1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * User interface for Resource Alloction in Resource Director Technology(RDT) 4 * 5 * Copyright (C) 2016 Intel Corporation 6 * 7 * Author: Fenghua Yu <fenghua.yu@intel.com> 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/cacheinfo.h> 16 #include <linux/cpu.h> 17 #include <linux/debugfs.h> 18 #include <linux/fs.h> 19 #include <linux/fs_parser.h> 20 #include <linux/sysfs.h> 21 #include <linux/kernfs.h> 22 #include <linux/seq_buf.h> 23 #include <linux/seq_file.h> 24 #include <linux/sched/signal.h> 25 #include <linux/sched/task.h> 26 #include <linux/slab.h> 27 #include <linux/task_work.h> 28 #include <linux/user_namespace.h> 29 30 #include <uapi/linux/magic.h> 31 32 #include <asm/resctrl_sched.h> 33 #include "internal.h" 34 35 DEFINE_STATIC_KEY_FALSE(rdt_enable_key); 36 DEFINE_STATIC_KEY_FALSE(rdt_mon_enable_key); 37 DEFINE_STATIC_KEY_FALSE(rdt_alloc_enable_key); 38 static struct kernfs_root *rdt_root; 39 struct rdtgroup rdtgroup_default; 40 LIST_HEAD(rdt_all_groups); 41 42 /* Kernel fs node for "info" directory under root */ 43 static struct kernfs_node *kn_info; 44 45 /* Kernel fs node for "mon_groups" directory under root */ 46 static struct kernfs_node *kn_mongrp; 47 48 /* Kernel fs node for "mon_data" directory under root */ 49 static struct kernfs_node *kn_mondata; 50 51 static struct seq_buf last_cmd_status; 52 static char last_cmd_status_buf[512]; 53 54 struct dentry *debugfs_resctrl; 55 56 void rdt_last_cmd_clear(void) 57 { 58 lockdep_assert_held(&rdtgroup_mutex); 59 seq_buf_clear(&last_cmd_status); 60 } 61 62 void rdt_last_cmd_puts(const char *s) 63 { 64 lockdep_assert_held(&rdtgroup_mutex); 65 seq_buf_puts(&last_cmd_status, s); 66 } 67 68 void rdt_last_cmd_printf(const char *fmt, ...) 69 { 70 va_list ap; 71 72 va_start(ap, fmt); 73 lockdep_assert_held(&rdtgroup_mutex); 74 seq_buf_vprintf(&last_cmd_status, fmt, ap); 75 va_end(ap); 76 } 77 78 /* 79 * Trivial allocator for CLOSIDs. Since h/w only supports a small number, 80 * we can keep a bitmap of free CLOSIDs in a single integer. 81 * 82 * Using a global CLOSID across all resources has some advantages and 83 * some drawbacks: 84 * + We can simply set "current->closid" to assign a task to a resource 85 * group. 86 * + Context switch code can avoid extra memory references deciding which 87 * CLOSID to load into the PQR_ASSOC MSR 88 * - We give up some options in configuring resource groups across multi-socket 89 * systems. 90 * - Our choices on how to configure each resource become progressively more 91 * limited as the number of resources grows. 92 */ 93 static int closid_free_map; 94 static int closid_free_map_len; 95 96 int closids_supported(void) 97 { 98 return closid_free_map_len; 99 } 100 101 static void closid_init(void) 102 { 103 struct rdt_resource *r; 104 int rdt_min_closid = 32; 105 106 /* Compute rdt_min_closid across all resources */ 107 for_each_alloc_enabled_rdt_resource(r) 108 rdt_min_closid = min(rdt_min_closid, r->num_closid); 109 110 closid_free_map = BIT_MASK(rdt_min_closid) - 1; 111 112 /* CLOSID 0 is always reserved for the default group */ 113 closid_free_map &= ~1; 114 closid_free_map_len = rdt_min_closid; 115 } 116 117 static int closid_alloc(void) 118 { 119 u32 closid = ffs(closid_free_map); 120 121 if (closid == 0) 122 return -ENOSPC; 123 closid--; 124 closid_free_map &= ~(1 << closid); 125 126 return closid; 127 } 128 129 void closid_free(int closid) 130 { 131 closid_free_map |= 1 << closid; 132 } 133 134 /** 135 * closid_allocated - test if provided closid is in use 136 * @closid: closid to be tested 137 * 138 * Return: true if @closid is currently associated with a resource group, 139 * false if @closid is free 140 */ 141 static bool closid_allocated(unsigned int closid) 142 { 143 return (closid_free_map & (1 << closid)) == 0; 144 } 145 146 /** 147 * rdtgroup_mode_by_closid - Return mode of resource group with closid 148 * @closid: closid if the resource group 149 * 150 * Each resource group is associated with a @closid. Here the mode 151 * of a resource group can be queried by searching for it using its closid. 152 * 153 * Return: mode as &enum rdtgrp_mode of resource group with closid @closid 154 */ 155 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid) 156 { 157 struct rdtgroup *rdtgrp; 158 159 list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) { 160 if (rdtgrp->closid == closid) 161 return rdtgrp->mode; 162 } 163 164 return RDT_NUM_MODES; 165 } 166 167 static const char * const rdt_mode_str[] = { 168 [RDT_MODE_SHAREABLE] = "shareable", 169 [RDT_MODE_EXCLUSIVE] = "exclusive", 170 [RDT_MODE_PSEUDO_LOCKSETUP] = "pseudo-locksetup", 171 [RDT_MODE_PSEUDO_LOCKED] = "pseudo-locked", 172 }; 173 174 /** 175 * rdtgroup_mode_str - Return the string representation of mode 176 * @mode: the resource group mode as &enum rdtgroup_mode 177 * 178 * Return: string representation of valid mode, "unknown" otherwise 179 */ 180 static const char *rdtgroup_mode_str(enum rdtgrp_mode mode) 181 { 182 if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES) 183 return "unknown"; 184 185 return rdt_mode_str[mode]; 186 } 187 188 /* set uid and gid of rdtgroup dirs and files to that of the creator */ 189 static int rdtgroup_kn_set_ugid(struct kernfs_node *kn) 190 { 191 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, 192 .ia_uid = current_fsuid(), 193 .ia_gid = current_fsgid(), }; 194 195 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && 196 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) 197 return 0; 198 199 return kernfs_setattr(kn, &iattr); 200 } 201 202 static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft) 203 { 204 struct kernfs_node *kn; 205 int ret; 206 207 kn = __kernfs_create_file(parent_kn, rft->name, rft->mode, 208 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 209 0, rft->kf_ops, rft, NULL, NULL); 210 if (IS_ERR(kn)) 211 return PTR_ERR(kn); 212 213 ret = rdtgroup_kn_set_ugid(kn); 214 if (ret) { 215 kernfs_remove(kn); 216 return ret; 217 } 218 219 return 0; 220 } 221 222 static int rdtgroup_seqfile_show(struct seq_file *m, void *arg) 223 { 224 struct kernfs_open_file *of = m->private; 225 struct rftype *rft = of->kn->priv; 226 227 if (rft->seq_show) 228 return rft->seq_show(of, m, arg); 229 return 0; 230 } 231 232 static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf, 233 size_t nbytes, loff_t off) 234 { 235 struct rftype *rft = of->kn->priv; 236 237 if (rft->write) 238 return rft->write(of, buf, nbytes, off); 239 240 return -EINVAL; 241 } 242 243 static struct kernfs_ops rdtgroup_kf_single_ops = { 244 .atomic_write_len = PAGE_SIZE, 245 .write = rdtgroup_file_write, 246 .seq_show = rdtgroup_seqfile_show, 247 }; 248 249 static struct kernfs_ops kf_mondata_ops = { 250 .atomic_write_len = PAGE_SIZE, 251 .seq_show = rdtgroup_mondata_show, 252 }; 253 254 static bool is_cpu_list(struct kernfs_open_file *of) 255 { 256 struct rftype *rft = of->kn->priv; 257 258 return rft->flags & RFTYPE_FLAGS_CPUS_LIST; 259 } 260 261 static int rdtgroup_cpus_show(struct kernfs_open_file *of, 262 struct seq_file *s, void *v) 263 { 264 struct rdtgroup *rdtgrp; 265 struct cpumask *mask; 266 int ret = 0; 267 268 rdtgrp = rdtgroup_kn_lock_live(of->kn); 269 270 if (rdtgrp) { 271 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { 272 if (!rdtgrp->plr->d) { 273 rdt_last_cmd_clear(); 274 rdt_last_cmd_puts("Cache domain offline\n"); 275 ret = -ENODEV; 276 } else { 277 mask = &rdtgrp->plr->d->cpu_mask; 278 seq_printf(s, is_cpu_list(of) ? 279 "%*pbl\n" : "%*pb\n", 280 cpumask_pr_args(mask)); 281 } 282 } else { 283 seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n", 284 cpumask_pr_args(&rdtgrp->cpu_mask)); 285 } 286 } else { 287 ret = -ENOENT; 288 } 289 rdtgroup_kn_unlock(of->kn); 290 291 return ret; 292 } 293 294 /* 295 * This is safe against resctrl_sched_in() called from __switch_to() 296 * because __switch_to() is executed with interrupts disabled. A local call 297 * from update_closid_rmid() is proteced against __switch_to() because 298 * preemption is disabled. 299 */ 300 static void update_cpu_closid_rmid(void *info) 301 { 302 struct rdtgroup *r = info; 303 304 if (r) { 305 this_cpu_write(pqr_state.default_closid, r->closid); 306 this_cpu_write(pqr_state.default_rmid, r->mon.rmid); 307 } 308 309 /* 310 * We cannot unconditionally write the MSR because the current 311 * executing task might have its own closid selected. Just reuse 312 * the context switch code. 313 */ 314 resctrl_sched_in(); 315 } 316 317 /* 318 * Update the PGR_ASSOC MSR on all cpus in @cpu_mask, 319 * 320 * Per task closids/rmids must have been set up before calling this function. 321 */ 322 static void 323 update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r) 324 { 325 int cpu = get_cpu(); 326 327 if (cpumask_test_cpu(cpu, cpu_mask)) 328 update_cpu_closid_rmid(r); 329 smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1); 330 put_cpu(); 331 } 332 333 static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, 334 cpumask_var_t tmpmask) 335 { 336 struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp; 337 struct list_head *head; 338 339 /* Check whether cpus belong to parent ctrl group */ 340 cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask); 341 if (cpumask_weight(tmpmask)) { 342 rdt_last_cmd_puts("Can only add CPUs to mongroup that belong to parent\n"); 343 return -EINVAL; 344 } 345 346 /* Check whether cpus are dropped from this group */ 347 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); 348 if (cpumask_weight(tmpmask)) { 349 /* Give any dropped cpus to parent rdtgroup */ 350 cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask); 351 update_closid_rmid(tmpmask, prgrp); 352 } 353 354 /* 355 * If we added cpus, remove them from previous group that owned them 356 * and update per-cpu rmid 357 */ 358 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); 359 if (cpumask_weight(tmpmask)) { 360 head = &prgrp->mon.crdtgrp_list; 361 list_for_each_entry(crgrp, head, mon.crdtgrp_list) { 362 if (crgrp == rdtgrp) 363 continue; 364 cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask, 365 tmpmask); 366 } 367 update_closid_rmid(tmpmask, rdtgrp); 368 } 369 370 /* Done pushing/pulling - update this group with new mask */ 371 cpumask_copy(&rdtgrp->cpu_mask, newmask); 372 373 return 0; 374 } 375 376 static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m) 377 { 378 struct rdtgroup *crgrp; 379 380 cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m); 381 /* update the child mon group masks as well*/ 382 list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list) 383 cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask); 384 } 385 386 static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, 387 cpumask_var_t tmpmask, cpumask_var_t tmpmask1) 388 { 389 struct rdtgroup *r, *crgrp; 390 struct list_head *head; 391 392 /* Check whether cpus are dropped from this group */ 393 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); 394 if (cpumask_weight(tmpmask)) { 395 /* Can't drop from default group */ 396 if (rdtgrp == &rdtgroup_default) { 397 rdt_last_cmd_puts("Can't drop CPUs from default group\n"); 398 return -EINVAL; 399 } 400 401 /* Give any dropped cpus to rdtgroup_default */ 402 cpumask_or(&rdtgroup_default.cpu_mask, 403 &rdtgroup_default.cpu_mask, tmpmask); 404 update_closid_rmid(tmpmask, &rdtgroup_default); 405 } 406 407 /* 408 * If we added cpus, remove them from previous group and 409 * the prev group's child groups that owned them 410 * and update per-cpu closid/rmid. 411 */ 412 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); 413 if (cpumask_weight(tmpmask)) { 414 list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) { 415 if (r == rdtgrp) 416 continue; 417 cpumask_and(tmpmask1, &r->cpu_mask, tmpmask); 418 if (cpumask_weight(tmpmask1)) 419 cpumask_rdtgrp_clear(r, tmpmask1); 420 } 421 update_closid_rmid(tmpmask, rdtgrp); 422 } 423 424 /* Done pushing/pulling - update this group with new mask */ 425 cpumask_copy(&rdtgrp->cpu_mask, newmask); 426 427 /* 428 * Clear child mon group masks since there is a new parent mask 429 * now and update the rmid for the cpus the child lost. 430 */ 431 head = &rdtgrp->mon.crdtgrp_list; 432 list_for_each_entry(crgrp, head, mon.crdtgrp_list) { 433 cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask); 434 update_closid_rmid(tmpmask, rdtgrp); 435 cpumask_clear(&crgrp->cpu_mask); 436 } 437 438 return 0; 439 } 440 441 static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of, 442 char *buf, size_t nbytes, loff_t off) 443 { 444 cpumask_var_t tmpmask, newmask, tmpmask1; 445 struct rdtgroup *rdtgrp; 446 int ret; 447 448 if (!buf) 449 return -EINVAL; 450 451 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) 452 return -ENOMEM; 453 if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) { 454 free_cpumask_var(tmpmask); 455 return -ENOMEM; 456 } 457 if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) { 458 free_cpumask_var(tmpmask); 459 free_cpumask_var(newmask); 460 return -ENOMEM; 461 } 462 463 rdtgrp = rdtgroup_kn_lock_live(of->kn); 464 if (!rdtgrp) { 465 ret = -ENOENT; 466 goto unlock; 467 } 468 469 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || 470 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 471 ret = -EINVAL; 472 rdt_last_cmd_puts("Pseudo-locking in progress\n"); 473 goto unlock; 474 } 475 476 if (is_cpu_list(of)) 477 ret = cpulist_parse(buf, newmask); 478 else 479 ret = cpumask_parse(buf, newmask); 480 481 if (ret) { 482 rdt_last_cmd_puts("Bad CPU list/mask\n"); 483 goto unlock; 484 } 485 486 /* check that user didn't specify any offline cpus */ 487 cpumask_andnot(tmpmask, newmask, cpu_online_mask); 488 if (cpumask_weight(tmpmask)) { 489 ret = -EINVAL; 490 rdt_last_cmd_puts("Can only assign online CPUs\n"); 491 goto unlock; 492 } 493 494 if (rdtgrp->type == RDTCTRL_GROUP) 495 ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1); 496 else if (rdtgrp->type == RDTMON_GROUP) 497 ret = cpus_mon_write(rdtgrp, newmask, tmpmask); 498 else 499 ret = -EINVAL; 500 501 unlock: 502 rdtgroup_kn_unlock(of->kn); 503 free_cpumask_var(tmpmask); 504 free_cpumask_var(newmask); 505 free_cpumask_var(tmpmask1); 506 507 return ret ?: nbytes; 508 } 509 510 struct task_move_callback { 511 struct callback_head work; 512 struct rdtgroup *rdtgrp; 513 }; 514 515 static void move_myself(struct callback_head *head) 516 { 517 struct task_move_callback *callback; 518 struct rdtgroup *rdtgrp; 519 520 callback = container_of(head, struct task_move_callback, work); 521 rdtgrp = callback->rdtgrp; 522 523 /* 524 * If resource group was deleted before this task work callback 525 * was invoked, then assign the task to root group and free the 526 * resource group. 527 */ 528 if (atomic_dec_and_test(&rdtgrp->waitcount) && 529 (rdtgrp->flags & RDT_DELETED)) { 530 current->closid = 0; 531 current->rmid = 0; 532 kfree(rdtgrp); 533 } 534 535 preempt_disable(); 536 /* update PQR_ASSOC MSR to make resource group go into effect */ 537 resctrl_sched_in(); 538 preempt_enable(); 539 540 kfree(callback); 541 } 542 543 static int __rdtgroup_move_task(struct task_struct *tsk, 544 struct rdtgroup *rdtgrp) 545 { 546 struct task_move_callback *callback; 547 int ret; 548 549 callback = kzalloc(sizeof(*callback), GFP_KERNEL); 550 if (!callback) 551 return -ENOMEM; 552 callback->work.func = move_myself; 553 callback->rdtgrp = rdtgrp; 554 555 /* 556 * Take a refcount, so rdtgrp cannot be freed before the 557 * callback has been invoked. 558 */ 559 atomic_inc(&rdtgrp->waitcount); 560 ret = task_work_add(tsk, &callback->work, true); 561 if (ret) { 562 /* 563 * Task is exiting. Drop the refcount and free the callback. 564 * No need to check the refcount as the group cannot be 565 * deleted before the write function unlocks rdtgroup_mutex. 566 */ 567 atomic_dec(&rdtgrp->waitcount); 568 kfree(callback); 569 rdt_last_cmd_puts("Task exited\n"); 570 } else { 571 /* 572 * For ctrl_mon groups move both closid and rmid. 573 * For monitor groups, can move the tasks only from 574 * their parent CTRL group. 575 */ 576 if (rdtgrp->type == RDTCTRL_GROUP) { 577 tsk->closid = rdtgrp->closid; 578 tsk->rmid = rdtgrp->mon.rmid; 579 } else if (rdtgrp->type == RDTMON_GROUP) { 580 if (rdtgrp->mon.parent->closid == tsk->closid) { 581 tsk->rmid = rdtgrp->mon.rmid; 582 } else { 583 rdt_last_cmd_puts("Can't move task to different control group\n"); 584 ret = -EINVAL; 585 } 586 } 587 } 588 return ret; 589 } 590 591 /** 592 * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group 593 * @r: Resource group 594 * 595 * Return: 1 if tasks have been assigned to @r, 0 otherwise 596 */ 597 int rdtgroup_tasks_assigned(struct rdtgroup *r) 598 { 599 struct task_struct *p, *t; 600 int ret = 0; 601 602 lockdep_assert_held(&rdtgroup_mutex); 603 604 rcu_read_lock(); 605 for_each_process_thread(p, t) { 606 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) || 607 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) { 608 ret = 1; 609 break; 610 } 611 } 612 rcu_read_unlock(); 613 614 return ret; 615 } 616 617 static int rdtgroup_task_write_permission(struct task_struct *task, 618 struct kernfs_open_file *of) 619 { 620 const struct cred *tcred = get_task_cred(task); 621 const struct cred *cred = current_cred(); 622 int ret = 0; 623 624 /* 625 * Even if we're attaching all tasks in the thread group, we only 626 * need to check permissions on one of them. 627 */ 628 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && 629 !uid_eq(cred->euid, tcred->uid) && 630 !uid_eq(cred->euid, tcred->suid)) { 631 rdt_last_cmd_printf("No permission to move task %d\n", task->pid); 632 ret = -EPERM; 633 } 634 635 put_cred(tcred); 636 return ret; 637 } 638 639 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp, 640 struct kernfs_open_file *of) 641 { 642 struct task_struct *tsk; 643 int ret; 644 645 rcu_read_lock(); 646 if (pid) { 647 tsk = find_task_by_vpid(pid); 648 if (!tsk) { 649 rcu_read_unlock(); 650 rdt_last_cmd_printf("No task %d\n", pid); 651 return -ESRCH; 652 } 653 } else { 654 tsk = current; 655 } 656 657 get_task_struct(tsk); 658 rcu_read_unlock(); 659 660 ret = rdtgroup_task_write_permission(tsk, of); 661 if (!ret) 662 ret = __rdtgroup_move_task(tsk, rdtgrp); 663 664 put_task_struct(tsk); 665 return ret; 666 } 667 668 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of, 669 char *buf, size_t nbytes, loff_t off) 670 { 671 struct rdtgroup *rdtgrp; 672 int ret = 0; 673 pid_t pid; 674 675 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) 676 return -EINVAL; 677 rdtgrp = rdtgroup_kn_lock_live(of->kn); 678 if (!rdtgrp) { 679 rdtgroup_kn_unlock(of->kn); 680 return -ENOENT; 681 } 682 rdt_last_cmd_clear(); 683 684 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || 685 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 686 ret = -EINVAL; 687 rdt_last_cmd_puts("Pseudo-locking in progress\n"); 688 goto unlock; 689 } 690 691 ret = rdtgroup_move_task(pid, rdtgrp, of); 692 693 unlock: 694 rdtgroup_kn_unlock(of->kn); 695 696 return ret ?: nbytes; 697 } 698 699 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s) 700 { 701 struct task_struct *p, *t; 702 703 rcu_read_lock(); 704 for_each_process_thread(p, t) { 705 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) || 706 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) 707 seq_printf(s, "%d\n", t->pid); 708 } 709 rcu_read_unlock(); 710 } 711 712 static int rdtgroup_tasks_show(struct kernfs_open_file *of, 713 struct seq_file *s, void *v) 714 { 715 struct rdtgroup *rdtgrp; 716 int ret = 0; 717 718 rdtgrp = rdtgroup_kn_lock_live(of->kn); 719 if (rdtgrp) 720 show_rdt_tasks(rdtgrp, s); 721 else 722 ret = -ENOENT; 723 rdtgroup_kn_unlock(of->kn); 724 725 return ret; 726 } 727 728 static int rdt_last_cmd_status_show(struct kernfs_open_file *of, 729 struct seq_file *seq, void *v) 730 { 731 int len; 732 733 mutex_lock(&rdtgroup_mutex); 734 len = seq_buf_used(&last_cmd_status); 735 if (len) 736 seq_printf(seq, "%.*s", len, last_cmd_status_buf); 737 else 738 seq_puts(seq, "ok\n"); 739 mutex_unlock(&rdtgroup_mutex); 740 return 0; 741 } 742 743 static int rdt_num_closids_show(struct kernfs_open_file *of, 744 struct seq_file *seq, void *v) 745 { 746 struct rdt_resource *r = of->kn->parent->priv; 747 748 seq_printf(seq, "%d\n", r->num_closid); 749 return 0; 750 } 751 752 static int rdt_default_ctrl_show(struct kernfs_open_file *of, 753 struct seq_file *seq, void *v) 754 { 755 struct rdt_resource *r = of->kn->parent->priv; 756 757 seq_printf(seq, "%x\n", r->default_ctrl); 758 return 0; 759 } 760 761 static int rdt_min_cbm_bits_show(struct kernfs_open_file *of, 762 struct seq_file *seq, void *v) 763 { 764 struct rdt_resource *r = of->kn->parent->priv; 765 766 seq_printf(seq, "%u\n", r->cache.min_cbm_bits); 767 return 0; 768 } 769 770 static int rdt_shareable_bits_show(struct kernfs_open_file *of, 771 struct seq_file *seq, void *v) 772 { 773 struct rdt_resource *r = of->kn->parent->priv; 774 775 seq_printf(seq, "%x\n", r->cache.shareable_bits); 776 return 0; 777 } 778 779 /** 780 * rdt_bit_usage_show - Display current usage of resources 781 * 782 * A domain is a shared resource that can now be allocated differently. Here 783 * we display the current regions of the domain as an annotated bitmask. 784 * For each domain of this resource its allocation bitmask 785 * is annotated as below to indicate the current usage of the corresponding bit: 786 * 0 - currently unused 787 * X - currently available for sharing and used by software and hardware 788 * H - currently used by hardware only but available for software use 789 * S - currently used and shareable by software only 790 * E - currently used exclusively by one resource group 791 * P - currently pseudo-locked by one resource group 792 */ 793 static int rdt_bit_usage_show(struct kernfs_open_file *of, 794 struct seq_file *seq, void *v) 795 { 796 struct rdt_resource *r = of->kn->parent->priv; 797 /* 798 * Use unsigned long even though only 32 bits are used to ensure 799 * test_bit() is used safely. 800 */ 801 unsigned long sw_shareable = 0, hw_shareable = 0; 802 unsigned long exclusive = 0, pseudo_locked = 0; 803 struct rdt_domain *dom; 804 int i, hwb, swb, excl, psl; 805 enum rdtgrp_mode mode; 806 bool sep = false; 807 u32 *ctrl; 808 809 mutex_lock(&rdtgroup_mutex); 810 hw_shareable = r->cache.shareable_bits; 811 list_for_each_entry(dom, &r->domains, list) { 812 if (sep) 813 seq_putc(seq, ';'); 814 ctrl = dom->ctrl_val; 815 sw_shareable = 0; 816 exclusive = 0; 817 seq_printf(seq, "%d=", dom->id); 818 for (i = 0; i < closids_supported(); i++, ctrl++) { 819 if (!closid_allocated(i)) 820 continue; 821 mode = rdtgroup_mode_by_closid(i); 822 switch (mode) { 823 case RDT_MODE_SHAREABLE: 824 sw_shareable |= *ctrl; 825 break; 826 case RDT_MODE_EXCLUSIVE: 827 exclusive |= *ctrl; 828 break; 829 case RDT_MODE_PSEUDO_LOCKSETUP: 830 /* 831 * RDT_MODE_PSEUDO_LOCKSETUP is possible 832 * here but not included since the CBM 833 * associated with this CLOSID in this mode 834 * is not initialized and no task or cpu can be 835 * assigned this CLOSID. 836 */ 837 break; 838 case RDT_MODE_PSEUDO_LOCKED: 839 case RDT_NUM_MODES: 840 WARN(1, 841 "invalid mode for closid %d\n", i); 842 break; 843 } 844 } 845 for (i = r->cache.cbm_len - 1; i >= 0; i--) { 846 pseudo_locked = dom->plr ? dom->plr->cbm : 0; 847 hwb = test_bit(i, &hw_shareable); 848 swb = test_bit(i, &sw_shareable); 849 excl = test_bit(i, &exclusive); 850 psl = test_bit(i, &pseudo_locked); 851 if (hwb && swb) 852 seq_putc(seq, 'X'); 853 else if (hwb && !swb) 854 seq_putc(seq, 'H'); 855 else if (!hwb && swb) 856 seq_putc(seq, 'S'); 857 else if (excl) 858 seq_putc(seq, 'E'); 859 else if (psl) 860 seq_putc(seq, 'P'); 861 else /* Unused bits remain */ 862 seq_putc(seq, '0'); 863 } 864 sep = true; 865 } 866 seq_putc(seq, '\n'); 867 mutex_unlock(&rdtgroup_mutex); 868 return 0; 869 } 870 871 static int rdt_min_bw_show(struct kernfs_open_file *of, 872 struct seq_file *seq, void *v) 873 { 874 struct rdt_resource *r = of->kn->parent->priv; 875 876 seq_printf(seq, "%u\n", r->membw.min_bw); 877 return 0; 878 } 879 880 static int rdt_num_rmids_show(struct kernfs_open_file *of, 881 struct seq_file *seq, void *v) 882 { 883 struct rdt_resource *r = of->kn->parent->priv; 884 885 seq_printf(seq, "%d\n", r->num_rmid); 886 887 return 0; 888 } 889 890 static int rdt_mon_features_show(struct kernfs_open_file *of, 891 struct seq_file *seq, void *v) 892 { 893 struct rdt_resource *r = of->kn->parent->priv; 894 struct mon_evt *mevt; 895 896 list_for_each_entry(mevt, &r->evt_list, list) 897 seq_printf(seq, "%s\n", mevt->name); 898 899 return 0; 900 } 901 902 static int rdt_bw_gran_show(struct kernfs_open_file *of, 903 struct seq_file *seq, void *v) 904 { 905 struct rdt_resource *r = of->kn->parent->priv; 906 907 seq_printf(seq, "%u\n", r->membw.bw_gran); 908 return 0; 909 } 910 911 static int rdt_delay_linear_show(struct kernfs_open_file *of, 912 struct seq_file *seq, void *v) 913 { 914 struct rdt_resource *r = of->kn->parent->priv; 915 916 seq_printf(seq, "%u\n", r->membw.delay_linear); 917 return 0; 918 } 919 920 static int max_threshold_occ_show(struct kernfs_open_file *of, 921 struct seq_file *seq, void *v) 922 { 923 struct rdt_resource *r = of->kn->parent->priv; 924 925 seq_printf(seq, "%u\n", resctrl_cqm_threshold * r->mon_scale); 926 927 return 0; 928 } 929 930 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of, 931 char *buf, size_t nbytes, loff_t off) 932 { 933 struct rdt_resource *r = of->kn->parent->priv; 934 unsigned int bytes; 935 int ret; 936 937 ret = kstrtouint(buf, 0, &bytes); 938 if (ret) 939 return ret; 940 941 if (bytes > (boot_cpu_data.x86_cache_size * 1024)) 942 return -EINVAL; 943 944 resctrl_cqm_threshold = bytes / r->mon_scale; 945 946 return nbytes; 947 } 948 949 /* 950 * rdtgroup_mode_show - Display mode of this resource group 951 */ 952 static int rdtgroup_mode_show(struct kernfs_open_file *of, 953 struct seq_file *s, void *v) 954 { 955 struct rdtgroup *rdtgrp; 956 957 rdtgrp = rdtgroup_kn_lock_live(of->kn); 958 if (!rdtgrp) { 959 rdtgroup_kn_unlock(of->kn); 960 return -ENOENT; 961 } 962 963 seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode)); 964 965 rdtgroup_kn_unlock(of->kn); 966 return 0; 967 } 968 969 /** 970 * rdt_cdp_peer_get - Retrieve CDP peer if it exists 971 * @r: RDT resource to which RDT domain @d belongs 972 * @d: Cache instance for which a CDP peer is requested 973 * @r_cdp: RDT resource that shares hardware with @r (RDT resource peer) 974 * Used to return the result. 975 * @d_cdp: RDT domain that shares hardware with @d (RDT domain peer) 976 * Used to return the result. 977 * 978 * RDT resources are managed independently and by extension the RDT domains 979 * (RDT resource instances) are managed independently also. The Code and 980 * Data Prioritization (CDP) RDT resources, while managed independently, 981 * could refer to the same underlying hardware. For example, 982 * RDT_RESOURCE_L2CODE and RDT_RESOURCE_L2DATA both refer to the L2 cache. 983 * 984 * When provided with an RDT resource @r and an instance of that RDT 985 * resource @d rdt_cdp_peer_get() will return if there is a peer RDT 986 * resource and the exact instance that shares the same hardware. 987 * 988 * Return: 0 if a CDP peer was found, <0 on error or if no CDP peer exists. 989 * If a CDP peer was found, @r_cdp will point to the peer RDT resource 990 * and @d_cdp will point to the peer RDT domain. 991 */ 992 static int rdt_cdp_peer_get(struct rdt_resource *r, struct rdt_domain *d, 993 struct rdt_resource **r_cdp, 994 struct rdt_domain **d_cdp) 995 { 996 struct rdt_resource *_r_cdp = NULL; 997 struct rdt_domain *_d_cdp = NULL; 998 int ret = 0; 999 1000 switch (r->rid) { 1001 case RDT_RESOURCE_L3DATA: 1002 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3CODE]; 1003 break; 1004 case RDT_RESOURCE_L3CODE: 1005 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3DATA]; 1006 break; 1007 case RDT_RESOURCE_L2DATA: 1008 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2CODE]; 1009 break; 1010 case RDT_RESOURCE_L2CODE: 1011 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2DATA]; 1012 break; 1013 default: 1014 ret = -ENOENT; 1015 goto out; 1016 } 1017 1018 /* 1019 * When a new CPU comes online and CDP is enabled then the new 1020 * RDT domains (if any) associated with both CDP RDT resources 1021 * are added in the same CPU online routine while the 1022 * rdtgroup_mutex is held. It should thus not happen for one 1023 * RDT domain to exist and be associated with its RDT CDP 1024 * resource but there is no RDT domain associated with the 1025 * peer RDT CDP resource. Hence the WARN. 1026 */ 1027 _d_cdp = rdt_find_domain(_r_cdp, d->id, NULL); 1028 if (WARN_ON(IS_ERR_OR_NULL(_d_cdp))) { 1029 _r_cdp = NULL; 1030 ret = -EINVAL; 1031 } 1032 1033 out: 1034 *r_cdp = _r_cdp; 1035 *d_cdp = _d_cdp; 1036 1037 return ret; 1038 } 1039 1040 /** 1041 * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other 1042 * @r: Resource to which domain instance @d belongs. 1043 * @d: The domain instance for which @closid is being tested. 1044 * @cbm: Capacity bitmask being tested. 1045 * @closid: Intended closid for @cbm. 1046 * @exclusive: Only check if overlaps with exclusive resource groups 1047 * 1048 * Checks if provided @cbm intended to be used for @closid on domain 1049 * @d overlaps with any other closids or other hardware usage associated 1050 * with this domain. If @exclusive is true then only overlaps with 1051 * resource groups in exclusive mode will be considered. If @exclusive 1052 * is false then overlaps with any resource group or hardware entities 1053 * will be considered. 1054 * 1055 * @cbm is unsigned long, even if only 32 bits are used, to make the 1056 * bitmap functions work correctly. 1057 * 1058 * Return: false if CBM does not overlap, true if it does. 1059 */ 1060 static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d, 1061 unsigned long cbm, int closid, bool exclusive) 1062 { 1063 enum rdtgrp_mode mode; 1064 unsigned long ctrl_b; 1065 u32 *ctrl; 1066 int i; 1067 1068 /* Check for any overlap with regions used by hardware directly */ 1069 if (!exclusive) { 1070 ctrl_b = r->cache.shareable_bits; 1071 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) 1072 return true; 1073 } 1074 1075 /* Check for overlap with other resource groups */ 1076 ctrl = d->ctrl_val; 1077 for (i = 0; i < closids_supported(); i++, ctrl++) { 1078 ctrl_b = *ctrl; 1079 mode = rdtgroup_mode_by_closid(i); 1080 if (closid_allocated(i) && i != closid && 1081 mode != RDT_MODE_PSEUDO_LOCKSETUP) { 1082 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) { 1083 if (exclusive) { 1084 if (mode == RDT_MODE_EXCLUSIVE) 1085 return true; 1086 continue; 1087 } 1088 return true; 1089 } 1090 } 1091 } 1092 1093 return false; 1094 } 1095 1096 /** 1097 * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware 1098 * @r: Resource to which domain instance @d belongs. 1099 * @d: The domain instance for which @closid is being tested. 1100 * @cbm: Capacity bitmask being tested. 1101 * @closid: Intended closid for @cbm. 1102 * @exclusive: Only check if overlaps with exclusive resource groups 1103 * 1104 * Resources that can be allocated using a CBM can use the CBM to control 1105 * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test 1106 * for overlap. Overlap test is not limited to the specific resource for 1107 * which the CBM is intended though - when dealing with CDP resources that 1108 * share the underlying hardware the overlap check should be performed on 1109 * the CDP resource sharing the hardware also. 1110 * 1111 * Refer to description of __rdtgroup_cbm_overlaps() for the details of the 1112 * overlap test. 1113 * 1114 * Return: true if CBM overlap detected, false if there is no overlap 1115 */ 1116 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d, 1117 unsigned long cbm, int closid, bool exclusive) 1118 { 1119 struct rdt_resource *r_cdp; 1120 struct rdt_domain *d_cdp; 1121 1122 if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, exclusive)) 1123 return true; 1124 1125 if (rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp) < 0) 1126 return false; 1127 1128 return __rdtgroup_cbm_overlaps(r_cdp, d_cdp, cbm, closid, exclusive); 1129 } 1130 1131 /** 1132 * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive 1133 * 1134 * An exclusive resource group implies that there should be no sharing of 1135 * its allocated resources. At the time this group is considered to be 1136 * exclusive this test can determine if its current schemata supports this 1137 * setting by testing for overlap with all other resource groups. 1138 * 1139 * Return: true if resource group can be exclusive, false if there is overlap 1140 * with allocations of other resource groups and thus this resource group 1141 * cannot be exclusive. 1142 */ 1143 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp) 1144 { 1145 int closid = rdtgrp->closid; 1146 struct rdt_resource *r; 1147 bool has_cache = false; 1148 struct rdt_domain *d; 1149 1150 for_each_alloc_enabled_rdt_resource(r) { 1151 if (r->rid == RDT_RESOURCE_MBA) 1152 continue; 1153 has_cache = true; 1154 list_for_each_entry(d, &r->domains, list) { 1155 if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid], 1156 rdtgrp->closid, false)) { 1157 rdt_last_cmd_puts("Schemata overlaps\n"); 1158 return false; 1159 } 1160 } 1161 } 1162 1163 if (!has_cache) { 1164 rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n"); 1165 return false; 1166 } 1167 1168 return true; 1169 } 1170 1171 /** 1172 * rdtgroup_mode_write - Modify the resource group's mode 1173 * 1174 */ 1175 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of, 1176 char *buf, size_t nbytes, loff_t off) 1177 { 1178 struct rdtgroup *rdtgrp; 1179 enum rdtgrp_mode mode; 1180 int ret = 0; 1181 1182 /* Valid input requires a trailing newline */ 1183 if (nbytes == 0 || buf[nbytes - 1] != '\n') 1184 return -EINVAL; 1185 buf[nbytes - 1] = '\0'; 1186 1187 rdtgrp = rdtgroup_kn_lock_live(of->kn); 1188 if (!rdtgrp) { 1189 rdtgroup_kn_unlock(of->kn); 1190 return -ENOENT; 1191 } 1192 1193 rdt_last_cmd_clear(); 1194 1195 mode = rdtgrp->mode; 1196 1197 if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) || 1198 (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) || 1199 (!strcmp(buf, "pseudo-locksetup") && 1200 mode == RDT_MODE_PSEUDO_LOCKSETUP) || 1201 (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED)) 1202 goto out; 1203 1204 if (mode == RDT_MODE_PSEUDO_LOCKED) { 1205 rdt_last_cmd_puts("Cannot change pseudo-locked group\n"); 1206 ret = -EINVAL; 1207 goto out; 1208 } 1209 1210 if (!strcmp(buf, "shareable")) { 1211 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 1212 ret = rdtgroup_locksetup_exit(rdtgrp); 1213 if (ret) 1214 goto out; 1215 } 1216 rdtgrp->mode = RDT_MODE_SHAREABLE; 1217 } else if (!strcmp(buf, "exclusive")) { 1218 if (!rdtgroup_mode_test_exclusive(rdtgrp)) { 1219 ret = -EINVAL; 1220 goto out; 1221 } 1222 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 1223 ret = rdtgroup_locksetup_exit(rdtgrp); 1224 if (ret) 1225 goto out; 1226 } 1227 rdtgrp->mode = RDT_MODE_EXCLUSIVE; 1228 } else if (!strcmp(buf, "pseudo-locksetup")) { 1229 ret = rdtgroup_locksetup_enter(rdtgrp); 1230 if (ret) 1231 goto out; 1232 rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP; 1233 } else { 1234 rdt_last_cmd_puts("Unknown or unsupported mode\n"); 1235 ret = -EINVAL; 1236 } 1237 1238 out: 1239 rdtgroup_kn_unlock(of->kn); 1240 return ret ?: nbytes; 1241 } 1242 1243 /** 1244 * rdtgroup_cbm_to_size - Translate CBM to size in bytes 1245 * @r: RDT resource to which @d belongs. 1246 * @d: RDT domain instance. 1247 * @cbm: bitmask for which the size should be computed. 1248 * 1249 * The bitmask provided associated with the RDT domain instance @d will be 1250 * translated into how many bytes it represents. The size in bytes is 1251 * computed by first dividing the total cache size by the CBM length to 1252 * determine how many bytes each bit in the bitmask represents. The result 1253 * is multiplied with the number of bits set in the bitmask. 1254 * 1255 * @cbm is unsigned long, even if only 32 bits are used to make the 1256 * bitmap functions work correctly. 1257 */ 1258 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, 1259 struct rdt_domain *d, unsigned long cbm) 1260 { 1261 struct cpu_cacheinfo *ci; 1262 unsigned int size = 0; 1263 int num_b, i; 1264 1265 num_b = bitmap_weight(&cbm, r->cache.cbm_len); 1266 ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask)); 1267 for (i = 0; i < ci->num_leaves; i++) { 1268 if (ci->info_list[i].level == r->cache_level) { 1269 size = ci->info_list[i].size / r->cache.cbm_len * num_b; 1270 break; 1271 } 1272 } 1273 1274 return size; 1275 } 1276 1277 /** 1278 * rdtgroup_size_show - Display size in bytes of allocated regions 1279 * 1280 * The "size" file mirrors the layout of the "schemata" file, printing the 1281 * size in bytes of each region instead of the capacity bitmask. 1282 * 1283 */ 1284 static int rdtgroup_size_show(struct kernfs_open_file *of, 1285 struct seq_file *s, void *v) 1286 { 1287 struct rdtgroup *rdtgrp; 1288 struct rdt_resource *r; 1289 struct rdt_domain *d; 1290 unsigned int size; 1291 int ret = 0; 1292 bool sep; 1293 u32 ctrl; 1294 1295 rdtgrp = rdtgroup_kn_lock_live(of->kn); 1296 if (!rdtgrp) { 1297 rdtgroup_kn_unlock(of->kn); 1298 return -ENOENT; 1299 } 1300 1301 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { 1302 if (!rdtgrp->plr->d) { 1303 rdt_last_cmd_clear(); 1304 rdt_last_cmd_puts("Cache domain offline\n"); 1305 ret = -ENODEV; 1306 } else { 1307 seq_printf(s, "%*s:", max_name_width, 1308 rdtgrp->plr->r->name); 1309 size = rdtgroup_cbm_to_size(rdtgrp->plr->r, 1310 rdtgrp->plr->d, 1311 rdtgrp->plr->cbm); 1312 seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size); 1313 } 1314 goto out; 1315 } 1316 1317 for_each_alloc_enabled_rdt_resource(r) { 1318 sep = false; 1319 seq_printf(s, "%*s:", max_name_width, r->name); 1320 list_for_each_entry(d, &r->domains, list) { 1321 if (sep) 1322 seq_putc(s, ';'); 1323 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 1324 size = 0; 1325 } else { 1326 ctrl = (!is_mba_sc(r) ? 1327 d->ctrl_val[rdtgrp->closid] : 1328 d->mbps_val[rdtgrp->closid]); 1329 if (r->rid == RDT_RESOURCE_MBA) 1330 size = ctrl; 1331 else 1332 size = rdtgroup_cbm_to_size(r, d, ctrl); 1333 } 1334 seq_printf(s, "%d=%u", d->id, size); 1335 sep = true; 1336 } 1337 seq_putc(s, '\n'); 1338 } 1339 1340 out: 1341 rdtgroup_kn_unlock(of->kn); 1342 1343 return ret; 1344 } 1345 1346 /* rdtgroup information files for one cache resource. */ 1347 static struct rftype res_common_files[] = { 1348 { 1349 .name = "last_cmd_status", 1350 .mode = 0444, 1351 .kf_ops = &rdtgroup_kf_single_ops, 1352 .seq_show = rdt_last_cmd_status_show, 1353 .fflags = RF_TOP_INFO, 1354 }, 1355 { 1356 .name = "num_closids", 1357 .mode = 0444, 1358 .kf_ops = &rdtgroup_kf_single_ops, 1359 .seq_show = rdt_num_closids_show, 1360 .fflags = RF_CTRL_INFO, 1361 }, 1362 { 1363 .name = "mon_features", 1364 .mode = 0444, 1365 .kf_ops = &rdtgroup_kf_single_ops, 1366 .seq_show = rdt_mon_features_show, 1367 .fflags = RF_MON_INFO, 1368 }, 1369 { 1370 .name = "num_rmids", 1371 .mode = 0444, 1372 .kf_ops = &rdtgroup_kf_single_ops, 1373 .seq_show = rdt_num_rmids_show, 1374 .fflags = RF_MON_INFO, 1375 }, 1376 { 1377 .name = "cbm_mask", 1378 .mode = 0444, 1379 .kf_ops = &rdtgroup_kf_single_ops, 1380 .seq_show = rdt_default_ctrl_show, 1381 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, 1382 }, 1383 { 1384 .name = "min_cbm_bits", 1385 .mode = 0444, 1386 .kf_ops = &rdtgroup_kf_single_ops, 1387 .seq_show = rdt_min_cbm_bits_show, 1388 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, 1389 }, 1390 { 1391 .name = "shareable_bits", 1392 .mode = 0444, 1393 .kf_ops = &rdtgroup_kf_single_ops, 1394 .seq_show = rdt_shareable_bits_show, 1395 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, 1396 }, 1397 { 1398 .name = "bit_usage", 1399 .mode = 0444, 1400 .kf_ops = &rdtgroup_kf_single_ops, 1401 .seq_show = rdt_bit_usage_show, 1402 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, 1403 }, 1404 { 1405 .name = "min_bandwidth", 1406 .mode = 0444, 1407 .kf_ops = &rdtgroup_kf_single_ops, 1408 .seq_show = rdt_min_bw_show, 1409 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, 1410 }, 1411 { 1412 .name = "bandwidth_gran", 1413 .mode = 0444, 1414 .kf_ops = &rdtgroup_kf_single_ops, 1415 .seq_show = rdt_bw_gran_show, 1416 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, 1417 }, 1418 { 1419 .name = "delay_linear", 1420 .mode = 0444, 1421 .kf_ops = &rdtgroup_kf_single_ops, 1422 .seq_show = rdt_delay_linear_show, 1423 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, 1424 }, 1425 { 1426 .name = "max_threshold_occupancy", 1427 .mode = 0644, 1428 .kf_ops = &rdtgroup_kf_single_ops, 1429 .write = max_threshold_occ_write, 1430 .seq_show = max_threshold_occ_show, 1431 .fflags = RF_MON_INFO | RFTYPE_RES_CACHE, 1432 }, 1433 { 1434 .name = "cpus", 1435 .mode = 0644, 1436 .kf_ops = &rdtgroup_kf_single_ops, 1437 .write = rdtgroup_cpus_write, 1438 .seq_show = rdtgroup_cpus_show, 1439 .fflags = RFTYPE_BASE, 1440 }, 1441 { 1442 .name = "cpus_list", 1443 .mode = 0644, 1444 .kf_ops = &rdtgroup_kf_single_ops, 1445 .write = rdtgroup_cpus_write, 1446 .seq_show = rdtgroup_cpus_show, 1447 .flags = RFTYPE_FLAGS_CPUS_LIST, 1448 .fflags = RFTYPE_BASE, 1449 }, 1450 { 1451 .name = "tasks", 1452 .mode = 0644, 1453 .kf_ops = &rdtgroup_kf_single_ops, 1454 .write = rdtgroup_tasks_write, 1455 .seq_show = rdtgroup_tasks_show, 1456 .fflags = RFTYPE_BASE, 1457 }, 1458 { 1459 .name = "schemata", 1460 .mode = 0644, 1461 .kf_ops = &rdtgroup_kf_single_ops, 1462 .write = rdtgroup_schemata_write, 1463 .seq_show = rdtgroup_schemata_show, 1464 .fflags = RF_CTRL_BASE, 1465 }, 1466 { 1467 .name = "mode", 1468 .mode = 0644, 1469 .kf_ops = &rdtgroup_kf_single_ops, 1470 .write = rdtgroup_mode_write, 1471 .seq_show = rdtgroup_mode_show, 1472 .fflags = RF_CTRL_BASE, 1473 }, 1474 { 1475 .name = "size", 1476 .mode = 0444, 1477 .kf_ops = &rdtgroup_kf_single_ops, 1478 .seq_show = rdtgroup_size_show, 1479 .fflags = RF_CTRL_BASE, 1480 }, 1481 1482 }; 1483 1484 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags) 1485 { 1486 struct rftype *rfts, *rft; 1487 int ret, len; 1488 1489 rfts = res_common_files; 1490 len = ARRAY_SIZE(res_common_files); 1491 1492 lockdep_assert_held(&rdtgroup_mutex); 1493 1494 for (rft = rfts; rft < rfts + len; rft++) { 1495 if ((fflags & rft->fflags) == rft->fflags) { 1496 ret = rdtgroup_add_file(kn, rft); 1497 if (ret) 1498 goto error; 1499 } 1500 } 1501 1502 return 0; 1503 error: 1504 pr_warn("Failed to add %s, err=%d\n", rft->name, ret); 1505 while (--rft >= rfts) { 1506 if ((fflags & rft->fflags) == rft->fflags) 1507 kernfs_remove_by_name(kn, rft->name); 1508 } 1509 return ret; 1510 } 1511 1512 /** 1513 * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file 1514 * @r: The resource group with which the file is associated. 1515 * @name: Name of the file 1516 * 1517 * The permissions of named resctrl file, directory, or link are modified 1518 * to not allow read, write, or execute by any user. 1519 * 1520 * WARNING: This function is intended to communicate to the user that the 1521 * resctrl file has been locked down - that it is not relevant to the 1522 * particular state the system finds itself in. It should not be relied 1523 * on to protect from user access because after the file's permissions 1524 * are restricted the user can still change the permissions using chmod 1525 * from the command line. 1526 * 1527 * Return: 0 on success, <0 on failure. 1528 */ 1529 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name) 1530 { 1531 struct iattr iattr = {.ia_valid = ATTR_MODE,}; 1532 struct kernfs_node *kn; 1533 int ret = 0; 1534 1535 kn = kernfs_find_and_get_ns(r->kn, name, NULL); 1536 if (!kn) 1537 return -ENOENT; 1538 1539 switch (kernfs_type(kn)) { 1540 case KERNFS_DIR: 1541 iattr.ia_mode = S_IFDIR; 1542 break; 1543 case KERNFS_FILE: 1544 iattr.ia_mode = S_IFREG; 1545 break; 1546 case KERNFS_LINK: 1547 iattr.ia_mode = S_IFLNK; 1548 break; 1549 } 1550 1551 ret = kernfs_setattr(kn, &iattr); 1552 kernfs_put(kn); 1553 return ret; 1554 } 1555 1556 /** 1557 * rdtgroup_kn_mode_restore - Restore user access to named resctrl file 1558 * @r: The resource group with which the file is associated. 1559 * @name: Name of the file 1560 * @mask: Mask of permissions that should be restored 1561 * 1562 * Restore the permissions of the named file. If @name is a directory the 1563 * permissions of its parent will be used. 1564 * 1565 * Return: 0 on success, <0 on failure. 1566 */ 1567 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name, 1568 umode_t mask) 1569 { 1570 struct iattr iattr = {.ia_valid = ATTR_MODE,}; 1571 struct kernfs_node *kn, *parent; 1572 struct rftype *rfts, *rft; 1573 int ret, len; 1574 1575 rfts = res_common_files; 1576 len = ARRAY_SIZE(res_common_files); 1577 1578 for (rft = rfts; rft < rfts + len; rft++) { 1579 if (!strcmp(rft->name, name)) 1580 iattr.ia_mode = rft->mode & mask; 1581 } 1582 1583 kn = kernfs_find_and_get_ns(r->kn, name, NULL); 1584 if (!kn) 1585 return -ENOENT; 1586 1587 switch (kernfs_type(kn)) { 1588 case KERNFS_DIR: 1589 parent = kernfs_get_parent(kn); 1590 if (parent) { 1591 iattr.ia_mode |= parent->mode; 1592 kernfs_put(parent); 1593 } 1594 iattr.ia_mode |= S_IFDIR; 1595 break; 1596 case KERNFS_FILE: 1597 iattr.ia_mode |= S_IFREG; 1598 break; 1599 case KERNFS_LINK: 1600 iattr.ia_mode |= S_IFLNK; 1601 break; 1602 } 1603 1604 ret = kernfs_setattr(kn, &iattr); 1605 kernfs_put(kn); 1606 return ret; 1607 } 1608 1609 static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name, 1610 unsigned long fflags) 1611 { 1612 struct kernfs_node *kn_subdir; 1613 int ret; 1614 1615 kn_subdir = kernfs_create_dir(kn_info, name, 1616 kn_info->mode, r); 1617 if (IS_ERR(kn_subdir)) 1618 return PTR_ERR(kn_subdir); 1619 1620 kernfs_get(kn_subdir); 1621 ret = rdtgroup_kn_set_ugid(kn_subdir); 1622 if (ret) 1623 return ret; 1624 1625 ret = rdtgroup_add_files(kn_subdir, fflags); 1626 if (!ret) 1627 kernfs_activate(kn_subdir); 1628 1629 return ret; 1630 } 1631 1632 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn) 1633 { 1634 struct rdt_resource *r; 1635 unsigned long fflags; 1636 char name[32]; 1637 int ret; 1638 1639 /* create the directory */ 1640 kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL); 1641 if (IS_ERR(kn_info)) 1642 return PTR_ERR(kn_info); 1643 kernfs_get(kn_info); 1644 1645 ret = rdtgroup_add_files(kn_info, RF_TOP_INFO); 1646 if (ret) 1647 goto out_destroy; 1648 1649 for_each_alloc_enabled_rdt_resource(r) { 1650 fflags = r->fflags | RF_CTRL_INFO; 1651 ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags); 1652 if (ret) 1653 goto out_destroy; 1654 } 1655 1656 for_each_mon_enabled_rdt_resource(r) { 1657 fflags = r->fflags | RF_MON_INFO; 1658 sprintf(name, "%s_MON", r->name); 1659 ret = rdtgroup_mkdir_info_resdir(r, name, fflags); 1660 if (ret) 1661 goto out_destroy; 1662 } 1663 1664 /* 1665 * This extra ref will be put in kernfs_remove() and guarantees 1666 * that @rdtgrp->kn is always accessible. 1667 */ 1668 kernfs_get(kn_info); 1669 1670 ret = rdtgroup_kn_set_ugid(kn_info); 1671 if (ret) 1672 goto out_destroy; 1673 1674 kernfs_activate(kn_info); 1675 1676 return 0; 1677 1678 out_destroy: 1679 kernfs_remove(kn_info); 1680 return ret; 1681 } 1682 1683 static int 1684 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp, 1685 char *name, struct kernfs_node **dest_kn) 1686 { 1687 struct kernfs_node *kn; 1688 int ret; 1689 1690 /* create the directory */ 1691 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); 1692 if (IS_ERR(kn)) 1693 return PTR_ERR(kn); 1694 1695 if (dest_kn) 1696 *dest_kn = kn; 1697 1698 /* 1699 * This extra ref will be put in kernfs_remove() and guarantees 1700 * that @rdtgrp->kn is always accessible. 1701 */ 1702 kernfs_get(kn); 1703 1704 ret = rdtgroup_kn_set_ugid(kn); 1705 if (ret) 1706 goto out_destroy; 1707 1708 kernfs_activate(kn); 1709 1710 return 0; 1711 1712 out_destroy: 1713 kernfs_remove(kn); 1714 return ret; 1715 } 1716 1717 static void l3_qos_cfg_update(void *arg) 1718 { 1719 bool *enable = arg; 1720 1721 wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL); 1722 } 1723 1724 static void l2_qos_cfg_update(void *arg) 1725 { 1726 bool *enable = arg; 1727 1728 wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL); 1729 } 1730 1731 static inline bool is_mba_linear(void) 1732 { 1733 return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear; 1734 } 1735 1736 static int set_cache_qos_cfg(int level, bool enable) 1737 { 1738 void (*update)(void *arg); 1739 struct rdt_resource *r_l; 1740 cpumask_var_t cpu_mask; 1741 struct rdt_domain *d; 1742 int cpu; 1743 1744 if (level == RDT_RESOURCE_L3) 1745 update = l3_qos_cfg_update; 1746 else if (level == RDT_RESOURCE_L2) 1747 update = l2_qos_cfg_update; 1748 else 1749 return -EINVAL; 1750 1751 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) 1752 return -ENOMEM; 1753 1754 r_l = &rdt_resources_all[level]; 1755 list_for_each_entry(d, &r_l->domains, list) { 1756 /* Pick one CPU from each domain instance to update MSR */ 1757 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask); 1758 } 1759 cpu = get_cpu(); 1760 /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */ 1761 if (cpumask_test_cpu(cpu, cpu_mask)) 1762 update(&enable); 1763 /* Update QOS_CFG MSR on all other cpus in cpu_mask. */ 1764 smp_call_function_many(cpu_mask, update, &enable, 1); 1765 put_cpu(); 1766 1767 free_cpumask_var(cpu_mask); 1768 1769 return 0; 1770 } 1771 1772 /* 1773 * Enable or disable the MBA software controller 1774 * which helps user specify bandwidth in MBps. 1775 * MBA software controller is supported only if 1776 * MBM is supported and MBA is in linear scale. 1777 */ 1778 static int set_mba_sc(bool mba_sc) 1779 { 1780 struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA]; 1781 struct rdt_domain *d; 1782 1783 if (!is_mbm_enabled() || !is_mba_linear() || 1784 mba_sc == is_mba_sc(r)) 1785 return -EINVAL; 1786 1787 r->membw.mba_sc = mba_sc; 1788 list_for_each_entry(d, &r->domains, list) 1789 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val); 1790 1791 return 0; 1792 } 1793 1794 static int cdp_enable(int level, int data_type, int code_type) 1795 { 1796 struct rdt_resource *r_ldata = &rdt_resources_all[data_type]; 1797 struct rdt_resource *r_lcode = &rdt_resources_all[code_type]; 1798 struct rdt_resource *r_l = &rdt_resources_all[level]; 1799 int ret; 1800 1801 if (!r_l->alloc_capable || !r_ldata->alloc_capable || 1802 !r_lcode->alloc_capable) 1803 return -EINVAL; 1804 1805 ret = set_cache_qos_cfg(level, true); 1806 if (!ret) { 1807 r_l->alloc_enabled = false; 1808 r_ldata->alloc_enabled = true; 1809 r_lcode->alloc_enabled = true; 1810 } 1811 return ret; 1812 } 1813 1814 static int cdpl3_enable(void) 1815 { 1816 return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, 1817 RDT_RESOURCE_L3CODE); 1818 } 1819 1820 static int cdpl2_enable(void) 1821 { 1822 return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, 1823 RDT_RESOURCE_L2CODE); 1824 } 1825 1826 static void cdp_disable(int level, int data_type, int code_type) 1827 { 1828 struct rdt_resource *r = &rdt_resources_all[level]; 1829 1830 r->alloc_enabled = r->alloc_capable; 1831 1832 if (rdt_resources_all[data_type].alloc_enabled) { 1833 rdt_resources_all[data_type].alloc_enabled = false; 1834 rdt_resources_all[code_type].alloc_enabled = false; 1835 set_cache_qos_cfg(level, false); 1836 } 1837 } 1838 1839 static void cdpl3_disable(void) 1840 { 1841 cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE); 1842 } 1843 1844 static void cdpl2_disable(void) 1845 { 1846 cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE); 1847 } 1848 1849 static void cdp_disable_all(void) 1850 { 1851 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled) 1852 cdpl3_disable(); 1853 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled) 1854 cdpl2_disable(); 1855 } 1856 1857 /* 1858 * We don't allow rdtgroup directories to be created anywhere 1859 * except the root directory. Thus when looking for the rdtgroup 1860 * structure for a kernfs node we are either looking at a directory, 1861 * in which case the rdtgroup structure is pointed at by the "priv" 1862 * field, otherwise we have a file, and need only look to the parent 1863 * to find the rdtgroup. 1864 */ 1865 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn) 1866 { 1867 if (kernfs_type(kn) == KERNFS_DIR) { 1868 /* 1869 * All the resource directories use "kn->priv" 1870 * to point to the "struct rdtgroup" for the 1871 * resource. "info" and its subdirectories don't 1872 * have rdtgroup structures, so return NULL here. 1873 */ 1874 if (kn == kn_info || kn->parent == kn_info) 1875 return NULL; 1876 else 1877 return kn->priv; 1878 } else { 1879 return kn->parent->priv; 1880 } 1881 } 1882 1883 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn) 1884 { 1885 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); 1886 1887 if (!rdtgrp) 1888 return NULL; 1889 1890 atomic_inc(&rdtgrp->waitcount); 1891 kernfs_break_active_protection(kn); 1892 1893 mutex_lock(&rdtgroup_mutex); 1894 1895 /* Was this group deleted while we waited? */ 1896 if (rdtgrp->flags & RDT_DELETED) 1897 return NULL; 1898 1899 return rdtgrp; 1900 } 1901 1902 void rdtgroup_kn_unlock(struct kernfs_node *kn) 1903 { 1904 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); 1905 1906 if (!rdtgrp) 1907 return; 1908 1909 mutex_unlock(&rdtgroup_mutex); 1910 1911 if (atomic_dec_and_test(&rdtgrp->waitcount) && 1912 (rdtgrp->flags & RDT_DELETED)) { 1913 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || 1914 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) 1915 rdtgroup_pseudo_lock_remove(rdtgrp); 1916 kernfs_unbreak_active_protection(kn); 1917 kernfs_put(rdtgrp->kn); 1918 kfree(rdtgrp); 1919 } else { 1920 kernfs_unbreak_active_protection(kn); 1921 } 1922 } 1923 1924 static int mkdir_mondata_all(struct kernfs_node *parent_kn, 1925 struct rdtgroup *prgrp, 1926 struct kernfs_node **mon_data_kn); 1927 1928 static int rdt_enable_ctx(struct rdt_fs_context *ctx) 1929 { 1930 int ret = 0; 1931 1932 if (ctx->enable_cdpl2) 1933 ret = cdpl2_enable(); 1934 1935 if (!ret && ctx->enable_cdpl3) 1936 ret = cdpl3_enable(); 1937 1938 if (!ret && ctx->enable_mba_mbps) 1939 ret = set_mba_sc(true); 1940 1941 return ret; 1942 } 1943 1944 static int rdt_get_tree(struct fs_context *fc) 1945 { 1946 struct rdt_fs_context *ctx = rdt_fc2context(fc); 1947 struct rdt_domain *dom; 1948 struct rdt_resource *r; 1949 int ret; 1950 1951 cpus_read_lock(); 1952 mutex_lock(&rdtgroup_mutex); 1953 /* 1954 * resctrl file system can only be mounted once. 1955 */ 1956 if (static_branch_unlikely(&rdt_enable_key)) { 1957 ret = -EBUSY; 1958 goto out; 1959 } 1960 1961 ret = rdt_enable_ctx(ctx); 1962 if (ret < 0) 1963 goto out_cdp; 1964 1965 closid_init(); 1966 1967 ret = rdtgroup_create_info_dir(rdtgroup_default.kn); 1968 if (ret < 0) 1969 goto out_mba; 1970 1971 if (rdt_mon_capable) { 1972 ret = mongroup_create_dir(rdtgroup_default.kn, 1973 &rdtgroup_default, "mon_groups", 1974 &kn_mongrp); 1975 if (ret < 0) 1976 goto out_info; 1977 kernfs_get(kn_mongrp); 1978 1979 ret = mkdir_mondata_all(rdtgroup_default.kn, 1980 &rdtgroup_default, &kn_mondata); 1981 if (ret < 0) 1982 goto out_mongrp; 1983 kernfs_get(kn_mondata); 1984 rdtgroup_default.mon.mon_data_kn = kn_mondata; 1985 } 1986 1987 ret = rdt_pseudo_lock_init(); 1988 if (ret) 1989 goto out_mondata; 1990 1991 ret = kernfs_get_tree(fc); 1992 if (ret < 0) 1993 goto out_psl; 1994 1995 if (rdt_alloc_capable) 1996 static_branch_enable_cpuslocked(&rdt_alloc_enable_key); 1997 if (rdt_mon_capable) 1998 static_branch_enable_cpuslocked(&rdt_mon_enable_key); 1999 2000 if (rdt_alloc_capable || rdt_mon_capable) 2001 static_branch_enable_cpuslocked(&rdt_enable_key); 2002 2003 if (is_mbm_enabled()) { 2004 r = &rdt_resources_all[RDT_RESOURCE_L3]; 2005 list_for_each_entry(dom, &r->domains, list) 2006 mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL); 2007 } 2008 2009 goto out; 2010 2011 out_psl: 2012 rdt_pseudo_lock_release(); 2013 out_mondata: 2014 if (rdt_mon_capable) 2015 kernfs_remove(kn_mondata); 2016 out_mongrp: 2017 if (rdt_mon_capable) 2018 kernfs_remove(kn_mongrp); 2019 out_info: 2020 kernfs_remove(kn_info); 2021 out_mba: 2022 if (ctx->enable_mba_mbps) 2023 set_mba_sc(false); 2024 out_cdp: 2025 cdp_disable_all(); 2026 out: 2027 rdt_last_cmd_clear(); 2028 mutex_unlock(&rdtgroup_mutex); 2029 cpus_read_unlock(); 2030 return ret; 2031 } 2032 2033 enum rdt_param { 2034 Opt_cdp, 2035 Opt_cdpl2, 2036 Opt_mba_mbps, 2037 nr__rdt_params 2038 }; 2039 2040 static const struct fs_parameter_spec rdt_param_specs[] = { 2041 fsparam_flag("cdp", Opt_cdp), 2042 fsparam_flag("cdpl2", Opt_cdpl2), 2043 fsparam_flag("mba_MBps", Opt_mba_mbps), 2044 {} 2045 }; 2046 2047 static const struct fs_parameter_description rdt_fs_parameters = { 2048 .name = "rdt", 2049 .specs = rdt_param_specs, 2050 }; 2051 2052 static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param) 2053 { 2054 struct rdt_fs_context *ctx = rdt_fc2context(fc); 2055 struct fs_parse_result result; 2056 int opt; 2057 2058 opt = fs_parse(fc, &rdt_fs_parameters, param, &result); 2059 if (opt < 0) 2060 return opt; 2061 2062 switch (opt) { 2063 case Opt_cdp: 2064 ctx->enable_cdpl3 = true; 2065 return 0; 2066 case Opt_cdpl2: 2067 ctx->enable_cdpl2 = true; 2068 return 0; 2069 case Opt_mba_mbps: 2070 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) 2071 return -EINVAL; 2072 ctx->enable_mba_mbps = true; 2073 return 0; 2074 } 2075 2076 return -EINVAL; 2077 } 2078 2079 static void rdt_fs_context_free(struct fs_context *fc) 2080 { 2081 struct rdt_fs_context *ctx = rdt_fc2context(fc); 2082 2083 kernfs_free_fs_context(fc); 2084 kfree(ctx); 2085 } 2086 2087 static const struct fs_context_operations rdt_fs_context_ops = { 2088 .free = rdt_fs_context_free, 2089 .parse_param = rdt_parse_param, 2090 .get_tree = rdt_get_tree, 2091 }; 2092 2093 static int rdt_init_fs_context(struct fs_context *fc) 2094 { 2095 struct rdt_fs_context *ctx; 2096 2097 ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL); 2098 if (!ctx) 2099 return -ENOMEM; 2100 2101 ctx->kfc.root = rdt_root; 2102 ctx->kfc.magic = RDTGROUP_SUPER_MAGIC; 2103 fc->fs_private = &ctx->kfc; 2104 fc->ops = &rdt_fs_context_ops; 2105 put_user_ns(fc->user_ns); 2106 fc->user_ns = get_user_ns(&init_user_ns); 2107 fc->global = true; 2108 return 0; 2109 } 2110 2111 static int reset_all_ctrls(struct rdt_resource *r) 2112 { 2113 struct msr_param msr_param; 2114 cpumask_var_t cpu_mask; 2115 struct rdt_domain *d; 2116 int i, cpu; 2117 2118 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) 2119 return -ENOMEM; 2120 2121 msr_param.res = r; 2122 msr_param.low = 0; 2123 msr_param.high = r->num_closid; 2124 2125 /* 2126 * Disable resource control for this resource by setting all 2127 * CBMs in all domains to the maximum mask value. Pick one CPU 2128 * from each domain to update the MSRs below. 2129 */ 2130 list_for_each_entry(d, &r->domains, list) { 2131 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask); 2132 2133 for (i = 0; i < r->num_closid; i++) 2134 d->ctrl_val[i] = r->default_ctrl; 2135 } 2136 cpu = get_cpu(); 2137 /* Update CBM on this cpu if it's in cpu_mask. */ 2138 if (cpumask_test_cpu(cpu, cpu_mask)) 2139 rdt_ctrl_update(&msr_param); 2140 /* Update CBM on all other cpus in cpu_mask. */ 2141 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1); 2142 put_cpu(); 2143 2144 free_cpumask_var(cpu_mask); 2145 2146 return 0; 2147 } 2148 2149 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r) 2150 { 2151 return (rdt_alloc_capable && 2152 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid)); 2153 } 2154 2155 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r) 2156 { 2157 return (rdt_mon_capable && 2158 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid)); 2159 } 2160 2161 /* 2162 * Move tasks from one to the other group. If @from is NULL, then all tasks 2163 * in the systems are moved unconditionally (used for teardown). 2164 * 2165 * If @mask is not NULL the cpus on which moved tasks are running are set 2166 * in that mask so the update smp function call is restricted to affected 2167 * cpus. 2168 */ 2169 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to, 2170 struct cpumask *mask) 2171 { 2172 struct task_struct *p, *t; 2173 2174 read_lock(&tasklist_lock); 2175 for_each_process_thread(p, t) { 2176 if (!from || is_closid_match(t, from) || 2177 is_rmid_match(t, from)) { 2178 t->closid = to->closid; 2179 t->rmid = to->mon.rmid; 2180 2181 #ifdef CONFIG_SMP 2182 /* 2183 * This is safe on x86 w/o barriers as the ordering 2184 * of writing to task_cpu() and t->on_cpu is 2185 * reverse to the reading here. The detection is 2186 * inaccurate as tasks might move or schedule 2187 * before the smp function call takes place. In 2188 * such a case the function call is pointless, but 2189 * there is no other side effect. 2190 */ 2191 if (mask && t->on_cpu) 2192 cpumask_set_cpu(task_cpu(t), mask); 2193 #endif 2194 } 2195 } 2196 read_unlock(&tasklist_lock); 2197 } 2198 2199 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp) 2200 { 2201 struct rdtgroup *sentry, *stmp; 2202 struct list_head *head; 2203 2204 head = &rdtgrp->mon.crdtgrp_list; 2205 list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) { 2206 free_rmid(sentry->mon.rmid); 2207 list_del(&sentry->mon.crdtgrp_list); 2208 2209 if (atomic_read(&sentry->waitcount) != 0) 2210 sentry->flags = RDT_DELETED; 2211 else 2212 kfree(sentry); 2213 } 2214 } 2215 2216 /* 2217 * Forcibly remove all of subdirectories under root. 2218 */ 2219 static void rmdir_all_sub(void) 2220 { 2221 struct rdtgroup *rdtgrp, *tmp; 2222 2223 /* Move all tasks to the default resource group */ 2224 rdt_move_group_tasks(NULL, &rdtgroup_default, NULL); 2225 2226 list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) { 2227 /* Free any child rmids */ 2228 free_all_child_rdtgrp(rdtgrp); 2229 2230 /* Remove each rdtgroup other than root */ 2231 if (rdtgrp == &rdtgroup_default) 2232 continue; 2233 2234 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || 2235 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) 2236 rdtgroup_pseudo_lock_remove(rdtgrp); 2237 2238 /* 2239 * Give any CPUs back to the default group. We cannot copy 2240 * cpu_online_mask because a CPU might have executed the 2241 * offline callback already, but is still marked online. 2242 */ 2243 cpumask_or(&rdtgroup_default.cpu_mask, 2244 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); 2245 2246 free_rmid(rdtgrp->mon.rmid); 2247 2248 kernfs_remove(rdtgrp->kn); 2249 list_del(&rdtgrp->rdtgroup_list); 2250 2251 if (atomic_read(&rdtgrp->waitcount) != 0) 2252 rdtgrp->flags = RDT_DELETED; 2253 else 2254 kfree(rdtgrp); 2255 } 2256 /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */ 2257 update_closid_rmid(cpu_online_mask, &rdtgroup_default); 2258 2259 kernfs_remove(kn_info); 2260 kernfs_remove(kn_mongrp); 2261 kernfs_remove(kn_mondata); 2262 } 2263 2264 static void rdt_kill_sb(struct super_block *sb) 2265 { 2266 struct rdt_resource *r; 2267 2268 cpus_read_lock(); 2269 mutex_lock(&rdtgroup_mutex); 2270 2271 set_mba_sc(false); 2272 2273 /*Put everything back to default values. */ 2274 for_each_alloc_enabled_rdt_resource(r) 2275 reset_all_ctrls(r); 2276 cdp_disable_all(); 2277 rmdir_all_sub(); 2278 rdt_pseudo_lock_release(); 2279 rdtgroup_default.mode = RDT_MODE_SHAREABLE; 2280 static_branch_disable_cpuslocked(&rdt_alloc_enable_key); 2281 static_branch_disable_cpuslocked(&rdt_mon_enable_key); 2282 static_branch_disable_cpuslocked(&rdt_enable_key); 2283 kernfs_kill_sb(sb); 2284 mutex_unlock(&rdtgroup_mutex); 2285 cpus_read_unlock(); 2286 } 2287 2288 static struct file_system_type rdt_fs_type = { 2289 .name = "resctrl", 2290 .init_fs_context = rdt_init_fs_context, 2291 .parameters = &rdt_fs_parameters, 2292 .kill_sb = rdt_kill_sb, 2293 }; 2294 2295 static int mon_addfile(struct kernfs_node *parent_kn, const char *name, 2296 void *priv) 2297 { 2298 struct kernfs_node *kn; 2299 int ret = 0; 2300 2301 kn = __kernfs_create_file(parent_kn, name, 0444, 2302 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0, 2303 &kf_mondata_ops, priv, NULL, NULL); 2304 if (IS_ERR(kn)) 2305 return PTR_ERR(kn); 2306 2307 ret = rdtgroup_kn_set_ugid(kn); 2308 if (ret) { 2309 kernfs_remove(kn); 2310 return ret; 2311 } 2312 2313 return ret; 2314 } 2315 2316 /* 2317 * Remove all subdirectories of mon_data of ctrl_mon groups 2318 * and monitor groups with given domain id. 2319 */ 2320 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id) 2321 { 2322 struct rdtgroup *prgrp, *crgrp; 2323 char name[32]; 2324 2325 if (!r->mon_enabled) 2326 return; 2327 2328 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { 2329 sprintf(name, "mon_%s_%02d", r->name, dom_id); 2330 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name); 2331 2332 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list) 2333 kernfs_remove_by_name(crgrp->mon.mon_data_kn, name); 2334 } 2335 } 2336 2337 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn, 2338 struct rdt_domain *d, 2339 struct rdt_resource *r, struct rdtgroup *prgrp) 2340 { 2341 union mon_data_bits priv; 2342 struct kernfs_node *kn; 2343 struct mon_evt *mevt; 2344 struct rmid_read rr; 2345 char name[32]; 2346 int ret; 2347 2348 sprintf(name, "mon_%s_%02d", r->name, d->id); 2349 /* create the directory */ 2350 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); 2351 if (IS_ERR(kn)) 2352 return PTR_ERR(kn); 2353 2354 /* 2355 * This extra ref will be put in kernfs_remove() and guarantees 2356 * that kn is always accessible. 2357 */ 2358 kernfs_get(kn); 2359 ret = rdtgroup_kn_set_ugid(kn); 2360 if (ret) 2361 goto out_destroy; 2362 2363 if (WARN_ON(list_empty(&r->evt_list))) { 2364 ret = -EPERM; 2365 goto out_destroy; 2366 } 2367 2368 priv.u.rid = r->rid; 2369 priv.u.domid = d->id; 2370 list_for_each_entry(mevt, &r->evt_list, list) { 2371 priv.u.evtid = mevt->evtid; 2372 ret = mon_addfile(kn, mevt->name, priv.priv); 2373 if (ret) 2374 goto out_destroy; 2375 2376 if (is_mbm_event(mevt->evtid)) 2377 mon_event_read(&rr, d, prgrp, mevt->evtid, true); 2378 } 2379 kernfs_activate(kn); 2380 return 0; 2381 2382 out_destroy: 2383 kernfs_remove(kn); 2384 return ret; 2385 } 2386 2387 /* 2388 * Add all subdirectories of mon_data for "ctrl_mon" groups 2389 * and "monitor" groups with given domain id. 2390 */ 2391 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, 2392 struct rdt_domain *d) 2393 { 2394 struct kernfs_node *parent_kn; 2395 struct rdtgroup *prgrp, *crgrp; 2396 struct list_head *head; 2397 2398 if (!r->mon_enabled) 2399 return; 2400 2401 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { 2402 parent_kn = prgrp->mon.mon_data_kn; 2403 mkdir_mondata_subdir(parent_kn, d, r, prgrp); 2404 2405 head = &prgrp->mon.crdtgrp_list; 2406 list_for_each_entry(crgrp, head, mon.crdtgrp_list) { 2407 parent_kn = crgrp->mon.mon_data_kn; 2408 mkdir_mondata_subdir(parent_kn, d, r, crgrp); 2409 } 2410 } 2411 } 2412 2413 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn, 2414 struct rdt_resource *r, 2415 struct rdtgroup *prgrp) 2416 { 2417 struct rdt_domain *dom; 2418 int ret; 2419 2420 list_for_each_entry(dom, &r->domains, list) { 2421 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp); 2422 if (ret) 2423 return ret; 2424 } 2425 2426 return 0; 2427 } 2428 2429 /* 2430 * This creates a directory mon_data which contains the monitored data. 2431 * 2432 * mon_data has one directory for each domain whic are named 2433 * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data 2434 * with L3 domain looks as below: 2435 * ./mon_data: 2436 * mon_L3_00 2437 * mon_L3_01 2438 * mon_L3_02 2439 * ... 2440 * 2441 * Each domain directory has one file per event: 2442 * ./mon_L3_00/: 2443 * llc_occupancy 2444 * 2445 */ 2446 static int mkdir_mondata_all(struct kernfs_node *parent_kn, 2447 struct rdtgroup *prgrp, 2448 struct kernfs_node **dest_kn) 2449 { 2450 struct rdt_resource *r; 2451 struct kernfs_node *kn; 2452 int ret; 2453 2454 /* 2455 * Create the mon_data directory first. 2456 */ 2457 ret = mongroup_create_dir(parent_kn, prgrp, "mon_data", &kn); 2458 if (ret) 2459 return ret; 2460 2461 if (dest_kn) 2462 *dest_kn = kn; 2463 2464 /* 2465 * Create the subdirectories for each domain. Note that all events 2466 * in a domain like L3 are grouped into a resource whose domain is L3 2467 */ 2468 for_each_mon_enabled_rdt_resource(r) { 2469 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp); 2470 if (ret) 2471 goto out_destroy; 2472 } 2473 2474 return 0; 2475 2476 out_destroy: 2477 kernfs_remove(kn); 2478 return ret; 2479 } 2480 2481 /** 2482 * cbm_ensure_valid - Enforce validity on provided CBM 2483 * @_val: Candidate CBM 2484 * @r: RDT resource to which the CBM belongs 2485 * 2486 * The provided CBM represents all cache portions available for use. This 2487 * may be represented by a bitmap that does not consist of contiguous ones 2488 * and thus be an invalid CBM. 2489 * Here the provided CBM is forced to be a valid CBM by only considering 2490 * the first set of contiguous bits as valid and clearing all bits. 2491 * The intention here is to provide a valid default CBM with which a new 2492 * resource group is initialized. The user can follow this with a 2493 * modification to the CBM if the default does not satisfy the 2494 * requirements. 2495 */ 2496 static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r) 2497 { 2498 unsigned int cbm_len = r->cache.cbm_len; 2499 unsigned long first_bit, zero_bit; 2500 unsigned long val = _val; 2501 2502 if (!val) 2503 return 0; 2504 2505 first_bit = find_first_bit(&val, cbm_len); 2506 zero_bit = find_next_zero_bit(&val, cbm_len, first_bit); 2507 2508 /* Clear any remaining bits to ensure contiguous region */ 2509 bitmap_clear(&val, zero_bit, cbm_len - zero_bit); 2510 return (u32)val; 2511 } 2512 2513 /* 2514 * Initialize cache resources per RDT domain 2515 * 2516 * Set the RDT domain up to start off with all usable allocations. That is, 2517 * all shareable and unused bits. All-zero CBM is invalid. 2518 */ 2519 static int __init_one_rdt_domain(struct rdt_domain *d, struct rdt_resource *r, 2520 u32 closid) 2521 { 2522 struct rdt_resource *r_cdp = NULL; 2523 struct rdt_domain *d_cdp = NULL; 2524 u32 used_b = 0, unused_b = 0; 2525 unsigned long tmp_cbm; 2526 enum rdtgrp_mode mode; 2527 u32 peer_ctl, *ctrl; 2528 int i; 2529 2530 rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp); 2531 d->have_new_ctrl = false; 2532 d->new_ctrl = r->cache.shareable_bits; 2533 used_b = r->cache.shareable_bits; 2534 ctrl = d->ctrl_val; 2535 for (i = 0; i < closids_supported(); i++, ctrl++) { 2536 if (closid_allocated(i) && i != closid) { 2537 mode = rdtgroup_mode_by_closid(i); 2538 if (mode == RDT_MODE_PSEUDO_LOCKSETUP) 2539 /* 2540 * ctrl values for locksetup aren't relevant 2541 * until the schemata is written, and the mode 2542 * becomes RDT_MODE_PSEUDO_LOCKED. 2543 */ 2544 continue; 2545 /* 2546 * If CDP is active include peer domain's 2547 * usage to ensure there is no overlap 2548 * with an exclusive group. 2549 */ 2550 if (d_cdp) 2551 peer_ctl = d_cdp->ctrl_val[i]; 2552 else 2553 peer_ctl = 0; 2554 used_b |= *ctrl | peer_ctl; 2555 if (mode == RDT_MODE_SHAREABLE) 2556 d->new_ctrl |= *ctrl | peer_ctl; 2557 } 2558 } 2559 if (d->plr && d->plr->cbm > 0) 2560 used_b |= d->plr->cbm; 2561 unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1); 2562 unused_b &= BIT_MASK(r->cache.cbm_len) - 1; 2563 d->new_ctrl |= unused_b; 2564 /* 2565 * Force the initial CBM to be valid, user can 2566 * modify the CBM based on system availability. 2567 */ 2568 d->new_ctrl = cbm_ensure_valid(d->new_ctrl, r); 2569 /* 2570 * Assign the u32 CBM to an unsigned long to ensure that 2571 * bitmap_weight() does not access out-of-bound memory. 2572 */ 2573 tmp_cbm = d->new_ctrl; 2574 if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) { 2575 rdt_last_cmd_printf("No space on %s:%d\n", r->name, d->id); 2576 return -ENOSPC; 2577 } 2578 d->have_new_ctrl = true; 2579 2580 return 0; 2581 } 2582 2583 /* 2584 * Initialize cache resources with default values. 2585 * 2586 * A new RDT group is being created on an allocation capable (CAT) 2587 * supporting system. Set this group up to start off with all usable 2588 * allocations. 2589 * 2590 * If there are no more shareable bits available on any domain then 2591 * the entire allocation will fail. 2592 */ 2593 static int rdtgroup_init_cat(struct rdt_resource *r, u32 closid) 2594 { 2595 struct rdt_domain *d; 2596 int ret; 2597 2598 list_for_each_entry(d, &r->domains, list) { 2599 ret = __init_one_rdt_domain(d, r, closid); 2600 if (ret < 0) 2601 return ret; 2602 } 2603 2604 return 0; 2605 } 2606 2607 /* Initialize MBA resource with default values. */ 2608 static void rdtgroup_init_mba(struct rdt_resource *r) 2609 { 2610 struct rdt_domain *d; 2611 2612 list_for_each_entry(d, &r->domains, list) { 2613 d->new_ctrl = is_mba_sc(r) ? MBA_MAX_MBPS : r->default_ctrl; 2614 d->have_new_ctrl = true; 2615 } 2616 } 2617 2618 /* Initialize the RDT group's allocations. */ 2619 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp) 2620 { 2621 struct rdt_resource *r; 2622 int ret; 2623 2624 for_each_alloc_enabled_rdt_resource(r) { 2625 if (r->rid == RDT_RESOURCE_MBA) { 2626 rdtgroup_init_mba(r); 2627 } else { 2628 ret = rdtgroup_init_cat(r, rdtgrp->closid); 2629 if (ret < 0) 2630 return ret; 2631 } 2632 2633 ret = update_domains(r, rdtgrp->closid); 2634 if (ret < 0) { 2635 rdt_last_cmd_puts("Failed to initialize allocations\n"); 2636 return ret; 2637 } 2638 2639 } 2640 2641 rdtgrp->mode = RDT_MODE_SHAREABLE; 2642 2643 return 0; 2644 } 2645 2646 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn, 2647 const char *name, umode_t mode, 2648 enum rdt_group_type rtype, struct rdtgroup **r) 2649 { 2650 struct rdtgroup *prdtgrp, *rdtgrp; 2651 struct kernfs_node *kn; 2652 uint files = 0; 2653 int ret; 2654 2655 prdtgrp = rdtgroup_kn_lock_live(parent_kn); 2656 if (!prdtgrp) { 2657 ret = -ENODEV; 2658 goto out_unlock; 2659 } 2660 2661 if (rtype == RDTMON_GROUP && 2662 (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || 2663 prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) { 2664 ret = -EINVAL; 2665 rdt_last_cmd_puts("Pseudo-locking in progress\n"); 2666 goto out_unlock; 2667 } 2668 2669 /* allocate the rdtgroup. */ 2670 rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL); 2671 if (!rdtgrp) { 2672 ret = -ENOSPC; 2673 rdt_last_cmd_puts("Kernel out of memory\n"); 2674 goto out_unlock; 2675 } 2676 *r = rdtgrp; 2677 rdtgrp->mon.parent = prdtgrp; 2678 rdtgrp->type = rtype; 2679 INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list); 2680 2681 /* kernfs creates the directory for rdtgrp */ 2682 kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp); 2683 if (IS_ERR(kn)) { 2684 ret = PTR_ERR(kn); 2685 rdt_last_cmd_puts("kernfs create error\n"); 2686 goto out_free_rgrp; 2687 } 2688 rdtgrp->kn = kn; 2689 2690 /* 2691 * kernfs_remove() will drop the reference count on "kn" which 2692 * will free it. But we still need it to stick around for the 2693 * rdtgroup_kn_unlock(kn} call below. Take one extra reference 2694 * here, which will be dropped inside rdtgroup_kn_unlock(). 2695 */ 2696 kernfs_get(kn); 2697 2698 ret = rdtgroup_kn_set_ugid(kn); 2699 if (ret) { 2700 rdt_last_cmd_puts("kernfs perm error\n"); 2701 goto out_destroy; 2702 } 2703 2704 files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype); 2705 ret = rdtgroup_add_files(kn, files); 2706 if (ret) { 2707 rdt_last_cmd_puts("kernfs fill error\n"); 2708 goto out_destroy; 2709 } 2710 2711 if (rdt_mon_capable) { 2712 ret = alloc_rmid(); 2713 if (ret < 0) { 2714 rdt_last_cmd_puts("Out of RMIDs\n"); 2715 goto out_destroy; 2716 } 2717 rdtgrp->mon.rmid = ret; 2718 2719 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn); 2720 if (ret) { 2721 rdt_last_cmd_puts("kernfs subdir error\n"); 2722 goto out_idfree; 2723 } 2724 } 2725 kernfs_activate(kn); 2726 2727 /* 2728 * The caller unlocks the parent_kn upon success. 2729 */ 2730 return 0; 2731 2732 out_idfree: 2733 free_rmid(rdtgrp->mon.rmid); 2734 out_destroy: 2735 kernfs_remove(rdtgrp->kn); 2736 out_free_rgrp: 2737 kfree(rdtgrp); 2738 out_unlock: 2739 rdtgroup_kn_unlock(parent_kn); 2740 return ret; 2741 } 2742 2743 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp) 2744 { 2745 kernfs_remove(rgrp->kn); 2746 free_rmid(rgrp->mon.rmid); 2747 kfree(rgrp); 2748 } 2749 2750 /* 2751 * Create a monitor group under "mon_groups" directory of a control 2752 * and monitor group(ctrl_mon). This is a resource group 2753 * to monitor a subset of tasks and cpus in its parent ctrl_mon group. 2754 */ 2755 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn, 2756 const char *name, umode_t mode) 2757 { 2758 struct rdtgroup *rdtgrp, *prgrp; 2759 int ret; 2760 2761 ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTMON_GROUP, &rdtgrp); 2762 if (ret) 2763 return ret; 2764 2765 prgrp = rdtgrp->mon.parent; 2766 rdtgrp->closid = prgrp->closid; 2767 2768 /* 2769 * Add the rdtgrp to the list of rdtgrps the parent 2770 * ctrl_mon group has to track. 2771 */ 2772 list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list); 2773 2774 rdtgroup_kn_unlock(parent_kn); 2775 return ret; 2776 } 2777 2778 /* 2779 * These are rdtgroups created under the root directory. Can be used 2780 * to allocate and monitor resources. 2781 */ 2782 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn, 2783 const char *name, umode_t mode) 2784 { 2785 struct rdtgroup *rdtgrp; 2786 struct kernfs_node *kn; 2787 u32 closid; 2788 int ret; 2789 2790 ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTCTRL_GROUP, &rdtgrp); 2791 if (ret) 2792 return ret; 2793 2794 kn = rdtgrp->kn; 2795 ret = closid_alloc(); 2796 if (ret < 0) { 2797 rdt_last_cmd_puts("Out of CLOSIDs\n"); 2798 goto out_common_fail; 2799 } 2800 closid = ret; 2801 ret = 0; 2802 2803 rdtgrp->closid = closid; 2804 ret = rdtgroup_init_alloc(rdtgrp); 2805 if (ret < 0) 2806 goto out_id_free; 2807 2808 list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups); 2809 2810 if (rdt_mon_capable) { 2811 /* 2812 * Create an empty mon_groups directory to hold the subset 2813 * of tasks and cpus to monitor. 2814 */ 2815 ret = mongroup_create_dir(kn, rdtgrp, "mon_groups", NULL); 2816 if (ret) { 2817 rdt_last_cmd_puts("kernfs subdir error\n"); 2818 goto out_del_list; 2819 } 2820 } 2821 2822 goto out_unlock; 2823 2824 out_del_list: 2825 list_del(&rdtgrp->rdtgroup_list); 2826 out_id_free: 2827 closid_free(closid); 2828 out_common_fail: 2829 mkdir_rdt_prepare_clean(rdtgrp); 2830 out_unlock: 2831 rdtgroup_kn_unlock(parent_kn); 2832 return ret; 2833 } 2834 2835 /* 2836 * We allow creating mon groups only with in a directory called "mon_groups" 2837 * which is present in every ctrl_mon group. Check if this is a valid 2838 * "mon_groups" directory. 2839 * 2840 * 1. The directory should be named "mon_groups". 2841 * 2. The mon group itself should "not" be named "mon_groups". 2842 * This makes sure "mon_groups" directory always has a ctrl_mon group 2843 * as parent. 2844 */ 2845 static bool is_mon_groups(struct kernfs_node *kn, const char *name) 2846 { 2847 return (!strcmp(kn->name, "mon_groups") && 2848 strcmp(name, "mon_groups")); 2849 } 2850 2851 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name, 2852 umode_t mode) 2853 { 2854 /* Do not accept '\n' to avoid unparsable situation. */ 2855 if (strchr(name, '\n')) 2856 return -EINVAL; 2857 2858 /* 2859 * If the parent directory is the root directory and RDT 2860 * allocation is supported, add a control and monitoring 2861 * subdirectory 2862 */ 2863 if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn) 2864 return rdtgroup_mkdir_ctrl_mon(parent_kn, name, mode); 2865 2866 /* 2867 * If RDT monitoring is supported and the parent directory is a valid 2868 * "mon_groups" directory, add a monitoring subdirectory. 2869 */ 2870 if (rdt_mon_capable && is_mon_groups(parent_kn, name)) 2871 return rdtgroup_mkdir_mon(parent_kn, name, mode); 2872 2873 return -EPERM; 2874 } 2875 2876 static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp, 2877 cpumask_var_t tmpmask) 2878 { 2879 struct rdtgroup *prdtgrp = rdtgrp->mon.parent; 2880 int cpu; 2881 2882 /* Give any tasks back to the parent group */ 2883 rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask); 2884 2885 /* Update per cpu rmid of the moved CPUs first */ 2886 for_each_cpu(cpu, &rdtgrp->cpu_mask) 2887 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid; 2888 /* 2889 * Update the MSR on moved CPUs and CPUs which have moved 2890 * task running on them. 2891 */ 2892 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); 2893 update_closid_rmid(tmpmask, NULL); 2894 2895 rdtgrp->flags = RDT_DELETED; 2896 free_rmid(rdtgrp->mon.rmid); 2897 2898 /* 2899 * Remove the rdtgrp from the parent ctrl_mon group's list 2900 */ 2901 WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list)); 2902 list_del(&rdtgrp->mon.crdtgrp_list); 2903 2904 /* 2905 * one extra hold on this, will drop when we kfree(rdtgrp) 2906 * in rdtgroup_kn_unlock() 2907 */ 2908 kernfs_get(kn); 2909 kernfs_remove(rdtgrp->kn); 2910 2911 return 0; 2912 } 2913 2914 static int rdtgroup_ctrl_remove(struct kernfs_node *kn, 2915 struct rdtgroup *rdtgrp) 2916 { 2917 rdtgrp->flags = RDT_DELETED; 2918 list_del(&rdtgrp->rdtgroup_list); 2919 2920 /* 2921 * one extra hold on this, will drop when we kfree(rdtgrp) 2922 * in rdtgroup_kn_unlock() 2923 */ 2924 kernfs_get(kn); 2925 kernfs_remove(rdtgrp->kn); 2926 return 0; 2927 } 2928 2929 static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp, 2930 cpumask_var_t tmpmask) 2931 { 2932 int cpu; 2933 2934 /* Give any tasks back to the default group */ 2935 rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask); 2936 2937 /* Give any CPUs back to the default group */ 2938 cpumask_or(&rdtgroup_default.cpu_mask, 2939 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); 2940 2941 /* Update per cpu closid and rmid of the moved CPUs first */ 2942 for_each_cpu(cpu, &rdtgrp->cpu_mask) { 2943 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid; 2944 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid; 2945 } 2946 2947 /* 2948 * Update the MSR on moved CPUs and CPUs which have moved 2949 * task running on them. 2950 */ 2951 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); 2952 update_closid_rmid(tmpmask, NULL); 2953 2954 closid_free(rdtgrp->closid); 2955 free_rmid(rdtgrp->mon.rmid); 2956 2957 rdtgroup_ctrl_remove(kn, rdtgrp); 2958 2959 /* 2960 * Free all the child monitor group rmids. 2961 */ 2962 free_all_child_rdtgrp(rdtgrp); 2963 2964 return 0; 2965 } 2966 2967 static int rdtgroup_rmdir(struct kernfs_node *kn) 2968 { 2969 struct kernfs_node *parent_kn = kn->parent; 2970 struct rdtgroup *rdtgrp; 2971 cpumask_var_t tmpmask; 2972 int ret = 0; 2973 2974 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) 2975 return -ENOMEM; 2976 2977 rdtgrp = rdtgroup_kn_lock_live(kn); 2978 if (!rdtgrp) { 2979 ret = -EPERM; 2980 goto out; 2981 } 2982 2983 /* 2984 * If the rdtgroup is a ctrl_mon group and parent directory 2985 * is the root directory, remove the ctrl_mon group. 2986 * 2987 * If the rdtgroup is a mon group and parent directory 2988 * is a valid "mon_groups" directory, remove the mon group. 2989 */ 2990 if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn) { 2991 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || 2992 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { 2993 ret = rdtgroup_ctrl_remove(kn, rdtgrp); 2994 } else { 2995 ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask); 2996 } 2997 } else if (rdtgrp->type == RDTMON_GROUP && 2998 is_mon_groups(parent_kn, kn->name)) { 2999 ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask); 3000 } else { 3001 ret = -EPERM; 3002 } 3003 3004 out: 3005 rdtgroup_kn_unlock(kn); 3006 free_cpumask_var(tmpmask); 3007 return ret; 3008 } 3009 3010 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf) 3011 { 3012 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled) 3013 seq_puts(seq, ",cdp"); 3014 3015 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled) 3016 seq_puts(seq, ",cdpl2"); 3017 3018 if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA])) 3019 seq_puts(seq, ",mba_MBps"); 3020 3021 return 0; 3022 } 3023 3024 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = { 3025 .mkdir = rdtgroup_mkdir, 3026 .rmdir = rdtgroup_rmdir, 3027 .show_options = rdtgroup_show_options, 3028 }; 3029 3030 static int __init rdtgroup_setup_root(void) 3031 { 3032 int ret; 3033 3034 rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops, 3035 KERNFS_ROOT_CREATE_DEACTIVATED | 3036 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK, 3037 &rdtgroup_default); 3038 if (IS_ERR(rdt_root)) 3039 return PTR_ERR(rdt_root); 3040 3041 mutex_lock(&rdtgroup_mutex); 3042 3043 rdtgroup_default.closid = 0; 3044 rdtgroup_default.mon.rmid = 0; 3045 rdtgroup_default.type = RDTCTRL_GROUP; 3046 INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list); 3047 3048 list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups); 3049 3050 ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE); 3051 if (ret) { 3052 kernfs_destroy_root(rdt_root); 3053 goto out; 3054 } 3055 3056 rdtgroup_default.kn = rdt_root->kn; 3057 kernfs_activate(rdtgroup_default.kn); 3058 3059 out: 3060 mutex_unlock(&rdtgroup_mutex); 3061 3062 return ret; 3063 } 3064 3065 /* 3066 * rdtgroup_init - rdtgroup initialization 3067 * 3068 * Setup resctrl file system including set up root, create mount point, 3069 * register rdtgroup filesystem, and initialize files under root directory. 3070 * 3071 * Return: 0 on success or -errno 3072 */ 3073 int __init rdtgroup_init(void) 3074 { 3075 int ret = 0; 3076 3077 seq_buf_init(&last_cmd_status, last_cmd_status_buf, 3078 sizeof(last_cmd_status_buf)); 3079 3080 ret = rdtgroup_setup_root(); 3081 if (ret) 3082 return ret; 3083 3084 ret = sysfs_create_mount_point(fs_kobj, "resctrl"); 3085 if (ret) 3086 goto cleanup_root; 3087 3088 ret = register_filesystem(&rdt_fs_type); 3089 if (ret) 3090 goto cleanup_mountpoint; 3091 3092 /* 3093 * Adding the resctrl debugfs directory here may not be ideal since 3094 * it would let the resctrl debugfs directory appear on the debugfs 3095 * filesystem before the resctrl filesystem is mounted. 3096 * It may also be ok since that would enable debugging of RDT before 3097 * resctrl is mounted. 3098 * The reason why the debugfs directory is created here and not in 3099 * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and 3100 * during the debugfs directory creation also &sb->s_type->i_mutex_key 3101 * (the lockdep class of inode->i_rwsem). Other filesystem 3102 * interactions (eg. SyS_getdents) have the lock ordering: 3103 * &sb->s_type->i_mutex_key --> &mm->mmap_sem 3104 * During mmap(), called with &mm->mmap_sem, the rdtgroup_mutex 3105 * is taken, thus creating dependency: 3106 * &mm->mmap_sem --> rdtgroup_mutex for the latter that can cause 3107 * issues considering the other two lock dependencies. 3108 * By creating the debugfs directory here we avoid a dependency 3109 * that may cause deadlock (even though file operations cannot 3110 * occur until the filesystem is mounted, but I do not know how to 3111 * tell lockdep that). 3112 */ 3113 debugfs_resctrl = debugfs_create_dir("resctrl", NULL); 3114 3115 return 0; 3116 3117 cleanup_mountpoint: 3118 sysfs_remove_mount_point(fs_kobj, "resctrl"); 3119 cleanup_root: 3120 kernfs_destroy_root(rdt_root); 3121 3122 return ret; 3123 } 3124 3125 void __exit rdtgroup_exit(void) 3126 { 3127 debugfs_remove_recursive(debugfs_resctrl); 3128 unregister_filesystem(&rdt_fs_type); 3129 sysfs_remove_mount_point(fs_kobj, "resctrl"); 3130 kernfs_destroy_root(rdt_root); 3131 } 3132