]>
Commit | Line | Data |
---|---|---|
2025cf9e | 1 | // SPDX-License-Identifier: GPL-2.0-only |
5ff193fb FY |
2 | /* |
3 | * User interface for Resource Alloction in Resource Director Technology(RDT) | |
4 | * | |
5 | * Copyright (C) 2016 Intel Corporation | |
6 | * | |
7 | * Author: Fenghua Yu <fenghua.yu@intel.com> | |
8 | * | |
5ff193fb FY |
9 | * More information about RDT be found in the Intel (R) x86 Architecture |
10 | * Software Developer Manual. | |
11 | */ | |
12 | ||
13 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
14 | ||
d9b48c86 | 15 | #include <linux/cacheinfo.h> |
12e0110c | 16 | #include <linux/cpu.h> |
37707ec6 | 17 | #include <linux/debugfs.h> |
5ff193fb | 18 | #include <linux/fs.h> |
23bf1b6b | 19 | #include <linux/fs_parser.h> |
5ff193fb FY |
20 | #include <linux/sysfs.h> |
21 | #include <linux/kernfs.h> | |
9b3a7fd0 | 22 | #include <linux/seq_buf.h> |
4e978d06 | 23 | #include <linux/seq_file.h> |
3f07c014 | 24 | #include <linux/sched/signal.h> |
29930025 | 25 | #include <linux/sched/task.h> |
5ff193fb | 26 | #include <linux/slab.h> |
e02737d5 | 27 | #include <linux/task_work.h> |
23bf1b6b | 28 | #include <linux/user_namespace.h> |
5ff193fb FY |
29 | |
30 | #include <uapi/linux/magic.h> | |
31 | ||
8dd97c65 | 32 | #include <asm/resctrl.h> |
fa7d9493 | 33 | #include "internal.h" |
5ff193fb | 34 | |
4af4a88e VS |
35 | DEFINE_STATIC_KEY_FALSE(rdt_enable_key); |
36 | DEFINE_STATIC_KEY_FALSE(rdt_mon_enable_key); | |
1b5c0b75 | 37 | DEFINE_STATIC_KEY_FALSE(rdt_alloc_enable_key); |
cb2200e9 | 38 | static struct kernfs_root *rdt_root; |
5ff193fb FY |
39 | struct rdtgroup rdtgroup_default; |
40 | LIST_HEAD(rdt_all_groups); | |
41 | ||
4e978d06 FY |
42 | /* Kernel fs node for "info" directory under root */ |
43 | static struct kernfs_node *kn_info; | |
44 | ||
4af4a88e VS |
45 | /* Kernel fs node for "mon_groups" directory under root */ |
46 | static struct kernfs_node *kn_mongrp; | |
47 | ||
48 | /* Kernel fs node for "mon_data" directory under root */ | |
49 | static struct kernfs_node *kn_mondata; | |
50 | ||
9b3a7fd0 TL |
51 | static struct seq_buf last_cmd_status; |
52 | static char last_cmd_status_buf[512]; | |
53 | ||
37707ec6 RC |
54 | struct dentry *debugfs_resctrl; |
55 | ||
9b3a7fd0 TL |
56 | void rdt_last_cmd_clear(void) |
57 | { | |
58 | lockdep_assert_held(&rdtgroup_mutex); | |
59 | seq_buf_clear(&last_cmd_status); | |
60 | } | |
61 | ||
62 | void rdt_last_cmd_puts(const char *s) | |
63 | { | |
64 | lockdep_assert_held(&rdtgroup_mutex); | |
65 | seq_buf_puts(&last_cmd_status, s); | |
66 | } | |
67 | ||
68 | void rdt_last_cmd_printf(const char *fmt, ...) | |
69 | { | |
70 | va_list ap; | |
71 | ||
72 | va_start(ap, fmt); | |
73 | lockdep_assert_held(&rdtgroup_mutex); | |
74 | seq_buf_vprintf(&last_cmd_status, fmt, ap); | |
75 | va_end(ap); | |
76 | } | |
77 | ||
60cf5e10 FY |
78 | /* |
79 | * Trivial allocator for CLOSIDs. Since h/w only supports a small number, | |
80 | * we can keep a bitmap of free CLOSIDs in a single integer. | |
81 | * | |
82 | * Using a global CLOSID across all resources has some advantages and | |
83 | * some drawbacks: | |
84 | * + We can simply set "current->closid" to assign a task to a resource | |
85 | * group. | |
86 | * + Context switch code can avoid extra memory references deciding which | |
87 | * CLOSID to load into the PQR_ASSOC MSR | |
88 | * - We give up some options in configuring resource groups across multi-socket | |
89 | * systems. | |
90 | * - Our choices on how to configure each resource become progressively more | |
91 | * limited as the number of resources grows. | |
92 | */ | |
93 | static int closid_free_map; | |
c793da8e RC |
94 | static int closid_free_map_len; |
95 | ||
96 | int closids_supported(void) | |
97 | { | |
98 | return closid_free_map_len; | |
99 | } | |
60cf5e10 FY |
100 | |
101 | static void closid_init(void) | |
102 | { | |
103 | struct rdt_resource *r; | |
104 | int rdt_min_closid = 32; | |
105 | ||
106 | /* Compute rdt_min_closid across all resources */ | |
1b5c0b75 | 107 | for_each_alloc_enabled_rdt_resource(r) |
60cf5e10 FY |
108 | rdt_min_closid = min(rdt_min_closid, r->num_closid); |
109 | ||
110 | closid_free_map = BIT_MASK(rdt_min_closid) - 1; | |
111 | ||
112 | /* CLOSID 0 is always reserved for the default group */ | |
113 | closid_free_map &= ~1; | |
c793da8e | 114 | closid_free_map_len = rdt_min_closid; |
60cf5e10 FY |
115 | } |
116 | ||
cb2200e9 | 117 | static int closid_alloc(void) |
60cf5e10 | 118 | { |
0734ded1 | 119 | u32 closid = ffs(closid_free_map); |
60cf5e10 FY |
120 | |
121 | if (closid == 0) | |
122 | return -ENOSPC; | |
123 | closid--; | |
124 | closid_free_map &= ~(1 << closid); | |
125 | ||
126 | return closid; | |
127 | } | |
128 | ||
024d15be | 129 | void closid_free(int closid) |
60cf5e10 FY |
130 | { |
131 | closid_free_map |= 1 << closid; | |
132 | } | |
133 | ||
0b9aa656 RC |
134 | /** |
135 | * closid_allocated - test if provided closid is in use | |
136 | * @closid: closid to be tested | |
137 | * | |
138 | * Return: true if @closid is currently associated with a resource group, | |
139 | * false if @closid is free | |
140 | */ | |
95f0b77e | 141 | static bool closid_allocated(unsigned int closid) |
0b9aa656 RC |
142 | { |
143 | return (closid_free_map & (1 << closid)) == 0; | |
144 | } | |
145 | ||
472ef09b RC |
146 | /** |
147 | * rdtgroup_mode_by_closid - Return mode of resource group with closid | |
148 | * @closid: closid if the resource group | |
149 | * | |
150 | * Each resource group is associated with a @closid. Here the mode | |
151 | * of a resource group can be queried by searching for it using its closid. | |
152 | * | |
153 | * Return: mode as &enum rdtgrp_mode of resource group with closid @closid | |
154 | */ | |
155 | enum rdtgrp_mode rdtgroup_mode_by_closid(int closid) | |
156 | { | |
157 | struct rdtgroup *rdtgrp; | |
158 | ||
159 | list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) { | |
160 | if (rdtgrp->closid == closid) | |
161 | return rdtgrp->mode; | |
162 | } | |
163 | ||
164 | return RDT_NUM_MODES; | |
165 | } | |
166 | ||
d48d7a57 | 167 | static const char * const rdt_mode_str[] = { |
bb9fec69 RC |
168 | [RDT_MODE_SHAREABLE] = "shareable", |
169 | [RDT_MODE_EXCLUSIVE] = "exclusive", | |
170 | [RDT_MODE_PSEUDO_LOCKSETUP] = "pseudo-locksetup", | |
171 | [RDT_MODE_PSEUDO_LOCKED] = "pseudo-locked", | |
d48d7a57 RC |
172 | }; |
173 | ||
174 | /** | |
175 | * rdtgroup_mode_str - Return the string representation of mode | |
176 | * @mode: the resource group mode as &enum rdtgroup_mode | |
177 | * | |
178 | * Return: string representation of valid mode, "unknown" otherwise | |
179 | */ | |
180 | static const char *rdtgroup_mode_str(enum rdtgrp_mode mode) | |
181 | { | |
182 | if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES) | |
183 | return "unknown"; | |
184 | ||
185 | return rdt_mode_str[mode]; | |
186 | } | |
187 | ||
4e978d06 FY |
188 | /* set uid and gid of rdtgroup dirs and files to that of the creator */ |
189 | static int rdtgroup_kn_set_ugid(struct kernfs_node *kn) | |
190 | { | |
191 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
192 | .ia_uid = current_fsuid(), | |
193 | .ia_gid = current_fsgid(), }; | |
194 | ||
195 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
196 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
197 | return 0; | |
198 | ||
199 | return kernfs_setattr(kn, &iattr); | |
200 | } | |
201 | ||
202 | static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft) | |
203 | { | |
204 | struct kernfs_node *kn; | |
205 | int ret; | |
206 | ||
207 | kn = __kernfs_create_file(parent_kn, rft->name, rft->mode, | |
488dee96 | 208 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, |
4e978d06 FY |
209 | 0, rft->kf_ops, rft, NULL, NULL); |
210 | if (IS_ERR(kn)) | |
211 | return PTR_ERR(kn); | |
212 | ||
213 | ret = rdtgroup_kn_set_ugid(kn); | |
214 | if (ret) { | |
215 | kernfs_remove(kn); | |
216 | return ret; | |
217 | } | |
218 | ||
219 | return 0; | |
220 | } | |
221 | ||
4e978d06 FY |
222 | static int rdtgroup_seqfile_show(struct seq_file *m, void *arg) |
223 | { | |
224 | struct kernfs_open_file *of = m->private; | |
225 | struct rftype *rft = of->kn->priv; | |
226 | ||
227 | if (rft->seq_show) | |
228 | return rft->seq_show(of, m, arg); | |
229 | return 0; | |
230 | } | |
231 | ||
232 | static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf, | |
233 | size_t nbytes, loff_t off) | |
234 | { | |
235 | struct rftype *rft = of->kn->priv; | |
236 | ||
237 | if (rft->write) | |
238 | return rft->write(of, buf, nbytes, off); | |
239 | ||
240 | return -EINVAL; | |
241 | } | |
242 | ||
243 | static struct kernfs_ops rdtgroup_kf_single_ops = { | |
244 | .atomic_write_len = PAGE_SIZE, | |
245 | .write = rdtgroup_file_write, | |
246 | .seq_show = rdtgroup_seqfile_show, | |
247 | }; | |
248 | ||
d89b7379 VS |
249 | static struct kernfs_ops kf_mondata_ops = { |
250 | .atomic_write_len = PAGE_SIZE, | |
251 | .seq_show = rdtgroup_mondata_show, | |
252 | }; | |
253 | ||
4ffa3c97 JO |
254 | static bool is_cpu_list(struct kernfs_open_file *of) |
255 | { | |
256 | struct rftype *rft = of->kn->priv; | |
257 | ||
258 | return rft->flags & RFTYPE_FLAGS_CPUS_LIST; | |
259 | } | |
260 | ||
12e0110c TL |
261 | static int rdtgroup_cpus_show(struct kernfs_open_file *of, |
262 | struct seq_file *s, void *v) | |
263 | { | |
264 | struct rdtgroup *rdtgrp; | |
b61b8bba | 265 | struct cpumask *mask; |
12e0110c TL |
266 | int ret = 0; |
267 | ||
268 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
269 | ||
4ffa3c97 | 270 | if (rdtgrp) { |
b61b8bba JJ |
271 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { |
272 | if (!rdtgrp->plr->d) { | |
273 | rdt_last_cmd_clear(); | |
274 | rdt_last_cmd_puts("Cache domain offline\n"); | |
275 | ret = -ENODEV; | |
276 | } else { | |
277 | mask = &rdtgrp->plr->d->cpu_mask; | |
278 | seq_printf(s, is_cpu_list(of) ? | |
279 | "%*pbl\n" : "%*pb\n", | |
280 | cpumask_pr_args(mask)); | |
281 | } | |
282 | } else { | |
33dc3e41 RC |
283 | seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n", |
284 | cpumask_pr_args(&rdtgrp->cpu_mask)); | |
b61b8bba | 285 | } |
4ffa3c97 | 286 | } else { |
12e0110c | 287 | ret = -ENOENT; |
4ffa3c97 | 288 | } |
12e0110c TL |
289 | rdtgroup_kn_unlock(of->kn); |
290 | ||
291 | return ret; | |
292 | } | |
293 | ||
f4107702 | 294 | /* |
352940ec | 295 | * This is safe against resctrl_sched_in() called from __switch_to() |
f4107702 | 296 | * because __switch_to() is executed with interrupts disabled. A local call |
a9fcf862 | 297 | * from update_closid_rmid() is proteced against __switch_to() because |
f4107702 FY |
298 | * preemption is disabled. |
299 | */ | |
a9fcf862 | 300 | static void update_cpu_closid_rmid(void *info) |
f4107702 | 301 | { |
b09d981b VS |
302 | struct rdtgroup *r = info; |
303 | ||
a9fcf862 | 304 | if (r) { |
a9110b55 VS |
305 | this_cpu_write(pqr_state.default_closid, r->closid); |
306 | this_cpu_write(pqr_state.default_rmid, r->mon.rmid); | |
a9fcf862 | 307 | } |
b09d981b | 308 | |
f4107702 FY |
309 | /* |
310 | * We cannot unconditionally write the MSR because the current | |
311 | * executing task might have its own closid selected. Just reuse | |
312 | * the context switch code. | |
313 | */ | |
352940ec | 314 | resctrl_sched_in(); |
f4107702 FY |
315 | } |
316 | ||
0efc89be FY |
317 | /* |
318 | * Update the PGR_ASSOC MSR on all cpus in @cpu_mask, | |
319 | * | |
b09d981b | 320 | * Per task closids/rmids must have been set up before calling this function. |
0efc89be FY |
321 | */ |
322 | static void | |
a9fcf862 | 323 | update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r) |
f4107702 FY |
324 | { |
325 | int cpu = get_cpu(); | |
326 | ||
327 | if (cpumask_test_cpu(cpu, cpu_mask)) | |
a9fcf862 VS |
328 | update_cpu_closid_rmid(r); |
329 | smp_call_function_many(cpu_mask, update_cpu_closid_rmid, r, 1); | |
f4107702 FY |
330 | put_cpu(); |
331 | } | |
332 | ||
a9fcf862 VS |
333 | static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, |
334 | cpumask_var_t tmpmask) | |
335 | { | |
336 | struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp; | |
337 | struct list_head *head; | |
338 | ||
339 | /* Check whether cpus belong to parent ctrl group */ | |
340 | cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask); | |
94457b36 | 341 | if (cpumask_weight(tmpmask)) { |
723f1a0d | 342 | rdt_last_cmd_puts("Can only add CPUs to mongroup that belong to parent\n"); |
a9fcf862 | 343 | return -EINVAL; |
94457b36 | 344 | } |
a9fcf862 VS |
345 | |
346 | /* Check whether cpus are dropped from this group */ | |
347 | cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); | |
348 | if (cpumask_weight(tmpmask)) { | |
349 | /* Give any dropped cpus to parent rdtgroup */ | |
350 | cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask); | |
351 | update_closid_rmid(tmpmask, prgrp); | |
352 | } | |
353 | ||
354 | /* | |
355 | * If we added cpus, remove them from previous group that owned them | |
356 | * and update per-cpu rmid | |
357 | */ | |
358 | cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); | |
359 | if (cpumask_weight(tmpmask)) { | |
360 | head = &prgrp->mon.crdtgrp_list; | |
361 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) { | |
362 | if (crgrp == rdtgrp) | |
363 | continue; | |
364 | cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask, | |
365 | tmpmask); | |
366 | } | |
367 | update_closid_rmid(tmpmask, rdtgrp); | |
368 | } | |
369 | ||
370 | /* Done pushing/pulling - update this group with new mask */ | |
371 | cpumask_copy(&rdtgrp->cpu_mask, newmask); | |
372 | ||
373 | return 0; | |
374 | } | |
375 | ||
376 | static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m) | |
377 | { | |
378 | struct rdtgroup *crgrp; | |
379 | ||
380 | cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m); | |
381 | /* update the child mon group masks as well*/ | |
382 | list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list) | |
383 | cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask); | |
384 | } | |
385 | ||
b09d981b | 386 | static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, |
a9fcf862 | 387 | cpumask_var_t tmpmask, cpumask_var_t tmpmask1) |
b09d981b | 388 | { |
a9fcf862 VS |
389 | struct rdtgroup *r, *crgrp; |
390 | struct list_head *head; | |
b09d981b VS |
391 | |
392 | /* Check whether cpus are dropped from this group */ | |
393 | cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); | |
394 | if (cpumask_weight(tmpmask)) { | |
395 | /* Can't drop from default group */ | |
94457b36 TL |
396 | if (rdtgrp == &rdtgroup_default) { |
397 | rdt_last_cmd_puts("Can't drop CPUs from default group\n"); | |
b09d981b | 398 | return -EINVAL; |
94457b36 | 399 | } |
b09d981b VS |
400 | |
401 | /* Give any dropped cpus to rdtgroup_default */ | |
402 | cpumask_or(&rdtgroup_default.cpu_mask, | |
403 | &rdtgroup_default.cpu_mask, tmpmask); | |
a9fcf862 | 404 | update_closid_rmid(tmpmask, &rdtgroup_default); |
b09d981b VS |
405 | } |
406 | ||
407 | /* | |
a9fcf862 VS |
408 | * If we added cpus, remove them from previous group and |
409 | * the prev group's child groups that owned them | |
410 | * and update per-cpu closid/rmid. | |
b09d981b VS |
411 | */ |
412 | cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); | |
413 | if (cpumask_weight(tmpmask)) { | |
414 | list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) { | |
415 | if (r == rdtgrp) | |
416 | continue; | |
a9fcf862 VS |
417 | cpumask_and(tmpmask1, &r->cpu_mask, tmpmask); |
418 | if (cpumask_weight(tmpmask1)) | |
419 | cpumask_rdtgrp_clear(r, tmpmask1); | |
b09d981b | 420 | } |
a9fcf862 | 421 | update_closid_rmid(tmpmask, rdtgrp); |
b09d981b VS |
422 | } |
423 | ||
424 | /* Done pushing/pulling - update this group with new mask */ | |
425 | cpumask_copy(&rdtgrp->cpu_mask, newmask); | |
426 | ||
a9fcf862 VS |
427 | /* |
428 | * Clear child mon group masks since there is a new parent mask | |
429 | * now and update the rmid for the cpus the child lost. | |
430 | */ | |
431 | head = &rdtgrp->mon.crdtgrp_list; | |
432 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) { | |
433 | cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask); | |
434 | update_closid_rmid(tmpmask, rdtgrp); | |
435 | cpumask_clear(&crgrp->cpu_mask); | |
436 | } | |
437 | ||
b09d981b VS |
438 | return 0; |
439 | } | |
440 | ||
12e0110c TL |
441 | static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of, |
442 | char *buf, size_t nbytes, loff_t off) | |
443 | { | |
a9fcf862 | 444 | cpumask_var_t tmpmask, newmask, tmpmask1; |
b09d981b | 445 | struct rdtgroup *rdtgrp; |
f4107702 | 446 | int ret; |
12e0110c TL |
447 | |
448 | if (!buf) | |
449 | return -EINVAL; | |
450 | ||
451 | if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
452 | return -ENOMEM; | |
453 | if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) { | |
454 | free_cpumask_var(tmpmask); | |
455 | return -ENOMEM; | |
456 | } | |
a9fcf862 VS |
457 | if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) { |
458 | free_cpumask_var(tmpmask); | |
459 | free_cpumask_var(newmask); | |
460 | return -ENOMEM; | |
461 | } | |
a2584e1d | 462 | |
12e0110c TL |
463 | rdtgrp = rdtgroup_kn_lock_live(of->kn); |
464 | if (!rdtgrp) { | |
465 | ret = -ENOENT; | |
466 | goto unlock; | |
467 | } | |
468 | ||
c966dac8 RC |
469 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || |
470 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { | |
471 | ret = -EINVAL; | |
723f1a0d | 472 | rdt_last_cmd_puts("Pseudo-locking in progress\n"); |
c966dac8 RC |
473 | goto unlock; |
474 | } | |
475 | ||
4ffa3c97 JO |
476 | if (is_cpu_list(of)) |
477 | ret = cpulist_parse(buf, newmask); | |
478 | else | |
479 | ret = cpumask_parse(buf, newmask); | |
480 | ||
94457b36 | 481 | if (ret) { |
723f1a0d | 482 | rdt_last_cmd_puts("Bad CPU list/mask\n"); |
12e0110c | 483 | goto unlock; |
94457b36 | 484 | } |
12e0110c | 485 | |
12e0110c TL |
486 | /* check that user didn't specify any offline cpus */ |
487 | cpumask_andnot(tmpmask, newmask, cpu_online_mask); | |
488 | if (cpumask_weight(tmpmask)) { | |
489 | ret = -EINVAL; | |
723f1a0d | 490 | rdt_last_cmd_puts("Can only assign online CPUs\n"); |
a2584e1d | 491 | goto unlock; |
12e0110c TL |
492 | } |
493 | ||
b09d981b | 494 | if (rdtgrp->type == RDTCTRL_GROUP) |
a9fcf862 VS |
495 | ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1); |
496 | else if (rdtgrp->type == RDTMON_GROUP) | |
497 | ret = cpus_mon_write(rdtgrp, newmask, tmpmask); | |
b09d981b VS |
498 | else |
499 | ret = -EINVAL; | |
12e0110c | 500 | |
12e0110c TL |
501 | unlock: |
502 | rdtgroup_kn_unlock(of->kn); | |
503 | free_cpumask_var(tmpmask); | |
504 | free_cpumask_var(newmask); | |
a9fcf862 | 505 | free_cpumask_var(tmpmask1); |
12e0110c TL |
506 | |
507 | return ret ?: nbytes; | |
508 | } | |
509 | ||
e02737d5 FY |
510 | struct task_move_callback { |
511 | struct callback_head work; | |
512 | struct rdtgroup *rdtgrp; | |
513 | }; | |
514 | ||
515 | static void move_myself(struct callback_head *head) | |
516 | { | |
517 | struct task_move_callback *callback; | |
518 | struct rdtgroup *rdtgrp; | |
519 | ||
520 | callback = container_of(head, struct task_move_callback, work); | |
521 | rdtgrp = callback->rdtgrp; | |
522 | ||
523 | /* | |
524 | * If resource group was deleted before this task work callback | |
525 | * was invoked, then assign the task to root group and free the | |
526 | * resource group. | |
527 | */ | |
528 | if (atomic_dec_and_test(&rdtgrp->waitcount) && | |
529 | (rdtgrp->flags & RDT_DELETED)) { | |
530 | current->closid = 0; | |
d6aaba61 | 531 | current->rmid = 0; |
e02737d5 FY |
532 | kfree(rdtgrp); |
533 | } | |
534 | ||
dc433797 XS |
535 | if (unlikely(current->flags & PF_EXITING)) |
536 | goto out; | |
537 | ||
74fcdae1 | 538 | preempt_disable(); |
4f341a5e | 539 | /* update PQR_ASSOC MSR to make resource group go into effect */ |
352940ec | 540 | resctrl_sched_in(); |
74fcdae1 | 541 | preempt_enable(); |
4f341a5e | 542 | |
dc433797 | 543 | out: |
e02737d5 FY |
544 | kfree(callback); |
545 | } | |
546 | ||
547 | static int __rdtgroup_move_task(struct task_struct *tsk, | |
548 | struct rdtgroup *rdtgrp) | |
549 | { | |
550 | struct task_move_callback *callback; | |
551 | int ret; | |
552 | ||
553 | callback = kzalloc(sizeof(*callback), GFP_KERNEL); | |
554 | if (!callback) | |
555 | return -ENOMEM; | |
556 | callback->work.func = move_myself; | |
557 | callback->rdtgrp = rdtgrp; | |
558 | ||
559 | /* | |
560 | * Take a refcount, so rdtgrp cannot be freed before the | |
561 | * callback has been invoked. | |
562 | */ | |
563 | atomic_inc(&rdtgrp->waitcount); | |
564 | ret = task_work_add(tsk, &callback->work, true); | |
565 | if (ret) { | |
566 | /* | |
567 | * Task is exiting. Drop the refcount and free the callback. | |
568 | * No need to check the refcount as the group cannot be | |
569 | * deleted before the write function unlocks rdtgroup_mutex. | |
570 | */ | |
571 | atomic_dec(&rdtgrp->waitcount); | |
572 | kfree(callback); | |
723f1a0d | 573 | rdt_last_cmd_puts("Task exited\n"); |
e02737d5 | 574 | } else { |
d6aaba61 VS |
575 | /* |
576 | * For ctrl_mon groups move both closid and rmid. | |
577 | * For monitor groups, can move the tasks only from | |
578 | * their parent CTRL group. | |
579 | */ | |
580 | if (rdtgrp->type == RDTCTRL_GROUP) { | |
581 | tsk->closid = rdtgrp->closid; | |
582 | tsk->rmid = rdtgrp->mon.rmid; | |
583 | } else if (rdtgrp->type == RDTMON_GROUP) { | |
29e74f35 | 584 | if (rdtgrp->mon.parent->closid == tsk->closid) { |
d6aaba61 | 585 | tsk->rmid = rdtgrp->mon.rmid; |
29e74f35 TL |
586 | } else { |
587 | rdt_last_cmd_puts("Can't move task to different control group\n"); | |
d6aaba61 | 588 | ret = -EINVAL; |
29e74f35 | 589 | } |
d6aaba61 | 590 | } |
e02737d5 FY |
591 | } |
592 | return ret; | |
593 | } | |
594 | ||
f7a6e3f6 RC |
595 | /** |
596 | * rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group | |
597 | * @r: Resource group | |
598 | * | |
599 | * Return: 1 if tasks have been assigned to @r, 0 otherwise | |
600 | */ | |
601 | int rdtgroup_tasks_assigned(struct rdtgroup *r) | |
602 | { | |
603 | struct task_struct *p, *t; | |
604 | int ret = 0; | |
605 | ||
606 | lockdep_assert_held(&rdtgroup_mutex); | |
607 | ||
608 | rcu_read_lock(); | |
609 | for_each_process_thread(p, t) { | |
610 | if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) || | |
611 | (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) { | |
612 | ret = 1; | |
613 | break; | |
614 | } | |
615 | } | |
616 | rcu_read_unlock(); | |
617 | ||
618 | return ret; | |
619 | } | |
620 | ||
e02737d5 FY |
621 | static int rdtgroup_task_write_permission(struct task_struct *task, |
622 | struct kernfs_open_file *of) | |
623 | { | |
624 | const struct cred *tcred = get_task_cred(task); | |
625 | const struct cred *cred = current_cred(); | |
626 | int ret = 0; | |
627 | ||
628 | /* | |
629 | * Even if we're attaching all tasks in the thread group, we only | |
630 | * need to check permissions on one of them. | |
631 | */ | |
632 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
633 | !uid_eq(cred->euid, tcred->uid) && | |
29e74f35 TL |
634 | !uid_eq(cred->euid, tcred->suid)) { |
635 | rdt_last_cmd_printf("No permission to move task %d\n", task->pid); | |
e02737d5 | 636 | ret = -EPERM; |
29e74f35 | 637 | } |
e02737d5 FY |
638 | |
639 | put_cred(tcred); | |
640 | return ret; | |
641 | } | |
642 | ||
643 | static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp, | |
644 | struct kernfs_open_file *of) | |
645 | { | |
646 | struct task_struct *tsk; | |
647 | int ret; | |
648 | ||
649 | rcu_read_lock(); | |
650 | if (pid) { | |
651 | tsk = find_task_by_vpid(pid); | |
652 | if (!tsk) { | |
653 | rcu_read_unlock(); | |
29e74f35 | 654 | rdt_last_cmd_printf("No task %d\n", pid); |
e02737d5 FY |
655 | return -ESRCH; |
656 | } | |
657 | } else { | |
658 | tsk = current; | |
659 | } | |
660 | ||
661 | get_task_struct(tsk); | |
662 | rcu_read_unlock(); | |
663 | ||
664 | ret = rdtgroup_task_write_permission(tsk, of); | |
665 | if (!ret) | |
666 | ret = __rdtgroup_move_task(tsk, rdtgrp); | |
667 | ||
668 | put_task_struct(tsk); | |
669 | return ret; | |
670 | } | |
671 | ||
672 | static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of, | |
673 | char *buf, size_t nbytes, loff_t off) | |
674 | { | |
675 | struct rdtgroup *rdtgrp; | |
676 | int ret = 0; | |
677 | pid_t pid; | |
678 | ||
679 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) | |
680 | return -EINVAL; | |
681 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
c966dac8 RC |
682 | if (!rdtgrp) { |
683 | rdtgroup_kn_unlock(of->kn); | |
684 | return -ENOENT; | |
685 | } | |
29e74f35 | 686 | rdt_last_cmd_clear(); |
e02737d5 | 687 | |
c966dac8 RC |
688 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || |
689 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { | |
690 | ret = -EINVAL; | |
723f1a0d | 691 | rdt_last_cmd_puts("Pseudo-locking in progress\n"); |
c966dac8 RC |
692 | goto unlock; |
693 | } | |
694 | ||
695 | ret = rdtgroup_move_task(pid, rdtgrp, of); | |
e02737d5 | 696 | |
c966dac8 | 697 | unlock: |
e02737d5 FY |
698 | rdtgroup_kn_unlock(of->kn); |
699 | ||
700 | return ret ?: nbytes; | |
701 | } | |
702 | ||
703 | static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s) | |
704 | { | |
705 | struct task_struct *p, *t; | |
706 | ||
707 | rcu_read_lock(); | |
708 | for_each_process_thread(p, t) { | |
d6aaba61 VS |
709 | if ((r->type == RDTCTRL_GROUP && t->closid == r->closid) || |
710 | (r->type == RDTMON_GROUP && t->rmid == r->mon.rmid)) | |
e02737d5 FY |
711 | seq_printf(s, "%d\n", t->pid); |
712 | } | |
713 | rcu_read_unlock(); | |
714 | } | |
715 | ||
716 | static int rdtgroup_tasks_show(struct kernfs_open_file *of, | |
717 | struct seq_file *s, void *v) | |
718 | { | |
719 | struct rdtgroup *rdtgrp; | |
720 | int ret = 0; | |
721 | ||
722 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
723 | if (rdtgrp) | |
724 | show_rdt_tasks(rdtgrp, s); | |
725 | else | |
726 | ret = -ENOENT; | |
727 | rdtgroup_kn_unlock(of->kn); | |
728 | ||
729 | return ret; | |
730 | } | |
731 | ||
e79f15a4 CY |
732 | #ifdef CONFIG_PROC_CPU_RESCTRL |
733 | ||
734 | /* | |
735 | * A task can only be part of one resctrl control group and of one monitor | |
736 | * group which is associated to that control group. | |
737 | * | |
738 | * 1) res: | |
739 | * mon: | |
740 | * | |
741 | * resctrl is not available. | |
742 | * | |
743 | * 2) res:/ | |
744 | * mon: | |
745 | * | |
746 | * Task is part of the root resctrl control group, and it is not associated | |
747 | * to any monitor group. | |
748 | * | |
749 | * 3) res:/ | |
750 | * mon:mon0 | |
751 | * | |
752 | * Task is part of the root resctrl control group and monitor group mon0. | |
753 | * | |
754 | * 4) res:group0 | |
755 | * mon: | |
756 | * | |
757 | * Task is part of resctrl control group group0, and it is not associated | |
758 | * to any monitor group. | |
759 | * | |
760 | * 5) res:group0 | |
761 | * mon:mon1 | |
762 | * | |
763 | * Task is part of resctrl control group group0 and monitor group mon1. | |
764 | */ | |
765 | int proc_resctrl_show(struct seq_file *s, struct pid_namespace *ns, | |
766 | struct pid *pid, struct task_struct *tsk) | |
767 | { | |
768 | struct rdtgroup *rdtg; | |
769 | int ret = 0; | |
770 | ||
771 | mutex_lock(&rdtgroup_mutex); | |
772 | ||
773 | /* Return empty if resctrl has not been mounted. */ | |
774 | if (!static_branch_unlikely(&rdt_enable_key)) { | |
775 | seq_puts(s, "res:\nmon:\n"); | |
776 | goto unlock; | |
777 | } | |
778 | ||
779 | list_for_each_entry(rdtg, &rdt_all_groups, rdtgroup_list) { | |
780 | struct rdtgroup *crg; | |
781 | ||
782 | /* | |
783 | * Task information is only relevant for shareable | |
784 | * and exclusive groups. | |
785 | */ | |
786 | if (rdtg->mode != RDT_MODE_SHAREABLE && | |
787 | rdtg->mode != RDT_MODE_EXCLUSIVE) | |
788 | continue; | |
789 | ||
790 | if (rdtg->closid != tsk->closid) | |
791 | continue; | |
792 | ||
793 | seq_printf(s, "res:%s%s\n", (rdtg == &rdtgroup_default) ? "/" : "", | |
794 | rdtg->kn->name); | |
795 | seq_puts(s, "mon:"); | |
796 | list_for_each_entry(crg, &rdtg->mon.crdtgrp_list, | |
797 | mon.crdtgrp_list) { | |
798 | if (tsk->rmid != crg->mon.rmid) | |
799 | continue; | |
800 | seq_printf(s, "%s", crg->kn->name); | |
801 | break; | |
802 | } | |
803 | seq_putc(s, '\n'); | |
804 | goto unlock; | |
805 | } | |
806 | /* | |
807 | * The above search should succeed. Otherwise return | |
808 | * with an error. | |
809 | */ | |
810 | ret = -ENOENT; | |
811 | unlock: | |
812 | mutex_unlock(&rdtgroup_mutex); | |
813 | ||
814 | return ret; | |
815 | } | |
816 | #endif | |
817 | ||
9b3a7fd0 TL |
818 | static int rdt_last_cmd_status_show(struct kernfs_open_file *of, |
819 | struct seq_file *seq, void *v) | |
820 | { | |
821 | int len; | |
822 | ||
823 | mutex_lock(&rdtgroup_mutex); | |
824 | len = seq_buf_used(&last_cmd_status); | |
825 | if (len) | |
826 | seq_printf(seq, "%.*s", len, last_cmd_status_buf); | |
827 | else | |
828 | seq_puts(seq, "ok\n"); | |
829 | mutex_unlock(&rdtgroup_mutex); | |
830 | return 0; | |
831 | } | |
832 | ||
4e978d06 FY |
833 | static int rdt_num_closids_show(struct kernfs_open_file *of, |
834 | struct seq_file *seq, void *v) | |
835 | { | |
836 | struct rdt_resource *r = of->kn->parent->priv; | |
837 | ||
838 | seq_printf(seq, "%d\n", r->num_closid); | |
4e978d06 FY |
839 | return 0; |
840 | } | |
841 | ||
2545e9f5 | 842 | static int rdt_default_ctrl_show(struct kernfs_open_file *of, |
4e978d06 FY |
843 | struct seq_file *seq, void *v) |
844 | { | |
845 | struct rdt_resource *r = of->kn->parent->priv; | |
846 | ||
2545e9f5 | 847 | seq_printf(seq, "%x\n", r->default_ctrl); |
4e978d06 FY |
848 | return 0; |
849 | } | |
850 | ||
53a114a6 SL |
851 | static int rdt_min_cbm_bits_show(struct kernfs_open_file *of, |
852 | struct seq_file *seq, void *v) | |
853 | { | |
854 | struct rdt_resource *r = of->kn->parent->priv; | |
855 | ||
d3e11b4d | 856 | seq_printf(seq, "%u\n", r->cache.min_cbm_bits); |
db69ef65 VS |
857 | return 0; |
858 | } | |
859 | ||
0dd2d749 FY |
860 | static int rdt_shareable_bits_show(struct kernfs_open_file *of, |
861 | struct seq_file *seq, void *v) | |
862 | { | |
863 | struct rdt_resource *r = of->kn->parent->priv; | |
864 | ||
865 | seq_printf(seq, "%x\n", r->cache.shareable_bits); | |
866 | return 0; | |
867 | } | |
868 | ||
e6519011 RC |
869 | /** |
870 | * rdt_bit_usage_show - Display current usage of resources | |
871 | * | |
872 | * A domain is a shared resource that can now be allocated differently. Here | |
873 | * we display the current regions of the domain as an annotated bitmask. | |
874 | * For each domain of this resource its allocation bitmask | |
875 | * is annotated as below to indicate the current usage of the corresponding bit: | |
876 | * 0 - currently unused | |
877 | * X - currently available for sharing and used by software and hardware | |
878 | * H - currently used by hardware only but available for software use | |
879 | * S - currently used and shareable by software only | |
880 | * E - currently used exclusively by one resource group | |
f4e80d67 | 881 | * P - currently pseudo-locked by one resource group |
e6519011 RC |
882 | */ |
883 | static int rdt_bit_usage_show(struct kernfs_open_file *of, | |
884 | struct seq_file *seq, void *v) | |
885 | { | |
886 | struct rdt_resource *r = of->kn->parent->priv; | |
32f010de RC |
887 | /* |
888 | * Use unsigned long even though only 32 bits are used to ensure | |
889 | * test_bit() is used safely. | |
890 | */ | |
891 | unsigned long sw_shareable = 0, hw_shareable = 0; | |
892 | unsigned long exclusive = 0, pseudo_locked = 0; | |
e6519011 | 893 | struct rdt_domain *dom; |
f4e80d67 | 894 | int i, hwb, swb, excl, psl; |
e6519011 RC |
895 | enum rdtgrp_mode mode; |
896 | bool sep = false; | |
897 | u32 *ctrl; | |
898 | ||
899 | mutex_lock(&rdtgroup_mutex); | |
900 | hw_shareable = r->cache.shareable_bits; | |
901 | list_for_each_entry(dom, &r->domains, list) { | |
902 | if (sep) | |
903 | seq_putc(seq, ';'); | |
904 | ctrl = dom->ctrl_val; | |
905 | sw_shareable = 0; | |
906 | exclusive = 0; | |
907 | seq_printf(seq, "%d=", dom->id); | |
47d53b18 | 908 | for (i = 0; i < closids_supported(); i++, ctrl++) { |
e6519011 RC |
909 | if (!closid_allocated(i)) |
910 | continue; | |
911 | mode = rdtgroup_mode_by_closid(i); | |
912 | switch (mode) { | |
913 | case RDT_MODE_SHAREABLE: | |
914 | sw_shareable |= *ctrl; | |
915 | break; | |
916 | case RDT_MODE_EXCLUSIVE: | |
917 | exclusive |= *ctrl; | |
918 | break; | |
f4e80d67 | 919 | case RDT_MODE_PSEUDO_LOCKSETUP: |
bb9fec69 | 920 | /* |
f4e80d67 RC |
921 | * RDT_MODE_PSEUDO_LOCKSETUP is possible |
922 | * here but not included since the CBM | |
923 | * associated with this CLOSID in this mode | |
924 | * is not initialized and no task or cpu can be | |
925 | * assigned this CLOSID. | |
bb9fec69 | 926 | */ |
f4e80d67 | 927 | break; |
bb9fec69 | 928 | case RDT_MODE_PSEUDO_LOCKED: |
e6519011 RC |
929 | case RDT_NUM_MODES: |
930 | WARN(1, | |
931 | "invalid mode for closid %d\n", i); | |
932 | break; | |
933 | } | |
934 | } | |
935 | for (i = r->cache.cbm_len - 1; i >= 0; i--) { | |
f4e80d67 | 936 | pseudo_locked = dom->plr ? dom->plr->cbm : 0; |
32f010de RC |
937 | hwb = test_bit(i, &hw_shareable); |
938 | swb = test_bit(i, &sw_shareable); | |
939 | excl = test_bit(i, &exclusive); | |
940 | psl = test_bit(i, &pseudo_locked); | |
e6519011 RC |
941 | if (hwb && swb) |
942 | seq_putc(seq, 'X'); | |
943 | else if (hwb && !swb) | |
944 | seq_putc(seq, 'H'); | |
945 | else if (!hwb && swb) | |
946 | seq_putc(seq, 'S'); | |
947 | else if (excl) | |
948 | seq_putc(seq, 'E'); | |
f4e80d67 RC |
949 | else if (psl) |
950 | seq_putc(seq, 'P'); | |
e6519011 RC |
951 | else /* Unused bits remain */ |
952 | seq_putc(seq, '0'); | |
953 | } | |
954 | sep = true; | |
955 | } | |
956 | seq_putc(seq, '\n'); | |
957 | mutex_unlock(&rdtgroup_mutex); | |
958 | return 0; | |
959 | } | |
960 | ||
db69ef65 VS |
961 | static int rdt_min_bw_show(struct kernfs_open_file *of, |
962 | struct seq_file *seq, void *v) | |
963 | { | |
964 | struct rdt_resource *r = of->kn->parent->priv; | |
53a114a6 | 965 | |
db69ef65 VS |
966 | seq_printf(seq, "%u\n", r->membw.min_bw); |
967 | return 0; | |
968 | } | |
969 | ||
d4ab3320 VS |
970 | static int rdt_num_rmids_show(struct kernfs_open_file *of, |
971 | struct seq_file *seq, void *v) | |
972 | { | |
973 | struct rdt_resource *r = of->kn->parent->priv; | |
974 | ||
975 | seq_printf(seq, "%d\n", r->num_rmid); | |
976 | ||
977 | return 0; | |
978 | } | |
979 | ||
980 | static int rdt_mon_features_show(struct kernfs_open_file *of, | |
981 | struct seq_file *seq, void *v) | |
982 | { | |
983 | struct rdt_resource *r = of->kn->parent->priv; | |
984 | struct mon_evt *mevt; | |
985 | ||
986 | list_for_each_entry(mevt, &r->evt_list, list) | |
987 | seq_printf(seq, "%s\n", mevt->name); | |
988 | ||
989 | return 0; | |
990 | } | |
991 | ||
db69ef65 VS |
992 | static int rdt_bw_gran_show(struct kernfs_open_file *of, |
993 | struct seq_file *seq, void *v) | |
994 | { | |
995 | struct rdt_resource *r = of->kn->parent->priv; | |
996 | ||
997 | seq_printf(seq, "%u\n", r->membw.bw_gran); | |
998 | return 0; | |
999 | } | |
1000 | ||
1001 | static int rdt_delay_linear_show(struct kernfs_open_file *of, | |
1002 | struct seq_file *seq, void *v) | |
1003 | { | |
1004 | struct rdt_resource *r = of->kn->parent->priv; | |
1005 | ||
1006 | seq_printf(seq, "%u\n", r->membw.delay_linear); | |
53a114a6 SL |
1007 | return 0; |
1008 | } | |
1009 | ||
d4ab3320 VS |
1010 | static int max_threshold_occ_show(struct kernfs_open_file *of, |
1011 | struct seq_file *seq, void *v) | |
1012 | { | |
1013 | struct rdt_resource *r = of->kn->parent->priv; | |
1014 | ||
352940ec | 1015 | seq_printf(seq, "%u\n", resctrl_cqm_threshold * r->mon_scale); |
d4ab3320 VS |
1016 | |
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | static ssize_t max_threshold_occ_write(struct kernfs_open_file *of, | |
1021 | char *buf, size_t nbytes, loff_t off) | |
1022 | { | |
1023 | struct rdt_resource *r = of->kn->parent->priv; | |
1024 | unsigned int bytes; | |
1025 | int ret; | |
1026 | ||
1027 | ret = kstrtouint(buf, 0, &bytes); | |
1028 | if (ret) | |
1029 | return ret; | |
1030 | ||
1031 | if (bytes > (boot_cpu_data.x86_cache_size * 1024)) | |
1032 | return -EINVAL; | |
1033 | ||
352940ec | 1034 | resctrl_cqm_threshold = bytes / r->mon_scale; |
d4ab3320 | 1035 | |
5707b46a | 1036 | return nbytes; |
d4ab3320 VS |
1037 | } |
1038 | ||
d48d7a57 RC |
1039 | /* |
1040 | * rdtgroup_mode_show - Display mode of this resource group | |
1041 | */ | |
1042 | static int rdtgroup_mode_show(struct kernfs_open_file *of, | |
1043 | struct seq_file *s, void *v) | |
1044 | { | |
1045 | struct rdtgroup *rdtgrp; | |
1046 | ||
1047 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
1048 | if (!rdtgrp) { | |
1049 | rdtgroup_kn_unlock(of->kn); | |
1050 | return -ENOENT; | |
1051 | } | |
1052 | ||
1053 | seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode)); | |
1054 | ||
1055 | rdtgroup_kn_unlock(of->kn); | |
1056 | return 0; | |
1057 | } | |
1058 | ||
521348b0 RC |
1059 | /** |
1060 | * rdt_cdp_peer_get - Retrieve CDP peer if it exists | |
1061 | * @r: RDT resource to which RDT domain @d belongs | |
1062 | * @d: Cache instance for which a CDP peer is requested | |
1063 | * @r_cdp: RDT resource that shares hardware with @r (RDT resource peer) | |
1064 | * Used to return the result. | |
1065 | * @d_cdp: RDT domain that shares hardware with @d (RDT domain peer) | |
1066 | * Used to return the result. | |
1067 | * | |
1068 | * RDT resources are managed independently and by extension the RDT domains | |
1069 | * (RDT resource instances) are managed independently also. The Code and | |
1070 | * Data Prioritization (CDP) RDT resources, while managed independently, | |
1071 | * could refer to the same underlying hardware. For example, | |
1072 | * RDT_RESOURCE_L2CODE and RDT_RESOURCE_L2DATA both refer to the L2 cache. | |
1073 | * | |
1074 | * When provided with an RDT resource @r and an instance of that RDT | |
1075 | * resource @d rdt_cdp_peer_get() will return if there is a peer RDT | |
1076 | * resource and the exact instance that shares the same hardware. | |
1077 | * | |
1078 | * Return: 0 if a CDP peer was found, <0 on error or if no CDP peer exists. | |
1079 | * If a CDP peer was found, @r_cdp will point to the peer RDT resource | |
1080 | * and @d_cdp will point to the peer RDT domain. | |
1081 | */ | |
e5f3530c RC |
1082 | static int rdt_cdp_peer_get(struct rdt_resource *r, struct rdt_domain *d, |
1083 | struct rdt_resource **r_cdp, | |
1084 | struct rdt_domain **d_cdp) | |
521348b0 RC |
1085 | { |
1086 | struct rdt_resource *_r_cdp = NULL; | |
1087 | struct rdt_domain *_d_cdp = NULL; | |
1088 | int ret = 0; | |
1089 | ||
1090 | switch (r->rid) { | |
1091 | case RDT_RESOURCE_L3DATA: | |
1092 | _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3CODE]; | |
1093 | break; | |
1094 | case RDT_RESOURCE_L3CODE: | |
1095 | _r_cdp = &rdt_resources_all[RDT_RESOURCE_L3DATA]; | |
1096 | break; | |
1097 | case RDT_RESOURCE_L2DATA: | |
1098 | _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2CODE]; | |
1099 | break; | |
1100 | case RDT_RESOURCE_L2CODE: | |
1101 | _r_cdp = &rdt_resources_all[RDT_RESOURCE_L2DATA]; | |
1102 | break; | |
1103 | default: | |
1104 | ret = -ENOENT; | |
1105 | goto out; | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * When a new CPU comes online and CDP is enabled then the new | |
1110 | * RDT domains (if any) associated with both CDP RDT resources | |
1111 | * are added in the same CPU online routine while the | |
1112 | * rdtgroup_mutex is held. It should thus not happen for one | |
1113 | * RDT domain to exist and be associated with its RDT CDP | |
1114 | * resource but there is no RDT domain associated with the | |
1115 | * peer RDT CDP resource. Hence the WARN. | |
1116 | */ | |
1117 | _d_cdp = rdt_find_domain(_r_cdp, d->id, NULL); | |
52eb7433 | 1118 | if (WARN_ON(IS_ERR_OR_NULL(_d_cdp))) { |
521348b0 RC |
1119 | _r_cdp = NULL; |
1120 | ret = -EINVAL; | |
1121 | } | |
1122 | ||
1123 | out: | |
1124 | *r_cdp = _r_cdp; | |
1125 | *d_cdp = _d_cdp; | |
1126 | ||
1127 | return ret; | |
1128 | } | |
1129 | ||
49f7b4ef | 1130 | /** |
e5f3530c | 1131 | * __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other |
49f7b4ef RC |
1132 | * @r: Resource to which domain instance @d belongs. |
1133 | * @d: The domain instance for which @closid is being tested. | |
1134 | * @cbm: Capacity bitmask being tested. | |
1135 | * @closid: Intended closid for @cbm. | |
1136 | * @exclusive: Only check if overlaps with exclusive resource groups | |
1137 | * | |
1138 | * Checks if provided @cbm intended to be used for @closid on domain | |
1139 | * @d overlaps with any other closids or other hardware usage associated | |
1140 | * with this domain. If @exclusive is true then only overlaps with | |
1141 | * resource groups in exclusive mode will be considered. If @exclusive | |
1142 | * is false then overlaps with any resource group or hardware entities | |
1143 | * will be considered. | |
1144 | * | |
49e00eee RC |
1145 | * @cbm is unsigned long, even if only 32 bits are used, to make the |
1146 | * bitmap functions work correctly. | |
1147 | * | |
49f7b4ef RC |
1148 | * Return: false if CBM does not overlap, true if it does. |
1149 | */ | |
e5f3530c RC |
1150 | static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d, |
1151 | unsigned long cbm, int closid, bool exclusive) | |
49f7b4ef | 1152 | { |
49f7b4ef | 1153 | enum rdtgrp_mode mode; |
49e00eee | 1154 | unsigned long ctrl_b; |
49f7b4ef RC |
1155 | u32 *ctrl; |
1156 | int i; | |
1157 | ||
1158 | /* Check for any overlap with regions used by hardware directly */ | |
1159 | if (!exclusive) { | |
49e00eee RC |
1160 | ctrl_b = r->cache.shareable_bits; |
1161 | if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) | |
49f7b4ef RC |
1162 | return true; |
1163 | } | |
1164 | ||
1165 | /* Check for overlap with other resource groups */ | |
1166 | ctrl = d->ctrl_val; | |
f0df4e1a | 1167 | for (i = 0; i < closids_supported(); i++, ctrl++) { |
49e00eee | 1168 | ctrl_b = *ctrl; |
dfe9674b RC |
1169 | mode = rdtgroup_mode_by_closid(i); |
1170 | if (closid_allocated(i) && i != closid && | |
1171 | mode != RDT_MODE_PSEUDO_LOCKSETUP) { | |
49e00eee | 1172 | if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) { |
49f7b4ef RC |
1173 | if (exclusive) { |
1174 | if (mode == RDT_MODE_EXCLUSIVE) | |
1175 | return true; | |
1176 | continue; | |
1177 | } | |
1178 | return true; | |
1179 | } | |
1180 | } | |
1181 | } | |
1182 | ||
1183 | return false; | |
1184 | } | |
1185 | ||
e5f3530c RC |
1186 | /** |
1187 | * rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware | |
1188 | * @r: Resource to which domain instance @d belongs. | |
1189 | * @d: The domain instance for which @closid is being tested. | |
1190 | * @cbm: Capacity bitmask being tested. | |
1191 | * @closid: Intended closid for @cbm. | |
1192 | * @exclusive: Only check if overlaps with exclusive resource groups | |
1193 | * | |
1194 | * Resources that can be allocated using a CBM can use the CBM to control | |
1195 | * the overlap of these allocations. rdtgroup_cmb_overlaps() is the test | |
1196 | * for overlap. Overlap test is not limited to the specific resource for | |
1197 | * which the CBM is intended though - when dealing with CDP resources that | |
1198 | * share the underlying hardware the overlap check should be performed on | |
1199 | * the CDP resource sharing the hardware also. | |
1200 | * | |
1201 | * Refer to description of __rdtgroup_cbm_overlaps() for the details of the | |
1202 | * overlap test. | |
1203 | * | |
1204 | * Return: true if CBM overlap detected, false if there is no overlap | |
1205 | */ | |
1206 | bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d, | |
1207 | unsigned long cbm, int closid, bool exclusive) | |
1208 | { | |
1209 | struct rdt_resource *r_cdp; | |
1210 | struct rdt_domain *d_cdp; | |
1211 | ||
1212 | if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, exclusive)) | |
1213 | return true; | |
1214 | ||
1215 | if (rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp) < 0) | |
1216 | return false; | |
1217 | ||
1218 | return __rdtgroup_cbm_overlaps(r_cdp, d_cdp, cbm, closid, exclusive); | |
1219 | } | |
1220 | ||
49f7b4ef RC |
1221 | /** |
1222 | * rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive | |
1223 | * | |
1224 | * An exclusive resource group implies that there should be no sharing of | |
1225 | * its allocated resources. At the time this group is considered to be | |
1226 | * exclusive this test can determine if its current schemata supports this | |
1227 | * setting by testing for overlap with all other resource groups. | |
1228 | * | |
1229 | * Return: true if resource group can be exclusive, false if there is overlap | |
1230 | * with allocations of other resource groups and thus this resource group | |
1231 | * cannot be exclusive. | |
1232 | */ | |
1233 | static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp) | |
1234 | { | |
1235 | int closid = rdtgrp->closid; | |
1236 | struct rdt_resource *r; | |
939b90b2 | 1237 | bool has_cache = false; |
49f7b4ef RC |
1238 | struct rdt_domain *d; |
1239 | ||
1240 | for_each_alloc_enabled_rdt_resource(r) { | |
939b90b2 RC |
1241 | if (r->rid == RDT_RESOURCE_MBA) |
1242 | continue; | |
1243 | has_cache = true; | |
49f7b4ef RC |
1244 | list_for_each_entry(d, &r->domains, list) { |
1245 | if (rdtgroup_cbm_overlaps(r, d, d->ctrl_val[closid], | |
939b90b2 | 1246 | rdtgrp->closid, false)) { |
723f1a0d | 1247 | rdt_last_cmd_puts("Schemata overlaps\n"); |
49f7b4ef | 1248 | return false; |
939b90b2 | 1249 | } |
49f7b4ef RC |
1250 | } |
1251 | } | |
1252 | ||
939b90b2 | 1253 | if (!has_cache) { |
723f1a0d | 1254 | rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n"); |
939b90b2 RC |
1255 | return false; |
1256 | } | |
1257 | ||
49f7b4ef RC |
1258 | return true; |
1259 | } | |
1260 | ||
1261 | /** | |
1262 | * rdtgroup_mode_write - Modify the resource group's mode | |
1263 | * | |
1264 | */ | |
d48d7a57 RC |
1265 | static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of, |
1266 | char *buf, size_t nbytes, loff_t off) | |
1267 | { | |
1268 | struct rdtgroup *rdtgrp; | |
1269 | enum rdtgrp_mode mode; | |
1270 | int ret = 0; | |
1271 | ||
1272 | /* Valid input requires a trailing newline */ | |
1273 | if (nbytes == 0 || buf[nbytes - 1] != '\n') | |
1274 | return -EINVAL; | |
1275 | buf[nbytes - 1] = '\0'; | |
1276 | ||
1277 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
1278 | if (!rdtgrp) { | |
1279 | rdtgroup_kn_unlock(of->kn); | |
1280 | return -ENOENT; | |
1281 | } | |
1282 | ||
1283 | rdt_last_cmd_clear(); | |
1284 | ||
1285 | mode = rdtgrp->mode; | |
1286 | ||
49f7b4ef | 1287 | if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) || |
dfe9674b RC |
1288 | (!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) || |
1289 | (!strcmp(buf, "pseudo-locksetup") && | |
1290 | mode == RDT_MODE_PSEUDO_LOCKSETUP) || | |
1291 | (!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED)) | |
d48d7a57 RC |
1292 | goto out; |
1293 | ||
dfe9674b | 1294 | if (mode == RDT_MODE_PSEUDO_LOCKED) { |
45682489 | 1295 | rdt_last_cmd_puts("Cannot change pseudo-locked group\n"); |
dfe9674b RC |
1296 | ret = -EINVAL; |
1297 | goto out; | |
1298 | } | |
1299 | ||
d48d7a57 | 1300 | if (!strcmp(buf, "shareable")) { |
dfe9674b RC |
1301 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
1302 | ret = rdtgroup_locksetup_exit(rdtgrp); | |
1303 | if (ret) | |
1304 | goto out; | |
1305 | } | |
d48d7a57 | 1306 | rdtgrp->mode = RDT_MODE_SHAREABLE; |
49f7b4ef RC |
1307 | } else if (!strcmp(buf, "exclusive")) { |
1308 | if (!rdtgroup_mode_test_exclusive(rdtgrp)) { | |
49f7b4ef RC |
1309 | ret = -EINVAL; |
1310 | goto out; | |
1311 | } | |
dfe9674b RC |
1312 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
1313 | ret = rdtgroup_locksetup_exit(rdtgrp); | |
1314 | if (ret) | |
1315 | goto out; | |
1316 | } | |
49f7b4ef | 1317 | rdtgrp->mode = RDT_MODE_EXCLUSIVE; |
dfe9674b RC |
1318 | } else if (!strcmp(buf, "pseudo-locksetup")) { |
1319 | ret = rdtgroup_locksetup_enter(rdtgrp); | |
1320 | if (ret) | |
1321 | goto out; | |
1322 | rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP; | |
d48d7a57 | 1323 | } else { |
45682489 | 1324 | rdt_last_cmd_puts("Unknown or unsupported mode\n"); |
d48d7a57 RC |
1325 | ret = -EINVAL; |
1326 | } | |
1327 | ||
1328 | out: | |
1329 | rdtgroup_kn_unlock(of->kn); | |
1330 | return ret ?: nbytes; | |
1331 | } | |
1332 | ||
d9b48c86 RC |
1333 | /** |
1334 | * rdtgroup_cbm_to_size - Translate CBM to size in bytes | |
1335 | * @r: RDT resource to which @d belongs. | |
1336 | * @d: RDT domain instance. | |
1337 | * @cbm: bitmask for which the size should be computed. | |
1338 | * | |
1339 | * The bitmask provided associated with the RDT domain instance @d will be | |
1340 | * translated into how many bytes it represents. The size in bytes is | |
1341 | * computed by first dividing the total cache size by the CBM length to | |
1342 | * determine how many bytes each bit in the bitmask represents. The result | |
1343 | * is multiplied with the number of bits set in the bitmask. | |
49e00eee RC |
1344 | * |
1345 | * @cbm is unsigned long, even if only 32 bits are used to make the | |
1346 | * bitmap functions work correctly. | |
d9b48c86 RC |
1347 | */ |
1348 | unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, | |
49e00eee | 1349 | struct rdt_domain *d, unsigned long cbm) |
d9b48c86 RC |
1350 | { |
1351 | struct cpu_cacheinfo *ci; | |
1352 | unsigned int size = 0; | |
1353 | int num_b, i; | |
1354 | ||
49e00eee | 1355 | num_b = bitmap_weight(&cbm, r->cache.cbm_len); |
d9b48c86 RC |
1356 | ci = get_cpu_cacheinfo(cpumask_any(&d->cpu_mask)); |
1357 | for (i = 0; i < ci->num_leaves; i++) { | |
1358 | if (ci->info_list[i].level == r->cache_level) { | |
1359 | size = ci->info_list[i].size / r->cache.cbm_len * num_b; | |
1360 | break; | |
1361 | } | |
1362 | } | |
1363 | ||
1364 | return size; | |
1365 | } | |
1366 | ||
1367 | /** | |
1368 | * rdtgroup_size_show - Display size in bytes of allocated regions | |
1369 | * | |
1370 | * The "size" file mirrors the layout of the "schemata" file, printing the | |
1371 | * size in bytes of each region instead of the capacity bitmask. | |
1372 | * | |
1373 | */ | |
1374 | static int rdtgroup_size_show(struct kernfs_open_file *of, | |
1375 | struct seq_file *s, void *v) | |
1376 | { | |
1377 | struct rdtgroup *rdtgrp; | |
1378 | struct rdt_resource *r; | |
1379 | struct rdt_domain *d; | |
1380 | unsigned int size; | |
b61b8bba | 1381 | int ret = 0; |
f968dc11 RC |
1382 | bool sep; |
1383 | u32 ctrl; | |
d9b48c86 RC |
1384 | |
1385 | rdtgrp = rdtgroup_kn_lock_live(of->kn); | |
1386 | if (!rdtgrp) { | |
1387 | rdtgroup_kn_unlock(of->kn); | |
1388 | return -ENOENT; | |
1389 | } | |
1390 | ||
f4e80d67 | 1391 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { |
b61b8bba JJ |
1392 | if (!rdtgrp->plr->d) { |
1393 | rdt_last_cmd_clear(); | |
1394 | rdt_last_cmd_puts("Cache domain offline\n"); | |
1395 | ret = -ENODEV; | |
1396 | } else { | |
1397 | seq_printf(s, "%*s:", max_name_width, | |
1398 | rdtgrp->plr->r->name); | |
1399 | size = rdtgroup_cbm_to_size(rdtgrp->plr->r, | |
1400 | rdtgrp->plr->d, | |
1401 | rdtgrp->plr->cbm); | |
1402 | seq_printf(s, "%d=%u\n", rdtgrp->plr->d->id, size); | |
1403 | } | |
f4e80d67 RC |
1404 | goto out; |
1405 | } | |
1406 | ||
d9b48c86 | 1407 | for_each_alloc_enabled_rdt_resource(r) { |
f968dc11 | 1408 | sep = false; |
d9b48c86 RC |
1409 | seq_printf(s, "%*s:", max_name_width, r->name); |
1410 | list_for_each_entry(d, &r->domains, list) { | |
1411 | if (sep) | |
1412 | seq_putc(s, ';'); | |
dfe9674b RC |
1413 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
1414 | size = 0; | |
1415 | } else { | |
f968dc11 RC |
1416 | ctrl = (!is_mba_sc(r) ? |
1417 | d->ctrl_val[rdtgrp->closid] : | |
1418 | d->mbps_val[rdtgrp->closid]); | |
1419 | if (r->rid == RDT_RESOURCE_MBA) | |
1420 | size = ctrl; | |
1421 | else | |
1422 | size = rdtgroup_cbm_to_size(r, d, ctrl); | |
dfe9674b | 1423 | } |
d9b48c86 RC |
1424 | seq_printf(s, "%d=%u", d->id, size); |
1425 | sep = true; | |
1426 | } | |
1427 | seq_putc(s, '\n'); | |
1428 | } | |
1429 | ||
f4e80d67 | 1430 | out: |
d9b48c86 RC |
1431 | rdtgroup_kn_unlock(of->kn); |
1432 | ||
b61b8bba | 1433 | return ret; |
d9b48c86 RC |
1434 | } |
1435 | ||
4e978d06 | 1436 | /* rdtgroup information files for one cache resource. */ |
5dc1d5c6 | 1437 | static struct rftype res_common_files[] = { |
9b3a7fd0 TL |
1438 | { |
1439 | .name = "last_cmd_status", | |
1440 | .mode = 0444, | |
1441 | .kf_ops = &rdtgroup_kf_single_ops, | |
1442 | .seq_show = rdt_last_cmd_status_show, | |
1443 | .fflags = RF_TOP_INFO, | |
1444 | }, | |
4e978d06 FY |
1445 | { |
1446 | .name = "num_closids", | |
1447 | .mode = 0444, | |
1448 | .kf_ops = &rdtgroup_kf_single_ops, | |
1449 | .seq_show = rdt_num_closids_show, | |
5dc1d5c6 | 1450 | .fflags = RF_CTRL_INFO, |
4e978d06 | 1451 | }, |
d4ab3320 VS |
1452 | { |
1453 | .name = "mon_features", | |
1454 | .mode = 0444, | |
1455 | .kf_ops = &rdtgroup_kf_single_ops, | |
1456 | .seq_show = rdt_mon_features_show, | |
1457 | .fflags = RF_MON_INFO, | |
1458 | }, | |
1459 | { | |
1460 | .name = "num_rmids", | |
1461 | .mode = 0444, | |
1462 | .kf_ops = &rdtgroup_kf_single_ops, | |
1463 | .seq_show = rdt_num_rmids_show, | |
1464 | .fflags = RF_MON_INFO, | |
1465 | }, | |
4e978d06 FY |
1466 | { |
1467 | .name = "cbm_mask", | |
1468 | .mode = 0444, | |
1469 | .kf_ops = &rdtgroup_kf_single_ops, | |
2545e9f5 | 1470 | .seq_show = rdt_default_ctrl_show, |
5dc1d5c6 | 1471 | .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, |
4e978d06 | 1472 | }, |
53a114a6 SL |
1473 | { |
1474 | .name = "min_cbm_bits", | |
1475 | .mode = 0444, | |
1476 | .kf_ops = &rdtgroup_kf_single_ops, | |
1477 | .seq_show = rdt_min_cbm_bits_show, | |
5dc1d5c6 | 1478 | .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, |
db69ef65 | 1479 | }, |
0dd2d749 FY |
1480 | { |
1481 | .name = "shareable_bits", | |
1482 | .mode = 0444, | |
1483 | .kf_ops = &rdtgroup_kf_single_ops, | |
1484 | .seq_show = rdt_shareable_bits_show, | |
1485 | .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, | |
1486 | }, | |
e6519011 RC |
1487 | { |
1488 | .name = "bit_usage", | |
1489 | .mode = 0444, | |
1490 | .kf_ops = &rdtgroup_kf_single_ops, | |
1491 | .seq_show = rdt_bit_usage_show, | |
1492 | .fflags = RF_CTRL_INFO | RFTYPE_RES_CACHE, | |
1493 | }, | |
db69ef65 VS |
1494 | { |
1495 | .name = "min_bandwidth", | |
1496 | .mode = 0444, | |
1497 | .kf_ops = &rdtgroup_kf_single_ops, | |
1498 | .seq_show = rdt_min_bw_show, | |
5dc1d5c6 | 1499 | .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, |
db69ef65 VS |
1500 | }, |
1501 | { | |
1502 | .name = "bandwidth_gran", | |
1503 | .mode = 0444, | |
1504 | .kf_ops = &rdtgroup_kf_single_ops, | |
1505 | .seq_show = rdt_bw_gran_show, | |
5dc1d5c6 | 1506 | .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, |
db69ef65 VS |
1507 | }, |
1508 | { | |
1509 | .name = "delay_linear", | |
1510 | .mode = 0444, | |
1511 | .kf_ops = &rdtgroup_kf_single_ops, | |
1512 | .seq_show = rdt_delay_linear_show, | |
5dc1d5c6 TL |
1513 | .fflags = RF_CTRL_INFO | RFTYPE_RES_MB, |
1514 | }, | |
d4ab3320 VS |
1515 | { |
1516 | .name = "max_threshold_occupancy", | |
1517 | .mode = 0644, | |
1518 | .kf_ops = &rdtgroup_kf_single_ops, | |
1519 | .write = max_threshold_occ_write, | |
1520 | .seq_show = max_threshold_occ_show, | |
1521 | .fflags = RF_MON_INFO | RFTYPE_RES_CACHE, | |
1522 | }, | |
5dc1d5c6 TL |
1523 | { |
1524 | .name = "cpus", | |
1525 | .mode = 0644, | |
1526 | .kf_ops = &rdtgroup_kf_single_ops, | |
1527 | .write = rdtgroup_cpus_write, | |
1528 | .seq_show = rdtgroup_cpus_show, | |
1529 | .fflags = RFTYPE_BASE, | |
1530 | }, | |
1531 | { | |
1532 | .name = "cpus_list", | |
1533 | .mode = 0644, | |
1534 | .kf_ops = &rdtgroup_kf_single_ops, | |
1535 | .write = rdtgroup_cpus_write, | |
1536 | .seq_show = rdtgroup_cpus_show, | |
1537 | .flags = RFTYPE_FLAGS_CPUS_LIST, | |
1538 | .fflags = RFTYPE_BASE, | |
1539 | }, | |
1540 | { | |
1541 | .name = "tasks", | |
1542 | .mode = 0644, | |
1543 | .kf_ops = &rdtgroup_kf_single_ops, | |
1544 | .write = rdtgroup_tasks_write, | |
1545 | .seq_show = rdtgroup_tasks_show, | |
1546 | .fflags = RFTYPE_BASE, | |
1547 | }, | |
1548 | { | |
1549 | .name = "schemata", | |
1550 | .mode = 0644, | |
1551 | .kf_ops = &rdtgroup_kf_single_ops, | |
1552 | .write = rdtgroup_schemata_write, | |
1553 | .seq_show = rdtgroup_schemata_show, | |
1554 | .fflags = RF_CTRL_BASE, | |
db69ef65 | 1555 | }, |
d48d7a57 RC |
1556 | { |
1557 | .name = "mode", | |
1558 | .mode = 0644, | |
1559 | .kf_ops = &rdtgroup_kf_single_ops, | |
1560 | .write = rdtgroup_mode_write, | |
1561 | .seq_show = rdtgroup_mode_show, | |
1562 | .fflags = RF_CTRL_BASE, | |
1563 | }, | |
d9b48c86 RC |
1564 | { |
1565 | .name = "size", | |
1566 | .mode = 0444, | |
1567 | .kf_ops = &rdtgroup_kf_single_ops, | |
1568 | .seq_show = rdtgroup_size_show, | |
1569 | .fflags = RF_CTRL_BASE, | |
1570 | }, | |
1571 | ||
db69ef65 VS |
1572 | }; |
1573 | ||
5dc1d5c6 | 1574 | static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags) |
db69ef65 | 1575 | { |
5dc1d5c6 TL |
1576 | struct rftype *rfts, *rft; |
1577 | int ret, len; | |
1578 | ||
1579 | rfts = res_common_files; | |
1580 | len = ARRAY_SIZE(res_common_files); | |
1581 | ||
1582 | lockdep_assert_held(&rdtgroup_mutex); | |
1583 | ||
1584 | for (rft = rfts; rft < rfts + len; rft++) { | |
1585 | if ((fflags & rft->fflags) == rft->fflags) { | |
1586 | ret = rdtgroup_add_file(kn, rft); | |
1587 | if (ret) | |
1588 | goto error; | |
1589 | } | |
1590 | } | |
1591 | ||
1592 | return 0; | |
1593 | error: | |
1594 | pr_warn("Failed to add %s, err=%d\n", rft->name, ret); | |
1595 | while (--rft >= rfts) { | |
1596 | if ((fflags & rft->fflags) == rft->fflags) | |
1597 | kernfs_remove_by_name(kn, rft->name); | |
1598 | } | |
1599 | return ret; | |
db69ef65 VS |
1600 | } |
1601 | ||
125db711 RC |
1602 | /** |
1603 | * rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file | |
1604 | * @r: The resource group with which the file is associated. | |
1605 | * @name: Name of the file | |
1606 | * | |
1607 | * The permissions of named resctrl file, directory, or link are modified | |
1608 | * to not allow read, write, or execute by any user. | |
1609 | * | |
1610 | * WARNING: This function is intended to communicate to the user that the | |
1611 | * resctrl file has been locked down - that it is not relevant to the | |
1612 | * particular state the system finds itself in. It should not be relied | |
1613 | * on to protect from user access because after the file's permissions | |
1614 | * are restricted the user can still change the permissions using chmod | |
1615 | * from the command line. | |
1616 | * | |
1617 | * Return: 0 on success, <0 on failure. | |
1618 | */ | |
1619 | int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name) | |
1620 | { | |
1621 | struct iattr iattr = {.ia_valid = ATTR_MODE,}; | |
1622 | struct kernfs_node *kn; | |
1623 | int ret = 0; | |
1624 | ||
1625 | kn = kernfs_find_and_get_ns(r->kn, name, NULL); | |
1626 | if (!kn) | |
1627 | return -ENOENT; | |
1628 | ||
1629 | switch (kernfs_type(kn)) { | |
1630 | case KERNFS_DIR: | |
1631 | iattr.ia_mode = S_IFDIR; | |
1632 | break; | |
1633 | case KERNFS_FILE: | |
1634 | iattr.ia_mode = S_IFREG; | |
1635 | break; | |
1636 | case KERNFS_LINK: | |
1637 | iattr.ia_mode = S_IFLNK; | |
1638 | break; | |
1639 | } | |
1640 | ||
1641 | ret = kernfs_setattr(kn, &iattr); | |
1642 | kernfs_put(kn); | |
1643 | return ret; | |
1644 | } | |
1645 | ||
1646 | /** | |
1647 | * rdtgroup_kn_mode_restore - Restore user access to named resctrl file | |
1648 | * @r: The resource group with which the file is associated. | |
1649 | * @name: Name of the file | |
392487de | 1650 | * @mask: Mask of permissions that should be restored |
125db711 RC |
1651 | * |
1652 | * Restore the permissions of the named file. If @name is a directory the | |
1653 | * permissions of its parent will be used. | |
1654 | * | |
1655 | * Return: 0 on success, <0 on failure. | |
1656 | */ | |
392487de RC |
1657 | int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name, |
1658 | umode_t mask) | |
125db711 RC |
1659 | { |
1660 | struct iattr iattr = {.ia_valid = ATTR_MODE,}; | |
1661 | struct kernfs_node *kn, *parent; | |
1662 | struct rftype *rfts, *rft; | |
1663 | int ret, len; | |
1664 | ||
1665 | rfts = res_common_files; | |
1666 | len = ARRAY_SIZE(res_common_files); | |
1667 | ||
1668 | for (rft = rfts; rft < rfts + len; rft++) { | |
1669 | if (!strcmp(rft->name, name)) | |
392487de | 1670 | iattr.ia_mode = rft->mode & mask; |
125db711 RC |
1671 | } |
1672 | ||
1673 | kn = kernfs_find_and_get_ns(r->kn, name, NULL); | |
1674 | if (!kn) | |
1675 | return -ENOENT; | |
1676 | ||
1677 | switch (kernfs_type(kn)) { | |
1678 | case KERNFS_DIR: | |
1679 | parent = kernfs_get_parent(kn); | |
1680 | if (parent) { | |
1681 | iattr.ia_mode |= parent->mode; | |
1682 | kernfs_put(parent); | |
1683 | } | |
1684 | iattr.ia_mode |= S_IFDIR; | |
1685 | break; | |
1686 | case KERNFS_FILE: | |
1687 | iattr.ia_mode |= S_IFREG; | |
1688 | break; | |
1689 | case KERNFS_LINK: | |
1690 | iattr.ia_mode |= S_IFLNK; | |
1691 | break; | |
1692 | } | |
1693 | ||
1694 | ret = kernfs_setattr(kn, &iattr); | |
1695 | kernfs_put(kn); | |
1696 | return ret; | |
1697 | } | |
1698 | ||
5dc1d5c6 TL |
1699 | static int rdtgroup_mkdir_info_resdir(struct rdt_resource *r, char *name, |
1700 | unsigned long fflags) | |
6a507a6a | 1701 | { |
5dc1d5c6 TL |
1702 | struct kernfs_node *kn_subdir; |
1703 | int ret; | |
1704 | ||
1705 | kn_subdir = kernfs_create_dir(kn_info, name, | |
1706 | kn_info->mode, r); | |
1707 | if (IS_ERR(kn_subdir)) | |
1708 | return PTR_ERR(kn_subdir); | |
1709 | ||
1710 | kernfs_get(kn_subdir); | |
1711 | ret = rdtgroup_kn_set_ugid(kn_subdir); | |
1712 | if (ret) | |
1713 | return ret; | |
1714 | ||
1715 | ret = rdtgroup_add_files(kn_subdir, fflags); | |
1716 | if (!ret) | |
1717 | kernfs_activate(kn_subdir); | |
1718 | ||
1719 | return ret; | |
6a507a6a VS |
1720 | } |
1721 | ||
4e978d06 FY |
1722 | static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn) |
1723 | { | |
4e978d06 | 1724 | struct rdt_resource *r; |
5dc1d5c6 | 1725 | unsigned long fflags; |
d4ab3320 | 1726 | char name[32]; |
5dc1d5c6 | 1727 | int ret; |
4e978d06 FY |
1728 | |
1729 | /* create the directory */ | |
1730 | kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL); | |
1731 | if (IS_ERR(kn_info)) | |
1732 | return PTR_ERR(kn_info); | |
1733 | kernfs_get(kn_info); | |
1734 | ||
9b3a7fd0 TL |
1735 | ret = rdtgroup_add_files(kn_info, RF_TOP_INFO); |
1736 | if (ret) | |
1737 | goto out_destroy; | |
1738 | ||
1b5c0b75 | 1739 | for_each_alloc_enabled_rdt_resource(r) { |
5dc1d5c6 TL |
1740 | fflags = r->fflags | RF_CTRL_INFO; |
1741 | ret = rdtgroup_mkdir_info_resdir(r, r->name, fflags); | |
4e978d06 FY |
1742 | if (ret) |
1743 | goto out_destroy; | |
4e978d06 | 1744 | } |
d4ab3320 VS |
1745 | |
1746 | for_each_mon_enabled_rdt_resource(r) { | |
1747 | fflags = r->fflags | RF_MON_INFO; | |
1748 | sprintf(name, "%s_MON", r->name); | |
1749 | ret = rdtgroup_mkdir_info_resdir(r, name, fflags); | |
1750 | if (ret) | |
1751 | goto out_destroy; | |
1752 | } | |
1753 | ||
4e978d06 FY |
1754 | /* |
1755 | * This extra ref will be put in kernfs_remove() and guarantees | |
1756 | * that @rdtgrp->kn is always accessible. | |
1757 | */ | |
1758 | kernfs_get(kn_info); | |
1759 | ||
1760 | ret = rdtgroup_kn_set_ugid(kn_info); | |
1761 | if (ret) | |
1762 | goto out_destroy; | |
1763 | ||
1764 | kernfs_activate(kn_info); | |
1765 | ||
1766 | return 0; | |
1767 | ||
1768 | out_destroy: | |
1769 | kernfs_remove(kn_info); | |
1770 | return ret; | |
1771 | } | |
1772 | ||
c7d9aac6 VS |
1773 | static int |
1774 | mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp, | |
1775 | char *name, struct kernfs_node **dest_kn) | |
1776 | { | |
1777 | struct kernfs_node *kn; | |
1778 | int ret; | |
1779 | ||
1780 | /* create the directory */ | |
1781 | kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); | |
1782 | if (IS_ERR(kn)) | |
1783 | return PTR_ERR(kn); | |
1784 | ||
1785 | if (dest_kn) | |
1786 | *dest_kn = kn; | |
1787 | ||
1788 | /* | |
1789 | * This extra ref will be put in kernfs_remove() and guarantees | |
1790 | * that @rdtgrp->kn is always accessible. | |
1791 | */ | |
1792 | kernfs_get(kn); | |
1793 | ||
1794 | ret = rdtgroup_kn_set_ugid(kn); | |
1795 | if (ret) | |
1796 | goto out_destroy; | |
1797 | ||
1798 | kernfs_activate(kn); | |
1799 | ||
1800 | return 0; | |
1801 | ||
1802 | out_destroy: | |
1803 | kernfs_remove(kn); | |
1804 | return ret; | |
1805 | } | |
99adde9b | 1806 | |
5ff193fb FY |
1807 | static void l3_qos_cfg_update(void *arg) |
1808 | { | |
1809 | bool *enable = arg; | |
1810 | ||
aa50453a | 1811 | wrmsrl(MSR_IA32_L3_QOS_CFG, *enable ? L3_QOS_CDP_ENABLE : 0ULL); |
5ff193fb FY |
1812 | } |
1813 | ||
99adde9b | 1814 | static void l2_qos_cfg_update(void *arg) |
5ff193fb | 1815 | { |
99adde9b FY |
1816 | bool *enable = arg; |
1817 | ||
aa50453a | 1818 | wrmsrl(MSR_IA32_L2_QOS_CFG, *enable ? L2_QOS_CDP_ENABLE : 0ULL); |
99adde9b FY |
1819 | } |
1820 | ||
19c635ab VS |
1821 | static inline bool is_mba_linear(void) |
1822 | { | |
1823 | return rdt_resources_all[RDT_RESOURCE_MBA].membw.delay_linear; | |
1824 | } | |
1825 | ||
99adde9b FY |
1826 | static int set_cache_qos_cfg(int level, bool enable) |
1827 | { | |
1828 | void (*update)(void *arg); | |
1829 | struct rdt_resource *r_l; | |
5ff193fb FY |
1830 | cpumask_var_t cpu_mask; |
1831 | struct rdt_domain *d; | |
1832 | int cpu; | |
1833 | ||
99adde9b FY |
1834 | if (level == RDT_RESOURCE_L3) |
1835 | update = l3_qos_cfg_update; | |
1836 | else if (level == RDT_RESOURCE_L2) | |
1837 | update = l2_qos_cfg_update; | |
1838 | else | |
1839 | return -EINVAL; | |
1840 | ||
ab6a2114 SB |
1841 | if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) |
1842 | return -ENOMEM; | |
1843 | ||
99adde9b FY |
1844 | r_l = &rdt_resources_all[level]; |
1845 | list_for_each_entry(d, &r_l->domains, list) { | |
5ff193fb FY |
1846 | /* Pick one CPU from each domain instance to update MSR */ |
1847 | cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask); | |
1848 | } | |
1849 | cpu = get_cpu(); | |
1850 | /* Update QOS_CFG MSR on this cpu if it's in cpu_mask. */ | |
1851 | if (cpumask_test_cpu(cpu, cpu_mask)) | |
99adde9b | 1852 | update(&enable); |
5ff193fb | 1853 | /* Update QOS_CFG MSR on all other cpus in cpu_mask. */ |
99adde9b | 1854 | smp_call_function_many(cpu_mask, update, &enable, 1); |
5ff193fb FY |
1855 | put_cpu(); |
1856 | ||
1857 | free_cpumask_var(cpu_mask); | |
1858 | ||
1859 | return 0; | |
1860 | } | |
1861 | ||
9fe04507 JM |
1862 | /* Restore the qos cfg state when a domain comes online */ |
1863 | void rdt_domain_reconfigure_cdp(struct rdt_resource *r) | |
1864 | { | |
1865 | if (!r->alloc_capable) | |
1866 | return; | |
1867 | ||
1868 | if (r == &rdt_resources_all[RDT_RESOURCE_L2DATA]) | |
1869 | l2_qos_cfg_update(&r->alloc_enabled); | |
1870 | ||
1871 | if (r == &rdt_resources_all[RDT_RESOURCE_L3DATA]) | |
1872 | l3_qos_cfg_update(&r->alloc_enabled); | |
1873 | } | |
1874 | ||
19c635ab VS |
1875 | /* |
1876 | * Enable or disable the MBA software controller | |
1877 | * which helps user specify bandwidth in MBps. | |
1878 | * MBA software controller is supported only if | |
1879 | * MBM is supported and MBA is in linear scale. | |
1880 | */ | |
1881 | static int set_mba_sc(bool mba_sc) | |
1882 | { | |
1883 | struct rdt_resource *r = &rdt_resources_all[RDT_RESOURCE_MBA]; | |
1bd2a63b | 1884 | struct rdt_domain *d; |
19c635ab VS |
1885 | |
1886 | if (!is_mbm_enabled() || !is_mba_linear() || | |
1887 | mba_sc == is_mba_sc(r)) | |
1888 | return -EINVAL; | |
1889 | ||
1890 | r->membw.mba_sc = mba_sc; | |
1bd2a63b VS |
1891 | list_for_each_entry(d, &r->domains, list) |
1892 | setup_default_ctrlval(r, d->ctrl_val, d->mbps_val); | |
19c635ab VS |
1893 | |
1894 | return 0; | |
1895 | } | |
1896 | ||
99adde9b | 1897 | static int cdp_enable(int level, int data_type, int code_type) |
5ff193fb | 1898 | { |
99adde9b FY |
1899 | struct rdt_resource *r_ldata = &rdt_resources_all[data_type]; |
1900 | struct rdt_resource *r_lcode = &rdt_resources_all[code_type]; | |
1901 | struct rdt_resource *r_l = &rdt_resources_all[level]; | |
5ff193fb FY |
1902 | int ret; |
1903 | ||
99adde9b FY |
1904 | if (!r_l->alloc_capable || !r_ldata->alloc_capable || |
1905 | !r_lcode->alloc_capable) | |
5ff193fb FY |
1906 | return -EINVAL; |
1907 | ||
99adde9b | 1908 | ret = set_cache_qos_cfg(level, true); |
5ff193fb | 1909 | if (!ret) { |
99adde9b FY |
1910 | r_l->alloc_enabled = false; |
1911 | r_ldata->alloc_enabled = true; | |
1912 | r_lcode->alloc_enabled = true; | |
5ff193fb FY |
1913 | } |
1914 | return ret; | |
1915 | } | |
1916 | ||
99adde9b FY |
1917 | static int cdpl3_enable(void) |
1918 | { | |
1919 | return cdp_enable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, | |
1920 | RDT_RESOURCE_L3CODE); | |
1921 | } | |
1922 | ||
1923 | static int cdpl2_enable(void) | |
5ff193fb | 1924 | { |
99adde9b FY |
1925 | return cdp_enable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, |
1926 | RDT_RESOURCE_L2CODE); | |
1927 | } | |
1928 | ||
1929 | static void cdp_disable(int level, int data_type, int code_type) | |
1930 | { | |
1931 | struct rdt_resource *r = &rdt_resources_all[level]; | |
5ff193fb | 1932 | |
1b5c0b75 | 1933 | r->alloc_enabled = r->alloc_capable; |
5ff193fb | 1934 | |
99adde9b FY |
1935 | if (rdt_resources_all[data_type].alloc_enabled) { |
1936 | rdt_resources_all[data_type].alloc_enabled = false; | |
1937 | rdt_resources_all[code_type].alloc_enabled = false; | |
1938 | set_cache_qos_cfg(level, false); | |
5ff193fb FY |
1939 | } |
1940 | } | |
1941 | ||
99adde9b FY |
1942 | static void cdpl3_disable(void) |
1943 | { | |
1944 | cdp_disable(RDT_RESOURCE_L3, RDT_RESOURCE_L3DATA, RDT_RESOURCE_L3CODE); | |
1945 | } | |
1946 | ||
1947 | static void cdpl2_disable(void) | |
1948 | { | |
1949 | cdp_disable(RDT_RESOURCE_L2, RDT_RESOURCE_L2DATA, RDT_RESOURCE_L2CODE); | |
1950 | } | |
1951 | ||
1952 | static void cdp_disable_all(void) | |
1953 | { | |
1954 | if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled) | |
1955 | cdpl3_disable(); | |
1956 | if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled) | |
1957 | cdpl2_disable(); | |
1958 | } | |
1959 | ||
60cf5e10 FY |
1960 | /* |
1961 | * We don't allow rdtgroup directories to be created anywhere | |
1962 | * except the root directory. Thus when looking for the rdtgroup | |
1963 | * structure for a kernfs node we are either looking at a directory, | |
1964 | * in which case the rdtgroup structure is pointed at by the "priv" | |
1965 | * field, otherwise we have a file, and need only look to the parent | |
1966 | * to find the rdtgroup. | |
1967 | */ | |
1968 | static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn) | |
1969 | { | |
f57b3087 FY |
1970 | if (kernfs_type(kn) == KERNFS_DIR) { |
1971 | /* | |
1972 | * All the resource directories use "kn->priv" | |
1973 | * to point to the "struct rdtgroup" for the | |
1974 | * resource. "info" and its subdirectories don't | |
1975 | * have rdtgroup structures, so return NULL here. | |
1976 | */ | |
1977 | if (kn == kn_info || kn->parent == kn_info) | |
1978 | return NULL; | |
1979 | else | |
1980 | return kn->priv; | |
1981 | } else { | |
60cf5e10 | 1982 | return kn->parent->priv; |
f57b3087 | 1983 | } |
60cf5e10 FY |
1984 | } |
1985 | ||
1986 | struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn) | |
1987 | { | |
1988 | struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); | |
1989 | ||
f57b3087 FY |
1990 | if (!rdtgrp) |
1991 | return NULL; | |
1992 | ||
60cf5e10 FY |
1993 | atomic_inc(&rdtgrp->waitcount); |
1994 | kernfs_break_active_protection(kn); | |
1995 | ||
1996 | mutex_lock(&rdtgroup_mutex); | |
1997 | ||
1998 | /* Was this group deleted while we waited? */ | |
1999 | if (rdtgrp->flags & RDT_DELETED) | |
2000 | return NULL; | |
2001 | ||
2002 | return rdtgrp; | |
2003 | } | |
2004 | ||
2005 | void rdtgroup_kn_unlock(struct kernfs_node *kn) | |
2006 | { | |
2007 | struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); | |
2008 | ||
f57b3087 FY |
2009 | if (!rdtgrp) |
2010 | return; | |
2011 | ||
60cf5e10 FY |
2012 | mutex_unlock(&rdtgroup_mutex); |
2013 | ||
2014 | if (atomic_dec_and_test(&rdtgrp->waitcount) && | |
2015 | (rdtgrp->flags & RDT_DELETED)) { | |
e0bdfe8e RC |
2016 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
2017 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) | |
2018 | rdtgroup_pseudo_lock_remove(rdtgrp); | |
60cf5e10 | 2019 | kernfs_unbreak_active_protection(kn); |
49ec8f5b | 2020 | kernfs_put(rdtgrp->kn); |
60cf5e10 FY |
2021 | kfree(rdtgrp); |
2022 | } else { | |
2023 | kernfs_unbreak_active_protection(kn); | |
2024 | } | |
2025 | } | |
2026 | ||
4af4a88e VS |
2027 | static int mkdir_mondata_all(struct kernfs_node *parent_kn, |
2028 | struct rdtgroup *prgrp, | |
2029 | struct kernfs_node **mon_data_kn); | |
2030 | ||
23bf1b6b | 2031 | static int rdt_enable_ctx(struct rdt_fs_context *ctx) |
5ff193fb | 2032 | { |
23bf1b6b DH |
2033 | int ret = 0; |
2034 | ||
2035 | if (ctx->enable_cdpl2) | |
2036 | ret = cdpl2_enable(); | |
2037 | ||
2038 | if (!ret && ctx->enable_cdpl3) | |
2039 | ret = cdpl3_enable(); | |
2040 | ||
2041 | if (!ret && ctx->enable_mba_mbps) | |
2042 | ret = set_mba_sc(true); | |
2043 | ||
2044 | return ret; | |
2045 | } | |
2046 | ||
2047 | static int rdt_get_tree(struct fs_context *fc) | |
2048 | { | |
2049 | struct rdt_fs_context *ctx = rdt_fc2context(fc); | |
e3302683 VS |
2050 | struct rdt_domain *dom; |
2051 | struct rdt_resource *r; | |
5ff193fb FY |
2052 | int ret; |
2053 | ||
87943db7 | 2054 | cpus_read_lock(); |
5ff193fb FY |
2055 | mutex_lock(&rdtgroup_mutex); |
2056 | /* | |
2057 | * resctrl file system can only be mounted once. | |
2058 | */ | |
4af4a88e | 2059 | if (static_branch_unlikely(&rdt_enable_key)) { |
23bf1b6b | 2060 | ret = -EBUSY; |
5ff193fb FY |
2061 | goto out; |
2062 | } | |
2063 | ||
23bf1b6b DH |
2064 | ret = rdt_enable_ctx(ctx); |
2065 | if (ret < 0) | |
5ff193fb | 2066 | goto out_cdp; |
5ff193fb | 2067 | |
60cf5e10 FY |
2068 | closid_init(); |
2069 | ||
4e978d06 | 2070 | ret = rdtgroup_create_info_dir(rdtgroup_default.kn); |
23bf1b6b DH |
2071 | if (ret < 0) |
2072 | goto out_mba; | |
4e978d06 | 2073 | |
4af4a88e VS |
2074 | if (rdt_mon_capable) { |
2075 | ret = mongroup_create_dir(rdtgroup_default.kn, | |
334b0f4e | 2076 | &rdtgroup_default, "mon_groups", |
4af4a88e | 2077 | &kn_mongrp); |
23bf1b6b | 2078 | if (ret < 0) |
4af4a88e | 2079 | goto out_info; |
4af4a88e VS |
2080 | kernfs_get(kn_mongrp); |
2081 | ||
2082 | ret = mkdir_mondata_all(rdtgroup_default.kn, | |
2083 | &rdtgroup_default, &kn_mondata); | |
23bf1b6b | 2084 | if (ret < 0) |
4af4a88e | 2085 | goto out_mongrp; |
4af4a88e VS |
2086 | kernfs_get(kn_mondata); |
2087 | rdtgroup_default.mon.mon_data_kn = kn_mondata; | |
2088 | } | |
2089 | ||
32206ab3 | 2090 | ret = rdt_pseudo_lock_init(); |
23bf1b6b | 2091 | if (ret) |
32206ab3 | 2092 | goto out_mondata; |
32206ab3 | 2093 | |
23bf1b6b DH |
2094 | ret = kernfs_get_tree(fc); |
2095 | if (ret < 0) | |
32206ab3 | 2096 | goto out_psl; |
4af4a88e VS |
2097 | |
2098 | if (rdt_alloc_capable) | |
87943db7 | 2099 | static_branch_enable_cpuslocked(&rdt_alloc_enable_key); |
4af4a88e | 2100 | if (rdt_mon_capable) |
87943db7 | 2101 | static_branch_enable_cpuslocked(&rdt_mon_enable_key); |
5ff193fb | 2102 | |
4af4a88e | 2103 | if (rdt_alloc_capable || rdt_mon_capable) |
87943db7 | 2104 | static_branch_enable_cpuslocked(&rdt_enable_key); |
e3302683 VS |
2105 | |
2106 | if (is_mbm_enabled()) { | |
2107 | r = &rdt_resources_all[RDT_RESOURCE_L3]; | |
2108 | list_for_each_entry(dom, &r->domains, list) | |
bbc4615e | 2109 | mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL); |
e3302683 VS |
2110 | } |
2111 | ||
5ff193fb FY |
2112 | goto out; |
2113 | ||
32206ab3 RC |
2114 | out_psl: |
2115 | rdt_pseudo_lock_release(); | |
4af4a88e VS |
2116 | out_mondata: |
2117 | if (rdt_mon_capable) | |
2118 | kernfs_remove(kn_mondata); | |
2119 | out_mongrp: | |
2120 | if (rdt_mon_capable) | |
2121 | kernfs_remove(kn_mongrp); | |
2122 | out_info: | |
79298acc | 2123 | kernfs_remove(kn_info); |
23bf1b6b DH |
2124 | out_mba: |
2125 | if (ctx->enable_mba_mbps) | |
2126 | set_mba_sc(false); | |
5ff193fb | 2127 | out_cdp: |
99adde9b | 2128 | cdp_disable_all(); |
5ff193fb | 2129 | out: |
9b3a7fd0 | 2130 | rdt_last_cmd_clear(); |
5ff193fb | 2131 | mutex_unlock(&rdtgroup_mutex); |
87943db7 | 2132 | cpus_read_unlock(); |
23bf1b6b DH |
2133 | return ret; |
2134 | } | |
2135 | ||
2136 | enum rdt_param { | |
2137 | Opt_cdp, | |
2138 | Opt_cdpl2, | |
faa3604e | 2139 | Opt_mba_mbps, |
23bf1b6b DH |
2140 | nr__rdt_params |
2141 | }; | |
2142 | ||
d7167b14 | 2143 | static const struct fs_parameter_spec rdt_fs_parameters[] = { |
23bf1b6b DH |
2144 | fsparam_flag("cdp", Opt_cdp), |
2145 | fsparam_flag("cdpl2", Opt_cdpl2), | |
faa3604e | 2146 | fsparam_flag("mba_MBps", Opt_mba_mbps), |
23bf1b6b DH |
2147 | {} |
2148 | }; | |
2149 | ||
23bf1b6b DH |
2150 | static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param) |
2151 | { | |
2152 | struct rdt_fs_context *ctx = rdt_fc2context(fc); | |
2153 | struct fs_parse_result result; | |
2154 | int opt; | |
2155 | ||
d7167b14 | 2156 | opt = fs_parse(fc, rdt_fs_parameters, param, &result); |
23bf1b6b DH |
2157 | if (opt < 0) |
2158 | return opt; | |
2159 | ||
2160 | switch (opt) { | |
2161 | case Opt_cdp: | |
2162 | ctx->enable_cdpl3 = true; | |
2163 | return 0; | |
2164 | case Opt_cdpl2: | |
2165 | ctx->enable_cdpl2 = true; | |
2166 | return 0; | |
faa3604e | 2167 | case Opt_mba_mbps: |
23bf1b6b DH |
2168 | if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) |
2169 | return -EINVAL; | |
2170 | ctx->enable_mba_mbps = true; | |
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | return -EINVAL; | |
2175 | } | |
2176 | ||
2177 | static void rdt_fs_context_free(struct fs_context *fc) | |
2178 | { | |
2179 | struct rdt_fs_context *ctx = rdt_fc2context(fc); | |
5ff193fb | 2180 | |
23bf1b6b DH |
2181 | kernfs_free_fs_context(fc); |
2182 | kfree(ctx); | |
2183 | } | |
2184 | ||
2185 | static const struct fs_context_operations rdt_fs_context_ops = { | |
2186 | .free = rdt_fs_context_free, | |
2187 | .parse_param = rdt_parse_param, | |
2188 | .get_tree = rdt_get_tree, | |
2189 | }; | |
2190 | ||
2191 | static int rdt_init_fs_context(struct fs_context *fc) | |
2192 | { | |
2193 | struct rdt_fs_context *ctx; | |
2194 | ||
2195 | ctx = kzalloc(sizeof(struct rdt_fs_context), GFP_KERNEL); | |
2196 | if (!ctx) | |
2197 | return -ENOMEM; | |
2198 | ||
2199 | ctx->kfc.root = rdt_root; | |
2200 | ctx->kfc.magic = RDTGROUP_SUPER_MAGIC; | |
2201 | fc->fs_private = &ctx->kfc; | |
2202 | fc->ops = &rdt_fs_context_ops; | |
f7a99451 | 2203 | put_user_ns(fc->user_ns); |
23bf1b6b DH |
2204 | fc->user_ns = get_user_ns(&init_user_ns); |
2205 | fc->global = true; | |
2206 | return 0; | |
5ff193fb FY |
2207 | } |
2208 | ||
2545e9f5 | 2209 | static int reset_all_ctrls(struct rdt_resource *r) |
5ff193fb FY |
2210 | { |
2211 | struct msr_param msr_param; | |
2212 | cpumask_var_t cpu_mask; | |
2213 | struct rdt_domain *d; | |
2214 | int i, cpu; | |
2215 | ||
2216 | if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) | |
2217 | return -ENOMEM; | |
2218 | ||
2219 | msr_param.res = r; | |
2220 | msr_param.low = 0; | |
2221 | msr_param.high = r->num_closid; | |
2222 | ||
2223 | /* | |
2224 | * Disable resource control for this resource by setting all | |
2225 | * CBMs in all domains to the maximum mask value. Pick one CPU | |
2226 | * from each domain to update the MSRs below. | |
2227 | */ | |
2228 | list_for_each_entry(d, &r->domains, list) { | |
2229 | cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask); | |
2230 | ||
2231 | for (i = 0; i < r->num_closid; i++) | |
2545e9f5 | 2232 | d->ctrl_val[i] = r->default_ctrl; |
5ff193fb FY |
2233 | } |
2234 | cpu = get_cpu(); | |
2235 | /* Update CBM on this cpu if it's in cpu_mask. */ | |
2236 | if (cpumask_test_cpu(cpu, cpu_mask)) | |
2545e9f5 | 2237 | rdt_ctrl_update(&msr_param); |
5ff193fb | 2238 | /* Update CBM on all other cpus in cpu_mask. */ |
2545e9f5 | 2239 | smp_call_function_many(cpu_mask, rdt_ctrl_update, &msr_param, 1); |
5ff193fb FY |
2240 | put_cpu(); |
2241 | ||
2242 | free_cpumask_var(cpu_mask); | |
2243 | ||
2244 | return 0; | |
2245 | } | |
2246 | ||
f3cbeaca VS |
2247 | static bool is_closid_match(struct task_struct *t, struct rdtgroup *r) |
2248 | { | |
2249 | return (rdt_alloc_capable && | |
2250 | (r->type == RDTCTRL_GROUP) && (t->closid == r->closid)); | |
2251 | } | |
2252 | ||
2253 | static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r) | |
2254 | { | |
2255 | return (rdt_mon_capable && | |
2256 | (r->type == RDTMON_GROUP) && (t->rmid == r->mon.rmid)); | |
2257 | } | |
2258 | ||
4e978d06 | 2259 | /* |
0efc89be FY |
2260 | * Move tasks from one to the other group. If @from is NULL, then all tasks |
2261 | * in the systems are moved unconditionally (used for teardown). | |
2262 | * | |
2263 | * If @mask is not NULL the cpus on which moved tasks are running are set | |
2264 | * in that mask so the update smp function call is restricted to affected | |
2265 | * cpus. | |
4e978d06 | 2266 | */ |
0efc89be FY |
2267 | static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to, |
2268 | struct cpumask *mask) | |
4e978d06 | 2269 | { |
e02737d5 FY |
2270 | struct task_struct *p, *t; |
2271 | ||
e02737d5 | 2272 | read_lock(&tasklist_lock); |
0efc89be | 2273 | for_each_process_thread(p, t) { |
f3cbeaca VS |
2274 | if (!from || is_closid_match(t, from) || |
2275 | is_rmid_match(t, from)) { | |
0efc89be | 2276 | t->closid = to->closid; |
f3cbeaca VS |
2277 | t->rmid = to->mon.rmid; |
2278 | ||
0efc89be FY |
2279 | #ifdef CONFIG_SMP |
2280 | /* | |
2281 | * This is safe on x86 w/o barriers as the ordering | |
2282 | * of writing to task_cpu() and t->on_cpu is | |
2283 | * reverse to the reading here. The detection is | |
2284 | * inaccurate as tasks might move or schedule | |
2285 | * before the smp function call takes place. In | |
2286 | * such a case the function call is pointless, but | |
2287 | * there is no other side effect. | |
2288 | */ | |
2289 | if (mask && t->on_cpu) | |
2290 | cpumask_set_cpu(task_cpu(t), mask); | |
2291 | #endif | |
2292 | } | |
2293 | } | |
e02737d5 | 2294 | read_unlock(&tasklist_lock); |
0efc89be FY |
2295 | } |
2296 | ||
f3cbeaca VS |
2297 | static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp) |
2298 | { | |
2299 | struct rdtgroup *sentry, *stmp; | |
2300 | struct list_head *head; | |
2301 | ||
2302 | head = &rdtgrp->mon.crdtgrp_list; | |
2303 | list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) { | |
2304 | free_rmid(sentry->mon.rmid); | |
2305 | list_del(&sentry->mon.crdtgrp_list); | |
b8511ccc XS |
2306 | |
2307 | if (atomic_read(&sentry->waitcount) != 0) | |
2308 | sentry->flags = RDT_DELETED; | |
2309 | else | |
2310 | kfree(sentry); | |
f3cbeaca VS |
2311 | } |
2312 | } | |
2313 | ||
0efc89be FY |
2314 | /* |
2315 | * Forcibly remove all of subdirectories under root. | |
2316 | */ | |
2317 | static void rmdir_all_sub(void) | |
2318 | { | |
2319 | struct rdtgroup *rdtgrp, *tmp; | |
2320 | ||
2321 | /* Move all tasks to the default resource group */ | |
2322 | rdt_move_group_tasks(NULL, &rdtgroup_default, NULL); | |
60cf5e10 | 2323 | |
60cf5e10 | 2324 | list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) { |
4af4a88e VS |
2325 | /* Free any child rmids */ |
2326 | free_all_child_rdtgrp(rdtgrp); | |
2327 | ||
60cf5e10 FY |
2328 | /* Remove each rdtgroup other than root */ |
2329 | if (rdtgrp == &rdtgroup_default) | |
2330 | continue; | |
c7cc0cc1 | 2331 | |
e0bdfe8e RC |
2332 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
2333 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) | |
2334 | rdtgroup_pseudo_lock_remove(rdtgrp); | |
2335 | ||
c7cc0cc1 FY |
2336 | /* |
2337 | * Give any CPUs back to the default group. We cannot copy | |
2338 | * cpu_online_mask because a CPU might have executed the | |
2339 | * offline callback already, but is still marked online. | |
2340 | */ | |
2341 | cpumask_or(&rdtgroup_default.cpu_mask, | |
2342 | &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); | |
2343 | ||
4af4a88e VS |
2344 | free_rmid(rdtgrp->mon.rmid); |
2345 | ||
60cf5e10 FY |
2346 | kernfs_remove(rdtgrp->kn); |
2347 | list_del(&rdtgrp->rdtgroup_list); | |
b8511ccc XS |
2348 | |
2349 | if (atomic_read(&rdtgrp->waitcount) != 0) | |
2350 | rdtgrp->flags = RDT_DELETED; | |
2351 | else | |
2352 | kfree(rdtgrp); | |
60cf5e10 | 2353 | } |
0efc89be | 2354 | /* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */ |
a9fcf862 | 2355 | update_closid_rmid(cpu_online_mask, &rdtgroup_default); |
0efc89be | 2356 | |
4e978d06 | 2357 | kernfs_remove(kn_info); |
4af4a88e VS |
2358 | kernfs_remove(kn_mongrp); |
2359 | kernfs_remove(kn_mondata); | |
4e978d06 FY |
2360 | } |
2361 | ||
5ff193fb FY |
2362 | static void rdt_kill_sb(struct super_block *sb) |
2363 | { | |
2364 | struct rdt_resource *r; | |
2365 | ||
36b6f9fc | 2366 | cpus_read_lock(); |
5ff193fb FY |
2367 | mutex_lock(&rdtgroup_mutex); |
2368 | ||
19c635ab VS |
2369 | set_mba_sc(false); |
2370 | ||
5ff193fb | 2371 | /*Put everything back to default values. */ |
1b5c0b75 | 2372 | for_each_alloc_enabled_rdt_resource(r) |
2545e9f5 | 2373 | reset_all_ctrls(r); |
99adde9b | 2374 | cdp_disable_all(); |
4e978d06 | 2375 | rmdir_all_sub(); |
746e0859 | 2376 | rdt_pseudo_lock_release(); |
472ef09b | 2377 | rdtgroup_default.mode = RDT_MODE_SHAREABLE; |
36b6f9fc RC |
2378 | static_branch_disable_cpuslocked(&rdt_alloc_enable_key); |
2379 | static_branch_disable_cpuslocked(&rdt_mon_enable_key); | |
2380 | static_branch_disable_cpuslocked(&rdt_enable_key); | |
5ff193fb FY |
2381 | kernfs_kill_sb(sb); |
2382 | mutex_unlock(&rdtgroup_mutex); | |
36b6f9fc | 2383 | cpus_read_unlock(); |
5ff193fb FY |
2384 | } |
2385 | ||
2386 | static struct file_system_type rdt_fs_type = { | |
23bf1b6b DH |
2387 | .name = "resctrl", |
2388 | .init_fs_context = rdt_init_fs_context, | |
d7167b14 | 2389 | .parameters = rdt_fs_parameters, |
23bf1b6b | 2390 | .kill_sb = rdt_kill_sb, |
5ff193fb FY |
2391 | }; |
2392 | ||
d89b7379 VS |
2393 | static int mon_addfile(struct kernfs_node *parent_kn, const char *name, |
2394 | void *priv) | |
2395 | { | |
2396 | struct kernfs_node *kn; | |
2397 | int ret = 0; | |
2398 | ||
488dee96 DT |
2399 | kn = __kernfs_create_file(parent_kn, name, 0444, |
2400 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0, | |
d89b7379 VS |
2401 | &kf_mondata_ops, priv, NULL, NULL); |
2402 | if (IS_ERR(kn)) | |
2403 | return PTR_ERR(kn); | |
2404 | ||
2405 | ret = rdtgroup_kn_set_ugid(kn); | |
2406 | if (ret) { | |
2407 | kernfs_remove(kn); | |
2408 | return ret; | |
2409 | } | |
2410 | ||
2411 | return ret; | |
2412 | } | |
2413 | ||
895c663e VS |
2414 | /* |
2415 | * Remove all subdirectories of mon_data of ctrl_mon groups | |
2416 | * and monitor groups with given domain id. | |
2417 | */ | |
2418 | void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id) | |
2419 | { | |
2420 | struct rdtgroup *prgrp, *crgrp; | |
2421 | char name[32]; | |
2422 | ||
2423 | if (!r->mon_enabled) | |
2424 | return; | |
2425 | ||
2426 | list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { | |
2427 | sprintf(name, "mon_%s_%02d", r->name, dom_id); | |
2428 | kernfs_remove_by_name(prgrp->mon.mon_data_kn, name); | |
2429 | ||
2430 | list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list) | |
2431 | kernfs_remove_by_name(crgrp->mon.mon_data_kn, name); | |
2432 | } | |
2433 | } | |
2434 | ||
d89b7379 VS |
2435 | static int mkdir_mondata_subdir(struct kernfs_node *parent_kn, |
2436 | struct rdt_domain *d, | |
2437 | struct rdt_resource *r, struct rdtgroup *prgrp) | |
2438 | { | |
2439 | union mon_data_bits priv; | |
2440 | struct kernfs_node *kn; | |
2441 | struct mon_evt *mevt; | |
a4de1dfd | 2442 | struct rmid_read rr; |
d89b7379 VS |
2443 | char name[32]; |
2444 | int ret; | |
2445 | ||
2446 | sprintf(name, "mon_%s_%02d", r->name, d->id); | |
2447 | /* create the directory */ | |
2448 | kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); | |
2449 | if (IS_ERR(kn)) | |
2450 | return PTR_ERR(kn); | |
2451 | ||
2452 | /* | |
2453 | * This extra ref will be put in kernfs_remove() and guarantees | |
2454 | * that kn is always accessible. | |
2455 | */ | |
2456 | kernfs_get(kn); | |
2457 | ret = rdtgroup_kn_set_ugid(kn); | |
2458 | if (ret) | |
2459 | goto out_destroy; | |
2460 | ||
2461 | if (WARN_ON(list_empty(&r->evt_list))) { | |
2462 | ret = -EPERM; | |
2463 | goto out_destroy; | |
2464 | } | |
2465 | ||
2466 | priv.u.rid = r->rid; | |
2467 | priv.u.domid = d->id; | |
2468 | list_for_each_entry(mevt, &r->evt_list, list) { | |
2469 | priv.u.evtid = mevt->evtid; | |
2470 | ret = mon_addfile(kn, mevt->name, priv.priv); | |
2471 | if (ret) | |
2472 | goto out_destroy; | |
a4de1dfd VS |
2473 | |
2474 | if (is_mbm_event(mevt->evtid)) | |
46637d45 | 2475 | mon_event_read(&rr, r, d, prgrp, mevt->evtid, true); |
d89b7379 VS |
2476 | } |
2477 | kernfs_activate(kn); | |
2478 | return 0; | |
2479 | ||
2480 | out_destroy: | |
2481 | kernfs_remove(kn); | |
2482 | return ret; | |
2483 | } | |
2484 | ||
895c663e VS |
2485 | /* |
2486 | * Add all subdirectories of mon_data for "ctrl_mon" groups | |
2487 | * and "monitor" groups with given domain id. | |
2488 | */ | |
2489 | void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, | |
2490 | struct rdt_domain *d) | |
2491 | { | |
2492 | struct kernfs_node *parent_kn; | |
2493 | struct rdtgroup *prgrp, *crgrp; | |
2494 | struct list_head *head; | |
2495 | ||
2496 | if (!r->mon_enabled) | |
2497 | return; | |
2498 | ||
2499 | list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { | |
2500 | parent_kn = prgrp->mon.mon_data_kn; | |
2501 | mkdir_mondata_subdir(parent_kn, d, r, prgrp); | |
2502 | ||
2503 | head = &prgrp->mon.crdtgrp_list; | |
2504 | list_for_each_entry(crgrp, head, mon.crdtgrp_list) { | |
2505 | parent_kn = crgrp->mon.mon_data_kn; | |
2506 | mkdir_mondata_subdir(parent_kn, d, r, crgrp); | |
2507 | } | |
2508 | } | |
2509 | } | |
2510 | ||
d89b7379 VS |
2511 | static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn, |
2512 | struct rdt_resource *r, | |
2513 | struct rdtgroup *prgrp) | |
2514 | { | |
2515 | struct rdt_domain *dom; | |
2516 | int ret; | |
2517 | ||
2518 | list_for_each_entry(dom, &r->domains, list) { | |
2519 | ret = mkdir_mondata_subdir(parent_kn, dom, r, prgrp); | |
2520 | if (ret) | |
2521 | return ret; | |
2522 | } | |
2523 | ||
2524 | return 0; | |
2525 | } | |
2526 | ||
2527 | /* | |
2528 | * This creates a directory mon_data which contains the monitored data. | |
2529 | * | |
2530 | * mon_data has one directory for each domain whic are named | |
2531 | * in the format mon_<domain_name>_<domain_id>. For ex: A mon_data | |
2532 | * with L3 domain looks as below: | |
2533 | * ./mon_data: | |
2534 | * mon_L3_00 | |
2535 | * mon_L3_01 | |
2536 | * mon_L3_02 | |
2537 | * ... | |
2538 | * | |
2539 | * Each domain directory has one file per event: | |
2540 | * ./mon_L3_00/: | |
2541 | * llc_occupancy | |
2542 | * | |
2543 | */ | |
2544 | static int mkdir_mondata_all(struct kernfs_node *parent_kn, | |
2545 | struct rdtgroup *prgrp, | |
2546 | struct kernfs_node **dest_kn) | |
2547 | { | |
2548 | struct rdt_resource *r; | |
2549 | struct kernfs_node *kn; | |
2550 | int ret; | |
2551 | ||
2552 | /* | |
2553 | * Create the mon_data directory first. | |
2554 | */ | |
334b0f4e | 2555 | ret = mongroup_create_dir(parent_kn, prgrp, "mon_data", &kn); |
d89b7379 VS |
2556 | if (ret) |
2557 | return ret; | |
2558 | ||
2559 | if (dest_kn) | |
2560 | *dest_kn = kn; | |
2561 | ||
2562 | /* | |
2563 | * Create the subdirectories for each domain. Note that all events | |
2564 | * in a domain like L3 are grouped into a resource whose domain is L3 | |
2565 | */ | |
2566 | for_each_mon_enabled_rdt_resource(r) { | |
2567 | ret = mkdir_mondata_subdir_alldom(kn, r, prgrp); | |
2568 | if (ret) | |
2569 | goto out_destroy; | |
2570 | } | |
2571 | ||
2572 | return 0; | |
2573 | ||
2574 | out_destroy: | |
2575 | kernfs_remove(kn); | |
2576 | return ret; | |
2577 | } | |
2578 | ||
95f0b77e RC |
2579 | /** |
2580 | * cbm_ensure_valid - Enforce validity on provided CBM | |
2581 | * @_val: Candidate CBM | |
2582 | * @r: RDT resource to which the CBM belongs | |
2583 | * | |
2584 | * The provided CBM represents all cache portions available for use. This | |
2585 | * may be represented by a bitmap that does not consist of contiguous ones | |
2586 | * and thus be an invalid CBM. | |
2587 | * Here the provided CBM is forced to be a valid CBM by only considering | |
2588 | * the first set of contiguous bits as valid and clearing all bits. | |
2589 | * The intention here is to provide a valid default CBM with which a new | |
2590 | * resource group is initialized. The user can follow this with a | |
2591 | * modification to the CBM if the default does not satisfy the | |
2592 | * requirements. | |
2593 | */ | |
2ef085bd | 2594 | static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r) |
95f0b77e | 2595 | { |
95f0b77e RC |
2596 | unsigned int cbm_len = r->cache.cbm_len; |
2597 | unsigned long first_bit, zero_bit; | |
2ef085bd | 2598 | unsigned long val = _val; |
95f0b77e | 2599 | |
2ef085bd RC |
2600 | if (!val) |
2601 | return 0; | |
95f0b77e | 2602 | |
32f010de RC |
2603 | first_bit = find_first_bit(&val, cbm_len); |
2604 | zero_bit = find_next_zero_bit(&val, cbm_len, first_bit); | |
95f0b77e RC |
2605 | |
2606 | /* Clear any remaining bits to ensure contiguous region */ | |
32f010de | 2607 | bitmap_clear(&val, zero_bit, cbm_len - zero_bit); |
2ef085bd | 2608 | return (u32)val; |
95f0b77e RC |
2609 | } |
2610 | ||
7390619a XS |
2611 | /* |
2612 | * Initialize cache resources per RDT domain | |
2613 | * | |
2614 | * Set the RDT domain up to start off with all usable allocations. That is, | |
2615 | * all shareable and unused bits. All-zero CBM is invalid. | |
2616 | */ | |
2617 | static int __init_one_rdt_domain(struct rdt_domain *d, struct rdt_resource *r, | |
2618 | u32 closid) | |
2619 | { | |
2620 | struct rdt_resource *r_cdp = NULL; | |
2621 | struct rdt_domain *d_cdp = NULL; | |
2622 | u32 used_b = 0, unused_b = 0; | |
2623 | unsigned long tmp_cbm; | |
2624 | enum rdtgrp_mode mode; | |
2625 | u32 peer_ctl, *ctrl; | |
2626 | int i; | |
2627 | ||
2628 | rdt_cdp_peer_get(r, d, &r_cdp, &d_cdp); | |
2629 | d->have_new_ctrl = false; | |
2630 | d->new_ctrl = r->cache.shareable_bits; | |
2631 | used_b = r->cache.shareable_bits; | |
2632 | ctrl = d->ctrl_val; | |
2633 | for (i = 0; i < closids_supported(); i++, ctrl++) { | |
2634 | if (closid_allocated(i) && i != closid) { | |
2635 | mode = rdtgroup_mode_by_closid(i); | |
2636 | if (mode == RDT_MODE_PSEUDO_LOCKSETUP) | |
87d3aa28 JM |
2637 | /* |
2638 | * ctrl values for locksetup aren't relevant | |
2639 | * until the schemata is written, and the mode | |
2640 | * becomes RDT_MODE_PSEUDO_LOCKED. | |
2641 | */ | |
2642 | continue; | |
7390619a XS |
2643 | /* |
2644 | * If CDP is active include peer domain's | |
2645 | * usage to ensure there is no overlap | |
2646 | * with an exclusive group. | |
2647 | */ | |
2648 | if (d_cdp) | |
2649 | peer_ctl = d_cdp->ctrl_val[i]; | |
2650 | else | |
2651 | peer_ctl = 0; | |
2652 | used_b |= *ctrl | peer_ctl; | |
2653 | if (mode == RDT_MODE_SHAREABLE) | |
2654 | d->new_ctrl |= *ctrl | peer_ctl; | |
2655 | } | |
2656 | } | |
2657 | if (d->plr && d->plr->cbm > 0) | |
2658 | used_b |= d->plr->cbm; | |
2659 | unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1); | |
2660 | unused_b &= BIT_MASK(r->cache.cbm_len) - 1; | |
2661 | d->new_ctrl |= unused_b; | |
2662 | /* | |
2663 | * Force the initial CBM to be valid, user can | |
2664 | * modify the CBM based on system availability. | |
2665 | */ | |
2ef085bd | 2666 | d->new_ctrl = cbm_ensure_valid(d->new_ctrl, r); |
7390619a XS |
2667 | /* |
2668 | * Assign the u32 CBM to an unsigned long to ensure that | |
2669 | * bitmap_weight() does not access out-of-bound memory. | |
2670 | */ | |
2671 | tmp_cbm = d->new_ctrl; | |
2672 | if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) { | |
2673 | rdt_last_cmd_printf("No space on %s:%d\n", r->name, d->id); | |
2674 | return -ENOSPC; | |
2675 | } | |
2676 | d->have_new_ctrl = true; | |
2677 | ||
2678 | return 0; | |
2679 | } | |
2680 | ||
47820e73 XS |
2681 | /* |
2682 | * Initialize cache resources with default values. | |
95f0b77e RC |
2683 | * |
2684 | * A new RDT group is being created on an allocation capable (CAT) | |
2685 | * supporting system. Set this group up to start off with all usable | |
7390619a | 2686 | * allocations. |
95f0b77e | 2687 | * |
7390619a XS |
2688 | * If there are no more shareable bits available on any domain then |
2689 | * the entire allocation will fail. | |
95f0b77e | 2690 | */ |
47820e73 XS |
2691 | static int rdtgroup_init_cat(struct rdt_resource *r, u32 closid) |
2692 | { | |
2693 | struct rdt_domain *d; | |
2694 | int ret; | |
2695 | ||
2696 | list_for_each_entry(d, &r->domains, list) { | |
2697 | ret = __init_one_rdt_domain(d, r, closid); | |
2698 | if (ret < 0) | |
2699 | return ret; | |
2700 | } | |
2701 | ||
2702 | return 0; | |
2703 | } | |
2704 | ||
2705 | /* Initialize MBA resource with default values. */ | |
2706 | static void rdtgroup_init_mba(struct rdt_resource *r) | |
2707 | { | |
2708 | struct rdt_domain *d; | |
2709 | ||
2710 | list_for_each_entry(d, &r->domains, list) { | |
2711 | d->new_ctrl = is_mba_sc(r) ? MBA_MAX_MBPS : r->default_ctrl; | |
2712 | d->have_new_ctrl = true; | |
2713 | } | |
2714 | } | |
2715 | ||
2716 | /* Initialize the RDT group's allocations. */ | |
95f0b77e RC |
2717 | static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp) |
2718 | { | |
95f0b77e | 2719 | struct rdt_resource *r; |
7390619a | 2720 | int ret; |
95f0b77e RC |
2721 | |
2722 | for_each_alloc_enabled_rdt_resource(r) { | |
47820e73 XS |
2723 | if (r->rid == RDT_RESOURCE_MBA) { |
2724 | rdtgroup_init_mba(r); | |
2725 | } else { | |
2726 | ret = rdtgroup_init_cat(r, rdtgrp->closid); | |
7390619a XS |
2727 | if (ret < 0) |
2728 | return ret; | |
95f0b77e | 2729 | } |
95f0b77e | 2730 | |
95f0b77e RC |
2731 | ret = update_domains(r, rdtgrp->closid); |
2732 | if (ret < 0) { | |
723f1a0d | 2733 | rdt_last_cmd_puts("Failed to initialize allocations\n"); |
95f0b77e RC |
2734 | return ret; |
2735 | } | |
47820e73 | 2736 | |
95f0b77e RC |
2737 | } |
2738 | ||
40fba00f XS |
2739 | rdtgrp->mode = RDT_MODE_SHAREABLE; |
2740 | ||
95f0b77e RC |
2741 | return 0; |
2742 | } | |
2743 | ||
65b4f403 | 2744 | static int mkdir_rdt_prepare(struct kernfs_node *parent_kn, |
65b4f403 | 2745 | const char *name, umode_t mode, |
c7d9aac6 | 2746 | enum rdt_group_type rtype, struct rdtgroup **r) |
60cf5e10 | 2747 | { |
65b4f403 | 2748 | struct rdtgroup *prdtgrp, *rdtgrp; |
60cf5e10 | 2749 | struct kernfs_node *kn; |
65b4f403 VS |
2750 | uint files = 0; |
2751 | int ret; | |
60cf5e10 | 2752 | |
334b0f4e | 2753 | prdtgrp = rdtgroup_kn_lock_live(parent_kn); |
65b4f403 | 2754 | if (!prdtgrp) { |
60cf5e10 FY |
2755 | ret = -ENODEV; |
2756 | goto out_unlock; | |
2757 | } | |
2758 | ||
c966dac8 RC |
2759 | if (rtype == RDTMON_GROUP && |
2760 | (prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || | |
2761 | prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) { | |
2762 | ret = -EINVAL; | |
723f1a0d | 2763 | rdt_last_cmd_puts("Pseudo-locking in progress\n"); |
c966dac8 RC |
2764 | goto out_unlock; |
2765 | } | |
2766 | ||
60cf5e10 FY |
2767 | /* allocate the rdtgroup. */ |
2768 | rdtgrp = kzalloc(sizeof(*rdtgrp), GFP_KERNEL); | |
2769 | if (!rdtgrp) { | |
2770 | ret = -ENOSPC; | |
723f1a0d | 2771 | rdt_last_cmd_puts("Kernel out of memory\n"); |
65b4f403 | 2772 | goto out_unlock; |
60cf5e10 | 2773 | } |
65b4f403 | 2774 | *r = rdtgrp; |
c7d9aac6 VS |
2775 | rdtgrp->mon.parent = prdtgrp; |
2776 | rdtgrp->type = rtype; | |
2777 | INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list); | |
60cf5e10 FY |
2778 | |
2779 | /* kernfs creates the directory for rdtgrp */ | |
65b4f403 | 2780 | kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp); |
60cf5e10 FY |
2781 | if (IS_ERR(kn)) { |
2782 | ret = PTR_ERR(kn); | |
cfd0f34e | 2783 | rdt_last_cmd_puts("kernfs create error\n"); |
65b4f403 | 2784 | goto out_free_rgrp; |
60cf5e10 FY |
2785 | } |
2786 | rdtgrp->kn = kn; | |
2787 | ||
2788 | /* | |
2789 | * kernfs_remove() will drop the reference count on "kn" which | |
2790 | * will free it. But we still need it to stick around for the | |
2791 | * rdtgroup_kn_unlock(kn} call below. Take one extra reference | |
2792 | * here, which will be dropped inside rdtgroup_kn_unlock(). | |
2793 | */ | |
2794 | kernfs_get(kn); | |
2795 | ||
2796 | ret = rdtgroup_kn_set_ugid(kn); | |
cfd0f34e TL |
2797 | if (ret) { |
2798 | rdt_last_cmd_puts("kernfs perm error\n"); | |
60cf5e10 | 2799 | goto out_destroy; |
cfd0f34e | 2800 | } |
60cf5e10 | 2801 | |
c7d9aac6 | 2802 | files = RFTYPE_BASE | BIT(RF_CTRLSHIFT + rtype); |
65b4f403 | 2803 | ret = rdtgroup_add_files(kn, files); |
cfd0f34e TL |
2804 | if (ret) { |
2805 | rdt_last_cmd_puts("kernfs fill error\n"); | |
12e0110c | 2806 | goto out_destroy; |
cfd0f34e | 2807 | } |
12e0110c | 2808 | |
c7d9aac6 VS |
2809 | if (rdt_mon_capable) { |
2810 | ret = alloc_rmid(); | |
cfd0f34e | 2811 | if (ret < 0) { |
723f1a0d | 2812 | rdt_last_cmd_puts("Out of RMIDs\n"); |
c7d9aac6 | 2813 | goto out_destroy; |
cfd0f34e | 2814 | } |
c7d9aac6 | 2815 | rdtgrp->mon.rmid = ret; |
d89b7379 VS |
2816 | |
2817 | ret = mkdir_mondata_all(kn, rdtgrp, &rdtgrp->mon.mon_data_kn); | |
cfd0f34e TL |
2818 | if (ret) { |
2819 | rdt_last_cmd_puts("kernfs subdir error\n"); | |
d89b7379 | 2820 | goto out_idfree; |
cfd0f34e | 2821 | } |
c7d9aac6 | 2822 | } |
60cf5e10 FY |
2823 | kernfs_activate(kn); |
2824 | ||
65b4f403 | 2825 | /* |
334b0f4e | 2826 | * The caller unlocks the parent_kn upon success. |
65b4f403 VS |
2827 | */ |
2828 | return 0; | |
60cf5e10 | 2829 | |
d89b7379 VS |
2830 | out_idfree: |
2831 | free_rmid(rdtgrp->mon.rmid); | |
60cf5e10 FY |
2832 | out_destroy: |
2833 | kernfs_remove(rdtgrp->kn); | |
65b4f403 | 2834 | out_free_rgrp: |
60cf5e10 | 2835 | kfree(rdtgrp); |
60cf5e10 | 2836 | out_unlock: |
334b0f4e | 2837 | rdtgroup_kn_unlock(parent_kn); |
65b4f403 VS |
2838 | return ret; |
2839 | } | |
2840 | ||
2841 | static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp) | |
2842 | { | |
2843 | kernfs_remove(rgrp->kn); | |
c7d9aac6 | 2844 | free_rmid(rgrp->mon.rmid); |
65b4f403 VS |
2845 | kfree(rgrp); |
2846 | } | |
2847 | ||
c7d9aac6 VS |
2848 | /* |
2849 | * Create a monitor group under "mon_groups" directory of a control | |
2850 | * and monitor group(ctrl_mon). This is a resource group | |
2851 | * to monitor a subset of tasks and cpus in its parent ctrl_mon group. | |
2852 | */ | |
2853 | static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn, | |
32ada3b9 | 2854 | const char *name, umode_t mode) |
c7d9aac6 VS |
2855 | { |
2856 | struct rdtgroup *rdtgrp, *prgrp; | |
2857 | int ret; | |
2858 | ||
32ada3b9 | 2859 | ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTMON_GROUP, &rdtgrp); |
c7d9aac6 VS |
2860 | if (ret) |
2861 | return ret; | |
2862 | ||
2863 | prgrp = rdtgrp->mon.parent; | |
2864 | rdtgrp->closid = prgrp->closid; | |
2865 | ||
2866 | /* | |
2867 | * Add the rdtgrp to the list of rdtgrps the parent | |
2868 | * ctrl_mon group has to track. | |
2869 | */ | |
2870 | list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list); | |
2871 | ||
334b0f4e | 2872 | rdtgroup_kn_unlock(parent_kn); |
c7d9aac6 VS |
2873 | return ret; |
2874 | } | |
2875 | ||
65b4f403 VS |
2876 | /* |
2877 | * These are rdtgroups created under the root directory. Can be used | |
c7d9aac6 | 2878 | * to allocate and monitor resources. |
65b4f403 | 2879 | */ |
c7d9aac6 | 2880 | static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn, |
c7d9aac6 | 2881 | const char *name, umode_t mode) |
65b4f403 VS |
2882 | { |
2883 | struct rdtgroup *rdtgrp; | |
2884 | struct kernfs_node *kn; | |
2885 | u32 closid; | |
2886 | int ret; | |
2887 | ||
32ada3b9 | 2888 | ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTCTRL_GROUP, &rdtgrp); |
65b4f403 VS |
2889 | if (ret) |
2890 | return ret; | |
2891 | ||
2892 | kn = rdtgrp->kn; | |
2893 | ret = closid_alloc(); | |
cfd0f34e | 2894 | if (ret < 0) { |
723f1a0d | 2895 | rdt_last_cmd_puts("Out of CLOSIDs\n"); |
65b4f403 | 2896 | goto out_common_fail; |
cfd0f34e | 2897 | } |
65b4f403 | 2898 | closid = ret; |
36e74d35 | 2899 | ret = 0; |
65b4f403 VS |
2900 | |
2901 | rdtgrp->closid = closid; | |
95f0b77e RC |
2902 | ret = rdtgroup_init_alloc(rdtgrp); |
2903 | if (ret < 0) | |
2904 | goto out_id_free; | |
2905 | ||
65b4f403 VS |
2906 | list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups); |
2907 | ||
c7d9aac6 VS |
2908 | if (rdt_mon_capable) { |
2909 | /* | |
2910 | * Create an empty mon_groups directory to hold the subset | |
2911 | * of tasks and cpus to monitor. | |
2912 | */ | |
334b0f4e | 2913 | ret = mongroup_create_dir(kn, rdtgrp, "mon_groups", NULL); |
cfd0f34e TL |
2914 | if (ret) { |
2915 | rdt_last_cmd_puts("kernfs subdir error\n"); | |
95f0b77e | 2916 | goto out_del_list; |
cfd0f34e | 2917 | } |
c7d9aac6 VS |
2918 | } |
2919 | ||
65b4f403 VS |
2920 | goto out_unlock; |
2921 | ||
95f0b77e RC |
2922 | out_del_list: |
2923 | list_del(&rdtgrp->rdtgroup_list); | |
c7d9aac6 VS |
2924 | out_id_free: |
2925 | closid_free(closid); | |
65b4f403 VS |
2926 | out_common_fail: |
2927 | mkdir_rdt_prepare_clean(rdtgrp); | |
2928 | out_unlock: | |
334b0f4e | 2929 | rdtgroup_kn_unlock(parent_kn); |
60cf5e10 FY |
2930 | return ret; |
2931 | } | |
2932 | ||
c7d9aac6 VS |
2933 | /* |
2934 | * We allow creating mon groups only with in a directory called "mon_groups" | |
2935 | * which is present in every ctrl_mon group. Check if this is a valid | |
2936 | * "mon_groups" directory. | |
2937 | * | |
2938 | * 1. The directory should be named "mon_groups". | |
2939 | * 2. The mon group itself should "not" be named "mon_groups". | |
2940 | * This makes sure "mon_groups" directory always has a ctrl_mon group | |
2941 | * as parent. | |
2942 | */ | |
2943 | static bool is_mon_groups(struct kernfs_node *kn, const char *name) | |
2944 | { | |
2945 | return (!strcmp(kn->name, "mon_groups") && | |
2946 | strcmp(name, "mon_groups")); | |
2947 | } | |
2948 | ||
65b4f403 VS |
2949 | static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
2950 | umode_t mode) | |
2951 | { | |
2952 | /* Do not accept '\n' to avoid unparsable situation. */ | |
2953 | if (strchr(name, '\n')) | |
2954 | return -EINVAL; | |
2955 | ||
2956 | /* | |
2957 | * If the parent directory is the root directory and RDT | |
c7d9aac6 VS |
2958 | * allocation is supported, add a control and monitoring |
2959 | * subdirectory | |
65b4f403 VS |
2960 | */ |
2961 | if (rdt_alloc_capable && parent_kn == rdtgroup_default.kn) | |
32ada3b9 | 2962 | return rdtgroup_mkdir_ctrl_mon(parent_kn, name, mode); |
c7d9aac6 VS |
2963 | |
2964 | /* | |
2965 | * If RDT monitoring is supported and the parent directory is a valid | |
2966 | * "mon_groups" directory, add a monitoring subdirectory. | |
2967 | */ | |
2968 | if (rdt_mon_capable && is_mon_groups(parent_kn, name)) | |
32ada3b9 | 2969 | return rdtgroup_mkdir_mon(parent_kn, name, mode); |
65b4f403 VS |
2970 | |
2971 | return -EPERM; | |
2972 | } | |
2973 | ||
f3cbeaca VS |
2974 | static int rdtgroup_rmdir_mon(struct kernfs_node *kn, struct rdtgroup *rdtgrp, |
2975 | cpumask_var_t tmpmask) | |
2976 | { | |
2977 | struct rdtgroup *prdtgrp = rdtgrp->mon.parent; | |
2978 | int cpu; | |
2979 | ||
2980 | /* Give any tasks back to the parent group */ | |
2981 | rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask); | |
2982 | ||
2983 | /* Update per cpu rmid of the moved CPUs first */ | |
2984 | for_each_cpu(cpu, &rdtgrp->cpu_mask) | |
a9110b55 | 2985 | per_cpu(pqr_state.default_rmid, cpu) = prdtgrp->mon.rmid; |
f3cbeaca VS |
2986 | /* |
2987 | * Update the MSR on moved CPUs and CPUs which have moved | |
2988 | * task running on them. | |
2989 | */ | |
2990 | cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); | |
2991 | update_closid_rmid(tmpmask, NULL); | |
2992 | ||
2993 | rdtgrp->flags = RDT_DELETED; | |
2994 | free_rmid(rdtgrp->mon.rmid); | |
2995 | ||
2996 | /* | |
2997 | * Remove the rdtgrp from the parent ctrl_mon group's list | |
2998 | */ | |
2999 | WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list)); | |
3000 | list_del(&rdtgrp->mon.crdtgrp_list); | |
3001 | ||
3002 | /* | |
3003 | * one extra hold on this, will drop when we kfree(rdtgrp) | |
3004 | * in rdtgroup_kn_unlock() | |
3005 | */ | |
3006 | kernfs_get(kn); | |
3007 | kernfs_remove(rdtgrp->kn); | |
3008 | ||
3009 | return 0; | |
3010 | } | |
3011 | ||
17eafd07 RC |
3012 | static int rdtgroup_ctrl_remove(struct kernfs_node *kn, |
3013 | struct rdtgroup *rdtgrp) | |
3014 | { | |
3015 | rdtgrp->flags = RDT_DELETED; | |
3016 | list_del(&rdtgrp->rdtgroup_list); | |
3017 | ||
3018 | /* | |
3019 | * one extra hold on this, will drop when we kfree(rdtgrp) | |
3020 | * in rdtgroup_kn_unlock() | |
3021 | */ | |
3022 | kernfs_get(kn); | |
3023 | kernfs_remove(rdtgrp->kn); | |
3024 | return 0; | |
3025 | } | |
3026 | ||
f9049547 VS |
3027 | static int rdtgroup_rmdir_ctrl(struct kernfs_node *kn, struct rdtgroup *rdtgrp, |
3028 | cpumask_var_t tmpmask) | |
60cf5e10 | 3029 | { |
f9049547 | 3030 | int cpu; |
60cf5e10 | 3031 | |
e02737d5 | 3032 | /* Give any tasks back to the default group */ |
0efc89be | 3033 | rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask); |
e02737d5 | 3034 | |
12e0110c TL |
3035 | /* Give any CPUs back to the default group */ |
3036 | cpumask_or(&rdtgroup_default.cpu_mask, | |
3037 | &rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); | |
0efc89be | 3038 | |
f3cbeaca VS |
3039 | /* Update per cpu closid and rmid of the moved CPUs first */ |
3040 | for_each_cpu(cpu, &rdtgrp->cpu_mask) { | |
a9110b55 VS |
3041 | per_cpu(pqr_state.default_closid, cpu) = rdtgroup_default.closid; |
3042 | per_cpu(pqr_state.default_rmid, cpu) = rdtgroup_default.mon.rmid; | |
f3cbeaca VS |
3043 | } |
3044 | ||
0efc89be FY |
3045 | /* |
3046 | * Update the MSR on moved CPUs and CPUs which have moved | |
3047 | * task running on them. | |
3048 | */ | |
3049 | cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); | |
a9fcf862 | 3050 | update_closid_rmid(tmpmask, NULL); |
12e0110c | 3051 | |
60cf5e10 | 3052 | closid_free(rdtgrp->closid); |
f3cbeaca VS |
3053 | free_rmid(rdtgrp->mon.rmid); |
3054 | ||
074fadee XS |
3055 | rdtgroup_ctrl_remove(kn, rdtgrp); |
3056 | ||
f3cbeaca VS |
3057 | /* |
3058 | * Free all the child monitor group rmids. | |
3059 | */ | |
3060 | free_all_child_rdtgrp(rdtgrp); | |
3061 | ||
f9049547 VS |
3062 | return 0; |
3063 | } | |
3064 | ||
3065 | static int rdtgroup_rmdir(struct kernfs_node *kn) | |
3066 | { | |
3067 | struct kernfs_node *parent_kn = kn->parent; | |
3068 | struct rdtgroup *rdtgrp; | |
3069 | cpumask_var_t tmpmask; | |
3070 | int ret = 0; | |
3071 | ||
3072 | if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
3073 | return -ENOMEM; | |
3074 | ||
3075 | rdtgrp = rdtgroup_kn_lock_live(kn); | |
3076 | if (!rdtgrp) { | |
3077 | ret = -EPERM; | |
3078 | goto out; | |
3079 | } | |
3080 | ||
3081 | /* | |
3082 | * If the rdtgroup is a ctrl_mon group and parent directory | |
f3cbeaca VS |
3083 | * is the root directory, remove the ctrl_mon group. |
3084 | * | |
3085 | * If the rdtgroup is a mon group and parent directory | |
3086 | * is a valid "mon_groups" directory, remove the mon group. | |
f9049547 | 3087 | */ |
b0151da5 RC |
3088 | if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn && |
3089 | rdtgrp != &rdtgroup_default) { | |
e0bdfe8e RC |
3090 | if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
3091 | rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { | |
3092 | ret = rdtgroup_ctrl_remove(kn, rdtgrp); | |
3093 | } else { | |
3094 | ret = rdtgroup_rmdir_ctrl(kn, rdtgrp, tmpmask); | |
3095 | } | |
3096 | } else if (rdtgrp->type == RDTMON_GROUP && | |
3097 | is_mon_groups(parent_kn, kn->name)) { | |
f3cbeaca | 3098 | ret = rdtgroup_rmdir_mon(kn, rdtgrp, tmpmask); |
e0bdfe8e | 3099 | } else { |
f9049547 | 3100 | ret = -EPERM; |
e0bdfe8e | 3101 | } |
f9049547 | 3102 | |
0efc89be | 3103 | out: |
60cf5e10 | 3104 | rdtgroup_kn_unlock(kn); |
0efc89be FY |
3105 | free_cpumask_var(tmpmask); |
3106 | return ret; | |
60cf5e10 FY |
3107 | } |
3108 | ||
76ae054c SL |
3109 | static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf) |
3110 | { | |
1b5c0b75 | 3111 | if (rdt_resources_all[RDT_RESOURCE_L3DATA].alloc_enabled) |
76ae054c | 3112 | seq_puts(seq, ",cdp"); |
2cc81c69 XS |
3113 | |
3114 | if (rdt_resources_all[RDT_RESOURCE_L2DATA].alloc_enabled) | |
3115 | seq_puts(seq, ",cdpl2"); | |
3116 | ||
3117 | if (is_mba_sc(&rdt_resources_all[RDT_RESOURCE_MBA])) | |
3118 | seq_puts(seq, ",mba_MBps"); | |
3119 | ||
76ae054c SL |
3120 | return 0; |
3121 | } | |
3122 | ||
5ff193fb | 3123 | static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = { |
76ae054c SL |
3124 | .mkdir = rdtgroup_mkdir, |
3125 | .rmdir = rdtgroup_rmdir, | |
3126 | .show_options = rdtgroup_show_options, | |
5ff193fb FY |
3127 | }; |
3128 | ||
3129 | static int __init rdtgroup_setup_root(void) | |
3130 | { | |
12e0110c TL |
3131 | int ret; |
3132 | ||
5ff193fb | 3133 | rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops, |
21220bb1 RC |
3134 | KERNFS_ROOT_CREATE_DEACTIVATED | |
3135 | KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK, | |
5ff193fb FY |
3136 | &rdtgroup_default); |
3137 | if (IS_ERR(rdt_root)) | |
3138 | return PTR_ERR(rdt_root); | |
3139 | ||
3140 | mutex_lock(&rdtgroup_mutex); | |
3141 | ||
3142 | rdtgroup_default.closid = 0; | |
c7d9aac6 VS |
3143 | rdtgroup_default.mon.rmid = 0; |
3144 | rdtgroup_default.type = RDTCTRL_GROUP; | |
3145 | INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list); | |
3146 | ||
5ff193fb FY |
3147 | list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups); |
3148 | ||
5dc1d5c6 | 3149 | ret = rdtgroup_add_files(rdt_root->kn, RF_CTRL_BASE); |
12e0110c TL |
3150 | if (ret) { |
3151 | kernfs_destroy_root(rdt_root); | |
3152 | goto out; | |
3153 | } | |
3154 | ||
5ff193fb FY |
3155 | rdtgroup_default.kn = rdt_root->kn; |
3156 | kernfs_activate(rdtgroup_default.kn); | |
3157 | ||
12e0110c | 3158 | out: |
5ff193fb FY |
3159 | mutex_unlock(&rdtgroup_mutex); |
3160 | ||
12e0110c | 3161 | return ret; |
5ff193fb FY |
3162 | } |
3163 | ||
3164 | /* | |
3165 | * rdtgroup_init - rdtgroup initialization | |
3166 | * | |
3167 | * Setup resctrl file system including set up root, create mount point, | |
3168 | * register rdtgroup filesystem, and initialize files under root directory. | |
3169 | * | |
3170 | * Return: 0 on success or -errno | |
3171 | */ | |
3172 | int __init rdtgroup_init(void) | |
3173 | { | |
3174 | int ret = 0; | |
3175 | ||
9b3a7fd0 TL |
3176 | seq_buf_init(&last_cmd_status, last_cmd_status_buf, |
3177 | sizeof(last_cmd_status_buf)); | |
3178 | ||
5ff193fb FY |
3179 | ret = rdtgroup_setup_root(); |
3180 | if (ret) | |
3181 | return ret; | |
3182 | ||
3183 | ret = sysfs_create_mount_point(fs_kobj, "resctrl"); | |
3184 | if (ret) | |
3185 | goto cleanup_root; | |
3186 | ||
3187 | ret = register_filesystem(&rdt_fs_type); | |
3188 | if (ret) | |
3189 | goto cleanup_mountpoint; | |
3190 | ||
37707ec6 RC |
3191 | /* |
3192 | * Adding the resctrl debugfs directory here may not be ideal since | |
3193 | * it would let the resctrl debugfs directory appear on the debugfs | |
3194 | * filesystem before the resctrl filesystem is mounted. | |
3195 | * It may also be ok since that would enable debugging of RDT before | |
3196 | * resctrl is mounted. | |
3197 | * The reason why the debugfs directory is created here and not in | |
3198 | * rdt_mount() is because rdt_mount() takes rdtgroup_mutex and | |
3199 | * during the debugfs directory creation also &sb->s_type->i_mutex_key | |
3200 | * (the lockdep class of inode->i_rwsem). Other filesystem | |
3201 | * interactions (eg. SyS_getdents) have the lock ordering: | |
c1e8d7c6 ML |
3202 | * &sb->s_type->i_mutex_key --> &mm->mmap_lock |
3203 | * During mmap(), called with &mm->mmap_lock, the rdtgroup_mutex | |
37707ec6 | 3204 | * is taken, thus creating dependency: |
c1e8d7c6 | 3205 | * &mm->mmap_lock --> rdtgroup_mutex for the latter that can cause |
37707ec6 RC |
3206 | * issues considering the other two lock dependencies. |
3207 | * By creating the debugfs directory here we avoid a dependency | |
3208 | * that may cause deadlock (even though file operations cannot | |
3209 | * occur until the filesystem is mounted, but I do not know how to | |
3210 | * tell lockdep that). | |
3211 | */ | |
3212 | debugfs_resctrl = debugfs_create_dir("resctrl", NULL); | |
3213 | ||
5ff193fb FY |
3214 | return 0; |
3215 | ||
3216 | cleanup_mountpoint: | |
3217 | sysfs_remove_mount_point(fs_kobj, "resctrl"); | |
3218 | cleanup_root: | |
3219 | kernfs_destroy_root(rdt_root); | |
3220 | ||
3221 | return ret; | |
3222 | } | |
0af6a48d RC |
3223 | |
3224 | void __exit rdtgroup_exit(void) | |
3225 | { | |
37707ec6 | 3226 | debugfs_remove_recursive(debugfs_resctrl); |
0af6a48d RC |
3227 | unregister_filesystem(&rdt_fs_type); |
3228 | sysfs_remove_mount_point(fs_kobj, "resctrl"); | |
3229 | kernfs_destroy_root(rdt_root); | |
3230 | } |