2 * User interface for Resource Alloction in Resource Director Technology(RDT)
4 * Copyright (C) 2016 Intel Corporation
6 * Author: Fenghua Yu <fenghua.yu@intel.com>
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms and conditions of the GNU General Public License,
10 * version 2, as published by the Free Software Foundation.
12 * This program is distributed in the hope it will be useful, but WITHOUT
13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
17 * More information about RDT be found in the Intel (R) x86 Architecture
18 * Software Developer Manual.
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23 #include <linux/cacheinfo.h>
24 #include <linux/cpu.h>
25 #include <linux/debugfs.h>
27 #include <linux/sysfs.h>
28 #include <linux/kernfs.h>
29 #include <linux/seq_buf.h>
30 #include <linux/seq_file.h>
31 #include <linux/sched/signal.h>
32 #include <linux/sched/task.h>
33 #include <linux/slab.h>
34 #include <linux/task_work.h>
36 #include <uapi/linux/magic.h>
38 #include <asm/intel_rdt_sched.h>
39 #include "intel_rdt.h"
41 DEFINE_STATIC_KEY_FALSE(rdt_enable_key);
42 DEFINE_STATIC_KEY_FALSE(rdt_mon_enable_key);
43 DEFINE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
44 static struct kernfs_root *rdt_root;
45 struct rdtgroup rdtgroup_default;
46 LIST_HEAD(rdt_all_groups);
48 /* Kernel fs node for "info" directory under root */
49 static struct kernfs_node *kn_info;
51 /* Kernel fs node for "mon_groups" directory under root */
52 static struct kernfs_node *kn_mongrp;
54 /* Kernel fs node for "mon_data" directory under root */
55 static struct kernfs_node *kn_mondata;
57 static struct seq_buf last_cmd_status;
58 static char last_cmd_status_buf[512];
60 struct dentry *debugfs_resctrl;
62 void rdt_last_cmd_clear(void)
64 lockdep_assert_held(&rdtgroup_mutex);
65 seq_buf_clear(&last_cmd_status);
68 void rdt_last_cmd_puts(const char *s)
70 lockdep_assert_held(&rdtgroup_mutex);
71 seq_buf_puts(&last_cmd_status, s);
74 void rdt_last_cmd_printf(const char *fmt, ...)
79 lockdep_assert_held(&rdtgroup_mutex);
80 seq_buf_vprintf(&last_cmd_status, fmt, ap);
85 * Trivial allocator for CLOSIDs. Since h/w only supports a small number,
86 * we can keep a bitmap of free CLOSIDs in a single integer.
88 * Using a global CLOSID across all resources has some advantages and
90 * + We can simply set "current->closid" to assign a task to a resource
92 * + Context switch code can avoid extra memory references deciding which
93 * CLOSID to load into the PQR_ASSOC MSR
94 * - We give up some options in configuring resource groups across multi-socket
96 * - Our choices on how to configure each resource become progressively more
97 * limited as the number of resources grows.
99 static int closid_free_map;
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;
116 static int closid_alloc(void)
118 u32 closid = ffs(closid_free_map);
123 closid_free_map &= ~(1 << closid);
128 void closid_free(int closid)
130 closid_free_map |= 1 << closid;
134 * closid_allocated - test if provided closid is in use
135 * @closid: closid to be tested
137 * Return: true if @closid is currently associated with a resource group,
138 * false if @closid is free
140 static bool closid_allocated(unsigned int closid)
142 return (closid_free_map & (1 << closid)) == 0;
146 * rdtgroup_mode_by_closid - Return mode of resource group with closid
147 * @closid: closid if the resource group
149 * Each resource group is associated with a @closid. Here the mode
150 * of a resource group can be queried by searching for it using its closid.
152 * Return: mode as &enum rdtgrp_mode of resource group with closid @closid
154 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid)
156 struct rdtgroup *rdtgrp;
158 list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
159 if (rdtgrp->closid == closid)
163 return RDT_NUM_MODES;
166 static const char * const rdt_mode_str[] = {
167 [RDT_MODE_SHAREABLE] = "shareable",
168 [RDT_MODE_EXCLUSIVE] = "exclusive",
169 [RDT_MODE_PSEUDO_LOCKSETUP] = "pseudo-locksetup",
170 [RDT_MODE_PSEUDO_LOCKED] = "pseudo-locked",
174 * rdtgroup_mode_str - Return the string representation of mode
175 * @mode: the resource group mode as &enum rdtgroup_mode
177 * Return: string representation of valid mode, "unknown" otherwise
179 static const char *rdtgroup_mode_str(enum rdtgrp_mode mode)
181 if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES)
184 return rdt_mode_str[mode];
187 /* set uid and gid of rdtgroup dirs and files to that of the creator */
188 static int rdtgroup_kn_set_ugid(struct kernfs_node *kn)
190 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
191 .ia_uid = current_fsuid(),
192 .ia_gid = current_fsgid(), };
194 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
195 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
198 return kernfs_setattr(kn, &iattr);
201 static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft)
203 struct kernfs_node *kn;
206 kn = __kernfs_create_file(parent_kn, rft->name, rft->mode,
207 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
208 0, rft->kf_ops, rft, NULL, NULL);
212 ret = rdtgroup_kn_set_ugid(kn);
221 static int rdtgroup_seqfile_show(struct seq_file *m, void *arg)
223 struct kernfs_open_file *of = m->private;
224 struct rftype *rft = of->kn->priv;
227 return rft->seq_show(of, m, arg);
231 static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf,
232 size_t nbytes, loff_t off)
234 struct rftype *rft = of->kn->priv;
237 return rft->write(of, buf, nbytes, off);
242 static struct kernfs_ops rdtgroup_kf_single_ops = {
243 .atomic_write_len = PAGE_SIZE,
244 .write = rdtgroup_file_write,
245 .seq_show = rdtgroup_seqfile_show,
248 static struct kernfs_ops kf_mondata_ops = {
249 .atomic_write_len = PAGE_SIZE,
250 .seq_show = rdtgroup_mondata_show,
253 static bool is_cpu_list(struct kernfs_open_file *of)
255 struct rftype *rft = of->kn->priv;
257 return rft->flags & RFTYPE_FLAGS_CPUS_LIST;
260 static int rdtgroup_cpus_show(struct kernfs_open_file *of,
261 struct seq_file *s, void *v)
263 struct rdtgroup *rdtgrp;
266 rdtgrp = rdtgroup_kn_lock_live(of->kn);
269 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
270 seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n",
271 cpumask_pr_args(&rdtgrp->plr->d->cpu_mask));
273 seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n",
274 cpumask_pr_args(&rdtgrp->cpu_mask));
278 rdtgroup_kn_unlock(of->kn);
284 * This is safe against intel_rdt_sched_in() called from __switch_to()
285 * because __switch_to() is executed with interrupts disabled. A local call
286 * from update_closid_rmid() is proteced against __switch_to() because
287 * preemption is disabled.
289 static void update_cpu_closid_rmid(void *info)
291 struct rdtgroup *r = info;
294 this_cpu_write(pqr_state.default_closid, r->closid);
295 this_cpu_write(pqr_state.default_rmid, r->mon.rmid);
299 * We cannot unconditionally write the MSR because the current
300 * executing task might have its own closid selected. Just reuse
301 * the context switch code.
303 intel_rdt_sched_in();
307 * Update the PGR_ASSOC MSR on all cpus in @cpu_mask,
309 * Per task closids/rmids must have been set up before calling this function.
312 update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r)
316 if (cpumask_test_cpu(cpu, cpu_mask))
317 update_cpu_closid_rmid(r);
318 smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1);
322 static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
323 cpumask_var_t tmpmask)
325 struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp;
326 struct list_head *head;
328 /* Check whether cpus belong to parent ctrl group */
329 cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask);
330 if (cpumask_weight(tmpmask)) {
331 rdt_last_cmd_puts("can only add CPUs to mongroup that belong to parent\n");
335 /* Check whether cpus are dropped from this group */
336 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
337 if (cpumask_weight(tmpmask)) {
338 /* Give any dropped cpus to parent rdtgroup */
339 cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask);
340 update_closid_rmid(tmpmask, prgrp);
344 * If we added cpus, remove them from previous group that owned them
345 * and update per-cpu rmid
347 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
348 if (cpumask_weight(tmpmask)) {
349 head = &prgrp->mon.crdtgrp_list;
350 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
353 cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask,
356 update_closid_rmid(tmpmask, rdtgrp);
359 /* Done pushing/pulling - update this group with new mask */
360 cpumask_copy(&rdtgrp->cpu_mask, newmask);
365 static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m)
367 struct rdtgroup *crgrp;
369 cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m);
370 /* update the child mon group masks as well*/
371 list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list)
372 cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask);
375 static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
376 cpumask_var_t tmpmask, cpumask_var_t tmpmask1)
378 struct rdtgroup *r, *crgrp;
379 struct list_head *head;
381 /* Check whether cpus are dropped from this group */
382 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
383 if (cpumask_weight(tmpmask)) {
384 /* Can't drop from default group */
385 if (rdtgrp == &rdtgroup_default) {
386 rdt_last_cmd_puts("Can't drop CPUs from default group\n");
390 /* Give any dropped cpus to rdtgroup_default */
391 cpumask_or(&rdtgroup_default.cpu_mask,
392 &rdtgroup_default.cpu_mask, tmpmask);
393 update_closid_rmid(tmpmask, &rdtgroup_default);
397 * If we added cpus, remove them from previous group and
398 * the prev group's child groups that owned them
399 * and update per-cpu closid/rmid.
401 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
402 if (cpumask_weight(tmpmask)) {
403 list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) {
406 cpumask_and(tmpmask1, &r->cpu_mask, tmpmask);
407 if (cpumask_weight(tmpmask1))
408 cpumask_rdtgrp_clear(r, tmpmask1);
410 update_closid_rmid(tmpmask, rdtgrp);
413 /* Done pushing/pulling - update this group with new mask */
414 cpumask_copy(&rdtgrp->cpu_mask, newmask);
417 * Clear child mon group masks since there is a new parent mask
418 * now and update the rmid for the cpus the child lost.
420 head = &rdtgrp->mon.crdtgrp_list;
421 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
422 cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask);
423 update_closid_rmid(tmpmask, rdtgrp);
424 cpumask_clear(&crgrp->cpu_mask);
430 static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of,
431 char *buf, size_t nbytes, loff_t off)
433 cpumask_var_t tmpmask, newmask, tmpmask1;
434 struct rdtgroup *rdtgrp;
440 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
442 if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) {
443 free_cpumask_var(tmpmask);
446 if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) {
447 free_cpumask_var(tmpmask);
448 free_cpumask_var(newmask);
452 rdtgrp = rdtgroup_kn_lock_live(of->kn);
453 rdt_last_cmd_clear();
456 rdt_last_cmd_puts("directory was removed\n");
460 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
461 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
463 rdt_last_cmd_puts("pseudo-locking in progress\n");
468 ret = cpulist_parse(buf, newmask);
470 ret = cpumask_parse(buf, newmask);
473 rdt_last_cmd_puts("bad cpu list/mask\n");
477 /* check that user didn't specify any offline cpus */
478 cpumask_andnot(tmpmask, newmask, cpu_online_mask);
479 if (cpumask_weight(tmpmask)) {
481 rdt_last_cmd_puts("can only assign online cpus\n");
485 if (rdtgrp->type == RDTCTRL_GROUP)
486 ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1);
487 else if (rdtgrp->type == RDTMON_GROUP)
488 ret = cpus_mon_write(rdtgrp, newmask, tmpmask);
493 rdtgroup_kn_unlock(of->kn);
494 free_cpumask_var(tmpmask);
495 free_cpumask_var(newmask);
496 free_cpumask_var(tmpmask1);
498 return ret ?: nbytes;
501 struct task_move_callback {
502 struct callback_head work;
503 struct rdtgroup *rdtgrp;
506 static void move_myself(struct callback_head *head)
508 struct task_move_callback *callback;
509 struct rdtgroup *rdtgrp;
511 callback = container_of(head, struct task_move_callback, work);
512 rdtgrp = callback->rdtgrp;
515 * If resource group was deleted before this task work callback
516 * was invoked, then assign the task to root group and free the
519 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
520 (rdtgrp->flags & RDT_DELETED)) {
527 /* update PQR_ASSOC MSR to make resource group go into effect */
528 intel_rdt_sched_in();
534 static int __rdtgroup_move_task(struct task_struct *tsk,
535 struct rdtgroup *rdtgrp)
537 struct task_move_callback *callback;
540 callback = kzalloc(sizeof(*callback), GFP_KERNEL);
543 callback->work.func = move_myself;
544 callback->rdtgrp = rdtgrp;
547 * Take a refcount, so rdtgrp cannot be freed before the
548 * callback has been invoked.
550 atomic_inc(&rdtgrp->waitcount);
551 ret = task_work_add(tsk, &callback->work, true);
554 * Task is exiting. Drop the refcount and free the callback.
555 * No need to check the refcount as the group cannot be
556 * deleted before the write function unlocks rdtgroup_mutex.
558 atomic_dec(&rdtgrp->waitcount);
560 rdt_last_cmd_puts("task exited\n");
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.
567 if (rdtgrp->type == RDTCTRL_GROUP) {
568 tsk->closid = rdtgrp->closid;
569 tsk->rmid = rdtgrp->mon.rmid;
570 } else if (rdtgrp->type == RDTMON_GROUP) {
571 if (rdtgrp->mon.parent->closid == tsk->closid) {
572 tsk->rmid = rdtgrp->mon.rmid;
574 rdt_last_cmd_puts("Can't move task to different control group\n");
583 * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group
586 * Return: 1 if tasks have been assigned to @r, 0 otherwise
588 int rdtgroup_tasks_assigned(struct rdtgroup *r)
590 struct task_struct *p, *t;
593 lockdep_assert_held(&rdtgroup_mutex);
596 for_each_process_thread(p, t) {
597 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
598 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) {
608 static int rdtgroup_task_write_permission(struct task_struct *task,
609 struct kernfs_open_file *of)
611 const struct cred *tcred = get_task_cred(task);
612 const struct cred *cred = current_cred();
616 * Even if we're attaching all tasks in the thread group, we only
617 * need to check permissions on one of them.
619 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
620 !uid_eq(cred->euid, tcred->uid) &&
621 !uid_eq(cred->euid, tcred->suid)) {
622 rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
630 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
631 struct kernfs_open_file *of)
633 struct task_struct *tsk;
638 tsk = find_task_by_vpid(pid);
641 rdt_last_cmd_printf("No task %d\n", pid);
648 get_task_struct(tsk);
651 ret = rdtgroup_task_write_permission(tsk, of);
653 ret = __rdtgroup_move_task(tsk, rdtgrp);
655 put_task_struct(tsk);
659 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
660 char *buf, size_t nbytes, loff_t off)
662 struct rdtgroup *rdtgrp;
666 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
668 rdtgrp = rdtgroup_kn_lock_live(of->kn);
670 rdtgroup_kn_unlock(of->kn);
673 rdt_last_cmd_clear();
675 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
676 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
678 rdt_last_cmd_puts("pseudo-locking in progress\n");
682 ret = rdtgroup_move_task(pid, rdtgrp, of);
685 rdtgroup_kn_unlock(of->kn);
687 return ret ?: nbytes;
690 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
692 struct task_struct *p, *t;
695 for_each_process_thread(p, t) {
696 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
697 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid))
698 seq_printf(s, "%d\n", t->pid);
703 static int rdtgroup_tasks_show(struct kernfs_open_file *of,
704 struct seq_file *s, void *v)
706 struct rdtgroup *rdtgrp;
709 rdtgrp = rdtgroup_kn_lock_live(of->kn);
711 show_rdt_tasks(rdtgrp, s);
714 rdtgroup_kn_unlock(of->kn);
719 static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
720 struct seq_file *seq, void *v)
724 mutex_lock(&rdtgroup_mutex);
725 len = seq_buf_used(&last_cmd_status);
727 seq_printf(seq, "%.*s", len, last_cmd_status_buf);
729 seq_puts(seq, "ok\n");
730 mutex_unlock(&rdtgroup_mutex);
734 static int rdt_num_closids_show(struct kernfs_open_file *of,
735 struct seq_file *seq, void *v)
737 struct rdt_resource *r = of->kn->parent->priv;
739 seq_printf(seq, "%d\n", r->num_closid);
743 static int rdt_default_ctrl_show(struct kernfs_open_file *of,
744 struct seq_file *seq, void *v)
746 struct rdt_resource *r = of->kn->parent->priv;
748 seq_printf(seq, "%x\n", r->default_ctrl);
752 static int rdt_min_cbm_bits_show(struct kernfs_open_file *of,
753 struct seq_file *seq, void *v)
755 struct rdt_resource *r = of->kn->parent->priv;
757 seq_printf(seq, "%u\n", r->cache.min_cbm_bits);
761 static int rdt_shareable_bits_show(struct kernfs_open_file *of,
762 struct seq_file *seq, void *v)
764 struct rdt_resource *r = of->kn->parent->priv;
766 seq_printf(seq, "%x\n", r->cache.shareable_bits);
771 * rdt_bit_usage_show - Display current usage of resources
773 * A domain is a shared resource that can now be allocated differently. Here
774 * we display the current regions of the domain as an annotated bitmask.
775 * For each domain of this resource its allocation bitmask
776 * is annotated as below to indicate the current usage of the corresponding bit:
777 * 0 - currently unused
778 * X - currently available for sharing and used by software and hardware
779 * H - currently used by hardware only but available for software use
780 * S - currently used and shareable by software only
781 * E - currently used exclusively by one resource group
782 * P - currently pseudo-locked by one resource group
784 static int rdt_bit_usage_show(struct kernfs_open_file *of,
785 struct seq_file *seq, void *v)
787 struct rdt_resource *r = of->kn->parent->priv;
788 u32 sw_shareable = 0, hw_shareable = 0;
789 u32 exclusive = 0, pseudo_locked = 0;
790 struct rdt_domain *dom;
791 int i, hwb, swb, excl, psl;
792 enum rdtgrp_mode mode;
796 mutex_lock(&rdtgroup_mutex);
797 hw_shareable = r->cache.shareable_bits;
798 list_for_each_entry(dom, &r->domains, list) {
801 ctrl = dom->ctrl_val;
804 seq_printf(seq, "%d=", dom->id);
805 for (i = 0; i < r->num_closid; i++, ctrl++) {
806 if (!closid_allocated(i))
808 mode = rdtgroup_mode_by_closid(i);
810 case RDT_MODE_SHAREABLE:
811 sw_shareable |= *ctrl;
813 case RDT_MODE_EXCLUSIVE:
816 case RDT_MODE_PSEUDO_LOCKSETUP:
818 * RDT_MODE_PSEUDO_LOCKSETUP is possible
819 * here but not included since the CBM
820 * associated with this CLOSID in this mode
821 * is not initialized and no task or cpu can be
822 * assigned this CLOSID.
825 case RDT_MODE_PSEUDO_LOCKED:
828 "invalid mode for closid %d\n", i);
832 for (i = r->cache.cbm_len - 1; i >= 0; i--) {
833 pseudo_locked = dom->plr ? dom->plr->cbm : 0;
834 hwb = test_bit(i, (unsigned long *)&hw_shareable);
835 swb = test_bit(i, (unsigned long *)&sw_shareable);
836 excl = test_bit(i, (unsigned long *)&exclusive);
837 psl = test_bit(i, (unsigned long *)&pseudo_locked);
840 else if (hwb && !swb)
842 else if (!hwb && swb)
848 else /* Unused bits remain */
854 mutex_unlock(&rdtgroup_mutex);
858 static int rdt_min_bw_show(struct kernfs_open_file *of,
859 struct seq_file *seq, void *v)
861 struct rdt_resource *r = of->kn->parent->priv;
863 seq_printf(seq, "%u\n", r->membw.min_bw);
867 static int rdt_num_rmids_show(struct kernfs_open_file *of,
868 struct seq_file *seq, void *v)
870 struct rdt_resource *r = of->kn->parent->priv;
872 seq_printf(seq, "%d\n", r->num_rmid);
877 static int rdt_mon_features_show(struct kernfs_open_file *of,
878 struct seq_file *seq, void *v)
880 struct rdt_resource *r = of->kn->parent->priv;
881 struct mon_evt *mevt;
883 list_for_each_entry(mevt, &r->evt_list, list)
884 seq_printf(seq, "%s\n", mevt->name);
889 static int rdt_bw_gran_show(struct kernfs_open_file *of,
890 struct seq_file *seq, void *v)
892 struct rdt_resource *r = of->kn->parent->priv;
894 seq_printf(seq, "%u\n", r->membw.bw_gran);
898 static int rdt_delay_linear_show(struct kernfs_open_file *of,
899 struct seq_file *seq, void *v)
901 struct rdt_resource *r = of->kn->parent->priv;
903 seq_printf(seq, "%u\n", r->membw.delay_linear);
907 static int max_threshold_occ_show(struct kernfs_open_file *of,
908 struct seq_file *seq, void *v)
910 struct rdt_resource *r = of->kn->parent->priv;
912 seq_printf(seq, "%u\n", intel_cqm_threshold * r->mon_scale);
917 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
918 char *buf, size_t nbytes, loff_t off)
920 struct rdt_resource *r = of->kn->parent->priv;
924 ret = kstrtouint(buf, 0, &bytes);
928 if (bytes > (boot_cpu_data.x86_cache_size * 1024))
931 intel_cqm_threshold = bytes / r->mon_scale;
937 * rdtgroup_mode_show - Display mode of this resource group
939 static int rdtgroup_mode_show(struct kernfs_open_file *of,
940 struct seq_file *s, void *v)
942 struct rdtgroup *rdtgrp;
944 rdtgrp = rdtgroup_kn_lock_live(of->kn);
946 rdtgroup_kn_unlock(of->kn);
950 seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode));
952 rdtgroup_kn_unlock(of->kn);
957 * rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other
958 * @r: Resource to which domain instance @d belongs.
959 * @d: The domain instance for which @closid is being tested.
960 * @cbm: Capacity bitmask being tested.
961 * @closid: Intended closid for @cbm.
962 * @exclusive: Only check if overlaps with exclusive resource groups
964 * Checks if provided @cbm intended to be used for @closid on domain
965 * @d overlaps with any other closids or other hardware usage associated
966 * with this domain. If @exclusive is true then only overlaps with
967 * resource groups in exclusive mode will be considered. If @exclusive
968 * is false then overlaps with any resource group or hardware entities
969 * will be considered.
971 * Return: false if CBM does not overlap, true if it does.
973 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
974 u32 _cbm, int closid, bool exclusive)
976 unsigned long *cbm = (unsigned long *)&_cbm;
977 unsigned long *ctrl_b;
978 enum rdtgrp_mode mode;
982 /* Check for any overlap with regions used by hardware directly */
984 if (bitmap_intersects(cbm,
985 (unsigned long *)&r->cache.shareable_bits,
990 /* Check for overlap with other resource groups */
992 for (i = 0; i < r->num_closid; i++, ctrl++) {
993 ctrl_b = (unsigned long *)ctrl;
994 mode = rdtgroup_mode_by_closid(i);
995 if (closid_allocated(i) && i != closid &&
996 mode != RDT_MODE_PSEUDO_LOCKSETUP) {
997 if (bitmap_intersects(cbm, ctrl_b, r->cache.cbm_len)) {
999 if (mode == RDT_MODE_EXCLUSIVE)
1012 * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive
1014 * An exclusive resource group implies that there should be no sharing of
1015 * its allocated resources. At the time this group is considered to be
1016 * exclusive this test can determine if its current schemata supports this
1017 * setting by testing for overlap with all other resource groups.
1019 * Return: true if resource group can be exclusive, false if there is overlap
1020 * with allocations of other resource groups and thus this resource group
1021 * cannot be exclusive.
1023 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp)
1025 int closid = rdtgrp->closid;
1026 struct rdt_resource *r;
1027 struct rdt_domain *d;
1029 for_each_alloc_enabled_rdt_resource(r) {
1030 list_for_each_entry(d, &r->domains, list) {
1031 if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid],
1032 rdtgrp->closid, false))
1041 * rdtgroup_mode_write - Modify the resource group's mode
1044 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of,
1045 char *buf, size_t nbytes, loff_t off)
1047 struct rdtgroup *rdtgrp;
1048 enum rdtgrp_mode mode;
1051 /* Valid input requires a trailing newline */
1052 if (nbytes == 0 || buf[nbytes - 1] != '\n')
1054 buf[nbytes - 1] = '\0';
1056 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1058 rdtgroup_kn_unlock(of->kn);
1062 rdt_last_cmd_clear();
1064 mode = rdtgrp->mode;
1066 if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) ||
1067 (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) ||
1068 (!strcmp(buf, "pseudo-locksetup") &&
1069 mode == RDT_MODE_PSEUDO_LOCKSETUP) ||
1070 (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED))
1073 if (mode == RDT_MODE_PSEUDO_LOCKED) {
1074 rdt_last_cmd_printf("cannot change pseudo-locked group\n");
1079 if (!strcmp(buf, "shareable")) {
1080 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1081 ret = rdtgroup_locksetup_exit(rdtgrp);
1085 rdtgrp->mode = RDT_MODE_SHAREABLE;
1086 } else if (!strcmp(buf, "exclusive")) {
1087 if (!rdtgroup_mode_test_exclusive(rdtgrp)) {
1088 rdt_last_cmd_printf("schemata overlaps\n");
1092 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1093 ret = rdtgroup_locksetup_exit(rdtgrp);
1097 rdtgrp->mode = RDT_MODE_EXCLUSIVE;
1098 } else if (!strcmp(buf, "pseudo-locksetup")) {
1099 ret = rdtgroup_locksetup_enter(rdtgrp);
1102 rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP;
1104 rdt_last_cmd_printf("unknown/unsupported mode\n");
1109 rdtgroup_kn_unlock(of->kn);
1110 return ret ?: nbytes;
1114 * rdtgroup_cbm_to_size - Translate CBM to size in bytes
1115 * @r: RDT resource to which @d belongs.
1116 * @d: RDT domain instance.
1117 * @cbm: bitmask for which the size should be computed.
1119 * The bitmask provided associated with the RDT domain instance @d will be
1120 * translated into how many bytes it represents. The size in bytes is
1121 * computed by first dividing the total cache size by the CBM length to
1122 * determine how many bytes each bit in the bitmask represents. The result
1123 * is multiplied with the number of bits set in the bitmask.
1125 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r,
1126 struct rdt_domain *d, u32 cbm)
1128 struct cpu_cacheinfo *ci;
1129 unsigned int size = 0;
1132 num_b = bitmap_weight((unsigned long *)&cbm, r->cache.cbm_len);
1133 ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask));
1134 for (i = 0; i < ci->num_leaves; i++) {
1135 if (ci->info_list[i].level == r->cache_level) {
1136 size = ci->info_list[i].size / r->cache.cbm_len * num_b;
1145 * rdtgroup_size_show - Display size in bytes of allocated regions
1147 * The "size" file mirrors the layout of the "schemata" file, printing the
1148 * size in bytes of each region instead of the capacity bitmask.
1151 static int rdtgroup_size_show(struct kernfs_open_file *of,
1152 struct seq_file *s, void *v)
1154 struct rdtgroup *rdtgrp;
1155 struct rdt_resource *r;
1156 struct rdt_domain *d;
1161 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1163 rdtgroup_kn_unlock(of->kn);
1167 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
1168 seq_printf(s, "%*s:", max_name_width, rdtgrp->plr->r->name);
1169 size = rdtgroup_cbm_to_size(rdtgrp->plr->r,
1172 seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size);
1176 for_each_alloc_enabled_rdt_resource(r) {
1177 seq_printf(s, "%*s:", max_name_width, r->name);
1178 list_for_each_entry(d, &r->domains, list) {
1181 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1184 cbm = d->ctrl_val[rdtgrp->closid];
1185 size = rdtgroup_cbm_to_size(r, d, cbm);
1187 seq_printf(s, "%d=%u", d->id, size);
1194 rdtgroup_kn_unlock(of->kn);
1199 /* rdtgroup information files for one cache resource. */
1200 static struct rftype res_common_files[] = {
1202 .name = "last_cmd_status",
1204 .kf_ops = &rdtgroup_kf_single_ops,
1205 .seq_show = rdt_last_cmd_status_show,
1206 .fflags = RF_TOP_INFO,
1209 .name = "num_closids",
1211 .kf_ops = &rdtgroup_kf_single_ops,
1212 .seq_show = rdt_num_closids_show,
1213 .fflags = RF_CTRL_INFO,
1216 .name = "mon_features",
1218 .kf_ops = &rdtgroup_kf_single_ops,
1219 .seq_show = rdt_mon_features_show,
1220 .fflags = RF_MON_INFO,
1223 .name = "num_rmids",
1225 .kf_ops = &rdtgroup_kf_single_ops,
1226 .seq_show = rdt_num_rmids_show,
1227 .fflags = RF_MON_INFO,
1232 .kf_ops = &rdtgroup_kf_single_ops,
1233 .seq_show = rdt_default_ctrl_show,
1234 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1237 .name = "min_cbm_bits",
1239 .kf_ops = &rdtgroup_kf_single_ops,
1240 .seq_show = rdt_min_cbm_bits_show,
1241 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1244 .name = "shareable_bits",
1246 .kf_ops = &rdtgroup_kf_single_ops,
1247 .seq_show = rdt_shareable_bits_show,
1248 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1251 .name = "bit_usage",
1253 .kf_ops = &rdtgroup_kf_single_ops,
1254 .seq_show = rdt_bit_usage_show,
1255 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1258 .name = "min_bandwidth",
1260 .kf_ops = &rdtgroup_kf_single_ops,
1261 .seq_show = rdt_min_bw_show,
1262 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1265 .name = "bandwidth_gran",
1267 .kf_ops = &rdtgroup_kf_single_ops,
1268 .seq_show = rdt_bw_gran_show,
1269 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1272 .name = "delay_linear",
1274 .kf_ops = &rdtgroup_kf_single_ops,
1275 .seq_show = rdt_delay_linear_show,
1276 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1279 .name = "max_threshold_occupancy",
1281 .kf_ops = &rdtgroup_kf_single_ops,
1282 .write = max_threshold_occ_write,
1283 .seq_show = max_threshold_occ_show,
1284 .fflags = RF_MON_INFO | RFTYPE_RES_CACHE,
1289 .kf_ops = &rdtgroup_kf_single_ops,
1290 .write = rdtgroup_cpus_write,
1291 .seq_show = rdtgroup_cpus_show,
1292 .fflags = RFTYPE_BASE,
1295 .name = "cpus_list",
1297 .kf_ops = &rdtgroup_kf_single_ops,
1298 .write = rdtgroup_cpus_write,
1299 .seq_show = rdtgroup_cpus_show,
1300 .flags = RFTYPE_FLAGS_CPUS_LIST,
1301 .fflags = RFTYPE_BASE,
1306 .kf_ops = &rdtgroup_kf_single_ops,
1307 .write = rdtgroup_tasks_write,
1308 .seq_show = rdtgroup_tasks_show,
1309 .fflags = RFTYPE_BASE,
1314 .kf_ops = &rdtgroup_kf_single_ops,
1315 .write = rdtgroup_schemata_write,
1316 .seq_show = rdtgroup_schemata_show,
1317 .fflags = RF_CTRL_BASE,
1322 .kf_ops = &rdtgroup_kf_single_ops,
1323 .write = rdtgroup_mode_write,
1324 .seq_show = rdtgroup_mode_show,
1325 .fflags = RF_CTRL_BASE,
1330 .kf_ops = &rdtgroup_kf_single_ops,
1331 .seq_show = rdtgroup_size_show,
1332 .fflags = RF_CTRL_BASE,
1337 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
1339 struct rftype *rfts, *rft;
1342 rfts = res_common_files;
1343 len = ARRAY_SIZE(res_common_files);
1345 lockdep_assert_held(&rdtgroup_mutex);
1347 for (rft = rfts; rft < rfts + len; rft++) {
1348 if ((fflags & rft->fflags) == rft->fflags) {
1349 ret = rdtgroup_add_file(kn, rft);
1357 pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
1358 while (--rft >= rfts) {
1359 if ((fflags & rft->fflags) == rft->fflags)
1360 kernfs_remove_by_name(kn, rft->name);
1366 * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file
1367 * @r: The resource group with which the file is associated.
1368 * @name: Name of the file
1370 * The permissions of named resctrl file, directory, or link are modified
1371 * to not allow read, write, or execute by any user.
1373 * WARNING: This function is intended to communicate to the user that the
1374 * resctrl file has been locked down - that it is not relevant to the
1375 * particular state the system finds itself in. It should not be relied
1376 * on to protect from user access because after the file's permissions
1377 * are restricted the user can still change the permissions using chmod
1378 * from the command line.
1380 * Return: 0 on success, <0 on failure.
1382 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name)
1384 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1385 struct kernfs_node *kn;
1388 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1392 switch (kernfs_type(kn)) {
1394 iattr.ia_mode = S_IFDIR;
1397 iattr.ia_mode = S_IFREG;
1400 iattr.ia_mode = S_IFLNK;
1404 ret = kernfs_setattr(kn, &iattr);
1410 * rdtgroup_kn_mode_restore - Restore user access to named resctrl file
1411 * @r: The resource group with which the file is associated.
1412 * @name: Name of the file
1413 * @mask: Mask of permissions that should be restored
1415 * Restore the permissions of the named file. If @name is a directory the
1416 * permissions of its parent will be used.
1418 * Return: 0 on success, <0 on failure.
1420 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
1423 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1424 struct kernfs_node *kn, *parent;
1425 struct rftype *rfts, *rft;
1428 rfts = res_common_files;
1429 len = ARRAY_SIZE(res_common_files);
1431 for (rft = rfts; rft < rfts + len; rft++) {
1432 if (!strcmp(rft->name, name))
1433 iattr.ia_mode = rft->mode & mask;
1436 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1440 switch (kernfs_type(kn)) {
1442 parent = kernfs_get_parent(kn);
1444 iattr.ia_mode |= parent->mode;
1447 iattr.ia_mode |= S_IFDIR;
1450 iattr.ia_mode |= S_IFREG;
1453 iattr.ia_mode |= S_IFLNK;
1457 ret = kernfs_setattr(kn, &iattr);
1462 static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name,
1463 unsigned long fflags)
1465 struct kernfs_node *kn_subdir;
1468 kn_subdir = kernfs_create_dir(kn_info, name,
1470 if (IS_ERR(kn_subdir))
1471 return PTR_ERR(kn_subdir);
1473 kernfs_get(kn_subdir);
1474 ret = rdtgroup_kn_set_ugid(kn_subdir);
1478 ret = rdtgroup_add_files(kn_subdir, fflags);
1480 kernfs_activate(kn_subdir);
1485 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
1487 struct rdt_resource *r;
1488 unsigned long fflags;
1492 /* create the directory */
1493 kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
1494 if (IS_ERR(kn_info))
1495 return PTR_ERR(kn_info);
1496 kernfs_get(kn_info);
1498 ret = rdtgroup_add_files(kn_info, RF_TOP_INFO);
1502 for_each_alloc_enabled_rdt_resource(r) {
1503 fflags = r->fflags | RF_CTRL_INFO;
1504 ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags);
1509 for_each_mon_enabled_rdt_resource(r) {
1510 fflags = r->fflags | RF_MON_INFO;
1511 sprintf(name, "%s_MON", r->name);
1512 ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
1518 * This extra ref will be put in kernfs_remove() and guarantees
1519 * that @rdtgrp->kn is always accessible.
1521 kernfs_get(kn_info);
1523 ret = rdtgroup_kn_set_ugid(kn_info);
1527 kernfs_activate(kn_info);
1532 kernfs_remove(kn_info);
1537 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
1538 char *name, struct kernfs_node **dest_kn)
1540 struct kernfs_node *kn;
1543 /* create the directory */
1544 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1552 * This extra ref will be put in kernfs_remove() and guarantees
1553 * that @rdtgrp->kn is always accessible.
1557 ret = rdtgroup_kn_set_ugid(kn);
1561 kernfs_activate(kn);
1570 static void l3_qos_cfg_update(void *arg)
1574 wrmsrl(IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL);
1577 static void l2_qos_cfg_update(void *arg)
1581 wrmsrl(IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL);
1584 static inline bool is_mba_linear(void)
1586 return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear;
1589 static int set_cache_qos_cfg(int level, bool enable)
1591 void (*update)(void *arg);
1592 struct rdt_resource *r_l;
1593 cpumask_var_t cpu_mask;
1594 struct rdt_domain *d;
1597 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1600 if (level == RDT_RESOURCE_L3)
1601 update = l3_qos_cfg_update;
1602 else if (level == RDT_RESOURCE_L2)
1603 update = l2_qos_cfg_update;
1607 r_l = &rdt_resources_all[level];
1608 list_for_each_entry(d, &r_l->domains, list) {
1609 /* Pick one CPU from each domain instance to update MSR */
1610 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1613 /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1614 if (cpumask_test_cpu(cpu, cpu_mask))
1616 /* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1617 smp_call_function_many(cpu_mask, update, &enable, 1);
1620 free_cpumask_var(cpu_mask);
1626 * Enable or disable the MBA software controller
1627 * which helps user specify bandwidth in MBps.
1628 * MBA software controller is supported only if
1629 * MBM is supported and MBA is in linear scale.
1631 static int set_mba_sc(bool mba_sc)
1633 struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA];
1634 struct rdt_domain *d;
1636 if (!is_mbm_enabled() || !is_mba_linear() ||
1637 mba_sc == is_mba_sc(r))
1640 r->membw.mba_sc = mba_sc;
1641 list_for_each_entry(d, &r->domains, list)
1642 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val);
1647 static int cdp_enable(int level, int data_type, int code_type)
1649 struct rdt_resource *r_ldata = &rdt_resources_all[data_type];
1650 struct rdt_resource *r_lcode = &rdt_resources_all[code_type];
1651 struct rdt_resource *r_l = &rdt_resources_all[level];
1654 if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1655 !r_lcode->alloc_capable)
1658 ret = set_cache_qos_cfg(level, true);
1660 r_l->alloc_enabled = false;
1661 r_ldata->alloc_enabled = true;
1662 r_lcode->alloc_enabled = true;
1667 static int cdpl3_enable(void)
1669 return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1670 RDT_RESOURCE_L3CODE);
1673 static int cdpl2_enable(void)
1675 return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1676 RDT_RESOURCE_L2CODE);
1679 static void cdp_disable(int level, int data_type, int code_type)
1681 struct rdt_resource *r = &rdt_resources_all[level];
1683 r->alloc_enabled = r->alloc_capable;
1685 if (rdt_resources_all[data_type].alloc_enabled) {
1686 rdt_resources_all[data_type].alloc_enabled = false;
1687 rdt_resources_all[code_type].alloc_enabled = false;
1688 set_cache_qos_cfg(level, false);
1692 static void cdpl3_disable(void)
1694 cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
1697 static void cdpl2_disable(void)
1699 cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
1702 static void cdp_disable_all(void)
1704 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
1706 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
1710 static int parse_rdtgroupfs_options(char *data)
1712 char *token, *o = data;
1715 while ((token = strsep(&o, ",")) != NULL) {
1721 if (!strcmp(token, "cdp")) {
1722 ret = cdpl3_enable();
1725 } else if (!strcmp(token, "cdpl2")) {
1726 ret = cdpl2_enable();
1729 } else if (!strcmp(token, "mba_MBps")) {
1730 ret = set_mba_sc(true);
1742 pr_err("Invalid mount option \"%s\"\n", token);
1748 * We don't allow rdtgroup directories to be created anywhere
1749 * except the root directory. Thus when looking for the rdtgroup
1750 * structure for a kernfs node we are either looking at a directory,
1751 * in which case the rdtgroup structure is pointed at by the "priv"
1752 * field, otherwise we have a file, and need only look to the parent
1753 * to find the rdtgroup.
1755 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
1757 if (kernfs_type(kn) == KERNFS_DIR) {
1759 * All the resource directories use "kn->priv"
1760 * to point to the "struct rdtgroup" for the
1761 * resource. "info" and its subdirectories don't
1762 * have rdtgroup structures, so return NULL here.
1764 if (kn == kn_info || kn->parent == kn_info)
1769 return kn->parent->priv;
1773 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
1775 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1780 atomic_inc(&rdtgrp->waitcount);
1781 kernfs_break_active_protection(kn);
1783 mutex_lock(&rdtgroup_mutex);
1785 /* Was this group deleted while we waited? */
1786 if (rdtgrp->flags & RDT_DELETED)
1792 void rdtgroup_kn_unlock(struct kernfs_node *kn)
1794 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1799 mutex_unlock(&rdtgroup_mutex);
1801 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
1802 (rdtgrp->flags & RDT_DELETED)) {
1803 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
1804 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
1805 rdtgroup_pseudo_lock_remove(rdtgrp);
1806 kernfs_unbreak_active_protection(kn);
1807 kernfs_put(rdtgrp->kn);
1810 kernfs_unbreak_active_protection(kn);
1814 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1815 struct rdtgroup *prgrp,
1816 struct kernfs_node **mon_data_kn);
1818 static struct dentry *rdt_mount(struct file_system_type *fs_type,
1819 int flags, const char *unused_dev_name,
1822 struct rdt_domain *dom;
1823 struct rdt_resource *r;
1824 struct dentry *dentry;
1828 mutex_lock(&rdtgroup_mutex);
1830 * resctrl file system can only be mounted once.
1832 if (static_branch_unlikely(&rdt_enable_key)) {
1833 dentry = ERR_PTR(-EBUSY);
1837 ret = parse_rdtgroupfs_options(data);
1839 dentry = ERR_PTR(ret);
1845 ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
1847 dentry = ERR_PTR(ret);
1851 if (rdt_mon_capable) {
1852 ret = mongroup_create_dir(rdtgroup_default.kn,
1856 dentry = ERR_PTR(ret);
1859 kernfs_get(kn_mongrp);
1861 ret = mkdir_mondata_all(rdtgroup_default.kn,
1862 &rdtgroup_default, &kn_mondata);
1864 dentry = ERR_PTR(ret);
1867 kernfs_get(kn_mondata);
1868 rdtgroup_default.mon.mon_data_kn = kn_mondata;
1871 ret = rdt_pseudo_lock_init();
1873 dentry = ERR_PTR(ret);
1877 dentry = kernfs_mount(fs_type, flags, rdt_root,
1878 RDTGROUP_SUPER_MAGIC, NULL);
1882 if (rdt_alloc_capable)
1883 static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
1884 if (rdt_mon_capable)
1885 static_branch_enable_cpuslocked(&rdt_mon_enable_key);
1887 if (rdt_alloc_capable || rdt_mon_capable)
1888 static_branch_enable_cpuslocked(&rdt_enable_key);
1890 if (is_mbm_enabled()) {
1891 r = &rdt_resources_all[RDT_RESOURCE_L3];
1892 list_for_each_entry(dom, &r->domains, list)
1893 mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
1899 rdt_pseudo_lock_release();
1901 if (rdt_mon_capable)
1902 kernfs_remove(kn_mondata);
1904 if (rdt_mon_capable)
1905 kernfs_remove(kn_mongrp);
1907 kernfs_remove(kn_info);
1911 rdt_last_cmd_clear();
1912 mutex_unlock(&rdtgroup_mutex);
1918 static int reset_all_ctrls(struct rdt_resource *r)
1920 struct msr_param msr_param;
1921 cpumask_var_t cpu_mask;
1922 struct rdt_domain *d;
1925 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1930 msr_param.high = r->num_closid;
1933 * Disable resource control for this resource by setting all
1934 * CBMs in all domains to the maximum mask value. Pick one CPU
1935 * from each domain to update the MSRs below.
1937 list_for_each_entry(d, &r->domains, list) {
1938 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1940 for (i = 0; i < r->num_closid; i++)
1941 d->ctrl_val[i] = r->default_ctrl;
1944 /* Update CBM on this cpu if it's in cpu_mask. */
1945 if (cpumask_test_cpu(cpu, cpu_mask))
1946 rdt_ctrl_update(&msr_param);
1947 /* Update CBM on all other cpus in cpu_mask. */
1948 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
1951 free_cpumask_var(cpu_mask);
1956 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
1958 return (rdt_alloc_capable &&
1959 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
1962 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
1964 return (rdt_mon_capable &&
1965 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
1969 * Move tasks from one to the other group. If @from is NULL, then all tasks
1970 * in the systems are moved unconditionally (used for teardown).
1972 * If @mask is not NULL the cpus on which moved tasks are running are set
1973 * in that mask so the update smp function call is restricted to affected
1976 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
1977 struct cpumask *mask)
1979 struct task_struct *p, *t;
1981 read_lock(&tasklist_lock);
1982 for_each_process_thread(p, t) {
1983 if (!from || is_closid_match(t, from) ||
1984 is_rmid_match(t, from)) {
1985 t->closid = to->closid;
1986 t->rmid = to->mon.rmid;
1990 * This is safe on x86 w/o barriers as the ordering
1991 * of writing to task_cpu() and t->on_cpu is
1992 * reverse to the reading here. The detection is
1993 * inaccurate as tasks might move or schedule
1994 * before the smp function call takes place. In
1995 * such a case the function call is pointless, but
1996 * there is no other side effect.
1998 if (mask && t->on_cpu)
1999 cpumask_set_cpu(task_cpu(t), mask);
2003 read_unlock(&tasklist_lock);
2006 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
2008 struct rdtgroup *sentry, *stmp;
2009 struct list_head *head;
2011 head = &rdtgrp->mon.crdtgrp_list;
2012 list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
2013 free_rmid(sentry->mon.rmid);
2014 list_del(&sentry->mon.crdtgrp_list);
2020 * Forcibly remove all of subdirectories under root.
2022 static void rmdir_all_sub(void)
2024 struct rdtgroup *rdtgrp, *tmp;
2026 /* Move all tasks to the default resource group */
2027 rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
2029 list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
2030 /* Free any child rmids */
2031 free_all_child_rdtgrp(rdtgrp);
2033 /* Remove each rdtgroup other than root */
2034 if (rdtgrp == &rdtgroup_default)
2037 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2038 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2039 rdtgroup_pseudo_lock_remove(rdtgrp);
2042 * Give any CPUs back to the default group. We cannot copy
2043 * cpu_online_mask because a CPU might have executed the
2044 * offline callback already, but is still marked online.
2046 cpumask_or(&rdtgroup_default.cpu_mask,
2047 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2049 free_rmid(rdtgrp->mon.rmid);
2051 kernfs_remove(rdtgrp->kn);
2052 list_del(&rdtgrp->rdtgroup_list);
2055 /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
2056 update_closid_rmid(cpu_online_mask, &rdtgroup_default);
2058 kernfs_remove(kn_info);
2059 kernfs_remove(kn_mongrp);
2060 kernfs_remove(kn_mondata);
2063 static void rdt_kill_sb(struct super_block *sb)
2065 struct rdt_resource *r;
2068 mutex_lock(&rdtgroup_mutex);
2072 /*Put everything back to default values. */
2073 for_each_alloc_enabled_rdt_resource(r)
2077 rdt_pseudo_lock_release();
2078 rdtgroup_default.mode = RDT_MODE_SHAREABLE;
2079 static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
2080 static_branch_disable_cpuslocked(&rdt_mon_enable_key);
2081 static_branch_disable_cpuslocked(&rdt_enable_key);
2083 mutex_unlock(&rdtgroup_mutex);
2087 static struct file_system_type rdt_fs_type = {
2090 .kill_sb = rdt_kill_sb,
2093 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
2096 struct kernfs_node *kn;
2099 kn = __kernfs_create_file(parent_kn, name, 0444,
2100 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
2101 &kf_mondata_ops, priv, NULL, NULL);
2105 ret = rdtgroup_kn_set_ugid(kn);
2115 * Remove all subdirectories of mon_data of ctrl_mon groups
2116 * and monitor groups with given domain id.
2118 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
2120 struct rdtgroup *prgrp, *crgrp;
2123 if (!r->mon_enabled)
2126 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2127 sprintf(name, "mon_%s_%02d", r->name, dom_id);
2128 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
2130 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
2131 kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
2135 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
2136 struct rdt_domain *d,
2137 struct rdt_resource *r, struct rdtgroup *prgrp)
2139 union mon_data_bits priv;
2140 struct kernfs_node *kn;
2141 struct mon_evt *mevt;
2142 struct rmid_read rr;
2146 sprintf(name, "mon_%s_%02d", r->name, d->id);
2147 /* create the directory */
2148 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
2153 * This extra ref will be put in kernfs_remove() and guarantees
2154 * that kn is always accessible.
2157 ret = rdtgroup_kn_set_ugid(kn);
2161 if (WARN_ON(list_empty(&r->evt_list))) {
2166 priv.u.rid = r->rid;
2167 priv.u.domid = d->id;
2168 list_for_each_entry(mevt, &r->evt_list, list) {
2169 priv.u.evtid = mevt->evtid;
2170 ret = mon_addfile(kn, mevt->name, priv.priv);
2174 if (is_mbm_event(mevt->evtid))
2175 mon_event_read(&rr, d, prgrp, mevt->evtid, true);
2177 kernfs_activate(kn);
2186 * Add all subdirectories of mon_data for "ctrl_mon" groups
2187 * and "monitor" groups with given domain id.
2189 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
2190 struct rdt_domain *d)
2192 struct kernfs_node *parent_kn;
2193 struct rdtgroup *prgrp, *crgrp;
2194 struct list_head *head;
2196 if (!r->mon_enabled)
2199 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2200 parent_kn = prgrp->mon.mon_data_kn;
2201 mkdir_mondata_subdir(parent_kn, d, r, prgrp);
2203 head = &prgrp->mon.crdtgrp_list;
2204 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
2205 parent_kn = crgrp->mon.mon_data_kn;
2206 mkdir_mondata_subdir(parent_kn, d, r, crgrp);
2211 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
2212 struct rdt_resource *r,
2213 struct rdtgroup *prgrp)
2215 struct rdt_domain *dom;
2218 list_for_each_entry(dom, &r->domains, list) {
2219 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
2228 * This creates a directory mon_data which contains the monitored data.
2230 * mon_data has one directory for each domain whic are named
2231 * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
2232 * with L3 domain looks as below:
2239 * Each domain directory has one file per event:
2244 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2245 struct rdtgroup *prgrp,
2246 struct kernfs_node **dest_kn)
2248 struct rdt_resource *r;
2249 struct kernfs_node *kn;
2253 * Create the mon_data directory first.
2255 ret = mongroup_create_dir(parent_kn, NULL, "mon_data", &kn);
2263 * Create the subdirectories for each domain. Note that all events
2264 * in a domain like L3 are grouped into a resource whose domain is L3
2266 for_each_mon_enabled_rdt_resource(r) {
2267 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
2280 * cbm_ensure_valid - Enforce validity on provided CBM
2281 * @_val: Candidate CBM
2282 * @r: RDT resource to which the CBM belongs
2284 * The provided CBM represents all cache portions available for use. This
2285 * may be represented by a bitmap that does not consist of contiguous ones
2286 * and thus be an invalid CBM.
2287 * Here the provided CBM is forced to be a valid CBM by only considering
2288 * the first set of contiguous bits as valid and clearing all bits.
2289 * The intention here is to provide a valid default CBM with which a new
2290 * resource group is initialized. The user can follow this with a
2291 * modification to the CBM if the default does not satisfy the
2294 static void cbm_ensure_valid(u32 *_val, struct rdt_resource *r)
2297 * Convert the u32 _val to an unsigned long required by all the bit
2298 * operations within this function. No more than 32 bits of this
2299 * converted value can be accessed because all bit operations are
2300 * additionally provided with cbm_len that is initialized during
2301 * hardware enumeration using five bits from the EAX register and
2302 * thus never can exceed 32 bits.
2304 unsigned long *val = (unsigned long *)_val;
2305 unsigned int cbm_len = r->cache.cbm_len;
2306 unsigned long first_bit, zero_bit;
2311 first_bit = find_first_bit(val, cbm_len);
2312 zero_bit = find_next_zero_bit(val, cbm_len, first_bit);
2314 /* Clear any remaining bits to ensure contiguous region */
2315 bitmap_clear(val, zero_bit, cbm_len - zero_bit);
2319 * rdtgroup_init_alloc - Initialize the new RDT group's allocations
2321 * A new RDT group is being created on an allocation capable (CAT)
2322 * supporting system. Set this group up to start off with all usable
2323 * allocations. That is, all shareable and unused bits.
2325 * All-zero CBM is invalid. If there are no more shareable bits available
2326 * on any domain then the entire allocation will fail.
2328 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
2330 u32 used_b = 0, unused_b = 0;
2331 u32 closid = rdtgrp->closid;
2332 struct rdt_resource *r;
2333 enum rdtgrp_mode mode;
2334 struct rdt_domain *d;
2338 for_each_alloc_enabled_rdt_resource(r) {
2339 list_for_each_entry(d, &r->domains, list) {
2340 d->have_new_ctrl = false;
2341 d->new_ctrl = r->cache.shareable_bits;
2342 used_b = r->cache.shareable_bits;
2344 for (i = 0; i < r->num_closid; i++, ctrl++) {
2345 if (closid_allocated(i) && i != closid) {
2346 mode = rdtgroup_mode_by_closid(i);
2347 if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
2350 if (mode == RDT_MODE_SHAREABLE)
2351 d->new_ctrl |= *ctrl;
2354 if (d->plr && d->plr->cbm > 0)
2355 used_b |= d->plr->cbm;
2356 unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
2357 unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
2358 d->new_ctrl |= unused_b;
2360 * Force the initial CBM to be valid, user can
2361 * modify the CBM based on system availability.
2363 cbm_ensure_valid(&d->new_ctrl, r);
2364 if (bitmap_weight((unsigned long *) &d->new_ctrl,
2366 r->cache.min_cbm_bits) {
2367 rdt_last_cmd_printf("no space on %s:%d\n",
2371 d->have_new_ctrl = true;
2375 for_each_alloc_enabled_rdt_resource(r) {
2376 ret = update_domains(r, rdtgrp->closid);
2378 rdt_last_cmd_puts("failed to initialize allocations\n");
2381 rdtgrp->mode = RDT_MODE_SHAREABLE;
2387 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
2388 struct kernfs_node *prgrp_kn,
2389 const char *name, umode_t mode,
2390 enum rdt_group_type rtype, struct rdtgroup **r)
2392 struct rdtgroup *prdtgrp, *rdtgrp;
2393 struct kernfs_node *kn;
2397 prdtgrp = rdtgroup_kn_lock_live(prgrp_kn);
2398 rdt_last_cmd_clear();
2401 rdt_last_cmd_puts("directory was removed\n");
2405 if (rtype == RDTMON_GROUP &&
2406 (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2407 prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
2409 rdt_last_cmd_puts("pseudo-locking in progress\n");
2413 /* allocate the rdtgroup. */
2414 rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
2417 rdt_last_cmd_puts("kernel out of memory\n");
2421 rdtgrp->mon.parent = prdtgrp;
2422 rdtgrp->type = rtype;
2423 INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
2425 /* kernfs creates the directory for rdtgrp */
2426 kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
2429 rdt_last_cmd_puts("kernfs create error\n");
2435 * kernfs_remove() will drop the reference count on "kn" which
2436 * will free it. But we still need it to stick around for the
2437 * rdtgroup_kn_unlock(kn} call below. Take one extra reference
2438 * here, which will be dropped inside rdtgroup_kn_unlock().
2442 ret = rdtgroup_kn_set_ugid(kn);
2444 rdt_last_cmd_puts("kernfs perm error\n");
2448 files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
2449 ret = rdtgroup_add_files(kn, files);
2451 rdt_last_cmd_puts("kernfs fill error\n");
2455 if (rdt_mon_capable) {
2458 rdt_last_cmd_puts("out of RMIDs\n");
2461 rdtgrp->mon.rmid = ret;
2463 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
2465 rdt_last_cmd_puts("kernfs subdir error\n");
2469 kernfs_activate(kn);
2472 * The caller unlocks the prgrp_kn upon success.
2477 free_rmid(rdtgrp->mon.rmid);
2479 kernfs_remove(rdtgrp->kn);
2483 rdtgroup_kn_unlock(prgrp_kn);
2487 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
2489 kernfs_remove(rgrp->kn);
2490 free_rmid(rgrp->mon.rmid);
2495 * Create a monitor group under "mon_groups" directory of a control
2496 * and monitor group(ctrl_mon). This is a resource group
2497 * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
2499 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
2500 struct kernfs_node *prgrp_kn,
2504 struct rdtgroup *rdtgrp, *prgrp;
2507 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP,
2512 prgrp = rdtgrp->mon.parent;
2513 rdtgrp->closid = prgrp->closid;
2516 * Add the rdtgrp to the list of rdtgrps the parent
2517 * ctrl_mon group has to track.
2519 list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
2521 rdtgroup_kn_unlock(prgrp_kn);
2526 * These are rdtgroups created under the root directory. Can be used
2527 * to allocate and monitor resources.
2529 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
2530 struct kernfs_node *prgrp_kn,
2531 const char *name, umode_t mode)
2533 struct rdtgroup *rdtgrp;
2534 struct kernfs_node *kn;
2538 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP,
2544 ret = closid_alloc();
2546 rdt_last_cmd_puts("out of CLOSIDs\n");
2547 goto out_common_fail;
2552 rdtgrp->closid = closid;
2553 ret = rdtgroup_init_alloc(rdtgrp);
2557 list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
2559 if (rdt_mon_capable) {
2561 * Create an empty mon_groups directory to hold the subset
2562 * of tasks and cpus to monitor.
2564 ret = mongroup_create_dir(kn, NULL, "mon_groups", NULL);
2566 rdt_last_cmd_puts("kernfs subdir error\n");
2574 list_del(&rdtgrp->rdtgroup_list);
2576 closid_free(closid);
2578 mkdir_rdt_prepare_clean(rdtgrp);
2580 rdtgroup_kn_unlock(prgrp_kn);
2585 * We allow creating mon groups only with in a directory called "mon_groups"
2586 * which is present in every ctrl_mon group. Check if this is a valid
2587 * "mon_groups" directory.
2589 * 1. The directory should be named "mon_groups".
2590 * 2. The mon group itself should "not" be named "mon_groups".
2591 * This makes sure "mon_groups" directory always has a ctrl_mon group
2594 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
2596 return (!strcmp(kn->name, "mon_groups") &&
2597 strcmp(name, "mon_groups"));
2600 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
2603 /* Do not accept '\n' to avoid unparsable situation. */
2604 if (strchr(name, '\n'))
2608 * If the parent directory is the root directory and RDT
2609 * allocation is supported, add a control and monitoring
2612 if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
2613 return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode);
2616 * If RDT monitoring is supported and the parent directory is a valid
2617 * "mon_groups" directory, add a monitoring subdirectory.
2619 if (rdt_mon_capable && is_mon_groups(parent_kn, name))
2620 return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode);
2625 static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2626 cpumask_var_t tmpmask)
2628 struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
2631 /* Give any tasks back to the parent group */
2632 rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
2634 /* Update per cpu rmid of the moved CPUs first */
2635 for_each_cpu(cpu, &rdtgrp->cpu_mask)
2636 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
2638 * Update the MSR on moved CPUs and CPUs which have moved
2639 * task running on them.
2641 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2642 update_closid_rmid(tmpmask, NULL);
2644 rdtgrp->flags = RDT_DELETED;
2645 free_rmid(rdtgrp->mon.rmid);
2648 * Remove the rdtgrp from the parent ctrl_mon group's list
2650 WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
2651 list_del(&rdtgrp->mon.crdtgrp_list);
2654 * one extra hold on this, will drop when we kfree(rdtgrp)
2655 * in rdtgroup_kn_unlock()
2658 kernfs_remove(rdtgrp->kn);
2663 static int rdtgroup_ctrl_remove(struct kernfs_node *kn,
2664 struct rdtgroup *rdtgrp)
2666 rdtgrp->flags = RDT_DELETED;
2667 list_del(&rdtgrp->rdtgroup_list);
2670 * one extra hold on this, will drop when we kfree(rdtgrp)
2671 * in rdtgroup_kn_unlock()
2674 kernfs_remove(rdtgrp->kn);
2678 static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2679 cpumask_var_t tmpmask)
2683 /* Give any tasks back to the default group */
2684 rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
2686 /* Give any CPUs back to the default group */
2687 cpumask_or(&rdtgroup_default.cpu_mask,
2688 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2690 /* Update per cpu closid and rmid of the moved CPUs first */
2691 for_each_cpu(cpu, &rdtgrp->cpu_mask) {
2692 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
2693 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
2697 * Update the MSR on moved CPUs and CPUs which have moved
2698 * task running on them.
2700 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2701 update_closid_rmid(tmpmask, NULL);
2703 closid_free(rdtgrp->closid);
2704 free_rmid(rdtgrp->mon.rmid);
2707 * Free all the child monitor group rmids.
2709 free_all_child_rdtgrp(rdtgrp);
2711 rdtgroup_ctrl_remove(kn, rdtgrp);
2716 static int rdtgroup_rmdir(struct kernfs_node *kn)
2718 struct kernfs_node *parent_kn = kn->parent;
2719 struct rdtgroup *rdtgrp;
2720 cpumask_var_t tmpmask;
2723 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
2726 rdtgrp = rdtgroup_kn_lock_live(kn);
2733 * If the rdtgroup is a ctrl_mon group and parent directory
2734 * is the root directory, remove the ctrl_mon group.
2736 * If the rdtgroup is a mon group and parent directory
2737 * is a valid "mon_groups" directory, remove the mon group.
2739 if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn) {
2740 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2741 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
2742 ret = rdtgroup_ctrl_remove(kn, rdtgrp);
2744 ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask);
2746 } else if (rdtgrp->type == RDTMON_GROUP &&
2747 is_mon_groups(parent_kn, kn->name)) {
2748 ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask);
2754 rdtgroup_kn_unlock(kn);
2755 free_cpumask_var(tmpmask);
2759 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
2761 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
2762 seq_puts(seq, ",cdp");
2766 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
2767 .mkdir = rdtgroup_mkdir,
2768 .rmdir = rdtgroup_rmdir,
2769 .show_options = rdtgroup_show_options,
2772 static int __init rdtgroup_setup_root(void)
2776 rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
2777 KERNFS_ROOT_CREATE_DEACTIVATED |
2778 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
2780 if (IS_ERR(rdt_root))
2781 return PTR_ERR(rdt_root);
2783 mutex_lock(&rdtgroup_mutex);
2785 rdtgroup_default.closid = 0;
2786 rdtgroup_default.mon.rmid = 0;
2787 rdtgroup_default.type = RDTCTRL_GROUP;
2788 INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
2790 list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
2792 ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
2794 kernfs_destroy_root(rdt_root);
2798 rdtgroup_default.kn = rdt_root->kn;
2799 kernfs_activate(rdtgroup_default.kn);
2802 mutex_unlock(&rdtgroup_mutex);
2808 * rdtgroup_init - rdtgroup initialization
2810 * Setup resctrl file system including set up root, create mount point,
2811 * register rdtgroup filesystem, and initialize files under root directory.
2813 * Return: 0 on success or -errno
2815 int __init rdtgroup_init(void)
2819 seq_buf_init(&last_cmd_status, last_cmd_status_buf,
2820 sizeof(last_cmd_status_buf));
2822 ret = rdtgroup_setup_root();
2826 ret = sysfs_create_mount_point(fs_kobj, "resctrl");
2830 ret = register_filesystem(&rdt_fs_type);
2832 goto cleanup_mountpoint;
2835 * Adding the resctrl debugfs directory here may not be ideal since
2836 * it would let the resctrl debugfs directory appear on the debugfs
2837 * filesystem before the resctrl filesystem is mounted.
2838 * It may also be ok since that would enable debugging of RDT before
2839 * resctrl is mounted.
2840 * The reason why the debugfs directory is created here and not in
2841 * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and
2842 * during the debugfs directory creation also &sb->s_type->i_mutex_key
2843 * (the lockdep class of inode->i_rwsem). Other filesystem
2844 * interactions (eg. SyS_getdents) have the lock ordering:
2845 * &sb->s_type->i_mutex_key --> &mm->mmap_sem
2846 * During mmap(), called with &mm->mmap_sem, the rdtgroup_mutex
2847 * is taken, thus creating dependency:
2848 * &mm->mmap_sem --> rdtgroup_mutex for the latter that can cause
2849 * issues considering the other two lock dependencies.
2850 * By creating the debugfs directory here we avoid a dependency
2851 * that may cause deadlock (even though file operations cannot
2852 * occur until the filesystem is mounted, but I do not know how to
2853 * tell lockdep that).
2855 debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
2860 sysfs_remove_mount_point(fs_kobj, "resctrl");
2862 kernfs_destroy_root(rdt_root);
2867 void __exit rdtgroup_exit(void)
2869 debugfs_remove_recursive(debugfs_resctrl);
2870 unregister_filesystem(&rdt_fs_type);
2871 sysfs_remove_mount_point(fs_kobj, "resctrl");
2872 kernfs_destroy_root(rdt_root);