| 1 | |
| 2 | |
| 3 | * User interface for Resource Allocation in Resource Director Technology(RDT)
|
| 4 | *
|
| 5 | * Copyright (C) 2016 Intel Corporation
|
| 6 | *
|
| 7 | * Author: Fenghua Yu <[email protected]>
|
| 8 | *
|
| 9 | * More information about RDT be found in the Intel (R) x86 Architecture
|
| 10 | * Software Developer Manual.
|
| 11 |
|
| 12 |
|
| 13 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
|
| 14 |
|
| 15 | #include <linux/cpu.h>
|
| 16 | #include <linux/debugfs.h>
|
| 17 | #include <linux/fs.h>
|
| 18 | #include <linux/fs_parser.h>
|
| 19 | #include <linux/sysfs.h>
|
| 20 | #include <linux/kernfs.h>
|
| 21 | #include <linux/once.h>
|
| 22 | #include <linux/resctrl.h>
|
| 23 | #include <linux/seq_buf.h>
|
| 24 | #include <linux/seq_file.h>
|
| 25 | #include <linux/sched/task.h>
|
| 26 | #include <linux/slab.h>
|
| 27 | #include <linux/user_namespace.h>
|
| 28 |
|
| 29 | #include <uapi/linux/magic.h>
|
| 30 |
|
| 31 | #include "internal.h"
|
| 32 |
|
| 33 |
|
| 34 | DEFINE_MUTEX(rdtgroup_mutex);
|
| 35 |
|
| 36 | static struct kernfs_root *rdt_root;
|
| 37 |
|
| 38 | struct rdtgroup rdtgroup_default;
|
| 39 |
|
| 40 | LIST_HEAD(rdt_all_groups);
|
| 41 |
|
| 42 |
|
| 43 | LIST_HEAD(resctrl_schema_all);
|
| 44 |
|
| 45 | |
| 46 | * List of struct mon_data containing private data of event files for use by
|
| 47 | * rdtgroup_mondata_show(). Protected by rdtgroup_mutex.
|
| 48 |
|
| 49 | static LIST_HEAD(mon_data_kn_priv_list);
|
| 50 |
|
| 51 |
|
| 52 | bool resctrl_mounted;
|
| 53 |
|
| 54 |
|
| 55 | static struct kernfs_node *kn_info;
|
| 56 |
|
| 57 |
|
| 58 | static struct kernfs_node *kn_mongrp;
|
| 59 |
|
| 60 |
|
| 61 | static struct kernfs_node *kn_mondata;
|
| 62 |
|
| 63 | |
| 64 | * Used to store the max resource name width to display the schemata names in
|
| 65 | * a tabular format.
|
| 66 |
|
| 67 | int max_name_width;
|
| 68 |
|
| 69 | static struct seq_buf last_cmd_status;
|
| 70 |
|
| 71 | static char last_cmd_status_buf[512];
|
| 72 |
|
| 73 | static int rdtgroup_setup_root(struct rdt_fs_context *ctx);
|
| 74 |
|
| 75 | static void rdtgroup_destroy_root(void);
|
| 76 |
|
| 77 | struct dentry *debugfs_resctrl;
|
| 78 |
|
| 79 | |
| 80 | * Memory bandwidth monitoring event to use for the default CTRL_MON group
|
| 81 | * and each new CTRL_MON group created by the user. Only relevant when
|
| 82 | * the filesystem is mounted with the "mba_MBps" option so it does not
|
| 83 | * matter that it remains uninitialized on systems that do not support
|
| 84 | * the "mba_MBps" option.
|
| 85 |
|
| 86 | enum resctrl_event_id mba_mbps_default_event;
|
| 87 |
|
| 88 | static bool resctrl_debug;
|
| 89 |
|
| 90 | void rdt_last_cmd_clear(void)
|
| 91 | {
|
| 92 | lockdep_assert_held(&rdtgroup_mutex);
|
| 93 | seq_buf_clear(&last_cmd_status);
|
| 94 | }
|
| 95 |
|
| 96 | void rdt_last_cmd_puts(const char *s)
|
| 97 | {
|
| 98 | lockdep_assert_held(&rdtgroup_mutex);
|
| 99 | seq_buf_puts(&last_cmd_status, s);
|
| 100 | }
|
| 101 |
|
| 102 | void rdt_last_cmd_printf(const char *fmt, ...)
|
| 103 | {
|
| 104 | va_list ap;
|
| 105 |
|
| 106 | va_start(ap, fmt);
|
| 107 | lockdep_assert_held(&rdtgroup_mutex);
|
| 108 | seq_buf_vprintf(&last_cmd_status, fmt, ap);
|
| 109 | va_end(ap);
|
| 110 | }
|
| 111 |
|
| 112 | void rdt_staged_configs_clear(void)
|
| 113 | {
|
| 114 | struct rdt_ctrl_domain *dom;
|
| 115 | struct rdt_resource *r;
|
| 116 |
|
| 117 | lockdep_assert_held(&rdtgroup_mutex);
|
| 118 |
|
| 119 | for_each_alloc_capable_rdt_resource(r) {
|
| 120 | list_for_each_entry_rcu(dom, &r->ctrl_domains, hdr.list, lockdep_is_cpus_held())
|
| 121 | memset(dom->staged_config, 0, sizeof(dom->staged_config));
|
| 122 | }
|
| 123 | }
|
| 124 |
|
| 125 | static bool resctrl_is_mbm_enabled(void)
|
| 126 | {
|
| 127 | return (resctrl_is_mon_event_enabled(QOS_L3_MBM_TOTAL_EVENT_ID) ||
|
| 128 | resctrl_is_mon_event_enabled(QOS_L3_MBM_LOCAL_EVENT_ID));
|
| 129 | }
|
| 130 |
|
| 131 | |
| 132 | * Trivial allocator for CLOSIDs. Use BITMAP APIs to manipulate a bitmap
|
| 133 | * of free CLOSIDs.
|
| 134 | *
|
| 135 | * Using a global CLOSID across all resources has some advantages and
|
| 136 | * some drawbacks:
|
| 137 | * + We can simply set current's closid to assign a task to a resource
|
| 138 | * group.
|
| 139 | * + Context switch code can avoid extra memory references deciding which
|
| 140 | * CLOSID to load into the PQR_ASSOC MSR
|
| 141 | * - We give up some options in configuring resource groups across multi-socket
|
| 142 | * systems.
|
| 143 | * - Our choices on how to configure each resource become progressively more
|
| 144 | * limited as the number of resources grows.
|
| 145 |
|
| 146 | static unsigned long *closid_free_map;
|
| 147 |
|
| 148 | static int closid_free_map_len;
|
| 149 |
|
| 150 | int closids_supported(void)
|
| 151 | {
|
| 152 | return closid_free_map_len;
|
| 153 | }
|
| 154 |
|
| 155 | static int closid_init(void)
|
| 156 | {
|
| 157 | struct resctrl_schema *s;
|
| 158 | u32 rdt_min_closid = ~0;
|
| 159 |
|
| 160 |
|
| 161 | if (list_empty(&resctrl_schema_all))
|
| 162 | return 0;
|
| 163 |
|
| 164 |
|
| 165 | list_for_each_entry(s, &resctrl_schema_all, list)
|
| 166 | rdt_min_closid = min(rdt_min_closid, s->num_closid);
|
| 167 |
|
| 168 | closid_free_map = bitmap_alloc(rdt_min_closid, GFP_KERNEL);
|
| 169 | if (!closid_free_map)
|
| 170 | return -ENOMEM;
|
| 171 | bitmap_fill(closid_free_map, rdt_min_closid);
|
| 172 |
|
| 173 |
|
| 174 | __clear_bit(RESCTRL_RESERVED_CLOSID, closid_free_map);
|
| 175 | closid_free_map_len = rdt_min_closid;
|
| 176 |
|
| 177 | return 0;
|
| 178 | }
|
| 179 |
|
| 180 | static void closid_exit(void)
|
| 181 | {
|
| 182 | bitmap_free(closid_free_map);
|
| 183 | closid_free_map = NULL;
|
| 184 | }
|
| 185 |
|
| 186 | static int closid_alloc(void)
|
| 187 | {
|
| 188 | int cleanest_closid;
|
| 189 | u32 closid;
|
| 190 |
|
| 191 | lockdep_assert_held(&rdtgroup_mutex);
|
| 192 |
|
| 193 | if (IS_ENABLED(CONFIG_RESCTRL_RMID_DEPENDS_ON_CLOSID) &&
|
| 194 | resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID)) {
|
| 195 | cleanest_closid = resctrl_find_cleanest_closid();
|
| 196 | if (cleanest_closid < 0)
|
| 197 | return cleanest_closid;
|
| 198 | closid = cleanest_closid;
|
| 199 | } else {
|
| 200 | closid = find_first_bit(closid_free_map, closid_free_map_len);
|
| 201 | if (closid == closid_free_map_len)
|
| 202 | return -ENOSPC;
|
| 203 | }
|
| 204 | __clear_bit(closid, closid_free_map);
|
| 205 |
|
| 206 | return closid;
|
| 207 | }
|
| 208 |
|
| 209 | void closid_free(int closid)
|
| 210 | {
|
| 211 | lockdep_assert_held(&rdtgroup_mutex);
|
| 212 |
|
| 213 | __set_bit(closid, closid_free_map);
|
| 214 | }
|
| 215 |
|
| 216 | |
| 217 | * closid_allocated - test if provided closid is in use
|
| 218 | * @closid: closid to be tested
|
| 219 | *
|
| 220 | * Return: true if @closid is currently associated with a resource group,
|
| 221 | * false if @closid is free
|
| 222 |
|
| 223 | bool closid_allocated(unsigned int closid)
|
| 224 | {
|
| 225 | lockdep_assert_held(&rdtgroup_mutex);
|
| 226 |
|
| 227 | return !test_bit(closid, closid_free_map);
|
| 228 | }
|
| 229 |
|
| 230 | bool closid_alloc_fixed(u32 closid)
|
| 231 | {
|
| 232 | return __test_and_clear_bit(closid, closid_free_map);
|
| 233 | }
|
| 234 |
|
| 235 | |
| 236 | * rdtgroup_mode_by_closid - Return mode of resource group with closid
|
| 237 | * @closid: closid if the resource group
|
| 238 | *
|
| 239 | * Each resource group is associated with a @closid. Here the mode
|
| 240 | * of a resource group can be queried by searching for it using its closid.
|
| 241 | *
|
| 242 | * Return: mode as &enum rdtgrp_mode of resource group with closid @closid
|
| 243 |
|
| 244 | enum rdtgrp_mode rdtgroup_mode_by_closid(int closid)
|
| 245 | {
|
| 246 | struct rdtgroup *rdtgrp;
|
| 247 |
|
| 248 | list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
|
| 249 | if (rdtgrp->closid == closid)
|
| 250 | return rdtgrp->mode;
|
| 251 | }
|
| 252 |
|
| 253 | return RDT_NUM_MODES;
|
| 254 | }
|
| 255 |
|
| 256 | static const char * const rdt_mode_str[] = {
|
| 257 | [RDT_MODE_SHAREABLE] = "shareable",
|
| 258 | [RDT_MODE_EXCLUSIVE] = "exclusive",
|
| 259 | [RDT_MODE_PSEUDO_LOCKSETUP] = "pseudo-locksetup",
|
| 260 | [RDT_MODE_PSEUDO_LOCKED] = "pseudo-locked",
|
| 261 | };
|
| 262 |
|
| 263 | |
| 264 | * rdtgroup_mode_str - Return the string representation of mode
|
| 265 | * @mode: the resource group mode as &enum rdtgroup_mode
|
| 266 | *
|
| 267 | * Return: string representation of valid mode, "unknown" otherwise
|
| 268 |
|
| 269 | static const char *rdtgroup_mode_str(enum rdtgrp_mode mode)
|
| 270 | {
|
| 271 | if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES)
|
| 272 | return "unknown";
|
| 273 |
|
| 274 | return rdt_mode_str[mode];
|
| 275 | }
|
| 276 |
|
| 277 |
|
| 278 | static int rdtgroup_kn_set_ugid(struct kernfs_node *kn)
|
| 279 | {
|
| 280 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
|
| 281 | .ia_uid = current_fsuid(),
|
| 282 | .ia_gid = current_fsgid(), };
|
| 283 |
|
| 284 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
|
| 285 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
|
| 286 | return 0;
|
| 287 |
|
| 288 | return kernfs_setattr(kn, &iattr);
|
| 289 | }
|
| 290 |
|
| 291 | static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft)
|
| 292 | {
|
| 293 | struct kernfs_node *kn;
|
| 294 | int ret;
|
| 295 |
|
| 296 | kn = __kernfs_create_file(parent_kn, rft->name, rft->mode,
|
| 297 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
|
| 298 | 0, rft->kf_ops, rft, NULL, NULL);
|
| 299 | if (IS_ERR(kn))
|
| 300 | return PTR_ERR(kn);
|
| 301 |
|
| 302 | ret = rdtgroup_kn_set_ugid(kn);
|
| 303 | if (ret) {
|
| 304 | kernfs_remove(kn);
|
| 305 | return ret;
|
| 306 | }
|
| 307 |
|
| 308 | return 0;
|
| 309 | }
|
| 310 |
|
| 311 | static int rdtgroup_seqfile_show(struct seq_file *m, void *arg)
|
| 312 | {
|
| 313 | struct kernfs_open_file *of = m->private;
|
| 314 | struct rftype *rft = of->kn->priv;
|
| 315 |
|
| 316 | if (rft->seq_show)
|
| 317 | return rft->seq_show(of, m, arg);
|
| 318 | return 0;
|
| 319 | }
|
| 320 |
|
| 321 | static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf,
|
| 322 | size_t nbytes, loff_t off)
|
| 323 | {
|
| 324 | struct rftype *rft = of->kn->priv;
|
| 325 |
|
| 326 | if (rft->write)
|
| 327 | return rft->write(of, buf, nbytes, off);
|
| 328 |
|
| 329 | return -EINVAL;
|
| 330 | }
|
| 331 |
|
| 332 | static const struct kernfs_ops rdtgroup_kf_single_ops = {
|
| 333 | .atomic_write_len = PAGE_SIZE,
|
| 334 | .write = rdtgroup_file_write,
|
| 335 | .seq_show = rdtgroup_seqfile_show,
|
| 336 | };
|
| 337 |
|
| 338 | static const struct kernfs_ops kf_mondata_ops = {
|
| 339 | .atomic_write_len = PAGE_SIZE,
|
| 340 | .seq_show = rdtgroup_mondata_show,
|
| 341 | };
|
| 342 |
|
| 343 | static bool is_cpu_list(struct kernfs_open_file *of)
|
| 344 | {
|
| 345 | struct rftype *rft = of->kn->priv;
|
| 346 |
|
| 347 | return rft->flags & RFTYPE_FLAGS_CPUS_LIST;
|
| 348 | }
|
| 349 |
|
| 350 | static int rdtgroup_cpus_show(struct kernfs_open_file *of,
|
| 351 | struct seq_file *s, void *v)
|
| 352 | {
|
| 353 | struct rdtgroup *rdtgrp;
|
| 354 | struct cpumask *mask;
|
| 355 | int ret = 0;
|
| 356 |
|
| 357 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 358 |
|
| 359 | if (rdtgrp) {
|
| 360 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
|
| 361 | if (!rdtgrp->plr->d) {
|
| 362 | rdt_last_cmd_puts("Cache domain offline\n");
|
| 363 | ret = -ENODEV;
|
| 364 | } else {
|
| 365 | mask = &rdtgrp->plr->d->hdr.cpu_mask;
|
| 366 | seq_printf(s, is_cpu_list(of) ?
|
| 367 | "%*pbl\n" : "%*pb\n",
|
| 368 | cpumask_pr_args(mask));
|
| 369 | }
|
| 370 | } else {
|
| 371 | seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n",
|
| 372 | cpumask_pr_args(&rdtgrp->cpu_mask));
|
| 373 | }
|
| 374 | } else {
|
| 375 | ret = -ENOENT;
|
| 376 | }
|
| 377 | rdtgroup_kn_unlock(of->kn);
|
| 378 |
|
| 379 | return ret;
|
| 380 | }
|
| 381 |
|
| 382 | |
| 383 | * Update the PGR_ASSOC MSR on all cpus in @cpu_mask,
|
| 384 | *
|
| 385 | * Per task closids/rmids must have been set up before calling this function.
|
| 386 | * @r may be NULL.
|
| 387 |
|
| 388 | static void
|
| 389 | update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r)
|
| 390 | {
|
| 391 | struct resctrl_cpu_defaults defaults, *p = NULL;
|
| 392 |
|
| 393 | if (r) {
|
| 394 | defaults.closid = r->closid;
|
| 395 | defaults.rmid = r->mon.rmid;
|
| 396 | p = &defaults;
|
| 397 | }
|
| 398 |
|
| 399 | on_each_cpu_mask(cpu_mask, resctrl_arch_sync_cpu_closid_rmid, p, 1);
|
| 400 | }
|
| 401 |
|
| 402 | static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
|
| 403 | cpumask_var_t tmpmask)
|
| 404 | {
|
| 405 | struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp;
|
| 406 | struct list_head *head;
|
| 407 |
|
| 408 |
|
| 409 | cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask);
|
| 410 | if (!cpumask_empty(tmpmask)) {
|
| 411 | rdt_last_cmd_puts("Can only add CPUs to mongroup that belong to parent\n");
|
| 412 | return -EINVAL;
|
| 413 | }
|
| 414 |
|
| 415 |
|
| 416 | cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
|
| 417 | if (!cpumask_empty(tmpmask)) {
|
| 418 |
|
| 419 | cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask);
|
| 420 | update_closid_rmid(tmpmask, prgrp);
|
| 421 | }
|
| 422 |
|
| 423 | |
| 424 | * If we added cpus, remove them from previous group that owned them
|
| 425 | * and update per-cpu rmid
|
| 426 |
|
| 427 | cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
|
| 428 | if (!cpumask_empty(tmpmask)) {
|
| 429 | head = &prgrp->mon.crdtgrp_list;
|
| 430 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
|
| 431 | if (crgrp == rdtgrp)
|
| 432 | continue;
|
| 433 | cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask,
|
| 434 | tmpmask);
|
| 435 | }
|
| 436 | update_closid_rmid(tmpmask, rdtgrp);
|
| 437 | }
|
| 438 |
|
| 439 |
|
| 440 | cpumask_copy(&rdtgrp->cpu_mask, newmask);
|
| 441 |
|
| 442 | return 0;
|
| 443 | }
|
| 444 |
|
| 445 | static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m)
|
| 446 | {
|
| 447 | struct rdtgroup *crgrp;
|
| 448 |
|
| 449 | cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m);
|
| 450 |
|
| 451 | list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list)
|
| 452 | cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask);
|
| 453 | }
|
| 454 |
|
| 455 | static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
|
| 456 | cpumask_var_t tmpmask, cpumask_var_t tmpmask1)
|
| 457 | {
|
| 458 | struct rdtgroup *r, *crgrp;
|
| 459 | struct list_head *head;
|
| 460 |
|
| 461 |
|
| 462 | cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
|
| 463 | if (!cpumask_empty(tmpmask)) {
|
| 464 |
|
| 465 | if (rdtgrp == &rdtgroup_default) {
|
| 466 | rdt_last_cmd_puts("Can't drop CPUs from default group\n");
|
| 467 | return -EINVAL;
|
| 468 | }
|
| 469 |
|
| 470 |
|
| 471 | cpumask_or(&rdtgroup_default.cpu_mask,
|
| 472 | &rdtgroup_default.cpu_mask, tmpmask);
|
| 473 | update_closid_rmid(tmpmask, &rdtgroup_default);
|
| 474 | }
|
| 475 |
|
| 476 | |
| 477 | * If we added cpus, remove them from previous group and
|
| 478 | * the prev group's child groups that owned them
|
| 479 | * and update per-cpu closid/rmid.
|
| 480 |
|
| 481 | cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
|
| 482 | if (!cpumask_empty(tmpmask)) {
|
| 483 | list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) {
|
| 484 | if (r == rdtgrp)
|
| 485 | continue;
|
| 486 | cpumask_and(tmpmask1, &r->cpu_mask, tmpmask);
|
| 487 | if (!cpumask_empty(tmpmask1))
|
| 488 | cpumask_rdtgrp_clear(r, tmpmask1);
|
| 489 | }
|
| 490 | update_closid_rmid(tmpmask, rdtgrp);
|
| 491 | }
|
| 492 |
|
| 493 |
|
| 494 | cpumask_copy(&rdtgrp->cpu_mask, newmask);
|
| 495 |
|
| 496 | |
| 497 | * Clear child mon group masks since there is a new parent mask
|
| 498 | * now and update the rmid for the cpus the child lost.
|
| 499 |
|
| 500 | head = &rdtgrp->mon.crdtgrp_list;
|
| 501 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
|
| 502 | cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask);
|
| 503 | update_closid_rmid(tmpmask, rdtgrp);
|
| 504 | cpumask_clear(&crgrp->cpu_mask);
|
| 505 | }
|
| 506 |
|
| 507 | return 0;
|
| 508 | }
|
| 509 |
|
| 510 | static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of,
|
| 511 | char *buf, size_t nbytes, loff_t off)
|
| 512 | {
|
| 513 | cpumask_var_t tmpmask = CPUMASK_VAR_NULL, newmask = CPUMASK_VAR_NULL;
|
| 514 | cpumask_var_t tmpmask1 = CPUMASK_VAR_NULL;
|
| 515 | struct rdtgroup *rdtgrp;
|
| 516 | int ret;
|
| 517 |
|
| 518 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 519 | if (!rdtgrp) {
|
| 520 | ret = -ENOENT;
|
| 521 | goto out_unlock;
|
| 522 | }
|
| 523 |
|
| 524 | if (!buf) {
|
| 525 | rdt_last_cmd_printf("%s: Invalid input\n",
|
| 526 | is_cpu_list(of) ? "cpus_list" : "cpus");
|
| 527 | ret = -EINVAL;
|
| 528 | goto out_unlock;
|
| 529 | }
|
| 530 |
|
| 531 | if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL) ||
|
| 532 | !zalloc_cpumask_var(&newmask, GFP_KERNEL) ||
|
| 533 | !zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) {
|
| 534 | rdt_last_cmd_printf("%s: Kernel allocation failure\n",
|
| 535 | is_cpu_list(of) ? "cpus_list" : "cpus");
|
| 536 | ret = -ENOMEM;
|
| 537 | goto out_free;
|
| 538 | }
|
| 539 |
|
| 540 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
|
| 541 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
|
| 542 | ret = -EINVAL;
|
| 543 | rdt_last_cmd_puts("Pseudo-locking in progress\n");
|
| 544 | goto out_free;
|
| 545 | }
|
| 546 |
|
| 547 | if (is_cpu_list(of))
|
| 548 | ret = cpulist_parse(buf, newmask);
|
| 549 | else
|
| 550 | ret = cpumask_parse(buf, newmask);
|
| 551 |
|
| 552 | if (ret) {
|
| 553 | rdt_last_cmd_puts("Bad CPU list/mask\n");
|
| 554 | goto out_free;
|
| 555 | }
|
| 556 |
|
| 557 |
|
| 558 | cpumask_andnot(tmpmask, newmask, cpu_online_mask);
|
| 559 | if (!cpumask_empty(tmpmask)) {
|
| 560 | ret = -EINVAL;
|
| 561 | rdt_last_cmd_puts("Can only assign online CPUs\n");
|
| 562 | goto out_free;
|
| 563 | }
|
| 564 |
|
| 565 | if (rdtgrp->type == RDTCTRL_GROUP)
|
| 566 | ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1);
|
| 567 | else if (rdtgrp->type == RDTMON_GROUP)
|
| 568 | ret = cpus_mon_write(rdtgrp, newmask, tmpmask);
|
| 569 | else
|
| 570 | ret = -EINVAL;
|
| 571 |
|
| 572 | out_free:
|
| 573 | free_cpumask_var(tmpmask);
|
| 574 | free_cpumask_var(newmask);
|
| 575 | free_cpumask_var(tmpmask1);
|
| 576 | out_unlock:
|
| 577 | rdtgroup_kn_unlock(of->kn);
|
| 578 |
|
| 579 | return ret ?: nbytes;
|
| 580 | }
|
| 581 |
|
| 582 | |
| 583 | * rdtgroup_remove - the helper to remove resource group safely
|
| 584 | * @rdtgrp: resource group to remove
|
| 585 | *
|
| 586 | * On resource group creation via a mkdir, an extra kernfs_node reference is
|
| 587 | * taken to ensure that the rdtgroup structure remains accessible for the
|
| 588 | * rdtgroup_kn_unlock() calls where it is removed. The default group is
|
| 589 | * statically allocated: it does not have an extra reference but will have
|
| 590 | * RDT_DELETED set on unmount to support safe access to its associated files
|
| 591 | * via rdtgroup_kn_lock_live/rdtgroup_kn_unlock().
|
| 592 | *
|
| 593 | * For all but the default group: drop the extra reference, then free the
|
| 594 | * rdtgroup structure.
|
| 595 | *
|
| 596 | * Return: void
|
| 597 |
|
| 598 | static void rdtgroup_remove(struct rdtgroup *rdtgrp)
|
| 599 | {
|
| 600 | if (rdtgrp == &rdtgroup_default)
|
| 601 | return;
|
| 602 | kernfs_put(rdtgrp->kn);
|
| 603 | kfree(rdtgrp);
|
| 604 | }
|
| 605 |
|
| 606 | static void _update_task_closid_rmid(void *task)
|
| 607 | {
|
| 608 | |
| 609 | * If the task is still current on this CPU, update PQR_ASSOC MSR.
|
| 610 | * Otherwise, the MSR is updated when the task is scheduled in.
|
| 611 |
|
| 612 | if (task == current)
|
| 613 | resctrl_arch_sched_in(task);
|
| 614 | }
|
| 615 |
|
| 616 | static void update_task_closid_rmid(struct task_struct *t)
|
| 617 | {
|
| 618 | if (IS_ENABLED(CONFIG_SMP) && task_curr(t))
|
| 619 | smp_call_function_single(task_cpu(t), _update_task_closid_rmid, t, 1);
|
| 620 | else
|
| 621 | _update_task_closid_rmid(t);
|
| 622 | }
|
| 623 |
|
| 624 | static bool task_in_rdtgroup(struct task_struct *tsk, struct rdtgroup *rdtgrp)
|
| 625 | {
|
| 626 | u32 closid, rmid = rdtgrp->mon.rmid;
|
| 627 |
|
| 628 | if (rdtgrp->type == RDTCTRL_GROUP)
|
| 629 | closid = rdtgrp->closid;
|
| 630 | else if (rdtgrp->type == RDTMON_GROUP)
|
| 631 | closid = rdtgrp->mon.parent->closid;
|
| 632 | else
|
| 633 | return false;
|
| 634 |
|
| 635 | return resctrl_arch_match_closid(tsk, closid) &&
|
| 636 | resctrl_arch_match_rmid(tsk, closid, rmid);
|
| 637 | }
|
| 638 |
|
| 639 | static int __rdtgroup_move_task(struct task_struct *tsk,
|
| 640 | struct rdtgroup *rdtgrp)
|
| 641 | {
|
| 642 |
|
| 643 | if (task_in_rdtgroup(tsk, rdtgrp))
|
| 644 | return 0;
|
| 645 |
|
| 646 | |
| 647 | * Set the task's closid/rmid before the PQR_ASSOC MSR can be
|
| 648 | * updated by them.
|
| 649 | *
|
| 650 | * For ctrl_mon groups, move both closid and rmid.
|
| 651 | * For monitor groups, can move the tasks only from
|
| 652 | * their parent CTRL group.
|
| 653 |
|
| 654 | if (rdtgrp->type == RDTMON_GROUP &&
|
| 655 | !resctrl_arch_match_closid(tsk, rdtgrp->mon.parent->closid)) {
|
| 656 | rdt_last_cmd_puts("Can't move task to different control group\n");
|
| 657 | return -EINVAL;
|
| 658 | }
|
| 659 |
|
| 660 | if (rdtgrp->type == RDTMON_GROUP)
|
| 661 | resctrl_arch_set_closid_rmid(tsk, rdtgrp->mon.parent->closid,
|
| 662 | rdtgrp->mon.rmid);
|
| 663 | else
|
| 664 | resctrl_arch_set_closid_rmid(tsk, rdtgrp->closid,
|
| 665 | rdtgrp->mon.rmid);
|
| 666 |
|
| 667 | |
| 668 | * Ensure the task's closid and rmid are written before determining if
|
| 669 | * the task is current that will decide if it will be interrupted.
|
| 670 | * This pairs with the full barrier between the rq->curr update and
|
| 671 | * resctrl_arch_sched_in() during context switch.
|
| 672 |
|
| 673 | smp_mb();
|
| 674 |
|
| 675 | |
| 676 | * By now, the task's closid and rmid are set. If the task is current
|
| 677 | * on a CPU, the PQR_ASSOC MSR needs to be updated to make the resource
|
| 678 | * group go into effect. If the task is not current, the MSR will be
|
| 679 | * updated when the task is scheduled in.
|
| 680 |
|
| 681 | update_task_closid_rmid(tsk);
|
| 682 |
|
| 683 | return 0;
|
| 684 | }
|
| 685 |
|
| 686 | static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
|
| 687 | {
|
| 688 | return (r->type == RDTCTRL_GROUP &&
|
| 689 | resctrl_arch_match_closid(t, r->closid));
|
| 690 | }
|
| 691 |
|
| 692 | static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
|
| 693 | {
|
| 694 | return (resctrl_arch_mon_capable() && (r->type == RDTMON_GROUP) &&
|
| 695 | resctrl_arch_match_rmid(t, r->mon.parent->closid,
|
| 696 | r->mon.rmid));
|
| 697 | }
|
| 698 |
|
| 699 | |
| 700 | * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group
|
| 701 | * @r: Resource group
|
| 702 | *
|
| 703 | * Return: 1 if tasks have been assigned to @r, 0 otherwise
|
| 704 |
|
| 705 | int rdtgroup_tasks_assigned(struct rdtgroup *r)
|
| 706 | {
|
| 707 | struct task_struct *p, *t;
|
| 708 | int ret = 0;
|
| 709 |
|
| 710 | lockdep_assert_held(&rdtgroup_mutex);
|
| 711 |
|
| 712 | rcu_read_lock();
|
| 713 | for_each_process_thread(p, t) {
|
| 714 | if (is_closid_match(t, r) || is_rmid_match(t, r)) {
|
| 715 | ret = 1;
|
| 716 | break;
|
| 717 | }
|
| 718 | }
|
| 719 | rcu_read_unlock();
|
| 720 |
|
| 721 | return ret;
|
| 722 | }
|
| 723 |
|
| 724 | static int rdtgroup_task_write_permission(struct task_struct *task,
|
| 725 | struct kernfs_open_file *of)
|
| 726 | {
|
| 727 | const struct cred *tcred = get_task_cred(task);
|
| 728 | const struct cred *cred = current_cred();
|
| 729 | int ret = 0;
|
| 730 |
|
| 731 | |
| 732 | * Even if we're attaching all tasks in the thread group, we only
|
| 733 | * need to check permissions on one of them.
|
| 734 |
|
| 735 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
|
| 736 | !uid_eq(cred->euid, tcred->uid) &&
|
| 737 | !uid_eq(cred->euid, tcred->suid)) {
|
| 738 | rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
|
| 739 | ret = -EPERM;
|
| 740 | }
|
| 741 |
|
| 742 | put_cred(tcred);
|
| 743 | return ret;
|
| 744 | }
|
| 745 |
|
| 746 | static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
|
| 747 | struct kernfs_open_file *of)
|
| 748 | {
|
| 749 | struct task_struct *tsk;
|
| 750 | int ret;
|
| 751 |
|
| 752 | rcu_read_lock();
|
| 753 | if (pid) {
|
| 754 | tsk = find_task_by_vpid(pid);
|
| 755 | if (!tsk) {
|
| 756 | rcu_read_unlock();
|
| 757 | rdt_last_cmd_printf("No task %d\n", pid);
|
| 758 | return -ESRCH;
|
| 759 | }
|
| 760 | } else {
|
| 761 | tsk = current;
|
| 762 | }
|
| 763 |
|
| 764 | get_task_struct(tsk);
|
| 765 | rcu_read_unlock();
|
| 766 |
|
| 767 | ret = rdtgroup_task_write_permission(tsk, of);
|
| 768 | if (!ret)
|
| 769 | ret = __rdtgroup_move_task(tsk, rdtgrp);
|
| 770 |
|
| 771 | put_task_struct(tsk);
|
| 772 | return ret;
|
| 773 | }
|
| 774 |
|
| 775 | static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
|
| 776 | char *buf, size_t nbytes, loff_t off)
|
| 777 | {
|
| 778 | struct rdtgroup *rdtgrp;
|
| 779 | char *pid_str;
|
| 780 | int ret = 0;
|
| 781 | pid_t pid;
|
| 782 |
|
| 783 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 784 | if (!rdtgrp) {
|
| 785 | rdtgroup_kn_unlock(of->kn);
|
| 786 | return -ENOENT;
|
| 787 | }
|
| 788 |
|
| 789 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
|
| 790 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
|
| 791 | ret = -EINVAL;
|
| 792 | rdt_last_cmd_puts("Pseudo-locking in progress\n");
|
| 793 | goto unlock;
|
| 794 | }
|
| 795 |
|
| 796 | while (buf && buf[0] != '\0' && buf[0] != '\n') {
|
| 797 | pid_str = strim(strsep(&buf, ","));
|
| 798 |
|
| 799 | if (kstrtoint(pid_str, 0, &pid)) {
|
| 800 | rdt_last_cmd_printf("Task list parsing error pid %s\n", pid_str);
|
| 801 | ret = -EINVAL;
|
| 802 | break;
|
| 803 | }
|
| 804 |
|
| 805 | if (pid < 0) {
|
| 806 | rdt_last_cmd_printf("Invalid pid %d\n", pid);
|
| 807 | ret = -EINVAL;
|
| 808 | break;
|
| 809 | }
|
| 810 |
|
| 811 | ret = rdtgroup_move_task(pid, rdtgrp, of);
|
| 812 | if (ret) {
|
| 813 | rdt_last_cmd_printf("Error while processing task %d\n", pid);
|
| 814 | break;
|
| 815 | }
|
| 816 | }
|
| 817 |
|
| 818 | unlock:
|
| 819 | rdtgroup_kn_unlock(of->kn);
|
| 820 |
|
| 821 | return ret ?: nbytes;
|
| 822 | }
|
| 823 |
|
| 824 | static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
|
| 825 | {
|
| 826 | struct task_struct *p, *t;
|
| 827 | pid_t pid;
|
| 828 |
|
| 829 | rcu_read_lock();
|
| 830 | for_each_process_thread(p, t) {
|
| 831 | if (is_closid_match(t, r) || is_rmid_match(t, r)) {
|
| 832 | pid = task_pid_vnr(t);
|
| 833 | if (pid)
|
| 834 | seq_printf(s, "%d\n", pid);
|
| 835 | }
|
| 836 | }
|
| 837 | rcu_read_unlock();
|
| 838 | }
|
| 839 |
|
| 840 | static int rdtgroup_tasks_show(struct kernfs_open_file *of,
|
| 841 | struct seq_file *s, void *v)
|
| 842 | {
|
| 843 | struct rdtgroup *rdtgrp;
|
| 844 | int ret = 0;
|
| 845 |
|
| 846 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 847 | if (rdtgrp)
|
| 848 | show_rdt_tasks(rdtgrp, s);
|
| 849 | else
|
| 850 | ret = -ENOENT;
|
| 851 | rdtgroup_kn_unlock(of->kn);
|
| 852 |
|
| 853 | return ret;
|
| 854 | }
|
| 855 |
|
| 856 | static int rdtgroup_closid_show(struct kernfs_open_file *of,
|
| 857 | struct seq_file *s, void *v)
|
| 858 | {
|
| 859 | struct rdtgroup *rdtgrp;
|
| 860 | int ret = 0;
|
| 861 |
|
| 862 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 863 | if (rdtgrp)
|
| 864 | seq_printf(s, "%u\n", rdtgrp->closid);
|
| 865 | else
|
| 866 | ret = -ENOENT;
|
| 867 | rdtgroup_kn_unlock(of->kn);
|
| 868 |
|
| 869 | return ret;
|
| 870 | }
|
| 871 |
|
| 872 | static int rdtgroup_rmid_show(struct kernfs_open_file *of,
|
| 873 | struct seq_file *s, void *v)
|
| 874 | {
|
| 875 | struct rdtgroup *rdtgrp;
|
| 876 | int ret = 0;
|
| 877 |
|
| 878 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 879 | if (rdtgrp)
|
| 880 | seq_printf(s, "%u\n", rdtgrp->mon.rmid);
|
| 881 | else
|
| 882 | ret = -ENOENT;
|
| 883 | rdtgroup_kn_unlock(of->kn);
|
| 884 |
|
| 885 | return ret;
|
| 886 | }
|
| 887 |
|
| 888 | #ifdef CONFIG_PROC_CPU_RESCTRL
|
| 889 | |
| 890 | * A task can only be part of one resctrl control group and of one monitor
|
| 891 | * group which is associated to that control group.
|
| 892 | *
|
| 893 | * 1) res:
|
| 894 | * mon:
|
| 895 | *
|
| 896 | * resctrl is not available.
|
| 897 | *
|
| 898 | * 2) res:/
|
| 899 | * mon:
|
| 900 | *
|
| 901 | * Task is part of the root resctrl control group, and it is not associated
|
| 902 | * to any monitor group.
|
| 903 | *
|
| 904 | * 3) res:/
|
| 905 | * mon:mon0
|
| 906 | *
|
| 907 | * Task is part of the root resctrl control group and monitor group mon0.
|
| 908 | *
|
| 909 | * 4) res:group0
|
| 910 | * mon:
|
| 911 | *
|
| 912 | * Task is part of resctrl control group group0, and it is not associated
|
| 913 | * to any monitor group.
|
| 914 | *
|
| 915 | * 5) res:group0
|
| 916 | * mon:mon1
|
| 917 | *
|
| 918 | * Task is part of resctrl control group group0 and monitor group mon1.
|
| 919 |
|
| 920 | int proc_resctrl_show(struct seq_file *s, struct pid_namespace *ns,
|
| 921 | struct pid *pid, struct task_struct *tsk)
|
| 922 | {
|
| 923 | struct rdtgroup *rdtg;
|
| 924 | int ret = 0;
|
| 925 |
|
| 926 | mutex_lock(&rdtgroup_mutex);
|
| 927 |
|
| 928 |
|
| 929 | if (!resctrl_mounted) {
|
| 930 | seq_puts(s, "res:\nmon:\n");
|
| 931 | goto unlock;
|
| 932 | }
|
| 933 |
|
| 934 | list_for_each_entry(rdtg, &rdt_all_groups, rdtgroup_list) {
|
| 935 | struct rdtgroup *crg;
|
| 936 |
|
| 937 | |
| 938 | * Task information is only relevant for shareable
|
| 939 | * and exclusive groups.
|
| 940 |
|
| 941 | if (rdtg->mode != RDT_MODE_SHAREABLE &&
|
| 942 | rdtg->mode != RDT_MODE_EXCLUSIVE)
|
| 943 | continue;
|
| 944 |
|
| 945 | if (!resctrl_arch_match_closid(tsk, rdtg->closid))
|
| 946 | continue;
|
| 947 |
|
| 948 | seq_printf(s, "res:%s%s\n", (rdtg == &rdtgroup_default) ? "/" : "",
|
| 949 | rdt_kn_name(rdtg->kn));
|
| 950 | seq_puts(s, "mon:");
|
| 951 | list_for_each_entry(crg, &rdtg->mon.crdtgrp_list,
|
| 952 | mon.crdtgrp_list) {
|
| 953 | if (!resctrl_arch_match_rmid(tsk, crg->mon.parent->closid,
|
| 954 | crg->mon.rmid))
|
| 955 | continue;
|
| 956 | seq_printf(s, "%s", rdt_kn_name(crg->kn));
|
| 957 | break;
|
| 958 | }
|
| 959 | seq_putc(s, '\n');
|
| 960 | goto unlock;
|
| 961 | }
|
| 962 | |
| 963 | * The above search should succeed. Otherwise return
|
| 964 | * with an error.
|
| 965 |
|
| 966 | ret = -ENOENT;
|
| 967 | unlock:
|
| 968 | mutex_unlock(&rdtgroup_mutex);
|
| 969 |
|
| 970 | return ret;
|
| 971 | }
|
| 972 | #endif
|
| 973 |
|
| 974 | static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
|
| 975 | struct seq_file *seq, void *v)
|
| 976 | {
|
| 977 | int len;
|
| 978 |
|
| 979 | if (!info_kn_lock(of->kn))
|
| 980 | return -ENOENT;
|
| 981 | len = seq_buf_used(&last_cmd_status);
|
| 982 | if (len) {
|
| 983 | seq_printf(seq, "%.*s", len, last_cmd_status_buf);
|
| 984 | if (seq_buf_has_overflowed(&last_cmd_status))
|
| 985 | seq_puts(seq, "[truncated]\n");
|
| 986 | } else {
|
| 987 | seq_puts(seq, "ok\n");
|
| 988 | }
|
| 989 | info_kn_unlock(of->kn);
|
| 990 | return 0;
|
| 991 | }
|
| 992 |
|
| 993 | void *rdt_kn_parent_priv(struct kernfs_node *kn)
|
| 994 | {
|
| 995 | |
| 996 | * The parent pointer is only valid within RCU section since it can be
|
| 997 | * replaced.
|
| 998 |
|
| 999 | guard(rcu)();
|
| 1000 | return rcu_dereference(kn->__parent)->priv;
|
| 1001 | }
|
| 1002 |
|
| 1003 | static int rdt_num_closids_show(struct kernfs_open_file *of,
|
| 1004 | struct seq_file *seq, void *v)
|
| 1005 | {
|
| 1006 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1007 |
|
| 1008 | if (!info_kn_lock(of->kn))
|
| 1009 | return -ENOENT;
|
| 1010 | seq_printf(seq, "%u\n", s->num_closid);
|
| 1011 | info_kn_unlock(of->kn);
|
| 1012 |
|
| 1013 | return 0;
|
| 1014 | }
|
| 1015 |
|
| 1016 | static int rdt_default_ctrl_show(struct kernfs_open_file *of,
|
| 1017 | struct seq_file *seq, void *v)
|
| 1018 | {
|
| 1019 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1020 | struct rdt_resource *r;
|
| 1021 |
|
| 1022 | if (!info_kn_lock(of->kn))
|
| 1023 | return -ENOENT;
|
| 1024 | r = s->res;
|
| 1025 | seq_printf(seq, "%x\n", resctrl_get_default_ctrl(r));
|
| 1026 | info_kn_unlock(of->kn);
|
| 1027 |
|
| 1028 | return 0;
|
| 1029 | }
|
| 1030 |
|
| 1031 | static int rdt_min_cbm_bits_show(struct kernfs_open_file *of,
|
| 1032 | struct seq_file *seq, void *v)
|
| 1033 | {
|
| 1034 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1035 | struct rdt_resource *r;
|
| 1036 |
|
| 1037 | if (!info_kn_lock(of->kn))
|
| 1038 | return -ENOENT;
|
| 1039 | r = s->res;
|
| 1040 | seq_printf(seq, "%u\n", r->cache.min_cbm_bits);
|
| 1041 | info_kn_unlock(of->kn);
|
| 1042 |
|
| 1043 | return 0;
|
| 1044 | }
|
| 1045 |
|
| 1046 | static int rdt_shareable_bits_show(struct kernfs_open_file *of,
|
| 1047 | struct seq_file *seq, void *v)
|
| 1048 | {
|
| 1049 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1050 | struct rdt_resource *r;
|
| 1051 |
|
| 1052 | if (!info_kn_lock(of->kn))
|
| 1053 | return -ENOENT;
|
| 1054 | r = s->res;
|
| 1055 | seq_printf(seq, "%x\n", r->cache.shareable_bits);
|
| 1056 | info_kn_unlock(of->kn);
|
| 1057 |
|
| 1058 | return 0;
|
| 1059 | }
|
| 1060 |
|
| 1061 | |
| 1062 | * rdt_bit_usage_show - Display current usage of resources
|
| 1063 | *
|
| 1064 | * A domain is a shared resource that can now be allocated differently. Here
|
| 1065 | * we display the current regions of the domain as an annotated bitmask.
|
| 1066 | * For each domain of this resource its allocation bitmask
|
| 1067 | * is annotated as below to indicate the current usage of the corresponding bit:
|
| 1068 | * 0 - currently unused
|
| 1069 | * X - currently available for sharing and used by software and hardware
|
| 1070 | * H - currently used by hardware only but available for software use
|
| 1071 | * S - currently used and shareable by software only
|
| 1072 | * E - currently used exclusively by one resource group
|
| 1073 | * P - currently pseudo-locked by one resource group
|
| 1074 |
|
| 1075 | static int rdt_bit_usage_show(struct kernfs_open_file *of,
|
| 1076 | struct seq_file *seq, void *v)
|
| 1077 | {
|
| 1078 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1079 | |
| 1080 | * Use unsigned long even though only 32 bits are used to ensure
|
| 1081 | * test_bit() is used safely.
|
| 1082 |
|
| 1083 | unsigned long sw_shareable = 0, hw_shareable = 0;
|
| 1084 | unsigned long exclusive = 0, pseudo_locked = 0;
|
| 1085 | struct rdt_ctrl_domain *dom;
|
| 1086 | int i, hwb, swb, excl, psl;
|
| 1087 | struct rdt_resource *r;
|
| 1088 | enum rdtgrp_mode mode;
|
| 1089 | bool sep = false;
|
| 1090 | u32 ctrl_val;
|
| 1091 |
|
| 1092 | if (!info_kn_lock(of->kn))
|
| 1093 | return -ENOENT;
|
| 1094 | r = s->res;
|
| 1095 | list_for_each_entry_rcu(dom, &r->ctrl_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 1096 | if (sep)
|
| 1097 | seq_putc(seq, ';');
|
| 1098 | hw_shareable = r->cache.shareable_bits;
|
| 1099 | sw_shareable = 0;
|
| 1100 | exclusive = 0;
|
| 1101 | seq_printf(seq, "%d=", dom->hdr.id);
|
| 1102 | for (i = 0; i < closids_supported(); i++) {
|
| 1103 | if (!closid_allocated(i) ||
|
| 1104 | (resctrl_arch_get_io_alloc_enabled(r) &&
|
| 1105 | i == resctrl_io_alloc_closid(r)))
|
| 1106 | continue;
|
| 1107 | ctrl_val = resctrl_arch_get_config(r, dom, i,
|
| 1108 | s->conf_type);
|
| 1109 | mode = rdtgroup_mode_by_closid(i);
|
| 1110 | switch (mode) {
|
| 1111 | case RDT_MODE_SHAREABLE:
|
| 1112 | sw_shareable |= ctrl_val;
|
| 1113 | break;
|
| 1114 | case RDT_MODE_EXCLUSIVE:
|
| 1115 | exclusive |= ctrl_val;
|
| 1116 | break;
|
| 1117 | case RDT_MODE_PSEUDO_LOCKSETUP:
|
| 1118 | |
| 1119 | * RDT_MODE_PSEUDO_LOCKSETUP is possible
|
| 1120 | * here but not included since the CBM
|
| 1121 | * associated with this CLOSID in this mode
|
| 1122 | * is not initialized and no task or cpu can be
|
| 1123 | * assigned this CLOSID.
|
| 1124 |
|
| 1125 | break;
|
| 1126 | case RDT_MODE_PSEUDO_LOCKED:
|
| 1127 | case RDT_NUM_MODES:
|
| 1128 | WARN(1,
|
| 1129 | "invalid mode for closid %d\n", i);
|
| 1130 | break;
|
| 1131 | }
|
| 1132 | }
|
| 1133 |
|
| 1134 | |
| 1135 | * When the "io_alloc" feature is enabled, a portion of the cache
|
| 1136 | * is configured for shared use between hardware and software.
|
| 1137 | * Also, when CDP is enabled the CBMs of CDP_CODE and CDP_DATA
|
| 1138 | * resources are kept in sync. So, the CBMs for "io_alloc" can
|
| 1139 | * be accessed through either resource.
|
| 1140 |
|
| 1141 | if (resctrl_arch_get_io_alloc_enabled(r)) {
|
| 1142 | ctrl_val = resctrl_arch_get_config(r, dom,
|
| 1143 | resctrl_io_alloc_closid(r),
|
| 1144 | s->conf_type);
|
| 1145 | hw_shareable |= ctrl_val;
|
| 1146 | }
|
| 1147 |
|
| 1148 | for (i = r->cache.cbm_len - 1; i >= 0; i--) {
|
| 1149 | pseudo_locked = dom->plr ? dom->plr->cbm : 0;
|
| 1150 | hwb = test_bit(i, &hw_shareable);
|
| 1151 | swb = test_bit(i, &sw_shareable);
|
| 1152 | excl = test_bit(i, &exclusive);
|
| 1153 | psl = test_bit(i, &pseudo_locked);
|
| 1154 | if (hwb && swb)
|
| 1155 | seq_putc(seq, 'X');
|
| 1156 | else if (hwb && !swb)
|
| 1157 | seq_putc(seq, 'H');
|
| 1158 | else if (!hwb && swb)
|
| 1159 | seq_putc(seq, 'S');
|
| 1160 | else if (excl)
|
| 1161 | seq_putc(seq, 'E');
|
| 1162 | else if (psl)
|
| 1163 | seq_putc(seq, 'P');
|
| 1164 | else
|
| 1165 | seq_putc(seq, '0');
|
| 1166 | }
|
| 1167 | sep = true;
|
| 1168 | }
|
| 1169 | seq_putc(seq, '\n');
|
| 1170 | info_kn_unlock(of->kn);
|
| 1171 | return 0;
|
| 1172 | }
|
| 1173 |
|
| 1174 | static int rdt_min_bw_show(struct kernfs_open_file *of,
|
| 1175 | struct seq_file *seq, void *v)
|
| 1176 | {
|
| 1177 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1178 | struct rdt_resource *r;
|
| 1179 |
|
| 1180 | if (!info_kn_lock(of->kn))
|
| 1181 | return -ENOENT;
|
| 1182 | r = s->res;
|
| 1183 | seq_printf(seq, "%u\n", r->membw.min_bw);
|
| 1184 | info_kn_unlock(of->kn);
|
| 1185 |
|
| 1186 | return 0;
|
| 1187 | }
|
| 1188 |
|
| 1189 | static int rdt_num_rmids_show(struct kernfs_open_file *of,
|
| 1190 | struct seq_file *seq, void *v)
|
| 1191 | {
|
| 1192 | struct rdt_resource *r = rdt_kn_parent_priv(of->kn);
|
| 1193 |
|
| 1194 | if (!info_kn_lock(of->kn))
|
| 1195 | return -ENOENT;
|
| 1196 | seq_printf(seq, "%u\n", r->mon.num_rmid);
|
| 1197 |
|
| 1198 | info_kn_unlock(of->kn);
|
| 1199 |
|
| 1200 | return 0;
|
| 1201 | }
|
| 1202 |
|
| 1203 | static int rdt_mon_features_show(struct kernfs_open_file *of,
|
| 1204 | struct seq_file *seq, void *v)
|
| 1205 | {
|
| 1206 | struct rdt_resource *r = rdt_kn_parent_priv(of->kn);
|
| 1207 | struct mon_evt *mevt;
|
| 1208 |
|
| 1209 | if (!info_kn_lock(of->kn))
|
| 1210 | return -ENOENT;
|
| 1211 | for_each_mon_event(mevt) {
|
| 1212 | if (mevt->rid != r->rid || !mevt->enabled)
|
| 1213 | continue;
|
| 1214 | seq_printf(seq, "%s\n", mevt->name);
|
| 1215 | if (mevt->configurable &&
|
| 1216 | !resctrl_arch_mbm_cntr_assign_enabled(r))
|
| 1217 | seq_printf(seq, "%s_config\n", mevt->name);
|
| 1218 | }
|
| 1219 |
|
| 1220 | info_kn_unlock(of->kn);
|
| 1221 |
|
| 1222 | return 0;
|
| 1223 | }
|
| 1224 |
|
| 1225 | static int rdt_bw_gran_show(struct kernfs_open_file *of,
|
| 1226 | struct seq_file *seq, void *v)
|
| 1227 | {
|
| 1228 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1229 | struct rdt_resource *r;
|
| 1230 |
|
| 1231 | if (!info_kn_lock(of->kn))
|
| 1232 | return -ENOENT;
|
| 1233 | r = s->res;
|
| 1234 | seq_printf(seq, "%u\n", r->membw.bw_gran);
|
| 1235 | info_kn_unlock(of->kn);
|
| 1236 |
|
| 1237 | return 0;
|
| 1238 | }
|
| 1239 |
|
| 1240 | static int rdt_delay_linear_show(struct kernfs_open_file *of,
|
| 1241 | struct seq_file *seq, void *v)
|
| 1242 | {
|
| 1243 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1244 | struct rdt_resource *r;
|
| 1245 |
|
| 1246 | if (!info_kn_lock(of->kn))
|
| 1247 | return -ENOENT;
|
| 1248 | r = s->res;
|
| 1249 | seq_printf(seq, "%u\n", r->membw.delay_linear);
|
| 1250 | info_kn_unlock(of->kn);
|
| 1251 |
|
| 1252 | return 0;
|
| 1253 | }
|
| 1254 |
|
| 1255 | static int max_threshold_occ_show(struct kernfs_open_file *of,
|
| 1256 | struct seq_file *seq, void *v)
|
| 1257 | {
|
| 1258 | if (!info_kn_lock(of->kn))
|
| 1259 | return -ENOENT;
|
| 1260 | seq_printf(seq, "%u\n", resctrl_rmid_realloc_threshold);
|
| 1261 | info_kn_unlock(of->kn);
|
| 1262 |
|
| 1263 | return 0;
|
| 1264 | }
|
| 1265 |
|
| 1266 | static int rdt_thread_throttle_mode_show(struct kernfs_open_file *of,
|
| 1267 | struct seq_file *seq, void *v)
|
| 1268 | {
|
| 1269 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1270 | struct rdt_resource *r;
|
| 1271 |
|
| 1272 | if (!info_kn_lock(of->kn))
|
| 1273 | return -ENOENT;
|
| 1274 |
|
| 1275 | r = s->res;
|
| 1276 | switch (r->membw.throttle_mode) {
|
| 1277 | case THREAD_THROTTLE_PER_THREAD:
|
| 1278 | seq_puts(seq, "per-thread\n");
|
| 1279 | break;
|
| 1280 | case THREAD_THROTTLE_MAX:
|
| 1281 | seq_puts(seq, "max\n");
|
| 1282 | break;
|
| 1283 | case THREAD_THROTTLE_UNDEFINED:
|
| 1284 | seq_puts(seq, "undefined\n");
|
| 1285 | break;
|
| 1286 | default:
|
| 1287 | WARN_ON_ONCE(1);
|
| 1288 | break;
|
| 1289 | }
|
| 1290 |
|
| 1291 | info_kn_unlock(of->kn);
|
| 1292 | return 0;
|
| 1293 | }
|
| 1294 |
|
| 1295 | static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
|
| 1296 | char *buf, size_t nbytes, loff_t off)
|
| 1297 | {
|
| 1298 | unsigned int bytes;
|
| 1299 | int ret;
|
| 1300 |
|
| 1301 | if (!info_kn_lock(of->kn))
|
| 1302 | return -ENOENT;
|
| 1303 |
|
| 1304 | rdt_last_cmd_clear();
|
| 1305 |
|
| 1306 | ret = kstrtouint(buf, 0, &bytes);
|
| 1307 | if (ret) {
|
| 1308 | rdt_last_cmd_puts("max_threshold_occupancy: Invalid input\n");
|
| 1309 | goto out_unlock;
|
| 1310 | }
|
| 1311 |
|
| 1312 | if (bytes > resctrl_rmid_realloc_limit) {
|
| 1313 | rdt_last_cmd_printf("max_threshold_occupancy: Exceeds limit (before adjustment) of %u bytes\n",
|
| 1314 | resctrl_rmid_realloc_limit);
|
| 1315 | ret = -EINVAL;
|
| 1316 | goto out_unlock;
|
| 1317 | }
|
| 1318 |
|
| 1319 | resctrl_rmid_realloc_threshold = resctrl_arch_round_mon_val(bytes);
|
| 1320 |
|
| 1321 | out_unlock:
|
| 1322 | info_kn_unlock(of->kn);
|
| 1323 |
|
| 1324 | return ret ?: nbytes;
|
| 1325 | }
|
| 1326 |
|
| 1327 | |
| 1328 | * rdtgroup_mode_show - Display mode of this resource group
|
| 1329 |
|
| 1330 | static int rdtgroup_mode_show(struct kernfs_open_file *of,
|
| 1331 | struct seq_file *s, void *v)
|
| 1332 | {
|
| 1333 | struct rdtgroup *rdtgrp;
|
| 1334 |
|
| 1335 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 1336 | if (!rdtgrp) {
|
| 1337 | rdtgroup_kn_unlock(of->kn);
|
| 1338 | return -ENOENT;
|
| 1339 | }
|
| 1340 |
|
| 1341 | seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode));
|
| 1342 |
|
| 1343 | rdtgroup_kn_unlock(of->kn);
|
| 1344 | return 0;
|
| 1345 | }
|
| 1346 |
|
| 1347 | enum resctrl_conf_type resctrl_peer_type(enum resctrl_conf_type my_type)
|
| 1348 | {
|
| 1349 | switch (my_type) {
|
| 1350 | case CDP_CODE:
|
| 1351 | return CDP_DATA;
|
| 1352 | case CDP_DATA:
|
| 1353 | return CDP_CODE;
|
| 1354 | default:
|
| 1355 | case CDP_NONE:
|
| 1356 | return CDP_NONE;
|
| 1357 | }
|
| 1358 | }
|
| 1359 |
|
| 1360 | static int rdt_has_sparse_bitmasks_show(struct kernfs_open_file *of,
|
| 1361 | struct seq_file *seq, void *v)
|
| 1362 | {
|
| 1363 | struct resctrl_schema *s = rdt_kn_parent_priv(of->kn);
|
| 1364 | struct rdt_resource *r;
|
| 1365 |
|
| 1366 | if (!info_kn_lock(of->kn))
|
| 1367 | return -ENOENT;
|
| 1368 | r = s->res;
|
| 1369 | seq_printf(seq, "%u\n", r->cache.arch_has_sparse_bitmasks);
|
| 1370 |
|
| 1371 | info_kn_unlock(of->kn);
|
| 1372 |
|
| 1373 | return 0;
|
| 1374 | }
|
| 1375 |
|
| 1376 | |
| 1377 | * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other
|
| 1378 | * @r: Resource to which domain instance @d belongs.
|
| 1379 | * @d: The domain instance for which @closid is being tested.
|
| 1380 | * @cbm: Capacity bitmask being tested.
|
| 1381 | * @closid: Intended closid for @cbm.
|
| 1382 | * @type: CDP type of @r.
|
| 1383 | * @exclusive: Only check if overlaps with exclusive resource groups
|
| 1384 | *
|
| 1385 | * Checks if provided @cbm intended to be used for @closid on domain
|
| 1386 | * @d overlaps with any other closids or other hardware usage associated
|
| 1387 | * with this domain. If @exclusive is true then only overlaps with
|
| 1388 | * resource groups in exclusive mode will be considered. If @exclusive
|
| 1389 | * is false then overlaps with any resource group or hardware entities
|
| 1390 | * will be considered.
|
| 1391 | *
|
| 1392 | * @cbm is unsigned long, even if only 32 bits are used, to make the
|
| 1393 | * bitmap functions work correctly.
|
| 1394 | *
|
| 1395 | * Return: false if CBM does not overlap, true if it does.
|
| 1396 |
|
| 1397 | static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_ctrl_domain *d,
|
| 1398 | unsigned long cbm, int closid,
|
| 1399 | enum resctrl_conf_type type, bool exclusive)
|
| 1400 | {
|
| 1401 | enum rdtgrp_mode mode;
|
| 1402 | unsigned long ctrl_b;
|
| 1403 | int i;
|
| 1404 |
|
| 1405 |
|
| 1406 | if (!exclusive) {
|
| 1407 | ctrl_b = r->cache.shareable_bits;
|
| 1408 | if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len))
|
| 1409 | return true;
|
| 1410 | }
|
| 1411 |
|
| 1412 |
|
| 1413 | for (i = 0; i < closids_supported(); i++) {
|
| 1414 | ctrl_b = resctrl_arch_get_config(r, d, i, type);
|
| 1415 | mode = rdtgroup_mode_by_closid(i);
|
| 1416 | if (closid_allocated(i) && i != closid &&
|
| 1417 | mode != RDT_MODE_PSEUDO_LOCKSETUP) {
|
| 1418 | if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) {
|
| 1419 | if (exclusive) {
|
| 1420 | if (mode == RDT_MODE_EXCLUSIVE)
|
| 1421 | return true;
|
| 1422 | continue;
|
| 1423 | }
|
| 1424 | return true;
|
| 1425 | }
|
| 1426 | }
|
| 1427 | }
|
| 1428 |
|
| 1429 | return false;
|
| 1430 | }
|
| 1431 |
|
| 1432 | |
| 1433 | * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware
|
| 1434 | * @s: Schema for the resource to which domain instance @d belongs.
|
| 1435 | * @d: The domain instance for which @closid is being tested.
|
| 1436 | * @cbm: Capacity bitmask being tested.
|
| 1437 | * @closid: Intended closid for @cbm.
|
| 1438 | * @exclusive: Only check if overlaps with exclusive resource groups
|
| 1439 | *
|
| 1440 | * Resources that can be allocated using a CBM can use the CBM to control
|
| 1441 | * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test
|
| 1442 | * for overlap. Overlap test is not limited to the specific resource for
|
| 1443 | * which the CBM is intended though - when dealing with CDP resources that
|
| 1444 | * share the underlying hardware the overlap check should be performed on
|
| 1445 | * the CDP resource sharing the hardware also.
|
| 1446 | *
|
| 1447 | * Refer to description of __rdtgroup_cbm_overlaps() for the details of the
|
| 1448 | * overlap test.
|
| 1449 | *
|
| 1450 | * Return: true if CBM overlap detected, false if there is no overlap
|
| 1451 |
|
| 1452 | bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_ctrl_domain *d,
|
| 1453 | unsigned long cbm, int closid, bool exclusive)
|
| 1454 | {
|
| 1455 | enum resctrl_conf_type peer_type = resctrl_peer_type(s->conf_type);
|
| 1456 | struct rdt_resource *r = s->res;
|
| 1457 |
|
| 1458 | if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, s->conf_type,
|
| 1459 | exclusive))
|
| 1460 | return true;
|
| 1461 |
|
| 1462 | if (!resctrl_arch_get_cdp_enabled(r->rid))
|
| 1463 | return false;
|
| 1464 | return __rdtgroup_cbm_overlaps(r, d, cbm, closid, peer_type, exclusive);
|
| 1465 | }
|
| 1466 |
|
| 1467 | |
| 1468 | * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive
|
| 1469 | * @rdtgrp: Resource group identified through its closid.
|
| 1470 | *
|
| 1471 | * An exclusive resource group implies that there should be no sharing of
|
| 1472 | * its allocated resources. At the time this group is considered to be
|
| 1473 | * exclusive this test can determine if its current schemata supports this
|
| 1474 | * setting by testing for overlap with all other resource groups.
|
| 1475 | *
|
| 1476 | * Return: true if resource group can be exclusive, false if there is overlap
|
| 1477 | * with allocations of other resource groups and thus this resource group
|
| 1478 | * cannot be exclusive.
|
| 1479 |
|
| 1480 | static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp)
|
| 1481 | {
|
| 1482 | int closid = rdtgrp->closid;
|
| 1483 | struct rdt_ctrl_domain *d;
|
| 1484 | struct resctrl_schema *s;
|
| 1485 | struct rdt_resource *r;
|
| 1486 | bool has_cache = false;
|
| 1487 | u32 ctrl;
|
| 1488 |
|
| 1489 |
|
| 1490 | lockdep_assert_cpus_held();
|
| 1491 |
|
| 1492 | list_for_each_entry(s, &resctrl_schema_all, list) {
|
| 1493 | r = s->res;
|
| 1494 | if (r->rid == RDT_RESOURCE_MBA || r->rid == RDT_RESOURCE_SMBA)
|
| 1495 | continue;
|
| 1496 | has_cache = true;
|
| 1497 | list_for_each_entry_rcu(d, &r->ctrl_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 1498 | ctrl = resctrl_arch_get_config(r, d, closid,
|
| 1499 | s->conf_type);
|
| 1500 | if (rdtgroup_cbm_overlaps(s, d, ctrl, closid, false)) {
|
| 1501 | rdt_last_cmd_puts("Schemata overlaps\n");
|
| 1502 | return false;
|
| 1503 | }
|
| 1504 | }
|
| 1505 | }
|
| 1506 |
|
| 1507 | if (!has_cache) {
|
| 1508 | rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n");
|
| 1509 | return false;
|
| 1510 | }
|
| 1511 |
|
| 1512 | return true;
|
| 1513 | }
|
| 1514 |
|
| 1515 | |
| 1516 | * rdtgroup_mode_write - Modify the resource group's mode
|
| 1517 |
|
| 1518 | static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of,
|
| 1519 | char *buf, size_t nbytes, loff_t off)
|
| 1520 | {
|
| 1521 | struct rdtgroup *rdtgrp;
|
| 1522 | enum rdtgrp_mode mode;
|
| 1523 | int ret = 0;
|
| 1524 |
|
| 1525 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 1526 | if (!rdtgrp) {
|
| 1527 | rdtgroup_kn_unlock(of->kn);
|
| 1528 | return -ENOENT;
|
| 1529 | }
|
| 1530 |
|
| 1531 |
|
| 1532 | if (nbytes == 0 || buf[nbytes - 1] != '\n') {
|
| 1533 | rdt_last_cmd_puts("mode: Invalid input\n");
|
| 1534 | ret = -EINVAL;
|
| 1535 | goto out;
|
| 1536 | }
|
| 1537 |
|
| 1538 | buf[nbytes - 1] = '\0';
|
| 1539 |
|
| 1540 | mode = rdtgrp->mode;
|
| 1541 |
|
| 1542 | if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) ||
|
| 1543 | (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) ||
|
| 1544 | (!strcmp(buf, "pseudo-locksetup") &&
|
| 1545 | mode == RDT_MODE_PSEUDO_LOCKSETUP) ||
|
| 1546 | (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED))
|
| 1547 | goto out;
|
| 1548 |
|
| 1549 | if (mode == RDT_MODE_PSEUDO_LOCKED) {
|
| 1550 | rdt_last_cmd_puts("Cannot change pseudo-locked group\n");
|
| 1551 | ret = -EINVAL;
|
| 1552 | goto out;
|
| 1553 | }
|
| 1554 |
|
| 1555 | if (!strcmp(buf, "shareable")) {
|
| 1556 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
|
| 1557 | ret = rdtgroup_locksetup_exit(rdtgrp);
|
| 1558 | if (ret)
|
| 1559 | goto out;
|
| 1560 | }
|
| 1561 | rdtgrp->mode = RDT_MODE_SHAREABLE;
|
| 1562 | } else if (!strcmp(buf, "exclusive")) {
|
| 1563 | if (!rdtgroup_mode_test_exclusive(rdtgrp)) {
|
| 1564 | ret = -EINVAL;
|
| 1565 | goto out;
|
| 1566 | }
|
| 1567 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
|
| 1568 | ret = rdtgroup_locksetup_exit(rdtgrp);
|
| 1569 | if (ret)
|
| 1570 | goto out;
|
| 1571 | }
|
| 1572 | rdtgrp->mode = RDT_MODE_EXCLUSIVE;
|
| 1573 | } else if (IS_ENABLED(CONFIG_RESCTRL_FS_PSEUDO_LOCK) &&
|
| 1574 | !strcmp(buf, "pseudo-locksetup")) {
|
| 1575 | ret = rdtgroup_locksetup_enter(rdtgrp);
|
| 1576 | if (ret)
|
| 1577 | goto out;
|
| 1578 | rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP;
|
| 1579 | } else {
|
| 1580 | rdt_last_cmd_puts("Unknown or unsupported mode\n");
|
| 1581 | ret = -EINVAL;
|
| 1582 | }
|
| 1583 |
|
| 1584 | out:
|
| 1585 | rdtgroup_kn_unlock(of->kn);
|
| 1586 | return ret ?: nbytes;
|
| 1587 | }
|
| 1588 |
|
| 1589 | |
| 1590 | * rdtgroup_cbm_to_size - Translate CBM to size in bytes
|
| 1591 | * @r: RDT resource to which @d belongs.
|
| 1592 | * @d: RDT domain instance.
|
| 1593 | * @cbm: bitmask for which the size should be computed.
|
| 1594 | *
|
| 1595 | * The bitmask provided associated with the RDT domain instance @d will be
|
| 1596 | * translated into how many bytes it represents. The size in bytes is
|
| 1597 | * computed by first dividing the total cache size by the CBM length to
|
| 1598 | * determine how many bytes each bit in the bitmask represents. The result
|
| 1599 | * is multiplied with the number of bits set in the bitmask.
|
| 1600 | *
|
| 1601 | * @cbm is unsigned long, even if only 32 bits are used to make the
|
| 1602 | * bitmap functions work correctly.
|
| 1603 | *
|
| 1604 | * Return: Size (in bytes) of cache portion represented by CBM, 0 on failure.
|
| 1605 |
|
| 1606 | unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r,
|
| 1607 | struct rdt_ctrl_domain *d, unsigned long cbm)
|
| 1608 | {
|
| 1609 | unsigned int size = 0;
|
| 1610 | struct cacheinfo *ci;
|
| 1611 | int num_b;
|
| 1612 |
|
| 1613 | if (WARN_ON_ONCE(r->ctrl_scope != RESCTRL_L2_CACHE && r->ctrl_scope != RESCTRL_L3_CACHE))
|
| 1614 | return size;
|
| 1615 |
|
| 1616 | num_b = bitmap_weight(&cbm, r->cache.cbm_len);
|
| 1617 | ci = get_cpu_cacheinfo_level(cpumask_any(&d->hdr.cpu_mask), r->ctrl_scope);
|
| 1618 | if (ci)
|
| 1619 | size = ci->size / r->cache.cbm_len * num_b;
|
| 1620 |
|
| 1621 | return size;
|
| 1622 | }
|
| 1623 |
|
| 1624 | bool is_mba_sc(struct rdt_resource *r)
|
| 1625 | {
|
| 1626 | if (!r)
|
| 1627 | r = resctrl_arch_get_resource(RDT_RESOURCE_MBA);
|
| 1628 |
|
| 1629 | |
| 1630 | * The software controller support is only applicable to MBA resource.
|
| 1631 | * Make sure to check for resource type.
|
| 1632 |
|
| 1633 | if (r->rid != RDT_RESOURCE_MBA)
|
| 1634 | return false;
|
| 1635 |
|
| 1636 | return r->membw.mba_sc;
|
| 1637 | }
|
| 1638 |
|
| 1639 | |
| 1640 | * rdtgroup_size_show - Display size in bytes of allocated regions
|
| 1641 | *
|
| 1642 | * The "size" file mirrors the layout of the "schemata" file, printing the
|
| 1643 | * size in bytes of each region instead of the capacity bitmask.
|
| 1644 |
|
| 1645 | static int rdtgroup_size_show(struct kernfs_open_file *of,
|
| 1646 | struct seq_file *s, void *v)
|
| 1647 | {
|
| 1648 | struct resctrl_schema *schema;
|
| 1649 | enum resctrl_conf_type type;
|
| 1650 | struct rdt_ctrl_domain *d;
|
| 1651 | struct rdtgroup *rdtgrp;
|
| 1652 | struct rdt_resource *r;
|
| 1653 | unsigned int size;
|
| 1654 | int ret = 0;
|
| 1655 | u32 closid;
|
| 1656 | bool sep;
|
| 1657 | u32 ctrl;
|
| 1658 |
|
| 1659 | rdtgrp = rdtgroup_kn_lock_live(of->kn);
|
| 1660 | if (!rdtgrp) {
|
| 1661 | rdtgroup_kn_unlock(of->kn);
|
| 1662 | return -ENOENT;
|
| 1663 | }
|
| 1664 |
|
| 1665 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
|
| 1666 | if (!rdtgrp->plr->d) {
|
| 1667 | rdt_last_cmd_puts("Cache domain offline\n");
|
| 1668 | ret = -ENODEV;
|
| 1669 | } else {
|
| 1670 | seq_printf(s, "%*s:", max_name_width,
|
| 1671 | rdtgrp->plr->s->name);
|
| 1672 | size = rdtgroup_cbm_to_size(rdtgrp->plr->s->res,
|
| 1673 | rdtgrp->plr->d,
|
| 1674 | rdtgrp->plr->cbm);
|
| 1675 | seq_printf(s, "%d=%u\n", rdtgrp->plr->d->hdr.id, size);
|
| 1676 | }
|
| 1677 | goto out;
|
| 1678 | }
|
| 1679 |
|
| 1680 | closid = rdtgrp->closid;
|
| 1681 |
|
| 1682 | list_for_each_entry(schema, &resctrl_schema_all, list) {
|
| 1683 | r = schema->res;
|
| 1684 | type = schema->conf_type;
|
| 1685 | sep = false;
|
| 1686 | seq_printf(s, "%*s:", max_name_width, schema->name);
|
| 1687 | list_for_each_entry_rcu(d, &r->ctrl_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 1688 | if (sep)
|
| 1689 | seq_putc(s, ';');
|
| 1690 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
|
| 1691 | size = 0;
|
| 1692 | } else {
|
| 1693 | if (is_mba_sc(r))
|
| 1694 | ctrl = d->mbps_val[closid];
|
| 1695 | else
|
| 1696 | ctrl = resctrl_arch_get_config(r, d,
|
| 1697 | closid,
|
| 1698 | type);
|
| 1699 | if (r->rid == RDT_RESOURCE_MBA ||
|
| 1700 | r->rid == RDT_RESOURCE_SMBA)
|
| 1701 | size = ctrl;
|
| 1702 | else
|
| 1703 | size = rdtgroup_cbm_to_size(r, d, ctrl);
|
| 1704 | }
|
| 1705 | seq_printf(s, "%d=%u", d->hdr.id, size);
|
| 1706 | sep = true;
|
| 1707 | }
|
| 1708 | seq_putc(s, '\n');
|
| 1709 | }
|
| 1710 |
|
| 1711 | out:
|
| 1712 | rdtgroup_kn_unlock(of->kn);
|
| 1713 |
|
| 1714 | return ret;
|
| 1715 | }
|
| 1716 |
|
| 1717 | static void mondata_config_read(struct resctrl_mon_config_info *mon_info)
|
| 1718 | {
|
| 1719 | smp_call_function_any(&mon_info->d->hdr.cpu_mask,
|
| 1720 | resctrl_arch_mon_event_config_read, mon_info, 1);
|
| 1721 | }
|
| 1722 |
|
| 1723 | static int mbm_config_show(struct seq_file *s, struct rdt_resource *r, u32 evtid)
|
| 1724 | {
|
| 1725 | struct resctrl_mon_config_info mon_info;
|
| 1726 | struct rdt_l3_mon_domain *dom;
|
| 1727 | bool sep = false;
|
| 1728 |
|
| 1729 | lockdep_assert_cpus_held();
|
| 1730 | lockdep_assert_held(&rdtgroup_mutex);
|
| 1731 |
|
| 1732 | list_for_each_entry_rcu(dom, &r->mon_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 1733 | if (sep)
|
| 1734 | seq_puts(s, ";");
|
| 1735 |
|
| 1736 | memset(&mon_info, 0, sizeof(struct resctrl_mon_config_info));
|
| 1737 | mon_info.r = r;
|
| 1738 | mon_info.d = dom;
|
| 1739 | mon_info.evtid = evtid;
|
| 1740 | mondata_config_read(&mon_info);
|
| 1741 |
|
| 1742 | seq_printf(s, "%d=0x%02x", dom->hdr.id, mon_info.mon_config);
|
| 1743 | sep = true;
|
| 1744 | }
|
| 1745 | seq_puts(s, "\n");
|
| 1746 |
|
| 1747 | return 0;
|
| 1748 | }
|
| 1749 |
|
| 1750 | static int mbm_total_bytes_config_show(struct kernfs_open_file *of,
|
| 1751 | struct seq_file *seq, void *v)
|
| 1752 | {
|
| 1753 | struct rdt_resource *r = rdt_kn_parent_priv(of->kn);
|
| 1754 |
|
| 1755 | if (!info_kn_lock(of->kn))
|
| 1756 | return -ENOENT;
|
| 1757 |
|
| 1758 | mbm_config_show(seq, r, QOS_L3_MBM_TOTAL_EVENT_ID);
|
| 1759 |
|
| 1760 | info_kn_unlock(of->kn);
|
| 1761 | return 0;
|
| 1762 | }
|
| 1763 |
|
| 1764 | static int mbm_local_bytes_config_show(struct kernfs_open_file *of,
|
| 1765 | struct seq_file *seq, void *v)
|
| 1766 | {
|
| 1767 | struct rdt_resource *r = rdt_kn_parent_priv(of->kn);
|
| 1768 |
|
| 1769 | if (!info_kn_lock(of->kn))
|
| 1770 | return -ENOENT;
|
| 1771 |
|
| 1772 | mbm_config_show(seq, r, QOS_L3_MBM_LOCAL_EVENT_ID);
|
| 1773 |
|
| 1774 | info_kn_unlock(of->kn);
|
| 1775 | return 0;
|
| 1776 | }
|
| 1777 |
|
| 1778 | static void mbm_config_write_domain(struct rdt_resource *r,
|
| 1779 | struct rdt_l3_mon_domain *d, u32 evtid, u32 val)
|
| 1780 | {
|
| 1781 | struct resctrl_mon_config_info mon_info = {0};
|
| 1782 |
|
| 1783 | |
| 1784 | * Read the current config value first. If both are the same then
|
| 1785 | * no need to write it again.
|
| 1786 |
|
| 1787 | mon_info.r = r;
|
| 1788 | mon_info.d = d;
|
| 1789 | mon_info.evtid = evtid;
|
| 1790 | mondata_config_read(&mon_info);
|
| 1791 | if (mon_info.mon_config == val)
|
| 1792 | return;
|
| 1793 |
|
| 1794 | mon_info.mon_config = val;
|
| 1795 |
|
| 1796 | |
| 1797 | * Update MSR_IA32_EVT_CFG_BASE MSR on one of the CPUs in the
|
| 1798 | * domain. The MSRs offset from MSR MSR_IA32_EVT_CFG_BASE
|
| 1799 | * are scoped at the domain level. Writing any of these MSRs
|
| 1800 | * on one CPU is observed by all the CPUs in the domain.
|
| 1801 |
|
| 1802 | smp_call_function_any(&d->hdr.cpu_mask, resctrl_arch_mon_event_config_write,
|
| 1803 | &mon_info, 1);
|
| 1804 |
|
| 1805 | |
| 1806 | * When an Event Configuration is changed, the bandwidth counters
|
| 1807 | * for all RMIDs and Events will be cleared by the hardware. The
|
| 1808 | * hardware also sets MSR_IA32_QM_CTR.Unavailable (bit 62) for
|
| 1809 | * every RMID on the next read to any event for every RMID.
|
| 1810 | * Subsequent reads will have MSR_IA32_QM_CTR.Unavailable (bit 62)
|
| 1811 | * cleared while it is tracked by the hardware. Clear the
|
| 1812 | * mbm_local and mbm_total counts for all the RMIDs.
|
| 1813 |
|
| 1814 | resctrl_arch_reset_rmid_all(r, d);
|
| 1815 | }
|
| 1816 |
|
| 1817 | static int mon_config_write(struct rdt_resource *r, char *tok, u32 evtid)
|
| 1818 | {
|
| 1819 | char *dom_str = NULL, *id_str;
|
| 1820 | struct rdt_l3_mon_domain *d;
|
| 1821 | unsigned long dom_id, val;
|
| 1822 |
|
| 1823 |
|
| 1824 | lockdep_assert_cpus_held();
|
| 1825 |
|
| 1826 | next:
|
| 1827 | if (!tok || tok[0] == '\0')
|
| 1828 | return 0;
|
| 1829 |
|
| 1830 |
|
| 1831 | dom_str = strim(strsep(&tok, ";"));
|
| 1832 | id_str = strsep(&dom_str, "=");
|
| 1833 |
|
| 1834 | if (!id_str || kstrtoul(id_str, 10, &dom_id)) {
|
| 1835 | rdt_last_cmd_puts("Missing '=' or non-numeric domain id\n");
|
| 1836 | return -EINVAL;
|
| 1837 | }
|
| 1838 |
|
| 1839 | if (!dom_str || kstrtoul(dom_str, 16, &val)) {
|
| 1840 | rdt_last_cmd_puts("Non-numeric event configuration value\n");
|
| 1841 | return -EINVAL;
|
| 1842 | }
|
| 1843 |
|
| 1844 |
|
| 1845 | if ((val & r->mon.mbm_cfg_mask) != val) {
|
| 1846 | rdt_last_cmd_printf("Invalid event configuration: max valid mask is 0x%02x\n",
|
| 1847 | r->mon.mbm_cfg_mask);
|
| 1848 | return -EINVAL;
|
| 1849 | }
|
| 1850 |
|
| 1851 | list_for_each_entry_rcu(d, &r->mon_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 1852 | if (d->hdr.id == dom_id) {
|
| 1853 | mbm_config_write_domain(r, d, evtid, val);
|
| 1854 | goto next;
|
| 1855 | }
|
| 1856 | }
|
| 1857 |
|
| 1858 | return -EINVAL;
|
| 1859 | }
|
| 1860 |
|
| 1861 | static ssize_t mbm_total_bytes_config_write(struct kernfs_open_file *of,
|
| 1862 | char *buf, size_t nbytes,
|
| 1863 | loff_t off)
|
| 1864 | {
|
| 1865 | struct rdt_resource *r = rdt_kn_parent_priv(of->kn);
|
| 1866 | int ret;
|
| 1867 |
|
| 1868 | if (!info_kn_lock(of->kn))
|
| 1869 | return -ENOENT;
|
| 1870 |
|
| 1871 | rdt_last_cmd_clear();
|
| 1872 |
|
| 1873 |
|
| 1874 | if (nbytes == 0 || buf[nbytes - 1] != '\n') {
|
| 1875 | rdt_last_cmd_puts("mbm_total_bytes_config: Invalid input\n");
|
| 1876 | ret = -EINVAL;
|
| 1877 | goto out_unlock;
|
| 1878 | }
|
| 1879 |
|
| 1880 | buf[nbytes - 1] = '\0';
|
| 1881 |
|
| 1882 | ret = mon_config_write(r, buf, QOS_L3_MBM_TOTAL_EVENT_ID);
|
| 1883 |
|
| 1884 | out_unlock:
|
| 1885 | info_kn_unlock(of->kn);
|
| 1886 |
|
| 1887 | return ret ?: nbytes;
|
| 1888 | }
|
| 1889 |
|
| 1890 | static ssize_t mbm_local_bytes_config_write(struct kernfs_open_file *of,
|
| 1891 | char *buf, size_t nbytes,
|
| 1892 | loff_t off)
|
| 1893 | {
|
| 1894 | struct rdt_resource *r = rdt_kn_parent_priv(of->kn);
|
| 1895 | int ret;
|
| 1896 |
|
| 1897 | if (!info_kn_lock(of->kn))
|
| 1898 | return -ENOENT;
|
| 1899 |
|
| 1900 | rdt_last_cmd_clear();
|
| 1901 |
|
| 1902 |
|
| 1903 | if (nbytes == 0 || buf[nbytes - 1] != '\n') {
|
| 1904 | rdt_last_cmd_puts("mbm_local_bytes_config: Invalid input\n");
|
| 1905 | ret = -EINVAL;
|
| 1906 | goto out_unlock;
|
| 1907 | }
|
| 1908 |
|
| 1909 | buf[nbytes - 1] = '\0';
|
| 1910 |
|
| 1911 | ret = mon_config_write(r, buf, QOS_L3_MBM_LOCAL_EVENT_ID);
|
| 1912 |
|
| 1913 | out_unlock:
|
| 1914 | info_kn_unlock(of->kn);
|
| 1915 |
|
| 1916 | return ret ?: nbytes;
|
| 1917 | }
|
| 1918 |
|
| 1919 | |
| 1920 | * resctrl_bmec_files_show() — Controls the visibility of BMEC-related resctrl
|
| 1921 | * files. When @show is true, the files are displayed; when false, the files
|
| 1922 | * are hidden.
|
| 1923 | * Don't treat kernfs_find_and_get failure as an error, since this function may
|
| 1924 | * be called regardless of whether BMEC is supported or the event is enabled.
|
| 1925 |
|
| 1926 | void resctrl_bmec_files_show(struct rdt_resource *r, struct kernfs_node *l3_mon_kn,
|
| 1927 | bool show)
|
| 1928 | {
|
| 1929 | struct kernfs_node *kn_config, *mon_kn = NULL;
|
| 1930 | char name[32];
|
| 1931 |
|
| 1932 | if (!l3_mon_kn) {
|
| 1933 | sprintf(name, "%s_MON", r->name);
|
| 1934 | mon_kn = kernfs_find_and_get(kn_info, name);
|
| 1935 | if (!mon_kn)
|
| 1936 | return;
|
| 1937 | l3_mon_kn = mon_kn;
|
| 1938 | }
|
| 1939 |
|
| 1940 | kn_config = kernfs_find_and_get(l3_mon_kn, "mbm_total_bytes_config");
|
| 1941 | if (kn_config) {
|
| 1942 | kernfs_show(kn_config, show);
|
| 1943 | kernfs_put(kn_config);
|
| 1944 | }
|
| 1945 |
|
| 1946 | kn_config = kernfs_find_and_get(l3_mon_kn, "mbm_local_bytes_config");
|
| 1947 | if (kn_config) {
|
| 1948 | kernfs_show(kn_config, show);
|
| 1949 | kernfs_put(kn_config);
|
| 1950 | }
|
| 1951 |
|
| 1952 |
|
| 1953 | if (mon_kn)
|
| 1954 | kernfs_put(mon_kn);
|
| 1955 | }
|
| 1956 |
|
| 1957 | const char *rdtgroup_name_by_closid(u32 closid)
|
| 1958 | {
|
| 1959 | struct rdtgroup *rdtgrp;
|
| 1960 |
|
| 1961 | list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
|
| 1962 | if (rdtgrp->closid == closid)
|
| 1963 | return rdt_kn_name(rdtgrp->kn);
|
| 1964 | }
|
| 1965 |
|
| 1966 | return NULL;
|
| 1967 | }
|
| 1968 |
|
| 1969 |
|
| 1970 | static struct rftype res_common_files[] = {
|
| 1971 | {
|
| 1972 | .name = "last_cmd_status",
|
| 1973 | .mode = 0444,
|
| 1974 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 1975 | .seq_show = rdt_last_cmd_status_show,
|
| 1976 | .fflags = RFTYPE_TOP_INFO,
|
| 1977 | },
|
| 1978 | {
|
| 1979 | .name = "mbm_assign_on_mkdir",
|
| 1980 | .mode = 0644,
|
| 1981 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 1982 | .seq_show = resctrl_mbm_assign_on_mkdir_show,
|
| 1983 | .write = resctrl_mbm_assign_on_mkdir_write,
|
| 1984 | },
|
| 1985 | {
|
| 1986 | .name = "num_closids",
|
| 1987 | .mode = 0444,
|
| 1988 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 1989 | .seq_show = rdt_num_closids_show,
|
| 1990 | .fflags = RFTYPE_CTRL_INFO,
|
| 1991 | },
|
| 1992 | {
|
| 1993 | .name = "mon_features",
|
| 1994 | .mode = 0444,
|
| 1995 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 1996 | .seq_show = rdt_mon_features_show,
|
| 1997 | .fflags = RFTYPE_MON_INFO,
|
| 1998 | },
|
| 1999 | {
|
| 2000 | .name = "available_mbm_cntrs",
|
| 2001 | .mode = 0444,
|
| 2002 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2003 | .seq_show = resctrl_available_mbm_cntrs_show,
|
| 2004 | },
|
| 2005 | {
|
| 2006 | .name = "num_rmids",
|
| 2007 | .mode = 0444,
|
| 2008 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2009 | .seq_show = rdt_num_rmids_show,
|
| 2010 | .fflags = RFTYPE_MON_INFO,
|
| 2011 | },
|
| 2012 | {
|
| 2013 | .name = "cbm_mask",
|
| 2014 | .mode = 0444,
|
| 2015 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2016 | .seq_show = rdt_default_ctrl_show,
|
| 2017 | .fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE,
|
| 2018 | },
|
| 2019 | {
|
| 2020 | .name = "num_mbm_cntrs",
|
| 2021 | .mode = 0444,
|
| 2022 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2023 | .seq_show = resctrl_num_mbm_cntrs_show,
|
| 2024 | },
|
| 2025 | {
|
| 2026 | .name = "min_cbm_bits",
|
| 2027 | .mode = 0444,
|
| 2028 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2029 | .seq_show = rdt_min_cbm_bits_show,
|
| 2030 | .fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE,
|
| 2031 | },
|
| 2032 | {
|
| 2033 | .name = "shareable_bits",
|
| 2034 | .mode = 0444,
|
| 2035 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2036 | .seq_show = rdt_shareable_bits_show,
|
| 2037 | .fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE,
|
| 2038 | },
|
| 2039 | {
|
| 2040 | .name = "bit_usage",
|
| 2041 | .mode = 0444,
|
| 2042 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2043 | .seq_show = rdt_bit_usage_show,
|
| 2044 | .fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE,
|
| 2045 | },
|
| 2046 | {
|
| 2047 | .name = "min_bandwidth",
|
| 2048 | .mode = 0444,
|
| 2049 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2050 | .seq_show = rdt_min_bw_show,
|
| 2051 | .fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_MB,
|
| 2052 | },
|
| 2053 | {
|
| 2054 | .name = "bandwidth_gran",
|
| 2055 | .mode = 0444,
|
| 2056 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2057 | .seq_show = rdt_bw_gran_show,
|
| 2058 | .fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_MB,
|
| 2059 | },
|
| 2060 | {
|
| 2061 | .name = "delay_linear",
|
| 2062 | .mode = 0444,
|
| 2063 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2064 | .seq_show = rdt_delay_linear_show,
|
| 2065 | .fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_MB,
|
| 2066 | },
|
| 2067 | |
| 2068 | * Platform specific which (if any) capabilities are provided by
|
| 2069 | * thread_throttle_mode. Defer "fflags" initialization to platform
|
| 2070 | * discovery.
|
| 2071 |
|
| 2072 | {
|
| 2073 | .name = "thread_throttle_mode",
|
| 2074 | .mode = 0444,
|
| 2075 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2076 | .seq_show = rdt_thread_throttle_mode_show,
|
| 2077 | },
|
| 2078 | {
|
| 2079 | .name = "io_alloc",
|
| 2080 | .mode = 0644,
|
| 2081 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2082 | .seq_show = resctrl_io_alloc_show,
|
| 2083 | .write = resctrl_io_alloc_write,
|
| 2084 | },
|
| 2085 | {
|
| 2086 | .name = "io_alloc_cbm",
|
| 2087 | .mode = 0644,
|
| 2088 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2089 | .seq_show = resctrl_io_alloc_cbm_show,
|
| 2090 | .write = resctrl_io_alloc_cbm_write,
|
| 2091 | },
|
| 2092 | {
|
| 2093 | .name = "max_threshold_occupancy",
|
| 2094 | .mode = 0644,
|
| 2095 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2096 | .write = max_threshold_occ_write,
|
| 2097 | .seq_show = max_threshold_occ_show,
|
| 2098 | .fflags = RFTYPE_MON_INFO | RFTYPE_RES_CACHE,
|
| 2099 | },
|
| 2100 | {
|
| 2101 | .name = "mbm_total_bytes_config",
|
| 2102 | .mode = 0644,
|
| 2103 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2104 | .seq_show = mbm_total_bytes_config_show,
|
| 2105 | .write = mbm_total_bytes_config_write,
|
| 2106 | },
|
| 2107 | {
|
| 2108 | .name = "mbm_local_bytes_config",
|
| 2109 | .mode = 0644,
|
| 2110 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2111 | .seq_show = mbm_local_bytes_config_show,
|
| 2112 | .write = mbm_local_bytes_config_write,
|
| 2113 | },
|
| 2114 | {
|
| 2115 | .name = "event_filter",
|
| 2116 | .mode = 0444,
|
| 2117 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2118 | .seq_show = event_filter_show,
|
| 2119 | .write = event_filter_write,
|
| 2120 | },
|
| 2121 | {
|
| 2122 | .name = "mbm_L3_assignments",
|
| 2123 | .mode = 0644,
|
| 2124 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2125 | .seq_show = mbm_L3_assignments_show,
|
| 2126 | .write = mbm_L3_assignments_write,
|
| 2127 | },
|
| 2128 | {
|
| 2129 | .name = "mbm_assign_mode",
|
| 2130 | .mode = 0644,
|
| 2131 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2132 | .seq_show = resctrl_mbm_assign_mode_show,
|
| 2133 | .write = resctrl_mbm_assign_mode_write,
|
| 2134 | .fflags = RFTYPE_MON_INFO | RFTYPE_RES_CACHE,
|
| 2135 | },
|
| 2136 | {
|
| 2137 | .name = "cpus",
|
| 2138 | .mode = 0644,
|
| 2139 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2140 | .write = rdtgroup_cpus_write,
|
| 2141 | .seq_show = rdtgroup_cpus_show,
|
| 2142 | .fflags = RFTYPE_BASE,
|
| 2143 | },
|
| 2144 | {
|
| 2145 | .name = "cpus_list",
|
| 2146 | .mode = 0644,
|
| 2147 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2148 | .write = rdtgroup_cpus_write,
|
| 2149 | .seq_show = rdtgroup_cpus_show,
|
| 2150 | .flags = RFTYPE_FLAGS_CPUS_LIST,
|
| 2151 | .fflags = RFTYPE_BASE,
|
| 2152 | },
|
| 2153 | {
|
| 2154 | .name = "tasks",
|
| 2155 | .mode = 0644,
|
| 2156 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2157 | .write = rdtgroup_tasks_write,
|
| 2158 | .seq_show = rdtgroup_tasks_show,
|
| 2159 | .fflags = RFTYPE_BASE,
|
| 2160 | },
|
| 2161 | {
|
| 2162 | .name = "mon_hw_id",
|
| 2163 | .mode = 0444,
|
| 2164 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2165 | .seq_show = rdtgroup_rmid_show,
|
| 2166 | .fflags = RFTYPE_MON_BASE | RFTYPE_DEBUG,
|
| 2167 | },
|
| 2168 | {
|
| 2169 | .name = "schemata",
|
| 2170 | .mode = 0644,
|
| 2171 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2172 | .write = rdtgroup_schemata_write,
|
| 2173 | .seq_show = rdtgroup_schemata_show,
|
| 2174 | .fflags = RFTYPE_CTRL_BASE,
|
| 2175 | },
|
| 2176 | {
|
| 2177 | .name = "mba_MBps_event",
|
| 2178 | .mode = 0644,
|
| 2179 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2180 | .write = rdtgroup_mba_mbps_event_write,
|
| 2181 | .seq_show = rdtgroup_mba_mbps_event_show,
|
| 2182 | },
|
| 2183 | {
|
| 2184 | .name = "mode",
|
| 2185 | .mode = 0644,
|
| 2186 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2187 | .write = rdtgroup_mode_write,
|
| 2188 | .seq_show = rdtgroup_mode_show,
|
| 2189 | .fflags = RFTYPE_CTRL_BASE,
|
| 2190 | },
|
| 2191 | {
|
| 2192 | .name = "size",
|
| 2193 | .mode = 0444,
|
| 2194 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2195 | .seq_show = rdtgroup_size_show,
|
| 2196 | .fflags = RFTYPE_CTRL_BASE,
|
| 2197 | },
|
| 2198 | {
|
| 2199 | .name = "sparse_masks",
|
| 2200 | .mode = 0444,
|
| 2201 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2202 | .seq_show = rdt_has_sparse_bitmasks_show,
|
| 2203 | .fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE,
|
| 2204 | },
|
| 2205 | {
|
| 2206 | .name = "ctrl_hw_id",
|
| 2207 | .mode = 0444,
|
| 2208 | .kf_ops = &rdtgroup_kf_single_ops,
|
| 2209 | .seq_show = rdtgroup_closid_show,
|
| 2210 | .fflags = RFTYPE_CTRL_BASE | RFTYPE_DEBUG,
|
| 2211 | },
|
| 2212 | };
|
| 2213 |
|
| 2214 | static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
|
| 2215 | {
|
| 2216 | struct rftype *rfts, *rft;
|
| 2217 | int ret, len;
|
| 2218 |
|
| 2219 | rfts = res_common_files;
|
| 2220 | len = ARRAY_SIZE(res_common_files);
|
| 2221 |
|
| 2222 | lockdep_assert_held(&rdtgroup_mutex);
|
| 2223 |
|
| 2224 | if (resctrl_debug)
|
| 2225 | fflags |= RFTYPE_DEBUG;
|
| 2226 |
|
| 2227 | for (rft = rfts; rft < rfts + len; rft++) {
|
| 2228 | if (rft->fflags && ((fflags & rft->fflags) == rft->fflags)) {
|
| 2229 | ret = rdtgroup_add_file(kn, rft);
|
| 2230 | if (ret)
|
| 2231 | goto error;
|
| 2232 | }
|
| 2233 | }
|
| 2234 |
|
| 2235 | return 0;
|
| 2236 | error:
|
| 2237 | pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
|
| 2238 | while (--rft >= rfts) {
|
| 2239 | if ((fflags & rft->fflags) == rft->fflags)
|
| 2240 | kernfs_remove_by_name(kn, rft->name);
|
| 2241 | }
|
| 2242 | return ret;
|
| 2243 | }
|
| 2244 |
|
| 2245 | static struct rftype *rdtgroup_get_rftype_by_name(const char *name)
|
| 2246 | {
|
| 2247 | struct rftype *rfts, *rft;
|
| 2248 | int len;
|
| 2249 |
|
| 2250 | rfts = res_common_files;
|
| 2251 | len = ARRAY_SIZE(res_common_files);
|
| 2252 |
|
| 2253 | for (rft = rfts; rft < rfts + len; rft++) {
|
| 2254 | if (!strcmp(rft->name, name))
|
| 2255 | return rft;
|
| 2256 | }
|
| 2257 |
|
| 2258 | return NULL;
|
| 2259 | }
|
| 2260 |
|
| 2261 | static void thread_throttle_mode_init(void)
|
| 2262 | {
|
| 2263 | enum membw_throttle_mode throttle_mode = THREAD_THROTTLE_UNDEFINED;
|
| 2264 | struct rdt_resource *r_mba, *r_smba;
|
| 2265 |
|
| 2266 | r_mba = resctrl_arch_get_resource(RDT_RESOURCE_MBA);
|
| 2267 | if (r_mba->alloc_capable &&
|
| 2268 | r_mba->membw.throttle_mode != THREAD_THROTTLE_UNDEFINED)
|
| 2269 | throttle_mode = r_mba->membw.throttle_mode;
|
| 2270 |
|
| 2271 | r_smba = resctrl_arch_get_resource(RDT_RESOURCE_SMBA);
|
| 2272 | if (r_smba->alloc_capable &&
|
| 2273 | r_smba->membw.throttle_mode != THREAD_THROTTLE_UNDEFINED)
|
| 2274 | throttle_mode = r_smba->membw.throttle_mode;
|
| 2275 |
|
| 2276 | if (throttle_mode == THREAD_THROTTLE_UNDEFINED)
|
| 2277 | return;
|
| 2278 |
|
| 2279 | resctrl_file_fflags_init("thread_throttle_mode",
|
| 2280 | RFTYPE_CTRL_INFO | RFTYPE_RES_MB);
|
| 2281 | }
|
| 2282 |
|
| 2283 | |
| 2284 | * The resctrl file "io_alloc" is added using L3 resource. However, it results
|
| 2285 | * in this file being visible for *all* cache resources (eg. L2 cache),
|
| 2286 | * whether it supports "io_alloc" or not.
|
| 2287 |
|
| 2288 | static void io_alloc_init(void)
|
| 2289 | {
|
| 2290 | struct rdt_resource *r = resctrl_arch_get_resource(RDT_RESOURCE_L3);
|
| 2291 |
|
| 2292 | if (r->cache.io_alloc_capable) {
|
| 2293 | resctrl_file_fflags_init("io_alloc", RFTYPE_CTRL_INFO |
|
| 2294 | RFTYPE_RES_CACHE);
|
| 2295 | resctrl_file_fflags_init("io_alloc_cbm",
|
| 2296 | RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE);
|
| 2297 | }
|
| 2298 | }
|
| 2299 |
|
| 2300 | void resctrl_file_fflags_init(const char *config, unsigned long fflags)
|
| 2301 | {
|
| 2302 | struct rftype *rft;
|
| 2303 |
|
| 2304 | rft = rdtgroup_get_rftype_by_name(config);
|
| 2305 | if (rft)
|
| 2306 | rft->fflags = fflags;
|
| 2307 | }
|
| 2308 |
|
| 2309 | void resctrl_file_mode_init(const char *config, umode_t mode)
|
| 2310 | {
|
| 2311 | struct rftype *rft;
|
| 2312 |
|
| 2313 | rft = rdtgroup_get_rftype_by_name(config);
|
| 2314 | if (rft)
|
| 2315 | rft->mode = mode;
|
| 2316 | }
|
| 2317 |
|
| 2318 | |
| 2319 | * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file
|
| 2320 | * @r: The resource group with which the file is associated.
|
| 2321 | * @name: Name of the file
|
| 2322 | *
|
| 2323 | * The permissions of named resctrl file, directory, or link are modified
|
| 2324 | * to not allow read, write, or execute by any user.
|
| 2325 | *
|
| 2326 | * WARNING: This function is intended to communicate to the user that the
|
| 2327 | * resctrl file has been locked down - that it is not relevant to the
|
| 2328 | * particular state the system finds itself in. It should not be relied
|
| 2329 | * on to protect from user access because after the file's permissions
|
| 2330 | * are restricted the user can still change the permissions using chmod
|
| 2331 | * from the command line.
|
| 2332 | *
|
| 2333 | * Return: 0 on success, <0 on failure.
|
| 2334 |
|
| 2335 | int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name)
|
| 2336 | {
|
| 2337 | struct iattr iattr = {.ia_valid = ATTR_MODE,};
|
| 2338 | struct kernfs_node *kn;
|
| 2339 | int ret = 0;
|
| 2340 |
|
| 2341 | kn = kernfs_find_and_get_ns(r->kn, name, NULL);
|
| 2342 | if (!kn)
|
| 2343 | return -ENOENT;
|
| 2344 |
|
| 2345 | switch (kernfs_type(kn)) {
|
| 2346 | case KERNFS_DIR:
|
| 2347 | iattr.ia_mode = S_IFDIR;
|
| 2348 | break;
|
| 2349 | case KERNFS_FILE:
|
| 2350 | iattr.ia_mode = S_IFREG;
|
| 2351 | break;
|
| 2352 | case KERNFS_LINK:
|
| 2353 | iattr.ia_mode = S_IFLNK;
|
| 2354 | break;
|
| 2355 | }
|
| 2356 |
|
| 2357 | ret = kernfs_setattr(kn, &iattr);
|
| 2358 | kernfs_put(kn);
|
| 2359 | return ret;
|
| 2360 | }
|
| 2361 |
|
| 2362 | |
| 2363 | * rdtgroup_kn_mode_restore - Restore user access to named resctrl file
|
| 2364 | * @r: The resource group with which the file is associated.
|
| 2365 | * @name: Name of the file
|
| 2366 | * @mask: Mask of permissions that should be restored
|
| 2367 | *
|
| 2368 | * Restore the permissions of the named file. If @name is a directory the
|
| 2369 | * permissions of its parent will be used.
|
| 2370 | *
|
| 2371 | * Return: 0 on success, <0 on failure.
|
| 2372 |
|
| 2373 | int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
|
| 2374 | umode_t mask)
|
| 2375 | {
|
| 2376 | struct iattr iattr = {.ia_valid = ATTR_MODE,};
|
| 2377 | struct kernfs_node *kn, *parent;
|
| 2378 | struct rftype *rfts, *rft;
|
| 2379 | int ret, len;
|
| 2380 |
|
| 2381 | rfts = res_common_files;
|
| 2382 | len = ARRAY_SIZE(res_common_files);
|
| 2383 |
|
| 2384 | for (rft = rfts; rft < rfts + len; rft++) {
|
| 2385 | if (!strcmp(rft->name, name))
|
| 2386 | iattr.ia_mode = rft->mode & mask;
|
| 2387 | }
|
| 2388 |
|
| 2389 | kn = kernfs_find_and_get_ns(r->kn, name, NULL);
|
| 2390 | if (!kn)
|
| 2391 | return -ENOENT;
|
| 2392 |
|
| 2393 | switch (kernfs_type(kn)) {
|
| 2394 | case KERNFS_DIR:
|
| 2395 | parent = kernfs_get_parent(kn);
|
| 2396 | if (parent) {
|
| 2397 | iattr.ia_mode |= parent->mode;
|
| 2398 | kernfs_put(parent);
|
| 2399 | }
|
| 2400 | iattr.ia_mode |= S_IFDIR;
|
| 2401 | break;
|
| 2402 | case KERNFS_FILE:
|
| 2403 | iattr.ia_mode |= S_IFREG;
|
| 2404 | break;
|
| 2405 | case KERNFS_LINK:
|
| 2406 | iattr.ia_mode |= S_IFLNK;
|
| 2407 | break;
|
| 2408 | }
|
| 2409 |
|
| 2410 | ret = kernfs_setattr(kn, &iattr);
|
| 2411 | kernfs_put(kn);
|
| 2412 | return ret;
|
| 2413 | }
|
| 2414 |
|
| 2415 | static int resctrl_mkdir_event_configs(struct rdt_resource *r, struct kernfs_node *l3_mon_kn)
|
| 2416 | {
|
| 2417 | struct kernfs_node *kn_subdir, *kn_subdir2;
|
| 2418 | struct mon_evt *mevt;
|
| 2419 | int ret;
|
| 2420 |
|
| 2421 | kn_subdir = kernfs_create_dir(l3_mon_kn, "event_configs", l3_mon_kn->mode, NULL);
|
| 2422 | if (IS_ERR(kn_subdir))
|
| 2423 | return PTR_ERR(kn_subdir);
|
| 2424 |
|
| 2425 | ret = rdtgroup_kn_set_ugid(kn_subdir);
|
| 2426 | if (ret)
|
| 2427 | return ret;
|
| 2428 |
|
| 2429 | for_each_mon_event(mevt) {
|
| 2430 | if (mevt->rid != r->rid || !mevt->enabled || !resctrl_is_mbm_event(mevt->evtid))
|
| 2431 | continue;
|
| 2432 |
|
| 2433 | kn_subdir2 = kernfs_create_dir(kn_subdir, mevt->name, kn_subdir->mode, mevt);
|
| 2434 | if (IS_ERR(kn_subdir2))
|
| 2435 | return PTR_ERR(kn_subdir2);
|
| 2436 |
|
| 2437 | ret = rdtgroup_kn_set_ugid(kn_subdir2);
|
| 2438 | if (ret)
|
| 2439 | return ret;
|
| 2440 |
|
| 2441 | ret = rdtgroup_add_files(kn_subdir2, RFTYPE_ASSIGN_CONFIG);
|
| 2442 | if (ret)
|
| 2443 | return ret;
|
| 2444 | }
|
| 2445 |
|
| 2446 | return 0;
|
| 2447 | }
|
| 2448 |
|
| 2449 | static int rdtgroup_mkdir_info_resdir(void *priv, char *name,
|
| 2450 | unsigned long fflags)
|
| 2451 | {
|
| 2452 | struct kernfs_node *kn_subdir;
|
| 2453 | struct rdt_resource *r;
|
| 2454 | int ret;
|
| 2455 |
|
| 2456 | kn_subdir = kernfs_create_dir(kn_info, name,
|
| 2457 | kn_info->mode, priv);
|
| 2458 | if (IS_ERR(kn_subdir))
|
| 2459 | return PTR_ERR(kn_subdir);
|
| 2460 |
|
| 2461 | ret = rdtgroup_kn_set_ugid(kn_subdir);
|
| 2462 | if (ret)
|
| 2463 | return ret;
|
| 2464 |
|
| 2465 | ret = rdtgroup_add_files(kn_subdir, fflags);
|
| 2466 | if (ret)
|
| 2467 | return ret;
|
| 2468 |
|
| 2469 | if ((fflags & RFTYPE_MON_INFO) == RFTYPE_MON_INFO) {
|
| 2470 | r = priv;
|
| 2471 | if (r->mon.mbm_cntr_assignable) {
|
| 2472 | ret = resctrl_mkdir_event_configs(r, kn_subdir);
|
| 2473 | if (ret)
|
| 2474 | return ret;
|
| 2475 | |
| 2476 | * Hide BMEC related files if mbm_event mode
|
| 2477 | * is enabled.
|
| 2478 |
|
| 2479 | if (resctrl_arch_mbm_cntr_assign_enabled(r))
|
| 2480 | resctrl_bmec_files_show(r, kn_subdir, false);
|
| 2481 | }
|
| 2482 | }
|
| 2483 |
|
| 2484 | kernfs_activate(kn_subdir);
|
| 2485 |
|
| 2486 | return ret;
|
| 2487 | }
|
| 2488 |
|
| 2489 | static unsigned long fflags_from_resource(struct rdt_resource *r)
|
| 2490 | {
|
| 2491 | switch (r->rid) {
|
| 2492 | case RDT_RESOURCE_L3:
|
| 2493 | case RDT_RESOURCE_L2:
|
| 2494 | return RFTYPE_RES_CACHE;
|
| 2495 | case RDT_RESOURCE_MBA:
|
| 2496 | case RDT_RESOURCE_SMBA:
|
| 2497 | return RFTYPE_RES_MB;
|
| 2498 | case RDT_RESOURCE_PERF_PKG:
|
| 2499 | return RFTYPE_RES_PERF_PKG;
|
| 2500 | }
|
| 2501 |
|
| 2502 | return WARN_ON_ONCE(1);
|
| 2503 | }
|
| 2504 |
|
| 2505 | static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
|
| 2506 | {
|
| 2507 | struct resctrl_schema *s;
|
| 2508 | struct rdt_resource *r;
|
| 2509 | unsigned long fflags;
|
| 2510 | char name[32];
|
| 2511 | int ret;
|
| 2512 |
|
| 2513 |
|
| 2514 | kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
|
| 2515 | if (IS_ERR(kn_info))
|
| 2516 | return PTR_ERR(kn_info);
|
| 2517 |
|
| 2518 | ret = rdtgroup_add_files(kn_info, RFTYPE_TOP_INFO);
|
| 2519 | if (ret)
|
| 2520 | goto out_destroy;
|
| 2521 |
|
| 2522 |
|
| 2523 | list_for_each_entry(s, &resctrl_schema_all, list) {
|
| 2524 | r = s->res;
|
| 2525 | fflags = fflags_from_resource(r) | RFTYPE_CTRL_INFO;
|
| 2526 | ret = rdtgroup_mkdir_info_resdir(s, s->name, fflags);
|
| 2527 | if (ret)
|
| 2528 | goto out_destroy;
|
| 2529 | }
|
| 2530 |
|
| 2531 | for_each_mon_capable_rdt_resource(r) {
|
| 2532 | fflags = fflags_from_resource(r) | RFTYPE_MON_INFO;
|
| 2533 | sprintf(name, "%s_MON", r->name);
|
| 2534 | ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
|
| 2535 | if (ret)
|
| 2536 | goto out_destroy;
|
| 2537 | }
|
| 2538 |
|
| 2539 | ret = rdtgroup_kn_set_ugid(kn_info);
|
| 2540 | if (ret)
|
| 2541 | goto out_destroy;
|
| 2542 |
|
| 2543 | kernfs_activate(kn_info);
|
| 2544 |
|
| 2545 | return 0;
|
| 2546 |
|
| 2547 | out_destroy:
|
| 2548 | kernfs_remove(kn_info);
|
| 2549 | return ret;
|
| 2550 | }
|
| 2551 |
|
| 2552 | static int
|
| 2553 | mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
|
| 2554 | char *name, struct kernfs_node **dest_kn)
|
| 2555 | {
|
| 2556 | struct kernfs_node *kn;
|
| 2557 | int ret;
|
| 2558 |
|
| 2559 |
|
| 2560 | kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
|
| 2561 | if (IS_ERR(kn))
|
| 2562 | return PTR_ERR(kn);
|
| 2563 |
|
| 2564 | if (dest_kn)
|
| 2565 | *dest_kn = kn;
|
| 2566 |
|
| 2567 | ret = rdtgroup_kn_set_ugid(kn);
|
| 2568 | if (ret)
|
| 2569 | goto out_destroy;
|
| 2570 |
|
| 2571 | kernfs_activate(kn);
|
| 2572 |
|
| 2573 | return 0;
|
| 2574 |
|
| 2575 | out_destroy:
|
| 2576 | kernfs_remove(kn);
|
| 2577 | return ret;
|
| 2578 | }
|
| 2579 |
|
| 2580 | static inline bool is_mba_linear(void)
|
| 2581 | {
|
| 2582 | return resctrl_arch_get_resource(RDT_RESOURCE_MBA)->membw.delay_linear;
|
| 2583 | }
|
| 2584 |
|
| 2585 | static int mba_sc_domain_allocate(struct rdt_resource *r, struct rdt_ctrl_domain *d)
|
| 2586 | {
|
| 2587 | u32 num_closid = resctrl_arch_get_num_closid(r);
|
| 2588 | int cpu = cpumask_any(&d->hdr.cpu_mask);
|
| 2589 | int i;
|
| 2590 |
|
| 2591 | d->mbps_val = kcalloc_node(num_closid, sizeof(*d->mbps_val),
|
| 2592 | GFP_KERNEL, cpu_to_node(cpu));
|
| 2593 | if (!d->mbps_val)
|
| 2594 | return -ENOMEM;
|
| 2595 |
|
| 2596 | for (i = 0; i < num_closid; i++)
|
| 2597 | d->mbps_val[i] = MBA_MAX_MBPS;
|
| 2598 |
|
| 2599 | return 0;
|
| 2600 | }
|
| 2601 |
|
| 2602 | static void mba_sc_domain_destroy(struct rdt_resource *r,
|
| 2603 | struct rdt_ctrl_domain *d)
|
| 2604 | {
|
| 2605 | kfree(d->mbps_val);
|
| 2606 | d->mbps_val = NULL;
|
| 2607 | }
|
| 2608 |
|
| 2609 | |
| 2610 | * The MBA software controller is supported only if MBM is supported and MBA is
|
| 2611 | * in linear scale, and the MBM monitor scope is the same as MBA control scope.
|
| 2612 | *
|
| 2613 | * The software controller cannot be supported when the MBM counters are
|
| 2614 | * assignable. There is no guarantee that MBM counters are assigned to the
|
| 2615 | * event backing the software controller in all monitoring domains of all
|
| 2616 | * monitoring groups.
|
| 2617 |
|
| 2618 | static bool supports_mba_mbps(void)
|
| 2619 | {
|
| 2620 | struct rdt_resource *rmbm = resctrl_arch_get_resource(RDT_RESOURCE_L3);
|
| 2621 | struct rdt_resource *r = resctrl_arch_get_resource(RDT_RESOURCE_MBA);
|
| 2622 |
|
| 2623 | return (resctrl_is_mbm_enabled() &&
|
| 2624 | r->alloc_capable && is_mba_linear() &&
|
| 2625 | r->ctrl_scope == rmbm->mon_scope &&
|
| 2626 | !rmbm->mon.mbm_cntr_assignable);
|
| 2627 | }
|
| 2628 |
|
| 2629 | |
| 2630 | * Enable or disable the MBA software controller
|
| 2631 | * which helps user specify bandwidth in MBps.
|
| 2632 |
|
| 2633 | static int set_mba_sc(bool mba_sc)
|
| 2634 | {
|
| 2635 | struct rdt_resource *r = resctrl_arch_get_resource(RDT_RESOURCE_MBA);
|
| 2636 | u32 num_closid = resctrl_arch_get_num_closid(r);
|
| 2637 | struct rdt_ctrl_domain *d;
|
| 2638 | unsigned long fflags;
|
| 2639 | int i;
|
| 2640 |
|
| 2641 | if (!supports_mba_mbps() || mba_sc == is_mba_sc(r))
|
| 2642 | return -EINVAL;
|
| 2643 |
|
| 2644 | r->membw.mba_sc = mba_sc;
|
| 2645 |
|
| 2646 | rdtgroup_default.mba_mbps_event = mba_mbps_default_event;
|
| 2647 |
|
| 2648 | list_for_each_entry_rcu(d, &r->ctrl_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 2649 | for (i = 0; i < num_closid; i++)
|
| 2650 | d->mbps_val[i] = MBA_MAX_MBPS;
|
| 2651 | }
|
| 2652 |
|
| 2653 | fflags = mba_sc ? RFTYPE_CTRL_BASE | RFTYPE_MON_BASE : 0;
|
| 2654 | resctrl_file_fflags_init("mba_MBps_event", fflags);
|
| 2655 |
|
| 2656 | return 0;
|
| 2657 | }
|
| 2658 |
|
| 2659 | |
| 2660 | * We don't allow rdtgroup directories to be created anywhere
|
| 2661 | * except the root directory. Thus when looking for the rdtgroup
|
| 2662 | * structure for a kernfs node we are either looking at a directory,
|
| 2663 | * in which case the rdtgroup structure is pointed at by the "priv"
|
| 2664 | * field, otherwise we have a file, and need only look to the parent
|
| 2665 | * to find the rdtgroup.
|
| 2666 |
|
| 2667 | static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
|
| 2668 | {
|
| 2669 | if (kernfs_type(kn) == KERNFS_DIR) {
|
| 2670 | |
| 2671 | * All the resource directories use "kn->priv"
|
| 2672 | * to point to the "struct rdtgroup" for the
|
| 2673 | * resource. "info" and its subdirectories don't
|
| 2674 | * have rdtgroup structures, so return NULL here.
|
| 2675 |
|
| 2676 | if (kn == kn_info ||
|
| 2677 | rcu_access_pointer(kn->__parent) == kn_info)
|
| 2678 | return NULL;
|
| 2679 | else
|
| 2680 | return kn->priv;
|
| 2681 | } else {
|
| 2682 | return rdt_kn_parent_priv(kn);
|
| 2683 | }
|
| 2684 | }
|
| 2685 |
|
| 2686 | static void rdtgroup_kn_get(struct rdtgroup *rdtgrp, struct kernfs_node *kn)
|
| 2687 | {
|
| 2688 | atomic_inc(&rdtgrp->waitcount);
|
| 2689 | kernfs_break_active_protection(kn);
|
| 2690 | }
|
| 2691 |
|
| 2692 | static void rdtgroup_kn_put(struct rdtgroup *rdtgrp, struct kernfs_node *kn)
|
| 2693 | {
|
| 2694 | bool needs_free;
|
| 2695 |
|
| 2696 | if (!atomic_dec_and_mutex_lock(&rdtgrp->waitcount, &rdtgroup_mutex)) {
|
| 2697 | kernfs_unbreak_active_protection(kn);
|
| 2698 | return;
|
| 2699 | }
|
| 2700 |
|
| 2701 | needs_free = rdtgrp->flags & RDT_DELETED;
|
| 2702 |
|
| 2703 | mutex_unlock(&rdtgroup_mutex);
|
| 2704 |
|
| 2705 | kernfs_unbreak_active_protection(kn);
|
| 2706 |
|
| 2707 | if (needs_free) {
|
| 2708 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
|
| 2709 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
|
| 2710 | rdtgroup_pseudo_lock_remove(rdtgrp);
|
| 2711 | rdtgroup_remove(rdtgrp);
|
| 2712 | }
|
| 2713 | }
|
| 2714 |
|
| 2715 | struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
|
| 2716 | {
|
| 2717 | struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
|
| 2718 |
|
| 2719 | if (!rdtgrp)
|
| 2720 | return NULL;
|
| 2721 |
|
| 2722 | rdtgroup_kn_get(rdtgrp, kn);
|
| 2723 |
|
| 2724 | cpus_read_lock();
|
| 2725 | mutex_lock(&rdtgroup_mutex);
|
| 2726 | rdt_last_cmd_clear();
|
| 2727 |
|
| 2728 |
|
| 2729 | if (rdtgrp->flags & RDT_DELETED) {
|
| 2730 | |
| 2731 | * It is safe to dereference kn to obtain the resource group's
|
| 2732 | * name because one extra reference to kn is obtained
|
| 2733 | * during resource group creation that will be released by
|
| 2734 | * rdtgroup_remove() called by rdtgroup_kn_put().
|
| 2735 |
|
| 2736 | rdt_last_cmd_printf("Resource group %s deleted. No commands possible.\n",
|
| 2737 | rdt_kn_name(rdtgrp->kn));
|
| 2738 | return NULL;
|
| 2739 | }
|
| 2740 |
|
| 2741 | return rdtgrp;
|
| 2742 | }
|
| 2743 |
|
| 2744 | void rdtgroup_kn_unlock(struct kernfs_node *kn)
|
| 2745 | {
|
| 2746 | struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
|
| 2747 |
|
| 2748 | if (!rdtgrp)
|
| 2749 | return;
|
| 2750 |
|
| 2751 | mutex_unlock(&rdtgroup_mutex);
|
| 2752 | cpus_read_unlock();
|
| 2753 |
|
| 2754 | rdtgroup_kn_put(rdtgrp, kn);
|
| 2755 | }
|
| 2756 |
|
| 2757 | |
| 2758 | * Accessing the kn after breaking active protection is safe since the open
|
| 2759 | * of resctrl file holds a kernfs base reference (different from active
|
| 2760 | * protection) on the kn ensuring that it remains accessible even if it was
|
| 2761 | * unlinked. Each kn in turn holds base reference to parent so the kn's
|
| 2762 | * genealogy remains in memory until all base references dropped.
|
| 2763 |
|
| 2764 | static bool is_active_resctrl_node(struct kernfs_node *kn)
|
| 2765 | {
|
| 2766 | struct kernfs_node *p;
|
| 2767 | bool match = false;
|
| 2768 |
|
| 2769 | guard(rcu)();
|
| 2770 | p = kn;
|
| 2771 | while (p) {
|
| 2772 | if (p == rdtgroup_default.kn) {
|
| 2773 | match = true;
|
| 2774 | break;
|
| 2775 | }
|
| 2776 | p = rcu_dereference(p->__parent);
|
| 2777 | }
|
| 2778 |
|
| 2779 | return match;
|
| 2780 | }
|
| 2781 |
|
| 2782 | bool info_kn_lock(struct kernfs_node *kn)
|
| 2783 | {
|
| 2784 | kernfs_break_active_protection(kn);
|
| 2785 | cpus_read_lock();
|
| 2786 | mutex_lock(&rdtgroup_mutex);
|
| 2787 |
|
| 2788 | |
| 2789 | * Check both if resctrl is torn down (!rdtgroup_default.kn) and
|
| 2790 | * if the reader's kernfs_node originates from a dead mount.
|
| 2791 |
|
| 2792 | if (!rdtgroup_default.kn || !is_active_resctrl_node(kn)) {
|
| 2793 | mutex_unlock(&rdtgroup_mutex);
|
| 2794 | cpus_read_unlock();
|
| 2795 | kernfs_unbreak_active_protection(kn);
|
| 2796 | return false;
|
| 2797 | }
|
| 2798 |
|
| 2799 | return true;
|
| 2800 | }
|
| 2801 |
|
| 2802 | void info_kn_unlock(struct kernfs_node *kn)
|
| 2803 | {
|
| 2804 | mutex_unlock(&rdtgroup_mutex);
|
| 2805 | cpus_read_unlock();
|
| 2806 | kernfs_unbreak_active_protection(kn);
|
| 2807 | }
|
| 2808 |
|
| 2809 | static int mkdir_mondata_all(struct kernfs_node *parent_kn,
|
| 2810 | struct rdtgroup *prgrp,
|
| 2811 | struct kernfs_node **mon_data_kn);
|
| 2812 |
|
| 2813 | static void rdt_disable_ctx(void)
|
| 2814 | {
|
| 2815 | resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L3, false);
|
| 2816 | resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L2, false);
|
| 2817 | set_mba_sc(false);
|
| 2818 |
|
| 2819 | resctrl_debug = false;
|
| 2820 | }
|
| 2821 |
|
| 2822 | static int rdt_enable_ctx(struct rdt_fs_context *ctx)
|
| 2823 | {
|
| 2824 | int ret = 0;
|
| 2825 |
|
| 2826 | if (ctx->enable_cdpl2) {
|
| 2827 | ret = resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L2, true);
|
| 2828 | if (ret)
|
| 2829 | goto out_done;
|
| 2830 | }
|
| 2831 |
|
| 2832 | if (ctx->enable_cdpl3) {
|
| 2833 | ret = resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L3, true);
|
| 2834 | if (ret)
|
| 2835 | goto out_cdpl2;
|
| 2836 | }
|
| 2837 |
|
| 2838 | if (ctx->enable_mba_mbps) {
|
| 2839 | ret = set_mba_sc(true);
|
| 2840 | if (ret)
|
| 2841 | goto out_cdpl3;
|
| 2842 | }
|
| 2843 |
|
| 2844 | if (ctx->enable_debug)
|
| 2845 | resctrl_debug = true;
|
| 2846 |
|
| 2847 | return 0;
|
| 2848 |
|
| 2849 | out_cdpl3:
|
| 2850 | resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L3, false);
|
| 2851 | out_cdpl2:
|
| 2852 | resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L2, false);
|
| 2853 | out_done:
|
| 2854 | return ret;
|
| 2855 | }
|
| 2856 |
|
| 2857 | static int schemata_list_add(struct rdt_resource *r, enum resctrl_conf_type type)
|
| 2858 | {
|
| 2859 | struct resctrl_schema *s;
|
| 2860 | const char *suffix = "";
|
| 2861 | int cl;
|
| 2862 |
|
| 2863 | s = kzalloc_obj(*s);
|
| 2864 | if (!s)
|
| 2865 | return -ENOMEM;
|
| 2866 |
|
| 2867 | s->res = r;
|
| 2868 | s->num_closid = resctrl_arch_get_num_closid(r);
|
| 2869 | if (resctrl_arch_get_cdp_enabled(r->rid))
|
| 2870 | s->num_closid /= 2;
|
| 2871 |
|
| 2872 | s->conf_type = type;
|
| 2873 | switch (type) {
|
| 2874 | case CDP_CODE:
|
| 2875 | suffix = "CODE";
|
| 2876 | break;
|
| 2877 | case CDP_DATA:
|
| 2878 | suffix = "DATA";
|
| 2879 | break;
|
| 2880 | case CDP_NONE:
|
| 2881 | suffix = "";
|
| 2882 | break;
|
| 2883 | }
|
| 2884 |
|
| 2885 | cl = snprintf(s->name, sizeof(s->name), "%s%s", r->name, suffix);
|
| 2886 | if (cl >= sizeof(s->name)) {
|
| 2887 | kfree(s);
|
| 2888 | return -EINVAL;
|
| 2889 | }
|
| 2890 |
|
| 2891 | |
| 2892 | * If CDP is supported by this resource, but not enabled,
|
| 2893 | * include the suffix. This ensures the tabular format of the
|
| 2894 | * schemata file does not change between mounts of the filesystem.
|
| 2895 |
|
| 2896 | if (r->cdp_capable && !resctrl_arch_get_cdp_enabled(r->rid))
|
| 2897 | cl += 4;
|
| 2898 |
|
| 2899 | if (cl > max_name_width)
|
| 2900 | max_name_width = cl;
|
| 2901 |
|
| 2902 | switch (r->schema_fmt) {
|
| 2903 | case RESCTRL_SCHEMA_BITMAP:
|
| 2904 | s->fmt_str = "%d=%x";
|
| 2905 | break;
|
| 2906 | case RESCTRL_SCHEMA_RANGE:
|
| 2907 | s->fmt_str = "%d=%u";
|
| 2908 | break;
|
| 2909 | }
|
| 2910 |
|
| 2911 | if (WARN_ON_ONCE(!s->fmt_str)) {
|
| 2912 | kfree(s);
|
| 2913 | return -EINVAL;
|
| 2914 | }
|
| 2915 |
|
| 2916 | INIT_LIST_HEAD(&s->list);
|
| 2917 | list_add(&s->list, &resctrl_schema_all);
|
| 2918 |
|
| 2919 | return 0;
|
| 2920 | }
|
| 2921 |
|
| 2922 | static int schemata_list_create(void)
|
| 2923 | {
|
| 2924 | struct rdt_resource *r;
|
| 2925 | int ret = 0;
|
| 2926 |
|
| 2927 | for_each_alloc_capable_rdt_resource(r) {
|
| 2928 | if (resctrl_arch_get_cdp_enabled(r->rid)) {
|
| 2929 | ret = schemata_list_add(r, CDP_CODE);
|
| 2930 | if (ret)
|
| 2931 | break;
|
| 2932 |
|
| 2933 | ret = schemata_list_add(r, CDP_DATA);
|
| 2934 | } else {
|
| 2935 | ret = schemata_list_add(r, CDP_NONE);
|
| 2936 | }
|
| 2937 |
|
| 2938 | if (ret)
|
| 2939 | break;
|
| 2940 | }
|
| 2941 |
|
| 2942 | return ret;
|
| 2943 | }
|
| 2944 |
|
| 2945 | static void schemata_list_destroy(void)
|
| 2946 | {
|
| 2947 | struct resctrl_schema *s, *tmp;
|
| 2948 |
|
| 2949 | list_for_each_entry_safe(s, tmp, &resctrl_schema_all, list) {
|
| 2950 | list_del(&s->list);
|
| 2951 | kfree(s);
|
| 2952 | }
|
| 2953 | }
|
| 2954 |
|
| 2955 | |
| 2956 | * Move tasks from one to the other group. If @from is NULL, then all tasks
|
| 2957 | * in the systems are moved unconditionally (used for teardown).
|
| 2958 | *
|
| 2959 | * If @mask is not NULL the cpus on which moved tasks are running are set
|
| 2960 | * in that mask so the update smp function call is restricted to affected
|
| 2961 | * cpus.
|
| 2962 |
|
| 2963 | static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
|
| 2964 | struct cpumask *mask)
|
| 2965 | {
|
| 2966 | struct task_struct *p, *t;
|
| 2967 |
|
| 2968 | read_lock(&tasklist_lock);
|
| 2969 | for_each_process_thread(p, t) {
|
| 2970 | if (!from || is_closid_match(t, from) ||
|
| 2971 | is_rmid_match(t, from)) {
|
| 2972 | resctrl_arch_set_closid_rmid(t, to->closid,
|
| 2973 | to->mon.rmid);
|
| 2974 |
|
| 2975 | |
| 2976 | * Order the closid/rmid stores above before the loads
|
| 2977 | * in task_curr(). This pairs with the full barrier
|
| 2978 | * between the rq->curr update and
|
| 2979 | * resctrl_arch_sched_in() during context switch.
|
| 2980 |
|
| 2981 | smp_mb();
|
| 2982 |
|
| 2983 | |
| 2984 | * If the task is on a CPU, set the CPU in the mask.
|
| 2985 | * The detection is inaccurate as tasks might move or
|
| 2986 | * schedule before the smp function call takes place.
|
| 2987 | * In such a case the function call is pointless, but
|
| 2988 | * there is no other side effect.
|
| 2989 |
|
| 2990 | if (IS_ENABLED(CONFIG_SMP) && mask && task_curr(t))
|
| 2991 | cpumask_set_cpu(task_cpu(t), mask);
|
| 2992 | }
|
| 2993 | }
|
| 2994 | read_unlock(&tasklist_lock);
|
| 2995 | }
|
| 2996 |
|
| 2997 | static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
|
| 2998 | {
|
| 2999 | struct rdtgroup *sentry, *stmp;
|
| 3000 | struct list_head *head;
|
| 3001 |
|
| 3002 | head = &rdtgrp->mon.crdtgrp_list;
|
| 3003 | list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
|
| 3004 | rdtgroup_unassign_cntrs(sentry);
|
| 3005 | free_rmid(sentry->closid, sentry->mon.rmid);
|
| 3006 | list_del(&sentry->mon.crdtgrp_list);
|
| 3007 |
|
| 3008 | if (atomic_read(&sentry->waitcount) != 0)
|
| 3009 | sentry->flags = RDT_DELETED;
|
| 3010 | else
|
| 3011 | rdtgroup_remove(sentry);
|
| 3012 | }
|
| 3013 | }
|
| 3014 |
|
| 3015 | |
| 3016 | * Forcibly remove all of subdirectories under root.
|
| 3017 |
|
| 3018 | static void rmdir_all_sub(void)
|
| 3019 | {
|
| 3020 | struct rdtgroup *rdtgrp, *tmp;
|
| 3021 |
|
| 3022 |
|
| 3023 | rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
|
| 3024 |
|
| 3025 | list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
|
| 3026 |
|
| 3027 | free_all_child_rdtgrp(rdtgrp);
|
| 3028 |
|
| 3029 |
|
| 3030 | if (rdtgrp == &rdtgroup_default)
|
| 3031 | continue;
|
| 3032 |
|
| 3033 | |
| 3034 | * Give any CPUs back to the default group. We cannot copy
|
| 3035 | * cpu_online_mask because a CPU might have executed the
|
| 3036 | * offline callback already, but is still marked online.
|
| 3037 |
|
| 3038 | cpumask_or(&rdtgroup_default.cpu_mask,
|
| 3039 | &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
|
| 3040 |
|
| 3041 | rdtgroup_unassign_cntrs(rdtgrp);
|
| 3042 |
|
| 3043 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
|
| 3044 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
|
| 3045 | rdtgroup_pseudo_lock_remove(rdtgrp);
|
| 3046 | } else {
|
| 3047 |
|
| 3048 | free_rmid(rdtgrp->closid, rdtgrp->mon.rmid);
|
| 3049 | }
|
| 3050 |
|
| 3051 | kernfs_remove(rdtgrp->kn);
|
| 3052 | list_del(&rdtgrp->rdtgroup_list);
|
| 3053 |
|
| 3054 | if (atomic_read(&rdtgrp->waitcount) != 0)
|
| 3055 | rdtgrp->flags = RDT_DELETED;
|
| 3056 | else
|
| 3057 | rdtgroup_remove(rdtgrp);
|
| 3058 | }
|
| 3059 |
|
| 3060 | update_closid_rmid(cpu_online_mask, &rdtgroup_default);
|
| 3061 |
|
| 3062 | kernfs_remove(kn_info);
|
| 3063 | kernfs_remove(kn_mongrp);
|
| 3064 | kernfs_remove(kn_mondata);
|
| 3065 | }
|
| 3066 |
|
| 3067 | |
| 3068 | * mon_get_kn_priv() - Get the mon_data priv data for this event.
|
| 3069 | *
|
| 3070 | * The same values are used across the mon_data directories of all control and
|
| 3071 | * monitor groups for the same event in the same domain. Keep a list of
|
| 3072 | * allocated structures and re-use an existing one with the same values for
|
| 3073 | * @rid, @domid, etc.
|
| 3074 | *
|
| 3075 | * @rid: The resource id for the event file being created.
|
| 3076 | * @domid: The domain id for the event file being created.
|
| 3077 | * @mevt: The type of event file being created.
|
| 3078 | * @do_sum: Whether SNC summing monitors are being created. Only set
|
| 3079 | * when @rid == RDT_RESOURCE_L3.
|
| 3080 | *
|
| 3081 | * Return: Pointer to mon_data private data of the event, NULL on failure.
|
| 3082 |
|
| 3083 | static struct mon_data *mon_get_kn_priv(enum resctrl_res_level rid, int domid,
|
| 3084 | struct mon_evt *mevt,
|
| 3085 | bool do_sum)
|
| 3086 | {
|
| 3087 | struct mon_data *priv;
|
| 3088 |
|
| 3089 | lockdep_assert_held(&rdtgroup_mutex);
|
| 3090 |
|
| 3091 | list_for_each_entry(priv, &mon_data_kn_priv_list, list) {
|
| 3092 | if (priv->rid == rid && priv->domid == domid &&
|
| 3093 | priv->sum == do_sum && priv->evt == mevt)
|
| 3094 | return priv;
|
| 3095 | }
|
| 3096 |
|
| 3097 | priv = kzalloc_obj(*priv);
|
| 3098 | if (!priv)
|
| 3099 | return NULL;
|
| 3100 |
|
| 3101 | priv->rid = rid;
|
| 3102 | priv->domid = domid;
|
| 3103 | priv->sum = do_sum;
|
| 3104 | priv->evt = mevt;
|
| 3105 | list_add_tail(&priv->list, &mon_data_kn_priv_list);
|
| 3106 |
|
| 3107 | return priv;
|
| 3108 | }
|
| 3109 |
|
| 3110 | |
| 3111 | * mon_put_kn_priv() - Free all allocated mon_data structures.
|
| 3112 | *
|
| 3113 | * Called when resctrl file system is unmounted.
|
| 3114 |
|
| 3115 | static void mon_put_kn_priv(void)
|
| 3116 | {
|
| 3117 | struct mon_data *priv, *tmp;
|
| 3118 |
|
| 3119 | lockdep_assert_held(&rdtgroup_mutex);
|
| 3120 |
|
| 3121 | list_for_each_entry_safe(priv, tmp, &mon_data_kn_priv_list, list) {
|
| 3122 | list_del(&priv->list);
|
| 3123 | kfree(priv);
|
| 3124 | }
|
| 3125 | }
|
| 3126 |
|
| 3127 | static void resctrl_fs_teardown(void)
|
| 3128 | {
|
| 3129 | lockdep_assert_held(&rdtgroup_mutex);
|
| 3130 |
|
| 3131 |
|
| 3132 | if (!rdtgroup_default.kn)
|
| 3133 | return;
|
| 3134 |
|
| 3135 | rmdir_all_sub();
|
| 3136 | rdtgroup_unassign_cntrs(&rdtgroup_default);
|
| 3137 | mon_put_kn_priv();
|
| 3138 | rdt_pseudo_lock_release();
|
| 3139 | rdtgroup_default.mode = RDT_MODE_SHAREABLE;
|
| 3140 | rdtgroup_default.flags = RDT_DELETED;
|
| 3141 | closid_exit();
|
| 3142 | schemata_list_destroy();
|
| 3143 | rdtgroup_destroy_root();
|
| 3144 | }
|
| 3145 |
|
| 3146 | static void resctrl_unmount(void)
|
| 3147 | {
|
| 3148 | struct rdt_resource *r;
|
| 3149 |
|
| 3150 | cpus_read_lock();
|
| 3151 | mutex_lock(&rdtgroup_mutex);
|
| 3152 |
|
| 3153 | rdt_disable_ctx();
|
| 3154 |
|
| 3155 |
|
| 3156 | for_each_alloc_capable_rdt_resource(r)
|
| 3157 | resctrl_arch_reset_all_ctrls(r);
|
| 3158 |
|
| 3159 | resctrl_fs_teardown();
|
| 3160 | if (resctrl_arch_alloc_capable())
|
| 3161 | resctrl_arch_disable_alloc();
|
| 3162 | if (resctrl_arch_mon_capable())
|
| 3163 | resctrl_arch_disable_mon();
|
| 3164 | resctrl_mounted = false;
|
| 3165 | mutex_unlock(&rdtgroup_mutex);
|
| 3166 | cpus_read_unlock();
|
| 3167 | }
|
| 3168 |
|
| 3169 | static int rdt_get_tree(struct fs_context *fc)
|
| 3170 | {
|
| 3171 | struct rdt_fs_context *ctx = rdt_fc2context(fc);
|
| 3172 | unsigned long flags = RFTYPE_CTRL_BASE;
|
| 3173 | struct kernfs_node *rdt_root_kn;
|
| 3174 | struct rdt_l3_mon_domain *dom;
|
| 3175 | struct rdt_resource *r;
|
| 3176 | int ret;
|
| 3177 |
|
| 3178 | DO_ONCE_SLEEPABLE(resctrl_arch_pre_mount);
|
| 3179 |
|
| 3180 | cpus_read_lock();
|
| 3181 | mutex_lock(&rdtgroup_mutex);
|
| 3182 | |
| 3183 | * resctrl file system can only be mounted once.
|
| 3184 |
|
| 3185 | if (resctrl_mounted) {
|
| 3186 | ret = -EBUSY;
|
| 3187 | goto out;
|
| 3188 | }
|
| 3189 |
|
| 3190 |
|
| 3191 | if (atomic_read(&rdtgroup_default.waitcount) != 0) {
|
| 3192 | ret = -EBUSY;
|
| 3193 | goto out;
|
| 3194 | }
|
| 3195 |
|
| 3196 | ret = setup_rmid_lru_list();
|
| 3197 | if (ret)
|
| 3198 | goto out;
|
| 3199 |
|
| 3200 | ret = rdtgroup_setup_root(ctx);
|
| 3201 | if (ret)
|
| 3202 | goto out;
|
| 3203 |
|
| 3204 | ret = rdt_enable_ctx(ctx);
|
| 3205 | if (ret)
|
| 3206 | goto out_root;
|
| 3207 |
|
| 3208 | ret = schemata_list_create();
|
| 3209 | if (ret)
|
| 3210 | goto out_schemata_free;
|
| 3211 |
|
| 3212 | ret = closid_init();
|
| 3213 | if (ret)
|
| 3214 | goto out_schemata_free;
|
| 3215 |
|
| 3216 | if (resctrl_arch_mon_capable())
|
| 3217 | flags |= RFTYPE_MON;
|
| 3218 |
|
| 3219 | ret = rdtgroup_add_files(rdtgroup_default.kn, flags);
|
| 3220 | if (ret)
|
| 3221 | goto out_closid_exit;
|
| 3222 |
|
| 3223 | kernfs_activate(rdtgroup_default.kn);
|
| 3224 |
|
| 3225 | ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
|
| 3226 | if (ret < 0)
|
| 3227 | goto out_closid_exit;
|
| 3228 |
|
| 3229 | if (resctrl_arch_mon_capable()) {
|
| 3230 | ret = mongroup_create_dir(rdtgroup_default.kn,
|
| 3231 | &rdtgroup_default, "mon_groups",
|
| 3232 | &kn_mongrp);
|
| 3233 | if (ret < 0)
|
| 3234 | goto out_info;
|
| 3235 |
|
| 3236 | rdtgroup_assign_cntrs(&rdtgroup_default);
|
| 3237 |
|
| 3238 | ret = mkdir_mondata_all(rdtgroup_default.kn,
|
| 3239 | &rdtgroup_default, &kn_mondata);
|
| 3240 | if (ret < 0)
|
| 3241 | goto out_mongrp;
|
| 3242 | rdtgroup_default.mon.mon_data_kn = kn_mondata;
|
| 3243 | }
|
| 3244 |
|
| 3245 | ret = rdt_pseudo_lock_init();
|
| 3246 | if (ret)
|
| 3247 | goto out_mondata;
|
| 3248 |
|
| 3249 | if (resctrl_arch_alloc_capable())
|
| 3250 | resctrl_arch_enable_alloc();
|
| 3251 | if (resctrl_arch_mon_capable())
|
| 3252 | resctrl_arch_enable_mon();
|
| 3253 |
|
| 3254 | if (resctrl_arch_alloc_capable() || resctrl_arch_mon_capable())
|
| 3255 | resctrl_mounted = true;
|
| 3256 |
|
| 3257 | if (resctrl_is_mbm_enabled()) {
|
| 3258 | r = resctrl_arch_get_resource(RDT_RESOURCE_L3);
|
| 3259 | list_for_each_entry_rcu(dom, &r->mon_domains, hdr.list, lockdep_is_cpus_held())
|
| 3260 | mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL,
|
| 3261 | RESCTRL_PICK_ANY_CPU);
|
| 3262 | }
|
| 3263 |
|
| 3264 | |
| 3265 | * Ensure root remains accessible after mutex is unlocked so that
|
| 3266 | * kernfs_kill_sb() can run safely if called by kernfs_get_tree()'s
|
| 3267 | * failure path after creating a superblock but before taking reference
|
| 3268 | * on root kn (for example, if unable to get inode for root kn).
|
| 3269 |
|
| 3270 | kernfs_get(rdtgroup_default.kn);
|
| 3271 |
|
| 3272 | |
| 3273 | * Make backup of the current root kn being created to be used in
|
| 3274 | * kernfs_put(). The additional reference taken above will prevent the
|
| 3275 | * kn from being freed before kernfs_kill_sb() can run but
|
| 3276 | * rdtgroup_default.kn may be set to NULL via rdtgroup_destroy_root()
|
| 3277 | * and its backing root (rdt_root) could be overwritten before
|
| 3278 | * kernfs_put() can run.
|
| 3279 |
|
| 3280 | rdt_root_kn = rdtgroup_default.kn;
|
| 3281 |
|
| 3282 | rdt_last_cmd_clear();
|
| 3283 | mutex_unlock(&rdtgroup_mutex);
|
| 3284 | cpus_read_unlock();
|
| 3285 |
|
| 3286 | ret = kernfs_get_tree(fc);
|
| 3287 | |
| 3288 | * resctrl can only be mounted once, new superblock only expected
|
| 3289 | * to be created once.
|
| 3290 |
|
| 3291 | if (!ctx->kfc.new_sb_created)
|
| 3292 | resctrl_unmount();
|
| 3293 | kernfs_put(rdt_root_kn);
|
| 3294 | return ret;
|
| 3295 |
|
| 3296 | out_mondata:
|
| 3297 | if (resctrl_arch_mon_capable())
|
| 3298 | kernfs_remove(kn_mondata);
|
| 3299 | out_mongrp:
|
| 3300 | if (resctrl_arch_mon_capable()) {
|
| 3301 | mon_put_kn_priv();
|
| 3302 | rdtgroup_unassign_cntrs(&rdtgroup_default);
|
| 3303 | kernfs_remove(kn_mongrp);
|
| 3304 | }
|
| 3305 | out_info:
|
| 3306 | kernfs_remove(kn_info);
|
| 3307 | out_closid_exit:
|
| 3308 | closid_exit();
|
| 3309 | out_schemata_free:
|
| 3310 | schemata_list_destroy();
|
| 3311 | rdt_disable_ctx();
|
| 3312 | out_root:
|
| 3313 | rdtgroup_destroy_root();
|
| 3314 | out:
|
| 3315 | mutex_unlock(&rdtgroup_mutex);
|
| 3316 | cpus_read_unlock();
|
| 3317 | return ret;
|
| 3318 | }
|
| 3319 |
|
| 3320 | enum rdt_param {
|
| 3321 | Opt_cdp,
|
| 3322 | Opt_cdpl2,
|
| 3323 | Opt_mba_mbps,
|
| 3324 | Opt_debug,
|
| 3325 | nr__rdt_params
|
| 3326 | };
|
| 3327 |
|
| 3328 | static const struct fs_parameter_spec rdt_fs_parameters[] = {
|
| 3329 | fsparam_flag("cdp", Opt_cdp),
|
| 3330 | fsparam_flag("cdpl2", Opt_cdpl2),
|
| 3331 | fsparam_flag("mba_MBps", Opt_mba_mbps),
|
| 3332 | fsparam_flag("debug", Opt_debug),
|
| 3333 | {}
|
| 3334 | };
|
| 3335 |
|
| 3336 | static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param)
|
| 3337 | {
|
| 3338 | struct rdt_fs_context *ctx = rdt_fc2context(fc);
|
| 3339 | struct fs_parse_result result;
|
| 3340 | const char *msg;
|
| 3341 | int opt;
|
| 3342 |
|
| 3343 | opt = fs_parse(fc, rdt_fs_parameters, param, &result);
|
| 3344 | if (opt < 0)
|
| 3345 | return opt;
|
| 3346 |
|
| 3347 | switch (opt) {
|
| 3348 | case Opt_cdp:
|
| 3349 | ctx->enable_cdpl3 = true;
|
| 3350 | return 0;
|
| 3351 | case Opt_cdpl2:
|
| 3352 | ctx->enable_cdpl2 = true;
|
| 3353 | return 0;
|
| 3354 | case Opt_mba_mbps:
|
| 3355 | msg = "mba_MBps requires MBM (mbm_event mode not supported) and linear scale MBA at L3 scope";
|
| 3356 | if (!supports_mba_mbps())
|
| 3357 | return invalfc(fc, msg);
|
| 3358 | ctx->enable_mba_mbps = true;
|
| 3359 | return 0;
|
| 3360 | case Opt_debug:
|
| 3361 | ctx->enable_debug = true;
|
| 3362 | return 0;
|
| 3363 | }
|
| 3364 |
|
| 3365 | return -EINVAL;
|
| 3366 | }
|
| 3367 |
|
| 3368 | static void rdt_fs_context_free(struct fs_context *fc)
|
| 3369 | {
|
| 3370 | struct rdt_fs_context *ctx = rdt_fc2context(fc);
|
| 3371 |
|
| 3372 | kernfs_free_fs_context(fc);
|
| 3373 | kfree(ctx);
|
| 3374 | }
|
| 3375 |
|
| 3376 | static const struct fs_context_operations rdt_fs_context_ops = {
|
| 3377 | .free = rdt_fs_context_free,
|
| 3378 | .parse_param = rdt_parse_param,
|
| 3379 | .get_tree = rdt_get_tree,
|
| 3380 | };
|
| 3381 |
|
| 3382 | static int rdt_init_fs_context(struct fs_context *fc)
|
| 3383 | {
|
| 3384 | struct rdt_fs_context *ctx;
|
| 3385 |
|
| 3386 | ctx = kzalloc_obj(*ctx);
|
| 3387 | if (!ctx)
|
| 3388 | return -ENOMEM;
|
| 3389 |
|
| 3390 | ctx->kfc.magic = RDTGROUP_SUPER_MAGIC;
|
| 3391 | fc->fs_private = &ctx->kfc;
|
| 3392 | fc->ops = &rdt_fs_context_ops;
|
| 3393 | put_user_ns(fc->user_ns);
|
| 3394 | fc->user_ns = get_user_ns(&init_user_ns);
|
| 3395 | fc->global = true;
|
| 3396 | return 0;
|
| 3397 | }
|
| 3398 |
|
| 3399 | static void rdt_kill_sb(struct super_block *sb)
|
| 3400 | {
|
| 3401 | resctrl_unmount();
|
| 3402 | kernfs_kill_sb(sb);
|
| 3403 | }
|
| 3404 |
|
| 3405 | static struct file_system_type rdt_fs_type = {
|
| 3406 | .name = "resctrl",
|
| 3407 | .init_fs_context = rdt_init_fs_context,
|
| 3408 | .parameters = rdt_fs_parameters,
|
| 3409 | .kill_sb = rdt_kill_sb,
|
| 3410 | };
|
| 3411 |
|
| 3412 | static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
|
| 3413 | void *priv)
|
| 3414 | {
|
| 3415 | struct kernfs_node *kn;
|
| 3416 | int ret = 0;
|
| 3417 |
|
| 3418 | kn = __kernfs_create_file(parent_kn, name, 0444,
|
| 3419 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
|
| 3420 | &kf_mondata_ops, priv, NULL, NULL);
|
| 3421 | if (IS_ERR(kn))
|
| 3422 | return PTR_ERR(kn);
|
| 3423 |
|
| 3424 | ret = rdtgroup_kn_set_ugid(kn);
|
| 3425 | if (ret) {
|
| 3426 | kernfs_remove(kn);
|
| 3427 | return ret;
|
| 3428 | }
|
| 3429 |
|
| 3430 | return ret;
|
| 3431 | }
|
| 3432 |
|
| 3433 | static void mon_rmdir_one_subdir(struct kernfs_node *pkn, char *name, char *subname)
|
| 3434 | {
|
| 3435 | struct kernfs_node *kn;
|
| 3436 |
|
| 3437 | kn = kernfs_find_and_get(pkn, name);
|
| 3438 | if (!kn)
|
| 3439 | return;
|
| 3440 | kernfs_put(kn);
|
| 3441 |
|
| 3442 | if (kn->dir.subdirs <= 1)
|
| 3443 | kernfs_remove(kn);
|
| 3444 | else
|
| 3445 | kernfs_remove_by_name(kn, subname);
|
| 3446 | }
|
| 3447 |
|
| 3448 | |
| 3449 | * Remove files and directories for one SNC node. If it is the last node
|
| 3450 | * sharing an L3 cache, then remove the upper level directory containing
|
| 3451 | * the "sum" files too.
|
| 3452 |
|
| 3453 | static void rmdir_mondata_subdir_allrdtgrp_snc(struct rdt_resource *r,
|
| 3454 | struct rdt_domain_hdr *hdr)
|
| 3455 | {
|
| 3456 | struct rdtgroup *prgrp, *crgrp;
|
| 3457 | struct rdt_l3_mon_domain *d;
|
| 3458 | char subname[32];
|
| 3459 | char name[32];
|
| 3460 |
|
| 3461 | if (!domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3))
|
| 3462 | return;
|
| 3463 |
|
| 3464 | d = container_of(hdr, struct rdt_l3_mon_domain, hdr);
|
| 3465 | sprintf(name, "mon_%s_%02d", r->name, d->ci_id);
|
| 3466 | sprintf(subname, "mon_sub_%s_%02d", r->name, hdr->id);
|
| 3467 |
|
| 3468 | list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
|
| 3469 | mon_rmdir_one_subdir(prgrp->mon.mon_data_kn, name, subname);
|
| 3470 |
|
| 3471 | list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
|
| 3472 | mon_rmdir_one_subdir(crgrp->mon.mon_data_kn, name, subname);
|
| 3473 | }
|
| 3474 | }
|
| 3475 |
|
| 3476 | |
| 3477 | * Remove all subdirectories of mon_data of ctrl_mon groups
|
| 3478 | * and monitor groups for the given domain.
|
| 3479 |
|
| 3480 | static void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
|
| 3481 | struct rdt_domain_hdr *hdr)
|
| 3482 | {
|
| 3483 | struct rdtgroup *prgrp, *crgrp;
|
| 3484 | char name[32];
|
| 3485 |
|
| 3486 | if (r->rid == RDT_RESOURCE_L3 && r->mon_scope == RESCTRL_L3_NODE) {
|
| 3487 | rmdir_mondata_subdir_allrdtgrp_snc(r, hdr);
|
| 3488 | return;
|
| 3489 | }
|
| 3490 |
|
| 3491 | sprintf(name, "mon_%s_%02d", r->name, hdr->id);
|
| 3492 | list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
|
| 3493 | kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
|
| 3494 |
|
| 3495 | list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
|
| 3496 | kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
|
| 3497 | }
|
| 3498 | }
|
| 3499 |
|
| 3500 | |
| 3501 | * Create a directory for a domain and populate it with monitor files. Create
|
| 3502 | * summing monitors when @hdr is NULL. No need to initialize summing monitors.
|
| 3503 |
|
| 3504 | static struct kernfs_node *_mkdir_mondata_subdir(struct kernfs_node *parent_kn, char *name,
|
| 3505 | struct rdt_domain_hdr *hdr,
|
| 3506 | struct rdt_resource *r,
|
| 3507 | struct rdtgroup *prgrp, int domid)
|
| 3508 | {
|
| 3509 | struct rmid_read rr = {0};
|
| 3510 | struct kernfs_node *kn;
|
| 3511 | struct mon_data *priv;
|
| 3512 | struct mon_evt *mevt;
|
| 3513 | int ret;
|
| 3514 |
|
| 3515 | kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
|
| 3516 | if (IS_ERR(kn))
|
| 3517 | return kn;
|
| 3518 |
|
| 3519 | ret = rdtgroup_kn_set_ugid(kn);
|
| 3520 | if (ret)
|
| 3521 | goto out_destroy;
|
| 3522 |
|
| 3523 | for_each_mon_event(mevt) {
|
| 3524 | if (mevt->rid != r->rid || !mevt->enabled)
|
| 3525 | continue;
|
| 3526 | priv = mon_get_kn_priv(r->rid, domid, mevt, !hdr);
|
| 3527 | if (WARN_ON_ONCE(!priv)) {
|
| 3528 | ret = -EINVAL;
|
| 3529 | goto out_destroy;
|
| 3530 | }
|
| 3531 |
|
| 3532 | ret = mon_addfile(kn, mevt->name, priv);
|
| 3533 | if (ret)
|
| 3534 | goto out_destroy;
|
| 3535 |
|
| 3536 | if (hdr && resctrl_is_mbm_event(mevt->evtid))
|
| 3537 | mon_event_read(&rr, r, hdr, prgrp, &hdr->cpu_mask, mevt, true);
|
| 3538 | }
|
| 3539 |
|
| 3540 | return kn;
|
| 3541 | out_destroy:
|
| 3542 | kernfs_remove(kn);
|
| 3543 | return ERR_PTR(ret);
|
| 3544 | }
|
| 3545 |
|
| 3546 | static int mkdir_mondata_subdir_snc(struct kernfs_node *parent_kn,
|
| 3547 | struct rdt_domain_hdr *hdr,
|
| 3548 | struct rdt_resource *r, struct rdtgroup *prgrp)
|
| 3549 | {
|
| 3550 | struct kernfs_node *ckn, *kn;
|
| 3551 | struct rdt_l3_mon_domain *d;
|
| 3552 | char name[32];
|
| 3553 |
|
| 3554 | if (!domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3))
|
| 3555 | return -EINVAL;
|
| 3556 |
|
| 3557 | d = container_of(hdr, struct rdt_l3_mon_domain, hdr);
|
| 3558 | sprintf(name, "mon_%s_%02d", r->name, d->ci_id);
|
| 3559 | kn = kernfs_find_and_get(parent_kn, name);
|
| 3560 | if (kn) {
|
| 3561 | |
| 3562 | * rdtgroup_mutex will prevent this directory from being
|
| 3563 | * removed. No need to keep this hold.
|
| 3564 |
|
| 3565 | kernfs_put(kn);
|
| 3566 | } else {
|
| 3567 | kn = _mkdir_mondata_subdir(parent_kn, name, NULL, r, prgrp, d->ci_id);
|
| 3568 | if (IS_ERR(kn))
|
| 3569 | return PTR_ERR(kn);
|
| 3570 | }
|
| 3571 |
|
| 3572 | sprintf(name, "mon_sub_%s_%02d", r->name, hdr->id);
|
| 3573 | ckn = _mkdir_mondata_subdir(kn, name, hdr, r, prgrp, hdr->id);
|
| 3574 | if (IS_ERR(ckn)) {
|
| 3575 | kernfs_remove(kn);
|
| 3576 | return PTR_ERR(ckn);
|
| 3577 | }
|
| 3578 |
|
| 3579 | kernfs_activate(kn);
|
| 3580 | return 0;
|
| 3581 | }
|
| 3582 |
|
| 3583 | static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
|
| 3584 | struct rdt_domain_hdr *hdr,
|
| 3585 | struct rdt_resource *r, struct rdtgroup *prgrp)
|
| 3586 | {
|
| 3587 | struct kernfs_node *kn;
|
| 3588 | char name[32];
|
| 3589 |
|
| 3590 | lockdep_assert_held(&rdtgroup_mutex);
|
| 3591 |
|
| 3592 | if (r->rid == RDT_RESOURCE_L3 && r->mon_scope == RESCTRL_L3_NODE)
|
| 3593 | return mkdir_mondata_subdir_snc(parent_kn, hdr, r, prgrp);
|
| 3594 |
|
| 3595 | sprintf(name, "mon_%s_%02d", r->name, hdr->id);
|
| 3596 | kn = _mkdir_mondata_subdir(parent_kn, name, hdr, r, prgrp, hdr->id);
|
| 3597 | if (IS_ERR(kn))
|
| 3598 | return PTR_ERR(kn);
|
| 3599 |
|
| 3600 | kernfs_activate(kn);
|
| 3601 | return 0;
|
| 3602 | }
|
| 3603 |
|
| 3604 | |
| 3605 | * Add all subdirectories of mon_data for "ctrl_mon" groups
|
| 3606 | * and "monitor" groups with given domain id.
|
| 3607 |
|
| 3608 | static void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
|
| 3609 | struct rdt_domain_hdr *hdr)
|
| 3610 | {
|
| 3611 | struct kernfs_node *parent_kn;
|
| 3612 | struct rdtgroup *prgrp, *crgrp;
|
| 3613 | struct list_head *head;
|
| 3614 |
|
| 3615 | list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
|
| 3616 | parent_kn = prgrp->mon.mon_data_kn;
|
| 3617 | mkdir_mondata_subdir(parent_kn, hdr, r, prgrp);
|
| 3618 |
|
| 3619 | head = &prgrp->mon.crdtgrp_list;
|
| 3620 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
|
| 3621 | parent_kn = crgrp->mon.mon_data_kn;
|
| 3622 | mkdir_mondata_subdir(parent_kn, hdr, r, crgrp);
|
| 3623 | }
|
| 3624 | }
|
| 3625 | }
|
| 3626 |
|
| 3627 | static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
|
| 3628 | struct rdt_resource *r,
|
| 3629 | struct rdtgroup *prgrp)
|
| 3630 | {
|
| 3631 | struct rdt_domain_hdr *hdr;
|
| 3632 | int ret;
|
| 3633 |
|
| 3634 |
|
| 3635 | lockdep_assert_cpus_held();
|
| 3636 |
|
| 3637 | list_for_each_entry_rcu(hdr, &r->mon_domains, list, lockdep_is_cpus_held()) {
|
| 3638 | ret = mkdir_mondata_subdir(parent_kn, hdr, r, prgrp);
|
| 3639 | if (ret)
|
| 3640 | return ret;
|
| 3641 | }
|
| 3642 |
|
| 3643 | return 0;
|
| 3644 | }
|
| 3645 |
|
| 3646 | |
| 3647 | * This creates a directory mon_data which contains the monitored data.
|
| 3648 | *
|
| 3649 | * mon_data has one directory for each domain which are named
|
| 3650 | * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
|
| 3651 | * with L3 domain looks as below:
|
| 3652 | * ./mon_data:
|
| 3653 | * mon_L3_00
|
| 3654 | * mon_L3_01
|
| 3655 | * mon_L3_02
|
| 3656 | * ...
|
| 3657 | *
|
| 3658 | * Each domain directory has one file per event:
|
| 3659 | * ./mon_L3_00/:
|
| 3660 | * llc_occupancy
|
| 3661 | *
|
| 3662 |
|
| 3663 | static int mkdir_mondata_all(struct kernfs_node *parent_kn,
|
| 3664 | struct rdtgroup *prgrp,
|
| 3665 | struct kernfs_node **dest_kn)
|
| 3666 | {
|
| 3667 | struct rdt_resource *r;
|
| 3668 | struct kernfs_node *kn;
|
| 3669 | int ret;
|
| 3670 |
|
| 3671 | |
| 3672 | * Create the mon_data directory first.
|
| 3673 |
|
| 3674 | ret = mongroup_create_dir(parent_kn, prgrp, "mon_data", &kn);
|
| 3675 | if (ret)
|
| 3676 | return ret;
|
| 3677 |
|
| 3678 | if (dest_kn)
|
| 3679 | *dest_kn = kn;
|
| 3680 |
|
| 3681 | |
| 3682 | * Create the subdirectories for each domain. Note that all events
|
| 3683 | * in a domain like L3 are grouped into a resource whose domain is L3
|
| 3684 |
|
| 3685 | for_each_mon_capable_rdt_resource(r) {
|
| 3686 | ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
|
| 3687 | if (ret)
|
| 3688 | goto out_destroy;
|
| 3689 | }
|
| 3690 |
|
| 3691 | return 0;
|
| 3692 |
|
| 3693 | out_destroy:
|
| 3694 | kernfs_remove(kn);
|
| 3695 | return ret;
|
| 3696 | }
|
| 3697 |
|
| 3698 | |
| 3699 | * cbm_ensure_valid - Enforce validity on provided CBM
|
| 3700 | * @_val: Candidate CBM
|
| 3701 | * @r: RDT resource to which the CBM belongs
|
| 3702 | *
|
| 3703 | * The provided CBM represents all cache portions available for use. This
|
| 3704 | * may be represented by a bitmap that does not consist of contiguous ones
|
| 3705 | * and thus be an invalid CBM.
|
| 3706 | * Here the provided CBM is forced to be a valid CBM by only considering
|
| 3707 | * the first set of contiguous bits as valid and clearing all bits.
|
| 3708 | * The intention here is to provide a valid default CBM with which a new
|
| 3709 | * resource group is initialized. The user can follow this with a
|
| 3710 | * modification to the CBM if the default does not satisfy the
|
| 3711 | * requirements.
|
| 3712 | *
|
| 3713 | * Return: A CBM that is valid for resource @r.
|
| 3714 |
|
| 3715 | static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r)
|
| 3716 | {
|
| 3717 | unsigned int cbm_len = r->cache.cbm_len;
|
| 3718 | unsigned long first_bit, zero_bit;
|
| 3719 | unsigned long val;
|
| 3720 |
|
| 3721 | if (!_val || r->cache.arch_has_sparse_bitmasks)
|
| 3722 | return _val;
|
| 3723 |
|
| 3724 | val = _val;
|
| 3725 | first_bit = find_first_bit(&val, cbm_len);
|
| 3726 | zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
|
| 3727 |
|
| 3728 |
|
| 3729 | bitmap_clear(&val, zero_bit, cbm_len - zero_bit);
|
| 3730 | return (u32)val;
|
| 3731 | }
|
| 3732 |
|
| 3733 | |
| 3734 | * Initialize cache resources per RDT domain
|
| 3735 | *
|
| 3736 | * Set the RDT domain up to start off with all usable allocations. That is,
|
| 3737 | * all shareable and unused bits. All-zero CBM is invalid.
|
| 3738 |
|
| 3739 | static int __init_one_rdt_domain(struct rdt_ctrl_domain *d, struct resctrl_schema *s,
|
| 3740 | u32 closid)
|
| 3741 | {
|
| 3742 | enum resctrl_conf_type peer_type = resctrl_peer_type(s->conf_type);
|
| 3743 | enum resctrl_conf_type t = s->conf_type;
|
| 3744 | struct resctrl_staged_config *cfg;
|
| 3745 | struct rdt_resource *r = s->res;
|
| 3746 | u32 used_b = 0, unused_b = 0;
|
| 3747 | unsigned long tmp_cbm;
|
| 3748 | enum rdtgrp_mode mode;
|
| 3749 | u32 peer_ctl, ctrl_val;
|
| 3750 | int i;
|
| 3751 |
|
| 3752 | cfg = &d->staged_config[t];
|
| 3753 | cfg->have_new_ctrl = false;
|
| 3754 | cfg->new_ctrl = r->cache.shareable_bits;
|
| 3755 | used_b = r->cache.shareable_bits;
|
| 3756 | for (i = 0; i < closids_supported(); i++) {
|
| 3757 | if (closid_allocated(i) && i != closid) {
|
| 3758 | mode = rdtgroup_mode_by_closid(i);
|
| 3759 | if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
|
| 3760 | |
| 3761 | * ctrl values for locksetup aren't relevant
|
| 3762 | * until the schemata is written, and the mode
|
| 3763 | * becomes RDT_MODE_PSEUDO_LOCKED.
|
| 3764 |
|
| 3765 | continue;
|
| 3766 | |
| 3767 | * If CDP is active include peer domain's
|
| 3768 | * usage to ensure there is no overlap
|
| 3769 | * with an exclusive group.
|
| 3770 |
|
| 3771 | if (resctrl_arch_get_cdp_enabled(r->rid))
|
| 3772 | peer_ctl = resctrl_arch_get_config(r, d, i,
|
| 3773 | peer_type);
|
| 3774 | else
|
| 3775 | peer_ctl = 0;
|
| 3776 | ctrl_val = resctrl_arch_get_config(r, d, i,
|
| 3777 | s->conf_type);
|
| 3778 | used_b |= ctrl_val | peer_ctl;
|
| 3779 | if (mode == RDT_MODE_SHAREABLE)
|
| 3780 | cfg->new_ctrl |= ctrl_val | peer_ctl;
|
| 3781 | }
|
| 3782 | }
|
| 3783 | if (d->plr && d->plr->cbm > 0)
|
| 3784 | used_b |= d->plr->cbm;
|
| 3785 | unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
|
| 3786 | unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
|
| 3787 | cfg->new_ctrl |= unused_b;
|
| 3788 | |
| 3789 | * Force the initial CBM to be valid, user can
|
| 3790 | * modify the CBM based on system availability.
|
| 3791 |
|
| 3792 | cfg->new_ctrl = cbm_ensure_valid(cfg->new_ctrl, r);
|
| 3793 | |
| 3794 | * Assign the u32 CBM to an unsigned long to ensure that
|
| 3795 | * bitmap_weight() does not access out-of-bound memory.
|
| 3796 |
|
| 3797 | tmp_cbm = cfg->new_ctrl;
|
| 3798 | if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) {
|
| 3799 | rdt_last_cmd_printf("No space on %s:%d\n", s->name, d->hdr.id);
|
| 3800 | return -ENOSPC;
|
| 3801 | }
|
| 3802 | cfg->have_new_ctrl = true;
|
| 3803 |
|
| 3804 | return 0;
|
| 3805 | }
|
| 3806 |
|
| 3807 | |
| 3808 | * Initialize cache resources with default values.
|
| 3809 | *
|
| 3810 | * A new RDT group is being created on an allocation capable (CAT)
|
| 3811 | * supporting system. Set this group up to start off with all usable
|
| 3812 | * allocations.
|
| 3813 | *
|
| 3814 | * If there are no more shareable bits available on any domain then
|
| 3815 | * the entire allocation will fail.
|
| 3816 |
|
| 3817 | int rdtgroup_init_cat(struct resctrl_schema *s, u32 closid)
|
| 3818 | {
|
| 3819 | struct rdt_ctrl_domain *d;
|
| 3820 | int ret;
|
| 3821 |
|
| 3822 | list_for_each_entry_rcu(d, &s->res->ctrl_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 3823 | ret = __init_one_rdt_domain(d, s, closid);
|
| 3824 | if (ret < 0)
|
| 3825 | return ret;
|
| 3826 | }
|
| 3827 |
|
| 3828 | return 0;
|
| 3829 | }
|
| 3830 |
|
| 3831 |
|
| 3832 | static void rdtgroup_init_mba(struct rdt_resource *r, u32 closid)
|
| 3833 | {
|
| 3834 | struct resctrl_staged_config *cfg;
|
| 3835 | struct rdt_ctrl_domain *d;
|
| 3836 |
|
| 3837 | list_for_each_entry_rcu(d, &r->ctrl_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 3838 | if (is_mba_sc(r)) {
|
| 3839 | d->mbps_val[closid] = MBA_MAX_MBPS;
|
| 3840 | continue;
|
| 3841 | }
|
| 3842 |
|
| 3843 | cfg = &d->staged_config[CDP_NONE];
|
| 3844 | cfg->new_ctrl = resctrl_get_default_ctrl(r);
|
| 3845 | cfg->have_new_ctrl = true;
|
| 3846 | }
|
| 3847 | }
|
| 3848 |
|
| 3849 |
|
| 3850 | static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
|
| 3851 | {
|
| 3852 | struct resctrl_schema *s;
|
| 3853 | struct rdt_resource *r;
|
| 3854 | int ret = 0;
|
| 3855 |
|
| 3856 | rdt_staged_configs_clear();
|
| 3857 |
|
| 3858 | list_for_each_entry(s, &resctrl_schema_all, list) {
|
| 3859 | r = s->res;
|
| 3860 | if (r->rid == RDT_RESOURCE_MBA ||
|
| 3861 | r->rid == RDT_RESOURCE_SMBA) {
|
| 3862 | rdtgroup_init_mba(r, rdtgrp->closid);
|
| 3863 | if (is_mba_sc(r))
|
| 3864 | continue;
|
| 3865 | } else {
|
| 3866 | ret = rdtgroup_init_cat(s, rdtgrp->closid);
|
| 3867 | if (ret < 0)
|
| 3868 | goto out;
|
| 3869 | }
|
| 3870 |
|
| 3871 | ret = resctrl_arch_update_domains(r, rdtgrp->closid);
|
| 3872 | if (ret < 0) {
|
| 3873 | rdt_last_cmd_puts("Failed to initialize allocations\n");
|
| 3874 | goto out;
|
| 3875 | }
|
| 3876 | }
|
| 3877 |
|
| 3878 | rdtgrp->mode = RDT_MODE_SHAREABLE;
|
| 3879 |
|
| 3880 | out:
|
| 3881 | rdt_staged_configs_clear();
|
| 3882 | return ret;
|
| 3883 | }
|
| 3884 |
|
| 3885 | static int mkdir_rdt_prepare_rmid_alloc(struct rdtgroup *rdtgrp)
|
| 3886 | {
|
| 3887 | int ret;
|
| 3888 |
|
| 3889 | if (!resctrl_arch_mon_capable())
|
| 3890 | return 0;
|
| 3891 |
|
| 3892 | ret = alloc_rmid(rdtgrp->closid);
|
| 3893 | if (ret < 0) {
|
| 3894 | rdt_last_cmd_puts("Out of RMIDs\n");
|
| 3895 | return ret;
|
| 3896 | }
|
| 3897 | rdtgrp->mon.rmid = ret;
|
| 3898 |
|
| 3899 | rdtgroup_assign_cntrs(rdtgrp);
|
| 3900 |
|
| 3901 | ret = mkdir_mondata_all(rdtgrp->kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
|
| 3902 | if (ret) {
|
| 3903 | rdt_last_cmd_puts("kernfs subdir error\n");
|
| 3904 | rdtgroup_unassign_cntrs(rdtgrp);
|
| 3905 | free_rmid(rdtgrp->closid, rdtgrp->mon.rmid);
|
| 3906 | return ret;
|
| 3907 | }
|
| 3908 |
|
| 3909 | return 0;
|
| 3910 | }
|
| 3911 |
|
| 3912 | static void mkdir_rdt_prepare_rmid_free(struct rdtgroup *rgrp)
|
| 3913 | {
|
| 3914 | if (resctrl_arch_mon_capable()) {
|
| 3915 | rdtgroup_unassign_cntrs(rgrp);
|
| 3916 | free_rmid(rgrp->closid, rgrp->mon.rmid);
|
| 3917 | }
|
| 3918 | }
|
| 3919 |
|
| 3920 | |
| 3921 | * We allow creating mon groups only with in a directory called "mon_groups"
|
| 3922 | * which is present in every ctrl_mon group. Check if this is a valid
|
| 3923 | * "mon_groups" directory.
|
| 3924 | *
|
| 3925 | * 1. The directory should be named "mon_groups".
|
| 3926 | * 2. The mon group itself should "not" be named "mon_groups".
|
| 3927 | * This makes sure "mon_groups" directory always has a ctrl_mon group
|
| 3928 | * as parent.
|
| 3929 |
|
| 3930 | static bool is_mon_groups(struct kernfs_node *kn, const char *name)
|
| 3931 | {
|
| 3932 | return (!strcmp(rdt_kn_name(kn), "mon_groups") &&
|
| 3933 | strcmp(name, "mon_groups"));
|
| 3934 | }
|
| 3935 |
|
| 3936 | static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
|
| 3937 | const char *name, umode_t mode,
|
| 3938 | enum rdt_group_type rtype, struct rdtgroup **r)
|
| 3939 | {
|
| 3940 | struct rdtgroup *prdtgrp, *rdtgrp;
|
| 3941 | unsigned long files = 0;
|
| 3942 | struct kernfs_node *kn;
|
| 3943 | int ret;
|
| 3944 |
|
| 3945 | prdtgrp = rdtgroup_kn_lock_live(parent_kn);
|
| 3946 | if (!prdtgrp) {
|
| 3947 | ret = -ENODEV;
|
| 3948 | goto out_unlock;
|
| 3949 | }
|
| 3950 |
|
| 3951 | |
| 3952 | * Check that the parent directory for a monitor group is a "mon_groups"
|
| 3953 | * directory.
|
| 3954 |
|
| 3955 | if (rtype == RDTMON_GROUP && !is_mon_groups(parent_kn, name)) {
|
| 3956 | ret = -EPERM;
|
| 3957 | goto out_unlock;
|
| 3958 | }
|
| 3959 |
|
| 3960 | if (rtype == RDTMON_GROUP &&
|
| 3961 | (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
|
| 3962 | prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
|
| 3963 | ret = -EINVAL;
|
| 3964 | rdt_last_cmd_puts("Pseudo-locking in progress\n");
|
| 3965 | goto out_unlock;
|
| 3966 | }
|
| 3967 |
|
| 3968 |
|
| 3969 | rdtgrp = kzalloc_obj(*rdtgrp);
|
| 3970 | if (!rdtgrp) {
|
| 3971 | ret = -ENOSPC;
|
| 3972 | rdt_last_cmd_puts("Kernel out of memory\n");
|
| 3973 | goto out_unlock;
|
| 3974 | }
|
| 3975 | *r = rdtgrp;
|
| 3976 | rdtgrp->mon.parent = prdtgrp;
|
| 3977 | rdtgrp->type = rtype;
|
| 3978 | INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
|
| 3979 |
|
| 3980 |
|
| 3981 | kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
|
| 3982 | if (IS_ERR(kn)) {
|
| 3983 | ret = PTR_ERR(kn);
|
| 3984 | rdt_last_cmd_puts("kernfs create error\n");
|
| 3985 | goto out_free_rgrp;
|
| 3986 | }
|
| 3987 | rdtgrp->kn = kn;
|
| 3988 |
|
| 3989 | |
| 3990 | * kernfs_remove() will drop the reference count on "kn" which
|
| 3991 | * will free it. But we still need it to stick around for the
|
| 3992 | * rdtgroup_kn_unlock(kn) call. Take one extra reference here,
|
| 3993 | * which will be dropped by kernfs_put() in rdtgroup_remove().
|
| 3994 |
|
| 3995 | kernfs_get(kn);
|
| 3996 |
|
| 3997 | ret = rdtgroup_kn_set_ugid(kn);
|
| 3998 | if (ret) {
|
| 3999 | rdt_last_cmd_puts("kernfs perm error\n");
|
| 4000 | goto out_destroy;
|
| 4001 | }
|
| 4002 |
|
| 4003 | if (rtype == RDTCTRL_GROUP) {
|
| 4004 | files = RFTYPE_BASE | RFTYPE_CTRL;
|
| 4005 | if (resctrl_arch_mon_capable())
|
| 4006 | files |= RFTYPE_MON;
|
| 4007 | } else {
|
| 4008 | files = RFTYPE_BASE | RFTYPE_MON;
|
| 4009 | }
|
| 4010 |
|
| 4011 | ret = rdtgroup_add_files(kn, files);
|
| 4012 | if (ret) {
|
| 4013 | rdt_last_cmd_puts("kernfs fill error\n");
|
| 4014 | goto out_destroy;
|
| 4015 | }
|
| 4016 |
|
| 4017 | |
| 4018 | * The caller unlocks the parent_kn upon success.
|
| 4019 |
|
| 4020 | return 0;
|
| 4021 |
|
| 4022 | out_destroy:
|
| 4023 | kernfs_put(rdtgrp->kn);
|
| 4024 | kernfs_remove(rdtgrp->kn);
|
| 4025 | out_free_rgrp:
|
| 4026 | kfree(rdtgrp);
|
| 4027 | out_unlock:
|
| 4028 | rdtgroup_kn_unlock(parent_kn);
|
| 4029 | return ret;
|
| 4030 | }
|
| 4031 |
|
| 4032 | static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
|
| 4033 | {
|
| 4034 | kernfs_remove(rgrp->kn);
|
| 4035 | rdtgroup_remove(rgrp);
|
| 4036 | }
|
| 4037 |
|
| 4038 | |
| 4039 | * Create a monitor group under "mon_groups" directory of a control
|
| 4040 | * and monitor group(ctrl_mon). This is a resource group
|
| 4041 | * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
|
| 4042 |
|
| 4043 | static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
|
| 4044 | const char *name, umode_t mode)
|
| 4045 | {
|
| 4046 | struct rdtgroup *rdtgrp, *prgrp;
|
| 4047 | int ret;
|
| 4048 |
|
| 4049 | ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTMON_GROUP, &rdtgrp);
|
| 4050 | if (ret)
|
| 4051 | return ret;
|
| 4052 |
|
| 4053 | prgrp = rdtgrp->mon.parent;
|
| 4054 | rdtgrp->closid = prgrp->closid;
|
| 4055 |
|
| 4056 | ret = mkdir_rdt_prepare_rmid_alloc(rdtgrp);
|
| 4057 | if (ret) {
|
| 4058 | mkdir_rdt_prepare_clean(rdtgrp);
|
| 4059 | goto out_unlock;
|
| 4060 | }
|
| 4061 |
|
| 4062 | kernfs_activate(rdtgrp->kn);
|
| 4063 |
|
| 4064 | |
| 4065 | * Add the rdtgrp to the list of rdtgrps the parent
|
| 4066 | * ctrl_mon group has to track.
|
| 4067 |
|
| 4068 | list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
|
| 4069 |
|
| 4070 | out_unlock:
|
| 4071 | rdtgroup_kn_unlock(parent_kn);
|
| 4072 | return ret;
|
| 4073 | }
|
| 4074 |
|
| 4075 | |
| 4076 | * These are rdtgroups created under the root directory. Can be used
|
| 4077 | * to allocate and monitor resources.
|
| 4078 |
|
| 4079 | static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
|
| 4080 | const char *name, umode_t mode)
|
| 4081 | {
|
| 4082 | struct rdtgroup *rdtgrp;
|
| 4083 | struct kernfs_node *kn;
|
| 4084 | u32 closid;
|
| 4085 | int ret;
|
| 4086 |
|
| 4087 | ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTCTRL_GROUP, &rdtgrp);
|
| 4088 | if (ret)
|
| 4089 | return ret;
|
| 4090 |
|
| 4091 | kn = rdtgrp->kn;
|
| 4092 | ret = closid_alloc();
|
| 4093 | if (ret < 0) {
|
| 4094 | rdt_last_cmd_puts("Out of CLOSIDs\n");
|
| 4095 | goto out_common_fail;
|
| 4096 | }
|
| 4097 | closid = ret;
|
| 4098 | ret = 0;
|
| 4099 |
|
| 4100 | rdtgrp->closid = closid;
|
| 4101 |
|
| 4102 | ret = mkdir_rdt_prepare_rmid_alloc(rdtgrp);
|
| 4103 | if (ret)
|
| 4104 | goto out_closid_free;
|
| 4105 |
|
| 4106 | kernfs_activate(rdtgrp->kn);
|
| 4107 |
|
| 4108 | ret = rdtgroup_init_alloc(rdtgrp);
|
| 4109 | if (ret < 0)
|
| 4110 | goto out_rmid_free;
|
| 4111 |
|
| 4112 | list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
|
| 4113 |
|
| 4114 | if (resctrl_arch_mon_capable()) {
|
| 4115 | |
| 4116 | * Create an empty mon_groups directory to hold the subset
|
| 4117 | * of tasks and cpus to monitor.
|
| 4118 |
|
| 4119 | ret = mongroup_create_dir(kn, rdtgrp, "mon_groups", NULL);
|
| 4120 | if (ret) {
|
| 4121 | rdt_last_cmd_puts("kernfs subdir error\n");
|
| 4122 | goto out_del_list;
|
| 4123 | }
|
| 4124 | if (is_mba_sc(NULL))
|
| 4125 | rdtgrp->mba_mbps_event = mba_mbps_default_event;
|
| 4126 | }
|
| 4127 |
|
| 4128 | goto out_unlock;
|
| 4129 |
|
| 4130 | out_del_list:
|
| 4131 | list_del(&rdtgrp->rdtgroup_list);
|
| 4132 | out_rmid_free:
|
| 4133 | mkdir_rdt_prepare_rmid_free(rdtgrp);
|
| 4134 | out_closid_free:
|
| 4135 | closid_free(closid);
|
| 4136 | out_common_fail:
|
| 4137 | mkdir_rdt_prepare_clean(rdtgrp);
|
| 4138 | out_unlock:
|
| 4139 | rdtgroup_kn_unlock(parent_kn);
|
| 4140 | return ret;
|
| 4141 | }
|
| 4142 |
|
| 4143 | static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
|
| 4144 | umode_t mode)
|
| 4145 | {
|
| 4146 |
|
| 4147 | if (strchr(name, '\n'))
|
| 4148 | return -EINVAL;
|
| 4149 |
|
| 4150 | |
| 4151 | * If the parent directory is the root directory and RDT
|
| 4152 | * allocation is supported, add a control and monitoring
|
| 4153 | * subdirectory
|
| 4154 |
|
| 4155 | if (resctrl_arch_alloc_capable() && parent_kn == rdtgroup_default.kn)
|
| 4156 | return rdtgroup_mkdir_ctrl_mon(parent_kn, name, mode);
|
| 4157 |
|
| 4158 |
|
| 4159 | if (resctrl_arch_mon_capable())
|
| 4160 | return rdtgroup_mkdir_mon(parent_kn, name, mode);
|
| 4161 |
|
| 4162 | return -EPERM;
|
| 4163 | }
|
| 4164 |
|
| 4165 | static int rdtgroup_rmdir_mon(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask)
|
| 4166 | {
|
| 4167 | struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
|
| 4168 | u32 closid, rmid;
|
| 4169 | int cpu;
|
| 4170 |
|
| 4171 |
|
| 4172 | rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
|
| 4173 |
|
| 4174 | |
| 4175 | * Update per cpu closid/rmid of the moved CPUs first.
|
| 4176 | * Note: the closid will not change, but the arch code still needs it.
|
| 4177 |
|
| 4178 | closid = prdtgrp->closid;
|
| 4179 | rmid = prdtgrp->mon.rmid;
|
| 4180 | for_each_cpu(cpu, &rdtgrp->cpu_mask)
|
| 4181 | resctrl_arch_set_cpu_default_closid_rmid(cpu, closid, rmid);
|
| 4182 |
|
| 4183 | |
| 4184 | * Update the MSR on moved CPUs and CPUs which have moved
|
| 4185 | * task running on them.
|
| 4186 |
|
| 4187 | cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
|
| 4188 | update_closid_rmid(tmpmask, NULL);
|
| 4189 |
|
| 4190 | rdtgrp->flags = RDT_DELETED;
|
| 4191 |
|
| 4192 | rdtgroup_unassign_cntrs(rdtgrp);
|
| 4193 |
|
| 4194 | free_rmid(rdtgrp->closid, rdtgrp->mon.rmid);
|
| 4195 |
|
| 4196 | |
| 4197 | * Remove the rdtgrp from the parent ctrl_mon group's list
|
| 4198 |
|
| 4199 | WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
|
| 4200 | list_del(&rdtgrp->mon.crdtgrp_list);
|
| 4201 |
|
| 4202 | kernfs_remove(rdtgrp->kn);
|
| 4203 |
|
| 4204 | return 0;
|
| 4205 | }
|
| 4206 |
|
| 4207 | static int rdtgroup_ctrl_remove(struct rdtgroup *rdtgrp)
|
| 4208 | {
|
| 4209 | rdtgrp->flags = RDT_DELETED;
|
| 4210 | list_del(&rdtgrp->rdtgroup_list);
|
| 4211 |
|
| 4212 | kernfs_remove(rdtgrp->kn);
|
| 4213 | return 0;
|
| 4214 | }
|
| 4215 |
|
| 4216 | static int rdtgroup_rmdir_ctrl(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask)
|
| 4217 | {
|
| 4218 | u32 closid, rmid;
|
| 4219 | int cpu;
|
| 4220 |
|
| 4221 |
|
| 4222 | rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
|
| 4223 |
|
| 4224 |
|
| 4225 | cpumask_or(&rdtgroup_default.cpu_mask,
|
| 4226 | &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
|
| 4227 |
|
| 4228 |
|
| 4229 | closid = rdtgroup_default.closid;
|
| 4230 | rmid = rdtgroup_default.mon.rmid;
|
| 4231 | for_each_cpu(cpu, &rdtgrp->cpu_mask)
|
| 4232 | resctrl_arch_set_cpu_default_closid_rmid(cpu, closid, rmid);
|
| 4233 |
|
| 4234 | |
| 4235 | * Update the MSR on moved CPUs and CPUs which have moved
|
| 4236 | * task running on them.
|
| 4237 |
|
| 4238 | cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
|
| 4239 | update_closid_rmid(tmpmask, NULL);
|
| 4240 |
|
| 4241 | rdtgroup_unassign_cntrs(rdtgrp);
|
| 4242 |
|
| 4243 | free_rmid(rdtgrp->closid, rdtgrp->mon.rmid);
|
| 4244 | closid_free(rdtgrp->closid);
|
| 4245 |
|
| 4246 | rdtgroup_ctrl_remove(rdtgrp);
|
| 4247 |
|
| 4248 | |
| 4249 | * Free all the child monitor group rmids.
|
| 4250 |
|
| 4251 | free_all_child_rdtgrp(rdtgrp);
|
| 4252 |
|
| 4253 | return 0;
|
| 4254 | }
|
| 4255 |
|
| 4256 | static struct kernfs_node *rdt_kn_parent(struct kernfs_node *kn)
|
| 4257 | {
|
| 4258 | |
| 4259 | * Valid within the RCU section it was obtained or while rdtgroup_mutex
|
| 4260 | * is held.
|
| 4261 |
|
| 4262 | return rcu_dereference_check(kn->__parent, lockdep_is_held(&rdtgroup_mutex));
|
| 4263 | }
|
| 4264 |
|
| 4265 | static int rdtgroup_rmdir(struct kernfs_node *kn)
|
| 4266 | {
|
| 4267 | struct kernfs_node *parent_kn;
|
| 4268 | struct rdtgroup *rdtgrp;
|
| 4269 | cpumask_var_t tmpmask;
|
| 4270 | int ret = 0;
|
| 4271 |
|
| 4272 | if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
|
| 4273 | return -ENOMEM;
|
| 4274 |
|
| 4275 | rdtgrp = rdtgroup_kn_lock_live(kn);
|
| 4276 | if (!rdtgrp) {
|
| 4277 | ret = -EPERM;
|
| 4278 | goto out;
|
| 4279 | }
|
| 4280 | parent_kn = rdt_kn_parent(kn);
|
| 4281 |
|
| 4282 | |
| 4283 | * If the rdtgroup is a ctrl_mon group and parent directory
|
| 4284 | * is the root directory, remove the ctrl_mon group.
|
| 4285 | *
|
| 4286 | * If the rdtgroup is a mon group and parent directory
|
| 4287 | * is a valid "mon_groups" directory, remove the mon group.
|
| 4288 |
|
| 4289 | if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn &&
|
| 4290 | rdtgrp != &rdtgroup_default) {
|
| 4291 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
|
| 4292 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
|
| 4293 | ret = rdtgroup_ctrl_remove(rdtgrp);
|
| 4294 | } else {
|
| 4295 | ret = rdtgroup_rmdir_ctrl(rdtgrp, tmpmask);
|
| 4296 | }
|
| 4297 | } else if (rdtgrp->type == RDTMON_GROUP &&
|
| 4298 | is_mon_groups(parent_kn, rdt_kn_name(kn))) {
|
| 4299 | ret = rdtgroup_rmdir_mon(rdtgrp, tmpmask);
|
| 4300 | } else {
|
| 4301 | ret = -EPERM;
|
| 4302 | }
|
| 4303 |
|
| 4304 | out:
|
| 4305 | rdtgroup_kn_unlock(kn);
|
| 4306 | free_cpumask_var(tmpmask);
|
| 4307 | return ret;
|
| 4308 | }
|
| 4309 |
|
| 4310 | |
| 4311 | * mongrp_reparent() - replace parent CTRL_MON group of a MON group
|
| 4312 | * @rdtgrp: the MON group whose parent should be replaced
|
| 4313 | * @new_prdtgrp: replacement parent CTRL_MON group for @rdtgrp
|
| 4314 | * @cpus: cpumask provided by the caller for use during this call
|
| 4315 | *
|
| 4316 | * Replaces the parent CTRL_MON group for a MON group, resulting in all member
|
| 4317 | * tasks' CLOSID immediately changing to that of the new parent group.
|
| 4318 | * Monitoring data for the group is unaffected by this operation.
|
| 4319 |
|
| 4320 | static void mongrp_reparent(struct rdtgroup *rdtgrp,
|
| 4321 | struct rdtgroup *new_prdtgrp,
|
| 4322 | cpumask_var_t cpus)
|
| 4323 | {
|
| 4324 | struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
|
| 4325 |
|
| 4326 | WARN_ON(rdtgrp->type != RDTMON_GROUP);
|
| 4327 | WARN_ON(new_prdtgrp->type != RDTCTRL_GROUP);
|
| 4328 |
|
| 4329 |
|
| 4330 | if (prdtgrp == new_prdtgrp)
|
| 4331 | return;
|
| 4332 |
|
| 4333 | WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
|
| 4334 | list_move_tail(&rdtgrp->mon.crdtgrp_list,
|
| 4335 | &new_prdtgrp->mon.crdtgrp_list);
|
| 4336 |
|
| 4337 | rdtgrp->mon.parent = new_prdtgrp;
|
| 4338 | rdtgrp->closid = new_prdtgrp->closid;
|
| 4339 |
|
| 4340 |
|
| 4341 | rdt_move_group_tasks(rdtgrp, rdtgrp, cpus);
|
| 4342 |
|
| 4343 | update_closid_rmid(cpus, NULL);
|
| 4344 | }
|
| 4345 |
|
| 4346 | static int rdtgroup_rename(struct kernfs_node *kn,
|
| 4347 | struct kernfs_node *new_parent, const char *new_name)
|
| 4348 | {
|
| 4349 | struct kernfs_node *kn_parent;
|
| 4350 | struct rdtgroup *new_prdtgrp;
|
| 4351 | struct rdtgroup *rdtgrp;
|
| 4352 | cpumask_var_t tmpmask;
|
| 4353 | int ret;
|
| 4354 |
|
| 4355 | rdtgrp = kernfs_to_rdtgroup(kn);
|
| 4356 | new_prdtgrp = kernfs_to_rdtgroup(new_parent);
|
| 4357 | if (!rdtgrp || !new_prdtgrp)
|
| 4358 | return -ENOENT;
|
| 4359 |
|
| 4360 |
|
| 4361 | rdtgroup_kn_get(rdtgrp, kn);
|
| 4362 | rdtgroup_kn_get(new_prdtgrp, new_parent);
|
| 4363 |
|
| 4364 | mutex_lock(&rdtgroup_mutex);
|
| 4365 |
|
| 4366 | rdt_last_cmd_clear();
|
| 4367 |
|
| 4368 | |
| 4369 | * Don't allow kernfs_to_rdtgroup() to return a parent rdtgroup if
|
| 4370 | * either kernfs_node is a file.
|
| 4371 |
|
| 4372 | if (kernfs_type(kn) != KERNFS_DIR ||
|
| 4373 | kernfs_type(new_parent) != KERNFS_DIR) {
|
| 4374 | rdt_last_cmd_puts("Source and destination must be directories");
|
| 4375 | ret = -EPERM;
|
| 4376 | goto out;
|
| 4377 | }
|
| 4378 |
|
| 4379 | if ((rdtgrp->flags & RDT_DELETED) || (new_prdtgrp->flags & RDT_DELETED)) {
|
| 4380 | ret = -ENOENT;
|
| 4381 | goto out;
|
| 4382 | }
|
| 4383 |
|
| 4384 | kn_parent = rdt_kn_parent(kn);
|
| 4385 | if (rdtgrp->type != RDTMON_GROUP || !kn_parent ||
|
| 4386 | !is_mon_groups(kn_parent, rdt_kn_name(kn))) {
|
| 4387 | rdt_last_cmd_puts("Source must be a MON group\n");
|
| 4388 | ret = -EPERM;
|
| 4389 | goto out;
|
| 4390 | }
|
| 4391 |
|
| 4392 | if (!is_mon_groups(new_parent, new_name)) {
|
| 4393 | rdt_last_cmd_puts("Destination must be a mon_groups subdirectory\n");
|
| 4394 | ret = -EPERM;
|
| 4395 | goto out;
|
| 4396 | }
|
| 4397 |
|
| 4398 | |
| 4399 | * If the MON group is monitoring CPUs, the CPUs must be assigned to the
|
| 4400 | * current parent CTRL_MON group and therefore cannot be assigned to
|
| 4401 | * the new parent, making the move illegal.
|
| 4402 |
|
| 4403 | if (!cpumask_empty(&rdtgrp->cpu_mask) &&
|
| 4404 | rdtgrp->mon.parent != new_prdtgrp) {
|
| 4405 | rdt_last_cmd_puts("Cannot move a MON group that monitors CPUs\n");
|
| 4406 | ret = -EPERM;
|
| 4407 | goto out;
|
| 4408 | }
|
| 4409 |
|
| 4410 | |
| 4411 | * Allocate the cpumask for use in mongrp_reparent() to avoid the
|
| 4412 | * possibility of failing to allocate it after kernfs_rename() has
|
| 4413 | * succeeded.
|
| 4414 |
|
| 4415 | if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) {
|
| 4416 | ret = -ENOMEM;
|
| 4417 | goto out;
|
| 4418 | }
|
| 4419 |
|
| 4420 | |
| 4421 | * Perform all input validation and allocations needed to ensure
|
| 4422 | * mongrp_reparent() will succeed before calling kernfs_rename(),
|
| 4423 | * otherwise it would be necessary to revert this call if
|
| 4424 | * mongrp_reparent() failed.
|
| 4425 |
|
| 4426 | ret = kernfs_rename(kn, new_parent, new_name);
|
| 4427 | if (!ret)
|
| 4428 | mongrp_reparent(rdtgrp, new_prdtgrp, tmpmask);
|
| 4429 |
|
| 4430 | free_cpumask_var(tmpmask);
|
| 4431 |
|
| 4432 | out:
|
| 4433 | mutex_unlock(&rdtgroup_mutex);
|
| 4434 | rdtgroup_kn_put(rdtgrp, kn);
|
| 4435 | rdtgroup_kn_put(new_prdtgrp, new_parent);
|
| 4436 | return ret;
|
| 4437 | }
|
| 4438 |
|
| 4439 | static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
|
| 4440 | {
|
| 4441 | if (resctrl_arch_get_cdp_enabled(RDT_RESOURCE_L3))
|
| 4442 | seq_puts(seq, ",cdp");
|
| 4443 |
|
| 4444 | if (resctrl_arch_get_cdp_enabled(RDT_RESOURCE_L2))
|
| 4445 | seq_puts(seq, ",cdpl2");
|
| 4446 |
|
| 4447 | if (is_mba_sc(resctrl_arch_get_resource(RDT_RESOURCE_MBA)))
|
| 4448 | seq_puts(seq, ",mba_MBps");
|
| 4449 |
|
| 4450 | if (resctrl_debug)
|
| 4451 | seq_puts(seq, ",debug");
|
| 4452 |
|
| 4453 | return 0;
|
| 4454 | }
|
| 4455 |
|
| 4456 | static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
|
| 4457 | .mkdir = rdtgroup_mkdir,
|
| 4458 | .rmdir = rdtgroup_rmdir,
|
| 4459 | .rename = rdtgroup_rename,
|
| 4460 | .show_options = rdtgroup_show_options,
|
| 4461 | };
|
| 4462 |
|
| 4463 | static int rdtgroup_setup_root(struct rdt_fs_context *ctx)
|
| 4464 | {
|
| 4465 | rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
|
| 4466 | KERNFS_ROOT_CREATE_DEACTIVATED |
|
| 4467 | KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
|
| 4468 | &rdtgroup_default);
|
| 4469 | if (IS_ERR(rdt_root))
|
| 4470 | return PTR_ERR(rdt_root);
|
| 4471 |
|
| 4472 | ctx->kfc.root = rdt_root;
|
| 4473 | rdtgroup_default.kn = kernfs_root_to_node(rdt_root);
|
| 4474 | rdtgroup_default.flags = 0;
|
| 4475 |
|
| 4476 | return 0;
|
| 4477 | }
|
| 4478 |
|
| 4479 | static void rdtgroup_destroy_root(void)
|
| 4480 | {
|
| 4481 | lockdep_assert_held(&rdtgroup_mutex);
|
| 4482 |
|
| 4483 | kernfs_destroy_root(rdt_root);
|
| 4484 | rdtgroup_default.kn = NULL;
|
| 4485 | }
|
| 4486 |
|
| 4487 | static void rdtgroup_setup_default(void)
|
| 4488 | {
|
| 4489 | mutex_lock(&rdtgroup_mutex);
|
| 4490 |
|
| 4491 | rdtgroup_default.closid = RESCTRL_RESERVED_CLOSID;
|
| 4492 | rdtgroup_default.mon.rmid = RESCTRL_RESERVED_RMID;
|
| 4493 | rdtgroup_default.type = RDTCTRL_GROUP;
|
| 4494 | INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
|
| 4495 |
|
| 4496 | list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
|
| 4497 |
|
| 4498 | mutex_unlock(&rdtgroup_mutex);
|
| 4499 | }
|
| 4500 |
|
| 4501 | static void domain_destroy_l3_mon_state(struct rdt_l3_mon_domain *d)
|
| 4502 | {
|
| 4503 | int idx;
|
| 4504 |
|
| 4505 | kfree(d->cntr_cfg);
|
| 4506 | bitmap_free(d->rmid_busy_llc);
|
| 4507 | for_each_mbm_idx(idx) {
|
| 4508 | kfree(d->mbm_states[idx]);
|
| 4509 | d->mbm_states[idx] = NULL;
|
| 4510 | }
|
| 4511 | }
|
| 4512 |
|
| 4513 | void resctrl_offline_ctrl_domain(struct rdt_resource *r, struct rdt_ctrl_domain *d)
|
| 4514 | {
|
| 4515 | |
| 4516 | * mbm_handle_overflow() may dereference this ctrl domain via
|
| 4517 | * update_mba_bw()->get_sc_ctrl_domain_from_cpu(). The architecture has
|
| 4518 | * unlinked the domain from the RCU list and waited a grace period, so
|
| 4519 | * no new worker iteration can find it; drain any worker that already
|
| 4520 | * holds a pointer to it before the architecture frees the domain.
|
| 4521 | *
|
| 4522 | * Software controller is enabled/disabled on mount/unmount with
|
| 4523 | * cpus_read_lock() held. Running here with cpus_write_lock() so
|
| 4524 | * there are no concurrent changes to software controller status.
|
| 4525 |
|
| 4526 | if (r->rid == RDT_RESOURCE_MBA && is_mba_sc(r)) {
|
| 4527 | struct rdt_resource *l3 = resctrl_arch_get_resource(RDT_RESOURCE_L3);
|
| 4528 | struct rdt_l3_mon_domain *mon_d;
|
| 4529 |
|
| 4530 | list_for_each_entry_rcu(mon_d, &l3->mon_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 4531 | if (mon_d->hdr.id == d->hdr.id) {
|
| 4532 | cancel_delayed_work_sync(&mon_d->mbm_over);
|
| 4533 | break;
|
| 4534 | }
|
| 4535 | }
|
| 4536 | }
|
| 4537 |
|
| 4538 | mutex_lock(&rdtgroup_mutex);
|
| 4539 |
|
| 4540 | if (supports_mba_mbps() && r->rid == RDT_RESOURCE_MBA)
|
| 4541 | mba_sc_domain_destroy(r, d);
|
| 4542 |
|
| 4543 | mutex_unlock(&rdtgroup_mutex);
|
| 4544 | }
|
| 4545 |
|
| 4546 | void resctrl_offline_mon_domain(struct rdt_resource *r, struct rdt_domain_hdr *hdr)
|
| 4547 | {
|
| 4548 | struct rdt_l3_mon_domain *d;
|
| 4549 |
|
| 4550 | |
| 4551 | * Called by architecture under CPU hotplug lock as it prepares to remove
|
| 4552 | * the domain which is guaranteed to be accessible here.
|
| 4553 | * The domain has been unlinked from the RCU list and a grace period
|
| 4554 | * has elapsed, so no new worker can be scheduled. Drain any worker that
|
| 4555 | * is in flight or pending before letting architecture proceed to free
|
| 4556 | * the domain that has the workers' struct delayed_work embedded.
|
| 4557 | * Do so before taking rdtgroup_mutex since the workers also acquire it.
|
| 4558 |
|
| 4559 | if (r->rid == RDT_RESOURCE_L3 &&
|
| 4560 | domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3)) {
|
| 4561 | d = container_of(hdr, struct rdt_l3_mon_domain, hdr);
|
| 4562 | if (resctrl_is_mbm_enabled())
|
| 4563 | cancel_delayed_work_sync(&d->mbm_over);
|
| 4564 | if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID))
|
| 4565 | cancel_delayed_work_sync(&d->cqm_limbo);
|
| 4566 | }
|
| 4567 |
|
| 4568 | mutex_lock(&rdtgroup_mutex);
|
| 4569 |
|
| 4570 | |
| 4571 | * If resctrl is mounted, remove all the
|
| 4572 | * per domain monitor data directories.
|
| 4573 |
|
| 4574 | if (resctrl_mounted && resctrl_arch_mon_capable())
|
| 4575 | rmdir_mondata_subdir_allrdtgrp(r, hdr);
|
| 4576 |
|
| 4577 | if (r->rid != RDT_RESOURCE_L3)
|
| 4578 | goto out_unlock;
|
| 4579 |
|
| 4580 | if (!domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3))
|
| 4581 | goto out_unlock;
|
| 4582 |
|
| 4583 | d = container_of(hdr, struct rdt_l3_mon_domain, hdr);
|
| 4584 | if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID) && has_busy_rmid(d)) {
|
| 4585 | |
| 4586 | * When a package is going down, forcefully
|
| 4587 | * decrement rmid->ebusy. There is no way to know
|
| 4588 | * that the L3 was flushed and hence may lead to
|
| 4589 | * incorrect counts in rare scenarios, but leaving
|
| 4590 | * the RMID as busy creates RMID leaks if the
|
| 4591 | * package never comes back.
|
| 4592 |
|
| 4593 | __check_limbo(d, true);
|
| 4594 | }
|
| 4595 |
|
| 4596 | domain_destroy_l3_mon_state(d);
|
| 4597 | out_unlock:
|
| 4598 | mutex_unlock(&rdtgroup_mutex);
|
| 4599 | }
|
| 4600 |
|
| 4601 | |
| 4602 | * domain_setup_l3_mon_state() - Initialise domain monitoring structures.
|
| 4603 | * @r: The resource for the newly online domain.
|
| 4604 | * @d: The newly online domain.
|
| 4605 | *
|
| 4606 | * Allocate monitor resources that belong to this domain.
|
| 4607 | * Called when the first CPU of a domain comes online, regardless of whether
|
| 4608 | * the filesystem is mounted.
|
| 4609 | * During boot this may be called before global allocations have been made by
|
| 4610 | * resctrl_l3_mon_resource_init().
|
| 4611 | *
|
| 4612 | * Called during CPU online that may run as soon as CPU online callbacks
|
| 4613 | * are set up during resctrl initialization. The number of supported RMIDs
|
| 4614 | * may be reduced if additional mon_capable resources are enumerated
|
| 4615 | * at mount time. This means the rdt_l3_mon_domain::mbm_states[] and
|
| 4616 | * rdt_l3_mon_domain::rmid_busy_llc allocations may be larger than needed.
|
| 4617 | *
|
| 4618 | * Return: 0 for success, or -ENOMEM.
|
| 4619 |
|
| 4620 | static int domain_setup_l3_mon_state(struct rdt_resource *r, struct rdt_l3_mon_domain *d)
|
| 4621 | {
|
| 4622 | u32 idx_limit = resctrl_arch_system_num_rmid_idx();
|
| 4623 | size_t tsize = sizeof(*d->mbm_states[0]);
|
| 4624 | enum resctrl_event_id eventid;
|
| 4625 | int idx;
|
| 4626 |
|
| 4627 | if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID)) {
|
| 4628 | d->rmid_busy_llc = bitmap_zalloc(idx_limit, GFP_KERNEL);
|
| 4629 | if (!d->rmid_busy_llc)
|
| 4630 | return -ENOMEM;
|
| 4631 | }
|
| 4632 |
|
| 4633 | for_each_mbm_event_id(eventid) {
|
| 4634 | if (!resctrl_is_mon_event_enabled(eventid))
|
| 4635 | continue;
|
| 4636 | idx = MBM_STATE_IDX(eventid);
|
| 4637 | d->mbm_states[idx] = kcalloc(idx_limit, tsize, GFP_KERNEL);
|
| 4638 | if (!d->mbm_states[idx])
|
| 4639 | goto cleanup;
|
| 4640 | }
|
| 4641 |
|
| 4642 | if (resctrl_is_mbm_enabled() && r->mon.mbm_cntr_assignable) {
|
| 4643 | tsize = sizeof(*d->cntr_cfg);
|
| 4644 | d->cntr_cfg = kcalloc(r->mon.num_mbm_cntrs, tsize, GFP_KERNEL);
|
| 4645 | if (!d->cntr_cfg)
|
| 4646 | goto cleanup;
|
| 4647 | }
|
| 4648 |
|
| 4649 | return 0;
|
| 4650 | cleanup:
|
| 4651 | bitmap_free(d->rmid_busy_llc);
|
| 4652 | for_each_mbm_idx(idx) {
|
| 4653 | kfree(d->mbm_states[idx]);
|
| 4654 | d->mbm_states[idx] = NULL;
|
| 4655 | }
|
| 4656 |
|
| 4657 | return -ENOMEM;
|
| 4658 | }
|
| 4659 |
|
| 4660 | int resctrl_online_ctrl_domain(struct rdt_resource *r, struct rdt_ctrl_domain *d)
|
| 4661 | {
|
| 4662 | int err = 0;
|
| 4663 |
|
| 4664 | mutex_lock(&rdtgroup_mutex);
|
| 4665 |
|
| 4666 | if (supports_mba_mbps() && r->rid == RDT_RESOURCE_MBA) {
|
| 4667 |
|
| 4668 | err = mba_sc_domain_allocate(r, d);
|
| 4669 | }
|
| 4670 |
|
| 4671 | mutex_unlock(&rdtgroup_mutex);
|
| 4672 |
|
| 4673 | return err;
|
| 4674 | }
|
| 4675 |
|
| 4676 | int resctrl_online_mon_domain(struct rdt_resource *r, struct rdt_domain_hdr *hdr)
|
| 4677 | {
|
| 4678 | struct rdt_l3_mon_domain *d;
|
| 4679 | int err = -EINVAL;
|
| 4680 |
|
| 4681 | mutex_lock(&rdtgroup_mutex);
|
| 4682 |
|
| 4683 | if (r->rid != RDT_RESOURCE_L3)
|
| 4684 | goto mkdir;
|
| 4685 |
|
| 4686 | if (!domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3))
|
| 4687 | goto out_unlock;
|
| 4688 |
|
| 4689 | d = container_of(hdr, struct rdt_l3_mon_domain, hdr);
|
| 4690 | err = domain_setup_l3_mon_state(r, d);
|
| 4691 | if (err)
|
| 4692 | goto out_unlock;
|
| 4693 |
|
| 4694 | if (resctrl_is_mbm_enabled()) {
|
| 4695 | INIT_DELAYED_WORK(&d->mbm_over, mbm_handle_overflow);
|
| 4696 | mbm_setup_overflow_handler(d, MBM_OVERFLOW_INTERVAL,
|
| 4697 | RESCTRL_PICK_ANY_CPU);
|
| 4698 | }
|
| 4699 |
|
| 4700 | if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID))
|
| 4701 | INIT_DELAYED_WORK(&d->cqm_limbo, cqm_handle_limbo);
|
| 4702 |
|
| 4703 | mkdir:
|
| 4704 | err = 0;
|
| 4705 | |
| 4706 | * If the filesystem is not mounted then only the default resource group
|
| 4707 | * exists. Creation of its directories is deferred until mount time
|
| 4708 | * by rdt_get_tree() calling mkdir_mondata_all().
|
| 4709 | * If resctrl is mounted, add per domain monitor data directories.
|
| 4710 |
|
| 4711 | if (resctrl_mounted && resctrl_arch_mon_capable())
|
| 4712 | mkdir_mondata_subdir_allrdtgrp(r, hdr);
|
| 4713 |
|
| 4714 | out_unlock:
|
| 4715 | mutex_unlock(&rdtgroup_mutex);
|
| 4716 |
|
| 4717 | return err;
|
| 4718 | }
|
| 4719 |
|
| 4720 | void resctrl_online_cpu(unsigned int cpu)
|
| 4721 | {
|
| 4722 | mutex_lock(&rdtgroup_mutex);
|
| 4723 |
|
| 4724 | cpumask_set_cpu(cpu, &rdtgroup_default.cpu_mask);
|
| 4725 | mutex_unlock(&rdtgroup_mutex);
|
| 4726 | }
|
| 4727 |
|
| 4728 | static void clear_childcpus(struct rdtgroup *r, unsigned int cpu)
|
| 4729 | {
|
| 4730 | struct rdtgroup *cr;
|
| 4731 |
|
| 4732 | list_for_each_entry(cr, &r->mon.crdtgrp_list, mon.crdtgrp_list) {
|
| 4733 | if (cpumask_test_and_clear_cpu(cpu, &cr->cpu_mask))
|
| 4734 | break;
|
| 4735 | }
|
| 4736 | }
|
| 4737 |
|
| 4738 | static struct rdt_l3_mon_domain *get_mon_domain_from_cpu(int cpu,
|
| 4739 | struct rdt_resource *r)
|
| 4740 | {
|
| 4741 | struct rdt_l3_mon_domain *d;
|
| 4742 |
|
| 4743 | lockdep_assert_cpus_held();
|
| 4744 |
|
| 4745 | list_for_each_entry_rcu(d, &r->mon_domains, hdr.list, lockdep_is_cpus_held()) {
|
| 4746 |
|
| 4747 | if (cpumask_test_cpu(cpu, &d->hdr.cpu_mask))
|
| 4748 | return d;
|
| 4749 | }
|
| 4750 |
|
| 4751 | return NULL;
|
| 4752 | }
|
| 4753 |
|
| 4754 | void resctrl_offline_cpu(unsigned int cpu)
|
| 4755 | {
|
| 4756 | struct rdt_resource *l3 = resctrl_arch_get_resource(RDT_RESOURCE_L3);
|
| 4757 | struct rdt_l3_mon_domain *d;
|
| 4758 | struct rdtgroup *rdtgrp;
|
| 4759 |
|
| 4760 | mutex_lock(&rdtgroup_mutex);
|
| 4761 | list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
|
| 4762 | if (cpumask_test_and_clear_cpu(cpu, &rdtgrp->cpu_mask)) {
|
| 4763 | clear_childcpus(rdtgrp, cpu);
|
| 4764 | break;
|
| 4765 | }
|
| 4766 | }
|
| 4767 |
|
| 4768 | if (!l3->mon_capable)
|
| 4769 | goto out_unlock;
|
| 4770 |
|
| 4771 | d = get_mon_domain_from_cpu(cpu, l3);
|
| 4772 | if (d) {
|
| 4773 | if (resctrl_is_mbm_enabled() && cpu == d->mbm_work_cpu) {
|
| 4774 | mutex_unlock(&rdtgroup_mutex);
|
| 4775 | cancel_delayed_work_sync(&d->mbm_over);
|
| 4776 | mutex_lock(&rdtgroup_mutex);
|
| 4777 | mbm_setup_overflow_handler(d, 0, cpu);
|
| 4778 | }
|
| 4779 | if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID) &&
|
| 4780 | cpu == d->cqm_work_cpu && has_busy_rmid(d)) {
|
| 4781 | mutex_unlock(&rdtgroup_mutex);
|
| 4782 | cancel_delayed_work_sync(&d->cqm_limbo);
|
| 4783 | mutex_lock(&rdtgroup_mutex);
|
| 4784 | cqm_setup_limbo_handler(d, 0, cpu);
|
| 4785 | }
|
| 4786 | }
|
| 4787 |
|
| 4788 | out_unlock:
|
| 4789 | mutex_unlock(&rdtgroup_mutex);
|
| 4790 | }
|
| 4791 |
|
| 4792 | |
| 4793 | * resctrl_init - resctrl filesystem initialization
|
| 4794 | *
|
| 4795 | * Setup resctrl file system including set up root, create mount point,
|
| 4796 | * register resctrl filesystem, and initialize files under root directory.
|
| 4797 | *
|
| 4798 | * Return: 0 on success or -errno
|
| 4799 |
|
| 4800 | int resctrl_init(void)
|
| 4801 | {
|
| 4802 | int ret = 0;
|
| 4803 |
|
| 4804 | seq_buf_init(&last_cmd_status, last_cmd_status_buf,
|
| 4805 | sizeof(last_cmd_status_buf));
|
| 4806 |
|
| 4807 | rdtgroup_setup_default();
|
| 4808 |
|
| 4809 | thread_throttle_mode_init();
|
| 4810 |
|
| 4811 | io_alloc_init();
|
| 4812 |
|
| 4813 | ret = resctrl_l3_mon_resource_init();
|
| 4814 | if (ret)
|
| 4815 | return ret;
|
| 4816 |
|
| 4817 | ret = sysfs_create_mount_point(fs_kobj, "resctrl");
|
| 4818 | if (ret) {
|
| 4819 | resctrl_l3_mon_resource_exit();
|
| 4820 | return ret;
|
| 4821 | }
|
| 4822 |
|
| 4823 | ret = register_filesystem(&rdt_fs_type);
|
| 4824 | if (ret)
|
| 4825 | goto cleanup_mountpoint;
|
| 4826 |
|
| 4827 | |
| 4828 | * Adding the resctrl debugfs directory here may not be ideal since
|
| 4829 | * it would let the resctrl debugfs directory appear on the debugfs
|
| 4830 | * filesystem before the resctrl filesystem is mounted.
|
| 4831 | * It may also be ok since that would enable debugging of RDT before
|
| 4832 | * resctrl is mounted.
|
| 4833 | * The reason why the debugfs directory is created here and not in
|
| 4834 | * rdt_get_tree() is because rdt_get_tree() takes rdtgroup_mutex and
|
| 4835 | * during the debugfs directory creation also &sb->s_type->i_mutex_key
|
| 4836 | * (the lockdep class of inode->i_rwsem). Other filesystem
|
| 4837 | * interactions (eg. SyS_getdents) have the lock ordering:
|
| 4838 | * &sb->s_type->i_mutex_key --> &mm->mmap_lock
|
| 4839 | * During mmap(), called with &mm->mmap_lock, the rdtgroup_mutex
|
| 4840 | * is taken, thus creating dependency:
|
| 4841 | * &mm->mmap_lock --> rdtgroup_mutex for the latter that can cause
|
| 4842 | * issues considering the other two lock dependencies.
|
| 4843 | * By creating the debugfs directory here we avoid a dependency
|
| 4844 | * that may cause deadlock (even though file operations cannot
|
| 4845 | * occur until the filesystem is mounted, but I do not know how to
|
| 4846 | * tell lockdep that).
|
| 4847 |
|
| 4848 | debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
|
| 4849 |
|
| 4850 | return 0;
|
| 4851 |
|
| 4852 | cleanup_mountpoint:
|
| 4853 | sysfs_remove_mount_point(fs_kobj, "resctrl");
|
| 4854 | resctrl_l3_mon_resource_exit();
|
| 4855 |
|
| 4856 | return ret;
|
| 4857 | }
|
| 4858 |
|
| 4859 | static bool resctrl_online_domains_exist(void)
|
| 4860 | {
|
| 4861 | struct rdt_resource *r;
|
| 4862 |
|
| 4863 | |
| 4864 | * Only walk capable resources to allow resctrl_arch_get_resource()
|
| 4865 | * to return dummy 'not capable' resources.
|
| 4866 |
|
| 4867 | for_each_alloc_capable_rdt_resource(r) {
|
| 4868 | if (!list_empty(&r->ctrl_domains))
|
| 4869 | return true;
|
| 4870 | }
|
| 4871 |
|
| 4872 | for_each_mon_capable_rdt_resource(r) {
|
| 4873 | if (!list_empty(&r->mon_domains))
|
| 4874 | return true;
|
| 4875 | }
|
| 4876 |
|
| 4877 | return false;
|
| 4878 | }
|
| 4879 |
|
| 4880 | |
| 4881 | * resctrl_exit() - Remove the resctrl filesystem and free resources.
|
| 4882 | *
|
| 4883 | * Called by the architecture code in response to a fatal error.
|
| 4884 | * Removes resctrl files and structures from kernfs to prevent further
|
| 4885 | * configuration.
|
| 4886 | *
|
| 4887 | * When called by the architecture code, all CPUs and resctrl domains must be
|
| 4888 | * offline. This ensures the limbo and overflow handlers are not scheduled to
|
| 4889 | * run, meaning the data structures they access can be freed by
|
| 4890 | * resctrl_l3_mon_resource_exit().
|
| 4891 | *
|
| 4892 | * After resctrl_exit() returns, the architecture code should return an
|
| 4893 | * error from all resctrl_arch_ functions that can do this.
|
| 4894 | * resctrl_arch_get_resource() must continue to return struct rdt_resources
|
| 4895 | * with the correct rid field to ensure the filesystem can be unmounted.
|
| 4896 |
|
| 4897 | void resctrl_exit(void)
|
| 4898 | {
|
| 4899 | cpus_read_lock();
|
| 4900 | WARN_ON_ONCE(resctrl_online_domains_exist());
|
| 4901 |
|
| 4902 | mutex_lock(&rdtgroup_mutex);
|
| 4903 | resctrl_fs_teardown();
|
| 4904 | mutex_unlock(&rdtgroup_mutex);
|
| 4905 |
|
| 4906 | cpus_read_unlock();
|
| 4907 |
|
| 4908 | debugfs_remove_recursive(debugfs_resctrl);
|
| 4909 | debugfs_resctrl = NULL;
|
| 4910 | unregister_filesystem(&rdt_fs_type);
|
| 4911 |
|
| 4912 | |
| 4913 | * Do not remove the sysfs mount point added by resctrl_init() so that
|
| 4914 | * it can be used to umount resctrl.
|
| 4915 |
|
| 4916 |
|
| 4917 | resctrl_l3_mon_resource_exit();
|
| 4918 | free_rmid_lru_list();
|
| 4919 | }
|