Loading...
Note: File does not exist in v5.4.
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/cpu.h>
24#include <linux/fs.h>
25#include <linux/sysfs.h>
26#include <linux/kernfs.h>
27#include <linux/seq_buf.h>
28#include <linux/seq_file.h>
29#include <linux/sched/signal.h>
30#include <linux/sched/task.h>
31#include <linux/slab.h>
32#include <linux/task_work.h>
33
34#include <uapi/linux/magic.h>
35
36#include <asm/intel_rdt_sched.h>
37#include "intel_rdt.h"
38
39DEFINE_STATIC_KEY_FALSE(rdt_enable_key);
40DEFINE_STATIC_KEY_FALSE(rdt_mon_enable_key);
41DEFINE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
42static struct kernfs_root *rdt_root;
43struct rdtgroup rdtgroup_default;
44LIST_HEAD(rdt_all_groups);
45
46/* Kernel fs node for "info" directory under root */
47static struct kernfs_node *kn_info;
48
49/* Kernel fs node for "mon_groups" directory under root */
50static struct kernfs_node *kn_mongrp;
51
52/* Kernel fs node for "mon_data" directory under root */
53static struct kernfs_node *kn_mondata;
54
55static struct seq_buf last_cmd_status;
56static char last_cmd_status_buf[512];
57
58void rdt_last_cmd_clear(void)
59{
60 lockdep_assert_held(&rdtgroup_mutex);
61 seq_buf_clear(&last_cmd_status);
62}
63
64void rdt_last_cmd_puts(const char *s)
65{
66 lockdep_assert_held(&rdtgroup_mutex);
67 seq_buf_puts(&last_cmd_status, s);
68}
69
70void rdt_last_cmd_printf(const char *fmt, ...)
71{
72 va_list ap;
73
74 va_start(ap, fmt);
75 lockdep_assert_held(&rdtgroup_mutex);
76 seq_buf_vprintf(&last_cmd_status, fmt, ap);
77 va_end(ap);
78}
79
80/*
81 * Trivial allocator for CLOSIDs. Since h/w only supports a small number,
82 * we can keep a bitmap of free CLOSIDs in a single integer.
83 *
84 * Using a global CLOSID across all resources has some advantages and
85 * some drawbacks:
86 * + We can simply set "current->closid" to assign a task to a resource
87 * group.
88 * + Context switch code can avoid extra memory references deciding which
89 * CLOSID to load into the PQR_ASSOC MSR
90 * - We give up some options in configuring resource groups across multi-socket
91 * systems.
92 * - Our choices on how to configure each resource become progressively more
93 * limited as the number of resources grows.
94 */
95static int closid_free_map;
96
97static void closid_init(void)
98{
99 struct rdt_resource *r;
100 int rdt_min_closid = 32;
101
102 /* Compute rdt_min_closid across all resources */
103 for_each_alloc_enabled_rdt_resource(r)
104 rdt_min_closid = min(rdt_min_closid, r->num_closid);
105
106 closid_free_map = BIT_MASK(rdt_min_closid) - 1;
107
108 /* CLOSID 0 is always reserved for the default group */
109 closid_free_map &= ~1;
110}
111
112static int closid_alloc(void)
113{
114 u32 closid = ffs(closid_free_map);
115
116 if (closid == 0)
117 return -ENOSPC;
118 closid--;
119 closid_free_map &= ~(1 << closid);
120
121 return closid;
122}
123
124static void closid_free(int closid)
125{
126 closid_free_map |= 1 << closid;
127}
128
129/* set uid and gid of rdtgroup dirs and files to that of the creator */
130static int rdtgroup_kn_set_ugid(struct kernfs_node *kn)
131{
132 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
133 .ia_uid = current_fsuid(),
134 .ia_gid = current_fsgid(), };
135
136 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
137 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
138 return 0;
139
140 return kernfs_setattr(kn, &iattr);
141}
142
143static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft)
144{
145 struct kernfs_node *kn;
146 int ret;
147
148 kn = __kernfs_create_file(parent_kn, rft->name, rft->mode,
149 0, rft->kf_ops, rft, NULL, NULL);
150 if (IS_ERR(kn))
151 return PTR_ERR(kn);
152
153 ret = rdtgroup_kn_set_ugid(kn);
154 if (ret) {
155 kernfs_remove(kn);
156 return ret;
157 }
158
159 return 0;
160}
161
162static int rdtgroup_seqfile_show(struct seq_file *m, void *arg)
163{
164 struct kernfs_open_file *of = m->private;
165 struct rftype *rft = of->kn->priv;
166
167 if (rft->seq_show)
168 return rft->seq_show(of, m, arg);
169 return 0;
170}
171
172static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf,
173 size_t nbytes, loff_t off)
174{
175 struct rftype *rft = of->kn->priv;
176
177 if (rft->write)
178 return rft->write(of, buf, nbytes, off);
179
180 return -EINVAL;
181}
182
183static struct kernfs_ops rdtgroup_kf_single_ops = {
184 .atomic_write_len = PAGE_SIZE,
185 .write = rdtgroup_file_write,
186 .seq_show = rdtgroup_seqfile_show,
187};
188
189static struct kernfs_ops kf_mondata_ops = {
190 .atomic_write_len = PAGE_SIZE,
191 .seq_show = rdtgroup_mondata_show,
192};
193
194static bool is_cpu_list(struct kernfs_open_file *of)
195{
196 struct rftype *rft = of->kn->priv;
197
198 return rft->flags & RFTYPE_FLAGS_CPUS_LIST;
199}
200
201static int rdtgroup_cpus_show(struct kernfs_open_file *of,
202 struct seq_file *s, void *v)
203{
204 struct rdtgroup *rdtgrp;
205 int ret = 0;
206
207 rdtgrp = rdtgroup_kn_lock_live(of->kn);
208
209 if (rdtgrp) {
210 seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n",
211 cpumask_pr_args(&rdtgrp->cpu_mask));
212 } else {
213 ret = -ENOENT;
214 }
215 rdtgroup_kn_unlock(of->kn);
216
217 return ret;
218}
219
220/*
221 * This is safe against intel_rdt_sched_in() called from __switch_to()
222 * because __switch_to() is executed with interrupts disabled. A local call
223 * from update_closid_rmid() is proteced against __switch_to() because
224 * preemption is disabled.
225 */
226static void update_cpu_closid_rmid(void *info)
227{
228 struct rdtgroup *r = info;
229
230 if (r) {
231 this_cpu_write(pqr_state.default_closid, r->closid);
232 this_cpu_write(pqr_state.default_rmid, r->mon.rmid);
233 }
234
235 /*
236 * We cannot unconditionally write the MSR because the current
237 * executing task might have its own closid selected. Just reuse
238 * the context switch code.
239 */
240 intel_rdt_sched_in();
241}
242
243/*
244 * Update the PGR_ASSOC MSR on all cpus in @cpu_mask,
245 *
246 * Per task closids/rmids must have been set up before calling this function.
247 */
248static void
249update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r)
250{
251 int cpu = get_cpu();
252
253 if (cpumask_test_cpu(cpu, cpu_mask))
254 update_cpu_closid_rmid(r);
255 smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1);
256 put_cpu();
257}
258
259static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
260 cpumask_var_t tmpmask)
261{
262 struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp;
263 struct list_head *head;
264
265 /* Check whether cpus belong to parent ctrl group */
266 cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask);
267 if (cpumask_weight(tmpmask)) {
268 rdt_last_cmd_puts("can only add CPUs to mongroup that belong to parent\n");
269 return -EINVAL;
270 }
271
272 /* Check whether cpus are dropped from this group */
273 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
274 if (cpumask_weight(tmpmask)) {
275 /* Give any dropped cpus to parent rdtgroup */
276 cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask);
277 update_closid_rmid(tmpmask, prgrp);
278 }
279
280 /*
281 * If we added cpus, remove them from previous group that owned them
282 * and update per-cpu rmid
283 */
284 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
285 if (cpumask_weight(tmpmask)) {
286 head = &prgrp->mon.crdtgrp_list;
287 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
288 if (crgrp == rdtgrp)
289 continue;
290 cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask,
291 tmpmask);
292 }
293 update_closid_rmid(tmpmask, rdtgrp);
294 }
295
296 /* Done pushing/pulling - update this group with new mask */
297 cpumask_copy(&rdtgrp->cpu_mask, newmask);
298
299 return 0;
300}
301
302static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m)
303{
304 struct rdtgroup *crgrp;
305
306 cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m);
307 /* update the child mon group masks as well*/
308 list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list)
309 cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask);
310}
311
312static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask,
313 cpumask_var_t tmpmask, cpumask_var_t tmpmask1)
314{
315 struct rdtgroup *r, *crgrp;
316 struct list_head *head;
317
318 /* Check whether cpus are dropped from this group */
319 cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask);
320 if (cpumask_weight(tmpmask)) {
321 /* Can't drop from default group */
322 if (rdtgrp == &rdtgroup_default) {
323 rdt_last_cmd_puts("Can't drop CPUs from default group\n");
324 return -EINVAL;
325 }
326
327 /* Give any dropped cpus to rdtgroup_default */
328 cpumask_or(&rdtgroup_default.cpu_mask,
329 &rdtgroup_default.cpu_mask, tmpmask);
330 update_closid_rmid(tmpmask, &rdtgroup_default);
331 }
332
333 /*
334 * If we added cpus, remove them from previous group and
335 * the prev group's child groups that owned them
336 * and update per-cpu closid/rmid.
337 */
338 cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask);
339 if (cpumask_weight(tmpmask)) {
340 list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) {
341 if (r == rdtgrp)
342 continue;
343 cpumask_and(tmpmask1, &r->cpu_mask, tmpmask);
344 if (cpumask_weight(tmpmask1))
345 cpumask_rdtgrp_clear(r, tmpmask1);
346 }
347 update_closid_rmid(tmpmask, rdtgrp);
348 }
349
350 /* Done pushing/pulling - update this group with new mask */
351 cpumask_copy(&rdtgrp->cpu_mask, newmask);
352
353 /*
354 * Clear child mon group masks since there is a new parent mask
355 * now and update the rmid for the cpus the child lost.
356 */
357 head = &rdtgrp->mon.crdtgrp_list;
358 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
359 cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask);
360 update_closid_rmid(tmpmask, rdtgrp);
361 cpumask_clear(&crgrp->cpu_mask);
362 }
363
364 return 0;
365}
366
367static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of,
368 char *buf, size_t nbytes, loff_t off)
369{
370 cpumask_var_t tmpmask, newmask, tmpmask1;
371 struct rdtgroup *rdtgrp;
372 int ret;
373
374 if (!buf)
375 return -EINVAL;
376
377 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
378 return -ENOMEM;
379 if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) {
380 free_cpumask_var(tmpmask);
381 return -ENOMEM;
382 }
383 if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) {
384 free_cpumask_var(tmpmask);
385 free_cpumask_var(newmask);
386 return -ENOMEM;
387 }
388
389 rdtgrp = rdtgroup_kn_lock_live(of->kn);
390 rdt_last_cmd_clear();
391 if (!rdtgrp) {
392 ret = -ENOENT;
393 rdt_last_cmd_puts("directory was removed\n");
394 goto unlock;
395 }
396
397 if (is_cpu_list(of))
398 ret = cpulist_parse(buf, newmask);
399 else
400 ret = cpumask_parse(buf, newmask);
401
402 if (ret) {
403 rdt_last_cmd_puts("bad cpu list/mask\n");
404 goto unlock;
405 }
406
407 /* check that user didn't specify any offline cpus */
408 cpumask_andnot(tmpmask, newmask, cpu_online_mask);
409 if (cpumask_weight(tmpmask)) {
410 ret = -EINVAL;
411 rdt_last_cmd_puts("can only assign online cpus\n");
412 goto unlock;
413 }
414
415 if (rdtgrp->type == RDTCTRL_GROUP)
416 ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1);
417 else if (rdtgrp->type == RDTMON_GROUP)
418 ret = cpus_mon_write(rdtgrp, newmask, tmpmask);
419 else
420 ret = -EINVAL;
421
422unlock:
423 rdtgroup_kn_unlock(of->kn);
424 free_cpumask_var(tmpmask);
425 free_cpumask_var(newmask);
426 free_cpumask_var(tmpmask1);
427
428 return ret ?: nbytes;
429}
430
431struct task_move_callback {
432 struct callback_head work;
433 struct rdtgroup *rdtgrp;
434};
435
436static void move_myself(struct callback_head *head)
437{
438 struct task_move_callback *callback;
439 struct rdtgroup *rdtgrp;
440
441 callback = container_of(head, struct task_move_callback, work);
442 rdtgrp = callback->rdtgrp;
443
444 /*
445 * If resource group was deleted before this task work callback
446 * was invoked, then assign the task to root group and free the
447 * resource group.
448 */
449 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
450 (rdtgrp->flags & RDT_DELETED)) {
451 current->closid = 0;
452 current->rmid = 0;
453 kfree(rdtgrp);
454 }
455
456 preempt_disable();
457 /* update PQR_ASSOC MSR to make resource group go into effect */
458 intel_rdt_sched_in();
459 preempt_enable();
460
461 kfree(callback);
462}
463
464static int __rdtgroup_move_task(struct task_struct *tsk,
465 struct rdtgroup *rdtgrp)
466{
467 struct task_move_callback *callback;
468 int ret;
469
470 callback = kzalloc(sizeof(*callback), GFP_KERNEL);
471 if (!callback)
472 return -ENOMEM;
473 callback->work.func = move_myself;
474 callback->rdtgrp = rdtgrp;
475
476 /*
477 * Take a refcount, so rdtgrp cannot be freed before the
478 * callback has been invoked.
479 */
480 atomic_inc(&rdtgrp->waitcount);
481 ret = task_work_add(tsk, &callback->work, true);
482 if (ret) {
483 /*
484 * Task is exiting. Drop the refcount and free the callback.
485 * No need to check the refcount as the group cannot be
486 * deleted before the write function unlocks rdtgroup_mutex.
487 */
488 atomic_dec(&rdtgrp->waitcount);
489 kfree(callback);
490 rdt_last_cmd_puts("task exited\n");
491 } else {
492 /*
493 * For ctrl_mon groups move both closid and rmid.
494 * For monitor groups, can move the tasks only from
495 * their parent CTRL group.
496 */
497 if (rdtgrp->type == RDTCTRL_GROUP) {
498 tsk->closid = rdtgrp->closid;
499 tsk->rmid = rdtgrp->mon.rmid;
500 } else if (rdtgrp->type == RDTMON_GROUP) {
501 if (rdtgrp->mon.parent->closid == tsk->closid) {
502 tsk->rmid = rdtgrp->mon.rmid;
503 } else {
504 rdt_last_cmd_puts("Can't move task to different control group\n");
505 ret = -EINVAL;
506 }
507 }
508 }
509 return ret;
510}
511
512static int rdtgroup_task_write_permission(struct task_struct *task,
513 struct kernfs_open_file *of)
514{
515 const struct cred *tcred = get_task_cred(task);
516 const struct cred *cred = current_cred();
517 int ret = 0;
518
519 /*
520 * Even if we're attaching all tasks in the thread group, we only
521 * need to check permissions on one of them.
522 */
523 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
524 !uid_eq(cred->euid, tcred->uid) &&
525 !uid_eq(cred->euid, tcred->suid)) {
526 rdt_last_cmd_printf("No permission to move task %d\n", task->pid);
527 ret = -EPERM;
528 }
529
530 put_cred(tcred);
531 return ret;
532}
533
534static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp,
535 struct kernfs_open_file *of)
536{
537 struct task_struct *tsk;
538 int ret;
539
540 rcu_read_lock();
541 if (pid) {
542 tsk = find_task_by_vpid(pid);
543 if (!tsk) {
544 rcu_read_unlock();
545 rdt_last_cmd_printf("No task %d\n", pid);
546 return -ESRCH;
547 }
548 } else {
549 tsk = current;
550 }
551
552 get_task_struct(tsk);
553 rcu_read_unlock();
554
555 ret = rdtgroup_task_write_permission(tsk, of);
556 if (!ret)
557 ret = __rdtgroup_move_task(tsk, rdtgrp);
558
559 put_task_struct(tsk);
560 return ret;
561}
562
563static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of,
564 char *buf, size_t nbytes, loff_t off)
565{
566 struct rdtgroup *rdtgrp;
567 int ret = 0;
568 pid_t pid;
569
570 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
571 return -EINVAL;
572 rdtgrp = rdtgroup_kn_lock_live(of->kn);
573 rdt_last_cmd_clear();
574
575 if (rdtgrp)
576 ret = rdtgroup_move_task(pid, rdtgrp, of);
577 else
578 ret = -ENOENT;
579
580 rdtgroup_kn_unlock(of->kn);
581
582 return ret ?: nbytes;
583}
584
585static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s)
586{
587 struct task_struct *p, *t;
588
589 rcu_read_lock();
590 for_each_process_thread(p, t) {
591 if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) ||
592 (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid))
593 seq_printf(s, "%d\n", t->pid);
594 }
595 rcu_read_unlock();
596}
597
598static int rdtgroup_tasks_show(struct kernfs_open_file *of,
599 struct seq_file *s, void *v)
600{
601 struct rdtgroup *rdtgrp;
602 int ret = 0;
603
604 rdtgrp = rdtgroup_kn_lock_live(of->kn);
605 if (rdtgrp)
606 show_rdt_tasks(rdtgrp, s);
607 else
608 ret = -ENOENT;
609 rdtgroup_kn_unlock(of->kn);
610
611 return ret;
612}
613
614static int rdt_last_cmd_status_show(struct kernfs_open_file *of,
615 struct seq_file *seq, void *v)
616{
617 int len;
618
619 mutex_lock(&rdtgroup_mutex);
620 len = seq_buf_used(&last_cmd_status);
621 if (len)
622 seq_printf(seq, "%.*s", len, last_cmd_status_buf);
623 else
624 seq_puts(seq, "ok\n");
625 mutex_unlock(&rdtgroup_mutex);
626 return 0;
627}
628
629static int rdt_num_closids_show(struct kernfs_open_file *of,
630 struct seq_file *seq, void *v)
631{
632 struct rdt_resource *r = of->kn->parent->priv;
633
634 seq_printf(seq, "%d\n", r->num_closid);
635 return 0;
636}
637
638static int rdt_default_ctrl_show(struct kernfs_open_file *of,
639 struct seq_file *seq, void *v)
640{
641 struct rdt_resource *r = of->kn->parent->priv;
642
643 seq_printf(seq, "%x\n", r->default_ctrl);
644 return 0;
645}
646
647static int rdt_min_cbm_bits_show(struct kernfs_open_file *of,
648 struct seq_file *seq, void *v)
649{
650 struct rdt_resource *r = of->kn->parent->priv;
651
652 seq_printf(seq, "%u\n", r->cache.min_cbm_bits);
653 return 0;
654}
655
656static int rdt_shareable_bits_show(struct kernfs_open_file *of,
657 struct seq_file *seq, void *v)
658{
659 struct rdt_resource *r = of->kn->parent->priv;
660
661 seq_printf(seq, "%x\n", r->cache.shareable_bits);
662 return 0;
663}
664
665static int rdt_min_bw_show(struct kernfs_open_file *of,
666 struct seq_file *seq, void *v)
667{
668 struct rdt_resource *r = of->kn->parent->priv;
669
670 seq_printf(seq, "%u\n", r->membw.min_bw);
671 return 0;
672}
673
674static int rdt_num_rmids_show(struct kernfs_open_file *of,
675 struct seq_file *seq, void *v)
676{
677 struct rdt_resource *r = of->kn->parent->priv;
678
679 seq_printf(seq, "%d\n", r->num_rmid);
680
681 return 0;
682}
683
684static int rdt_mon_features_show(struct kernfs_open_file *of,
685 struct seq_file *seq, void *v)
686{
687 struct rdt_resource *r = of->kn->parent->priv;
688 struct mon_evt *mevt;
689
690 list_for_each_entry(mevt, &r->evt_list, list)
691 seq_printf(seq, "%s\n", mevt->name);
692
693 return 0;
694}
695
696static int rdt_bw_gran_show(struct kernfs_open_file *of,
697 struct seq_file *seq, void *v)
698{
699 struct rdt_resource *r = of->kn->parent->priv;
700
701 seq_printf(seq, "%u\n", r->membw.bw_gran);
702 return 0;
703}
704
705static int rdt_delay_linear_show(struct kernfs_open_file *of,
706 struct seq_file *seq, void *v)
707{
708 struct rdt_resource *r = of->kn->parent->priv;
709
710 seq_printf(seq, "%u\n", r->membw.delay_linear);
711 return 0;
712}
713
714static int max_threshold_occ_show(struct kernfs_open_file *of,
715 struct seq_file *seq, void *v)
716{
717 struct rdt_resource *r = of->kn->parent->priv;
718
719 seq_printf(seq, "%u\n", intel_cqm_threshold * r->mon_scale);
720
721 return 0;
722}
723
724static ssize_t max_threshold_occ_write(struct kernfs_open_file *of,
725 char *buf, size_t nbytes, loff_t off)
726{
727 struct rdt_resource *r = of->kn->parent->priv;
728 unsigned int bytes;
729 int ret;
730
731 ret = kstrtouint(buf, 0, &bytes);
732 if (ret)
733 return ret;
734
735 if (bytes > (boot_cpu_data.x86_cache_size * 1024))
736 return -EINVAL;
737
738 intel_cqm_threshold = bytes / r->mon_scale;
739
740 return nbytes;
741}
742
743/* rdtgroup information files for one cache resource. */
744static struct rftype res_common_files[] = {
745 {
746 .name = "last_cmd_status",
747 .mode = 0444,
748 .kf_ops = &rdtgroup_kf_single_ops,
749 .seq_show = rdt_last_cmd_status_show,
750 .fflags = RF_TOP_INFO,
751 },
752 {
753 .name = "num_closids",
754 .mode = 0444,
755 .kf_ops = &rdtgroup_kf_single_ops,
756 .seq_show = rdt_num_closids_show,
757 .fflags = RF_CTRL_INFO,
758 },
759 {
760 .name = "mon_features",
761 .mode = 0444,
762 .kf_ops = &rdtgroup_kf_single_ops,
763 .seq_show = rdt_mon_features_show,
764 .fflags = RF_MON_INFO,
765 },
766 {
767 .name = "num_rmids",
768 .mode = 0444,
769 .kf_ops = &rdtgroup_kf_single_ops,
770 .seq_show = rdt_num_rmids_show,
771 .fflags = RF_MON_INFO,
772 },
773 {
774 .name = "cbm_mask",
775 .mode = 0444,
776 .kf_ops = &rdtgroup_kf_single_ops,
777 .seq_show = rdt_default_ctrl_show,
778 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
779 },
780 {
781 .name = "min_cbm_bits",
782 .mode = 0444,
783 .kf_ops = &rdtgroup_kf_single_ops,
784 .seq_show = rdt_min_cbm_bits_show,
785 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
786 },
787 {
788 .name = "shareable_bits",
789 .mode = 0444,
790 .kf_ops = &rdtgroup_kf_single_ops,
791 .seq_show = rdt_shareable_bits_show,
792 .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE,
793 },
794 {
795 .name = "min_bandwidth",
796 .mode = 0444,
797 .kf_ops = &rdtgroup_kf_single_ops,
798 .seq_show = rdt_min_bw_show,
799 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
800 },
801 {
802 .name = "bandwidth_gran",
803 .mode = 0444,
804 .kf_ops = &rdtgroup_kf_single_ops,
805 .seq_show = rdt_bw_gran_show,
806 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
807 },
808 {
809 .name = "delay_linear",
810 .mode = 0444,
811 .kf_ops = &rdtgroup_kf_single_ops,
812 .seq_show = rdt_delay_linear_show,
813 .fflags = RF_CTRL_INFO | RFTYPE_RES_MB,
814 },
815 {
816 .name = "max_threshold_occupancy",
817 .mode = 0644,
818 .kf_ops = &rdtgroup_kf_single_ops,
819 .write = max_threshold_occ_write,
820 .seq_show = max_threshold_occ_show,
821 .fflags = RF_MON_INFO | RFTYPE_RES_CACHE,
822 },
823 {
824 .name = "cpus",
825 .mode = 0644,
826 .kf_ops = &rdtgroup_kf_single_ops,
827 .write = rdtgroup_cpus_write,
828 .seq_show = rdtgroup_cpus_show,
829 .fflags = RFTYPE_BASE,
830 },
831 {
832 .name = "cpus_list",
833 .mode = 0644,
834 .kf_ops = &rdtgroup_kf_single_ops,
835 .write = rdtgroup_cpus_write,
836 .seq_show = rdtgroup_cpus_show,
837 .flags = RFTYPE_FLAGS_CPUS_LIST,
838 .fflags = RFTYPE_BASE,
839 },
840 {
841 .name = "tasks",
842 .mode = 0644,
843 .kf_ops = &rdtgroup_kf_single_ops,
844 .write = rdtgroup_tasks_write,
845 .seq_show = rdtgroup_tasks_show,
846 .fflags = RFTYPE_BASE,
847 },
848 {
849 .name = "schemata",
850 .mode = 0644,
851 .kf_ops = &rdtgroup_kf_single_ops,
852 .write = rdtgroup_schemata_write,
853 .seq_show = rdtgroup_schemata_show,
854 .fflags = RF_CTRL_BASE,
855 },
856};
857
858static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags)
859{
860 struct rftype *rfts, *rft;
861 int ret, len;
862
863 rfts = res_common_files;
864 len = ARRAY_SIZE(res_common_files);
865
866 lockdep_assert_held(&rdtgroup_mutex);
867
868 for (rft = rfts; rft < rfts + len; rft++) {
869 if ((fflags & rft->fflags) == rft->fflags) {
870 ret = rdtgroup_add_file(kn, rft);
871 if (ret)
872 goto error;
873 }
874 }
875
876 return 0;
877error:
878 pr_warn("Failed to add %s, err=%d\n", rft->name, ret);
879 while (--rft >= rfts) {
880 if ((fflags & rft->fflags) == rft->fflags)
881 kernfs_remove_by_name(kn, rft->name);
882 }
883 return ret;
884}
885
886static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name,
887 unsigned long fflags)
888{
889 struct kernfs_node *kn_subdir;
890 int ret;
891
892 kn_subdir = kernfs_create_dir(kn_info, name,
893 kn_info->mode, r);
894 if (IS_ERR(kn_subdir))
895 return PTR_ERR(kn_subdir);
896
897 kernfs_get(kn_subdir);
898 ret = rdtgroup_kn_set_ugid(kn_subdir);
899 if (ret)
900 return ret;
901
902 ret = rdtgroup_add_files(kn_subdir, fflags);
903 if (!ret)
904 kernfs_activate(kn_subdir);
905
906 return ret;
907}
908
909static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn)
910{
911 struct rdt_resource *r;
912 unsigned long fflags;
913 char name[32];
914 int ret;
915
916 /* create the directory */
917 kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL);
918 if (IS_ERR(kn_info))
919 return PTR_ERR(kn_info);
920 kernfs_get(kn_info);
921
922 ret = rdtgroup_add_files(kn_info, RF_TOP_INFO);
923 if (ret)
924 goto out_destroy;
925
926 for_each_alloc_enabled_rdt_resource(r) {
927 fflags = r->fflags | RF_CTRL_INFO;
928 ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags);
929 if (ret)
930 goto out_destroy;
931 }
932
933 for_each_mon_enabled_rdt_resource(r) {
934 fflags = r->fflags | RF_MON_INFO;
935 sprintf(name, "%s_MON", r->name);
936 ret = rdtgroup_mkdir_info_resdir(r, name, fflags);
937 if (ret)
938 goto out_destroy;
939 }
940
941 /*
942 * This extra ref will be put in kernfs_remove() and guarantees
943 * that @rdtgrp->kn is always accessible.
944 */
945 kernfs_get(kn_info);
946
947 ret = rdtgroup_kn_set_ugid(kn_info);
948 if (ret)
949 goto out_destroy;
950
951 kernfs_activate(kn_info);
952
953 return 0;
954
955out_destroy:
956 kernfs_remove(kn_info);
957 return ret;
958}
959
960static int
961mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp,
962 char *name, struct kernfs_node **dest_kn)
963{
964 struct kernfs_node *kn;
965 int ret;
966
967 /* create the directory */
968 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
969 if (IS_ERR(kn))
970 return PTR_ERR(kn);
971
972 if (dest_kn)
973 *dest_kn = kn;
974
975 /*
976 * This extra ref will be put in kernfs_remove() and guarantees
977 * that @rdtgrp->kn is always accessible.
978 */
979 kernfs_get(kn);
980
981 ret = rdtgroup_kn_set_ugid(kn);
982 if (ret)
983 goto out_destroy;
984
985 kernfs_activate(kn);
986
987 return 0;
988
989out_destroy:
990 kernfs_remove(kn);
991 return ret;
992}
993
994static void l3_qos_cfg_update(void *arg)
995{
996 bool *enable = arg;
997
998 wrmsrl(IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL);
999}
1000
1001static void l2_qos_cfg_update(void *arg)
1002{
1003 bool *enable = arg;
1004
1005 wrmsrl(IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL);
1006}
1007
1008static int set_cache_qos_cfg(int level, bool enable)
1009{
1010 void (*update)(void *arg);
1011 struct rdt_resource *r_l;
1012 cpumask_var_t cpu_mask;
1013 struct rdt_domain *d;
1014 int cpu;
1015
1016 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1017 return -ENOMEM;
1018
1019 if (level == RDT_RESOURCE_L3)
1020 update = l3_qos_cfg_update;
1021 else if (level == RDT_RESOURCE_L2)
1022 update = l2_qos_cfg_update;
1023 else
1024 return -EINVAL;
1025
1026 r_l = &rdt_resources_all[level];
1027 list_for_each_entry(d, &r_l->domains, list) {
1028 /* Pick one CPU from each domain instance to update MSR */
1029 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1030 }
1031 cpu = get_cpu();
1032 /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1033 if (cpumask_test_cpu(cpu, cpu_mask))
1034 update(&enable);
1035 /* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1036 smp_call_function_many(cpu_mask, update, &enable, 1);
1037 put_cpu();
1038
1039 free_cpumask_var(cpu_mask);
1040
1041 return 0;
1042}
1043
1044static int cdp_enable(int level, int data_type, int code_type)
1045{
1046 struct rdt_resource *r_ldata = &rdt_resources_all[data_type];
1047 struct rdt_resource *r_lcode = &rdt_resources_all[code_type];
1048 struct rdt_resource *r_l = &rdt_resources_all[level];
1049 int ret;
1050
1051 if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1052 !r_lcode->alloc_capable)
1053 return -EINVAL;
1054
1055 ret = set_cache_qos_cfg(level, true);
1056 if (!ret) {
1057 r_l->alloc_enabled = false;
1058 r_ldata->alloc_enabled = true;
1059 r_lcode->alloc_enabled = true;
1060 }
1061 return ret;
1062}
1063
1064static int cdpl3_enable(void)
1065{
1066 return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1067 RDT_RESOURCE_L3CODE);
1068}
1069
1070static int cdpl2_enable(void)
1071{
1072 return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1073 RDT_RESOURCE_L2CODE);
1074}
1075
1076static void cdp_disable(int level, int data_type, int code_type)
1077{
1078 struct rdt_resource *r = &rdt_resources_all[level];
1079
1080 r->alloc_enabled = r->alloc_capable;
1081
1082 if (rdt_resources_all[data_type].alloc_enabled) {
1083 rdt_resources_all[data_type].alloc_enabled = false;
1084 rdt_resources_all[code_type].alloc_enabled = false;
1085 set_cache_qos_cfg(level, false);
1086 }
1087}
1088
1089static void cdpl3_disable(void)
1090{
1091 cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
1092}
1093
1094static void cdpl2_disable(void)
1095{
1096 cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
1097}
1098
1099static void cdp_disable_all(void)
1100{
1101 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
1102 cdpl3_disable();
1103 if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
1104 cdpl2_disable();
1105}
1106
1107static int parse_rdtgroupfs_options(char *data)
1108{
1109 char *token, *o = data;
1110 int ret = 0;
1111
1112 while ((token = strsep(&o, ",")) != NULL) {
1113 if (!*token) {
1114 ret = -EINVAL;
1115 goto out;
1116 }
1117
1118 if (!strcmp(token, "cdp")) {
1119 ret = cdpl3_enable();
1120 if (ret)
1121 goto out;
1122 } else if (!strcmp(token, "cdpl2")) {
1123 ret = cdpl2_enable();
1124 if (ret)
1125 goto out;
1126 } else {
1127 ret = -EINVAL;
1128 goto out;
1129 }
1130 }
1131
1132 return 0;
1133
1134out:
1135 pr_err("Invalid mount option \"%s\"\n", token);
1136
1137 return ret;
1138}
1139
1140/*
1141 * We don't allow rdtgroup directories to be created anywhere
1142 * except the root directory. Thus when looking for the rdtgroup
1143 * structure for a kernfs node we are either looking at a directory,
1144 * in which case the rdtgroup structure is pointed at by the "priv"
1145 * field, otherwise we have a file, and need only look to the parent
1146 * to find the rdtgroup.
1147 */
1148static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
1149{
1150 if (kernfs_type(kn) == KERNFS_DIR) {
1151 /*
1152 * All the resource directories use "kn->priv"
1153 * to point to the "struct rdtgroup" for the
1154 * resource. "info" and its subdirectories don't
1155 * have rdtgroup structures, so return NULL here.
1156 */
1157 if (kn == kn_info || kn->parent == kn_info)
1158 return NULL;
1159 else
1160 return kn->priv;
1161 } else {
1162 return kn->parent->priv;
1163 }
1164}
1165
1166struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
1167{
1168 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1169
1170 if (!rdtgrp)
1171 return NULL;
1172
1173 atomic_inc(&rdtgrp->waitcount);
1174 kernfs_break_active_protection(kn);
1175
1176 mutex_lock(&rdtgroup_mutex);
1177
1178 /* Was this group deleted while we waited? */
1179 if (rdtgrp->flags & RDT_DELETED)
1180 return NULL;
1181
1182 return rdtgrp;
1183}
1184
1185void rdtgroup_kn_unlock(struct kernfs_node *kn)
1186{
1187 struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1188
1189 if (!rdtgrp)
1190 return;
1191
1192 mutex_unlock(&rdtgroup_mutex);
1193
1194 if (atomic_dec_and_test(&rdtgrp->waitcount) &&
1195 (rdtgrp->flags & RDT_DELETED)) {
1196 kernfs_unbreak_active_protection(kn);
1197 kernfs_put(rdtgrp->kn);
1198 kfree(rdtgrp);
1199 } else {
1200 kernfs_unbreak_active_protection(kn);
1201 }
1202}
1203
1204static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1205 struct rdtgroup *prgrp,
1206 struct kernfs_node **mon_data_kn);
1207
1208static struct dentry *rdt_mount(struct file_system_type *fs_type,
1209 int flags, const char *unused_dev_name,
1210 void *data)
1211{
1212 struct rdt_domain *dom;
1213 struct rdt_resource *r;
1214 struct dentry *dentry;
1215 int ret;
1216
1217 cpus_read_lock();
1218 mutex_lock(&rdtgroup_mutex);
1219 /*
1220 * resctrl file system can only be mounted once.
1221 */
1222 if (static_branch_unlikely(&rdt_enable_key)) {
1223 dentry = ERR_PTR(-EBUSY);
1224 goto out;
1225 }
1226
1227 ret = parse_rdtgroupfs_options(data);
1228 if (ret) {
1229 dentry = ERR_PTR(ret);
1230 goto out_cdp;
1231 }
1232
1233 closid_init();
1234
1235 ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
1236 if (ret) {
1237 dentry = ERR_PTR(ret);
1238 goto out_cdp;
1239 }
1240
1241 if (rdt_mon_capable) {
1242 ret = mongroup_create_dir(rdtgroup_default.kn,
1243 NULL, "mon_groups",
1244 &kn_mongrp);
1245 if (ret) {
1246 dentry = ERR_PTR(ret);
1247 goto out_info;
1248 }
1249 kernfs_get(kn_mongrp);
1250
1251 ret = mkdir_mondata_all(rdtgroup_default.kn,
1252 &rdtgroup_default, &kn_mondata);
1253 if (ret) {
1254 dentry = ERR_PTR(ret);
1255 goto out_mongrp;
1256 }
1257 kernfs_get(kn_mondata);
1258 rdtgroup_default.mon.mon_data_kn = kn_mondata;
1259 }
1260
1261 dentry = kernfs_mount(fs_type, flags, rdt_root,
1262 RDTGROUP_SUPER_MAGIC, NULL);
1263 if (IS_ERR(dentry))
1264 goto out_mondata;
1265
1266 if (rdt_alloc_capable)
1267 static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
1268 if (rdt_mon_capable)
1269 static_branch_enable_cpuslocked(&rdt_mon_enable_key);
1270
1271 if (rdt_alloc_capable || rdt_mon_capable)
1272 static_branch_enable_cpuslocked(&rdt_enable_key);
1273
1274 if (is_mbm_enabled()) {
1275 r = &rdt_resources_all[RDT_RESOURCE_L3];
1276 list_for_each_entry(dom, &r->domains, list)
1277 mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
1278 }
1279
1280 goto out;
1281
1282out_mondata:
1283 if (rdt_mon_capable)
1284 kernfs_remove(kn_mondata);
1285out_mongrp:
1286 if (rdt_mon_capable)
1287 kernfs_remove(kn_mongrp);
1288out_info:
1289 kernfs_remove(kn_info);
1290out_cdp:
1291 cdp_disable_all();
1292out:
1293 rdt_last_cmd_clear();
1294 mutex_unlock(&rdtgroup_mutex);
1295 cpus_read_unlock();
1296
1297 return dentry;
1298}
1299
1300static int reset_all_ctrls(struct rdt_resource *r)
1301{
1302 struct msr_param msr_param;
1303 cpumask_var_t cpu_mask;
1304 struct rdt_domain *d;
1305 int i, cpu;
1306
1307 if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1308 return -ENOMEM;
1309
1310 msr_param.res = r;
1311 msr_param.low = 0;
1312 msr_param.high = r->num_closid;
1313
1314 /*
1315 * Disable resource control for this resource by setting all
1316 * CBMs in all domains to the maximum mask value. Pick one CPU
1317 * from each domain to update the MSRs below.
1318 */
1319 list_for_each_entry(d, &r->domains, list) {
1320 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1321
1322 for (i = 0; i < r->num_closid; i++)
1323 d->ctrl_val[i] = r->default_ctrl;
1324 }
1325 cpu = get_cpu();
1326 /* Update CBM on this cpu if it's in cpu_mask. */
1327 if (cpumask_test_cpu(cpu, cpu_mask))
1328 rdt_ctrl_update(&msr_param);
1329 /* Update CBM on all other cpus in cpu_mask. */
1330 smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
1331 put_cpu();
1332
1333 free_cpumask_var(cpu_mask);
1334
1335 return 0;
1336}
1337
1338static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
1339{
1340 return (rdt_alloc_capable &&
1341 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
1342}
1343
1344static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
1345{
1346 return (rdt_mon_capable &&
1347 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
1348}
1349
1350/*
1351 * Move tasks from one to the other group. If @from is NULL, then all tasks
1352 * in the systems are moved unconditionally (used for teardown).
1353 *
1354 * If @mask is not NULL the cpus on which moved tasks are running are set
1355 * in that mask so the update smp function call is restricted to affected
1356 * cpus.
1357 */
1358static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
1359 struct cpumask *mask)
1360{
1361 struct task_struct *p, *t;
1362
1363 read_lock(&tasklist_lock);
1364 for_each_process_thread(p, t) {
1365 if (!from || is_closid_match(t, from) ||
1366 is_rmid_match(t, from)) {
1367 t->closid = to->closid;
1368 t->rmid = to->mon.rmid;
1369
1370#ifdef CONFIG_SMP
1371 /*
1372 * This is safe on x86 w/o barriers as the ordering
1373 * of writing to task_cpu() and t->on_cpu is
1374 * reverse to the reading here. The detection is
1375 * inaccurate as tasks might move or schedule
1376 * before the smp function call takes place. In
1377 * such a case the function call is pointless, but
1378 * there is no other side effect.
1379 */
1380 if (mask && t->on_cpu)
1381 cpumask_set_cpu(task_cpu(t), mask);
1382#endif
1383 }
1384 }
1385 read_unlock(&tasklist_lock);
1386}
1387
1388static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
1389{
1390 struct rdtgroup *sentry, *stmp;
1391 struct list_head *head;
1392
1393 head = &rdtgrp->mon.crdtgrp_list;
1394 list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
1395 free_rmid(sentry->mon.rmid);
1396 list_del(&sentry->mon.crdtgrp_list);
1397 kfree(sentry);
1398 }
1399}
1400
1401/*
1402 * Forcibly remove all of subdirectories under root.
1403 */
1404static void rmdir_all_sub(void)
1405{
1406 struct rdtgroup *rdtgrp, *tmp;
1407
1408 /* Move all tasks to the default resource group */
1409 rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
1410
1411 list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
1412 /* Free any child rmids */
1413 free_all_child_rdtgrp(rdtgrp);
1414
1415 /* Remove each rdtgroup other than root */
1416 if (rdtgrp == &rdtgroup_default)
1417 continue;
1418
1419 /*
1420 * Give any CPUs back to the default group. We cannot copy
1421 * cpu_online_mask because a CPU might have executed the
1422 * offline callback already, but is still marked online.
1423 */
1424 cpumask_or(&rdtgroup_default.cpu_mask,
1425 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
1426
1427 free_rmid(rdtgrp->mon.rmid);
1428
1429 kernfs_remove(rdtgrp->kn);
1430 list_del(&rdtgrp->rdtgroup_list);
1431 kfree(rdtgrp);
1432 }
1433 /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
1434 update_closid_rmid(cpu_online_mask, &rdtgroup_default);
1435
1436 kernfs_remove(kn_info);
1437 kernfs_remove(kn_mongrp);
1438 kernfs_remove(kn_mondata);
1439}
1440
1441static void rdt_kill_sb(struct super_block *sb)
1442{
1443 struct rdt_resource *r;
1444
1445 cpus_read_lock();
1446 mutex_lock(&rdtgroup_mutex);
1447
1448 /*Put everything back to default values. */
1449 for_each_alloc_enabled_rdt_resource(r)
1450 reset_all_ctrls(r);
1451 cdp_disable_all();
1452 rmdir_all_sub();
1453 static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
1454 static_branch_disable_cpuslocked(&rdt_mon_enable_key);
1455 static_branch_disable_cpuslocked(&rdt_enable_key);
1456 kernfs_kill_sb(sb);
1457 mutex_unlock(&rdtgroup_mutex);
1458 cpus_read_unlock();
1459}
1460
1461static struct file_system_type rdt_fs_type = {
1462 .name = "resctrl",
1463 .mount = rdt_mount,
1464 .kill_sb = rdt_kill_sb,
1465};
1466
1467static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
1468 void *priv)
1469{
1470 struct kernfs_node *kn;
1471 int ret = 0;
1472
1473 kn = __kernfs_create_file(parent_kn, name, 0444, 0,
1474 &kf_mondata_ops, priv, NULL, NULL);
1475 if (IS_ERR(kn))
1476 return PTR_ERR(kn);
1477
1478 ret = rdtgroup_kn_set_ugid(kn);
1479 if (ret) {
1480 kernfs_remove(kn);
1481 return ret;
1482 }
1483
1484 return ret;
1485}
1486
1487/*
1488 * Remove all subdirectories of mon_data of ctrl_mon groups
1489 * and monitor groups with given domain id.
1490 */
1491void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
1492{
1493 struct rdtgroup *prgrp, *crgrp;
1494 char name[32];
1495
1496 if (!r->mon_enabled)
1497 return;
1498
1499 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
1500 sprintf(name, "mon_%s_%02d", r->name, dom_id);
1501 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
1502
1503 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
1504 kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
1505 }
1506}
1507
1508static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
1509 struct rdt_domain *d,
1510 struct rdt_resource *r, struct rdtgroup *prgrp)
1511{
1512 union mon_data_bits priv;
1513 struct kernfs_node *kn;
1514 struct mon_evt *mevt;
1515 struct rmid_read rr;
1516 char name[32];
1517 int ret;
1518
1519 sprintf(name, "mon_%s_%02d", r->name, d->id);
1520 /* create the directory */
1521 kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1522 if (IS_ERR(kn))
1523 return PTR_ERR(kn);
1524
1525 /*
1526 * This extra ref will be put in kernfs_remove() and guarantees
1527 * that kn is always accessible.
1528 */
1529 kernfs_get(kn);
1530 ret = rdtgroup_kn_set_ugid(kn);
1531 if (ret)
1532 goto out_destroy;
1533
1534 if (WARN_ON(list_empty(&r->evt_list))) {
1535 ret = -EPERM;
1536 goto out_destroy;
1537 }
1538
1539 priv.u.rid = r->rid;
1540 priv.u.domid = d->id;
1541 list_for_each_entry(mevt, &r->evt_list, list) {
1542 priv.u.evtid = mevt->evtid;
1543 ret = mon_addfile(kn, mevt->name, priv.priv);
1544 if (ret)
1545 goto out_destroy;
1546
1547 if (is_mbm_event(mevt->evtid))
1548 mon_event_read(&rr, d, prgrp, mevt->evtid, true);
1549 }
1550 kernfs_activate(kn);
1551 return 0;
1552
1553out_destroy:
1554 kernfs_remove(kn);
1555 return ret;
1556}
1557
1558/*
1559 * Add all subdirectories of mon_data for "ctrl_mon" groups
1560 * and "monitor" groups with given domain id.
1561 */
1562void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
1563 struct rdt_domain *d)
1564{
1565 struct kernfs_node *parent_kn;
1566 struct rdtgroup *prgrp, *crgrp;
1567 struct list_head *head;
1568
1569 if (!r->mon_enabled)
1570 return;
1571
1572 list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
1573 parent_kn = prgrp->mon.mon_data_kn;
1574 mkdir_mondata_subdir(parent_kn, d, r, prgrp);
1575
1576 head = &prgrp->mon.crdtgrp_list;
1577 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
1578 parent_kn = crgrp->mon.mon_data_kn;
1579 mkdir_mondata_subdir(parent_kn, d, r, crgrp);
1580 }
1581 }
1582}
1583
1584static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
1585 struct rdt_resource *r,
1586 struct rdtgroup *prgrp)
1587{
1588 struct rdt_domain *dom;
1589 int ret;
1590
1591 list_for_each_entry(dom, &r->domains, list) {
1592 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
1593 if (ret)
1594 return ret;
1595 }
1596
1597 return 0;
1598}
1599
1600/*
1601 * This creates a directory mon_data which contains the monitored data.
1602 *
1603 * mon_data has one directory for each domain whic are named
1604 * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
1605 * with L3 domain looks as below:
1606 * ./mon_data:
1607 * mon_L3_00
1608 * mon_L3_01
1609 * mon_L3_02
1610 * ...
1611 *
1612 * Each domain directory has one file per event:
1613 * ./mon_L3_00/:
1614 * llc_occupancy
1615 *
1616 */
1617static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1618 struct rdtgroup *prgrp,
1619 struct kernfs_node **dest_kn)
1620{
1621 struct rdt_resource *r;
1622 struct kernfs_node *kn;
1623 int ret;
1624
1625 /*
1626 * Create the mon_data directory first.
1627 */
1628 ret = mongroup_create_dir(parent_kn, NULL, "mon_data", &kn);
1629 if (ret)
1630 return ret;
1631
1632 if (dest_kn)
1633 *dest_kn = kn;
1634
1635 /*
1636 * Create the subdirectories for each domain. Note that all events
1637 * in a domain like L3 are grouped into a resource whose domain is L3
1638 */
1639 for_each_mon_enabled_rdt_resource(r) {
1640 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
1641 if (ret)
1642 goto out_destroy;
1643 }
1644
1645 return 0;
1646
1647out_destroy:
1648 kernfs_remove(kn);
1649 return ret;
1650}
1651
1652static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
1653 struct kernfs_node *prgrp_kn,
1654 const char *name, umode_t mode,
1655 enum rdt_group_type rtype, struct rdtgroup **r)
1656{
1657 struct rdtgroup *prdtgrp, *rdtgrp;
1658 struct kernfs_node *kn;
1659 uint files = 0;
1660 int ret;
1661
1662 prdtgrp = rdtgroup_kn_lock_live(prgrp_kn);
1663 rdt_last_cmd_clear();
1664 if (!prdtgrp) {
1665 ret = -ENODEV;
1666 rdt_last_cmd_puts("directory was removed\n");
1667 goto out_unlock;
1668 }
1669
1670 /* allocate the rdtgroup. */
1671 rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
1672 if (!rdtgrp) {
1673 ret = -ENOSPC;
1674 rdt_last_cmd_puts("kernel out of memory\n");
1675 goto out_unlock;
1676 }
1677 *r = rdtgrp;
1678 rdtgrp->mon.parent = prdtgrp;
1679 rdtgrp->type = rtype;
1680 INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
1681
1682 /* kernfs creates the directory for rdtgrp */
1683 kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
1684 if (IS_ERR(kn)) {
1685 ret = PTR_ERR(kn);
1686 rdt_last_cmd_puts("kernfs create error\n");
1687 goto out_free_rgrp;
1688 }
1689 rdtgrp->kn = kn;
1690
1691 /*
1692 * kernfs_remove() will drop the reference count on "kn" which
1693 * will free it. But we still need it to stick around for the
1694 * rdtgroup_kn_unlock(kn} call below. Take one extra reference
1695 * here, which will be dropped inside rdtgroup_kn_unlock().
1696 */
1697 kernfs_get(kn);
1698
1699 ret = rdtgroup_kn_set_ugid(kn);
1700 if (ret) {
1701 rdt_last_cmd_puts("kernfs perm error\n");
1702 goto out_destroy;
1703 }
1704
1705 files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
1706 ret = rdtgroup_add_files(kn, files);
1707 if (ret) {
1708 rdt_last_cmd_puts("kernfs fill error\n");
1709 goto out_destroy;
1710 }
1711
1712 if (rdt_mon_capable) {
1713 ret = alloc_rmid();
1714 if (ret < 0) {
1715 rdt_last_cmd_puts("out of RMIDs\n");
1716 goto out_destroy;
1717 }
1718 rdtgrp->mon.rmid = ret;
1719
1720 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
1721 if (ret) {
1722 rdt_last_cmd_puts("kernfs subdir error\n");
1723 goto out_idfree;
1724 }
1725 }
1726 kernfs_activate(kn);
1727
1728 /*
1729 * The caller unlocks the prgrp_kn upon success.
1730 */
1731 return 0;
1732
1733out_idfree:
1734 free_rmid(rdtgrp->mon.rmid);
1735out_destroy:
1736 kernfs_remove(rdtgrp->kn);
1737out_free_rgrp:
1738 kfree(rdtgrp);
1739out_unlock:
1740 rdtgroup_kn_unlock(prgrp_kn);
1741 return ret;
1742}
1743
1744static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
1745{
1746 kernfs_remove(rgrp->kn);
1747 free_rmid(rgrp->mon.rmid);
1748 kfree(rgrp);
1749}
1750
1751/*
1752 * Create a monitor group under "mon_groups" directory of a control
1753 * and monitor group(ctrl_mon). This is a resource group
1754 * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
1755 */
1756static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
1757 struct kernfs_node *prgrp_kn,
1758 const char *name,
1759 umode_t mode)
1760{
1761 struct rdtgroup *rdtgrp, *prgrp;
1762 int ret;
1763
1764 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP,
1765 &rdtgrp);
1766 if (ret)
1767 return ret;
1768
1769 prgrp = rdtgrp->mon.parent;
1770 rdtgrp->closid = prgrp->closid;
1771
1772 /*
1773 * Add the rdtgrp to the list of rdtgrps the parent
1774 * ctrl_mon group has to track.
1775 */
1776 list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
1777
1778 rdtgroup_kn_unlock(prgrp_kn);
1779 return ret;
1780}
1781
1782/*
1783 * These are rdtgroups created under the root directory. Can be used
1784 * to allocate and monitor resources.
1785 */
1786static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
1787 struct kernfs_node *prgrp_kn,
1788 const char *name, umode_t mode)
1789{
1790 struct rdtgroup *rdtgrp;
1791 struct kernfs_node *kn;
1792 u32 closid;
1793 int ret;
1794
1795 ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP,
1796 &rdtgrp);
1797 if (ret)
1798 return ret;
1799
1800 kn = rdtgrp->kn;
1801 ret = closid_alloc();
1802 if (ret < 0) {
1803 rdt_last_cmd_puts("out of CLOSIDs\n");
1804 goto out_common_fail;
1805 }
1806 closid = ret;
1807 ret = 0;
1808
1809 rdtgrp->closid = closid;
1810 list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
1811
1812 if (rdt_mon_capable) {
1813 /*
1814 * Create an empty mon_groups directory to hold the subset
1815 * of tasks and cpus to monitor.
1816 */
1817 ret = mongroup_create_dir(kn, NULL, "mon_groups", NULL);
1818 if (ret) {
1819 rdt_last_cmd_puts("kernfs subdir error\n");
1820 goto out_id_free;
1821 }
1822 }
1823
1824 goto out_unlock;
1825
1826out_id_free:
1827 closid_free(closid);
1828 list_del(&rdtgrp->rdtgroup_list);
1829out_common_fail:
1830 mkdir_rdt_prepare_clean(rdtgrp);
1831out_unlock:
1832 rdtgroup_kn_unlock(prgrp_kn);
1833 return ret;
1834}
1835
1836/*
1837 * We allow creating mon groups only with in a directory called "mon_groups"
1838 * which is present in every ctrl_mon group. Check if this is a valid
1839 * "mon_groups" directory.
1840 *
1841 * 1. The directory should be named "mon_groups".
1842 * 2. The mon group itself should "not" be named "mon_groups".
1843 * This makes sure "mon_groups" directory always has a ctrl_mon group
1844 * as parent.
1845 */
1846static bool is_mon_groups(struct kernfs_node *kn, const char *name)
1847{
1848 return (!strcmp(kn->name, "mon_groups") &&
1849 strcmp(name, "mon_groups"));
1850}
1851
1852static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
1853 umode_t mode)
1854{
1855 /* Do not accept '\n' to avoid unparsable situation. */
1856 if (strchr(name, '\n'))
1857 return -EINVAL;
1858
1859 /*
1860 * If the parent directory is the root directory and RDT
1861 * allocation is supported, add a control and monitoring
1862 * subdirectory
1863 */
1864 if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
1865 return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode);
1866
1867 /*
1868 * If RDT monitoring is supported and the parent directory is a valid
1869 * "mon_groups" directory, add a monitoring subdirectory.
1870 */
1871 if (rdt_mon_capable && is_mon_groups(parent_kn, name))
1872 return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode);
1873
1874 return -EPERM;
1875}
1876
1877static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
1878 cpumask_var_t tmpmask)
1879{
1880 struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
1881 int cpu;
1882
1883 /* Give any tasks back to the parent group */
1884 rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
1885
1886 /* Update per cpu rmid of the moved CPUs first */
1887 for_each_cpu(cpu, &rdtgrp->cpu_mask)
1888 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
1889 /*
1890 * Update the MSR on moved CPUs and CPUs which have moved
1891 * task running on them.
1892 */
1893 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
1894 update_closid_rmid(tmpmask, NULL);
1895
1896 rdtgrp->flags = RDT_DELETED;
1897 free_rmid(rdtgrp->mon.rmid);
1898
1899 /*
1900 * Remove the rdtgrp from the parent ctrl_mon group's list
1901 */
1902 WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
1903 list_del(&rdtgrp->mon.crdtgrp_list);
1904
1905 /*
1906 * one extra hold on this, will drop when we kfree(rdtgrp)
1907 * in rdtgroup_kn_unlock()
1908 */
1909 kernfs_get(kn);
1910 kernfs_remove(rdtgrp->kn);
1911
1912 return 0;
1913}
1914
1915static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
1916 cpumask_var_t tmpmask)
1917{
1918 int cpu;
1919
1920 /* Give any tasks back to the default group */
1921 rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
1922
1923 /* Give any CPUs back to the default group */
1924 cpumask_or(&rdtgroup_default.cpu_mask,
1925 &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
1926
1927 /* Update per cpu closid and rmid of the moved CPUs first */
1928 for_each_cpu(cpu, &rdtgrp->cpu_mask) {
1929 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
1930 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
1931 }
1932
1933 /*
1934 * Update the MSR on moved CPUs and CPUs which have moved
1935 * task running on them.
1936 */
1937 cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
1938 update_closid_rmid(tmpmask, NULL);
1939
1940 rdtgrp->flags = RDT_DELETED;
1941 closid_free(rdtgrp->closid);
1942 free_rmid(rdtgrp->mon.rmid);
1943
1944 /*
1945 * Free all the child monitor group rmids.
1946 */
1947 free_all_child_rdtgrp(rdtgrp);
1948
1949 list_del(&rdtgrp->rdtgroup_list);
1950
1951 /*
1952 * one extra hold on this, will drop when we kfree(rdtgrp)
1953 * in rdtgroup_kn_unlock()
1954 */
1955 kernfs_get(kn);
1956 kernfs_remove(rdtgrp->kn);
1957
1958 return 0;
1959}
1960
1961static int rdtgroup_rmdir(struct kernfs_node *kn)
1962{
1963 struct kernfs_node *parent_kn = kn->parent;
1964 struct rdtgroup *rdtgrp;
1965 cpumask_var_t tmpmask;
1966 int ret = 0;
1967
1968 if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
1969 return -ENOMEM;
1970
1971 rdtgrp = rdtgroup_kn_lock_live(kn);
1972 if (!rdtgrp) {
1973 ret = -EPERM;
1974 goto out;
1975 }
1976
1977 /*
1978 * If the rdtgroup is a ctrl_mon group and parent directory
1979 * is the root directory, remove the ctrl_mon group.
1980 *
1981 * If the rdtgroup is a mon group and parent directory
1982 * is a valid "mon_groups" directory, remove the mon group.
1983 */
1984 if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn)
1985 ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask);
1986 else if (rdtgrp->type == RDTMON_GROUP &&
1987 is_mon_groups(parent_kn, kn->name))
1988 ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask);
1989 else
1990 ret = -EPERM;
1991
1992out:
1993 rdtgroup_kn_unlock(kn);
1994 free_cpumask_var(tmpmask);
1995 return ret;
1996}
1997
1998static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
1999{
2000 if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
2001 seq_puts(seq, ",cdp");
2002 return 0;
2003}
2004
2005static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
2006 .mkdir = rdtgroup_mkdir,
2007 .rmdir = rdtgroup_rmdir,
2008 .show_options = rdtgroup_show_options,
2009};
2010
2011static int __init rdtgroup_setup_root(void)
2012{
2013 int ret;
2014
2015 rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
2016 KERNFS_ROOT_CREATE_DEACTIVATED,
2017 &rdtgroup_default);
2018 if (IS_ERR(rdt_root))
2019 return PTR_ERR(rdt_root);
2020
2021 mutex_lock(&rdtgroup_mutex);
2022
2023 rdtgroup_default.closid = 0;
2024 rdtgroup_default.mon.rmid = 0;
2025 rdtgroup_default.type = RDTCTRL_GROUP;
2026 INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
2027
2028 list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
2029
2030 ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
2031 if (ret) {
2032 kernfs_destroy_root(rdt_root);
2033 goto out;
2034 }
2035
2036 rdtgroup_default.kn = rdt_root->kn;
2037 kernfs_activate(rdtgroup_default.kn);
2038
2039out:
2040 mutex_unlock(&rdtgroup_mutex);
2041
2042 return ret;
2043}
2044
2045/*
2046 * rdtgroup_init - rdtgroup initialization
2047 *
2048 * Setup resctrl file system including set up root, create mount point,
2049 * register rdtgroup filesystem, and initialize files under root directory.
2050 *
2051 * Return: 0 on success or -errno
2052 */
2053int __init rdtgroup_init(void)
2054{
2055 int ret = 0;
2056
2057 seq_buf_init(&last_cmd_status, last_cmd_status_buf,
2058 sizeof(last_cmd_status_buf));
2059
2060 ret = rdtgroup_setup_root();
2061 if (ret)
2062 return ret;
2063
2064 ret = sysfs_create_mount_point(fs_kobj, "resctrl");
2065 if (ret)
2066 goto cleanup_root;
2067
2068 ret = register_filesystem(&rdt_fs_type);
2069 if (ret)
2070 goto cleanup_mountpoint;
2071
2072 return 0;
2073
2074cleanup_mountpoint:
2075 sysfs_remove_mount_point(fs_kobj, "resctrl");
2076cleanup_root:
2077 kernfs_destroy_root(rdt_root);
2078
2079 return ret;
2080}