 break;
692 case 2:
693 new_l2 = this_leaf.size/1024;
694 num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
695 index_msb = get_count_order(num_threads_sharing);
696 l2_id = c->apicid & ~((1 << index_msb) - 1);
697 break;
698 case 3:
699 new_l3 = this_leaf.size/1024;
700 num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
701 index_msb = get_count_order(num_threads_sharing);
702 l3_id = c->apicid & ~((1 << index_msb) - 1);
703 break;
704 default:
705 break;
706 }
707 }
708 }
709 /*
710 * Don't use cpuid2 if cpuid4 is supported. For P4, we use cpuid2 for
711 * trace cache
712 */
713 if ((num_cache_leaves == 0 || c->x86 == 15) && c->cpuid_level > 1) {
714 /* supports eax=2 call */
715 int j, n;
716 unsigned int regs[4];
717 unsigned char *dp = (unsigned char *)regs;
718 int only_trace = 0;
719
720 if (num_cache_leaves != 0 && c->x86 == 15)
721 only_trace = 1;
722
723 /* Number of times to iterate */
724 n = cpuid_eax(2) & 0xFF;
725
726 for (i = 0 ; i < n ; i++) {
727 cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
728
729 /* If bit 31 is set, this is an unknown format */
730 for (j = 0 ; j < 3 ; j++)
731 if (regs[j] & (1 << 31))
732 regs[j] = 0;
733
734 /* Byte 0 is level count, not a descriptor */
735 for (j = 1 ; j < 16 ; j++) {
736 unsigned char des = dp[j];
737 unsigned char k = 0;
738
739 /* look up this descriptor in the table */
740 while (cache_table[k].descriptor != 0) {
741 if (cache_table[k].descriptor == des) {
742 if (only_trace && cache_table[k].cache_type != LVL_TRACE)
743 break;
744 switch (cache_table[k].cache_type) {
745 case LVL_1_INST:
746 l1i += cache_table[k].size;
747 break;
748 case LVL_1_DATA:
749 l1d += cache_table[k].size;
750 break;
751 case LVL_2:
752 l2 += cache_table[k].size;
753 break;
754 case LVL_3:
755 l3 += cache_table[k].size;
756 break;
757 case LVL_TRACE:
758 trace += cache_table[k].size;
759 break;
760 }
761
762 break;
763 }
764
765 k++;
766 }
767 }
768 }
769 }
770
771 if (new_l1d)
772 l1d = new_l1d;
773
774 if (new_l1i)
775 l1i = new_l1i;
776
777 if (new_l2) {
778 l2 = new_l2;
779 #ifdef CONFIG_SMP
780 per_cpu(cpu_llc_id, cpu) = l2_id;
781 #endif
782 }
783
784 if (new_l3) {
785 l3 = new_l3;
786 #ifdef CONFIG_SMP
787 per_cpu(cpu_llc_id, cpu) = l3_id;
788 #endif
789 }
790
791 #ifdef CONFIG_SMP
792 /*
793 * If cpu_llc_id is not yet set, this means cpuid_level < 4 which in
794 * turns means that the only possibility is SMT (as indicated in
795 * cpuid1). Since cpuid2 doesn't specify shared caches, and we know
796 * that SMT shares all caches, we can unconditionally set cpu_llc_id to
797 * c->phys_proc_id.
798 */
799 if (per_cpu(cpu_llc_id, cpu) == BAD_APICID)
800 per_cpu(cpu_llc_id, cpu) = c->phys_proc_id;
801 #endif
802
803 c->x86_cache_size = l3 ? l3 : (l2 ? l2 : (l1i+l1d));
804
805 return l2;
806 }
807
808 static int __cache_amd_cpumap_setup(unsigned int cpu, int index,
809 struct _cpuid4_info_regs *base)
810 {
811 struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
812 struct cacheinfo *this_leaf;
813 int i, sibling;
814
815 /*
816 * For L3, always use the pre-calculated cpu_llc_shared_mask
817 * to derive shared_cpu_map.
818 */
819 if (index == 3) {
820 for_each_cpu(i, cpu_llc_shared_mask(cpu)) {
821 this_cpu_ci = get_cpu_cacheinfo(i);
822 if (!this_cpu_ci->info_list)
823 continue;
824 this_leaf = this_cpu_ci->info_list + index;
825 for_each_cpu(sibling, cpu_llc_shared_mask(cpu)) {
826 if (!cpu_online(sibling))
827 continue;
828 cpumask_set_cpu(sibling,
829 &this_leaf->shared_cpu_map);
830 }
831 }
832 } else if (boot_cpu_has(X86_FEATURE_TOPOEXT)) {
833 unsigned int apicid, nshared, first, last;
834
835 nshared = base->eax.split.num_threads_sharing + 1;
836 apicid = cpu_data(cpu).apicid;
837 first = apicid - (apicid % nshared);
838 last = first + nshared - 1;
839
840 for_each_online_cpu(i) {
841 this_cpu_ci = get_cpu_cacheinfo(i);
842 if (!this_cpu_ci->info_list)
843 continue;
844
845 apicid = cpu_data(i).apicid;
846 if ((apicid < first) || (apicid > last))
847 continue;
848
849 this_leaf = this_cpu_ci->info_list + index;
850
851 for_each_online_cpu(sibling) {
852 apicid = cpu_data(sibling).apicid;
853 if ((apicid < first) || (apicid > last))
854 continue;
855 cpumask_set_cpu(sibling,
856 &this_leaf->shared_cpu_map);
857 }
858 }
859 } else
860 return 0;
861
862 return 1;
863 }
864
865 static void __cache_cpumap_setup(unsigned int cpu, int index,
866 struct _cpuid4_info_regs *base)
867 {
868 struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
869 struct cacheinfo *this_leaf, *sibling_leaf;
870 unsigned long num_threads_sharing;
871 int index_msb, i;
872 struct cpuinfo_x86 *c = &cpu_data(cpu);
873
874 if (c->x86_vendor == X86_VENDOR_AMD) {
875 if (__cache_amd_cpumap_setup(cpu, index, base))
876 return;
877 }
878
879 this_leaf = this_cpu_ci->info_list + index;
880 num_threads_sharing = 1 + base->eax.split.num_threads_sharing;
881
882 cpumask_set_cpu(cpu, &this_leaf->shared_cpu_map);
883 if (num_threads_sharing == 1)
884 return;
885
886 index_msb = get_count_order(num_threads_sharing);
887
888 for_each_online_cpu(i)
889 if (cpu_data(i).apicid >> index_msb == c->apicid >> index_msb) {
890 struct cpu_cacheinfo *sib_cpu_ci = get_cpu_cacheinfo(i);
891
892 if (i == cpu || !sib_cpu_ci->info_list)
893 continue;/* skip if itself or no cacheinfo */
894 sibling_leaf = sib_cpu_ci->info_list + index;
895 cpumask_set_cpu(i, &this_leaf->shared_cpu_map);
896 cpumask_set_cpu(cpu, &sibling_leaf->shared_cpu_map);
897 }
898 }
899
900 static void ci_leaf_init(struct cacheinfo *this_leaf,
901 struct _cpuid4_info_regs *base)
902 {
903 this_leaf->id = base->id;
904 this_leaf->attributes = CACHE_ID;
905 this_leaf->level = base->eax.split.level;
906 this_leaf->type = cache_type_map[base->eax.split.type];
907 this_leaf->coherency_line_size =
908 base->ebx.split.coherency_line_size + 1;
909 this_leaf->ways_of_associativity =
910 base->ebx.split.ways_of_associativity + 1;
911 this_leaf->size = base->size;
912 this_leaf->number_of_sets = base->ecx.split.number_of_sets + 1;
913 this_leaf->physical_line_partition =
914 base->ebx.split.physical_line_partition + 1;
915 this_leaf->priv = base->nb;
916 }
917
918 static int __init_cache_level(unsigned int cpu)
919 {
920 struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
921
922 if (!num_cache_leaves)
923 return -ENOENT;
924 if (!this_cpu_ci)
925 return -EINVAL;
926 this_cpu_ci->num_levels = 3;
927 this_cpu_ci->num_leaves = num_cache_leaves;
928 return 0;
929 }
930
931 /*
932 * The max shared threads number comes from CPUID.4:EAX[25-14] with input
933 * ECX as cache index. Then right shift apicid by the number's order to get
934 * cache id for this cache node.
935 */
936 static void get_cache_id(int cpu, struct _cpuid4_info_regs *id4_regs)
937 {
938 struct cpuinfo_x86 *c = &cpu_data(cpu);
939 unsigned long num_threads_sharing;
940 int index_msb;
941
942 num_threads_sharing = 1 + id4_regs->eax.split.num_threads_sharing;
943 index_msb = get_count_order(num_threads_sharing);
944 id4_regs->id = c->apicid >> index_msb;
945 }
946
947 static int __populate_cache_leaves(unsigned int cpu)
948 {
949 unsigned int idx, ret;
950 struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
951 struct cacheinfo *this_leaf = this_cpu_ci->info_list;
952 struct _cpuid4_info_regs id4_regs = {};
953
954 for (idx = 0; idx < this_cpu_ci->num_leaves; idx++) {
955 ret = cpuid4_cache_lookup_regs(idx, &id4_regs);
956 if (ret)
957 return ret;
958 get_cache_id(cpu, &id4_regs);
959 ci_leaf_init(this_leaf++, &id4_regs);
960 __cache_cpumap_setup(cpu, idx, &id4_regs);
961 }
962 this_cpu_ci->cpu_map_populated = true;
963
964 return 0;
965 }
966
967 DEFINE_SMP_CALL_CACHE_FUNCTION(init_cache_level)
968 DEFINE_SMP_CALL_CACHE_FUNCTION(populate_cache_leaves)
CRIS linux kernel tree