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_resource *r;
104 int rdt_min_closid = 32;
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);
110 closid_free_map = BIT_MASK(rdt_min_closid) - 1;
112 /* CLOSID 0 is always reserved for the default group */
113 closid_free_map &= ~1;
114 closid_free_map_len = rdt_min_closid;
117 static int closid_alloc(void)
119 u32 closid = ffs(closid_free_map);
124 closid_free_map &= ~(1 << closid);
129 void closid_free(int closid)
131 closid_free_map |= 1 << closid;
135 * closid_allocated - test if provided closid is in use
136 * @closid: closid to be tested
138 * Return: true if @closid is currently associated with a resource group,
139 * false if @closid is free
141 static bool closid_allocated(unsigned int closid)
143 return (closid_free_map & (1 << closid)) == 0;
147 * rdtgroup_mode_by_closid - Return mode of resource group with closid
148 * @closid: closid if the resource group
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.
153 * Return: mode as &enum rdtgrp_mode of resource group with closid @closid
155 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid)
157 struct rdtgroup *rdtgrp;
159 list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
160 if (rdtgrp->closid == closid)
164 return RDT_NUM_MODES;
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",
175 * rdtgroup_mode_str - Return the string representation of mode
176 * @mode: the resource group mode as &enum rdtgroup_mode
178 * Return: string representation of valid mode, "unknown" otherwise
180 static const char *rdtgroup_mode_str(enum rdtgrp_mode mode)
182 if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES)
185 return rdt_mode_str[mode];
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)
191 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
192 .ia_uid = current_fsuid(),
193 .ia_gid = current_fsgid(), };
195 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
196 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
199 return kernfs_setattr(kn, &iattr);
202 static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft)
204 struct kernfs_node *kn;
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);
213 ret = rdtgroup_kn_set_ugid(kn);
222 static int rdtgroup_seqfile_show(struct seq_file *m, void *arg)
224 struct kernfs_open_file *of = m->private;
225 struct rftype *rft = of->kn->priv;
228 return rft->seq_show(of, m, arg);
232 static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf,
233 size_t nbytes, loff_t off)
235 struct rftype *rft = of->kn->priv;
238 return rft->write(of, buf, nbytes, off);
243 static const struct kernfs_ops rdtgroup_kf_single_ops = {
244 .atomic_write_len = PAGE_SIZE,
245 .write = rdtgroup_file_write,
246 .seq_show = rdtgroup_seqfile_show,
249 static const struct kernfs_ops kf_mondata_ops = {
250 .atomic_write_len = PAGE_SIZE,
251 .seq_show = rdtgroup_mondata_show,
254 static bool is_cpu_list(struct kernfs_open_file *of)
256 struct rftype *rft = of->kn->priv;
258 return rft->flags & RFTYPE_FLAGS_CPUS_LIST;
261 static int rdtgroup_cpus_show(struct kernfs_open_file *of,
262 struct seq_file *s, void *v)
264 struct rdtgroup *rdtgrp;
265 struct cpumask *mask;
268 rdtgrp = rdtgroup_kn_lock_live(of->kn);
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");
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));
283 seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n",
284 cpumask_pr_args(&rdtgrp->cpu_mask));
289 rdtgroup_kn_unlock(of->kn);
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 protected against __switch_to() because
298 * preemption is disabled.
300 static void update_cpu_closid_rmid(void *info)
302 struct rdtgroup *r = info;
305 this_cpu_write(pqr_state.default_closid, r->closid);
306 this_cpu_write(pqr_state.default_rmid, r->mon.rmid);
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.
318 * Update the PGR_ASSOC MSR on all cpus in @cpu_mask,
320 * Per task closids/rmids must have been set up before calling this function.
323 update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r)
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);
333 static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
334 cpumask_var_t tmpmask)
336 struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp;
337 struct list_head *head;
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");
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);
355 * If we added cpus, remove them from previous group that owned them
356 * and update per-cpu rmid
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) {
364 cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask,
367 update_closid_rmid(tmpmask, rdtgrp);
370 /* Done pushing/pulling - update this group with new mask */
371 cpumask_copy(&rdtgrp->cpu_mask, newmask);
376 static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m)
378 struct rdtgroup *crgrp;
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);
386 static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
387 cpumask_var_t tmpmask, cpumask_var_t tmpmask1)
389 struct rdtgroup *r, *crgrp;
390 struct list_head *head;
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");
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);
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.
412 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
413 if (cpumask_weight(tmpmask)) {
414 list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) {
417 cpumask_and(tmpmask1, &r->cpu_mask, tmpmask);
418 if (cpumask_weight(tmpmask1))
419 cpumask_rdtgrp_clear(r, tmpmask1);
421 update_closid_rmid(tmpmask, rdtgrp);
424 /* Done pushing/pulling - update this group with new mask */
425 cpumask_copy(&rdtgrp->cpu_mask, newmask);
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.
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);
441 static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of,
442 char *buf, size_t nbytes, loff_t off)
444 cpumask_var_t tmpmask, newmask, tmpmask1;
445 struct rdtgroup *rdtgrp;
451 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
453 if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) {
454 free_cpumask_var(tmpmask);
457 if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) {
458 free_cpumask_var(tmpmask);
459 free_cpumask_var(newmask);
463 rdtgrp = rdtgroup_kn_lock_live(of->kn);
469 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
470 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
472 rdt_last_cmd_puts("Pseudo-locking in progress\n");
477 ret = cpulist_parse(buf, newmask);
479 ret = cpumask_parse(buf, newmask);
482 rdt_last_cmd_puts("Bad CPU list/mask\n");
486 /* check that user didn't specify any offline cpus */
487 cpumask_andnot(tmpmask, newmask, cpu_online_mask);
488 if (cpumask_weight(tmpmask)) {
490 rdt_last_cmd_puts("Can only assign online CPUs\n");
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);
502 rdtgroup_kn_unlock(of->kn);
503 free_cpumask_var(tmpmask);
504 free_cpumask_var(newmask);
505 free_cpumask_var(tmpmask1);
507 return ret ?: nbytes;
511 * rdtgroup_remove - the helper to remove resource group safely
512 * @rdtgrp: resource group to remove
514 * On resource group creation via a mkdir, an extra kernfs_node reference is
515 * taken to ensure that the rdtgroup structure remains accessible for the
516 * rdtgroup_kn_unlock() calls where it is removed.
518 * Drop the extra reference here, then free the rdtgroup structure.
522 static void rdtgroup_remove(struct rdtgroup *rdtgrp)
524 kernfs_put(rdtgrp->kn);
528 static void _update_task_closid_rmid(void *task)
531 * If the task is still current on this CPU, update PQR_ASSOC MSR.
532 * Otherwise, the MSR is updated when the task is scheduled in.
538 static void update_task_closid_rmid(struct task_struct *t)
540 if (IS_ENABLED(CONFIG_SMP) && task_curr(t))
541 smp_call_function_single(task_cpu(t), _update_task_closid_rmid, t, 1);
543 _update_task_closid_rmid(t);
546 static int __rdtgroup_move_task(struct task_struct *tsk,
547 struct rdtgroup *rdtgrp)
549 /* If the task is already in rdtgrp, no need to move the task. */
550 if ((rdtgrp->type == RDTCTRL_GROUP && tsk->closid == rdtgrp->closid &&
551 tsk->rmid == rdtgrp->mon.rmid) ||
552 (rdtgrp->type == RDTMON_GROUP && tsk->rmid == rdtgrp->mon.rmid &&
553 tsk->closid == rdtgrp->mon.parent->closid))
557 * Set the task's closid/rmid before the PQR_ASSOC MSR can be
560 * For ctrl_mon groups, move both closid and rmid.
561 * For monitor groups, can move the tasks only from
562 * their parent CTRL group.
565 if (rdtgrp->type == RDTCTRL_GROUP) {
566 WRITE_ONCE(tsk->closid, rdtgrp->closid);
567 WRITE_ONCE(tsk->rmid, rdtgrp->mon.rmid);
568 } else if (rdtgrp->type == RDTMON_GROUP) {
569 if (rdtgrp->mon.parent->closid == tsk->closid) {
570 WRITE_ONCE(tsk->rmid, rdtgrp->mon.rmid);
572 rdt_last_cmd_puts("Can't move task to different control group\n");
578 * Ensure the task's closid and rmid are written before determining if
579 * the task is current that will decide if it will be interrupted.
584 * By now, the task's closid and rmid are set. If the task is current
585 * on a CPU, the PQR_ASSOC MSR needs to be updated to make the resource
586 * group go into effect. If the task is not current, the MSR will be
587 * updated when the task is scheduled in.
589 update_task_closid_rmid(tsk);
594 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
596 return (rdt_alloc_capable &&
597 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
600 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
602 return (rdt_mon_capable &&
603 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
607 * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group
610 * Return: 1 if tasks have been assigned to @r, 0 otherwise
612 int rdtgroup_tasks_assigned(struct rdtgroup *r)
614 struct task_struct *p, *t;
617 lockdep_assert_held(&rdtgroup_mutex);
620 for_each_process_thread(p, t) {
621 if (is_closid_match(t, r) || is_rmid_match(t, r)) {
631 static int rdtgroup_task_write_permission(struct task_struct *task,
632 struct kernfs_open_file *of)
634 const struct cred *tcred = get_task_cred(task);
635 const struct cred *cred = current_cred();
639 * Even if we're attaching all tasks in the thread group, we only
640 * need to check permissions on one of them.
642 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
643 !uid_eq(cred->euid, tcred->uid) &&
644 !uid_eq(cred->euid, tcred->suid)) {
645 rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
653 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
654 struct kernfs_open_file *of)
656 struct task_struct *tsk;
661 tsk = find_task_by_vpid(pid);
664 rdt_last_cmd_printf("No task %d\n", pid);
671 get_task_struct(tsk);
674 ret = rdtgroup_task_write_permission(tsk, of);
676 ret = __rdtgroup_move_task(tsk, rdtgrp);
678 put_task_struct(tsk);
682 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
683 char *buf, size_t nbytes, loff_t off)
685 struct rdtgroup *rdtgrp;
689 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
691 rdtgrp = rdtgroup_kn_lock_live(of->kn);
693 rdtgroup_kn_unlock(of->kn);
696 rdt_last_cmd_clear();
698 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
699 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
701 rdt_last_cmd_puts("Pseudo-locking in progress\n");
705 ret = rdtgroup_move_task(pid, rdtgrp, of);
708 rdtgroup_kn_unlock(of->kn);
710 return ret ?: nbytes;
713 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
715 struct task_struct *p, *t;
718 for_each_process_thread(p, t) {
719 if (is_closid_match(t, r) || is_rmid_match(t, r))
720 seq_printf(s, "%d\n", t->pid);
725 static int rdtgroup_tasks_show(struct kernfs_open_file *of,
726 struct seq_file *s, void *v)
728 struct rdtgroup *rdtgrp;
731 rdtgrp = rdtgroup_kn_lock_live(of->kn);
733 show_rdt_tasks(rdtgrp, s);
736 rdtgroup_kn_unlock(of->kn);
741 #ifdef CONFIG_PROC_CPU_RESCTRL
744 * A task can only be part of one resctrl control group and of one monitor
745 * group which is associated to that control group.
750 * resctrl is not available.
755 * Task is part of the root resctrl control group, and it is not associated
756 * to any monitor group.
761 * Task is part of the root resctrl control group and monitor group mon0.
766 * Task is part of resctrl control group group0, and it is not associated
767 * to any monitor group.
772 * Task is part of resctrl control group group0 and monitor group mon1.
774 int proc_resctrl_show(struct seq_file *s, struct pid_namespace *ns,
775 struct pid *pid, struct task_struct *tsk)
777 struct rdtgroup *rdtg;
780 mutex_lock(&rdtgroup_mutex);
782 /* Return empty if resctrl has not been mounted. */
783 if (!static_branch_unlikely(&rdt_enable_key)) {
784 seq_puts(s, "res:\nmon:\n");
788 list_for_each_entry(rdtg, &rdt_all_groups, rdtgroup_list) {
789 struct rdtgroup *crg;
792 * Task information is only relevant for shareable
793 * and exclusive groups.
795 if (rdtg->mode != RDT_MODE_SHAREABLE &&
796 rdtg->mode != RDT_MODE_EXCLUSIVE)
799 if (rdtg->closid != tsk->closid)
802 seq_printf(s, "res:%s%s\n", (rdtg == &rdtgroup_default) ? "/" : "",
805 list_for_each_entry(crg, &rdtg->mon.crdtgrp_list,
807 if (tsk->rmid != crg->mon.rmid)
809 seq_printf(s, "%s", crg->kn->name);
816 * The above search should succeed. Otherwise return
821 mutex_unlock(&rdtgroup_mutex);
827 static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
828 struct seq_file *seq, void *v)
832 mutex_lock(&rdtgroup_mutex);
833 len = seq_buf_used(&last_cmd_status);
835 seq_printf(seq, "%.*s", len, last_cmd_status_buf);
837 seq_puts(seq, "ok\n");
838 mutex_unlock(&rdtgroup_mutex);
842 static int rdt_num_closids_show(struct kernfs_open_file *of,
843 struct seq_file *seq, void *v)
845 struct rdt_resource *r = of->kn->parent->priv;
847 seq_printf(seq, "%d\n", r->num_closid);
851 static int rdt_default_ctrl_show(struct kernfs_open_file *of,
852 struct seq_file *seq, void *v)
854 struct rdt_resource *r = of->kn->parent->priv;
856 seq_printf(seq, "%x\n", r->default_ctrl);
860 static int rdt_min_cbm_bits_show(struct kernfs_open_file *of,
861 struct seq_file *seq, void *v)
863 struct rdt_resource *r = of->kn->parent->priv;
865 seq_printf(seq, "%u\n", r->cache.min_cbm_bits);
869 static int rdt_shareable_bits_show(struct kernfs_open_file *of,
870 struct seq_file *seq, void *v)
872 struct rdt_resource *r = of->kn->parent->priv;
874 seq_printf(seq, "%x\n", r->cache.shareable_bits);
879 * rdt_bit_usage_show - Display current usage of resources
881 * A domain is a shared resource that can now be allocated differently. Here
882 * we display the current regions of the domain as an annotated bitmask.
883 * For each domain of this resource its allocation bitmask
884 * is annotated as below to indicate the current usage of the corresponding bit:
885 * 0 - currently unused
886 * X - currently available for sharing and used by software and hardware
887 * H - currently used by hardware only but available for software use
888 * S - currently used and shareable by software only
889 * E - currently used exclusively by one resource group
890 * P - currently pseudo-locked by one resource group
892 static int rdt_bit_usage_show(struct kernfs_open_file *of,
893 struct seq_file *seq, void *v)
895 struct rdt_resource *r = of->kn->parent->priv;
897 * Use unsigned long even though only 32 bits are used to ensure
898 * test_bit() is used safely.
900 unsigned long sw_shareable = 0, hw_shareable = 0;
901 unsigned long exclusive = 0, pseudo_locked = 0;
902 struct rdt_domain *dom;
903 int i, hwb, swb, excl, psl;
904 enum rdtgrp_mode mode;
908 mutex_lock(&rdtgroup_mutex);
909 hw_shareable = r->cache.shareable_bits;
910 list_for_each_entry(dom, &r->domains, list) {
913 ctrl = dom->ctrl_val;
916 seq_printf(seq, "%d=", dom->id);
917 for (i = 0; i < closids_supported(); i++, ctrl++) {
918 if (!closid_allocated(i))
920 mode = rdtgroup_mode_by_closid(i);
922 case RDT_MODE_SHAREABLE:
923 sw_shareable |= *ctrl;
925 case RDT_MODE_EXCLUSIVE:
928 case RDT_MODE_PSEUDO_LOCKSETUP:
930 * RDT_MODE_PSEUDO_LOCKSETUP is possible
931 * here but not included since the CBM
932 * associated with this CLOSID in this mode
933 * is not initialized and no task or cpu can be
934 * assigned this CLOSID.
937 case RDT_MODE_PSEUDO_LOCKED:
940 "invalid mode for closid %d\n", i);
944 for (i = r->cache.cbm_len - 1; i >= 0; i--) {
945 pseudo_locked = dom->plr ? dom->plr->cbm : 0;
946 hwb = test_bit(i, &hw_shareable);
947 swb = test_bit(i, &sw_shareable);
948 excl = test_bit(i, &exclusive);
949 psl = test_bit(i, &pseudo_locked);
952 else if (hwb && !swb)
954 else if (!hwb && swb)
960 else /* Unused bits remain */
966 mutex_unlock(&rdtgroup_mutex);
970 static int rdt_min_bw_show(struct kernfs_open_file *of,
971 struct seq_file *seq, void *v)
973 struct rdt_resource *r = of->kn->parent->priv;
975 seq_printf(seq, "%u\n", r->membw.min_bw);
979 static int rdt_num_rmids_show(struct kernfs_open_file *of,
980 struct seq_file *seq, void *v)
982 struct rdt_resource *r = of->kn->parent->priv;
984 seq_printf(seq, "%d\n", r->num_rmid);
989 static int rdt_mon_features_show(struct kernfs_open_file *of,
990 struct seq_file *seq, void *v)
992 struct rdt_resource *r = of->kn->parent->priv;
993 struct mon_evt *mevt;
995 list_for_each_entry(mevt, &r->evt_list, list)
996 seq_printf(seq, "%s\n", mevt->name);
1001 static int rdt_bw_gran_show(struct kernfs_open_file *of,
1002 struct seq_file *seq, void *v)
1004 struct rdt_resource *r = of->kn->parent->priv;
1006 seq_printf(seq, "%u\n", r->membw.bw_gran);
1010 static int rdt_delay_linear_show(struct kernfs_open_file *of,
1011 struct seq_file *seq, void *v)
1013 struct rdt_resource *r = of->kn->parent->priv;
1015 seq_printf(seq, "%u\n", r->membw.delay_linear);
1019 static int max_threshold_occ_show(struct kernfs_open_file *of,
1020 struct seq_file *seq, void *v)
1022 struct rdt_resource *r = of->kn->parent->priv;
1024 seq_printf(seq, "%u\n", resctrl_cqm_threshold * r->mon_scale);
1029 static int rdt_thread_throttle_mode_show(struct kernfs_open_file *of,
1030 struct seq_file *seq, void *v)
1032 struct rdt_resource *r = of->kn->parent->priv;
1034 if (r->membw.throttle_mode == THREAD_THROTTLE_PER_THREAD)
1035 seq_puts(seq, "per-thread\n");
1037 seq_puts(seq, "max\n");
1042 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
1043 char *buf, size_t nbytes, loff_t off)
1045 struct rdt_resource *r = of->kn->parent->priv;
1049 ret = kstrtouint(buf, 0, &bytes);
1053 if (bytes > (boot_cpu_data.x86_cache_size * 1024))
1056 resctrl_cqm_threshold = bytes / r->mon_scale;
1062 * rdtgroup_mode_show - Display mode of this resource group
1064 static int rdtgroup_mode_show(struct kernfs_open_file *of,
1065 struct seq_file *s, void *v)
1067 struct rdtgroup *rdtgrp;
1069 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1071 rdtgroup_kn_unlock(of->kn);
1075 seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode));
1077 rdtgroup_kn_unlock(of->kn);
1082 * rdt_cdp_peer_get - Retrieve CDP peer if it exists
1083 * @r: RDT resource to which RDT domain @d belongs
1084 * @d: Cache instance for which a CDP peer is requested
1085 * @r_cdp: RDT resource that shares hardware with @r (RDT resource peer)
1086 * Used to return the result.
1087 * @d_cdp: RDT domain that shares hardware with @d (RDT domain peer)
1088 * Used to return the result.
1090 * RDT resources are managed independently and by extension the RDT domains
1091 * (RDT resource instances) are managed independently also. The Code and
1092 * Data Prioritization (CDP) RDT resources, while managed independently,
1093 * could refer to the same underlying hardware. For example,
1094 * RDT_RESOURCE_L2CODE and RDT_RESOURCE_L2DATA both refer to the L2 cache.
1096 * When provided with an RDT resource @r and an instance of that RDT
1097 * resource @d rdt_cdp_peer_get() will return if there is a peer RDT
1098 * resource and the exact instance that shares the same hardware.
1100 * Return: 0 if a CDP peer was found, <0 on error or if no CDP peer exists.
1101 * If a CDP peer was found, @r_cdp will point to the peer RDT resource
1102 * and @d_cdp will point to the peer RDT domain.
1104 static int rdt_cdp_peer_get(struct rdt_resource *r, struct rdt_domain *d,
1105 struct rdt_resource **r_cdp,
1106 struct rdt_domain **d_cdp)
1108 struct rdt_resource *_r_cdp = NULL;
1109 struct rdt_domain *_d_cdp = NULL;
1113 case RDT_RESOURCE_L3DATA:
1114 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3CODE];
1116 case RDT_RESOURCE_L3CODE:
1117 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3DATA];
1119 case RDT_RESOURCE_L2DATA:
1120 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2CODE];
1122 case RDT_RESOURCE_L2CODE:
1123 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2DATA];
1131 * When a new CPU comes online and CDP is enabled then the new
1132 * RDT domains (if any) associated with both CDP RDT resources
1133 * are added in the same CPU online routine while the
1134 * rdtgroup_mutex is held. It should thus not happen for one
1135 * RDT domain to exist and be associated with its RDT CDP
1136 * resource but there is no RDT domain associated with the
1137 * peer RDT CDP resource. Hence the WARN.
1139 _d_cdp = rdt_find_domain(_r_cdp, d->id, NULL);
1140 if (WARN_ON(IS_ERR_OR_NULL(_d_cdp))) {
1154 * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other
1155 * @r: Resource to which domain instance @d belongs.
1156 * @d: The domain instance for which @closid is being tested.
1157 * @cbm: Capacity bitmask being tested.
1158 * @closid: Intended closid for @cbm.
1159 * @exclusive: Only check if overlaps with exclusive resource groups
1161 * Checks if provided @cbm intended to be used for @closid on domain
1162 * @d overlaps with any other closids or other hardware usage associated
1163 * with this domain. If @exclusive is true then only overlaps with
1164 * resource groups in exclusive mode will be considered. If @exclusive
1165 * is false then overlaps with any resource group or hardware entities
1166 * will be considered.
1168 * @cbm is unsigned long, even if only 32 bits are used, to make the
1169 * bitmap functions work correctly.
1171 * Return: false if CBM does not overlap, true if it does.
1173 static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1174 unsigned long cbm, int closid, bool exclusive)
1176 enum rdtgrp_mode mode;
1177 unsigned long ctrl_b;
1181 /* Check for any overlap with regions used by hardware directly */
1183 ctrl_b = r->cache.shareable_bits;
1184 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len))
1188 /* Check for overlap with other resource groups */
1190 for (i = 0; i < closids_supported(); i++, ctrl++) {
1192 mode = rdtgroup_mode_by_closid(i);
1193 if (closid_allocated(i) && i != closid &&
1194 mode != RDT_MODE_PSEUDO_LOCKSETUP) {
1195 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) {
1197 if (mode == RDT_MODE_EXCLUSIVE)
1210 * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware
1211 * @r: Resource to which domain instance @d belongs.
1212 * @d: The domain instance for which @closid is being tested.
1213 * @cbm: Capacity bitmask being tested.
1214 * @closid: Intended closid for @cbm.
1215 * @exclusive: Only check if overlaps with exclusive resource groups
1217 * Resources that can be allocated using a CBM can use the CBM to control
1218 * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test
1219 * for overlap. Overlap test is not limited to the specific resource for
1220 * which the CBM is intended though - when dealing with CDP resources that
1221 * share the underlying hardware the overlap check should be performed on
1222 * the CDP resource sharing the hardware also.
1224 * Refer to description of __rdtgroup_cbm_overlaps() for the details of the
1227 * Return: true if CBM overlap detected, false if there is no overlap
1229 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1230 unsigned long cbm, int closid, bool exclusive)
1232 struct rdt_resource *r_cdp;
1233 struct rdt_domain *d_cdp;
1235 if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, exclusive))
1238 if (rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp) < 0)
1241 return __rdtgroup_cbm_overlaps(r_cdp, d_cdp, cbm, closid, exclusive);
1245 * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive
1247 * An exclusive resource group implies that there should be no sharing of
1248 * its allocated resources. At the time this group is considered to be
1249 * exclusive this test can determine if its current schemata supports this
1250 * setting by testing for overlap with all other resource groups.
1252 * Return: true if resource group can be exclusive, false if there is overlap
1253 * with allocations of other resource groups and thus this resource group
1254 * cannot be exclusive.
1256 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp)
1258 int closid = rdtgrp->closid;
1259 struct rdt_resource *r;
1260 bool has_cache = false;
1261 struct rdt_domain *d;
1263 for_each_alloc_enabled_rdt_resource(r) {
1264 if (r->rid == RDT_RESOURCE_MBA)
1267 list_for_each_entry(d, &r->domains, list) {
1268 if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid],
1269 rdtgrp->closid, false)) {
1270 rdt_last_cmd_puts("Schemata overlaps\n");
1277 rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n");
1285 * rdtgroup_mode_write - Modify the resource group's mode
1288 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of,
1289 char *buf, size_t nbytes, loff_t off)
1291 struct rdtgroup *rdtgrp;
1292 enum rdtgrp_mode mode;
1295 /* Valid input requires a trailing newline */
1296 if (nbytes == 0 || buf[nbytes - 1] != '\n')
1298 buf[nbytes - 1] = '\0';
1300 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1302 rdtgroup_kn_unlock(of->kn);
1306 rdt_last_cmd_clear();
1308 mode = rdtgrp->mode;
1310 if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) ||
1311 (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) ||
1312 (!strcmp(buf, "pseudo-locksetup") &&
1313 mode == RDT_MODE_PSEUDO_LOCKSETUP) ||
1314 (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED))
1317 if (mode == RDT_MODE_PSEUDO_LOCKED) {
1318 rdt_last_cmd_puts("Cannot change pseudo-locked group\n");
1323 if (!strcmp(buf, "shareable")) {
1324 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1325 ret = rdtgroup_locksetup_exit(rdtgrp);
1329 rdtgrp->mode = RDT_MODE_SHAREABLE;
1330 } else if (!strcmp(buf, "exclusive")) {
1331 if (!rdtgroup_mode_test_exclusive(rdtgrp)) {
1335 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1336 ret = rdtgroup_locksetup_exit(rdtgrp);
1340 rdtgrp->mode = RDT_MODE_EXCLUSIVE;
1341 } else if (!strcmp(buf, "pseudo-locksetup")) {
1342 ret = rdtgroup_locksetup_enter(rdtgrp);
1345 rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP;
1347 rdt_last_cmd_puts("Unknown or unsupported mode\n");
1352 rdtgroup_kn_unlock(of->kn);
1353 return ret ?: nbytes;
1357 * rdtgroup_cbm_to_size - Translate CBM to size in bytes
1358 * @r: RDT resource to which @d belongs.
1359 * @d: RDT domain instance.
1360 * @cbm: bitmask for which the size should be computed.
1362 * The bitmask provided associated with the RDT domain instance @d will be
1363 * translated into how many bytes it represents. The size in bytes is
1364 * computed by first dividing the total cache size by the CBM length to
1365 * determine how many bytes each bit in the bitmask represents. The result
1366 * is multiplied with the number of bits set in the bitmask.
1368 * @cbm is unsigned long, even if only 32 bits are used to make the
1369 * bitmap functions work correctly.
1371 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r,
1372 struct rdt_domain *d, unsigned long cbm)
1374 struct cpu_cacheinfo *ci;
1375 unsigned int size = 0;
1378 num_b = bitmap_weight(&cbm, r->cache.cbm_len);
1379 ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask));
1380 for (i = 0; i < ci->num_leaves; i++) {
1381 if (ci->info_list[i].level == r->cache_level) {
1382 size = ci->info_list[i].size / r->cache.cbm_len * num_b;
1391 * rdtgroup_size_show - Display size in bytes of allocated regions
1393 * The "size" file mirrors the layout of the "schemata" file, printing the
1394 * size in bytes of each region instead of the capacity bitmask.
1397 static int rdtgroup_size_show(struct kernfs_open_file *of,
1398 struct seq_file *s, void *v)
1400 struct rdtgroup *rdtgrp;
1401 struct rdt_resource *r;
1402 struct rdt_domain *d;
1408 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1410 rdtgroup_kn_unlock(of->kn);
1414 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
1415 if (!rdtgrp->plr->d) {
1416 rdt_last_cmd_clear();
1417 rdt_last_cmd_puts("Cache domain offline\n");
1420 seq_printf(s, "%*s:", max_name_width,
1421 rdtgrp->plr->r->name);
1422 size = rdtgroup_cbm_to_size(rdtgrp->plr->r,
1425 seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size);
1430 for_each_alloc_enabled_rdt_resource(r) {
1432 seq_printf(s, "%*s:", max_name_width, r->name);
1433 list_for_each_entry(d, &r->domains, list) {
1436 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1439 ctrl = (!is_mba_sc(r) ?
1440 d->ctrl_val[rdtgrp->closid] :
1441 d->mbps_val[rdtgrp->closid]);
1442 if (r->rid == RDT_RESOURCE_MBA)
1445 size = rdtgroup_cbm_to_size(r, d, ctrl);
1447 seq_printf(s, "%d=%u", d->id, size);
1454 rdtgroup_kn_unlock(of->kn);
1459 /* rdtgroup information files for one cache resource. */
1460 static struct rftype res_common_files[] = {
1462 .name = "last_cmd_status",
1464 .kf_ops = &rdtgroup_kf_single_ops,
1465 .seq_show = rdt_last_cmd_status_show,
1466 .fflags = RF_TOP_INFO,
1469 .name = "num_closids",
1471 .kf_ops = &rdtgroup_kf_single_ops,
1472 .seq_show = rdt_num_closids_show,
1473 .fflags = RF_CTRL_INFO,
1476 .name = "mon_features",
1478 .kf_ops = &rdtgroup_kf_single_ops,
1479 .seq_show = rdt_mon_features_show,
1480 .fflags = RF_MON_INFO,
1483 .name = "num_rmids",
1485 .kf_ops = &rdtgroup_kf_single_ops,
1486 .seq_show = rdt_num_rmids_show,
1487 .fflags = RF_MON_INFO,
1492 .kf_ops = &rdtgroup_kf_single_ops,
1493 .seq_show = rdt_default_ctrl_show,
1494 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1497 .name = "min_cbm_bits",
1499 .kf_ops = &rdtgroup_kf_single_ops,
1500 .seq_show = rdt_min_cbm_bits_show,
1501 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1504 .name = "shareable_bits",
1506 .kf_ops = &rdtgroup_kf_single_ops,
1507 .seq_show = rdt_shareable_bits_show,
1508 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1511 .name = "bit_usage",
1513 .kf_ops = &rdtgroup_kf_single_ops,
1514 .seq_show = rdt_bit_usage_show,
1515 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1518 .name = "min_bandwidth",
1520 .kf_ops = &rdtgroup_kf_single_ops,
1521 .seq_show = rdt_min_bw_show,
1522 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1525 .name = "bandwidth_gran",
1527 .kf_ops = &rdtgroup_kf_single_ops,
1528 .seq_show = rdt_bw_gran_show,
1529 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1532 .name = "delay_linear",
1534 .kf_ops = &rdtgroup_kf_single_ops,
1535 .seq_show = rdt_delay_linear_show,
1536 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1539 * Platform specific which (if any) capabilities are provided by
1540 * thread_throttle_mode. Defer "fflags" initialization to platform
1544 .name = "thread_throttle_mode",
1546 .kf_ops = &rdtgroup_kf_single_ops,
1547 .seq_show = rdt_thread_throttle_mode_show,
1550 .name = "max_threshold_occupancy",
1552 .kf_ops = &rdtgroup_kf_single_ops,
1553 .write = max_threshold_occ_write,
1554 .seq_show = max_threshold_occ_show,
1555 .fflags = RF_MON_INFO | RFTYPE_RES_CACHE,
1560 .kf_ops = &rdtgroup_kf_single_ops,
1561 .write = rdtgroup_cpus_write,
1562 .seq_show = rdtgroup_cpus_show,
1563 .fflags = RFTYPE_BASE,
1566 .name = "cpus_list",
1568 .kf_ops = &rdtgroup_kf_single_ops,
1569 .write = rdtgroup_cpus_write,
1570 .seq_show = rdtgroup_cpus_show,
1571 .flags = RFTYPE_FLAGS_CPUS_LIST,
1572 .fflags = RFTYPE_BASE,
1577 .kf_ops = &rdtgroup_kf_single_ops,
1578 .write = rdtgroup_tasks_write,
1579 .seq_show = rdtgroup_tasks_show,
1580 .fflags = RFTYPE_BASE,
1585 .kf_ops = &rdtgroup_kf_single_ops,
1586 .write = rdtgroup_schemata_write,
1587 .seq_show = rdtgroup_schemata_show,
1588 .fflags = RF_CTRL_BASE,
1593 .kf_ops = &rdtgroup_kf_single_ops,
1594 .write = rdtgroup_mode_write,
1595 .seq_show = rdtgroup_mode_show,
1596 .fflags = RF_CTRL_BASE,
1601 .kf_ops = &rdtgroup_kf_single_ops,
1602 .seq_show = rdtgroup_size_show,
1603 .fflags = RF_CTRL_BASE,
1608 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
1610 struct rftype *rfts, *rft;
1613 rfts = res_common_files;
1614 len = ARRAY_SIZE(res_common_files);
1616 lockdep_assert_held(&rdtgroup_mutex);
1618 for (rft = rfts; rft < rfts + len; rft++) {
1619 if (rft->fflags && ((fflags & rft->fflags) == rft->fflags)) {
1620 ret = rdtgroup_add_file(kn, rft);
1628 pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
1629 while (--rft >= rfts) {
1630 if ((fflags & rft->fflags) == rft->fflags)
1631 kernfs_remove_by_name(kn, rft->name);
1636 static struct rftype *rdtgroup_get_rftype_by_name(const char *name)
1638 struct rftype *rfts, *rft;
1641 rfts = res_common_files;
1642 len = ARRAY_SIZE(res_common_files);
1644 for (rft = rfts; rft < rfts + len; rft++) {
1645 if (!strcmp(rft->name, name))
1652 void __init thread_throttle_mode_init(void)
1656 rft = rdtgroup_get_rftype_by_name("thread_throttle_mode");
1660 rft->fflags = RF_CTRL_INFO | RFTYPE_RES_MB;
1664 * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file
1665 * @r: The resource group with which the file is associated.
1666 * @name: Name of the file
1668 * The permissions of named resctrl file, directory, or link are modified
1669 * to not allow read, write, or execute by any user.
1671 * WARNING: This function is intended to communicate to the user that the
1672 * resctrl file has been locked down - that it is not relevant to the
1673 * particular state the system finds itself in. It should not be relied
1674 * on to protect from user access because after the file's permissions
1675 * are restricted the user can still change the permissions using chmod
1676 * from the command line.
1678 * Return: 0 on success, <0 on failure.
1680 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name)
1682 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1683 struct kernfs_node *kn;
1686 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1690 switch (kernfs_type(kn)) {
1692 iattr.ia_mode = S_IFDIR;
1695 iattr.ia_mode = S_IFREG;
1698 iattr.ia_mode = S_IFLNK;
1702 ret = kernfs_setattr(kn, &iattr);
1708 * rdtgroup_kn_mode_restore - Restore user access to named resctrl file
1709 * @r: The resource group with which the file is associated.
1710 * @name: Name of the file
1711 * @mask: Mask of permissions that should be restored
1713 * Restore the permissions of the named file. If @name is a directory the
1714 * permissions of its parent will be used.
1716 * Return: 0 on success, <0 on failure.
1718 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
1721 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1722 struct kernfs_node *kn, *parent;
1723 struct rftype *rfts, *rft;
1726 rfts = res_common_files;
1727 len = ARRAY_SIZE(res_common_files);
1729 for (rft = rfts; rft < rfts + len; rft++) {
1730 if (!strcmp(rft->name, name))
1731 iattr.ia_mode = rft->mode & mask;
1734 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1738 switch (kernfs_type(kn)) {
1740 parent = kernfs_get_parent(kn);
1742 iattr.ia_mode |= parent->mode;
1745 iattr.ia_mode |= S_IFDIR;
1748 iattr.ia_mode |= S_IFREG;
1751 iattr.ia_mode |= S_IFLNK;
1755 ret = kernfs_setattr(kn, &iattr);
1760 static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name,
1761 unsigned long fflags)
1763 struct kernfs_node *kn_subdir;
1766 kn_subdir = kernfs_create_dir(kn_info, name,
1768 if (IS_ERR(kn_subdir))
1769 return PTR_ERR(kn_subdir);
1771 ret = rdtgroup_kn_set_ugid(kn_subdir);
1775 ret = rdtgroup_add_files(kn_subdir, fflags);
1777 kernfs_activate(kn_subdir);
1782 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
1784 struct rdt_resource *r;
1785 unsigned long fflags;
1789 /* create the directory */
1790 kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
1791 if (IS_ERR(kn_info))
1792 return PTR_ERR(kn_info);
1794 ret = rdtgroup_add_files(kn_info, RF_TOP_INFO);
1798 for_each_alloc_enabled_rdt_resource(r) {
1799 fflags = r->fflags | RF_CTRL_INFO;
1800 ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags);
1805 for_each_mon_enabled_rdt_resource(r) {
1806 fflags = r->fflags | RF_MON_INFO;
1807 sprintf(name, "%s_MON", r->name);
1808 ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
1813 ret = rdtgroup_kn_set_ugid(kn_info);
1817 kernfs_activate(kn_info);
1822 kernfs_remove(kn_info);
1827 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
1828 char *name, struct kernfs_node **dest_kn)
1830 struct kernfs_node *kn;
1833 /* create the directory */
1834 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1841 ret = rdtgroup_kn_set_ugid(kn);
1845 kernfs_activate(kn);
1854 static void l3_qos_cfg_update(void *arg)
1858 wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL);
1861 static void l2_qos_cfg_update(void *arg)
1865 wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL);
1868 static inline bool is_mba_linear(void)
1870 return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear;
1873 static int set_cache_qos_cfg(int level, bool enable)
1875 void (*update)(void *arg);
1876 struct rdt_resource *r_l;
1877 cpumask_var_t cpu_mask;
1878 struct rdt_domain *d;
1881 if (level == RDT_RESOURCE_L3)
1882 update = l3_qos_cfg_update;
1883 else if (level == RDT_RESOURCE_L2)
1884 update = l2_qos_cfg_update;
1888 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1891 r_l = &rdt_resources_all[level];
1892 list_for_each_entry(d, &r_l->domains, list) {
1893 if (r_l->cache.arch_has_per_cpu_cfg)
1894 /* Pick all the CPUs in the domain instance */
1895 for_each_cpu(cpu, &d->cpu_mask)
1896 cpumask_set_cpu(cpu, cpu_mask);
1898 /* Pick one CPU from each domain instance to update MSR */
1899 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1902 /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1903 if (cpumask_test_cpu(cpu, cpu_mask))
1905 /* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1906 smp_call_function_many(cpu_mask, update, &enable, 1);
1909 free_cpumask_var(cpu_mask);
1914 /* Restore the qos cfg state when a domain comes online */
1915 void rdt_domain_reconfigure_cdp(struct rdt_resource *r)
1917 if (!r->alloc_capable)
1920 if (r == &rdt_resources_all[RDT_RESOURCE_L2DATA])
1921 l2_qos_cfg_update(&r->alloc_enabled);
1923 if (r == &rdt_resources_all[RDT_RESOURCE_L3DATA])
1924 l3_qos_cfg_update(&r->alloc_enabled);
1928 * Enable or disable the MBA software controller
1929 * which helps user specify bandwidth in MBps.
1930 * MBA software controller is supported only if
1931 * MBM is supported and MBA is in linear scale.
1933 static int set_mba_sc(bool mba_sc)
1935 struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA];
1936 struct rdt_domain *d;
1938 if (!is_mbm_enabled() || !is_mba_linear() ||
1939 mba_sc == is_mba_sc(r))
1942 r->membw.mba_sc = mba_sc;
1943 list_for_each_entry(d, &r->domains, list)
1944 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val);
1949 static int cdp_enable(int level, int data_type, int code_type)
1951 struct rdt_resource *r_ldata = &rdt_resources_all[data_type];
1952 struct rdt_resource *r_lcode = &rdt_resources_all[code_type];
1953 struct rdt_resource *r_l = &rdt_resources_all[level];
1956 if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1957 !r_lcode->alloc_capable)
1960 ret = set_cache_qos_cfg(level, true);
1962 r_l->alloc_enabled = false;
1963 r_ldata->alloc_enabled = true;
1964 r_lcode->alloc_enabled = true;
1969 static int cdpl3_enable(void)
1971 return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1972 RDT_RESOURCE_L3CODE);
1975 static int cdpl2_enable(void)
1977 return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1978 RDT_RESOURCE_L2CODE);
1981 static void cdp_disable(int level, int data_type, int code_type)
1983 struct rdt_resource *r = &rdt_resources_all[level];
1985 r->alloc_enabled = r->alloc_capable;
1987 if (rdt_resources_all[data_type].alloc_enabled) {
1988 rdt_resources_all[data_type].alloc_enabled = false;
1989 rdt_resources_all[code_type].alloc_enabled = false;
1990 set_cache_qos_cfg(level, false);
1994 static void cdpl3_disable(void)
1996 cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
1999 static void cdpl2_disable(void)
2001 cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
2004 static void cdp_disable_all(void)
2006 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
2008 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
2013 * We don't allow rdtgroup directories to be created anywhere
2014 * except the root directory. Thus when looking for the rdtgroup
2015 * structure for a kernfs node we are either looking at a directory,
2016 * in which case the rdtgroup structure is pointed at by the "priv"
2017 * field, otherwise we have a file, and need only look to the parent
2018 * to find the rdtgroup.
2020 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
2022 if (kernfs_type(kn) == KERNFS_DIR) {
2024 * All the resource directories use "kn->priv"
2025 * to point to the "struct rdtgroup" for the
2026 * resource. "info" and its subdirectories don't
2027 * have rdtgroup structures, so return NULL here.
2029 if (kn == kn_info || kn->parent == kn_info)
2034 return kn->parent->priv;
2038 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
2040 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
2045 atomic_inc(&rdtgrp->waitcount);
2046 kernfs_break_active_protection(kn);
2048 mutex_lock(&rdtgroup_mutex);
2050 /* Was this group deleted while we waited? */
2051 if (rdtgrp->flags & RDT_DELETED)
2057 void rdtgroup_kn_unlock(struct kernfs_node *kn)
2059 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
2064 mutex_unlock(&rdtgroup_mutex);
2066 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
2067 (rdtgrp->flags & RDT_DELETED)) {
2068 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2069 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2070 rdtgroup_pseudo_lock_remove(rdtgrp);
2071 kernfs_unbreak_active_protection(kn);
2072 rdtgroup_remove(rdtgrp);
2074 kernfs_unbreak_active_protection(kn);
2078 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2079 struct rdtgroup *prgrp,
2080 struct kernfs_node **mon_data_kn);
2082 static int rdt_enable_ctx(struct rdt_fs_context *ctx)
2086 if (ctx->enable_cdpl2)
2087 ret = cdpl2_enable();
2089 if (!ret && ctx->enable_cdpl3)
2090 ret = cdpl3_enable();
2092 if (!ret && ctx->enable_mba_mbps)
2093 ret = set_mba_sc(true);
2098 static int rdt_get_tree(struct fs_context *fc)
2100 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2101 struct rdt_domain *dom;
2102 struct rdt_resource *r;
2106 mutex_lock(&rdtgroup_mutex);
2108 * resctrl file system can only be mounted once.
2110 if (static_branch_unlikely(&rdt_enable_key)) {
2115 ret = rdt_enable_ctx(ctx);
2121 ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
2125 if (rdt_mon_capable) {
2126 ret = mongroup_create_dir(rdtgroup_default.kn,
2127 &rdtgroup_default, "mon_groups",
2132 ret = mkdir_mondata_all(rdtgroup_default.kn,
2133 &rdtgroup_default, &kn_mondata);
2136 rdtgroup_default.mon.mon_data_kn = kn_mondata;
2139 ret = rdt_pseudo_lock_init();
2143 ret = kernfs_get_tree(fc);
2147 if (rdt_alloc_capable)
2148 static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
2149 if (rdt_mon_capable)
2150 static_branch_enable_cpuslocked(&rdt_mon_enable_key);
2152 if (rdt_alloc_capable || rdt_mon_capable)
2153 static_branch_enable_cpuslocked(&rdt_enable_key);
2155 if (is_mbm_enabled()) {
2156 r = &rdt_resources_all[RDT_RESOURCE_L3];
2157 list_for_each_entry(dom, &r->domains, list)
2158 mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
2164 rdt_pseudo_lock_release();
2166 if (rdt_mon_capable)
2167 kernfs_remove(kn_mondata);
2169 if (rdt_mon_capable)
2170 kernfs_remove(kn_mongrp);
2172 kernfs_remove(kn_info);
2174 if (ctx->enable_mba_mbps)
2179 rdt_last_cmd_clear();
2180 mutex_unlock(&rdtgroup_mutex);
2192 static const struct fs_parameter_spec rdt_fs_parameters[] = {
2193 fsparam_flag("cdp", Opt_cdp),
2194 fsparam_flag("cdpl2", Opt_cdpl2),
2195 fsparam_flag("mba_MBps", Opt_mba_mbps),
2199 static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param)
2201 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2202 struct fs_parse_result result;
2205 opt = fs_parse(fc, rdt_fs_parameters, param, &result);
2211 ctx->enable_cdpl3 = true;
2214 ctx->enable_cdpl2 = true;
2217 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2219 ctx->enable_mba_mbps = true;
2226 static void rdt_fs_context_free(struct fs_context *fc)
2228 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2230 kernfs_free_fs_context(fc);
2234 static const struct fs_context_operations rdt_fs_context_ops = {
2235 .free = rdt_fs_context_free,
2236 .parse_param = rdt_parse_param,
2237 .get_tree = rdt_get_tree,
2240 static int rdt_init_fs_context(struct fs_context *fc)
2242 struct rdt_fs_context *ctx;
2244 ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL);
2248 ctx->kfc.root = rdt_root;
2249 ctx->kfc.magic = RDTGROUP_SUPER_MAGIC;
2250 fc->fs_private = &ctx->kfc;
2251 fc->ops = &rdt_fs_context_ops;
2252 put_user_ns(fc->user_ns);
2253 fc->user_ns = get_user_ns(&init_user_ns);
2258 static int reset_all_ctrls(struct rdt_resource *r)
2260 struct msr_param msr_param;
2261 cpumask_var_t cpu_mask;
2262 struct rdt_domain *d;
2265 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
2270 msr_param.high = r->num_closid;
2273 * Disable resource control for this resource by setting all
2274 * CBMs in all domains to the maximum mask value. Pick one CPU
2275 * from each domain to update the MSRs below.
2277 list_for_each_entry(d, &r->domains, list) {
2278 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
2280 for (i = 0; i < r->num_closid; i++)
2281 d->ctrl_val[i] = r->default_ctrl;
2284 /* Update CBM on this cpu if it's in cpu_mask. */
2285 if (cpumask_test_cpu(cpu, cpu_mask))
2286 rdt_ctrl_update(&msr_param);
2287 /* Update CBM on all other cpus in cpu_mask. */
2288 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
2291 free_cpumask_var(cpu_mask);
2297 * Move tasks from one to the other group. If @from is NULL, then all tasks
2298 * in the systems are moved unconditionally (used for teardown).
2300 * If @mask is not NULL the cpus on which moved tasks are running are set
2301 * in that mask so the update smp function call is restricted to affected
2304 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
2305 struct cpumask *mask)
2307 struct task_struct *p, *t;
2309 read_lock(&tasklist_lock);
2310 for_each_process_thread(p, t) {
2311 if (!from || is_closid_match(t, from) ||
2312 is_rmid_match(t, from)) {
2313 WRITE_ONCE(t->closid, to->closid);
2314 WRITE_ONCE(t->rmid, to->mon.rmid);
2317 * If the task is on a CPU, set the CPU in the mask.
2318 * The detection is inaccurate as tasks might move or
2319 * schedule before the smp function call takes place.
2320 * In such a case the function call is pointless, but
2321 * there is no other side effect.
2323 if (IS_ENABLED(CONFIG_SMP) && mask && task_curr(t))
2324 cpumask_set_cpu(task_cpu(t), mask);
2327 read_unlock(&tasklist_lock);
2330 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
2332 struct rdtgroup *sentry, *stmp;
2333 struct list_head *head;
2335 head = &rdtgrp->mon.crdtgrp_list;
2336 list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
2337 free_rmid(sentry->mon.rmid);
2338 list_del(&sentry->mon.crdtgrp_list);
2340 if (atomic_read(&sentry->waitcount) != 0)
2341 sentry->flags = RDT_DELETED;
2343 rdtgroup_remove(sentry);
2348 * Forcibly remove all of subdirectories under root.
2350 static void rmdir_all_sub(void)
2352 struct rdtgroup *rdtgrp, *tmp;
2354 /* Move all tasks to the default resource group */
2355 rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
2357 list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
2358 /* Free any child rmids */
2359 free_all_child_rdtgrp(rdtgrp);
2361 /* Remove each rdtgroup other than root */
2362 if (rdtgrp == &rdtgroup_default)
2365 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2366 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2367 rdtgroup_pseudo_lock_remove(rdtgrp);
2370 * Give any CPUs back to the default group. We cannot copy
2371 * cpu_online_mask because a CPU might have executed the
2372 * offline callback already, but is still marked online.
2374 cpumask_or(&rdtgroup_default.cpu_mask,
2375 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2377 free_rmid(rdtgrp->mon.rmid);
2379 kernfs_remove(rdtgrp->kn);
2380 list_del(&rdtgrp->rdtgroup_list);
2382 if (atomic_read(&rdtgrp->waitcount) != 0)
2383 rdtgrp->flags = RDT_DELETED;
2385 rdtgroup_remove(rdtgrp);
2387 /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
2388 update_closid_rmid(cpu_online_mask, &rdtgroup_default);
2390 kernfs_remove(kn_info);
2391 kernfs_remove(kn_mongrp);
2392 kernfs_remove(kn_mondata);
2395 static void rdt_kill_sb(struct super_block *sb)
2397 struct rdt_resource *r;
2400 mutex_lock(&rdtgroup_mutex);
2404 /*Put everything back to default values. */
2405 for_each_alloc_enabled_rdt_resource(r)
2409 rdt_pseudo_lock_release();
2410 rdtgroup_default.mode = RDT_MODE_SHAREABLE;
2411 static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
2412 static_branch_disable_cpuslocked(&rdt_mon_enable_key);
2413 static_branch_disable_cpuslocked(&rdt_enable_key);
2415 mutex_unlock(&rdtgroup_mutex);
2419 static struct file_system_type rdt_fs_type = {
2421 .init_fs_context = rdt_init_fs_context,
2422 .parameters = rdt_fs_parameters,
2423 .kill_sb = rdt_kill_sb,
2426 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
2429 struct kernfs_node *kn;
2432 kn = __kernfs_create_file(parent_kn, name, 0444,
2433 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
2434 &kf_mondata_ops, priv, NULL, NULL);
2438 ret = rdtgroup_kn_set_ugid(kn);
2448 * Remove all subdirectories of mon_data of ctrl_mon groups
2449 * and monitor groups with given domain id.
2451 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
2453 struct rdtgroup *prgrp, *crgrp;
2456 if (!r->mon_enabled)
2459 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2460 sprintf(name, "mon_%s_%02d", r->name, dom_id);
2461 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
2463 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
2464 kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
2468 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
2469 struct rdt_domain *d,
2470 struct rdt_resource *r, struct rdtgroup *prgrp)
2472 union mon_data_bits priv;
2473 struct kernfs_node *kn;
2474 struct mon_evt *mevt;
2475 struct rmid_read rr;
2479 sprintf(name, "mon_%s_%02d", r->name, d->id);
2480 /* create the directory */
2481 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
2485 ret = rdtgroup_kn_set_ugid(kn);
2489 if (WARN_ON(list_empty(&r->evt_list))) {
2494 priv.u.rid = r->rid;
2495 priv.u.domid = d->id;
2496 list_for_each_entry(mevt, &r->evt_list, list) {
2497 priv.u.evtid = mevt->evtid;
2498 ret = mon_addfile(kn, mevt->name, priv.priv);
2502 if (is_mbm_event(mevt->evtid))
2503 mon_event_read(&rr, r, d, prgrp, mevt->evtid, true);
2505 kernfs_activate(kn);
2514 * Add all subdirectories of mon_data for "ctrl_mon" groups
2515 * and "monitor" groups with given domain id.
2517 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
2518 struct rdt_domain *d)
2520 struct kernfs_node *parent_kn;
2521 struct rdtgroup *prgrp, *crgrp;
2522 struct list_head *head;
2524 if (!r->mon_enabled)
2527 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2528 parent_kn = prgrp->mon.mon_data_kn;
2529 mkdir_mondata_subdir(parent_kn, d, r, prgrp);
2531 head = &prgrp->mon.crdtgrp_list;
2532 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
2533 parent_kn = crgrp->mon.mon_data_kn;
2534 mkdir_mondata_subdir(parent_kn, d, r, crgrp);
2539 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
2540 struct rdt_resource *r,
2541 struct rdtgroup *prgrp)
2543 struct rdt_domain *dom;
2546 list_for_each_entry(dom, &r->domains, list) {
2547 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
2556 * This creates a directory mon_data which contains the monitored data.
2558 * mon_data has one directory for each domain which are named
2559 * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
2560 * with L3 domain looks as below:
2567 * Each domain directory has one file per event:
2572 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2573 struct rdtgroup *prgrp,
2574 struct kernfs_node **dest_kn)
2576 struct rdt_resource *r;
2577 struct kernfs_node *kn;
2581 * Create the mon_data directory first.
2583 ret = mongroup_create_dir(parent_kn, prgrp, "mon_data", &kn);
2591 * Create the subdirectories for each domain. Note that all events
2592 * in a domain like L3 are grouped into a resource whose domain is L3
2594 for_each_mon_enabled_rdt_resource(r) {
2595 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
2608 * cbm_ensure_valid - Enforce validity on provided CBM
2609 * @_val: Candidate CBM
2610 * @r: RDT resource to which the CBM belongs
2612 * The provided CBM represents all cache portions available for use. This
2613 * may be represented by a bitmap that does not consist of contiguous ones
2614 * and thus be an invalid CBM.
2615 * Here the provided CBM is forced to be a valid CBM by only considering
2616 * the first set of contiguous bits as valid and clearing all bits.
2617 * The intention here is to provide a valid default CBM with which a new
2618 * resource group is initialized. The user can follow this with a
2619 * modification to the CBM if the default does not satisfy the
2622 static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r)
2624 unsigned int cbm_len = r->cache.cbm_len;
2625 unsigned long first_bit, zero_bit;
2626 unsigned long val = _val;
2631 first_bit = find_first_bit(&val, cbm_len);
2632 zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
2634 /* Clear any remaining bits to ensure contiguous region */
2635 bitmap_clear(&val, zero_bit, cbm_len - zero_bit);
2640 * Initialize cache resources per RDT domain
2642 * Set the RDT domain up to start off with all usable allocations. That is,
2643 * all shareable and unused bits. All-zero CBM is invalid.
2645 static int __init_one_rdt_domain(struct rdt_domain *d, struct rdt_resource *r,
2648 struct rdt_resource *r_cdp = NULL;
2649 struct rdt_domain *d_cdp = NULL;
2650 u32 used_b = 0, unused_b = 0;
2651 unsigned long tmp_cbm;
2652 enum rdtgrp_mode mode;
2653 u32 peer_ctl, *ctrl;
2656 rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp);
2657 d->have_new_ctrl = false;
2658 d->new_ctrl = r->cache.shareable_bits;
2659 used_b = r->cache.shareable_bits;
2661 for (i = 0; i < closids_supported(); i++, ctrl++) {
2662 if (closid_allocated(i) && i != closid) {
2663 mode = rdtgroup_mode_by_closid(i);
2664 if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
2666 * ctrl values for locksetup aren't relevant
2667 * until the schemata is written, and the mode
2668 * becomes RDT_MODE_PSEUDO_LOCKED.
2672 * If CDP is active include peer domain's
2673 * usage to ensure there is no overlap
2674 * with an exclusive group.
2677 peer_ctl = d_cdp->ctrl_val[i];
2680 used_b |= *ctrl | peer_ctl;
2681 if (mode == RDT_MODE_SHAREABLE)
2682 d->new_ctrl |= *ctrl | peer_ctl;
2685 if (d->plr && d->plr->cbm > 0)
2686 used_b |= d->plr->cbm;
2687 unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
2688 unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
2689 d->new_ctrl |= unused_b;
2691 * Force the initial CBM to be valid, user can
2692 * modify the CBM based on system availability.
2694 d->new_ctrl = cbm_ensure_valid(d->new_ctrl, r);
2696 * Assign the u32 CBM to an unsigned long to ensure that
2697 * bitmap_weight() does not access out-of-bound memory.
2699 tmp_cbm = d->new_ctrl;
2700 if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) {
2701 rdt_last_cmd_printf("No space on %s:%d\n", r->name, d->id);
2704 d->have_new_ctrl = true;
2710 * Initialize cache resources with default values.
2712 * A new RDT group is being created on an allocation capable (CAT)
2713 * supporting system. Set this group up to start off with all usable
2716 * If there are no more shareable bits available on any domain then
2717 * the entire allocation will fail.
2719 static int rdtgroup_init_cat(struct rdt_resource *r, u32 closid)
2721 struct rdt_domain *d;
2724 list_for_each_entry(d, &r->domains, list) {
2725 ret = __init_one_rdt_domain(d, r, closid);
2733 /* Initialize MBA resource with default values. */
2734 static void rdtgroup_init_mba(struct rdt_resource *r)
2736 struct rdt_domain *d;
2738 list_for_each_entry(d, &r->domains, list) {
2739 d->new_ctrl = is_mba_sc(r) ? MBA_MAX_MBPS : r->default_ctrl;
2740 d->have_new_ctrl = true;
2744 /* Initialize the RDT group's allocations. */
2745 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
2747 struct rdt_resource *r;
2750 for_each_alloc_enabled_rdt_resource(r) {
2751 if (r->rid == RDT_RESOURCE_MBA) {
2752 rdtgroup_init_mba(r);
2754 ret = rdtgroup_init_cat(r, rdtgrp->closid);
2759 ret = update_domains(r, rdtgrp->closid);
2761 rdt_last_cmd_puts("Failed to initialize allocations\n");
2767 rdtgrp->mode = RDT_MODE_SHAREABLE;
2772 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
2773 const char *name, umode_t mode,
2774 enum rdt_group_type rtype, struct rdtgroup **r)
2776 struct rdtgroup *prdtgrp, *rdtgrp;
2777 struct kernfs_node *kn;
2781 prdtgrp = rdtgroup_kn_lock_live(parent_kn);
2787 if (rtype == RDTMON_GROUP &&
2788 (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2789 prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
2791 rdt_last_cmd_puts("Pseudo-locking in progress\n");
2795 /* allocate the rdtgroup. */
2796 rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
2799 rdt_last_cmd_puts("Kernel out of memory\n");
2803 rdtgrp->mon.parent = prdtgrp;
2804 rdtgrp->type = rtype;
2805 INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
2807 /* kernfs creates the directory for rdtgrp */
2808 kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
2811 rdt_last_cmd_puts("kernfs create error\n");
2817 * kernfs_remove() will drop the reference count on "kn" which
2818 * will free it. But we still need it to stick around for the
2819 * rdtgroup_kn_unlock(kn) call. Take one extra reference here,
2820 * which will be dropped by kernfs_put() in rdtgroup_remove().
2824 ret = rdtgroup_kn_set_ugid(kn);
2826 rdt_last_cmd_puts("kernfs perm error\n");
2830 files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
2831 ret = rdtgroup_add_files(kn, files);
2833 rdt_last_cmd_puts("kernfs fill error\n");
2837 if (rdt_mon_capable) {
2840 rdt_last_cmd_puts("Out of RMIDs\n");
2843 rdtgrp->mon.rmid = ret;
2845 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
2847 rdt_last_cmd_puts("kernfs subdir error\n");
2851 kernfs_activate(kn);
2854 * The caller unlocks the parent_kn upon success.
2859 free_rmid(rdtgrp->mon.rmid);
2861 kernfs_put(rdtgrp->kn);
2862 kernfs_remove(rdtgrp->kn);
2866 rdtgroup_kn_unlock(parent_kn);
2870 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
2872 kernfs_remove(rgrp->kn);
2873 free_rmid(rgrp->mon.rmid);
2874 rdtgroup_remove(rgrp);
2878 * Create a monitor group under "mon_groups" directory of a control
2879 * and monitor group(ctrl_mon). This is a resource group
2880 * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
2882 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
2883 const char *name, umode_t mode)
2885 struct rdtgroup *rdtgrp, *prgrp;
2888 ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTMON_GROUP, &rdtgrp);
2892 prgrp = rdtgrp->mon.parent;
2893 rdtgrp->closid = prgrp->closid;
2896 * Add the rdtgrp to the list of rdtgrps the parent
2897 * ctrl_mon group has to track.
2899 list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
2901 rdtgroup_kn_unlock(parent_kn);
2906 * These are rdtgroups created under the root directory. Can be used
2907 * to allocate and monitor resources.
2909 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
2910 const char *name, umode_t mode)
2912 struct rdtgroup *rdtgrp;
2913 struct kernfs_node *kn;
2917 ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTCTRL_GROUP, &rdtgrp);
2922 ret = closid_alloc();
2924 rdt_last_cmd_puts("Out of CLOSIDs\n");
2925 goto out_common_fail;
2930 rdtgrp->closid = closid;
2931 ret = rdtgroup_init_alloc(rdtgrp);
2935 list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
2937 if (rdt_mon_capable) {
2939 * Create an empty mon_groups directory to hold the subset
2940 * of tasks and cpus to monitor.
2942 ret = mongroup_create_dir(kn, rdtgrp, "mon_groups", NULL);
2944 rdt_last_cmd_puts("kernfs subdir error\n");
2952 list_del(&rdtgrp->rdtgroup_list);
2954 closid_free(closid);
2956 mkdir_rdt_prepare_clean(rdtgrp);
2958 rdtgroup_kn_unlock(parent_kn);
2963 * We allow creating mon groups only with in a directory called "mon_groups"
2964 * which is present in every ctrl_mon group. Check if this is a valid
2965 * "mon_groups" directory.
2967 * 1. The directory should be named "mon_groups".
2968 * 2. The mon group itself should "not" be named "mon_groups".
2969 * This makes sure "mon_groups" directory always has a ctrl_mon group
2972 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
2974 return (!strcmp(kn->name, "mon_groups") &&
2975 strcmp(name, "mon_groups"));
2978 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
2981 /* Do not accept '\n' to avoid unparsable situation. */
2982 if (strchr(name, '\n'))
2986 * If the parent directory is the root directory and RDT
2987 * allocation is supported, add a control and monitoring
2990 if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
2991 return rdtgroup_mkdir_ctrl_mon(parent_kn, name, mode);
2994 * If RDT monitoring is supported and the parent directory is a valid
2995 * "mon_groups" directory, add a monitoring subdirectory.
2997 if (rdt_mon_capable && is_mon_groups(parent_kn, name))
2998 return rdtgroup_mkdir_mon(parent_kn, name, mode);
3003 static int rdtgroup_rmdir_mon(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask)
3005 struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
3008 /* Give any tasks back to the parent group */
3009 rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
3011 /* Update per cpu rmid of the moved CPUs first */
3012 for_each_cpu(cpu, &rdtgrp->cpu_mask)
3013 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
3015 * Update the MSR on moved CPUs and CPUs which have moved
3016 * task running on them.
3018 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
3019 update_closid_rmid(tmpmask, NULL);
3021 rdtgrp->flags = RDT_DELETED;
3022 free_rmid(rdtgrp->mon.rmid);
3025 * Remove the rdtgrp from the parent ctrl_mon group's list
3027 WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
3028 list_del(&rdtgrp->mon.crdtgrp_list);
3030 kernfs_remove(rdtgrp->kn);
3035 static int rdtgroup_ctrl_remove(struct rdtgroup *rdtgrp)
3037 rdtgrp->flags = RDT_DELETED;
3038 list_del(&rdtgrp->rdtgroup_list);
3040 kernfs_remove(rdtgrp->kn);
3044 static int rdtgroup_rmdir_ctrl(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask)
3048 /* Give any tasks back to the default group */
3049 rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
3051 /* Give any CPUs back to the default group */
3052 cpumask_or(&rdtgroup_default.cpu_mask,
3053 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
3055 /* Update per cpu closid and rmid of the moved CPUs first */
3056 for_each_cpu(cpu, &rdtgrp->cpu_mask) {
3057 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
3058 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
3062 * Update the MSR on moved CPUs and CPUs which have moved
3063 * task running on them.
3065 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
3066 update_closid_rmid(tmpmask, NULL);
3068 closid_free(rdtgrp->closid);
3069 free_rmid(rdtgrp->mon.rmid);
3071 rdtgroup_ctrl_remove(rdtgrp);
3074 * Free all the child monitor group rmids.
3076 free_all_child_rdtgrp(rdtgrp);
3081 static int rdtgroup_rmdir(struct kernfs_node *kn)
3083 struct kernfs_node *parent_kn = kn->parent;
3084 struct rdtgroup *rdtgrp;
3085 cpumask_var_t tmpmask;
3088 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
3091 rdtgrp = rdtgroup_kn_lock_live(kn);
3098 * If the rdtgroup is a ctrl_mon group and parent directory
3099 * is the root directory, remove the ctrl_mon group.
3101 * If the rdtgroup is a mon group and parent directory
3102 * is a valid "mon_groups" directory, remove the mon group.
3104 if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn &&
3105 rdtgrp != &rdtgroup_default) {
3106 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
3107 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
3108 ret = rdtgroup_ctrl_remove(rdtgrp);
3110 ret = rdtgroup_rmdir_ctrl(rdtgrp, tmpmask);
3112 } else if (rdtgrp->type == RDTMON_GROUP &&
3113 is_mon_groups(parent_kn, kn->name)) {
3114 ret = rdtgroup_rmdir_mon(rdtgrp, tmpmask);
3120 rdtgroup_kn_unlock(kn);
3121 free_cpumask_var(tmpmask);
3125 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
3127 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
3128 seq_puts(seq, ",cdp");
3130 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
3131 seq_puts(seq, ",cdpl2");
3133 if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA]))
3134 seq_puts(seq, ",mba_MBps");
3139 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
3140 .mkdir = rdtgroup_mkdir,
3141 .rmdir = rdtgroup_rmdir,
3142 .show_options = rdtgroup_show_options,
3145 static int __init rdtgroup_setup_root(void)
3149 rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
3150 KERNFS_ROOT_CREATE_DEACTIVATED |
3151 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
3153 if (IS_ERR(rdt_root))
3154 return PTR_ERR(rdt_root);
3156 mutex_lock(&rdtgroup_mutex);
3158 rdtgroup_default.closid = 0;
3159 rdtgroup_default.mon.rmid = 0;
3160 rdtgroup_default.type = RDTCTRL_GROUP;
3161 INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
3163 list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
3165 ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
3167 kernfs_destroy_root(rdt_root);
3171 rdtgroup_default.kn = rdt_root->kn;
3172 kernfs_activate(rdtgroup_default.kn);
3175 mutex_unlock(&rdtgroup_mutex);
3181 * rdtgroup_init - rdtgroup initialization
3183 * Setup resctrl file system including set up root, create mount point,
3184 * register rdtgroup filesystem, and initialize files under root directory.
3186 * Return: 0 on success or -errno
3188 int __init rdtgroup_init(void)
3192 seq_buf_init(&last_cmd_status, last_cmd_status_buf,
3193 sizeof(last_cmd_status_buf));
3195 ret = rdtgroup_setup_root();
3199 ret = sysfs_create_mount_point(fs_kobj, "resctrl");
3203 ret = register_filesystem(&rdt_fs_type);
3205 goto cleanup_mountpoint;
3208 * Adding the resctrl debugfs directory here may not be ideal since
3209 * it would let the resctrl debugfs directory appear on the debugfs
3210 * filesystem before the resctrl filesystem is mounted.
3211 * It may also be ok since that would enable debugging of RDT before
3212 * resctrl is mounted.
3213 * The reason why the debugfs directory is created here and not in
3214 * rdt_get_tree() is because rdt_get_tree() takes rdtgroup_mutex and
3215 * during the debugfs directory creation also &sb->s_type->i_mutex_key
3216 * (the lockdep class of inode->i_rwsem). Other filesystem
3217 * interactions (eg. SyS_getdents) have the lock ordering:
3218 * &sb->s_type->i_mutex_key --> &mm->mmap_lock
3219 * During mmap(), called with &mm->mmap_lock, the rdtgroup_mutex
3220 * is taken, thus creating dependency:
3221 * &mm->mmap_lock --> rdtgroup_mutex for the latter that can cause
3222 * issues considering the other two lock dependencies.
3223 * By creating the debugfs directory here we avoid a dependency
3224 * that may cause deadlock (even though file operations cannot
3225 * occur until the filesystem is mounted, but I do not know how to
3226 * tell lockdep that).
3228 debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
3233 sysfs_remove_mount_point(fs_kobj, "resctrl");
3235 kernfs_destroy_root(rdt_root);
3240 void __exit rdtgroup_exit(void)
3242 debugfs_remove_recursive(debugfs_resctrl);
3243 unregister_filesystem(&rdt_fs_type);
3244 sysfs_remove_mount_point(fs_kobj, "resctrl");
3245 kernfs_destroy_root(rdt_root);