~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/kernel/cpu/intel_rdt_rdtgroup.c

Version: ~ [ linux-4.18-rc5 ] ~ [ linux-4.17.6 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.55 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.112 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.140 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.115 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.57 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 
 39 DEFINE_STATIC_KEY_FALSE(rdt_enable_key);
 40 DEFINE_STATIC_KEY_FALSE(rdt_mon_enable_key);
 41 DEFINE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
 42 static struct kernfs_root *rdt_root;
 43 struct rdtgroup rdtgroup_default;
 44 LIST_HEAD(rdt_all_groups);
 45 
 46 /* Kernel fs node for "info" directory under root */
 47 static struct kernfs_node *kn_info;
 48 
 49 /* Kernel fs node for "mon_groups" directory under root */
 50 static struct kernfs_node *kn_mongrp;
 51 
 52 /* Kernel fs node for "mon_data" directory under root */
 53 static struct kernfs_node *kn_mondata;
 54 
 55 static struct seq_buf last_cmd_status;
 56 static char last_cmd_status_buf[512];
 57 
 58 void rdt_last_cmd_clear(void)
 59 {
 60         lockdep_assert_held(&rdtgroup_mutex);
 61         seq_buf_clear(&last_cmd_status);
 62 }
 63 
 64 void rdt_last_cmd_puts(const char *s)
 65 {
 66         lockdep_assert_held(&rdtgroup_mutex);
 67         seq_buf_puts(&last_cmd_status, s);
 68 }
 69 
 70 void 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  */
 95 static int closid_free_map;
 96 
 97 static 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 
112 static 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 
124 static 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 */
130 static 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 
143 static 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 
162 static 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 
172 static 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 
183 static 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 
189 static struct kernfs_ops kf_mondata_ops = {
190         .atomic_write_len       = PAGE_SIZE,
191         .seq_show               = rdtgroup_mondata_show,
192 };
193 
194 static 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 
201 static 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  */
226 static 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  */
248 static void
249 update_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 
259 static 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 
302 static 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 
312 static 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 
367 static 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 
422 unlock:
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 
431 struct task_move_callback {
432         struct callback_head    work;
433         struct rdtgroup         *rdtgrp;
434 };
435 
436 static 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 
464 static 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 
512 static 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 
534 static 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 
563 static 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 
585 static 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 
598 static 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 
614 static 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 
629 static 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 
638 static 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 
647 static 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 
656 static 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 
665 static 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 
674 static 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 
684 static 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 
696 static 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 
705 static 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 
714 static 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 
724 static 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. */
744 static 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 
858 static 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;
877 error:
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 
886 static 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 
909 static 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 
955 out_destroy:
956         kernfs_remove(kn_info);
957         return ret;
958 }
959 
960 static int
961 mongroup_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 
989 out_destroy:
990         kernfs_remove(kn);
991         return ret;
992 }
993 
994 static 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 
1001 static 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 
1008 static inline bool is_mba_linear(void)
1009 {
1010         return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear;
1011 }
1012 
1013 static int set_cache_qos_cfg(int level, bool enable)
1014 {
1015         void (*update)(void *arg);
1016         struct rdt_resource *r_l;
1017         cpumask_var_t cpu_mask;
1018         struct rdt_domain *d;
1019         int cpu;
1020 
1021         if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1022                 return -ENOMEM;
1023 
1024         if (level == RDT_RESOURCE_L3)
1025                 update = l3_qos_cfg_update;
1026         else if (level == RDT_RESOURCE_L2)
1027                 update = l2_qos_cfg_update;
1028         else
1029                 return -EINVAL;
1030 
1031         r_l = &rdt_resources_all[level];
1032         list_for_each_entry(d, &r_l->domains, list) {
1033                 /* Pick one CPU from each domain instance to update MSR */
1034                 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1035         }
1036         cpu = get_cpu();
1037         /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */
1038         if (cpumask_test_cpu(cpu, cpu_mask))
1039                 update(&enable);
1040         /* Update QOS_CFG MSR on all other cpus in cpu_mask. */
1041         smp_call_function_many(cpu_mask, update, &enable, 1);
1042         put_cpu();
1043 
1044         free_cpumask_var(cpu_mask);
1045 
1046         return 0;
1047 }
1048 
1049 /*
1050  * Enable or disable the MBA software controller
1051  * which helps user specify bandwidth in MBps.
1052  * MBA software controller is supported only if
1053  * MBM is supported and MBA is in linear scale.
1054  */
1055 static int set_mba_sc(bool mba_sc)
1056 {
1057         struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA];
1058         struct rdt_domain *d;
1059 
1060         if (!is_mbm_enabled() || !is_mba_linear() ||
1061             mba_sc == is_mba_sc(r))
1062                 return -EINVAL;
1063 
1064         r->membw.mba_sc = mba_sc;
1065         list_for_each_entry(d, &r->domains, list)
1066                 setup_default_ctrlval(r, d->ctrl_val, d->mbps_val);
1067 
1068         return 0;
1069 }
1070 
1071 static int cdp_enable(int level, int data_type, int code_type)
1072 {
1073         struct rdt_resource *r_ldata = &rdt_resources_all[data_type];
1074         struct rdt_resource *r_lcode = &rdt_resources_all[code_type];
1075         struct rdt_resource *r_l = &rdt_resources_all[level];
1076         int ret;
1077 
1078         if (!r_l->alloc_capable || !r_ldata->alloc_capable ||
1079             !r_lcode->alloc_capable)
1080                 return -EINVAL;
1081 
1082         ret = set_cache_qos_cfg(level, true);
1083         if (!ret) {
1084                 r_l->alloc_enabled = false;
1085                 r_ldata->alloc_enabled = true;
1086                 r_lcode->alloc_enabled = true;
1087         }
1088         return ret;
1089 }
1090 
1091 static int cdpl3_enable(void)
1092 {
1093         return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA,
1094                           RDT_RESOURCE_L3CODE);
1095 }
1096 
1097 static int cdpl2_enable(void)
1098 {
1099         return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA,
1100                           RDT_RESOURCE_L2CODE);
1101 }
1102 
1103 static void cdp_disable(int level, int data_type, int code_type)
1104 {
1105         struct rdt_resource *r = &rdt_resources_all[level];
1106 
1107         r->alloc_enabled = r->alloc_capable;
1108 
1109         if (rdt_resources_all[data_type].alloc_enabled) {
1110                 rdt_resources_all[data_type].alloc_enabled = false;
1111                 rdt_resources_all[code_type].alloc_enabled = false;
1112                 set_cache_qos_cfg(level, false);
1113         }
1114 }
1115 
1116 static void cdpl3_disable(void)
1117 {
1118         cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE);
1119 }
1120 
1121 static void cdpl2_disable(void)
1122 {
1123         cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE);
1124 }
1125 
1126 static void cdp_disable_all(void)
1127 {
1128         if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
1129                 cdpl3_disable();
1130         if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled)
1131                 cdpl2_disable();
1132 }
1133 
1134 static int parse_rdtgroupfs_options(char *data)
1135 {
1136         char *token, *o = data;
1137         int ret = 0;
1138 
1139         while ((token = strsep(&o, ",")) != NULL) {
1140                 if (!*token) {
1141                         ret = -EINVAL;
1142                         goto out;
1143                 }
1144 
1145                 if (!strcmp(token, "cdp")) {
1146                         ret = cdpl3_enable();
1147                         if (ret)
1148                                 goto out;
1149                 } else if (!strcmp(token, "cdpl2")) {
1150                         ret = cdpl2_enable();
1151                         if (ret)
1152                                 goto out;
1153                 } else if (!strcmp(token, "mba_MBps")) {
1154                         ret = set_mba_sc(true);
1155                         if (ret)
1156                                 goto out;
1157                 } else {
1158                         ret = -EINVAL;
1159                         goto out;
1160                 }
1161         }
1162 
1163         return 0;
1164 
1165 out:
1166         pr_err("Invalid mount option \"%s\"\n", token);
1167 
1168         return ret;
1169 }
1170 
1171 /*
1172  * We don't allow rdtgroup directories to be created anywhere
1173  * except the root directory. Thus when looking for the rdtgroup
1174  * structure for a kernfs node we are either looking at a directory,
1175  * in which case the rdtgroup structure is pointed at by the "priv"
1176  * field, otherwise we have a file, and need only look to the parent
1177  * to find the rdtgroup.
1178  */
1179 static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn)
1180 {
1181         if (kernfs_type(kn) == KERNFS_DIR) {
1182                 /*
1183                  * All the resource directories use "kn->priv"
1184                  * to point to the "struct rdtgroup" for the
1185                  * resource. "info" and its subdirectories don't
1186                  * have rdtgroup structures, so return NULL here.
1187                  */
1188                 if (kn == kn_info || kn->parent == kn_info)
1189                         return NULL;
1190                 else
1191                         return kn->priv;
1192         } else {
1193                 return kn->parent->priv;
1194         }
1195 }
1196 
1197 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn)
1198 {
1199         struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1200 
1201         if (!rdtgrp)
1202                 return NULL;
1203 
1204         atomic_inc(&rdtgrp->waitcount);
1205         kernfs_break_active_protection(kn);
1206 
1207         mutex_lock(&rdtgroup_mutex);
1208 
1209         /* Was this group deleted while we waited? */
1210         if (rdtgrp->flags & RDT_DELETED)
1211                 return NULL;
1212 
1213         return rdtgrp;
1214 }
1215 
1216 void rdtgroup_kn_unlock(struct kernfs_node *kn)
1217 {
1218         struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn);
1219 
1220         if (!rdtgrp)
1221                 return;
1222 
1223         mutex_unlock(&rdtgroup_mutex);
1224 
1225         if (atomic_dec_and_test(&rdtgrp->waitcount) &&
1226             (rdtgrp->flags & RDT_DELETED)) {
1227                 kernfs_unbreak_active_protection(kn);
1228                 kernfs_put(rdtgrp->kn);
1229                 kfree(rdtgrp);
1230         } else {
1231                 kernfs_unbreak_active_protection(kn);
1232         }
1233 }
1234 
1235 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1236                              struct rdtgroup *prgrp,
1237                              struct kernfs_node **mon_data_kn);
1238 
1239 static struct dentry *rdt_mount(struct file_system_type *fs_type,
1240                                 int flags, const char *unused_dev_name,
1241                                 void *data)
1242 {
1243         struct rdt_domain *dom;
1244         struct rdt_resource *r;
1245         struct dentry *dentry;
1246         int ret;
1247 
1248         cpus_read_lock();
1249         mutex_lock(&rdtgroup_mutex);
1250         /*
1251          * resctrl file system can only be mounted once.
1252          */
1253         if (static_branch_unlikely(&rdt_enable_key)) {
1254                 dentry = ERR_PTR(-EBUSY);
1255                 goto out;
1256         }
1257 
1258         ret = parse_rdtgroupfs_options(data);
1259         if (ret) {
1260                 dentry = ERR_PTR(ret);
1261                 goto out_cdp;
1262         }
1263 
1264         closid_init();
1265 
1266         ret = rdtgroup_create_info_dir(rdtgroup_default.kn);
1267         if (ret) {
1268                 dentry = ERR_PTR(ret);
1269                 goto out_cdp;
1270         }
1271 
1272         if (rdt_mon_capable) {
1273                 ret = mongroup_create_dir(rdtgroup_default.kn,
1274                                           NULL, "mon_groups",
1275                                           &kn_mongrp);
1276                 if (ret) {
1277                         dentry = ERR_PTR(ret);
1278                         goto out_info;
1279                 }
1280                 kernfs_get(kn_mongrp);
1281 
1282                 ret = mkdir_mondata_all(rdtgroup_default.kn,
1283                                         &rdtgroup_default, &kn_mondata);
1284                 if (ret) {
1285                         dentry = ERR_PTR(ret);
1286                         goto out_mongrp;
1287                 }
1288                 kernfs_get(kn_mondata);
1289                 rdtgroup_default.mon.mon_data_kn = kn_mondata;
1290         }
1291 
1292         dentry = kernfs_mount(fs_type, flags, rdt_root,
1293                               RDTGROUP_SUPER_MAGIC, NULL);
1294         if (IS_ERR(dentry))
1295                 goto out_mondata;
1296 
1297         if (rdt_alloc_capable)
1298                 static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
1299         if (rdt_mon_capable)
1300                 static_branch_enable_cpuslocked(&rdt_mon_enable_key);
1301 
1302         if (rdt_alloc_capable || rdt_mon_capable)
1303                 static_branch_enable_cpuslocked(&rdt_enable_key);
1304 
1305         if (is_mbm_enabled()) {
1306                 r = &rdt_resources_all[RDT_RESOURCE_L3];
1307                 list_for_each_entry(dom, &r->domains, list)
1308                         mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL);
1309         }
1310 
1311         goto out;
1312 
1313 out_mondata:
1314         if (rdt_mon_capable)
1315                 kernfs_remove(kn_mondata);
1316 out_mongrp:
1317         if (rdt_mon_capable)
1318                 kernfs_remove(kn_mongrp);
1319 out_info:
1320         kernfs_remove(kn_info);
1321 out_cdp:
1322         cdp_disable_all();
1323 out:
1324         rdt_last_cmd_clear();
1325         mutex_unlock(&rdtgroup_mutex);
1326         cpus_read_unlock();
1327 
1328         return dentry;
1329 }
1330 
1331 static int reset_all_ctrls(struct rdt_resource *r)
1332 {
1333         struct msr_param msr_param;
1334         cpumask_var_t cpu_mask;
1335         struct rdt_domain *d;
1336         int i, cpu;
1337 
1338         if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL))
1339                 return -ENOMEM;
1340 
1341         msr_param.res = r;
1342         msr_param.low = 0;
1343         msr_param.high = r->num_closid;
1344 
1345         /*
1346          * Disable resource control for this resource by setting all
1347          * CBMs in all domains to the maximum mask value. Pick one CPU
1348          * from each domain to update the MSRs below.
1349          */
1350         list_for_each_entry(d, &r->domains, list) {
1351                 cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
1352 
1353                 for (i = 0; i < r->num_closid; i++)
1354                         d->ctrl_val[i] = r->default_ctrl;
1355         }
1356         cpu = get_cpu();
1357         /* Update CBM on this cpu if it's in cpu_mask. */
1358         if (cpumask_test_cpu(cpu, cpu_mask))
1359                 rdt_ctrl_update(&msr_param);
1360         /* Update CBM on all other cpus in cpu_mask. */
1361         smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1);
1362         put_cpu();
1363 
1364         free_cpumask_var(cpu_mask);
1365 
1366         return 0;
1367 }
1368 
1369 static bool is_closid_match(struct task_struct *t, struct rdtgroup *r)
1370 {
1371         return (rdt_alloc_capable &&
1372                 (r->type == RDTCTRL_GROUP) && (t->closid == r->closid));
1373 }
1374 
1375 static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r)
1376 {
1377         return (rdt_mon_capable &&
1378                 (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid));
1379 }
1380 
1381 /*
1382  * Move tasks from one to the other group. If @from is NULL, then all tasks
1383  * in the systems are moved unconditionally (used for teardown).
1384  *
1385  * If @mask is not NULL the cpus on which moved tasks are running are set
1386  * in that mask so the update smp function call is restricted to affected
1387  * cpus.
1388  */
1389 static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to,
1390                                  struct cpumask *mask)
1391 {
1392         struct task_struct *p, *t;
1393 
1394         read_lock(&tasklist_lock);
1395         for_each_process_thread(p, t) {
1396                 if (!from || is_closid_match(t, from) ||
1397                     is_rmid_match(t, from)) {
1398                         t->closid = to->closid;
1399                         t->rmid = to->mon.rmid;
1400 
1401 #ifdef CONFIG_SMP
1402                         /*
1403                          * This is safe on x86 w/o barriers as the ordering
1404                          * of writing to task_cpu() and t->on_cpu is
1405                          * reverse to the reading here. The detection is
1406                          * inaccurate as tasks might move or schedule
1407                          * before the smp function call takes place. In
1408                          * such a case the function call is pointless, but
1409                          * there is no other side effect.
1410                          */
1411                         if (mask && t->on_cpu)
1412                                 cpumask_set_cpu(task_cpu(t), mask);
1413 #endif
1414                 }
1415         }
1416         read_unlock(&tasklist_lock);
1417 }
1418 
1419 static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp)
1420 {
1421         struct rdtgroup *sentry, *stmp;
1422         struct list_head *head;
1423 
1424         head = &rdtgrp->mon.crdtgrp_list;
1425         list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) {
1426                 free_rmid(sentry->mon.rmid);
1427                 list_del(&sentry->mon.crdtgrp_list);
1428                 kfree(sentry);
1429         }
1430 }
1431 
1432 /*
1433  * Forcibly remove all of subdirectories under root.
1434  */
1435 static void rmdir_all_sub(void)
1436 {
1437         struct rdtgroup *rdtgrp, *tmp;
1438 
1439         /* Move all tasks to the default resource group */
1440         rdt_move_group_tasks(NULL, &rdtgroup_default, NULL);
1441 
1442         list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) {
1443                 /* Free any child rmids */
1444                 free_all_child_rdtgrp(rdtgrp);
1445 
1446                 /* Remove each rdtgroup other than root */
1447                 if (rdtgrp == &rdtgroup_default)
1448                         continue;
1449 
1450                 /*
1451                  * Give any CPUs back to the default group. We cannot copy
1452                  * cpu_online_mask because a CPU might have executed the
1453                  * offline callback already, but is still marked online.
1454                  */
1455                 cpumask_or(&rdtgroup_default.cpu_mask,
1456                            &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
1457 
1458                 free_rmid(rdtgrp->mon.rmid);
1459 
1460                 kernfs_remove(rdtgrp->kn);
1461                 list_del(&rdtgrp->rdtgroup_list);
1462                 kfree(rdtgrp);
1463         }
1464         /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
1465         update_closid_rmid(cpu_online_mask, &rdtgroup_default);
1466 
1467         kernfs_remove(kn_info);
1468         kernfs_remove(kn_mongrp);
1469         kernfs_remove(kn_mondata);
1470 }
1471 
1472 static void rdt_kill_sb(struct super_block *sb)
1473 {
1474         struct rdt_resource *r;
1475 
1476         cpus_read_lock();
1477         mutex_lock(&rdtgroup_mutex);
1478 
1479         set_mba_sc(false);
1480 
1481         /*Put everything back to default values. */
1482         for_each_alloc_enabled_rdt_resource(r)
1483                 reset_all_ctrls(r);
1484         cdp_disable_all();
1485         rmdir_all_sub();
1486         static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
1487         static_branch_disable_cpuslocked(&rdt_mon_enable_key);
1488         static_branch_disable_cpuslocked(&rdt_enable_key);
1489         kernfs_kill_sb(sb);
1490         mutex_unlock(&rdtgroup_mutex);
1491         cpus_read_unlock();
1492 }
1493 
1494 static struct file_system_type rdt_fs_type = {
1495         .name    = "resctrl",
1496         .mount   = rdt_mount,
1497         .kill_sb = rdt_kill_sb,
1498 };
1499 
1500 static int mon_addfile(struct kernfs_node *parent_kn, const char *name,
1501                        void *priv)
1502 {
1503         struct kernfs_node *kn;
1504         int ret = 0;
1505 
1506         kn = __kernfs_create_file(parent_kn, name, 0444, 0,
1507                                   &kf_mondata_ops, priv, NULL, NULL);
1508         if (IS_ERR(kn))
1509                 return PTR_ERR(kn);
1510 
1511         ret = rdtgroup_kn_set_ugid(kn);
1512         if (ret) {
1513                 kernfs_remove(kn);
1514                 return ret;
1515         }
1516 
1517         return ret;
1518 }
1519 
1520 /*
1521  * Remove all subdirectories of mon_data of ctrl_mon groups
1522  * and monitor groups with given domain id.
1523  */
1524 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id)
1525 {
1526         struct rdtgroup *prgrp, *crgrp;
1527         char name[32];
1528 
1529         if (!r->mon_enabled)
1530                 return;
1531 
1532         list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
1533                 sprintf(name, "mon_%s_%02d", r->name, dom_id);
1534                 kernfs_remove_by_name(prgrp->mon.mon_data_kn, name);
1535 
1536                 list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list)
1537                         kernfs_remove_by_name(crgrp->mon.mon_data_kn, name);
1538         }
1539 }
1540 
1541 static int mkdir_mondata_subdir(struct kernfs_node *parent_kn,
1542                                 struct rdt_domain *d,
1543                                 struct rdt_resource *r, struct rdtgroup *prgrp)
1544 {
1545         union mon_data_bits priv;
1546         struct kernfs_node *kn;
1547         struct mon_evt *mevt;
1548         struct rmid_read rr;
1549         char name[32];
1550         int ret;
1551 
1552         sprintf(name, "mon_%s_%02d", r->name, d->id);
1553         /* create the directory */
1554         kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp);
1555         if (IS_ERR(kn))
1556                 return PTR_ERR(kn);
1557 
1558         /*
1559          * This extra ref will be put in kernfs_remove() and guarantees
1560          * that kn is always accessible.
1561          */
1562         kernfs_get(kn);
1563         ret = rdtgroup_kn_set_ugid(kn);
1564         if (ret)
1565                 goto out_destroy;
1566 
1567         if (WARN_ON(list_empty(&r->evt_list))) {
1568                 ret = -EPERM;
1569                 goto out_destroy;
1570         }
1571 
1572         priv.u.rid = r->rid;
1573         priv.u.domid = d->id;
1574         list_for_each_entry(mevt, &r->evt_list, list) {
1575                 priv.u.evtid = mevt->evtid;
1576                 ret = mon_addfile(kn, mevt->name, priv.priv);
1577                 if (ret)
1578                         goto out_destroy;
1579 
1580                 if (is_mbm_event(mevt->evtid))
1581                         mon_event_read(&rr, d, prgrp, mevt->evtid, true);
1582         }
1583         kernfs_activate(kn);
1584         return 0;
1585 
1586 out_destroy:
1587         kernfs_remove(kn);
1588         return ret;
1589 }
1590 
1591 /*
1592  * Add all subdirectories of mon_data for "ctrl_mon" groups
1593  * and "monitor" groups with given domain id.
1594  */
1595 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
1596                                     struct rdt_domain *d)
1597 {
1598         struct kernfs_node *parent_kn;
1599         struct rdtgroup *prgrp, *crgrp;
1600         struct list_head *head;
1601 
1602         if (!r->mon_enabled)
1603                 return;
1604 
1605         list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) {
1606                 parent_kn = prgrp->mon.mon_data_kn;
1607                 mkdir_mondata_subdir(parent_kn, d, r, prgrp);
1608 
1609                 head = &prgrp->mon.crdtgrp_list;
1610                 list_for_each_entry(crgrp, head, mon.crdtgrp_list) {
1611                         parent_kn = crgrp->mon.mon_data_kn;
1612                         mkdir_mondata_subdir(parent_kn, d, r, crgrp);
1613                 }
1614         }
1615 }
1616 
1617 static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn,
1618                                        struct rdt_resource *r,
1619                                        struct rdtgroup *prgrp)
1620 {
1621         struct rdt_domain *dom;
1622         int ret;
1623 
1624         list_for_each_entry(dom, &r->domains, list) {
1625                 ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp);
1626                 if (ret)
1627                         return ret;
1628         }
1629 
1630         return 0;
1631 }
1632 
1633 /*
1634  * This creates a directory mon_data which contains the monitored data.
1635  *
1636  * mon_data has one directory for each domain whic are named
1637  * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data
1638  * with L3 domain looks as below:
1639  * ./mon_data:
1640  * mon_L3_00
1641  * mon_L3_01
1642  * mon_L3_02
1643  * ...
1644  *
1645  * Each domain directory has one file per event:
1646  * ./mon_L3_00/:
1647  * llc_occupancy
1648  *
1649  */
1650 static int mkdir_mondata_all(struct kernfs_node *parent_kn,
1651                              struct rdtgroup *prgrp,
1652                              struct kernfs_node **dest_kn)
1653 {
1654         struct rdt_resource *r;
1655         struct kernfs_node *kn;
1656         int ret;
1657 
1658         /*
1659          * Create the mon_data directory first.
1660          */
1661         ret = mongroup_create_dir(parent_kn, NULL, "mon_data", &kn);
1662         if (ret)
1663                 return ret;
1664 
1665         if (dest_kn)
1666                 *dest_kn = kn;
1667 
1668         /*
1669          * Create the subdirectories for each domain. Note that all events
1670          * in a domain like L3 are grouped into a resource whose domain is L3
1671          */
1672         for_each_mon_enabled_rdt_resource(r) {
1673                 ret = mkdir_mondata_subdir_alldom(kn, r, prgrp);
1674                 if (ret)
1675                         goto out_destroy;
1676         }
1677 
1678         return 0;
1679 
1680 out_destroy:
1681         kernfs_remove(kn);
1682         return ret;
1683 }
1684 
1685 static int mkdir_rdt_prepare(struct kernfs_node *parent_kn,
1686                              struct kernfs_node *prgrp_kn,
1687                              const char *name, umode_t mode,
1688                              enum rdt_group_type rtype, struct rdtgroup **r)
1689 {
1690         struct rdtgroup *prdtgrp, *rdtgrp;
1691         struct kernfs_node *kn;
1692         uint files = 0;
1693         int ret;
1694 
1695         prdtgrp = rdtgroup_kn_lock_live(prgrp_kn);
1696         rdt_last_cmd_clear();
1697         if (!prdtgrp) {
1698                 ret = -ENODEV;
1699                 rdt_last_cmd_puts("directory was removed\n");
1700                 goto out_unlock;
1701         }
1702 
1703         /* allocate the rdtgroup. */
1704         rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL);
1705         if (!rdtgrp) {
1706                 ret = -ENOSPC;
1707                 rdt_last_cmd_puts("kernel out of memory\n");
1708                 goto out_unlock;
1709         }
1710         *r = rdtgrp;
1711         rdtgrp->mon.parent = prdtgrp;
1712         rdtgrp->type = rtype;
1713         INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list);
1714 
1715         /* kernfs creates the directory for rdtgrp */
1716         kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp);
1717         if (IS_ERR(kn)) {
1718                 ret = PTR_ERR(kn);
1719                 rdt_last_cmd_puts("kernfs create error\n");
1720                 goto out_free_rgrp;
1721         }
1722         rdtgrp->kn = kn;
1723 
1724         /*
1725          * kernfs_remove() will drop the reference count on "kn" which
1726          * will free it. But we still need it to stick around for the
1727          * rdtgroup_kn_unlock(kn} call below. Take one extra reference
1728          * here, which will be dropped inside rdtgroup_kn_unlock().
1729          */
1730         kernfs_get(kn);
1731 
1732         ret = rdtgroup_kn_set_ugid(kn);
1733         if (ret) {
1734                 rdt_last_cmd_puts("kernfs perm error\n");
1735                 goto out_destroy;
1736         }
1737 
1738         files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype);
1739         ret = rdtgroup_add_files(kn, files);
1740         if (ret) {
1741                 rdt_last_cmd_puts("kernfs fill error\n");
1742                 goto out_destroy;
1743         }
1744 
1745         if (rdt_mon_capable) {
1746                 ret = alloc_rmid();
1747                 if (ret < 0) {
1748                         rdt_last_cmd_puts("out of RMIDs\n");
1749                         goto out_destroy;
1750                 }
1751                 rdtgrp->mon.rmid = ret;
1752 
1753                 ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn);
1754                 if (ret) {
1755                         rdt_last_cmd_puts("kernfs subdir error\n");
1756                         goto out_idfree;
1757                 }
1758         }
1759         kernfs_activate(kn);
1760 
1761         /*
1762          * The caller unlocks the prgrp_kn upon success.
1763          */
1764         return 0;
1765 
1766 out_idfree:
1767         free_rmid(rdtgrp->mon.rmid);
1768 out_destroy:
1769         kernfs_remove(rdtgrp->kn);
1770 out_free_rgrp:
1771         kfree(rdtgrp);
1772 out_unlock:
1773         rdtgroup_kn_unlock(prgrp_kn);
1774         return ret;
1775 }
1776 
1777 static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp)
1778 {
1779         kernfs_remove(rgrp->kn);
1780         free_rmid(rgrp->mon.rmid);
1781         kfree(rgrp);
1782 }
1783 
1784 /*
1785  * Create a monitor group under "mon_groups" directory of a control
1786  * and monitor group(ctrl_mon). This is a resource group
1787  * to monitor a subset of tasks and cpus in its parent ctrl_mon group.
1788  */
1789 static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn,
1790                               struct kernfs_node *prgrp_kn,
1791                               const char *name,
1792                               umode_t mode)
1793 {
1794         struct rdtgroup *rdtgrp, *prgrp;
1795         int ret;
1796 
1797         ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP,
1798                                 &rdtgrp);
1799         if (ret)
1800                 return ret;
1801 
1802         prgrp = rdtgrp->mon.parent;
1803         rdtgrp->closid = prgrp->closid;
1804 
1805         /*
1806          * Add the rdtgrp to the list of rdtgrps the parent
1807          * ctrl_mon group has to track.
1808          */
1809         list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list);
1810 
1811         rdtgroup_kn_unlock(prgrp_kn);
1812         return ret;
1813 }
1814 
1815 /*
1816  * These are rdtgroups created under the root directory. Can be used
1817  * to allocate and monitor resources.
1818  */
1819 static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn,
1820                                    struct kernfs_node *prgrp_kn,
1821                                    const char *name, umode_t mode)
1822 {
1823         struct rdtgroup *rdtgrp;
1824         struct kernfs_node *kn;
1825         u32 closid;
1826         int ret;
1827 
1828         ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP,
1829                                 &rdtgrp);
1830         if (ret)
1831                 return ret;
1832 
1833         kn = rdtgrp->kn;
1834         ret = closid_alloc();
1835         if (ret < 0) {
1836                 rdt_last_cmd_puts("out of CLOSIDs\n");
1837                 goto out_common_fail;
1838         }
1839         closid = ret;
1840         ret = 0;
1841 
1842         rdtgrp->closid = closid;
1843         list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups);
1844 
1845         if (rdt_mon_capable) {
1846                 /*
1847                  * Create an empty mon_groups directory to hold the subset
1848                  * of tasks and cpus to monitor.
1849                  */
1850                 ret = mongroup_create_dir(kn, NULL, "mon_groups", NULL);
1851                 if (ret) {
1852                         rdt_last_cmd_puts("kernfs subdir error\n");
1853                         goto out_id_free;
1854                 }
1855         }
1856 
1857         goto out_unlock;
1858 
1859 out_id_free:
1860         closid_free(closid);
1861         list_del(&rdtgrp->rdtgroup_list);
1862 out_common_fail:
1863         mkdir_rdt_prepare_clean(rdtgrp);
1864 out_unlock:
1865         rdtgroup_kn_unlock(prgrp_kn);
1866         return ret;
1867 }
1868 
1869 /*
1870  * We allow creating mon groups only with in a directory called "mon_groups"
1871  * which is present in every ctrl_mon group. Check if this is a valid
1872  * "mon_groups" directory.
1873  *
1874  * 1. The directory should be named "mon_groups".
1875  * 2. The mon group itself should "not" be named "mon_groups".
1876  *   This makes sure "mon_groups" directory always has a ctrl_mon group
1877  *   as parent.
1878  */
1879 static bool is_mon_groups(struct kernfs_node *kn, const char *name)
1880 {
1881         return (!strcmp(kn->name, "mon_groups") &&
1882                 strcmp(name, "mon_groups"));
1883 }
1884 
1885 static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
1886                           umode_t mode)
1887 {
1888         /* Do not accept '\n' to avoid unparsable situation. */
1889         if (strchr(name, '\n'))
1890                 return -EINVAL;
1891 
1892         /*
1893          * If the parent directory is the root directory and RDT
1894          * allocation is supported, add a control and monitoring
1895          * subdirectory
1896          */
1897         if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn)
1898                 return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode);
1899 
1900         /*
1901          * If RDT monitoring is supported and the parent directory is a valid
1902          * "mon_groups" directory, add a monitoring subdirectory.
1903          */
1904         if (rdt_mon_capable && is_mon_groups(parent_kn, name))
1905                 return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode);
1906 
1907         return -EPERM;
1908 }
1909 
1910 static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
1911                               cpumask_var_t tmpmask)
1912 {
1913         struct rdtgroup *prdtgrp = rdtgrp->mon.parent;
1914         int cpu;
1915 
1916         /* Give any tasks back to the parent group */
1917         rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask);
1918 
1919         /* Update per cpu rmid of the moved CPUs first */
1920         for_each_cpu(cpu, &rdtgrp->cpu_mask)
1921                 per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid;
1922         /*
1923          * Update the MSR on moved CPUs and CPUs which have moved
1924          * task running on them.
1925          */
1926         cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
1927         update_closid_rmid(tmpmask, NULL);
1928 
1929         rdtgrp->flags = RDT_DELETED;
1930         free_rmid(rdtgrp->mon.rmid);
1931 
1932         /*
1933          * Remove the rdtgrp from the parent ctrl_mon group's list
1934          */
1935         WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list));
1936         list_del(&rdtgrp->mon.crdtgrp_list);
1937 
1938         /*
1939          * one extra hold on this, will drop when we kfree(rdtgrp)
1940          * in rdtgroup_kn_unlock()
1941          */
1942         kernfs_get(kn);
1943         kernfs_remove(rdtgrp->kn);
1944 
1945         return 0;
1946 }
1947 
1948 static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp,
1949                                cpumask_var_t tmpmask)
1950 {
1951         int cpu;
1952 
1953         /* Give any tasks back to the default group */
1954         rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask);
1955 
1956         /* Give any CPUs back to the default group */
1957         cpumask_or(&rdtgroup_default.cpu_mask,
1958                    &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask);
1959 
1960         /* Update per cpu closid and rmid of the moved CPUs first */
1961         for_each_cpu(cpu, &rdtgrp->cpu_mask) {
1962                 per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid;
1963                 per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid;
1964         }
1965 
1966         /*
1967          * Update the MSR on moved CPUs and CPUs which have moved
1968          * task running on them.
1969          */
1970         cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask);
1971         update_closid_rmid(tmpmask, NULL);
1972 
1973         rdtgrp->flags = RDT_DELETED;
1974         closid_free(rdtgrp->closid);
1975         free_rmid(rdtgrp->mon.rmid);
1976 
1977         /*
1978          * Free all the child monitor group rmids.
1979          */
1980         free_all_child_rdtgrp(rdtgrp);
1981 
1982         list_del(&rdtgrp->rdtgroup_list);
1983 
1984         /*
1985          * one extra hold on this, will drop when we kfree(rdtgrp)
1986          * in rdtgroup_kn_unlock()
1987          */
1988         kernfs_get(kn);
1989         kernfs_remove(rdtgrp->kn);
1990 
1991         return 0;
1992 }
1993 
1994 static int rdtgroup_rmdir(struct kernfs_node *kn)
1995 {
1996         struct kernfs_node *parent_kn = kn->parent;
1997         struct rdtgroup *rdtgrp;
1998         cpumask_var_t tmpmask;
1999         int ret = 0;
2000 
2001         if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL))
2002                 return -ENOMEM;
2003 
2004         rdtgrp = rdtgroup_kn_lock_live(kn);
2005         if (!rdtgrp) {
2006                 ret = -EPERM;
2007                 goto out;
2008         }
2009 
2010         /*
2011          * If the rdtgroup is a ctrl_mon group and parent directory
2012          * is the root directory, remove the ctrl_mon group.
2013          *
2014          * If the rdtgroup is a mon group and parent directory
2015          * is a valid "mon_groups" directory, remove the mon group.
2016          */
2017         if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn)
2018                 ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask);
2019         else if (rdtgrp->type == RDTMON_GROUP &&
2020                  is_mon_groups(parent_kn, kn->name))
2021                 ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask);
2022         else
2023                 ret = -EPERM;
2024 
2025 out:
2026         rdtgroup_kn_unlock(kn);
2027         free_cpumask_var(tmpmask);
2028         return ret;
2029 }
2030 
2031 static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf)
2032 {
2033         if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled)
2034                 seq_puts(seq, ",cdp");
2035         return 0;
2036 }
2037 
2038 static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = {
2039         .mkdir          = rdtgroup_mkdir,
2040         .rmdir          = rdtgroup_rmdir,
2041         .show_options   = rdtgroup_show_options,
2042 };
2043 
2044 static int __init rdtgroup_setup_root(void)
2045 {
2046         int ret;
2047 
2048         rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops,
2049                                       KERNFS_ROOT_CREATE_DEACTIVATED,
2050                                       &rdtgroup_default);
2051         if (IS_ERR(rdt_root))
2052                 return PTR_ERR(rdt_root);
2053 
2054         mutex_lock(&rdtgroup_mutex);
2055 
2056         rdtgroup_default.closid = 0;
2057         rdtgroup_default.mon.rmid = 0;
2058         rdtgroup_default.type = RDTCTRL_GROUP;
2059         INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list);
2060 
2061         list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups);
2062 
2063         ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE);
2064         if (ret) {
2065                 kernfs_destroy_root(rdt_root);
2066                 goto out;
2067         }
2068 
2069         rdtgroup_default.kn = rdt_root->kn;
2070         kernfs_activate(rdtgroup_default.kn);
2071 
2072 out:
2073         mutex_unlock(&rdtgroup_mutex);
2074 
2075         return ret;
2076 }
2077 
2078 /*
2079  * rdtgroup_init - rdtgroup initialization
2080  *
2081  * Setup resctrl file system including set up root, create mount point,
2082  * register rdtgroup filesystem, and initialize files under root directory.
2083  *
2084  * Return: 0 on success or -errno
2085  */
2086 int __init rdtgroup_init(void)
2087 {
2088         int ret = 0;
2089 
2090         seq_buf_init(&last_cmd_status, last_cmd_status_buf,
2091                      sizeof(last_cmd_status_buf));
2092 
2093         ret = rdtgroup_setup_root();
2094         if (ret)
2095                 return ret;
2096 
2097         ret = sysfs_create_mount_point(fs_kobj, "resctrl");
2098         if (ret)
2099                 goto cleanup_root;
2100 
2101         ret = register_filesystem(&rdt_fs_type);
2102         if (ret)
2103                 goto cleanup_mountpoint;
2104 
2105         return 0;
2106 
2107 cleanup_mountpoint:
2108         sysfs_remove_mount_point(fs_kobj, "resctrl");
2109 cleanup_root:
2110         kernfs_destroy_root(rdt_root);
2111 
2112         return ret;
2113 }
2114 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp