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);
464 rdt_last_cmd_clear();
467 rdt_last_cmd_puts("Directory was removed\n");
471 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
472 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
474 rdt_last_cmd_puts("Pseudo-locking in progress\n");
479 ret = cpulist_parse(buf, newmask);
481 ret = cpumask_parse(buf, newmask);
484 rdt_last_cmd_puts("Bad CPU list/mask\n");
488 /* check that user didn't specify any offline cpus */
489 cpumask_andnot(tmpmask, newmask, cpu_online_mask);
490 if (cpumask_weight(tmpmask)) {
492 rdt_last_cmd_puts("Can only assign online CPUs\n");
496 if (rdtgrp->type == RDTCTRL_GROUP)
497 ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1);
498 else if (rdtgrp->type == RDTMON_GROUP)
499 ret = cpus_mon_write(rdtgrp, newmask, tmpmask);
504 rdtgroup_kn_unlock(of->kn);
505 free_cpumask_var(tmpmask);
506 free_cpumask_var(newmask);
507 free_cpumask_var(tmpmask1);
509 return ret ?: nbytes;
512 struct task_move_callback {
513 struct callback_head work;
514 struct rdtgroup *rdtgrp;
517 static void move_myself(struct callback_head *head)
519 struct task_move_callback *callback;
520 struct rdtgroup *rdtgrp;
522 callback = container_of(head, struct task_move_callback, work);
523 rdtgrp = callback->rdtgrp;
526 * If resource group was deleted before this task work callback
527 * was invoked, then assign the task to root group and free the
530 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
531 (rdtgrp->flags & RDT_DELETED)) {
538 /* update PQR_ASSOC MSR to make resource group go into effect */
545 static int __rdtgroup_move_task(struct task_struct *tsk,
546 struct rdtgroup *rdtgrp)
548 struct task_move_callback *callback;
551 callback = kzalloc(sizeof(*callback), GFP_KERNEL);
554 callback->work.func = move_myself;
555 callback->rdtgrp = rdtgrp;
558 * Take a refcount, so rdtgrp cannot be freed before the
559 * callback has been invoked.
561 atomic_inc(&rdtgrp->waitcount);
562 ret = task_work_add(tsk, &callback->work, true);
565 * Task is exiting. Drop the refcount and free the callback.
566 * No need to check the refcount as the group cannot be
567 * deleted before the write function unlocks rdtgroup_mutex.
569 atomic_dec(&rdtgrp->waitcount);
571 rdt_last_cmd_puts("Task exited\n");
574 * For ctrl_mon groups move both closid and rmid.
575 * For monitor groups, can move the tasks only from
576 * their parent CTRL group.
578 if (rdtgrp->type == RDTCTRL_GROUP) {
579 tsk->closid = rdtgrp->closid;
580 tsk->rmid = rdtgrp->mon.rmid;
581 } else if (rdtgrp->type == RDTMON_GROUP) {
582 if (rdtgrp->mon.parent->closid == tsk->closid) {
583 tsk->rmid = rdtgrp->mon.rmid;
585 rdt_last_cmd_puts("Can't move task to different control group\n");
594 * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group
597 * Return: 1 if tasks have been assigned to @r, 0 otherwise
599 int rdtgroup_tasks_assigned(struct rdtgroup *r)
601 struct task_struct *p, *t;
604 lockdep_assert_held(&rdtgroup_mutex);
607 for_each_process_thread(p, t) {
608 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
609 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) {
619 static int rdtgroup_task_write_permission(struct task_struct *task,
620 struct kernfs_open_file *of)
622 const struct cred *tcred = get_task_cred(task);
623 const struct cred *cred = current_cred();
627 * Even if we're attaching all tasks in the thread group, we only
628 * need to check permissions on one of them.
630 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
631 !uid_eq(cred->euid, tcred->uid) &&
632 !uid_eq(cred->euid, tcred->suid)) {
633 rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
641 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
642 struct kernfs_open_file *of)
644 struct task_struct *tsk;
649 tsk = find_task_by_vpid(pid);
652 rdt_last_cmd_printf("No task %d\n", pid);
659 get_task_struct(tsk);
662 ret = rdtgroup_task_write_permission(tsk, of);
664 ret = __rdtgroup_move_task(tsk, rdtgrp);
666 put_task_struct(tsk);
670 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
671 char *buf, size_t nbytes, loff_t off)
673 struct rdtgroup *rdtgrp;
677 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
679 rdtgrp = rdtgroup_kn_lock_live(of->kn);
681 rdtgroup_kn_unlock(of->kn);
684 rdt_last_cmd_clear();
686 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
687 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
689 rdt_last_cmd_puts("Pseudo-locking in progress\n");
693 ret = rdtgroup_move_task(pid, rdtgrp, of);
696 rdtgroup_kn_unlock(of->kn);
698 return ret ?: nbytes;
701 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
703 struct task_struct *p, *t;
706 for_each_process_thread(p, t) {
707 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
708 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid))
709 seq_printf(s, "%d\n", t->pid);
714 static int rdtgroup_tasks_show(struct kernfs_open_file *of,
715 struct seq_file *s, void *v)
717 struct rdtgroup *rdtgrp;
720 rdtgrp = rdtgroup_kn_lock_live(of->kn);
722 show_rdt_tasks(rdtgrp, s);
725 rdtgroup_kn_unlock(of->kn);
730 static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
731 struct seq_file *seq, void *v)
735 mutex_lock(&rdtgroup_mutex);
736 len = seq_buf_used(&last_cmd_status);
738 seq_printf(seq, "%.*s", len, last_cmd_status_buf);
740 seq_puts(seq, "ok\n");
741 mutex_unlock(&rdtgroup_mutex);
745 static int rdt_num_closids_show(struct kernfs_open_file *of,
746 struct seq_file *seq, void *v)
748 struct rdt_resource *r = of->kn->parent->priv;
750 seq_printf(seq, "%d\n", r->num_closid);
754 static int rdt_default_ctrl_show(struct kernfs_open_file *of,
755 struct seq_file *seq, void *v)
757 struct rdt_resource *r = of->kn->parent->priv;
759 seq_printf(seq, "%x\n", r->default_ctrl);
763 static int rdt_min_cbm_bits_show(struct kernfs_open_file *of,
764 struct seq_file *seq, void *v)
766 struct rdt_resource *r = of->kn->parent->priv;
768 seq_printf(seq, "%u\n", r->cache.min_cbm_bits);
772 static int rdt_shareable_bits_show(struct kernfs_open_file *of,
773 struct seq_file *seq, void *v)
775 struct rdt_resource *r = of->kn->parent->priv;
777 seq_printf(seq, "%x\n", r->cache.shareable_bits);
782 * rdt_bit_usage_show - Display current usage of resources
784 * A domain is a shared resource that can now be allocated differently. Here
785 * we display the current regions of the domain as an annotated bitmask.
786 * For each domain of this resource its allocation bitmask
787 * is annotated as below to indicate the current usage of the corresponding bit:
788 * 0 - currently unused
789 * X - currently available for sharing and used by software and hardware
790 * H - currently used by hardware only but available for software use
791 * S - currently used and shareable by software only
792 * E - currently used exclusively by one resource group
793 * P - currently pseudo-locked by one resource group
795 static int rdt_bit_usage_show(struct kernfs_open_file *of,
796 struct seq_file *seq, void *v)
798 struct rdt_resource *r = of->kn->parent->priv;
800 * Use unsigned long even though only 32 bits are used to ensure
801 * test_bit() is used safely.
803 unsigned long sw_shareable = 0, hw_shareable = 0;
804 unsigned long exclusive = 0, pseudo_locked = 0;
805 struct rdt_domain *dom;
806 int i, hwb, swb, excl, psl;
807 enum rdtgrp_mode mode;
811 mutex_lock(&rdtgroup_mutex);
812 hw_shareable = r->cache.shareable_bits;
813 list_for_each_entry(dom, &r->domains, list) {
816 ctrl = dom->ctrl_val;
819 seq_printf(seq, "%d=", dom->id);
820 for (i = 0; i < closids_supported(); i++, ctrl++) {
821 if (!closid_allocated(i))
823 mode = rdtgroup_mode_by_closid(i);
825 case RDT_MODE_SHAREABLE:
826 sw_shareable |= *ctrl;
828 case RDT_MODE_EXCLUSIVE:
831 case RDT_MODE_PSEUDO_LOCKSETUP:
833 * RDT_MODE_PSEUDO_LOCKSETUP is possible
834 * here but not included since the CBM
835 * associated with this CLOSID in this mode
836 * is not initialized and no task or cpu can be
837 * assigned this CLOSID.
840 case RDT_MODE_PSEUDO_LOCKED:
843 "invalid mode for closid %d\n", i);
847 for (i = r->cache.cbm_len - 1; i >= 0; i--) {
848 pseudo_locked = dom->plr ? dom->plr->cbm : 0;
849 hwb = test_bit(i, &hw_shareable);
850 swb = test_bit(i, &sw_shareable);
851 excl = test_bit(i, &exclusive);
852 psl = test_bit(i, &pseudo_locked);
855 else if (hwb && !swb)
857 else if (!hwb && swb)
863 else /* Unused bits remain */
869 mutex_unlock(&rdtgroup_mutex);
873 static int rdt_min_bw_show(struct kernfs_open_file *of,
874 struct seq_file *seq, void *v)
876 struct rdt_resource *r = of->kn->parent->priv;
878 seq_printf(seq, "%u\n", r->membw.min_bw);
882 static int rdt_num_rmids_show(struct kernfs_open_file *of,
883 struct seq_file *seq, void *v)
885 struct rdt_resource *r = of->kn->parent->priv;
887 seq_printf(seq, "%d\n", r->num_rmid);
892 static int rdt_mon_features_show(struct kernfs_open_file *of,
893 struct seq_file *seq, void *v)
895 struct rdt_resource *r = of->kn->parent->priv;
896 struct mon_evt *mevt;
898 list_for_each_entry(mevt, &r->evt_list, list)
899 seq_printf(seq, "%s\n", mevt->name);
904 static int rdt_bw_gran_show(struct kernfs_open_file *of,
905 struct seq_file *seq, void *v)
907 struct rdt_resource *r = of->kn->parent->priv;
909 seq_printf(seq, "%u\n", r->membw.bw_gran);
913 static int rdt_delay_linear_show(struct kernfs_open_file *of,
914 struct seq_file *seq, void *v)
916 struct rdt_resource *r = of->kn->parent->priv;
918 seq_printf(seq, "%u\n", r->membw.delay_linear);
922 static int max_threshold_occ_show(struct kernfs_open_file *of,
923 struct seq_file *seq, void *v)
925 struct rdt_resource *r = of->kn->parent->priv;
927 seq_printf(seq, "%u\n", resctrl_cqm_threshold * r->mon_scale);
932 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
933 char *buf, size_t nbytes, loff_t off)
935 struct rdt_resource *r = of->kn->parent->priv;
939 ret = kstrtouint(buf, 0, &bytes);
943 if (bytes > (boot_cpu_data.x86_cache_size * 1024))
946 resctrl_cqm_threshold = bytes / r->mon_scale;
952 * rdtgroup_mode_show - Display mode of this resource group
954 static int rdtgroup_mode_show(struct kernfs_open_file *of,
955 struct seq_file *s, void *v)
957 struct rdtgroup *rdtgrp;
959 rdtgrp = rdtgroup_kn_lock_live(of->kn);
961 rdtgroup_kn_unlock(of->kn);
965 seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode));
967 rdtgroup_kn_unlock(of->kn);
972 * rdt_cdp_peer_get - Retrieve CDP peer if it exists
973 * @r: RDT resource to which RDT domain @d belongs
974 * @d: Cache instance for which a CDP peer is requested
975 * @r_cdp: RDT resource that shares hardware with @r (RDT resource peer)
976 * Used to return the result.
977 * @d_cdp: RDT domain that shares hardware with @d (RDT domain peer)
978 * Used to return the result.
980 * RDT resources are managed independently and by extension the RDT domains
981 * (RDT resource instances) are managed independently also. The Code and
982 * Data Prioritization (CDP) RDT resources, while managed independently,
983 * could refer to the same underlying hardware. For example,
984 * RDT_RESOURCE_L2CODE and RDT_RESOURCE_L2DATA both refer to the L2 cache.
986 * When provided with an RDT resource @r and an instance of that RDT
987 * resource @d rdt_cdp_peer_get() will return if there is a peer RDT
988 * resource and the exact instance that shares the same hardware.
990 * Return: 0 if a CDP peer was found, <0 on error or if no CDP peer exists.
991 * If a CDP peer was found, @r_cdp will point to the peer RDT resource
992 * and @d_cdp will point to the peer RDT domain.
994 static int rdt_cdp_peer_get(struct rdt_resource *r, struct rdt_domain *d,
995 struct rdt_resource **r_cdp,
996 struct rdt_domain **d_cdp)
998 struct rdt_resource *_r_cdp = NULL;
999 struct rdt_domain *_d_cdp = NULL;
1003 case RDT_RESOURCE_L3DATA:
1004 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3CODE];
1006 case RDT_RESOURCE_L3CODE:
1007 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3DATA];
1009 case RDT_RESOURCE_L2DATA:
1010 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2CODE];
1012 case RDT_RESOURCE_L2CODE:
1013 _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2DATA];
1021 * When a new CPU comes online and CDP is enabled then the new
1022 * RDT domains (if any) associated with both CDP RDT resources
1023 * are added in the same CPU online routine while the
1024 * rdtgroup_mutex is held. It should thus not happen for one
1025 * RDT domain to exist and be associated with its RDT CDP
1026 * resource but there is no RDT domain associated with the
1027 * peer RDT CDP resource. Hence the WARN.
1029 _d_cdp = rdt_find_domain(_r_cdp, d->id, NULL);
1030 if (WARN_ON(IS_ERR_OR_NULL(_d_cdp))) {
1043 * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other
1044 * @r: Resource to which domain instance @d belongs.
1045 * @d: The domain instance for which @closid is being tested.
1046 * @cbm: Capacity bitmask being tested.
1047 * @closid: Intended closid for @cbm.
1048 * @exclusive: Only check if overlaps with exclusive resource groups
1050 * Checks if provided @cbm intended to be used for @closid on domain
1051 * @d overlaps with any other closids or other hardware usage associated
1052 * with this domain. If @exclusive is true then only overlaps with
1053 * resource groups in exclusive mode will be considered. If @exclusive
1054 * is false then overlaps with any resource group or hardware entities
1055 * will be considered.
1057 * @cbm is unsigned long, even if only 32 bits are used, to make the
1058 * bitmap functions work correctly.
1060 * Return: false if CBM does not overlap, true if it does.
1062 static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1063 unsigned long cbm, int closid, bool exclusive)
1065 enum rdtgrp_mode mode;
1066 unsigned long ctrl_b;
1070 /* Check for any overlap with regions used by hardware directly */
1072 ctrl_b = r->cache.shareable_bits;
1073 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len))
1077 /* Check for overlap with other resource groups */
1079 for (i = 0; i < closids_supported(); i++, ctrl++) {
1081 mode = rdtgroup_mode_by_closid(i);
1082 if (closid_allocated(i) && i != closid &&
1083 mode != RDT_MODE_PSEUDO_LOCKSETUP) {
1084 if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) {
1086 if (mode == RDT_MODE_EXCLUSIVE)
1099 * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware
1100 * @r: Resource to which domain instance @d belongs.
1101 * @d: The domain instance for which @closid is being tested.
1102 * @cbm: Capacity bitmask being tested.
1103 * @closid: Intended closid for @cbm.
1104 * @exclusive: Only check if overlaps with exclusive resource groups
1106 * Resources that can be allocated using a CBM can use the CBM to control
1107 * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test
1108 * for overlap. Overlap test is not limited to the specific resource for
1109 * which the CBM is intended though - when dealing with CDP resources that
1110 * share the underlying hardware the overlap check should be performed on
1111 * the CDP resource sharing the hardware also.
1113 * Refer to description of __rdtgroup_cbm_overlaps() for the details of the
1116 * Return: true if CBM overlap detected, false if there is no overlap
1118 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1119 unsigned long cbm, int closid, bool exclusive)
1121 struct rdt_resource *r_cdp;
1122 struct rdt_domain *d_cdp;
1124 if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, exclusive))
1127 if (rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp) < 0)
1130 return __rdtgroup_cbm_overlaps(r_cdp, d_cdp, cbm, closid, exclusive);
1134 * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive
1136 * An exclusive resource group implies that there should be no sharing of
1137 * its allocated resources. At the time this group is considered to be
1138 * exclusive this test can determine if its current schemata supports this
1139 * setting by testing for overlap with all other resource groups.
1141 * Return: true if resource group can be exclusive, false if there is overlap
1142 * with allocations of other resource groups and thus this resource group
1143 * cannot be exclusive.
1145 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp)
1147 int closid = rdtgrp->closid;
1148 struct rdt_resource *r;
1149 bool has_cache = false;
1150 struct rdt_domain *d;
1152 for_each_alloc_enabled_rdt_resource(r) {
1153 if (r->rid == RDT_RESOURCE_MBA)
1156 list_for_each_entry(d, &r->domains, list) {
1157 if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid],
1158 rdtgrp->closid, false)) {
1159 rdt_last_cmd_puts("Schemata overlaps\n");
1166 rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n");
1174 * rdtgroup_mode_write - Modify the resource group's mode
1177 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of,
1178 char *buf, size_t nbytes, loff_t off)
1180 struct rdtgroup *rdtgrp;
1181 enum rdtgrp_mode mode;
1184 /* Valid input requires a trailing newline */
1185 if (nbytes == 0 || buf[nbytes - 1] != '\n')
1187 buf[nbytes - 1] = '\0';
1189 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1191 rdtgroup_kn_unlock(of->kn);
1195 rdt_last_cmd_clear();
1197 mode = rdtgrp->mode;
1199 if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) ||
1200 (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) ||
1201 (!strcmp(buf, "pseudo-locksetup") &&
1202 mode == RDT_MODE_PSEUDO_LOCKSETUP) ||
1203 (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED))
1206 if (mode == RDT_MODE_PSEUDO_LOCKED) {
1207 rdt_last_cmd_puts("Cannot change pseudo-locked group\n");
1212 if (!strcmp(buf, "shareable")) {
1213 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1214 ret = rdtgroup_locksetup_exit(rdtgrp);
1218 rdtgrp->mode = RDT_MODE_SHAREABLE;
1219 } else if (!strcmp(buf, "exclusive")) {
1220 if (!rdtgroup_mode_test_exclusive(rdtgrp)) {
1224 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1225 ret = rdtgroup_locksetup_exit(rdtgrp);
1229 rdtgrp->mode = RDT_MODE_EXCLUSIVE;
1230 } else if (!strcmp(buf, "pseudo-locksetup")) {
1231 ret = rdtgroup_locksetup_enter(rdtgrp);
1234 rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP;
1236 rdt_last_cmd_puts("Unknown or unsupported mode\n");
1241 rdtgroup_kn_unlock(of->kn);
1242 return ret ?: nbytes;
1246 * rdtgroup_cbm_to_size - Translate CBM to size in bytes
1247 * @r: RDT resource to which @d belongs.
1248 * @d: RDT domain instance.
1249 * @cbm: bitmask for which the size should be computed.
1251 * The bitmask provided associated with the RDT domain instance @d will be
1252 * translated into how many bytes it represents. The size in bytes is
1253 * computed by first dividing the total cache size by the CBM length to
1254 * determine how many bytes each bit in the bitmask represents. The result
1255 * is multiplied with the number of bits set in the bitmask.
1257 * @cbm is unsigned long, even if only 32 bits are used to make the
1258 * bitmap functions work correctly.
1260 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r,
1261 struct rdt_domain *d, unsigned long cbm)
1263 struct cpu_cacheinfo *ci;
1264 unsigned int size = 0;
1267 num_b = bitmap_weight(&cbm, r->cache.cbm_len);
1268 ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask));
1269 for (i = 0; i < ci->num_leaves; i++) {
1270 if (ci->info_list[i].level == r->cache_level) {
1271 size = ci->info_list[i].size / r->cache.cbm_len * num_b;
1280 * rdtgroup_size_show - Display size in bytes of allocated regions
1282 * The "size" file mirrors the layout of the "schemata" file, printing the
1283 * size in bytes of each region instead of the capacity bitmask.
1286 static int rdtgroup_size_show(struct kernfs_open_file *of,
1287 struct seq_file *s, void *v)
1289 struct rdtgroup *rdtgrp;
1290 struct rdt_resource *r;
1291 struct rdt_domain *d;
1297 rdtgrp = rdtgroup_kn_lock_live(of->kn);
1299 rdtgroup_kn_unlock(of->kn);
1303 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
1304 if (!rdtgrp->plr->d) {
1305 rdt_last_cmd_clear();
1306 rdt_last_cmd_puts("Cache domain offline\n");
1309 seq_printf(s, "%*s:", max_name_width,
1310 rdtgrp->plr->r->name);
1311 size = rdtgroup_cbm_to_size(rdtgrp->plr->r,
1314 seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size);
1319 for_each_alloc_enabled_rdt_resource(r) {
1321 seq_printf(s, "%*s:", max_name_width, r->name);
1322 list_for_each_entry(d, &r->domains, list) {
1325 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1328 ctrl = (!is_mba_sc(r) ?
1329 d->ctrl_val[rdtgrp->closid] :
1330 d->mbps_val[rdtgrp->closid]);
1331 if (r->rid == RDT_RESOURCE_MBA)
1334 size = rdtgroup_cbm_to_size(r, d, ctrl);
1336 seq_printf(s, "%d=%u", d->id, size);
1343 rdtgroup_kn_unlock(of->kn);
1348 /* rdtgroup information files for one cache resource. */
1349 static struct rftype res_common_files[] = {
1351 .name = "last_cmd_status",
1353 .kf_ops = &rdtgroup_kf_single_ops,
1354 .seq_show = rdt_last_cmd_status_show,
1355 .fflags = RF_TOP_INFO,
1358 .name = "num_closids",
1360 .kf_ops = &rdtgroup_kf_single_ops,
1361 .seq_show = rdt_num_closids_show,
1362 .fflags = RF_CTRL_INFO,
1365 .name = "mon_features",
1367 .kf_ops = &rdtgroup_kf_single_ops,
1368 .seq_show = rdt_mon_features_show,
1369 .fflags = RF_MON_INFO,
1372 .name = "num_rmids",
1374 .kf_ops = &rdtgroup_kf_single_ops,
1375 .seq_show = rdt_num_rmids_show,
1376 .fflags = RF_MON_INFO,
1381 .kf_ops = &rdtgroup_kf_single_ops,
1382 .seq_show = rdt_default_ctrl_show,
1383 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1386 .name = "min_cbm_bits",
1388 .kf_ops = &rdtgroup_kf_single_ops,
1389 .seq_show = rdt_min_cbm_bits_show,
1390 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1393 .name = "shareable_bits",
1395 .kf_ops = &rdtgroup_kf_single_ops,
1396 .seq_show = rdt_shareable_bits_show,
1397 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1400 .name = "bit_usage",
1402 .kf_ops = &rdtgroup_kf_single_ops,
1403 .seq_show = rdt_bit_usage_show,
1404 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
1407 .name = "min_bandwidth",
1409 .kf_ops = &rdtgroup_kf_single_ops,
1410 .seq_show = rdt_min_bw_show,
1411 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1414 .name = "bandwidth_gran",
1416 .kf_ops = &rdtgroup_kf_single_ops,
1417 .seq_show = rdt_bw_gran_show,
1418 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1421 .name = "delay_linear",
1423 .kf_ops = &rdtgroup_kf_single_ops,
1424 .seq_show = rdt_delay_linear_show,
1425 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
1428 .name = "max_threshold_occupancy",
1430 .kf_ops = &rdtgroup_kf_single_ops,
1431 .write = max_threshold_occ_write,
1432 .seq_show = max_threshold_occ_show,
1433 .fflags = RF_MON_INFO | RFTYPE_RES_CACHE,
1438 .kf_ops = &rdtgroup_kf_single_ops,
1439 .write = rdtgroup_cpus_write,
1440 .seq_show = rdtgroup_cpus_show,
1441 .fflags = RFTYPE_BASE,
1444 .name = "cpus_list",
1446 .kf_ops = &rdtgroup_kf_single_ops,
1447 .write = rdtgroup_cpus_write,
1448 .seq_show = rdtgroup_cpus_show,
1449 .flags = RFTYPE_FLAGS_CPUS_LIST,
1450 .fflags = RFTYPE_BASE,
1455 .kf_ops = &rdtgroup_kf_single_ops,
1456 .write = rdtgroup_tasks_write,
1457 .seq_show = rdtgroup_tasks_show,
1458 .fflags = RFTYPE_BASE,
1463 .kf_ops = &rdtgroup_kf_single_ops,
1464 .write = rdtgroup_schemata_write,
1465 .seq_show = rdtgroup_schemata_show,
1466 .fflags = RF_CTRL_BASE,
1471 .kf_ops = &rdtgroup_kf_single_ops,
1472 .write = rdtgroup_mode_write,
1473 .seq_show = rdtgroup_mode_show,
1474 .fflags = RF_CTRL_BASE,
1479 .kf_ops = &rdtgroup_kf_single_ops,
1480 .seq_show = rdtgroup_size_show,
1481 .fflags = RF_CTRL_BASE,
1486 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
1488 struct rftype *rfts, *rft;
1491 rfts = res_common_files;
1492 len = ARRAY_SIZE(res_common_files);
1494 lockdep_assert_held(&rdtgroup_mutex);
1496 for (rft = rfts; rft < rfts + len; rft++) {
1497 if ((fflags & rft->fflags) == rft->fflags) {
1498 ret = rdtgroup_add_file(kn, rft);
1506 pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
1507 while (--rft >= rfts) {
1508 if ((fflags & rft->fflags) == rft->fflags)
1509 kernfs_remove_by_name(kn, rft->name);
1515 * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file
1516 * @r: The resource group with which the file is associated.
1517 * @name: Name of the file
1519 * The permissions of named resctrl file, directory, or link are modified
1520 * to not allow read, write, or execute by any user.
1522 * WARNING: This function is intended to communicate to the user that the
1523 * resctrl file has been locked down - that it is not relevant to the
1524 * particular state the system finds itself in. It should not be relied
1525 * on to protect from user access because after the file's permissions
1526 * are restricted the user can still change the permissions using chmod
1527 * from the command line.
1529 * Return: 0 on success, <0 on failure.
1531 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name)
1533 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1534 struct kernfs_node *kn;
1537 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1541 switch (kernfs_type(kn)) {
1543 iattr.ia_mode = S_IFDIR;
1546 iattr.ia_mode = S_IFREG;
1549 iattr.ia_mode = S_IFLNK;
1553 ret = kernfs_setattr(kn, &iattr);
1559 * rdtgroup_kn_mode_restore - Restore user access to named resctrl file
1560 * @r: The resource group with which the file is associated.
1561 * @name: Name of the file
1562 * @mask: Mask of permissions that should be restored
1564 * Restore the permissions of the named file. If @name is a directory the
1565 * permissions of its parent will be used.
1567 * Return: 0 on success, <0 on failure.
1569 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
1572 struct iattr iattr = {.ia_valid = ATTR_MODE,};
1573 struct kernfs_node *kn, *parent;
1574 struct rftype *rfts, *rft;
1577 rfts = res_common_files;
1578 len = ARRAY_SIZE(res_common_files);
1580 for (rft = rfts; rft < rfts + len; rft++) {
1581 if (!strcmp(rft->name, name))
1582 iattr.ia_mode = rft->mode & mask;
1585 kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1589 switch (kernfs_type(kn)) {
1591 parent = kernfs_get_parent(kn);
1593 iattr.ia_mode |= parent->mode;
1596 iattr.ia_mode |= S_IFDIR;
1599 iattr.ia_mode |= S_IFREG;
1602 iattr.ia_mode |= S_IFLNK;
1606 ret = kernfs_setattr(kn, &iattr);
1611 static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name,
1612 unsigned long fflags)
1614 struct kernfs_node *kn_subdir;
1617 kn_subdir = kernfs_create_dir(kn_info, name,
1619 if (IS_ERR(kn_subdir))
1620 return PTR_ERR(kn_subdir);
1622 kernfs_get(kn_subdir);
1623 ret = rdtgroup_kn_set_ugid(kn_subdir);
1627 ret = rdtgroup_add_files(kn_subdir, fflags);
1629 kernfs_activate(kn_subdir);
1634 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
1636 struct rdt_resource *r;
1637 unsigned long fflags;
1641 /* create the directory */
1642 kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
1643 if (IS_ERR(kn_info))
1644 return PTR_ERR(kn_info);
1645 kernfs_get(kn_info);
1647 ret = rdtgroup_add_files(kn_info, RF_TOP_INFO);
1651 for_each_alloc_enabled_rdt_resource(r) {
1652 fflags = r->fflags | RF_CTRL_INFO;
1653 ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags);
1658 for_each_mon_enabled_rdt_resource(r) {
1659 fflags = r->fflags | RF_MON_INFO;
1660 sprintf(name, "%s_MON", r->name);
1661 ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
1667 * This extra ref will be put in kernfs_remove() and guarantees
1668 * that @rdtgrp->kn is always accessible.
1670 kernfs_get(kn_info);
1672 ret = rdtgroup_kn_set_ugid(kn_info);
1676 kernfs_activate(kn_info);
1681 kernfs_remove(kn_info);
1686 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
1687 char *name, struct kernfs_node **dest_kn)
1689 struct kernfs_node *kn;
1692 /* create the directory */
1693 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1701 * This extra ref will be put in kernfs_remove() and guarantees
1702 * that @rdtgrp->kn is always accessible.
1706 ret = rdtgroup_kn_set_ugid(kn);
1710 kernfs_activate(kn);
1719 static void l3_qos_cfg_update(void *arg)
1723 wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL);
1726 static void l2_qos_cfg_update(void *arg)
1730 wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL);
1733 static inline bool is_mba_linear(void)
1735 return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear;
1738 static int set_cache_qos_cfg(int level, bool enable)
1740 void (*update)(void *arg);
1741 struct rdt_resource *r_l;
1742 cpumask_var_t cpu_mask;
1743 struct rdt_domain *d;
1746 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1749 if (level == RDT_RESOURCE_L3)
1750 update = l3_qos_cfg_update;
1751 else if (level == RDT_RESOURCE_L2)
1752 update = l2_qos_cfg_update;
1756 r_l = &rdt_resources_all[level];
1757 list_for_each_entry(d, &r_l->domains, list) {
1758 /* Pick one CPU from each domain instance to update MSR */
1759 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1762 /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1763 if (cpumask_test_cpu(cpu, cpu_mask))
1765 /* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1766 smp_call_function_many(cpu_mask, update, &enable, 1);
1769 free_cpumask_var(cpu_mask);
1775 * Enable or disable the MBA software controller
1776 * which helps user specify bandwidth in MBps.
1777 * MBA software controller is supported only if
1778 * MBM is supported and MBA is in linear scale.
1780 static int set_mba_sc(bool mba_sc)
1782 struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA];
1783 struct rdt_domain *d;
1785 if (!is_mbm_enabled() || !is_mba_linear() ||
1786 mba_sc == is_mba_sc(r))
1789 r->membw.mba_sc = mba_sc;
1790 list_for_each_entry(d, &r->domains, list)
1791 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val);
1796 static int cdp_enable(int level, int data_type, int code_type)
1798 struct rdt_resource *r_ldata = &rdt_resources_all[data_type];
1799 struct rdt_resource *r_lcode = &rdt_resources_all[code_type];
1800 struct rdt_resource *r_l = &rdt_resources_all[level];
1803 if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1804 !r_lcode->alloc_capable)
1807 ret = set_cache_qos_cfg(level, true);
1809 r_l->alloc_enabled = false;
1810 r_ldata->alloc_enabled = true;
1811 r_lcode->alloc_enabled = true;
1816 static int cdpl3_enable(void)
1818 return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1819 RDT_RESOURCE_L3CODE);
1822 static int cdpl2_enable(void)
1824 return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1825 RDT_RESOURCE_L2CODE);
1828 static void cdp_disable(int level, int data_type, int code_type)
1830 struct rdt_resource *r = &rdt_resources_all[level];
1832 r->alloc_enabled = r->alloc_capable;
1834 if (rdt_resources_all[data_type].alloc_enabled) {
1835 rdt_resources_all[data_type].alloc_enabled = false;
1836 rdt_resources_all[code_type].alloc_enabled = false;
1837 set_cache_qos_cfg(level, false);
1841 static void cdpl3_disable(void)
1843 cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
1846 static void cdpl2_disable(void)
1848 cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
1851 static void cdp_disable_all(void)
1853 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
1855 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
1860 * We don't allow rdtgroup directories to be created anywhere
1861 * except the root directory. Thus when looking for the rdtgroup
1862 * structure for a kernfs node we are either looking at a directory,
1863 * in which case the rdtgroup structure is pointed at by the "priv"
1864 * field, otherwise we have a file, and need only look to the parent
1865 * to find the rdtgroup.
1867 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
1869 if (kernfs_type(kn) == KERNFS_DIR) {
1871 * All the resource directories use "kn->priv"
1872 * to point to the "struct rdtgroup" for the
1873 * resource. "info" and its subdirectories don't
1874 * have rdtgroup structures, so return NULL here.
1876 if (kn == kn_info || kn->parent == kn_info)
1881 return kn->parent->priv;
1885 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
1887 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1892 atomic_inc(&rdtgrp->waitcount);
1893 kernfs_break_active_protection(kn);
1895 mutex_lock(&rdtgroup_mutex);
1897 /* Was this group deleted while we waited? */
1898 if (rdtgrp->flags & RDT_DELETED)
1904 void rdtgroup_kn_unlock(struct kernfs_node *kn)
1906 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1911 mutex_unlock(&rdtgroup_mutex);
1913 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
1914 (rdtgrp->flags & RDT_DELETED)) {
1915 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
1916 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
1917 rdtgroup_pseudo_lock_remove(rdtgrp);
1918 kernfs_unbreak_active_protection(kn);
1919 kernfs_put(rdtgrp->kn);
1922 kernfs_unbreak_active_protection(kn);
1926 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1927 struct rdtgroup *prgrp,
1928 struct kernfs_node **mon_data_kn);
1930 static int rdt_enable_ctx(struct rdt_fs_context *ctx)
1934 if (ctx->enable_cdpl2)
1935 ret = cdpl2_enable();
1937 if (!ret && ctx->enable_cdpl3)
1938 ret = cdpl3_enable();
1940 if (!ret && ctx->enable_mba_mbps)
1941 ret = set_mba_sc(true);
1946 static int rdt_get_tree(struct fs_context *fc)
1948 struct rdt_fs_context *ctx = rdt_fc2context(fc);
1949 struct rdt_domain *dom;
1950 struct rdt_resource *r;
1954 mutex_lock(&rdtgroup_mutex);
1956 * resctrl file system can only be mounted once.
1958 if (static_branch_unlikely(&rdt_enable_key)) {
1963 ret = rdt_enable_ctx(ctx);
1969 ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
1973 if (rdt_mon_capable) {
1974 ret = mongroup_create_dir(rdtgroup_default.kn,
1979 kernfs_get(kn_mongrp);
1981 ret = mkdir_mondata_all(rdtgroup_default.kn,
1982 &rdtgroup_default, &kn_mondata);
1985 kernfs_get(kn_mondata);
1986 rdtgroup_default.mon.mon_data_kn = kn_mondata;
1989 ret = rdt_pseudo_lock_init();
1993 ret = kernfs_get_tree(fc);
1997 if (rdt_alloc_capable)
1998 static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
1999 if (rdt_mon_capable)
2000 static_branch_enable_cpuslocked(&rdt_mon_enable_key);
2002 if (rdt_alloc_capable || rdt_mon_capable)
2003 static_branch_enable_cpuslocked(&rdt_enable_key);
2005 if (is_mbm_enabled()) {
2006 r = &rdt_resources_all[RDT_RESOURCE_L3];
2007 list_for_each_entry(dom, &r->domains, list)
2008 mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
2014 rdt_pseudo_lock_release();
2016 if (rdt_mon_capable)
2017 kernfs_remove(kn_mondata);
2019 if (rdt_mon_capable)
2020 kernfs_remove(kn_mongrp);
2022 kernfs_remove(kn_info);
2024 if (ctx->enable_mba_mbps)
2029 rdt_last_cmd_clear();
2030 mutex_unlock(&rdtgroup_mutex);
2042 static const struct fs_parameter_spec rdt_param_specs[] = {
2043 fsparam_flag("cdp", Opt_cdp),
2044 fsparam_flag("cdpl2", Opt_cdpl2),
2045 fsparam_flag("mba_MBps", Opt_mba_mbps),
2049 static const struct fs_parameter_description rdt_fs_parameters = {
2051 .specs = rdt_param_specs,
2054 static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param)
2056 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2057 struct fs_parse_result result;
2060 opt = fs_parse(fc, &rdt_fs_parameters, param, &result);
2066 ctx->enable_cdpl3 = true;
2069 ctx->enable_cdpl2 = true;
2072 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2074 ctx->enable_mba_mbps = true;
2081 static void rdt_fs_context_free(struct fs_context *fc)
2083 struct rdt_fs_context *ctx = rdt_fc2context(fc);
2085 kernfs_free_fs_context(fc);
2089 static const struct fs_context_operations rdt_fs_context_ops = {
2090 .free = rdt_fs_context_free,
2091 .parse_param = rdt_parse_param,
2092 .get_tree = rdt_get_tree,
2095 static int rdt_init_fs_context(struct fs_context *fc)
2097 struct rdt_fs_context *ctx;
2099 ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL);
2103 ctx->kfc.root = rdt_root;
2104 ctx->kfc.magic = RDTGROUP_SUPER_MAGIC;
2105 fc->fs_private = &ctx->kfc;
2106 fc->ops = &rdt_fs_context_ops;
2107 put_user_ns(fc->user_ns);
2108 fc->user_ns = get_user_ns(&init_user_ns);
2113 static int reset_all_ctrls(struct rdt_resource *r)
2115 struct msr_param msr_param;
2116 cpumask_var_t cpu_mask;
2117 struct rdt_domain *d;
2120 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
2125 msr_param.high = r->num_closid;
2128 * Disable resource control for this resource by setting all
2129 * CBMs in all domains to the maximum mask value. Pick one CPU
2130 * from each domain to update the MSRs below.
2132 list_for_each_entry(d, &r->domains, list) {
2133 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
2135 for (i = 0; i < r->num_closid; i++)
2136 d->ctrl_val[i] = r->default_ctrl;
2139 /* Update CBM on this cpu if it's in cpu_mask. */
2140 if (cpumask_test_cpu(cpu, cpu_mask))
2141 rdt_ctrl_update(&msr_param);
2142 /* Update CBM on all other cpus in cpu_mask. */
2143 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
2146 free_cpumask_var(cpu_mask);
2151 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
2153 return (rdt_alloc_capable &&
2154 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
2157 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
2159 return (rdt_mon_capable &&
2160 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
2164 * Move tasks from one to the other group. If @from is NULL, then all tasks
2165 * in the systems are moved unconditionally (used for teardown).
2167 * If @mask is not NULL the cpus on which moved tasks are running are set
2168 * in that mask so the update smp function call is restricted to affected
2171 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
2172 struct cpumask *mask)
2174 struct task_struct *p, *t;
2176 read_lock(&tasklist_lock);
2177 for_each_process_thread(p, t) {
2178 if (!from || is_closid_match(t, from) ||
2179 is_rmid_match(t, from)) {
2180 t->closid = to->closid;
2181 t->rmid = to->mon.rmid;
2185 * This is safe on x86 w/o barriers as the ordering
2186 * of writing to task_cpu() and t->on_cpu is
2187 * reverse to the reading here. The detection is
2188 * inaccurate as tasks might move or schedule
2189 * before the smp function call takes place. In
2190 * such a case the function call is pointless, but
2191 * there is no other side effect.
2193 if (mask && t->on_cpu)
2194 cpumask_set_cpu(task_cpu(t), mask);
2198 read_unlock(&tasklist_lock);
2201 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
2203 struct rdtgroup *sentry, *stmp;
2204 struct list_head *head;
2206 head = &rdtgrp->mon.crdtgrp_list;
2207 list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
2208 free_rmid(sentry->mon.rmid);
2209 list_del(&sentry->mon.crdtgrp_list);
2215 * Forcibly remove all of subdirectories under root.
2217 static void rmdir_all_sub(void)
2219 struct rdtgroup *rdtgrp, *tmp;
2221 /* Move all tasks to the default resource group */
2222 rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
2224 list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
2225 /* Free any child rmids */
2226 free_all_child_rdtgrp(rdtgrp);
2228 /* Remove each rdtgroup other than root */
2229 if (rdtgrp == &rdtgroup_default)
2232 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2233 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2234 rdtgroup_pseudo_lock_remove(rdtgrp);
2237 * Give any CPUs back to the default group. We cannot copy
2238 * cpu_online_mask because a CPU might have executed the
2239 * offline callback already, but is still marked online.
2241 cpumask_or(&rdtgroup_default.cpu_mask,
2242 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2244 free_rmid(rdtgrp->mon.rmid);
2246 kernfs_remove(rdtgrp->kn);
2247 list_del(&rdtgrp->rdtgroup_list);
2250 /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
2251 update_closid_rmid(cpu_online_mask, &rdtgroup_default);
2253 kernfs_remove(kn_info);
2254 kernfs_remove(kn_mongrp);
2255 kernfs_remove(kn_mondata);
2258 static void rdt_kill_sb(struct super_block *sb)
2260 struct rdt_resource *r;
2263 mutex_lock(&rdtgroup_mutex);
2267 /*Put everything back to default values. */
2268 for_each_alloc_enabled_rdt_resource(r)
2272 rdt_pseudo_lock_release();
2273 rdtgroup_default.mode = RDT_MODE_SHAREABLE;
2274 static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
2275 static_branch_disable_cpuslocked(&rdt_mon_enable_key);
2276 static_branch_disable_cpuslocked(&rdt_enable_key);
2278 mutex_unlock(&rdtgroup_mutex);
2282 static struct file_system_type rdt_fs_type = {
2284 .init_fs_context = rdt_init_fs_context,
2285 .parameters = &rdt_fs_parameters,
2286 .kill_sb = rdt_kill_sb,
2289 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
2292 struct kernfs_node *kn;
2295 kn = __kernfs_create_file(parent_kn, name, 0444,
2296 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
2297 &kf_mondata_ops, priv, NULL, NULL);
2301 ret = rdtgroup_kn_set_ugid(kn);
2311 * Remove all subdirectories of mon_data of ctrl_mon groups
2312 * and monitor groups with given domain id.
2314 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
2316 struct rdtgroup *prgrp, *crgrp;
2319 if (!r->mon_enabled)
2322 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2323 sprintf(name, "mon_%s_%02d", r->name, dom_id);
2324 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
2326 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
2327 kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
2331 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
2332 struct rdt_domain *d,
2333 struct rdt_resource *r, struct rdtgroup *prgrp)
2335 union mon_data_bits priv;
2336 struct kernfs_node *kn;
2337 struct mon_evt *mevt;
2338 struct rmid_read rr;
2342 sprintf(name, "mon_%s_%02d", r->name, d->id);
2343 /* create the directory */
2344 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
2349 * This extra ref will be put in kernfs_remove() and guarantees
2350 * that kn is always accessible.
2353 ret = rdtgroup_kn_set_ugid(kn);
2357 if (WARN_ON(list_empty(&r->evt_list))) {
2362 priv.u.rid = r->rid;
2363 priv.u.domid = d->id;
2364 list_for_each_entry(mevt, &r->evt_list, list) {
2365 priv.u.evtid = mevt->evtid;
2366 ret = mon_addfile(kn, mevt->name, priv.priv);
2370 if (is_mbm_event(mevt->evtid))
2371 mon_event_read(&rr, d, prgrp, mevt->evtid, true);
2373 kernfs_activate(kn);
2382 * Add all subdirectories of mon_data for "ctrl_mon" groups
2383 * and "monitor" groups with given domain id.
2385 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
2386 struct rdt_domain *d)
2388 struct kernfs_node *parent_kn;
2389 struct rdtgroup *prgrp, *crgrp;
2390 struct list_head *head;
2392 if (!r->mon_enabled)
2395 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2396 parent_kn = prgrp->mon.mon_data_kn;
2397 mkdir_mondata_subdir(parent_kn, d, r, prgrp);
2399 head = &prgrp->mon.crdtgrp_list;
2400 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
2401 parent_kn = crgrp->mon.mon_data_kn;
2402 mkdir_mondata_subdir(parent_kn, d, r, crgrp);
2407 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
2408 struct rdt_resource *r,
2409 struct rdtgroup *prgrp)
2411 struct rdt_domain *dom;
2414 list_for_each_entry(dom, &r->domains, list) {
2415 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
2424 * This creates a directory mon_data which contains the monitored data.
2426 * mon_data has one directory for each domain whic are named
2427 * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
2428 * with L3 domain looks as below:
2435 * Each domain directory has one file per event:
2440 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2441 struct rdtgroup *prgrp,
2442 struct kernfs_node **dest_kn)
2444 struct rdt_resource *r;
2445 struct kernfs_node *kn;
2449 * Create the mon_data directory first.
2451 ret = mongroup_create_dir(parent_kn, NULL, "mon_data", &kn);
2459 * Create the subdirectories for each domain. Note that all events
2460 * in a domain like L3 are grouped into a resource whose domain is L3
2462 for_each_mon_enabled_rdt_resource(r) {
2463 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
2476 * cbm_ensure_valid - Enforce validity on provided CBM
2477 * @_val: Candidate CBM
2478 * @r: RDT resource to which the CBM belongs
2480 * The provided CBM represents all cache portions available for use. This
2481 * may be represented by a bitmap that does not consist of contiguous ones
2482 * and thus be an invalid CBM.
2483 * Here the provided CBM is forced to be a valid CBM by only considering
2484 * the first set of contiguous bits as valid and clearing all bits.
2485 * The intention here is to provide a valid default CBM with which a new
2486 * resource group is initialized. The user can follow this with a
2487 * modification to the CBM if the default does not satisfy the
2490 static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r)
2492 unsigned int cbm_len = r->cache.cbm_len;
2493 unsigned long first_bit, zero_bit;
2494 unsigned long val = _val;
2499 first_bit = find_first_bit(&val, cbm_len);
2500 zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
2502 /* Clear any remaining bits to ensure contiguous region */
2503 bitmap_clear(&val, zero_bit, cbm_len - zero_bit);
2508 * Initialize cache resources per RDT domain
2510 * Set the RDT domain up to start off with all usable allocations. That is,
2511 * all shareable and unused bits. All-zero CBM is invalid.
2513 static int __init_one_rdt_domain(struct rdt_domain *d, struct rdt_resource *r,
2516 struct rdt_resource *r_cdp = NULL;
2517 struct rdt_domain *d_cdp = NULL;
2518 u32 used_b = 0, unused_b = 0;
2519 unsigned long tmp_cbm;
2520 enum rdtgrp_mode mode;
2521 u32 peer_ctl, *ctrl;
2524 rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp);
2525 d->have_new_ctrl = false;
2526 d->new_ctrl = r->cache.shareable_bits;
2527 used_b = r->cache.shareable_bits;
2529 for (i = 0; i < closids_supported(); i++, ctrl++) {
2530 if (closid_allocated(i) && i != closid) {
2531 mode = rdtgroup_mode_by_closid(i);
2532 if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
2534 * ctrl values for locksetup aren't relevant
2535 * until the schemata is written, and the mode
2536 * becomes RDT_MODE_PSEUDO_LOCKED.
2540 * If CDP is active include peer domain's
2541 * usage to ensure there is no overlap
2542 * with an exclusive group.
2545 peer_ctl = d_cdp->ctrl_val[i];
2548 used_b |= *ctrl | peer_ctl;
2549 if (mode == RDT_MODE_SHAREABLE)
2550 d->new_ctrl |= *ctrl | peer_ctl;
2553 if (d->plr && d->plr->cbm > 0)
2554 used_b |= d->plr->cbm;
2555 unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
2556 unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
2557 d->new_ctrl |= unused_b;
2559 * Force the initial CBM to be valid, user can
2560 * modify the CBM based on system availability.
2562 d->new_ctrl = cbm_ensure_valid(d->new_ctrl, r);
2564 * Assign the u32 CBM to an unsigned long to ensure that
2565 * bitmap_weight() does not access out-of-bound memory.
2567 tmp_cbm = d->new_ctrl;
2568 if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) {
2569 rdt_last_cmd_printf("No space on %s:%d\n", r->name, d->id);
2572 d->have_new_ctrl = true;
2578 * Initialize cache resources with default values.
2580 * A new RDT group is being created on an allocation capable (CAT)
2581 * supporting system. Set this group up to start off with all usable
2584 * If there are no more shareable bits available on any domain then
2585 * the entire allocation will fail.
2587 static int rdtgroup_init_cat(struct rdt_resource *r, u32 closid)
2589 struct rdt_domain *d;
2592 list_for_each_entry(d, &r->domains, list) {
2593 ret = __init_one_rdt_domain(d, r, closid);
2601 /* Initialize MBA resource with default values. */
2602 static void rdtgroup_init_mba(struct rdt_resource *r)
2604 struct rdt_domain *d;
2606 list_for_each_entry(d, &r->domains, list) {
2607 d->new_ctrl = is_mba_sc(r) ? MBA_MAX_MBPS : r->default_ctrl;
2608 d->have_new_ctrl = true;
2612 /* Initialize the RDT group's allocations. */
2613 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
2615 struct rdt_resource *r;
2618 for_each_alloc_enabled_rdt_resource(r) {
2619 if (r->rid == RDT_RESOURCE_MBA) {
2620 rdtgroup_init_mba(r);
2622 ret = rdtgroup_init_cat(r, rdtgrp->closid);
2627 ret = update_domains(r, rdtgrp->closid);
2629 rdt_last_cmd_puts("Failed to initialize allocations\n");
2635 rdtgrp->mode = RDT_MODE_SHAREABLE;
2640 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
2641 struct kernfs_node *prgrp_kn,
2642 const char *name, umode_t mode,
2643 enum rdt_group_type rtype, struct rdtgroup **r)
2645 struct rdtgroup *prdtgrp, *rdtgrp;
2646 struct kernfs_node *kn;
2650 prdtgrp = rdtgroup_kn_lock_live(prgrp_kn);
2651 rdt_last_cmd_clear();
2654 rdt_last_cmd_puts("Directory was removed\n");
2658 if (rtype == RDTMON_GROUP &&
2659 (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2660 prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
2662 rdt_last_cmd_puts("Pseudo-locking in progress\n");
2666 /* allocate the rdtgroup. */
2667 rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
2670 rdt_last_cmd_puts("Kernel out of memory\n");
2674 rdtgrp->mon.parent = prdtgrp;
2675 rdtgrp->type = rtype;
2676 INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
2678 /* kernfs creates the directory for rdtgrp */
2679 kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
2682 rdt_last_cmd_puts("kernfs create error\n");
2688 * kernfs_remove() will drop the reference count on "kn" which
2689 * will free it. But we still need it to stick around for the
2690 * rdtgroup_kn_unlock(kn} call below. Take one extra reference
2691 * here, which will be dropped inside rdtgroup_kn_unlock().
2695 ret = rdtgroup_kn_set_ugid(kn);
2697 rdt_last_cmd_puts("kernfs perm error\n");
2701 files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
2702 ret = rdtgroup_add_files(kn, files);
2704 rdt_last_cmd_puts("kernfs fill error\n");
2708 if (rdt_mon_capable) {
2711 rdt_last_cmd_puts("Out of RMIDs\n");
2714 rdtgrp->mon.rmid = ret;
2716 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
2718 rdt_last_cmd_puts("kernfs subdir error\n");
2722 kernfs_activate(kn);
2725 * The caller unlocks the prgrp_kn upon success.
2730 free_rmid(rdtgrp->mon.rmid);
2732 kernfs_remove(rdtgrp->kn);
2736 rdtgroup_kn_unlock(prgrp_kn);
2740 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
2742 kernfs_remove(rgrp->kn);
2743 free_rmid(rgrp->mon.rmid);
2748 * Create a monitor group under "mon_groups" directory of a control
2749 * and monitor group(ctrl_mon). This is a resource group
2750 * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
2752 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
2753 struct kernfs_node *prgrp_kn,
2757 struct rdtgroup *rdtgrp, *prgrp;
2760 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP,
2765 prgrp = rdtgrp->mon.parent;
2766 rdtgrp->closid = prgrp->closid;
2769 * Add the rdtgrp to the list of rdtgrps the parent
2770 * ctrl_mon group has to track.
2772 list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
2774 rdtgroup_kn_unlock(prgrp_kn);
2779 * These are rdtgroups created under the root directory. Can be used
2780 * to allocate and monitor resources.
2782 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
2783 struct kernfs_node *prgrp_kn,
2784 const char *name, umode_t mode)
2786 struct rdtgroup *rdtgrp;
2787 struct kernfs_node *kn;
2791 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP,
2797 ret = closid_alloc();
2799 rdt_last_cmd_puts("Out of CLOSIDs\n");
2800 goto out_common_fail;
2805 rdtgrp->closid = closid;
2806 ret = rdtgroup_init_alloc(rdtgrp);
2810 list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
2812 if (rdt_mon_capable) {
2814 * Create an empty mon_groups directory to hold the subset
2815 * of tasks and cpus to monitor.
2817 ret = mongroup_create_dir(kn, NULL, "mon_groups", NULL);
2819 rdt_last_cmd_puts("kernfs subdir error\n");
2827 list_del(&rdtgrp->rdtgroup_list);
2829 closid_free(closid);
2831 mkdir_rdt_prepare_clean(rdtgrp);
2833 rdtgroup_kn_unlock(prgrp_kn);
2838 * We allow creating mon groups only with in a directory called "mon_groups"
2839 * which is present in every ctrl_mon group. Check if this is a valid
2840 * "mon_groups" directory.
2842 * 1. The directory should be named "mon_groups".
2843 * 2. The mon group itself should "not" be named "mon_groups".
2844 * This makes sure "mon_groups" directory always has a ctrl_mon group
2847 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
2849 return (!strcmp(kn->name, "mon_groups") &&
2850 strcmp(name, "mon_groups"));
2853 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
2856 /* Do not accept '\n' to avoid unparsable situation. */
2857 if (strchr(name, '\n'))
2861 * If the parent directory is the root directory and RDT
2862 * allocation is supported, add a control and monitoring
2865 if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
2866 return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode);
2869 * If RDT monitoring is supported and the parent directory is a valid
2870 * "mon_groups" directory, add a monitoring subdirectory.
2872 if (rdt_mon_capable && is_mon_groups(parent_kn, name))
2873 return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode);
2878 static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2879 cpumask_var_t tmpmask)
2881 struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
2884 /* Give any tasks back to the parent group */
2885 rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
2887 /* Update per cpu rmid of the moved CPUs first */
2888 for_each_cpu(cpu, &rdtgrp->cpu_mask)
2889 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
2891 * Update the MSR on moved CPUs and CPUs which have moved
2892 * task running on them.
2894 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2895 update_closid_rmid(tmpmask, NULL);
2897 rdtgrp->flags = RDT_DELETED;
2898 free_rmid(rdtgrp->mon.rmid);
2901 * Remove the rdtgrp from the parent ctrl_mon group's list
2903 WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
2904 list_del(&rdtgrp->mon.crdtgrp_list);
2907 * one extra hold on this, will drop when we kfree(rdtgrp)
2908 * in rdtgroup_kn_unlock()
2911 kernfs_remove(rdtgrp->kn);
2916 static int rdtgroup_ctrl_remove(struct kernfs_node *kn,
2917 struct rdtgroup *rdtgrp)
2919 rdtgrp->flags = RDT_DELETED;
2920 list_del(&rdtgrp->rdtgroup_list);
2923 * one extra hold on this, will drop when we kfree(rdtgrp)
2924 * in rdtgroup_kn_unlock()
2927 kernfs_remove(rdtgrp->kn);
2931 static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2932 cpumask_var_t tmpmask)
2936 /* Give any tasks back to the default group */
2937 rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
2939 /* Give any CPUs back to the default group */
2940 cpumask_or(&rdtgroup_default.cpu_mask,
2941 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2943 /* Update per cpu closid and rmid of the moved CPUs first */
2944 for_each_cpu(cpu, &rdtgrp->cpu_mask) {
2945 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
2946 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
2950 * Update the MSR on moved CPUs and CPUs which have moved
2951 * task running on them.
2953 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2954 update_closid_rmid(tmpmask, NULL);
2956 closid_free(rdtgrp->closid);
2957 free_rmid(rdtgrp->mon.rmid);
2960 * Free all the child monitor group rmids.
2962 free_all_child_rdtgrp(rdtgrp);
2964 rdtgroup_ctrl_remove(kn, rdtgrp);
2969 static int rdtgroup_rmdir(struct kernfs_node *kn)
2971 struct kernfs_node *parent_kn = kn->parent;
2972 struct rdtgroup *rdtgrp;
2973 cpumask_var_t tmpmask;
2976 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
2979 rdtgrp = rdtgroup_kn_lock_live(kn);
2986 * If the rdtgroup is a ctrl_mon group and parent directory
2987 * is the root directory, remove the ctrl_mon group.
2989 * If the rdtgroup is a mon group and parent directory
2990 * is a valid "mon_groups" directory, remove the mon group.
2992 if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn) {
2993 if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2994 rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
2995 ret = rdtgroup_ctrl_remove(kn, rdtgrp);
2997 ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask);
2999 } else if (rdtgrp->type == RDTMON_GROUP &&
3000 is_mon_groups(parent_kn, kn->name)) {
3001 ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask);
3007 rdtgroup_kn_unlock(kn);
3008 free_cpumask_var(tmpmask);
3012 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
3014 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
3015 seq_puts(seq, ",cdp");
3017 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
3018 seq_puts(seq, ",cdpl2");
3020 if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA]))
3021 seq_puts(seq, ",mba_MBps");
3026 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
3027 .mkdir = rdtgroup_mkdir,
3028 .rmdir = rdtgroup_rmdir,
3029 .show_options = rdtgroup_show_options,
3032 static int __init rdtgroup_setup_root(void)
3036 rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
3037 KERNFS_ROOT_CREATE_DEACTIVATED |
3038 KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
3040 if (IS_ERR(rdt_root))
3041 return PTR_ERR(rdt_root);
3043 mutex_lock(&rdtgroup_mutex);
3045 rdtgroup_default.closid = 0;
3046 rdtgroup_default.mon.rmid = 0;
3047 rdtgroup_default.type = RDTCTRL_GROUP;
3048 INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
3050 list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
3052 ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
3054 kernfs_destroy_root(rdt_root);
3058 rdtgroup_default.kn = rdt_root->kn;
3059 kernfs_activate(rdtgroup_default.kn);
3062 mutex_unlock(&rdtgroup_mutex);
3068 * rdtgroup_init - rdtgroup initialization
3070 * Setup resctrl file system including set up root, create mount point,
3071 * register rdtgroup filesystem, and initialize files under root directory.
3073 * Return: 0 on success or -errno
3075 int __init rdtgroup_init(void)
3079 seq_buf_init(&last_cmd_status, last_cmd_status_buf,
3080 sizeof(last_cmd_status_buf));
3082 ret = rdtgroup_setup_root();
3086 ret = sysfs_create_mount_point(fs_kobj, "resctrl");
3090 ret = register_filesystem(&rdt_fs_type);
3092 goto cleanup_mountpoint;
3095 * Adding the resctrl debugfs directory here may not be ideal since
3096 * it would let the resctrl debugfs directory appear on the debugfs
3097 * filesystem before the resctrl filesystem is mounted.
3098 * It may also be ok since that would enable debugging of RDT before
3099 * resctrl is mounted.
3100 * The reason why the debugfs directory is created here and not in
3101 * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and
3102 * during the debugfs directory creation also &sb->s_type->i_mutex_key
3103 * (the lockdep class of inode->i_rwsem). Other filesystem
3104 * interactions (eg. SyS_getdents) have the lock ordering:
3105 * &sb->s_type->i_mutex_key --> &mm->mmap_sem
3106 * During mmap(), called with &mm->mmap_sem, the rdtgroup_mutex
3107 * is taken, thus creating dependency:
3108 * &mm->mmap_sem --> rdtgroup_mutex for the latter that can cause
3109 * issues considering the other two lock dependencies.
3110 * By creating the debugfs directory here we avoid a dependency
3111 * that may cause deadlock (even though file operations cannot
3112 * occur until the filesystem is mounted, but I do not know how to
3113 * tell lockdep that).
3115 debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
3120 sysfs_remove_mount_point(fs_kobj, "resctrl");
3122 kernfs_destroy_root(rdt_root);
3127 void __exit rdtgroup_exit(void)
3129 debugfs_remove_recursive(debugfs_resctrl);
3130 unregister_filesystem(&rdt_fs_type);
3131 sysfs_remove_mount_point(fs_kobj, "resctrl");
3132 kernfs_destroy_root(rdt_root);