| 1 | |
| 2 | |
| 3 | * cacheinfo support - processor cache information via sysfs
|
| 4 | *
|
| 5 | * Based on arch/x86/kernel/cpu/intel_cacheinfo.c
|
| 6 | * Author: Sudeep Holla <[email protected]>
|
| 7 |
|
| 8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
|
| 9 |
|
| 10 | #include <linux/acpi.h>
|
| 11 | #include <linux/bitops.h>
|
| 12 | #include <linux/cacheinfo.h>
|
| 13 | #include <linux/compiler.h>
|
| 14 | #include <linux/cpu.h>
|
| 15 | #include <linux/device.h>
|
| 16 | #include <linux/init.h>
|
| 17 | #include <linux/of.h>
|
| 18 | #include <linux/sched.h>
|
| 19 | #include <linux/slab.h>
|
| 20 | #include <linux/smp.h>
|
| 21 | #include <linux/sysfs.h>
|
| 22 |
|
| 23 |
|
| 24 | static DEFINE_PER_CPU(struct cpu_cacheinfo, ci_cpu_cacheinfo);
|
| 25 | #define ci_cacheinfo(cpu) (&per_cpu(ci_cpu_cacheinfo, cpu))
|
| 26 | #define cache_leaves(cpu) (ci_cacheinfo(cpu)->num_leaves)
|
| 27 | #define per_cpu_cacheinfo(cpu) (ci_cacheinfo(cpu)->info_list)
|
| 28 | #define per_cpu_cacheinfo_idx(cpu, idx) \
|
| 29 | (per_cpu_cacheinfo(cpu) + (idx))
|
| 30 |
|
| 31 |
|
| 32 | static bool use_arch_info;
|
| 33 |
|
| 34 | struct cpu_cacheinfo *get_cpu_cacheinfo(unsigned int cpu)
|
| 35 | {
|
| 36 | return ci_cacheinfo(cpu);
|
| 37 | }
|
| 38 |
|
| 39 | static inline bool cache_leaves_are_shared(struct cacheinfo *this_leaf,
|
| 40 | struct cacheinfo *sib_leaf)
|
| 41 | {
|
| 42 | |
| 43 | * For non DT/ACPI systems, assume unique level 1 caches,
|
| 44 | * system-wide shared caches for all other levels.
|
| 45 |
|
| 46 | if (!(IS_ENABLED(CONFIG_OF) || IS_ENABLED(CONFIG_ACPI)) ||
|
| 47 | use_arch_info)
|
| 48 | return (this_leaf->level != 1) && (sib_leaf->level != 1);
|
| 49 |
|
| 50 | if ((sib_leaf->attributes & CACHE_ID) &&
|
| 51 | (this_leaf->attributes & CACHE_ID))
|
| 52 | return sib_leaf->id == this_leaf->id;
|
| 53 |
|
| 54 | return sib_leaf->fw_token == this_leaf->fw_token;
|
| 55 | }
|
| 56 |
|
| 57 | bool last_level_cache_is_valid(unsigned int cpu)
|
| 58 | {
|
| 59 | struct cacheinfo *llc;
|
| 60 |
|
| 61 | if (!cache_leaves(cpu) || !per_cpu_cacheinfo(cpu))
|
| 62 | return false;
|
| 63 |
|
| 64 | llc = per_cpu_cacheinfo_idx(cpu, cache_leaves(cpu) - 1);
|
| 65 |
|
| 66 | return (llc->attributes & CACHE_ID) || !!llc->fw_token;
|
| 67 |
|
| 68 | }
|
| 69 |
|
| 70 | bool last_level_cache_is_shared(unsigned int cpu_x, unsigned int cpu_y)
|
| 71 | {
|
| 72 | struct cacheinfo *llc_x, *llc_y;
|
| 73 |
|
| 74 | if (!last_level_cache_is_valid(cpu_x) ||
|
| 75 | !last_level_cache_is_valid(cpu_y))
|
| 76 | return false;
|
| 77 |
|
| 78 | llc_x = per_cpu_cacheinfo_idx(cpu_x, cache_leaves(cpu_x) - 1);
|
| 79 | llc_y = per_cpu_cacheinfo_idx(cpu_y, cache_leaves(cpu_y) - 1);
|
| 80 |
|
| 81 | return cache_leaves_are_shared(llc_x, llc_y);
|
| 82 | }
|
| 83 |
|
| 84 | #ifdef CONFIG_OF
|
| 85 |
|
| 86 | static bool of_check_cache_nodes(struct device_node *np);
|
| 87 |
|
| 88 |
|
| 89 | struct cache_type_info {
|
| 90 | const char *size_prop;
|
| 91 | const char *line_size_props[2];
|
| 92 | const char *nr_sets_prop;
|
| 93 | };
|
| 94 |
|
| 95 | static const struct cache_type_info cache_type_info[] = {
|
| 96 | {
|
| 97 | .size_prop = "cache-size",
|
| 98 | .line_size_props = { "cache-line-size",
|
| 99 | "cache-block-size", },
|
| 100 | .nr_sets_prop = "cache-sets",
|
| 101 | }, {
|
| 102 | .size_prop = "i-cache-size",
|
| 103 | .line_size_props = { "i-cache-line-size",
|
| 104 | "i-cache-block-size", },
|
| 105 | .nr_sets_prop = "i-cache-sets",
|
| 106 | }, {
|
| 107 | .size_prop = "d-cache-size",
|
| 108 | .line_size_props = { "d-cache-line-size",
|
| 109 | "d-cache-block-size", },
|
| 110 | .nr_sets_prop = "d-cache-sets",
|
| 111 | },
|
| 112 | };
|
| 113 |
|
| 114 | static inline int get_cacheinfo_idx(enum cache_type type)
|
| 115 | {
|
| 116 | if (type == CACHE_TYPE_UNIFIED)
|
| 117 | return 0;
|
| 118 | return type;
|
| 119 | }
|
| 120 |
|
| 121 | static void cache_size(struct cacheinfo *this_leaf, struct device_node *np)
|
| 122 | {
|
| 123 | const char *propname;
|
| 124 | int ct_idx;
|
| 125 |
|
| 126 | ct_idx = get_cacheinfo_idx(this_leaf->type);
|
| 127 | propname = cache_type_info[ct_idx].size_prop;
|
| 128 |
|
| 129 | of_property_read_u32(np, propname, &this_leaf->size);
|
| 130 | }
|
| 131 |
|
| 132 |
|
| 133 | static void cache_get_line_size(struct cacheinfo *this_leaf,
|
| 134 | struct device_node *np)
|
| 135 | {
|
| 136 | int i, lim, ct_idx;
|
| 137 |
|
| 138 | ct_idx = get_cacheinfo_idx(this_leaf->type);
|
| 139 | lim = ARRAY_SIZE(cache_type_info[ct_idx].line_size_props);
|
| 140 |
|
| 141 | for (i = 0; i < lim; i++) {
|
| 142 | int ret;
|
| 143 | u32 line_size;
|
| 144 | const char *propname;
|
| 145 |
|
| 146 | propname = cache_type_info[ct_idx].line_size_props[i];
|
| 147 | ret = of_property_read_u32(np, propname, &line_size);
|
| 148 | if (!ret) {
|
| 149 | this_leaf->coherency_line_size = line_size;
|
| 150 | break;
|
| 151 | }
|
| 152 | }
|
| 153 | }
|
| 154 |
|
| 155 | static void cache_nr_sets(struct cacheinfo *this_leaf, struct device_node *np)
|
| 156 | {
|
| 157 | const char *propname;
|
| 158 | int ct_idx;
|
| 159 |
|
| 160 | ct_idx = get_cacheinfo_idx(this_leaf->type);
|
| 161 | propname = cache_type_info[ct_idx].nr_sets_prop;
|
| 162 |
|
| 163 | of_property_read_u32(np, propname, &this_leaf->number_of_sets);
|
| 164 | }
|
| 165 |
|
| 166 | static void cache_associativity(struct cacheinfo *this_leaf)
|
| 167 | {
|
| 168 | unsigned int line_size = this_leaf->coherency_line_size;
|
| 169 | unsigned int nr_sets = this_leaf->number_of_sets;
|
| 170 | unsigned int size = this_leaf->size;
|
| 171 |
|
| 172 | |
| 173 | * If the cache is fully associative, there is no need to
|
| 174 | * check the other properties.
|
| 175 |
|
| 176 | if (!(nr_sets == 1) && (nr_sets > 0 && size > 0 && line_size > 0))
|
| 177 | this_leaf->ways_of_associativity = (size / nr_sets) / line_size;
|
| 178 | }
|
| 179 |
|
| 180 | static bool cache_node_is_unified(struct cacheinfo *this_leaf,
|
| 181 | struct device_node *np)
|
| 182 | {
|
| 183 | return of_property_read_bool(np, "cache-unified");
|
| 184 | }
|
| 185 |
|
| 186 | static void cache_of_set_props(struct cacheinfo *this_leaf,
|
| 187 | struct device_node *np)
|
| 188 | {
|
| 189 | |
| 190 | * init_cache_level must setup the cache level correctly
|
| 191 | * overriding the architecturally specified levels, so
|
| 192 | * if type is NONE at this stage, it should be unified
|
| 193 |
|
| 194 | if (this_leaf->type == CACHE_TYPE_NOCACHE &&
|
| 195 | cache_node_is_unified(this_leaf, np))
|
| 196 | this_leaf->type = CACHE_TYPE_UNIFIED;
|
| 197 | cache_size(this_leaf, np);
|
| 198 | cache_get_line_size(this_leaf, np);
|
| 199 | cache_nr_sets(this_leaf, np);
|
| 200 | cache_associativity(this_leaf);
|
| 201 | }
|
| 202 |
|
| 203 | static int cache_setup_of_node(unsigned int cpu)
|
| 204 | {
|
| 205 | struct cacheinfo *this_leaf;
|
| 206 | unsigned int index = 0;
|
| 207 |
|
| 208 | struct device_node *np __free(device_node) = of_cpu_device_node_get(cpu);
|
| 209 | if (!np) {
|
| 210 | pr_err("Failed to find cpu%d device node\n", cpu);
|
| 211 | return -ENOENT;
|
| 212 | }
|
| 213 |
|
| 214 | if (!of_check_cache_nodes(np)) {
|
| 215 | return -ENOENT;
|
| 216 | }
|
| 217 |
|
| 218 | while (index < cache_leaves(cpu)) {
|
| 219 | this_leaf = per_cpu_cacheinfo_idx(cpu, index);
|
| 220 | if (this_leaf->level != 1) {
|
| 221 | struct device_node *prev __free(device_node) = np;
|
| 222 | np = of_find_next_cache_node(np);
|
| 223 | if (!np)
|
| 224 | break;
|
| 225 | }
|
| 226 | cache_of_set_props(this_leaf, np);
|
| 227 | this_leaf->fw_token = np;
|
| 228 | index++;
|
| 229 | }
|
| 230 |
|
| 231 | if (index != cache_leaves(cpu))
|
| 232 | return -ENOENT;
|
| 233 |
|
| 234 | return 0;
|
| 235 | }
|
| 236 |
|
| 237 | static bool of_check_cache_nodes(struct device_node *np)
|
| 238 | {
|
| 239 | if (of_property_present(np, "cache-size") ||
|
| 240 | of_property_present(np, "i-cache-size") ||
|
| 241 | of_property_present(np, "d-cache-size") ||
|
| 242 | of_property_present(np, "cache-unified"))
|
| 243 | return true;
|
| 244 |
|
| 245 | struct device_node *next __free(device_node) = of_find_next_cache_node(np);
|
| 246 | if (next) {
|
| 247 | return true;
|
| 248 | }
|
| 249 |
|
| 250 | return false;
|
| 251 | }
|
| 252 |
|
| 253 | static int of_count_cache_leaves(struct device_node *np)
|
| 254 | {
|
| 255 | unsigned int leaves = 0;
|
| 256 |
|
| 257 | if (of_property_present(np, "cache-size"))
|
| 258 | ++leaves;
|
| 259 | if (of_property_present(np, "i-cache-size"))
|
| 260 | ++leaves;
|
| 261 | if (of_property_present(np, "d-cache-size"))
|
| 262 | ++leaves;
|
| 263 |
|
| 264 | if (!leaves) {
|
| 265 | |
| 266 | * if absent, fallback on the 'cache-unified' property.
|
| 267 |
|
| 268 | if (of_property_read_bool(np, "cache-unified"))
|
| 269 | return 1;
|
| 270 | else
|
| 271 | return 2;
|
| 272 | }
|
| 273 |
|
| 274 | return leaves;
|
| 275 | }
|
| 276 |
|
| 277 | int init_of_cache_level(unsigned int cpu)
|
| 278 | {
|
| 279 | struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
|
| 280 | struct device_node *np __free(device_node) = of_cpu_device_node_get(cpu);
|
| 281 | unsigned int levels = 0, leaves, level;
|
| 282 |
|
| 283 | if (!of_check_cache_nodes(np)) {
|
| 284 | return -ENOENT;
|
| 285 | }
|
| 286 |
|
| 287 | leaves = of_count_cache_leaves(np);
|
| 288 | if (leaves > 0)
|
| 289 | levels = 1;
|
| 290 |
|
| 291 | while (1) {
|
| 292 | struct device_node *prev __free(device_node) = np;
|
| 293 | np = of_find_next_cache_node(np);
|
| 294 | if (!np)
|
| 295 | break;
|
| 296 |
|
| 297 | if (!of_device_is_compatible(np, "cache"))
|
| 298 | return -EINVAL;
|
| 299 | if (of_property_read_u32(np, "cache-level", &level))
|
| 300 | return -EINVAL;
|
| 301 | if (level <= levels)
|
| 302 | return -EINVAL;
|
| 303 |
|
| 304 | leaves += of_count_cache_leaves(np);
|
| 305 | levels = level;
|
| 306 | }
|
| 307 |
|
| 308 | this_cpu_ci->num_levels = levels;
|
| 309 | this_cpu_ci->num_leaves = leaves;
|
| 310 |
|
| 311 | return 0;
|
| 312 | }
|
| 313 |
|
| 314 | #else
|
| 315 | static inline int cache_setup_of_node(unsigned int cpu) { return 0; }
|
| 316 | int init_of_cache_level(unsigned int cpu) { return 0; }
|
| 317 | #endif
|
| 318 |
|
| 319 | int __weak cache_setup_acpi(unsigned int cpu)
|
| 320 | {
|
| 321 | return -ENOTSUPP;
|
| 322 | }
|
| 323 |
|
| 324 | unsigned int coherency_max_size;
|
| 325 |
|
| 326 | static int cache_setup_properties(unsigned int cpu)
|
| 327 | {
|
| 328 | int ret = 0;
|
| 329 |
|
| 330 | if (of_have_populated_dt())
|
| 331 | ret = cache_setup_of_node(cpu);
|
| 332 | else if (!acpi_disabled)
|
| 333 | ret = cache_setup_acpi(cpu);
|
| 334 |
|
| 335 | |
| 336 | if (ret && use_arch_cache_info())
|
| 337 | use_arch_info = true;
|
| 338 |
|
| 339 | return ret;
|
| 340 | }
|
| 341 |
|
| 342 | static int cache_shared_cpu_map_setup(unsigned int cpu)
|
| 343 | {
|
| 344 | struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
|
| 345 | struct cacheinfo *this_leaf, *sib_leaf;
|
| 346 | unsigned int index, sib_index;
|
| 347 | int ret = 0;
|
| 348 |
|
| 349 | if (this_cpu_ci->cpu_map_populated)
|
| 350 | return 0;
|
| 351 |
|
| 352 | |
| 353 | * skip setting up cache properties if LLC is valid, just need
|
| 354 | * to update the shared cpu_map if the cache attributes were
|
| 355 | * populated early before all the cpus are brought online
|
| 356 |
|
| 357 | if (!last_level_cache_is_valid(cpu) && !use_arch_info) {
|
| 358 | ret = cache_setup_properties(cpu);
|
| 359 | if (ret)
|
| 360 | return ret;
|
| 361 | }
|
| 362 |
|
| 363 | for (index = 0; index < cache_leaves(cpu); index++) {
|
| 364 | unsigned int i;
|
| 365 |
|
| 366 | this_leaf = per_cpu_cacheinfo_idx(cpu, index);
|
| 367 |
|
| 368 | cpumask_set_cpu(cpu, &this_leaf->shared_cpu_map);
|
| 369 | for_each_online_cpu(i) {
|
| 370 | if (i == cpu || !per_cpu_cacheinfo(i))
|
| 371 | continue;
|
| 372 | for (sib_index = 0; sib_index < cache_leaves(i); sib_index++) {
|
| 373 | sib_leaf = per_cpu_cacheinfo_idx(i, sib_index);
|
| 374 |
|
| 375 | |
| 376 | * Comparing cache IDs only makes sense if the leaves
|
| 377 | * belong to the same cache level of same type. Skip
|
| 378 | * the check if level and type do not match.
|
| 379 |
|
| 380 | if (sib_leaf->level != this_leaf->level ||
|
| 381 | sib_leaf->type != this_leaf->type)
|
| 382 | continue;
|
| 383 |
|
| 384 | if (cache_leaves_are_shared(this_leaf, sib_leaf)) {
|
| 385 | cpumask_set_cpu(cpu, &sib_leaf->shared_cpu_map);
|
| 386 | cpumask_set_cpu(i, &this_leaf->shared_cpu_map);
|
| 387 | break;
|
| 388 | }
|
| 389 | }
|
| 390 | }
|
| 391 |
|
| 392 | if (this_leaf->coherency_line_size > coherency_max_size)
|
| 393 | coherency_max_size = this_leaf->coherency_line_size;
|
| 394 | }
|
| 395 |
|
| 396 |
|
| 397 | this_cpu_ci->cpu_map_populated = true;
|
| 398 | return 0;
|
| 399 | }
|
| 400 |
|
| 401 | static void cache_shared_cpu_map_remove(unsigned int cpu)
|
| 402 | {
|
| 403 | struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
|
| 404 | struct cacheinfo *this_leaf, *sib_leaf;
|
| 405 | unsigned int sibling, index, sib_index;
|
| 406 |
|
| 407 | for (index = 0; index < cache_leaves(cpu); index++) {
|
| 408 | this_leaf = per_cpu_cacheinfo_idx(cpu, index);
|
| 409 | for_each_cpu(sibling, &this_leaf->shared_cpu_map) {
|
| 410 | if (sibling == cpu || !per_cpu_cacheinfo(sibling))
|
| 411 | continue;
|
| 412 |
|
| 413 | for (sib_index = 0; sib_index < cache_leaves(sibling); sib_index++) {
|
| 414 | sib_leaf = per_cpu_cacheinfo_idx(sibling, sib_index);
|
| 415 |
|
| 416 | |
| 417 | * Comparing cache IDs only makes sense if the leaves
|
| 418 | * belong to the same cache level of same type. Skip
|
| 419 | * the check if level and type do not match.
|
| 420 |
|
| 421 | if (sib_leaf->level != this_leaf->level ||
|
| 422 | sib_leaf->type != this_leaf->type)
|
| 423 | continue;
|
| 424 |
|
| 425 | if (cache_leaves_are_shared(this_leaf, sib_leaf)) {
|
| 426 | cpumask_clear_cpu(cpu, &sib_leaf->shared_cpu_map);
|
| 427 | cpumask_clear_cpu(sibling, &this_leaf->shared_cpu_map);
|
| 428 | break;
|
| 429 | }
|
| 430 | }
|
| 431 | }
|
| 432 | }
|
| 433 |
|
| 434 |
|
| 435 | this_cpu_ci->cpu_map_populated = false;
|
| 436 | }
|
| 437 |
|
| 438 | static void free_cache_attributes(unsigned int cpu)
|
| 439 | {
|
| 440 | if (!per_cpu_cacheinfo(cpu))
|
| 441 | return;
|
| 442 |
|
| 443 | cache_shared_cpu_map_remove(cpu);
|
| 444 | }
|
| 445 |
|
| 446 | int __weak early_cache_level(unsigned int cpu)
|
| 447 | {
|
| 448 | return -ENOENT;
|
| 449 | }
|
| 450 |
|
| 451 | int __weak init_cache_level(unsigned int cpu)
|
| 452 | {
|
| 453 | return -ENOENT;
|
| 454 | }
|
| 455 |
|
| 456 | int __weak populate_cache_leaves(unsigned int cpu)
|
| 457 | {
|
| 458 | return -ENOENT;
|
| 459 | }
|
| 460 |
|
| 461 | static inline int allocate_cache_info(int cpu)
|
| 462 | {
|
| 463 | per_cpu_cacheinfo(cpu) = kcalloc(cache_leaves(cpu), sizeof(struct cacheinfo), GFP_ATOMIC);
|
| 464 | if (!per_cpu_cacheinfo(cpu)) {
|
| 465 | cache_leaves(cpu) = 0;
|
| 466 | return -ENOMEM;
|
| 467 | }
|
| 468 |
|
| 469 | return 0;
|
| 470 | }
|
| 471 |
|
| 472 | int fetch_cache_info(unsigned int cpu)
|
| 473 | {
|
| 474 | struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
|
| 475 | unsigned int levels = 0, split_levels = 0;
|
| 476 | int ret;
|
| 477 |
|
| 478 | if (acpi_disabled) {
|
| 479 | ret = init_of_cache_level(cpu);
|
| 480 | } else {
|
| 481 | ret = acpi_get_cache_info(cpu, &levels, &split_levels);
|
| 482 | if (!ret) {
|
| 483 | this_cpu_ci->num_levels = levels;
|
| 484 | |
| 485 | * This assumes that:
|
| 486 | * - there cannot be any split caches (data/instruction)
|
| 487 | * above a unified cache
|
| 488 | * - data/instruction caches come by pair
|
| 489 |
|
| 490 | this_cpu_ci->num_leaves = levels + split_levels;
|
| 491 | }
|
| 492 | }
|
| 493 |
|
| 494 | if (ret || !cache_leaves(cpu)) {
|
| 495 | ret = early_cache_level(cpu);
|
| 496 | if (ret)
|
| 497 | return ret;
|
| 498 |
|
| 499 | if (!cache_leaves(cpu))
|
| 500 | return -ENOENT;
|
| 501 |
|
| 502 | this_cpu_ci->early_ci_levels = true;
|
| 503 | }
|
| 504 |
|
| 505 | return allocate_cache_info(cpu);
|
| 506 | }
|
| 507 |
|
| 508 | static inline int init_level_allocate_ci(unsigned int cpu)
|
| 509 | {
|
| 510 | unsigned int early_leaves = cache_leaves(cpu);
|
| 511 |
|
| 512 | |
| 513 | * via fetch_cache_info() and this also gets called as CPU hotplug
|
| 514 | * callbacks via cacheinfo_cpu_online, the init/alloc can be skipped
|
| 515 | * as it will happen only once (the cacheinfo memory is never freed).
|
| 516 | * Just populate the cacheinfo. However, if the cacheinfo has been
|
| 517 | * allocated early through the arch-specific early_cache_level() call,
|
| 518 | * there is a chance the info is wrong (this can happen on arm64). In
|
| 519 | * that case, call init_cache_level() anyway to give the arch-specific
|
| 520 | * code a chance to make things right.
|
| 521 |
|
| 522 | if (per_cpu_cacheinfo(cpu) && !ci_cacheinfo(cpu)->early_ci_levels)
|
| 523 | return 0;
|
| 524 |
|
| 525 | if (init_cache_level(cpu) || !cache_leaves(cpu))
|
| 526 | return -ENOENT;
|
| 527 |
|
| 528 | |
| 529 | * Now that we have properly initialized the cache level info, make
|
| 530 | * sure we don't try to do that again the next time we are called
|
| 531 | * (e.g. as CPU hotplug callbacks).
|
| 532 |
|
| 533 | ci_cacheinfo(cpu)->early_ci_levels = false;
|
| 534 |
|
| 535 | |
| 536 | * Some architectures (e.g., x86) do not use early initialization.
|
| 537 | * Allocate memory now in such case.
|
| 538 |
|
| 539 | if (cache_leaves(cpu) <= early_leaves && per_cpu_cacheinfo(cpu))
|
| 540 | return 0;
|
| 541 |
|
| 542 | kfree(per_cpu_cacheinfo(cpu));
|
| 543 | return allocate_cache_info(cpu);
|
| 544 | }
|
| 545 |
|
| 546 | int detect_cache_attributes(unsigned int cpu)
|
| 547 | {
|
| 548 | int ret;
|
| 549 |
|
| 550 | ret = init_level_allocate_ci(cpu);
|
| 551 | if (ret)
|
| 552 | return ret;
|
| 553 |
|
| 554 | |
| 555 | * If LLC is valid the cache leaves were already populated so just go to
|
| 556 | * update the cpu map.
|
| 557 |
|
| 558 | if (!last_level_cache_is_valid(cpu)) {
|
| 559 | |
| 560 | * populate_cache_leaves() may completely setup the cache leaves and
|
| 561 | * shared_cpu_map or it may leave it partially setup.
|
| 562 |
|
| 563 | ret = populate_cache_leaves(cpu);
|
| 564 | if (ret)
|
| 565 | goto free_ci;
|
| 566 | }
|
| 567 |
|
| 568 | |
| 569 | * For systems using DT for cache hierarchy, fw_token
|
| 570 | * and shared_cpu_map will be set up here only if they are
|
| 571 | * not populated already
|
| 572 |
|
| 573 | ret = cache_shared_cpu_map_setup(cpu);
|
| 574 | if (ret) {
|
| 575 | pr_warn("Unable to detect cache hierarchy for CPU %d\n", cpu);
|
| 576 | goto free_ci;
|
| 577 | }
|
| 578 |
|
| 579 | return 0;
|
| 580 |
|
| 581 | free_ci:
|
| 582 | free_cache_attributes(cpu);
|
| 583 | return ret;
|
| 584 | }
|
| 585 |
|
| 586 |
|
| 587 | static DEFINE_PER_CPU(struct device *, ci_cache_dev);
|
| 588 | #define per_cpu_cache_dev(cpu) (per_cpu(ci_cache_dev, cpu))
|
| 589 |
|
| 590 | static cpumask_t cache_dev_map;
|
| 591 |
|
| 592 |
|
| 593 | static DEFINE_PER_CPU(struct device **, ci_index_dev);
|
| 594 | #define per_cpu_index_dev(cpu) (per_cpu(ci_index_dev, cpu))
|
| 595 | #define per_cache_index_dev(cpu, idx) ((per_cpu_index_dev(cpu))[idx])
|
| 596 |
|
| 597 | #define show_one(file_name, object) \
|
| 598 | static ssize_t file_name##_show(struct device *dev, \
|
| 599 | struct device_attribute *attr, char *buf) \
|
| 600 | { \
|
| 601 | struct cacheinfo *this_leaf = dev_get_drvdata(dev); \
|
| 602 | return sysfs_emit(buf, "%u\n", this_leaf->object); \
|
| 603 | }
|
| 604 |
|
| 605 | show_one(id, id);
|
| 606 | show_one(level, level);
|
| 607 | show_one(coherency_line_size, coherency_line_size);
|
| 608 | show_one(number_of_sets, number_of_sets);
|
| 609 | show_one(physical_line_partition, physical_line_partition);
|
| 610 | show_one(ways_of_associativity, ways_of_associativity);
|
| 611 |
|
| 612 | static ssize_t size_show(struct device *dev,
|
| 613 | struct device_attribute *attr, char *buf)
|
| 614 | {
|
| 615 | struct cacheinfo *this_leaf = dev_get_drvdata(dev);
|
| 616 |
|
| 617 | return sysfs_emit(buf, "%uK\n", this_leaf->size >> 10);
|
| 618 | }
|
| 619 |
|
| 620 | static ssize_t shared_cpu_map_show(struct device *dev,
|
| 621 | struct device_attribute *attr, char *buf)
|
| 622 | {
|
| 623 | struct cacheinfo *this_leaf = dev_get_drvdata(dev);
|
| 624 | const struct cpumask *mask = &this_leaf->shared_cpu_map;
|
| 625 |
|
| 626 | return sysfs_emit(buf, "%*pb\n", nr_cpu_ids, mask);
|
| 627 | }
|
| 628 |
|
| 629 | static ssize_t shared_cpu_list_show(struct device *dev,
|
| 630 | struct device_attribute *attr, char *buf)
|
| 631 | {
|
| 632 | struct cacheinfo *this_leaf = dev_get_drvdata(dev);
|
| 633 | const struct cpumask *mask = &this_leaf->shared_cpu_map;
|
| 634 |
|
| 635 | return sysfs_emit(buf, "%*pbl\n", nr_cpu_ids, mask);
|
| 636 | }
|
| 637 |
|
| 638 | static ssize_t type_show(struct device *dev,
|
| 639 | struct device_attribute *attr, char *buf)
|
| 640 | {
|
| 641 | struct cacheinfo *this_leaf = dev_get_drvdata(dev);
|
| 642 | const char *output;
|
| 643 |
|
| 644 | switch (this_leaf->type) {
|
| 645 | case CACHE_TYPE_DATA:
|
| 646 | output = "Data";
|
| 647 | break;
|
| 648 | case CACHE_TYPE_INST:
|
| 649 | output = "Instruction";
|
| 650 | break;
|
| 651 | case CACHE_TYPE_UNIFIED:
|
| 652 | output = "Unified";
|
| 653 | break;
|
| 654 | default:
|
| 655 | return -EINVAL;
|
| 656 | }
|
| 657 |
|
| 658 | return sysfs_emit(buf, "%s\n", output);
|
| 659 | }
|
| 660 |
|
| 661 | static ssize_t allocation_policy_show(struct device *dev,
|
| 662 | struct device_attribute *attr, char *buf)
|
| 663 | {
|
| 664 | struct cacheinfo *this_leaf = dev_get_drvdata(dev);
|
| 665 | unsigned int ci_attr = this_leaf->attributes;
|
| 666 | const char *output;
|
| 667 |
|
| 668 | if ((ci_attr & CACHE_READ_ALLOCATE) && (ci_attr & CACHE_WRITE_ALLOCATE))
|
| 669 | output = "ReadWriteAllocate";
|
| 670 | else if (ci_attr & CACHE_READ_ALLOCATE)
|
| 671 | output = "ReadAllocate";
|
| 672 | else if (ci_attr & CACHE_WRITE_ALLOCATE)
|
| 673 | output = "WriteAllocate";
|
| 674 | else
|
| 675 | return 0;
|
| 676 |
|
| 677 | return sysfs_emit(buf, "%s\n", output);
|
| 678 | }
|
| 679 |
|
| 680 | static ssize_t write_policy_show(struct device *dev,
|
| 681 | struct device_attribute *attr, char *buf)
|
| 682 | {
|
| 683 | struct cacheinfo *this_leaf = dev_get_drvdata(dev);
|
| 684 | unsigned int ci_attr = this_leaf->attributes;
|
| 685 | int n = 0;
|
| 686 |
|
| 687 | if (ci_attr & CACHE_WRITE_THROUGH)
|
| 688 | n = sysfs_emit(buf, "WriteThrough\n");
|
| 689 | else if (ci_attr & CACHE_WRITE_BACK)
|
| 690 | n = sysfs_emit(buf, "WriteBack\n");
|
| 691 | return n;
|
| 692 | }
|
| 693 |
|
| 694 | static DEVICE_ATTR_RO(id);
|
| 695 | static DEVICE_ATTR_RO(level);
|
| 696 | static DEVICE_ATTR_RO(type);
|
| 697 | static DEVICE_ATTR_RO(coherency_line_size);
|
| 698 | static DEVICE_ATTR_RO(ways_of_associativity);
|
| 699 | static DEVICE_ATTR_RO(number_of_sets);
|
| 700 | static DEVICE_ATTR_RO(size);
|
| 701 | static DEVICE_ATTR_RO(allocation_policy);
|
| 702 | static DEVICE_ATTR_RO(write_policy);
|
| 703 | static DEVICE_ATTR_RO(shared_cpu_map);
|
| 704 | static DEVICE_ATTR_RO(shared_cpu_list);
|
| 705 | static DEVICE_ATTR_RO(physical_line_partition);
|
| 706 |
|
| 707 | static struct attribute *cache_default_attrs[] = {
|
| 708 | &dev_attr_id.attr,
|
| 709 | &dev_attr_type.attr,
|
| 710 | &dev_attr_level.attr,
|
| 711 | &dev_attr_shared_cpu_map.attr,
|
| 712 | &dev_attr_shared_cpu_list.attr,
|
| 713 | &dev_attr_coherency_line_size.attr,
|
| 714 | &dev_attr_ways_of_associativity.attr,
|
| 715 | &dev_attr_number_of_sets.attr,
|
| 716 | &dev_attr_size.attr,
|
| 717 | &dev_attr_allocation_policy.attr,
|
| 718 | &dev_attr_write_policy.attr,
|
| 719 | &dev_attr_physical_line_partition.attr,
|
| 720 | NULL
|
| 721 | };
|
| 722 |
|
| 723 | static umode_t
|
| 724 | cache_default_attrs_is_visible(struct kobject *kobj,
|
| 725 | struct attribute *attr, int unused)
|
| 726 | {
|
| 727 | struct device *dev = kobj_to_dev(kobj);
|
| 728 | struct cacheinfo *this_leaf = dev_get_drvdata(dev);
|
| 729 | const struct cpumask *mask = &this_leaf->shared_cpu_map;
|
| 730 | umode_t mode = attr->mode;
|
| 731 |
|
| 732 | if ((attr == &dev_attr_id.attr) && (this_leaf->attributes & CACHE_ID))
|
| 733 | return mode;
|
| 734 | if ((attr == &dev_attr_type.attr) && this_leaf->type)
|
| 735 | return mode;
|
| 736 | if ((attr == &dev_attr_level.attr) && this_leaf->level)
|
| 737 | return mode;
|
| 738 | if ((attr == &dev_attr_shared_cpu_map.attr) && !cpumask_empty(mask))
|
| 739 | return mode;
|
| 740 | if ((attr == &dev_attr_shared_cpu_list.attr) && !cpumask_empty(mask))
|
| 741 | return mode;
|
| 742 | if ((attr == &dev_attr_coherency_line_size.attr) &&
|
| 743 | this_leaf->coherency_line_size)
|
| 744 | return mode;
|
| 745 | if ((attr == &dev_attr_ways_of_associativity.attr) &&
|
| 746 | this_leaf->size)
|
| 747 | return mode;
|
| 748 | if ((attr == &dev_attr_number_of_sets.attr) &&
|
| 749 | this_leaf->number_of_sets)
|
| 750 | return mode;
|
| 751 | if ((attr == &dev_attr_size.attr) && this_leaf->size)
|
| 752 | return mode;
|
| 753 | if ((attr == &dev_attr_write_policy.attr) &&
|
| 754 | (this_leaf->attributes & CACHE_WRITE_POLICY_MASK))
|
| 755 | return mode;
|
| 756 | if ((attr == &dev_attr_allocation_policy.attr) &&
|
| 757 | (this_leaf->attributes & CACHE_ALLOCATE_POLICY_MASK))
|
| 758 | return mode;
|
| 759 | if ((attr == &dev_attr_physical_line_partition.attr) &&
|
| 760 | this_leaf->physical_line_partition)
|
| 761 | return mode;
|
| 762 |
|
| 763 | return 0;
|
| 764 | }
|
| 765 |
|
| 766 | static const struct attribute_group cache_default_group = {
|
| 767 | .attrs = cache_default_attrs,
|
| 768 | .is_visible = cache_default_attrs_is_visible,
|
| 769 | };
|
| 770 |
|
| 771 | static const struct attribute_group *cache_default_groups[] = {
|
| 772 | &cache_default_group,
|
| 773 | NULL,
|
| 774 | };
|
| 775 |
|
| 776 | static const struct attribute_group *cache_private_groups[] = {
|
| 777 | &cache_default_group,
|
| 778 | NULL,
|
| 779 | NULL,
|
| 780 | };
|
| 781 |
|
| 782 | const struct attribute_group *
|
| 783 | __weak cache_get_priv_group(struct cacheinfo *this_leaf)
|
| 784 | {
|
| 785 | return NULL;
|
| 786 | }
|
| 787 |
|
| 788 | static const struct attribute_group **
|
| 789 | cache_get_attribute_groups(struct cacheinfo *this_leaf)
|
| 790 | {
|
| 791 | const struct attribute_group *priv_group =
|
| 792 | cache_get_priv_group(this_leaf);
|
| 793 |
|
| 794 | if (!priv_group)
|
| 795 | return cache_default_groups;
|
| 796 |
|
| 797 | if (!cache_private_groups[1])
|
| 798 | cache_private_groups[1] = priv_group;
|
| 799 |
|
| 800 | return cache_private_groups;
|
| 801 | }
|
| 802 |
|
| 803 |
|
| 804 | static void cpu_cache_sysfs_exit(unsigned int cpu)
|
| 805 | {
|
| 806 | int i;
|
| 807 | struct device *ci_dev;
|
| 808 |
|
| 809 | if (per_cpu_index_dev(cpu)) {
|
| 810 | for (i = 0; i < cache_leaves(cpu); i++) {
|
| 811 | ci_dev = per_cache_index_dev(cpu, i);
|
| 812 | if (!ci_dev)
|
| 813 | continue;
|
| 814 | device_unregister(ci_dev);
|
| 815 | }
|
| 816 | kfree(per_cpu_index_dev(cpu));
|
| 817 | per_cpu_index_dev(cpu) = NULL;
|
| 818 | }
|
| 819 | device_unregister(per_cpu_cache_dev(cpu));
|
| 820 | per_cpu_cache_dev(cpu) = NULL;
|
| 821 | }
|
| 822 |
|
| 823 | static int cpu_cache_sysfs_init(unsigned int cpu)
|
| 824 | {
|
| 825 | struct device *dev = get_cpu_device(cpu);
|
| 826 |
|
| 827 | if (per_cpu_cacheinfo(cpu) == NULL)
|
| 828 | return -ENOENT;
|
| 829 |
|
| 830 | per_cpu_cache_dev(cpu) = cpu_device_create(dev, NULL, NULL, "cache");
|
| 831 | if (IS_ERR(per_cpu_cache_dev(cpu)))
|
| 832 | return PTR_ERR(per_cpu_cache_dev(cpu));
|
| 833 |
|
| 834 |
|
| 835 | per_cpu_index_dev(cpu) = kcalloc(cache_leaves(cpu),
|
| 836 | sizeof(struct device *), GFP_KERNEL);
|
| 837 | if (unlikely(per_cpu_index_dev(cpu) == NULL))
|
| 838 | goto err_out;
|
| 839 |
|
| 840 | return 0;
|
| 841 |
|
| 842 | err_out:
|
| 843 | cpu_cache_sysfs_exit(cpu);
|
| 844 | return -ENOMEM;
|
| 845 | }
|
| 846 |
|
| 847 | static int cache_add_dev(unsigned int cpu)
|
| 848 | {
|
| 849 | unsigned int i;
|
| 850 | int rc;
|
| 851 | struct device *ci_dev, *parent;
|
| 852 | struct cacheinfo *this_leaf;
|
| 853 | const struct attribute_group **cache_groups;
|
| 854 |
|
| 855 | rc = cpu_cache_sysfs_init(cpu);
|
| 856 | if (unlikely(rc < 0))
|
| 857 | return rc;
|
| 858 |
|
| 859 | parent = per_cpu_cache_dev(cpu);
|
| 860 | for (i = 0; i < cache_leaves(cpu); i++) {
|
| 861 | this_leaf = per_cpu_cacheinfo_idx(cpu, i);
|
| 862 | if (this_leaf->disable_sysfs)
|
| 863 | continue;
|
| 864 | if (this_leaf->type == CACHE_TYPE_NOCACHE)
|
| 865 | break;
|
| 866 | cache_groups = cache_get_attribute_groups(this_leaf);
|
| 867 | ci_dev = cpu_device_create(parent, this_leaf, cache_groups,
|
| 868 | "index%1u", i);
|
| 869 | if (IS_ERR(ci_dev)) {
|
| 870 | rc = PTR_ERR(ci_dev);
|
| 871 | goto err;
|
| 872 | }
|
| 873 | per_cache_index_dev(cpu, i) = ci_dev;
|
| 874 | }
|
| 875 | cpumask_set_cpu(cpu, &cache_dev_map);
|
| 876 |
|
| 877 | return 0;
|
| 878 | err:
|
| 879 | cpu_cache_sysfs_exit(cpu);
|
| 880 | return rc;
|
| 881 | }
|
| 882 |
|
| 883 | static unsigned int cpu_map_shared_cache(bool online, unsigned int cpu,
|
| 884 | cpumask_t **map)
|
| 885 | {
|
| 886 | struct cacheinfo *llc, *sib_llc;
|
| 887 | unsigned int sibling;
|
| 888 |
|
| 889 | if (!last_level_cache_is_valid(cpu))
|
| 890 | return 0;
|
| 891 |
|
| 892 | llc = per_cpu_cacheinfo_idx(cpu, cache_leaves(cpu) - 1);
|
| 893 |
|
| 894 | if (llc->type != CACHE_TYPE_DATA && llc->type != CACHE_TYPE_UNIFIED)
|
| 895 | return 0;
|
| 896 |
|
| 897 | if (online) {
|
| 898 | *map = &llc->shared_cpu_map;
|
| 899 | return cpumask_weight(*map);
|
| 900 | }
|
| 901 |
|
| 902 |
|
| 903 | for_each_cpu(sibling, &llc->shared_cpu_map) {
|
| 904 | if (sibling == cpu || !last_level_cache_is_valid(sibling))
|
| 905 | continue;
|
| 906 | sib_llc = per_cpu_cacheinfo_idx(sibling, cache_leaves(sibling) - 1);
|
| 907 | *map = &sib_llc->shared_cpu_map;
|
| 908 | return cpumask_weight(*map);
|
| 909 | }
|
| 910 |
|
| 911 | return 0;
|
| 912 | }
|
| 913 |
|
| 914 | |
| 915 | * Calculate the size of the per-CPU data cache slice. This can be
|
| 916 | * used to estimate the size of the data cache slice that can be used
|
| 917 | * by one CPU under ideal circumstances. UNIFIED caches are counted
|
| 918 | * in addition to DATA caches. So, please consider code cache usage
|
| 919 | * when use the result.
|
| 920 | *
|
| 921 | * Because the cache inclusive/non-inclusive information isn't
|
| 922 | * available, we just use the size of the per-CPU slice of LLC to make
|
| 923 | * the result more predictable across architectures.
|
| 924 |
|
| 925 | static void update_per_cpu_data_slice_size_cpu(unsigned int cpu)
|
| 926 | {
|
| 927 | struct cpu_cacheinfo *ci;
|
| 928 | struct cacheinfo *llc;
|
| 929 | unsigned int nr_shared;
|
| 930 |
|
| 931 | if (!last_level_cache_is_valid(cpu))
|
| 932 | return;
|
| 933 |
|
| 934 | ci = ci_cacheinfo(cpu);
|
| 935 | llc = per_cpu_cacheinfo_idx(cpu, cache_leaves(cpu) - 1);
|
| 936 |
|
| 937 | if (llc->type != CACHE_TYPE_DATA && llc->type != CACHE_TYPE_UNIFIED)
|
| 938 | return;
|
| 939 |
|
| 940 | nr_shared = cpumask_weight(&llc->shared_cpu_map);
|
| 941 | if (nr_shared)
|
| 942 | ci->per_cpu_data_slice_size = llc->size / nr_shared;
|
| 943 | }
|
| 944 |
|
| 945 | static void update_per_cpu_data_slice_size(bool cpu_online, unsigned int cpu,
|
| 946 | cpumask_t *cpu_map)
|
| 947 | {
|
| 948 | unsigned int icpu;
|
| 949 |
|
| 950 | for_each_cpu(icpu, cpu_map) {
|
| 951 | if (!cpu_online && icpu == cpu)
|
| 952 | continue;
|
| 953 | update_per_cpu_data_slice_size_cpu(icpu);
|
| 954 | setup_pcp_cacheinfo(icpu);
|
| 955 | }
|
| 956 | }
|
| 957 |
|
| 958 | static int cacheinfo_cpu_online(unsigned int cpu)
|
| 959 | {
|
| 960 | int rc = detect_cache_attributes(cpu);
|
| 961 | cpumask_t *cpu_map;
|
| 962 |
|
| 963 | if (rc)
|
| 964 | return rc;
|
| 965 | rc = cache_add_dev(cpu);
|
| 966 | if (rc)
|
| 967 | goto err;
|
| 968 | if (cpu_map_shared_cache(true, cpu, &cpu_map))
|
| 969 | update_per_cpu_data_slice_size(true, cpu, cpu_map);
|
| 970 | return 0;
|
| 971 | err:
|
| 972 | free_cache_attributes(cpu);
|
| 973 | return rc;
|
| 974 | }
|
| 975 |
|
| 976 | static int cacheinfo_cpu_pre_down(unsigned int cpu)
|
| 977 | {
|
| 978 | cpumask_t *cpu_map;
|
| 979 | unsigned int nr_shared;
|
| 980 |
|
| 981 | nr_shared = cpu_map_shared_cache(false, cpu, &cpu_map);
|
| 982 | if (cpumask_test_and_clear_cpu(cpu, &cache_dev_map))
|
| 983 | cpu_cache_sysfs_exit(cpu);
|
| 984 |
|
| 985 | free_cache_attributes(cpu);
|
| 986 | if (nr_shared > 1)
|
| 987 | update_per_cpu_data_slice_size(false, cpu, cpu_map);
|
| 988 | return 0;
|
| 989 | }
|
| 990 |
|
| 991 | static int __init cacheinfo_sysfs_init(void)
|
| 992 | {
|
| 993 | return cpuhp_setup_state(CPUHP_AP_BASE_CACHEINFO_ONLINE,
|
| 994 | "base/cacheinfo:online",
|
| 995 | cacheinfo_cpu_online, cacheinfo_cpu_pre_down);
|
| 996 | }
|
| 997 | device_initcall(cacheinfo_sysfs_init);
|