Commit | Line | Data |
---|---|---|
5ff193fb FY |
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 | ||
d9b48c86 | 23 | #include <linux/cacheinfo.h> |
12e0110c | 24 | #include <linux/cpu.h> |
37707ec6 | 25 | #include <linux/debugfs.h> |
5ff193fb FY |
26 | #include <linux/fs.h> |
27 | #include <linux/sysfs.h> | |
28 | #include <linux/kernfs.h> | |
9b3a7fd0 | 29 | #include <linux/seq_buf.h> |
4e978d06 | 30 | #include <linux/seq_file.h> |
3f07c014 | 31 | #include <linux/sched/signal.h> |
29930025 | 32 | #include <linux/sched/task.h> |
5ff193fb | 33 | #include <linux/slab.h> |
e02737d5 | 34 | #include <linux/task_work.h> |
5ff193fb FY |
35 | |
36 | #include <uapi/linux/magic.h> | |
37 | ||
05830204 VS |
38 | #include <asm/intel_rdt_sched.h> |
39 | #include "intel_rdt.h" | |
5ff193fb | 40 | |
4af4a88e VS |
41 | DEFINE_STATIC_KEY_FALSE(rdt_enable_key); |
42 | DEFINE_STATIC_KEY_FALSE(rdt_mon_enable_key); | |
1b5c0b75 | 43 | DEFINE_STATIC_KEY_FALSE(rdt_alloc_enable_key); |
cb2200e9 | 44 | static struct kernfs_root *rdt_root; |
5ff193fb FY |
45 | struct rdtgroup rdtgroup_default; |
46 | LIST_HEAD(rdt_all_groups); | |
47 | ||
4e978d06 FY |
48 | /* Kernel fs node for "info" directory under root */ |
49 | static struct kernfs_node *kn_info; | |
50 | ||
4af4a88e VS |
51 | /* Kernel fs node for "mon_groups" directory under root */ |
52 | static struct kernfs_node *kn_mongrp; | |
53 | ||
54 | /* Kernel fs node for "mon_data" directory under root */ | |
55 | static struct kernfs_node *kn_mondata; | |
56 | ||
9b3a7fd0 TL |
57 | static struct seq_buf last_cmd_status; |
58 | static char last_cmd_status_buf[512]; | |
59 | ||
37707ec6 RC |
60 | struct dentry *debugfs_resctrl; |
61 | ||
9b3a7fd0 TL |
62 | void rdt_last_cmd_clear(void) |
63 | { | |
64 | lockdep_assert_held(&rdtgroup_mutex); | |
65 | seq_buf_clear(&last_cmd_status); | |
66 | } | |
67 | ||
68 | void rdt_last_cmd_puts(const char *s) | |
69 | { | |
70 | lockdep_assert_held(&rdtgroup_mutex); | |
71 | seq_buf_puts(&last_cmd_status, s); | |
72 | } | |
73 | ||
74 | void rdt_last_cmd_printf(const char *fmt, ...) | |
75 | { | |
76 | va_list ap; | |
77 | ||
78 | va_start(ap, fmt); | |
79 | lockdep_assert_held(&rdtgroup_mutex); | |
80 | seq_buf_vprintf(&last_cmd_status, fmt, ap); | |
81 | va_end(ap); | |
82 | } | |
83 | ||
60cf5e10 FY |
84 | /* |
85 | * Trivial allocator for CLOSIDs. Since h/w only supports a small number, | |
86 | * we can keep a bitmap of free CLOSIDs in a single integer. | |
87 | * | |
88 | * Using a global CLOSID across all resources has some advantages and | |
89 | * some drawbacks: | |
90 | * + We can simply set "current->closid" to assign a task to a resource | |
91 | * group. | |
92 | * + Context switch code can avoid extra memory references deciding which | |
93 | * CLOSID to load into the PQR_ASSOC MSR | |
94 | * - We give up some options in configuring resource groups across multi-socket | |
95 | * systems. | |
96 | * - Our choices on how to configure each resource become progressively more | |
97 | * limited as the number of resources grows. | |
98 | */ | |
99 | static int closid_free_map; | |
c793da8e RC |
100 | static int closid_free_map_len; |
101 | ||
102 | int closids_supported(void) | |
103 | { | |
104 | return closid_free_map_len; | |
105 | } | |
60cf5e10 FY |
106 | |
107 | static void closid_init(void) | |
108 | { | |
109 | struct rdt_resource *r; | |
110 | int rdt_min_closid = 32; | |
111 | ||
112 | /* Compute rdt_min_closid across all resources */ | |
1b5c0b75 | 113 | for_each_alloc_enabled_rdt_resource(r) |
60cf5e10 FY |
114 | rdt_min_closid = min(rdt_min_closid, r->num_closid); |
115 | ||
116 | closid_free_map = BIT_MASK(rdt_min_closid) - 1; | |
117 | ||
118 | /* CLOSID 0 is always reserved for the default group */ | |
119 | closid_free_map &= ~1; | |
c793da8e | 120 | closid_free_map_len = rdt_min_closid; |
60cf5e10 FY |
121 | } |
122 | ||
cb2200e9 | 123 | static int closid_alloc(void) |
60cf5e10 | 124 | { |
0734ded1 | 125 | u32 closid = ffs(closid_free_map); |
60cf5e10 FY |
126 | |
127 | if (closid == 0) | |
128 | return -ENOSPC; | |
129 | closid--; | |
130 | closid_free_map &= ~(1 << closid); | |
131 | ||
132 | return closid; | |
133 | } | |
134 | ||
024d15be | 135 | void closid_free(int closid) |
60cf5e10 FY |
136 | { |
137 | closid_free_map |= 1 << closid; | |
138 | } | |
139 | ||
0b9aa656 RC |
140 | /** |
141 | * closid_allocated - test if provided closid is in use | |
142 | * @closid: closid to be tested | |
143 | * | |
144 | * Return: true if @closid is currently associated with a resource group, | |
145 | * false if @closid is free | |
146 | */ | |
95f0b77e | 147 | static bool closid_allocated(unsigned int closid) |
0b9aa656 RC |
148 | { |
149 | return (closid_free_map & (1 << closid)) == 0; | |
150 | } | |
151 | ||
472ef09b RC |
152 | /** |
153 | * rdtgroup_mode_by_closid - Return mode of resource group with closid | |
154 | * @closid: closid if the resource group | |
155 | * | |
156 | * Each resource group is associated with a @closid. Here the mode | |
157 | * of a resource group can be queried by searching for it using its closid. | |
158 | * | |
159 | * Return: mode as &enum rdtgrp_mode of resource group with closid @closid | |
160 | */ | |
161 | enum rdtgrp_mode rdtgroup_mode_by_closid(int closid) | |
162 | { | |
163 | struct rdtgroup *rdtgrp; | |
164 | ||
165 | list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) { | |
166 | if (rdtgrp->closid == closid) | |
167 | return rdtgrp->mode; | |
168 | } | |
169 | ||
170 | return RDT_NUM_MODES; | |
171 | } | |
172 | ||
d48d7a57 | 173 | static const char * const rdt_mode_str[] = { |
bb9fec69 RC |
174 | [RDT_MODE_SHAREABLE] = "shareable", |
175 | [RDT_MODE_EXCLUSIVE] = "exclusive", | |
176 | [RDT_MODE_PSEUDO_LOCKSETUP] = "pseudo-locksetup", | |
177 | [RDT_MODE_PSEUDO_LOCKED] = "pseudo-locked", | |
d48d7a57 RC |
178 | }; |
179 | ||
180 | /** | |
181 | * rdtgroup_mode_str - Return the string representation of mode | |
182 | * @mode: the resource group mode as &enum rdtgroup_mode | |
183 | * | |
184 | * Return: string representation of valid mode, "unknown" otherwise | |
185 | */ | |
186 | static const char *rdtgroup_mode_str(enum rdtgrp_mode mode) | |
187 | { | |
188 | if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES) | |
189 | return "unknown"; | |
190 | ||
191 | return rdt_mode_str[mode]; | |
192 | } | |
193 | ||
4e978d06 FY |
194 | /* set uid and gid of rdtgroup dirs and files to that of the creator */ |
195 | static int rdtgroup_kn_set_ugid(struct kernfs_node *kn) | |
196 | { | |
197 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
198 | .ia_uid = current_fsuid(), | |
199 | .ia_gid = current_fsgid(), }; | |
200 | ||
201 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
202 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
203 | return 0; | |
204 | ||
205 | return kernfs_setattr(kn, &iattr); | |
206 | } | |
207 | ||
208 | static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft) | |
209 | { | |
210 | struct kernfs_node *kn; | |
211 | int ret; | |
212 | ||
213 | kn = __kernfs_create_file(parent_kn, rft->name, rft->mode, | |
488dee96 | 214 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, |
4e978d06 FY |
215 | 0, rft->kf_ops, rft, NULL, NULL); |
216 | if (IS_ERR(kn)) | |
217 | return PTR_ERR(kn); | |
218 | ||
219 | ret = rdtgroup_kn_set_ugid(kn); | |
220 | if (ret) { | |
221 | kernfs_remove(kn); | |
222 | return ret; | |
223 | } | |
224 | ||
225 | return 0; | |
226 | } | |
227 | ||
4e978d06 FY |
228 | static int rdtgroup_seqfile_show(struct seq_file *m, void *arg) |
229 | { | |
230 | struct kernfs_open_file *of = m->private; | |
231 | struct rftype *rft = of->kn->priv; | |
232 | ||
233 | if (rft->seq_show) | |
234 | return rft->seq_show(of, m, arg); | |
235 | return 0; | |
236 | } | |
237 | ||
238 | static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf, | |
239 | size_t nbytes, loff_t off) | |
240 | { | |
241 | struct rftype *rft = of->kn->priv; | |
242 | ||
243 | if (rft->write) | |
244 | return rft->write(of, buf, nbytes, off); | |
245 | ||
246 | return -EINVAL; | |
247 | } | |
248 | ||
249 | static struct kernfs_ops rdtgroup_kf_single_ops = { | |
250 | .atomic_write_len = PAGE_SIZE, | |
251 | .write = rdtgroup_file_write, | |
252 | .seq_show = rdtgroup_seqfile_show, | |
253 | }; | |
254 | ||
d89b7379 VS |
255 | static struct kernfs_ops kf_mondata_ops = { |
256 | .atomic_write_len = PAGE_SIZE, | |
257 | .seq_show = rdtgroup_mondata_show, | |
258 | }; | |
259 | ||
4ffa3c97 JO |
260 | static bool is_cpu_list(struct kernfs_open_file *of) |
261 | { | |
262 | struct rftype *rft = of->kn->priv; | |
263 | ||
264 | return rft->flags & RFTYPE_FLAGS_CPUS_LIST; | |
265 | } | |
266 | ||
12e0110c TL |
267 | static int rdtgroup_cpus_show(struct kernfs_open_file *of, |
268 | struct seq_file *s, void *v) | |
269 | { | |
270 | struct rdtgroup *rdtgrp; | |
271 | int ret = 0; | |
272 | ||
273 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
274 | ||
4ffa3c97 | 275 | if (rdtgrp) { |
33dc3e41 RC |
276 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) |
277 | seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n", | |
278 | cpumask_pr_args(&rdtgrp->plr->d->cpu_mask)); | |
279 | else | |
280 | seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n", | |
281 | cpumask_pr_args(&rdtgrp->cpu_mask)); | |
4ffa3c97 | 282 | } else { |
12e0110c | 283 | ret = -ENOENT; |
4ffa3c97 | 284 | } |
12e0110c TL |
285 | rdtgroup_kn_unlock(of->kn); |
286 | ||
287 | return ret; | |
288 | } | |
289 | ||
f4107702 FY |
290 | /* |
291 | * This is safe against intel_rdt_sched_in() called from __switch_to() | |
292 | * because __switch_to() is executed with interrupts disabled. A local call | |
a9fcf862 | 293 | * from update_closid_rmid() is proteced against __switch_to() because |
f4107702 FY |
294 | * preemption is disabled. |
295 | */ | |
a9fcf862 | 296 | static void update_cpu_closid_rmid(void *info) |
f4107702 | 297 | { |
b09d981b VS |
298 | struct rdtgroup *r = info; |
299 | ||
a9fcf862 | 300 | if (r) { |
a9110b55 VS |
301 | this_cpu_write(pqr_state.default_closid, r->closid); |
302 | this_cpu_write(pqr_state.default_rmid, r->mon.rmid); | |
a9fcf862 | 303 | } |
b09d981b | 304 | |
f4107702 FY |
305 | /* |
306 | * We cannot unconditionally write the MSR because the current | |
307 | * executing task might have its own closid selected. Just reuse | |
308 | * the context switch code. | |
309 | */ | |
310 | intel_rdt_sched_in(); | |
311 | } | |
312 | ||
0efc89be FY |
313 | /* |
314 | * Update the PGR_ASSOC MSR on all cpus in @cpu_mask, | |
315 | * | |
b09d981b | 316 | * Per task closids/rmids must have been set up before calling this function. |
0efc89be FY |
317 | */ |
318 | static void | |
a9fcf862 | 319 | update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r) |
f4107702 FY |
320 | { |
321 | int cpu = get_cpu(); | |
322 | ||
323 | if (cpumask_test_cpu(cpu, cpu_mask)) | |
a9fcf862 VS |
324 | update_cpu_closid_rmid(r); |
325 | smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1); | |
f4107702 FY |
326 | put_cpu(); |
327 | } | |
328 | ||
a9fcf862 VS |
329 | static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, |
330 | cpumask_var_t tmpmask) | |
331 | { | |
332 | struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp; | |
333 | struct list_head *head; | |
334 | ||
335 | /* Check whether cpus belong to parent ctrl group */ | |
336 | cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask); | |
94457b36 TL |
337 | if (cpumask_weight(tmpmask)) { |
338 | rdt_last_cmd_puts("can only add CPUs to mongroup that belong to parent\n"); | |
a9fcf862 | 339 | return -EINVAL; |
94457b36 | 340 | } |
a9fcf862 VS |
341 | |
342 | /* Check whether cpus are dropped from this group */ | |
343 | cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); | |
344 | if (cpumask_weight(tmpmask)) { | |
345 | /* Give any dropped cpus to parent rdtgroup */ | |
346 | cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask); | |
347 | update_closid_rmid(tmpmask, prgrp); | |
348 | } | |
349 | ||
350 | /* | |
351 | * If we added cpus, remove them from previous group that owned them | |
352 | * and update per-cpu rmid | |
353 | */ | |
354 | cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); | |
355 | if (cpumask_weight(tmpmask)) { | |
356 | head = &prgrp->mon.crdtgrp_list; | |
357 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) { | |
358 | if (crgrp == rdtgrp) | |
359 | continue; | |
360 | cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask, | |
361 | tmpmask); | |
362 | } | |
363 | update_closid_rmid(tmpmask, rdtgrp); | |
364 | } | |
365 | ||
366 | /* Done pushing/pulling - update this group with new mask */ | |
367 | cpumask_copy(&rdtgrp->cpu_mask, newmask); | |
368 | ||
369 | return 0; | |
370 | } | |
371 | ||
372 | static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m) | |
373 | { | |
374 | struct rdtgroup *crgrp; | |
375 | ||
376 | cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m); | |
377 | /* update the child mon group masks as well*/ | |
378 | list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list) | |
379 | cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask); | |
380 | } | |
381 | ||
b09d981b | 382 | static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, |
a9fcf862 | 383 | cpumask_var_t tmpmask, cpumask_var_t tmpmask1) |
b09d981b | 384 | { |
a9fcf862 VS |
385 | struct rdtgroup *r, *crgrp; |
386 | struct list_head *head; | |
b09d981b VS |
387 | |
388 | /* Check whether cpus are dropped from this group */ | |
389 | cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); | |
390 | if (cpumask_weight(tmpmask)) { | |
391 | /* Can't drop from default group */ | |
94457b36 TL |
392 | if (rdtgrp == &rdtgroup_default) { |
393 | rdt_last_cmd_puts("Can't drop CPUs from default group\n"); | |
b09d981b | 394 | return -EINVAL; |
94457b36 | 395 | } |
b09d981b VS |
396 | |
397 | /* Give any dropped cpus to rdtgroup_default */ | |
398 | cpumask_or(&rdtgroup_default.cpu_mask, | |
399 | &rdtgroup_default.cpu_mask, tmpmask); | |
a9fcf862 | 400 | update_closid_rmid(tmpmask, &rdtgroup_default); |
b09d981b VS |
401 | } |
402 | ||
403 | /* | |
a9fcf862 VS |
404 | * If we added cpus, remove them from previous group and |
405 | * the prev group's child groups that owned them | |
406 | * and update per-cpu closid/rmid. | |
b09d981b VS |
407 | */ |
408 | cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); | |
409 | if (cpumask_weight(tmpmask)) { | |
410 | list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) { | |
411 | if (r == rdtgrp) | |
412 | continue; | |
a9fcf862 VS |
413 | cpumask_and(tmpmask1, &r->cpu_mask, tmpmask); |
414 | if (cpumask_weight(tmpmask1)) | |
415 | cpumask_rdtgrp_clear(r, tmpmask1); | |
b09d981b | 416 | } |
a9fcf862 | 417 | update_closid_rmid(tmpmask, rdtgrp); |
b09d981b VS |
418 | } |
419 | ||
420 | /* Done pushing/pulling - update this group with new mask */ | |
421 | cpumask_copy(&rdtgrp->cpu_mask, newmask); | |
422 | ||
a9fcf862 VS |
423 | /* |
424 | * Clear child mon group masks since there is a new parent mask | |
425 | * now and update the rmid for the cpus the child lost. | |
426 | */ | |
427 | head = &rdtgrp->mon.crdtgrp_list; | |
428 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) { | |
429 | cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask); | |
430 | update_closid_rmid(tmpmask, rdtgrp); | |
431 | cpumask_clear(&crgrp->cpu_mask); | |
432 | } | |
433 | ||
b09d981b VS |
434 | return 0; |
435 | } | |
436 | ||
12e0110c TL |
437 | static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of, |
438 | char *buf, size_t nbytes, loff_t off) | |
439 | { | |
a9fcf862 | 440 | cpumask_var_t tmpmask, newmask, tmpmask1; |
b09d981b | 441 | struct rdtgroup *rdtgrp; |
f4107702 | 442 | int ret; |
12e0110c TL |
443 | |
444 | if (!buf) | |
445 | return -EINVAL; | |
446 | ||
447 | if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
448 | return -ENOMEM; | |
449 | if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) { | |
450 | free_cpumask_var(tmpmask); | |
451 | return -ENOMEM; | |
452 | } | |
a9fcf862 VS |
453 | if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) { |
454 | free_cpumask_var(tmpmask); | |
455 | free_cpumask_var(newmask); | |
456 | return -ENOMEM; | |
457 | } | |
a2584e1d | 458 | |
12e0110c | 459 | rdtgrp = rdtgroup_kn_lock_live(of->kn); |
94457b36 | 460 | rdt_last_cmd_clear(); |
12e0110c TL |
461 | if (!rdtgrp) { |
462 | ret = -ENOENT; | |
94457b36 | 463 | rdt_last_cmd_puts("directory was removed\n"); |
12e0110c TL |
464 | goto unlock; |
465 | } | |
466 | ||
c966dac8 RC |
467 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || |
468 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { | |
469 | ret = -EINVAL; | |
470 | rdt_last_cmd_puts("pseudo-locking in progress\n"); | |
471 | goto unlock; | |
472 | } | |
473 | ||
4ffa3c97 JO |
474 | if (is_cpu_list(of)) |
475 | ret = cpulist_parse(buf, newmask); | |
476 | else | |
477 | ret = cpumask_parse(buf, newmask); | |
478 | ||
94457b36 TL |
479 | if (ret) { |
480 | rdt_last_cmd_puts("bad cpu list/mask\n"); | |
12e0110c | 481 | goto unlock; |
94457b36 | 482 | } |
12e0110c | 483 | |
12e0110c TL |
484 | /* check that user didn't specify any offline cpus */ |
485 | cpumask_andnot(tmpmask, newmask, cpu_online_mask); | |
486 | if (cpumask_weight(tmpmask)) { | |
487 | ret = -EINVAL; | |
94457b36 | 488 | rdt_last_cmd_puts("can only assign online cpus\n"); |
a2584e1d | 489 | goto unlock; |
12e0110c TL |
490 | } |
491 | ||
b09d981b | 492 | if (rdtgrp->type == RDTCTRL_GROUP) |
a9fcf862 VS |
493 | ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1); |
494 | else if (rdtgrp->type == RDTMON_GROUP) | |
495 | ret = cpus_mon_write(rdtgrp, newmask, tmpmask); | |
b09d981b VS |
496 | else |
497 | ret = -EINVAL; | |
12e0110c | 498 | |
12e0110c TL |
499 | unlock: |
500 | rdtgroup_kn_unlock(of->kn); | |
501 | free_cpumask_var(tmpmask); | |
502 | free_cpumask_var(newmask); | |
a9fcf862 | 503 | free_cpumask_var(tmpmask1); |
12e0110c TL |
504 | |
505 | return ret ?: nbytes; | |
506 | } | |
507 | ||
e02737d5 FY |
508 | struct task_move_callback { |
509 | struct callback_head work; | |
510 | struct rdtgroup *rdtgrp; | |
511 | }; | |
512 | ||
513 | static void move_myself(struct callback_head *head) | |
514 | { | |
515 | struct task_move_callback *callback; | |
516 | struct rdtgroup *rdtgrp; | |
517 | ||
518 | callback = container_of(head, struct task_move_callback, work); | |
519 | rdtgrp = callback->rdtgrp; | |
520 | ||
521 | /* | |
522 | * If resource group was deleted before this task work callback | |
523 | * was invoked, then assign the task to root group and free the | |
524 | * resource group. | |
525 | */ | |
526 | if (atomic_dec_and_test(&rdtgrp->waitcount) && | |
527 | (rdtgrp->flags & RDT_DELETED)) { | |
528 | current->closid = 0; | |
d6aaba61 | 529 | current->rmid = 0; |
e02737d5 FY |
530 | kfree(rdtgrp); |
531 | } | |
532 | ||
74fcdae1 | 533 | preempt_disable(); |
4f341a5e FY |
534 | /* update PQR_ASSOC MSR to make resource group go into effect */ |
535 | intel_rdt_sched_in(); | |
74fcdae1 | 536 | preempt_enable(); |
4f341a5e | 537 | |
e02737d5 FY |
538 | kfree(callback); |
539 | } | |
540 | ||
541 | static int __rdtgroup_move_task(struct task_struct *tsk, | |
542 | struct rdtgroup *rdtgrp) | |
543 | { | |
544 | struct task_move_callback *callback; | |
545 | int ret; | |
546 | ||
547 | callback = kzalloc(sizeof(*callback), GFP_KERNEL); | |
548 | if (!callback) | |
549 | return -ENOMEM; | |
550 | callback->work.func = move_myself; | |
551 | callback->rdtgrp = rdtgrp; | |
552 | ||
553 | /* | |
554 | * Take a refcount, so rdtgrp cannot be freed before the | |
555 | * callback has been invoked. | |
556 | */ | |
557 | atomic_inc(&rdtgrp->waitcount); | |
558 | ret = task_work_add(tsk, &callback->work, true); | |
559 | if (ret) { | |
560 | /* | |
561 | * Task is exiting. Drop the refcount and free the callback. | |
562 | * No need to check the refcount as the group cannot be | |
563 | * deleted before the write function unlocks rdtgroup_mutex. | |
564 | */ | |
565 | atomic_dec(&rdtgrp->waitcount); | |
566 | kfree(callback); | |
29e74f35 | 567 | rdt_last_cmd_puts("task exited\n"); |
e02737d5 | 568 | } else { |
d6aaba61 VS |
569 | /* |
570 | * For ctrl_mon groups move both closid and rmid. | |
571 | * For monitor groups, can move the tasks only from | |
572 | * their parent CTRL group. | |
573 | */ | |
574 | if (rdtgrp->type == RDTCTRL_GROUP) { | |
575 | tsk->closid = rdtgrp->closid; | |
576 | tsk->rmid = rdtgrp->mon.rmid; | |
577 | } else if (rdtgrp->type == RDTMON_GROUP) { | |
29e74f35 | 578 | if (rdtgrp->mon.parent->closid == tsk->closid) { |
d6aaba61 | 579 | tsk->rmid = rdtgrp->mon.rmid; |
29e74f35 TL |
580 | } else { |
581 | rdt_last_cmd_puts("Can't move task to different control group\n"); | |
d6aaba61 | 582 | ret = -EINVAL; |
29e74f35 | 583 | } |
d6aaba61 | 584 | } |
e02737d5 FY |
585 | } |
586 | return ret; | |
587 | } | |
588 | ||
f7a6e3f6 RC |
589 | /** |
590 | * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group | |
591 | * @r: Resource group | |
592 | * | |
593 | * Return: 1 if tasks have been assigned to @r, 0 otherwise | |
594 | */ | |
595 | int rdtgroup_tasks_assigned(struct rdtgroup *r) | |
596 | { | |
597 | struct task_struct *p, *t; | |
598 | int ret = 0; | |
599 | ||
600 | lockdep_assert_held(&rdtgroup_mutex); | |
601 | ||
602 | rcu_read_lock(); | |
603 | for_each_process_thread(p, t) { | |
604 | if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) || | |
605 | (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) { | |
606 | ret = 1; | |
607 | break; | |
608 | } | |
609 | } | |
610 | rcu_read_unlock(); | |
611 | ||
612 | return ret; | |
613 | } | |
614 | ||
e02737d5 FY |
615 | static int rdtgroup_task_write_permission(struct task_struct *task, |
616 | struct kernfs_open_file *of) | |
617 | { | |
618 | const struct cred *tcred = get_task_cred(task); | |
619 | const struct cred *cred = current_cred(); | |
620 | int ret = 0; | |
621 | ||
622 | /* | |
623 | * Even if we're attaching all tasks in the thread group, we only | |
624 | * need to check permissions on one of them. | |
625 | */ | |
626 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
627 | !uid_eq(cred->euid, tcred->uid) && | |
29e74f35 TL |
628 | !uid_eq(cred->euid, tcred->suid)) { |
629 | rdt_last_cmd_printf("No permission to move task %d\n", task->pid); | |
e02737d5 | 630 | ret = -EPERM; |
29e74f35 | 631 | } |
e02737d5 FY |
632 | |
633 | put_cred(tcred); | |
634 | return ret; | |
635 | } | |
636 | ||
637 | static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp, | |
638 | struct kernfs_open_file *of) | |
639 | { | |
640 | struct task_struct *tsk; | |
641 | int ret; | |
642 | ||
643 | rcu_read_lock(); | |
644 | if (pid) { | |
645 | tsk = find_task_by_vpid(pid); | |
646 | if (!tsk) { | |
647 | rcu_read_unlock(); | |
29e74f35 | 648 | rdt_last_cmd_printf("No task %d\n", pid); |
e02737d5 FY |
649 | return -ESRCH; |
650 | } | |
651 | } else { | |
652 | tsk = current; | |
653 | } | |
654 | ||
655 | get_task_struct(tsk); | |
656 | rcu_read_unlock(); | |
657 | ||
658 | ret = rdtgroup_task_write_permission(tsk, of); | |
659 | if (!ret) | |
660 | ret = __rdtgroup_move_task(tsk, rdtgrp); | |
661 | ||
662 | put_task_struct(tsk); | |
663 | return ret; | |
664 | } | |
665 | ||
666 | static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of, | |
667 | char *buf, size_t nbytes, loff_t off) | |
668 | { | |
669 | struct rdtgroup *rdtgrp; | |
670 | int ret = 0; | |
671 | pid_t pid; | |
672 | ||
673 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) | |
674 | return -EINVAL; | |
675 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
c966dac8 RC |
676 | if (!rdtgrp) { |
677 | rdtgroup_kn_unlock(of->kn); | |
678 | return -ENOENT; | |
679 | } | |
29e74f35 | 680 | rdt_last_cmd_clear(); |
e02737d5 | 681 | |
c966dac8 RC |
682 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || |
683 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { | |
684 | ret = -EINVAL; | |
685 | rdt_last_cmd_puts("pseudo-locking in progress\n"); | |
686 | goto unlock; | |
687 | } | |
688 | ||
689 | ret = rdtgroup_move_task(pid, rdtgrp, of); | |
e02737d5 | 690 | |
c966dac8 | 691 | unlock: |
e02737d5 FY |
692 | rdtgroup_kn_unlock(of->kn); |
693 | ||
694 | return ret ?: nbytes; | |
695 | } | |
696 | ||
697 | static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s) | |
698 | { | |
699 | struct task_struct *p, *t; | |
700 | ||
701 | rcu_read_lock(); | |
702 | for_each_process_thread(p, t) { | |
d6aaba61 VS |
703 | if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) || |
704 | (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) | |
e02737d5 FY |
705 | seq_printf(s, "%d\n", t->pid); |
706 | } | |
707 | rcu_read_unlock(); | |
708 | } | |
709 | ||
710 | static int rdtgroup_tasks_show(struct kernfs_open_file *of, | |
711 | struct seq_file *s, void *v) | |
712 | { | |
713 | struct rdtgroup *rdtgrp; | |
714 | int ret = 0; | |
715 | ||
716 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
717 | if (rdtgrp) | |
718 | show_rdt_tasks(rdtgrp, s); | |
719 | else | |
720 | ret = -ENOENT; | |
721 | rdtgroup_kn_unlock(of->kn); | |
722 | ||
723 | return ret; | |
724 | } | |
725 | ||
9b3a7fd0 TL |
726 | static int rdt_last_cmd_status_show(struct kernfs_open_file *of, |
727 | struct seq_file *seq, void *v) | |
728 | { | |
729 | int len; | |
730 | ||
731 | mutex_lock(&rdtgroup_mutex); | |
732 | len = seq_buf_used(&last_cmd_status); | |
733 | if (len) | |
734 | seq_printf(seq, "%.*s", len, last_cmd_status_buf); | |
735 | else | |
736 | seq_puts(seq, "ok\n"); | |
737 | mutex_unlock(&rdtgroup_mutex); | |
738 | return 0; | |
739 | } | |
740 | ||
4e978d06 FY |
741 | static int rdt_num_closids_show(struct kernfs_open_file *of, |
742 | struct seq_file *seq, void *v) | |
743 | { | |
744 | struct rdt_resource *r = of->kn->parent->priv; | |
745 | ||
746 | seq_printf(seq, "%d\n", r->num_closid); | |
4e978d06 FY |
747 | return 0; |
748 | } | |
749 | ||
2545e9f5 | 750 | static int rdt_default_ctrl_show(struct kernfs_open_file *of, |
4e978d06 FY |
751 | struct seq_file *seq, void *v) |
752 | { | |
753 | struct rdt_resource *r = of->kn->parent->priv; | |
754 | ||
2545e9f5 | 755 | seq_printf(seq, "%x\n", r->default_ctrl); |
4e978d06 FY |
756 | return 0; |
757 | } | |
758 | ||
53a114a6 SL |
759 | static int rdt_min_cbm_bits_show(struct kernfs_open_file *of, |
760 | struct seq_file *seq, void *v) | |
761 | { | |
762 | struct rdt_resource *r = of->kn->parent->priv; | |
763 | ||
d3e11b4d | 764 | seq_printf(seq, "%u\n", r->cache.min_cbm_bits); |
db69ef65 VS |
765 | return 0; |
766 | } | |
767 | ||
0dd2d749 FY |
768 | static int rdt_shareable_bits_show(struct kernfs_open_file *of, |
769 | struct seq_file *seq, void *v) | |
770 | { | |
771 | struct rdt_resource *r = of->kn->parent->priv; | |
772 | ||
773 | seq_printf(seq, "%x\n", r->cache.shareable_bits); | |
774 | return 0; | |
775 | } | |
776 | ||
e6519011 RC |
777 | /** |
778 | * rdt_bit_usage_show - Display current usage of resources | |
779 | * | |
780 | * A domain is a shared resource that can now be allocated differently. Here | |
781 | * we display the current regions of the domain as an annotated bitmask. | |
782 | * For each domain of this resource its allocation bitmask | |
783 | * is annotated as below to indicate the current usage of the corresponding bit: | |
784 | * 0 - currently unused | |
785 | * X - currently available for sharing and used by software and hardware | |
786 | * H - currently used by hardware only but available for software use | |
787 | * S - currently used and shareable by software only | |
788 | * E - currently used exclusively by one resource group | |
f4e80d67 | 789 | * P - currently pseudo-locked by one resource group |
e6519011 RC |
790 | */ |
791 | static int rdt_bit_usage_show(struct kernfs_open_file *of, | |
792 | struct seq_file *seq, void *v) | |
793 | { | |
794 | struct rdt_resource *r = of->kn->parent->priv; | |
f4e80d67 RC |
795 | u32 sw_shareable = 0, hw_shareable = 0; |
796 | u32 exclusive = 0, pseudo_locked = 0; | |
e6519011 | 797 | struct rdt_domain *dom; |
f4e80d67 | 798 | int i, hwb, swb, excl, psl; |
e6519011 RC |
799 | enum rdtgrp_mode mode; |
800 | bool sep = false; | |
801 | u32 *ctrl; | |
802 | ||
803 | mutex_lock(&rdtgroup_mutex); | |
804 | hw_shareable = r->cache.shareable_bits; | |
805 | list_for_each_entry(dom, &r->domains, list) { | |
806 | if (sep) | |
807 | seq_putc(seq, ';'); | |
808 | ctrl = dom->ctrl_val; | |
809 | sw_shareable = 0; | |
810 | exclusive = 0; | |
811 | seq_printf(seq, "%d=", dom->id); | |
47d53b18 | 812 | for (i = 0; i < closids_supported(); i++, ctrl++) { |
e6519011 RC |
813 | if (!closid_allocated(i)) |
814 | continue; | |
815 | mode = rdtgroup_mode_by_closid(i); | |
816 | switch (mode) { | |
817 | case RDT_MODE_SHAREABLE: | |
818 | sw_shareable |= *ctrl; | |
819 | break; | |
820 | case RDT_MODE_EXCLUSIVE: | |
821 | exclusive |= *ctrl; | |
822 | break; | |
f4e80d67 | 823 | case RDT_MODE_PSEUDO_LOCKSETUP: |
bb9fec69 | 824 | /* |
f4e80d67 RC |
825 | * RDT_MODE_PSEUDO_LOCKSETUP is possible |
826 | * here but not included since the CBM | |
827 | * associated with this CLOSID in this mode | |
828 | * is not initialized and no task or cpu can be | |
829 | * assigned this CLOSID. | |
bb9fec69 | 830 | */ |
f4e80d67 | 831 | break; |
bb9fec69 | 832 | case RDT_MODE_PSEUDO_LOCKED: |
e6519011 RC |
833 | case RDT_NUM_MODES: |
834 | WARN(1, | |
835 | "invalid mode for closid %d\n", i); | |
836 | break; | |
837 | } | |
838 | } | |
839 | for (i = r->cache.cbm_len - 1; i >= 0; i--) { | |
f4e80d67 | 840 | pseudo_locked = dom->plr ? dom->plr->cbm : 0; |
e6519011 RC |
841 | hwb = test_bit(i, (unsigned long *)&hw_shareable); |
842 | swb = test_bit(i, (unsigned long *)&sw_shareable); | |
843 | excl = test_bit(i, (unsigned long *)&exclusive); | |
f4e80d67 | 844 | psl = test_bit(i, (unsigned long *)&pseudo_locked); |
e6519011 RC |
845 | if (hwb && swb) |
846 | seq_putc(seq, 'X'); | |
847 | else if (hwb && !swb) | |
848 | seq_putc(seq, 'H'); | |
849 | else if (!hwb && swb) | |
850 | seq_putc(seq, 'S'); | |
851 | else if (excl) | |
852 | seq_putc(seq, 'E'); | |
f4e80d67 RC |
853 | else if (psl) |
854 | seq_putc(seq, 'P'); | |
e6519011 RC |
855 | else /* Unused bits remain */ |
856 | seq_putc(seq, '0'); | |
857 | } | |
858 | sep = true; | |
859 | } | |
860 | seq_putc(seq, '\n'); | |
861 | mutex_unlock(&rdtgroup_mutex); | |
862 | return 0; | |
863 | } | |
864 | ||
db69ef65 VS |
865 | static int rdt_min_bw_show(struct kernfs_open_file *of, |
866 | struct seq_file *seq, void *v) | |
867 | { | |
868 | struct rdt_resource *r = of->kn->parent->priv; | |
53a114a6 | 869 | |
db69ef65 VS |
870 | seq_printf(seq, "%u\n", r->membw.min_bw); |
871 | return 0; | |
872 | } | |
873 | ||
d4ab3320 VS |
874 | static int rdt_num_rmids_show(struct kernfs_open_file *of, |
875 | struct seq_file *seq, void *v) | |
876 | { | |
877 | struct rdt_resource *r = of->kn->parent->priv; | |
878 | ||
879 | seq_printf(seq, "%d\n", r->num_rmid); | |
880 | ||
881 | return 0; | |
882 | } | |
883 | ||
884 | static int rdt_mon_features_show(struct kernfs_open_file *of, | |
885 | struct seq_file *seq, void *v) | |
886 | { | |
887 | struct rdt_resource *r = of->kn->parent->priv; | |
888 | struct mon_evt *mevt; | |
889 | ||
890 | list_for_each_entry(mevt, &r->evt_list, list) | |
891 | seq_printf(seq, "%s\n", mevt->name); | |
892 | ||
893 | return 0; | |
894 | } | |
895 | ||
db69ef65 VS |
896 | static int rdt_bw_gran_show(struct kernfs_open_file *of, |
897 | struct seq_file *seq, void *v) | |
898 | { | |
899 | struct rdt_resource *r = of->kn->parent->priv; | |
900 | ||
901 | seq_printf(seq, "%u\n", r->membw.bw_gran); | |
902 | return 0; | |
903 | } | |
904 | ||
905 | static int rdt_delay_linear_show(struct kernfs_open_file *of, | |
906 | struct seq_file *seq, void *v) | |
907 | { | |
908 | struct rdt_resource *r = of->kn->parent->priv; | |
909 | ||
910 | seq_printf(seq, "%u\n", r->membw.delay_linear); | |
53a114a6 SL |
911 | return 0; |
912 | } | |
913 | ||
d4ab3320 VS |
914 | static int max_threshold_occ_show(struct kernfs_open_file *of, |
915 | struct seq_file *seq, void *v) | |
916 | { | |
917 | struct rdt_resource *r = of->kn->parent->priv; | |
918 | ||
919 | seq_printf(seq, "%u\n", intel_cqm_threshold * r->mon_scale); | |
920 | ||
921 | return 0; | |
922 | } | |
923 | ||
924 | static ssize_t max_threshold_occ_write(struct kernfs_open_file *of, | |
925 | char *buf, size_t nbytes, loff_t off) | |
926 | { | |
927 | struct rdt_resource *r = of->kn->parent->priv; | |
928 | unsigned int bytes; | |
929 | int ret; | |
930 | ||
931 | ret = kstrtouint(buf, 0, &bytes); | |
932 | if (ret) | |
933 | return ret; | |
934 | ||
935 | if (bytes > (boot_cpu_data.x86_cache_size * 1024)) | |
936 | return -EINVAL; | |
937 | ||
938 | intel_cqm_threshold = bytes / r->mon_scale; | |
939 | ||
5707b46a | 940 | return nbytes; |
d4ab3320 VS |
941 | } |
942 | ||
d48d7a57 RC |
943 | /* |
944 | * rdtgroup_mode_show - Display mode of this resource group | |
945 | */ | |
946 | static int rdtgroup_mode_show(struct kernfs_open_file *of, | |
947 | struct seq_file *s, void *v) | |
948 | { | |
949 | struct rdtgroup *rdtgrp; | |
950 | ||
951 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
952 | if (!rdtgrp) { | |
953 | rdtgroup_kn_unlock(of->kn); | |
954 | return -ENOENT; | |
955 | } | |
956 | ||
957 | seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode)); | |
958 | ||
959 | rdtgroup_kn_unlock(of->kn); | |
960 | return 0; | |
961 | } | |
962 | ||
49f7b4ef RC |
963 | /** |
964 | * rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other | |
965 | * @r: Resource to which domain instance @d belongs. | |
966 | * @d: The domain instance for which @closid is being tested. | |
967 | * @cbm: Capacity bitmask being tested. | |
968 | * @closid: Intended closid for @cbm. | |
969 | * @exclusive: Only check if overlaps with exclusive resource groups | |
970 | * | |
971 | * Checks if provided @cbm intended to be used for @closid on domain | |
972 | * @d overlaps with any other closids or other hardware usage associated | |
973 | * with this domain. If @exclusive is true then only overlaps with | |
974 | * resource groups in exclusive mode will be considered. If @exclusive | |
975 | * is false then overlaps with any resource group or hardware entities | |
976 | * will be considered. | |
977 | * | |
49e00eee RC |
978 | * @cbm is unsigned long, even if only 32 bits are used, to make the |
979 | * bitmap functions work correctly. | |
980 | * | |
49f7b4ef RC |
981 | * Return: false if CBM does not overlap, true if it does. |
982 | */ | |
9ab9aa15 | 983 | bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d, |
49e00eee | 984 | unsigned long cbm, int closid, bool exclusive) |
49f7b4ef | 985 | { |
49f7b4ef | 986 | enum rdtgrp_mode mode; |
49e00eee | 987 | unsigned long ctrl_b; |
49f7b4ef RC |
988 | u32 *ctrl; |
989 | int i; | |
990 | ||
991 | /* Check for any overlap with regions used by hardware directly */ | |
992 | if (!exclusive) { | |
49e00eee RC |
993 | ctrl_b = r->cache.shareable_bits; |
994 | if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) | |
49f7b4ef RC |
995 | return true; |
996 | } | |
997 | ||
998 | /* Check for overlap with other resource groups */ | |
999 | ctrl = d->ctrl_val; | |
f0df4e1a | 1000 | for (i = 0; i < closids_supported(); i++, ctrl++) { |
49e00eee | 1001 | ctrl_b = *ctrl; |
dfe9674b RC |
1002 | mode = rdtgroup_mode_by_closid(i); |
1003 | if (closid_allocated(i) && i != closid && | |
1004 | mode != RDT_MODE_PSEUDO_LOCKSETUP) { | |
49e00eee | 1005 | if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) { |
49f7b4ef RC |
1006 | if (exclusive) { |
1007 | if (mode == RDT_MODE_EXCLUSIVE) | |
1008 | return true; | |
1009 | continue; | |
1010 | } | |
1011 | return true; | |
1012 | } | |
1013 | } | |
1014 | } | |
1015 | ||
1016 | return false; | |
1017 | } | |
1018 | ||
1019 | /** | |
1020 | * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive | |
1021 | * | |
1022 | * An exclusive resource group implies that there should be no sharing of | |
1023 | * its allocated resources. At the time this group is considered to be | |
1024 | * exclusive this test can determine if its current schemata supports this | |
1025 | * setting by testing for overlap with all other resource groups. | |
1026 | * | |
1027 | * Return: true if resource group can be exclusive, false if there is overlap | |
1028 | * with allocations of other resource groups and thus this resource group | |
1029 | * cannot be exclusive. | |
1030 | */ | |
1031 | static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp) | |
1032 | { | |
1033 | int closid = rdtgrp->closid; | |
1034 | struct rdt_resource *r; | |
939b90b2 | 1035 | bool has_cache = false; |
49f7b4ef RC |
1036 | struct rdt_domain *d; |
1037 | ||
1038 | for_each_alloc_enabled_rdt_resource(r) { | |
939b90b2 RC |
1039 | if (r->rid == RDT_RESOURCE_MBA) |
1040 | continue; | |
1041 | has_cache = true; | |
49f7b4ef RC |
1042 | list_for_each_entry(d, &r->domains, list) { |
1043 | if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid], | |
939b90b2 RC |
1044 | rdtgrp->closid, false)) { |
1045 | rdt_last_cmd_puts("schemata overlaps\n"); | |
49f7b4ef | 1046 | return false; |
939b90b2 | 1047 | } |
49f7b4ef RC |
1048 | } |
1049 | } | |
1050 | ||
939b90b2 RC |
1051 | if (!has_cache) { |
1052 | rdt_last_cmd_puts("cannot be exclusive without CAT/CDP\n"); | |
1053 | return false; | |
1054 | } | |
1055 | ||
49f7b4ef RC |
1056 | return true; |
1057 | } | |
1058 | ||
1059 | /** | |
1060 | * rdtgroup_mode_write - Modify the resource group's mode | |
1061 | * | |
1062 | */ | |
d48d7a57 RC |
1063 | static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of, |
1064 | char *buf, size_t nbytes, loff_t off) | |
1065 | { | |
1066 | struct rdtgroup *rdtgrp; | |
1067 | enum rdtgrp_mode mode; | |
1068 | int ret = 0; | |
1069 | ||
1070 | /* Valid input requires a trailing newline */ | |
1071 | if (nbytes == 0 || buf[nbytes - 1] != '\n') | |
1072 | return -EINVAL; | |
1073 | buf[nbytes - 1] = '\0'; | |
1074 | ||
1075 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
1076 | if (!rdtgrp) { | |
1077 | rdtgroup_kn_unlock(of->kn); | |
1078 | return -ENOENT; | |
1079 | } | |
1080 | ||
1081 | rdt_last_cmd_clear(); | |
1082 | ||
1083 | mode = rdtgrp->mode; | |
1084 | ||
49f7b4ef | 1085 | if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) || |
dfe9674b RC |
1086 | (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) || |
1087 | (!strcmp(buf, "pseudo-locksetup") && | |
1088 | mode == RDT_MODE_PSEUDO_LOCKSETUP) || | |
1089 | (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED)) | |
d48d7a57 RC |
1090 | goto out; |
1091 | ||
dfe9674b RC |
1092 | if (mode == RDT_MODE_PSEUDO_LOCKED) { |
1093 | rdt_last_cmd_printf("cannot change pseudo-locked group\n"); | |
1094 | ret = -EINVAL; | |
1095 | goto out; | |
1096 | } | |
1097 | ||
d48d7a57 | 1098 | if (!strcmp(buf, "shareable")) { |
dfe9674b RC |
1099 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
1100 | ret = rdtgroup_locksetup_exit(rdtgrp); | |
1101 | if (ret) | |
1102 | goto out; | |
1103 | } | |
d48d7a57 | 1104 | rdtgrp->mode = RDT_MODE_SHAREABLE; |
49f7b4ef RC |
1105 | } else if (!strcmp(buf, "exclusive")) { |
1106 | if (!rdtgroup_mode_test_exclusive(rdtgrp)) { | |
49f7b4ef RC |
1107 | ret = -EINVAL; |
1108 | goto out; | |
1109 | } | |
dfe9674b RC |
1110 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
1111 | ret = rdtgroup_locksetup_exit(rdtgrp); | |
1112 | if (ret) | |
1113 | goto out; | |
1114 | } | |
49f7b4ef | 1115 | rdtgrp->mode = RDT_MODE_EXCLUSIVE; |
dfe9674b RC |
1116 | } else if (!strcmp(buf, "pseudo-locksetup")) { |
1117 | ret = rdtgroup_locksetup_enter(rdtgrp); | |
1118 | if (ret) | |
1119 | goto out; | |
1120 | rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP; | |
d48d7a57 RC |
1121 | } else { |
1122 | rdt_last_cmd_printf("unknown/unsupported mode\n"); | |
1123 | ret = -EINVAL; | |
1124 | } | |
1125 | ||
1126 | out: | |
1127 | rdtgroup_kn_unlock(of->kn); | |
1128 | return ret ?: nbytes; | |
1129 | } | |
1130 | ||
d9b48c86 RC |
1131 | /** |
1132 | * rdtgroup_cbm_to_size - Translate CBM to size in bytes | |
1133 | * @r: RDT resource to which @d belongs. | |
1134 | * @d: RDT domain instance. | |
1135 | * @cbm: bitmask for which the size should be computed. | |
1136 | * | |
1137 | * The bitmask provided associated with the RDT domain instance @d will be | |
1138 | * translated into how many bytes it represents. The size in bytes is | |
1139 | * computed by first dividing the total cache size by the CBM length to | |
1140 | * determine how many bytes each bit in the bitmask represents. The result | |
1141 | * is multiplied with the number of bits set in the bitmask. | |
49e00eee RC |
1142 | * |
1143 | * @cbm is unsigned long, even if only 32 bits are used to make the | |
1144 | * bitmap functions work correctly. | |
d9b48c86 RC |
1145 | */ |
1146 | unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, | |
49e00eee | 1147 | struct rdt_domain *d, unsigned long cbm) |
d9b48c86 RC |
1148 | { |
1149 | struct cpu_cacheinfo *ci; | |
1150 | unsigned int size = 0; | |
1151 | int num_b, i; | |
1152 | ||
49e00eee | 1153 | num_b = bitmap_weight(&cbm, r->cache.cbm_len); |
d9b48c86 RC |
1154 | ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask)); |
1155 | for (i = 0; i < ci->num_leaves; i++) { | |
1156 | if (ci->info_list[i].level == r->cache_level) { | |
1157 | size = ci->info_list[i].size / r->cache.cbm_len * num_b; | |
1158 | break; | |
1159 | } | |
1160 | } | |
1161 | ||
1162 | return size; | |
1163 | } | |
1164 | ||
1165 | /** | |
1166 | * rdtgroup_size_show - Display size in bytes of allocated regions | |
1167 | * | |
1168 | * The "size" file mirrors the layout of the "schemata" file, printing the | |
1169 | * size in bytes of each region instead of the capacity bitmask. | |
1170 | * | |
1171 | */ | |
1172 | static int rdtgroup_size_show(struct kernfs_open_file *of, | |
1173 | struct seq_file *s, void *v) | |
1174 | { | |
1175 | struct rdtgroup *rdtgrp; | |
1176 | struct rdt_resource *r; | |
1177 | struct rdt_domain *d; | |
1178 | unsigned int size; | |
f968dc11 RC |
1179 | bool sep; |
1180 | u32 ctrl; | |
d9b48c86 RC |
1181 | |
1182 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
1183 | if (!rdtgrp) { | |
1184 | rdtgroup_kn_unlock(of->kn); | |
1185 | return -ENOENT; | |
1186 | } | |
1187 | ||
f4e80d67 RC |
1188 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { |
1189 | seq_printf(s, "%*s:", max_name_width, rdtgrp->plr->r->name); | |
1190 | size = rdtgroup_cbm_to_size(rdtgrp->plr->r, | |
1191 | rdtgrp->plr->d, | |
1192 | rdtgrp->plr->cbm); | |
1193 | seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size); | |
1194 | goto out; | |
1195 | } | |
1196 | ||
d9b48c86 | 1197 | for_each_alloc_enabled_rdt_resource(r) { |
f968dc11 | 1198 | sep = false; |
d9b48c86 RC |
1199 | seq_printf(s, "%*s:", max_name_width, r->name); |
1200 | list_for_each_entry(d, &r->domains, list) { | |
1201 | if (sep) | |
1202 | seq_putc(s, ';'); | |
dfe9674b RC |
1203 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
1204 | size = 0; | |
1205 | } else { | |
f968dc11 RC |
1206 | ctrl = (!is_mba_sc(r) ? |
1207 | d->ctrl_val[rdtgrp->closid] : | |
1208 | d->mbps_val[rdtgrp->closid]); | |
1209 | if (r->rid == RDT_RESOURCE_MBA) | |
1210 | size = ctrl; | |
1211 | else | |
1212 | size = rdtgroup_cbm_to_size(r, d, ctrl); | |
dfe9674b | 1213 | } |
d9b48c86 RC |
1214 | seq_printf(s, "%d=%u", d->id, size); |
1215 | sep = true; | |
1216 | } | |
1217 | seq_putc(s, '\n'); | |
1218 | } | |
1219 | ||
f4e80d67 | 1220 | out: |
d9b48c86 RC |
1221 | rdtgroup_kn_unlock(of->kn); |
1222 | ||
1223 | return 0; | |
1224 | } | |
1225 | ||
4e978d06 | 1226 | /* rdtgroup information files for one cache resource. */ |
5dc1d5c6 | 1227 | static struct rftype res_common_files[] = { |
9b3a7fd0 TL |
1228 | { |
1229 | .name = "last_cmd_status", | |
1230 | .mode = 0444, | |
1231 | .kf_ops = &rdtgroup_kf_single_ops, | |
1232 | .seq_show = rdt_last_cmd_status_show, | |
1233 | .fflags = RF_TOP_INFO, | |
1234 | }, | |
4e978d06 FY |
1235 | { |
1236 | .name = "num_closids", | |
1237 | .mode = 0444, | |
1238 | .kf_ops = &rdtgroup_kf_single_ops, | |
1239 | .seq_show = rdt_num_closids_show, | |
5dc1d5c6 | 1240 | .fflags = RF_CTRL_INFO, |
4e978d06 | 1241 | }, |
d4ab3320 VS |
1242 | { |
1243 | .name = "mon_features", | |
1244 | .mode = 0444, | |
1245 | .kf_ops = &rdtgroup_kf_single_ops, | |
1246 | .seq_show = rdt_mon_features_show, | |
1247 | .fflags = RF_MON_INFO, | |
1248 | }, | |
1249 | { | |
1250 | .name = "num_rmids", | |
1251 | .mode = 0444, | |
1252 | .kf_ops = &rdtgroup_kf_single_ops, | |
1253 | .seq_show = rdt_num_rmids_show, | |
1254 | .fflags = RF_MON_INFO, | |
1255 | }, | |
4e978d06 FY |
1256 | { |
1257 | .name = "cbm_mask", | |
1258 | .mode = 0444, | |
1259 | .kf_ops = &rdtgroup_kf_single_ops, | |
2545e9f5 | 1260 | .seq_show = rdt_default_ctrl_show, |
5dc1d5c6 | 1261 | .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, |
4e978d06 | 1262 | }, |
53a114a6 SL |
1263 | { |
1264 | .name = "min_cbm_bits", | |
1265 | .mode = 0444, | |
1266 | .kf_ops = &rdtgroup_kf_single_ops, | |
1267 | .seq_show = rdt_min_cbm_bits_show, | |
5dc1d5c6 | 1268 | .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, |
db69ef65 | 1269 | }, |
0dd2d749 FY |
1270 | { |
1271 | .name = "shareable_bits", | |
1272 | .mode = 0444, | |
1273 | .kf_ops = &rdtgroup_kf_single_ops, | |
1274 | .seq_show = rdt_shareable_bits_show, | |
1275 | .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, | |
1276 | }, | |
e6519011 RC |
1277 | { |
1278 | .name = "bit_usage", | |
1279 | .mode = 0444, | |
1280 | .kf_ops = &rdtgroup_kf_single_ops, | |
1281 | .seq_show = rdt_bit_usage_show, | |
1282 | .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, | |
1283 | }, | |
db69ef65 VS |
1284 | { |
1285 | .name = "min_bandwidth", | |
1286 | .mode = 0444, | |
1287 | .kf_ops = &rdtgroup_kf_single_ops, | |
1288 | .seq_show = rdt_min_bw_show, | |
5dc1d5c6 | 1289 | .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, |
db69ef65 VS |
1290 | }, |
1291 | { | |
1292 | .name = "bandwidth_gran", | |
1293 | .mode = 0444, | |
1294 | .kf_ops = &rdtgroup_kf_single_ops, | |
1295 | .seq_show = rdt_bw_gran_show, | |
5dc1d5c6 | 1296 | .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, |
db69ef65 VS |
1297 | }, |
1298 | { | |
1299 | .name = "delay_linear", | |
1300 | .mode = 0444, | |
1301 | .kf_ops = &rdtgroup_kf_single_ops, | |
1302 | .seq_show = rdt_delay_linear_show, | |
5dc1d5c6 T |
1303 | .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, |
1304 | }, | |
d4ab3320 VS |
1305 | { |
1306 | .name = "max_threshold_occupancy", | |
1307 | .mode = 0644, | |
1308 | .kf_ops = &rdtgroup_kf_single_ops, | |
1309 | .write = max_threshold_occ_write, | |
1310 | .seq_show = max_threshold_occ_show, | |
1311 | .fflags = RF_MON_INFO | RFTYPE_RES_CACHE, | |
1312 | }, | |
5dc1d5c6 T |
1313 | { |
1314 | .name = "cpus", | |
1315 | .mode = 0644, | |
1316 | .kf_ops = &rdtgroup_kf_single_ops, | |
1317 | .write = rdtgroup_cpus_write, | |
1318 | .seq_show = rdtgroup_cpus_show, | |
1319 | .fflags = RFTYPE_BASE, | |
1320 | }, | |
1321 | { | |
1322 | .name = "cpus_list", | |
1323 | .mode = 0644, | |
1324 | .kf_ops = &rdtgroup_kf_single_ops, | |
1325 | .write = rdtgroup_cpus_write, | |
1326 | .seq_show = rdtgroup_cpus_show, | |
1327 | .flags = RFTYPE_FLAGS_CPUS_LIST, | |
1328 | .fflags = RFTYPE_BASE, | |
1329 | }, | |
1330 | { | |
1331 | .name = "tasks", | |
1332 | .mode = 0644, | |
1333 | .kf_ops = &rdtgroup_kf_single_ops, | |
1334 | .write = rdtgroup_tasks_write, | |
1335 | .seq_show = rdtgroup_tasks_show, | |
1336 | .fflags = RFTYPE_BASE, | |
1337 | }, | |
1338 | { | |
1339 | .name = "schemata", | |
1340 | .mode = 0644, | |
1341 | .kf_ops = &rdtgroup_kf_single_ops, | |
1342 | .write = rdtgroup_schemata_write, | |
1343 | .seq_show = rdtgroup_schemata_show, | |
1344 | .fflags = RF_CTRL_BASE, | |
db69ef65 | 1345 | }, |
d48d7a57 RC |
1346 | { |
1347 | .name = "mode", | |
1348 | .mode = 0644, | |
1349 | .kf_ops = &rdtgroup_kf_single_ops, | |
1350 | .write = rdtgroup_mode_write, | |
1351 | .seq_show = rdtgroup_mode_show, | |
1352 | .fflags = RF_CTRL_BASE, | |
1353 | }, | |
d9b48c86 RC |
1354 | { |
1355 | .name = "size", | |
1356 | .mode = 0444, | |
1357 | .kf_ops = &rdtgroup_kf_single_ops, | |
1358 | .seq_show = rdtgroup_size_show, | |
1359 | .fflags = RF_CTRL_BASE, | |
1360 | }, | |
1361 | ||
db69ef65 VS |
1362 | }; |
1363 | ||
5dc1d5c6 | 1364 | static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags) |
db69ef65 | 1365 | { |
5dc1d5c6 T |
1366 | struct rftype *rfts, *rft; |
1367 | int ret, len; | |
1368 | ||
1369 | rfts = res_common_files; | |
1370 | len = ARRAY_SIZE(res_common_files); | |
1371 | ||
1372 | lockdep_assert_held(&rdtgroup_mutex); | |
1373 | ||
1374 | for (rft = rfts; rft < rfts + len; rft++) { | |
1375 | if ((fflags & rft->fflags) == rft->fflags) { | |
1376 | ret = rdtgroup_add_file(kn, rft); | |
1377 | if (ret) | |
1378 | goto error; | |
1379 | } | |
1380 | } | |
1381 | ||
1382 | return 0; | |
1383 | error: | |
1384 | pr_warn("Failed to add %s, err=%d\n", rft->name, ret); | |
1385 | while (--rft >= rfts) { | |
1386 | if ((fflags & rft->fflags) == rft->fflags) | |
1387 | kernfs_remove_by_name(kn, rft->name); | |
1388 | } | |
1389 | return ret; | |
db69ef65 VS |
1390 | } |
1391 | ||
125db711 RC |
1392 | /** |
1393 | * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file | |
1394 | * @r: The resource group with which the file is associated. | |
1395 | * @name: Name of the file | |
1396 | * | |
1397 | * The permissions of named resctrl file, directory, or link are modified | |
1398 | * to not allow read, write, or execute by any user. | |
1399 | * | |
1400 | * WARNING: This function is intended to communicate to the user that the | |
1401 | * resctrl file has been locked down - that it is not relevant to the | |
1402 | * particular state the system finds itself in. It should not be relied | |
1403 | * on to protect from user access because after the file's permissions | |
1404 | * are restricted the user can still change the permissions using chmod | |
1405 | * from the command line. | |
1406 | * | |
1407 | * Return: 0 on success, <0 on failure. | |
1408 | */ | |
1409 | int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name) | |
1410 | { | |
1411 | struct iattr iattr = {.ia_valid = ATTR_MODE,}; | |
1412 | struct kernfs_node *kn; | |
1413 | int ret = 0; | |
1414 | ||
1415 | kn = kernfs_find_and_get_ns(r->kn, name, NULL); | |
1416 | if (!kn) | |
1417 | return -ENOENT; | |
1418 | ||
1419 | switch (kernfs_type(kn)) { | |
1420 | case KERNFS_DIR: | |
1421 | iattr.ia_mode = S_IFDIR; | |
1422 | break; | |
1423 | case KERNFS_FILE: | |
1424 | iattr.ia_mode = S_IFREG; | |
1425 | break; | |
1426 | case KERNFS_LINK: | |
1427 | iattr.ia_mode = S_IFLNK; | |
1428 | break; | |
1429 | } | |
1430 | ||
1431 | ret = kernfs_setattr(kn, &iattr); | |
1432 | kernfs_put(kn); | |
1433 | return ret; | |
1434 | } | |
1435 | ||
1436 | /** | |
1437 | * rdtgroup_kn_mode_restore - Restore user access to named resctrl file | |
1438 | * @r: The resource group with which the file is associated. | |
1439 | * @name: Name of the file | |
392487de | 1440 | * @mask: Mask of permissions that should be restored |
125db711 RC |
1441 | * |
1442 | * Restore the permissions of the named file. If @name is a directory the | |
1443 | * permissions of its parent will be used. | |
1444 | * | |
1445 | * Return: 0 on success, <0 on failure. | |
1446 | */ | |
392487de RC |
1447 | int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name, |
1448 | umode_t mask) | |
125db711 RC |
1449 | { |
1450 | struct iattr iattr = {.ia_valid = ATTR_MODE,}; | |
1451 | struct kernfs_node *kn, *parent; | |
1452 | struct rftype *rfts, *rft; | |
1453 | int ret, len; | |
1454 | ||
1455 | rfts = res_common_files; | |
1456 | len = ARRAY_SIZE(res_common_files); | |
1457 | ||
1458 | for (rft = rfts; rft < rfts + len; rft++) { | |
1459 | if (!strcmp(rft->name, name)) | |
392487de | 1460 | iattr.ia_mode = rft->mode & mask; |
125db711 RC |
1461 | } |
1462 | ||
1463 | kn = kernfs_find_and_get_ns(r->kn, name, NULL); | |
1464 | if (!kn) | |
1465 | return -ENOENT; | |
1466 | ||
1467 | switch (kernfs_type(kn)) { | |
1468 | case KERNFS_DIR: | |
1469 | parent = kernfs_get_parent(kn); | |
1470 | if (parent) { | |
1471 | iattr.ia_mode |= parent->mode; | |
1472 | kernfs_put(parent); | |
1473 | } | |
1474 | iattr.ia_mode |= S_IFDIR; | |
1475 | break; | |
1476 | case KERNFS_FILE: | |
1477 | iattr.ia_mode |= S_IFREG; | |
1478 | break; | |
1479 | case KERNFS_LINK: | |
1480 | iattr.ia_mode |= S_IFLNK; | |
1481 | break; | |
1482 | } | |
1483 | ||
1484 | ret = kernfs_setattr(kn, &iattr); | |
1485 | kernfs_put(kn); | |
1486 | return ret; | |
1487 | } | |
1488 | ||
5dc1d5c6 T |
1489 | static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name, |
1490 | unsigned long fflags) | |
6a507a6a | 1491 | { |
5dc1d5c6 T |
1492 | struct kernfs_node *kn_subdir; |
1493 | int ret; | |
1494 | ||
1495 | kn_subdir = kernfs_create_dir(kn_info, name, | |
1496 | kn_info->mode, r); | |
1497 | if (IS_ERR(kn_subdir)) | |
1498 | return PTR_ERR(kn_subdir); | |
1499 | ||
1500 | kernfs_get(kn_subdir); | |
1501 | ret = rdtgroup_kn_set_ugid(kn_subdir); | |
1502 | if (ret) | |
1503 | return ret; | |
1504 | ||
1505 | ret = rdtgroup_add_files(kn_subdir, fflags); | |
1506 | if (!ret) | |
1507 | kernfs_activate(kn_subdir); | |
1508 | ||
1509 | return ret; | |
6a507a6a VS |
1510 | } |
1511 | ||
4e978d06 FY |
1512 | static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn) |
1513 | { | |
4e978d06 | 1514 | struct rdt_resource *r; |
5dc1d5c6 | 1515 | unsigned long fflags; |
d4ab3320 | 1516 | char name[32]; |
5dc1d5c6 | 1517 | int ret; |
4e978d06 FY |
1518 | |
1519 | /* create the directory */ | |
1520 | kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL); | |
1521 | if (IS_ERR(kn_info)) | |
1522 | return PTR_ERR(kn_info); | |
1523 | kernfs_get(kn_info); | |
1524 | ||
9b3a7fd0 TL |
1525 | ret = rdtgroup_add_files(kn_info, RF_TOP_INFO); |
1526 | if (ret) | |
1527 | goto out_destroy; | |
1528 | ||
1b5c0b75 | 1529 | for_each_alloc_enabled_rdt_resource(r) { |
5dc1d5c6 T |
1530 | fflags = r->fflags | RF_CTRL_INFO; |
1531 | ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags); | |
4e978d06 FY |
1532 | if (ret) |
1533 | goto out_destroy; | |
4e978d06 | 1534 | } |
d4ab3320 VS |
1535 | |
1536 | for_each_mon_enabled_rdt_resource(r) { | |
1537 | fflags = r->fflags | RF_MON_INFO; | |
1538 | sprintf(name, "%s_MON", r->name); | |
1539 | ret = rdtgroup_mkdir_info_resdir(r, name, fflags); | |
1540 | if (ret) | |
1541 | goto out_destroy; | |
1542 | } | |
1543 | ||
4e978d06 FY |
1544 | /* |
1545 | * This extra ref will be put in kernfs_remove() and guarantees | |
1546 | * that @rdtgrp->kn is always accessible. | |
1547 | */ | |
1548 | kernfs_get(kn_info); | |
1549 | ||
1550 | ret = rdtgroup_kn_set_ugid(kn_info); | |
1551 | if (ret) | |
1552 | goto out_destroy; | |
1553 | ||
1554 | kernfs_activate(kn_info); | |
1555 | ||
1556 | return 0; | |
1557 | ||
1558 | out_destroy: | |
1559 | kernfs_remove(kn_info); | |
1560 | return ret; | |
1561 | } | |
1562 | ||
c7d9aac6 VS |
1563 | static int |
1564 | mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp, | |
1565 | char *name, struct kernfs_node **dest_kn) | |
1566 | { | |
1567 | struct kernfs_node *kn; | |
1568 | int ret; | |
1569 | ||
1570 | /* create the directory */ | |
1571 | kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); | |
1572 | if (IS_ERR(kn)) | |
1573 | return PTR_ERR(kn); | |
1574 | ||
1575 | if (dest_kn) | |
1576 | *dest_kn = kn; | |
1577 | ||
1578 | /* | |
1579 | * This extra ref will be put in kernfs_remove() and guarantees | |
1580 | * that @rdtgrp->kn is always accessible. | |
1581 | */ | |
1582 | kernfs_get(kn); | |
1583 | ||
1584 | ret = rdtgroup_kn_set_ugid(kn); | |
1585 | if (ret) | |
1586 | goto out_destroy; | |
1587 | ||
1588 | kernfs_activate(kn); | |
1589 | ||
1590 | return 0; | |
1591 | ||
1592 | out_destroy: | |
1593 | kernfs_remove(kn); | |
1594 | return ret; | |
1595 | } | |
99adde9b | 1596 | |
5ff193fb FY |
1597 | static void l3_qos_cfg_update(void *arg) |
1598 | { | |
1599 | bool *enable = arg; | |
1600 | ||
1601 | wrmsrl(IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL); | |
1602 | } | |
1603 | ||
99adde9b | 1604 | static void l2_qos_cfg_update(void *arg) |
5ff193fb | 1605 | { |
99adde9b FY |
1606 | bool *enable = arg; |
1607 | ||
1608 | wrmsrl(IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL); | |
1609 | } | |
1610 | ||
19c635ab VS |
1611 | static inline bool is_mba_linear(void) |
1612 | { | |
1613 | return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear; | |
1614 | } | |
1615 | ||
99adde9b FY |
1616 | static int set_cache_qos_cfg(int level, bool enable) |
1617 | { | |
1618 | void (*update)(void *arg); | |
1619 | struct rdt_resource *r_l; | |
5ff193fb FY |
1620 | cpumask_var_t cpu_mask; |
1621 | struct rdt_domain *d; | |
1622 | int cpu; | |
1623 | ||
1624 | if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) | |
1625 | return -ENOMEM; | |
1626 | ||
99adde9b FY |
1627 | if (level == RDT_RESOURCE_L3) |
1628 | update = l3_qos_cfg_update; | |
1629 | else if (level == RDT_RESOURCE_L2) | |
1630 | update = l2_qos_cfg_update; | |
1631 | else | |
1632 | return -EINVAL; | |
1633 | ||
1634 | r_l = &rdt_resources_all[level]; | |
1635 | list_for_each_entry(d, &r_l->domains, list) { | |
5ff193fb FY |
1636 | /* Pick one CPU from each domain instance to update MSR */ |
1637 | cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask); | |
1638 | } | |
1639 | cpu = get_cpu(); | |
1640 | /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */ | |
1641 | if (cpumask_test_cpu(cpu, cpu_mask)) | |
99adde9b | 1642 | update(&enable); |
5ff193fb | 1643 | /* Update QOS_CFG MSR on all other cpus in cpu_mask. */ |
99adde9b | 1644 | smp_call_function_many(cpu_mask, update, &enable, 1); |
5ff193fb FY |
1645 | put_cpu(); |
1646 | ||
1647 | free_cpumask_var(cpu_mask); | |
1648 | ||
1649 | return 0; | |
1650 | } | |
1651 | ||
19c635ab VS |
1652 | /* |
1653 | * Enable or disable the MBA software controller | |
1654 | * which helps user specify bandwidth in MBps. | |
1655 | * MBA software controller is supported only if | |
1656 | * MBM is supported and MBA is in linear scale. | |
1657 | */ | |
1658 | static int set_mba_sc(bool mba_sc) | |
1659 | { | |
1660 | struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA]; | |
1bd2a63b | 1661 | struct rdt_domain *d; |
19c635ab VS |
1662 | |
1663 | if (!is_mbm_enabled() || !is_mba_linear() || | |
1664 | mba_sc == is_mba_sc(r)) | |
1665 | return -EINVAL; | |
1666 | ||
1667 | r->membw.mba_sc = mba_sc; | |
1bd2a63b VS |
1668 | list_for_each_entry(d, &r->domains, list) |
1669 | setup_default_ctrlval(r, d->ctrl_val, d->mbps_val); | |
19c635ab VS |
1670 | |
1671 | return 0; | |
1672 | } | |
1673 | ||
99adde9b | 1674 | static int cdp_enable(int level, int data_type, int code_type) |
5ff193fb | 1675 | { |
99adde9b FY |
1676 | struct rdt_resource *r_ldata = &rdt_resources_all[data_type]; |
1677 | struct rdt_resource *r_lcode = &rdt_resources_all[code_type]; | |
1678 | struct rdt_resource *r_l = &rdt_resources_all[level]; | |
5ff193fb FY |
1679 | int ret; |
1680 | ||
99adde9b FY |
1681 | if (!r_l->alloc_capable || !r_ldata->alloc_capable || |
1682 | !r_lcode->alloc_capable) | |
5ff193fb FY |
1683 | return -EINVAL; |
1684 | ||
99adde9b | 1685 | ret = set_cache_qos_cfg(level, true); |
5ff193fb | 1686 | if (!ret) { |
99adde9b FY |
1687 | r_l->alloc_enabled = false; |
1688 | r_ldata->alloc_enabled = true; | |
1689 | r_lcode->alloc_enabled = true; | |
5ff193fb FY |
1690 | } |
1691 | return ret; | |
1692 | } | |
1693 | ||
99adde9b FY |
1694 | static int cdpl3_enable(void) |
1695 | { | |
1696 | return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, | |
1697 | RDT_RESOURCE_L3CODE); | |
1698 | } | |
1699 | ||
1700 | static int cdpl2_enable(void) | |
5ff193fb | 1701 | { |
99adde9b FY |
1702 | return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, |
1703 | RDT_RESOURCE_L2CODE); | |
1704 | } | |
1705 | ||
1706 | static void cdp_disable(int level, int data_type, int code_type) | |
1707 | { | |
1708 | struct rdt_resource *r = &rdt_resources_all[level]; | |
5ff193fb | 1709 | |
1b5c0b75 | 1710 | r->alloc_enabled = r->alloc_capable; |
5ff193fb | 1711 | |
99adde9b FY |
1712 | if (rdt_resources_all[data_type].alloc_enabled) { |
1713 | rdt_resources_all[data_type].alloc_enabled = false; | |
1714 | rdt_resources_all[code_type].alloc_enabled = false; | |
1715 | set_cache_qos_cfg(level, false); | |
5ff193fb FY |
1716 | } |
1717 | } | |
1718 | ||
99adde9b FY |
1719 | static void cdpl3_disable(void) |
1720 | { | |
1721 | cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE); | |
1722 | } | |
1723 | ||
1724 | static void cdpl2_disable(void) | |
1725 | { | |
1726 | cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE); | |
1727 | } | |
1728 | ||
1729 | static void cdp_disable_all(void) | |
1730 | { | |
1731 | if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled) | |
1732 | cdpl3_disable(); | |
1733 | if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled) | |
1734 | cdpl2_disable(); | |
1735 | } | |
1736 | ||
5ff193fb FY |
1737 | static int parse_rdtgroupfs_options(char *data) |
1738 | { | |
1739 | char *token, *o = data; | |
1740 | int ret = 0; | |
1741 | ||
1742 | while ((token = strsep(&o, ",")) != NULL) { | |
99adde9b FY |
1743 | if (!*token) { |
1744 | ret = -EINVAL; | |
1745 | goto out; | |
1746 | } | |
5ff193fb | 1747 | |
99adde9b FY |
1748 | if (!strcmp(token, "cdp")) { |
1749 | ret = cdpl3_enable(); | |
1750 | if (ret) | |
1751 | goto out; | |
1752 | } else if (!strcmp(token, "cdpl2")) { | |
1753 | ret = cdpl2_enable(); | |
1754 | if (ret) | |
1755 | goto out; | |
19c635ab VS |
1756 | } else if (!strcmp(token, "mba_MBps")) { |
1757 | ret = set_mba_sc(true); | |
1758 | if (ret) | |
1759 | goto out; | |
99adde9b FY |
1760 | } else { |
1761 | ret = -EINVAL; | |
1762 | goto out; | |
1763 | } | |
5ff193fb FY |
1764 | } |
1765 | ||
99adde9b FY |
1766 | return 0; |
1767 | ||
1768 | out: | |
1769 | pr_err("Invalid mount option \"%s\"\n", token); | |
1770 | ||
5ff193fb FY |
1771 | return ret; |
1772 | } | |
1773 | ||
60cf5e10 FY |
1774 | /* |
1775 | * We don't allow rdtgroup directories to be created anywhere | |
1776 | * except the root directory. Thus when looking for the rdtgroup | |
1777 | * structure for a kernfs node we are either looking at a directory, | |
1778 | * in which case the rdtgroup structure is pointed at by the "priv" | |
1779 | * field, otherwise we have a file, and need only look to the parent | |
1780 | * to find the rdtgroup. | |
1781 | */ | |
1782 | static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn) | |
1783 | { | |
f57b3087 FY |
1784 | if (kernfs_type(kn) == KERNFS_DIR) { |
1785 | /* | |
1786 | * All the resource directories use "kn->priv" | |
1787 | * to point to the "struct rdtgroup" for the | |
1788 | * resource. "info" and its subdirectories don't | |
1789 | * have rdtgroup structures, so return NULL here. | |
1790 | */ | |
1791 | if (kn == kn_info || kn->parent == kn_info) | |
1792 | return NULL; | |
1793 | else | |
1794 | return kn->priv; | |
1795 | } else { | |
60cf5e10 | 1796 | return kn->parent->priv; |
f57b3087 | 1797 | } |
60cf5e10 FY |
1798 | } |
1799 | ||
1800 | struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn) | |
1801 | { | |
1802 | struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); | |
1803 | ||
f57b3087 FY |
1804 | if (!rdtgrp) |
1805 | return NULL; | |
1806 | ||
60cf5e10 FY |
1807 | atomic_inc(&rdtgrp->waitcount); |
1808 | kernfs_break_active_protection(kn); | |
1809 | ||
1810 | mutex_lock(&rdtgroup_mutex); | |
1811 | ||
1812 | /* Was this group deleted while we waited? */ | |
1813 | if (rdtgrp->flags & RDT_DELETED) | |
1814 | return NULL; | |
1815 | ||
1816 | return rdtgrp; | |
1817 | } | |
1818 | ||
1819 | void rdtgroup_kn_unlock(struct kernfs_node *kn) | |
1820 | { | |
1821 | struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); | |
1822 | ||
f57b3087 FY |
1823 | if (!rdtgrp) |
1824 | return; | |
1825 | ||
60cf5e10 FY |
1826 | mutex_unlock(&rdtgroup_mutex); |
1827 | ||
1828 | if (atomic_dec_and_test(&rdtgrp->waitcount) && | |
1829 | (rdtgrp->flags & RDT_DELETED)) { | |
e0bdfe8e RC |
1830 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
1831 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) | |
1832 | rdtgroup_pseudo_lock_remove(rdtgrp); | |
60cf5e10 | 1833 | kernfs_unbreak_active_protection(kn); |
49ec8f5b | 1834 | kernfs_put(rdtgrp->kn); |
60cf5e10 FY |
1835 | kfree(rdtgrp); |
1836 | } else { | |
1837 | kernfs_unbreak_active_protection(kn); | |
1838 | } | |
1839 | } | |
1840 | ||
4af4a88e VS |
1841 | static int mkdir_mondata_all(struct kernfs_node *parent_kn, |
1842 | struct rdtgroup *prgrp, | |
1843 | struct kernfs_node **mon_data_kn); | |
1844 | ||
5ff193fb FY |
1845 | static struct dentry *rdt_mount(struct file_system_type *fs_type, |
1846 | int flags, const char *unused_dev_name, | |
1847 | void *data) | |
1848 | { | |
e3302683 VS |
1849 | struct rdt_domain *dom; |
1850 | struct rdt_resource *r; | |
5ff193fb FY |
1851 | struct dentry *dentry; |
1852 | int ret; | |
1853 | ||
87943db7 | 1854 | cpus_read_lock(); |
5ff193fb FY |
1855 | mutex_lock(&rdtgroup_mutex); |
1856 | /* | |
1857 | * resctrl file system can only be mounted once. | |
1858 | */ | |
4af4a88e | 1859 | if (static_branch_unlikely(&rdt_enable_key)) { |
5ff193fb FY |
1860 | dentry = ERR_PTR(-EBUSY); |
1861 | goto out; | |
1862 | } | |
1863 | ||
1864 | ret = parse_rdtgroupfs_options(data); | |
1865 | if (ret) { | |
1866 | dentry = ERR_PTR(ret); | |
1867 | goto out_cdp; | |
1868 | } | |
1869 | ||
60cf5e10 FY |
1870 | closid_init(); |
1871 | ||
4e978d06 | 1872 | ret = rdtgroup_create_info_dir(rdtgroup_default.kn); |
7bff0af5 SL |
1873 | if (ret) { |
1874 | dentry = ERR_PTR(ret); | |
4e978d06 | 1875 | goto out_cdp; |
7bff0af5 | 1876 | } |
4e978d06 | 1877 | |
4af4a88e VS |
1878 | if (rdt_mon_capable) { |
1879 | ret = mongroup_create_dir(rdtgroup_default.kn, | |
1880 | NULL, "mon_groups", | |
1881 | &kn_mongrp); | |
1882 | if (ret) { | |
1883 | dentry = ERR_PTR(ret); | |
1884 | goto out_info; | |
1885 | } | |
1886 | kernfs_get(kn_mongrp); | |
1887 | ||
1888 | ret = mkdir_mondata_all(rdtgroup_default.kn, | |
1889 | &rdtgroup_default, &kn_mondata); | |
1890 | if (ret) { | |
1891 | dentry = ERR_PTR(ret); | |
1892 | goto out_mongrp; | |
1893 | } | |
1894 | kernfs_get(kn_mondata); | |
1895 | rdtgroup_default.mon.mon_data_kn = kn_mondata; | |
1896 | } | |
1897 | ||
32206ab3 RC |
1898 | ret = rdt_pseudo_lock_init(); |
1899 | if (ret) { | |
1900 | dentry = ERR_PTR(ret); | |
1901 | goto out_mondata; | |
1902 | } | |
1903 | ||
5ff193fb FY |
1904 | dentry = kernfs_mount(fs_type, flags, rdt_root, |
1905 | RDTGROUP_SUPER_MAGIC, NULL); | |
1906 | if (IS_ERR(dentry)) | |
32206ab3 | 1907 | goto out_psl; |
4af4a88e VS |
1908 | |
1909 | if (rdt_alloc_capable) | |
87943db7 | 1910 | static_branch_enable_cpuslocked(&rdt_alloc_enable_key); |
4af4a88e | 1911 | if (rdt_mon_capable) |
87943db7 | 1912 | static_branch_enable_cpuslocked(&rdt_mon_enable_key); |
5ff193fb | 1913 | |
4af4a88e | 1914 | if (rdt_alloc_capable || rdt_mon_capable) |
87943db7 | 1915 | static_branch_enable_cpuslocked(&rdt_enable_key); |
e3302683 VS |
1916 | |
1917 | if (is_mbm_enabled()) { | |
1918 | r = &rdt_resources_all[RDT_RESOURCE_L3]; | |
1919 | list_for_each_entry(dom, &r->domains, list) | |
bbc4615e | 1920 | mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL); |
e3302683 VS |
1921 | } |
1922 | ||
5ff193fb FY |
1923 | goto out; |
1924 | ||
32206ab3 RC |
1925 | out_psl: |
1926 | rdt_pseudo_lock_release(); | |
4af4a88e VS |
1927 | out_mondata: |
1928 | if (rdt_mon_capable) | |
1929 | kernfs_remove(kn_mondata); | |
1930 | out_mongrp: | |
1931 | if (rdt_mon_capable) | |
1932 | kernfs_remove(kn_mongrp); | |
1933 | out_info: | |
79298acc | 1934 | kernfs_remove(kn_info); |
5ff193fb | 1935 | out_cdp: |
99adde9b | 1936 | cdp_disable_all(); |
5ff193fb | 1937 | out: |
9b3a7fd0 | 1938 | rdt_last_cmd_clear(); |
5ff193fb | 1939 | mutex_unlock(&rdtgroup_mutex); |
87943db7 | 1940 | cpus_read_unlock(); |
5ff193fb FY |
1941 | |
1942 | return dentry; | |
1943 | } | |
1944 | ||
2545e9f5 | 1945 | static int reset_all_ctrls(struct rdt_resource *r) |
5ff193fb FY |
1946 | { |
1947 | struct msr_param msr_param; | |
1948 | cpumask_var_t cpu_mask; | |
1949 | struct rdt_domain *d; | |
1950 | int i, cpu; | |
1951 | ||
1952 | if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) | |
1953 | return -ENOMEM; | |
1954 | ||
1955 | msr_param.res = r; | |
1956 | msr_param.low = 0; | |
1957 | msr_param.high = r->num_closid; | |
1958 | ||
1959 | /* | |
1960 | * Disable resource control for this resource by setting all | |
1961 | * CBMs in all domains to the maximum mask value. Pick one CPU | |
1962 | * from each domain to update the MSRs below. | |
1963 | */ | |
1964 | list_for_each_entry(d, &r->domains, list) { | |
1965 | cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask); | |
1966 | ||
1967 | for (i = 0; i < r->num_closid; i++) | |
2545e9f5 | 1968 | d->ctrl_val[i] = r->default_ctrl; |
5ff193fb FY |
1969 | } |
1970 | cpu = get_cpu(); | |
1971 | /* Update CBM on this cpu if it's in cpu_mask. */ | |
1972 | if (cpumask_test_cpu(cpu, cpu_mask)) | |
2545e9f5 | 1973 | rdt_ctrl_update(&msr_param); |
5ff193fb | 1974 | /* Update CBM on all other cpus in cpu_mask. */ |
2545e9f5 | 1975 | smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1); |
5ff193fb FY |
1976 | put_cpu(); |
1977 | ||
1978 | free_cpumask_var(cpu_mask); | |
1979 | ||
1980 | return 0; | |
1981 | } | |
1982 | ||
f3cbeaca VS |
1983 | static bool is_closid_match(struct task_struct *t, struct rdtgroup *r) |
1984 | { | |
1985 | return (rdt_alloc_capable && | |
1986 | (r->type == RDTCTRL_GROUP) && (t->closid == r->closid)); | |
1987 | } | |
1988 | ||
1989 | static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r) | |
1990 | { | |
1991 | return (rdt_mon_capable && | |
1992 | (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid)); | |
1993 | } | |
1994 | ||
4e978d06 | 1995 | /* |
0efc89be FY |
1996 | * Move tasks from one to the other group. If @from is NULL, then all tasks |
1997 | * in the systems are moved unconditionally (used for teardown). | |
1998 | * | |
1999 | * If @mask is not NULL the cpus on which moved tasks are running are set | |
2000 | * in that mask so the update smp function call is restricted to affected | |
2001 | * cpus. | |
4e978d06 | 2002 | */ |
0efc89be FY |
2003 | static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to, |
2004 | struct cpumask *mask) | |
4e978d06 | 2005 | { |
e02737d5 FY |
2006 | struct task_struct *p, *t; |
2007 | ||
e02737d5 | 2008 | read_lock(&tasklist_lock); |
0efc89be | 2009 | for_each_process_thread(p, t) { |
f3cbeaca VS |
2010 | if (!from || is_closid_match(t, from) || |
2011 | is_rmid_match(t, from)) { | |
0efc89be | 2012 | t->closid = to->closid; |
f3cbeaca VS |
2013 | t->rmid = to->mon.rmid; |
2014 | ||
0efc89be FY |
2015 | #ifdef CONFIG_SMP |
2016 | /* | |
2017 | * This is safe on x86 w/o barriers as the ordering | |
2018 | * of writing to task_cpu() and t->on_cpu is | |
2019 | * reverse to the reading here. The detection is | |
2020 | * inaccurate as tasks might move or schedule | |
2021 | * before the smp function call takes place. In | |
2022 | * such a case the function call is pointless, but | |
2023 | * there is no other side effect. | |
2024 | */ | |
2025 | if (mask && t->on_cpu) | |
2026 | cpumask_set_cpu(task_cpu(t), mask); | |
2027 | #endif | |
2028 | } | |
2029 | } | |
e02737d5 | 2030 | read_unlock(&tasklist_lock); |
0efc89be FY |
2031 | } |
2032 | ||
f3cbeaca VS |
2033 | static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp) |
2034 | { | |
2035 | struct rdtgroup *sentry, *stmp; | |
2036 | struct list_head *head; | |
2037 | ||
2038 | head = &rdtgrp->mon.crdtgrp_list; | |
2039 | list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) { | |
2040 | free_rmid(sentry->mon.rmid); | |
2041 | list_del(&sentry->mon.crdtgrp_list); | |
2042 | kfree(sentry); | |
2043 | } | |
2044 | } | |
2045 | ||
0efc89be FY |
2046 | /* |
2047 | * Forcibly remove all of subdirectories under root. | |
2048 | */ | |
2049 | static void rmdir_all_sub(void) | |
2050 | { | |
2051 | struct rdtgroup *rdtgrp, *tmp; | |
2052 | ||
2053 | /* Move all tasks to the default resource group */ | |
2054 | rdt_move_group_tasks(NULL, &rdtgroup_default, NULL); | |
60cf5e10 | 2055 | |
60cf5e10 | 2056 | list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) { |
4af4a88e VS |
2057 | /* Free any child rmids */ |
2058 | free_all_child_rdtgrp(rdtgrp); | |
2059 | ||
60cf5e10 FY |
2060 | /* Remove each rdtgroup other than root */ |
2061 | if (rdtgrp == &rdtgroup_default) | |
2062 | continue; | |
c7cc0cc1 | 2063 | |
e0bdfe8e RC |
2064 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
2065 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) | |
2066 | rdtgroup_pseudo_lock_remove(rdtgrp); | |
2067 | ||
c7cc0cc1 FY |
2068 | /* |
2069 | * Give any CPUs back to the default group. We cannot copy | |
2070 | * cpu_online_mask because a CPU might have executed the | |
2071 | * offline callback already, but is still marked online. | |
2072 | */ | |
2073 | cpumask_or(&rdtgroup_default.cpu_mask, | |
2074 | &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); | |
2075 | ||
4af4a88e VS |
2076 | free_rmid(rdtgrp->mon.rmid); |
2077 | ||
60cf5e10 FY |
2078 | kernfs_remove(rdtgrp->kn); |
2079 | list_del(&rdtgrp->rdtgroup_list); | |
2080 | kfree(rdtgrp); | |
2081 | } | |
0efc89be | 2082 | /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */ |
a9fcf862 | 2083 | update_closid_rmid(cpu_online_mask, &rdtgroup_default); |
0efc89be | 2084 | |
4e978d06 | 2085 | kernfs_remove(kn_info); |
4af4a88e VS |
2086 | kernfs_remove(kn_mongrp); |
2087 | kernfs_remove(kn_mondata); | |
4e978d06 FY |
2088 | } |
2089 | ||
5ff193fb FY |
2090 | static void rdt_kill_sb(struct super_block *sb) |
2091 | { | |
2092 | struct rdt_resource *r; | |
2093 | ||
36b6f9fc | 2094 | cpus_read_lock(); |
5ff193fb FY |
2095 | mutex_lock(&rdtgroup_mutex); |
2096 | ||
19c635ab VS |
2097 | set_mba_sc(false); |
2098 | ||
5ff193fb | 2099 | /*Put everything back to default values. */ |
1b5c0b75 | 2100 | for_each_alloc_enabled_rdt_resource(r) |
2545e9f5 | 2101 | reset_all_ctrls(r); |
99adde9b | 2102 | cdp_disable_all(); |
4e978d06 | 2103 | rmdir_all_sub(); |
746e0859 | 2104 | rdt_pseudo_lock_release(); |
472ef09b | 2105 | rdtgroup_default.mode = RDT_MODE_SHAREABLE; |
36b6f9fc RC |
2106 | static_branch_disable_cpuslocked(&rdt_alloc_enable_key); |
2107 | static_branch_disable_cpuslocked(&rdt_mon_enable_key); | |
2108 | static_branch_disable_cpuslocked(&rdt_enable_key); | |
5ff193fb FY |
2109 | kernfs_kill_sb(sb); |
2110 | mutex_unlock(&rdtgroup_mutex); | |
36b6f9fc | 2111 | cpus_read_unlock(); |
5ff193fb FY |
2112 | } |
2113 | ||
2114 | static struct file_system_type rdt_fs_type = { | |
2115 | .name = "resctrl", | |
2116 | .mount = rdt_mount, | |
2117 | .kill_sb = rdt_kill_sb, | |
2118 | }; | |
2119 | ||
d89b7379 VS |
2120 | static int mon_addfile(struct kernfs_node *parent_kn, const char *name, |
2121 | void *priv) | |
2122 | { | |
2123 | struct kernfs_node *kn; | |
2124 | int ret = 0; | |
2125 | ||
488dee96 DT |
2126 | kn = __kernfs_create_file(parent_kn, name, 0444, |
2127 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0, | |
d89b7379 VS |
2128 | &kf_mondata_ops, priv, NULL, NULL); |
2129 | if (IS_ERR(kn)) | |
2130 | return PTR_ERR(kn); | |
2131 | ||
2132 | ret = rdtgroup_kn_set_ugid(kn); | |
2133 | if (ret) { | |
2134 | kernfs_remove(kn); | |
2135 | return ret; | |
2136 | } | |
2137 | ||
2138 | return ret; | |
2139 | } | |
2140 | ||
895c663e VS |
2141 | /* |
2142 | * Remove all subdirectories of mon_data of ctrl_mon groups | |
2143 | * and monitor groups with given domain id. | |
2144 | */ | |
2145 | void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id) | |
2146 | { | |
2147 | struct rdtgroup *prgrp, *crgrp; | |
2148 | char name[32]; | |
2149 | ||
2150 | if (!r->mon_enabled) | |
2151 | return; | |
2152 | ||
2153 | list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { | |
2154 | sprintf(name, "mon_%s_%02d", r->name, dom_id); | |
2155 | kernfs_remove_by_name(prgrp->mon.mon_data_kn, name); | |
2156 | ||
2157 | list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list) | |
2158 | kernfs_remove_by_name(crgrp->mon.mon_data_kn, name); | |
2159 | } | |
2160 | } | |
2161 | ||
d89b7379 VS |
2162 | static int mkdir_mondata_subdir(struct kernfs_node *parent_kn, |
2163 | struct rdt_domain *d, | |
2164 | struct rdt_resource *r, struct rdtgroup *prgrp) | |
2165 | { | |
2166 | union mon_data_bits priv; | |
2167 | struct kernfs_node *kn; | |
2168 | struct mon_evt *mevt; | |
a4de1dfd | 2169 | struct rmid_read rr; |
d89b7379 VS |
2170 | char name[32]; |
2171 | int ret; | |
2172 | ||
2173 | sprintf(name, "mon_%s_%02d", r->name, d->id); | |
2174 | /* create the directory */ | |
2175 | kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); | |
2176 | if (IS_ERR(kn)) | |
2177 | return PTR_ERR(kn); | |
2178 | ||
2179 | /* | |
2180 | * This extra ref will be put in kernfs_remove() and guarantees | |
2181 | * that kn is always accessible. | |
2182 | */ | |
2183 | kernfs_get(kn); | |
2184 | ret = rdtgroup_kn_set_ugid(kn); | |
2185 | if (ret) | |
2186 | goto out_destroy; | |
2187 | ||
2188 | if (WARN_ON(list_empty(&r->evt_list))) { | |
2189 | ret = -EPERM; | |
2190 | goto out_destroy; | |
2191 | } | |
2192 | ||
2193 | priv.u.rid = r->rid; | |
2194 | priv.u.domid = d->id; | |
2195 | list_for_each_entry(mevt, &r->evt_list, list) { | |
2196 | priv.u.evtid = mevt->evtid; | |
2197 | ret = mon_addfile(kn, mevt->name, priv.priv); | |
2198 | if (ret) | |
2199 | goto out_destroy; | |
a4de1dfd VS |
2200 | |
2201 | if (is_mbm_event(mevt->evtid)) | |
2202 | mon_event_read(&rr, d, prgrp, mevt->evtid, true); | |
d89b7379 VS |
2203 | } |
2204 | kernfs_activate(kn); | |
2205 | return 0; | |
2206 | ||
2207 | out_destroy: | |
2208 | kernfs_remove(kn); | |
2209 | return ret; | |
2210 | } | |
2211 | ||
895c663e VS |
2212 | /* |
2213 | * Add all subdirectories of mon_data for "ctrl_mon" groups | |
2214 | * and "monitor" groups with given domain id. | |
2215 | */ | |
2216 | void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, | |
2217 | struct rdt_domain *d) | |
2218 | { | |
2219 | struct kernfs_node *parent_kn; | |
2220 | struct rdtgroup *prgrp, *crgrp; | |
2221 | struct list_head *head; | |
2222 | ||
2223 | if (!r->mon_enabled) | |
2224 | return; | |
2225 | ||
2226 | list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { | |
2227 | parent_kn = prgrp->mon.mon_data_kn; | |
2228 | mkdir_mondata_subdir(parent_kn, d, r, prgrp); | |
2229 | ||
2230 | head = &prgrp->mon.crdtgrp_list; | |
2231 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) { | |
2232 | parent_kn = crgrp->mon.mon_data_kn; | |
2233 | mkdir_mondata_subdir(parent_kn, d, r, crgrp); | |
2234 | } | |
2235 | } | |
2236 | } | |
2237 | ||
d89b7379 VS |
2238 | static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn, |
2239 | struct rdt_resource *r, | |
2240 | struct rdtgroup *prgrp) | |
2241 | { | |
2242 | struct rdt_domain *dom; | |
2243 | int ret; | |
2244 | ||
2245 | list_for_each_entry(dom, &r->domains, list) { | |
2246 | ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp); | |
2247 | if (ret) | |
2248 | return ret; | |
2249 | } | |
2250 | ||
2251 | return 0; | |
2252 | } | |
2253 | ||
2254 | /* | |
2255 | * This creates a directory mon_data which contains the monitored data. | |
2256 | * | |
2257 | * mon_data has one directory for each domain whic are named | |
2258 | * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data | |
2259 | * with L3 domain looks as below: | |
2260 | * ./mon_data: | |
2261 | * mon_L3_00 | |
2262 | * mon_L3_01 | |
2263 | * mon_L3_02 | |
2264 | * ... | |
2265 | * | |
2266 | * Each domain directory has one file per event: | |
2267 | * ./mon_L3_00/: | |
2268 | * llc_occupancy | |
2269 | * | |
2270 | */ | |
2271 | static int mkdir_mondata_all(struct kernfs_node *parent_kn, | |
2272 | struct rdtgroup *prgrp, | |
2273 | struct kernfs_node **dest_kn) | |
2274 | { | |
2275 | struct rdt_resource *r; | |
2276 | struct kernfs_node *kn; | |
2277 | int ret; | |
2278 | ||
2279 | /* | |
2280 | * Create the mon_data directory first. | |
2281 | */ | |
2282 | ret = mongroup_create_dir(parent_kn, NULL, "mon_data", &kn); | |
2283 | if (ret) | |
2284 | return ret; | |
2285 | ||
2286 | if (dest_kn) | |
2287 | *dest_kn = kn; | |
2288 | ||
2289 | /* | |
2290 | * Create the subdirectories for each domain. Note that all events | |
2291 | * in a domain like L3 are grouped into a resource whose domain is L3 | |
2292 | */ | |
2293 | for_each_mon_enabled_rdt_resource(r) { | |
2294 | ret = mkdir_mondata_subdir_alldom(kn, r, prgrp); | |
2295 | if (ret) | |
2296 | goto out_destroy; | |
2297 | } | |
2298 | ||
2299 | return 0; | |
2300 | ||
2301 | out_destroy: | |
2302 | kernfs_remove(kn); | |
2303 | return ret; | |
2304 | } | |
2305 | ||
95f0b77e RC |
2306 | /** |
2307 | * cbm_ensure_valid - Enforce validity on provided CBM | |
2308 | * @_val: Candidate CBM | |
2309 | * @r: RDT resource to which the CBM belongs | |
2310 | * | |
2311 | * The provided CBM represents all cache portions available for use. This | |
2312 | * may be represented by a bitmap that does not consist of contiguous ones | |
2313 | * and thus be an invalid CBM. | |
2314 | * Here the provided CBM is forced to be a valid CBM by only considering | |
2315 | * the first set of contiguous bits as valid and clearing all bits. | |
2316 | * The intention here is to provide a valid default CBM with which a new | |
2317 | * resource group is initialized. The user can follow this with a | |
2318 | * modification to the CBM if the default does not satisfy the | |
2319 | * requirements. | |
2320 | */ | |
2321 | static void cbm_ensure_valid(u32 *_val, struct rdt_resource *r) | |
2322 | { | |
2323 | /* | |
2324 | * Convert the u32 _val to an unsigned long required by all the bit | |
2325 | * operations within this function. No more than 32 bits of this | |
2326 | * converted value can be accessed because all bit operations are | |
2327 | * additionally provided with cbm_len that is initialized during | |
2328 | * hardware enumeration using five bits from the EAX register and | |
2329 | * thus never can exceed 32 bits. | |
2330 | */ | |
2331 | unsigned long *val = (unsigned long *)_val; | |
2332 | unsigned int cbm_len = r->cache.cbm_len; | |
2333 | unsigned long first_bit, zero_bit; | |
2334 | ||
2335 | if (*val == 0) | |
2336 | return; | |
2337 | ||
2338 | first_bit = find_first_bit(val, cbm_len); | |
2339 | zero_bit = find_next_zero_bit(val, cbm_len, first_bit); | |
2340 | ||
2341 | /* Clear any remaining bits to ensure contiguous region */ | |
2342 | bitmap_clear(val, zero_bit, cbm_len - zero_bit); | |
2343 | } | |
2344 | ||
2345 | /** | |
2346 | * rdtgroup_init_alloc - Initialize the new RDT group's allocations | |
2347 | * | |
2348 | * A new RDT group is being created on an allocation capable (CAT) | |
2349 | * supporting system. Set this group up to start off with all usable | |
2350 | * allocations. That is, all shareable and unused bits. | |
2351 | * | |
2352 | * All-zero CBM is invalid. If there are no more shareable bits available | |
2353 | * on any domain then the entire allocation will fail. | |
2354 | */ | |
2355 | static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp) | |
2356 | { | |
2357 | u32 used_b = 0, unused_b = 0; | |
2358 | u32 closid = rdtgrp->closid; | |
2359 | struct rdt_resource *r; | |
49e00eee | 2360 | unsigned long tmp_cbm; |
95f0b77e RC |
2361 | enum rdtgrp_mode mode; |
2362 | struct rdt_domain *d; | |
2363 | int i, ret; | |
2364 | u32 *ctrl; | |
2365 | ||
2366 | for_each_alloc_enabled_rdt_resource(r) { | |
70479c01 RC |
2367 | /* |
2368 | * Only initialize default allocations for CBM cache | |
2369 | * resources | |
2370 | */ | |
2371 | if (r->rid == RDT_RESOURCE_MBA) | |
2372 | continue; | |
95f0b77e RC |
2373 | list_for_each_entry(d, &r->domains, list) { |
2374 | d->have_new_ctrl = false; | |
2375 | d->new_ctrl = r->cache.shareable_bits; | |
2376 | used_b = r->cache.shareable_bits; | |
2377 | ctrl = d->ctrl_val; | |
ffb2315f | 2378 | for (i = 0; i < closids_supported(); i++, ctrl++) { |
95f0b77e RC |
2379 | if (closid_allocated(i) && i != closid) { |
2380 | mode = rdtgroup_mode_by_closid(i); | |
dfe9674b RC |
2381 | if (mode == RDT_MODE_PSEUDO_LOCKSETUP) |
2382 | break; | |
95f0b77e RC |
2383 | used_b |= *ctrl; |
2384 | if (mode == RDT_MODE_SHAREABLE) | |
2385 | d->new_ctrl |= *ctrl; | |
2386 | } | |
2387 | } | |
e0bdfe8e RC |
2388 | if (d->plr && d->plr->cbm > 0) |
2389 | used_b |= d->plr->cbm; | |
95f0b77e RC |
2390 | unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1); |
2391 | unused_b &= BIT_MASK(r->cache.cbm_len) - 1; | |
2392 | d->new_ctrl |= unused_b; | |
2393 | /* | |
2394 | * Force the initial CBM to be valid, user can | |
2395 | * modify the CBM based on system availability. | |
2396 | */ | |
2397 | cbm_ensure_valid(&d->new_ctrl, r); | |
49e00eee RC |
2398 | /* |
2399 | * Assign the u32 CBM to an unsigned long to ensure | |
2400 | * that bitmap_weight() does not access out-of-bound | |
2401 | * memory. | |
2402 | */ | |
2403 | tmp_cbm = d->new_ctrl; | |
2404 | if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < | |
2405 | r->cache.min_cbm_bits) { | |
95f0b77e RC |
2406 | rdt_last_cmd_printf("no space on %s:%d\n", |
2407 | r->name, d->id); | |
2408 | return -ENOSPC; | |
2409 | } | |
2410 | d->have_new_ctrl = true; | |
2411 | } | |
2412 | } | |
2413 | ||
2414 | for_each_alloc_enabled_rdt_resource(r) { | |
70479c01 RC |
2415 | /* |
2416 | * Only initialize default allocations for CBM cache | |
2417 | * resources | |
2418 | */ | |
2419 | if (r->rid == RDT_RESOURCE_MBA) | |
2420 | continue; | |
95f0b77e RC |
2421 | ret = update_domains(r, rdtgrp->closid); |
2422 | if (ret < 0) { | |
2423 | rdt_last_cmd_puts("failed to initialize allocations\n"); | |
2424 | return ret; | |
2425 | } | |
2426 | rdtgrp->mode = RDT_MODE_SHAREABLE; | |
2427 | } | |
2428 | ||
2429 | return 0; | |
2430 | } | |
2431 | ||
65b4f403 VS |
2432 | static int mkdir_rdt_prepare(struct kernfs_node *parent_kn, |
2433 | struct kernfs_node *prgrp_kn, | |
2434 | const char *name, umode_t mode, | |
c7d9aac6 | 2435 | enum rdt_group_type rtype, struct rdtgroup **r) |
60cf5e10 | 2436 | { |
65b4f403 | 2437 | struct rdtgroup *prdtgrp, *rdtgrp; |
60cf5e10 | 2438 | struct kernfs_node *kn; |
65b4f403 VS |
2439 | uint files = 0; |
2440 | int ret; | |
60cf5e10 | 2441 | |
65b4f403 | 2442 | prdtgrp = rdtgroup_kn_lock_live(prgrp_kn); |
cfd0f34e | 2443 | rdt_last_cmd_clear(); |
65b4f403 | 2444 | if (!prdtgrp) { |
60cf5e10 | 2445 | ret = -ENODEV; |
cfd0f34e | 2446 | rdt_last_cmd_puts("directory was removed\n"); |
60cf5e10 FY |
2447 | goto out_unlock; |
2448 | } | |
2449 | ||
c966dac8 RC |
2450 | if (rtype == RDTMON_GROUP && |
2451 | (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || | |
2452 | prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) { | |
2453 | ret = -EINVAL; | |
2454 | rdt_last_cmd_puts("pseudo-locking in progress\n"); | |
2455 | goto out_unlock; | |
2456 | } | |
2457 | ||
60cf5e10 FY |
2458 | /* allocate the rdtgroup. */ |
2459 | rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL); | |
2460 | if (!rdtgrp) { | |
2461 | ret = -ENOSPC; | |
cfd0f34e | 2462 | rdt_last_cmd_puts("kernel out of memory\n"); |
65b4f403 | 2463 | goto out_unlock; |
60cf5e10 | 2464 | } |
65b4f403 | 2465 | *r = rdtgrp; |
c7d9aac6 VS |
2466 | rdtgrp->mon.parent = prdtgrp; |
2467 | rdtgrp->type = rtype; | |
2468 | INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list); | |
60cf5e10 FY |
2469 | |
2470 | /* kernfs creates the directory for rdtgrp */ | |
65b4f403 | 2471 | kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp); |
60cf5e10 FY |
2472 | if (IS_ERR(kn)) { |
2473 | ret = PTR_ERR(kn); | |
cfd0f34e | 2474 | rdt_last_cmd_puts("kernfs create error\n"); |
65b4f403 | 2475 | goto out_free_rgrp; |
60cf5e10 FY |
2476 | } |
2477 | rdtgrp->kn = kn; | |
2478 | ||
2479 | /* | |
2480 | * kernfs_remove() will drop the reference count on "kn" which | |
2481 | * will free it. But we still need it to stick around for the | |
2482 | * rdtgroup_kn_unlock(kn} call below. Take one extra reference | |
2483 | * here, which will be dropped inside rdtgroup_kn_unlock(). | |
2484 | */ | |
2485 | kernfs_get(kn); | |
2486 | ||
2487 | ret = rdtgroup_kn_set_ugid(kn); | |
cfd0f34e TL |
2488 | if (ret) { |
2489 | rdt_last_cmd_puts("kernfs perm error\n"); | |
60cf5e10 | 2490 | goto out_destroy; |
cfd0f34e | 2491 | } |
60cf5e10 | 2492 | |
c7d9aac6 | 2493 | files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype); |
65b4f403 | 2494 | ret = rdtgroup_add_files(kn, files); |
cfd0f34e TL |
2495 | if (ret) { |
2496 | rdt_last_cmd_puts("kernfs fill error\n"); | |
12e0110c | 2497 | goto out_destroy; |
cfd0f34e | 2498 | } |
12e0110c | 2499 | |
c7d9aac6 VS |
2500 | if (rdt_mon_capable) { |
2501 | ret = alloc_rmid(); | |
cfd0f34e TL |
2502 | if (ret < 0) { |
2503 | rdt_last_cmd_puts("out of RMIDs\n"); | |
c7d9aac6 | 2504 | goto out_destroy; |
cfd0f34e | 2505 | } |
c7d9aac6 | 2506 | rdtgrp->mon.rmid = ret; |
d89b7379 VS |
2507 | |
2508 | ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn); | |
cfd0f34e TL |
2509 | if (ret) { |
2510 | rdt_last_cmd_puts("kernfs subdir error\n"); | |
d89b7379 | 2511 | goto out_idfree; |
cfd0f34e | 2512 | } |
c7d9aac6 | 2513 | } |
60cf5e10 FY |
2514 | kernfs_activate(kn); |
2515 | ||
65b4f403 VS |
2516 | /* |
2517 | * The caller unlocks the prgrp_kn upon success. | |
2518 | */ | |
2519 | return 0; | |
60cf5e10 | 2520 | |
d89b7379 VS |
2521 | out_idfree: |
2522 | free_rmid(rdtgrp->mon.rmid); | |
60cf5e10 FY |
2523 | out_destroy: |
2524 | kernfs_remove(rdtgrp->kn); | |
65b4f403 | 2525 | out_free_rgrp: |
60cf5e10 | 2526 | kfree(rdtgrp); |
60cf5e10 | 2527 | out_unlock: |
65b4f403 VS |
2528 | rdtgroup_kn_unlock(prgrp_kn); |
2529 | return ret; | |
2530 | } | |
2531 | ||
2532 | static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp) | |
2533 | { | |
2534 | kernfs_remove(rgrp->kn); | |
c7d9aac6 | 2535 | free_rmid(rgrp->mon.rmid); |
65b4f403 VS |
2536 | kfree(rgrp); |
2537 | } | |
2538 | ||
c7d9aac6 VS |
2539 | /* |
2540 | * Create a monitor group under "mon_groups" directory of a control | |
2541 | * and monitor group(ctrl_mon). This is a resource group | |
2542 | * to monitor a subset of tasks and cpus in its parent ctrl_mon group. | |
2543 | */ | |
2544 | static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn, | |
2545 | struct kernfs_node *prgrp_kn, | |
2546 | const char *name, | |
2547 | umode_t mode) | |
2548 | { | |
2549 | struct rdtgroup *rdtgrp, *prgrp; | |
2550 | int ret; | |
2551 | ||
2552 | ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTMON_GROUP, | |
2553 | &rdtgrp); | |
2554 | if (ret) | |
2555 | return ret; | |
2556 | ||
2557 | prgrp = rdtgrp->mon.parent; | |
2558 | rdtgrp->closid = prgrp->closid; | |
2559 | ||
2560 | /* | |
2561 | * Add the rdtgrp to the list of rdtgrps the parent | |
2562 | * ctrl_mon group has to track. | |
2563 | */ | |
2564 | list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list); | |
2565 | ||
2566 | rdtgroup_kn_unlock(prgrp_kn); | |
2567 | return ret; | |
2568 | } | |
2569 | ||
65b4f403 VS |
2570 | /* |
2571 | * These are rdtgroups created under the root directory. Can be used | |
c7d9aac6 | 2572 | * to allocate and monitor resources. |
65b4f403 | 2573 | */ |
c7d9aac6 VS |
2574 | static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn, |
2575 | struct kernfs_node *prgrp_kn, | |
2576 | const char *name, umode_t mode) | |
65b4f403 VS |
2577 | { |
2578 | struct rdtgroup *rdtgrp; | |
2579 | struct kernfs_node *kn; | |
2580 | u32 closid; | |
2581 | int ret; | |
2582 | ||
c7d9aac6 VS |
2583 | ret = mkdir_rdt_prepare(parent_kn, prgrp_kn, name, mode, RDTCTRL_GROUP, |
2584 | &rdtgrp); | |
65b4f403 VS |
2585 | if (ret) |
2586 | return ret; | |
2587 | ||
2588 | kn = rdtgrp->kn; | |
2589 | ret = closid_alloc(); | |
cfd0f34e TL |
2590 | if (ret < 0) { |
2591 | rdt_last_cmd_puts("out of CLOSIDs\n"); | |
65b4f403 | 2592 | goto out_common_fail; |
cfd0f34e | 2593 | } |
65b4f403 | 2594 | closid = ret; |
36e74d35 | 2595 | ret = 0; |
65b4f403 VS |
2596 | |
2597 | rdtgrp->closid = closid; | |
95f0b77e RC |
2598 | ret = rdtgroup_init_alloc(rdtgrp); |
2599 | if (ret < 0) | |
2600 | goto out_id_free; | |
2601 | ||
65b4f403 VS |
2602 | list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups); |
2603 | ||
c7d9aac6 VS |
2604 | if (rdt_mon_capable) { |
2605 | /* | |
2606 | * Create an empty mon_groups directory to hold the subset | |
2607 | * of tasks and cpus to monitor. | |
2608 | */ | |
2609 | ret = mongroup_create_dir(kn, NULL, "mon_groups", NULL); | |
cfd0f34e TL |
2610 | if (ret) { |
2611 | rdt_last_cmd_puts("kernfs subdir error\n"); | |
95f0b77e | 2612 | goto out_del_list; |
cfd0f34e | 2613 | } |
c7d9aac6 VS |
2614 | } |
2615 | ||
65b4f403 VS |
2616 | goto out_unlock; |
2617 | ||
95f0b77e RC |
2618 | out_del_list: |
2619 | list_del(&rdtgrp->rdtgroup_list); | |
c7d9aac6 VS |
2620 | out_id_free: |
2621 | closid_free(closid); | |
65b4f403 VS |
2622 | out_common_fail: |
2623 | mkdir_rdt_prepare_clean(rdtgrp); | |
2624 | out_unlock: | |
2625 | rdtgroup_kn_unlock(prgrp_kn); | |
60cf5e10 FY |
2626 | return ret; |
2627 | } | |
2628 | ||
c7d9aac6 VS |
2629 | /* |
2630 | * We allow creating mon groups only with in a directory called "mon_groups" | |
2631 | * which is present in every ctrl_mon group. Check if this is a valid | |
2632 | * "mon_groups" directory. | |
2633 | * | |
2634 | * 1. The directory should be named "mon_groups". | |
2635 | * 2. The mon group itself should "not" be named "mon_groups". | |
2636 | * This makes sure "mon_groups" directory always has a ctrl_mon group | |
2637 | * as parent. | |
2638 | */ | |
2639 | static bool is_mon_groups(struct kernfs_node *kn, const char *name) | |
2640 | { | |
2641 | return (!strcmp(kn->name, "mon_groups") && | |
2642 | strcmp(name, "mon_groups")); | |
2643 | } | |
2644 | ||
65b4f403 VS |
2645 | static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
2646 | umode_t mode) | |
2647 | { | |
2648 | /* Do not accept '\n' to avoid unparsable situation. */ | |
2649 | if (strchr(name, '\n')) | |
2650 | return -EINVAL; | |
2651 | ||
2652 | /* | |
2653 | * If the parent directory is the root directory and RDT | |
c7d9aac6 VS |
2654 | * allocation is supported, add a control and monitoring |
2655 | * subdirectory | |
65b4f403 VS |
2656 | */ |
2657 | if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn) | |
c7d9aac6 VS |
2658 | return rdtgroup_mkdir_ctrl_mon(parent_kn, parent_kn, name, mode); |
2659 | ||
2660 | /* | |
2661 | * If RDT monitoring is supported and the parent directory is a valid | |
2662 | * "mon_groups" directory, add a monitoring subdirectory. | |
2663 | */ | |
2664 | if (rdt_mon_capable && is_mon_groups(parent_kn, name)) | |
2665 | return rdtgroup_mkdir_mon(parent_kn, parent_kn->parent, name, mode); | |
65b4f403 VS |
2666 | |
2667 | return -EPERM; | |
2668 | } | |
2669 | ||
f3cbeaca VS |
2670 | static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp, |
2671 | cpumask_var_t tmpmask) | |
2672 | { | |
2673 | struct rdtgroup *prdtgrp = rdtgrp->mon.parent; | |
2674 | int cpu; | |
2675 | ||
2676 | /* Give any tasks back to the parent group */ | |
2677 | rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask); | |
2678 | ||
2679 | /* Update per cpu rmid of the moved CPUs first */ | |
2680 | for_each_cpu(cpu, &rdtgrp->cpu_mask) | |
a9110b55 | 2681 | per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid; |
f3cbeaca VS |
2682 | /* |
2683 | * Update the MSR on moved CPUs and CPUs which have moved | |
2684 | * task running on them. | |
2685 | */ | |
2686 | cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); | |
2687 | update_closid_rmid(tmpmask, NULL); | |
2688 | ||
2689 | rdtgrp->flags = RDT_DELETED; | |
2690 | free_rmid(rdtgrp->mon.rmid); | |
2691 | ||
2692 | /* | |
2693 | * Remove the rdtgrp from the parent ctrl_mon group's list | |
2694 | */ | |
2695 | WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list)); | |
2696 | list_del(&rdtgrp->mon.crdtgrp_list); | |
2697 | ||
2698 | /* | |
2699 | * one extra hold on this, will drop when we kfree(rdtgrp) | |
2700 | * in rdtgroup_kn_unlock() | |
2701 | */ | |
2702 | kernfs_get(kn); | |
2703 | kernfs_remove(rdtgrp->kn); | |
2704 | ||
2705 | return 0; | |
2706 | } | |
2707 | ||
17eafd07 RC |
2708 | static int rdtgroup_ctrl_remove(struct kernfs_node *kn, |
2709 | struct rdtgroup *rdtgrp) | |
2710 | { | |
2711 | rdtgrp->flags = RDT_DELETED; | |
2712 | list_del(&rdtgrp->rdtgroup_list); | |
2713 | ||
2714 | /* | |
2715 | * one extra hold on this, will drop when we kfree(rdtgrp) | |
2716 | * in rdtgroup_kn_unlock() | |
2717 | */ | |
2718 | kernfs_get(kn); | |
2719 | kernfs_remove(rdtgrp->kn); | |
2720 | return 0; | |
2721 | } | |
2722 | ||
f9049547 VS |
2723 | static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp, |
2724 | cpumask_var_t tmpmask) | |
60cf5e10 | 2725 | { |
f9049547 | 2726 | int cpu; |
60cf5e10 | 2727 | |
e02737d5 | 2728 | /* Give any tasks back to the default group */ |
0efc89be | 2729 | rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask); |
e02737d5 | 2730 | |
12e0110c TL |
2731 | /* Give any CPUs back to the default group */ |
2732 | cpumask_or(&rdtgroup_default.cpu_mask, | |
2733 | &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); | |
0efc89be | 2734 | |
f3cbeaca VS |
2735 | /* Update per cpu closid and rmid of the moved CPUs first */ |
2736 | for_each_cpu(cpu, &rdtgrp->cpu_mask) { | |
a9110b55 VS |
2737 | per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid; |
2738 | per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid; | |
f3cbeaca VS |
2739 | } |
2740 | ||
0efc89be FY |
2741 | /* |
2742 | * Update the MSR on moved CPUs and CPUs which have moved | |
2743 | * task running on them. | |
2744 | */ | |
2745 | cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); | |
a9fcf862 | 2746 | update_closid_rmid(tmpmask, NULL); |
12e0110c | 2747 | |
60cf5e10 | 2748 | closid_free(rdtgrp->closid); |
f3cbeaca VS |
2749 | free_rmid(rdtgrp->mon.rmid); |
2750 | ||
2751 | /* | |
2752 | * Free all the child monitor group rmids. | |
2753 | */ | |
2754 | free_all_child_rdtgrp(rdtgrp); | |
2755 | ||
17eafd07 | 2756 | rdtgroup_ctrl_remove(kn, rdtgrp); |
f9049547 VS |
2757 | |
2758 | return 0; | |
2759 | } | |
2760 | ||
2761 | static int rdtgroup_rmdir(struct kernfs_node *kn) | |
2762 | { | |
2763 | struct kernfs_node *parent_kn = kn->parent; | |
2764 | struct rdtgroup *rdtgrp; | |
2765 | cpumask_var_t tmpmask; | |
2766 | int ret = 0; | |
2767 | ||
2768 | if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
2769 | return -ENOMEM; | |
2770 | ||
2771 | rdtgrp = rdtgroup_kn_lock_live(kn); | |
2772 | if (!rdtgrp) { | |
2773 | ret = -EPERM; | |
2774 | goto out; | |
2775 | } | |
2776 | ||
2777 | /* | |
2778 | * If the rdtgroup is a ctrl_mon group and parent directory | |
f3cbeaca VS |
2779 | * is the root directory, remove the ctrl_mon group. |
2780 | * | |
2781 | * If the rdtgroup is a mon group and parent directory | |
2782 | * is a valid "mon_groups" directory, remove the mon group. | |
f9049547 | 2783 | */ |
e0bdfe8e RC |
2784 | if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn) { |
2785 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || | |
2786 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { | |
2787 | ret = rdtgroup_ctrl_remove(kn, rdtgrp); | |
2788 | } else { | |
2789 | ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask); | |
2790 | } | |
2791 | } else if (rdtgrp->type == RDTMON_GROUP && | |
2792 | is_mon_groups(parent_kn, kn->name)) { | |
f3cbeaca | 2793 | ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask); |
e0bdfe8e | 2794 | } else { |
f9049547 | 2795 | ret = -EPERM; |
e0bdfe8e | 2796 | } |
f9049547 | 2797 | |
0efc89be | 2798 | out: |
60cf5e10 | 2799 | rdtgroup_kn_unlock(kn); |
0efc89be FY |
2800 | free_cpumask_var(tmpmask); |
2801 | return ret; | |
60cf5e10 FY |
2802 | } |
2803 | ||
76ae054c SL |
2804 | static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf) |
2805 | { | |
1b5c0b75 | 2806 | if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled) |
76ae054c SL |
2807 | seq_puts(seq, ",cdp"); |
2808 | return 0; | |
2809 | } | |
2810 | ||
5ff193fb | 2811 | static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = { |
76ae054c SL |
2812 | .mkdir = rdtgroup_mkdir, |
2813 | .rmdir = rdtgroup_rmdir, | |
2814 | .show_options = rdtgroup_show_options, | |
5ff193fb FY |
2815 | }; |
2816 | ||
2817 | static int __init rdtgroup_setup_root(void) | |
2818 | { | |
12e0110c TL |
2819 | int ret; |
2820 | ||
5ff193fb | 2821 | rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops, |
21220bb1 RC |
2822 | KERNFS_ROOT_CREATE_DEACTIVATED | |
2823 | KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK, | |
5ff193fb FY |
2824 | &rdtgroup_default); |
2825 | if (IS_ERR(rdt_root)) | |
2826 | return PTR_ERR(rdt_root); | |
2827 | ||
2828 | mutex_lock(&rdtgroup_mutex); | |
2829 | ||
2830 | rdtgroup_default.closid = 0; | |
c7d9aac6 VS |
2831 | rdtgroup_default.mon.rmid = 0; |
2832 | rdtgroup_default.type = RDTCTRL_GROUP; | |
2833 | INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list); | |
2834 | ||
5ff193fb FY |
2835 | list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups); |
2836 | ||
5dc1d5c6 | 2837 | ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE); |
12e0110c TL |
2838 | if (ret) { |
2839 | kernfs_destroy_root(rdt_root); | |
2840 | goto out; | |
2841 | } | |
2842 | ||
5ff193fb FY |
2843 | rdtgroup_default.kn = rdt_root->kn; |
2844 | kernfs_activate(rdtgroup_default.kn); | |
2845 | ||
12e0110c | 2846 | out: |
5ff193fb FY |
2847 | mutex_unlock(&rdtgroup_mutex); |
2848 | ||
12e0110c | 2849 | return ret; |
5ff193fb FY |
2850 | } |
2851 | ||
2852 | /* | |
2853 | * rdtgroup_init - rdtgroup initialization | |
2854 | * | |
2855 | * Setup resctrl file system including set up root, create mount point, | |
2856 | * register rdtgroup filesystem, and initialize files under root directory. | |
2857 | * | |
2858 | * Return: 0 on success or -errno | |
2859 | */ | |
2860 | int __init rdtgroup_init(void) | |
2861 | { | |
2862 | int ret = 0; | |
2863 | ||
9b3a7fd0 TL |
2864 | seq_buf_init(&last_cmd_status, last_cmd_status_buf, |
2865 | sizeof(last_cmd_status_buf)); | |
2866 | ||
5ff193fb FY |
2867 | ret = rdtgroup_setup_root(); |
2868 | if (ret) | |
2869 | return ret; | |
2870 | ||
2871 | ret = sysfs_create_mount_point(fs_kobj, "resctrl"); | |
2872 | if (ret) | |
2873 | goto cleanup_root; | |
2874 | ||
2875 | ret = register_filesystem(&rdt_fs_type); | |
2876 | if (ret) | |
2877 | goto cleanup_mountpoint; | |
2878 | ||
37707ec6 RC |
2879 | /* |
2880 | * Adding the resctrl debugfs directory here may not be ideal since | |
2881 | * it would let the resctrl debugfs directory appear on the debugfs | |
2882 | * filesystem before the resctrl filesystem is mounted. | |
2883 | * It may also be ok since that would enable debugging of RDT before | |
2884 | * resctrl is mounted. | |
2885 | * The reason why the debugfs directory is created here and not in | |
2886 | * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and | |
2887 | * during the debugfs directory creation also &sb->s_type->i_mutex_key | |
2888 | * (the lockdep class of inode->i_rwsem). Other filesystem | |
2889 | * interactions (eg. SyS_getdents) have the lock ordering: | |
2890 | * &sb->s_type->i_mutex_key --> &mm->mmap_sem | |
2891 | * During mmap(), called with &mm->mmap_sem, the rdtgroup_mutex | |
2892 | * is taken, thus creating dependency: | |
2893 | * &mm->mmap_sem --> rdtgroup_mutex for the latter that can cause | |
2894 | * issues considering the other two lock dependencies. | |
2895 | * By creating the debugfs directory here we avoid a dependency | |
2896 | * that may cause deadlock (even though file operations cannot | |
2897 | * occur until the filesystem is mounted, but I do not know how to | |
2898 | * tell lockdep that). | |
2899 | */ | |
2900 | debugfs_resctrl = debugfs_create_dir("resctrl", NULL); | |
2901 | ||
5ff193fb FY |
2902 | return 0; |
2903 | ||
2904 | cleanup_mountpoint: | |
2905 | sysfs_remove_mount_point(fs_kobj, "resctrl"); | |
2906 | cleanup_root: | |
2907 | kernfs_destroy_root(rdt_root); | |
2908 | ||
2909 | return ret; | |
2910 | } | |
0af6a48d RC |
2911 | |
2912 | void __exit rdtgroup_exit(void) | |
2913 | { | |
37707ec6 | 2914 | debugfs_remove_recursive(debugfs_resctrl); |
0af6a48d RC |
2915 | unregister_filesystem(&rdt_fs_type); |
2916 | sysfs_remove_mount_point(fs_kobj, "resctrl"); | |
2917 | kernfs_destroy_root(rdt_root); | |
2918 | } |