1 // SPDX-License-Identifier: GPL-2.0-only
3 * User interface for Resource Allocation in Resource Director Technology(RDT)
5 * Copyright (C) 2016 Intel Corporation
7 * Author: Fenghua Yu <fenghua.yu@intel.com>
9 * More information about RDT be found in the Intel (R) x86 Architecture
10 * Software Developer Manual.
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/cacheinfo.h>
16 #include <linux/cpu.h>
17 #include <linux/debugfs.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>
30 #include <uapi/linux/magic.h>
32 #include <asm/resctrl.h>
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);
42 /* Kernel fs node for "info" directory under root */
43 static struct kernfs_node *kn_info;
45 /* Kernel fs node for "mon_groups" directory under root */
46 static struct kernfs_node *kn_mongrp;
48 /* Kernel fs node for "mon_data" directory under root */
49 static struct kernfs_node *kn_mondata;
51 static struct seq_buf last_cmd_status;
52 static char last_cmd_status_buf[512];
54 struct dentry *debugfs_resctrl;
56 void rdt_last_cmd_clear(void)
58 lockdep_assert_held(&rdtgroup_mutex);
59 seq_buf_clear(&last_cmd_status);
62 void rdt_last_cmd_puts(const char *s)
64 lockdep_assert_held(&rdtgroup_mutex);
65 seq_buf_puts(&last_cmd_status, s);
68 void rdt_last_cmd_printf(const char *fmt, ...)
73 lockdep_assert_held(&rdtgroup_mutex);
74 seq_buf_vprintf(&last_cmd_status, fmt, ap);
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.
82 * Using a global CLOSID across all resources has some advantages and
84 * + We can simply set "current->closid" to assign a task to a resource
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
90 * - Our choices on how to configure each resource become progressively more
91 * limited as the number of resources grows.
93 static int closid_free_map;
94 static int closid_free_map_len;
96 int closids_supported(void)
98 return closid_free_map_len;
101 static void closid_init(void)
103 struct rdt_hw_resource *hw_res;
104 struct rdt_resource *r;
105 int rdt_min_closid = 32;
107 /* Compute rdt_min_closid across all resources */
108 for_each_alloc_enabled_rdt_resource(r) {
109 hw_res = resctrl_to_arch_res(r);
110 rdt_min_closid = min(rdt_min_closid, hw_res->num_closid);
113 closid_free_map = BIT_MASK(rdt_min_closid) - 1;
115 /* CLOSID 0 is always reserved for the default group */
116 closid_free_map &= ~1;
117 closid_free_map_len = rdt_min_closid;
120 static int closid_alloc(void)
122 u32 closid = ffs(closid_free_map);
127 closid_free_map &= ~(1 << closid);
132 void closid_free(int closid)
134 closid_free_map |= 1 << closid;
138 * closid_allocated - test if provided closid is in use
139 * @closid: closid to be tested
141 * Return: true if @closid is currently associated with a resource group,
142 * false if @closid is free
144 static bool closid_allocated(unsigned int closid)
146 return (closid_free_map & (1 << closid)) == 0;
150 * rdtgroup_mode_by_closid - Return mode of resource group with closid
151 * @closid: closid if the resource group
153 * Each resource group is associated with a @closid. Here the mode
154 * of a resource group can be queried by searching for it using its closid.
156 * Return: mode as &enum rdtgrp_mode of resource group with closid @closid
158 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid)
160 struct rdtgroup *rdtgrp;
162 list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
163 if (rdtgrp->closid == closid)
167 return RDT_NUM_MODES;
170 static const char * const rdt_mode_str[] = {
171 [RDT_MODE_SHAREABLE] = "shareable",
172 [RDT_MODE_EXCLUSIVE] = "exclusive",
173 [RDT_MODE_PSEUDO_LOCKSETUP] = "pseudo-locksetup",
174 [RDT_MODE_PSEUDO_LOCKED] = "pseudo-locked",
178 * rdtgroup_mode_str - Return the string representation of mode
179 * @mode: the resource group mode as &enum rdtgroup_mode
181 * Return: string representation of valid mode, "unknown" otherwise
183 static const char *rdtgroup_mode_str(enum rdtgrp_mode mode)
185 if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES)
188 return rdt_mode_str[mode];
191 /* set uid and gid of rdtgroup dirs and files to that of the creator */
192 static int rdtgroup_kn_set_ugid(struct kernfs_node *kn)
194 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
195 .ia_uid = current_fsuid(),
196 .ia_gid = current_fsgid(), };
198 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
199 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
202 return kernfs_setattr(kn, &iattr);
205 static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft)
207 struct kernfs_node *kn;
210 kn = __kernfs_create_file(parent_kn, rft->name, rft->mode,
211 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
212 0, rft->kf_ops, rft, NULL, NULL);
216 ret = rdtgroup_kn_set_ugid(kn);
225 static int rdtgroup_seqfile_show(struct seq_file *m, void *arg)
227 struct kernfs_open_file *of = m->private;
228 struct rftype *rft = of->kn->priv;
231 return rft->seq_show(of, m, arg);
235 static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf,
236 size_t nbytes, loff_t off)
238 struct rftype *rft = of->kn->priv;
241 return rft->write(of, buf, nbytes, off);
246 static const struct kernfs_ops rdtgroup_kf_single_ops = {
247 .atomic_write_len = PAGE_SIZE,
248 .write = rdtgroup_file_write,
249 .seq_show = rdtgroup_seqfile_show,
252 static const struct kernfs_ops kf_mondata_ops = {
253 .atomic_write_len = PAGE_SIZE,
254 .seq_show = rdtgroup_mondata_show,
257 static bool is_cpu_list(struct kernfs_open_file *of)
259 struct rftype *rft = of->kn->priv;
261 return rft->flags & RFTYPE_FLAGS_CPUS_LIST;
264 static int rdtgroup_cpus_show(struct kernfs_open_file *of,
265 struct seq_file *s, void *v)
267 struct rdtgroup *rdtgrp;
268 struct cpumask *mask;
271 rdtgrp = rdtgroup_kn_lock_live(of->kn);
274 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
275 if (!rdtgrp->plr->d) {
276 rdt_last_cmd_clear();
277 rdt_last_cmd_puts("Cache domain offline\n");
280 mask = &rdtgrp->plr->d->cpu_mask;
281 seq_printf(s, is_cpu_list(of) ?
282 "%*pbl\n" : "%*pb\n",
283 cpumask_pr_args(mask));
286 seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n",
287 cpumask_pr_args(&rdtgrp->cpu_mask));
292 rdtgroup_kn_unlock(of->kn);
298 * This is safe against resctrl_sched_in() called from __switch_to()
299 * because __switch_to() is executed with interrupts disabled. A local call
300 * from update_closid_rmid() is protected against __switch_to() because
301 * preemption is disabled.
303 static void update_cpu_closid_rmid(void *info)
305 struct rdtgroup *r = info;
308 this_cpu_write(pqr_state.default_closid, r->closid);
309 this_cpu_write(pqr_state.default_rmid, r->mon.rmid);
313 * We cannot unconditionally write the MSR because the current
314 * executing task might have its own closid selected. Just reuse
315 * the context switch code.
321 * Update the PGR_ASSOC MSR on all cpus in @cpu_mask,
323 * Per task closids/rmids must have been set up before calling this function.
326 update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r)
330 if (cpumask_test_cpu(cpu, cpu_mask))
331 update_cpu_closid_rmid(r);
332 smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1);
336 static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
337 cpumask_var_t tmpmask)
339 struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp;
340 struct list_head *head;
342 /* Check whether cpus belong to parent ctrl group */
343 cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask);
344 if (cpumask_weight(tmpmask)) {
345 rdt_last_cmd_puts("Can only add CPUs to mongroup that belong to parent\n");
349 /* Check whether cpus are dropped from this group */
350 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
351 if (cpumask_weight(tmpmask)) {
352 /* Give any dropped cpus to parent rdtgroup */
353 cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask);
354 update_closid_rmid(tmpmask, prgrp);
358 * If we added cpus, remove them from previous group that owned them
359 * and update per-cpu rmid
361 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
362 if (cpumask_weight(tmpmask)) {
363 head = &prgrp->mon.crdtgrp_list;
364 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
367 cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask,
370 update_closid_rmid(tmpmask, rdtgrp);
373 /* Done pushing/pulling - update this group with new mask */
374 cpumask_copy(&rdtgrp->cpu_mask, newmask);
379 static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m)
381 struct rdtgroup *crgrp;
383 cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m);
384 /* update the child mon group masks as well*/
385 list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list)
386 cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask);
389 static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
390 cpumask_var_t tmpmask, cpumask_var_t tmpmask1)
392 struct rdtgroup *r, *crgrp;
393 struct list_head *head;
395 /* Check whether cpus are dropped from this group */
396 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
397 if (cpumask_weight(tmpmask)) {
398 /* Can't drop from default group */
399 if (rdtgrp == &rdtgroup_default) {
400 rdt_last_cmd_puts("Can't drop CPUs from default group\n");
404 /* Give any dropped cpus to rdtgroup_default */
405 cpumask_or(&rdtgroup_default.cpu_mask,
406 &rdtgroup_default.cpu_mask, tmpmask);
407 update_closid_rmid(tmpmask, &rdtgroup_default);
411 * If we added cpus, remove them from previous group and
412 * the prev group's child groups that owned them
413 * and update per-cpu closid/rmid.
415 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
416 if (cpumask_weight(tmpmask)) {
417 list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) {
420 cpumask_and(tmpmask1, &r->cpu_mask, tmpmask);
421 if (cpumask_weight(tmpmask1))
422 cpumask_rdtgrp_clear(r, tmpmask1);
424 update_closid_rmid(tmpmask, rdtgrp);
427 /* Done pushing/pulling - update this group with new mask */
428 cpumask_copy(&rdtgrp->cpu_mask, newmask);
431 * Clear child mon group masks since there is a new parent mask
432 * now and update the rmid for the cpus the child lost.
434 head = &rdtgrp->mon.crdtgrp_list;
435 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
436 cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask);
437 update_closid_rmid(tmpmask, rdtgrp);
438 cpumask_clear(&crgrp->cpu_mask);
444 static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of,
445 char *buf, size_t nbytes, loff_t off)
447 cpumask_var_t tmpmask, newmask, tmpmask1;
448 struct rdtgroup *rdtgrp;
454 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
456 if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) {
457 free_cpumask_var(tmpmask);
460 if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) {
461 free_cpumask_var(tmpmask);
462 free_cpumask_var(newmask);
466 rdtgrp = rdtgroup_kn_lock_live(of->kn);
472 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
473 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
475 rdt_last_cmd_puts("Pseudo-locking in progress\n");
480 ret = cpulist_parse(buf, newmask);
482 ret = cpumask_parse(buf, newmask);
485 rdt_last_cmd_puts("Bad CPU list/mask\n");
489 /* check that user didn't specify any offline cpus */
490 cpumask_andnot(tmpmask, newmask, cpu_online_mask);
491 if (cpumask_weight(tmpmask)) {
493 rdt_last_cmd_puts("Can only assign online CPUs\n");
497 if (rdtgrp->type == RDTCTRL_GROUP)
498 ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1);
499 else if (rdtgrp->type == RDTMON_GROUP)
500 ret = cpus_mon_write(rdtgrp, newmask, tmpmask);
505 rdtgroup_kn_unlock(of->kn);
506 free_cpumask_var(tmpmask);
507 free_cpumask_var(newmask);
508 free_cpumask_var(tmpmask1);
510 return ret ?: nbytes;
514 * rdtgroup_remove - the helper to remove resource group safely
515 * @rdtgrp: resource group to remove
517 * On resource group creation via a mkdir, an extra kernfs_node reference is
518 * taken to ensure that the rdtgroup structure remains accessible for the
519 * rdtgroup_kn_unlock() calls where it is removed.
521 * Drop the extra reference here, then free the rdtgroup structure.
525 static void rdtgroup_remove(struct rdtgroup *rdtgrp)
527 kernfs_put(rdtgrp->kn);
531 static void _update_task_closid_rmid(void *task)
534 * If the task is still current on this CPU, update PQR_ASSOC MSR.
535 * Otherwise, the MSR is updated when the task is scheduled in.
541 static void update_task_closid_rmid(struct task_struct *t)
543 if (IS_ENABLED(CONFIG_SMP) && task_curr(t))
544 smp_call_function_single(task_cpu(t), _update_task_closid_rmid, t, 1);
546 _update_task_closid_rmid(t);
549 static int __rdtgroup_move_task(struct task_struct *tsk,
550 struct rdtgroup *rdtgrp)
552 /* If the task is already in rdtgrp, no need to move the task. */
553 if ((rdtgrp->type == RDTCTRL_GROUP && tsk->closid == rdtgrp->closid &&
554 tsk->rmid == rdtgrp->mon.rmid) ||
555 (rdtgrp->type == RDTMON_GROUP && tsk->rmid == rdtgrp->mon.rmid &&
556 tsk->closid == rdtgrp->mon.parent->closid))
560 * Set the task's closid/rmid before the PQR_ASSOC MSR can be
563 * For ctrl_mon groups, move both closid and rmid.
564 * For monitor groups, can move the tasks only from
565 * their parent CTRL group.
568 if (rdtgrp->type == RDTCTRL_GROUP) {
569 WRITE_ONCE(tsk->closid, rdtgrp->closid);
570 WRITE_ONCE(tsk->rmid, rdtgrp->mon.rmid);
571 } else if (rdtgrp->type == RDTMON_GROUP) {
572 if (rdtgrp->mon.parent->closid == tsk->closid) {
573 WRITE_ONCE(tsk->rmid, rdtgrp->mon.rmid);
575 rdt_last_cmd_puts("Can't move task to different control group\n");
581 * Ensure the task's closid and rmid are written before determining if
582 * the task is current that will decide if it will be interrupted.
587 * By now, the task's closid and rmid are set. If the task is current
588 * on a CPU, the PQR_ASSOC MSR needs to be updated to make the resource
589 * group go into effect. If the task is not current, the MSR will be
590 * updated when the task is scheduled in.
592 update_task_closid_rmid(tsk);
597 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
599 return (rdt_alloc_capable &&
600 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
603 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
605 return (rdt_mon_capable &&
606 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
610 * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group
613 * Return: 1 if tasks have been assigned to @r, 0 otherwise
615 int rdtgroup_tasks_assigned(struct rdtgroup *r)
617 struct task_struct *p, *t;
620 lockdep_assert_held(&rdtgroup_mutex);
623 for_each_process_thread(p, t) {
624 if (is_closid_match(t, r) || is_rmid_match(t, r)) {
634 static int rdtgroup_task_write_permission(struct task_struct *task,
635 struct kernfs_open_file *of)
637 const struct cred *tcred = get_task_cred(task);
638 const struct cred *cred = current_cred();
642 * Even if we're attaching all tasks in the thread group, we only
643 * need to check permissions on one of them.
645 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
646 !uid_eq(cred->euid, tcred->uid) &&
647 !uid_eq(cred->euid, tcred->suid)) {
648 rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
656 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
657 struct kernfs_open_file *of)
659 struct task_struct *tsk;
664 tsk = find_task_by_vpid(pid);
667 rdt_last_cmd_printf("No task %d\n", pid);
674 get_task_struct(tsk);
677 ret = rdtgroup_task_write_permission(tsk, of);
679 ret = __rdtgroup_move_task(tsk, rdtgrp);
681 put_task_struct(tsk);
685 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
686 char *buf, size_t nbytes, loff_t off)
688 struct rdtgroup *rdtgrp;
692 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
694 rdtgrp = rdtgroup_kn_lock_live(of->kn);
696 rdtgroup_kn_unlock(of->kn);
699 rdt_last_cmd_clear();
701 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
702 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
704 rdt_last_cmd_puts("Pseudo-locking in progress\n");
708 ret = rdtgroup_move_task(pid, rdtgrp, of);
711 rdtgroup_kn_unlock(of->kn);
713 return ret ?: nbytes;
716 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
718 struct task_struct *p, *t;
721 for_each_process_thread(p, t) {
722 if (is_closid_match(t, r) || is_rmid_match(t, r))
723 seq_printf(s, "%d\n", t->pid);
728 static int rdtgroup_tasks_show(struct kernfs_open_file *of,
729 struct seq_file *s, void *v)
731 struct rdtgroup *rdtgrp;
734 rdtgrp = rdtgroup_kn_lock_live(of->kn);
736 show_rdt_tasks(rdtgrp, s);
739 rdtgroup_kn_unlock(of->kn);
744 #ifdef CONFIG_PROC_CPU_RESCTRL
747 * A task can only be part of one resctrl control group and of one monitor
748 * group which is associated to that control group.
753 * resctrl is not available.
758 * Task is part of the root resctrl control group, and it is not associated
759 * to any monitor group.
764 * Task is part of the root resctrl control group and monitor group mon0.
769 * Task is part of resctrl control group group0, and it is not associated
770 * to any monitor group.
775 * Task is part of resctrl control group group0 and monitor group mon1.
777 int proc_resctrl_show(struct seq_file *s, struct pid_namespace *ns,
778 struct pid *pid, struct task_struct *tsk)
780 struct rdtgroup *rdtg;
783 mutex_lock(&rdtgroup_mutex);
785 /* Return empty if resctrl has not been mounted. */
786 if (!static_branch_unlikely(&rdt_enable_key)) {
787 seq_puts(s, "res:\nmon:\n");
791 list_for_each_entry(rdtg, &rdt_all_groups, rdtgroup_list) {
792 struct rdtgroup *crg;
795 * Task information is only relevant for shareable
796 * and exclusive groups.
798 if (rdtg->mode != RDT_MODE_SHAREABLE &&
799 rdtg->mode != RDT_MODE_EXCLUSIVE)
802 if (rdtg->closid != tsk->closid)
805 seq_printf(s, "res:%s%s\n", (rdtg == &rdtgroup_default) ? "/" : "",
808 list_for_each_entry(crg, &rdtg->mon.crdtgrp_list,
810 if (tsk->rmid != crg->mon.rmid)
812 seq_printf(s, "%s", crg->kn->name);
819 * The above search should succeed. Otherwise return
824 mutex_unlock(&rdtgroup_mutex);
830 static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
831 struct seq_file *seq, void *v)
835 mutex_lock(&rdtgroup_mutex);
836 len = seq_buf_used(&last_cmd_status);
838 seq_printf(seq, "%.*s", len, last_cmd_status_buf);
840 seq_puts(seq, "ok\n");
841 mutex_unlock(&rdtgroup_mutex);
845 static int rdt_num_closids_show(struct kernfs_open_file *of,
846 struct seq_file *seq, void *v)
848 struct rdt_resource *r = of->kn->parent->priv;
849 struct rdt_hw_resource *hw_res;
851 hw_res = resctrl_to_arch_res(r);
852 seq_printf(seq, "%d\n", hw_res->num_closid);
856 static int rdt_default_ctrl_show(struct kernfs_open_file *of,
857 struct seq_file *seq, void *v)
859 struct rdt_resource *r = of->kn->parent->priv;
861 seq_printf(seq, "%x\n", r->default_ctrl);
865 static int rdt_min_cbm_bits_show(struct kernfs_open_file *of,
866 struct seq_file *seq, void *v)
868 struct rdt_resource *r = of->kn->parent->priv;
870 seq_printf(seq, "%u\n", r->cache.min_cbm_bits);
874 static int rdt_shareable_bits_show(struct kernfs_open_file *of,
875 struct seq_file *seq, void *v)
877 struct rdt_resource *r = of->kn->parent->priv;
879 seq_printf(seq, "%x\n", r->cache.shareable_bits);
884 * rdt_bit_usage_show - Display current usage of resources
886 * A domain is a shared resource that can now be allocated differently. Here
887 * we display the current regions of the domain as an annotated bitmask.
888 * For each domain of this resource its allocation bitmask
889 * is annotated as below to indicate the current usage of the corresponding bit:
890 * 0 - currently unused
891 * X - currently available for sharing and used by software and hardware
892 * H - currently used by hardware only but available for software use
893 * S - currently used and shareable by software only
894 * E - currently used exclusively by one resource group
895 * P - currently pseudo-locked by one resource group
897 static int rdt_bit_usage_show(struct kernfs_open_file *of,
898 struct seq_file *seq, void *v)
900 struct rdt_resource *r = of->kn->parent->priv;
902 * Use unsigned long even though only 32 bits are used to ensure
903 * test_bit() is used safely.
905 unsigned long sw_shareable = 0, hw_shareable = 0;
906 unsigned long exclusive = 0, pseudo_locked = 0;
907 struct rdt_domain *dom;
908 int i, hwb, swb, excl, psl;
909 enum rdtgrp_mode mode;
913 mutex_lock(&rdtgroup_mutex);
914 hw_shareable = r->cache.shareable_bits;
915 list_for_each_entry(dom, &r->domains, list) {
918 ctrl = dom->ctrl_val;
921 seq_printf(seq, "%d=", dom->id);
922 for (i = 0; i < closids_supported(); i++, ctrl++) {
923 if (!closid_allocated(i))
925 mode = rdtgroup_mode_by_closid(i);
927 case RDT_MODE_SHAREABLE:
928 sw_shareable |= *ctrl;
930 case RDT_MODE_EXCLUSIVE:
933 case RDT_MODE_PSEUDO_LOCKSETUP:
935 * RDT_MODE_PSEUDO_LOCKSETUP is possible
936 * here but not included since the CBM
937 * associated with this CLOSID in this mode
938 * is not initialized and no task or cpu can be
939 * assigned this CLOSID.
942 case RDT_MODE_PSEUDO_LOCKED:
945 "invalid mode for closid %d\n", i);
949 for (i = r->cache.cbm_len - 1; i >= 0; i--) {
950 pseudo_locked = dom->plr ? dom->plr->cbm : 0;
951 hwb = test_bit(i, &hw_shareable);
952 swb = test_bit(i, &sw_shareable);
953 excl = test_bit(i, &exclusive);
954 psl = test_bit(i, &pseudo_locked);
957 else if (hwb && !swb)
959 else if (!hwb && swb)
965 else /* Unused bits remain */
971 mutex_unlock(&rdtgroup_mutex);
975 static int rdt_min_bw_show(struct kernfs_open_file *of,
976 struct seq_file *seq, void *v)
978 struct rdt_resource *r = of->kn->parent->priv;
980 seq_printf(seq, "%u\n", r->membw.min_bw);
984 static int rdt_num_rmids_show(struct kernfs_open_file *of,
985 struct seq_file *seq, void *v)
987 struct rdt_resource *r = of->kn->parent->priv;
989 seq_printf(seq, "%d\n", r->num_rmid);
994 static int rdt_mon_features_show(struct kernfs_open_file *of,
995 struct seq_file *seq, void *v)
997 struct rdt_resource *r = of->kn->parent->priv;
998 struct mon_evt *mevt;
1000 list_for_each_entry(mevt, &r->evt_list, list)
1001 seq_printf(seq, "%s\n", mevt->name);
1006 static int rdt_bw_gran_show(struct kernfs_open_file *of,
1007 struct seq_file *seq, void *v)
1009 struct rdt_resource *r = of->kn->parent->priv;
1011 seq_printf(seq, "%u\n", r->membw.bw_gran);
1015 static int rdt_delay_linear_show(struct kernfs_open_file *of,
1016 struct seq_file *seq, void *v)
1018 struct rdt_resource *r = of->kn->parent->priv;
1020 seq_printf(seq, "%u\n", r->membw.delay_linear);
1024 static int max_threshold_occ_show(struct kernfs_open_file *of,
1025 struct seq_file *seq, void *v)
1027 struct rdt_resource *r = of->kn->parent->priv;
1028 struct rdt_hw_resource *hw_res = resctrl_to_arch_res(r);
1030 seq_printf(seq, "%u\n", resctrl_cqm_threshold * hw_res->mon_scale);
1035 static int rdt_thread_throttle_mode_show(struct kernfs_open_file *of,
1036 struct seq_file *seq, void *v)
1038 struct rdt_resource *r = of->kn->parent->priv;
1040 if (r->membw.throttle_mode == THREAD_THROTTLE_PER_THREAD)
1041 seq_puts(seq, "per-thread\n");
1043 seq_puts(seq, "max\n");
1048 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
1049 char *buf, size_t nbytes, loff_t off)
1051 struct rdt_hw_resource *hw_res;
1055 ret = kstrtouint(buf, 0, &bytes);
1059 if (bytes > (boot_cpu_data.x86_cache_size * 1024))
1062 hw_res = resctrl_to_arch_res(of->kn->parent->priv);
1063 resctrl_cqm_threshold = bytes / hw_res->mon_scale;
1069 * rdtgroup_mode_show - Display mode of this resource group
1071 static int rdtgroup_mode_show(struct kernfs_open_file *of,
1072 struct seq_file *s, void *v)
1074 struct rdtgroup *rdtgrp;
1076 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1078 rdtgroup_kn_unlock(of->kn);
1082 seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode));
1084 rdtgroup_kn_unlock(of->kn);
1089 * rdt_cdp_peer_get - Retrieve CDP peer if it exists
1090 * @r: RDT resource to which RDT domain @d belongs
1091 * @d: Cache instance for which a CDP peer is requested
1092 * @r_cdp: RDT resource that shares hardware with @r (RDT resource peer)
1093 * Used to return the result.
1094 * @d_cdp: RDT domain that shares hardware with @d (RDT domain peer)
1095 * Used to return the result.
1097 * RDT resources are managed independently and by extension the RDT domains
1098 * (RDT resource instances) are managed independently also. The Code and
1099 * Data Prioritization (CDP) RDT resources, while managed independently,
1100 * could refer to the same underlying hardware. For example,
1101 * RDT_RESOURCE_L2CODE and RDT_RESOURCE_L2DATA both refer to the L2 cache.
1103 * When provided with an RDT resource @r and an instance of that RDT
1104 * resource @d rdt_cdp_peer_get() will return if there is a peer RDT
1105 * resource and the exact instance that shares the same hardware.
1107 * Return: 0 if a CDP peer was found, <0 on error or if no CDP peer exists.
1108 * If a CDP peer was found, @r_cdp will point to the peer RDT resource
1109 * and @d_cdp will point to the peer RDT domain.
1111 static int rdt_cdp_peer_get(struct rdt_resource *r, struct rdt_domain *d,
1112 struct rdt_resource **r_cdp,
1113 struct rdt_domain **d_cdp)
1115 struct rdt_resource *_r_cdp = NULL;
1116 struct rdt_domain *_d_cdp = NULL;
1120 case RDT_RESOURCE_L3DATA:
1121 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3CODE].r_resctrl;
1123 case RDT_RESOURCE_L3CODE:
1124 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3DATA].r_resctrl;
1126 case RDT_RESOURCE_L2DATA:
1127 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2CODE].r_resctrl;
1129 case RDT_RESOURCE_L2CODE:
1130 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2DATA].r_resctrl;
1138 * When a new CPU comes online and CDP is enabled then the new
1139 * RDT domains (if any) associated with both CDP RDT resources
1140 * are added in the same CPU online routine while the
1141 * rdtgroup_mutex is held. It should thus not happen for one
1142 * RDT domain to exist and be associated with its RDT CDP
1143 * resource but there is no RDT domain associated with the
1144 * peer RDT CDP resource. Hence the WARN.
1146 _d_cdp = rdt_find_domain(_r_cdp, d->id, NULL);
1147 if (WARN_ON(IS_ERR_OR_NULL(_d_cdp))) {
1161 * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other
1162 * @r: Resource to which domain instance @d belongs.
1163 * @d: The domain instance for which @closid is being tested.
1164 * @cbm: Capacity bitmask being tested.
1165 * @closid: Intended closid for @cbm.
1166 * @exclusive: Only check if overlaps with exclusive resource groups
1168 * Checks if provided @cbm intended to be used for @closid on domain
1169 * @d overlaps with any other closids or other hardware usage associated
1170 * with this domain. If @exclusive is true then only overlaps with
1171 * resource groups in exclusive mode will be considered. If @exclusive
1172 * is false then overlaps with any resource group or hardware entities
1173 * will be considered.
1175 * @cbm is unsigned long, even if only 32 bits are used, to make the
1176 * bitmap functions work correctly.
1178 * Return: false if CBM does not overlap, true if it does.
1180 static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1181 unsigned long cbm, int closid, bool exclusive)
1183 enum rdtgrp_mode mode;
1184 unsigned long ctrl_b;
1188 /* Check for any overlap with regions used by hardware directly */
1190 ctrl_b = r->cache.shareable_bits;
1191 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len))
1195 /* Check for overlap with other resource groups */
1197 for (i = 0; i < closids_supported(); i++, ctrl++) {
1199 mode = rdtgroup_mode_by_closid(i);
1200 if (closid_allocated(i) && i != closid &&
1201 mode != RDT_MODE_PSEUDO_LOCKSETUP) {
1202 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) {
1204 if (mode == RDT_MODE_EXCLUSIVE)
1217 * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware
1218 * @r: Resource to which domain instance @d belongs.
1219 * @d: The domain instance for which @closid is being tested.
1220 * @cbm: Capacity bitmask being tested.
1221 * @closid: Intended closid for @cbm.
1222 * @exclusive: Only check if overlaps with exclusive resource groups
1224 * Resources that can be allocated using a CBM can use the CBM to control
1225 * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test
1226 * for overlap. Overlap test is not limited to the specific resource for
1227 * which the CBM is intended though - when dealing with CDP resources that
1228 * share the underlying hardware the overlap check should be performed on
1229 * the CDP resource sharing the hardware also.
1231 * Refer to description of __rdtgroup_cbm_overlaps() for the details of the
1234 * Return: true if CBM overlap detected, false if there is no overlap
1236 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1237 unsigned long cbm, int closid, bool exclusive)
1239 struct rdt_resource *r_cdp;
1240 struct rdt_domain *d_cdp;
1242 if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, exclusive))
1245 if (rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp) < 0)
1248 return __rdtgroup_cbm_overlaps(r_cdp, d_cdp, cbm, closid, exclusive);
1252 * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive
1254 * An exclusive resource group implies that there should be no sharing of
1255 * its allocated resources. At the time this group is considered to be
1256 * exclusive this test can determine if its current schemata supports this
1257 * setting by testing for overlap with all other resource groups.
1259 * Return: true if resource group can be exclusive, false if there is overlap
1260 * with allocations of other resource groups and thus this resource group
1261 * cannot be exclusive.
1263 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp)
1265 int closid = rdtgrp->closid;
1266 struct rdt_resource *r;
1267 bool has_cache = false;
1268 struct rdt_domain *d;
1270 for_each_alloc_enabled_rdt_resource(r) {
1271 if (r->rid == RDT_RESOURCE_MBA)
1274 list_for_each_entry(d, &r->domains, list) {
1275 if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid],
1276 rdtgrp->closid, false)) {
1277 rdt_last_cmd_puts("Schemata overlaps\n");
1284 rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n");
1292 * rdtgroup_mode_write - Modify the resource group's mode
1295 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of,
1296 char *buf, size_t nbytes, loff_t off)
1298 struct rdtgroup *rdtgrp;
1299 enum rdtgrp_mode mode;
1302 /* Valid input requires a trailing newline */
1303 if (nbytes == 0 || buf[nbytes - 1] != '\n')
1305 buf[nbytes - 1] = '\0';
1307 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1309 rdtgroup_kn_unlock(of->kn);
1313 rdt_last_cmd_clear();
1315 mode = rdtgrp->mode;
1317 if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) ||
1318 (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) ||
1319 (!strcmp(buf, "pseudo-locksetup") &&
1320 mode == RDT_MODE_PSEUDO_LOCKSETUP) ||
1321 (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED))
1324 if (mode == RDT_MODE_PSEUDO_LOCKED) {
1325 rdt_last_cmd_puts("Cannot change pseudo-locked group\n");
1330 if (!strcmp(buf, "shareable")) {
1331 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1332 ret = rdtgroup_locksetup_exit(rdtgrp);
1336 rdtgrp->mode = RDT_MODE_SHAREABLE;
1337 } else if (!strcmp(buf, "exclusive")) {
1338 if (!rdtgroup_mode_test_exclusive(rdtgrp)) {
1342 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1343 ret = rdtgroup_locksetup_exit(rdtgrp);
1347 rdtgrp->mode = RDT_MODE_EXCLUSIVE;
1348 } else if (!strcmp(buf, "pseudo-locksetup")) {
1349 ret = rdtgroup_locksetup_enter(rdtgrp);
1352 rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP;
1354 rdt_last_cmd_puts("Unknown or unsupported mode\n");
1359 rdtgroup_kn_unlock(of->kn);
1360 return ret ?: nbytes;
1364 * rdtgroup_cbm_to_size - Translate CBM to size in bytes
1365 * @r: RDT resource to which @d belongs.
1366 * @d: RDT domain instance.
1367 * @cbm: bitmask for which the size should be computed.
1369 * The bitmask provided associated with the RDT domain instance @d will be
1370 * translated into how many bytes it represents. The size in bytes is
1371 * computed by first dividing the total cache size by the CBM length to
1372 * determine how many bytes each bit in the bitmask represents. The result
1373 * is multiplied with the number of bits set in the bitmask.
1375 * @cbm is unsigned long, even if only 32 bits are used to make the
1376 * bitmap functions work correctly.
1378 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r,
1379 struct rdt_domain *d, unsigned long cbm)
1381 struct cpu_cacheinfo *ci;
1382 unsigned int size = 0;
1385 num_b = bitmap_weight(&cbm, r->cache.cbm_len);
1386 ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask));
1387 for (i = 0; i < ci->num_leaves; i++) {
1388 if (ci->info_list[i].level == r->cache_level) {
1389 size = ci->info_list[i].size / r->cache.cbm_len * num_b;
1398 * rdtgroup_size_show - Display size in bytes of allocated regions
1400 * The "size" file mirrors the layout of the "schemata" file, printing the
1401 * size in bytes of each region instead of the capacity bitmask.
1404 static int rdtgroup_size_show(struct kernfs_open_file *of,
1405 struct seq_file *s, void *v)
1407 struct rdtgroup *rdtgrp;
1408 struct rdt_resource *r;
1409 struct rdt_domain *d;
1415 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1417 rdtgroup_kn_unlock(of->kn);
1421 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
1422 if (!rdtgrp->plr->d) {
1423 rdt_last_cmd_clear();
1424 rdt_last_cmd_puts("Cache domain offline\n");
1427 seq_printf(s, "%*s:", max_name_width,
1428 rdtgrp->plr->r->name);
1429 size = rdtgroup_cbm_to_size(rdtgrp->plr->r,
1432 seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size);
1437 for_each_alloc_enabled_rdt_resource(r) {
1439 seq_printf(s, "%*s:", max_name_width, r->name);
1440 list_for_each_entry(d, &r->domains, list) {
1443 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1446 ctrl = (!is_mba_sc(r) ?
1447 d->ctrl_val[rdtgrp->closid] :
1448 d->mbps_val[rdtgrp->closid]);
1449 if (r->rid == RDT_RESOURCE_MBA)
1452 size = rdtgroup_cbm_to_size(r, d, ctrl);
1454 seq_printf(s, "%d=%u", d->id, size);
1461 rdtgroup_kn_unlock(of->kn);
1466 /* rdtgroup information files for one cache resource. */
1467 static struct rftype res_common_files[] = {
1469 .name = "last_cmd_status",
1471 .kf_ops = &rdtgroup_kf_single_ops,
1472 .seq_show = rdt_last_cmd_status_show,
1473 .fflags = RF_TOP_INFO,
1476 .name = "num_closids",
1478 .kf_ops = &rdtgroup_kf_single_ops,
1479 .seq_show = rdt_num_closids_show,
1480 .fflags = RF_CTRL_INFO,
1483 .name = "mon_features",
1485 .kf_ops = &rdtgroup_kf_single_ops,
1486 .seq_show = rdt_mon_features_show,
1487 .fflags = RF_MON_INFO,
1490 .name = "num_rmids",
1492 .kf_ops = &rdtgroup_kf_single_ops,
1493 .seq_show = rdt_num_rmids_show,
1494 .fflags = RF_MON_INFO,
1499 .kf_ops = &rdtgroup_kf_single_ops,
1500 .seq_show = rdt_default_ctrl_show,
1501 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1504 .name = "min_cbm_bits",
1506 .kf_ops = &rdtgroup_kf_single_ops,
1507 .seq_show = rdt_min_cbm_bits_show,
1508 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1511 .name = "shareable_bits",
1513 .kf_ops = &rdtgroup_kf_single_ops,
1514 .seq_show = rdt_shareable_bits_show,
1515 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1518 .name = "bit_usage",
1520 .kf_ops = &rdtgroup_kf_single_ops,
1521 .seq_show = rdt_bit_usage_show,
1522 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1525 .name = "min_bandwidth",
1527 .kf_ops = &rdtgroup_kf_single_ops,
1528 .seq_show = rdt_min_bw_show,
1529 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1532 .name = "bandwidth_gran",
1534 .kf_ops = &rdtgroup_kf_single_ops,
1535 .seq_show = rdt_bw_gran_show,
1536 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1539 .name = "delay_linear",
1541 .kf_ops = &rdtgroup_kf_single_ops,
1542 .seq_show = rdt_delay_linear_show,
1543 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1546 * Platform specific which (if any) capabilities are provided by
1547 * thread_throttle_mode. Defer "fflags" initialization to platform
1551 .name = "thread_throttle_mode",
1553 .kf_ops = &rdtgroup_kf_single_ops,
1554 .seq_show = rdt_thread_throttle_mode_show,
1557 .name = "max_threshold_occupancy",
1559 .kf_ops = &rdtgroup_kf_single_ops,
1560 .write = max_threshold_occ_write,
1561 .seq_show = max_threshold_occ_show,
1562 .fflags = RF_MON_INFO | RFTYPE_RES_CACHE,
1567 .kf_ops = &rdtgroup_kf_single_ops,
1568 .write = rdtgroup_cpus_write,
1569 .seq_show = rdtgroup_cpus_show,
1570 .fflags = RFTYPE_BASE,
1573 .name = "cpus_list",
1575 .kf_ops = &rdtgroup_kf_single_ops,
1576 .write = rdtgroup_cpus_write,
1577 .seq_show = rdtgroup_cpus_show,
1578 .flags = RFTYPE_FLAGS_CPUS_LIST,
1579 .fflags = RFTYPE_BASE,
1584 .kf_ops = &rdtgroup_kf_single_ops,
1585 .write = rdtgroup_tasks_write,
1586 .seq_show = rdtgroup_tasks_show,
1587 .fflags = RFTYPE_BASE,
1592 .kf_ops = &rdtgroup_kf_single_ops,
1593 .write = rdtgroup_schemata_write,
1594 .seq_show = rdtgroup_schemata_show,
1595 .fflags = RF_CTRL_BASE,
1600 .kf_ops = &rdtgroup_kf_single_ops,
1601 .write = rdtgroup_mode_write,
1602 .seq_show = rdtgroup_mode_show,
1603 .fflags = RF_CTRL_BASE,
1608 .kf_ops = &rdtgroup_kf_single_ops,
1609 .seq_show = rdtgroup_size_show,
1610 .fflags = RF_CTRL_BASE,
1615 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
1617 struct rftype *rfts, *rft;
1620 rfts = res_common_files;
1621 len = ARRAY_SIZE(res_common_files);
1623 lockdep_assert_held(&rdtgroup_mutex);
1625 for (rft = rfts; rft < rfts + len; rft++) {
1626 if (rft->fflags && ((fflags & rft->fflags) == rft->fflags)) {
1627 ret = rdtgroup_add_file(kn, rft);
1635 pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
1636 while (--rft >= rfts) {
1637 if ((fflags & rft->fflags) == rft->fflags)
1638 kernfs_remove_by_name(kn, rft->name);
1643 static struct rftype *rdtgroup_get_rftype_by_name(const char *name)
1645 struct rftype *rfts, *rft;
1648 rfts = res_common_files;
1649 len = ARRAY_SIZE(res_common_files);
1651 for (rft = rfts; rft < rfts + len; rft++) {
1652 if (!strcmp(rft->name, name))
1659 void __init thread_throttle_mode_init(void)
1663 rft = rdtgroup_get_rftype_by_name("thread_throttle_mode");
1667 rft->fflags = RF_CTRL_INFO | RFTYPE_RES_MB;
1671 * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file
1672 * @r: The resource group with which the file is associated.
1673 * @name: Name of the file
1675 * The permissions of named resctrl file, directory, or link are modified
1676 * to not allow read, write, or execute by any user.
1678 * WARNING: This function is intended to communicate to the user that the
1679 * resctrl file has been locked down - that it is not relevant to the
1680 * particular state the system finds itself in. It should not be relied
1681 * on to protect from user access because after the file's permissions
1682 * are restricted the user can still change the permissions using chmod
1683 * from the command line.
1685 * Return: 0 on success, <0 on failure.
1687 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name)
1689 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1690 struct kernfs_node *kn;
1693 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1697 switch (kernfs_type(kn)) {
1699 iattr.ia_mode = S_IFDIR;
1702 iattr.ia_mode = S_IFREG;
1705 iattr.ia_mode = S_IFLNK;
1709 ret = kernfs_setattr(kn, &iattr);
1715 * rdtgroup_kn_mode_restore - Restore user access to named resctrl file
1716 * @r: The resource group with which the file is associated.
1717 * @name: Name of the file
1718 * @mask: Mask of permissions that should be restored
1720 * Restore the permissions of the named file. If @name is a directory the
1721 * permissions of its parent will be used.
1723 * Return: 0 on success, <0 on failure.
1725 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
1728 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1729 struct kernfs_node *kn, *parent;
1730 struct rftype *rfts, *rft;
1733 rfts = res_common_files;
1734 len = ARRAY_SIZE(res_common_files);
1736 for (rft = rfts; rft < rfts + len; rft++) {
1737 if (!strcmp(rft->name, name))
1738 iattr.ia_mode = rft->mode & mask;
1741 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1745 switch (kernfs_type(kn)) {
1747 parent = kernfs_get_parent(kn);
1749 iattr.ia_mode |= parent->mode;
1752 iattr.ia_mode |= S_IFDIR;
1755 iattr.ia_mode |= S_IFREG;
1758 iattr.ia_mode |= S_IFLNK;
1762 ret = kernfs_setattr(kn, &iattr);
1767 static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name,
1768 unsigned long fflags)
1770 struct kernfs_node *kn_subdir;
1773 kn_subdir = kernfs_create_dir(kn_info, name,
1775 if (IS_ERR(kn_subdir))
1776 return PTR_ERR(kn_subdir);
1778 ret = rdtgroup_kn_set_ugid(kn_subdir);
1782 ret = rdtgroup_add_files(kn_subdir, fflags);
1784 kernfs_activate(kn_subdir);
1789 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
1791 struct rdt_resource *r;
1792 unsigned long fflags;
1796 /* create the directory */
1797 kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
1798 if (IS_ERR(kn_info))
1799 return PTR_ERR(kn_info);
1801 ret = rdtgroup_add_files(kn_info, RF_TOP_INFO);
1805 for_each_alloc_enabled_rdt_resource(r) {
1806 fflags = r->fflags | RF_CTRL_INFO;
1807 ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags);
1812 for_each_mon_enabled_rdt_resource(r) {
1813 fflags = r->fflags | RF_MON_INFO;
1814 sprintf(name, "%s_MON", r->name);
1815 ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
1820 ret = rdtgroup_kn_set_ugid(kn_info);
1824 kernfs_activate(kn_info);
1829 kernfs_remove(kn_info);
1834 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
1835 char *name, struct kernfs_node **dest_kn)
1837 struct kernfs_node *kn;
1840 /* create the directory */
1841 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1848 ret = rdtgroup_kn_set_ugid(kn);
1852 kernfs_activate(kn);
1861 static void l3_qos_cfg_update(void *arg)
1865 wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL);
1868 static void l2_qos_cfg_update(void *arg)
1872 wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL);
1875 static inline bool is_mba_linear(void)
1877 return rdt_resources_all[RDT_RESOURCE_MBA].r_resctrl.membw.delay_linear;
1880 static int set_cache_qos_cfg(int level, bool enable)
1882 void (*update)(void *arg);
1883 struct rdt_resource *r_l;
1884 cpumask_var_t cpu_mask;
1885 struct rdt_domain *d;
1888 if (level == RDT_RESOURCE_L3)
1889 update = l3_qos_cfg_update;
1890 else if (level == RDT_RESOURCE_L2)
1891 update = l2_qos_cfg_update;
1895 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1898 r_l = &rdt_resources_all[level].r_resctrl;
1899 list_for_each_entry(d, &r_l->domains, list) {
1900 if (r_l->cache.arch_has_per_cpu_cfg)
1901 /* Pick all the CPUs in the domain instance */
1902 for_each_cpu(cpu, &d->cpu_mask)
1903 cpumask_set_cpu(cpu, cpu_mask);
1905 /* Pick one CPU from each domain instance to update MSR */
1906 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1909 /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1910 if (cpumask_test_cpu(cpu, cpu_mask))
1912 /* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1913 smp_call_function_many(cpu_mask, update, &enable, 1);
1916 free_cpumask_var(cpu_mask);
1921 /* Restore the qos cfg state when a domain comes online */
1922 void rdt_domain_reconfigure_cdp(struct rdt_resource *r)
1924 if (!r->alloc_capable)
1927 if (r == &rdt_resources_all[RDT_RESOURCE_L2DATA].r_resctrl)
1928 l2_qos_cfg_update(&r->alloc_enabled);
1930 if (r == &rdt_resources_all[RDT_RESOURCE_L3DATA].r_resctrl)
1931 l3_qos_cfg_update(&r->alloc_enabled);
1935 * Enable or disable the MBA software controller
1936 * which helps user specify bandwidth in MBps.
1937 * MBA software controller is supported only if
1938 * MBM is supported and MBA is in linear scale.
1940 static int set_mba_sc(bool mba_sc)
1942 struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA].r_resctrl;
1943 struct rdt_domain *d;
1945 if (!is_mbm_enabled() || !is_mba_linear() ||
1946 mba_sc == is_mba_sc(r))
1949 r->membw.mba_sc = mba_sc;
1950 list_for_each_entry(d, &r->domains, list)
1951 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val);
1956 static int cdp_enable(int level, int data_type, int code_type)
1958 struct rdt_resource *r_ldata = &rdt_resources_all[data_type].r_resctrl;
1959 struct rdt_resource *r_lcode = &rdt_resources_all[code_type].r_resctrl;
1960 struct rdt_resource *r_l = &rdt_resources_all[level].r_resctrl;
1963 if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1964 !r_lcode->alloc_capable)
1967 ret = set_cache_qos_cfg(level, true);
1969 r_l->alloc_enabled = false;
1970 r_ldata->alloc_enabled = true;
1971 r_lcode->alloc_enabled = true;
1976 static int cdpl3_enable(void)
1978 return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1979 RDT_RESOURCE_L3CODE);
1982 static int cdpl2_enable(void)
1984 return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1985 RDT_RESOURCE_L2CODE);
1988 static void cdp_disable(int level, int data_type, int code_type)
1990 struct rdt_resource *r = &rdt_resources_all[level].r_resctrl;
1992 r->alloc_enabled = r->alloc_capable;
1994 if (rdt_resources_all[data_type].r_resctrl.alloc_enabled) {
1995 rdt_resources_all[data_type].r_resctrl.alloc_enabled = false;
1996 rdt_resources_all[code_type].r_resctrl.alloc_enabled = false;
1997 set_cache_qos_cfg(level, false);
2001 static void cdpl3_disable(void)
2003 cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
2006 static void cdpl2_disable(void)
2008 cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
2011 static void cdp_disable_all(void)
2013 if (rdt_resources_all[RDT_RESOURCE_L3DATA].r_resctrl.alloc_enabled)
2015 if (rdt_resources_all[RDT_RESOURCE_L2DATA].r_resctrl.alloc_enabled)
2020 * We don't allow rdtgroup directories to be created anywhere
2021 * except the root directory. Thus when looking for the rdtgroup
2022 * structure for a kernfs node we are either looking at a directory,
2023 * in which case the rdtgroup structure is pointed at by the "priv"
2024 * field, otherwise we have a file, and need only look to the parent
2025 * to find the rdtgroup.
2027 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
2029 if (kernfs_type(kn) == KERNFS_DIR) {
2031 * All the resource directories use "kn->priv"
2032 * to point to the "struct rdtgroup" for the
2033 * resource. "info" and its subdirectories don't
2034 * have rdtgroup structures, so return NULL here.
2036 if (kn == kn_info || kn->parent == kn_info)
2041 return kn->parent->priv;
2045 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
2047 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
2052 atomic_inc(&rdtgrp->waitcount);
2053 kernfs_break_active_protection(kn);
2055 mutex_lock(&rdtgroup_mutex);
2057 /* Was this group deleted while we waited? */
2058 if (rdtgrp->flags & RDT_DELETED)
2064 void rdtgroup_kn_unlock(struct kernfs_node *kn)
2066 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
2071 mutex_unlock(&rdtgroup_mutex);
2073 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
2074 (rdtgrp->flags & RDT_DELETED)) {
2075 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2076 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2077 rdtgroup_pseudo_lock_remove(rdtgrp);
2078 kernfs_unbreak_active_protection(kn);
2079 rdtgroup_remove(rdtgrp);
2081 kernfs_unbreak_active_protection(kn);
2085 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2086 struct rdtgroup *prgrp,
2087 struct kernfs_node **mon_data_kn);
2089 static int rdt_enable_ctx(struct rdt_fs_context *ctx)
2093 if (ctx->enable_cdpl2)
2094 ret = cdpl2_enable();
2096 if (!ret && ctx->enable_cdpl3)
2097 ret = cdpl3_enable();
2099 if (!ret && ctx->enable_mba_mbps)
2100 ret = set_mba_sc(true);
2105 static int rdt_get_tree(struct fs_context *fc)
2107 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2108 struct rdt_domain *dom;
2109 struct rdt_resource *r;
2113 mutex_lock(&rdtgroup_mutex);
2115 * resctrl file system can only be mounted once.
2117 if (static_branch_unlikely(&rdt_enable_key)) {
2122 ret = rdt_enable_ctx(ctx);
2128 ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
2132 if (rdt_mon_capable) {
2133 ret = mongroup_create_dir(rdtgroup_default.kn,
2134 &rdtgroup_default, "mon_groups",
2139 ret = mkdir_mondata_all(rdtgroup_default.kn,
2140 &rdtgroup_default, &kn_mondata);
2143 rdtgroup_default.mon.mon_data_kn = kn_mondata;
2146 ret = rdt_pseudo_lock_init();
2150 ret = kernfs_get_tree(fc);
2154 if (rdt_alloc_capable)
2155 static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
2156 if (rdt_mon_capable)
2157 static_branch_enable_cpuslocked(&rdt_mon_enable_key);
2159 if (rdt_alloc_capable || rdt_mon_capable)
2160 static_branch_enable_cpuslocked(&rdt_enable_key);
2162 if (is_mbm_enabled()) {
2163 r = &rdt_resources_all[RDT_RESOURCE_L3].r_resctrl;
2164 list_for_each_entry(dom, &r->domains, list)
2165 mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
2171 rdt_pseudo_lock_release();
2173 if (rdt_mon_capable)
2174 kernfs_remove(kn_mondata);
2176 if (rdt_mon_capable)
2177 kernfs_remove(kn_mongrp);
2179 kernfs_remove(kn_info);
2181 if (ctx->enable_mba_mbps)
2186 rdt_last_cmd_clear();
2187 mutex_unlock(&rdtgroup_mutex);
2199 static const struct fs_parameter_spec rdt_fs_parameters[] = {
2200 fsparam_flag("cdp", Opt_cdp),
2201 fsparam_flag("cdpl2", Opt_cdpl2),
2202 fsparam_flag("mba_MBps", Opt_mba_mbps),
2206 static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param)
2208 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2209 struct fs_parse_result result;
2212 opt = fs_parse(fc, rdt_fs_parameters, param, &result);
2218 ctx->enable_cdpl3 = true;
2221 ctx->enable_cdpl2 = true;
2224 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2226 ctx->enable_mba_mbps = true;
2233 static void rdt_fs_context_free(struct fs_context *fc)
2235 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2237 kernfs_free_fs_context(fc);
2241 static const struct fs_context_operations rdt_fs_context_ops = {
2242 .free = rdt_fs_context_free,
2243 .parse_param = rdt_parse_param,
2244 .get_tree = rdt_get_tree,
2247 static int rdt_init_fs_context(struct fs_context *fc)
2249 struct rdt_fs_context *ctx;
2251 ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL);
2255 ctx->kfc.root = rdt_root;
2256 ctx->kfc.magic = RDTGROUP_SUPER_MAGIC;
2257 fc->fs_private = &ctx->kfc;
2258 fc->ops = &rdt_fs_context_ops;
2259 put_user_ns(fc->user_ns);
2260 fc->user_ns = get_user_ns(&init_user_ns);
2265 static int reset_all_ctrls(struct rdt_resource *r)
2267 struct rdt_hw_resource *hw_res = resctrl_to_arch_res(r);
2268 struct msr_param msr_param;
2269 cpumask_var_t cpu_mask;
2270 struct rdt_domain *d;
2273 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
2278 msr_param.high = hw_res->num_closid;
2281 * Disable resource control for this resource by setting all
2282 * CBMs in all domains to the maximum mask value. Pick one CPU
2283 * from each domain to update the MSRs below.
2285 list_for_each_entry(d, &r->domains, list) {
2286 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
2288 for (i = 0; i < hw_res->num_closid; i++)
2289 d->ctrl_val[i] = r->default_ctrl;
2292 /* Update CBM on this cpu if it's in cpu_mask. */
2293 if (cpumask_test_cpu(cpu, cpu_mask))
2294 rdt_ctrl_update(&msr_param);
2295 /* Update CBM on all other cpus in cpu_mask. */
2296 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
2299 free_cpumask_var(cpu_mask);
2305 * Move tasks from one to the other group. If @from is NULL, then all tasks
2306 * in the systems are moved unconditionally (used for teardown).
2308 * If @mask is not NULL the cpus on which moved tasks are running are set
2309 * in that mask so the update smp function call is restricted to affected
2312 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
2313 struct cpumask *mask)
2315 struct task_struct *p, *t;
2317 read_lock(&tasklist_lock);
2318 for_each_process_thread(p, t) {
2319 if (!from || is_closid_match(t, from) ||
2320 is_rmid_match(t, from)) {
2321 WRITE_ONCE(t->closid, to->closid);
2322 WRITE_ONCE(t->rmid, to->mon.rmid);
2325 * If the task is on a CPU, set the CPU in the mask.
2326 * The detection is inaccurate as tasks might move or
2327 * schedule before the smp function call takes place.
2328 * In such a case the function call is pointless, but
2329 * there is no other side effect.
2331 if (IS_ENABLED(CONFIG_SMP) && mask && task_curr(t))
2332 cpumask_set_cpu(task_cpu(t), mask);
2335 read_unlock(&tasklist_lock);
2338 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
2340 struct rdtgroup *sentry, *stmp;
2341 struct list_head *head;
2343 head = &rdtgrp->mon.crdtgrp_list;
2344 list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
2345 free_rmid(sentry->mon.rmid);
2346 list_del(&sentry->mon.crdtgrp_list);
2348 if (atomic_read(&sentry->waitcount) != 0)
2349 sentry->flags = RDT_DELETED;
2351 rdtgroup_remove(sentry);
2356 * Forcibly remove all of subdirectories under root.
2358 static void rmdir_all_sub(void)
2360 struct rdtgroup *rdtgrp, *tmp;
2362 /* Move all tasks to the default resource group */
2363 rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
2365 list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
2366 /* Free any child rmids */
2367 free_all_child_rdtgrp(rdtgrp);
2369 /* Remove each rdtgroup other than root */
2370 if (rdtgrp == &rdtgroup_default)
2373 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2374 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2375 rdtgroup_pseudo_lock_remove(rdtgrp);
2378 * Give any CPUs back to the default group. We cannot copy
2379 * cpu_online_mask because a CPU might have executed the
2380 * offline callback already, but is still marked online.
2382 cpumask_or(&rdtgroup_default.cpu_mask,
2383 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2385 free_rmid(rdtgrp->mon.rmid);
2387 kernfs_remove(rdtgrp->kn);
2388 list_del(&rdtgrp->rdtgroup_list);
2390 if (atomic_read(&rdtgrp->waitcount) != 0)
2391 rdtgrp->flags = RDT_DELETED;
2393 rdtgroup_remove(rdtgrp);
2395 /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
2396 update_closid_rmid(cpu_online_mask, &rdtgroup_default);
2398 kernfs_remove(kn_info);
2399 kernfs_remove(kn_mongrp);
2400 kernfs_remove(kn_mondata);
2403 static void rdt_kill_sb(struct super_block *sb)
2405 struct rdt_resource *r;
2408 mutex_lock(&rdtgroup_mutex);
2412 /*Put everything back to default values. */
2413 for_each_alloc_enabled_rdt_resource(r)
2417 rdt_pseudo_lock_release();
2418 rdtgroup_default.mode = RDT_MODE_SHAREABLE;
2419 static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
2420 static_branch_disable_cpuslocked(&rdt_mon_enable_key);
2421 static_branch_disable_cpuslocked(&rdt_enable_key);
2423 mutex_unlock(&rdtgroup_mutex);
2427 static struct file_system_type rdt_fs_type = {
2429 .init_fs_context = rdt_init_fs_context,
2430 .parameters = rdt_fs_parameters,
2431 .kill_sb = rdt_kill_sb,
2434 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
2437 struct kernfs_node *kn;
2440 kn = __kernfs_create_file(parent_kn, name, 0444,
2441 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
2442 &kf_mondata_ops, priv, NULL, NULL);
2446 ret = rdtgroup_kn_set_ugid(kn);
2456 * Remove all subdirectories of mon_data of ctrl_mon groups
2457 * and monitor groups with given domain id.
2459 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
2461 struct rdtgroup *prgrp, *crgrp;
2464 if (!r->mon_enabled)
2467 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2468 sprintf(name, "mon_%s_%02d", r->name, dom_id);
2469 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
2471 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
2472 kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
2476 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
2477 struct rdt_domain *d,
2478 struct rdt_resource *r, struct rdtgroup *prgrp)
2480 union mon_data_bits priv;
2481 struct kernfs_node *kn;
2482 struct mon_evt *mevt;
2483 struct rmid_read rr;
2487 sprintf(name, "mon_%s_%02d", r->name, d->id);
2488 /* create the directory */
2489 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
2493 ret = rdtgroup_kn_set_ugid(kn);
2497 if (WARN_ON(list_empty(&r->evt_list))) {
2502 priv.u.rid = r->rid;
2503 priv.u.domid = d->id;
2504 list_for_each_entry(mevt, &r->evt_list, list) {
2505 priv.u.evtid = mevt->evtid;
2506 ret = mon_addfile(kn, mevt->name, priv.priv);
2510 if (is_mbm_event(mevt->evtid))
2511 mon_event_read(&rr, r, d, prgrp, mevt->evtid, true);
2513 kernfs_activate(kn);
2522 * Add all subdirectories of mon_data for "ctrl_mon" groups
2523 * and "monitor" groups with given domain id.
2525 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
2526 struct rdt_domain *d)
2528 struct kernfs_node *parent_kn;
2529 struct rdtgroup *prgrp, *crgrp;
2530 struct list_head *head;
2532 if (!r->mon_enabled)
2535 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2536 parent_kn = prgrp->mon.mon_data_kn;
2537 mkdir_mondata_subdir(parent_kn, d, r, prgrp);
2539 head = &prgrp->mon.crdtgrp_list;
2540 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
2541 parent_kn = crgrp->mon.mon_data_kn;
2542 mkdir_mondata_subdir(parent_kn, d, r, crgrp);
2547 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
2548 struct rdt_resource *r,
2549 struct rdtgroup *prgrp)
2551 struct rdt_domain *dom;
2554 list_for_each_entry(dom, &r->domains, list) {
2555 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
2564 * This creates a directory mon_data which contains the monitored data.
2566 * mon_data has one directory for each domain which are named
2567 * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
2568 * with L3 domain looks as below:
2575 * Each domain directory has one file per event:
2580 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2581 struct rdtgroup *prgrp,
2582 struct kernfs_node **dest_kn)
2584 struct rdt_resource *r;
2585 struct kernfs_node *kn;
2589 * Create the mon_data directory first.
2591 ret = mongroup_create_dir(parent_kn, prgrp, "mon_data", &kn);
2599 * Create the subdirectories for each domain. Note that all events
2600 * in a domain like L3 are grouped into a resource whose domain is L3
2602 for_each_mon_enabled_rdt_resource(r) {
2603 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
2616 * cbm_ensure_valid - Enforce validity on provided CBM
2617 * @_val: Candidate CBM
2618 * @r: RDT resource to which the CBM belongs
2620 * The provided CBM represents all cache portions available for use. This
2621 * may be represented by a bitmap that does not consist of contiguous ones
2622 * and thus be an invalid CBM.
2623 * Here the provided CBM is forced to be a valid CBM by only considering
2624 * the first set of contiguous bits as valid and clearing all bits.
2625 * The intention here is to provide a valid default CBM with which a new
2626 * resource group is initialized. The user can follow this with a
2627 * modification to the CBM if the default does not satisfy the
2630 static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r)
2632 unsigned int cbm_len = r->cache.cbm_len;
2633 unsigned long first_bit, zero_bit;
2634 unsigned long val = _val;
2639 first_bit = find_first_bit(&val, cbm_len);
2640 zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
2642 /* Clear any remaining bits to ensure contiguous region */
2643 bitmap_clear(&val, zero_bit, cbm_len - zero_bit);
2648 * Initialize cache resources per RDT domain
2650 * Set the RDT domain up to start off with all usable allocations. That is,
2651 * all shareable and unused bits. All-zero CBM is invalid.
2653 static int __init_one_rdt_domain(struct rdt_domain *d, struct rdt_resource *r,
2656 struct rdt_resource *r_cdp = NULL;
2657 struct rdt_domain *d_cdp = NULL;
2658 u32 used_b = 0, unused_b = 0;
2659 unsigned long tmp_cbm;
2660 enum rdtgrp_mode mode;
2661 u32 peer_ctl, *ctrl;
2664 rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp);
2665 d->have_new_ctrl = false;
2666 d->new_ctrl = r->cache.shareable_bits;
2667 used_b = r->cache.shareable_bits;
2669 for (i = 0; i < closids_supported(); i++, ctrl++) {
2670 if (closid_allocated(i) && i != closid) {
2671 mode = rdtgroup_mode_by_closid(i);
2672 if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
2674 * ctrl values for locksetup aren't relevant
2675 * until the schemata is written, and the mode
2676 * becomes RDT_MODE_PSEUDO_LOCKED.
2680 * If CDP is active include peer domain's
2681 * usage to ensure there is no overlap
2682 * with an exclusive group.
2685 peer_ctl = d_cdp->ctrl_val[i];
2688 used_b |= *ctrl | peer_ctl;
2689 if (mode == RDT_MODE_SHAREABLE)
2690 d->new_ctrl |= *ctrl | peer_ctl;
2693 if (d->plr && d->plr->cbm > 0)
2694 used_b |= d->plr->cbm;
2695 unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
2696 unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
2697 d->new_ctrl |= unused_b;
2699 * Force the initial CBM to be valid, user can
2700 * modify the CBM based on system availability.
2702 d->new_ctrl = cbm_ensure_valid(d->new_ctrl, r);
2704 * Assign the u32 CBM to an unsigned long to ensure that
2705 * bitmap_weight() does not access out-of-bound memory.
2707 tmp_cbm = d->new_ctrl;
2708 if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) {
2709 rdt_last_cmd_printf("No space on %s:%d\n", r->name, d->id);
2712 d->have_new_ctrl = true;
2718 * Initialize cache resources with default values.
2720 * A new RDT group is being created on an allocation capable (CAT)
2721 * supporting system. Set this group up to start off with all usable
2724 * If there are no more shareable bits available on any domain then
2725 * the entire allocation will fail.
2727 static int rdtgroup_init_cat(struct rdt_resource *r, u32 closid)
2729 struct rdt_domain *d;
2732 list_for_each_entry(d, &r->domains, list) {
2733 ret = __init_one_rdt_domain(d, r, closid);
2741 /* Initialize MBA resource with default values. */
2742 static void rdtgroup_init_mba(struct rdt_resource *r)
2744 struct rdt_domain *d;
2746 list_for_each_entry(d, &r->domains, list) {
2747 d->new_ctrl = is_mba_sc(r) ? MBA_MAX_MBPS : r->default_ctrl;
2748 d->have_new_ctrl = true;
2752 /* Initialize the RDT group's allocations. */
2753 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
2755 struct rdt_resource *r;
2758 for_each_alloc_enabled_rdt_resource(r) {
2759 if (r->rid == RDT_RESOURCE_MBA) {
2760 rdtgroup_init_mba(r);
2762 ret = rdtgroup_init_cat(r, rdtgrp->closid);
2767 ret = update_domains(r, rdtgrp->closid);
2769 rdt_last_cmd_puts("Failed to initialize allocations\n");
2775 rdtgrp->mode = RDT_MODE_SHAREABLE;
2780 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
2781 const char *name, umode_t mode,
2782 enum rdt_group_type rtype, struct rdtgroup **r)
2784 struct rdtgroup *prdtgrp, *rdtgrp;
2785 struct kernfs_node *kn;
2789 prdtgrp = rdtgroup_kn_lock_live(parent_kn);
2795 if (rtype == RDTMON_GROUP &&
2796 (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2797 prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
2799 rdt_last_cmd_puts("Pseudo-locking in progress\n");
2803 /* allocate the rdtgroup. */
2804 rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
2807 rdt_last_cmd_puts("Kernel out of memory\n");
2811 rdtgrp->mon.parent = prdtgrp;
2812 rdtgrp->type = rtype;
2813 INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
2815 /* kernfs creates the directory for rdtgrp */
2816 kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
2819 rdt_last_cmd_puts("kernfs create error\n");
2825 * kernfs_remove() will drop the reference count on "kn" which
2826 * will free it. But we still need it to stick around for the
2827 * rdtgroup_kn_unlock(kn) call. Take one extra reference here,
2828 * which will be dropped by kernfs_put() in rdtgroup_remove().
2832 ret = rdtgroup_kn_set_ugid(kn);
2834 rdt_last_cmd_puts("kernfs perm error\n");
2838 files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
2839 ret = rdtgroup_add_files(kn, files);
2841 rdt_last_cmd_puts("kernfs fill error\n");
2845 if (rdt_mon_capable) {
2848 rdt_last_cmd_puts("Out of RMIDs\n");
2851 rdtgrp->mon.rmid = ret;
2853 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
2855 rdt_last_cmd_puts("kernfs subdir error\n");
2859 kernfs_activate(kn);
2862 * The caller unlocks the parent_kn upon success.
2867 free_rmid(rdtgrp->mon.rmid);
2869 kernfs_put(rdtgrp->kn);
2870 kernfs_remove(rdtgrp->kn);
2874 rdtgroup_kn_unlock(parent_kn);
2878 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
2880 kernfs_remove(rgrp->kn);
2881 free_rmid(rgrp->mon.rmid);
2882 rdtgroup_remove(rgrp);
2886 * Create a monitor group under "mon_groups" directory of a control
2887 * and monitor group(ctrl_mon). This is a resource group
2888 * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
2890 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
2891 const char *name, umode_t mode)
2893 struct rdtgroup *rdtgrp, *prgrp;
2896 ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTMON_GROUP, &rdtgrp);
2900 prgrp = rdtgrp->mon.parent;
2901 rdtgrp->closid = prgrp->closid;
2904 * Add the rdtgrp to the list of rdtgrps the parent
2905 * ctrl_mon group has to track.
2907 list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
2909 rdtgroup_kn_unlock(parent_kn);
2914 * These are rdtgroups created under the root directory. Can be used
2915 * to allocate and monitor resources.
2917 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
2918 const char *name, umode_t mode)
2920 struct rdtgroup *rdtgrp;
2921 struct kernfs_node *kn;
2925 ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTCTRL_GROUP, &rdtgrp);
2930 ret = closid_alloc();
2932 rdt_last_cmd_puts("Out of CLOSIDs\n");
2933 goto out_common_fail;
2938 rdtgrp->closid = closid;
2939 ret = rdtgroup_init_alloc(rdtgrp);
2943 list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
2945 if (rdt_mon_capable) {
2947 * Create an empty mon_groups directory to hold the subset
2948 * of tasks and cpus to monitor.
2950 ret = mongroup_create_dir(kn, rdtgrp, "mon_groups", NULL);
2952 rdt_last_cmd_puts("kernfs subdir error\n");
2960 list_del(&rdtgrp->rdtgroup_list);
2962 closid_free(closid);
2964 mkdir_rdt_prepare_clean(rdtgrp);
2966 rdtgroup_kn_unlock(parent_kn);
2971 * We allow creating mon groups only with in a directory called "mon_groups"
2972 * which is present in every ctrl_mon group. Check if this is a valid
2973 * "mon_groups" directory.
2975 * 1. The directory should be named "mon_groups".
2976 * 2. The mon group itself should "not" be named "mon_groups".
2977 * This makes sure "mon_groups" directory always has a ctrl_mon group
2980 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
2982 return (!strcmp(kn->name, "mon_groups") &&
2983 strcmp(name, "mon_groups"));
2986 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
2989 /* Do not accept '\n' to avoid unparsable situation. */
2990 if (strchr(name, '\n'))
2994 * If the parent directory is the root directory and RDT
2995 * allocation is supported, add a control and monitoring
2998 if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
2999 return rdtgroup_mkdir_ctrl_mon(parent_kn, name, mode);
3002 * If RDT monitoring is supported and the parent directory is a valid
3003 * "mon_groups" directory, add a monitoring subdirectory.
3005 if (rdt_mon_capable && is_mon_groups(parent_kn, name))
3006 return rdtgroup_mkdir_mon(parent_kn, name, mode);
3011 static int rdtgroup_rmdir_mon(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask)
3013 struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
3016 /* Give any tasks back to the parent group */
3017 rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
3019 /* Update per cpu rmid of the moved CPUs first */
3020 for_each_cpu(cpu, &rdtgrp->cpu_mask)
3021 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
3023 * Update the MSR on moved CPUs and CPUs which have moved
3024 * task running on them.
3026 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
3027 update_closid_rmid(tmpmask, NULL);
3029 rdtgrp->flags = RDT_DELETED;
3030 free_rmid(rdtgrp->mon.rmid);
3033 * Remove the rdtgrp from the parent ctrl_mon group's list
3035 WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
3036 list_del(&rdtgrp->mon.crdtgrp_list);
3038 kernfs_remove(rdtgrp->kn);
3043 static int rdtgroup_ctrl_remove(struct rdtgroup *rdtgrp)
3045 rdtgrp->flags = RDT_DELETED;
3046 list_del(&rdtgrp->rdtgroup_list);
3048 kernfs_remove(rdtgrp->kn);
3052 static int rdtgroup_rmdir_ctrl(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask)
3056 /* Give any tasks back to the default group */
3057 rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
3059 /* Give any CPUs back to the default group */
3060 cpumask_or(&rdtgroup_default.cpu_mask,
3061 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
3063 /* Update per cpu closid and rmid of the moved CPUs first */
3064 for_each_cpu(cpu, &rdtgrp->cpu_mask) {
3065 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
3066 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
3070 * Update the MSR on moved CPUs and CPUs which have moved
3071 * task running on them.
3073 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
3074 update_closid_rmid(tmpmask, NULL);
3076 closid_free(rdtgrp->closid);
3077 free_rmid(rdtgrp->mon.rmid);
3079 rdtgroup_ctrl_remove(rdtgrp);
3082 * Free all the child monitor group rmids.
3084 free_all_child_rdtgrp(rdtgrp);
3089 static int rdtgroup_rmdir(struct kernfs_node *kn)
3091 struct kernfs_node *parent_kn = kn->parent;
3092 struct rdtgroup *rdtgrp;
3093 cpumask_var_t tmpmask;
3096 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
3099 rdtgrp = rdtgroup_kn_lock_live(kn);
3106 * If the rdtgroup is a ctrl_mon group and parent directory
3107 * is the root directory, remove the ctrl_mon group.
3109 * If the rdtgroup is a mon group and parent directory
3110 * is a valid "mon_groups" directory, remove the mon group.
3112 if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn &&
3113 rdtgrp != &rdtgroup_default) {
3114 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
3115 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
3116 ret = rdtgroup_ctrl_remove(rdtgrp);
3118 ret = rdtgroup_rmdir_ctrl(rdtgrp, tmpmask);
3120 } else if (rdtgrp->type == RDTMON_GROUP &&
3121 is_mon_groups(parent_kn, kn->name)) {
3122 ret = rdtgroup_rmdir_mon(rdtgrp, tmpmask);
3128 rdtgroup_kn_unlock(kn);
3129 free_cpumask_var(tmpmask);
3133 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
3135 if (rdt_resources_all[RDT_RESOURCE_L3DATA].r_resctrl.alloc_enabled)
3136 seq_puts(seq, ",cdp");
3138 if (rdt_resources_all[RDT_RESOURCE_L2DATA].r_resctrl.alloc_enabled)
3139 seq_puts(seq, ",cdpl2");
3141 if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA].r_resctrl))
3142 seq_puts(seq, ",mba_MBps");
3147 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
3148 .mkdir = rdtgroup_mkdir,
3149 .rmdir = rdtgroup_rmdir,
3150 .show_options = rdtgroup_show_options,
3153 static int __init rdtgroup_setup_root(void)
3157 rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
3158 KERNFS_ROOT_CREATE_DEACTIVATED |
3159 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
3161 if (IS_ERR(rdt_root))
3162 return PTR_ERR(rdt_root);
3164 mutex_lock(&rdtgroup_mutex);
3166 rdtgroup_default.closid = 0;
3167 rdtgroup_default.mon.rmid = 0;
3168 rdtgroup_default.type = RDTCTRL_GROUP;
3169 INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
3171 list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
3173 ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
3175 kernfs_destroy_root(rdt_root);
3179 rdtgroup_default.kn = rdt_root->kn;
3180 kernfs_activate(rdtgroup_default.kn);
3183 mutex_unlock(&rdtgroup_mutex);
3189 * rdtgroup_init - rdtgroup initialization
3191 * Setup resctrl file system including set up root, create mount point,
3192 * register rdtgroup filesystem, and initialize files under root directory.
3194 * Return: 0 on success or -errno
3196 int __init rdtgroup_init(void)
3200 seq_buf_init(&last_cmd_status, last_cmd_status_buf,
3201 sizeof(last_cmd_status_buf));
3203 ret = rdtgroup_setup_root();
3207 ret = sysfs_create_mount_point(fs_kobj, "resctrl");
3211 ret = register_filesystem(&rdt_fs_type);
3213 goto cleanup_mountpoint;
3216 * Adding the resctrl debugfs directory here may not be ideal since
3217 * it would let the resctrl debugfs directory appear on the debugfs
3218 * filesystem before the resctrl filesystem is mounted.
3219 * It may also be ok since that would enable debugging of RDT before
3220 * resctrl is mounted.
3221 * The reason why the debugfs directory is created here and not in
3222 * rdt_get_tree() is because rdt_get_tree() takes rdtgroup_mutex and
3223 * during the debugfs directory creation also &sb->s_type->i_mutex_key
3224 * (the lockdep class of inode->i_rwsem). Other filesystem
3225 * interactions (eg. SyS_getdents) have the lock ordering:
3226 * &sb->s_type->i_mutex_key --> &mm->mmap_lock
3227 * During mmap(), called with &mm->mmap_lock, the rdtgroup_mutex
3228 * is taken, thus creating dependency:
3229 * &mm->mmap_lock --> rdtgroup_mutex for the latter that can cause
3230 * issues considering the other two lock dependencies.
3231 * By creating the debugfs directory here we avoid a dependency
3232 * that may cause deadlock (even though file operations cannot
3233 * occur until the filesystem is mounted, but I do not know how to
3234 * tell lockdep that).
3236 debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
3241 sysfs_remove_mount_point(fs_kobj, "resctrl");
3243 kernfs_destroy_root(rdt_root);
3248 void __exit rdtgroup_exit(void)
3250 debugfs_remove_recursive(debugfs_resctrl);
3251 unregister_filesystem(&rdt_fs_type);
3252 sysfs_remove_mount_point(fs_kobj, "resctrl");
3253 kernfs_destroy_root(rdt_root);