1 /* 2 * User interface for Resource Alloction in Resource Director Technology(RDT) 3 * 4 * Copyright (C) 2016 Intel Corporation 5 * 6 * Author: Fenghua Yu <fenghua.yu@intel.com> 7 * 8 * This program is free software; you can redistribute it and/or modify it 9 * under the terms and conditions of the GNU General Public License, 10 * version 2, as published by the Free Software Foundation. 11 * 12 * This program is distributed in the hope it will be useful, but WITHOUT 13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 15 * more details. 16 * 17 * More information about RDT be found in the Intel (R) x86 Architecture 18 * Software Developer Manual. 19 */ 20 21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 22 23 #include <linux/cacheinfo.h> 24 #include <linux/cpu.h> 25 #include <linux/debugfs.h> 26 #include <linux/fs.h> 27 #include <linux/sysfs.h> 28 #include <linux/kernfs.h> 29 #include <linux/seq_buf.h> 30 #include <linux/seq_file.h> 31 #include <linux/sched/signal.h> 32 #include <linux/sched/task.h> 33 #include <linux/slab.h> 34 #include <linux/task_work.h> 35 36 #include <uapi/linux/magic.h> 37 38 #include <asm/resctrl_sched.h> 39 #include "internal.h" 40 41 DEFINE_STATIC_KEY_FALSE(rdt_enable_key); 42 DEFINE_STATIC_KEY_FALSE(rdt_mon_enable_key); 43 DEFINE_STATIC_KEY_FALSE(rdt_alloc_enable_key); 44 static struct kernfs_root *rdt_root; 45 struct rdtgroup rdtgroup_default; 46 LIST_HEAD(rdt_all_groups); 47 48 /* Kernel fs node for "info" directory under root */ 49 static struct kernfs_node *kn_info; 50 51 /* Kernel fs node for "mon_groups" directory under root */ 52 static struct kernfs_node *kn_mongrp; 53 54 /* Kernel fs node for "mon_data" directory under root */ 55 static struct kernfs_node *kn_mondata; 56 57 static struct seq_buf last_cmd_status; 58 static char last_cmd_status_buf[512]; 59 60 struct dentry *debugfs_resctrl; 61 62 void rdt_last_cmd_clear(void) 63 { 64 lockdep_assert_held(&rdtgroup_mutex); 65 seq_buf_clear(&last_cmd_status); 66 } 67 68 void rdt_last_cmd_puts(const char *s) 69 { 70 lockdep_assert_held(&rdtgroup_mutex); 71 seq_buf_puts(&last_cmd_status, s); 72 } 73 74 void rdt_last_cmd_printf(const char *fmt, ...) 75 { 76 va_list ap; 77 78 va_start(ap, fmt); 79 lockdep_assert_held(&rdtgroup_mutex); 80 seq_buf_vprintf(&last_cmd_status, fmt, ap); 81 va_end(ap); 82 } 83 84 /* 85 * Trivial allocator for CLOSIDs. Since h/w only supports a small number, 86 * we can keep a bitmap of free CLOSIDs in a single integer. 87 * 88 * Using a global CLOSID across all resources has some advantages and 89 * some drawbacks: 90 * + We can simply set "current->closid" to assign a task to a resource 91 * group. 92 * + Context switch code can avoid extra memory references deciding which 93 * CLOSID to load into the PQR_ASSOC MSR 94 * - We give up some options in configuring resource groups across multi-socket 95 * systems. 96 * - Our choices on how to configure each resource become progressively more 97 * limited as the number of resources grows. 98 */ 99 static int closid_free_map; 100 static int closid_free_map_len; 101 102 int closids_supported(void) 103 { 104 return closid_free_map_len; 105 } 106 107 static void closid_init(void) 108 { 109 struct rdt_resource *r; 110 int rdt_min_closid = 32; 111 112 /* Compute rdt_min_closid across all resources */ 113 for_each_alloc_enabled_rdt_resource(r) 114 rdt_min_closid = min(rdt_min_closid, r->num_closid); 115 116 closid_free_map = BIT_MASK(rdt_min_closid) - 1; 117 118 /* CLOSID 0 is always reserved for the default group */ 119 closid_free_map &= ~1; 120 closid_free_map_len = rdt_min_closid; 121 } 122 123 static int closid_alloc(void) 124 { 125 u32 closid = ffs(closid_free_map); 126 127 if (closid == 0) 128 return -ENOSPC; 129 closid--; 130 closid_free_map &= ~(1 << closid); 131 132 return closid; 133 } 134 135 void closid_free(int closid) 136 { 137 closid_free_map |= 1 << closid; 138 } 139 140 /** 141 * closid_allocated - test if provided closid is in use 142 * @closid: closid to be tested 143 * 144 * Return: true if @closid is currently associated with a resource group, 145 * false if @closid is free 146 */ 147 static bool closid_allocated(unsigned int closid) 148 { 149 return (closid_free_map & (1 << closid)) == 0; 150 } 151 152 /** 153 * rdtgroup_mode_by_closid - Return mode of resource group with closid 154 * @closid: closid if the resource group 155 * 156 * Each resource group is associated with a @closid. Here the mode 157 * of a resource group can be queried by searching for it using its closid. 158 * 159 * Return: mode as &enum rdtgrp_mode of resource group with closid @closid 160 */ 161 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid) 162 { 163 struct rdtgroup *rdtgrp; 164 165 list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) { 166 if (rdtgrp->closid == closid) 167 return rdtgrp->mode; 168 } 169 170 return RDT_NUM_MODES; 171 } 172 173 static const char * const rdt_mode_str[] = { 174 [RDT_MODE_SHAREABLE] = "shareable", 175 [RDT_MODE_EXCLUSIVE] = "exclusive", 176 [RDT_MODE_PSEUDO_LOCKSETUP] = "pseudo-locksetup", 177 [RDT_MODE_PSEUDO_LOCKED] = "pseudo-locked", 178 }; 179 180 /** 181 * rdtgroup_mode_str - Return the string representation of mode 182 * @mode: the resource group mode as &enum rdtgroup_mode 183 * 184 * Return: string representation of valid mode, "unknown" otherwise 185 */ 186 static const char *rdtgroup_mode_str(enum rdtgrp_mode mode) 187 { 188 if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES) 189 return "unknown"; 190 191 return rdt_mode_str[mode]; 192 } 193 194 /* set uid and gid of rdtgroup dirs and files to that of the creator */ 195 static int rdtgroup_kn_set_ugid(struct kernfs_node *kn) 196 { 197 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, 198 .ia_uid = current_fsuid(), 199 .ia_gid = current_fsgid(), }; 200 201 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && 202 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) 203 return 0; 204 205 return kernfs_setattr(kn, &iattr); 206 } 207 208 static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft) 209 { 210 struct kernfs_node *kn; 211 int ret; 212 213 kn = __kernfs_create_file(parent_kn, rft->name, rft->mode, 214 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 215 0, rft->kf_ops, rft, NULL, NULL); 216 if (IS_ERR(kn)) 217 return PTR_ERR(kn); 218 219 ret = rdtgroup_kn_set_ugid(kn); 220 if (ret) { 221 kernfs_remove(kn); 222 return ret; 223 } 224 225 return 0; 226 } 227 228 static int rdtgroup_seqfile_show(struct seq_file *m, void *arg) 229 { 230 struct kernfs_open_file *of = m->private; 231 struct rftype *rft = of->kn->priv; 232 233 if (rft->seq_show) 234 return rft->seq_show(of, m, arg); 235 return 0; 236 } 237 238 static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf, 239 size_t nbytes, loff_t off) 240 { 241 struct rftype *rft = of->kn->priv; 242 243 if (rft->write) 244 return rft->write(of, buf, nbytes, off); 245 246 return -EINVAL; 247 } 248 249 static struct kernfs_ops rdtgroup_kf_single_ops = { 250 .atomic_write_len = PAGE_SIZE, 251 .write = rdtgroup_file_write, 252 .seq_show = rdtgroup_seqfile_show, 253 }; 254 255 static struct kernfs_ops kf_mondata_ops = { 256 .atomic_write_len = PAGE_SIZE, 257 .seq_show = rdtgroup_mondata_show, 258 }; 259 260 static bool is_cpu_list(struct kernfs_open_file *of) 261 { 262 struct rftype *rft = of->kn->priv; 263 264 return rft->flags & RFTYPE_FLAGS_CPUS_LIST; 265 } 266 267 static int rdtgroup_cpus_show(struct kernfs_open_file *of, 268 struct seq_file *s, void *v) 269 { 270 struct rdtgroup *rdtgrp; 271 struct cpumask *mask; 272 int ret = 0; 273 274 rdtgrp = rdtgroup_kn_lock_live(of->kn); 275 276 if (rdtgrp) { 277 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { 278 if (!rdtgrp->plr->d) { 279 rdt_last_cmd_clear(); 280 rdt_last_cmd_puts("Cache domain offline\n"); 281 ret = -ENODEV; 282 } else { 283 mask = &rdtgrp->plr->d->cpu_mask; 284 seq_printf(s, is_cpu_list(of) ? 285 "%*pbl\n" : "%*pb\n", 286 cpumask_pr_args(mask)); 287 } 288 } else { 289 seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n", 290 cpumask_pr_args(&rdtgrp->cpu_mask)); 291 } 292 } else { 293 ret = -ENOENT; 294 } 295 rdtgroup_kn_unlock(of->kn); 296 297 return ret; 298 } 299 300 /* 301 * This is safe against resctrl_sched_in() called from __switch_to() 302 * because __switch_to() is executed with interrupts disabled. A local call 303 * from update_closid_rmid() is proteced against __switch_to() because 304 * preemption is disabled. 305 */ 306 static void update_cpu_closid_rmid(void *info) 307 { 308 struct rdtgroup *r = info; 309 310 if (r) { 311 this_cpu_write(pqr_state.default_closid, r->closid); 312 this_cpu_write(pqr_state.default_rmid, r->mon.rmid); 313 } 314 315 /* 316 * We cannot unconditionally write the MSR because the current 317 * executing task might have its own closid selected. Just reuse 318 * the context switch code. 319 */ 320 resctrl_sched_in(); 321 } 322 323 /* 324 * Update the PGR_ASSOC MSR on all cpus in @cpu_mask, 325 * 326 * Per task closids/rmids must have been set up before calling this function. 327 */ 328 static void 329 update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r) 330 { 331 int cpu = get_cpu(); 332 333 if (cpumask_test_cpu(cpu, cpu_mask)) 334 update_cpu_closid_rmid(r); 335 smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1); 336 put_cpu(); 337 } 338 339 static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, 340 cpumask_var_t tmpmask) 341 { 342 struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp; 343 struct list_head *head; 344 345 /* Check whether cpus belong to parent ctrl group */ 346 cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask); 347 if (cpumask_weight(tmpmask)) { 348 rdt_last_cmd_puts("Can only add CPUs to mongroup that belong to parent\n"); 349 return -EINVAL; 350 } 351 352 /* Check whether cpus are dropped from this group */ 353 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); 354 if (cpumask_weight(tmpmask)) { 355 /* Give any dropped cpus to parent rdtgroup */ 356 cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask); 357 update_closid_rmid(tmpmask, prgrp); 358 } 359 360 /* 361 * If we added cpus, remove them from previous group that owned them 362 * and update per-cpu rmid 363 */ 364 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); 365 if (cpumask_weight(tmpmask)) { 366 head = &prgrp->mon.crdtgrp_list; 367 list_for_each_entry(crgrp, head, mon.crdtgrp_list) { 368 if (crgrp == rdtgrp) 369 continue; 370 cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask, 371 tmpmask); 372 } 373 update_closid_rmid(tmpmask, rdtgrp); 374 } 375 376 /* Done pushing/pulling - update this group with new mask */ 377 cpumask_copy(&rdtgrp->cpu_mask, newmask); 378 379 return 0; 380 } 381 382 static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m) 383 { 384 struct rdtgroup *crgrp; 385 386 cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m); 387 /* update the child mon group masks as well*/ 388 list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list) 389 cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask); 390 } 391 392 static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, 393 cpumask_var_t tmpmask, cpumask_var_t tmpmask1) 394 { 395 struct rdtgroup *r, *crgrp; 396 struct list_head *head; 397 398 /* Check whether cpus are dropped from this group */ 399 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); 400 if (cpumask_weight(tmpmask)) { 401 /* Can't drop from default group */ 402 if (rdtgrp == &rdtgroup_default) { 403 rdt_last_cmd_puts("Can't drop CPUs from default group\n"); 404 return -EINVAL; 405 } 406 407 /* Give any dropped cpus to rdtgroup_default */ 408 cpumask_or(&rdtgroup_default.cpu_mask, 409 &rdtgroup_default.cpu_mask, tmpmask); 410 update_closid_rmid(tmpmask, &rdtgroup_default); 411 } 412 413 /* 414 * If we added cpus, remove them from previous group and 415 * the prev group's child groups that owned them 416 * and update per-cpu closid/rmid. 417 */ 418 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); 419 if (cpumask_weight(tmpmask)) { 420 list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) { 421 if (r == rdtgrp) 422 continue; 423 cpumask_and(tmpmask1, &r->cpu_mask, tmpmask); 424 if (cpumask_weight(tmpmask1)) 425 cpumask_rdtgrp_clear(r, tmpmask1); 426 } 427 update_closid_rmid(tmpmask, rdtgrp); 428 } 429 430 /* Done pushing/pulling - update this group with new mask */ 431 cpumask_copy(&rdtgrp->cpu_mask, newmask); 432 433 /* 434 * Clear child mon group masks since there is a new parent mask 435 * now and update the rmid for the cpus the child lost. 436 */ 437 head = &rdtgrp->mon.crdtgrp_list; 438 list_for_each_entry(crgrp, head, mon.crdtgrp_list) { 439 cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask); 440 update_closid_rmid(tmpmask, rdtgrp); 441 cpumask_clear(&crgrp->cpu_mask); 442 } 443 444 return 0; 445 } 446 447 static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of, 448 char *buf, size_t nbytes, loff_t off) 449 { 450 cpumask_var_t tmpmask, newmask, tmpmask1; 451 struct rdtgroup *rdtgrp; 452 int ret; 453 454 if (!buf) 455 return -EINVAL; 456 457 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) 458 return -ENOMEM; 459 if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) { 460 free_cpumask_var(tmpmask); 461 return -ENOMEM; 462 } 463 if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) { 464 free_cpumask_var(tmpmask); 465 free_cpumask_var(newmask); 466 return -ENOMEM; 467 } 468 469 rdtgrp = rdtgroup_kn_lock_live(of->kn); 470 rdt_last_cmd_clear(); 471 if (!rdtgrp) { 472 ret = -ENOENT; 473 rdt_last_cmd_puts("Directory was removed\n"); 474 goto unlock; 475 } 476 477 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || 478 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 479 ret = -EINVAL; 480 rdt_last_cmd_puts("Pseudo-locking in progress\n"); 481 goto unlock; 482 } 483 484 if (is_cpu_list(of)) 485 ret = cpulist_parse(buf, newmask); 486 else 487 ret = cpumask_parse(buf, newmask); 488 489 if (ret) { 490 rdt_last_cmd_puts("Bad CPU list/mask\n"); 491 goto unlock; 492 } 493 494 /* check that user didn't specify any offline cpus */ 495 cpumask_andnot(tmpmask, newmask, cpu_online_mask); 496 if (cpumask_weight(tmpmask)) { 497 ret = -EINVAL; 498 rdt_last_cmd_puts("Can only assign online CPUs\n"); 499 goto unlock; 500 } 501 502 if (rdtgrp->type == RDTCTRL_GROUP) 503 ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1); 504 else if (rdtgrp->type == RDTMON_GROUP) 505 ret = cpus_mon_write(rdtgrp, newmask, tmpmask); 506 else 507 ret = -EINVAL; 508 509 unlock: 510 rdtgroup_kn_unlock(of->kn); 511 free_cpumask_var(tmpmask); 512 free_cpumask_var(newmask); 513 free_cpumask_var(tmpmask1); 514 515 return ret ?: nbytes; 516 } 517 518 struct task_move_callback { 519 struct callback_head work; 520 struct rdtgroup *rdtgrp; 521 }; 522 523 static void move_myself(struct callback_head *head) 524 { 525 struct task_move_callback *callback; 526 struct rdtgroup *rdtgrp; 527 528 callback = container_of(head, struct task_move_callback, work); 529 rdtgrp = callback->rdtgrp; 530 531 /* 532 * If resource group was deleted before this task work callback 533 * was invoked, then assign the task to root group and free the 534 * resource group. 535 */ 536 if (atomic_dec_and_test(&rdtgrp->waitcount) && 537 (rdtgrp->flags & RDT_DELETED)) { 538 current->closid = 0; 539 current->rmid = 0; 540 kfree(rdtgrp); 541 } 542 543 preempt_disable(); 544 /* update PQR_ASSOC MSR to make resource group go into effect */ 545 resctrl_sched_in(); 546 preempt_enable(); 547 548 kfree(callback); 549 } 550 551 static int __rdtgroup_move_task(struct task_struct *tsk, 552 struct rdtgroup *rdtgrp) 553 { 554 struct task_move_callback *callback; 555 int ret; 556 557 callback = kzalloc(sizeof(*callback), GFP_KERNEL); 558 if (!callback) 559 return -ENOMEM; 560 callback->work.func = move_myself; 561 callback->rdtgrp = rdtgrp; 562 563 /* 564 * Take a refcount, so rdtgrp cannot be freed before the 565 * callback has been invoked. 566 */ 567 atomic_inc(&rdtgrp->waitcount); 568 ret = task_work_add(tsk, &callback->work, true); 569 if (ret) { 570 /* 571 * Task is exiting. Drop the refcount and free the callback. 572 * No need to check the refcount as the group cannot be 573 * deleted before the write function unlocks rdtgroup_mutex. 574 */ 575 atomic_dec(&rdtgrp->waitcount); 576 kfree(callback); 577 rdt_last_cmd_puts("Task exited\n"); 578 } else { 579 /* 580 * For ctrl_mon groups move both closid and rmid. 581 * For monitor groups, can move the tasks only from 582 * their parent CTRL group. 583 */ 584 if (rdtgrp->type == RDTCTRL_GROUP) { 585 tsk->closid = rdtgrp->closid; 586 tsk->rmid = rdtgrp->mon.rmid; 587 } else if (rdtgrp->type == RDTMON_GROUP) { 588 if (rdtgrp->mon.parent->closid == tsk->closid) { 589 tsk->rmid = rdtgrp->mon.rmid; 590 } else { 591 rdt_last_cmd_puts("Can't move task to different control group\n"); 592 ret = -EINVAL; 593 } 594 } 595 } 596 return ret; 597 } 598 599 /** 600 * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group 601 * @r: Resource group 602 * 603 * Return: 1 if tasks have been assigned to @r, 0 otherwise 604 */ 605 int rdtgroup_tasks_assigned(struct rdtgroup *r) 606 { 607 struct task_struct *p, *t; 608 int ret = 0; 609 610 lockdep_assert_held(&rdtgroup_mutex); 611 612 rcu_read_lock(); 613 for_each_process_thread(p, t) { 614 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) || 615 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) { 616 ret = 1; 617 break; 618 } 619 } 620 rcu_read_unlock(); 621 622 return ret; 623 } 624 625 static int rdtgroup_task_write_permission(struct task_struct *task, 626 struct kernfs_open_file *of) 627 { 628 const struct cred *tcred = get_task_cred(task); 629 const struct cred *cred = current_cred(); 630 int ret = 0; 631 632 /* 633 * Even if we're attaching all tasks in the thread group, we only 634 * need to check permissions on one of them. 635 */ 636 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && 637 !uid_eq(cred->euid, tcred->uid) && 638 !uid_eq(cred->euid, tcred->suid)) { 639 rdt_last_cmd_printf("No permission to move task %d\n", task->pid); 640 ret = -EPERM; 641 } 642 643 put_cred(tcred); 644 return ret; 645 } 646 647 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp, 648 struct kernfs_open_file *of) 649 { 650 struct task_struct *tsk; 651 int ret; 652 653 rcu_read_lock(); 654 if (pid) { 655 tsk = find_task_by_vpid(pid); 656 if (!tsk) { 657 rcu_read_unlock(); 658 rdt_last_cmd_printf("No task %d\n", pid); 659 return -ESRCH; 660 } 661 } else { 662 tsk = current; 663 } 664 665 get_task_struct(tsk); 666 rcu_read_unlock(); 667 668 ret = rdtgroup_task_write_permission(tsk, of); 669 if (!ret) 670 ret = __rdtgroup_move_task(tsk, rdtgrp); 671 672 put_task_struct(tsk); 673 return ret; 674 } 675 676 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of, 677 char *buf, size_t nbytes, loff_t off) 678 { 679 struct rdtgroup *rdtgrp; 680 int ret = 0; 681 pid_t pid; 682 683 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) 684 return -EINVAL; 685 rdtgrp = rdtgroup_kn_lock_live(of->kn); 686 if (!rdtgrp) { 687 rdtgroup_kn_unlock(of->kn); 688 return -ENOENT; 689 } 690 rdt_last_cmd_clear(); 691 692 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || 693 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 694 ret = -EINVAL; 695 rdt_last_cmd_puts("Pseudo-locking in progress\n"); 696 goto unlock; 697 } 698 699 ret = rdtgroup_move_task(pid, rdtgrp, of); 700 701 unlock: 702 rdtgroup_kn_unlock(of->kn); 703 704 return ret ?: nbytes; 705 } 706 707 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s) 708 { 709 struct task_struct *p, *t; 710 711 rcu_read_lock(); 712 for_each_process_thread(p, t) { 713 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) || 714 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) 715 seq_printf(s, "%d\n", t->pid); 716 } 717 rcu_read_unlock(); 718 } 719 720 static int rdtgroup_tasks_show(struct kernfs_open_file *of, 721 struct seq_file *s, void *v) 722 { 723 struct rdtgroup *rdtgrp; 724 int ret = 0; 725 726 rdtgrp = rdtgroup_kn_lock_live(of->kn); 727 if (rdtgrp) 728 show_rdt_tasks(rdtgrp, s); 729 else 730 ret = -ENOENT; 731 rdtgroup_kn_unlock(of->kn); 732 733 return ret; 734 } 735 736 static int rdt_last_cmd_status_show(struct kernfs_open_file *of, 737 struct seq_file *seq, void *v) 738 { 739 int len; 740 741 mutex_lock(&rdtgroup_mutex); 742 len = seq_buf_used(&last_cmd_status); 743 if (len) 744 seq_printf(seq, "%.*s", len, last_cmd_status_buf); 745 else 746 seq_puts(seq, "ok\n"); 747 mutex_unlock(&rdtgroup_mutex); 748 return 0; 749 } 750 751 static int rdt_num_closids_show(struct kernfs_open_file *of, 752 struct seq_file *seq, void *v) 753 { 754 struct rdt_resource *r = of->kn->parent->priv; 755 756 seq_printf(seq, "%d\n", r->num_closid); 757 return 0; 758 } 759 760 static int rdt_default_ctrl_show(struct kernfs_open_file *of, 761 struct seq_file *seq, void *v) 762 { 763 struct rdt_resource *r = of->kn->parent->priv; 764 765 seq_printf(seq, "%x\n", r->default_ctrl); 766 return 0; 767 } 768 769 static int rdt_min_cbm_bits_show(struct kernfs_open_file *of, 770 struct seq_file *seq, void *v) 771 { 772 struct rdt_resource *r = of->kn->parent->priv; 773 774 seq_printf(seq, "%u\n", r->cache.min_cbm_bits); 775 return 0; 776 } 777 778 static int rdt_shareable_bits_show(struct kernfs_open_file *of, 779 struct seq_file *seq, void *v) 780 { 781 struct rdt_resource *r = of->kn->parent->priv; 782 783 seq_printf(seq, "%x\n", r->cache.shareable_bits); 784 return 0; 785 } 786 787 /** 788 * rdt_bit_usage_show - Display current usage of resources 789 * 790 * A domain is a shared resource that can now be allocated differently. Here 791 * we display the current regions of the domain as an annotated bitmask. 792 * For each domain of this resource its allocation bitmask 793 * is annotated as below to indicate the current usage of the corresponding bit: 794 * 0 - currently unused 795 * X - currently available for sharing and used by software and hardware 796 * H - currently used by hardware only but available for software use 797 * S - currently used and shareable by software only 798 * E - currently used exclusively by one resource group 799 * P - currently pseudo-locked by one resource group 800 */ 801 static int rdt_bit_usage_show(struct kernfs_open_file *of, 802 struct seq_file *seq, void *v) 803 { 804 struct rdt_resource *r = of->kn->parent->priv; 805 u32 sw_shareable = 0, hw_shareable = 0; 806 u32 exclusive = 0, pseudo_locked = 0; 807 struct rdt_domain *dom; 808 int i, hwb, swb, excl, psl; 809 enum rdtgrp_mode mode; 810 bool sep = false; 811 u32 *ctrl; 812 813 mutex_lock(&rdtgroup_mutex); 814 hw_shareable = r->cache.shareable_bits; 815 list_for_each_entry(dom, &r->domains, list) { 816 if (sep) 817 seq_putc(seq, ';'); 818 ctrl = dom->ctrl_val; 819 sw_shareable = 0; 820 exclusive = 0; 821 seq_printf(seq, "%d=", dom->id); 822 for (i = 0; i < closids_supported(); i++, ctrl++) { 823 if (!closid_allocated(i)) 824 continue; 825 mode = rdtgroup_mode_by_closid(i); 826 switch (mode) { 827 case RDT_MODE_SHAREABLE: 828 sw_shareable |= *ctrl; 829 break; 830 case RDT_MODE_EXCLUSIVE: 831 exclusive |= *ctrl; 832 break; 833 case RDT_MODE_PSEUDO_LOCKSETUP: 834 /* 835 * RDT_MODE_PSEUDO_LOCKSETUP is possible 836 * here but not included since the CBM 837 * associated with this CLOSID in this mode 838 * is not initialized and no task or cpu can be 839 * assigned this CLOSID. 840 */ 841 break; 842 case RDT_MODE_PSEUDO_LOCKED: 843 case RDT_NUM_MODES: 844 WARN(1, 845 "invalid mode for closid %d\n", i); 846 break; 847 } 848 } 849 for (i = r->cache.cbm_len - 1; i >= 0; i--) { 850 pseudo_locked = dom->plr ? dom->plr->cbm : 0; 851 hwb = test_bit(i, (unsigned long *)&hw_shareable); 852 swb = test_bit(i, (unsigned long *)&sw_shareable); 853 excl = test_bit(i, (unsigned long *)&exclusive); 854 psl = test_bit(i, (unsigned long *)&pseudo_locked); 855 if (hwb && swb) 856 seq_putc(seq, 'X'); 857 else if (hwb && !swb) 858 seq_putc(seq, 'H'); 859 else if (!hwb && swb) 860 seq_putc(seq, 'S'); 861 else if (excl) 862 seq_putc(seq, 'E'); 863 else if (psl) 864 seq_putc(seq, 'P'); 865 else /* Unused bits remain */ 866 seq_putc(seq, '0'); 867 } 868 sep = true; 869 } 870 seq_putc(seq, '\n'); 871 mutex_unlock(&rdtgroup_mutex); 872 return 0; 873 } 874 875 static int rdt_min_bw_show(struct kernfs_open_file *of, 876 struct seq_file *seq, void *v) 877 { 878 struct rdt_resource *r = of->kn->parent->priv; 879 880 seq_printf(seq, "%u\n", r->membw.min_bw); 881 return 0; 882 } 883 884 static int rdt_num_rmids_show(struct kernfs_open_file *of, 885 struct seq_file *seq, void *v) 886 { 887 struct rdt_resource *r = of->kn->parent->priv; 888 889 seq_printf(seq, "%d\n", r->num_rmid); 890 891 return 0; 892 } 893 894 static int rdt_mon_features_show(struct kernfs_open_file *of, 895 struct seq_file *seq, void *v) 896 { 897 struct rdt_resource *r = of->kn->parent->priv; 898 struct mon_evt *mevt; 899 900 list_for_each_entry(mevt, &r->evt_list, list) 901 seq_printf(seq, "%s\n", mevt->name); 902 903 return 0; 904 } 905 906 static int rdt_bw_gran_show(struct kernfs_open_file *of, 907 struct seq_file *seq, void *v) 908 { 909 struct rdt_resource *r = of->kn->parent->priv; 910 911 seq_printf(seq, "%u\n", r->membw.bw_gran); 912 return 0; 913 } 914 915 static int rdt_delay_linear_show(struct kernfs_open_file *of, 916 struct seq_file *seq, void *v) 917 { 918 struct rdt_resource *r = of->kn->parent->priv; 919 920 seq_printf(seq, "%u\n", r->membw.delay_linear); 921 return 0; 922 } 923 924 static int max_threshold_occ_show(struct kernfs_open_file *of, 925 struct seq_file *seq, void *v) 926 { 927 struct rdt_resource *r = of->kn->parent->priv; 928 929 seq_printf(seq, "%u\n", resctrl_cqm_threshold * r->mon_scale); 930 931 return 0; 932 } 933 934 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of, 935 char *buf, size_t nbytes, loff_t off) 936 { 937 struct rdt_resource *r = of->kn->parent->priv; 938 unsigned int bytes; 939 int ret; 940 941 ret = kstrtouint(buf, 0, &bytes); 942 if (ret) 943 return ret; 944 945 if (bytes > (boot_cpu_data.x86_cache_size * 1024)) 946 return -EINVAL; 947 948 resctrl_cqm_threshold = bytes / r->mon_scale; 949 950 return nbytes; 951 } 952 953 /* 954 * rdtgroup_mode_show - Display mode of this resource group 955 */ 956 static int rdtgroup_mode_show(struct kernfs_open_file *of, 957 struct seq_file *s, void *v) 958 { 959 struct rdtgroup *rdtgrp; 960 961 rdtgrp = rdtgroup_kn_lock_live(of->kn); 962 if (!rdtgrp) { 963 rdtgroup_kn_unlock(of->kn); 964 return -ENOENT; 965 } 966 967 seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode)); 968 969 rdtgroup_kn_unlock(of->kn); 970 return 0; 971 } 972 973 /** 974 * rdt_cdp_peer_get - Retrieve CDP peer if it exists 975 * @r: RDT resource to which RDT domain @d belongs 976 * @d: Cache instance for which a CDP peer is requested 977 * @r_cdp: RDT resource that shares hardware with @r (RDT resource peer) 978 * Used to return the result. 979 * @d_cdp: RDT domain that shares hardware with @d (RDT domain peer) 980 * Used to return the result. 981 * 982 * RDT resources are managed independently and by extension the RDT domains 983 * (RDT resource instances) are managed independently also. The Code and 984 * Data Prioritization (CDP) RDT resources, while managed independently, 985 * could refer to the same underlying hardware. For example, 986 * RDT_RESOURCE_L2CODE and RDT_RESOURCE_L2DATA both refer to the L2 cache. 987 * 988 * When provided with an RDT resource @r and an instance of that RDT 989 * resource @d rdt_cdp_peer_get() will return if there is a peer RDT 990 * resource and the exact instance that shares the same hardware. 991 * 992 * Return: 0 if a CDP peer was found, <0 on error or if no CDP peer exists. 993 * If a CDP peer was found, @r_cdp will point to the peer RDT resource 994 * and @d_cdp will point to the peer RDT domain. 995 */ 996 static int rdt_cdp_peer_get(struct rdt_resource *r, struct rdt_domain *d, 997 struct rdt_resource **r_cdp, 998 struct rdt_domain **d_cdp) 999 { 1000 struct rdt_resource *_r_cdp = NULL; 1001 struct rdt_domain *_d_cdp = NULL; 1002 int ret = 0; 1003 1004 switch (r->rid) { 1005 case RDT_RESOURCE_L3DATA: 1006 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3CODE]; 1007 break; 1008 case RDT_RESOURCE_L3CODE: 1009 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3DATA]; 1010 break; 1011 case RDT_RESOURCE_L2DATA: 1012 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2CODE]; 1013 break; 1014 case RDT_RESOURCE_L2CODE: 1015 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2DATA]; 1016 break; 1017 default: 1018 ret = -ENOENT; 1019 goto out; 1020 } 1021 1022 /* 1023 * When a new CPU comes online and CDP is enabled then the new 1024 * RDT domains (if any) associated with both CDP RDT resources 1025 * are added in the same CPU online routine while the 1026 * rdtgroup_mutex is held. It should thus not happen for one 1027 * RDT domain to exist and be associated with its RDT CDP 1028 * resource but there is no RDT domain associated with the 1029 * peer RDT CDP resource. Hence the WARN. 1030 */ 1031 _d_cdp = rdt_find_domain(_r_cdp, d->id, NULL); 1032 if (WARN_ON(!_d_cdp)) { 1033 _r_cdp = NULL; 1034 ret = -EINVAL; 1035 } 1036 1037 out: 1038 *r_cdp = _r_cdp; 1039 *d_cdp = _d_cdp; 1040 1041 return ret; 1042 } 1043 1044 /** 1045 * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other 1046 * @r: Resource to which domain instance @d belongs. 1047 * @d: The domain instance for which @closid is being tested. 1048 * @cbm: Capacity bitmask being tested. 1049 * @closid: Intended closid for @cbm. 1050 * @exclusive: Only check if overlaps with exclusive resource groups 1051 * 1052 * Checks if provided @cbm intended to be used for @closid on domain 1053 * @d overlaps with any other closids or other hardware usage associated 1054 * with this domain. If @exclusive is true then only overlaps with 1055 * resource groups in exclusive mode will be considered. If @exclusive 1056 * is false then overlaps with any resource group or hardware entities 1057 * will be considered. 1058 * 1059 * @cbm is unsigned long, even if only 32 bits are used, to make the 1060 * bitmap functions work correctly. 1061 * 1062 * Return: false if CBM does not overlap, true if it does. 1063 */ 1064 static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d, 1065 unsigned long cbm, int closid, bool exclusive) 1066 { 1067 enum rdtgrp_mode mode; 1068 unsigned long ctrl_b; 1069 u32 *ctrl; 1070 int i; 1071 1072 /* Check for any overlap with regions used by hardware directly */ 1073 if (!exclusive) { 1074 ctrl_b = r->cache.shareable_bits; 1075 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) 1076 return true; 1077 } 1078 1079 /* Check for overlap with other resource groups */ 1080 ctrl = d->ctrl_val; 1081 for (i = 0; i < closids_supported(); i++, ctrl++) { 1082 ctrl_b = *ctrl; 1083 mode = rdtgroup_mode_by_closid(i); 1084 if (closid_allocated(i) && i != closid && 1085 mode != RDT_MODE_PSEUDO_LOCKSETUP) { 1086 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) { 1087 if (exclusive) { 1088 if (mode == RDT_MODE_EXCLUSIVE) 1089 return true; 1090 continue; 1091 } 1092 return true; 1093 } 1094 } 1095 } 1096 1097 return false; 1098 } 1099 1100 /** 1101 * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware 1102 * @r: Resource to which domain instance @d belongs. 1103 * @d: The domain instance for which @closid is being tested. 1104 * @cbm: Capacity bitmask being tested. 1105 * @closid: Intended closid for @cbm. 1106 * @exclusive: Only check if overlaps with exclusive resource groups 1107 * 1108 * Resources that can be allocated using a CBM can use the CBM to control 1109 * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test 1110 * for overlap. Overlap test is not limited to the specific resource for 1111 * which the CBM is intended though - when dealing with CDP resources that 1112 * share the underlying hardware the overlap check should be performed on 1113 * the CDP resource sharing the hardware also. 1114 * 1115 * Refer to description of __rdtgroup_cbm_overlaps() for the details of the 1116 * overlap test. 1117 * 1118 * Return: true if CBM overlap detected, false if there is no overlap 1119 */ 1120 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d, 1121 unsigned long cbm, int closid, bool exclusive) 1122 { 1123 struct rdt_resource *r_cdp; 1124 struct rdt_domain *d_cdp; 1125 1126 if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, exclusive)) 1127 return true; 1128 1129 if (rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp) < 0) 1130 return false; 1131 1132 return __rdtgroup_cbm_overlaps(r_cdp, d_cdp, cbm, closid, exclusive); 1133 } 1134 1135 /** 1136 * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive 1137 * 1138 * An exclusive resource group implies that there should be no sharing of 1139 * its allocated resources. At the time this group is considered to be 1140 * exclusive this test can determine if its current schemata supports this 1141 * setting by testing for overlap with all other resource groups. 1142 * 1143 * Return: true if resource group can be exclusive, false if there is overlap 1144 * with allocations of other resource groups and thus this resource group 1145 * cannot be exclusive. 1146 */ 1147 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp) 1148 { 1149 int closid = rdtgrp->closid; 1150 struct rdt_resource *r; 1151 bool has_cache = false; 1152 struct rdt_domain *d; 1153 1154 for_each_alloc_enabled_rdt_resource(r) { 1155 if (r->rid == RDT_RESOURCE_MBA) 1156 continue; 1157 has_cache = true; 1158 list_for_each_entry(d, &r->domains, list) { 1159 if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid], 1160 rdtgrp->closid, false)) { 1161 rdt_last_cmd_puts("Schemata overlaps\n"); 1162 return false; 1163 } 1164 } 1165 } 1166 1167 if (!has_cache) { 1168 rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n"); 1169 return false; 1170 } 1171 1172 return true; 1173 } 1174 1175 /** 1176 * rdtgroup_mode_write - Modify the resource group's mode 1177 * 1178 */ 1179 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of, 1180 char *buf, size_t nbytes, loff_t off) 1181 { 1182 struct rdtgroup *rdtgrp; 1183 enum rdtgrp_mode mode; 1184 int ret = 0; 1185 1186 /* Valid input requires a trailing newline */ 1187 if (nbytes == 0 || buf[nbytes - 1] != '\n') 1188 return -EINVAL; 1189 buf[nbytes - 1] = '\0'; 1190 1191 rdtgrp = rdtgroup_kn_lock_live(of->kn); 1192 if (!rdtgrp) { 1193 rdtgroup_kn_unlock(of->kn); 1194 return -ENOENT; 1195 } 1196 1197 rdt_last_cmd_clear(); 1198 1199 mode = rdtgrp->mode; 1200 1201 if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) || 1202 (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) || 1203 (!strcmp(buf, "pseudo-locksetup") && 1204 mode == RDT_MODE_PSEUDO_LOCKSETUP) || 1205 (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED)) 1206 goto out; 1207 1208 if (mode == RDT_MODE_PSEUDO_LOCKED) { 1209 rdt_last_cmd_puts("Cannot change pseudo-locked group\n"); 1210 ret = -EINVAL; 1211 goto out; 1212 } 1213 1214 if (!strcmp(buf, "shareable")) { 1215 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 1216 ret = rdtgroup_locksetup_exit(rdtgrp); 1217 if (ret) 1218 goto out; 1219 } 1220 rdtgrp->mode = RDT_MODE_SHAREABLE; 1221 } else if (!strcmp(buf, "exclusive")) { 1222 if (!rdtgroup_mode_test_exclusive(rdtgrp)) { 1223 ret = -EINVAL; 1224 goto out; 1225 } 1226 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 1227 ret = rdtgroup_locksetup_exit(rdtgrp); 1228 if (ret) 1229 goto out; 1230 } 1231 rdtgrp->mode = RDT_MODE_EXCLUSIVE; 1232 } else if (!strcmp(buf, "pseudo-locksetup")) { 1233 ret = rdtgroup_locksetup_enter(rdtgrp); 1234 if (ret) 1235 goto out; 1236 rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP; 1237 } else { 1238 rdt_last_cmd_puts("Unknown or unsupported mode\n"); 1239 ret = -EINVAL; 1240 } 1241 1242 out: 1243 rdtgroup_kn_unlock(of->kn); 1244 return ret ?: nbytes; 1245 } 1246 1247 /** 1248 * rdtgroup_cbm_to_size - Translate CBM to size in bytes 1249 * @r: RDT resource to which @d belongs. 1250 * @d: RDT domain instance. 1251 * @cbm: bitmask for which the size should be computed. 1252 * 1253 * The bitmask provided associated with the RDT domain instance @d will be 1254 * translated into how many bytes it represents. The size in bytes is 1255 * computed by first dividing the total cache size by the CBM length to 1256 * determine how many bytes each bit in the bitmask represents. The result 1257 * is multiplied with the number of bits set in the bitmask. 1258 * 1259 * @cbm is unsigned long, even if only 32 bits are used to make the 1260 * bitmap functions work correctly. 1261 */ 1262 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, 1263 struct rdt_domain *d, unsigned long cbm) 1264 { 1265 struct cpu_cacheinfo *ci; 1266 unsigned int size = 0; 1267 int num_b, i; 1268 1269 num_b = bitmap_weight(&cbm, r->cache.cbm_len); 1270 ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask)); 1271 for (i = 0; i < ci->num_leaves; i++) { 1272 if (ci->info_list[i].level == r->cache_level) { 1273 size = ci->info_list[i].size / r->cache.cbm_len * num_b; 1274 break; 1275 } 1276 } 1277 1278 return size; 1279 } 1280 1281 /** 1282 * rdtgroup_size_show - Display size in bytes of allocated regions 1283 * 1284 * The "size" file mirrors the layout of the "schemata" file, printing the 1285 * size in bytes of each region instead of the capacity bitmask. 1286 * 1287 */ 1288 static int rdtgroup_size_show(struct kernfs_open_file *of, 1289 struct seq_file *s, void *v) 1290 { 1291 struct rdtgroup *rdtgrp; 1292 struct rdt_resource *r; 1293 struct rdt_domain *d; 1294 unsigned int size; 1295 int ret = 0; 1296 bool sep; 1297 u32 ctrl; 1298 1299 rdtgrp = rdtgroup_kn_lock_live(of->kn); 1300 if (!rdtgrp) { 1301 rdtgroup_kn_unlock(of->kn); 1302 return -ENOENT; 1303 } 1304 1305 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { 1306 if (!rdtgrp->plr->d) { 1307 rdt_last_cmd_clear(); 1308 rdt_last_cmd_puts("Cache domain offline\n"); 1309 ret = -ENODEV; 1310 } else { 1311 seq_printf(s, "%*s:", max_name_width, 1312 rdtgrp->plr->r->name); 1313 size = rdtgroup_cbm_to_size(rdtgrp->plr->r, 1314 rdtgrp->plr->d, 1315 rdtgrp->plr->cbm); 1316 seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size); 1317 } 1318 goto out; 1319 } 1320 1321 for_each_alloc_enabled_rdt_resource(r) { 1322 sep = false; 1323 seq_printf(s, "%*s:", max_name_width, r->name); 1324 list_for_each_entry(d, &r->domains, list) { 1325 if (sep) 1326 seq_putc(s, ';'); 1327 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { 1328 size = 0; 1329 } else { 1330 ctrl = (!is_mba_sc(r) ? 1331 d->ctrl_val[rdtgrp->closid] : 1332 d->mbps_val[rdtgrp->closid]); 1333 if (r->rid == RDT_RESOURCE_MBA) 1334 size = ctrl; 1335 else 1336 size = rdtgroup_cbm_to_size(r, d, ctrl); 1337 } 1338 seq_printf(s, "%d=%u", d->id, size); 1339 sep = true; 1340 } 1341 seq_putc(s, '\n'); 1342 } 1343 1344 out: 1345 rdtgroup_kn_unlock(of->kn); 1346 1347 return ret; 1348 } 1349 1350 /* rdtgroup information files for one cache resource. */ 1351 static struct rftype res_common_files[] = { 1352 { 1353 .name = "last_cmd_status", 1354 .mode = 0444, 1355 .kf_ops = &rdtgroup_kf_single_ops, 1356 .seq_show = rdt_last_cmd_status_show, 1357 .fflags = RF_TOP_INFO, 1358 }, 1359 { 1360 .name = "num_closids", 1361 .mode = 0444, 1362 .kf_ops = &rdtgroup_kf_single_ops, 1363 .seq_show = rdt_num_closids_show, 1364 .fflags = RF_CTRL_INFO, 1365 }, 1366 { 1367 .name = "mon_features", 1368 .mode = 0444, 1369 .kf_ops = &rdtgroup_kf_single_ops, 1370 .seq_show = rdt_mon_features_show, 1371 .fflags = RF_MON_INFO, 1372 }, 1373 { 1374 .name = "num_rmids", 1375 .mode = 0444, 1376 .kf_ops = &rdtgroup_kf_single_ops, 1377 .seq_show = rdt_num_rmids_show, 1378 .fflags = RF_MON_INFO, 1379 }, 1380 { 1381 .name = "cbm_mask", 1382 .mode = 0444, 1383 .kf_ops = &rdtgroup_kf_single_ops, 1384 .seq_show = rdt_default_ctrl_show, 1385 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, 1386 }, 1387 { 1388 .name = "min_cbm_bits", 1389 .mode = 0444, 1390 .kf_ops = &rdtgroup_kf_single_ops, 1391 .seq_show = rdt_min_cbm_bits_show, 1392 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, 1393 }, 1394 { 1395 .name = "shareable_bits", 1396 .mode = 0444, 1397 .kf_ops = &rdtgroup_kf_single_ops, 1398 .seq_show = rdt_shareable_bits_show, 1399 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, 1400 }, 1401 { 1402 .name = "bit_usage", 1403 .mode = 0444, 1404 .kf_ops = &rdtgroup_kf_single_ops, 1405 .seq_show = rdt_bit_usage_show, 1406 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, 1407 }, 1408 { 1409 .name = "min_bandwidth", 1410 .mode = 0444, 1411 .kf_ops = &rdtgroup_kf_single_ops, 1412 .seq_show = rdt_min_bw_show, 1413 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, 1414 }, 1415 { 1416 .name = "bandwidth_gran", 1417 .mode = 0444, 1418 .kf_ops = &rdtgroup_kf_single_ops, 1419 .seq_show = rdt_bw_gran_show, 1420 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, 1421 }, 1422 { 1423 .name = "delay_linear", 1424 .mode = 0444, 1425 .kf_ops = &rdtgroup_kf_single_ops, 1426 .seq_show = rdt_delay_linear_show, 1427 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, 1428 }, 1429 { 1430 .name = "max_threshold_occupancy", 1431 .mode = 0644, 1432 .kf_ops = &rdtgroup_kf_single_ops, 1433 .write = max_threshold_occ_write, 1434 .seq_show = max_threshold_occ_show, 1435 .fflags = RF_MON_INFO | RFTYPE_RES_CACHE, 1436 }, 1437 { 1438 .name = "cpus", 1439 .mode = 0644, 1440 .kf_ops = &rdtgroup_kf_single_ops, 1441 .write = rdtgroup_cpus_write, 1442 .seq_show = rdtgroup_cpus_show, 1443 .fflags = RFTYPE_BASE, 1444 }, 1445 { 1446 .name = "cpus_list", 1447 .mode = 0644, 1448 .kf_ops = &rdtgroup_kf_single_ops, 1449 .write = rdtgroup_cpus_write, 1450 .seq_show = rdtgroup_cpus_show, 1451 .flags = RFTYPE_FLAGS_CPUS_LIST, 1452 .fflags = RFTYPE_BASE, 1453 }, 1454 { 1455 .name = "tasks", 1456 .mode = 0644, 1457 .kf_ops = &rdtgroup_kf_single_ops, 1458 .write = rdtgroup_tasks_write, 1459 .seq_show = rdtgroup_tasks_show, 1460 .fflags = RFTYPE_BASE, 1461 }, 1462 { 1463 .name = "schemata", 1464 .mode = 0644, 1465 .kf_ops = &rdtgroup_kf_single_ops, 1466 .write = rdtgroup_schemata_write, 1467 .seq_show = rdtgroup_schemata_show, 1468 .fflags = RF_CTRL_BASE, 1469 }, 1470 { 1471 .name = "mode", 1472 .mode = 0644, 1473 .kf_ops = &rdtgroup_kf_single_ops, 1474 .write = rdtgroup_mode_write, 1475 .seq_show = rdtgroup_mode_show, 1476 .fflags = RF_CTRL_BASE, 1477 }, 1478 { 1479 .name = "size", 1480 .mode = 0444, 1481 .kf_ops = &rdtgroup_kf_single_ops, 1482 .seq_show = rdtgroup_size_show, 1483 .fflags = RF_CTRL_BASE, 1484 }, 1485 1486 }; 1487 1488 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags) 1489 { 1490 struct rftype *rfts, *rft; 1491 int ret, len; 1492 1493 rfts = res_common_files; 1494 len = ARRAY_SIZE(res_common_files); 1495 1496 lockdep_assert_held(&rdtgroup_mutex); 1497 1498 for (rft = rfts; rft < rfts + len; rft++) { 1499 if ((fflags & rft->fflags) == rft->fflags) { 1500 ret = rdtgroup_add_file(kn, rft); 1501 if (ret) 1502 goto error; 1503 } 1504 } 1505 1506 return 0; 1507 error: 1508 pr_warn("Failed to add %s, err=%d\n", rft->name, ret); 1509 while (--rft >= rfts) { 1510 if ((fflags & rft->fflags) == rft->fflags) 1511 kernfs_remove_by_name(kn, rft->name); 1512 } 1513 return ret; 1514 } 1515 1516 /** 1517 * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file 1518 * @r: The resource group with which the file is associated. 1519 * @name: Name of the file 1520 * 1521 * The permissions of named resctrl file, directory, or link are modified 1522 * to not allow read, write, or execute by any user. 1523 * 1524 * WARNING: This function is intended to communicate to the user that the 1525 * resctrl file has been locked down - that it is not relevant to the 1526 * particular state the system finds itself in. It should not be relied 1527 * on to protect from user access because after the file's permissions 1528 * are restricted the user can still change the permissions using chmod 1529 * from the command line. 1530 * 1531 * Return: 0 on success, <0 on failure. 1532 */ 1533 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name) 1534 { 1535 struct iattr iattr = {.ia_valid = ATTR_MODE,}; 1536 struct kernfs_node *kn; 1537 int ret = 0; 1538 1539 kn = kernfs_find_and_get_ns(r->kn, name, NULL); 1540 if (!kn) 1541 return -ENOENT; 1542 1543 switch (kernfs_type(kn)) { 1544 case KERNFS_DIR: 1545 iattr.ia_mode = S_IFDIR; 1546 break; 1547 case KERNFS_FILE: 1548 iattr.ia_mode = S_IFREG; 1549 break; 1550 case KERNFS_LINK: 1551 iattr.ia_mode = S_IFLNK; 1552 break; 1553 } 1554 1555 ret = kernfs_setattr(kn, &iattr); 1556 kernfs_put(kn); 1557 return ret; 1558 } 1559 1560 /** 1561 * rdtgroup_kn_mode_restore - Restore user access to named resctrl file 1562 * @r: The resource group with which the file is associated. 1563 * @name: Name of the file 1564 * @mask: Mask of permissions that should be restored 1565 * 1566 * Restore the permissions of the named file. If @name is a directory the 1567 * permissions of its parent will be used. 1568 * 1569 * Return: 0 on success, <0 on failure. 1570 */ 1571 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name, 1572 umode_t mask) 1573 { 1574 struct iattr iattr = {.ia_valid = ATTR_MODE,}; 1575 struct kernfs_node *kn, *parent; 1576 struct rftype *rfts, *rft; 1577 int ret, len; 1578 1579 rfts = res_common_files; 1580 len = ARRAY_SIZE(res_common_files); 1581 1582 for (rft = rfts; rft < rfts + len; rft++) { 1583 if (!strcmp(rft->name, name)) 1584 iattr.ia_mode = rft->mode & mask; 1585 } 1586 1587 kn = kernfs_find_and_get_ns(r->kn, name, NULL); 1588 if (!kn) 1589 return -ENOENT; 1590 1591 switch (kernfs_type(kn)) { 1592 case KERNFS_DIR: 1593 parent = kernfs_get_parent(kn); 1594 if (parent) { 1595 iattr.ia_mode |= parent->mode; 1596 kernfs_put(parent); 1597 } 1598 iattr.ia_mode |= S_IFDIR; 1599 break; 1600 case KERNFS_FILE: 1601 iattr.ia_mode |= S_IFREG; 1602 break; 1603 case KERNFS_LINK: 1604 iattr.ia_mode |= S_IFLNK; 1605 break; 1606 } 1607 1608 ret = kernfs_setattr(kn, &iattr); 1609 kernfs_put(kn); 1610 return ret; 1611 } 1612 1613 static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name, 1614 unsigned long fflags) 1615 { 1616 struct kernfs_node *kn_subdir; 1617 int ret; 1618 1619 kn_subdir = kernfs_create_dir(kn_info, name, 1620 kn_info->mode, r); 1621 if (IS_ERR(kn_subdir)) 1622 return PTR_ERR(kn_subdir); 1623 1624 kernfs_get(kn_subdir); 1625 ret = rdtgroup_kn_set_ugid(kn_subdir); 1626 if (ret) 1627 return ret; 1628 1629 ret = rdtgroup_add_files(kn_subdir, fflags); 1630 if (!ret) 1631 kernfs_activate(kn_subdir); 1632 1633 return ret; 1634 } 1635 1636 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn) 1637 { 1638 struct rdt_resource *r; 1639 unsigned long fflags; 1640 char name[32]; 1641 int ret; 1642 1643 /* create the directory */ 1644 kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL); 1645 if (IS_ERR(kn_info)) 1646 return PTR_ERR(kn_info); 1647 kernfs_get(kn_info); 1648 1649 ret = rdtgroup_add_files(kn_info, RF_TOP_INFO); 1650 if (ret) 1651 goto out_destroy; 1652 1653 for_each_alloc_enabled_rdt_resource(r) { 1654 fflags = r->fflags | RF_CTRL_INFO; 1655 ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags); 1656 if (ret) 1657 goto out_destroy; 1658 } 1659 1660 for_each_mon_enabled_rdt_resource(r) { 1661 fflags = r->fflags | RF_MON_INFO; 1662 sprintf(name, "%s_MON", r->name); 1663 ret = rdtgroup_mkdir_info_resdir(r, name, fflags); 1664 if (ret) 1665 goto out_destroy; 1666 } 1667 1668 /* 1669 * This extra ref will be put in kernfs_remove() and guarantees 1670 * that @rdtgrp->kn is always accessible. 1671 */ 1672 kernfs_get(kn_info); 1673 1674 ret = rdtgroup_kn_set_ugid(kn_info); 1675 if (ret) 1676 goto out_destroy; 1677 1678 kernfs_activate(kn_info); 1679 1680 return 0; 1681 1682 out_destroy: 1683 kernfs_remove(kn_info); 1684 return ret; 1685 } 1686 1687 static int 1688 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp, 1689 char *name, struct kernfs_node **dest_kn) 1690 { 1691 struct kernfs_node *kn; 1692 int ret; 1693 1694 /* create the directory */ 1695 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); 1696 if (IS_ERR(kn)) 1697 return PTR_ERR(kn); 1698 1699 if (dest_kn) 1700 *dest_kn = kn; 1701 1702 /* 1703 * This extra ref will be put in kernfs_remove() and guarantees 1704 * that @rdtgrp->kn is always accessible. 1705 */ 1706 kernfs_get(kn); 1707 1708 ret = rdtgroup_kn_set_ugid(kn); 1709 if (ret) 1710 goto out_destroy; 1711 1712 kernfs_activate(kn); 1713 1714 return 0; 1715 1716 out_destroy: 1717 kernfs_remove(kn); 1718 return ret; 1719 } 1720 1721 static void l3_qos_cfg_update(void *arg) 1722 { 1723 bool *enable = arg; 1724 1725 wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL); 1726 } 1727 1728 static void l2_qos_cfg_update(void *arg) 1729 { 1730 bool *enable = arg; 1731 1732 wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL); 1733 } 1734 1735 static inline bool is_mba_linear(void) 1736 { 1737 return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear; 1738 } 1739 1740 static int set_cache_qos_cfg(int level, bool enable) 1741 { 1742 void (*update)(void *arg); 1743 struct rdt_resource *r_l; 1744 cpumask_var_t cpu_mask; 1745 struct rdt_domain *d; 1746 int cpu; 1747 1748 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) 1749 return -ENOMEM; 1750 1751 if (level == RDT_RESOURCE_L3) 1752 update = l3_qos_cfg_update; 1753 else if (level == RDT_RESOURCE_L2) 1754 update = l2_qos_cfg_update; 1755 else 1756 return -EINVAL; 1757 1758 r_l = &rdt_resources_all[level]; 1759 list_for_each_entry(d, &r_l->domains, list) { 1760 /* Pick one CPU from each domain instance to update MSR */ 1761 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask); 1762 } 1763 cpu = get_cpu(); 1764 /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */ 1765 if (cpumask_test_cpu(cpu, cpu_mask)) 1766 update(&enable); 1767 /* Update QOS_CFG MSR on all other cpus in cpu_mask. */ 1768 smp_call_function_many(cpu_mask, update, &enable, 1); 1769 put_cpu(); 1770 1771 free_cpumask_var(cpu_mask); 1772 1773 return 0; 1774 } 1775 1776 /* 1777 * Enable or disable the MBA software controller 1778 * which helps user specify bandwidth in MBps. 1779 * MBA software controller is supported only if 1780 * MBM is supported and MBA is in linear scale. 1781 */ 1782 static int set_mba_sc(bool mba_sc) 1783 { 1784 struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA]; 1785 struct rdt_domain *d; 1786 1787 if (!is_mbm_enabled() || !is_mba_linear() || 1788 mba_sc == is_mba_sc(r)) 1789 return -EINVAL; 1790 1791 r->membw.mba_sc = mba_sc; 1792 list_for_each_entry(d, &r->domains, list) 1793 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val); 1794 1795 return 0; 1796 } 1797 1798 static int cdp_enable(int level, int data_type, int code_type) 1799 { 1800 struct rdt_resource *r_ldata = &rdt_resources_all[data_type]; 1801 struct rdt_resource *r_lcode = &rdt_resources_all[code_type]; 1802 struct rdt_resource *r_l = &rdt_resources_all[level]; 1803 int ret; 1804 1805 if (!r_l->alloc_capable || !r_ldata->alloc_capable || 1806 !r_lcode->alloc_capable) 1807 return -EINVAL; 1808 1809 ret = set_cache_qos_cfg(level, true); 1810 if (!ret) { 1811 r_l->alloc_enabled = false; 1812 r_ldata->alloc_enabled = true; 1813 r_lcode->alloc_enabled = true; 1814 } 1815 return ret; 1816 } 1817 1818 static int cdpl3_enable(void) 1819 { 1820 return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, 1821 RDT_RESOURCE_L3CODE); 1822 } 1823 1824 static int cdpl2_enable(void) 1825 { 1826 return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, 1827 RDT_RESOURCE_L2CODE); 1828 } 1829 1830 static void cdp_disable(int level, int data_type, int code_type) 1831 { 1832 struct rdt_resource *r = &rdt_resources_all[level]; 1833 1834 r->alloc_enabled = r->alloc_capable; 1835 1836 if (rdt_resources_all[data_type].alloc_enabled) { 1837 rdt_resources_all[data_type].alloc_enabled = false; 1838 rdt_resources_all[code_type].alloc_enabled = false; 1839 set_cache_qos_cfg(level, false); 1840 } 1841 } 1842 1843 static void cdpl3_disable(void) 1844 { 1845 cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE); 1846 } 1847 1848 static void cdpl2_disable(void) 1849 { 1850 cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE); 1851 } 1852 1853 static void cdp_disable_all(void) 1854 { 1855 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled) 1856 cdpl3_disable(); 1857 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled) 1858 cdpl2_disable(); 1859 } 1860 1861 static int parse_rdtgroupfs_options(char *data) 1862 { 1863 char *token, *o = data; 1864 int ret = 0; 1865 1866 while ((token = strsep(&o, ",")) != NULL) { 1867 if (!*token) { 1868 ret = -EINVAL; 1869 goto out; 1870 } 1871 1872 if (!strcmp(token, "cdp")) { 1873 ret = cdpl3_enable(); 1874 if (ret) 1875 goto out; 1876 } else if (!strcmp(token, "cdpl2")) { 1877 ret = cdpl2_enable(); 1878 if (ret) 1879 goto out; 1880 } else if (!strcmp(token, "mba_MBps")) { 1881 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) 1882 ret = set_mba_sc(true); 1883 else 1884 ret = -EINVAL; 1885 if (ret) 1886 goto out; 1887 } else { 1888 ret = -EINVAL; 1889 goto out; 1890 } 1891 } 1892 1893 return 0; 1894 1895 out: 1896 pr_err("Invalid mount option \"%s\"\n", token); 1897 1898 return ret; 1899 } 1900 1901 /* 1902 * We don't allow rdtgroup directories to be created anywhere 1903 * except the root directory. Thus when looking for the rdtgroup 1904 * structure for a kernfs node we are either looking at a directory, 1905 * in which case the rdtgroup structure is pointed at by the "priv" 1906 * field, otherwise we have a file, and need only look to the parent 1907 * to find the rdtgroup. 1908 */ 1909 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn) 1910 { 1911 if (kernfs_type(kn) == KERNFS_DIR) { 1912 /* 1913 * All the resource directories use "kn->priv" 1914 * to point to the "struct rdtgroup" for the 1915 * resource. "info" and its subdirectories don't 1916 * have rdtgroup structures, so return NULL here. 1917 */ 1918 if (kn == kn_info || kn->parent == kn_info) 1919 return NULL; 1920 else 1921 return kn->priv; 1922 } else { 1923 return kn->parent->priv; 1924 } 1925 } 1926 1927 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn) 1928 { 1929 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); 1930 1931 if (!rdtgrp) 1932 return NULL; 1933 1934 atomic_inc(&rdtgrp->waitcount); 1935 kernfs_break_active_protection(kn); 1936 1937 mutex_lock(&rdtgroup_mutex); 1938 1939 /* Was this group deleted while we waited? */ 1940 if (rdtgrp->flags & RDT_DELETED) 1941 return NULL; 1942 1943 return rdtgrp; 1944 } 1945 1946 void rdtgroup_kn_unlock(struct kernfs_node *kn) 1947 { 1948 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); 1949 1950 if (!rdtgrp) 1951 return; 1952 1953 mutex_unlock(&rdtgroup_mutex); 1954 1955 if (atomic_dec_and_test(&rdtgrp->waitcount) && 1956 (rdtgrp->flags & RDT_DELETED)) { 1957 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || 1958 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) 1959 rdtgroup_pseudo_lock_remove(rdtgrp); 1960 kernfs_unbreak_active_protection(kn); 1961 kernfs_put(rdtgrp->kn); 1962 kfree(rdtgrp); 1963 } else { 1964 kernfs_unbreak_active_protection(kn); 1965 } 1966 } 1967 1968 static int mkdir_mondata_all(struct kernfs_node *parent_kn, 1969 struct rdtgroup *prgrp, 1970 struct kernfs_node **mon_data_kn); 1971 1972 static struct dentry *rdt_mount(struct file_system_type *fs_type, 1973 int flags, const char *unused_dev_name, 1974 void *data) 1975 { 1976 struct rdt_domain *dom; 1977 struct rdt_resource *r; 1978 struct dentry *dentry; 1979 int ret; 1980 1981 cpus_read_lock(); 1982 mutex_lock(&rdtgroup_mutex); 1983 /* 1984 * resctrl file system can only be mounted once. 1985 */ 1986 if (static_branch_unlikely(&rdt_enable_key)) { 1987 dentry = ERR_PTR(-EBUSY); 1988 goto out; 1989 } 1990 1991 ret = parse_rdtgroupfs_options(data); 1992 if (ret) { 1993 dentry = ERR_PTR(ret); 1994 goto out_cdp; 1995 } 1996 1997 closid_init(); 1998 1999 ret = rdtgroup_create_info_dir(rdtgroup_default.kn); 2000 if (ret) { 2001 dentry = ERR_PTR(ret); 2002 goto out_cdp; 2003 } 2004 2005 if (rdt_mon_capable) { 2006 ret = mongroup_create_dir(rdtgroup_default.kn, 2007 NULL, "mon_groups", 2008 &kn_mongrp); 2009 if (ret) { 2010 dentry = ERR_PTR(ret); 2011 goto out_info; 2012 } 2013 kernfs_get(kn_mongrp); 2014 2015 ret = mkdir_mondata_all(rdtgroup_default.kn, 2016 &rdtgroup_default, &kn_mondata); 2017 if (ret) { 2018 dentry = ERR_PTR(ret); 2019 goto out_mongrp; 2020 } 2021 kernfs_get(kn_mondata); 2022 rdtgroup_default.mon.mon_data_kn = kn_mondata; 2023 } 2024 2025 ret = rdt_pseudo_lock_init(); 2026 if (ret) { 2027 dentry = ERR_PTR(ret); 2028 goto out_mondata; 2029 } 2030 2031 dentry = kernfs_mount(fs_type, flags, rdt_root, 2032 RDTGROUP_SUPER_MAGIC, NULL); 2033 if (IS_ERR(dentry)) 2034 goto out_psl; 2035 2036 if (rdt_alloc_capable) 2037 static_branch_enable_cpuslocked(&rdt_alloc_enable_key); 2038 if (rdt_mon_capable) 2039 static_branch_enable_cpuslocked(&rdt_mon_enable_key); 2040 2041 if (rdt_alloc_capable || rdt_mon_capable) 2042 static_branch_enable_cpuslocked(&rdt_enable_key); 2043 2044 if (is_mbm_enabled()) { 2045 r = &rdt_resources_all[RDT_RESOURCE_L3]; 2046 list_for_each_entry(dom, &r->domains, list) 2047 mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL); 2048 } 2049 2050 goto out; 2051 2052 out_psl: 2053 rdt_pseudo_lock_release(); 2054 out_mondata: 2055 if (rdt_mon_capable) 2056 kernfs_remove(kn_mondata); 2057 out_mongrp: 2058 if (rdt_mon_capable) 2059 kernfs_remove(kn_mongrp); 2060 out_info: 2061 kernfs_remove(kn_info); 2062 out_cdp: 2063 cdp_disable_all(); 2064 out: 2065 rdt_last_cmd_clear(); 2066 mutex_unlock(&rdtgroup_mutex); 2067 cpus_read_unlock(); 2068 2069 return dentry; 2070 } 2071 2072 static int reset_all_ctrls(struct rdt_resource *r) 2073 { 2074 struct msr_param msr_param; 2075 cpumask_var_t cpu_mask; 2076 struct rdt_domain *d; 2077 int i, cpu; 2078 2079 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) 2080 return -ENOMEM; 2081 2082 msr_param.res = r; 2083 msr_param.low = 0; 2084 msr_param.high = r->num_closid; 2085 2086 /* 2087 * Disable resource control for this resource by setting all 2088 * CBMs in all domains to the maximum mask value. Pick one CPU 2089 * from each domain to update the MSRs below. 2090 */ 2091 list_for_each_entry(d, &r->domains, list) { 2092 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask); 2093 2094 for (i = 0; i < r->num_closid; i++) 2095 d->ctrl_val[i] = r->default_ctrl; 2096 } 2097 cpu = get_cpu(); 2098 /* Update CBM on this cpu if it's in cpu_mask. */ 2099 if (cpumask_test_cpu(cpu, cpu_mask)) 2100 rdt_ctrl_update(&msr_param); 2101 /* Update CBM on all other cpus in cpu_mask. */ 2102 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1); 2103 put_cpu(); 2104 2105 free_cpumask_var(cpu_mask); 2106 2107 return 0; 2108 } 2109 2110 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r) 2111 { 2112 return (rdt_alloc_capable && 2113 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid)); 2114 } 2115 2116 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r) 2117 { 2118 return (rdt_mon_capable && 2119 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid)); 2120 } 2121 2122 /* 2123 * Move tasks from one to the other group. If @from is NULL, then all tasks 2124 * in the systems are moved unconditionally (used for teardown). 2125 * 2126 * If @mask is not NULL the cpus on which moved tasks are running are set 2127 * in that mask so the update smp function call is restricted to affected 2128 * cpus. 2129 */ 2130 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to, 2131 struct cpumask *mask) 2132 { 2133 struct task_struct *p, *t; 2134 2135 read_lock(&tasklist_lock); 2136 for_each_process_thread(p, t) { 2137 if (!from || is_closid_match(t, from) || 2138 is_rmid_match(t, from)) { 2139 t->closid = to->closid; 2140 t->rmid = to->mon.rmid; 2141 2142 #ifdef CONFIG_SMP 2143 /* 2144 * This is safe on x86 w/o barriers as the ordering 2145 * of writing to task_cpu() and t->on_cpu is 2146 * reverse to the reading here. The detection is 2147 * inaccurate as tasks might move or schedule 2148 * before the smp function call takes place. In 2149 * such a case the function call is pointless, but 2150 * there is no other side effect. 2151 */ 2152 if (mask && t->on_cpu) 2153 cpumask_set_cpu(task_cpu(t), mask); 2154 #endif 2155 } 2156 } 2157 read_unlock(&tasklist_lock); 2158 } 2159 2160 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp) 2161 { 2162 struct rdtgroup *sentry, *stmp; 2163 struct list_head *head; 2164 2165 head = &rdtgrp->mon.crdtgrp_list; 2166 list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) { 2167 free_rmid(sentry->mon.rmid); 2168 list_del(&sentry->mon.crdtgrp_list); 2169 kfree(sentry); 2170 } 2171 } 2172 2173 /* 2174 * Forcibly remove all of subdirectories under root. 2175 */ 2176 static void rmdir_all_sub(void) 2177 { 2178 struct rdtgroup *rdtgrp, *tmp; 2179 2180 /* Move all tasks to the default resource group */ 2181 rdt_move_group_tasks(NULL, &rdtgroup_default, NULL); 2182 2183 list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) { 2184 /* Free any child rmids */ 2185 free_all_child_rdtgrp(rdtgrp); 2186 2187 /* Remove each rdtgroup other than root */ 2188 if (rdtgrp == &rdtgroup_default) 2189 continue; 2190 2191 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || 2192 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) 2193 rdtgroup_pseudo_lock_remove(rdtgrp); 2194 2195 /* 2196 * Give any CPUs back to the default group. We cannot copy 2197 * cpu_online_mask because a CPU might have executed the 2198 * offline callback already, but is still marked online. 2199 */ 2200 cpumask_or(&rdtgroup_default.cpu_mask, 2201 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); 2202 2203 free_rmid(rdtgrp->mon.rmid); 2204 2205 kernfs_remove(rdtgrp->kn); 2206 list_del(&rdtgrp->rdtgroup_list); 2207 kfree(rdtgrp); 2208 } 2209 /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */ 2210 update_closid_rmid(cpu_online_mask, &rdtgroup_default); 2211 2212 kernfs_remove(kn_info); 2213 kernfs_remove(kn_mongrp); 2214 kernfs_remove(kn_mondata); 2215 } 2216 2217 static void rdt_kill_sb(struct super_block *sb) 2218 { 2219 struct rdt_resource *r; 2220 2221 cpus_read_lock(); 2222 mutex_lock(&rdtgroup_mutex); 2223 2224 set_mba_sc(false); 2225 2226 /*Put everything back to default values. */ 2227 for_each_alloc_enabled_rdt_resource(r) 2228 reset_all_ctrls(r); 2229 cdp_disable_all(); 2230 rmdir_all_sub(); 2231 rdt_pseudo_lock_release(); 2232 rdtgroup_default.mode = RDT_MODE_SHAREABLE; 2233 static_branch_disable_cpuslocked(&rdt_alloc_enable_key); 2234 static_branch_disable_cpuslocked(&rdt_mon_enable_key); 2235 static_branch_disable_cpuslocked(&rdt_enable_key); 2236 kernfs_kill_sb(sb); 2237 mutex_unlock(&rdtgroup_mutex); 2238 cpus_read_unlock(); 2239 } 2240 2241 static struct file_system_type rdt_fs_type = { 2242 .name = "resctrl", 2243 .mount = rdt_mount, 2244 .kill_sb = rdt_kill_sb, 2245 }; 2246 2247 static int mon_addfile(struct kernfs_node *parent_kn, const char *name, 2248 void *priv) 2249 { 2250 struct kernfs_node *kn; 2251 int ret = 0; 2252 2253 kn = __kernfs_create_file(parent_kn, name, 0444, 2254 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0, 2255 &kf_mondata_ops, priv, NULL, NULL); 2256 if (IS_ERR(kn)) 2257 return PTR_ERR(kn); 2258 2259 ret = rdtgroup_kn_set_ugid(kn); 2260 if (ret) { 2261 kernfs_remove(kn); 2262 return ret; 2263 } 2264 2265 return ret; 2266 } 2267 2268 /* 2269 * Remove all subdirectories of mon_data of ctrl_mon groups 2270 * and monitor groups with given domain id. 2271 */ 2272 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id) 2273 { 2274 struct rdtgroup *prgrp, *crgrp; 2275 char name[32]; 2276 2277 if (!r->mon_enabled) 2278 return; 2279 2280 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { 2281 sprintf(name, "mon_%s_%02d", r->name, dom_id); 2282 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name); 2283 2284 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list) 2285 kernfs_remove_by_name(crgrp->mon.mon_data_kn, name); 2286 } 2287 } 2288 2289 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn, 2290 struct rdt_domain *d, 2291 struct rdt_resource *r, struct rdtgroup *prgrp) 2292 { 2293 union mon_data_bits priv; 2294 struct kernfs_node *kn; 2295 struct mon_evt *mevt; 2296 struct rmid_read rr; 2297 char name[32]; 2298 int ret; 2299 2300 sprintf(name, "mon_%s_%02d", r->name, d->id); 2301 /* create the directory */ 2302 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); 2303 if (IS_ERR(kn)) 2304 return PTR_ERR(kn); 2305 2306 /* 2307 * This extra ref will be put in kernfs_remove() and guarantees 2308 * that kn is always accessible. 2309 */ 2310 kernfs_get(kn); 2311 ret = rdtgroup_kn_set_ugid(kn); 2312 if (ret) 2313 goto out_destroy; 2314 2315 if (WARN_ON(list_empty(&r->evt_list))) { 2316 ret = -EPERM; 2317 goto out_destroy; 2318 } 2319 2320 priv.u.rid = r->rid; 2321 priv.u.domid = d->id; 2322 list_for_each_entry(mevt, &r->evt_list, list) { 2323 priv.u.evtid = mevt->evtid; 2324 ret = mon_addfile(kn, mevt->name, priv.priv); 2325 if (ret) 2326 goto out_destroy; 2327 2328 if (is_mbm_event(mevt->evtid)) 2329 mon_event_read(&rr, d, prgrp, mevt->evtid, true); 2330 } 2331 kernfs_activate(kn); 2332 return 0; 2333 2334 out_destroy: 2335 kernfs_remove(kn); 2336 return ret; 2337 } 2338 2339 /* 2340 * Add all subdirectories of mon_data for "ctrl_mon" groups 2341 * and "monitor" groups with given domain id. 2342 */ 2343 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, 2344 struct rdt_domain *d) 2345 { 2346 struct kernfs_node *parent_kn; 2347 struct rdtgroup *prgrp, *crgrp; 2348 struct list_head *head; 2349 2350 if (!r->mon_enabled) 2351 return; 2352 2353 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { 2354 parent_kn = prgrp->mon.mon_data_kn; 2355 mkdir_mondata_subdir(parent_kn, d, r, prgrp); 2356 2357 head = &prgrp->mon.crdtgrp_list; 2358 list_for_each_entry(crgrp, head, mon.crdtgrp_list) { 2359 parent_kn = crgrp->mon.mon_data_kn; 2360 mkdir_mondata_subdir(parent_kn, d, r, crgrp); 2361 } 2362 } 2363 } 2364 2365 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn, 2366 struct rdt_resource *r, 2367 struct rdtgroup *prgrp) 2368 { 2369 struct rdt_domain *dom; 2370 int ret; 2371 2372 list_for_each_entry(dom, &r->domains, list) { 2373 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp); 2374 if (ret) 2375 return ret; 2376 } 2377 2378 return 0; 2379 } 2380 2381 /* 2382 * This creates a directory mon_data which contains the monitored data. 2383 * 2384 * mon_data has one directory for each domain whic are named 2385 * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data 2386 * with L3 domain looks as below: 2387 * ./mon_data: 2388 * mon_L3_00 2389 * mon_L3_01 2390 * mon_L3_02 2391 * ... 2392 * 2393 * Each domain directory has one file per event: 2394 * ./mon_L3_00/: 2395 * llc_occupancy 2396 * 2397 */ 2398 static int mkdir_mondata_all(struct kernfs_node *parent_kn, 2399 struct rdtgroup *prgrp, 2400 struct kernfs_node **dest_kn) 2401 { 2402 struct rdt_resource *r; 2403 struct kernfs_node *kn; 2404 int ret; 2405 2406 /* 2407 * Create the mon_data directory first. 2408 */ 2409 ret = mongroup_create_dir(parent_kn, NULL, "mon_data", &kn); 2410 if (ret) 2411 return ret; 2412 2413 if (dest_kn) 2414 *dest_kn = kn; 2415 2416 /* 2417 * Create the subdirectories for each domain. Note that all events 2418 * in a domain like L3 are grouped into a resource whose domain is L3 2419 */ 2420 for_each_mon_enabled_rdt_resource(r) { 2421 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp); 2422 if (ret) 2423 goto out_destroy; 2424 } 2425 2426 return 0; 2427 2428 out_destroy: 2429 kernfs_remove(kn); 2430 return ret; 2431 } 2432 2433 /** 2434 * cbm_ensure_valid - Enforce validity on provided CBM 2435 * @_val: Candidate CBM 2436 * @r: RDT resource to which the CBM belongs 2437 * 2438 * The provided CBM represents all cache portions available for use. This 2439 * may be represented by a bitmap that does not consist of contiguous ones 2440 * and thus be an invalid CBM. 2441 * Here the provided CBM is forced to be a valid CBM by only considering 2442 * the first set of contiguous bits as valid and clearing all bits. 2443 * The intention here is to provide a valid default CBM with which a new 2444 * resource group is initialized. The user can follow this with a 2445 * modification to the CBM if the default does not satisfy the 2446 * requirements. 2447 */ 2448 static void cbm_ensure_valid(u32 *_val, struct rdt_resource *r) 2449 { 2450 /* 2451 * Convert the u32 _val to an unsigned long required by all the bit 2452 * operations within this function. No more than 32 bits of this 2453 * converted value can be accessed because all bit operations are 2454 * additionally provided with cbm_len that is initialized during 2455 * hardware enumeration using five bits from the EAX register and 2456 * thus never can exceed 32 bits. 2457 */ 2458 unsigned long *val = (unsigned long *)_val; 2459 unsigned int cbm_len = r->cache.cbm_len; 2460 unsigned long first_bit, zero_bit; 2461 2462 if (*val == 0) 2463 return; 2464 2465 first_bit = find_first_bit(val, cbm_len); 2466 zero_bit = find_next_zero_bit(val, cbm_len, first_bit); 2467 2468 /* Clear any remaining bits to ensure contiguous region */ 2469 bitmap_clear(val, zero_bit, cbm_len - zero_bit); 2470 } 2471 2472 /** 2473 * rdtgroup_init_alloc - Initialize the new RDT group's allocations 2474 * 2475 * A new RDT group is being created on an allocation capable (CAT) 2476 * supporting system. Set this group up to start off with all usable 2477 * allocations. That is, all shareable and unused bits. 2478 * 2479 * All-zero CBM is invalid. If there are no more shareable bits available 2480 * on any domain then the entire allocation will fail. 2481 */ 2482 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp) 2483 { 2484 struct rdt_resource *r_cdp = NULL; 2485 struct rdt_domain *d_cdp = NULL; 2486 u32 used_b = 0, unused_b = 0; 2487 u32 closid = rdtgrp->closid; 2488 struct rdt_resource *r; 2489 unsigned long tmp_cbm; 2490 enum rdtgrp_mode mode; 2491 struct rdt_domain *d; 2492 u32 peer_ctl, *ctrl; 2493 int i, ret; 2494 2495 for_each_alloc_enabled_rdt_resource(r) { 2496 /* 2497 * Only initialize default allocations for CBM cache 2498 * resources 2499 */ 2500 if (r->rid == RDT_RESOURCE_MBA) 2501 continue; 2502 list_for_each_entry(d, &r->domains, list) { 2503 rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp); 2504 d->have_new_ctrl = false; 2505 d->new_ctrl = r->cache.shareable_bits; 2506 used_b = r->cache.shareable_bits; 2507 ctrl = d->ctrl_val; 2508 for (i = 0; i < closids_supported(); i++, ctrl++) { 2509 if (closid_allocated(i) && i != closid) { 2510 mode = rdtgroup_mode_by_closid(i); 2511 if (mode == RDT_MODE_PSEUDO_LOCKSETUP) 2512 break; 2513 /* 2514 * If CDP is active include peer 2515 * domain's usage to ensure there 2516 * is no overlap with an exclusive 2517 * group. 2518 */ 2519 if (d_cdp) 2520 peer_ctl = d_cdp->ctrl_val[i]; 2521 else 2522 peer_ctl = 0; 2523 used_b |= *ctrl | peer_ctl; 2524 if (mode == RDT_MODE_SHAREABLE) 2525 d->new_ctrl |= *ctrl | peer_ctl; 2526 } 2527 } 2528 if (d->plr && d->plr->cbm > 0) 2529 used_b |= d->plr->cbm; 2530 unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1); 2531 unused_b &= BIT_MASK(r->cache.cbm_len) - 1; 2532 d->new_ctrl |= unused_b; 2533 /* 2534 * Force the initial CBM to be valid, user can 2535 * modify the CBM based on system availability. 2536 */ 2537 cbm_ensure_valid(&d->new_ctrl, r); 2538 /* 2539 * Assign the u32 CBM to an unsigned long to ensure 2540 * that bitmap_weight() does not access out-of-bound 2541 * memory. 2542 */ 2543 tmp_cbm = d->new_ctrl; 2544 if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < 2545 r->cache.min_cbm_bits) { 2546 rdt_last_cmd_printf("No space on %s:%d\n", 2547 r->name, d->id); 2548 return -ENOSPC; 2549 } 2550 d->have_new_ctrl = true; 2551 } 2552 } 2553 2554 for_each_alloc_enabled_rdt_resource(r) { 2555 /* 2556 * Only initialize default allocations for CBM cache 2557 * resources 2558 */ 2559 if (r->rid == RDT_RESOURCE_MBA) 2560 continue; 2561 ret = update_domains(r, rdtgrp->closid); 2562 if (ret < 0) { 2563 rdt_last_cmd_puts("Failed to initialize allocations\n"); 2564 return ret; 2565 } 2566 rdtgrp->mode = RDT_MODE_SHAREABLE; 2567 } 2568 2569 return 0; 2570 } 2571 2572 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn, 2573 struct kernfs_node *prgrp_kn, 2574 const char *name, umode_t mode, 2575 enum rdt_group_type rtype, struct rdtgroup **r) 2576 { 2577 struct rdtgroup *prdtgrp, *rdtgrp; 2578 struct kernfs_node *kn; 2579 uint files = 0; 2580 int ret; 2581 2582 prdtgrp = rdtgroup_kn_lock_live(prgrp_kn); 2583 rdt_last_cmd_clear(); 2584 if (!prdtgrp) { 2585 ret = -ENODEV; 2586 rdt_last_cmd_puts("Directory was removed\n"); 2587 goto out_unlock; 2588 } 2589 2590 if (rtype == RDTMON_GROUP && 2591 (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || 2592 prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) { 2593 ret = -EINVAL; 2594 rdt_last_cmd_puts("Pseudo-locking in progress\n"); 2595 goto out_unlock; 2596 } 2597 2598 /* allocate the rdtgroup. */ 2599 rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL); 2600 if (!rdtgrp) { 2601 ret = -ENOSPC; 2602 rdt_last_cmd_puts("Kernel out of memory\n"); 2603 goto out_unlock; 2604 } 2605 *r = rdtgrp; 2606 rdtgrp->mon.parent = prdtgrp; 2607 rdtgrp->type = rtype; 2608 INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list); 2609 2610 /* kernfs creates the directory for rdtgrp */ 2611 kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp); 2612 if (IS_ERR(kn)) { 2613 ret = PTR_ERR(kn); 2614 rdt_last_cmd_puts("kernfs create error\n"); 2615 goto out_free_rgrp; 2616 } 2617 rdtgrp->kn = kn; 2618 2619 /* 2620 * kernfs_remove() will drop the reference count on "kn" which 2621 * will free it. But we still need it to stick around for the 2622 * rdtgroup_kn_unlock(kn} call below. Take one extra reference 2623 * here, which will be dropped inside rdtgroup_kn_unlock(). 2624 */ 2625 kernfs_get(kn); 2626 2627 ret = rdtgroup_kn_set_ugid(kn); 2628 if (ret) { 2629 rdt_last_cmd_puts("kernfs perm error\n"); 2630 goto out_destroy; 2631 } 2632 2633 files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype); 2634 ret = rdtgroup_add_files(kn, files); 2635 if (ret) { 2636 rdt_last_cmd_puts("kernfs fill error\n"); 2637 goto out_destroy; 2638 } 2639 2640 if (rdt_mon_capable) { 2641 ret = alloc_rmid(); 2642 if (ret < 0) { 2643 rdt_last_cmd_puts("Out of RMIDs\n"); 2644 goto out_destroy; 2645 } 2646 rdtgrp->mon.rmid = ret; 2647 2648 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn); 2649 if (ret) { 2650 rdt_last_cmd_puts("kernfs subdir error\n"); 2651 goto out_idfree; 2652 } 2653 } 2654 kernfs_activate(kn); 2655 2656 /* 2657 * The caller unlocks the prgrp_kn upon success. 2658 */ 2659 return 0; 2660 2661 out_idfree: 2662 free_rmid(rdtgrp->mon.rmid); 2663 out_destroy: 2664 kernfs_remove(rdtgrp->kn); 2665 out_free_rgrp: 2666 kfree(rdtgrp); 2667 out_unlock: 2668 rdtgroup_kn_unlock(prgrp_kn); 2669 return ret; 2670 } 2671 2672 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp) 2673 { 2674 kernfs_remove(rgrp->kn); 2675 free_rmid(rgrp->mon.rmid); 2676 kfree(rgrp); 2677 } 2678 2679 /* 2680 * Create a monitor group under "mon_groups" directory of a control 2681 * and monitor group(ctrl_mon). This is a resource group 2682 * to monitor a subset of tasks and cpus in its parent ctrl_mon group. 2683 */ 2684 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn, 2685 struct kernfs_node *prgrp_kn, 2686 const char *name, 2687 umode_t mode) 2688 { 2689 struct rdtgroup *rdtgrp, *prgrp; 2690 int ret; 2691 2692 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP, 2693 &rdtgrp); 2694 if (ret) 2695 return ret; 2696 2697 prgrp = rdtgrp->mon.parent; 2698 rdtgrp->closid = prgrp->closid; 2699 2700 /* 2701 * Add the rdtgrp to the list of rdtgrps the parent 2702 * ctrl_mon group has to track. 2703 */ 2704 list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list); 2705 2706 rdtgroup_kn_unlock(prgrp_kn); 2707 return ret; 2708 } 2709 2710 /* 2711 * These are rdtgroups created under the root directory. Can be used 2712 * to allocate and monitor resources. 2713 */ 2714 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn, 2715 struct kernfs_node *prgrp_kn, 2716 const char *name, umode_t mode) 2717 { 2718 struct rdtgroup *rdtgrp; 2719 struct kernfs_node *kn; 2720 u32 closid; 2721 int ret; 2722 2723 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP, 2724 &rdtgrp); 2725 if (ret) 2726 return ret; 2727 2728 kn = rdtgrp->kn; 2729 ret = closid_alloc(); 2730 if (ret < 0) { 2731 rdt_last_cmd_puts("Out of CLOSIDs\n"); 2732 goto out_common_fail; 2733 } 2734 closid = ret; 2735 ret = 0; 2736 2737 rdtgrp->closid = closid; 2738 ret = rdtgroup_init_alloc(rdtgrp); 2739 if (ret < 0) 2740 goto out_id_free; 2741 2742 list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups); 2743 2744 if (rdt_mon_capable) { 2745 /* 2746 * Create an empty mon_groups directory to hold the subset 2747 * of tasks and cpus to monitor. 2748 */ 2749 ret = mongroup_create_dir(kn, NULL, "mon_groups", NULL); 2750 if (ret) { 2751 rdt_last_cmd_puts("kernfs subdir error\n"); 2752 goto out_del_list; 2753 } 2754 } 2755 2756 goto out_unlock; 2757 2758 out_del_list: 2759 list_del(&rdtgrp->rdtgroup_list); 2760 out_id_free: 2761 closid_free(closid); 2762 out_common_fail: 2763 mkdir_rdt_prepare_clean(rdtgrp); 2764 out_unlock: 2765 rdtgroup_kn_unlock(prgrp_kn); 2766 return ret; 2767 } 2768 2769 /* 2770 * We allow creating mon groups only with in a directory called "mon_groups" 2771 * which is present in every ctrl_mon group. Check if this is a valid 2772 * "mon_groups" directory. 2773 * 2774 * 1. The directory should be named "mon_groups". 2775 * 2. The mon group itself should "not" be named "mon_groups". 2776 * This makes sure "mon_groups" directory always has a ctrl_mon group 2777 * as parent. 2778 */ 2779 static bool is_mon_groups(struct kernfs_node *kn, const char *name) 2780 { 2781 return (!strcmp(kn->name, "mon_groups") && 2782 strcmp(name, "mon_groups")); 2783 } 2784 2785 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name, 2786 umode_t mode) 2787 { 2788 /* Do not accept '\n' to avoid unparsable situation. */ 2789 if (strchr(name, '\n')) 2790 return -EINVAL; 2791 2792 /* 2793 * If the parent directory is the root directory and RDT 2794 * allocation is supported, add a control and monitoring 2795 * subdirectory 2796 */ 2797 if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn) 2798 return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode); 2799 2800 /* 2801 * If RDT monitoring is supported and the parent directory is a valid 2802 * "mon_groups" directory, add a monitoring subdirectory. 2803 */ 2804 if (rdt_mon_capable && is_mon_groups(parent_kn, name)) 2805 return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode); 2806 2807 return -EPERM; 2808 } 2809 2810 static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp, 2811 cpumask_var_t tmpmask) 2812 { 2813 struct rdtgroup *prdtgrp = rdtgrp->mon.parent; 2814 int cpu; 2815 2816 /* Give any tasks back to the parent group */ 2817 rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask); 2818 2819 /* Update per cpu rmid of the moved CPUs first */ 2820 for_each_cpu(cpu, &rdtgrp->cpu_mask) 2821 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid; 2822 /* 2823 * Update the MSR on moved CPUs and CPUs which have moved 2824 * task running on them. 2825 */ 2826 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); 2827 update_closid_rmid(tmpmask, NULL); 2828 2829 rdtgrp->flags = RDT_DELETED; 2830 free_rmid(rdtgrp->mon.rmid); 2831 2832 /* 2833 * Remove the rdtgrp from the parent ctrl_mon group's list 2834 */ 2835 WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list)); 2836 list_del(&rdtgrp->mon.crdtgrp_list); 2837 2838 /* 2839 * one extra hold on this, will drop when we kfree(rdtgrp) 2840 * in rdtgroup_kn_unlock() 2841 */ 2842 kernfs_get(kn); 2843 kernfs_remove(rdtgrp->kn); 2844 2845 return 0; 2846 } 2847 2848 static int rdtgroup_ctrl_remove(struct kernfs_node *kn, 2849 struct rdtgroup *rdtgrp) 2850 { 2851 rdtgrp->flags = RDT_DELETED; 2852 list_del(&rdtgrp->rdtgroup_list); 2853 2854 /* 2855 * one extra hold on this, will drop when we kfree(rdtgrp) 2856 * in rdtgroup_kn_unlock() 2857 */ 2858 kernfs_get(kn); 2859 kernfs_remove(rdtgrp->kn); 2860 return 0; 2861 } 2862 2863 static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp, 2864 cpumask_var_t tmpmask) 2865 { 2866 int cpu; 2867 2868 /* Give any tasks back to the default group */ 2869 rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask); 2870 2871 /* Give any CPUs back to the default group */ 2872 cpumask_or(&rdtgroup_default.cpu_mask, 2873 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); 2874 2875 /* Update per cpu closid and rmid of the moved CPUs first */ 2876 for_each_cpu(cpu, &rdtgrp->cpu_mask) { 2877 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid; 2878 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid; 2879 } 2880 2881 /* 2882 * Update the MSR on moved CPUs and CPUs which have moved 2883 * task running on them. 2884 */ 2885 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); 2886 update_closid_rmid(tmpmask, NULL); 2887 2888 closid_free(rdtgrp->closid); 2889 free_rmid(rdtgrp->mon.rmid); 2890 2891 /* 2892 * Free all the child monitor group rmids. 2893 */ 2894 free_all_child_rdtgrp(rdtgrp); 2895 2896 rdtgroup_ctrl_remove(kn, rdtgrp); 2897 2898 return 0; 2899 } 2900 2901 static int rdtgroup_rmdir(struct kernfs_node *kn) 2902 { 2903 struct kernfs_node *parent_kn = kn->parent; 2904 struct rdtgroup *rdtgrp; 2905 cpumask_var_t tmpmask; 2906 int ret = 0; 2907 2908 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) 2909 return -ENOMEM; 2910 2911 rdtgrp = rdtgroup_kn_lock_live(kn); 2912 if (!rdtgrp) { 2913 ret = -EPERM; 2914 goto out; 2915 } 2916 2917 /* 2918 * If the rdtgroup is a ctrl_mon group and parent directory 2919 * is the root directory, remove the ctrl_mon group. 2920 * 2921 * If the rdtgroup is a mon group and parent directory 2922 * is a valid "mon_groups" directory, remove the mon group. 2923 */ 2924 if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn) { 2925 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || 2926 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { 2927 ret = rdtgroup_ctrl_remove(kn, rdtgrp); 2928 } else { 2929 ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask); 2930 } 2931 } else if (rdtgrp->type == RDTMON_GROUP && 2932 is_mon_groups(parent_kn, kn->name)) { 2933 ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask); 2934 } else { 2935 ret = -EPERM; 2936 } 2937 2938 out: 2939 rdtgroup_kn_unlock(kn); 2940 free_cpumask_var(tmpmask); 2941 return ret; 2942 } 2943 2944 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf) 2945 { 2946 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled) 2947 seq_puts(seq, ",cdp"); 2948 2949 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled) 2950 seq_puts(seq, ",cdpl2"); 2951 2952 if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA])) 2953 seq_puts(seq, ",mba_MBps"); 2954 2955 return 0; 2956 } 2957 2958 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = { 2959 .mkdir = rdtgroup_mkdir, 2960 .rmdir = rdtgroup_rmdir, 2961 .show_options = rdtgroup_show_options, 2962 }; 2963 2964 static int __init rdtgroup_setup_root(void) 2965 { 2966 int ret; 2967 2968 rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops, 2969 KERNFS_ROOT_CREATE_DEACTIVATED | 2970 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK, 2971 &rdtgroup_default); 2972 if (IS_ERR(rdt_root)) 2973 return PTR_ERR(rdt_root); 2974 2975 mutex_lock(&rdtgroup_mutex); 2976 2977 rdtgroup_default.closid = 0; 2978 rdtgroup_default.mon.rmid = 0; 2979 rdtgroup_default.type = RDTCTRL_GROUP; 2980 INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list); 2981 2982 list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups); 2983 2984 ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE); 2985 if (ret) { 2986 kernfs_destroy_root(rdt_root); 2987 goto out; 2988 } 2989 2990 rdtgroup_default.kn = rdt_root->kn; 2991 kernfs_activate(rdtgroup_default.kn); 2992 2993 out: 2994 mutex_unlock(&rdtgroup_mutex); 2995 2996 return ret; 2997 } 2998 2999 /* 3000 * rdtgroup_init - rdtgroup initialization 3001 * 3002 * Setup resctrl file system including set up root, create mount point, 3003 * register rdtgroup filesystem, and initialize files under root directory. 3004 * 3005 * Return: 0 on success or -errno 3006 */ 3007 int __init rdtgroup_init(void) 3008 { 3009 int ret = 0; 3010 3011 seq_buf_init(&last_cmd_status, last_cmd_status_buf, 3012 sizeof(last_cmd_status_buf)); 3013 3014 ret = rdtgroup_setup_root(); 3015 if (ret) 3016 return ret; 3017 3018 ret = sysfs_create_mount_point(fs_kobj, "resctrl"); 3019 if (ret) 3020 goto cleanup_root; 3021 3022 ret = register_filesystem(&rdt_fs_type); 3023 if (ret) 3024 goto cleanup_mountpoint; 3025 3026 /* 3027 * Adding the resctrl debugfs directory here may not be ideal since 3028 * it would let the resctrl debugfs directory appear on the debugfs 3029 * filesystem before the resctrl filesystem is mounted. 3030 * It may also be ok since that would enable debugging of RDT before 3031 * resctrl is mounted. 3032 * The reason why the debugfs directory is created here and not in 3033 * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and 3034 * during the debugfs directory creation also &sb->s_type->i_mutex_key 3035 * (the lockdep class of inode->i_rwsem). Other filesystem 3036 * interactions (eg. SyS_getdents) have the lock ordering: 3037 * &sb->s_type->i_mutex_key --> &mm->mmap_sem 3038 * During mmap(), called with &mm->mmap_sem, the rdtgroup_mutex 3039 * is taken, thus creating dependency: 3040 * &mm->mmap_sem --> rdtgroup_mutex for the latter that can cause 3041 * issues considering the other two lock dependencies. 3042 * By creating the debugfs directory here we avoid a dependency 3043 * that may cause deadlock (even though file operations cannot 3044 * occur until the filesystem is mounted, but I do not know how to 3045 * tell lockdep that). 3046 */ 3047 debugfs_resctrl = debugfs_create_dir("resctrl", NULL); 3048 3049 return 0; 3050 3051 cleanup_mountpoint: 3052 sysfs_remove_mount_point(fs_kobj, "resctrl"); 3053 cleanup_root: 3054 kernfs_destroy_root(rdt_root); 3055 3056 return ret; 3057 } 3058 3059 void __exit rdtgroup_exit(void) 3060 { 3061 debugfs_remove_recursive(debugfs_resctrl); 3062 unregister_filesystem(&rdt_fs_type); 3063 sysfs_remove_mount_point(fs_kobj, "resctrl"); 3064 kernfs_destroy_root(rdt_root); 3065 } 3066