1 // SPDX-License-Identifier: GPL-2.0-only
3 * User interface for Resource Alloction 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_sched.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 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 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 proteced 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;
510 struct task_move_callback {
511 struct callback_head work;
512 struct rdtgroup *rdtgrp;
515 static void move_myself(struct callback_head *head)
517 struct task_move_callback *callback;
518 struct rdtgroup *rdtgrp;
520 callback = container_of(head, struct task_move_callback, work);
521 rdtgrp = callback->rdtgrp;
524 * If resource group was deleted before this task work callback
525 * was invoked, then assign the task to root group and free the
528 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
529 (rdtgrp->flags & RDT_DELETED)) {
536 /* update PQR_ASSOC MSR to make resource group go into effect */
543 static int __rdtgroup_move_task(struct task_struct *tsk,
544 struct rdtgroup *rdtgrp)
546 struct task_move_callback *callback;
549 callback = kzalloc(sizeof(*callback), GFP_KERNEL);
552 callback->work.func = move_myself;
553 callback->rdtgrp = rdtgrp;
556 * Take a refcount, so rdtgrp cannot be freed before the
557 * callback has been invoked.
559 atomic_inc(&rdtgrp->waitcount);
560 ret = task_work_add(tsk, &callback->work, true);
563 * Task is exiting. Drop the refcount and free the callback.
564 * No need to check the refcount as the group cannot be
565 * deleted before the write function unlocks rdtgroup_mutex.
567 atomic_dec(&rdtgrp->waitcount);
569 rdt_last_cmd_puts("Task exited\n");
572 * For ctrl_mon groups move both closid and rmid.
573 * For monitor groups, can move the tasks only from
574 * their parent CTRL group.
576 if (rdtgrp->type == RDTCTRL_GROUP) {
577 tsk->closid = rdtgrp->closid;
578 tsk->rmid = rdtgrp->mon.rmid;
579 } else if (rdtgrp->type == RDTMON_GROUP) {
580 if (rdtgrp->mon.parent->closid == tsk->closid) {
581 tsk->rmid = rdtgrp->mon.rmid;
583 rdt_last_cmd_puts("Can't move task to different control group\n");
592 * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group
595 * Return: 1 if tasks have been assigned to @r, 0 otherwise
597 int rdtgroup_tasks_assigned(struct rdtgroup *r)
599 struct task_struct *p, *t;
602 lockdep_assert_held(&rdtgroup_mutex);
605 for_each_process_thread(p, t) {
606 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
607 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) {
617 static int rdtgroup_task_write_permission(struct task_struct *task,
618 struct kernfs_open_file *of)
620 const struct cred *tcred = get_task_cred(task);
621 const struct cred *cred = current_cred();
625 * Even if we're attaching all tasks in the thread group, we only
626 * need to check permissions on one of them.
628 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
629 !uid_eq(cred->euid, tcred->uid) &&
630 !uid_eq(cred->euid, tcred->suid)) {
631 rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
639 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
640 struct kernfs_open_file *of)
642 struct task_struct *tsk;
647 tsk = find_task_by_vpid(pid);
650 rdt_last_cmd_printf("No task %d\n", pid);
657 get_task_struct(tsk);
660 ret = rdtgroup_task_write_permission(tsk, of);
662 ret = __rdtgroup_move_task(tsk, rdtgrp);
664 put_task_struct(tsk);
668 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
669 char *buf, size_t nbytes, loff_t off)
671 struct rdtgroup *rdtgrp;
675 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
677 rdtgrp = rdtgroup_kn_lock_live(of->kn);
679 rdtgroup_kn_unlock(of->kn);
682 rdt_last_cmd_clear();
684 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
685 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
687 rdt_last_cmd_puts("Pseudo-locking in progress\n");
691 ret = rdtgroup_move_task(pid, rdtgrp, of);
694 rdtgroup_kn_unlock(of->kn);
696 return ret ?: nbytes;
699 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
701 struct task_struct *p, *t;
704 for_each_process_thread(p, t) {
705 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
706 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid))
707 seq_printf(s, "%d\n", t->pid);
712 static int rdtgroup_tasks_show(struct kernfs_open_file *of,
713 struct seq_file *s, void *v)
715 struct rdtgroup *rdtgrp;
718 rdtgrp = rdtgroup_kn_lock_live(of->kn);
720 show_rdt_tasks(rdtgrp, s);
723 rdtgroup_kn_unlock(of->kn);
728 static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
729 struct seq_file *seq, void *v)
733 mutex_lock(&rdtgroup_mutex);
734 len = seq_buf_used(&last_cmd_status);
736 seq_printf(seq, "%.*s", len, last_cmd_status_buf);
738 seq_puts(seq, "ok\n");
739 mutex_unlock(&rdtgroup_mutex);
743 static int rdt_num_closids_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, "%d\n", r->num_closid);
752 static int rdt_default_ctrl_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, "%x\n", r->default_ctrl);
761 static int rdt_min_cbm_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, "%u\n", r->cache.min_cbm_bits);
770 static int rdt_shareable_bits_show(struct kernfs_open_file *of,
771 struct seq_file *seq, void *v)
773 struct rdt_resource *r = of->kn->parent->priv;
775 seq_printf(seq, "%x\n", r->cache.shareable_bits);
780 * rdt_bit_usage_show - Display current usage of resources
782 * A domain is a shared resource that can now be allocated differently. Here
783 * we display the current regions of the domain as an annotated bitmask.
784 * For each domain of this resource its allocation bitmask
785 * is annotated as below to indicate the current usage of the corresponding bit:
786 * 0 - currently unused
787 * X - currently available for sharing and used by software and hardware
788 * H - currently used by hardware only but available for software use
789 * S - currently used and shareable by software only
790 * E - currently used exclusively by one resource group
791 * P - currently pseudo-locked by one resource group
793 static int rdt_bit_usage_show(struct kernfs_open_file *of,
794 struct seq_file *seq, void *v)
796 struct rdt_resource *r = of->kn->parent->priv;
798 * Use unsigned long even though only 32 bits are used to ensure
799 * test_bit() is used safely.
801 unsigned long sw_shareable = 0, hw_shareable = 0;
802 unsigned long exclusive = 0, pseudo_locked = 0;
803 struct rdt_domain *dom;
804 int i, hwb, swb, excl, psl;
805 enum rdtgrp_mode mode;
809 mutex_lock(&rdtgroup_mutex);
810 hw_shareable = r->cache.shareable_bits;
811 list_for_each_entry(dom, &r->domains, list) {
814 ctrl = dom->ctrl_val;
817 seq_printf(seq, "%d=", dom->id);
818 for (i = 0; i < closids_supported(); i++, ctrl++) {
819 if (!closid_allocated(i))
821 mode = rdtgroup_mode_by_closid(i);
823 case RDT_MODE_SHAREABLE:
824 sw_shareable |= *ctrl;
826 case RDT_MODE_EXCLUSIVE:
829 case RDT_MODE_PSEUDO_LOCKSETUP:
831 * RDT_MODE_PSEUDO_LOCKSETUP is possible
832 * here but not included since the CBM
833 * associated with this CLOSID in this mode
834 * is not initialized and no task or cpu can be
835 * assigned this CLOSID.
838 case RDT_MODE_PSEUDO_LOCKED:
841 "invalid mode for closid %d\n", i);
845 for (i = r->cache.cbm_len - 1; i >= 0; i--) {
846 pseudo_locked = dom->plr ? dom->plr->cbm : 0;
847 hwb = test_bit(i, &hw_shareable);
848 swb = test_bit(i, &sw_shareable);
849 excl = test_bit(i, &exclusive);
850 psl = test_bit(i, &pseudo_locked);
853 else if (hwb && !swb)
855 else if (!hwb && swb)
861 else /* Unused bits remain */
867 mutex_unlock(&rdtgroup_mutex);
871 static int rdt_min_bw_show(struct kernfs_open_file *of,
872 struct seq_file *seq, void *v)
874 struct rdt_resource *r = of->kn->parent->priv;
876 seq_printf(seq, "%u\n", r->membw.min_bw);
880 static int rdt_num_rmids_show(struct kernfs_open_file *of,
881 struct seq_file *seq, void *v)
883 struct rdt_resource *r = of->kn->parent->priv;
885 seq_printf(seq, "%d\n", r->num_rmid);
890 static int rdt_mon_features_show(struct kernfs_open_file *of,
891 struct seq_file *seq, void *v)
893 struct rdt_resource *r = of->kn->parent->priv;
894 struct mon_evt *mevt;
896 list_for_each_entry(mevt, &r->evt_list, list)
897 seq_printf(seq, "%s\n", mevt->name);
902 static int rdt_bw_gran_show(struct kernfs_open_file *of,
903 struct seq_file *seq, void *v)
905 struct rdt_resource *r = of->kn->parent->priv;
907 seq_printf(seq, "%u\n", r->membw.bw_gran);
911 static int rdt_delay_linear_show(struct kernfs_open_file *of,
912 struct seq_file *seq, void *v)
914 struct rdt_resource *r = of->kn->parent->priv;
916 seq_printf(seq, "%u\n", r->membw.delay_linear);
920 static int max_threshold_occ_show(struct kernfs_open_file *of,
921 struct seq_file *seq, void *v)
923 struct rdt_resource *r = of->kn->parent->priv;
925 seq_printf(seq, "%u\n", resctrl_cqm_threshold * r->mon_scale);
930 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
931 char *buf, size_t nbytes, loff_t off)
933 struct rdt_resource *r = of->kn->parent->priv;
937 ret = kstrtouint(buf, 0, &bytes);
941 if (bytes > (boot_cpu_data.x86_cache_size * 1024))
944 resctrl_cqm_threshold = bytes / r->mon_scale;
950 * rdtgroup_mode_show - Display mode of this resource group
952 static int rdtgroup_mode_show(struct kernfs_open_file *of,
953 struct seq_file *s, void *v)
955 struct rdtgroup *rdtgrp;
957 rdtgrp = rdtgroup_kn_lock_live(of->kn);
959 rdtgroup_kn_unlock(of->kn);
963 seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode));
965 rdtgroup_kn_unlock(of->kn);
970 * rdt_cdp_peer_get - Retrieve CDP peer if it exists
971 * @r: RDT resource to which RDT domain @d belongs
972 * @d: Cache instance for which a CDP peer is requested
973 * @r_cdp: RDT resource that shares hardware with @r (RDT resource peer)
974 * Used to return the result.
975 * @d_cdp: RDT domain that shares hardware with @d (RDT domain peer)
976 * Used to return the result.
978 * RDT resources are managed independently and by extension the RDT domains
979 * (RDT resource instances) are managed independently also. The Code and
980 * Data Prioritization (CDP) RDT resources, while managed independently,
981 * could refer to the same underlying hardware. For example,
982 * RDT_RESOURCE_L2CODE and RDT_RESOURCE_L2DATA both refer to the L2 cache.
984 * When provided with an RDT resource @r and an instance of that RDT
985 * resource @d rdt_cdp_peer_get() will return if there is a peer RDT
986 * resource and the exact instance that shares the same hardware.
988 * Return: 0 if a CDP peer was found, <0 on error or if no CDP peer exists.
989 * If a CDP peer was found, @r_cdp will point to the peer RDT resource
990 * and @d_cdp will point to the peer RDT domain.
992 static int rdt_cdp_peer_get(struct rdt_resource *r, struct rdt_domain *d,
993 struct rdt_resource **r_cdp,
994 struct rdt_domain **d_cdp)
996 struct rdt_resource *_r_cdp = NULL;
997 struct rdt_domain *_d_cdp = NULL;
1001 case RDT_RESOURCE_L3DATA:
1002 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3CODE];
1004 case RDT_RESOURCE_L3CODE:
1005 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3DATA];
1007 case RDT_RESOURCE_L2DATA:
1008 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2CODE];
1010 case RDT_RESOURCE_L2CODE:
1011 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2DATA];
1019 * When a new CPU comes online and CDP is enabled then the new
1020 * RDT domains (if any) associated with both CDP RDT resources
1021 * are added in the same CPU online routine while the
1022 * rdtgroup_mutex is held. It should thus not happen for one
1023 * RDT domain to exist and be associated with its RDT CDP
1024 * resource but there is no RDT domain associated with the
1025 * peer RDT CDP resource. Hence the WARN.
1027 _d_cdp = rdt_find_domain(_r_cdp, d->id, NULL);
1028 if (WARN_ON(IS_ERR_OR_NULL(_d_cdp))) {
1041 * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other
1042 * @r: Resource to which domain instance @d belongs.
1043 * @d: The domain instance for which @closid is being tested.
1044 * @cbm: Capacity bitmask being tested.
1045 * @closid: Intended closid for @cbm.
1046 * @exclusive: Only check if overlaps with exclusive resource groups
1048 * Checks if provided @cbm intended to be used for @closid on domain
1049 * @d overlaps with any other closids or other hardware usage associated
1050 * with this domain. If @exclusive is true then only overlaps with
1051 * resource groups in exclusive mode will be considered. If @exclusive
1052 * is false then overlaps with any resource group or hardware entities
1053 * will be considered.
1055 * @cbm is unsigned long, even if only 32 bits are used, to make the
1056 * bitmap functions work correctly.
1058 * Return: false if CBM does not overlap, true if it does.
1060 static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1061 unsigned long cbm, int closid, bool exclusive)
1063 enum rdtgrp_mode mode;
1064 unsigned long ctrl_b;
1068 /* Check for any overlap with regions used by hardware directly */
1070 ctrl_b = r->cache.shareable_bits;
1071 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len))
1075 /* Check for overlap with other resource groups */
1077 for (i = 0; i < closids_supported(); i++, ctrl++) {
1079 mode = rdtgroup_mode_by_closid(i);
1080 if (closid_allocated(i) && i != closid &&
1081 mode != RDT_MODE_PSEUDO_LOCKSETUP) {
1082 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) {
1084 if (mode == RDT_MODE_EXCLUSIVE)
1097 * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware
1098 * @r: Resource to which domain instance @d belongs.
1099 * @d: The domain instance for which @closid is being tested.
1100 * @cbm: Capacity bitmask being tested.
1101 * @closid: Intended closid for @cbm.
1102 * @exclusive: Only check if overlaps with exclusive resource groups
1104 * Resources that can be allocated using a CBM can use the CBM to control
1105 * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test
1106 * for overlap. Overlap test is not limited to the specific resource for
1107 * which the CBM is intended though - when dealing with CDP resources that
1108 * share the underlying hardware the overlap check should be performed on
1109 * the CDP resource sharing the hardware also.
1111 * Refer to description of __rdtgroup_cbm_overlaps() for the details of the
1114 * Return: true if CBM overlap detected, false if there is no overlap
1116 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1117 unsigned long cbm, int closid, bool exclusive)
1119 struct rdt_resource *r_cdp;
1120 struct rdt_domain *d_cdp;
1122 if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, exclusive))
1125 if (rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp) < 0)
1128 return __rdtgroup_cbm_overlaps(r_cdp, d_cdp, cbm, closid, exclusive);
1132 * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive
1134 * An exclusive resource group implies that there should be no sharing of
1135 * its allocated resources. At the time this group is considered to be
1136 * exclusive this test can determine if its current schemata supports this
1137 * setting by testing for overlap with all other resource groups.
1139 * Return: true if resource group can be exclusive, false if there is overlap
1140 * with allocations of other resource groups and thus this resource group
1141 * cannot be exclusive.
1143 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp)
1145 int closid = rdtgrp->closid;
1146 struct rdt_resource *r;
1147 bool has_cache = false;
1148 struct rdt_domain *d;
1150 for_each_alloc_enabled_rdt_resource(r) {
1151 if (r->rid == RDT_RESOURCE_MBA)
1154 list_for_each_entry(d, &r->domains, list) {
1155 if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid],
1156 rdtgrp->closid, false)) {
1157 rdt_last_cmd_puts("Schemata overlaps\n");
1164 rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n");
1172 * rdtgroup_mode_write - Modify the resource group's mode
1175 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of,
1176 char *buf, size_t nbytes, loff_t off)
1178 struct rdtgroup *rdtgrp;
1179 enum rdtgrp_mode mode;
1182 /* Valid input requires a trailing newline */
1183 if (nbytes == 0 || buf[nbytes - 1] != '\n')
1185 buf[nbytes - 1] = '\0';
1187 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1189 rdtgroup_kn_unlock(of->kn);
1193 rdt_last_cmd_clear();
1195 mode = rdtgrp->mode;
1197 if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) ||
1198 (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) ||
1199 (!strcmp(buf, "pseudo-locksetup") &&
1200 mode == RDT_MODE_PSEUDO_LOCKSETUP) ||
1201 (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED))
1204 if (mode == RDT_MODE_PSEUDO_LOCKED) {
1205 rdt_last_cmd_puts("Cannot change pseudo-locked group\n");
1210 if (!strcmp(buf, "shareable")) {
1211 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1212 ret = rdtgroup_locksetup_exit(rdtgrp);
1216 rdtgrp->mode = RDT_MODE_SHAREABLE;
1217 } else if (!strcmp(buf, "exclusive")) {
1218 if (!rdtgroup_mode_test_exclusive(rdtgrp)) {
1222 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1223 ret = rdtgroup_locksetup_exit(rdtgrp);
1227 rdtgrp->mode = RDT_MODE_EXCLUSIVE;
1228 } else if (!strcmp(buf, "pseudo-locksetup")) {
1229 ret = rdtgroup_locksetup_enter(rdtgrp);
1232 rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP;
1234 rdt_last_cmd_puts("Unknown or unsupported mode\n");
1239 rdtgroup_kn_unlock(of->kn);
1240 return ret ?: nbytes;
1244 * rdtgroup_cbm_to_size - Translate CBM to size in bytes
1245 * @r: RDT resource to which @d belongs.
1246 * @d: RDT domain instance.
1247 * @cbm: bitmask for which the size should be computed.
1249 * The bitmask provided associated with the RDT domain instance @d will be
1250 * translated into how many bytes it represents. The size in bytes is
1251 * computed by first dividing the total cache size by the CBM length to
1252 * determine how many bytes each bit in the bitmask represents. The result
1253 * is multiplied with the number of bits set in the bitmask.
1255 * @cbm is unsigned long, even if only 32 bits are used to make the
1256 * bitmap functions work correctly.
1258 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r,
1259 struct rdt_domain *d, unsigned long cbm)
1261 struct cpu_cacheinfo *ci;
1262 unsigned int size = 0;
1265 num_b = bitmap_weight(&cbm, r->cache.cbm_len);
1266 ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask));
1267 for (i = 0; i < ci->num_leaves; i++) {
1268 if (ci->info_list[i].level == r->cache_level) {
1269 size = ci->info_list[i].size / r->cache.cbm_len * num_b;
1278 * rdtgroup_size_show - Display size in bytes of allocated regions
1280 * The "size" file mirrors the layout of the "schemata" file, printing the
1281 * size in bytes of each region instead of the capacity bitmask.
1284 static int rdtgroup_size_show(struct kernfs_open_file *of,
1285 struct seq_file *s, void *v)
1287 struct rdtgroup *rdtgrp;
1288 struct rdt_resource *r;
1289 struct rdt_domain *d;
1295 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1297 rdtgroup_kn_unlock(of->kn);
1301 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
1302 if (!rdtgrp->plr->d) {
1303 rdt_last_cmd_clear();
1304 rdt_last_cmd_puts("Cache domain offline\n");
1307 seq_printf(s, "%*s:", max_name_width,
1308 rdtgrp->plr->r->name);
1309 size = rdtgroup_cbm_to_size(rdtgrp->plr->r,
1312 seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size);
1317 for_each_alloc_enabled_rdt_resource(r) {
1319 seq_printf(s, "%*s:", max_name_width, r->name);
1320 list_for_each_entry(d, &r->domains, list) {
1323 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1326 ctrl = (!is_mba_sc(r) ?
1327 d->ctrl_val[rdtgrp->closid] :
1328 d->mbps_val[rdtgrp->closid]);
1329 if (r->rid == RDT_RESOURCE_MBA)
1332 size = rdtgroup_cbm_to_size(r, d, ctrl);
1334 seq_printf(s, "%d=%u", d->id, size);
1341 rdtgroup_kn_unlock(of->kn);
1346 /* rdtgroup information files for one cache resource. */
1347 static struct rftype res_common_files[] = {
1349 .name = "last_cmd_status",
1351 .kf_ops = &rdtgroup_kf_single_ops,
1352 .seq_show = rdt_last_cmd_status_show,
1353 .fflags = RF_TOP_INFO,
1356 .name = "num_closids",
1358 .kf_ops = &rdtgroup_kf_single_ops,
1359 .seq_show = rdt_num_closids_show,
1360 .fflags = RF_CTRL_INFO,
1363 .name = "mon_features",
1365 .kf_ops = &rdtgroup_kf_single_ops,
1366 .seq_show = rdt_mon_features_show,
1367 .fflags = RF_MON_INFO,
1370 .name = "num_rmids",
1372 .kf_ops = &rdtgroup_kf_single_ops,
1373 .seq_show = rdt_num_rmids_show,
1374 .fflags = RF_MON_INFO,
1379 .kf_ops = &rdtgroup_kf_single_ops,
1380 .seq_show = rdt_default_ctrl_show,
1381 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1384 .name = "min_cbm_bits",
1386 .kf_ops = &rdtgroup_kf_single_ops,
1387 .seq_show = rdt_min_cbm_bits_show,
1388 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1391 .name = "shareable_bits",
1393 .kf_ops = &rdtgroup_kf_single_ops,
1394 .seq_show = rdt_shareable_bits_show,
1395 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1398 .name = "bit_usage",
1400 .kf_ops = &rdtgroup_kf_single_ops,
1401 .seq_show = rdt_bit_usage_show,
1402 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1405 .name = "min_bandwidth",
1407 .kf_ops = &rdtgroup_kf_single_ops,
1408 .seq_show = rdt_min_bw_show,
1409 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1412 .name = "bandwidth_gran",
1414 .kf_ops = &rdtgroup_kf_single_ops,
1415 .seq_show = rdt_bw_gran_show,
1416 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1419 .name = "delay_linear",
1421 .kf_ops = &rdtgroup_kf_single_ops,
1422 .seq_show = rdt_delay_linear_show,
1423 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1426 .name = "max_threshold_occupancy",
1428 .kf_ops = &rdtgroup_kf_single_ops,
1429 .write = max_threshold_occ_write,
1430 .seq_show = max_threshold_occ_show,
1431 .fflags = RF_MON_INFO | RFTYPE_RES_CACHE,
1436 .kf_ops = &rdtgroup_kf_single_ops,
1437 .write = rdtgroup_cpus_write,
1438 .seq_show = rdtgroup_cpus_show,
1439 .fflags = RFTYPE_BASE,
1442 .name = "cpus_list",
1444 .kf_ops = &rdtgroup_kf_single_ops,
1445 .write = rdtgroup_cpus_write,
1446 .seq_show = rdtgroup_cpus_show,
1447 .flags = RFTYPE_FLAGS_CPUS_LIST,
1448 .fflags = RFTYPE_BASE,
1453 .kf_ops = &rdtgroup_kf_single_ops,
1454 .write = rdtgroup_tasks_write,
1455 .seq_show = rdtgroup_tasks_show,
1456 .fflags = RFTYPE_BASE,
1461 .kf_ops = &rdtgroup_kf_single_ops,
1462 .write = rdtgroup_schemata_write,
1463 .seq_show = rdtgroup_schemata_show,
1464 .fflags = RF_CTRL_BASE,
1469 .kf_ops = &rdtgroup_kf_single_ops,
1470 .write = rdtgroup_mode_write,
1471 .seq_show = rdtgroup_mode_show,
1472 .fflags = RF_CTRL_BASE,
1477 .kf_ops = &rdtgroup_kf_single_ops,
1478 .seq_show = rdtgroup_size_show,
1479 .fflags = RF_CTRL_BASE,
1484 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
1486 struct rftype *rfts, *rft;
1489 rfts = res_common_files;
1490 len = ARRAY_SIZE(res_common_files);
1492 lockdep_assert_held(&rdtgroup_mutex);
1494 for (rft = rfts; rft < rfts + len; rft++) {
1495 if ((fflags & rft->fflags) == rft->fflags) {
1496 ret = rdtgroup_add_file(kn, rft);
1504 pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
1505 while (--rft >= rfts) {
1506 if ((fflags & rft->fflags) == rft->fflags)
1507 kernfs_remove_by_name(kn, rft->name);
1513 * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file
1514 * @r: The resource group with which the file is associated.
1515 * @name: Name of the file
1517 * The permissions of named resctrl file, directory, or link are modified
1518 * to not allow read, write, or execute by any user.
1520 * WARNING: This function is intended to communicate to the user that the
1521 * resctrl file has been locked down - that it is not relevant to the
1522 * particular state the system finds itself in. It should not be relied
1523 * on to protect from user access because after the file's permissions
1524 * are restricted the user can still change the permissions using chmod
1525 * from the command line.
1527 * Return: 0 on success, <0 on failure.
1529 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name)
1531 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1532 struct kernfs_node *kn;
1535 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1539 switch (kernfs_type(kn)) {
1541 iattr.ia_mode = S_IFDIR;
1544 iattr.ia_mode = S_IFREG;
1547 iattr.ia_mode = S_IFLNK;
1551 ret = kernfs_setattr(kn, &iattr);
1557 * rdtgroup_kn_mode_restore - Restore user access to named resctrl file
1558 * @r: The resource group with which the file is associated.
1559 * @name: Name of the file
1560 * @mask: Mask of permissions that should be restored
1562 * Restore the permissions of the named file. If @name is a directory the
1563 * permissions of its parent will be used.
1565 * Return: 0 on success, <0 on failure.
1567 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
1570 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1571 struct kernfs_node *kn, *parent;
1572 struct rftype *rfts, *rft;
1575 rfts = res_common_files;
1576 len = ARRAY_SIZE(res_common_files);
1578 for (rft = rfts; rft < rfts + len; rft++) {
1579 if (!strcmp(rft->name, name))
1580 iattr.ia_mode = rft->mode & mask;
1583 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1587 switch (kernfs_type(kn)) {
1589 parent = kernfs_get_parent(kn);
1591 iattr.ia_mode |= parent->mode;
1594 iattr.ia_mode |= S_IFDIR;
1597 iattr.ia_mode |= S_IFREG;
1600 iattr.ia_mode |= S_IFLNK;
1604 ret = kernfs_setattr(kn, &iattr);
1609 static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name,
1610 unsigned long fflags)
1612 struct kernfs_node *kn_subdir;
1615 kn_subdir = kernfs_create_dir(kn_info, name,
1617 if (IS_ERR(kn_subdir))
1618 return PTR_ERR(kn_subdir);
1620 kernfs_get(kn_subdir);
1621 ret = rdtgroup_kn_set_ugid(kn_subdir);
1625 ret = rdtgroup_add_files(kn_subdir, fflags);
1627 kernfs_activate(kn_subdir);
1632 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
1634 struct rdt_resource *r;
1635 unsigned long fflags;
1639 /* create the directory */
1640 kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
1641 if (IS_ERR(kn_info))
1642 return PTR_ERR(kn_info);
1643 kernfs_get(kn_info);
1645 ret = rdtgroup_add_files(kn_info, RF_TOP_INFO);
1649 for_each_alloc_enabled_rdt_resource(r) {
1650 fflags = r->fflags | RF_CTRL_INFO;
1651 ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags);
1656 for_each_mon_enabled_rdt_resource(r) {
1657 fflags = r->fflags | RF_MON_INFO;
1658 sprintf(name, "%s_MON", r->name);
1659 ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
1665 * This extra ref will be put in kernfs_remove() and guarantees
1666 * that @rdtgrp->kn is always accessible.
1668 kernfs_get(kn_info);
1670 ret = rdtgroup_kn_set_ugid(kn_info);
1674 kernfs_activate(kn_info);
1679 kernfs_remove(kn_info);
1684 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
1685 char *name, struct kernfs_node **dest_kn)
1687 struct kernfs_node *kn;
1690 /* create the directory */
1691 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1699 * This extra ref will be put in kernfs_remove() and guarantees
1700 * that @rdtgrp->kn is always accessible.
1704 ret = rdtgroup_kn_set_ugid(kn);
1708 kernfs_activate(kn);
1717 static void l3_qos_cfg_update(void *arg)
1721 wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL);
1724 static void l2_qos_cfg_update(void *arg)
1728 wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL);
1731 static inline bool is_mba_linear(void)
1733 return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear;
1736 static int set_cache_qos_cfg(int level, bool enable)
1738 void (*update)(void *arg);
1739 struct rdt_resource *r_l;
1740 cpumask_var_t cpu_mask;
1741 struct rdt_domain *d;
1744 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1747 if (level == RDT_RESOURCE_L3)
1748 update = l3_qos_cfg_update;
1749 else if (level == RDT_RESOURCE_L2)
1750 update = l2_qos_cfg_update;
1754 r_l = &rdt_resources_all[level];
1755 list_for_each_entry(d, &r_l->domains, list) {
1756 /* Pick one CPU from each domain instance to update MSR */
1757 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1760 /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1761 if (cpumask_test_cpu(cpu, cpu_mask))
1763 /* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1764 smp_call_function_many(cpu_mask, update, &enable, 1);
1767 free_cpumask_var(cpu_mask);
1773 * Enable or disable the MBA software controller
1774 * which helps user specify bandwidth in MBps.
1775 * MBA software controller is supported only if
1776 * MBM is supported and MBA is in linear scale.
1778 static int set_mba_sc(bool mba_sc)
1780 struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA];
1781 struct rdt_domain *d;
1783 if (!is_mbm_enabled() || !is_mba_linear() ||
1784 mba_sc == is_mba_sc(r))
1787 r->membw.mba_sc = mba_sc;
1788 list_for_each_entry(d, &r->domains, list)
1789 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val);
1794 static int cdp_enable(int level, int data_type, int code_type)
1796 struct rdt_resource *r_ldata = &rdt_resources_all[data_type];
1797 struct rdt_resource *r_lcode = &rdt_resources_all[code_type];
1798 struct rdt_resource *r_l = &rdt_resources_all[level];
1801 if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1802 !r_lcode->alloc_capable)
1805 ret = set_cache_qos_cfg(level, true);
1807 r_l->alloc_enabled = false;
1808 r_ldata->alloc_enabled = true;
1809 r_lcode->alloc_enabled = true;
1814 static int cdpl3_enable(void)
1816 return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1817 RDT_RESOURCE_L3CODE);
1820 static int cdpl2_enable(void)
1822 return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1823 RDT_RESOURCE_L2CODE);
1826 static void cdp_disable(int level, int data_type, int code_type)
1828 struct rdt_resource *r = &rdt_resources_all[level];
1830 r->alloc_enabled = r->alloc_capable;
1832 if (rdt_resources_all[data_type].alloc_enabled) {
1833 rdt_resources_all[data_type].alloc_enabled = false;
1834 rdt_resources_all[code_type].alloc_enabled = false;
1835 set_cache_qos_cfg(level, false);
1839 static void cdpl3_disable(void)
1841 cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
1844 static void cdpl2_disable(void)
1846 cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
1849 static void cdp_disable_all(void)
1851 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
1853 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
1858 * We don't allow rdtgroup directories to be created anywhere
1859 * except the root directory. Thus when looking for the rdtgroup
1860 * structure for a kernfs node we are either looking at a directory,
1861 * in which case the rdtgroup structure is pointed at by the "priv"
1862 * field, otherwise we have a file, and need only look to the parent
1863 * to find the rdtgroup.
1865 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
1867 if (kernfs_type(kn) == KERNFS_DIR) {
1869 * All the resource directories use "kn->priv"
1870 * to point to the "struct rdtgroup" for the
1871 * resource. "info" and its subdirectories don't
1872 * have rdtgroup structures, so return NULL here.
1874 if (kn == kn_info || kn->parent == kn_info)
1879 return kn->parent->priv;
1883 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
1885 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1890 atomic_inc(&rdtgrp->waitcount);
1891 kernfs_break_active_protection(kn);
1893 mutex_lock(&rdtgroup_mutex);
1895 /* Was this group deleted while we waited? */
1896 if (rdtgrp->flags & RDT_DELETED)
1902 void rdtgroup_kn_unlock(struct kernfs_node *kn)
1904 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1909 mutex_unlock(&rdtgroup_mutex);
1911 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
1912 (rdtgrp->flags & RDT_DELETED)) {
1913 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
1914 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
1915 rdtgroup_pseudo_lock_remove(rdtgrp);
1916 kernfs_unbreak_active_protection(kn);
1917 kernfs_put(rdtgrp->kn);
1920 kernfs_unbreak_active_protection(kn);
1924 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1925 struct rdtgroup *prgrp,
1926 struct kernfs_node **mon_data_kn);
1928 static int rdt_enable_ctx(struct rdt_fs_context *ctx)
1932 if (ctx->enable_cdpl2)
1933 ret = cdpl2_enable();
1935 if (!ret && ctx->enable_cdpl3)
1936 ret = cdpl3_enable();
1938 if (!ret && ctx->enable_mba_mbps)
1939 ret = set_mba_sc(true);
1944 static int rdt_get_tree(struct fs_context *fc)
1946 struct rdt_fs_context *ctx = rdt_fc2context(fc);
1947 struct rdt_domain *dom;
1948 struct rdt_resource *r;
1952 mutex_lock(&rdtgroup_mutex);
1954 * resctrl file system can only be mounted once.
1956 if (static_branch_unlikely(&rdt_enable_key)) {
1961 ret = rdt_enable_ctx(ctx);
1967 ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
1971 if (rdt_mon_capable) {
1972 ret = mongroup_create_dir(rdtgroup_default.kn,
1977 kernfs_get(kn_mongrp);
1979 ret = mkdir_mondata_all(rdtgroup_default.kn,
1980 &rdtgroup_default, &kn_mondata);
1983 kernfs_get(kn_mondata);
1984 rdtgroup_default.mon.mon_data_kn = kn_mondata;
1987 ret = rdt_pseudo_lock_init();
1991 ret = kernfs_get_tree(fc);
1995 if (rdt_alloc_capable)
1996 static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
1997 if (rdt_mon_capable)
1998 static_branch_enable_cpuslocked(&rdt_mon_enable_key);
2000 if (rdt_alloc_capable || rdt_mon_capable)
2001 static_branch_enable_cpuslocked(&rdt_enable_key);
2003 if (is_mbm_enabled()) {
2004 r = &rdt_resources_all[RDT_RESOURCE_L3];
2005 list_for_each_entry(dom, &r->domains, list)
2006 mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
2012 rdt_pseudo_lock_release();
2014 if (rdt_mon_capable)
2015 kernfs_remove(kn_mondata);
2017 if (rdt_mon_capable)
2018 kernfs_remove(kn_mongrp);
2020 kernfs_remove(kn_info);
2022 if (ctx->enable_mba_mbps)
2027 rdt_last_cmd_clear();
2028 mutex_unlock(&rdtgroup_mutex);
2040 static const struct fs_parameter_spec rdt_fs_parameters[] = {
2041 fsparam_flag("cdp", Opt_cdp),
2042 fsparam_flag("cdpl2", Opt_cdpl2),
2043 fsparam_flag("mba_MBps", Opt_mba_mbps),
2047 static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param)
2049 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2050 struct fs_parse_result result;
2053 opt = fs_parse(fc, rdt_fs_parameters, param, &result);
2059 ctx->enable_cdpl3 = true;
2062 ctx->enable_cdpl2 = true;
2065 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2067 ctx->enable_mba_mbps = true;
2074 static void rdt_fs_context_free(struct fs_context *fc)
2076 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2078 kernfs_free_fs_context(fc);
2082 static const struct fs_context_operations rdt_fs_context_ops = {
2083 .free = rdt_fs_context_free,
2084 .parse_param = rdt_parse_param,
2085 .get_tree = rdt_get_tree,
2088 static int rdt_init_fs_context(struct fs_context *fc)
2090 struct rdt_fs_context *ctx;
2092 ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL);
2096 ctx->kfc.root = rdt_root;
2097 ctx->kfc.magic = RDTGROUP_SUPER_MAGIC;
2098 fc->fs_private = &ctx->kfc;
2099 fc->ops = &rdt_fs_context_ops;
2100 put_user_ns(fc->user_ns);
2101 fc->user_ns = get_user_ns(&init_user_ns);
2106 static int reset_all_ctrls(struct rdt_resource *r)
2108 struct msr_param msr_param;
2109 cpumask_var_t cpu_mask;
2110 struct rdt_domain *d;
2113 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
2118 msr_param.high = r->num_closid;
2121 * Disable resource control for this resource by setting all
2122 * CBMs in all domains to the maximum mask value. Pick one CPU
2123 * from each domain to update the MSRs below.
2125 list_for_each_entry(d, &r->domains, list) {
2126 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
2128 for (i = 0; i < r->num_closid; i++)
2129 d->ctrl_val[i] = r->default_ctrl;
2132 /* Update CBM on this cpu if it's in cpu_mask. */
2133 if (cpumask_test_cpu(cpu, cpu_mask))
2134 rdt_ctrl_update(&msr_param);
2135 /* Update CBM on all other cpus in cpu_mask. */
2136 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
2139 free_cpumask_var(cpu_mask);
2144 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
2146 return (rdt_alloc_capable &&
2147 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
2150 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
2152 return (rdt_mon_capable &&
2153 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
2157 * Move tasks from one to the other group. If @from is NULL, then all tasks
2158 * in the systems are moved unconditionally (used for teardown).
2160 * If @mask is not NULL the cpus on which moved tasks are running are set
2161 * in that mask so the update smp function call is restricted to affected
2164 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
2165 struct cpumask *mask)
2167 struct task_struct *p, *t;
2169 read_lock(&tasklist_lock);
2170 for_each_process_thread(p, t) {
2171 if (!from || is_closid_match(t, from) ||
2172 is_rmid_match(t, from)) {
2173 t->closid = to->closid;
2174 t->rmid = to->mon.rmid;
2178 * This is safe on x86 w/o barriers as the ordering
2179 * of writing to task_cpu() and t->on_cpu is
2180 * reverse to the reading here. The detection is
2181 * inaccurate as tasks might move or schedule
2182 * before the smp function call takes place. In
2183 * such a case the function call is pointless, but
2184 * there is no other side effect.
2186 if (mask && t->on_cpu)
2187 cpumask_set_cpu(task_cpu(t), mask);
2191 read_unlock(&tasklist_lock);
2194 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
2196 struct rdtgroup *sentry, *stmp;
2197 struct list_head *head;
2199 head = &rdtgrp->mon.crdtgrp_list;
2200 list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
2201 free_rmid(sentry->mon.rmid);
2202 list_del(&sentry->mon.crdtgrp_list);
2208 * Forcibly remove all of subdirectories under root.
2210 static void rmdir_all_sub(void)
2212 struct rdtgroup *rdtgrp, *tmp;
2214 /* Move all tasks to the default resource group */
2215 rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
2217 list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
2218 /* Free any child rmids */
2219 free_all_child_rdtgrp(rdtgrp);
2221 /* Remove each rdtgroup other than root */
2222 if (rdtgrp == &rdtgroup_default)
2225 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2226 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2227 rdtgroup_pseudo_lock_remove(rdtgrp);
2230 * Give any CPUs back to the default group. We cannot copy
2231 * cpu_online_mask because a CPU might have executed the
2232 * offline callback already, but is still marked online.
2234 cpumask_or(&rdtgroup_default.cpu_mask,
2235 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2237 free_rmid(rdtgrp->mon.rmid);
2239 kernfs_remove(rdtgrp->kn);
2240 list_del(&rdtgrp->rdtgroup_list);
2243 /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
2244 update_closid_rmid(cpu_online_mask, &rdtgroup_default);
2246 kernfs_remove(kn_info);
2247 kernfs_remove(kn_mongrp);
2248 kernfs_remove(kn_mondata);
2251 static void rdt_kill_sb(struct super_block *sb)
2253 struct rdt_resource *r;
2256 mutex_lock(&rdtgroup_mutex);
2260 /*Put everything back to default values. */
2261 for_each_alloc_enabled_rdt_resource(r)
2265 rdt_pseudo_lock_release();
2266 rdtgroup_default.mode = RDT_MODE_SHAREABLE;
2267 static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
2268 static_branch_disable_cpuslocked(&rdt_mon_enable_key);
2269 static_branch_disable_cpuslocked(&rdt_enable_key);
2271 mutex_unlock(&rdtgroup_mutex);
2275 static struct file_system_type rdt_fs_type = {
2277 .init_fs_context = rdt_init_fs_context,
2278 .parameters = rdt_fs_parameters,
2279 .kill_sb = rdt_kill_sb,
2282 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
2285 struct kernfs_node *kn;
2288 kn = __kernfs_create_file(parent_kn, name, 0444,
2289 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
2290 &kf_mondata_ops, priv, NULL, NULL);
2294 ret = rdtgroup_kn_set_ugid(kn);
2304 * Remove all subdirectories of mon_data of ctrl_mon groups
2305 * and monitor groups with given domain id.
2307 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
2309 struct rdtgroup *prgrp, *crgrp;
2312 if (!r->mon_enabled)
2315 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2316 sprintf(name, "mon_%s_%02d", r->name, dom_id);
2317 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
2319 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
2320 kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
2324 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
2325 struct rdt_domain *d,
2326 struct rdt_resource *r, struct rdtgroup *prgrp)
2328 union mon_data_bits priv;
2329 struct kernfs_node *kn;
2330 struct mon_evt *mevt;
2331 struct rmid_read rr;
2335 sprintf(name, "mon_%s_%02d", r->name, d->id);
2336 /* create the directory */
2337 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
2342 * This extra ref will be put in kernfs_remove() and guarantees
2343 * that kn is always accessible.
2346 ret = rdtgroup_kn_set_ugid(kn);
2350 if (WARN_ON(list_empty(&r->evt_list))) {
2355 priv.u.rid = r->rid;
2356 priv.u.domid = d->id;
2357 list_for_each_entry(mevt, &r->evt_list, list) {
2358 priv.u.evtid = mevt->evtid;
2359 ret = mon_addfile(kn, mevt->name, priv.priv);
2363 if (is_mbm_event(mevt->evtid))
2364 mon_event_read(&rr, d, prgrp, mevt->evtid, true);
2366 kernfs_activate(kn);
2375 * Add all subdirectories of mon_data for "ctrl_mon" groups
2376 * and "monitor" groups with given domain id.
2378 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
2379 struct rdt_domain *d)
2381 struct kernfs_node *parent_kn;
2382 struct rdtgroup *prgrp, *crgrp;
2383 struct list_head *head;
2385 if (!r->mon_enabled)
2388 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2389 parent_kn = prgrp->mon.mon_data_kn;
2390 mkdir_mondata_subdir(parent_kn, d, r, prgrp);
2392 head = &prgrp->mon.crdtgrp_list;
2393 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
2394 parent_kn = crgrp->mon.mon_data_kn;
2395 mkdir_mondata_subdir(parent_kn, d, r, crgrp);
2400 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
2401 struct rdt_resource *r,
2402 struct rdtgroup *prgrp)
2404 struct rdt_domain *dom;
2407 list_for_each_entry(dom, &r->domains, list) {
2408 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
2417 * This creates a directory mon_data which contains the monitored data.
2419 * mon_data has one directory for each domain whic are named
2420 * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
2421 * with L3 domain looks as below:
2428 * Each domain directory has one file per event:
2433 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2434 struct rdtgroup *prgrp,
2435 struct kernfs_node **dest_kn)
2437 struct rdt_resource *r;
2438 struct kernfs_node *kn;
2442 * Create the mon_data directory first.
2444 ret = mongroup_create_dir(parent_kn, NULL, "mon_data", &kn);
2452 * Create the subdirectories for each domain. Note that all events
2453 * in a domain like L3 are grouped into a resource whose domain is L3
2455 for_each_mon_enabled_rdt_resource(r) {
2456 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
2469 * cbm_ensure_valid - Enforce validity on provided CBM
2470 * @_val: Candidate CBM
2471 * @r: RDT resource to which the CBM belongs
2473 * The provided CBM represents all cache portions available for use. This
2474 * may be represented by a bitmap that does not consist of contiguous ones
2475 * and thus be an invalid CBM.
2476 * Here the provided CBM is forced to be a valid CBM by only considering
2477 * the first set of contiguous bits as valid and clearing all bits.
2478 * The intention here is to provide a valid default CBM with which a new
2479 * resource group is initialized. The user can follow this with a
2480 * modification to the CBM if the default does not satisfy the
2483 static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r)
2485 unsigned int cbm_len = r->cache.cbm_len;
2486 unsigned long first_bit, zero_bit;
2487 unsigned long val = _val;
2492 first_bit = find_first_bit(&val, cbm_len);
2493 zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
2495 /* Clear any remaining bits to ensure contiguous region */
2496 bitmap_clear(&val, zero_bit, cbm_len - zero_bit);
2501 * Initialize cache resources per RDT domain
2503 * Set the RDT domain up to start off with all usable allocations. That is,
2504 * all shareable and unused bits. All-zero CBM is invalid.
2506 static int __init_one_rdt_domain(struct rdt_domain *d, struct rdt_resource *r,
2509 struct rdt_resource *r_cdp = NULL;
2510 struct rdt_domain *d_cdp = NULL;
2511 u32 used_b = 0, unused_b = 0;
2512 unsigned long tmp_cbm;
2513 enum rdtgrp_mode mode;
2514 u32 peer_ctl, *ctrl;
2517 rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp);
2518 d->have_new_ctrl = false;
2519 d->new_ctrl = r->cache.shareable_bits;
2520 used_b = r->cache.shareable_bits;
2522 for (i = 0; i < closids_supported(); i++, ctrl++) {
2523 if (closid_allocated(i) && i != closid) {
2524 mode = rdtgroup_mode_by_closid(i);
2525 if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
2527 * ctrl values for locksetup aren't relevant
2528 * until the schemata is written, and the mode
2529 * becomes RDT_MODE_PSEUDO_LOCKED.
2533 * If CDP is active include peer domain's
2534 * usage to ensure there is no overlap
2535 * with an exclusive group.
2538 peer_ctl = d_cdp->ctrl_val[i];
2541 used_b |= *ctrl | peer_ctl;
2542 if (mode == RDT_MODE_SHAREABLE)
2543 d->new_ctrl |= *ctrl | peer_ctl;
2546 if (d->plr && d->plr->cbm > 0)
2547 used_b |= d->plr->cbm;
2548 unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
2549 unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
2550 d->new_ctrl |= unused_b;
2552 * Force the initial CBM to be valid, user can
2553 * modify the CBM based on system availability.
2555 d->new_ctrl = cbm_ensure_valid(d->new_ctrl, r);
2557 * Assign the u32 CBM to an unsigned long to ensure that
2558 * bitmap_weight() does not access out-of-bound memory.
2560 tmp_cbm = d->new_ctrl;
2561 if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) {
2562 rdt_last_cmd_printf("No space on %s:%d\n", r->name, d->id);
2565 d->have_new_ctrl = true;
2571 * Initialize cache resources with default values.
2573 * A new RDT group is being created on an allocation capable (CAT)
2574 * supporting system. Set this group up to start off with all usable
2577 * If there are no more shareable bits available on any domain then
2578 * the entire allocation will fail.
2580 static int rdtgroup_init_cat(struct rdt_resource *r, u32 closid)
2582 struct rdt_domain *d;
2585 list_for_each_entry(d, &r->domains, list) {
2586 ret = __init_one_rdt_domain(d, r, closid);
2594 /* Initialize MBA resource with default values. */
2595 static void rdtgroup_init_mba(struct rdt_resource *r)
2597 struct rdt_domain *d;
2599 list_for_each_entry(d, &r->domains, list) {
2600 d->new_ctrl = is_mba_sc(r) ? MBA_MAX_MBPS : r->default_ctrl;
2601 d->have_new_ctrl = true;
2605 /* Initialize the RDT group's allocations. */
2606 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
2608 struct rdt_resource *r;
2611 for_each_alloc_enabled_rdt_resource(r) {
2612 if (r->rid == RDT_RESOURCE_MBA) {
2613 rdtgroup_init_mba(r);
2615 ret = rdtgroup_init_cat(r, rdtgrp->closid);
2620 ret = update_domains(r, rdtgrp->closid);
2622 rdt_last_cmd_puts("Failed to initialize allocations\n");
2628 rdtgrp->mode = RDT_MODE_SHAREABLE;
2633 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
2634 struct kernfs_node *prgrp_kn,
2635 const char *name, umode_t mode,
2636 enum rdt_group_type rtype, struct rdtgroup **r)
2638 struct rdtgroup *prdtgrp, *rdtgrp;
2639 struct kernfs_node *kn;
2643 prdtgrp = rdtgroup_kn_lock_live(prgrp_kn);
2649 if (rtype == RDTMON_GROUP &&
2650 (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2651 prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
2653 rdt_last_cmd_puts("Pseudo-locking in progress\n");
2657 /* allocate the rdtgroup. */
2658 rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
2661 rdt_last_cmd_puts("Kernel out of memory\n");
2665 rdtgrp->mon.parent = prdtgrp;
2666 rdtgrp->type = rtype;
2667 INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
2669 /* kernfs creates the directory for rdtgrp */
2670 kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
2673 rdt_last_cmd_puts("kernfs create error\n");
2679 * kernfs_remove() will drop the reference count on "kn" which
2680 * will free it. But we still need it to stick around for the
2681 * rdtgroup_kn_unlock(kn} call below. Take one extra reference
2682 * here, which will be dropped inside rdtgroup_kn_unlock().
2686 ret = rdtgroup_kn_set_ugid(kn);
2688 rdt_last_cmd_puts("kernfs perm error\n");
2692 files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
2693 ret = rdtgroup_add_files(kn, files);
2695 rdt_last_cmd_puts("kernfs fill error\n");
2699 if (rdt_mon_capable) {
2702 rdt_last_cmd_puts("Out of RMIDs\n");
2705 rdtgrp->mon.rmid = ret;
2707 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
2709 rdt_last_cmd_puts("kernfs subdir error\n");
2713 kernfs_activate(kn);
2716 * The caller unlocks the prgrp_kn upon success.
2721 free_rmid(rdtgrp->mon.rmid);
2723 kernfs_remove(rdtgrp->kn);
2727 rdtgroup_kn_unlock(prgrp_kn);
2731 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
2733 kernfs_remove(rgrp->kn);
2734 free_rmid(rgrp->mon.rmid);
2739 * Create a monitor group under "mon_groups" directory of a control
2740 * and monitor group(ctrl_mon). This is a resource group
2741 * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
2743 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
2744 struct kernfs_node *prgrp_kn,
2748 struct rdtgroup *rdtgrp, *prgrp;
2751 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP,
2756 prgrp = rdtgrp->mon.parent;
2757 rdtgrp->closid = prgrp->closid;
2760 * Add the rdtgrp to the list of rdtgrps the parent
2761 * ctrl_mon group has to track.
2763 list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
2765 rdtgroup_kn_unlock(prgrp_kn);
2770 * These are rdtgroups created under the root directory. Can be used
2771 * to allocate and monitor resources.
2773 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
2774 struct kernfs_node *prgrp_kn,
2775 const char *name, umode_t mode)
2777 struct rdtgroup *rdtgrp;
2778 struct kernfs_node *kn;
2782 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP,
2788 ret = closid_alloc();
2790 rdt_last_cmd_puts("Out of CLOSIDs\n");
2791 goto out_common_fail;
2796 rdtgrp->closid = closid;
2797 ret = rdtgroup_init_alloc(rdtgrp);
2801 list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
2803 if (rdt_mon_capable) {
2805 * Create an empty mon_groups directory to hold the subset
2806 * of tasks and cpus to monitor.
2808 ret = mongroup_create_dir(kn, NULL, "mon_groups", NULL);
2810 rdt_last_cmd_puts("kernfs subdir error\n");
2818 list_del(&rdtgrp->rdtgroup_list);
2820 closid_free(closid);
2822 mkdir_rdt_prepare_clean(rdtgrp);
2824 rdtgroup_kn_unlock(prgrp_kn);
2829 * We allow creating mon groups only with in a directory called "mon_groups"
2830 * which is present in every ctrl_mon group. Check if this is a valid
2831 * "mon_groups" directory.
2833 * 1. The directory should be named "mon_groups".
2834 * 2. The mon group itself should "not" be named "mon_groups".
2835 * This makes sure "mon_groups" directory always has a ctrl_mon group
2838 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
2840 return (!strcmp(kn->name, "mon_groups") &&
2841 strcmp(name, "mon_groups"));
2844 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
2847 /* Do not accept '\n' to avoid unparsable situation. */
2848 if (strchr(name, '\n'))
2852 * If the parent directory is the root directory and RDT
2853 * allocation is supported, add a control and monitoring
2856 if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
2857 return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode);
2860 * If RDT monitoring is supported and the parent directory is a valid
2861 * "mon_groups" directory, add a monitoring subdirectory.
2863 if (rdt_mon_capable && is_mon_groups(parent_kn, name))
2864 return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode);
2869 static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2870 cpumask_var_t tmpmask)
2872 struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
2875 /* Give any tasks back to the parent group */
2876 rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
2878 /* Update per cpu rmid of the moved CPUs first */
2879 for_each_cpu(cpu, &rdtgrp->cpu_mask)
2880 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
2882 * Update the MSR on moved CPUs and CPUs which have moved
2883 * task running on them.
2885 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2886 update_closid_rmid(tmpmask, NULL);
2888 rdtgrp->flags = RDT_DELETED;
2889 free_rmid(rdtgrp->mon.rmid);
2892 * Remove the rdtgrp from the parent ctrl_mon group's list
2894 WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
2895 list_del(&rdtgrp->mon.crdtgrp_list);
2898 * one extra hold on this, will drop when we kfree(rdtgrp)
2899 * in rdtgroup_kn_unlock()
2902 kernfs_remove(rdtgrp->kn);
2907 static int rdtgroup_ctrl_remove(struct kernfs_node *kn,
2908 struct rdtgroup *rdtgrp)
2910 rdtgrp->flags = RDT_DELETED;
2911 list_del(&rdtgrp->rdtgroup_list);
2914 * one extra hold on this, will drop when we kfree(rdtgrp)
2915 * in rdtgroup_kn_unlock()
2918 kernfs_remove(rdtgrp->kn);
2922 static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2923 cpumask_var_t tmpmask)
2927 /* Give any tasks back to the default group */
2928 rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
2930 /* Give any CPUs back to the default group */
2931 cpumask_or(&rdtgroup_default.cpu_mask,
2932 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2934 /* Update per cpu closid and rmid of the moved CPUs first */
2935 for_each_cpu(cpu, &rdtgrp->cpu_mask) {
2936 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
2937 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
2941 * Update the MSR on moved CPUs and CPUs which have moved
2942 * task running on them.
2944 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2945 update_closid_rmid(tmpmask, NULL);
2947 closid_free(rdtgrp->closid);
2948 free_rmid(rdtgrp->mon.rmid);
2951 * Free all the child monitor group rmids.
2953 free_all_child_rdtgrp(rdtgrp);
2955 rdtgroup_ctrl_remove(kn, rdtgrp);
2960 static int rdtgroup_rmdir(struct kernfs_node *kn)
2962 struct kernfs_node *parent_kn = kn->parent;
2963 struct rdtgroup *rdtgrp;
2964 cpumask_var_t tmpmask;
2967 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
2970 rdtgrp = rdtgroup_kn_lock_live(kn);
2977 * If the rdtgroup is a ctrl_mon group and parent directory
2978 * is the root directory, remove the ctrl_mon group.
2980 * If the rdtgroup is a mon group and parent directory
2981 * is a valid "mon_groups" directory, remove the mon group.
2983 if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn) {
2984 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2985 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
2986 ret = rdtgroup_ctrl_remove(kn, rdtgrp);
2988 ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask);
2990 } else if (rdtgrp->type == RDTMON_GROUP &&
2991 is_mon_groups(parent_kn, kn->name)) {
2992 ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask);
2998 rdtgroup_kn_unlock(kn);
2999 free_cpumask_var(tmpmask);
3003 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
3005 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
3006 seq_puts(seq, ",cdp");
3008 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
3009 seq_puts(seq, ",cdpl2");
3011 if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA]))
3012 seq_puts(seq, ",mba_MBps");
3017 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
3018 .mkdir = rdtgroup_mkdir,
3019 .rmdir = rdtgroup_rmdir,
3020 .show_options = rdtgroup_show_options,
3023 static int __init rdtgroup_setup_root(void)
3027 rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
3028 KERNFS_ROOT_CREATE_DEACTIVATED |
3029 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
3031 if (IS_ERR(rdt_root))
3032 return PTR_ERR(rdt_root);
3034 mutex_lock(&rdtgroup_mutex);
3036 rdtgroup_default.closid = 0;
3037 rdtgroup_default.mon.rmid = 0;
3038 rdtgroup_default.type = RDTCTRL_GROUP;
3039 INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
3041 list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
3043 ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
3045 kernfs_destroy_root(rdt_root);
3049 rdtgroup_default.kn = rdt_root->kn;
3050 kernfs_activate(rdtgroup_default.kn);
3053 mutex_unlock(&rdtgroup_mutex);
3059 * rdtgroup_init - rdtgroup initialization
3061 * Setup resctrl file system including set up root, create mount point,
3062 * register rdtgroup filesystem, and initialize files under root directory.
3064 * Return: 0 on success or -errno
3066 int __init rdtgroup_init(void)
3070 seq_buf_init(&last_cmd_status, last_cmd_status_buf,
3071 sizeof(last_cmd_status_buf));
3073 ret = rdtgroup_setup_root();
3077 ret = sysfs_create_mount_point(fs_kobj, "resctrl");
3081 ret = register_filesystem(&rdt_fs_type);
3083 goto cleanup_mountpoint;
3086 * Adding the resctrl debugfs directory here may not be ideal since
3087 * it would let the resctrl debugfs directory appear on the debugfs
3088 * filesystem before the resctrl filesystem is mounted.
3089 * It may also be ok since that would enable debugging of RDT before
3090 * resctrl is mounted.
3091 * The reason why the debugfs directory is created here and not in
3092 * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and
3093 * during the debugfs directory creation also &sb->s_type->i_mutex_key
3094 * (the lockdep class of inode->i_rwsem). Other filesystem
3095 * interactions (eg. SyS_getdents) have the lock ordering:
3096 * &sb->s_type->i_mutex_key --> &mm->mmap_sem
3097 * During mmap(), called with &mm->mmap_sem, the rdtgroup_mutex
3098 * is taken, thus creating dependency:
3099 * &mm->mmap_sem --> rdtgroup_mutex for the latter that can cause
3100 * issues considering the other two lock dependencies.
3101 * By creating the debugfs directory here we avoid a dependency
3102 * that may cause deadlock (even though file operations cannot
3103 * occur until the filesystem is mounted, but I do not know how to
3104 * tell lockdep that).
3106 debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
3111 sysfs_remove_mount_point(fs_kobj, "resctrl");
3113 kernfs_destroy_root(rdt_root);
3118 void __exit rdtgroup_exit(void)
3120 debugfs_remove_recursive(debugfs_resctrl);
3121 unregister_filesystem(&rdt_fs_type);
3122 sysfs_remove_mount_point(fs_kobj, "resctrl");
3123 kernfs_destroy_root(rdt_root);