root/arch/x86/kernel/cpu/resctrl/rdtgroup.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. rdt_last_cmd_clear
  2. rdt_last_cmd_puts
  3. rdt_last_cmd_printf
  4. closids_supported
  5. closid_init
  6. closid_alloc
  7. closid_free
  8. closid_allocated
  9. rdtgroup_mode_by_closid
  10. rdtgroup_mode_str
  11. rdtgroup_kn_set_ugid
  12. rdtgroup_add_file
  13. rdtgroup_seqfile_show
  14. rdtgroup_file_write
  15. is_cpu_list
  16. rdtgroup_cpus_show
  17. update_cpu_closid_rmid
  18. update_closid_rmid
  19. cpus_mon_write
  20. cpumask_rdtgrp_clear
  21. cpus_ctrl_write
  22. rdtgroup_cpus_write
  23. move_myself
  24. __rdtgroup_move_task
  25. rdtgroup_tasks_assigned
  26. rdtgroup_task_write_permission
  27. rdtgroup_move_task
  28. rdtgroup_tasks_write
  29. show_rdt_tasks
  30. rdtgroup_tasks_show
  31. rdt_last_cmd_status_show
  32. rdt_num_closids_show
  33. rdt_default_ctrl_show
  34. rdt_min_cbm_bits_show
  35. rdt_shareable_bits_show
  36. rdt_bit_usage_show
  37. rdt_min_bw_show
  38. rdt_num_rmids_show
  39. rdt_mon_features_show
  40. rdt_bw_gran_show
  41. rdt_delay_linear_show
  42. max_threshold_occ_show
  43. max_threshold_occ_write
  44. rdtgroup_mode_show
  45. rdt_cdp_peer_get
  46. __rdtgroup_cbm_overlaps
  47. rdtgroup_cbm_overlaps
  48. rdtgroup_mode_test_exclusive
  49. rdtgroup_mode_write
  50. rdtgroup_cbm_to_size
  51. rdtgroup_size_show
  52. rdtgroup_add_files
  53. rdtgroup_kn_mode_restrict
  54. rdtgroup_kn_mode_restore
  55. rdtgroup_mkdir_info_resdir
  56. rdtgroup_create_info_dir
  57. mongroup_create_dir
  58. l3_qos_cfg_update
  59. l2_qos_cfg_update
  60. is_mba_linear
  61. set_cache_qos_cfg
  62. rdt_domain_reconfigure_cdp
  63. set_mba_sc
  64. cdp_enable
  65. cdpl3_enable
  66. cdpl2_enable
  67. cdp_disable
  68. cdpl3_disable
  69. cdpl2_disable
  70. cdp_disable_all
  71. kernfs_to_rdtgroup
  72. rdtgroup_kn_lock_live
  73. rdtgroup_kn_unlock
  74. rdt_enable_ctx
  75. rdt_get_tree
  76. rdt_parse_param
  77. rdt_fs_context_free
  78. rdt_init_fs_context
  79. reset_all_ctrls
  80. is_closid_match
  81. is_rmid_match
  82. rdt_move_group_tasks
  83. free_all_child_rdtgrp
  84. rmdir_all_sub
  85. rdt_kill_sb
  86. mon_addfile
  87. rmdir_mondata_subdir_allrdtgrp
  88. mkdir_mondata_subdir
  89. mkdir_mondata_subdir_allrdtgrp
  90. mkdir_mondata_subdir_alldom
  91. mkdir_mondata_all
  92. cbm_ensure_valid
  93. __init_one_rdt_domain
  94. rdtgroup_init_cat
  95. rdtgroup_init_mba
  96. rdtgroup_init_alloc
  97. mkdir_rdt_prepare
  98. mkdir_rdt_prepare_clean
  99. rdtgroup_mkdir_mon
  100. rdtgroup_mkdir_ctrl_mon
  101. is_mon_groups
  102. rdtgroup_mkdir
  103. rdtgroup_rmdir_mon
  104. rdtgroup_ctrl_remove
  105. rdtgroup_rmdir_ctrl
  106. rdtgroup_rmdir
  107. rdtgroup_show_options
  108. rdtgroup_setup_root
  109. rdtgroup_init
  110. rdtgroup_exit

   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 /* Restore the qos cfg state when a domain comes online */
1773 void rdt_domain_reconfigure_cdp(struct rdt_resource *r)
1774 {
1775         if (!r->alloc_capable)
1776                 return;
1777 
1778         if (r == &rdt_resources_all[RDT_RESOURCE_L2DATA])
1779                 l2_qos_cfg_update(&r->alloc_enabled);
1780 
1781         if (r == &rdt_resources_all[RDT_RESOURCE_L3DATA])
1782                 l3_qos_cfg_update(&r->alloc_enabled);
1783 }
1784 
1785 /*
1786  * Enable or disable the MBA software controller
1787  * which helps user specify bandwidth in MBps.
1788  * MBA software controller is supported only if
1789  * MBM is supported and MBA is in linear scale.
1790  */
1791 static int set_mba_sc(bool mba_sc)
1792 {
1793         struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA];
1794         struct rdt_domain *d;
1795 
1796         if (!is_mbm_enabled() || !is_mba_linear() ||
1797             mba_sc == is_mba_sc(r))
1798                 return -EINVAL;
1799 
1800         r->membw.mba_sc = mba_sc;
1801         list_for_each_entry(d, &r->domains, list)
1802                 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val);
1803 
1804         return 0;
1805 }
1806 
1807 static int cdp_enable(int level, int data_type, int code_type)
1808 {
1809         struct rdt_resource *r_ldata = &rdt_resources_all[data_type];
1810         struct rdt_resource *r_lcode = &rdt_resources_all[code_type];
1811         struct rdt_resource *r_l = &rdt_resources_all[level];
1812         int ret;
1813 
1814         if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1815             !r_lcode->alloc_capable)
1816                 return -EINVAL;
1817 
1818         ret = set_cache_qos_cfg(level, true);
1819         if (!ret) {
1820                 r_l->alloc_enabled = false;
1821                 r_ldata->alloc_enabled = true;
1822                 r_lcode->alloc_enabled = true;
1823         }
1824         return ret;
1825 }
1826 
1827 static int cdpl3_enable(void)
1828 {
1829         return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1830                           RDT_RESOURCE_L3CODE);
1831 }
1832 
1833 static int cdpl2_enable(void)
1834 {
1835         return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1836                           RDT_RESOURCE_L2CODE);
1837 }
1838 
1839 static void cdp_disable(int level, int data_type, int code_type)
1840 {
1841         struct rdt_resource *r = &rdt_resources_all[level];
1842 
1843         r->alloc_enabled = r->alloc_capable;
1844 
1845         if (rdt_resources_all[data_type].alloc_enabled) {
1846                 rdt_resources_all[data_type].alloc_enabled = false;
1847                 rdt_resources_all[code_type].alloc_enabled = false;
1848                 set_cache_qos_cfg(level, false);
1849         }
1850 }
1851 
1852 static void cdpl3_disable(void)
1853 {
1854         cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
1855 }
1856 
1857 static void cdpl2_disable(void)
1858 {
1859         cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
1860 }
1861 
1862 static void cdp_disable_all(void)
1863 {
1864         if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
1865                 cdpl3_disable();
1866         if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
1867                 cdpl2_disable();
1868 }
1869 
1870 /*
1871  * We don't allow rdtgroup directories to be created anywhere
1872  * except the root directory. Thus when looking for the rdtgroup
1873  * structure for a kernfs node we are either looking at a directory,
1874  * in which case the rdtgroup structure is pointed at by the "priv"
1875  * field, otherwise we have a file, and need only look to the parent
1876  * to find the rdtgroup.
1877  */
1878 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
1879 {
1880         if (kernfs_type(kn) == KERNFS_DIR) {
1881                 /*
1882                  * All the resource directories use "kn->priv"
1883                  * to point to the "struct rdtgroup" for the
1884                  * resource. "info" and its subdirectories don't
1885                  * have rdtgroup structures, so return NULL here.
1886                  */
1887                 if (kn == kn_info || kn->parent == kn_info)
1888                         return NULL;
1889                 else
1890                         return kn->priv;
1891         } else {
1892                 return kn->parent->priv;
1893         }
1894 }
1895 
1896 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
1897 {
1898         struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1899 
1900         if (!rdtgrp)
1901                 return NULL;
1902 
1903         atomic_inc(&rdtgrp->waitcount);
1904         kernfs_break_active_protection(kn);
1905 
1906         mutex_lock(&rdtgroup_mutex);
1907 
1908         /* Was this group deleted while we waited? */
1909         if (rdtgrp->flags & RDT_DELETED)
1910                 return NULL;
1911 
1912         return rdtgrp;
1913 }
1914 
1915 void rdtgroup_kn_unlock(struct kernfs_node *kn)
1916 {
1917         struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1918 
1919         if (!rdtgrp)
1920                 return;
1921 
1922         mutex_unlock(&rdtgroup_mutex);
1923 
1924         if (atomic_dec_and_test(&rdtgrp->waitcount) &&
1925             (rdtgrp->flags & RDT_DELETED)) {
1926                 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
1927                     rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
1928                         rdtgroup_pseudo_lock_remove(rdtgrp);
1929                 kernfs_unbreak_active_protection(kn);
1930                 kernfs_put(rdtgrp->kn);
1931                 kfree(rdtgrp);
1932         } else {
1933                 kernfs_unbreak_active_protection(kn);
1934         }
1935 }
1936 
1937 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1938                              struct rdtgroup *prgrp,
1939                              struct kernfs_node **mon_data_kn);
1940 
1941 static int rdt_enable_ctx(struct rdt_fs_context *ctx)
1942 {
1943         int ret = 0;
1944 
1945         if (ctx->enable_cdpl2)
1946                 ret = cdpl2_enable();
1947 
1948         if (!ret && ctx->enable_cdpl3)
1949                 ret = cdpl3_enable();
1950 
1951         if (!ret && ctx->enable_mba_mbps)
1952                 ret = set_mba_sc(true);
1953 
1954         return ret;
1955 }
1956 
1957 static int rdt_get_tree(struct fs_context *fc)
1958 {
1959         struct rdt_fs_context *ctx = rdt_fc2context(fc);
1960         struct rdt_domain *dom;
1961         struct rdt_resource *r;
1962         int ret;
1963 
1964         cpus_read_lock();
1965         mutex_lock(&rdtgroup_mutex);
1966         /*
1967          * resctrl file system can only be mounted once.
1968          */
1969         if (static_branch_unlikely(&rdt_enable_key)) {
1970                 ret = -EBUSY;
1971                 goto out;
1972         }
1973 
1974         ret = rdt_enable_ctx(ctx);
1975         if (ret < 0)
1976                 goto out_cdp;
1977 
1978         closid_init();
1979 
1980         ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
1981         if (ret < 0)
1982                 goto out_mba;
1983 
1984         if (rdt_mon_capable) {
1985                 ret = mongroup_create_dir(rdtgroup_default.kn,
1986                                           &rdtgroup_default, "mon_groups",
1987                                           &kn_mongrp);
1988                 if (ret < 0)
1989                         goto out_info;
1990                 kernfs_get(kn_mongrp);
1991 
1992                 ret = mkdir_mondata_all(rdtgroup_default.kn,
1993                                         &rdtgroup_default, &kn_mondata);
1994                 if (ret < 0)
1995                         goto out_mongrp;
1996                 kernfs_get(kn_mondata);
1997                 rdtgroup_default.mon.mon_data_kn = kn_mondata;
1998         }
1999 
2000         ret = rdt_pseudo_lock_init();
2001         if (ret)
2002                 goto out_mondata;
2003 
2004         ret = kernfs_get_tree(fc);
2005         if (ret < 0)
2006                 goto out_psl;
2007 
2008         if (rdt_alloc_capable)
2009                 static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
2010         if (rdt_mon_capable)
2011                 static_branch_enable_cpuslocked(&rdt_mon_enable_key);
2012 
2013         if (rdt_alloc_capable || rdt_mon_capable)
2014                 static_branch_enable_cpuslocked(&rdt_enable_key);
2015 
2016         if (is_mbm_enabled()) {
2017                 r = &rdt_resources_all[RDT_RESOURCE_L3];
2018                 list_for_each_entry(dom, &r->domains, list)
2019                         mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
2020         }
2021 
2022         goto out;
2023 
2024 out_psl:
2025         rdt_pseudo_lock_release();
2026 out_mondata:
2027         if (rdt_mon_capable)
2028                 kernfs_remove(kn_mondata);
2029 out_mongrp:
2030         if (rdt_mon_capable)
2031                 kernfs_remove(kn_mongrp);
2032 out_info:
2033         kernfs_remove(kn_info);
2034 out_mba:
2035         if (ctx->enable_mba_mbps)
2036                 set_mba_sc(false);
2037 out_cdp:
2038         cdp_disable_all();
2039 out:
2040         rdt_last_cmd_clear();
2041         mutex_unlock(&rdtgroup_mutex);
2042         cpus_read_unlock();
2043         return ret;
2044 }
2045 
2046 enum rdt_param {
2047         Opt_cdp,
2048         Opt_cdpl2,
2049         Opt_mba_mbps,
2050         nr__rdt_params
2051 };
2052 
2053 static const struct fs_parameter_spec rdt_param_specs[] = {
2054         fsparam_flag("cdp",             Opt_cdp),
2055         fsparam_flag("cdpl2",           Opt_cdpl2),
2056         fsparam_flag("mba_MBps",        Opt_mba_mbps),
2057         {}
2058 };
2059 
2060 static const struct fs_parameter_description rdt_fs_parameters = {
2061         .name           = "rdt",
2062         .specs          = rdt_param_specs,
2063 };
2064 
2065 static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param)
2066 {
2067         struct rdt_fs_context *ctx = rdt_fc2context(fc);
2068         struct fs_parse_result result;
2069         int opt;
2070 
2071         opt = fs_parse(fc, &rdt_fs_parameters, param, &result);
2072         if (opt < 0)
2073                 return opt;
2074 
2075         switch (opt) {
2076         case Opt_cdp:
2077                 ctx->enable_cdpl3 = true;
2078                 return 0;
2079         case Opt_cdpl2:
2080                 ctx->enable_cdpl2 = true;
2081                 return 0;
2082         case Opt_mba_mbps:
2083                 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2084                         return -EINVAL;
2085                 ctx->enable_mba_mbps = true;
2086                 return 0;
2087         }
2088 
2089         return -EINVAL;
2090 }
2091 
2092 static void rdt_fs_context_free(struct fs_context *fc)
2093 {
2094         struct rdt_fs_context *ctx = rdt_fc2context(fc);
2095 
2096         kernfs_free_fs_context(fc);
2097         kfree(ctx);
2098 }
2099 
2100 static const struct fs_context_operations rdt_fs_context_ops = {
2101         .free           = rdt_fs_context_free,
2102         .parse_param    = rdt_parse_param,
2103         .get_tree       = rdt_get_tree,
2104 };
2105 
2106 static int rdt_init_fs_context(struct fs_context *fc)
2107 {
2108         struct rdt_fs_context *ctx;
2109 
2110         ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL);
2111         if (!ctx)
2112                 return -ENOMEM;
2113 
2114         ctx->kfc.root = rdt_root;
2115         ctx->kfc.magic = RDTGROUP_SUPER_MAGIC;
2116         fc->fs_private = &ctx->kfc;
2117         fc->ops = &rdt_fs_context_ops;
2118         put_user_ns(fc->user_ns);
2119         fc->user_ns = get_user_ns(&init_user_ns);
2120         fc->global = true;
2121         return 0;
2122 }
2123 
2124 static int reset_all_ctrls(struct rdt_resource *r)
2125 {
2126         struct msr_param msr_param;
2127         cpumask_var_t cpu_mask;
2128         struct rdt_domain *d;
2129         int i, cpu;
2130 
2131         if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
2132                 return -ENOMEM;
2133 
2134         msr_param.res = r;
2135         msr_param.low = 0;
2136         msr_param.high = r->num_closid;
2137 
2138         /*
2139          * Disable resource control for this resource by setting all
2140          * CBMs in all domains to the maximum mask value. Pick one CPU
2141          * from each domain to update the MSRs below.
2142          */
2143         list_for_each_entry(d, &r->domains, list) {
2144                 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
2145 
2146                 for (i = 0; i < r->num_closid; i++)
2147                         d->ctrl_val[i] = r->default_ctrl;
2148         }
2149         cpu = get_cpu();
2150         /* Update CBM on this cpu if it's in cpu_mask. */
2151         if (cpumask_test_cpu(cpu, cpu_mask))
2152                 rdt_ctrl_update(&msr_param);
2153         /* Update CBM on all other cpus in cpu_mask. */
2154         smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
2155         put_cpu();
2156 
2157         free_cpumask_var(cpu_mask);
2158 
2159         return 0;
2160 }
2161 
2162 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
2163 {
2164         return (rdt_alloc_capable &&
2165                 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
2166 }
2167 
2168 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
2169 {
2170         return (rdt_mon_capable &&
2171                 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
2172 }
2173 
2174 /*
2175  * Move tasks from one to the other group. If @from is NULL, then all tasks
2176  * in the systems are moved unconditionally (used for teardown).
2177  *
2178  * If @mask is not NULL the cpus on which moved tasks are running are set
2179  * in that mask so the update smp function call is restricted to affected
2180  * cpus.
2181  */
2182 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
2183                                  struct cpumask *mask)
2184 {
2185         struct task_struct *p, *t;
2186 
2187         read_lock(&tasklist_lock);
2188         for_each_process_thread(p, t) {
2189                 if (!from || is_closid_match(t, from) ||
2190                     is_rmid_match(t, from)) {
2191                         t->closid = to->closid;
2192                         t->rmid = to->mon.rmid;
2193 
2194 #ifdef CONFIG_SMP
2195                         /*
2196                          * This is safe on x86 w/o barriers as the ordering
2197                          * of writing to task_cpu() and t->on_cpu is
2198                          * reverse to the reading here. The detection is
2199                          * inaccurate as tasks might move or schedule
2200                          * before the smp function call takes place. In
2201                          * such a case the function call is pointless, but
2202                          * there is no other side effect.
2203                          */
2204                         if (mask && t->on_cpu)
2205                                 cpumask_set_cpu(task_cpu(t), mask);
2206 #endif
2207                 }
2208         }
2209         read_unlock(&tasklist_lock);
2210 }
2211 
2212 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
2213 {
2214         struct rdtgroup *sentry, *stmp;
2215         struct list_head *head;
2216 
2217         head = &rdtgrp->mon.crdtgrp_list;
2218         list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
2219                 free_rmid(sentry->mon.rmid);
2220                 list_del(&sentry->mon.crdtgrp_list);
2221 
2222                 if (atomic_read(&sentry->waitcount) != 0)
2223                         sentry->flags = RDT_DELETED;
2224                 else
2225                         kfree(sentry);
2226         }
2227 }
2228 
2229 /*
2230  * Forcibly remove all of subdirectories under root.
2231  */
2232 static void rmdir_all_sub(void)
2233 {
2234         struct rdtgroup *rdtgrp, *tmp;
2235 
2236         /* Move all tasks to the default resource group */
2237         rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
2238 
2239         list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
2240                 /* Free any child rmids */
2241                 free_all_child_rdtgrp(rdtgrp);
2242 
2243                 /* Remove each rdtgroup other than root */
2244                 if (rdtgrp == &rdtgroup_default)
2245                         continue;
2246 
2247                 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2248                     rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2249                         rdtgroup_pseudo_lock_remove(rdtgrp);
2250 
2251                 /*
2252                  * Give any CPUs back to the default group. We cannot copy
2253                  * cpu_online_mask because a CPU might have executed the
2254                  * offline callback already, but is still marked online.
2255                  */
2256                 cpumask_or(&rdtgroup_default.cpu_mask,
2257                            &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2258 
2259                 free_rmid(rdtgrp->mon.rmid);
2260 
2261                 kernfs_remove(rdtgrp->kn);
2262                 list_del(&rdtgrp->rdtgroup_list);
2263 
2264                 if (atomic_read(&rdtgrp->waitcount) != 0)
2265                         rdtgrp->flags = RDT_DELETED;
2266                 else
2267                         kfree(rdtgrp);
2268         }
2269         /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
2270         update_closid_rmid(cpu_online_mask, &rdtgroup_default);
2271 
2272         kernfs_remove(kn_info);
2273         kernfs_remove(kn_mongrp);
2274         kernfs_remove(kn_mondata);
2275 }
2276 
2277 static void rdt_kill_sb(struct super_block *sb)
2278 {
2279         struct rdt_resource *r;
2280 
2281         cpus_read_lock();
2282         mutex_lock(&rdtgroup_mutex);
2283 
2284         set_mba_sc(false);
2285 
2286         /*Put everything back to default values. */
2287         for_each_alloc_enabled_rdt_resource(r)
2288                 reset_all_ctrls(r);
2289         cdp_disable_all();
2290         rmdir_all_sub();
2291         rdt_pseudo_lock_release();
2292         rdtgroup_default.mode = RDT_MODE_SHAREABLE;
2293         static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
2294         static_branch_disable_cpuslocked(&rdt_mon_enable_key);
2295         static_branch_disable_cpuslocked(&rdt_enable_key);
2296         kernfs_kill_sb(sb);
2297         mutex_unlock(&rdtgroup_mutex);
2298         cpus_read_unlock();
2299 }
2300 
2301 static struct file_system_type rdt_fs_type = {
2302         .name                   = "resctrl",
2303         .init_fs_context        = rdt_init_fs_context,
2304         .parameters             = &rdt_fs_parameters,
2305         .kill_sb                = rdt_kill_sb,
2306 };
2307 
2308 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
2309                        void *priv)
2310 {
2311         struct kernfs_node *kn;
2312         int ret = 0;
2313 
2314         kn = __kernfs_create_file(parent_kn, name, 0444,
2315                                   GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
2316                                   &kf_mondata_ops, priv, NULL, NULL);
2317         if (IS_ERR(kn))
2318                 return PTR_ERR(kn);
2319 
2320         ret = rdtgroup_kn_set_ugid(kn);
2321         if (ret) {
2322                 kernfs_remove(kn);
2323                 return ret;
2324         }
2325 
2326         return ret;
2327 }
2328 
2329 /*
2330  * Remove all subdirectories of mon_data of ctrl_mon groups
2331  * and monitor groups with given domain id.
2332  */
2333 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
2334 {
2335         struct rdtgroup *prgrp, *crgrp;
2336         char name[32];
2337 
2338         if (!r->mon_enabled)
2339                 return;
2340 
2341         list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2342                 sprintf(name, "mon_%s_%02d", r->name, dom_id);
2343                 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
2344 
2345                 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
2346                         kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
2347         }
2348 }
2349 
2350 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
2351                                 struct rdt_domain *d,
2352                                 struct rdt_resource *r, struct rdtgroup *prgrp)
2353 {
2354         union mon_data_bits priv;
2355         struct kernfs_node *kn;
2356         struct mon_evt *mevt;
2357         struct rmid_read rr;
2358         char name[32];
2359         int ret;
2360 
2361         sprintf(name, "mon_%s_%02d", r->name, d->id);
2362         /* create the directory */
2363         kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
2364         if (IS_ERR(kn))
2365                 return PTR_ERR(kn);
2366 
2367         /*
2368          * This extra ref will be put in kernfs_remove() and guarantees
2369          * that kn is always accessible.
2370          */
2371         kernfs_get(kn);
2372         ret = rdtgroup_kn_set_ugid(kn);
2373         if (ret)
2374                 goto out_destroy;
2375 
2376         if (WARN_ON(list_empty(&r->evt_list))) {
2377                 ret = -EPERM;
2378                 goto out_destroy;
2379         }
2380 
2381         priv.u.rid = r->rid;
2382         priv.u.domid = d->id;
2383         list_for_each_entry(mevt, &r->evt_list, list) {
2384                 priv.u.evtid = mevt->evtid;
2385                 ret = mon_addfile(kn, mevt->name, priv.priv);
2386                 if (ret)
2387                         goto out_destroy;
2388 
2389                 if (is_mbm_event(mevt->evtid))
2390                         mon_event_read(&rr, d, prgrp, mevt->evtid, true);
2391         }
2392         kernfs_activate(kn);
2393         return 0;
2394 
2395 out_destroy:
2396         kernfs_remove(kn);
2397         return ret;
2398 }
2399 
2400 /*
2401  * Add all subdirectories of mon_data for "ctrl_mon" groups
2402  * and "monitor" groups with given domain id.
2403  */
2404 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
2405                                     struct rdt_domain *d)
2406 {
2407         struct kernfs_node *parent_kn;
2408         struct rdtgroup *prgrp, *crgrp;
2409         struct list_head *head;
2410 
2411         if (!r->mon_enabled)
2412                 return;
2413 
2414         list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2415                 parent_kn = prgrp->mon.mon_data_kn;
2416                 mkdir_mondata_subdir(parent_kn, d, r, prgrp);
2417 
2418                 head = &prgrp->mon.crdtgrp_list;
2419                 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
2420                         parent_kn = crgrp->mon.mon_data_kn;
2421                         mkdir_mondata_subdir(parent_kn, d, r, crgrp);
2422                 }
2423         }
2424 }
2425 
2426 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
2427                                        struct rdt_resource *r,
2428                                        struct rdtgroup *prgrp)
2429 {
2430         struct rdt_domain *dom;
2431         int ret;
2432 
2433         list_for_each_entry(dom, &r->domains, list) {
2434                 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
2435                 if (ret)
2436                         return ret;
2437         }
2438 
2439         return 0;
2440 }
2441 
2442 /*
2443  * This creates a directory mon_data which contains the monitored data.
2444  *
2445  * mon_data has one directory for each domain whic are named
2446  * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
2447  * with L3 domain looks as below:
2448  * ./mon_data:
2449  * mon_L3_00
2450  * mon_L3_01
2451  * mon_L3_02
2452  * ...
2453  *
2454  * Each domain directory has one file per event:
2455  * ./mon_L3_00/:
2456  * llc_occupancy
2457  *
2458  */
2459 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2460                              struct rdtgroup *prgrp,
2461                              struct kernfs_node **dest_kn)
2462 {
2463         struct rdt_resource *r;
2464         struct kernfs_node *kn;
2465         int ret;
2466 
2467         /*
2468          * Create the mon_data directory first.
2469          */
2470         ret = mongroup_create_dir(parent_kn, prgrp, "mon_data", &kn);
2471         if (ret)
2472                 return ret;
2473 
2474         if (dest_kn)
2475                 *dest_kn = kn;
2476 
2477         /*
2478          * Create the subdirectories for each domain. Note that all events
2479          * in a domain like L3 are grouped into a resource whose domain is L3
2480          */
2481         for_each_mon_enabled_rdt_resource(r) {
2482                 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
2483                 if (ret)
2484                         goto out_destroy;
2485         }
2486 
2487         return 0;
2488 
2489 out_destroy:
2490         kernfs_remove(kn);
2491         return ret;
2492 }
2493 
2494 /**
2495  * cbm_ensure_valid - Enforce validity on provided CBM
2496  * @_val:       Candidate CBM
2497  * @r:          RDT resource to which the CBM belongs
2498  *
2499  * The provided CBM represents all cache portions available for use. This
2500  * may be represented by a bitmap that does not consist of contiguous ones
2501  * and thus be an invalid CBM.
2502  * Here the provided CBM is forced to be a valid CBM by only considering
2503  * the first set of contiguous bits as valid and clearing all bits.
2504  * The intention here is to provide a valid default CBM with which a new
2505  * resource group is initialized. The user can follow this with a
2506  * modification to the CBM if the default does not satisfy the
2507  * requirements.
2508  */
2509 static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r)
2510 {
2511         unsigned int cbm_len = r->cache.cbm_len;
2512         unsigned long first_bit, zero_bit;
2513         unsigned long val = _val;
2514 
2515         if (!val)
2516                 return 0;
2517 
2518         first_bit = find_first_bit(&val, cbm_len);
2519         zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
2520 
2521         /* Clear any remaining bits to ensure contiguous region */
2522         bitmap_clear(&val, zero_bit, cbm_len - zero_bit);
2523         return (u32)val;
2524 }
2525 
2526 /*
2527  * Initialize cache resources per RDT domain
2528  *
2529  * Set the RDT domain up to start off with all usable allocations. That is,
2530  * all shareable and unused bits. All-zero CBM is invalid.
2531  */
2532 static int __init_one_rdt_domain(struct rdt_domain *d, struct rdt_resource *r,
2533                                  u32 closid)
2534 {
2535         struct rdt_resource *r_cdp = NULL;
2536         struct rdt_domain *d_cdp = NULL;
2537         u32 used_b = 0, unused_b = 0;
2538         unsigned long tmp_cbm;
2539         enum rdtgrp_mode mode;
2540         u32 peer_ctl, *ctrl;
2541         int i;
2542 
2543         rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp);
2544         d->have_new_ctrl = false;
2545         d->new_ctrl = r->cache.shareable_bits;
2546         used_b = r->cache.shareable_bits;
2547         ctrl = d->ctrl_val;
2548         for (i = 0; i < closids_supported(); i++, ctrl++) {
2549                 if (closid_allocated(i) && i != closid) {
2550                         mode = rdtgroup_mode_by_closid(i);
2551                         if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
2552                                 /*
2553                                  * ctrl values for locksetup aren't relevant
2554                                  * until the schemata is written, and the mode
2555                                  * becomes RDT_MODE_PSEUDO_LOCKED.
2556                                  */
2557                                 continue;
2558                         /*
2559                          * If CDP is active include peer domain's
2560                          * usage to ensure there is no overlap
2561                          * with an exclusive group.
2562                          */
2563                         if (d_cdp)
2564                                 peer_ctl = d_cdp->ctrl_val[i];
2565                         else
2566                                 peer_ctl = 0;
2567                         used_b |= *ctrl | peer_ctl;
2568                         if (mode == RDT_MODE_SHAREABLE)
2569                                 d->new_ctrl |= *ctrl | peer_ctl;
2570                 }
2571         }
2572         if (d->plr && d->plr->cbm > 0)
2573                 used_b |= d->plr->cbm;
2574         unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
2575         unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
2576         d->new_ctrl |= unused_b;
2577         /*
2578          * Force the initial CBM to be valid, user can
2579          * modify the CBM based on system availability.
2580          */
2581         d->new_ctrl = cbm_ensure_valid(d->new_ctrl, r);
2582         /*
2583          * Assign the u32 CBM to an unsigned long to ensure that
2584          * bitmap_weight() does not access out-of-bound memory.
2585          */
2586         tmp_cbm = d->new_ctrl;
2587         if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) {
2588                 rdt_last_cmd_printf("No space on %s:%d\n", r->name, d->id);
2589                 return -ENOSPC;
2590         }
2591         d->have_new_ctrl = true;
2592 
2593         return 0;
2594 }
2595 
2596 /*
2597  * Initialize cache resources with default values.
2598  *
2599  * A new RDT group is being created on an allocation capable (CAT)
2600  * supporting system. Set this group up to start off with all usable
2601  * allocations.
2602  *
2603  * If there are no more shareable bits available on any domain then
2604  * the entire allocation will fail.
2605  */
2606 static int rdtgroup_init_cat(struct rdt_resource *r, u32 closid)
2607 {
2608         struct rdt_domain *d;
2609         int ret;
2610 
2611         list_for_each_entry(d, &r->domains, list) {
2612                 ret = __init_one_rdt_domain(d, r, closid);
2613                 if (ret < 0)
2614                         return ret;
2615         }
2616 
2617         return 0;
2618 }
2619 
2620 /* Initialize MBA resource with default values. */
2621 static void rdtgroup_init_mba(struct rdt_resource *r)
2622 {
2623         struct rdt_domain *d;
2624 
2625         list_for_each_entry(d, &r->domains, list) {
2626                 d->new_ctrl = is_mba_sc(r) ? MBA_MAX_MBPS : r->default_ctrl;
2627                 d->have_new_ctrl = true;
2628         }
2629 }
2630 
2631 /* Initialize the RDT group's allocations. */
2632 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
2633 {
2634         struct rdt_resource *r;
2635         int ret;
2636 
2637         for_each_alloc_enabled_rdt_resource(r) {
2638                 if (r->rid == RDT_RESOURCE_MBA) {
2639                         rdtgroup_init_mba(r);
2640                 } else {
2641                         ret = rdtgroup_init_cat(r, rdtgrp->closid);
2642                         if (ret < 0)
2643                                 return ret;
2644                 }
2645 
2646                 ret = update_domains(r, rdtgrp->closid);
2647                 if (ret < 0) {
2648                         rdt_last_cmd_puts("Failed to initialize allocations\n");
2649                         return ret;
2650                 }
2651 
2652         }
2653 
2654         rdtgrp->mode = RDT_MODE_SHAREABLE;
2655 
2656         return 0;
2657 }
2658 
2659 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
2660                              struct kernfs_node *prgrp_kn,
2661                              const char *name, umode_t mode,
2662                              enum rdt_group_type rtype, struct rdtgroup **r)
2663 {
2664         struct rdtgroup *prdtgrp, *rdtgrp;
2665         struct kernfs_node *kn;
2666         uint files = 0;
2667         int ret;
2668 
2669         prdtgrp = rdtgroup_kn_lock_live(parent_kn);
2670         if (!prdtgrp) {
2671                 ret = -ENODEV;
2672                 goto out_unlock;
2673         }
2674 
2675         if (rtype == RDTMON_GROUP &&
2676             (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2677              prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
2678                 ret = -EINVAL;
2679                 rdt_last_cmd_puts("Pseudo-locking in progress\n");
2680                 goto out_unlock;
2681         }
2682 
2683         /* allocate the rdtgroup. */
2684         rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
2685         if (!rdtgrp) {
2686                 ret = -ENOSPC;
2687                 rdt_last_cmd_puts("Kernel out of memory\n");
2688                 goto out_unlock;
2689         }
2690         *r = rdtgrp;
2691         rdtgrp->mon.parent = prdtgrp;
2692         rdtgrp->type = rtype;
2693         INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
2694 
2695         /* kernfs creates the directory for rdtgrp */
2696         kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
2697         if (IS_ERR(kn)) {
2698                 ret = PTR_ERR(kn);
2699                 rdt_last_cmd_puts("kernfs create error\n");
2700                 goto out_free_rgrp;
2701         }
2702         rdtgrp->kn = kn;
2703 
2704         /*
2705          * kernfs_remove() will drop the reference count on "kn" which
2706          * will free it. But we still need it to stick around for the
2707          * rdtgroup_kn_unlock(kn} call below. Take one extra reference
2708          * here, which will be dropped inside rdtgroup_kn_unlock().
2709          */
2710         kernfs_get(kn);
2711 
2712         ret = rdtgroup_kn_set_ugid(kn);
2713         if (ret) {
2714                 rdt_last_cmd_puts("kernfs perm error\n");
2715                 goto out_destroy;
2716         }
2717 
2718         files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
2719         ret = rdtgroup_add_files(kn, files);
2720         if (ret) {
2721                 rdt_last_cmd_puts("kernfs fill error\n");
2722                 goto out_destroy;
2723         }
2724 
2725         if (rdt_mon_capable) {
2726                 ret = alloc_rmid();
2727                 if (ret < 0) {
2728                         rdt_last_cmd_puts("Out of RMIDs\n");
2729                         goto out_destroy;
2730                 }
2731                 rdtgrp->mon.rmid = ret;
2732 
2733                 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
2734                 if (ret) {
2735                         rdt_last_cmd_puts("kernfs subdir error\n");
2736                         goto out_idfree;
2737                 }
2738         }
2739         kernfs_activate(kn);
2740 
2741         /*
2742          * The caller unlocks the parent_kn upon success.
2743          */
2744         return 0;
2745 
2746 out_idfree:
2747         free_rmid(rdtgrp->mon.rmid);
2748 out_destroy:
2749         kernfs_remove(rdtgrp->kn);
2750 out_free_rgrp:
2751         kfree(rdtgrp);
2752 out_unlock:
2753         rdtgroup_kn_unlock(parent_kn);
2754         return ret;
2755 }
2756 
2757 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
2758 {
2759         kernfs_remove(rgrp->kn);
2760         free_rmid(rgrp->mon.rmid);
2761         kfree(rgrp);
2762 }
2763 
2764 /*
2765  * Create a monitor group under "mon_groups" directory of a control
2766  * and monitor group(ctrl_mon). This is a resource group
2767  * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
2768  */
2769 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
2770                               struct kernfs_node *prgrp_kn,
2771                               const char *name,
2772                               umode_t mode)
2773 {
2774         struct rdtgroup *rdtgrp, *prgrp;
2775         int ret;
2776 
2777         ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP,
2778                                 &rdtgrp);
2779         if (ret)
2780                 return ret;
2781 
2782         prgrp = rdtgrp->mon.parent;
2783         rdtgrp->closid = prgrp->closid;
2784 
2785         /*
2786          * Add the rdtgrp to the list of rdtgrps the parent
2787          * ctrl_mon group has to track.
2788          */
2789         list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
2790 
2791         rdtgroup_kn_unlock(parent_kn);
2792         return ret;
2793 }
2794 
2795 /*
2796  * These are rdtgroups created under the root directory. Can be used
2797  * to allocate and monitor resources.
2798  */
2799 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
2800                                    struct kernfs_node *prgrp_kn,
2801                                    const char *name, umode_t mode)
2802 {
2803         struct rdtgroup *rdtgrp;
2804         struct kernfs_node *kn;
2805         u32 closid;
2806         int ret;
2807 
2808         ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP,
2809                                 &rdtgrp);
2810         if (ret)
2811                 return ret;
2812 
2813         kn = rdtgrp->kn;
2814         ret = closid_alloc();
2815         if (ret < 0) {
2816                 rdt_last_cmd_puts("Out of CLOSIDs\n");
2817                 goto out_common_fail;
2818         }
2819         closid = ret;
2820         ret = 0;
2821 
2822         rdtgrp->closid = closid;
2823         ret = rdtgroup_init_alloc(rdtgrp);
2824         if (ret < 0)
2825                 goto out_id_free;
2826 
2827         list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
2828 
2829         if (rdt_mon_capable) {
2830                 /*
2831                  * Create an empty mon_groups directory to hold the subset
2832                  * of tasks and cpus to monitor.
2833                  */
2834                 ret = mongroup_create_dir(kn, rdtgrp, "mon_groups", NULL);
2835                 if (ret) {
2836                         rdt_last_cmd_puts("kernfs subdir error\n");
2837                         goto out_del_list;
2838                 }
2839         }
2840 
2841         goto out_unlock;
2842 
2843 out_del_list:
2844         list_del(&rdtgrp->rdtgroup_list);
2845 out_id_free:
2846         closid_free(closid);
2847 out_common_fail:
2848         mkdir_rdt_prepare_clean(rdtgrp);
2849 out_unlock:
2850         rdtgroup_kn_unlock(parent_kn);
2851         return ret;
2852 }
2853 
2854 /*
2855  * We allow creating mon groups only with in a directory called "mon_groups"
2856  * which is present in every ctrl_mon group. Check if this is a valid
2857  * "mon_groups" directory.
2858  *
2859  * 1. The directory should be named "mon_groups".
2860  * 2. The mon group itself should "not" be named "mon_groups".
2861  *   This makes sure "mon_groups" directory always has a ctrl_mon group
2862  *   as parent.
2863  */
2864 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
2865 {
2866         return (!strcmp(kn->name, "mon_groups") &&
2867                 strcmp(name, "mon_groups"));
2868 }
2869 
2870 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
2871                           umode_t mode)
2872 {
2873         /* Do not accept '\n' to avoid unparsable situation. */
2874         if (strchr(name, '\n'))
2875                 return -EINVAL;
2876 
2877         /*
2878          * If the parent directory is the root directory and RDT
2879          * allocation is supported, add a control and monitoring
2880          * subdirectory
2881          */
2882         if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
2883                 return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode);
2884 
2885         /*
2886          * If RDT monitoring is supported and the parent directory is a valid
2887          * "mon_groups" directory, add a monitoring subdirectory.
2888          */
2889         if (rdt_mon_capable && is_mon_groups(parent_kn, name))
2890                 return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode);
2891 
2892         return -EPERM;
2893 }
2894 
2895 static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2896                               cpumask_var_t tmpmask)
2897 {
2898         struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
2899         int cpu;
2900 
2901         /* Give any tasks back to the parent group */
2902         rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
2903 
2904         /* Update per cpu rmid of the moved CPUs first */
2905         for_each_cpu(cpu, &rdtgrp->cpu_mask)
2906                 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
2907         /*
2908          * Update the MSR on moved CPUs and CPUs which have moved
2909          * task running on them.
2910          */
2911         cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2912         update_closid_rmid(tmpmask, NULL);
2913 
2914         rdtgrp->flags = RDT_DELETED;
2915         free_rmid(rdtgrp->mon.rmid);
2916 
2917         /*
2918          * Remove the rdtgrp from the parent ctrl_mon group's list
2919          */
2920         WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
2921         list_del(&rdtgrp->mon.crdtgrp_list);
2922 
2923         /*
2924          * one extra hold on this, will drop when we kfree(rdtgrp)
2925          * in rdtgroup_kn_unlock()
2926          */
2927         kernfs_get(kn);
2928         kernfs_remove(rdtgrp->kn);
2929 
2930         return 0;
2931 }
2932 
2933 static int rdtgroup_ctrl_remove(struct kernfs_node *kn,
2934                                 struct rdtgroup *rdtgrp)
2935 {
2936         rdtgrp->flags = RDT_DELETED;
2937         list_del(&rdtgrp->rdtgroup_list);
2938 
2939         /*
2940          * one extra hold on this, will drop when we kfree(rdtgrp)
2941          * in rdtgroup_kn_unlock()
2942          */
2943         kernfs_get(kn);
2944         kernfs_remove(rdtgrp->kn);
2945         return 0;
2946 }
2947 
2948 static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2949                                cpumask_var_t tmpmask)
2950 {
2951         int cpu;
2952 
2953         /* Give any tasks back to the default group */
2954         rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
2955 
2956         /* Give any CPUs back to the default group */
2957         cpumask_or(&rdtgroup_default.cpu_mask,
2958                    &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2959 
2960         /* Update per cpu closid and rmid of the moved CPUs first */
2961         for_each_cpu(cpu, &rdtgrp->cpu_mask) {
2962                 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
2963                 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
2964         }
2965 
2966         /*
2967          * Update the MSR on moved CPUs and CPUs which have moved
2968          * task running on them.
2969          */
2970         cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2971         update_closid_rmid(tmpmask, NULL);
2972 
2973         closid_free(rdtgrp->closid);
2974         free_rmid(rdtgrp->mon.rmid);
2975 
2976         rdtgroup_ctrl_remove(kn, rdtgrp);
2977 
2978         /*
2979          * Free all the child monitor group rmids.
2980          */
2981         free_all_child_rdtgrp(rdtgrp);
2982 
2983         return 0;
2984 }
2985 
2986 static int rdtgroup_rmdir(struct kernfs_node *kn)
2987 {
2988         struct kernfs_node *parent_kn = kn->parent;
2989         struct rdtgroup *rdtgrp;
2990         cpumask_var_t tmpmask;
2991         int ret = 0;
2992 
2993         if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
2994                 return -ENOMEM;
2995 
2996         rdtgrp = rdtgroup_kn_lock_live(kn);
2997         if (!rdtgrp) {
2998                 ret = -EPERM;
2999                 goto out;
3000         }
3001 
3002         /*
3003          * If the rdtgroup is a ctrl_mon group and parent directory
3004          * is the root directory, remove the ctrl_mon group.
3005          *
3006          * If the rdtgroup is a mon group and parent directory
3007          * is a valid "mon_groups" directory, remove the mon group.
3008          */
3009         if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn &&
3010             rdtgrp != &rdtgroup_default) {
3011                 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
3012                     rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
3013                         ret = rdtgroup_ctrl_remove(kn, rdtgrp);
3014                 } else {
3015                         ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask);
3016                 }
3017         } else if (rdtgrp->type == RDTMON_GROUP &&
3018                  is_mon_groups(parent_kn, kn->name)) {
3019                 ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask);
3020         } else {
3021                 ret = -EPERM;
3022         }
3023 
3024 out:
3025         rdtgroup_kn_unlock(kn);
3026         free_cpumask_var(tmpmask);
3027         return ret;
3028 }
3029 
3030 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
3031 {
3032         if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
3033                 seq_puts(seq, ",cdp");
3034 
3035         if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
3036                 seq_puts(seq, ",cdpl2");
3037 
3038         if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA]))
3039                 seq_puts(seq, ",mba_MBps");
3040 
3041         return 0;
3042 }
3043 
3044 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
3045         .mkdir          = rdtgroup_mkdir,
3046         .rmdir          = rdtgroup_rmdir,
3047         .show_options   = rdtgroup_show_options,
3048 };
3049 
3050 static int __init rdtgroup_setup_root(void)
3051 {
3052         int ret;
3053 
3054         rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
3055                                       KERNFS_ROOT_CREATE_DEACTIVATED |
3056                                       KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
3057                                       &rdtgroup_default);
3058         if (IS_ERR(rdt_root))
3059                 return PTR_ERR(rdt_root);
3060 
3061         mutex_lock(&rdtgroup_mutex);
3062 
3063         rdtgroup_default.closid = 0;
3064         rdtgroup_default.mon.rmid = 0;
3065         rdtgroup_default.type = RDTCTRL_GROUP;
3066         INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
3067 
3068         list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
3069 
3070         ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
3071         if (ret) {
3072                 kernfs_destroy_root(rdt_root);
3073                 goto out;
3074         }
3075 
3076         rdtgroup_default.kn = rdt_root->kn;
3077         kernfs_activate(rdtgroup_default.kn);
3078 
3079 out:
3080         mutex_unlock(&rdtgroup_mutex);
3081 
3082         return ret;
3083 }
3084 
3085 /*
3086  * rdtgroup_init - rdtgroup initialization
3087  *
3088  * Setup resctrl file system including set up root, create mount point,
3089  * register rdtgroup filesystem, and initialize files under root directory.
3090  *
3091  * Return: 0 on success or -errno
3092  */
3093 int __init rdtgroup_init(void)
3094 {
3095         int ret = 0;
3096 
3097         seq_buf_init(&last_cmd_status, last_cmd_status_buf,
3098                      sizeof(last_cmd_status_buf));
3099 
3100         ret = rdtgroup_setup_root();
3101         if (ret)
3102                 return ret;
3103 
3104         ret = sysfs_create_mount_point(fs_kobj, "resctrl");
3105         if (ret)
3106                 goto cleanup_root;
3107 
3108         ret = register_filesystem(&rdt_fs_type);
3109         if (ret)
3110                 goto cleanup_mountpoint;
3111 
3112         /*
3113          * Adding the resctrl debugfs directory here may not be ideal since
3114          * it would let the resctrl debugfs directory appear on the debugfs
3115          * filesystem before the resctrl filesystem is mounted.
3116          * It may also be ok since that would enable debugging of RDT before
3117          * resctrl is mounted.
3118          * The reason why the debugfs directory is created here and not in
3119          * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and
3120          * during the debugfs directory creation also &sb->s_type->i_mutex_key
3121          * (the lockdep class of inode->i_rwsem). Other filesystem
3122          * interactions (eg. SyS_getdents) have the lock ordering:
3123          * &sb->s_type->i_mutex_key --> &mm->mmap_sem
3124          * During mmap(), called with &mm->mmap_sem, the rdtgroup_mutex
3125          * is taken, thus creating dependency:
3126          * &mm->mmap_sem --> rdtgroup_mutex for the latter that can cause
3127          * issues considering the other two lock dependencies.
3128          * By creating the debugfs directory here we avoid a dependency
3129          * that may cause deadlock (even though file operations cannot
3130          * occur until the filesystem is mounted, but I do not know how to
3131          * tell lockdep that).
3132          */
3133         debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
3134 
3135         return 0;
3136 
3137 cleanup_mountpoint:
3138         sysfs_remove_mount_point(fs_kobj, "resctrl");
3139 cleanup_root:
3140         kernfs_destroy_root(rdt_root);
3141 
3142         return ret;
3143 }
3144 
3145 void __exit rdtgroup_exit(void)
3146 {
3147         debugfs_remove_recursive(debugfs_resctrl);
3148         unregister_filesystem(&rdt_fs_type);
3149         sysfs_remove_mount_point(fs_kobj, "resctrl");
3150         kernfs_destroy_root(rdt_root);
3151 }

/* [<][>][^][v][top][bottom][index][help] */