1 /*
2  * User interface for Resource Alloction in Resource Director Technology(RDT)
3  *
4  * Copyright (C) 2016 Intel Corporation
5  *
6  * Author: Fenghua Yu <fenghua.yu@intel.com>
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms and conditions of the GNU General Public License,
10  * version 2, as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope it will be useful, but WITHOUT
13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
15  * more details.
16  *
17  * More information about RDT be found in the Intel (R) x86 Architecture
18  * Software Developer Manual.
19  */
20 
21 #define pr_fmt(fmt)	KBUILD_MODNAME ": " fmt
22 
23 #include <linux/cacheinfo.h>
24 #include <linux/cpu.h>
25 #include <linux/debugfs.h>
26 #include <linux/fs.h>
27 #include <linux/fs_parser.h>
28 #include <linux/sysfs.h>
29 #include <linux/kernfs.h>
30 #include <linux/seq_buf.h>
31 #include <linux/seq_file.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/task.h>
34 #include <linux/slab.h>
35 #include <linux/task_work.h>
36 #include <linux/user_namespace.h>
37 
38 #include <uapi/linux/magic.h>
39 
40 #include <asm/resctrl_sched.h>
41 #include "internal.h"
42 
43 DEFINE_STATIC_KEY_FALSE(rdt_enable_key);
44 DEFINE_STATIC_KEY_FALSE(rdt_mon_enable_key);
45 DEFINE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
46 static struct kernfs_root *rdt_root;
47 struct rdtgroup rdtgroup_default;
48 LIST_HEAD(rdt_all_groups);
49 
50 /* Kernel fs node for "info" directory under root */
51 static struct kernfs_node *kn_info;
52 
53 /* Kernel fs node for "mon_groups" directory under root */
54 static struct kernfs_node *kn_mongrp;
55 
56 /* Kernel fs node for "mon_data" directory under root */
57 static struct kernfs_node *kn_mondata;
58 
59 static struct seq_buf last_cmd_status;
60 static char last_cmd_status_buf[512];
61 
62 struct dentry *debugfs_resctrl;
63 
64 void rdt_last_cmd_clear(void)
65 {
66 	lockdep_assert_held(&rdtgroup_mutex);
67 	seq_buf_clear(&last_cmd_status);
68 }
69 
70 void rdt_last_cmd_puts(const char *s)
71 {
72 	lockdep_assert_held(&rdtgroup_mutex);
73 	seq_buf_puts(&last_cmd_status, s);
74 }
75 
76 void rdt_last_cmd_printf(const char *fmt, ...)
77 {
78 	va_list ap;
79 
80 	va_start(ap, fmt);
81 	lockdep_assert_held(&rdtgroup_mutex);
82 	seq_buf_vprintf(&last_cmd_status, fmt, ap);
83 	va_end(ap);
84 }
85 
86 /*
87  * Trivial allocator for CLOSIDs. Since h/w only supports a small number,
88  * we can keep a bitmap of free CLOSIDs in a single integer.
89  *
90  * Using a global CLOSID across all resources has some advantages and
91  * some drawbacks:
92  * + We can simply set "current->closid" to assign a task to a resource
93  *   group.
94  * + Context switch code can avoid extra memory references deciding which
95  *   CLOSID to load into the PQR_ASSOC MSR
96  * - We give up some options in configuring resource groups across multi-socket
97  *   systems.
98  * - Our choices on how to configure each resource become progressively more
99  *   limited as the number of resources grows.
100  */
101 static int closid_free_map;
102 static int closid_free_map_len;
103 
104 int closids_supported(void)
105 {
106 	return closid_free_map_len;
107 }
108 
109 static void closid_init(void)
110 {
111 	struct rdt_resource *r;
112 	int rdt_min_closid = 32;
113 
114 	/* Compute rdt_min_closid across all resources */
115 	for_each_alloc_enabled_rdt_resource(r)
116 		rdt_min_closid = min(rdt_min_closid, r->num_closid);
117 
118 	closid_free_map = BIT_MASK(rdt_min_closid) - 1;
119 
120 	/* CLOSID 0 is always reserved for the default group */
121 	closid_free_map &= ~1;
122 	closid_free_map_len = rdt_min_closid;
123 }
124 
125 static int closid_alloc(void)
126 {
127 	u32 closid = ffs(closid_free_map);
128 
129 	if (closid == 0)
130 		return -ENOSPC;
131 	closid--;
132 	closid_free_map &= ~(1 << closid);
133 
134 	return closid;
135 }
136 
137 void closid_free(int closid)
138 {
139 	closid_free_map |= 1 << closid;
140 }
141 
142 /**
143  * closid_allocated - test if provided closid is in use
144  * @closid: closid to be tested
145  *
146  * Return: true if @closid is currently associated with a resource group,
147  * false if @closid is free
148  */
149 static bool closid_allocated(unsigned int closid)
150 {
151 	return (closid_free_map & (1 << closid)) == 0;
152 }
153 
154 /**
155  * rdtgroup_mode_by_closid - Return mode of resource group with closid
156  * @closid: closid if the resource group
157  *
158  * Each resource group is associated with a @closid. Here the mode
159  * of a resource group can be queried by searching for it using its closid.
160  *
161  * Return: mode as &enum rdtgrp_mode of resource group with closid @closid
162  */
163 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid)
164 {
165 	struct rdtgroup *rdtgrp;
166 
167 	list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) {
168 		if (rdtgrp->closid == closid)
169 			return rdtgrp->mode;
170 	}
171 
172 	return RDT_NUM_MODES;
173 }
174 
175 static const char * const rdt_mode_str[] = {
176 	[RDT_MODE_SHAREABLE]		= "shareable",
177 	[RDT_MODE_EXCLUSIVE]		= "exclusive",
178 	[RDT_MODE_PSEUDO_LOCKSETUP]	= "pseudo-locksetup",
179 	[RDT_MODE_PSEUDO_LOCKED]	= "pseudo-locked",
180 };
181 
182 /**
183  * rdtgroup_mode_str - Return the string representation of mode
184  * @mode: the resource group mode as &enum rdtgroup_mode
185  *
186  * Return: string representation of valid mode, "unknown" otherwise
187  */
188 static const char *rdtgroup_mode_str(enum rdtgrp_mode mode)
189 {
190 	if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES)
191 		return "unknown";
192 
193 	return rdt_mode_str[mode];
194 }
195 
196 /* set uid and gid of rdtgroup dirs and files to that of the creator */
197 static int rdtgroup_kn_set_ugid(struct kernfs_node *kn)
198 {
199 	struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
200 				.ia_uid = current_fsuid(),
201 				.ia_gid = current_fsgid(), };
202 
203 	if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
204 	    gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
205 		return 0;
206 
207 	return kernfs_setattr(kn, &iattr);
208 }
209 
210 static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft)
211 {
212 	struct kernfs_node *kn;
213 	int ret;
214 
215 	kn = __kernfs_create_file(parent_kn, rft->name, rft->mode,
216 				  GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
217 				  0, rft->kf_ops, rft, NULL, NULL);
218 	if (IS_ERR(kn))
219 		return PTR_ERR(kn);
220 
221 	ret = rdtgroup_kn_set_ugid(kn);
222 	if (ret) {
223 		kernfs_remove(kn);
224 		return ret;
225 	}
226 
227 	return 0;
228 }
229 
230 static int rdtgroup_seqfile_show(struct seq_file *m, void *arg)
231 {
232 	struct kernfs_open_file *of = m->private;
233 	struct rftype *rft = of->kn->priv;
234 
235 	if (rft->seq_show)
236 		return rft->seq_show(of, m, arg);
237 	return 0;
238 }
239 
240 static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf,
241 				   size_t nbytes, loff_t off)
242 {
243 	struct rftype *rft = of->kn->priv;
244 
245 	if (rft->write)
246 		return rft->write(of, buf, nbytes, off);
247 
248 	return -EINVAL;
249 }
250 
251 static struct kernfs_ops rdtgroup_kf_single_ops = {
252 	.atomic_write_len	= PAGE_SIZE,
253 	.write			= rdtgroup_file_write,
254 	.seq_show		= rdtgroup_seqfile_show,
255 };
256 
257 static struct kernfs_ops kf_mondata_ops = {
258 	.atomic_write_len	= PAGE_SIZE,
259 	.seq_show		= rdtgroup_mondata_show,
260 };
261 
262 static bool is_cpu_list(struct kernfs_open_file *of)
263 {
264 	struct rftype *rft = of->kn->priv;
265 
266 	return rft->flags & RFTYPE_FLAGS_CPUS_LIST;
267 }
268 
269 static int rdtgroup_cpus_show(struct kernfs_open_file *of,
270 			      struct seq_file *s, void *v)
271 {
272 	struct rdtgroup *rdtgrp;
273 	struct cpumask *mask;
274 	int ret = 0;
275 
276 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
277 
278 	if (rdtgrp) {
279 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
280 			if (!rdtgrp->plr->d) {
281 				rdt_last_cmd_clear();
282 				rdt_last_cmd_puts("Cache domain offline\n");
283 				ret = -ENODEV;
284 			} else {
285 				mask = &rdtgrp->plr->d->cpu_mask;
286 				seq_printf(s, is_cpu_list(of) ?
287 					   "%*pbl\n" : "%*pb\n",
288 					   cpumask_pr_args(mask));
289 			}
290 		} else {
291 			seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n",
292 				   cpumask_pr_args(&rdtgrp->cpu_mask));
293 		}
294 	} else {
295 		ret = -ENOENT;
296 	}
297 	rdtgroup_kn_unlock(of->kn);
298 
299 	return ret;
300 }
301 
302 /*
303  * This is safe against resctrl_sched_in() called from __switch_to()
304  * because __switch_to() is executed with interrupts disabled. A local call
305  * from update_closid_rmid() is proteced against __switch_to() because
306  * preemption is disabled.
307  */
308 static void update_cpu_closid_rmid(void *info)
309 {
310 	struct rdtgroup *r = info;
311 
312 	if (r) {
313 		this_cpu_write(pqr_state.default_closid, r->closid);
314 		this_cpu_write(pqr_state.default_rmid, r->mon.rmid);
315 	}
316 
317 	/*
318 	 * We cannot unconditionally write the MSR because the current
319 	 * executing task might have its own closid selected. Just reuse
320 	 * the context switch code.
321 	 */
322 	resctrl_sched_in();
323 }
324 
325 /*
326  * Update the PGR_ASSOC MSR on all cpus in @cpu_mask,
327  *
328  * Per task closids/rmids must have been set up before calling this function.
329  */
330 static void
331 update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r)
332 {
333 	int cpu = get_cpu();
334 
335 	if (cpumask_test_cpu(cpu, cpu_mask))
336 		update_cpu_closid_rmid(r);
337 	smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1);
338 	put_cpu();
339 }
340 
341 static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
342 			  cpumask_var_t tmpmask)
343 {
344 	struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp;
345 	struct list_head *head;
346 
347 	/* Check whether cpus belong to parent ctrl group */
348 	cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask);
349 	if (cpumask_weight(tmpmask)) {
350 		rdt_last_cmd_puts("Can only add CPUs to mongroup that belong to parent\n");
351 		return -EINVAL;
352 	}
353 
354 	/* Check whether cpus are dropped from this group */
355 	cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
356 	if (cpumask_weight(tmpmask)) {
357 		/* Give any dropped cpus to parent rdtgroup */
358 		cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask);
359 		update_closid_rmid(tmpmask, prgrp);
360 	}
361 
362 	/*
363 	 * If we added cpus, remove them from previous group that owned them
364 	 * and update per-cpu rmid
365 	 */
366 	cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
367 	if (cpumask_weight(tmpmask)) {
368 		head = &prgrp->mon.crdtgrp_list;
369 		list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
370 			if (crgrp == rdtgrp)
371 				continue;
372 			cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask,
373 				       tmpmask);
374 		}
375 		update_closid_rmid(tmpmask, rdtgrp);
376 	}
377 
378 	/* Done pushing/pulling - update this group with new mask */
379 	cpumask_copy(&rdtgrp->cpu_mask, newmask);
380 
381 	return 0;
382 }
383 
384 static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m)
385 {
386 	struct rdtgroup *crgrp;
387 
388 	cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m);
389 	/* update the child mon group masks as well*/
390 	list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list)
391 		cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask);
392 }
393 
394 static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
395 			   cpumask_var_t tmpmask, cpumask_var_t tmpmask1)
396 {
397 	struct rdtgroup *r, *crgrp;
398 	struct list_head *head;
399 
400 	/* Check whether cpus are dropped from this group */
401 	cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
402 	if (cpumask_weight(tmpmask)) {
403 		/* Can't drop from default group */
404 		if (rdtgrp == &rdtgroup_default) {
405 			rdt_last_cmd_puts("Can't drop CPUs from default group\n");
406 			return -EINVAL;
407 		}
408 
409 		/* Give any dropped cpus to rdtgroup_default */
410 		cpumask_or(&rdtgroup_default.cpu_mask,
411 			   &rdtgroup_default.cpu_mask, tmpmask);
412 		update_closid_rmid(tmpmask, &rdtgroup_default);
413 	}
414 
415 	/*
416 	 * If we added cpus, remove them from previous group and
417 	 * the prev group's child groups that owned them
418 	 * and update per-cpu closid/rmid.
419 	 */
420 	cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
421 	if (cpumask_weight(tmpmask)) {
422 		list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) {
423 			if (r == rdtgrp)
424 				continue;
425 			cpumask_and(tmpmask1, &r->cpu_mask, tmpmask);
426 			if (cpumask_weight(tmpmask1))
427 				cpumask_rdtgrp_clear(r, tmpmask1);
428 		}
429 		update_closid_rmid(tmpmask, rdtgrp);
430 	}
431 
432 	/* Done pushing/pulling - update this group with new mask */
433 	cpumask_copy(&rdtgrp->cpu_mask, newmask);
434 
435 	/*
436 	 * Clear child mon group masks since there is a new parent mask
437 	 * now and update the rmid for the cpus the child lost.
438 	 */
439 	head = &rdtgrp->mon.crdtgrp_list;
440 	list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
441 		cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask);
442 		update_closid_rmid(tmpmask, rdtgrp);
443 		cpumask_clear(&crgrp->cpu_mask);
444 	}
445 
446 	return 0;
447 }
448 
449 static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of,
450 				   char *buf, size_t nbytes, loff_t off)
451 {
452 	cpumask_var_t tmpmask, newmask, tmpmask1;
453 	struct rdtgroup *rdtgrp;
454 	int ret;
455 
456 	if (!buf)
457 		return -EINVAL;
458 
459 	if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
460 		return -ENOMEM;
461 	if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) {
462 		free_cpumask_var(tmpmask);
463 		return -ENOMEM;
464 	}
465 	if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) {
466 		free_cpumask_var(tmpmask);
467 		free_cpumask_var(newmask);
468 		return -ENOMEM;
469 	}
470 
471 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
472 	rdt_last_cmd_clear();
473 	if (!rdtgrp) {
474 		ret = -ENOENT;
475 		rdt_last_cmd_puts("Directory was removed\n");
476 		goto unlock;
477 	}
478 
479 	if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
480 	    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
481 		ret = -EINVAL;
482 		rdt_last_cmd_puts("Pseudo-locking in progress\n");
483 		goto unlock;
484 	}
485 
486 	if (is_cpu_list(of))
487 		ret = cpulist_parse(buf, newmask);
488 	else
489 		ret = cpumask_parse(buf, newmask);
490 
491 	if (ret) {
492 		rdt_last_cmd_puts("Bad CPU list/mask\n");
493 		goto unlock;
494 	}
495 
496 	/* check that user didn't specify any offline cpus */
497 	cpumask_andnot(tmpmask, newmask, cpu_online_mask);
498 	if (cpumask_weight(tmpmask)) {
499 		ret = -EINVAL;
500 		rdt_last_cmd_puts("Can only assign online CPUs\n");
501 		goto unlock;
502 	}
503 
504 	if (rdtgrp->type == RDTCTRL_GROUP)
505 		ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1);
506 	else if (rdtgrp->type == RDTMON_GROUP)
507 		ret = cpus_mon_write(rdtgrp, newmask, tmpmask);
508 	else
509 		ret = -EINVAL;
510 
511 unlock:
512 	rdtgroup_kn_unlock(of->kn);
513 	free_cpumask_var(tmpmask);
514 	free_cpumask_var(newmask);
515 	free_cpumask_var(tmpmask1);
516 
517 	return ret ?: nbytes;
518 }
519 
520 struct task_move_callback {
521 	struct callback_head	work;
522 	struct rdtgroup		*rdtgrp;
523 };
524 
525 static void move_myself(struct callback_head *head)
526 {
527 	struct task_move_callback *callback;
528 	struct rdtgroup *rdtgrp;
529 
530 	callback = container_of(head, struct task_move_callback, work);
531 	rdtgrp = callback->rdtgrp;
532 
533 	/*
534 	 * If resource group was deleted before this task work callback
535 	 * was invoked, then assign the task to root group and free the
536 	 * resource group.
537 	 */
538 	if (atomic_dec_and_test(&rdtgrp->waitcount) &&
539 	    (rdtgrp->flags & RDT_DELETED)) {
540 		current->closid = 0;
541 		current->rmid = 0;
542 		kfree(rdtgrp);
543 	}
544 
545 	preempt_disable();
546 	/* update PQR_ASSOC MSR to make resource group go into effect */
547 	resctrl_sched_in();
548 	preempt_enable();
549 
550 	kfree(callback);
551 }
552 
553 static int __rdtgroup_move_task(struct task_struct *tsk,
554 				struct rdtgroup *rdtgrp)
555 {
556 	struct task_move_callback *callback;
557 	int ret;
558 
559 	callback = kzalloc(sizeof(*callback), GFP_KERNEL);
560 	if (!callback)
561 		return -ENOMEM;
562 	callback->work.func = move_myself;
563 	callback->rdtgrp = rdtgrp;
564 
565 	/*
566 	 * Take a refcount, so rdtgrp cannot be freed before the
567 	 * callback has been invoked.
568 	 */
569 	atomic_inc(&rdtgrp->waitcount);
570 	ret = task_work_add(tsk, &callback->work, true);
571 	if (ret) {
572 		/*
573 		 * Task is exiting. Drop the refcount and free the callback.
574 		 * No need to check the refcount as the group cannot be
575 		 * deleted before the write function unlocks rdtgroup_mutex.
576 		 */
577 		atomic_dec(&rdtgrp->waitcount);
578 		kfree(callback);
579 		rdt_last_cmd_puts("Task exited\n");
580 	} else {
581 		/*
582 		 * For ctrl_mon groups move both closid and rmid.
583 		 * For monitor groups, can move the tasks only from
584 		 * their parent CTRL group.
585 		 */
586 		if (rdtgrp->type == RDTCTRL_GROUP) {
587 			tsk->closid = rdtgrp->closid;
588 			tsk->rmid = rdtgrp->mon.rmid;
589 		} else if (rdtgrp->type == RDTMON_GROUP) {
590 			if (rdtgrp->mon.parent->closid == tsk->closid) {
591 				tsk->rmid = rdtgrp->mon.rmid;
592 			} else {
593 				rdt_last_cmd_puts("Can't move task to different control group\n");
594 				ret = -EINVAL;
595 			}
596 		}
597 	}
598 	return ret;
599 }
600 
601 /**
602  * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group
603  * @r: Resource group
604  *
605  * Return: 1 if tasks have been assigned to @r, 0 otherwise
606  */
607 int rdtgroup_tasks_assigned(struct rdtgroup *r)
608 {
609 	struct task_struct *p, *t;
610 	int ret = 0;
611 
612 	lockdep_assert_held(&rdtgroup_mutex);
613 
614 	rcu_read_lock();
615 	for_each_process_thread(p, t) {
616 		if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
617 		    (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) {
618 			ret = 1;
619 			break;
620 		}
621 	}
622 	rcu_read_unlock();
623 
624 	return ret;
625 }
626 
627 static int rdtgroup_task_write_permission(struct task_struct *task,
628 					  struct kernfs_open_file *of)
629 {
630 	const struct cred *tcred = get_task_cred(task);
631 	const struct cred *cred = current_cred();
632 	int ret = 0;
633 
634 	/*
635 	 * Even if we're attaching all tasks in the thread group, we only
636 	 * need to check permissions on one of them.
637 	 */
638 	if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
639 	    !uid_eq(cred->euid, tcred->uid) &&
640 	    !uid_eq(cred->euid, tcred->suid)) {
641 		rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
642 		ret = -EPERM;
643 	}
644 
645 	put_cred(tcred);
646 	return ret;
647 }
648 
649 static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
650 			      struct kernfs_open_file *of)
651 {
652 	struct task_struct *tsk;
653 	int ret;
654 
655 	rcu_read_lock();
656 	if (pid) {
657 		tsk = find_task_by_vpid(pid);
658 		if (!tsk) {
659 			rcu_read_unlock();
660 			rdt_last_cmd_printf("No task %d\n", pid);
661 			return -ESRCH;
662 		}
663 	} else {
664 		tsk = current;
665 	}
666 
667 	get_task_struct(tsk);
668 	rcu_read_unlock();
669 
670 	ret = rdtgroup_task_write_permission(tsk, of);
671 	if (!ret)
672 		ret = __rdtgroup_move_task(tsk, rdtgrp);
673 
674 	put_task_struct(tsk);
675 	return ret;
676 }
677 
678 static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
679 				    char *buf, size_t nbytes, loff_t off)
680 {
681 	struct rdtgroup *rdtgrp;
682 	int ret = 0;
683 	pid_t pid;
684 
685 	if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
686 		return -EINVAL;
687 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
688 	if (!rdtgrp) {
689 		rdtgroup_kn_unlock(of->kn);
690 		return -ENOENT;
691 	}
692 	rdt_last_cmd_clear();
693 
694 	if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED ||
695 	    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
696 		ret = -EINVAL;
697 		rdt_last_cmd_puts("Pseudo-locking in progress\n");
698 		goto unlock;
699 	}
700 
701 	ret = rdtgroup_move_task(pid, rdtgrp, of);
702 
703 unlock:
704 	rdtgroup_kn_unlock(of->kn);
705 
706 	return ret ?: nbytes;
707 }
708 
709 static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
710 {
711 	struct task_struct *p, *t;
712 
713 	rcu_read_lock();
714 	for_each_process_thread(p, t) {
715 		if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
716 		    (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid))
717 			seq_printf(s, "%d\n", t->pid);
718 	}
719 	rcu_read_unlock();
720 }
721 
722 static int rdtgroup_tasks_show(struct kernfs_open_file *of,
723 			       struct seq_file *s, void *v)
724 {
725 	struct rdtgroup *rdtgrp;
726 	int ret = 0;
727 
728 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
729 	if (rdtgrp)
730 		show_rdt_tasks(rdtgrp, s);
731 	else
732 		ret = -ENOENT;
733 	rdtgroup_kn_unlock(of->kn);
734 
735 	return ret;
736 }
737 
738 static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
739 				    struct seq_file *seq, void *v)
740 {
741 	int len;
742 
743 	mutex_lock(&rdtgroup_mutex);
744 	len = seq_buf_used(&last_cmd_status);
745 	if (len)
746 		seq_printf(seq, "%.*s", len, last_cmd_status_buf);
747 	else
748 		seq_puts(seq, "ok\n");
749 	mutex_unlock(&rdtgroup_mutex);
750 	return 0;
751 }
752 
753 static int rdt_num_closids_show(struct kernfs_open_file *of,
754 				struct seq_file *seq, void *v)
755 {
756 	struct rdt_resource *r = of->kn->parent->priv;
757 
758 	seq_printf(seq, "%d\n", r->num_closid);
759 	return 0;
760 }
761 
762 static int rdt_default_ctrl_show(struct kernfs_open_file *of,
763 			     struct seq_file *seq, void *v)
764 {
765 	struct rdt_resource *r = of->kn->parent->priv;
766 
767 	seq_printf(seq, "%x\n", r->default_ctrl);
768 	return 0;
769 }
770 
771 static int rdt_min_cbm_bits_show(struct kernfs_open_file *of,
772 			     struct seq_file *seq, void *v)
773 {
774 	struct rdt_resource *r = of->kn->parent->priv;
775 
776 	seq_printf(seq, "%u\n", r->cache.min_cbm_bits);
777 	return 0;
778 }
779 
780 static int rdt_shareable_bits_show(struct kernfs_open_file *of,
781 				   struct seq_file *seq, void *v)
782 {
783 	struct rdt_resource *r = of->kn->parent->priv;
784 
785 	seq_printf(seq, "%x\n", r->cache.shareable_bits);
786 	return 0;
787 }
788 
789 /**
790  * rdt_bit_usage_show - Display current usage of resources
791  *
792  * A domain is a shared resource that can now be allocated differently. Here
793  * we display the current regions of the domain as an annotated bitmask.
794  * For each domain of this resource its allocation bitmask
795  * is annotated as below to indicate the current usage of the corresponding bit:
796  *   0 - currently unused
797  *   X - currently available for sharing and used by software and hardware
798  *   H - currently used by hardware only but available for software use
799  *   S - currently used and shareable by software only
800  *   E - currently used exclusively by one resource group
801  *   P - currently pseudo-locked by one resource group
802  */
803 static int rdt_bit_usage_show(struct kernfs_open_file *of,
804 			      struct seq_file *seq, void *v)
805 {
806 	struct rdt_resource *r = of->kn->parent->priv;
807 	u32 sw_shareable = 0, hw_shareable = 0;
808 	u32 exclusive = 0, pseudo_locked = 0;
809 	struct rdt_domain *dom;
810 	int i, hwb, swb, excl, psl;
811 	enum rdtgrp_mode mode;
812 	bool sep = false;
813 	u32 *ctrl;
814 
815 	mutex_lock(&rdtgroup_mutex);
816 	hw_shareable = r->cache.shareable_bits;
817 	list_for_each_entry(dom, &r->domains, list) {
818 		if (sep)
819 			seq_putc(seq, ';');
820 		ctrl = dom->ctrl_val;
821 		sw_shareable = 0;
822 		exclusive = 0;
823 		seq_printf(seq, "%d=", dom->id);
824 		for (i = 0; i < closids_supported(); i++, ctrl++) {
825 			if (!closid_allocated(i))
826 				continue;
827 			mode = rdtgroup_mode_by_closid(i);
828 			switch (mode) {
829 			case RDT_MODE_SHAREABLE:
830 				sw_shareable |= *ctrl;
831 				break;
832 			case RDT_MODE_EXCLUSIVE:
833 				exclusive |= *ctrl;
834 				break;
835 			case RDT_MODE_PSEUDO_LOCKSETUP:
836 			/*
837 			 * RDT_MODE_PSEUDO_LOCKSETUP is possible
838 			 * here but not included since the CBM
839 			 * associated with this CLOSID in this mode
840 			 * is not initialized and no task or cpu can be
841 			 * assigned this CLOSID.
842 			 */
843 				break;
844 			case RDT_MODE_PSEUDO_LOCKED:
845 			case RDT_NUM_MODES:
846 				WARN(1,
847 				     "invalid mode for closid %d\n", i);
848 				break;
849 			}
850 		}
851 		for (i = r->cache.cbm_len - 1; i >= 0; i--) {
852 			pseudo_locked = dom->plr ? dom->plr->cbm : 0;
853 			hwb = test_bit(i, (unsigned long *)&hw_shareable);
854 			swb = test_bit(i, (unsigned long *)&sw_shareable);
855 			excl = test_bit(i, (unsigned long *)&exclusive);
856 			psl = test_bit(i, (unsigned long *)&pseudo_locked);
857 			if (hwb && swb)
858 				seq_putc(seq, 'X');
859 			else if (hwb && !swb)
860 				seq_putc(seq, 'H');
861 			else if (!hwb && swb)
862 				seq_putc(seq, 'S');
863 			else if (excl)
864 				seq_putc(seq, 'E');
865 			else if (psl)
866 				seq_putc(seq, 'P');
867 			else /* Unused bits remain */
868 				seq_putc(seq, '0');
869 		}
870 		sep = true;
871 	}
872 	seq_putc(seq, '\n');
873 	mutex_unlock(&rdtgroup_mutex);
874 	return 0;
875 }
876 
877 static int rdt_min_bw_show(struct kernfs_open_file *of,
878 			     struct seq_file *seq, void *v)
879 {
880 	struct rdt_resource *r = of->kn->parent->priv;
881 
882 	seq_printf(seq, "%u\n", r->membw.min_bw);
883 	return 0;
884 }
885 
886 static int rdt_num_rmids_show(struct kernfs_open_file *of,
887 			      struct seq_file *seq, void *v)
888 {
889 	struct rdt_resource *r = of->kn->parent->priv;
890 
891 	seq_printf(seq, "%d\n", r->num_rmid);
892 
893 	return 0;
894 }
895 
896 static int rdt_mon_features_show(struct kernfs_open_file *of,
897 				 struct seq_file *seq, void *v)
898 {
899 	struct rdt_resource *r = of->kn->parent->priv;
900 	struct mon_evt *mevt;
901 
902 	list_for_each_entry(mevt, &r->evt_list, list)
903 		seq_printf(seq, "%s\n", mevt->name);
904 
905 	return 0;
906 }
907 
908 static int rdt_bw_gran_show(struct kernfs_open_file *of,
909 			     struct seq_file *seq, void *v)
910 {
911 	struct rdt_resource *r = of->kn->parent->priv;
912 
913 	seq_printf(seq, "%u\n", r->membw.bw_gran);
914 	return 0;
915 }
916 
917 static int rdt_delay_linear_show(struct kernfs_open_file *of,
918 			     struct seq_file *seq, void *v)
919 {
920 	struct rdt_resource *r = of->kn->parent->priv;
921 
922 	seq_printf(seq, "%u\n", r->membw.delay_linear);
923 	return 0;
924 }
925 
926 static int max_threshold_occ_show(struct kernfs_open_file *of,
927 				  struct seq_file *seq, void *v)
928 {
929 	struct rdt_resource *r = of->kn->parent->priv;
930 
931 	seq_printf(seq, "%u\n", resctrl_cqm_threshold * r->mon_scale);
932 
933 	return 0;
934 }
935 
936 static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
937 				       char *buf, size_t nbytes, loff_t off)
938 {
939 	struct rdt_resource *r = of->kn->parent->priv;
940 	unsigned int bytes;
941 	int ret;
942 
943 	ret = kstrtouint(buf, 0, &bytes);
944 	if (ret)
945 		return ret;
946 
947 	if (bytes > (boot_cpu_data.x86_cache_size * 1024))
948 		return -EINVAL;
949 
950 	resctrl_cqm_threshold = bytes / r->mon_scale;
951 
952 	return nbytes;
953 }
954 
955 /*
956  * rdtgroup_mode_show - Display mode of this resource group
957  */
958 static int rdtgroup_mode_show(struct kernfs_open_file *of,
959 			      struct seq_file *s, void *v)
960 {
961 	struct rdtgroup *rdtgrp;
962 
963 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
964 	if (!rdtgrp) {
965 		rdtgroup_kn_unlock(of->kn);
966 		return -ENOENT;
967 	}
968 
969 	seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode));
970 
971 	rdtgroup_kn_unlock(of->kn);
972 	return 0;
973 }
974 
975 /**
976  * rdt_cdp_peer_get - Retrieve CDP peer if it exists
977  * @r: RDT resource to which RDT domain @d belongs
978  * @d: Cache instance for which a CDP peer is requested
979  * @r_cdp: RDT resource that shares hardware with @r (RDT resource peer)
980  *         Used to return the result.
981  * @d_cdp: RDT domain that shares hardware with @d (RDT domain peer)
982  *         Used to return the result.
983  *
984  * RDT resources are managed independently and by extension the RDT domains
985  * (RDT resource instances) are managed independently also. The Code and
986  * Data Prioritization (CDP) RDT resources, while managed independently,
987  * could refer to the same underlying hardware. For example,
988  * RDT_RESOURCE_L2CODE and RDT_RESOURCE_L2DATA both refer to the L2 cache.
989  *
990  * When provided with an RDT resource @r and an instance of that RDT
991  * resource @d rdt_cdp_peer_get() will return if there is a peer RDT
992  * resource and the exact instance that shares the same hardware.
993  *
994  * Return: 0 if a CDP peer was found, <0 on error or if no CDP peer exists.
995  *         If a CDP peer was found, @r_cdp will point to the peer RDT resource
996  *         and @d_cdp will point to the peer RDT domain.
997  */
998 static int rdt_cdp_peer_get(struct rdt_resource *r, struct rdt_domain *d,
999 			    struct rdt_resource **r_cdp,
1000 			    struct rdt_domain **d_cdp)
1001 {
1002 	struct rdt_resource *_r_cdp = NULL;
1003 	struct rdt_domain *_d_cdp = NULL;
1004 	int ret = 0;
1005 
1006 	switch (r->rid) {
1007 	case RDT_RESOURCE_L3DATA:
1008 		_r_cdp = &rdt_resources_all[RDT_RESOURCE_L3CODE];
1009 		break;
1010 	case RDT_RESOURCE_L3CODE:
1011 		_r_cdp =  &rdt_resources_all[RDT_RESOURCE_L3DATA];
1012 		break;
1013 	case RDT_RESOURCE_L2DATA:
1014 		_r_cdp =  &rdt_resources_all[RDT_RESOURCE_L2CODE];
1015 		break;
1016 	case RDT_RESOURCE_L2CODE:
1017 		_r_cdp =  &rdt_resources_all[RDT_RESOURCE_L2DATA];
1018 		break;
1019 	default:
1020 		ret = -ENOENT;
1021 		goto out;
1022 	}
1023 
1024 	/*
1025 	 * When a new CPU comes online and CDP is enabled then the new
1026 	 * RDT domains (if any) associated with both CDP RDT resources
1027 	 * are added in the same CPU online routine while the
1028 	 * rdtgroup_mutex is held. It should thus not happen for one
1029 	 * RDT domain to exist and be associated with its RDT CDP
1030 	 * resource but there is no RDT domain associated with the
1031 	 * peer RDT CDP resource. Hence the WARN.
1032 	 */
1033 	_d_cdp = rdt_find_domain(_r_cdp, d->id, NULL);
1034 	if (WARN_ON(IS_ERR_OR_NULL(_d_cdp))) {
1035 		_r_cdp = NULL;
1036 		ret = -EINVAL;
1037 	}
1038 
1039 out:
1040 	*r_cdp = _r_cdp;
1041 	*d_cdp = _d_cdp;
1042 
1043 	return ret;
1044 }
1045 
1046 /**
1047  * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other
1048  * @r: Resource to which domain instance @d belongs.
1049  * @d: The domain instance for which @closid is being tested.
1050  * @cbm: Capacity bitmask being tested.
1051  * @closid: Intended closid for @cbm.
1052  * @exclusive: Only check if overlaps with exclusive resource groups
1053  *
1054  * Checks if provided @cbm intended to be used for @closid on domain
1055  * @d overlaps with any other closids or other hardware usage associated
1056  * with this domain. If @exclusive is true then only overlaps with
1057  * resource groups in exclusive mode will be considered. If @exclusive
1058  * is false then overlaps with any resource group or hardware entities
1059  * will be considered.
1060  *
1061  * @cbm is unsigned long, even if only 32 bits are used, to make the
1062  * bitmap functions work correctly.
1063  *
1064  * Return: false if CBM does not overlap, true if it does.
1065  */
1066 static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1067 				    unsigned long cbm, int closid, bool exclusive)
1068 {
1069 	enum rdtgrp_mode mode;
1070 	unsigned long ctrl_b;
1071 	u32 *ctrl;
1072 	int i;
1073 
1074 	/* Check for any overlap with regions used by hardware directly */
1075 	if (!exclusive) {
1076 		ctrl_b = r->cache.shareable_bits;
1077 		if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len))
1078 			return true;
1079 	}
1080 
1081 	/* Check for overlap with other resource groups */
1082 	ctrl = d->ctrl_val;
1083 	for (i = 0; i < closids_supported(); i++, ctrl++) {
1084 		ctrl_b = *ctrl;
1085 		mode = rdtgroup_mode_by_closid(i);
1086 		if (closid_allocated(i) && i != closid &&
1087 		    mode != RDT_MODE_PSEUDO_LOCKSETUP) {
1088 			if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) {
1089 				if (exclusive) {
1090 					if (mode == RDT_MODE_EXCLUSIVE)
1091 						return true;
1092 					continue;
1093 				}
1094 				return true;
1095 			}
1096 		}
1097 	}
1098 
1099 	return false;
1100 }
1101 
1102 /**
1103  * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware
1104  * @r: Resource to which domain instance @d belongs.
1105  * @d: The domain instance for which @closid is being tested.
1106  * @cbm: Capacity bitmask being tested.
1107  * @closid: Intended closid for @cbm.
1108  * @exclusive: Only check if overlaps with exclusive resource groups
1109  *
1110  * Resources that can be allocated using a CBM can use the CBM to control
1111  * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test
1112  * for overlap. Overlap test is not limited to the specific resource for
1113  * which the CBM is intended though - when dealing with CDP resources that
1114  * share the underlying hardware the overlap check should be performed on
1115  * the CDP resource sharing the hardware also.
1116  *
1117  * Refer to description of __rdtgroup_cbm_overlaps() for the details of the
1118  * overlap test.
1119  *
1120  * Return: true if CBM overlap detected, false if there is no overlap
1121  */
1122 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
1123 			   unsigned long cbm, int closid, bool exclusive)
1124 {
1125 	struct rdt_resource *r_cdp;
1126 	struct rdt_domain *d_cdp;
1127 
1128 	if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, exclusive))
1129 		return true;
1130 
1131 	if (rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp) < 0)
1132 		return false;
1133 
1134 	return  __rdtgroup_cbm_overlaps(r_cdp, d_cdp, cbm, closid, exclusive);
1135 }
1136 
1137 /**
1138  * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive
1139  *
1140  * An exclusive resource group implies that there should be no sharing of
1141  * its allocated resources. At the time this group is considered to be
1142  * exclusive this test can determine if its current schemata supports this
1143  * setting by testing for overlap with all other resource groups.
1144  *
1145  * Return: true if resource group can be exclusive, false if there is overlap
1146  * with allocations of other resource groups and thus this resource group
1147  * cannot be exclusive.
1148  */
1149 static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp)
1150 {
1151 	int closid = rdtgrp->closid;
1152 	struct rdt_resource *r;
1153 	bool has_cache = false;
1154 	struct rdt_domain *d;
1155 
1156 	for_each_alloc_enabled_rdt_resource(r) {
1157 		if (r->rid == RDT_RESOURCE_MBA)
1158 			continue;
1159 		has_cache = true;
1160 		list_for_each_entry(d, &r->domains, list) {
1161 			if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid],
1162 						  rdtgrp->closid, false)) {
1163 				rdt_last_cmd_puts("Schemata overlaps\n");
1164 				return false;
1165 			}
1166 		}
1167 	}
1168 
1169 	if (!has_cache) {
1170 		rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n");
1171 		return false;
1172 	}
1173 
1174 	return true;
1175 }
1176 
1177 /**
1178  * rdtgroup_mode_write - Modify the resource group's mode
1179  *
1180  */
1181 static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of,
1182 				   char *buf, size_t nbytes, loff_t off)
1183 {
1184 	struct rdtgroup *rdtgrp;
1185 	enum rdtgrp_mode mode;
1186 	int ret = 0;
1187 
1188 	/* Valid input requires a trailing newline */
1189 	if (nbytes == 0 || buf[nbytes - 1] != '\n')
1190 		return -EINVAL;
1191 	buf[nbytes - 1] = '\0';
1192 
1193 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
1194 	if (!rdtgrp) {
1195 		rdtgroup_kn_unlock(of->kn);
1196 		return -ENOENT;
1197 	}
1198 
1199 	rdt_last_cmd_clear();
1200 
1201 	mode = rdtgrp->mode;
1202 
1203 	if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) ||
1204 	    (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) ||
1205 	    (!strcmp(buf, "pseudo-locksetup") &&
1206 	     mode == RDT_MODE_PSEUDO_LOCKSETUP) ||
1207 	    (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED))
1208 		goto out;
1209 
1210 	if (mode == RDT_MODE_PSEUDO_LOCKED) {
1211 		rdt_last_cmd_puts("Cannot change pseudo-locked group\n");
1212 		ret = -EINVAL;
1213 		goto out;
1214 	}
1215 
1216 	if (!strcmp(buf, "shareable")) {
1217 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1218 			ret = rdtgroup_locksetup_exit(rdtgrp);
1219 			if (ret)
1220 				goto out;
1221 		}
1222 		rdtgrp->mode = RDT_MODE_SHAREABLE;
1223 	} else if (!strcmp(buf, "exclusive")) {
1224 		if (!rdtgroup_mode_test_exclusive(rdtgrp)) {
1225 			ret = -EINVAL;
1226 			goto out;
1227 		}
1228 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1229 			ret = rdtgroup_locksetup_exit(rdtgrp);
1230 			if (ret)
1231 				goto out;
1232 		}
1233 		rdtgrp->mode = RDT_MODE_EXCLUSIVE;
1234 	} else if (!strcmp(buf, "pseudo-locksetup")) {
1235 		ret = rdtgroup_locksetup_enter(rdtgrp);
1236 		if (ret)
1237 			goto out;
1238 		rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP;
1239 	} else {
1240 		rdt_last_cmd_puts("Unknown or unsupported mode\n");
1241 		ret = -EINVAL;
1242 	}
1243 
1244 out:
1245 	rdtgroup_kn_unlock(of->kn);
1246 	return ret ?: nbytes;
1247 }
1248 
1249 /**
1250  * rdtgroup_cbm_to_size - Translate CBM to size in bytes
1251  * @r: RDT resource to which @d belongs.
1252  * @d: RDT domain instance.
1253  * @cbm: bitmask for which the size should be computed.
1254  *
1255  * The bitmask provided associated with the RDT domain instance @d will be
1256  * translated into how many bytes it represents. The size in bytes is
1257  * computed by first dividing the total cache size by the CBM length to
1258  * determine how many bytes each bit in the bitmask represents. The result
1259  * is multiplied with the number of bits set in the bitmask.
1260  *
1261  * @cbm is unsigned long, even if only 32 bits are used to make the
1262  * bitmap functions work correctly.
1263  */
1264 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r,
1265 				  struct rdt_domain *d, unsigned long cbm)
1266 {
1267 	struct cpu_cacheinfo *ci;
1268 	unsigned int size = 0;
1269 	int num_b, i;
1270 
1271 	num_b = bitmap_weight(&cbm, r->cache.cbm_len);
1272 	ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask));
1273 	for (i = 0; i < ci->num_leaves; i++) {
1274 		if (ci->info_list[i].level == r->cache_level) {
1275 			size = ci->info_list[i].size / r->cache.cbm_len * num_b;
1276 			break;
1277 		}
1278 	}
1279 
1280 	return size;
1281 }
1282 
1283 /**
1284  * rdtgroup_size_show - Display size in bytes of allocated regions
1285  *
1286  * The "size" file mirrors the layout of the "schemata" file, printing the
1287  * size in bytes of each region instead of the capacity bitmask.
1288  *
1289  */
1290 static int rdtgroup_size_show(struct kernfs_open_file *of,
1291 			      struct seq_file *s, void *v)
1292 {
1293 	struct rdtgroup *rdtgrp;
1294 	struct rdt_resource *r;
1295 	struct rdt_domain *d;
1296 	unsigned int size;
1297 	int ret = 0;
1298 	bool sep;
1299 	u32 ctrl;
1300 
1301 	rdtgrp = rdtgroup_kn_lock_live(of->kn);
1302 	if (!rdtgrp) {
1303 		rdtgroup_kn_unlock(of->kn);
1304 		return -ENOENT;
1305 	}
1306 
1307 	if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
1308 		if (!rdtgrp->plr->d) {
1309 			rdt_last_cmd_clear();
1310 			rdt_last_cmd_puts("Cache domain offline\n");
1311 			ret = -ENODEV;
1312 		} else {
1313 			seq_printf(s, "%*s:", max_name_width,
1314 				   rdtgrp->plr->r->name);
1315 			size = rdtgroup_cbm_to_size(rdtgrp->plr->r,
1316 						    rdtgrp->plr->d,
1317 						    rdtgrp->plr->cbm);
1318 			seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size);
1319 		}
1320 		goto out;
1321 	}
1322 
1323 	for_each_alloc_enabled_rdt_resource(r) {
1324 		sep = false;
1325 		seq_printf(s, "%*s:", max_name_width, r->name);
1326 		list_for_each_entry(d, &r->domains, list) {
1327 			if (sep)
1328 				seq_putc(s, ';');
1329 			if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) {
1330 				size = 0;
1331 			} else {
1332 				ctrl = (!is_mba_sc(r) ?
1333 						d->ctrl_val[rdtgrp->closid] :
1334 						d->mbps_val[rdtgrp->closid]);
1335 				if (r->rid == RDT_RESOURCE_MBA)
1336 					size = ctrl;
1337 				else
1338 					size = rdtgroup_cbm_to_size(r, d, ctrl);
1339 			}
1340 			seq_printf(s, "%d=%u", d->id, size);
1341 			sep = true;
1342 		}
1343 		seq_putc(s, '\n');
1344 	}
1345 
1346 out:
1347 	rdtgroup_kn_unlock(of->kn);
1348 
1349 	return ret;
1350 }
1351 
1352 /* rdtgroup information files for one cache resource. */
1353 static struct rftype res_common_files[] = {
1354 	{
1355 		.name		= "last_cmd_status",
1356 		.mode		= 0444,
1357 		.kf_ops		= &rdtgroup_kf_single_ops,
1358 		.seq_show	= rdt_last_cmd_status_show,
1359 		.fflags		= RF_TOP_INFO,
1360 	},
1361 	{
1362 		.name		= "num_closids",
1363 		.mode		= 0444,
1364 		.kf_ops		= &rdtgroup_kf_single_ops,
1365 		.seq_show	= rdt_num_closids_show,
1366 		.fflags		= RF_CTRL_INFO,
1367 	},
1368 	{
1369 		.name		= "mon_features",
1370 		.mode		= 0444,
1371 		.kf_ops		= &rdtgroup_kf_single_ops,
1372 		.seq_show	= rdt_mon_features_show,
1373 		.fflags		= RF_MON_INFO,
1374 	},
1375 	{
1376 		.name		= "num_rmids",
1377 		.mode		= 0444,
1378 		.kf_ops		= &rdtgroup_kf_single_ops,
1379 		.seq_show	= rdt_num_rmids_show,
1380 		.fflags		= RF_MON_INFO,
1381 	},
1382 	{
1383 		.name		= "cbm_mask",
1384 		.mode		= 0444,
1385 		.kf_ops		= &rdtgroup_kf_single_ops,
1386 		.seq_show	= rdt_default_ctrl_show,
1387 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_CACHE,
1388 	},
1389 	{
1390 		.name		= "min_cbm_bits",
1391 		.mode		= 0444,
1392 		.kf_ops		= &rdtgroup_kf_single_ops,
1393 		.seq_show	= rdt_min_cbm_bits_show,
1394 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_CACHE,
1395 	},
1396 	{
1397 		.name		= "shareable_bits",
1398 		.mode		= 0444,
1399 		.kf_ops		= &rdtgroup_kf_single_ops,
1400 		.seq_show	= rdt_shareable_bits_show,
1401 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_CACHE,
1402 	},
1403 	{
1404 		.name		= "bit_usage",
1405 		.mode		= 0444,
1406 		.kf_ops		= &rdtgroup_kf_single_ops,
1407 		.seq_show	= rdt_bit_usage_show,
1408 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_CACHE,
1409 	},
1410 	{
1411 		.name		= "min_bandwidth",
1412 		.mode		= 0444,
1413 		.kf_ops		= &rdtgroup_kf_single_ops,
1414 		.seq_show	= rdt_min_bw_show,
1415 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_MB,
1416 	},
1417 	{
1418 		.name		= "bandwidth_gran",
1419 		.mode		= 0444,
1420 		.kf_ops		= &rdtgroup_kf_single_ops,
1421 		.seq_show	= rdt_bw_gran_show,
1422 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_MB,
1423 	},
1424 	{
1425 		.name		= "delay_linear",
1426 		.mode		= 0444,
1427 		.kf_ops		= &rdtgroup_kf_single_ops,
1428 		.seq_show	= rdt_delay_linear_show,
1429 		.fflags		= RF_CTRL_INFO | RFTYPE_RES_MB,
1430 	},
1431 	{
1432 		.name		= "max_threshold_occupancy",
1433 		.mode		= 0644,
1434 		.kf_ops		= &rdtgroup_kf_single_ops,
1435 		.write		= max_threshold_occ_write,
1436 		.seq_show	= max_threshold_occ_show,
1437 		.fflags		= RF_MON_INFO | RFTYPE_RES_CACHE,
1438 	},
1439 	{
1440 		.name		= "cpus",
1441 		.mode		= 0644,
1442 		.kf_ops		= &rdtgroup_kf_single_ops,
1443 		.write		= rdtgroup_cpus_write,
1444 		.seq_show	= rdtgroup_cpus_show,
1445 		.fflags		= RFTYPE_BASE,
1446 	},
1447 	{
1448 		.name		= "cpus_list",
1449 		.mode		= 0644,
1450 		.kf_ops		= &rdtgroup_kf_single_ops,
1451 		.write		= rdtgroup_cpus_write,
1452 		.seq_show	= rdtgroup_cpus_show,
1453 		.flags		= RFTYPE_FLAGS_CPUS_LIST,
1454 		.fflags		= RFTYPE_BASE,
1455 	},
1456 	{
1457 		.name		= "tasks",
1458 		.mode		= 0644,
1459 		.kf_ops		= &rdtgroup_kf_single_ops,
1460 		.write		= rdtgroup_tasks_write,
1461 		.seq_show	= rdtgroup_tasks_show,
1462 		.fflags		= RFTYPE_BASE,
1463 	},
1464 	{
1465 		.name		= "schemata",
1466 		.mode		= 0644,
1467 		.kf_ops		= &rdtgroup_kf_single_ops,
1468 		.write		= rdtgroup_schemata_write,
1469 		.seq_show	= rdtgroup_schemata_show,
1470 		.fflags		= RF_CTRL_BASE,
1471 	},
1472 	{
1473 		.name		= "mode",
1474 		.mode		= 0644,
1475 		.kf_ops		= &rdtgroup_kf_single_ops,
1476 		.write		= rdtgroup_mode_write,
1477 		.seq_show	= rdtgroup_mode_show,
1478 		.fflags		= RF_CTRL_BASE,
1479 	},
1480 	{
1481 		.name		= "size",
1482 		.mode		= 0444,
1483 		.kf_ops		= &rdtgroup_kf_single_ops,
1484 		.seq_show	= rdtgroup_size_show,
1485 		.fflags		= RF_CTRL_BASE,
1486 	},
1487 
1488 };
1489 
1490 static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
1491 {
1492 	struct rftype *rfts, *rft;
1493 	int ret, len;
1494 
1495 	rfts = res_common_files;
1496 	len = ARRAY_SIZE(res_common_files);
1497 
1498 	lockdep_assert_held(&rdtgroup_mutex);
1499 
1500 	for (rft = rfts; rft < rfts + len; rft++) {
1501 		if ((fflags & rft->fflags) == rft->fflags) {
1502 			ret = rdtgroup_add_file(kn, rft);
1503 			if (ret)
1504 				goto error;
1505 		}
1506 	}
1507 
1508 	return 0;
1509 error:
1510 	pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
1511 	while (--rft >= rfts) {
1512 		if ((fflags & rft->fflags) == rft->fflags)
1513 			kernfs_remove_by_name(kn, rft->name);
1514 	}
1515 	return ret;
1516 }
1517 
1518 /**
1519  * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file
1520  * @r: The resource group with which the file is associated.
1521  * @name: Name of the file
1522  *
1523  * The permissions of named resctrl file, directory, or link are modified
1524  * to not allow read, write, or execute by any user.
1525  *
1526  * WARNING: This function is intended to communicate to the user that the
1527  * resctrl file has been locked down - that it is not relevant to the
1528  * particular state the system finds itself in. It should not be relied
1529  * on to protect from user access because after the file's permissions
1530  * are restricted the user can still change the permissions using chmod
1531  * from the command line.
1532  *
1533  * Return: 0 on success, <0 on failure.
1534  */
1535 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name)
1536 {
1537 	struct iattr iattr = {.ia_valid = ATTR_MODE,};
1538 	struct kernfs_node *kn;
1539 	int ret = 0;
1540 
1541 	kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1542 	if (!kn)
1543 		return -ENOENT;
1544 
1545 	switch (kernfs_type(kn)) {
1546 	case KERNFS_DIR:
1547 		iattr.ia_mode = S_IFDIR;
1548 		break;
1549 	case KERNFS_FILE:
1550 		iattr.ia_mode = S_IFREG;
1551 		break;
1552 	case KERNFS_LINK:
1553 		iattr.ia_mode = S_IFLNK;
1554 		break;
1555 	}
1556 
1557 	ret = kernfs_setattr(kn, &iattr);
1558 	kernfs_put(kn);
1559 	return ret;
1560 }
1561 
1562 /**
1563  * rdtgroup_kn_mode_restore - Restore user access to named resctrl file
1564  * @r: The resource group with which the file is associated.
1565  * @name: Name of the file
1566  * @mask: Mask of permissions that should be restored
1567  *
1568  * Restore the permissions of the named file. If @name is a directory the
1569  * permissions of its parent will be used.
1570  *
1571  * Return: 0 on success, <0 on failure.
1572  */
1573 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
1574 			     umode_t mask)
1575 {
1576 	struct iattr iattr = {.ia_valid = ATTR_MODE,};
1577 	struct kernfs_node *kn, *parent;
1578 	struct rftype *rfts, *rft;
1579 	int ret, len;
1580 
1581 	rfts = res_common_files;
1582 	len = ARRAY_SIZE(res_common_files);
1583 
1584 	for (rft = rfts; rft < rfts + len; rft++) {
1585 		if (!strcmp(rft->name, name))
1586 			iattr.ia_mode = rft->mode & mask;
1587 	}
1588 
1589 	kn = kernfs_find_and_get_ns(r->kn, name, NULL);
1590 	if (!kn)
1591 		return -ENOENT;
1592 
1593 	switch (kernfs_type(kn)) {
1594 	case KERNFS_DIR:
1595 		parent = kernfs_get_parent(kn);
1596 		if (parent) {
1597 			iattr.ia_mode |= parent->mode;
1598 			kernfs_put(parent);
1599 		}
1600 		iattr.ia_mode |= S_IFDIR;
1601 		break;
1602 	case KERNFS_FILE:
1603 		iattr.ia_mode |= S_IFREG;
1604 		break;
1605 	case KERNFS_LINK:
1606 		iattr.ia_mode |= S_IFLNK;
1607 		break;
1608 	}
1609 
1610 	ret = kernfs_setattr(kn, &iattr);
1611 	kernfs_put(kn);
1612 	return ret;
1613 }
1614 
1615 static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name,
1616 				      unsigned long fflags)
1617 {
1618 	struct kernfs_node *kn_subdir;
1619 	int ret;
1620 
1621 	kn_subdir = kernfs_create_dir(kn_info, name,
1622 				      kn_info->mode, r);
1623 	if (IS_ERR(kn_subdir))
1624 		return PTR_ERR(kn_subdir);
1625 
1626 	kernfs_get(kn_subdir);
1627 	ret = rdtgroup_kn_set_ugid(kn_subdir);
1628 	if (ret)
1629 		return ret;
1630 
1631 	ret = rdtgroup_add_files(kn_subdir, fflags);
1632 	if (!ret)
1633 		kernfs_activate(kn_subdir);
1634 
1635 	return ret;
1636 }
1637 
1638 static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
1639 {
1640 	struct rdt_resource *r;
1641 	unsigned long fflags;
1642 	char name[32];
1643 	int ret;
1644 
1645 	/* create the directory */
1646 	kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
1647 	if (IS_ERR(kn_info))
1648 		return PTR_ERR(kn_info);
1649 	kernfs_get(kn_info);
1650 
1651 	ret = rdtgroup_add_files(kn_info, RF_TOP_INFO);
1652 	if (ret)
1653 		goto out_destroy;
1654 
1655 	for_each_alloc_enabled_rdt_resource(r) {
1656 		fflags =  r->fflags | RF_CTRL_INFO;
1657 		ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags);
1658 		if (ret)
1659 			goto out_destroy;
1660 	}
1661 
1662 	for_each_mon_enabled_rdt_resource(r) {
1663 		fflags =  r->fflags | RF_MON_INFO;
1664 		sprintf(name, "%s_MON", r->name);
1665 		ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
1666 		if (ret)
1667 			goto out_destroy;
1668 	}
1669 
1670 	/*
1671 	 * This extra ref will be put in kernfs_remove() and guarantees
1672 	 * that @rdtgrp->kn is always accessible.
1673 	 */
1674 	kernfs_get(kn_info);
1675 
1676 	ret = rdtgroup_kn_set_ugid(kn_info);
1677 	if (ret)
1678 		goto out_destroy;
1679 
1680 	kernfs_activate(kn_info);
1681 
1682 	return 0;
1683 
1684 out_destroy:
1685 	kernfs_remove(kn_info);
1686 	return ret;
1687 }
1688 
1689 static int
1690 mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
1691 		    char *name, struct kernfs_node **dest_kn)
1692 {
1693 	struct kernfs_node *kn;
1694 	int ret;
1695 
1696 	/* create the directory */
1697 	kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1698 	if (IS_ERR(kn))
1699 		return PTR_ERR(kn);
1700 
1701 	if (dest_kn)
1702 		*dest_kn = kn;
1703 
1704 	/*
1705 	 * This extra ref will be put in kernfs_remove() and guarantees
1706 	 * that @rdtgrp->kn is always accessible.
1707 	 */
1708 	kernfs_get(kn);
1709 
1710 	ret = rdtgroup_kn_set_ugid(kn);
1711 	if (ret)
1712 		goto out_destroy;
1713 
1714 	kernfs_activate(kn);
1715 
1716 	return 0;
1717 
1718 out_destroy:
1719 	kernfs_remove(kn);
1720 	return ret;
1721 }
1722 
1723 static void l3_qos_cfg_update(void *arg)
1724 {
1725 	bool *enable = arg;
1726 
1727 	wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL);
1728 }
1729 
1730 static void l2_qos_cfg_update(void *arg)
1731 {
1732 	bool *enable = arg;
1733 
1734 	wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL);
1735 }
1736 
1737 static inline bool is_mba_linear(void)
1738 {
1739 	return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear;
1740 }
1741 
1742 static int set_cache_qos_cfg(int level, bool enable)
1743 {
1744 	void (*update)(void *arg);
1745 	struct rdt_resource *r_l;
1746 	cpumask_var_t cpu_mask;
1747 	struct rdt_domain *d;
1748 	int cpu;
1749 
1750 	if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1751 		return -ENOMEM;
1752 
1753 	if (level == RDT_RESOURCE_L3)
1754 		update = l3_qos_cfg_update;
1755 	else if (level == RDT_RESOURCE_L2)
1756 		update = l2_qos_cfg_update;
1757 	else
1758 		return -EINVAL;
1759 
1760 	r_l = &rdt_resources_all[level];
1761 	list_for_each_entry(d, &r_l->domains, list) {
1762 		/* Pick one CPU from each domain instance to update MSR */
1763 		cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1764 	}
1765 	cpu = get_cpu();
1766 	/* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1767 	if (cpumask_test_cpu(cpu, cpu_mask))
1768 		update(&enable);
1769 	/* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1770 	smp_call_function_many(cpu_mask, update, &enable, 1);
1771 	put_cpu();
1772 
1773 	free_cpumask_var(cpu_mask);
1774 
1775 	return 0;
1776 }
1777 
1778 /*
1779  * Enable or disable the MBA software controller
1780  * which helps user specify bandwidth in MBps.
1781  * MBA software controller is supported only if
1782  * MBM is supported and MBA is in linear scale.
1783  */
1784 static int set_mba_sc(bool mba_sc)
1785 {
1786 	struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA];
1787 	struct rdt_domain *d;
1788 
1789 	if (!is_mbm_enabled() || !is_mba_linear() ||
1790 	    mba_sc == is_mba_sc(r))
1791 		return -EINVAL;
1792 
1793 	r->membw.mba_sc = mba_sc;
1794 	list_for_each_entry(d, &r->domains, list)
1795 		setup_default_ctrlval(r, d->ctrl_val, d->mbps_val);
1796 
1797 	return 0;
1798 }
1799 
1800 static int cdp_enable(int level, int data_type, int code_type)
1801 {
1802 	struct rdt_resource *r_ldata = &rdt_resources_all[data_type];
1803 	struct rdt_resource *r_lcode = &rdt_resources_all[code_type];
1804 	struct rdt_resource *r_l = &rdt_resources_all[level];
1805 	int ret;
1806 
1807 	if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1808 	    !r_lcode->alloc_capable)
1809 		return -EINVAL;
1810 
1811 	ret = set_cache_qos_cfg(level, true);
1812 	if (!ret) {
1813 		r_l->alloc_enabled = false;
1814 		r_ldata->alloc_enabled = true;
1815 		r_lcode->alloc_enabled = true;
1816 	}
1817 	return ret;
1818 }
1819 
1820 static int cdpl3_enable(void)
1821 {
1822 	return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1823 			  RDT_RESOURCE_L3CODE);
1824 }
1825 
1826 static int cdpl2_enable(void)
1827 {
1828 	return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1829 			  RDT_RESOURCE_L2CODE);
1830 }
1831 
1832 static void cdp_disable(int level, int data_type, int code_type)
1833 {
1834 	struct rdt_resource *r = &rdt_resources_all[level];
1835 
1836 	r->alloc_enabled = r->alloc_capable;
1837 
1838 	if (rdt_resources_all[data_type].alloc_enabled) {
1839 		rdt_resources_all[data_type].alloc_enabled = false;
1840 		rdt_resources_all[code_type].alloc_enabled = false;
1841 		set_cache_qos_cfg(level, false);
1842 	}
1843 }
1844 
1845 static void cdpl3_disable(void)
1846 {
1847 	cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
1848 }
1849 
1850 static void cdpl2_disable(void)
1851 {
1852 	cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
1853 }
1854 
1855 static void cdp_disable_all(void)
1856 {
1857 	if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
1858 		cdpl3_disable();
1859 	if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
1860 		cdpl2_disable();
1861 }
1862 
1863 /*
1864  * We don't allow rdtgroup directories to be created anywhere
1865  * except the root directory. Thus when looking for the rdtgroup
1866  * structure for a kernfs node we are either looking at a directory,
1867  * in which case the rdtgroup structure is pointed at by the "priv"
1868  * field, otherwise we have a file, and need only look to the parent
1869  * to find the rdtgroup.
1870  */
1871 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
1872 {
1873 	if (kernfs_type(kn) == KERNFS_DIR) {
1874 		/*
1875 		 * All the resource directories use "kn->priv"
1876 		 * to point to the "struct rdtgroup" for the
1877 		 * resource. "info" and its subdirectories don't
1878 		 * have rdtgroup structures, so return NULL here.
1879 		 */
1880 		if (kn == kn_info || kn->parent == kn_info)
1881 			return NULL;
1882 		else
1883 			return kn->priv;
1884 	} else {
1885 		return kn->parent->priv;
1886 	}
1887 }
1888 
1889 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
1890 {
1891 	struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1892 
1893 	if (!rdtgrp)
1894 		return NULL;
1895 
1896 	atomic_inc(&rdtgrp->waitcount);
1897 	kernfs_break_active_protection(kn);
1898 
1899 	mutex_lock(&rdtgroup_mutex);
1900 
1901 	/* Was this group deleted while we waited? */
1902 	if (rdtgrp->flags & RDT_DELETED)
1903 		return NULL;
1904 
1905 	return rdtgrp;
1906 }
1907 
1908 void rdtgroup_kn_unlock(struct kernfs_node *kn)
1909 {
1910 	struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1911 
1912 	if (!rdtgrp)
1913 		return;
1914 
1915 	mutex_unlock(&rdtgroup_mutex);
1916 
1917 	if (atomic_dec_and_test(&rdtgrp->waitcount) &&
1918 	    (rdtgrp->flags & RDT_DELETED)) {
1919 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
1920 		    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
1921 			rdtgroup_pseudo_lock_remove(rdtgrp);
1922 		kernfs_unbreak_active_protection(kn);
1923 		kernfs_put(rdtgrp->kn);
1924 		kfree(rdtgrp);
1925 	} else {
1926 		kernfs_unbreak_active_protection(kn);
1927 	}
1928 }
1929 
1930 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1931 			     struct rdtgroup *prgrp,
1932 			     struct kernfs_node **mon_data_kn);
1933 
1934 static int rdt_enable_ctx(struct rdt_fs_context *ctx)
1935 {
1936 	int ret = 0;
1937 
1938 	if (ctx->enable_cdpl2)
1939 		ret = cdpl2_enable();
1940 
1941 	if (!ret && ctx->enable_cdpl3)
1942 		ret = cdpl3_enable();
1943 
1944 	if (!ret && ctx->enable_mba_mbps)
1945 		ret = set_mba_sc(true);
1946 
1947 	return ret;
1948 }
1949 
1950 static int rdt_get_tree(struct fs_context *fc)
1951 {
1952 	struct rdt_fs_context *ctx = rdt_fc2context(fc);
1953 	struct rdt_domain *dom;
1954 	struct rdt_resource *r;
1955 	int ret;
1956 
1957 	cpus_read_lock();
1958 	mutex_lock(&rdtgroup_mutex);
1959 	/*
1960 	 * resctrl file system can only be mounted once.
1961 	 */
1962 	if (static_branch_unlikely(&rdt_enable_key)) {
1963 		ret = -EBUSY;
1964 		goto out;
1965 	}
1966 
1967 	ret = rdt_enable_ctx(ctx);
1968 	if (ret < 0)
1969 		goto out_cdp;
1970 
1971 	closid_init();
1972 
1973 	ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
1974 	if (ret < 0)
1975 		goto out_mba;
1976 
1977 	if (rdt_mon_capable) {
1978 		ret = mongroup_create_dir(rdtgroup_default.kn,
1979 					  NULL, "mon_groups",
1980 					  &kn_mongrp);
1981 		if (ret < 0)
1982 			goto out_info;
1983 		kernfs_get(kn_mongrp);
1984 
1985 		ret = mkdir_mondata_all(rdtgroup_default.kn,
1986 					&rdtgroup_default, &kn_mondata);
1987 		if (ret < 0)
1988 			goto out_mongrp;
1989 		kernfs_get(kn_mondata);
1990 		rdtgroup_default.mon.mon_data_kn = kn_mondata;
1991 	}
1992 
1993 	ret = rdt_pseudo_lock_init();
1994 	if (ret)
1995 		goto out_mondata;
1996 
1997 	ret = kernfs_get_tree(fc);
1998 	if (ret < 0)
1999 		goto out_psl;
2000 
2001 	if (rdt_alloc_capable)
2002 		static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
2003 	if (rdt_mon_capable)
2004 		static_branch_enable_cpuslocked(&rdt_mon_enable_key);
2005 
2006 	if (rdt_alloc_capable || rdt_mon_capable)
2007 		static_branch_enable_cpuslocked(&rdt_enable_key);
2008 
2009 	if (is_mbm_enabled()) {
2010 		r = &rdt_resources_all[RDT_RESOURCE_L3];
2011 		list_for_each_entry(dom, &r->domains, list)
2012 			mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
2013 	}
2014 
2015 	goto out;
2016 
2017 out_psl:
2018 	rdt_pseudo_lock_release();
2019 out_mondata:
2020 	if (rdt_mon_capable)
2021 		kernfs_remove(kn_mondata);
2022 out_mongrp:
2023 	if (rdt_mon_capable)
2024 		kernfs_remove(kn_mongrp);
2025 out_info:
2026 	kernfs_remove(kn_info);
2027 out_mba:
2028 	if (ctx->enable_mba_mbps)
2029 		set_mba_sc(false);
2030 out_cdp:
2031 	cdp_disable_all();
2032 out:
2033 	rdt_last_cmd_clear();
2034 	mutex_unlock(&rdtgroup_mutex);
2035 	cpus_read_unlock();
2036 	return ret;
2037 }
2038 
2039 enum rdt_param {
2040 	Opt_cdp,
2041 	Opt_cdpl2,
2042 	Opt_mba_mpbs,
2043 	nr__rdt_params
2044 };
2045 
2046 static const struct fs_parameter_spec rdt_param_specs[] = {
2047 	fsparam_flag("cdp",		Opt_cdp),
2048 	fsparam_flag("cdpl2",		Opt_cdpl2),
2049 	fsparam_flag("mba_mpbs",	Opt_mba_mpbs),
2050 	{}
2051 };
2052 
2053 static const struct fs_parameter_description rdt_fs_parameters = {
2054 	.name		= "rdt",
2055 	.specs		= rdt_param_specs,
2056 };
2057 
2058 static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param)
2059 {
2060 	struct rdt_fs_context *ctx = rdt_fc2context(fc);
2061 	struct fs_parse_result result;
2062 	int opt;
2063 
2064 	opt = fs_parse(fc, &rdt_fs_parameters, param, &result);
2065 	if (opt < 0)
2066 		return opt;
2067 
2068 	switch (opt) {
2069 	case Opt_cdp:
2070 		ctx->enable_cdpl3 = true;
2071 		return 0;
2072 	case Opt_cdpl2:
2073 		ctx->enable_cdpl2 = true;
2074 		return 0;
2075 	case Opt_mba_mpbs:
2076 		if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2077 			return -EINVAL;
2078 		ctx->enable_mba_mbps = true;
2079 		return 0;
2080 	}
2081 
2082 	return -EINVAL;
2083 }
2084 
2085 static void rdt_fs_context_free(struct fs_context *fc)
2086 {
2087 	struct rdt_fs_context *ctx = rdt_fc2context(fc);
2088 
2089 	kernfs_free_fs_context(fc);
2090 	kfree(ctx);
2091 }
2092 
2093 static const struct fs_context_operations rdt_fs_context_ops = {
2094 	.free		= rdt_fs_context_free,
2095 	.parse_param	= rdt_parse_param,
2096 	.get_tree	= rdt_get_tree,
2097 };
2098 
2099 static int rdt_init_fs_context(struct fs_context *fc)
2100 {
2101 	struct rdt_fs_context *ctx;
2102 
2103 	ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL);
2104 	if (!ctx)
2105 		return -ENOMEM;
2106 
2107 	ctx->kfc.root = rdt_root;
2108 	ctx->kfc.magic = RDTGROUP_SUPER_MAGIC;
2109 	fc->fs_private = &ctx->kfc;
2110 	fc->ops = &rdt_fs_context_ops;
2111 	if (fc->user_ns)
2112 		put_user_ns(fc->user_ns);
2113 	fc->user_ns = get_user_ns(&init_user_ns);
2114 	fc->global = true;
2115 	return 0;
2116 }
2117 
2118 static int reset_all_ctrls(struct rdt_resource *r)
2119 {
2120 	struct msr_param msr_param;
2121 	cpumask_var_t cpu_mask;
2122 	struct rdt_domain *d;
2123 	int i, cpu;
2124 
2125 	if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
2126 		return -ENOMEM;
2127 
2128 	msr_param.res = r;
2129 	msr_param.low = 0;
2130 	msr_param.high = r->num_closid;
2131 
2132 	/*
2133 	 * Disable resource control for this resource by setting all
2134 	 * CBMs in all domains to the maximum mask value. Pick one CPU
2135 	 * from each domain to update the MSRs below.
2136 	 */
2137 	list_for_each_entry(d, &r->domains, list) {
2138 		cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
2139 
2140 		for (i = 0; i < r->num_closid; i++)
2141 			d->ctrl_val[i] = r->default_ctrl;
2142 	}
2143 	cpu = get_cpu();
2144 	/* Update CBM on this cpu if it's in cpu_mask. */
2145 	if (cpumask_test_cpu(cpu, cpu_mask))
2146 		rdt_ctrl_update(&msr_param);
2147 	/* Update CBM on all other cpus in cpu_mask. */
2148 	smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
2149 	put_cpu();
2150 
2151 	free_cpumask_var(cpu_mask);
2152 
2153 	return 0;
2154 }
2155 
2156 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
2157 {
2158 	return (rdt_alloc_capable &&
2159 		(r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
2160 }
2161 
2162 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
2163 {
2164 	return (rdt_mon_capable &&
2165 		(r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
2166 }
2167 
2168 /*
2169  * Move tasks from one to the other group. If @from is NULL, then all tasks
2170  * in the systems are moved unconditionally (used for teardown).
2171  *
2172  * If @mask is not NULL the cpus on which moved tasks are running are set
2173  * in that mask so the update smp function call is restricted to affected
2174  * cpus.
2175  */
2176 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
2177 				 struct cpumask *mask)
2178 {
2179 	struct task_struct *p, *t;
2180 
2181 	read_lock(&tasklist_lock);
2182 	for_each_process_thread(p, t) {
2183 		if (!from || is_closid_match(t, from) ||
2184 		    is_rmid_match(t, from)) {
2185 			t->closid = to->closid;
2186 			t->rmid = to->mon.rmid;
2187 
2188 #ifdef CONFIG_SMP
2189 			/*
2190 			 * This is safe on x86 w/o barriers as the ordering
2191 			 * of writing to task_cpu() and t->on_cpu is
2192 			 * reverse to the reading here. The detection is
2193 			 * inaccurate as tasks might move or schedule
2194 			 * before the smp function call takes place. In
2195 			 * such a case the function call is pointless, but
2196 			 * there is no other side effect.
2197 			 */
2198 			if (mask && t->on_cpu)
2199 				cpumask_set_cpu(task_cpu(t), mask);
2200 #endif
2201 		}
2202 	}
2203 	read_unlock(&tasklist_lock);
2204 }
2205 
2206 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
2207 {
2208 	struct rdtgroup *sentry, *stmp;
2209 	struct list_head *head;
2210 
2211 	head = &rdtgrp->mon.crdtgrp_list;
2212 	list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
2213 		free_rmid(sentry->mon.rmid);
2214 		list_del(&sentry->mon.crdtgrp_list);
2215 		kfree(sentry);
2216 	}
2217 }
2218 
2219 /*
2220  * Forcibly remove all of subdirectories under root.
2221  */
2222 static void rmdir_all_sub(void)
2223 {
2224 	struct rdtgroup *rdtgrp, *tmp;
2225 
2226 	/* Move all tasks to the default resource group */
2227 	rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
2228 
2229 	list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
2230 		/* Free any child rmids */
2231 		free_all_child_rdtgrp(rdtgrp);
2232 
2233 		/* Remove each rdtgroup other than root */
2234 		if (rdtgrp == &rdtgroup_default)
2235 			continue;
2236 
2237 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2238 		    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)
2239 			rdtgroup_pseudo_lock_remove(rdtgrp);
2240 
2241 		/*
2242 		 * Give any CPUs back to the default group. We cannot copy
2243 		 * cpu_online_mask because a CPU might have executed the
2244 		 * offline callback already, but is still marked online.
2245 		 */
2246 		cpumask_or(&rdtgroup_default.cpu_mask,
2247 			   &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2248 
2249 		free_rmid(rdtgrp->mon.rmid);
2250 
2251 		kernfs_remove(rdtgrp->kn);
2252 		list_del(&rdtgrp->rdtgroup_list);
2253 		kfree(rdtgrp);
2254 	}
2255 	/* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
2256 	update_closid_rmid(cpu_online_mask, &rdtgroup_default);
2257 
2258 	kernfs_remove(kn_info);
2259 	kernfs_remove(kn_mongrp);
2260 	kernfs_remove(kn_mondata);
2261 }
2262 
2263 static void rdt_kill_sb(struct super_block *sb)
2264 {
2265 	struct rdt_resource *r;
2266 
2267 	cpus_read_lock();
2268 	mutex_lock(&rdtgroup_mutex);
2269 
2270 	set_mba_sc(false);
2271 
2272 	/*Put everything back to default values. */
2273 	for_each_alloc_enabled_rdt_resource(r)
2274 		reset_all_ctrls(r);
2275 	cdp_disable_all();
2276 	rmdir_all_sub();
2277 	rdt_pseudo_lock_release();
2278 	rdtgroup_default.mode = RDT_MODE_SHAREABLE;
2279 	static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
2280 	static_branch_disable_cpuslocked(&rdt_mon_enable_key);
2281 	static_branch_disable_cpuslocked(&rdt_enable_key);
2282 	kernfs_kill_sb(sb);
2283 	mutex_unlock(&rdtgroup_mutex);
2284 	cpus_read_unlock();
2285 }
2286 
2287 static struct file_system_type rdt_fs_type = {
2288 	.name			= "resctrl",
2289 	.init_fs_context	= rdt_init_fs_context,
2290 	.parameters		= &rdt_fs_parameters,
2291 	.kill_sb		= rdt_kill_sb,
2292 };
2293 
2294 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
2295 		       void *priv)
2296 {
2297 	struct kernfs_node *kn;
2298 	int ret = 0;
2299 
2300 	kn = __kernfs_create_file(parent_kn, name, 0444,
2301 				  GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0,
2302 				  &kf_mondata_ops, priv, NULL, NULL);
2303 	if (IS_ERR(kn))
2304 		return PTR_ERR(kn);
2305 
2306 	ret = rdtgroup_kn_set_ugid(kn);
2307 	if (ret) {
2308 		kernfs_remove(kn);
2309 		return ret;
2310 	}
2311 
2312 	return ret;
2313 }
2314 
2315 /*
2316  * Remove all subdirectories of mon_data of ctrl_mon groups
2317  * and monitor groups with given domain id.
2318  */
2319 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
2320 {
2321 	struct rdtgroup *prgrp, *crgrp;
2322 	char name[32];
2323 
2324 	if (!r->mon_enabled)
2325 		return;
2326 
2327 	list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2328 		sprintf(name, "mon_%s_%02d", r->name, dom_id);
2329 		kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
2330 
2331 		list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
2332 			kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
2333 	}
2334 }
2335 
2336 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
2337 				struct rdt_domain *d,
2338 				struct rdt_resource *r, struct rdtgroup *prgrp)
2339 {
2340 	union mon_data_bits priv;
2341 	struct kernfs_node *kn;
2342 	struct mon_evt *mevt;
2343 	struct rmid_read rr;
2344 	char name[32];
2345 	int ret;
2346 
2347 	sprintf(name, "mon_%s_%02d", r->name, d->id);
2348 	/* create the directory */
2349 	kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
2350 	if (IS_ERR(kn))
2351 		return PTR_ERR(kn);
2352 
2353 	/*
2354 	 * This extra ref will be put in kernfs_remove() and guarantees
2355 	 * that kn is always accessible.
2356 	 */
2357 	kernfs_get(kn);
2358 	ret = rdtgroup_kn_set_ugid(kn);
2359 	if (ret)
2360 		goto out_destroy;
2361 
2362 	if (WARN_ON(list_empty(&r->evt_list))) {
2363 		ret = -EPERM;
2364 		goto out_destroy;
2365 	}
2366 
2367 	priv.u.rid = r->rid;
2368 	priv.u.domid = d->id;
2369 	list_for_each_entry(mevt, &r->evt_list, list) {
2370 		priv.u.evtid = mevt->evtid;
2371 		ret = mon_addfile(kn, mevt->name, priv.priv);
2372 		if (ret)
2373 			goto out_destroy;
2374 
2375 		if (is_mbm_event(mevt->evtid))
2376 			mon_event_read(&rr, d, prgrp, mevt->evtid, true);
2377 	}
2378 	kernfs_activate(kn);
2379 	return 0;
2380 
2381 out_destroy:
2382 	kernfs_remove(kn);
2383 	return ret;
2384 }
2385 
2386 /*
2387  * Add all subdirectories of mon_data for "ctrl_mon" groups
2388  * and "monitor" groups with given domain id.
2389  */
2390 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
2391 				    struct rdt_domain *d)
2392 {
2393 	struct kernfs_node *parent_kn;
2394 	struct rdtgroup *prgrp, *crgrp;
2395 	struct list_head *head;
2396 
2397 	if (!r->mon_enabled)
2398 		return;
2399 
2400 	list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
2401 		parent_kn = prgrp->mon.mon_data_kn;
2402 		mkdir_mondata_subdir(parent_kn, d, r, prgrp);
2403 
2404 		head = &prgrp->mon.crdtgrp_list;
2405 		list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
2406 			parent_kn = crgrp->mon.mon_data_kn;
2407 			mkdir_mondata_subdir(parent_kn, d, r, crgrp);
2408 		}
2409 	}
2410 }
2411 
2412 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
2413 				       struct rdt_resource *r,
2414 				       struct rdtgroup *prgrp)
2415 {
2416 	struct rdt_domain *dom;
2417 	int ret;
2418 
2419 	list_for_each_entry(dom, &r->domains, list) {
2420 		ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
2421 		if (ret)
2422 			return ret;
2423 	}
2424 
2425 	return 0;
2426 }
2427 
2428 /*
2429  * This creates a directory mon_data which contains the monitored data.
2430  *
2431  * mon_data has one directory for each domain whic are named
2432  * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
2433  * with L3 domain looks as below:
2434  * ./mon_data:
2435  * mon_L3_00
2436  * mon_L3_01
2437  * mon_L3_02
2438  * ...
2439  *
2440  * Each domain directory has one file per event:
2441  * ./mon_L3_00/:
2442  * llc_occupancy
2443  *
2444  */
2445 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
2446 			     struct rdtgroup *prgrp,
2447 			     struct kernfs_node **dest_kn)
2448 {
2449 	struct rdt_resource *r;
2450 	struct kernfs_node *kn;
2451 	int ret;
2452 
2453 	/*
2454 	 * Create the mon_data directory first.
2455 	 */
2456 	ret = mongroup_create_dir(parent_kn, NULL, "mon_data", &kn);
2457 	if (ret)
2458 		return ret;
2459 
2460 	if (dest_kn)
2461 		*dest_kn = kn;
2462 
2463 	/*
2464 	 * Create the subdirectories for each domain. Note that all events
2465 	 * in a domain like L3 are grouped into a resource whose domain is L3
2466 	 */
2467 	for_each_mon_enabled_rdt_resource(r) {
2468 		ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
2469 		if (ret)
2470 			goto out_destroy;
2471 	}
2472 
2473 	return 0;
2474 
2475 out_destroy:
2476 	kernfs_remove(kn);
2477 	return ret;
2478 }
2479 
2480 /**
2481  * cbm_ensure_valid - Enforce validity on provided CBM
2482  * @_val:	Candidate CBM
2483  * @r:		RDT resource to which the CBM belongs
2484  *
2485  * The provided CBM represents all cache portions available for use. This
2486  * may be represented by a bitmap that does not consist of contiguous ones
2487  * and thus be an invalid CBM.
2488  * Here the provided CBM is forced to be a valid CBM by only considering
2489  * the first set of contiguous bits as valid and clearing all bits.
2490  * The intention here is to provide a valid default CBM with which a new
2491  * resource group is initialized. The user can follow this with a
2492  * modification to the CBM if the default does not satisfy the
2493  * requirements.
2494  */
2495 static void cbm_ensure_valid(u32 *_val, struct rdt_resource *r)
2496 {
2497 	/*
2498 	 * Convert the u32 _val to an unsigned long required by all the bit
2499 	 * operations within this function. No more than 32 bits of this
2500 	 * converted value can be accessed because all bit operations are
2501 	 * additionally provided with cbm_len that is initialized during
2502 	 * hardware enumeration using five bits from the EAX register and
2503 	 * thus never can exceed 32 bits.
2504 	 */
2505 	unsigned long *val = (unsigned long *)_val;
2506 	unsigned int cbm_len = r->cache.cbm_len;
2507 	unsigned long first_bit, zero_bit;
2508 
2509 	if (*val == 0)
2510 		return;
2511 
2512 	first_bit = find_first_bit(val, cbm_len);
2513 	zero_bit = find_next_zero_bit(val, cbm_len, first_bit);
2514 
2515 	/* Clear any remaining bits to ensure contiguous region */
2516 	bitmap_clear(val, zero_bit, cbm_len - zero_bit);
2517 }
2518 
2519 /**
2520  * rdtgroup_init_alloc - Initialize the new RDT group's allocations
2521  *
2522  * A new RDT group is being created on an allocation capable (CAT)
2523  * supporting system. Set this group up to start off with all usable
2524  * allocations. That is, all shareable and unused bits.
2525  *
2526  * All-zero CBM is invalid. If there are no more shareable bits available
2527  * on any domain then the entire allocation will fail.
2528  */
2529 static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp)
2530 {
2531 	struct rdt_resource *r_cdp = NULL;
2532 	struct rdt_domain *d_cdp = NULL;
2533 	u32 used_b = 0, unused_b = 0;
2534 	u32 closid = rdtgrp->closid;
2535 	struct rdt_resource *r;
2536 	unsigned long tmp_cbm;
2537 	enum rdtgrp_mode mode;
2538 	struct rdt_domain *d;
2539 	u32 peer_ctl, *ctrl;
2540 	int i, ret;
2541 
2542 	for_each_alloc_enabled_rdt_resource(r) {
2543 		/*
2544 		 * Only initialize default allocations for CBM cache
2545 		 * resources
2546 		 */
2547 		if (r->rid == RDT_RESOURCE_MBA)
2548 			continue;
2549 		list_for_each_entry(d, &r->domains, list) {
2550 			rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp);
2551 			d->have_new_ctrl = false;
2552 			d->new_ctrl = r->cache.shareable_bits;
2553 			used_b = r->cache.shareable_bits;
2554 			ctrl = d->ctrl_val;
2555 			for (i = 0; i < closids_supported(); i++, ctrl++) {
2556 				if (closid_allocated(i) && i != closid) {
2557 					mode = rdtgroup_mode_by_closid(i);
2558 					if (mode == RDT_MODE_PSEUDO_LOCKSETUP)
2559 						break;
2560 					/*
2561 					 * If CDP is active include peer
2562 					 * domain's usage to ensure there
2563 					 * is no overlap with an exclusive
2564 					 * group.
2565 					 */
2566 					if (d_cdp)
2567 						peer_ctl = d_cdp->ctrl_val[i];
2568 					else
2569 						peer_ctl = 0;
2570 					used_b |= *ctrl | peer_ctl;
2571 					if (mode == RDT_MODE_SHAREABLE)
2572 						d->new_ctrl |= *ctrl | peer_ctl;
2573 				}
2574 			}
2575 			if (d->plr && d->plr->cbm > 0)
2576 				used_b |= d->plr->cbm;
2577 			unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1);
2578 			unused_b &= BIT_MASK(r->cache.cbm_len) - 1;
2579 			d->new_ctrl |= unused_b;
2580 			/*
2581 			 * Force the initial CBM to be valid, user can
2582 			 * modify the CBM based on system availability.
2583 			 */
2584 			cbm_ensure_valid(&d->new_ctrl, r);
2585 			/*
2586 			 * Assign the u32 CBM to an unsigned long to ensure
2587 			 * that bitmap_weight() does not access out-of-bound
2588 			 * memory.
2589 			 */
2590 			tmp_cbm = d->new_ctrl;
2591 			if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) <
2592 			    r->cache.min_cbm_bits) {
2593 				rdt_last_cmd_printf("No space on %s:%d\n",
2594 						    r->name, d->id);
2595 				return -ENOSPC;
2596 			}
2597 			d->have_new_ctrl = true;
2598 		}
2599 	}
2600 
2601 	for_each_alloc_enabled_rdt_resource(r) {
2602 		/*
2603 		 * Only initialize default allocations for CBM cache
2604 		 * resources
2605 		 */
2606 		if (r->rid == RDT_RESOURCE_MBA)
2607 			continue;
2608 		ret = update_domains(r, rdtgrp->closid);
2609 		if (ret < 0) {
2610 			rdt_last_cmd_puts("Failed to initialize allocations\n");
2611 			return ret;
2612 		}
2613 		rdtgrp->mode = RDT_MODE_SHAREABLE;
2614 	}
2615 
2616 	return 0;
2617 }
2618 
2619 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
2620 			     struct kernfs_node *prgrp_kn,
2621 			     const char *name, umode_t mode,
2622 			     enum rdt_group_type rtype, struct rdtgroup **r)
2623 {
2624 	struct rdtgroup *prdtgrp, *rdtgrp;
2625 	struct kernfs_node *kn;
2626 	uint files = 0;
2627 	int ret;
2628 
2629 	prdtgrp = rdtgroup_kn_lock_live(prgrp_kn);
2630 	rdt_last_cmd_clear();
2631 	if (!prdtgrp) {
2632 		ret = -ENODEV;
2633 		rdt_last_cmd_puts("Directory was removed\n");
2634 		goto out_unlock;
2635 	}
2636 
2637 	if (rtype == RDTMON_GROUP &&
2638 	    (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2639 	     prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) {
2640 		ret = -EINVAL;
2641 		rdt_last_cmd_puts("Pseudo-locking in progress\n");
2642 		goto out_unlock;
2643 	}
2644 
2645 	/* allocate the rdtgroup. */
2646 	rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
2647 	if (!rdtgrp) {
2648 		ret = -ENOSPC;
2649 		rdt_last_cmd_puts("Kernel out of memory\n");
2650 		goto out_unlock;
2651 	}
2652 	*r = rdtgrp;
2653 	rdtgrp->mon.parent = prdtgrp;
2654 	rdtgrp->type = rtype;
2655 	INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
2656 
2657 	/* kernfs creates the directory for rdtgrp */
2658 	kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
2659 	if (IS_ERR(kn)) {
2660 		ret = PTR_ERR(kn);
2661 		rdt_last_cmd_puts("kernfs create error\n");
2662 		goto out_free_rgrp;
2663 	}
2664 	rdtgrp->kn = kn;
2665 
2666 	/*
2667 	 * kernfs_remove() will drop the reference count on "kn" which
2668 	 * will free it. But we still need it to stick around for the
2669 	 * rdtgroup_kn_unlock(kn} call below. Take one extra reference
2670 	 * here, which will be dropped inside rdtgroup_kn_unlock().
2671 	 */
2672 	kernfs_get(kn);
2673 
2674 	ret = rdtgroup_kn_set_ugid(kn);
2675 	if (ret) {
2676 		rdt_last_cmd_puts("kernfs perm error\n");
2677 		goto out_destroy;
2678 	}
2679 
2680 	files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
2681 	ret = rdtgroup_add_files(kn, files);
2682 	if (ret) {
2683 		rdt_last_cmd_puts("kernfs fill error\n");
2684 		goto out_destroy;
2685 	}
2686 
2687 	if (rdt_mon_capable) {
2688 		ret = alloc_rmid();
2689 		if (ret < 0) {
2690 			rdt_last_cmd_puts("Out of RMIDs\n");
2691 			goto out_destroy;
2692 		}
2693 		rdtgrp->mon.rmid = ret;
2694 
2695 		ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
2696 		if (ret) {
2697 			rdt_last_cmd_puts("kernfs subdir error\n");
2698 			goto out_idfree;
2699 		}
2700 	}
2701 	kernfs_activate(kn);
2702 
2703 	/*
2704 	 * The caller unlocks the prgrp_kn upon success.
2705 	 */
2706 	return 0;
2707 
2708 out_idfree:
2709 	free_rmid(rdtgrp->mon.rmid);
2710 out_destroy:
2711 	kernfs_remove(rdtgrp->kn);
2712 out_free_rgrp:
2713 	kfree(rdtgrp);
2714 out_unlock:
2715 	rdtgroup_kn_unlock(prgrp_kn);
2716 	return ret;
2717 }
2718 
2719 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
2720 {
2721 	kernfs_remove(rgrp->kn);
2722 	free_rmid(rgrp->mon.rmid);
2723 	kfree(rgrp);
2724 }
2725 
2726 /*
2727  * Create a monitor group under "mon_groups" directory of a control
2728  * and monitor group(ctrl_mon). This is a resource group
2729  * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
2730  */
2731 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
2732 			      struct kernfs_node *prgrp_kn,
2733 			      const char *name,
2734 			      umode_t mode)
2735 {
2736 	struct rdtgroup *rdtgrp, *prgrp;
2737 	int ret;
2738 
2739 	ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP,
2740 				&rdtgrp);
2741 	if (ret)
2742 		return ret;
2743 
2744 	prgrp = rdtgrp->mon.parent;
2745 	rdtgrp->closid = prgrp->closid;
2746 
2747 	/*
2748 	 * Add the rdtgrp to the list of rdtgrps the parent
2749 	 * ctrl_mon group has to track.
2750 	 */
2751 	list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
2752 
2753 	rdtgroup_kn_unlock(prgrp_kn);
2754 	return ret;
2755 }
2756 
2757 /*
2758  * These are rdtgroups created under the root directory. Can be used
2759  * to allocate and monitor resources.
2760  */
2761 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
2762 				   struct kernfs_node *prgrp_kn,
2763 				   const char *name, umode_t mode)
2764 {
2765 	struct rdtgroup *rdtgrp;
2766 	struct kernfs_node *kn;
2767 	u32 closid;
2768 	int ret;
2769 
2770 	ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP,
2771 				&rdtgrp);
2772 	if (ret)
2773 		return ret;
2774 
2775 	kn = rdtgrp->kn;
2776 	ret = closid_alloc();
2777 	if (ret < 0) {
2778 		rdt_last_cmd_puts("Out of CLOSIDs\n");
2779 		goto out_common_fail;
2780 	}
2781 	closid = ret;
2782 	ret = 0;
2783 
2784 	rdtgrp->closid = closid;
2785 	ret = rdtgroup_init_alloc(rdtgrp);
2786 	if (ret < 0)
2787 		goto out_id_free;
2788 
2789 	list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
2790 
2791 	if (rdt_mon_capable) {
2792 		/*
2793 		 * Create an empty mon_groups directory to hold the subset
2794 		 * of tasks and cpus to monitor.
2795 		 */
2796 		ret = mongroup_create_dir(kn, NULL, "mon_groups", NULL);
2797 		if (ret) {
2798 			rdt_last_cmd_puts("kernfs subdir error\n");
2799 			goto out_del_list;
2800 		}
2801 	}
2802 
2803 	goto out_unlock;
2804 
2805 out_del_list:
2806 	list_del(&rdtgrp->rdtgroup_list);
2807 out_id_free:
2808 	closid_free(closid);
2809 out_common_fail:
2810 	mkdir_rdt_prepare_clean(rdtgrp);
2811 out_unlock:
2812 	rdtgroup_kn_unlock(prgrp_kn);
2813 	return ret;
2814 }
2815 
2816 /*
2817  * We allow creating mon groups only with in a directory called "mon_groups"
2818  * which is present in every ctrl_mon group. Check if this is a valid
2819  * "mon_groups" directory.
2820  *
2821  * 1. The directory should be named "mon_groups".
2822  * 2. The mon group itself should "not" be named "mon_groups".
2823  *   This makes sure "mon_groups" directory always has a ctrl_mon group
2824  *   as parent.
2825  */
2826 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
2827 {
2828 	return (!strcmp(kn->name, "mon_groups") &&
2829 		strcmp(name, "mon_groups"));
2830 }
2831 
2832 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
2833 			  umode_t mode)
2834 {
2835 	/* Do not accept '\n' to avoid unparsable situation. */
2836 	if (strchr(name, '\n'))
2837 		return -EINVAL;
2838 
2839 	/*
2840 	 * If the parent directory is the root directory and RDT
2841 	 * allocation is supported, add a control and monitoring
2842 	 * subdirectory
2843 	 */
2844 	if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
2845 		return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode);
2846 
2847 	/*
2848 	 * If RDT monitoring is supported and the parent directory is a valid
2849 	 * "mon_groups" directory, add a monitoring subdirectory.
2850 	 */
2851 	if (rdt_mon_capable && is_mon_groups(parent_kn, name))
2852 		return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode);
2853 
2854 	return -EPERM;
2855 }
2856 
2857 static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2858 			      cpumask_var_t tmpmask)
2859 {
2860 	struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
2861 	int cpu;
2862 
2863 	/* Give any tasks back to the parent group */
2864 	rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
2865 
2866 	/* Update per cpu rmid of the moved CPUs first */
2867 	for_each_cpu(cpu, &rdtgrp->cpu_mask)
2868 		per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
2869 	/*
2870 	 * Update the MSR on moved CPUs and CPUs which have moved
2871 	 * task running on them.
2872 	 */
2873 	cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2874 	update_closid_rmid(tmpmask, NULL);
2875 
2876 	rdtgrp->flags = RDT_DELETED;
2877 	free_rmid(rdtgrp->mon.rmid);
2878 
2879 	/*
2880 	 * Remove the rdtgrp from the parent ctrl_mon group's list
2881 	 */
2882 	WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
2883 	list_del(&rdtgrp->mon.crdtgrp_list);
2884 
2885 	/*
2886 	 * one extra hold on this, will drop when we kfree(rdtgrp)
2887 	 * in rdtgroup_kn_unlock()
2888 	 */
2889 	kernfs_get(kn);
2890 	kernfs_remove(rdtgrp->kn);
2891 
2892 	return 0;
2893 }
2894 
2895 static int rdtgroup_ctrl_remove(struct kernfs_node *kn,
2896 				struct rdtgroup *rdtgrp)
2897 {
2898 	rdtgrp->flags = RDT_DELETED;
2899 	list_del(&rdtgrp->rdtgroup_list);
2900 
2901 	/*
2902 	 * one extra hold on this, will drop when we kfree(rdtgrp)
2903 	 * in rdtgroup_kn_unlock()
2904 	 */
2905 	kernfs_get(kn);
2906 	kernfs_remove(rdtgrp->kn);
2907 	return 0;
2908 }
2909 
2910 static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
2911 			       cpumask_var_t tmpmask)
2912 {
2913 	int cpu;
2914 
2915 	/* Give any tasks back to the default group */
2916 	rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
2917 
2918 	/* Give any CPUs back to the default group */
2919 	cpumask_or(&rdtgroup_default.cpu_mask,
2920 		   &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
2921 
2922 	/* Update per cpu closid and rmid of the moved CPUs first */
2923 	for_each_cpu(cpu, &rdtgrp->cpu_mask) {
2924 		per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
2925 		per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
2926 	}
2927 
2928 	/*
2929 	 * Update the MSR on moved CPUs and CPUs which have moved
2930 	 * task running on them.
2931 	 */
2932 	cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
2933 	update_closid_rmid(tmpmask, NULL);
2934 
2935 	closid_free(rdtgrp->closid);
2936 	free_rmid(rdtgrp->mon.rmid);
2937 
2938 	/*
2939 	 * Free all the child monitor group rmids.
2940 	 */
2941 	free_all_child_rdtgrp(rdtgrp);
2942 
2943 	rdtgroup_ctrl_remove(kn, rdtgrp);
2944 
2945 	return 0;
2946 }
2947 
2948 static int rdtgroup_rmdir(struct kernfs_node *kn)
2949 {
2950 	struct kernfs_node *parent_kn = kn->parent;
2951 	struct rdtgroup *rdtgrp;
2952 	cpumask_var_t tmpmask;
2953 	int ret = 0;
2954 
2955 	if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
2956 		return -ENOMEM;
2957 
2958 	rdtgrp = rdtgroup_kn_lock_live(kn);
2959 	if (!rdtgrp) {
2960 		ret = -EPERM;
2961 		goto out;
2962 	}
2963 
2964 	/*
2965 	 * If the rdtgroup is a ctrl_mon group and parent directory
2966 	 * is the root directory, remove the ctrl_mon group.
2967 	 *
2968 	 * If the rdtgroup is a mon group and parent directory
2969 	 * is a valid "mon_groups" directory, remove the mon group.
2970 	 */
2971 	if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn) {
2972 		if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP ||
2973 		    rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) {
2974 			ret = rdtgroup_ctrl_remove(kn, rdtgrp);
2975 		} else {
2976 			ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask);
2977 		}
2978 	} else if (rdtgrp->type == RDTMON_GROUP &&
2979 		 is_mon_groups(parent_kn, kn->name)) {
2980 		ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask);
2981 	} else {
2982 		ret = -EPERM;
2983 	}
2984 
2985 out:
2986 	rdtgroup_kn_unlock(kn);
2987 	free_cpumask_var(tmpmask);
2988 	return ret;
2989 }
2990 
2991 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
2992 {
2993 	if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
2994 		seq_puts(seq, ",cdp");
2995 
2996 	if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
2997 		seq_puts(seq, ",cdpl2");
2998 
2999 	if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA]))
3000 		seq_puts(seq, ",mba_MBps");
3001 
3002 	return 0;
3003 }
3004 
3005 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
3006 	.mkdir		= rdtgroup_mkdir,
3007 	.rmdir		= rdtgroup_rmdir,
3008 	.show_options	= rdtgroup_show_options,
3009 };
3010 
3011 static int __init rdtgroup_setup_root(void)
3012 {
3013 	int ret;
3014 
3015 	rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
3016 				      KERNFS_ROOT_CREATE_DEACTIVATED |
3017 				      KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK,
3018 				      &rdtgroup_default);
3019 	if (IS_ERR(rdt_root))
3020 		return PTR_ERR(rdt_root);
3021 
3022 	mutex_lock(&rdtgroup_mutex);
3023 
3024 	rdtgroup_default.closid = 0;
3025 	rdtgroup_default.mon.rmid = 0;
3026 	rdtgroup_default.type = RDTCTRL_GROUP;
3027 	INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
3028 
3029 	list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
3030 
3031 	ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
3032 	if (ret) {
3033 		kernfs_destroy_root(rdt_root);
3034 		goto out;
3035 	}
3036 
3037 	rdtgroup_default.kn = rdt_root->kn;
3038 	kernfs_activate(rdtgroup_default.kn);
3039 
3040 out:
3041 	mutex_unlock(&rdtgroup_mutex);
3042 
3043 	return ret;
3044 }
3045 
3046 /*
3047  * rdtgroup_init - rdtgroup initialization
3048  *
3049  * Setup resctrl file system including set up root, create mount point,
3050  * register rdtgroup filesystem, and initialize files under root directory.
3051  *
3052  * Return: 0 on success or -errno
3053  */
3054 int __init rdtgroup_init(void)
3055 {
3056 	int ret = 0;
3057 
3058 	seq_buf_init(&last_cmd_status, last_cmd_status_buf,
3059 		     sizeof(last_cmd_status_buf));
3060 
3061 	ret = rdtgroup_setup_root();
3062 	if (ret)
3063 		return ret;
3064 
3065 	ret = sysfs_create_mount_point(fs_kobj, "resctrl");
3066 	if (ret)
3067 		goto cleanup_root;
3068 
3069 	ret = register_filesystem(&rdt_fs_type);
3070 	if (ret)
3071 		goto cleanup_mountpoint;
3072 
3073 	/*
3074 	 * Adding the resctrl debugfs directory here may not be ideal since
3075 	 * it would let the resctrl debugfs directory appear on the debugfs
3076 	 * filesystem before the resctrl filesystem is mounted.
3077 	 * It may also be ok since that would enable debugging of RDT before
3078 	 * resctrl is mounted.
3079 	 * The reason why the debugfs directory is created here and not in
3080 	 * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and
3081 	 * during the debugfs directory creation also &sb->s_type->i_mutex_key
3082 	 * (the lockdep class of inode->i_rwsem). Other filesystem
3083 	 * interactions (eg. SyS_getdents) have the lock ordering:
3084 	 * &sb->s_type->i_mutex_key --> &mm->mmap_sem
3085 	 * During mmap(), called with &mm->mmap_sem, the rdtgroup_mutex
3086 	 * is taken, thus creating dependency:
3087 	 * &mm->mmap_sem --> rdtgroup_mutex for the latter that can cause
3088 	 * issues considering the other two lock dependencies.
3089 	 * By creating the debugfs directory here we avoid a dependency
3090 	 * that may cause deadlock (even though file operations cannot
3091 	 * occur until the filesystem is mounted, but I do not know how to
3092 	 * tell lockdep that).
3093 	 */
3094 	debugfs_resctrl = debugfs_create_dir("resctrl", NULL);
3095 
3096 	return 0;
3097 
3098 cleanup_mountpoint:
3099 	sysfs_remove_mount_point(fs_kobj, "resctrl");
3100 cleanup_root:
3101 	kernfs_destroy_root(rdt_root);
3102 
3103 	return ret;
3104 }
3105 
3106 void __exit rdtgroup_exit(void)
3107 {
3108 	debugfs_remove_recursive(debugfs_resctrl);
3109 	unregister_filesystem(&rdt_fs_type);
3110 	sysfs_remove_mount_point(fs_kobj, "resctrl");
3111 	kernfs_destroy_root(rdt_root);
3112 }
3113