1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * User interface for Resource Allocation in Resource Director Technology(RDT)
4  *
5  * Copyright (C) 2016 Intel Corporation
6  *
7  * Author: Fenghua Yu <fenghua.yu@intel.com>
8  *
9  * More information about RDT be found in the Intel (R) x86 Architecture
10  * Software Developer Manual.
11  */
12 
13 #define pr_fmt(fmt)	KBUILD_MODNAME ": " fmt
14 
15 #include <linux/cacheinfo.h>
16 #include <linux/cpu.h>
17 #include <linux/debugfs.h>
18 #include <linux/fs.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>
29 
30 #include <uapi/linux/magic.h>
31 
32 #include <asm/resctrl.h>
33 #include "internal.h"
34 
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);
41 
42 /* list of entries for the schemata file */
43 LIST_HEAD(resctrl_schema_all);
44 
45 /* Kernel fs node for "info" directory under root */
46 static struct kernfs_node *kn_info;
47 
48 /* Kernel fs node for "mon_groups" directory under root */
49 static struct kernfs_node *kn_mongrp;
50 
51 /* Kernel fs node for "mon_data" directory under root */
52 static struct kernfs_node *kn_mondata;
53 
54 static struct seq_buf last_cmd_status;
55 static char last_cmd_status_buf[512];
56 
57 struct dentry *debugfs_resctrl;
58 
59 void rdt_last_cmd_clear(void)
60 {
61 	lockdep_assert_held(&rdtgroup_mutex);
62 	seq_buf_clear(&last_cmd_status);
63 }
64 
65 void rdt_last_cmd_puts(const char *s)
66 {
67 	lockdep_assert_held(&rdtgroup_mutex);
68 	seq_buf_puts(&last_cmd_status, s);
69 }
70 
71 void rdt_last_cmd_printf(const char *fmt, ...)
72 {
73 	va_list ap;
74 
75 	va_start(ap, fmt);
76 	lockdep_assert_held(&rdtgroup_mutex);
77 	seq_buf_vprintf(&last_cmd_status, fmt, ap);
78 	va_end(ap);
79 }
80 
81 /*
82  * Trivial allocator for CLOSIDs. Since h/w only supports a small number,
83  * we can keep a bitmap of free CLOSIDs in a single integer.
84  *
85  * Using a global CLOSID across all resources has some advantages and
86  * some drawbacks:
87  * + We can simply set "current->closid" to assign a task to a resource
88  *   group.
89  * + Context switch code can avoid extra memory references deciding which
90  *   CLOSID to load into the PQR_ASSOC MSR
91  * - We give up some options in configuring resource groups across multi-socket
92  *   systems.
93  * - Our choices on how to configure each resource become progressively more
94  *   limited as the number of resources grows.
95  */
96 static int closid_free_map;
97 static int closid_free_map_len;
98 
99 int closids_supported(void)
100 {
101 	return closid_free_map_len;
102 }
103 
104 static void closid_init(void)
105 {
106 	struct resctrl_schema *s;
107 	u32 rdt_min_closid = 32;
108 
109 	/* Compute rdt_min_closid across all resources */
110 	list_for_each_entry(s, &resctrl_schema_all, list)
111 		rdt_min_closid = min(rdt_min_closid, s->num_closid);
112 
113 	closid_free_map = BIT_MASK(rdt_min_closid) - 1;
114 
115 	/* CLOSID 0 is always reserved for the default group */
116 	closid_free_map &= ~1;
117 	closid_free_map_len = rdt_min_closid;
118 }
119 
120 static int closid_alloc(void)
121 {
122 	u32 closid = ffs(closid_free_map);
123 
124 	if (closid == 0)
125 		return -ENOSPC;
126 	closid--;
127 	closid_free_map &= ~(1 << closid);
128 
129 	return closid;
130 }
131 
132 void closid_free(int closid)
133 {
134 	closid_free_map |= 1 << closid;
135 }
136 
137 /**
138  * closid_allocated - test if provided closid is in use
139  * @closid: closid to be tested
140  *
141  * Return: true if @closid is currently associated with a resource group,
142  * false if @closid is free
143  */
144 static bool closid_allocated(unsigned int closid)
145 {
146 	return (closid_free_map & (1 << closid)) == 0;
147 }
148 
149 /**
150  * rdtgroup_mode_by_closid - Return mode of resource group with closid
151  * @closid: closid if the resource group
152  *
153  * Each resource group is associated with a @closid. Here the mode
154  * of a resource group can be queried by searching for it using its closid.
155  *
156  * Return: mode as &enum rdtgrp_mode of resource group with closid @closid
157  */
158 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid)
159 {
160 	struct rdtgroup *rdtgrp;
161 
162 	list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
163 		if (rdtgrp->closid == closid)
164 			return rdtgrp->mode;
165 	}
166 
167 	return RDT_NUM_MODES;
168 }
169 
170 static const char * const rdt_mode_str[] = {
171 	[RDT_MODE_SHAREABLE]		= "shareable",
172 	[RDT_MODE_EXCLUSIVE]		= "exclusive",
173 	[RDT_MODE_PSEUDO_LOCKSETUP]	= "pseudo-locksetup",
174 	[RDT_MODE_PSEUDO_LOCKED]	= "pseudo-locked",
175 };
176 
177 /**
178  * rdtgroup_mode_str - Return the string representation of mode
179  * @mode: the resource group mode as &enum rdtgroup_mode
180  *
181  * Return: string representation of valid mode, "unknown" otherwise
182  */
183 static const char *rdtgroup_mode_str(enum rdtgrp_mode mode)
184 {
185 	if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES)
186 		return "unknown";
187 
188 	return rdt_mode_str[mode];
189 }
190 
191 /* set uid and gid of rdtgroup dirs and files to that of the creator */
192 static int rdtgroup_kn_set_ugid(struct kernfs_node *kn)
193 {
194 	struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
195 				.ia_uid = current_fsuid(),
196 				.ia_gid = current_fsgid(), };
197 
198 	if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
199 	    gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
200 		return 0;
201 
202 	return kernfs_setattr(kn, &iattr);
203 }
204 
205 static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft)
206 {
207 	struct kernfs_node *kn;
208 	int ret;
209 
210 	kn = __kernfs_create_file(parent_kn, rft->name, rft->mode,
211 				  GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
212 				  0, rft->kf_ops, rft, NULL, NULL);
213 	if (IS_ERR(kn))
214 		return PTR_ERR(kn);
215 
216 	ret = rdtgroup_kn_set_ugid(kn);
217 	if (ret) {
218 		kernfs_remove(kn);
219 		return ret;
220 	}
221 
222 	return 0;
223 }
224 
225 static int rdtgroup_seqfile_show(struct seq_file *m, void *arg)
226 {
227 	struct kernfs_open_file *of = m->private;
228 	struct rftype *rft = of->kn->priv;
229 
230 	if (rft->seq_show)
231 		return rft->seq_show(of, m, arg);
232 	return 0;
233 }
234 
235 static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf,
236 				   size_t nbytes, loff_t off)
237 {
238 	struct rftype *rft = of->kn->priv;
239 
240 	if (rft->write)
241 		return rft->write(of, buf, nbytes, off);
242 
243 	return -EINVAL;
244 }
245 
246 static const struct kernfs_ops rdtgroup_kf_single_ops = {
247 	.atomic_write_len	= PAGE_SIZE,
248 	.write			= rdtgroup_file_write,
249 	.seq_show		= rdtgroup_seqfile_show,
250 };
251 
252 static const struct kernfs_ops kf_mondata_ops = {
253 	.atomic_write_len	= PAGE_SIZE,
254 	.seq_show		= rdtgroup_mondata_show,
255 };
256 
257 static bool is_cpu_list(struct kernfs_open_file *of)
258 {
259 	struct rftype *rft = of->kn->priv;
260 
261 	return rft->flags & RFTYPE_FLAGS_CPUS_LIST;
262 }
263 
264 static int rdtgroup_cpus_show(struct kernfs_open_file *of,
265 			      struct seq_file *s, void *v)
266 {
267 	struct rdtgroup *rdtgrp;
268 	struct cpumask *mask;
269 	int ret = 0;
270 
271 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
272 
273 	if (rdtgrp) {
274 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
275 			if (!rdtgrp->plr->d) {
276 				rdt_last_cmd_clear();
277 				rdt_last_cmd_puts("Cache domain offline\n");
278 				ret = -ENODEV;
279 			} else {
280 				mask = &rdtgrp->plr->d->cpu_mask;
281 				seq_printf(s, is_cpu_list(of) ?
282 					   "%*pbl\n" : "%*pb\n",
283 					   cpumask_pr_args(mask));
284 			}
285 		} else {
286 			seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n",
287 				   cpumask_pr_args(&rdtgrp->cpu_mask));
288 		}
289 	} else {
290 		ret = -ENOENT;
291 	}
292 	rdtgroup_kn_unlock(of->kn);
293 
294 	return ret;
295 }
296 
297 /*
298  * This is safe against resctrl_sched_in() called from __switch_to()
299  * because __switch_to() is executed with interrupts disabled. A local call
300  * from update_closid_rmid() is protected against __switch_to() because
301  * preemption is disabled.
302  */
303 static void update_cpu_closid_rmid(void *info)
304 {
305 	struct rdtgroup *r = info;
306 
307 	if (r) {
308 		this_cpu_write(pqr_state.default_closid, r->closid);
309 		this_cpu_write(pqr_state.default_rmid, r->mon.rmid);
310 	}
311 
312 	/*
313 	 * We cannot unconditionally write the MSR because the current
314 	 * executing task might have its own closid selected. Just reuse
315 	 * the context switch code.
316 	 */
317 	resctrl_sched_in();
318 }
319 
320 /*
321  * Update the PGR_ASSOC MSR on all cpus in @cpu_mask,
322  *
323  * Per task closids/rmids must have been set up before calling this function.
324  */
325 static void
326 update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r)
327 {
328 	int cpu = get_cpu();
329 
330 	if (cpumask_test_cpu(cpu, cpu_mask))
331 		update_cpu_closid_rmid(r);
332 	smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1);
333 	put_cpu();
334 }
335 
336 static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
337 			  cpumask_var_t tmpmask)
338 {
339 	struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp;
340 	struct list_head *head;
341 
342 	/* Check whether cpus belong to parent ctrl group */
343 	cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask);
344 	if (!cpumask_empty(tmpmask)) {
345 		rdt_last_cmd_puts("Can only add CPUs to mongroup that belong to parent\n");
346 		return -EINVAL;
347 	}
348 
349 	/* Check whether cpus are dropped from this group */
350 	cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
351 	if (!cpumask_empty(tmpmask)) {
352 		/* Give any dropped cpus to parent rdtgroup */
353 		cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask);
354 		update_closid_rmid(tmpmask, prgrp);
355 	}
356 
357 	/*
358 	 * If we added cpus, remove them from previous group that owned them
359 	 * and update per-cpu rmid
360 	 */
361 	cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
362 	if (!cpumask_empty(tmpmask)) {
363 		head = &prgrp->mon.crdtgrp_list;
364 		list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
365 			if (crgrp == rdtgrp)
366 				continue;
367 			cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask,
368 				       tmpmask);
369 		}
370 		update_closid_rmid(tmpmask, rdtgrp);
371 	}
372 
373 	/* Done pushing/pulling - update this group with new mask */
374 	cpumask_copy(&rdtgrp->cpu_mask, newmask);
375 
376 	return 0;
377 }
378 
379 static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m)
380 {
381 	struct rdtgroup *crgrp;
382 
383 	cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m);
384 	/* update the child mon group masks as well*/
385 	list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list)
386 		cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask);
387 }
388 
389 static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
390 			   cpumask_var_t tmpmask, cpumask_var_t tmpmask1)
391 {
392 	struct rdtgroup *r, *crgrp;
393 	struct list_head *head;
394 
395 	/* Check whether cpus are dropped from this group */
396 	cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
397 	if (!cpumask_empty(tmpmask)) {
398 		/* Can't drop from default group */
399 		if (rdtgrp == &rdtgroup_default) {
400 			rdt_last_cmd_puts("Can't drop CPUs from default group\n");
401 			return -EINVAL;
402 		}
403 
404 		/* Give any dropped cpus to rdtgroup_default */
405 		cpumask_or(&rdtgroup_default.cpu_mask,
406 			   &rdtgroup_default.cpu_mask, tmpmask);
407 		update_closid_rmid(tmpmask, &rdtgroup_default);
408 	}
409 
410 	/*
411 	 * If we added cpus, remove them from previous group and
412 	 * the prev group's child groups that owned them
413 	 * and update per-cpu closid/rmid.
414 	 */
415 	cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
416 	if (!cpumask_empty(tmpmask)) {
417 		list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) {
418 			if (r == rdtgrp)
419 				continue;
420 			cpumask_and(tmpmask1, &r->cpu_mask, tmpmask);
421 			if (!cpumask_empty(tmpmask1))
422 				cpumask_rdtgrp_clear(r, tmpmask1);
423 		}
424 		update_closid_rmid(tmpmask, rdtgrp);
425 	}
426 
427 	/* Done pushing/pulling - update this group with new mask */
428 	cpumask_copy(&rdtgrp->cpu_mask, newmask);
429 
430 	/*
431 	 * Clear child mon group masks since there is a new parent mask
432 	 * now and update the rmid for the cpus the child lost.
433 	 */
434 	head = &rdtgrp->mon.crdtgrp_list;
435 	list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
436 		cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask);
437 		update_closid_rmid(tmpmask, rdtgrp);
438 		cpumask_clear(&crgrp->cpu_mask);
439 	}
440 
441 	return 0;
442 }
443 
444 static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of,
445 				   char *buf, size_t nbytes, loff_t off)
446 {
447 	cpumask_var_t tmpmask, newmask, tmpmask1;
448 	struct rdtgroup *rdtgrp;
449 	int ret;
450 
451 	if (!buf)
452 		return -EINVAL;
453 
454 	if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
455 		return -ENOMEM;
456 	if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) {
457 		free_cpumask_var(tmpmask);
458 		return -ENOMEM;
459 	}
460 	if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) {
461 		free_cpumask_var(tmpmask);
462 		free_cpumask_var(newmask);
463 		return -ENOMEM;
464 	}
465 
466 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
467 	if (!rdtgrp) {
468 		ret = -ENOENT;
469 		goto unlock;
470 	}
471 
472 	if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
473 	    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
474 		ret = -EINVAL;
475 		rdt_last_cmd_puts("Pseudo-locking in progress\n");
476 		goto unlock;
477 	}
478 
479 	if (is_cpu_list(of))
480 		ret = cpulist_parse(buf, newmask);
481 	else
482 		ret = cpumask_parse(buf, newmask);
483 
484 	if (ret) {
485 		rdt_last_cmd_puts("Bad CPU list/mask\n");
486 		goto unlock;
487 	}
488 
489 	/* check that user didn't specify any offline cpus */
490 	cpumask_andnot(tmpmask, newmask, cpu_online_mask);
491 	if (!cpumask_empty(tmpmask)) {
492 		ret = -EINVAL;
493 		rdt_last_cmd_puts("Can only assign online CPUs\n");
494 		goto unlock;
495 	}
496 
497 	if (rdtgrp->type == RDTCTRL_GROUP)
498 		ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1);
499 	else if (rdtgrp->type == RDTMON_GROUP)
500 		ret = cpus_mon_write(rdtgrp, newmask, tmpmask);
501 	else
502 		ret = -EINVAL;
503 
504 unlock:
505 	rdtgroup_kn_unlock(of->kn);
506 	free_cpumask_var(tmpmask);
507 	free_cpumask_var(newmask);
508 	free_cpumask_var(tmpmask1);
509 
510 	return ret ?: nbytes;
511 }
512 
513 /**
514  * rdtgroup_remove - the helper to remove resource group safely
515  * @rdtgrp: resource group to remove
516  *
517  * On resource group creation via a mkdir, an extra kernfs_node reference is
518  * taken to ensure that the rdtgroup structure remains accessible for the
519  * rdtgroup_kn_unlock() calls where it is removed.
520  *
521  * Drop the extra reference here, then free the rdtgroup structure.
522  *
523  * Return: void
524  */
525 static void rdtgroup_remove(struct rdtgroup *rdtgrp)
526 {
527 	kernfs_put(rdtgrp->kn);
528 	kfree(rdtgrp);
529 }
530 
531 static void _update_task_closid_rmid(void *task)
532 {
533 	/*
534 	 * If the task is still current on this CPU, update PQR_ASSOC MSR.
535 	 * Otherwise, the MSR is updated when the task is scheduled in.
536 	 */
537 	if (task == current)
538 		resctrl_sched_in();
539 }
540 
541 static void update_task_closid_rmid(struct task_struct *t)
542 {
543 	if (IS_ENABLED(CONFIG_SMP) && task_curr(t))
544 		smp_call_function_single(task_cpu(t), _update_task_closid_rmid, t, 1);
545 	else
546 		_update_task_closid_rmid(t);
547 }
548 
549 static int __rdtgroup_move_task(struct task_struct *tsk,
550 				struct rdtgroup *rdtgrp)
551 {
552 	/* If the task is already in rdtgrp, no need to move the task. */
553 	if ((rdtgrp->type == RDTCTRL_GROUP && tsk->closid == rdtgrp->closid &&
554 	     tsk->rmid == rdtgrp->mon.rmid) ||
555 	    (rdtgrp->type == RDTMON_GROUP && tsk->rmid == rdtgrp->mon.rmid &&
556 	     tsk->closid == rdtgrp->mon.parent->closid))
557 		return 0;
558 
559 	/*
560 	 * Set the task's closid/rmid before the PQR_ASSOC MSR can be
561 	 * updated by them.
562 	 *
563 	 * For ctrl_mon groups, move both closid and rmid.
564 	 * For monitor groups, can move the tasks only from
565 	 * their parent CTRL group.
566 	 */
567 
568 	if (rdtgrp->type == RDTCTRL_GROUP) {
569 		WRITE_ONCE(tsk->closid, rdtgrp->closid);
570 		WRITE_ONCE(tsk->rmid, rdtgrp->mon.rmid);
571 	} else if (rdtgrp->type == RDTMON_GROUP) {
572 		if (rdtgrp->mon.parent->closid == tsk->closid) {
573 			WRITE_ONCE(tsk->rmid, rdtgrp->mon.rmid);
574 		} else {
575 			rdt_last_cmd_puts("Can't move task to different control group\n");
576 			return -EINVAL;
577 		}
578 	}
579 
580 	/*
581 	 * Ensure the task's closid and rmid are written before determining if
582 	 * the task is current that will decide if it will be interrupted.
583 	 */
584 	barrier();
585 
586 	/*
587 	 * By now, the task's closid and rmid are set. If the task is current
588 	 * on a CPU, the PQR_ASSOC MSR needs to be updated to make the resource
589 	 * group go into effect. If the task is not current, the MSR will be
590 	 * updated when the task is scheduled in.
591 	 */
592 	update_task_closid_rmid(tsk);
593 
594 	return 0;
595 }
596 
597 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
598 {
599 	return (rdt_alloc_capable &&
600 	       (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
601 }
602 
603 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
604 {
605 	return (rdt_mon_capable &&
606 	       (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
607 }
608 
609 /**
610  * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group
611  * @r: Resource group
612  *
613  * Return: 1 if tasks have been assigned to @r, 0 otherwise
614  */
615 int rdtgroup_tasks_assigned(struct rdtgroup *r)
616 {
617 	struct task_struct *p, *t;
618 	int ret = 0;
619 
620 	lockdep_assert_held(&rdtgroup_mutex);
621 
622 	rcu_read_lock();
623 	for_each_process_thread(p, t) {
624 		if (is_closid_match(t, r) || is_rmid_match(t, r)) {
625 			ret = 1;
626 			break;
627 		}
628 	}
629 	rcu_read_unlock();
630 
631 	return ret;
632 }
633 
634 static int rdtgroup_task_write_permission(struct task_struct *task,
635 					  struct kernfs_open_file *of)
636 {
637 	const struct cred *tcred = get_task_cred(task);
638 	const struct cred *cred = current_cred();
639 	int ret = 0;
640 
641 	/*
642 	 * Even if we're attaching all tasks in the thread group, we only
643 	 * need to check permissions on one of them.
644 	 */
645 	if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
646 	    !uid_eq(cred->euid, tcred->uid) &&
647 	    !uid_eq(cred->euid, tcred->suid)) {
648 		rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
649 		ret = -EPERM;
650 	}
651 
652 	put_cred(tcred);
653 	return ret;
654 }
655 
656 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
657 			      struct kernfs_open_file *of)
658 {
659 	struct task_struct *tsk;
660 	int ret;
661 
662 	rcu_read_lock();
663 	if (pid) {
664 		tsk = find_task_by_vpid(pid);
665 		if (!tsk) {
666 			rcu_read_unlock();
667 			rdt_last_cmd_printf("No task %d\n", pid);
668 			return -ESRCH;
669 		}
670 	} else {
671 		tsk = current;
672 	}
673 
674 	get_task_struct(tsk);
675 	rcu_read_unlock();
676 
677 	ret = rdtgroup_task_write_permission(tsk, of);
678 	if (!ret)
679 		ret = __rdtgroup_move_task(tsk, rdtgrp);
680 
681 	put_task_struct(tsk);
682 	return ret;
683 }
684 
685 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
686 				    char *buf, size_t nbytes, loff_t off)
687 {
688 	struct rdtgroup *rdtgrp;
689 	int ret = 0;
690 	pid_t pid;
691 
692 	if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
693 		return -EINVAL;
694 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
695 	if (!rdtgrp) {
696 		rdtgroup_kn_unlock(of->kn);
697 		return -ENOENT;
698 	}
699 	rdt_last_cmd_clear();
700 
701 	if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
702 	    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
703 		ret = -EINVAL;
704 		rdt_last_cmd_puts("Pseudo-locking in progress\n");
705 		goto unlock;
706 	}
707 
708 	ret = rdtgroup_move_task(pid, rdtgrp, of);
709 
710 unlock:
711 	rdtgroup_kn_unlock(of->kn);
712 
713 	return ret ?: nbytes;
714 }
715 
716 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
717 {
718 	struct task_struct *p, *t;
719 
720 	rcu_read_lock();
721 	for_each_process_thread(p, t) {
722 		if (is_closid_match(t, r) || is_rmid_match(t, r))
723 			seq_printf(s, "%d\n", t->pid);
724 	}
725 	rcu_read_unlock();
726 }
727 
728 static int rdtgroup_tasks_show(struct kernfs_open_file *of,
729 			       struct seq_file *s, void *v)
730 {
731 	struct rdtgroup *rdtgrp;
732 	int ret = 0;
733 
734 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
735 	if (rdtgrp)
736 		show_rdt_tasks(rdtgrp, s);
737 	else
738 		ret = -ENOENT;
739 	rdtgroup_kn_unlock(of->kn);
740 
741 	return ret;
742 }
743 
744 #ifdef CONFIG_PROC_CPU_RESCTRL
745 
746 /*
747  * A task can only be part of one resctrl control group and of one monitor
748  * group which is associated to that control group.
749  *
750  * 1)   res:
751  *      mon:
752  *
753  *    resctrl is not available.
754  *
755  * 2)   res:/
756  *      mon:
757  *
758  *    Task is part of the root resctrl control group, and it is not associated
759  *    to any monitor group.
760  *
761  * 3)  res:/
762  *     mon:mon0
763  *
764  *    Task is part of the root resctrl control group and monitor group mon0.
765  *
766  * 4)  res:group0
767  *     mon:
768  *
769  *    Task is part of resctrl control group group0, and it is not associated
770  *    to any monitor group.
771  *
772  * 5) res:group0
773  *    mon:mon1
774  *
775  *    Task is part of resctrl control group group0 and monitor group mon1.
776  */
777 int proc_resctrl_show(struct seq_file *s, struct pid_namespace *ns,
778 		      struct pid *pid, struct task_struct *tsk)
779 {
780 	struct rdtgroup *rdtg;
781 	int ret = 0;
782 
783 	mutex_lock(&rdtgroup_mutex);
784 
785 	/* Return empty if resctrl has not been mounted. */
786 	if (!static_branch_unlikely(&rdt_enable_key)) {
787 		seq_puts(s, "res:\nmon:\n");
788 		goto unlock;
789 	}
790 
791 	list_for_each_entry(rdtg, &rdt_all_groups, rdtgroup_list) {
792 		struct rdtgroup *crg;
793 
794 		/*
795 		 * Task information is only relevant for shareable
796 		 * and exclusive groups.
797 		 */
798 		if (rdtg->mode != RDT_MODE_SHAREABLE &&
799 		    rdtg->mode != RDT_MODE_EXCLUSIVE)
800 			continue;
801 
802 		if (rdtg->closid != tsk->closid)
803 			continue;
804 
805 		seq_printf(s, "res:%s%s\n", (rdtg == &rdtgroup_default) ? "/" : "",
806 			   rdtg->kn->name);
807 		seq_puts(s, "mon:");
808 		list_for_each_entry(crg, &rdtg->mon.crdtgrp_list,
809 				    mon.crdtgrp_list) {
810 			if (tsk->rmid != crg->mon.rmid)
811 				continue;
812 			seq_printf(s, "%s", crg->kn->name);
813 			break;
814 		}
815 		seq_putc(s, '\n');
816 		goto unlock;
817 	}
818 	/*
819 	 * The above search should succeed. Otherwise return
820 	 * with an error.
821 	 */
822 	ret = -ENOENT;
823 unlock:
824 	mutex_unlock(&rdtgroup_mutex);
825 
826 	return ret;
827 }
828 #endif
829 
830 static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
831 				    struct seq_file *seq, void *v)
832 {
833 	int len;
834 
835 	mutex_lock(&rdtgroup_mutex);
836 	len = seq_buf_used(&last_cmd_status);
837 	if (len)
838 		seq_printf(seq, "%.*s", len, last_cmd_status_buf);
839 	else
840 		seq_puts(seq, "ok\n");
841 	mutex_unlock(&rdtgroup_mutex);
842 	return 0;
843 }
844 
845 static int rdt_num_closids_show(struct kernfs_open_file *of,
846 				struct seq_file *seq, void *v)
847 {
848 	struct resctrl_schema *s = of->kn->parent->priv;
849 
850 	seq_printf(seq, "%u\n", s->num_closid);
851 	return 0;
852 }
853 
854 static int rdt_default_ctrl_show(struct kernfs_open_file *of,
855 			     struct seq_file *seq, void *v)
856 {
857 	struct resctrl_schema *s = of->kn->parent->priv;
858 	struct rdt_resource *r = s->res;
859 
860 	seq_printf(seq, "%x\n", r->default_ctrl);
861 	return 0;
862 }
863 
864 static int rdt_min_cbm_bits_show(struct kernfs_open_file *of,
865 			     struct seq_file *seq, void *v)
866 {
867 	struct resctrl_schema *s = of->kn->parent->priv;
868 	struct rdt_resource *r = s->res;
869 
870 	seq_printf(seq, "%u\n", r->cache.min_cbm_bits);
871 	return 0;
872 }
873 
874 static int rdt_shareable_bits_show(struct kernfs_open_file *of,
875 				   struct seq_file *seq, void *v)
876 {
877 	struct resctrl_schema *s = of->kn->parent->priv;
878 	struct rdt_resource *r = s->res;
879 
880 	seq_printf(seq, "%x\n", r->cache.shareable_bits);
881 	return 0;
882 }
883 
884 /**
885  * rdt_bit_usage_show - Display current usage of resources
886  *
887  * A domain is a shared resource that can now be allocated differently. Here
888  * we display the current regions of the domain as an annotated bitmask.
889  * For each domain of this resource its allocation bitmask
890  * is annotated as below to indicate the current usage of the corresponding bit:
891  *   0 - currently unused
892  *   X - currently available for sharing and used by software and hardware
893  *   H - currently used by hardware only but available for software use
894  *   S - currently used and shareable by software only
895  *   E - currently used exclusively by one resource group
896  *   P - currently pseudo-locked by one resource group
897  */
898 static int rdt_bit_usage_show(struct kernfs_open_file *of,
899 			      struct seq_file *seq, void *v)
900 {
901 	struct resctrl_schema *s = of->kn->parent->priv;
902 	/*
903 	 * Use unsigned long even though only 32 bits are used to ensure
904 	 * test_bit() is used safely.
905 	 */
906 	unsigned long sw_shareable = 0, hw_shareable = 0;
907 	unsigned long exclusive = 0, pseudo_locked = 0;
908 	struct rdt_resource *r = s->res;
909 	struct rdt_domain *dom;
910 	int i, hwb, swb, excl, psl;
911 	enum rdtgrp_mode mode;
912 	bool sep = false;
913 	u32 ctrl_val;
914 
915 	mutex_lock(&rdtgroup_mutex);
916 	hw_shareable = r->cache.shareable_bits;
917 	list_for_each_entry(dom, &r->domains, list) {
918 		if (sep)
919 			seq_putc(seq, ';');
920 		sw_shareable = 0;
921 		exclusive = 0;
922 		seq_printf(seq, "%d=", dom->id);
923 		for (i = 0; i < closids_supported(); i++) {
924 			if (!closid_allocated(i))
925 				continue;
926 			ctrl_val = resctrl_arch_get_config(r, dom, i,
927 							   s->conf_type);
928 			mode = rdtgroup_mode_by_closid(i);
929 			switch (mode) {
930 			case RDT_MODE_SHAREABLE:
931 				sw_shareable |= ctrl_val;
932 				break;
933 			case RDT_MODE_EXCLUSIVE:
934 				exclusive |= ctrl_val;
935 				break;
936 			case RDT_MODE_PSEUDO_LOCKSETUP:
937 			/*
938 			 * RDT_MODE_PSEUDO_LOCKSETUP is possible
939 			 * here but not included since the CBM
940 			 * associated with this CLOSID in this mode
941 			 * is not initialized and no task or cpu can be
942 			 * assigned this CLOSID.
943 			 */
944 				break;
945 			case RDT_MODE_PSEUDO_LOCKED:
946 			case RDT_NUM_MODES:
947 				WARN(1,
948 				     "invalid mode for closid %d\n", i);
949 				break;
950 			}
951 		}
952 		for (i = r->cache.cbm_len - 1; i >= 0; i--) {
953 			pseudo_locked = dom->plr ? dom->plr->cbm : 0;
954 			hwb = test_bit(i, &hw_shareable);
955 			swb = test_bit(i, &sw_shareable);
956 			excl = test_bit(i, &exclusive);
957 			psl = test_bit(i, &pseudo_locked);
958 			if (hwb && swb)
959 				seq_putc(seq, 'X');
960 			else if (hwb && !swb)
961 				seq_putc(seq, 'H');
962 			else if (!hwb && swb)
963 				seq_putc(seq, 'S');
964 			else if (excl)
965 				seq_putc(seq, 'E');
966 			else if (psl)
967 				seq_putc(seq, 'P');
968 			else /* Unused bits remain */
969 				seq_putc(seq, '0');
970 		}
971 		sep = true;
972 	}
973 	seq_putc(seq, '\n');
974 	mutex_unlock(&rdtgroup_mutex);
975 	return 0;
976 }
977 
978 static int rdt_min_bw_show(struct kernfs_open_file *of,
979 			     struct seq_file *seq, void *v)
980 {
981 	struct resctrl_schema *s = of->kn->parent->priv;
982 	struct rdt_resource *r = s->res;
983 
984 	seq_printf(seq, "%u\n", r->membw.min_bw);
985 	return 0;
986 }
987 
988 static int rdt_num_rmids_show(struct kernfs_open_file *of,
989 			      struct seq_file *seq, void *v)
990 {
991 	struct rdt_resource *r = of->kn->parent->priv;
992 
993 	seq_printf(seq, "%d\n", r->num_rmid);
994 
995 	return 0;
996 }
997 
998 static int rdt_mon_features_show(struct kernfs_open_file *of,
999 				 struct seq_file *seq, void *v)
1000 {
1001 	struct rdt_resource *r = of->kn->parent->priv;
1002 	struct mon_evt *mevt;
1003 
1004 	list_for_each_entry(mevt, &r->evt_list, list)
1005 		seq_printf(seq, "%s\n", mevt->name);
1006 
1007 	return 0;
1008 }
1009 
1010 static int rdt_bw_gran_show(struct kernfs_open_file *of,
1011 			     struct seq_file *seq, void *v)
1012 {
1013 	struct resctrl_schema *s = of->kn->parent->priv;
1014 	struct rdt_resource *r = s->res;
1015 
1016 	seq_printf(seq, "%u\n", r->membw.bw_gran);
1017 	return 0;
1018 }
1019 
1020 static int rdt_delay_linear_show(struct kernfs_open_file *of,
1021 			     struct seq_file *seq, void *v)
1022 {
1023 	struct resctrl_schema *s = of->kn->parent->priv;
1024 	struct rdt_resource *r = s->res;
1025 
1026 	seq_printf(seq, "%u\n", r->membw.delay_linear);
1027 	return 0;
1028 }
1029 
1030 static int max_threshold_occ_show(struct kernfs_open_file *of,
1031 				  struct seq_file *seq, void *v)
1032 {
1033 	struct rdt_resource *r = of->kn->parent->priv;
1034 	struct rdt_hw_resource *hw_res = resctrl_to_arch_res(r);
1035 
1036 	seq_printf(seq, "%u\n", resctrl_cqm_threshold * hw_res->mon_scale);
1037 
1038 	return 0;
1039 }
1040 
1041 static int rdt_thread_throttle_mode_show(struct kernfs_open_file *of,
1042 					 struct seq_file *seq, void *v)
1043 {
1044 	struct resctrl_schema *s = of->kn->parent->priv;
1045 	struct rdt_resource *r = s->res;
1046 
1047 	if (r->membw.throttle_mode == THREAD_THROTTLE_PER_THREAD)
1048 		seq_puts(seq, "per-thread\n");
1049 	else
1050 		seq_puts(seq, "max\n");
1051 
1052 	return 0;
1053 }
1054 
1055 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
1056 				       char *buf, size_t nbytes, loff_t off)
1057 {
1058 	struct rdt_hw_resource *hw_res;
1059 	unsigned int bytes;
1060 	int ret;
1061 
1062 	ret = kstrtouint(buf, 0, &bytes);
1063 	if (ret)
1064 		return ret;
1065 
1066 	if (bytes > (boot_cpu_data.x86_cache_size * 1024))
1067 		return -EINVAL;
1068 
1069 	hw_res = resctrl_to_arch_res(of->kn->parent->priv);
1070 	resctrl_cqm_threshold = bytes / hw_res->mon_scale;
1071 
1072 	return nbytes;
1073 }
1074 
1075 /*
1076  * rdtgroup_mode_show - Display mode of this resource group
1077  */
1078 static int rdtgroup_mode_show(struct kernfs_open_file *of,
1079 			      struct seq_file *s, void *v)
1080 {
1081 	struct rdtgroup *rdtgrp;
1082 
1083 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
1084 	if (!rdtgrp) {
1085 		rdtgroup_kn_unlock(of->kn);
1086 		return -ENOENT;
1087 	}
1088 
1089 	seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode));
1090 
1091 	rdtgroup_kn_unlock(of->kn);
1092 	return 0;
1093 }
1094 
1095 static enum resctrl_conf_type resctrl_peer_type(enum resctrl_conf_type my_type)
1096 {
1097 	switch (my_type) {
1098 	case CDP_CODE:
1099 		return CDP_DATA;
1100 	case CDP_DATA:
1101 		return CDP_CODE;
1102 	default:
1103 	case CDP_NONE:
1104 		return CDP_NONE;
1105 	}
1106 }
1107 
1108 /**
1109  * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other
1110  * @r: Resource to which domain instance @d belongs.
1111  * @d: The domain instance for which @closid is being tested.
1112  * @cbm: Capacity bitmask being tested.
1113  * @closid: Intended closid for @cbm.
1114  * @exclusive: Only check if overlaps with exclusive resource groups
1115  *
1116  * Checks if provided @cbm intended to be used for @closid on domain
1117  * @d overlaps with any other closids or other hardware usage associated
1118  * with this domain. If @exclusive is true then only overlaps with
1119  * resource groups in exclusive mode will be considered. If @exclusive
1120  * is false then overlaps with any resource group or hardware entities
1121  * will be considered.
1122  *
1123  * @cbm is unsigned long, even if only 32 bits are used, to make the
1124  * bitmap functions work correctly.
1125  *
1126  * Return: false if CBM does not overlap, true if it does.
1127  */
1128 static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1129 				    unsigned long cbm, int closid,
1130 				    enum resctrl_conf_type type, bool exclusive)
1131 {
1132 	enum rdtgrp_mode mode;
1133 	unsigned long ctrl_b;
1134 	int i;
1135 
1136 	/* Check for any overlap with regions used by hardware directly */
1137 	if (!exclusive) {
1138 		ctrl_b = r->cache.shareable_bits;
1139 		if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len))
1140 			return true;
1141 	}
1142 
1143 	/* Check for overlap with other resource groups */
1144 	for (i = 0; i < closids_supported(); i++) {
1145 		ctrl_b = resctrl_arch_get_config(r, d, i, type);
1146 		mode = rdtgroup_mode_by_closid(i);
1147 		if (closid_allocated(i) && i != closid &&
1148 		    mode != RDT_MODE_PSEUDO_LOCKSETUP) {
1149 			if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) {
1150 				if (exclusive) {
1151 					if (mode == RDT_MODE_EXCLUSIVE)
1152 						return true;
1153 					continue;
1154 				}
1155 				return true;
1156 			}
1157 		}
1158 	}
1159 
1160 	return false;
1161 }
1162 
1163 /**
1164  * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware
1165  * @s: Schema for the resource to which domain instance @d belongs.
1166  * @d: The domain instance for which @closid is being tested.
1167  * @cbm: Capacity bitmask being tested.
1168  * @closid: Intended closid for @cbm.
1169  * @exclusive: Only check if overlaps with exclusive resource groups
1170  *
1171  * Resources that can be allocated using a CBM can use the CBM to control
1172  * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test
1173  * for overlap. Overlap test is not limited to the specific resource for
1174  * which the CBM is intended though - when dealing with CDP resources that
1175  * share the underlying hardware the overlap check should be performed on
1176  * the CDP resource sharing the hardware also.
1177  *
1178  * Refer to description of __rdtgroup_cbm_overlaps() for the details of the
1179  * overlap test.
1180  *
1181  * Return: true if CBM overlap detected, false if there is no overlap
1182  */
1183 bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_domain *d,
1184 			   unsigned long cbm, int closid, bool exclusive)
1185 {
1186 	enum resctrl_conf_type peer_type = resctrl_peer_type(s->conf_type);
1187 	struct rdt_resource *r = s->res;
1188 
1189 	if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, s->conf_type,
1190 				    exclusive))
1191 		return true;
1192 
1193 	if (!resctrl_arch_get_cdp_enabled(r->rid))
1194 		return false;
1195 	return  __rdtgroup_cbm_overlaps(r, d, cbm, closid, peer_type, exclusive);
1196 }
1197 
1198 /**
1199  * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive
1200  *
1201  * An exclusive resource group implies that there should be no sharing of
1202  * its allocated resources. At the time this group is considered to be
1203  * exclusive this test can determine if its current schemata supports this
1204  * setting by testing for overlap with all other resource groups.
1205  *
1206  * Return: true if resource group can be exclusive, false if there is overlap
1207  * with allocations of other resource groups and thus this resource group
1208  * cannot be exclusive.
1209  */
1210 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp)
1211 {
1212 	int closid = rdtgrp->closid;
1213 	struct resctrl_schema *s;
1214 	struct rdt_resource *r;
1215 	bool has_cache = false;
1216 	struct rdt_domain *d;
1217 	u32 ctrl;
1218 
1219 	list_for_each_entry(s, &resctrl_schema_all, list) {
1220 		r = s->res;
1221 		if (r->rid == RDT_RESOURCE_MBA)
1222 			continue;
1223 		has_cache = true;
1224 		list_for_each_entry(d, &r->domains, list) {
1225 			ctrl = resctrl_arch_get_config(r, d, closid,
1226 						       s->conf_type);
1227 			if (rdtgroup_cbm_overlaps(s, d, ctrl, closid, false)) {
1228 				rdt_last_cmd_puts("Schemata overlaps\n");
1229 				return false;
1230 			}
1231 		}
1232 	}
1233 
1234 	if (!has_cache) {
1235 		rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n");
1236 		return false;
1237 	}
1238 
1239 	return true;
1240 }
1241 
1242 /**
1243  * rdtgroup_mode_write - Modify the resource group's mode
1244  *
1245  */
1246 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of,
1247 				   char *buf, size_t nbytes, loff_t off)
1248 {
1249 	struct rdtgroup *rdtgrp;
1250 	enum rdtgrp_mode mode;
1251 	int ret = 0;
1252 
1253 	/* Valid input requires a trailing newline */
1254 	if (nbytes == 0 || buf[nbytes - 1] != '\n')
1255 		return -EINVAL;
1256 	buf[nbytes - 1] = '\0';
1257 
1258 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
1259 	if (!rdtgrp) {
1260 		rdtgroup_kn_unlock(of->kn);
1261 		return -ENOENT;
1262 	}
1263 
1264 	rdt_last_cmd_clear();
1265 
1266 	mode = rdtgrp->mode;
1267 
1268 	if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) ||
1269 	    (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) ||
1270 	    (!strcmp(buf, "pseudo-locksetup") &&
1271 	     mode == RDT_MODE_PSEUDO_LOCKSETUP) ||
1272 	    (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED))
1273 		goto out;
1274 
1275 	if (mode == RDT_MODE_PSEUDO_LOCKED) {
1276 		rdt_last_cmd_puts("Cannot change pseudo-locked group\n");
1277 		ret = -EINVAL;
1278 		goto out;
1279 	}
1280 
1281 	if (!strcmp(buf, "shareable")) {
1282 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1283 			ret = rdtgroup_locksetup_exit(rdtgrp);
1284 			if (ret)
1285 				goto out;
1286 		}
1287 		rdtgrp->mode = RDT_MODE_SHAREABLE;
1288 	} else if (!strcmp(buf, "exclusive")) {
1289 		if (!rdtgroup_mode_test_exclusive(rdtgrp)) {
1290 			ret = -EINVAL;
1291 			goto out;
1292 		}
1293 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1294 			ret = rdtgroup_locksetup_exit(rdtgrp);
1295 			if (ret)
1296 				goto out;
1297 		}
1298 		rdtgrp->mode = RDT_MODE_EXCLUSIVE;
1299 	} else if (!strcmp(buf, "pseudo-locksetup")) {
1300 		ret = rdtgroup_locksetup_enter(rdtgrp);
1301 		if (ret)
1302 			goto out;
1303 		rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP;
1304 	} else {
1305 		rdt_last_cmd_puts("Unknown or unsupported mode\n");
1306 		ret = -EINVAL;
1307 	}
1308 
1309 out:
1310 	rdtgroup_kn_unlock(of->kn);
1311 	return ret ?: nbytes;
1312 }
1313 
1314 /**
1315  * rdtgroup_cbm_to_size - Translate CBM to size in bytes
1316  * @r: RDT resource to which @d belongs.
1317  * @d: RDT domain instance.
1318  * @cbm: bitmask for which the size should be computed.
1319  *
1320  * The bitmask provided associated with the RDT domain instance @d will be
1321  * translated into how many bytes it represents. The size in bytes is
1322  * computed by first dividing the total cache size by the CBM length to
1323  * determine how many bytes each bit in the bitmask represents. The result
1324  * is multiplied with the number of bits set in the bitmask.
1325  *
1326  * @cbm is unsigned long, even if only 32 bits are used to make the
1327  * bitmap functions work correctly.
1328  */
1329 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r,
1330 				  struct rdt_domain *d, unsigned long cbm)
1331 {
1332 	struct cpu_cacheinfo *ci;
1333 	unsigned int size = 0;
1334 	int num_b, i;
1335 
1336 	num_b = bitmap_weight(&cbm, r->cache.cbm_len);
1337 	ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask));
1338 	for (i = 0; i < ci->num_leaves; i++) {
1339 		if (ci->info_list[i].level == r->cache_level) {
1340 			size = ci->info_list[i].size / r->cache.cbm_len * num_b;
1341 			break;
1342 		}
1343 	}
1344 
1345 	return size;
1346 }
1347 
1348 /**
1349  * rdtgroup_size_show - Display size in bytes of allocated regions
1350  *
1351  * The "size" file mirrors the layout of the "schemata" file, printing the
1352  * size in bytes of each region instead of the capacity bitmask.
1353  *
1354  */
1355 static int rdtgroup_size_show(struct kernfs_open_file *of,
1356 			      struct seq_file *s, void *v)
1357 {
1358 	struct resctrl_schema *schema;
1359 	struct rdtgroup *rdtgrp;
1360 	struct rdt_resource *r;
1361 	struct rdt_domain *d;
1362 	unsigned int size;
1363 	int ret = 0;
1364 	bool sep;
1365 	u32 ctrl;
1366 
1367 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
1368 	if (!rdtgrp) {
1369 		rdtgroup_kn_unlock(of->kn);
1370 		return -ENOENT;
1371 	}
1372 
1373 	if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
1374 		if (!rdtgrp->plr->d) {
1375 			rdt_last_cmd_clear();
1376 			rdt_last_cmd_puts("Cache domain offline\n");
1377 			ret = -ENODEV;
1378 		} else {
1379 			seq_printf(s, "%*s:", max_name_width,
1380 				   rdtgrp->plr->s->name);
1381 			size = rdtgroup_cbm_to_size(rdtgrp->plr->s->res,
1382 						    rdtgrp->plr->d,
1383 						    rdtgrp->plr->cbm);
1384 			seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size);
1385 		}
1386 		goto out;
1387 	}
1388 
1389 	list_for_each_entry(schema, &resctrl_schema_all, list) {
1390 		r = schema->res;
1391 		sep = false;
1392 		seq_printf(s, "%*s:", max_name_width, schema->name);
1393 		list_for_each_entry(d, &r->domains, list) {
1394 			if (sep)
1395 				seq_putc(s, ';');
1396 			if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1397 				size = 0;
1398 			} else {
1399 				ctrl = resctrl_arch_get_config(r, d,
1400 							       rdtgrp->closid,
1401 							       schema->conf_type);
1402 				if (r->rid == RDT_RESOURCE_MBA)
1403 					size = ctrl;
1404 				else
1405 					size = rdtgroup_cbm_to_size(r, d, ctrl);
1406 			}
1407 			seq_printf(s, "%d=%u", d->id, size);
1408 			sep = true;
1409 		}
1410 		seq_putc(s, '\n');
1411 	}
1412 
1413 out:
1414 	rdtgroup_kn_unlock(of->kn);
1415 
1416 	return ret;
1417 }
1418 
1419 /* rdtgroup information files for one cache resource. */
1420 static struct rftype res_common_files[] = {
1421 	{
1422 		.name		= "last_cmd_status",
1423 		.mode		= 0444,
1424 		.kf_ops		= &rdtgroup_kf_single_ops,
1425 		.seq_show	= rdt_last_cmd_status_show,
1426 		.fflags		= RF_TOP_INFO,
1427 	},
1428 	{
1429 		.name		= "num_closids",
1430 		.mode		= 0444,
1431 		.kf_ops		= &rdtgroup_kf_single_ops,
1432 		.seq_show	= rdt_num_closids_show,
1433 		.fflags		= RF_CTRL_INFO,
1434 	},
1435 	{
1436 		.name		= "mon_features",
1437 		.mode		= 0444,
1438 		.kf_ops		= &rdtgroup_kf_single_ops,
1439 		.seq_show	= rdt_mon_features_show,
1440 		.fflags		= RF_MON_INFO,
1441 	},
1442 	{
1443 		.name		= "num_rmids",
1444 		.mode		= 0444,
1445 		.kf_ops		= &rdtgroup_kf_single_ops,
1446 		.seq_show	= rdt_num_rmids_show,
1447 		.fflags		= RF_MON_INFO,
1448 	},
1449 	{
1450 		.name		= "cbm_mask",
1451 		.mode		= 0444,
1452 		.kf_ops		= &rdtgroup_kf_single_ops,
1453 		.seq_show	= rdt_default_ctrl_show,
1454 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_CACHE,
1455 	},
1456 	{
1457 		.name		= "min_cbm_bits",
1458 		.mode		= 0444,
1459 		.kf_ops		= &rdtgroup_kf_single_ops,
1460 		.seq_show	= rdt_min_cbm_bits_show,
1461 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_CACHE,
1462 	},
1463 	{
1464 		.name		= "shareable_bits",
1465 		.mode		= 0444,
1466 		.kf_ops		= &rdtgroup_kf_single_ops,
1467 		.seq_show	= rdt_shareable_bits_show,
1468 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_CACHE,
1469 	},
1470 	{
1471 		.name		= "bit_usage",
1472 		.mode		= 0444,
1473 		.kf_ops		= &rdtgroup_kf_single_ops,
1474 		.seq_show	= rdt_bit_usage_show,
1475 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_CACHE,
1476 	},
1477 	{
1478 		.name		= "min_bandwidth",
1479 		.mode		= 0444,
1480 		.kf_ops		= &rdtgroup_kf_single_ops,
1481 		.seq_show	= rdt_min_bw_show,
1482 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_MB,
1483 	},
1484 	{
1485 		.name		= "bandwidth_gran",
1486 		.mode		= 0444,
1487 		.kf_ops		= &rdtgroup_kf_single_ops,
1488 		.seq_show	= rdt_bw_gran_show,
1489 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_MB,
1490 	},
1491 	{
1492 		.name		= "delay_linear",
1493 		.mode		= 0444,
1494 		.kf_ops		= &rdtgroup_kf_single_ops,
1495 		.seq_show	= rdt_delay_linear_show,
1496 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_MB,
1497 	},
1498 	/*
1499 	 * Platform specific which (if any) capabilities are provided by
1500 	 * thread_throttle_mode. Defer "fflags" initialization to platform
1501 	 * discovery.
1502 	 */
1503 	{
1504 		.name		= "thread_throttle_mode",
1505 		.mode		= 0444,
1506 		.kf_ops		= &rdtgroup_kf_single_ops,
1507 		.seq_show	= rdt_thread_throttle_mode_show,
1508 	},
1509 	{
1510 		.name		= "max_threshold_occupancy",
1511 		.mode		= 0644,
1512 		.kf_ops		= &rdtgroup_kf_single_ops,
1513 		.write		= max_threshold_occ_write,
1514 		.seq_show	= max_threshold_occ_show,
1515 		.fflags		= RF_MON_INFO | RFTYPE_RES_CACHE,
1516 	},
1517 	{
1518 		.name		= "cpus",
1519 		.mode		= 0644,
1520 		.kf_ops		= &rdtgroup_kf_single_ops,
1521 		.write		= rdtgroup_cpus_write,
1522 		.seq_show	= rdtgroup_cpus_show,
1523 		.fflags		= RFTYPE_BASE,
1524 	},
1525 	{
1526 		.name		= "cpus_list",
1527 		.mode		= 0644,
1528 		.kf_ops		= &rdtgroup_kf_single_ops,
1529 		.write		= rdtgroup_cpus_write,
1530 		.seq_show	= rdtgroup_cpus_show,
1531 		.flags		= RFTYPE_FLAGS_CPUS_LIST,
1532 		.fflags		= RFTYPE_BASE,
1533 	},
1534 	{
1535 		.name		= "tasks",
1536 		.mode		= 0644,
1537 		.kf_ops		= &rdtgroup_kf_single_ops,
1538 		.write		= rdtgroup_tasks_write,
1539 		.seq_show	= rdtgroup_tasks_show,
1540 		.fflags		= RFTYPE_BASE,
1541 	},
1542 	{
1543 		.name		= "schemata",
1544 		.mode		= 0644,
1545 		.kf_ops		= &rdtgroup_kf_single_ops,
1546 		.write		= rdtgroup_schemata_write,
1547 		.seq_show	= rdtgroup_schemata_show,
1548 		.fflags		= RF_CTRL_BASE,
1549 	},
1550 	{
1551 		.name		= "mode",
1552 		.mode		= 0644,
1553 		.kf_ops		= &rdtgroup_kf_single_ops,
1554 		.write		= rdtgroup_mode_write,
1555 		.seq_show	= rdtgroup_mode_show,
1556 		.fflags		= RF_CTRL_BASE,
1557 	},
1558 	{
1559 		.name		= "size",
1560 		.mode		= 0444,
1561 		.kf_ops		= &rdtgroup_kf_single_ops,
1562 		.seq_show	= rdtgroup_size_show,
1563 		.fflags		= RF_CTRL_BASE,
1564 	},
1565 
1566 };
1567 
1568 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
1569 {
1570 	struct rftype *rfts, *rft;
1571 	int ret, len;
1572 
1573 	rfts = res_common_files;
1574 	len = ARRAY_SIZE(res_common_files);
1575 
1576 	lockdep_assert_held(&rdtgroup_mutex);
1577 
1578 	for (rft = rfts; rft < rfts + len; rft++) {
1579 		if (rft->fflags && ((fflags & rft->fflags) == rft->fflags)) {
1580 			ret = rdtgroup_add_file(kn, rft);
1581 			if (ret)
1582 				goto error;
1583 		}
1584 	}
1585 
1586 	return 0;
1587 error:
1588 	pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
1589 	while (--rft >= rfts) {
1590 		if ((fflags & rft->fflags) == rft->fflags)
1591 			kernfs_remove_by_name(kn, rft->name);
1592 	}
1593 	return ret;
1594 }
1595 
1596 static struct rftype *rdtgroup_get_rftype_by_name(const char *name)
1597 {
1598 	struct rftype *rfts, *rft;
1599 	int len;
1600 
1601 	rfts = res_common_files;
1602 	len = ARRAY_SIZE(res_common_files);
1603 
1604 	for (rft = rfts; rft < rfts + len; rft++) {
1605 		if (!strcmp(rft->name, name))
1606 			return rft;
1607 	}
1608 
1609 	return NULL;
1610 }
1611 
1612 void __init thread_throttle_mode_init(void)
1613 {
1614 	struct rftype *rft;
1615 
1616 	rft = rdtgroup_get_rftype_by_name("thread_throttle_mode");
1617 	if (!rft)
1618 		return;
1619 
1620 	rft->fflags = RF_CTRL_INFO | RFTYPE_RES_MB;
1621 }
1622 
1623 /**
1624  * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file
1625  * @r: The resource group with which the file is associated.
1626  * @name: Name of the file
1627  *
1628  * The permissions of named resctrl file, directory, or link are modified
1629  * to not allow read, write, or execute by any user.
1630  *
1631  * WARNING: This function is intended to communicate to the user that the
1632  * resctrl file has been locked down - that it is not relevant to the
1633  * particular state the system finds itself in. It should not be relied
1634  * on to protect from user access because after the file's permissions
1635  * are restricted the user can still change the permissions using chmod
1636  * from the command line.
1637  *
1638  * Return: 0 on success, <0 on failure.
1639  */
1640 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name)
1641 {
1642 	struct iattr iattr = {.ia_valid = ATTR_MODE,};
1643 	struct kernfs_node *kn;
1644 	int ret = 0;
1645 
1646 	kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1647 	if (!kn)
1648 		return -ENOENT;
1649 
1650 	switch (kernfs_type(kn)) {
1651 	case KERNFS_DIR:
1652 		iattr.ia_mode = S_IFDIR;
1653 		break;
1654 	case KERNFS_FILE:
1655 		iattr.ia_mode = S_IFREG;
1656 		break;
1657 	case KERNFS_LINK:
1658 		iattr.ia_mode = S_IFLNK;
1659 		break;
1660 	}
1661 
1662 	ret = kernfs_setattr(kn, &iattr);
1663 	kernfs_put(kn);
1664 	return ret;
1665 }
1666 
1667 /**
1668  * rdtgroup_kn_mode_restore - Restore user access to named resctrl file
1669  * @r: The resource group with which the file is associated.
1670  * @name: Name of the file
1671  * @mask: Mask of permissions that should be restored
1672  *
1673  * Restore the permissions of the named file. If @name is a directory the
1674  * permissions of its parent will be used.
1675  *
1676  * Return: 0 on success, <0 on failure.
1677  */
1678 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
1679 			     umode_t mask)
1680 {
1681 	struct iattr iattr = {.ia_valid = ATTR_MODE,};
1682 	struct kernfs_node *kn, *parent;
1683 	struct rftype *rfts, *rft;
1684 	int ret, len;
1685 
1686 	rfts = res_common_files;
1687 	len = ARRAY_SIZE(res_common_files);
1688 
1689 	for (rft = rfts; rft < rfts + len; rft++) {
1690 		if (!strcmp(rft->name, name))
1691 			iattr.ia_mode = rft->mode & mask;
1692 	}
1693 
1694 	kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1695 	if (!kn)
1696 		return -ENOENT;
1697 
1698 	switch (kernfs_type(kn)) {
1699 	case KERNFS_DIR:
1700 		parent = kernfs_get_parent(kn);
1701 		if (parent) {
1702 			iattr.ia_mode |= parent->mode;
1703 			kernfs_put(parent);
1704 		}
1705 		iattr.ia_mode |= S_IFDIR;
1706 		break;
1707 	case KERNFS_FILE:
1708 		iattr.ia_mode |= S_IFREG;
1709 		break;
1710 	case KERNFS_LINK:
1711 		iattr.ia_mode |= S_IFLNK;
1712 		break;
1713 	}
1714 
1715 	ret = kernfs_setattr(kn, &iattr);
1716 	kernfs_put(kn);
1717 	return ret;
1718 }
1719 
1720 static int rdtgroup_mkdir_info_resdir(void *priv, char *name,
1721 				      unsigned long fflags)
1722 {
1723 	struct kernfs_node *kn_subdir;
1724 	int ret;
1725 
1726 	kn_subdir = kernfs_create_dir(kn_info, name,
1727 				      kn_info->mode, priv);
1728 	if (IS_ERR(kn_subdir))
1729 		return PTR_ERR(kn_subdir);
1730 
1731 	ret = rdtgroup_kn_set_ugid(kn_subdir);
1732 	if (ret)
1733 		return ret;
1734 
1735 	ret = rdtgroup_add_files(kn_subdir, fflags);
1736 	if (!ret)
1737 		kernfs_activate(kn_subdir);
1738 
1739 	return ret;
1740 }
1741 
1742 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
1743 {
1744 	struct resctrl_schema *s;
1745 	struct rdt_resource *r;
1746 	unsigned long fflags;
1747 	char name[32];
1748 	int ret;
1749 
1750 	/* create the directory */
1751 	kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
1752 	if (IS_ERR(kn_info))
1753 		return PTR_ERR(kn_info);
1754 
1755 	ret = rdtgroup_add_files(kn_info, RF_TOP_INFO);
1756 	if (ret)
1757 		goto out_destroy;
1758 
1759 	/* loop over enabled controls, these are all alloc_enabled */
1760 	list_for_each_entry(s, &resctrl_schema_all, list) {
1761 		r = s->res;
1762 		fflags =  r->fflags | RF_CTRL_INFO;
1763 		ret = rdtgroup_mkdir_info_resdir(s, s->name, fflags);
1764 		if (ret)
1765 			goto out_destroy;
1766 	}
1767 
1768 	for_each_mon_enabled_rdt_resource(r) {
1769 		fflags =  r->fflags | RF_MON_INFO;
1770 		sprintf(name, "%s_MON", r->name);
1771 		ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
1772 		if (ret)
1773 			goto out_destroy;
1774 	}
1775 
1776 	ret = rdtgroup_kn_set_ugid(kn_info);
1777 	if (ret)
1778 		goto out_destroy;
1779 
1780 	kernfs_activate(kn_info);
1781 
1782 	return 0;
1783 
1784 out_destroy:
1785 	kernfs_remove(kn_info);
1786 	return ret;
1787 }
1788 
1789 static int
1790 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
1791 		    char *name, struct kernfs_node **dest_kn)
1792 {
1793 	struct kernfs_node *kn;
1794 	int ret;
1795 
1796 	/* create the directory */
1797 	kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1798 	if (IS_ERR(kn))
1799 		return PTR_ERR(kn);
1800 
1801 	if (dest_kn)
1802 		*dest_kn = kn;
1803 
1804 	ret = rdtgroup_kn_set_ugid(kn);
1805 	if (ret)
1806 		goto out_destroy;
1807 
1808 	kernfs_activate(kn);
1809 
1810 	return 0;
1811 
1812 out_destroy:
1813 	kernfs_remove(kn);
1814 	return ret;
1815 }
1816 
1817 static void l3_qos_cfg_update(void *arg)
1818 {
1819 	bool *enable = arg;
1820 
1821 	wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL);
1822 }
1823 
1824 static void l2_qos_cfg_update(void *arg)
1825 {
1826 	bool *enable = arg;
1827 
1828 	wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL);
1829 }
1830 
1831 static inline bool is_mba_linear(void)
1832 {
1833 	return rdt_resources_all[RDT_RESOURCE_MBA].r_resctrl.membw.delay_linear;
1834 }
1835 
1836 static int set_cache_qos_cfg(int level, bool enable)
1837 {
1838 	void (*update)(void *arg);
1839 	struct rdt_resource *r_l;
1840 	cpumask_var_t cpu_mask;
1841 	struct rdt_domain *d;
1842 	int cpu;
1843 
1844 	if (level == RDT_RESOURCE_L3)
1845 		update = l3_qos_cfg_update;
1846 	else if (level == RDT_RESOURCE_L2)
1847 		update = l2_qos_cfg_update;
1848 	else
1849 		return -EINVAL;
1850 
1851 	if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1852 		return -ENOMEM;
1853 
1854 	r_l = &rdt_resources_all[level].r_resctrl;
1855 	list_for_each_entry(d, &r_l->domains, list) {
1856 		if (r_l->cache.arch_has_per_cpu_cfg)
1857 			/* Pick all the CPUs in the domain instance */
1858 			for_each_cpu(cpu, &d->cpu_mask)
1859 				cpumask_set_cpu(cpu, cpu_mask);
1860 		else
1861 			/* Pick one CPU from each domain instance to update MSR */
1862 			cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1863 	}
1864 	cpu = get_cpu();
1865 	/* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1866 	if (cpumask_test_cpu(cpu, cpu_mask))
1867 		update(&enable);
1868 	/* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1869 	smp_call_function_many(cpu_mask, update, &enable, 1);
1870 	put_cpu();
1871 
1872 	free_cpumask_var(cpu_mask);
1873 
1874 	return 0;
1875 }
1876 
1877 /* Restore the qos cfg state when a domain comes online */
1878 void rdt_domain_reconfigure_cdp(struct rdt_resource *r)
1879 {
1880 	struct rdt_hw_resource *hw_res = resctrl_to_arch_res(r);
1881 
1882 	if (!r->cdp_capable)
1883 		return;
1884 
1885 	if (r->rid == RDT_RESOURCE_L2)
1886 		l2_qos_cfg_update(&hw_res->cdp_enabled);
1887 
1888 	if (r->rid == RDT_RESOURCE_L3)
1889 		l3_qos_cfg_update(&hw_res->cdp_enabled);
1890 }
1891 
1892 /*
1893  * Enable or disable the MBA software controller
1894  * which helps user specify bandwidth in MBps.
1895  * MBA software controller is supported only if
1896  * MBM is supported and MBA is in linear scale.
1897  */
1898 static int set_mba_sc(bool mba_sc)
1899 {
1900 	struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA].r_resctrl;
1901 	struct rdt_hw_domain *hw_dom;
1902 	struct rdt_domain *d;
1903 
1904 	if (!is_mbm_enabled() || !is_mba_linear() ||
1905 	    mba_sc == is_mba_sc(r))
1906 		return -EINVAL;
1907 
1908 	r->membw.mba_sc = mba_sc;
1909 	list_for_each_entry(d, &r->domains, list) {
1910 		hw_dom = resctrl_to_arch_dom(d);
1911 		setup_default_ctrlval(r, hw_dom->ctrl_val, hw_dom->mbps_val);
1912 	}
1913 
1914 	return 0;
1915 }
1916 
1917 static int cdp_enable(int level)
1918 {
1919 	struct rdt_resource *r_l = &rdt_resources_all[level].r_resctrl;
1920 	int ret;
1921 
1922 	if (!r_l->alloc_capable)
1923 		return -EINVAL;
1924 
1925 	ret = set_cache_qos_cfg(level, true);
1926 	if (!ret)
1927 		rdt_resources_all[level].cdp_enabled = true;
1928 
1929 	return ret;
1930 }
1931 
1932 static void cdp_disable(int level)
1933 {
1934 	struct rdt_hw_resource *r_hw = &rdt_resources_all[level];
1935 
1936 	if (r_hw->cdp_enabled) {
1937 		set_cache_qos_cfg(level, false);
1938 		r_hw->cdp_enabled = false;
1939 	}
1940 }
1941 
1942 int resctrl_arch_set_cdp_enabled(enum resctrl_res_level l, bool enable)
1943 {
1944 	struct rdt_hw_resource *hw_res = &rdt_resources_all[l];
1945 
1946 	if (!hw_res->r_resctrl.cdp_capable)
1947 		return -EINVAL;
1948 
1949 	if (enable)
1950 		return cdp_enable(l);
1951 
1952 	cdp_disable(l);
1953 
1954 	return 0;
1955 }
1956 
1957 static void cdp_disable_all(void)
1958 {
1959 	if (resctrl_arch_get_cdp_enabled(RDT_RESOURCE_L3))
1960 		resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L3, false);
1961 	if (resctrl_arch_get_cdp_enabled(RDT_RESOURCE_L2))
1962 		resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L2, false);
1963 }
1964 
1965 /*
1966  * We don't allow rdtgroup directories to be created anywhere
1967  * except the root directory. Thus when looking for the rdtgroup
1968  * structure for a kernfs node we are either looking at a directory,
1969  * in which case the rdtgroup structure is pointed at by the "priv"
1970  * field, otherwise we have a file, and need only look to the parent
1971  * to find the rdtgroup.
1972  */
1973 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
1974 {
1975 	if (kernfs_type(kn) == KERNFS_DIR) {
1976 		/*
1977 		 * All the resource directories use "kn->priv"
1978 		 * to point to the "struct rdtgroup" for the
1979 		 * resource. "info" and its subdirectories don't
1980 		 * have rdtgroup structures, so return NULL here.
1981 		 */
1982 		if (kn == kn_info || kn->parent == kn_info)
1983 			return NULL;
1984 		else
1985 			return kn->priv;
1986 	} else {
1987 		return kn->parent->priv;
1988 	}
1989 }
1990 
1991 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
1992 {
1993 	struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1994 
1995 	if (!rdtgrp)
1996 		return NULL;
1997 
1998 	atomic_inc(&rdtgrp->waitcount);
1999 	kernfs_break_active_protection(kn);
2000 
2001 	mutex_lock(&rdtgroup_mutex);
2002 
2003 	/* Was this group deleted while we waited? */
2004 	if (rdtgrp->flags & RDT_DELETED)
2005 		return NULL;
2006 
2007 	return rdtgrp;
2008 }
2009 
2010 void rdtgroup_kn_unlock(struct kernfs_node *kn)
2011 {
2012 	struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
2013 
2014 	if (!rdtgrp)
2015 		return;
2016 
2017 	mutex_unlock(&rdtgroup_mutex);
2018 
2019 	if (atomic_dec_and_test(&rdtgrp->waitcount) &&
2020 	    (rdtgrp->flags & RDT_DELETED)) {
2021 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2022 		    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2023 			rdtgroup_pseudo_lock_remove(rdtgrp);
2024 		kernfs_unbreak_active_protection(kn);
2025 		rdtgroup_remove(rdtgrp);
2026 	} else {
2027 		kernfs_unbreak_active_protection(kn);
2028 	}
2029 }
2030 
2031 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2032 			     struct rdtgroup *prgrp,
2033 			     struct kernfs_node **mon_data_kn);
2034 
2035 static int rdt_enable_ctx(struct rdt_fs_context *ctx)
2036 {
2037 	int ret = 0;
2038 
2039 	if (ctx->enable_cdpl2)
2040 		ret = resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L2, true);
2041 
2042 	if (!ret && ctx->enable_cdpl3)
2043 		ret = resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L3, true);
2044 
2045 	if (!ret && ctx->enable_mba_mbps)
2046 		ret = set_mba_sc(true);
2047 
2048 	return ret;
2049 }
2050 
2051 static int schemata_list_add(struct rdt_resource *r, enum resctrl_conf_type type)
2052 {
2053 	struct resctrl_schema *s;
2054 	const char *suffix = "";
2055 	int ret, cl;
2056 
2057 	s = kzalloc(sizeof(*s), GFP_KERNEL);
2058 	if (!s)
2059 		return -ENOMEM;
2060 
2061 	s->res = r;
2062 	s->num_closid = resctrl_arch_get_num_closid(r);
2063 	if (resctrl_arch_get_cdp_enabled(r->rid))
2064 		s->num_closid /= 2;
2065 
2066 	s->conf_type = type;
2067 	switch (type) {
2068 	case CDP_CODE:
2069 		suffix = "CODE";
2070 		break;
2071 	case CDP_DATA:
2072 		suffix = "DATA";
2073 		break;
2074 	case CDP_NONE:
2075 		suffix = "";
2076 		break;
2077 	}
2078 
2079 	ret = snprintf(s->name, sizeof(s->name), "%s%s", r->name, suffix);
2080 	if (ret >= sizeof(s->name)) {
2081 		kfree(s);
2082 		return -EINVAL;
2083 	}
2084 
2085 	cl = strlen(s->name);
2086 
2087 	/*
2088 	 * If CDP is supported by this resource, but not enabled,
2089 	 * include the suffix. This ensures the tabular format of the
2090 	 * schemata file does not change between mounts of the filesystem.
2091 	 */
2092 	if (r->cdp_capable && !resctrl_arch_get_cdp_enabled(r->rid))
2093 		cl += 4;
2094 
2095 	if (cl > max_name_width)
2096 		max_name_width = cl;
2097 
2098 	INIT_LIST_HEAD(&s->list);
2099 	list_add(&s->list, &resctrl_schema_all);
2100 
2101 	return 0;
2102 }
2103 
2104 static int schemata_list_create(void)
2105 {
2106 	struct rdt_resource *r;
2107 	int ret = 0;
2108 
2109 	for_each_alloc_enabled_rdt_resource(r) {
2110 		if (resctrl_arch_get_cdp_enabled(r->rid)) {
2111 			ret = schemata_list_add(r, CDP_CODE);
2112 			if (ret)
2113 				break;
2114 
2115 			ret = schemata_list_add(r, CDP_DATA);
2116 		} else {
2117 			ret = schemata_list_add(r, CDP_NONE);
2118 		}
2119 
2120 		if (ret)
2121 			break;
2122 	}
2123 
2124 	return ret;
2125 }
2126 
2127 static void schemata_list_destroy(void)
2128 {
2129 	struct resctrl_schema *s, *tmp;
2130 
2131 	list_for_each_entry_safe(s, tmp, &resctrl_schema_all, list) {
2132 		list_del(&s->list);
2133 		kfree(s);
2134 	}
2135 }
2136 
2137 static int rdt_get_tree(struct fs_context *fc)
2138 {
2139 	struct rdt_fs_context *ctx = rdt_fc2context(fc);
2140 	struct rdt_domain *dom;
2141 	struct rdt_resource *r;
2142 	int ret;
2143 
2144 	cpus_read_lock();
2145 	mutex_lock(&rdtgroup_mutex);
2146 	/*
2147 	 * resctrl file system can only be mounted once.
2148 	 */
2149 	if (static_branch_unlikely(&rdt_enable_key)) {
2150 		ret = -EBUSY;
2151 		goto out;
2152 	}
2153 
2154 	ret = rdt_enable_ctx(ctx);
2155 	if (ret < 0)
2156 		goto out_cdp;
2157 
2158 	ret = schemata_list_create();
2159 	if (ret) {
2160 		schemata_list_destroy();
2161 		goto out_mba;
2162 	}
2163 
2164 	closid_init();
2165 
2166 	ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
2167 	if (ret < 0)
2168 		goto out_schemata_free;
2169 
2170 	if (rdt_mon_capable) {
2171 		ret = mongroup_create_dir(rdtgroup_default.kn,
2172 					  &rdtgroup_default, "mon_groups",
2173 					  &kn_mongrp);
2174 		if (ret < 0)
2175 			goto out_info;
2176 
2177 		ret = mkdir_mondata_all(rdtgroup_default.kn,
2178 					&rdtgroup_default, &kn_mondata);
2179 		if (ret < 0)
2180 			goto out_mongrp;
2181 		rdtgroup_default.mon.mon_data_kn = kn_mondata;
2182 	}
2183 
2184 	ret = rdt_pseudo_lock_init();
2185 	if (ret)
2186 		goto out_mondata;
2187 
2188 	ret = kernfs_get_tree(fc);
2189 	if (ret < 0)
2190 		goto out_psl;
2191 
2192 	if (rdt_alloc_capable)
2193 		static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
2194 	if (rdt_mon_capable)
2195 		static_branch_enable_cpuslocked(&rdt_mon_enable_key);
2196 
2197 	if (rdt_alloc_capable || rdt_mon_capable)
2198 		static_branch_enable_cpuslocked(&rdt_enable_key);
2199 
2200 	if (is_mbm_enabled()) {
2201 		r = &rdt_resources_all[RDT_RESOURCE_L3].r_resctrl;
2202 		list_for_each_entry(dom, &r->domains, list)
2203 			mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
2204 	}
2205 
2206 	goto out;
2207 
2208 out_psl:
2209 	rdt_pseudo_lock_release();
2210 out_mondata:
2211 	if (rdt_mon_capable)
2212 		kernfs_remove(kn_mondata);
2213 out_mongrp:
2214 	if (rdt_mon_capable)
2215 		kernfs_remove(kn_mongrp);
2216 out_info:
2217 	kernfs_remove(kn_info);
2218 out_schemata_free:
2219 	schemata_list_destroy();
2220 out_mba:
2221 	if (ctx->enable_mba_mbps)
2222 		set_mba_sc(false);
2223 out_cdp:
2224 	cdp_disable_all();
2225 out:
2226 	rdt_last_cmd_clear();
2227 	mutex_unlock(&rdtgroup_mutex);
2228 	cpus_read_unlock();
2229 	return ret;
2230 }
2231 
2232 enum rdt_param {
2233 	Opt_cdp,
2234 	Opt_cdpl2,
2235 	Opt_mba_mbps,
2236 	nr__rdt_params
2237 };
2238 
2239 static const struct fs_parameter_spec rdt_fs_parameters[] = {
2240 	fsparam_flag("cdp",		Opt_cdp),
2241 	fsparam_flag("cdpl2",		Opt_cdpl2),
2242 	fsparam_flag("mba_MBps",	Opt_mba_mbps),
2243 	{}
2244 };
2245 
2246 static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param)
2247 {
2248 	struct rdt_fs_context *ctx = rdt_fc2context(fc);
2249 	struct fs_parse_result result;
2250 	int opt;
2251 
2252 	opt = fs_parse(fc, rdt_fs_parameters, param, &result);
2253 	if (opt < 0)
2254 		return opt;
2255 
2256 	switch (opt) {
2257 	case Opt_cdp:
2258 		ctx->enable_cdpl3 = true;
2259 		return 0;
2260 	case Opt_cdpl2:
2261 		ctx->enable_cdpl2 = true;
2262 		return 0;
2263 	case Opt_mba_mbps:
2264 		if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2265 			return -EINVAL;
2266 		ctx->enable_mba_mbps = true;
2267 		return 0;
2268 	}
2269 
2270 	return -EINVAL;
2271 }
2272 
2273 static void rdt_fs_context_free(struct fs_context *fc)
2274 {
2275 	struct rdt_fs_context *ctx = rdt_fc2context(fc);
2276 
2277 	kernfs_free_fs_context(fc);
2278 	kfree(ctx);
2279 }
2280 
2281 static const struct fs_context_operations rdt_fs_context_ops = {
2282 	.free		= rdt_fs_context_free,
2283 	.parse_param	= rdt_parse_param,
2284 	.get_tree	= rdt_get_tree,
2285 };
2286 
2287 static int rdt_init_fs_context(struct fs_context *fc)
2288 {
2289 	struct rdt_fs_context *ctx;
2290 
2291 	ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL);
2292 	if (!ctx)
2293 		return -ENOMEM;
2294 
2295 	ctx->kfc.root = rdt_root;
2296 	ctx->kfc.magic = RDTGROUP_SUPER_MAGIC;
2297 	fc->fs_private = &ctx->kfc;
2298 	fc->ops = &rdt_fs_context_ops;
2299 	put_user_ns(fc->user_ns);
2300 	fc->user_ns = get_user_ns(&init_user_ns);
2301 	fc->global = true;
2302 	return 0;
2303 }
2304 
2305 static int reset_all_ctrls(struct rdt_resource *r)
2306 {
2307 	struct rdt_hw_resource *hw_res = resctrl_to_arch_res(r);
2308 	struct rdt_hw_domain *hw_dom;
2309 	struct msr_param msr_param;
2310 	cpumask_var_t cpu_mask;
2311 	struct rdt_domain *d;
2312 	int i, cpu;
2313 
2314 	if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
2315 		return -ENOMEM;
2316 
2317 	msr_param.res = r;
2318 	msr_param.low = 0;
2319 	msr_param.high = hw_res->num_closid;
2320 
2321 	/*
2322 	 * Disable resource control for this resource by setting all
2323 	 * CBMs in all domains to the maximum mask value. Pick one CPU
2324 	 * from each domain to update the MSRs below.
2325 	 */
2326 	list_for_each_entry(d, &r->domains, list) {
2327 		hw_dom = resctrl_to_arch_dom(d);
2328 		cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
2329 
2330 		for (i = 0; i < hw_res->num_closid; i++)
2331 			hw_dom->ctrl_val[i] = r->default_ctrl;
2332 	}
2333 	cpu = get_cpu();
2334 	/* Update CBM on this cpu if it's in cpu_mask. */
2335 	if (cpumask_test_cpu(cpu, cpu_mask))
2336 		rdt_ctrl_update(&msr_param);
2337 	/* Update CBM on all other cpus in cpu_mask. */
2338 	smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
2339 	put_cpu();
2340 
2341 	free_cpumask_var(cpu_mask);
2342 
2343 	return 0;
2344 }
2345 
2346 /*
2347  * Move tasks from one to the other group. If @from is NULL, then all tasks
2348  * in the systems are moved unconditionally (used for teardown).
2349  *
2350  * If @mask is not NULL the cpus on which moved tasks are running are set
2351  * in that mask so the update smp function call is restricted to affected
2352  * cpus.
2353  */
2354 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
2355 				 struct cpumask *mask)
2356 {
2357 	struct task_struct *p, *t;
2358 
2359 	read_lock(&tasklist_lock);
2360 	for_each_process_thread(p, t) {
2361 		if (!from || is_closid_match(t, from) ||
2362 		    is_rmid_match(t, from)) {
2363 			WRITE_ONCE(t->closid, to->closid);
2364 			WRITE_ONCE(t->rmid, to->mon.rmid);
2365 
2366 			/*
2367 			 * If the task is on a CPU, set the CPU in the mask.
2368 			 * The detection is inaccurate as tasks might move or
2369 			 * schedule before the smp function call takes place.
2370 			 * In such a case the function call is pointless, but
2371 			 * there is no other side effect.
2372 			 */
2373 			if (IS_ENABLED(CONFIG_SMP) && mask && task_curr(t))
2374 				cpumask_set_cpu(task_cpu(t), mask);
2375 		}
2376 	}
2377 	read_unlock(&tasklist_lock);
2378 }
2379 
2380 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
2381 {
2382 	struct rdtgroup *sentry, *stmp;
2383 	struct list_head *head;
2384 
2385 	head = &rdtgrp->mon.crdtgrp_list;
2386 	list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
2387 		free_rmid(sentry->mon.rmid);
2388 		list_del(&sentry->mon.crdtgrp_list);
2389 
2390 		if (atomic_read(&sentry->waitcount) != 0)
2391 			sentry->flags = RDT_DELETED;
2392 		else
2393 			rdtgroup_remove(sentry);
2394 	}
2395 }
2396 
2397 /*
2398  * Forcibly remove all of subdirectories under root.
2399  */
2400 static void rmdir_all_sub(void)
2401 {
2402 	struct rdtgroup *rdtgrp, *tmp;
2403 
2404 	/* Move all tasks to the default resource group */
2405 	rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
2406 
2407 	list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
2408 		/* Free any child rmids */
2409 		free_all_child_rdtgrp(rdtgrp);
2410 
2411 		/* Remove each rdtgroup other than root */
2412 		if (rdtgrp == &rdtgroup_default)
2413 			continue;
2414 
2415 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2416 		    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2417 			rdtgroup_pseudo_lock_remove(rdtgrp);
2418 
2419 		/*
2420 		 * Give any CPUs back to the default group. We cannot copy
2421 		 * cpu_online_mask because a CPU might have executed the
2422 		 * offline callback already, but is still marked online.
2423 		 */
2424 		cpumask_or(&rdtgroup_default.cpu_mask,
2425 			   &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2426 
2427 		free_rmid(rdtgrp->mon.rmid);
2428 
2429 		kernfs_remove(rdtgrp->kn);
2430 		list_del(&rdtgrp->rdtgroup_list);
2431 
2432 		if (atomic_read(&rdtgrp->waitcount) != 0)
2433 			rdtgrp->flags = RDT_DELETED;
2434 		else
2435 			rdtgroup_remove(rdtgrp);
2436 	}
2437 	/* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
2438 	update_closid_rmid(cpu_online_mask, &rdtgroup_default);
2439 
2440 	kernfs_remove(kn_info);
2441 	kernfs_remove(kn_mongrp);
2442 	kernfs_remove(kn_mondata);
2443 }
2444 
2445 static void rdt_kill_sb(struct super_block *sb)
2446 {
2447 	struct rdt_resource *r;
2448 
2449 	cpus_read_lock();
2450 	mutex_lock(&rdtgroup_mutex);
2451 
2452 	set_mba_sc(false);
2453 
2454 	/*Put everything back to default values. */
2455 	for_each_alloc_enabled_rdt_resource(r)
2456 		reset_all_ctrls(r);
2457 	cdp_disable_all();
2458 	rmdir_all_sub();
2459 	rdt_pseudo_lock_release();
2460 	rdtgroup_default.mode = RDT_MODE_SHAREABLE;
2461 	schemata_list_destroy();
2462 	static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
2463 	static_branch_disable_cpuslocked(&rdt_mon_enable_key);
2464 	static_branch_disable_cpuslocked(&rdt_enable_key);
2465 	kernfs_kill_sb(sb);
2466 	mutex_unlock(&rdtgroup_mutex);
2467 	cpus_read_unlock();
2468 }
2469 
2470 static struct file_system_type rdt_fs_type = {
2471 	.name			= "resctrl",
2472 	.init_fs_context	= rdt_init_fs_context,
2473 	.parameters		= rdt_fs_parameters,
2474 	.kill_sb		= rdt_kill_sb,
2475 };
2476 
2477 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
2478 		       void *priv)
2479 {
2480 	struct kernfs_node *kn;
2481 	int ret = 0;
2482 
2483 	kn = __kernfs_create_file(parent_kn, name, 0444,
2484 				  GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
2485 				  &kf_mondata_ops, priv, NULL, NULL);
2486 	if (IS_ERR(kn))
2487 		return PTR_ERR(kn);
2488 
2489 	ret = rdtgroup_kn_set_ugid(kn);
2490 	if (ret) {
2491 		kernfs_remove(kn);
2492 		return ret;
2493 	}
2494 
2495 	return ret;
2496 }
2497 
2498 /*
2499  * Remove all subdirectories of mon_data of ctrl_mon groups
2500  * and monitor groups with given domain id.
2501  */
2502 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
2503 {
2504 	struct rdtgroup *prgrp, *crgrp;
2505 	char name[32];
2506 
2507 	if (!r->mon_enabled)
2508 		return;
2509 
2510 	list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2511 		sprintf(name, "mon_%s_%02d", r->name, dom_id);
2512 		kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
2513 
2514 		list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
2515 			kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
2516 	}
2517 }
2518 
2519 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
2520 				struct rdt_domain *d,
2521 				struct rdt_resource *r, struct rdtgroup *prgrp)
2522 {
2523 	union mon_data_bits priv;
2524 	struct kernfs_node *kn;
2525 	struct mon_evt *mevt;
2526 	struct rmid_read rr;
2527 	char name[32];
2528 	int ret;
2529 
2530 	sprintf(name, "mon_%s_%02d", r->name, d->id);
2531 	/* create the directory */
2532 	kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
2533 	if (IS_ERR(kn))
2534 		return PTR_ERR(kn);
2535 
2536 	ret = rdtgroup_kn_set_ugid(kn);
2537 	if (ret)
2538 		goto out_destroy;
2539 
2540 	if (WARN_ON(list_empty(&r->evt_list))) {
2541 		ret = -EPERM;
2542 		goto out_destroy;
2543 	}
2544 
2545 	priv.u.rid = r->rid;
2546 	priv.u.domid = d->id;
2547 	list_for_each_entry(mevt, &r->evt_list, list) {
2548 		priv.u.evtid = mevt->evtid;
2549 		ret = mon_addfile(kn, mevt->name, priv.priv);
2550 		if (ret)
2551 			goto out_destroy;
2552 
2553 		if (is_mbm_event(mevt->evtid))
2554 			mon_event_read(&rr, r, d, prgrp, mevt->evtid, true);
2555 	}
2556 	kernfs_activate(kn);
2557 	return 0;
2558 
2559 out_destroy:
2560 	kernfs_remove(kn);
2561 	return ret;
2562 }
2563 
2564 /*
2565  * Add all subdirectories of mon_data for "ctrl_mon" groups
2566  * and "monitor" groups with given domain id.
2567  */
2568 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
2569 				    struct rdt_domain *d)
2570 {
2571 	struct kernfs_node *parent_kn;
2572 	struct rdtgroup *prgrp, *crgrp;
2573 	struct list_head *head;
2574 
2575 	if (!r->mon_enabled)
2576 		return;
2577 
2578 	list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2579 		parent_kn = prgrp->mon.mon_data_kn;
2580 		mkdir_mondata_subdir(parent_kn, d, r, prgrp);
2581 
2582 		head = &prgrp->mon.crdtgrp_list;
2583 		list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
2584 			parent_kn = crgrp->mon.mon_data_kn;
2585 			mkdir_mondata_subdir(parent_kn, d, r, crgrp);
2586 		}
2587 	}
2588 }
2589 
2590 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
2591 				       struct rdt_resource *r,
2592 				       struct rdtgroup *prgrp)
2593 {
2594 	struct rdt_domain *dom;
2595 	int ret;
2596 
2597 	list_for_each_entry(dom, &r->domains, list) {
2598 		ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
2599 		if (ret)
2600 			return ret;
2601 	}
2602 
2603 	return 0;
2604 }
2605 
2606 /*
2607  * This creates a directory mon_data which contains the monitored data.
2608  *
2609  * mon_data has one directory for each domain which are named
2610  * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
2611  * with L3 domain looks as below:
2612  * ./mon_data:
2613  * mon_L3_00
2614  * mon_L3_01
2615  * mon_L3_02
2616  * ...
2617  *
2618  * Each domain directory has one file per event:
2619  * ./mon_L3_00/:
2620  * llc_occupancy
2621  *
2622  */
2623 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2624 			     struct rdtgroup *prgrp,
2625 			     struct kernfs_node **dest_kn)
2626 {
2627 	struct rdt_resource *r;
2628 	struct kernfs_node *kn;
2629 	int ret;
2630 
2631 	/*
2632 	 * Create the mon_data directory first.
2633 	 */
2634 	ret = mongroup_create_dir(parent_kn, prgrp, "mon_data", &kn);
2635 	if (ret)
2636 		return ret;
2637 
2638 	if (dest_kn)
2639 		*dest_kn = kn;
2640 
2641 	/*
2642 	 * Create the subdirectories for each domain. Note that all events
2643 	 * in a domain like L3 are grouped into a resource whose domain is L3
2644 	 */
2645 	for_each_mon_enabled_rdt_resource(r) {
2646 		ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
2647 		if (ret)
2648 			goto out_destroy;
2649 	}
2650 
2651 	return 0;
2652 
2653 out_destroy:
2654 	kernfs_remove(kn);
2655 	return ret;
2656 }
2657 
2658 /**
2659  * cbm_ensure_valid - Enforce validity on provided CBM
2660  * @_val:	Candidate CBM
2661  * @r:		RDT resource to which the CBM belongs
2662  *
2663  * The provided CBM represents all cache portions available for use. This
2664  * may be represented by a bitmap that does not consist of contiguous ones
2665  * and thus be an invalid CBM.
2666  * Here the provided CBM is forced to be a valid CBM by only considering
2667  * the first set of contiguous bits as valid and clearing all bits.
2668  * The intention here is to provide a valid default CBM with which a new
2669  * resource group is initialized. The user can follow this with a
2670  * modification to the CBM if the default does not satisfy the
2671  * requirements.
2672  */
2673 static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r)
2674 {
2675 	unsigned int cbm_len = r->cache.cbm_len;
2676 	unsigned long first_bit, zero_bit;
2677 	unsigned long val = _val;
2678 
2679 	if (!val)
2680 		return 0;
2681 
2682 	first_bit = find_first_bit(&val, cbm_len);
2683 	zero_bit = find_next_zero_bit(&val, cbm_len, first_bit);
2684 
2685 	/* Clear any remaining bits to ensure contiguous region */
2686 	bitmap_clear(&val, zero_bit, cbm_len - zero_bit);
2687 	return (u32)val;
2688 }
2689 
2690 /*
2691  * Initialize cache resources per RDT domain
2692  *
2693  * Set the RDT domain up to start off with all usable allocations. That is,
2694  * all shareable and unused bits. All-zero CBM is invalid.
2695  */
2696 static int __init_one_rdt_domain(struct rdt_domain *d, struct resctrl_schema *s,
2697 				 u32 closid)
2698 {
2699 	enum resctrl_conf_type peer_type = resctrl_peer_type(s->conf_type);
2700 	enum resctrl_conf_type t = s->conf_type;
2701 	struct resctrl_staged_config *cfg;
2702 	struct rdt_resource *r = s->res;
2703 	u32 used_b = 0, unused_b = 0;
2704 	unsigned long tmp_cbm;
2705 	enum rdtgrp_mode mode;
2706 	u32 peer_ctl, ctrl_val;
2707 	int i;
2708 
2709 	cfg = &d->staged_config[t];
2710 	cfg->have_new_ctrl = false;
2711 	cfg->new_ctrl = r->cache.shareable_bits;
2712 	used_b = r->cache.shareable_bits;
2713 	for (i = 0; i < closids_supported(); i++) {
2714 		if (closid_allocated(i) && i != closid) {
2715 			mode = rdtgroup_mode_by_closid(i);
2716 			if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
2717 				/*
2718 				 * ctrl values for locksetup aren't relevant
2719 				 * until the schemata is written, and the mode
2720 				 * becomes RDT_MODE_PSEUDO_LOCKED.
2721 				 */
2722 				continue;
2723 			/*
2724 			 * If CDP is active include peer domain's
2725 			 * usage to ensure there is no overlap
2726 			 * with an exclusive group.
2727 			 */
2728 			if (resctrl_arch_get_cdp_enabled(r->rid))
2729 				peer_ctl = resctrl_arch_get_config(r, d, i,
2730 								   peer_type);
2731 			else
2732 				peer_ctl = 0;
2733 			ctrl_val = resctrl_arch_get_config(r, d, i,
2734 							   s->conf_type);
2735 			used_b |= ctrl_val | peer_ctl;
2736 			if (mode == RDT_MODE_SHAREABLE)
2737 				cfg->new_ctrl |= ctrl_val | peer_ctl;
2738 		}
2739 	}
2740 	if (d->plr && d->plr->cbm > 0)
2741 		used_b |= d->plr->cbm;
2742 	unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
2743 	unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
2744 	cfg->new_ctrl |= unused_b;
2745 	/*
2746 	 * Force the initial CBM to be valid, user can
2747 	 * modify the CBM based on system availability.
2748 	 */
2749 	cfg->new_ctrl = cbm_ensure_valid(cfg->new_ctrl, r);
2750 	/*
2751 	 * Assign the u32 CBM to an unsigned long to ensure that
2752 	 * bitmap_weight() does not access out-of-bound memory.
2753 	 */
2754 	tmp_cbm = cfg->new_ctrl;
2755 	if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) {
2756 		rdt_last_cmd_printf("No space on %s:%d\n", s->name, d->id);
2757 		return -ENOSPC;
2758 	}
2759 	cfg->have_new_ctrl = true;
2760 
2761 	return 0;
2762 }
2763 
2764 /*
2765  * Initialize cache resources with default values.
2766  *
2767  * A new RDT group is being created on an allocation capable (CAT)
2768  * supporting system. Set this group up to start off with all usable
2769  * allocations.
2770  *
2771  * If there are no more shareable bits available on any domain then
2772  * the entire allocation will fail.
2773  */
2774 static int rdtgroup_init_cat(struct resctrl_schema *s, u32 closid)
2775 {
2776 	struct rdt_domain *d;
2777 	int ret;
2778 
2779 	list_for_each_entry(d, &s->res->domains, list) {
2780 		ret = __init_one_rdt_domain(d, s, closid);
2781 		if (ret < 0)
2782 			return ret;
2783 	}
2784 
2785 	return 0;
2786 }
2787 
2788 /* Initialize MBA resource with default values. */
2789 static void rdtgroup_init_mba(struct rdt_resource *r)
2790 {
2791 	struct resctrl_staged_config *cfg;
2792 	struct rdt_domain *d;
2793 
2794 	list_for_each_entry(d, &r->domains, list) {
2795 		cfg = &d->staged_config[CDP_NONE];
2796 		cfg->new_ctrl = is_mba_sc(r) ? MBA_MAX_MBPS : r->default_ctrl;
2797 		cfg->have_new_ctrl = true;
2798 	}
2799 }
2800 
2801 /* Initialize the RDT group's allocations. */
2802 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
2803 {
2804 	struct resctrl_schema *s;
2805 	struct rdt_resource *r;
2806 	int ret;
2807 
2808 	list_for_each_entry(s, &resctrl_schema_all, list) {
2809 		r = s->res;
2810 		if (r->rid == RDT_RESOURCE_MBA) {
2811 			rdtgroup_init_mba(r);
2812 		} else {
2813 			ret = rdtgroup_init_cat(s, rdtgrp->closid);
2814 			if (ret < 0)
2815 				return ret;
2816 		}
2817 
2818 		ret = resctrl_arch_update_domains(r, rdtgrp->closid);
2819 		if (ret < 0) {
2820 			rdt_last_cmd_puts("Failed to initialize allocations\n");
2821 			return ret;
2822 		}
2823 
2824 	}
2825 
2826 	rdtgrp->mode = RDT_MODE_SHAREABLE;
2827 
2828 	return 0;
2829 }
2830 
2831 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
2832 			     const char *name, umode_t mode,
2833 			     enum rdt_group_type rtype, struct rdtgroup **r)
2834 {
2835 	struct rdtgroup *prdtgrp, *rdtgrp;
2836 	struct kernfs_node *kn;
2837 	uint files = 0;
2838 	int ret;
2839 
2840 	prdtgrp = rdtgroup_kn_lock_live(parent_kn);
2841 	if (!prdtgrp) {
2842 		ret = -ENODEV;
2843 		goto out_unlock;
2844 	}
2845 
2846 	if (rtype == RDTMON_GROUP &&
2847 	    (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2848 	     prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
2849 		ret = -EINVAL;
2850 		rdt_last_cmd_puts("Pseudo-locking in progress\n");
2851 		goto out_unlock;
2852 	}
2853 
2854 	/* allocate the rdtgroup. */
2855 	rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
2856 	if (!rdtgrp) {
2857 		ret = -ENOSPC;
2858 		rdt_last_cmd_puts("Kernel out of memory\n");
2859 		goto out_unlock;
2860 	}
2861 	*r = rdtgrp;
2862 	rdtgrp->mon.parent = prdtgrp;
2863 	rdtgrp->type = rtype;
2864 	INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
2865 
2866 	/* kernfs creates the directory for rdtgrp */
2867 	kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
2868 	if (IS_ERR(kn)) {
2869 		ret = PTR_ERR(kn);
2870 		rdt_last_cmd_puts("kernfs create error\n");
2871 		goto out_free_rgrp;
2872 	}
2873 	rdtgrp->kn = kn;
2874 
2875 	/*
2876 	 * kernfs_remove() will drop the reference count on "kn" which
2877 	 * will free it. But we still need it to stick around for the
2878 	 * rdtgroup_kn_unlock(kn) call. Take one extra reference here,
2879 	 * which will be dropped by kernfs_put() in rdtgroup_remove().
2880 	 */
2881 	kernfs_get(kn);
2882 
2883 	ret = rdtgroup_kn_set_ugid(kn);
2884 	if (ret) {
2885 		rdt_last_cmd_puts("kernfs perm error\n");
2886 		goto out_destroy;
2887 	}
2888 
2889 	files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
2890 	ret = rdtgroup_add_files(kn, files);
2891 	if (ret) {
2892 		rdt_last_cmd_puts("kernfs fill error\n");
2893 		goto out_destroy;
2894 	}
2895 
2896 	if (rdt_mon_capable) {
2897 		ret = alloc_rmid();
2898 		if (ret < 0) {
2899 			rdt_last_cmd_puts("Out of RMIDs\n");
2900 			goto out_destroy;
2901 		}
2902 		rdtgrp->mon.rmid = ret;
2903 
2904 		ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
2905 		if (ret) {
2906 			rdt_last_cmd_puts("kernfs subdir error\n");
2907 			goto out_idfree;
2908 		}
2909 	}
2910 	kernfs_activate(kn);
2911 
2912 	/*
2913 	 * The caller unlocks the parent_kn upon success.
2914 	 */
2915 	return 0;
2916 
2917 out_idfree:
2918 	free_rmid(rdtgrp->mon.rmid);
2919 out_destroy:
2920 	kernfs_put(rdtgrp->kn);
2921 	kernfs_remove(rdtgrp->kn);
2922 out_free_rgrp:
2923 	kfree(rdtgrp);
2924 out_unlock:
2925 	rdtgroup_kn_unlock(parent_kn);
2926 	return ret;
2927 }
2928 
2929 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
2930 {
2931 	kernfs_remove(rgrp->kn);
2932 	free_rmid(rgrp->mon.rmid);
2933 	rdtgroup_remove(rgrp);
2934 }
2935 
2936 /*
2937  * Create a monitor group under "mon_groups" directory of a control
2938  * and monitor group(ctrl_mon). This is a resource group
2939  * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
2940  */
2941 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
2942 			      const char *name, umode_t mode)
2943 {
2944 	struct rdtgroup *rdtgrp, *prgrp;
2945 	int ret;
2946 
2947 	ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTMON_GROUP, &rdtgrp);
2948 	if (ret)
2949 		return ret;
2950 
2951 	prgrp = rdtgrp->mon.parent;
2952 	rdtgrp->closid = prgrp->closid;
2953 
2954 	/*
2955 	 * Add the rdtgrp to the list of rdtgrps the parent
2956 	 * ctrl_mon group has to track.
2957 	 */
2958 	list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
2959 
2960 	rdtgroup_kn_unlock(parent_kn);
2961 	return ret;
2962 }
2963 
2964 /*
2965  * These are rdtgroups created under the root directory. Can be used
2966  * to allocate and monitor resources.
2967  */
2968 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
2969 				   const char *name, umode_t mode)
2970 {
2971 	struct rdtgroup *rdtgrp;
2972 	struct kernfs_node *kn;
2973 	u32 closid;
2974 	int ret;
2975 
2976 	ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTCTRL_GROUP, &rdtgrp);
2977 	if (ret)
2978 		return ret;
2979 
2980 	kn = rdtgrp->kn;
2981 	ret = closid_alloc();
2982 	if (ret < 0) {
2983 		rdt_last_cmd_puts("Out of CLOSIDs\n");
2984 		goto out_common_fail;
2985 	}
2986 	closid = ret;
2987 	ret = 0;
2988 
2989 	rdtgrp->closid = closid;
2990 	ret = rdtgroup_init_alloc(rdtgrp);
2991 	if (ret < 0)
2992 		goto out_id_free;
2993 
2994 	list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
2995 
2996 	if (rdt_mon_capable) {
2997 		/*
2998 		 * Create an empty mon_groups directory to hold the subset
2999 		 * of tasks and cpus to monitor.
3000 		 */
3001 		ret = mongroup_create_dir(kn, rdtgrp, "mon_groups", NULL);
3002 		if (ret) {
3003 			rdt_last_cmd_puts("kernfs subdir error\n");
3004 			goto out_del_list;
3005 		}
3006 	}
3007 
3008 	goto out_unlock;
3009 
3010 out_del_list:
3011 	list_del(&rdtgrp->rdtgroup_list);
3012 out_id_free:
3013 	closid_free(closid);
3014 out_common_fail:
3015 	mkdir_rdt_prepare_clean(rdtgrp);
3016 out_unlock:
3017 	rdtgroup_kn_unlock(parent_kn);
3018 	return ret;
3019 }
3020 
3021 /*
3022  * We allow creating mon groups only with in a directory called "mon_groups"
3023  * which is present in every ctrl_mon group. Check if this is a valid
3024  * "mon_groups" directory.
3025  *
3026  * 1. The directory should be named "mon_groups".
3027  * 2. The mon group itself should "not" be named "mon_groups".
3028  *   This makes sure "mon_groups" directory always has a ctrl_mon group
3029  *   as parent.
3030  */
3031 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
3032 {
3033 	return (!strcmp(kn->name, "mon_groups") &&
3034 		strcmp(name, "mon_groups"));
3035 }
3036 
3037 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
3038 			  umode_t mode)
3039 {
3040 	/* Do not accept '\n' to avoid unparsable situation. */
3041 	if (strchr(name, '\n'))
3042 		return -EINVAL;
3043 
3044 	/*
3045 	 * If the parent directory is the root directory and RDT
3046 	 * allocation is supported, add a control and monitoring
3047 	 * subdirectory
3048 	 */
3049 	if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
3050 		return rdtgroup_mkdir_ctrl_mon(parent_kn, name, mode);
3051 
3052 	/*
3053 	 * If RDT monitoring is supported and the parent directory is a valid
3054 	 * "mon_groups" directory, add a monitoring subdirectory.
3055 	 */
3056 	if (rdt_mon_capable && is_mon_groups(parent_kn, name))
3057 		return rdtgroup_mkdir_mon(parent_kn, name, mode);
3058 
3059 	return -EPERM;
3060 }
3061 
3062 static int rdtgroup_rmdir_mon(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask)
3063 {
3064 	struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
3065 	int cpu;
3066 
3067 	/* Give any tasks back to the parent group */
3068 	rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
3069 
3070 	/* Update per cpu rmid of the moved CPUs first */
3071 	for_each_cpu(cpu, &rdtgrp->cpu_mask)
3072 		per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
3073 	/*
3074 	 * Update the MSR on moved CPUs and CPUs which have moved
3075 	 * task running on them.
3076 	 */
3077 	cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
3078 	update_closid_rmid(tmpmask, NULL);
3079 
3080 	rdtgrp->flags = RDT_DELETED;
3081 	free_rmid(rdtgrp->mon.rmid);
3082 
3083 	/*
3084 	 * Remove the rdtgrp from the parent ctrl_mon group's list
3085 	 */
3086 	WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
3087 	list_del(&rdtgrp->mon.crdtgrp_list);
3088 
3089 	kernfs_remove(rdtgrp->kn);
3090 
3091 	return 0;
3092 }
3093 
3094 static int rdtgroup_ctrl_remove(struct rdtgroup *rdtgrp)
3095 {
3096 	rdtgrp->flags = RDT_DELETED;
3097 	list_del(&rdtgrp->rdtgroup_list);
3098 
3099 	kernfs_remove(rdtgrp->kn);
3100 	return 0;
3101 }
3102 
3103 static int rdtgroup_rmdir_ctrl(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask)
3104 {
3105 	int cpu;
3106 
3107 	/* Give any tasks back to the default group */
3108 	rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
3109 
3110 	/* Give any CPUs back to the default group */
3111 	cpumask_or(&rdtgroup_default.cpu_mask,
3112 		   &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
3113 
3114 	/* Update per cpu closid and rmid of the moved CPUs first */
3115 	for_each_cpu(cpu, &rdtgrp->cpu_mask) {
3116 		per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
3117 		per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
3118 	}
3119 
3120 	/*
3121 	 * Update the MSR on moved CPUs and CPUs which have moved
3122 	 * task running on them.
3123 	 */
3124 	cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
3125 	update_closid_rmid(tmpmask, NULL);
3126 
3127 	closid_free(rdtgrp->closid);
3128 	free_rmid(rdtgrp->mon.rmid);
3129 
3130 	rdtgroup_ctrl_remove(rdtgrp);
3131 
3132 	/*
3133 	 * Free all the child monitor group rmids.
3134 	 */
3135 	free_all_child_rdtgrp(rdtgrp);
3136 
3137 	return 0;
3138 }
3139 
3140 static int rdtgroup_rmdir(struct kernfs_node *kn)
3141 {
3142 	struct kernfs_node *parent_kn = kn->parent;
3143 	struct rdtgroup *rdtgrp;
3144 	cpumask_var_t tmpmask;
3145 	int ret = 0;
3146 
3147 	if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
3148 		return -ENOMEM;
3149 
3150 	rdtgrp = rdtgroup_kn_lock_live(kn);
3151 	if (!rdtgrp) {
3152 		ret = -EPERM;
3153 		goto out;
3154 	}
3155 
3156 	/*
3157 	 * If the rdtgroup is a ctrl_mon group and parent directory
3158 	 * is the root directory, remove the ctrl_mon group.
3159 	 *
3160 	 * If the rdtgroup is a mon group and parent directory
3161 	 * is a valid "mon_groups" directory, remove the mon group.
3162 	 */
3163 	if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn &&
3164 	    rdtgrp != &rdtgroup_default) {
3165 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
3166 		    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
3167 			ret = rdtgroup_ctrl_remove(rdtgrp);
3168 		} else {
3169 			ret = rdtgroup_rmdir_ctrl(rdtgrp, tmpmask);
3170 		}
3171 	} else if (rdtgrp->type == RDTMON_GROUP &&
3172 		 is_mon_groups(parent_kn, kn->name)) {
3173 		ret = rdtgroup_rmdir_mon(rdtgrp, tmpmask);
3174 	} else {
3175 		ret = -EPERM;
3176 	}
3177 
3178 out:
3179 	rdtgroup_kn_unlock(kn);
3180 	free_cpumask_var(tmpmask);
3181 	return ret;
3182 }
3183 
3184 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
3185 {
3186 	if (resctrl_arch_get_cdp_enabled(RDT_RESOURCE_L3))
3187 		seq_puts(seq, ",cdp");
3188 
3189 	if (resctrl_arch_get_cdp_enabled(RDT_RESOURCE_L2))
3190 		seq_puts(seq, ",cdpl2");
3191 
3192 	if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA].r_resctrl))
3193 		seq_puts(seq, ",mba_MBps");
3194 
3195 	return 0;
3196 }
3197 
3198 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
3199 	.mkdir		= rdtgroup_mkdir,
3200 	.rmdir		= rdtgroup_rmdir,
3201 	.show_options	= rdtgroup_show_options,
3202 };
3203 
3204 static int __init rdtgroup_setup_root(void)
3205 {
3206 	int ret;
3207 
3208 	rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
3209 				      KERNFS_ROOT_CREATE_DEACTIVATED |
3210 				      KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
3211 				      &rdtgroup_default);
3212 	if (IS_ERR(rdt_root))
3213 		return PTR_ERR(rdt_root);
3214 
3215 	mutex_lock(&rdtgroup_mutex);
3216 
3217 	rdtgroup_default.closid = 0;
3218 	rdtgroup_default.mon.rmid = 0;
3219 	rdtgroup_default.type = RDTCTRL_GROUP;
3220 	INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
3221 
3222 	list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
3223 
3224 	ret = rdtgroup_add_files(kernfs_root_to_node(rdt_root), RF_CTRL_BASE);
3225 	if (ret) {
3226 		kernfs_destroy_root(rdt_root);
3227 		goto out;
3228 	}
3229 
3230 	rdtgroup_default.kn = kernfs_root_to_node(rdt_root);
3231 	kernfs_activate(rdtgroup_default.kn);
3232 
3233 out:
3234 	mutex_unlock(&rdtgroup_mutex);
3235 
3236 	return ret;
3237 }
3238 
3239 /*
3240  * rdtgroup_init - rdtgroup initialization
3241  *
3242  * Setup resctrl file system including set up root, create mount point,
3243  * register rdtgroup filesystem, and initialize files under root directory.
3244  *
3245  * Return: 0 on success or -errno
3246  */
3247 int __init rdtgroup_init(void)
3248 {
3249 	int ret = 0;
3250 
3251 	seq_buf_init(&last_cmd_status, last_cmd_status_buf,
3252 		     sizeof(last_cmd_status_buf));
3253 
3254 	ret = rdtgroup_setup_root();
3255 	if (ret)
3256 		return ret;
3257 
3258 	ret = sysfs_create_mount_point(fs_kobj, "resctrl");
3259 	if (ret)
3260 		goto cleanup_root;
3261 
3262 	ret = register_filesystem(&rdt_fs_type);
3263 	if (ret)
3264 		goto cleanup_mountpoint;
3265 
3266 	/*
3267 	 * Adding the resctrl debugfs directory here may not be ideal since
3268 	 * it would let the resctrl debugfs directory appear on the debugfs
3269 	 * filesystem before the resctrl filesystem is mounted.
3270 	 * It may also be ok since that would enable debugging of RDT before
3271 	 * resctrl is mounted.
3272 	 * The reason why the debugfs directory is created here and not in
3273 	 * rdt_get_tree() is because rdt_get_tree() takes rdtgroup_mutex and
3274 	 * during the debugfs directory creation also &sb->s_type->i_mutex_key
3275 	 * (the lockdep class of inode->i_rwsem). Other filesystem
3276 	 * interactions (eg. SyS_getdents) have the lock ordering:
3277 	 * &sb->s_type->i_mutex_key --> &mm->mmap_lock
3278 	 * During mmap(), called with &mm->mmap_lock, the rdtgroup_mutex
3279 	 * is taken, thus creating dependency:
3280 	 * &mm->mmap_lock --> rdtgroup_mutex for the latter that can cause
3281 	 * issues considering the other two lock dependencies.
3282 	 * By creating the debugfs directory here we avoid a dependency
3283 	 * that may cause deadlock (even though file operations cannot
3284 	 * occur until the filesystem is mounted, but I do not know how to
3285 	 * tell lockdep that).
3286 	 */
3287 	debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
3288 
3289 	return 0;
3290 
3291 cleanup_mountpoint:
3292 	sysfs_remove_mount_point(fs_kobj, "resctrl");
3293 cleanup_root:
3294 	kernfs_destroy_root(rdt_root);
3295 
3296 	return ret;
3297 }
3298 
3299 void __exit rdtgroup_exit(void)
3300 {
3301 	debugfs_remove_recursive(debugfs_resctrl);
3302 	unregister_filesystem(&rdt_fs_type);
3303 	sysfs_remove_mount_point(fs_kobj, "resctrl");
3304 	kernfs_destroy_root(rdt_root);
3305 }
3306