| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/* |
| 3 |
* User interface for Resource Allocation in Resource Director Technology(RDT) |
| 4 |
* |
| 5 |
* Copyright (C) 2016 Intel Corporation |
| 6 |
* |
| 7 |
* Author: Fenghua Yu <[email protected]> |
| 8 |
* |
| 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 |
|
| 15 |
#include <linux/cpu.h> |
| 16 |
#include <linux/debugfs.h> |
| 17 |
#include <linux/fs.h> |
| 18 |
#include <linux/fs_parser.h> |
| 19 |
#include <linux/sysfs.h> |
| 20 |
#include <linux/kernfs.h> |
| 21 |
#include <linux/once.h> |
| 22 |
#include <linux/resctrl.h> |
| 23 |
#include <linux/seq_buf.h> |
| 24 |
#include <linux/seq_file.h> |
| 25 |
#include <linux/sched/task.h> |
| 26 |
#include <linux/slab.h> |
| 27 |
#include <linux/user_namespace.h> |
| 28 |
|
| 29 |
#include <uapi/linux/magic.h> |
| 30 |
|
| 31 |
#include "internal.h" |
| 32 |
|
| 33 |
/* Mutex to protect rdtgroup access. */ |
| 34 |
DEFINE_MUTEX(rdtgroup_mutex); |
| 35 |
|
| 36 |
static struct kernfs_root *rdt_root; |
| 37 |
|
| 38 |
struct rdtgroup rdtgroup_default; |
| 39 |
|
| 40 |
LIST_HEAD(rdt_all_groups); |
| 41 |
|
| 42 |
/* list of entries for the schemata file */ |
| 43 |
LIST_HEAD(resctrl_schema_all); |
| 44 |
|
| 45 |
/* |
| 46 |
* List of struct mon_data containing private data of event files for use by |
| 47 |
* rdtgroup_mondata_show(). Protected by rdtgroup_mutex. |
| 48 |
*/ |
| 49 |
static LIST_HEAD(mon_data_kn_priv_list); |
| 50 |
|
| 51 |
/* The filesystem can only be mounted once. */ |
| 52 |
bool resctrl_mounted; |
| 53 |
|
| 54 |
/* Kernel fs node for "info" directory under root */ |
| 55 |
static struct kernfs_node *kn_info; |
| 56 |
|
| 57 |
/* Kernel fs node for "mon_groups" directory under root */ |
| 58 |
static struct kernfs_node *kn_mongrp; |
| 59 |
|
| 60 |
/* Kernel fs node for "mon_data" directory under root */ |
| 61 |
static struct kernfs_node *kn_mondata; |
| 62 |
|
| 63 |
/* |
| 64 |
* Used to store the max resource name width to display the schemata names in |
| 65 |
* a tabular format. |
| 66 |
*/ |
| 67 |
int max_name_width; |
| 68 |
|
| 69 |
static struct seq_buf last_cmd_status; |
| 70 |
|
| 71 |
static char last_cmd_status_buf[512]; |
| 72 |
|
| 73 |
static int rdtgroup_setup_root(struct rdt_fs_context *ctx); |
| 74 |
|
| 75 |
static void rdtgroup_destroy_root(void); |
| 76 |
|
| 77 |
struct dentry *debugfs_resctrl; |
| 78 |
|
| 79 |
/* |
| 80 |
* Memory bandwidth monitoring event to use for the default CTRL_MON group |
| 81 |
* and each new CTRL_MON group created by the user. Only relevant when |
| 82 |
* the filesystem is mounted with the "mba_MBps" option so it does not |
| 83 |
* matter that it remains uninitialized on systems that do not support |
| 84 |
* the "mba_MBps" option. |
| 85 |
*/ |
| 86 |
enum resctrl_event_id mba_mbps_default_event; |
| 87 |
|
| 88 |
static bool resctrl_debug; |
| 89 |
|
| 90 |
void rdt_last_cmd_clear(void) |
| 91 |
{ |
| 92 |
lockdep_assert_held(&rdtgroup_mutex); |
| 93 |
seq_buf_clear(&last_cmd_status); |
| 94 |
} |
| 95 |
|
| 96 |
void rdt_last_cmd_puts(const char *s) |
| 97 |
{ |
| 98 |
lockdep_assert_held(&rdtgroup_mutex); |
| 99 |
seq_buf_puts(&last_cmd_status, s); |
| 100 |
} |
| 101 |
|
| 102 |
void rdt_last_cmd_printf(const char *fmt, ...) |
| 103 |
{ |
| 104 |
va_list ap; |
| 105 |
|
| 106 |
va_start(ap, fmt); |
| 107 |
lockdep_assert_held(&rdtgroup_mutex); |
| 108 |
seq_buf_vprintf(&last_cmd_status, fmt, ap); |
| 109 |
va_end(ap); |
| 110 |
} |
| 111 |
|
| 112 |
void rdt_staged_configs_clear(void) |
| 113 |
{ |
| 114 |
struct rdt_ctrl_domain *dom; |
| 115 |
struct rdt_resource *r; |
| 116 |
|
| 117 |
lockdep_assert_held(&rdtgroup_mutex); |
| 118 |
|
| 119 |
for_each_alloc_capable_rdt_resource(r) { |
| 120 |
list_for_each_entry(dom, &r->ctrl_domains, hdr.list) |
| 121 |
memset(dom->staged_config, 0, sizeof(dom->staged_config)); |
| 122 |
} |
| 123 |
} |
| 124 |
|
| 125 |
static bool resctrl_is_mbm_enabled(void) |
| 126 |
{ |
| 127 |
return (resctrl_is_mon_event_enabled(QOS_L3_MBM_TOTAL_EVENT_ID) || |
| 128 |
resctrl_is_mon_event_enabled(QOS_L3_MBM_LOCAL_EVENT_ID)); |
| 129 |
} |
| 130 |
|
| 131 |
/* |
| 132 |
* Trivial allocator for CLOSIDs. Use BITMAP APIs to manipulate a bitmap |
| 133 |
* of free CLOSIDs. |
| 134 |
* |
| 135 |
* Using a global CLOSID across all resources has some advantages and |
| 136 |
* some drawbacks: |
| 137 |
* + We can simply set current's closid to assign a task to a resource |
| 138 |
* group. |
| 139 |
* + Context switch code can avoid extra memory references deciding which |
| 140 |
* CLOSID to load into the PQR_ASSOC MSR |
| 141 |
* - We give up some options in configuring resource groups across multi-socket |
| 142 |
* systems. |
| 143 |
* - Our choices on how to configure each resource become progressively more |
| 144 |
* limited as the number of resources grows. |
| 145 |
*/ |
| 146 |
static unsigned long *closid_free_map; |
| 147 |
|
| 148 |
static int closid_free_map_len; |
| 149 |
|
| 150 |
int closids_supported(void) |
| 151 |
{ |
| 152 |
return closid_free_map_len; |
| 153 |
} |
| 154 |
|
| 155 |
static int closid_init(void) |
| 156 |
{ |
| 157 |
struct resctrl_schema *s; |
| 158 |
u32 rdt_min_closid = ~0; |
| 159 |
|
| 160 |
/* Monitor only platforms still call closid_init() */ |
| 161 |
if (list_empty(&resctrl_schema_all)) |
| 162 |
return 0; |
| 163 |
|
| 164 |
/* Compute rdt_min_closid across all resources */ |
| 165 |
list_for_each_entry(s, &resctrl_schema_all, list) |
| 166 |
rdt_min_closid = min(rdt_min_closid, s->num_closid); |
| 167 |
|
| 168 |
closid_free_map = bitmap_alloc(rdt_min_closid, GFP_KERNEL); |
| 169 |
if (!closid_free_map) |
| 170 |
return -ENOMEM; |
| 171 |
bitmap_fill(closid_free_map, rdt_min_closid); |
| 172 |
|
| 173 |
/* RESCTRL_RESERVED_CLOSID is always reserved for the default group */ |
| 174 |
__clear_bit(RESCTRL_RESERVED_CLOSID, closid_free_map); |
| 175 |
closid_free_map_len = rdt_min_closid; |
| 176 |
|
| 177 |
return 0; |
| 178 |
} |
| 179 |
|
| 180 |
static void closid_exit(void) |
| 181 |
{ |
| 182 |
bitmap_free(closid_free_map); |
| 183 |
closid_free_map = NULL; |
| 184 |
} |
| 185 |
|
| 186 |
static int closid_alloc(void) |
| 187 |
{ |
| 188 |
int cleanest_closid; |
| 189 |
u32 closid; |
| 190 |
|
| 191 |
lockdep_assert_held(&rdtgroup_mutex); |
| 192 |
|
| 193 |
if (IS_ENABLED(CONFIG_RESCTRL_RMID_DEPENDS_ON_CLOSID) && |
| 194 |
resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID)) { |
| 195 |
cleanest_closid = resctrl_find_cleanest_closid(); |
| 196 |
if (cleanest_closid < 0) |
| 197 |
return cleanest_closid; |
| 198 |
closid = cleanest_closid; |
| 199 |
} else { |
| 200 |
closid = find_first_bit(closid_free_map, closid_free_map_len); |
| 201 |
if (closid == closid_free_map_len) |
| 202 |
return -ENOSPC; |
| 203 |
} |
| 204 |
__clear_bit(closid, closid_free_map); |
| 205 |
|
| 206 |
return closid; |
| 207 |
} |
| 208 |
|
| 209 |
void closid_free(int closid) |
| 210 |
{ |
| 211 |
lockdep_assert_held(&rdtgroup_mutex); |
| 212 |
|
| 213 |
__set_bit(closid, closid_free_map); |
| 214 |
} |
| 215 |
|
| 216 |
/** |
| 217 |
* closid_allocated - test if provided closid is in use |
| 218 |
* @closid: closid to be tested |
| 219 |
* |
| 220 |
* Return: true if @closid is currently associated with a resource group, |
| 221 |
* false if @closid is free |
| 222 |
*/ |
| 223 |
bool closid_allocated(unsigned int closid) |
| 224 |
{ |
| 225 |
lockdep_assert_held(&rdtgroup_mutex); |
| 226 |
|
| 227 |
return !test_bit(closid, closid_free_map); |
| 228 |
} |
| 229 |
|
| 230 |
bool closid_alloc_fixed(u32 closid) |
| 231 |
{ |
| 232 |
return __test_and_clear_bit(closid, closid_free_map); |
| 233 |
} |
| 234 |
|
| 235 |
/** |
| 236 |
* rdtgroup_mode_by_closid - Return mode of resource group with closid |
| 237 |
* @closid: closid if the resource group |
| 238 |
* |
| 239 |
* Each resource group is associated with a @closid. Here the mode |
| 240 |
* of a resource group can be queried by searching for it using its closid. |
| 241 |
* |
| 242 |
* Return: mode as &enum rdtgrp_mode of resource group with closid @closid |
| 243 |
*/ |
| 244 |
enum rdtgrp_mode rdtgroup_mode_by_closid(int closid) |
| 245 |
{ |
| 246 |
struct rdtgroup *rdtgrp; |
| 247 |
|
| 248 |
list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) { |
| 249 |
if (rdtgrp->closid == closid) |
| 250 |
return rdtgrp->mode; |
| 251 |
} |
| 252 |
|
| 253 |
return RDT_NUM_MODES; |
| 254 |
} |
| 255 |
|
| 256 |
static const char * const rdt_mode_str[] = { |
| 257 |
[RDT_MODE_SHAREABLE] = "shareable", |
| 258 |
[RDT_MODE_EXCLUSIVE] = "exclusive", |
| 259 |
[RDT_MODE_PSEUDO_LOCKSETUP] = "pseudo-locksetup", |
| 260 |
[RDT_MODE_PSEUDO_LOCKED] = "pseudo-locked", |
| 261 |
}; |
| 262 |
|
| 263 |
/** |
| 264 |
* rdtgroup_mode_str - Return the string representation of mode |
| 265 |
* @mode: the resource group mode as &enum rdtgroup_mode |
| 266 |
* |
| 267 |
* Return: string representation of valid mode, "unknown" otherwise |
| 268 |
*/ |
| 269 |
static const char *rdtgroup_mode_str(enum rdtgrp_mode mode) |
| 270 |
{ |
| 271 |
if (mode < RDT_MODE_SHAREABLE || mode >= RDT_NUM_MODES) |
| 272 |
return "unknown"; |
| 273 |
|
| 274 |
return rdt_mode_str[mode]; |
| 275 |
} |
| 276 |
|
| 277 |
/* set uid and gid of rdtgroup dirs and files to that of the creator */ |
| 278 |
static int rdtgroup_kn_set_ugid(struct kernfs_node *kn) |
| 279 |
{ |
| 280 |
struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, |
| 281 |
.ia_uid = current_fsuid(), |
| 282 |
.ia_gid = current_fsgid(), }; |
| 283 |
|
| 284 |
if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && |
| 285 |
gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) |
| 286 |
return 0; |
| 287 |
|
| 288 |
return kernfs_setattr(kn, &iattr); |
| 289 |
} |
| 290 |
|
| 291 |
static int rdtgroup_add_file(struct kernfs_node *parent_kn, struct rftype *rft) |
| 292 |
{ |
| 293 |
struct kernfs_node *kn; |
| 294 |
int ret; |
| 295 |
|
| 296 |
kn = __kernfs_create_file(parent_kn, rft->name, rft->mode, |
| 297 |
GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, |
| 298 |
0, rft->kf_ops, rft, NULL, NULL); |
| 299 |
if (IS_ERR(kn)) |
| 300 |
return PTR_ERR(kn); |
| 301 |
|
| 302 |
ret = rdtgroup_kn_set_ugid(kn); |
| 303 |
if (ret) { |
| 304 |
kernfs_remove(kn); |
| 305 |
return ret; |
| 306 |
} |
| 307 |
|
| 308 |
return 0; |
| 309 |
} |
| 310 |
|
| 311 |
static int rdtgroup_seqfile_show(struct seq_file *m, void *arg) |
| 312 |
{ |
| 313 |
struct kernfs_open_file *of = m->private; |
| 314 |
struct rftype *rft = of->kn->priv; |
| 315 |
|
| 316 |
if (rft->seq_show) |
| 317 |
return rft->seq_show(of, m, arg); |
| 318 |
return 0; |
| 319 |
} |
| 320 |
|
| 321 |
static ssize_t rdtgroup_file_write(struct kernfs_open_file *of, char *buf, |
| 322 |
size_t nbytes, loff_t off) |
| 323 |
{ |
| 324 |
struct rftype *rft = of->kn->priv; |
| 325 |
|
| 326 |
if (rft->write) |
| 327 |
return rft->write(of, buf, nbytes, off); |
| 328 |
|
| 329 |
return -EINVAL; |
| 330 |
} |
| 331 |
|
| 332 |
static const struct kernfs_ops rdtgroup_kf_single_ops = { |
| 333 |
.atomic_write_len = PAGE_SIZE, |
| 334 |
.write = rdtgroup_file_write, |
| 335 |
.seq_show = rdtgroup_seqfile_show, |
| 336 |
}; |
| 337 |
|
| 338 |
static const struct kernfs_ops kf_mondata_ops = { |
| 339 |
.atomic_write_len = PAGE_SIZE, |
| 340 |
.seq_show = rdtgroup_mondata_show, |
| 341 |
}; |
| 342 |
|
| 343 |
static bool is_cpu_list(struct kernfs_open_file *of) |
| 344 |
{ |
| 345 |
struct rftype *rft = of->kn->priv; |
| 346 |
|
| 347 |
return rft->flags & RFTYPE_FLAGS_CPUS_LIST; |
| 348 |
} |
| 349 |
|
| 350 |
static int rdtgroup_cpus_show(struct kernfs_open_file *of, |
| 351 |
struct seq_file *s, void *v) |
| 352 |
{ |
| 353 |
struct rdtgroup *rdtgrp; |
| 354 |
struct cpumask *mask; |
| 355 |
int ret = 0; |
| 356 |
|
| 357 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 358 |
|
| 359 |
if (rdtgrp) { |
| 360 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { |
| 361 |
if (!rdtgrp->plr->d) { |
| 362 |
rdt_last_cmd_clear(); |
| 363 |
rdt_last_cmd_puts("Cache domain offline\n"); |
| 364 |
ret = -ENODEV; |
| 365 |
} else { |
| 366 |
mask = &rdtgrp->plr->d->hdr.cpu_mask; |
| 367 |
seq_printf(s, is_cpu_list(of) ? |
| 368 |
"%*pbl\n" : "%*pb\n", |
| 369 |
cpumask_pr_args(mask)); |
| 370 |
} |
| 371 |
} else { |
| 372 |
seq_printf(s, is_cpu_list(of) ? "%*pbl\n" : "%*pb\n", |
| 373 |
cpumask_pr_args(&rdtgrp->cpu_mask)); |
| 374 |
} |
| 375 |
} else { |
| 376 |
ret = -ENOENT; |
| 377 |
} |
| 378 |
rdtgroup_kn_unlock(of->kn); |
| 379 |
|
| 380 |
return ret; |
| 381 |
} |
| 382 |
|
| 383 |
/* |
| 384 |
* Update the PGR_ASSOC MSR on all cpus in @cpu_mask, |
| 385 |
* |
| 386 |
* Per task closids/rmids must have been set up before calling this function. |
| 387 |
* @r may be NULL. |
| 388 |
*/ |
| 389 |
static void |
| 390 |
update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r) |
| 391 |
{ |
| 392 |
struct resctrl_cpu_defaults defaults, *p = NULL; |
| 393 |
|
| 394 |
if (r) { |
| 395 |
defaults.closid = r->closid; |
| 396 |
defaults.rmid = r->mon.rmid; |
| 397 |
p = &defaults; |
| 398 |
} |
| 399 |
|
| 400 |
on_each_cpu_mask(cpu_mask, resctrl_arch_sync_cpu_closid_rmid, p, 1); |
| 401 |
} |
| 402 |
|
| 403 |
static int cpus_mon_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, |
| 404 |
cpumask_var_t tmpmask) |
| 405 |
{ |
| 406 |
struct rdtgroup *prgrp = rdtgrp->mon.parent, *crgrp; |
| 407 |
struct list_head *head; |
| 408 |
|
| 409 |
/* Check whether cpus belong to parent ctrl group */ |
| 410 |
cpumask_andnot(tmpmask, newmask, &prgrp->cpu_mask); |
| 411 |
if (!cpumask_empty(tmpmask)) { |
| 412 |
rdt_last_cmd_puts("Can only add CPUs to mongroup that belong to parent\n"); |
| 413 |
return -EINVAL; |
| 414 |
} |
| 415 |
|
| 416 |
/* Check whether cpus are dropped from this group */ |
| 417 |
cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); |
| 418 |
if (!cpumask_empty(tmpmask)) { |
| 419 |
/* Give any dropped cpus to parent rdtgroup */ |
| 420 |
cpumask_or(&prgrp->cpu_mask, &prgrp->cpu_mask, tmpmask); |
| 421 |
update_closid_rmid(tmpmask, prgrp); |
| 422 |
} |
| 423 |
|
| 424 |
/* |
| 425 |
* If we added cpus, remove them from previous group that owned them |
| 426 |
* and update per-cpu rmid |
| 427 |
*/ |
| 428 |
cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); |
| 429 |
if (!cpumask_empty(tmpmask)) { |
| 430 |
head = &prgrp->mon.crdtgrp_list; |
| 431 |
list_for_each_entry(crgrp, head, mon.crdtgrp_list) { |
| 432 |
if (crgrp == rdtgrp) |
| 433 |
continue; |
| 434 |
cpumask_andnot(&crgrp->cpu_mask, &crgrp->cpu_mask, |
| 435 |
tmpmask); |
| 436 |
} |
| 437 |
update_closid_rmid(tmpmask, rdtgrp); |
| 438 |
} |
| 439 |
|
| 440 |
/* Done pushing/pulling - update this group with new mask */ |
| 441 |
cpumask_copy(&rdtgrp->cpu_mask, newmask); |
| 442 |
|
| 443 |
return 0; |
| 444 |
} |
| 445 |
|
| 446 |
static void cpumask_rdtgrp_clear(struct rdtgroup *r, struct cpumask *m) |
| 447 |
{ |
| 448 |
struct rdtgroup *crgrp; |
| 449 |
|
| 450 |
cpumask_andnot(&r->cpu_mask, &r->cpu_mask, m); |
| 451 |
/* update the child mon group masks as well*/ |
| 452 |
list_for_each_entry(crgrp, &r->mon.crdtgrp_list, mon.crdtgrp_list) |
| 453 |
cpumask_and(&crgrp->cpu_mask, &r->cpu_mask, &crgrp->cpu_mask); |
| 454 |
} |
| 455 |
|
| 456 |
static int cpus_ctrl_write(struct rdtgroup *rdtgrp, cpumask_var_t newmask, |
| 457 |
cpumask_var_t tmpmask, cpumask_var_t tmpmask1) |
| 458 |
{ |
| 459 |
struct rdtgroup *r, *crgrp; |
| 460 |
struct list_head *head; |
| 461 |
|
| 462 |
/* Check whether cpus are dropped from this group */ |
| 463 |
cpumask_andnot(tmpmask, &rdtgrp->cpu_mask, newmask); |
| 464 |
if (!cpumask_empty(tmpmask)) { |
| 465 |
/* Can't drop from default group */ |
| 466 |
if (rdtgrp == &rdtgroup_default) { |
| 467 |
rdt_last_cmd_puts("Can't drop CPUs from default group\n"); |
| 468 |
return -EINVAL; |
| 469 |
} |
| 470 |
|
| 471 |
/* Give any dropped cpus to rdtgroup_default */ |
| 472 |
cpumask_or(&rdtgroup_default.cpu_mask, |
| 473 |
&rdtgroup_default.cpu_mask, tmpmask); |
| 474 |
update_closid_rmid(tmpmask, &rdtgroup_default); |
| 475 |
} |
| 476 |
|
| 477 |
/* |
| 478 |
* If we added cpus, remove them from previous group and |
| 479 |
* the prev group's child groups that owned them |
| 480 |
* and update per-cpu closid/rmid. |
| 481 |
*/ |
| 482 |
cpumask_andnot(tmpmask, newmask, &rdtgrp->cpu_mask); |
| 483 |
if (!cpumask_empty(tmpmask)) { |
| 484 |
list_for_each_entry(r, &rdt_all_groups, rdtgroup_list) { |
| 485 |
if (r == rdtgrp) |
| 486 |
continue; |
| 487 |
cpumask_and(tmpmask1, &r->cpu_mask, tmpmask); |
| 488 |
if (!cpumask_empty(tmpmask1)) |
| 489 |
cpumask_rdtgrp_clear(r, tmpmask1); |
| 490 |
} |
| 491 |
update_closid_rmid(tmpmask, rdtgrp); |
| 492 |
} |
| 493 |
|
| 494 |
/* Done pushing/pulling - update this group with new mask */ |
| 495 |
cpumask_copy(&rdtgrp->cpu_mask, newmask); |
| 496 |
|
| 497 |
/* |
| 498 |
* Clear child mon group masks since there is a new parent mask |
| 499 |
* now and update the rmid for the cpus the child lost. |
| 500 |
*/ |
| 501 |
head = &rdtgrp->mon.crdtgrp_list; |
| 502 |
list_for_each_entry(crgrp, head, mon.crdtgrp_list) { |
| 503 |
cpumask_and(tmpmask, &rdtgrp->cpu_mask, &crgrp->cpu_mask); |
| 504 |
update_closid_rmid(tmpmask, rdtgrp); |
| 505 |
cpumask_clear(&crgrp->cpu_mask); |
| 506 |
} |
| 507 |
|
| 508 |
return 0; |
| 509 |
} |
| 510 |
|
| 511 |
static ssize_t rdtgroup_cpus_write(struct kernfs_open_file *of, |
| 512 |
char *buf, size_t nbytes, loff_t off) |
| 513 |
{ |
| 514 |
cpumask_var_t tmpmask, newmask, tmpmask1; |
| 515 |
struct rdtgroup *rdtgrp; |
| 516 |
int ret; |
| 517 |
|
| 518 |
if (!buf) |
| 519 |
return -EINVAL; |
| 520 |
|
| 521 |
if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) |
| 522 |
return -ENOMEM; |
| 523 |
if (!zalloc_cpumask_var(&newmask, GFP_KERNEL)) { |
| 524 |
free_cpumask_var(tmpmask); |
| 525 |
return -ENOMEM; |
| 526 |
} |
| 527 |
if (!zalloc_cpumask_var(&tmpmask1, GFP_KERNEL)) { |
| 528 |
free_cpumask_var(tmpmask); |
| 529 |
free_cpumask_var(newmask); |
| 530 |
return -ENOMEM; |
| 531 |
} |
| 532 |
|
| 533 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 534 |
if (!rdtgrp) { |
| 535 |
ret = -ENOENT; |
| 536 |
goto unlock; |
| 537 |
} |
| 538 |
|
| 539 |
rdt_last_cmd_clear(); |
| 540 |
|
| 541 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || |
| 542 |
rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
| 543 |
ret = -EINVAL; |
| 544 |
rdt_last_cmd_puts("Pseudo-locking in progress\n"); |
| 545 |
goto unlock; |
| 546 |
} |
| 547 |
|
| 548 |
if (is_cpu_list(of)) |
| 549 |
ret = cpulist_parse(buf, newmask); |
| 550 |
else |
| 551 |
ret = cpumask_parse(buf, newmask); |
| 552 |
|
| 553 |
if (ret) { |
| 554 |
rdt_last_cmd_puts("Bad CPU list/mask\n"); |
| 555 |
goto unlock; |
| 556 |
} |
| 557 |
|
| 558 |
/* check that user didn't specify any offline cpus */ |
| 559 |
cpumask_andnot(tmpmask, newmask, cpu_online_mask); |
| 560 |
if (!cpumask_empty(tmpmask)) { |
| 561 |
ret = -EINVAL; |
| 562 |
rdt_last_cmd_puts("Can only assign online CPUs\n"); |
| 563 |
goto unlock; |
| 564 |
} |
| 565 |
|
| 566 |
if (rdtgrp->type == RDTCTRL_GROUP) |
| 567 |
ret = cpus_ctrl_write(rdtgrp, newmask, tmpmask, tmpmask1); |
| 568 |
else if (rdtgrp->type == RDTMON_GROUP) |
| 569 |
ret = cpus_mon_write(rdtgrp, newmask, tmpmask); |
| 570 |
else |
| 571 |
ret = -EINVAL; |
| 572 |
|
| 573 |
unlock: |
| 574 |
rdtgroup_kn_unlock(of->kn); |
| 575 |
free_cpumask_var(tmpmask); |
| 576 |
free_cpumask_var(newmask); |
| 577 |
free_cpumask_var(tmpmask1); |
| 578 |
|
| 579 |
return ret ?: nbytes; |
| 580 |
} |
| 581 |
|
| 582 |
/** |
| 583 |
* rdtgroup_remove - the helper to remove resource group safely |
| 584 |
* @rdtgrp: resource group to remove |
| 585 |
* |
| 586 |
* On resource group creation via a mkdir, an extra kernfs_node reference is |
| 587 |
* taken to ensure that the rdtgroup structure remains accessible for the |
| 588 |
* rdtgroup_kn_unlock() calls where it is removed. |
| 589 |
* |
| 590 |
* Drop the extra reference here, then free the rdtgroup structure. |
| 591 |
* |
| 592 |
* Return: void |
| 593 |
*/ |
| 594 |
static void rdtgroup_remove(struct rdtgroup *rdtgrp) |
| 595 |
{ |
| 596 |
kernfs_put(rdtgrp->kn); |
| 597 |
kfree(rdtgrp); |
| 598 |
} |
| 599 |
|
| 600 |
static void _update_task_closid_rmid(void *task) |
| 601 |
{ |
| 602 |
/* |
| 603 |
* If the task is still current on this CPU, update PQR_ASSOC MSR. |
| 604 |
* Otherwise, the MSR is updated when the task is scheduled in. |
| 605 |
*/ |
| 606 |
if (task == current) |
| 607 |
resctrl_arch_sched_in(task); |
| 608 |
} |
| 609 |
|
| 610 |
static void update_task_closid_rmid(struct task_struct *t) |
| 611 |
{ |
| 612 |
if (IS_ENABLED(CONFIG_SMP) && task_curr(t)) |
| 613 |
smp_call_function_single(task_cpu(t), _update_task_closid_rmid, t, 1); |
| 614 |
else |
| 615 |
_update_task_closid_rmid(t); |
| 616 |
} |
| 617 |
|
| 618 |
static bool task_in_rdtgroup(struct task_struct *tsk, struct rdtgroup *rdtgrp) |
| 619 |
{ |
| 620 |
u32 closid, rmid = rdtgrp->mon.rmid; |
| 621 |
|
| 622 |
if (rdtgrp->type == RDTCTRL_GROUP) |
| 623 |
closid = rdtgrp->closid; |
| 624 |
else if (rdtgrp->type == RDTMON_GROUP) |
| 625 |
closid = rdtgrp->mon.parent->closid; |
| 626 |
else |
| 627 |
return false; |
| 628 |
|
| 629 |
return resctrl_arch_match_closid(tsk, closid) && |
| 630 |
resctrl_arch_match_rmid(tsk, closid, rmid); |
| 631 |
} |
| 632 |
|
| 633 |
static int __rdtgroup_move_task(struct task_struct *tsk, |
| 634 |
struct rdtgroup *rdtgrp) |
| 635 |
{ |
| 636 |
/* If the task is already in rdtgrp, no need to move the task. */ |
| 637 |
if (task_in_rdtgroup(tsk, rdtgrp)) |
| 638 |
return 0; |
| 639 |
|
| 640 |
/* |
| 641 |
* Set the task's closid/rmid before the PQR_ASSOC MSR can be |
| 642 |
* updated by them. |
| 643 |
* |
| 644 |
* For ctrl_mon groups, move both closid and rmid. |
| 645 |
* For monitor groups, can move the tasks only from |
| 646 |
* their parent CTRL group. |
| 647 |
*/ |
| 648 |
if (rdtgrp->type == RDTMON_GROUP && |
| 649 |
!resctrl_arch_match_closid(tsk, rdtgrp->mon.parent->closid)) { |
| 650 |
rdt_last_cmd_puts("Can't move task to different control group\n"); |
| 651 |
return -EINVAL; |
| 652 |
} |
| 653 |
|
| 654 |
if (rdtgrp->type == RDTMON_GROUP) |
| 655 |
resctrl_arch_set_closid_rmid(tsk, rdtgrp->mon.parent->closid, |
| 656 |
rdtgrp->mon.rmid); |
| 657 |
else |
| 658 |
resctrl_arch_set_closid_rmid(tsk, rdtgrp->closid, |
| 659 |
rdtgrp->mon.rmid); |
| 660 |
|
| 661 |
/* |
| 662 |
* Ensure the task's closid and rmid are written before determining if |
| 663 |
* the task is current that will decide if it will be interrupted. |
| 664 |
* This pairs with the full barrier between the rq->curr update and |
| 665 |
* resctrl_arch_sched_in() during context switch. |
| 666 |
*/ |
| 667 |
smp_mb(); |
| 668 |
|
| 669 |
/* |
| 670 |
* By now, the task's closid and rmid are set. If the task is current |
| 671 |
* on a CPU, the PQR_ASSOC MSR needs to be updated to make the resource |
| 672 |
* group go into effect. If the task is not current, the MSR will be |
| 673 |
* updated when the task is scheduled in. |
| 674 |
*/ |
| 675 |
update_task_closid_rmid(tsk); |
| 676 |
|
| 677 |
return 0; |
| 678 |
} |
| 679 |
|
| 680 |
static bool is_closid_match(struct task_struct *t, struct rdtgroup *r) |
| 681 |
{ |
| 682 |
return (resctrl_arch_alloc_capable() && (r->type == RDTCTRL_GROUP) && |
| 683 |
resctrl_arch_match_closid(t, r->closid)); |
| 684 |
} |
| 685 |
|
| 686 |
static bool is_rmid_match(struct task_struct *t, struct rdtgroup *r) |
| 687 |
{ |
| 688 |
return (resctrl_arch_mon_capable() && (r->type == RDTMON_GROUP) && |
| 689 |
resctrl_arch_match_rmid(t, r->mon.parent->closid, |
| 690 |
r->mon.rmid)); |
| 691 |
} |
| 692 |
|
| 693 |
/** |
| 694 |
* rdtgroup_tasks_assigned - Test if tasks have been assigned to resource group |
| 695 |
* @r: Resource group |
| 696 |
* |
| 697 |
* Return: 1 if tasks have been assigned to @r, 0 otherwise |
| 698 |
*/ |
| 699 |
int rdtgroup_tasks_assigned(struct rdtgroup *r) |
| 700 |
{ |
| 701 |
struct task_struct *p, *t; |
| 702 |
int ret = 0; |
| 703 |
|
| 704 |
lockdep_assert_held(&rdtgroup_mutex); |
| 705 |
|
| 706 |
rcu_read_lock(); |
| 707 |
for_each_process_thread(p, t) { |
| 708 |
if (is_closid_match(t, r) || is_rmid_match(t, r)) { |
| 709 |
ret = 1; |
| 710 |
break; |
| 711 |
} |
| 712 |
} |
| 713 |
rcu_read_unlock(); |
| 714 |
|
| 715 |
return ret; |
| 716 |
} |
| 717 |
|
| 718 |
static int rdtgroup_task_write_permission(struct task_struct *task, |
| 719 |
struct kernfs_open_file *of) |
| 720 |
{ |
| 721 |
const struct cred *tcred = get_task_cred(task); |
| 722 |
const struct cred *cred = current_cred(); |
| 723 |
int ret = 0; |
| 724 |
|
| 725 |
/* |
| 726 |
* Even if we're attaching all tasks in the thread group, we only |
| 727 |
* need to check permissions on one of them. |
| 728 |
*/ |
| 729 |
if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
| 730 |
!uid_eq(cred->euid, tcred->uid) && |
| 731 |
!uid_eq(cred->euid, tcred->suid)) { |
| 732 |
rdt_last_cmd_printf("No permission to move task %d\n", task->pid); |
| 733 |
ret = -EPERM; |
| 734 |
} |
| 735 |
|
| 736 |
put_cred(tcred); |
| 737 |
return ret; |
| 738 |
} |
| 739 |
|
| 740 |
static int rdtgroup_move_task(pid_t pid, struct rdtgroup *rdtgrp, |
| 741 |
struct kernfs_open_file *of) |
| 742 |
{ |
| 743 |
struct task_struct *tsk; |
| 744 |
int ret; |
| 745 |
|
| 746 |
rcu_read_lock(); |
| 747 |
if (pid) { |
| 748 |
tsk = find_task_by_vpid(pid); |
| 749 |
if (!tsk) { |
| 750 |
rcu_read_unlock(); |
| 751 |
rdt_last_cmd_printf("No task %d\n", pid); |
| 752 |
return -ESRCH; |
| 753 |
} |
| 754 |
} else { |
| 755 |
tsk = current; |
| 756 |
} |
| 757 |
|
| 758 |
get_task_struct(tsk); |
| 759 |
rcu_read_unlock(); |
| 760 |
|
| 761 |
ret = rdtgroup_task_write_permission(tsk, of); |
| 762 |
if (!ret) |
| 763 |
ret = __rdtgroup_move_task(tsk, rdtgrp); |
| 764 |
|
| 765 |
put_task_struct(tsk); |
| 766 |
return ret; |
| 767 |
} |
| 768 |
|
| 769 |
static ssize_t rdtgroup_tasks_write(struct kernfs_open_file *of, |
| 770 |
char *buf, size_t nbytes, loff_t off) |
| 771 |
{ |
| 772 |
struct rdtgroup *rdtgrp; |
| 773 |
char *pid_str; |
| 774 |
int ret = 0; |
| 775 |
pid_t pid; |
| 776 |
|
| 777 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 778 |
if (!rdtgrp) { |
| 779 |
rdtgroup_kn_unlock(of->kn); |
| 780 |
return -ENOENT; |
| 781 |
} |
| 782 |
rdt_last_cmd_clear(); |
| 783 |
|
| 784 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED || |
| 785 |
rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
| 786 |
ret = -EINVAL; |
| 787 |
rdt_last_cmd_puts("Pseudo-locking in progress\n"); |
| 788 |
goto unlock; |
| 789 |
} |
| 790 |
|
| 791 |
while (buf && buf[0] != '\0' && buf[0] != '\n') { |
| 792 |
pid_str = strim(strsep(&buf, ",")); |
| 793 |
|
| 794 |
if (kstrtoint(pid_str, 0, &pid)) { |
| 795 |
rdt_last_cmd_printf("Task list parsing error pid %s\n", pid_str); |
| 796 |
ret = -EINVAL; |
| 797 |
break; |
| 798 |
} |
| 799 |
|
| 800 |
if (pid < 0) { |
| 801 |
rdt_last_cmd_printf("Invalid pid %d\n", pid); |
| 802 |
ret = -EINVAL; |
| 803 |
break; |
| 804 |
} |
| 805 |
|
| 806 |
ret = rdtgroup_move_task(pid, rdtgrp, of); |
| 807 |
if (ret) { |
| 808 |
rdt_last_cmd_printf("Error while processing task %d\n", pid); |
| 809 |
break; |
| 810 |
} |
| 811 |
} |
| 812 |
|
| 813 |
unlock: |
| 814 |
rdtgroup_kn_unlock(of->kn); |
| 815 |
|
| 816 |
return ret ?: nbytes; |
| 817 |
} |
| 818 |
|
| 819 |
static void show_rdt_tasks(struct rdtgroup *r, struct seq_file *s) |
| 820 |
{ |
| 821 |
struct task_struct *p, *t; |
| 822 |
pid_t pid; |
| 823 |
|
| 824 |
rcu_read_lock(); |
| 825 |
for_each_process_thread(p, t) { |
| 826 |
if (is_closid_match(t, r) || is_rmid_match(t, r)) { |
| 827 |
pid = task_pid_vnr(t); |
| 828 |
if (pid) |
| 829 |
seq_printf(s, "%d\n", pid); |
| 830 |
} |
| 831 |
} |
| 832 |
rcu_read_unlock(); |
| 833 |
} |
| 834 |
|
| 835 |
static int rdtgroup_tasks_show(struct kernfs_open_file *of, |
| 836 |
struct seq_file *s, void *v) |
| 837 |
{ |
| 838 |
struct rdtgroup *rdtgrp; |
| 839 |
int ret = 0; |
| 840 |
|
| 841 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 842 |
if (rdtgrp) |
| 843 |
show_rdt_tasks(rdtgrp, s); |
| 844 |
else |
| 845 |
ret = -ENOENT; |
| 846 |
rdtgroup_kn_unlock(of->kn); |
| 847 |
|
| 848 |
return ret; |
| 849 |
} |
| 850 |
|
| 851 |
static int rdtgroup_closid_show(struct kernfs_open_file *of, |
| 852 |
struct seq_file *s, void *v) |
| 853 |
{ |
| 854 |
struct rdtgroup *rdtgrp; |
| 855 |
int ret = 0; |
| 856 |
|
| 857 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 858 |
if (rdtgrp) |
| 859 |
seq_printf(s, "%u\n", rdtgrp->closid); |
| 860 |
else |
| 861 |
ret = -ENOENT; |
| 862 |
rdtgroup_kn_unlock(of->kn); |
| 863 |
|
| 864 |
return ret; |
| 865 |
} |
| 866 |
|
| 867 |
static int rdtgroup_rmid_show(struct kernfs_open_file *of, |
| 868 |
struct seq_file *s, void *v) |
| 869 |
{ |
| 870 |
struct rdtgroup *rdtgrp; |
| 871 |
int ret = 0; |
| 872 |
|
| 873 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 874 |
if (rdtgrp) |
| 875 |
seq_printf(s, "%u\n", rdtgrp->mon.rmid); |
| 876 |
else |
| 877 |
ret = -ENOENT; |
| 878 |
rdtgroup_kn_unlock(of->kn); |
| 879 |
|
| 880 |
return ret; |
| 881 |
} |
| 882 |
|
| 883 |
#ifdef CONFIG_PROC_CPU_RESCTRL |
| 884 |
/* |
| 885 |
* A task can only be part of one resctrl control group and of one monitor |
| 886 |
* group which is associated to that control group. |
| 887 |
* |
| 888 |
* 1) res: |
| 889 |
* mon: |
| 890 |
* |
| 891 |
* resctrl is not available. |
| 892 |
* |
| 893 |
* 2) res:/ |
| 894 |
* mon: |
| 895 |
* |
| 896 |
* Task is part of the root resctrl control group, and it is not associated |
| 897 |
* to any monitor group. |
| 898 |
* |
| 899 |
* 3) res:/ |
| 900 |
* mon:mon0 |
| 901 |
* |
| 902 |
* Task is part of the root resctrl control group and monitor group mon0. |
| 903 |
* |
| 904 |
* 4) res:group0 |
| 905 |
* mon: |
| 906 |
* |
| 907 |
* Task is part of resctrl control group group0, and it is not associated |
| 908 |
* to any monitor group. |
| 909 |
* |
| 910 |
* 5) res:group0 |
| 911 |
* mon:mon1 |
| 912 |
* |
| 913 |
* Task is part of resctrl control group group0 and monitor group mon1. |
| 914 |
*/ |
| 915 |
int proc_resctrl_show(struct seq_file *s, struct pid_namespace *ns, |
| 916 |
struct pid *pid, struct task_struct *tsk) |
| 917 |
{ |
| 918 |
struct rdtgroup *rdtg; |
| 919 |
int ret = 0; |
| 920 |
|
| 921 |
mutex_lock(&rdtgroup_mutex); |
| 922 |
|
| 923 |
/* Return empty if resctrl has not been mounted. */ |
| 924 |
if (!resctrl_mounted) { |
| 925 |
seq_puts(s, "res:\nmon:\n"); |
| 926 |
goto unlock; |
| 927 |
} |
| 928 |
|
| 929 |
list_for_each_entry(rdtg, &rdt_all_groups, rdtgroup_list) { |
| 930 |
struct rdtgroup *crg; |
| 931 |
|
| 932 |
/* |
| 933 |
* Task information is only relevant for shareable |
| 934 |
* and exclusive groups. |
| 935 |
*/ |
| 936 |
if (rdtg->mode != RDT_MODE_SHAREABLE && |
| 937 |
rdtg->mode != RDT_MODE_EXCLUSIVE) |
| 938 |
continue; |
| 939 |
|
| 940 |
if (!resctrl_arch_match_closid(tsk, rdtg->closid)) |
| 941 |
continue; |
| 942 |
|
| 943 |
seq_printf(s, "res:%s%s\n", (rdtg == &rdtgroup_default) ? "/" : "", |
| 944 |
rdt_kn_name(rdtg->kn)); |
| 945 |
seq_puts(s, "mon:"); |
| 946 |
list_for_each_entry(crg, &rdtg->mon.crdtgrp_list, |
| 947 |
mon.crdtgrp_list) { |
| 948 |
if (!resctrl_arch_match_rmid(tsk, crg->mon.parent->closid, |
| 949 |
crg->mon.rmid)) |
| 950 |
continue; |
| 951 |
seq_printf(s, "%s", rdt_kn_name(crg->kn)); |
| 952 |
break; |
| 953 |
} |
| 954 |
seq_putc(s, '\n'); |
| 955 |
goto unlock; |
| 956 |
} |
| 957 |
/* |
| 958 |
* The above search should succeed. Otherwise return |
| 959 |
* with an error. |
| 960 |
*/ |
| 961 |
ret = -ENOENT; |
| 962 |
unlock: |
| 963 |
mutex_unlock(&rdtgroup_mutex); |
| 964 |
|
| 965 |
return ret; |
| 966 |
} |
| 967 |
#endif |
| 968 |
|
| 969 |
static int rdt_last_cmd_status_show(struct kernfs_open_file *of, |
| 970 |
struct seq_file *seq, void *v) |
| 971 |
{ |
| 972 |
int len; |
| 973 |
|
| 974 |
mutex_lock(&rdtgroup_mutex); |
| 975 |
len = seq_buf_used(&last_cmd_status); |
| 976 |
if (len) |
| 977 |
seq_printf(seq, "%.*s", len, last_cmd_status_buf); |
| 978 |
else |
| 979 |
seq_puts(seq, "ok\n"); |
| 980 |
mutex_unlock(&rdtgroup_mutex); |
| 981 |
return 0; |
| 982 |
} |
| 983 |
|
| 984 |
void *rdt_kn_parent_priv(struct kernfs_node *kn) |
| 985 |
{ |
| 986 |
/* |
| 987 |
* The parent pointer is only valid within RCU section since it can be |
| 988 |
* replaced. |
| 989 |
*/ |
| 990 |
guard(rcu)(); |
| 991 |
return rcu_dereference(kn->__parent)->priv; |
| 992 |
} |
| 993 |
|
| 994 |
static int rdt_num_closids_show(struct kernfs_open_file *of, |
| 995 |
struct seq_file *seq, void *v) |
| 996 |
{ |
| 997 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 998 |
|
| 999 |
seq_printf(seq, "%u\n", s->num_closid); |
| 1000 |
return 0; |
| 1001 |
} |
| 1002 |
|
| 1003 |
static int rdt_default_ctrl_show(struct kernfs_open_file *of, |
| 1004 |
struct seq_file *seq, void *v) |
| 1005 |
{ |
| 1006 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1007 |
struct rdt_resource *r = s->res; |
| 1008 |
|
| 1009 |
seq_printf(seq, "%x\n", resctrl_get_default_ctrl(r)); |
| 1010 |
return 0; |
| 1011 |
} |
| 1012 |
|
| 1013 |
static int rdt_min_cbm_bits_show(struct kernfs_open_file *of, |
| 1014 |
struct seq_file *seq, void *v) |
| 1015 |
{ |
| 1016 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1017 |
struct rdt_resource *r = s->res; |
| 1018 |
|
| 1019 |
seq_printf(seq, "%u\n", r->cache.min_cbm_bits); |
| 1020 |
return 0; |
| 1021 |
} |
| 1022 |
|
| 1023 |
static int rdt_shareable_bits_show(struct kernfs_open_file *of, |
| 1024 |
struct seq_file *seq, void *v) |
| 1025 |
{ |
| 1026 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1027 |
struct rdt_resource *r = s->res; |
| 1028 |
|
| 1029 |
seq_printf(seq, "%x\n", r->cache.shareable_bits); |
| 1030 |
return 0; |
| 1031 |
} |
| 1032 |
|
| 1033 |
/* |
| 1034 |
* rdt_bit_usage_show - Display current usage of resources |
| 1035 |
* |
| 1036 |
* A domain is a shared resource that can now be allocated differently. Here |
| 1037 |
* we display the current regions of the domain as an annotated bitmask. |
| 1038 |
* For each domain of this resource its allocation bitmask |
| 1039 |
* is annotated as below to indicate the current usage of the corresponding bit: |
| 1040 |
* 0 - currently unused |
| 1041 |
* X - currently available for sharing and used by software and hardware |
| 1042 |
* H - currently used by hardware only but available for software use |
| 1043 |
* S - currently used and shareable by software only |
| 1044 |
* E - currently used exclusively by one resource group |
| 1045 |
* P - currently pseudo-locked by one resource group |
| 1046 |
*/ |
| 1047 |
static int rdt_bit_usage_show(struct kernfs_open_file *of, |
| 1048 |
struct seq_file *seq, void *v) |
| 1049 |
{ |
| 1050 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1051 |
/* |
| 1052 |
* Use unsigned long even though only 32 bits are used to ensure |
| 1053 |
* test_bit() is used safely. |
| 1054 |
*/ |
| 1055 |
unsigned long sw_shareable = 0, hw_shareable = 0; |
| 1056 |
unsigned long exclusive = 0, pseudo_locked = 0; |
| 1057 |
struct rdt_resource *r = s->res; |
| 1058 |
struct rdt_ctrl_domain *dom; |
| 1059 |
int i, hwb, swb, excl, psl; |
| 1060 |
enum rdtgrp_mode mode; |
| 1061 |
bool sep = false; |
| 1062 |
u32 ctrl_val; |
| 1063 |
|
| 1064 |
cpus_read_lock(); |
| 1065 |
mutex_lock(&rdtgroup_mutex); |
| 1066 |
list_for_each_entry(dom, &r->ctrl_domains, hdr.list) { |
| 1067 |
if (sep) |
| 1068 |
seq_putc(seq, ';'); |
| 1069 |
hw_shareable = r->cache.shareable_bits; |
| 1070 |
sw_shareable = 0; |
| 1071 |
exclusive = 0; |
| 1072 |
seq_printf(seq, "%d=", dom->hdr.id); |
| 1073 |
for (i = 0; i < closids_supported(); i++) { |
| 1074 |
if (!closid_allocated(i) || |
| 1075 |
(resctrl_arch_get_io_alloc_enabled(r) && |
| 1076 |
i == resctrl_io_alloc_closid(r))) |
| 1077 |
continue; |
| 1078 |
ctrl_val = resctrl_arch_get_config(r, dom, i, |
| 1079 |
s->conf_type); |
| 1080 |
mode = rdtgroup_mode_by_closid(i); |
| 1081 |
switch (mode) { |
| 1082 |
case RDT_MODE_SHAREABLE: |
| 1083 |
sw_shareable |= ctrl_val; |
| 1084 |
break; |
| 1085 |
case RDT_MODE_EXCLUSIVE: |
| 1086 |
exclusive |= ctrl_val; |
| 1087 |
break; |
| 1088 |
case RDT_MODE_PSEUDO_LOCKSETUP: |
| 1089 |
/* |
| 1090 |
* RDT_MODE_PSEUDO_LOCKSETUP is possible |
| 1091 |
* here but not included since the CBM |
| 1092 |
* associated with this CLOSID in this mode |
| 1093 |
* is not initialized and no task or cpu can be |
| 1094 |
* assigned this CLOSID. |
| 1095 |
*/ |
| 1096 |
break; |
| 1097 |
case RDT_MODE_PSEUDO_LOCKED: |
| 1098 |
case RDT_NUM_MODES: |
| 1099 |
WARN(1, |
| 1100 |
"invalid mode for closid %d\n", i); |
| 1101 |
break; |
| 1102 |
} |
| 1103 |
} |
| 1104 |
|
| 1105 |
/* |
| 1106 |
* When the "io_alloc" feature is enabled, a portion of the cache |
| 1107 |
* is configured for shared use between hardware and software. |
| 1108 |
* Also, when CDP is enabled the CBMs of CDP_CODE and CDP_DATA |
| 1109 |
* resources are kept in sync. So, the CBMs for "io_alloc" can |
| 1110 |
* be accessed through either resource. |
| 1111 |
*/ |
| 1112 |
if (resctrl_arch_get_io_alloc_enabled(r)) { |
| 1113 |
ctrl_val = resctrl_arch_get_config(r, dom, |
| 1114 |
resctrl_io_alloc_closid(r), |
| 1115 |
s->conf_type); |
| 1116 |
hw_shareable |= ctrl_val; |
| 1117 |
} |
| 1118 |
|
| 1119 |
for (i = r->cache.cbm_len - 1; i >= 0; i--) { |
| 1120 |
pseudo_locked = dom->plr ? dom->plr->cbm : 0; |
| 1121 |
hwb = test_bit(i, &hw_shareable); |
| 1122 |
swb = test_bit(i, &sw_shareable); |
| 1123 |
excl = test_bit(i, &exclusive); |
| 1124 |
psl = test_bit(i, &pseudo_locked); |
| 1125 |
if (hwb && swb) |
| 1126 |
seq_putc(seq, 'X'); |
| 1127 |
else if (hwb && !swb) |
| 1128 |
seq_putc(seq, 'H'); |
| 1129 |
else if (!hwb && swb) |
| 1130 |
seq_putc(seq, 'S'); |
| 1131 |
else if (excl) |
| 1132 |
seq_putc(seq, 'E'); |
| 1133 |
else if (psl) |
| 1134 |
seq_putc(seq, 'P'); |
| 1135 |
else /* Unused bits remain */ |
| 1136 |
seq_putc(seq, '0'); |
| 1137 |
} |
| 1138 |
sep = true; |
| 1139 |
} |
| 1140 |
seq_putc(seq, '\n'); |
| 1141 |
mutex_unlock(&rdtgroup_mutex); |
| 1142 |
cpus_read_unlock(); |
| 1143 |
return 0; |
| 1144 |
} |
| 1145 |
|
| 1146 |
static int rdt_min_bw_show(struct kernfs_open_file *of, |
| 1147 |
struct seq_file *seq, void *v) |
| 1148 |
{ |
| 1149 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1150 |
struct rdt_resource *r = s->res; |
| 1151 |
|
| 1152 |
seq_printf(seq, "%u\n", r->membw.min_bw); |
| 1153 |
return 0; |
| 1154 |
} |
| 1155 |
|
| 1156 |
static int rdt_num_rmids_show(struct kernfs_open_file *of, |
| 1157 |
struct seq_file *seq, void *v) |
| 1158 |
{ |
| 1159 |
struct rdt_resource *r = rdt_kn_parent_priv(of->kn); |
| 1160 |
|
| 1161 |
seq_printf(seq, "%u\n", r->mon.num_rmid); |
| 1162 |
|
| 1163 |
return 0; |
| 1164 |
} |
| 1165 |
|
| 1166 |
static int rdt_mon_features_show(struct kernfs_open_file *of, |
| 1167 |
struct seq_file *seq, void *v) |
| 1168 |
{ |
| 1169 |
struct rdt_resource *r = rdt_kn_parent_priv(of->kn); |
| 1170 |
struct mon_evt *mevt; |
| 1171 |
|
| 1172 |
for_each_mon_event(mevt) { |
| 1173 |
if (mevt->rid != r->rid || !mevt->enabled) |
| 1174 |
continue; |
| 1175 |
seq_printf(seq, "%s\n", mevt->name); |
| 1176 |
if (mevt->configurable && |
| 1177 |
!resctrl_arch_mbm_cntr_assign_enabled(r)) |
| 1178 |
seq_printf(seq, "%s_config\n", mevt->name); |
| 1179 |
} |
| 1180 |
|
| 1181 |
return 0; |
| 1182 |
} |
| 1183 |
|
| 1184 |
static int rdt_bw_gran_show(struct kernfs_open_file *of, |
| 1185 |
struct seq_file *seq, void *v) |
| 1186 |
{ |
| 1187 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1188 |
struct rdt_resource *r = s->res; |
| 1189 |
|
| 1190 |
seq_printf(seq, "%u\n", r->membw.bw_gran); |
| 1191 |
return 0; |
| 1192 |
} |
| 1193 |
|
| 1194 |
static int rdt_delay_linear_show(struct kernfs_open_file *of, |
| 1195 |
struct seq_file *seq, void *v) |
| 1196 |
{ |
| 1197 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1198 |
struct rdt_resource *r = s->res; |
| 1199 |
|
| 1200 |
seq_printf(seq, "%u\n", r->membw.delay_linear); |
| 1201 |
return 0; |
| 1202 |
} |
| 1203 |
|
| 1204 |
static int max_threshold_occ_show(struct kernfs_open_file *of, |
| 1205 |
struct seq_file *seq, void *v) |
| 1206 |
{ |
| 1207 |
seq_printf(seq, "%u\n", resctrl_rmid_realloc_threshold); |
| 1208 |
|
| 1209 |
return 0; |
| 1210 |
} |
| 1211 |
|
| 1212 |
static int rdt_thread_throttle_mode_show(struct kernfs_open_file *of, |
| 1213 |
struct seq_file *seq, void *v) |
| 1214 |
{ |
| 1215 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1216 |
struct rdt_resource *r = s->res; |
| 1217 |
|
| 1218 |
switch (r->membw.throttle_mode) { |
| 1219 |
case THREAD_THROTTLE_PER_THREAD: |
| 1220 |
seq_puts(seq, "per-thread\n"); |
| 1221 |
return 0; |
| 1222 |
case THREAD_THROTTLE_MAX: |
| 1223 |
seq_puts(seq, "max\n"); |
| 1224 |
return 0; |
| 1225 |
case THREAD_THROTTLE_UNDEFINED: |
| 1226 |
seq_puts(seq, "undefined\n"); |
| 1227 |
return 0; |
| 1228 |
} |
| 1229 |
|
| 1230 |
WARN_ON_ONCE(1); |
| 1231 |
|
| 1232 |
return 0; |
| 1233 |
} |
| 1234 |
|
| 1235 |
static ssize_t max_threshold_occ_write(struct kernfs_open_file *of, |
| 1236 |
char *buf, size_t nbytes, loff_t off) |
| 1237 |
{ |
| 1238 |
unsigned int bytes; |
| 1239 |
int ret; |
| 1240 |
|
| 1241 |
ret = kstrtouint(buf, 0, &bytes); |
| 1242 |
if (ret) |
| 1243 |
return ret; |
| 1244 |
|
| 1245 |
if (bytes > resctrl_rmid_realloc_limit) |
| 1246 |
return -EINVAL; |
| 1247 |
|
| 1248 |
resctrl_rmid_realloc_threshold = resctrl_arch_round_mon_val(bytes); |
| 1249 |
|
| 1250 |
return nbytes; |
| 1251 |
} |
| 1252 |
|
| 1253 |
/* |
| 1254 |
* rdtgroup_mode_show - Display mode of this resource group |
| 1255 |
*/ |
| 1256 |
static int rdtgroup_mode_show(struct kernfs_open_file *of, |
| 1257 |
struct seq_file *s, void *v) |
| 1258 |
{ |
| 1259 |
struct rdtgroup *rdtgrp; |
| 1260 |
|
| 1261 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 1262 |
if (!rdtgrp) { |
| 1263 |
rdtgroup_kn_unlock(of->kn); |
| 1264 |
return -ENOENT; |
| 1265 |
} |
| 1266 |
|
| 1267 |
seq_printf(s, "%s\n", rdtgroup_mode_str(rdtgrp->mode)); |
| 1268 |
|
| 1269 |
rdtgroup_kn_unlock(of->kn); |
| 1270 |
return 0; |
| 1271 |
} |
| 1272 |
|
| 1273 |
enum resctrl_conf_type resctrl_peer_type(enum resctrl_conf_type my_type) |
| 1274 |
{ |
| 1275 |
switch (my_type) { |
| 1276 |
case CDP_CODE: |
| 1277 |
return CDP_DATA; |
| 1278 |
case CDP_DATA: |
| 1279 |
return CDP_CODE; |
| 1280 |
default: |
| 1281 |
case CDP_NONE: |
| 1282 |
return CDP_NONE; |
| 1283 |
} |
| 1284 |
} |
| 1285 |
|
| 1286 |
static int rdt_has_sparse_bitmasks_show(struct kernfs_open_file *of, |
| 1287 |
struct seq_file *seq, void *v) |
| 1288 |
{ |
| 1289 |
struct resctrl_schema *s = rdt_kn_parent_priv(of->kn); |
| 1290 |
struct rdt_resource *r = s->res; |
| 1291 |
|
| 1292 |
seq_printf(seq, "%u\n", r->cache.arch_has_sparse_bitmasks); |
| 1293 |
|
| 1294 |
return 0; |
| 1295 |
} |
| 1296 |
|
| 1297 |
/** |
| 1298 |
* __rdtgroup_cbm_overlaps - Does CBM for intended closid overlap with other |
| 1299 |
* @r: Resource to which domain instance @d belongs. |
| 1300 |
* @d: The domain instance for which @closid is being tested. |
| 1301 |
* @cbm: Capacity bitmask being tested. |
| 1302 |
* @closid: Intended closid for @cbm. |
| 1303 |
* @type: CDP type of @r. |
| 1304 |
* @exclusive: Only check if overlaps with exclusive resource groups |
| 1305 |
* |
| 1306 |
* Checks if provided @cbm intended to be used for @closid on domain |
| 1307 |
* @d overlaps with any other closids or other hardware usage associated |
| 1308 |
* with this domain. If @exclusive is true then only overlaps with |
| 1309 |
* resource groups in exclusive mode will be considered. If @exclusive |
| 1310 |
* is false then overlaps with any resource group or hardware entities |
| 1311 |
* will be considered. |
| 1312 |
* |
| 1313 |
* @cbm is unsigned long, even if only 32 bits are used, to make the |
| 1314 |
* bitmap functions work correctly. |
| 1315 |
* |
| 1316 |
* Return: false if CBM does not overlap, true if it does. |
| 1317 |
*/ |
| 1318 |
static bool __rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_ctrl_domain *d, |
| 1319 |
unsigned long cbm, int closid, |
| 1320 |
enum resctrl_conf_type type, bool exclusive) |
| 1321 |
{ |
| 1322 |
enum rdtgrp_mode mode; |
| 1323 |
unsigned long ctrl_b; |
| 1324 |
int i; |
| 1325 |
|
| 1326 |
/* Check for any overlap with regions used by hardware directly */ |
| 1327 |
if (!exclusive) { |
| 1328 |
ctrl_b = r->cache.shareable_bits; |
| 1329 |
if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) |
| 1330 |
return true; |
| 1331 |
} |
| 1332 |
|
| 1333 |
/* Check for overlap with other resource groups */ |
| 1334 |
for (i = 0; i < closids_supported(); i++) { |
| 1335 |
ctrl_b = resctrl_arch_get_config(r, d, i, type); |
| 1336 |
mode = rdtgroup_mode_by_closid(i); |
| 1337 |
if (closid_allocated(i) && i != closid && |
| 1338 |
mode != RDT_MODE_PSEUDO_LOCKSETUP) { |
| 1339 |
if (bitmap_intersects(&cbm, &ctrl_b, r->cache.cbm_len)) { |
| 1340 |
if (exclusive) { |
| 1341 |
if (mode == RDT_MODE_EXCLUSIVE) |
| 1342 |
return true; |
| 1343 |
continue; |
| 1344 |
} |
| 1345 |
return true; |
| 1346 |
} |
| 1347 |
} |
| 1348 |
} |
| 1349 |
|
| 1350 |
return false; |
| 1351 |
} |
| 1352 |
|
| 1353 |
/** |
| 1354 |
* rdtgroup_cbm_overlaps - Does CBM overlap with other use of hardware |
| 1355 |
* @s: Schema for the resource to which domain instance @d belongs. |
| 1356 |
* @d: The domain instance for which @closid is being tested. |
| 1357 |
* @cbm: Capacity bitmask being tested. |
| 1358 |
* @closid: Intended closid for @cbm. |
| 1359 |
* @exclusive: Only check if overlaps with exclusive resource groups |
| 1360 |
* |
| 1361 |
* Resources that can be allocated using a CBM can use the CBM to control |
| 1362 |
* the overlap of these allocations. rdtgroup_cmb_overlaps() is the test |
| 1363 |
* for overlap. Overlap test is not limited to the specific resource for |
| 1364 |
* which the CBM is intended though - when dealing with CDP resources that |
| 1365 |
* share the underlying hardware the overlap check should be performed on |
| 1366 |
* the CDP resource sharing the hardware also. |
| 1367 |
* |
| 1368 |
* Refer to description of __rdtgroup_cbm_overlaps() for the details of the |
| 1369 |
* overlap test. |
| 1370 |
* |
| 1371 |
* Return: true if CBM overlap detected, false if there is no overlap |
| 1372 |
*/ |
| 1373 |
bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_ctrl_domain *d, |
| 1374 |
unsigned long cbm, int closid, bool exclusive) |
| 1375 |
{ |
| 1376 |
enum resctrl_conf_type peer_type = resctrl_peer_type(s->conf_type); |
| 1377 |
struct rdt_resource *r = s->res; |
| 1378 |
|
| 1379 |
if (__rdtgroup_cbm_overlaps(r, d, cbm, closid, s->conf_type, |
| 1380 |
exclusive)) |
| 1381 |
return true; |
| 1382 |
|
| 1383 |
if (!resctrl_arch_get_cdp_enabled(r->rid)) |
| 1384 |
return false; |
| 1385 |
return __rdtgroup_cbm_overlaps(r, d, cbm, closid, peer_type, exclusive); |
| 1386 |
} |
| 1387 |
|
| 1388 |
/** |
| 1389 |
* rdtgroup_mode_test_exclusive - Test if this resource group can be exclusive |
| 1390 |
* @rdtgrp: Resource group identified through its closid. |
| 1391 |
* |
| 1392 |
* An exclusive resource group implies that there should be no sharing of |
| 1393 |
* its allocated resources. At the time this group is considered to be |
| 1394 |
* exclusive this test can determine if its current schemata supports this |
| 1395 |
* setting by testing for overlap with all other resource groups. |
| 1396 |
* |
| 1397 |
* Return: true if resource group can be exclusive, false if there is overlap |
| 1398 |
* with allocations of other resource groups and thus this resource group |
| 1399 |
* cannot be exclusive. |
| 1400 |
*/ |
| 1401 |
static bool rdtgroup_mode_test_exclusive(struct rdtgroup *rdtgrp) |
| 1402 |
{ |
| 1403 |
int closid = rdtgrp->closid; |
| 1404 |
struct rdt_ctrl_domain *d; |
| 1405 |
struct resctrl_schema *s; |
| 1406 |
struct rdt_resource *r; |
| 1407 |
bool has_cache = false; |
| 1408 |
u32 ctrl; |
| 1409 |
|
| 1410 |
/* Walking r->domains, ensure it can't race with cpuhp */ |
| 1411 |
lockdep_assert_cpus_held(); |
| 1412 |
|
| 1413 |
list_for_each_entry(s, &resctrl_schema_all, list) { |
| 1414 |
r = s->res; |
| 1415 |
if (r->rid == RDT_RESOURCE_MBA || r->rid == RDT_RESOURCE_SMBA) |
| 1416 |
continue; |
| 1417 |
has_cache = true; |
| 1418 |
list_for_each_entry(d, &r->ctrl_domains, hdr.list) { |
| 1419 |
ctrl = resctrl_arch_get_config(r, d, closid, |
| 1420 |
s->conf_type); |
| 1421 |
if (rdtgroup_cbm_overlaps(s, d, ctrl, closid, false)) { |
| 1422 |
rdt_last_cmd_puts("Schemata overlaps\n"); |
| 1423 |
return false; |
| 1424 |
} |
| 1425 |
} |
| 1426 |
} |
| 1427 |
|
| 1428 |
if (!has_cache) { |
| 1429 |
rdt_last_cmd_puts("Cannot be exclusive without CAT/CDP\n"); |
| 1430 |
return false; |
| 1431 |
} |
| 1432 |
|
| 1433 |
return true; |
| 1434 |
} |
| 1435 |
|
| 1436 |
/* |
| 1437 |
* rdtgroup_mode_write - Modify the resource group's mode |
| 1438 |
*/ |
| 1439 |
static ssize_t rdtgroup_mode_write(struct kernfs_open_file *of, |
| 1440 |
char *buf, size_t nbytes, loff_t off) |
| 1441 |
{ |
| 1442 |
struct rdtgroup *rdtgrp; |
| 1443 |
enum rdtgrp_mode mode; |
| 1444 |
int ret = 0; |
| 1445 |
|
| 1446 |
/* Valid input requires a trailing newline */ |
| 1447 |
if (nbytes == 0 || buf[nbytes - 1] != '\n') |
| 1448 |
return -EINVAL; |
| 1449 |
buf[nbytes - 1] = '\0'; |
| 1450 |
|
| 1451 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 1452 |
if (!rdtgrp) { |
| 1453 |
rdtgroup_kn_unlock(of->kn); |
| 1454 |
return -ENOENT; |
| 1455 |
} |
| 1456 |
|
| 1457 |
rdt_last_cmd_clear(); |
| 1458 |
|
| 1459 |
mode = rdtgrp->mode; |
| 1460 |
|
| 1461 |
if ((!strcmp(buf, "shareable") && mode == RDT_MODE_SHAREABLE) || |
| 1462 |
(!strcmp(buf, "exclusive") && mode == RDT_MODE_EXCLUSIVE) || |
| 1463 |
(!strcmp(buf, "pseudo-locksetup") && |
| 1464 |
mode == RDT_MODE_PSEUDO_LOCKSETUP) || |
| 1465 |
(!strcmp(buf, "pseudo-locked") && mode == RDT_MODE_PSEUDO_LOCKED)) |
| 1466 |
goto out; |
| 1467 |
|
| 1468 |
if (mode == RDT_MODE_PSEUDO_LOCKED) { |
| 1469 |
rdt_last_cmd_puts("Cannot change pseudo-locked group\n"); |
| 1470 |
ret = -EINVAL; |
| 1471 |
goto out; |
| 1472 |
} |
| 1473 |
|
| 1474 |
if (!strcmp(buf, "shareable")) { |
| 1475 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
| 1476 |
ret = rdtgroup_locksetup_exit(rdtgrp); |
| 1477 |
if (ret) |
| 1478 |
goto out; |
| 1479 |
} |
| 1480 |
rdtgrp->mode = RDT_MODE_SHAREABLE; |
| 1481 |
} else if (!strcmp(buf, "exclusive")) { |
| 1482 |
if (!rdtgroup_mode_test_exclusive(rdtgrp)) { |
| 1483 |
ret = -EINVAL; |
| 1484 |
goto out; |
| 1485 |
} |
| 1486 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
| 1487 |
ret = rdtgroup_locksetup_exit(rdtgrp); |
| 1488 |
if (ret) |
| 1489 |
goto out; |
| 1490 |
} |
| 1491 |
rdtgrp->mode = RDT_MODE_EXCLUSIVE; |
| 1492 |
} else if (IS_ENABLED(CONFIG_RESCTRL_FS_PSEUDO_LOCK) && |
| 1493 |
!strcmp(buf, "pseudo-locksetup")) { |
| 1494 |
ret = rdtgroup_locksetup_enter(rdtgrp); |
| 1495 |
if (ret) |
| 1496 |
goto out; |
| 1497 |
rdtgrp->mode = RDT_MODE_PSEUDO_LOCKSETUP; |
| 1498 |
} else { |
| 1499 |
rdt_last_cmd_puts("Unknown or unsupported mode\n"); |
| 1500 |
ret = -EINVAL; |
| 1501 |
} |
| 1502 |
|
| 1503 |
out: |
| 1504 |
rdtgroup_kn_unlock(of->kn); |
| 1505 |
return ret ?: nbytes; |
| 1506 |
} |
| 1507 |
|
| 1508 |
/** |
| 1509 |
* rdtgroup_cbm_to_size - Translate CBM to size in bytes |
| 1510 |
* @r: RDT resource to which @d belongs. |
| 1511 |
* @d: RDT domain instance. |
| 1512 |
* @cbm: bitmask for which the size should be computed. |
| 1513 |
* |
| 1514 |
* The bitmask provided associated with the RDT domain instance @d will be |
| 1515 |
* translated into how many bytes it represents. The size in bytes is |
| 1516 |
* computed by first dividing the total cache size by the CBM length to |
| 1517 |
* determine how many bytes each bit in the bitmask represents. The result |
| 1518 |
* is multiplied with the number of bits set in the bitmask. |
| 1519 |
* |
| 1520 |
* @cbm is unsigned long, even if only 32 bits are used to make the |
| 1521 |
* bitmap functions work correctly. |
| 1522 |
*/ |
| 1523 |
unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, |
| 1524 |
struct rdt_ctrl_domain *d, unsigned long cbm) |
| 1525 |
{ |
| 1526 |
unsigned int size = 0; |
| 1527 |
struct cacheinfo *ci; |
| 1528 |
int num_b; |
| 1529 |
|
| 1530 |
if (WARN_ON_ONCE(r->ctrl_scope != RESCTRL_L2_CACHE && r->ctrl_scope != RESCTRL_L3_CACHE)) |
| 1531 |
return size; |
| 1532 |
|
| 1533 |
num_b = bitmap_weight(&cbm, r->cache.cbm_len); |
| 1534 |
ci = get_cpu_cacheinfo_level(cpumask_any(&d->hdr.cpu_mask), r->ctrl_scope); |
| 1535 |
if (ci) |
| 1536 |
size = ci->size / r->cache.cbm_len * num_b; |
| 1537 |
|
| 1538 |
return size; |
| 1539 |
} |
| 1540 |
|
| 1541 |
bool is_mba_sc(struct rdt_resource *r) |
| 1542 |
{ |
| 1543 |
if (!r) |
| 1544 |
r = resctrl_arch_get_resource(RDT_RESOURCE_MBA); |
| 1545 |
|
| 1546 |
/* |
| 1547 |
* The software controller support is only applicable to MBA resource. |
| 1548 |
* Make sure to check for resource type. |
| 1549 |
*/ |
| 1550 |
if (r->rid != RDT_RESOURCE_MBA) |
| 1551 |
return false; |
| 1552 |
|
| 1553 |
return r->membw.mba_sc; |
| 1554 |
} |
| 1555 |
|
| 1556 |
/* |
| 1557 |
* rdtgroup_size_show - Display size in bytes of allocated regions |
| 1558 |
* |
| 1559 |
* The "size" file mirrors the layout of the "schemata" file, printing the |
| 1560 |
* size in bytes of each region instead of the capacity bitmask. |
| 1561 |
*/ |
| 1562 |
static int rdtgroup_size_show(struct kernfs_open_file *of, |
| 1563 |
struct seq_file *s, void *v) |
| 1564 |
{ |
| 1565 |
struct resctrl_schema *schema; |
| 1566 |
enum resctrl_conf_type type; |
| 1567 |
struct rdt_ctrl_domain *d; |
| 1568 |
struct rdtgroup *rdtgrp; |
| 1569 |
struct rdt_resource *r; |
| 1570 |
unsigned int size; |
| 1571 |
int ret = 0; |
| 1572 |
u32 closid; |
| 1573 |
bool sep; |
| 1574 |
u32 ctrl; |
| 1575 |
|
| 1576 |
rdtgrp = rdtgroup_kn_lock_live(of->kn); |
| 1577 |
if (!rdtgrp) { |
| 1578 |
rdtgroup_kn_unlock(of->kn); |
| 1579 |
return -ENOENT; |
| 1580 |
} |
| 1581 |
|
| 1582 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { |
| 1583 |
if (!rdtgrp->plr->d) { |
| 1584 |
rdt_last_cmd_clear(); |
| 1585 |
rdt_last_cmd_puts("Cache domain offline\n"); |
| 1586 |
ret = -ENODEV; |
| 1587 |
} else { |
| 1588 |
seq_printf(s, "%*s:", max_name_width, |
| 1589 |
rdtgrp->plr->s->name); |
| 1590 |
size = rdtgroup_cbm_to_size(rdtgrp->plr->s->res, |
| 1591 |
rdtgrp->plr->d, |
| 1592 |
rdtgrp->plr->cbm); |
| 1593 |
seq_printf(s, "%d=%u\n", rdtgrp->plr->d->hdr.id, size); |
| 1594 |
} |
| 1595 |
goto out; |
| 1596 |
} |
| 1597 |
|
| 1598 |
closid = rdtgrp->closid; |
| 1599 |
|
| 1600 |
list_for_each_entry(schema, &resctrl_schema_all, list) { |
| 1601 |
r = schema->res; |
| 1602 |
type = schema->conf_type; |
| 1603 |
sep = false; |
| 1604 |
seq_printf(s, "%*s:", max_name_width, schema->name); |
| 1605 |
list_for_each_entry(d, &r->ctrl_domains, hdr.list) { |
| 1606 |
if (sep) |
| 1607 |
seq_putc(s, ';'); |
| 1608 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP) { |
| 1609 |
size = 0; |
| 1610 |
} else { |
| 1611 |
if (is_mba_sc(r)) |
| 1612 |
ctrl = d->mbps_val[closid]; |
| 1613 |
else |
| 1614 |
ctrl = resctrl_arch_get_config(r, d, |
| 1615 |
closid, |
| 1616 |
type); |
| 1617 |
if (r->rid == RDT_RESOURCE_MBA || |
| 1618 |
r->rid == RDT_RESOURCE_SMBA) |
| 1619 |
size = ctrl; |
| 1620 |
else |
| 1621 |
size = rdtgroup_cbm_to_size(r, d, ctrl); |
| 1622 |
} |
| 1623 |
seq_printf(s, "%d=%u", d->hdr.id, size); |
| 1624 |
sep = true; |
| 1625 |
} |
| 1626 |
seq_putc(s, '\n'); |
| 1627 |
} |
| 1628 |
|
| 1629 |
out: |
| 1630 |
rdtgroup_kn_unlock(of->kn); |
| 1631 |
|
| 1632 |
return ret; |
| 1633 |
} |
| 1634 |
|
| 1635 |
static void mondata_config_read(struct resctrl_mon_config_info *mon_info) |
| 1636 |
{ |
| 1637 |
smp_call_function_any(&mon_info->d->hdr.cpu_mask, |
| 1638 |
resctrl_arch_mon_event_config_read, mon_info, 1); |
| 1639 |
} |
| 1640 |
|
| 1641 |
static int mbm_config_show(struct seq_file *s, struct rdt_resource *r, u32 evtid) |
| 1642 |
{ |
| 1643 |
struct resctrl_mon_config_info mon_info; |
| 1644 |
struct rdt_l3_mon_domain *dom; |
| 1645 |
bool sep = false; |
| 1646 |
|
| 1647 |
cpus_read_lock(); |
| 1648 |
mutex_lock(&rdtgroup_mutex); |
| 1649 |
|
| 1650 |
list_for_each_entry(dom, &r->mon_domains, hdr.list) { |
| 1651 |
if (sep) |
| 1652 |
seq_puts(s, ";"); |
| 1653 |
|
| 1654 |
memset(&mon_info, 0, sizeof(struct resctrl_mon_config_info)); |
| 1655 |
mon_info.r = r; |
| 1656 |
mon_info.d = dom; |
| 1657 |
mon_info.evtid = evtid; |
| 1658 |
mondata_config_read(&mon_info); |
| 1659 |
|
| 1660 |
seq_printf(s, "%d=0x%02x", dom->hdr.id, mon_info.mon_config); |
| 1661 |
sep = true; |
| 1662 |
} |
| 1663 |
seq_puts(s, "\n"); |
| 1664 |
|
| 1665 |
mutex_unlock(&rdtgroup_mutex); |
| 1666 |
cpus_read_unlock(); |
| 1667 |
|
| 1668 |
return 0; |
| 1669 |
} |
| 1670 |
|
| 1671 |
static int mbm_total_bytes_config_show(struct kernfs_open_file *of, |
| 1672 |
struct seq_file *seq, void *v) |
| 1673 |
{ |
| 1674 |
struct rdt_resource *r = rdt_kn_parent_priv(of->kn); |
| 1675 |
|
| 1676 |
mbm_config_show(seq, r, QOS_L3_MBM_TOTAL_EVENT_ID); |
| 1677 |
|
| 1678 |
return 0; |
| 1679 |
} |
| 1680 |
|
| 1681 |
static int mbm_local_bytes_config_show(struct kernfs_open_file *of, |
| 1682 |
struct seq_file *seq, void *v) |
| 1683 |
{ |
| 1684 |
struct rdt_resource *r = rdt_kn_parent_priv(of->kn); |
| 1685 |
|
| 1686 |
mbm_config_show(seq, r, QOS_L3_MBM_LOCAL_EVENT_ID); |
| 1687 |
|
| 1688 |
return 0; |
| 1689 |
} |
| 1690 |
|
| 1691 |
static void mbm_config_write_domain(struct rdt_resource *r, |
| 1692 |
struct rdt_l3_mon_domain *d, u32 evtid, u32 val) |
| 1693 |
{ |
| 1694 |
struct resctrl_mon_config_info mon_info = {0}; |
| 1695 |
|
| 1696 |
/* |
| 1697 |
* Read the current config value first. If both are the same then |
| 1698 |
* no need to write it again. |
| 1699 |
*/ |
| 1700 |
mon_info.r = r; |
| 1701 |
mon_info.d = d; |
| 1702 |
mon_info.evtid = evtid; |
| 1703 |
mondata_config_read(&mon_info); |
| 1704 |
if (mon_info.mon_config == val) |
| 1705 |
return; |
| 1706 |
|
| 1707 |
mon_info.mon_config = val; |
| 1708 |
|
| 1709 |
/* |
| 1710 |
* Update MSR_IA32_EVT_CFG_BASE MSR on one of the CPUs in the |
| 1711 |
* domain. The MSRs offset from MSR MSR_IA32_EVT_CFG_BASE |
| 1712 |
* are scoped at the domain level. Writing any of these MSRs |
| 1713 |
* on one CPU is observed by all the CPUs in the domain. |
| 1714 |
*/ |
| 1715 |
smp_call_function_any(&d->hdr.cpu_mask, resctrl_arch_mon_event_config_write, |
| 1716 |
&mon_info, 1); |
| 1717 |
|
| 1718 |
/* |
| 1719 |
* When an Event Configuration is changed, the bandwidth counters |
| 1720 |
* for all RMIDs and Events will be cleared by the hardware. The |
| 1721 |
* hardware also sets MSR_IA32_QM_CTR.Unavailable (bit 62) for |
| 1722 |
* every RMID on the next read to any event for every RMID. |
| 1723 |
* Subsequent reads will have MSR_IA32_QM_CTR.Unavailable (bit 62) |
| 1724 |
* cleared while it is tracked by the hardware. Clear the |
| 1725 |
* mbm_local and mbm_total counts for all the RMIDs. |
| 1726 |
*/ |
| 1727 |
resctrl_arch_reset_rmid_all(r, d); |
| 1728 |
} |
| 1729 |
|
| 1730 |
static int mon_config_write(struct rdt_resource *r, char *tok, u32 evtid) |
| 1731 |
{ |
| 1732 |
char *dom_str = NULL, *id_str; |
| 1733 |
struct rdt_l3_mon_domain *d; |
| 1734 |
unsigned long dom_id, val; |
| 1735 |
|
| 1736 |
/* Walking r->domains, ensure it can't race with cpuhp */ |
| 1737 |
lockdep_assert_cpus_held(); |
| 1738 |
|
| 1739 |
next: |
| 1740 |
if (!tok || tok[0] == '\0') |
| 1741 |
return 0; |
| 1742 |
|
| 1743 |
/* Start processing the strings for each domain */ |
| 1744 |
dom_str = strim(strsep(&tok, ";")); |
| 1745 |
id_str = strsep(&dom_str, "="); |
| 1746 |
|
| 1747 |
if (!id_str || kstrtoul(id_str, 10, &dom_id)) { |
| 1748 |
rdt_last_cmd_puts("Missing '=' or non-numeric domain id\n"); |
| 1749 |
return -EINVAL; |
| 1750 |
} |
| 1751 |
|
| 1752 |
if (!dom_str || kstrtoul(dom_str, 16, &val)) { |
| 1753 |
rdt_last_cmd_puts("Non-numeric event configuration value\n"); |
| 1754 |
return -EINVAL; |
| 1755 |
} |
| 1756 |
|
| 1757 |
/* Value from user cannot be more than the supported set of events */ |
| 1758 |
if ((val & r->mon.mbm_cfg_mask) != val) { |
| 1759 |
rdt_last_cmd_printf("Invalid event configuration: max valid mask is 0x%02x\n", |
| 1760 |
r->mon.mbm_cfg_mask); |
| 1761 |
return -EINVAL; |
| 1762 |
} |
| 1763 |
|
| 1764 |
list_for_each_entry(d, &r->mon_domains, hdr.list) { |
| 1765 |
if (d->hdr.id == dom_id) { |
| 1766 |
mbm_config_write_domain(r, d, evtid, val); |
| 1767 |
goto next; |
| 1768 |
} |
| 1769 |
} |
| 1770 |
|
| 1771 |
return -EINVAL; |
| 1772 |
} |
| 1773 |
|
| 1774 |
static ssize_t mbm_total_bytes_config_write(struct kernfs_open_file *of, |
| 1775 |
char *buf, size_t nbytes, |
| 1776 |
loff_t off) |
| 1777 |
{ |
| 1778 |
struct rdt_resource *r = rdt_kn_parent_priv(of->kn); |
| 1779 |
int ret; |
| 1780 |
|
| 1781 |
/* Valid input requires a trailing newline */ |
| 1782 |
if (nbytes == 0 || buf[nbytes - 1] != '\n') |
| 1783 |
return -EINVAL; |
| 1784 |
|
| 1785 |
cpus_read_lock(); |
| 1786 |
mutex_lock(&rdtgroup_mutex); |
| 1787 |
|
| 1788 |
rdt_last_cmd_clear(); |
| 1789 |
|
| 1790 |
buf[nbytes - 1] = '\0'; |
| 1791 |
|
| 1792 |
ret = mon_config_write(r, buf, QOS_L3_MBM_TOTAL_EVENT_ID); |
| 1793 |
|
| 1794 |
mutex_unlock(&rdtgroup_mutex); |
| 1795 |
cpus_read_unlock(); |
| 1796 |
|
| 1797 |
return ret ?: nbytes; |
| 1798 |
} |
| 1799 |
|
| 1800 |
static ssize_t mbm_local_bytes_config_write(struct kernfs_open_file *of, |
| 1801 |
char *buf, size_t nbytes, |
| 1802 |
loff_t off) |
| 1803 |
{ |
| 1804 |
struct rdt_resource *r = rdt_kn_parent_priv(of->kn); |
| 1805 |
int ret; |
| 1806 |
|
| 1807 |
/* Valid input requires a trailing newline */ |
| 1808 |
if (nbytes == 0 || buf[nbytes - 1] != '\n') |
| 1809 |
return -EINVAL; |
| 1810 |
|
| 1811 |
cpus_read_lock(); |
| 1812 |
mutex_lock(&rdtgroup_mutex); |
| 1813 |
|
| 1814 |
rdt_last_cmd_clear(); |
| 1815 |
|
| 1816 |
buf[nbytes - 1] = '\0'; |
| 1817 |
|
| 1818 |
ret = mon_config_write(r, buf, QOS_L3_MBM_LOCAL_EVENT_ID); |
| 1819 |
|
| 1820 |
mutex_unlock(&rdtgroup_mutex); |
| 1821 |
cpus_read_unlock(); |
| 1822 |
|
| 1823 |
return ret ?: nbytes; |
| 1824 |
} |
| 1825 |
|
| 1826 |
/* |
| 1827 |
* resctrl_bmec_files_show() — Controls the visibility of BMEC-related resctrl |
| 1828 |
* files. When @show is true, the files are displayed; when false, the files |
| 1829 |
* are hidden. |
| 1830 |
* Don't treat kernfs_find_and_get failure as an error, since this function may |
| 1831 |
* be called regardless of whether BMEC is supported or the event is enabled. |
| 1832 |
*/ |
| 1833 |
void resctrl_bmec_files_show(struct rdt_resource *r, struct kernfs_node *l3_mon_kn, |
| 1834 |
bool show) |
| 1835 |
{ |
| 1836 |
struct kernfs_node *kn_config, *mon_kn = NULL; |
| 1837 |
char name[32]; |
| 1838 |
|
| 1839 |
if (!l3_mon_kn) { |
| 1840 |
sprintf(name, "%s_MON", r->name); |
| 1841 |
mon_kn = kernfs_find_and_get(kn_info, name); |
| 1842 |
if (!mon_kn) |
| 1843 |
return; |
| 1844 |
l3_mon_kn = mon_kn; |
| 1845 |
} |
| 1846 |
|
| 1847 |
kn_config = kernfs_find_and_get(l3_mon_kn, "mbm_total_bytes_config"); |
| 1848 |
if (kn_config) { |
| 1849 |
kernfs_show(kn_config, show); |
| 1850 |
kernfs_put(kn_config); |
| 1851 |
} |
| 1852 |
|
| 1853 |
kn_config = kernfs_find_and_get(l3_mon_kn, "mbm_local_bytes_config"); |
| 1854 |
if (kn_config) { |
| 1855 |
kernfs_show(kn_config, show); |
| 1856 |
kernfs_put(kn_config); |
| 1857 |
} |
| 1858 |
|
| 1859 |
/* Release the reference only if it was acquired */ |
| 1860 |
if (mon_kn) |
| 1861 |
kernfs_put(mon_kn); |
| 1862 |
} |
| 1863 |
|
| 1864 |
const char *rdtgroup_name_by_closid(u32 closid) |
| 1865 |
{ |
| 1866 |
struct rdtgroup *rdtgrp; |
| 1867 |
|
| 1868 |
list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) { |
| 1869 |
if (rdtgrp->closid == closid) |
| 1870 |
return rdt_kn_name(rdtgrp->kn); |
| 1871 |
} |
| 1872 |
|
| 1873 |
return NULL; |
| 1874 |
} |
| 1875 |
|
| 1876 |
/* rdtgroup information files for one cache resource. */ |
| 1877 |
static struct rftype res_common_files[] = { |
| 1878 |
{ |
| 1879 |
.name = "last_cmd_status", |
| 1880 |
.mode = 0444, |
| 1881 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1882 |
.seq_show = rdt_last_cmd_status_show, |
| 1883 |
.fflags = RFTYPE_TOP_INFO, |
| 1884 |
}, |
| 1885 |
{ |
| 1886 |
.name = "mbm_assign_on_mkdir", |
| 1887 |
.mode = 0644, |
| 1888 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1889 |
.seq_show = resctrl_mbm_assign_on_mkdir_show, |
| 1890 |
.write = resctrl_mbm_assign_on_mkdir_write, |
| 1891 |
}, |
| 1892 |
{ |
| 1893 |
.name = "num_closids", |
| 1894 |
.mode = 0444, |
| 1895 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1896 |
.seq_show = rdt_num_closids_show, |
| 1897 |
.fflags = RFTYPE_CTRL_INFO, |
| 1898 |
}, |
| 1899 |
{ |
| 1900 |
.name = "mon_features", |
| 1901 |
.mode = 0444, |
| 1902 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1903 |
.seq_show = rdt_mon_features_show, |
| 1904 |
.fflags = RFTYPE_MON_INFO, |
| 1905 |
}, |
| 1906 |
{ |
| 1907 |
.name = "available_mbm_cntrs", |
| 1908 |
.mode = 0444, |
| 1909 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1910 |
.seq_show = resctrl_available_mbm_cntrs_show, |
| 1911 |
}, |
| 1912 |
{ |
| 1913 |
.name = "num_rmids", |
| 1914 |
.mode = 0444, |
| 1915 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1916 |
.seq_show = rdt_num_rmids_show, |
| 1917 |
.fflags = RFTYPE_MON_INFO, |
| 1918 |
}, |
| 1919 |
{ |
| 1920 |
.name = "cbm_mask", |
| 1921 |
.mode = 0444, |
| 1922 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1923 |
.seq_show = rdt_default_ctrl_show, |
| 1924 |
.fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE, |
| 1925 |
}, |
| 1926 |
{ |
| 1927 |
.name = "num_mbm_cntrs", |
| 1928 |
.mode = 0444, |
| 1929 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1930 |
.seq_show = resctrl_num_mbm_cntrs_show, |
| 1931 |
}, |
| 1932 |
{ |
| 1933 |
.name = "min_cbm_bits", |
| 1934 |
.mode = 0444, |
| 1935 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1936 |
.seq_show = rdt_min_cbm_bits_show, |
| 1937 |
.fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE, |
| 1938 |
}, |
| 1939 |
{ |
| 1940 |
.name = "shareable_bits", |
| 1941 |
.mode = 0444, |
| 1942 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1943 |
.seq_show = rdt_shareable_bits_show, |
| 1944 |
.fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE, |
| 1945 |
}, |
| 1946 |
{ |
| 1947 |
.name = "bit_usage", |
| 1948 |
.mode = 0444, |
| 1949 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1950 |
.seq_show = rdt_bit_usage_show, |
| 1951 |
.fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE, |
| 1952 |
}, |
| 1953 |
{ |
| 1954 |
.name = "min_bandwidth", |
| 1955 |
.mode = 0444, |
| 1956 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1957 |
.seq_show = rdt_min_bw_show, |
| 1958 |
.fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_MB, |
| 1959 |
}, |
| 1960 |
{ |
| 1961 |
.name = "bandwidth_gran", |
| 1962 |
.mode = 0444, |
| 1963 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1964 |
.seq_show = rdt_bw_gran_show, |
| 1965 |
.fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_MB, |
| 1966 |
}, |
| 1967 |
{ |
| 1968 |
.name = "delay_linear", |
| 1969 |
.mode = 0444, |
| 1970 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1971 |
.seq_show = rdt_delay_linear_show, |
| 1972 |
.fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_MB, |
| 1973 |
}, |
| 1974 |
/* |
| 1975 |
* Platform specific which (if any) capabilities are provided by |
| 1976 |
* thread_throttle_mode. Defer "fflags" initialization to platform |
| 1977 |
* discovery. |
| 1978 |
*/ |
| 1979 |
{ |
| 1980 |
.name = "thread_throttle_mode", |
| 1981 |
.mode = 0444, |
| 1982 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1983 |
.seq_show = rdt_thread_throttle_mode_show, |
| 1984 |
}, |
| 1985 |
{ |
| 1986 |
.name = "io_alloc", |
| 1987 |
.mode = 0644, |
| 1988 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1989 |
.seq_show = resctrl_io_alloc_show, |
| 1990 |
.write = resctrl_io_alloc_write, |
| 1991 |
}, |
| 1992 |
{ |
| 1993 |
.name = "io_alloc_cbm", |
| 1994 |
.mode = 0644, |
| 1995 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 1996 |
.seq_show = resctrl_io_alloc_cbm_show, |
| 1997 |
.write = resctrl_io_alloc_cbm_write, |
| 1998 |
}, |
| 1999 |
{ |
| 2000 |
.name = "max_threshold_occupancy", |
| 2001 |
.mode = 0644, |
| 2002 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2003 |
.write = max_threshold_occ_write, |
| 2004 |
.seq_show = max_threshold_occ_show, |
| 2005 |
.fflags = RFTYPE_MON_INFO | RFTYPE_RES_CACHE, |
| 2006 |
}, |
| 2007 |
{ |
| 2008 |
.name = "mbm_total_bytes_config", |
| 2009 |
.mode = 0644, |
| 2010 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2011 |
.seq_show = mbm_total_bytes_config_show, |
| 2012 |
.write = mbm_total_bytes_config_write, |
| 2013 |
}, |
| 2014 |
{ |
| 2015 |
.name = "mbm_local_bytes_config", |
| 2016 |
.mode = 0644, |
| 2017 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2018 |
.seq_show = mbm_local_bytes_config_show, |
| 2019 |
.write = mbm_local_bytes_config_write, |
| 2020 |
}, |
| 2021 |
{ |
| 2022 |
.name = "event_filter", |
| 2023 |
.mode = 0644, |
| 2024 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2025 |
.seq_show = event_filter_show, |
| 2026 |
.write = event_filter_write, |
| 2027 |
}, |
| 2028 |
{ |
| 2029 |
.name = "mbm_L3_assignments", |
| 2030 |
.mode = 0644, |
| 2031 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2032 |
.seq_show = mbm_L3_assignments_show, |
| 2033 |
.write = mbm_L3_assignments_write, |
| 2034 |
}, |
| 2035 |
{ |
| 2036 |
.name = "mbm_assign_mode", |
| 2037 |
.mode = 0644, |
| 2038 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2039 |
.seq_show = resctrl_mbm_assign_mode_show, |
| 2040 |
.write = resctrl_mbm_assign_mode_write, |
| 2041 |
.fflags = RFTYPE_MON_INFO | RFTYPE_RES_CACHE, |
| 2042 |
}, |
| 2043 |
{ |
| 2044 |
.name = "cpus", |
| 2045 |
.mode = 0644, |
| 2046 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2047 |
.write = rdtgroup_cpus_write, |
| 2048 |
.seq_show = rdtgroup_cpus_show, |
| 2049 |
.fflags = RFTYPE_BASE, |
| 2050 |
}, |
| 2051 |
{ |
| 2052 |
.name = "cpus_list", |
| 2053 |
.mode = 0644, |
| 2054 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2055 |
.write = rdtgroup_cpus_write, |
| 2056 |
.seq_show = rdtgroup_cpus_show, |
| 2057 |
.flags = RFTYPE_FLAGS_CPUS_LIST, |
| 2058 |
.fflags = RFTYPE_BASE, |
| 2059 |
}, |
| 2060 |
{ |
| 2061 |
.name = "tasks", |
| 2062 |
.mode = 0644, |
| 2063 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2064 |
.write = rdtgroup_tasks_write, |
| 2065 |
.seq_show = rdtgroup_tasks_show, |
| 2066 |
.fflags = RFTYPE_BASE, |
| 2067 |
}, |
| 2068 |
{ |
| 2069 |
.name = "mon_hw_id", |
| 2070 |
.mode = 0444, |
| 2071 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2072 |
.seq_show = rdtgroup_rmid_show, |
| 2073 |
.fflags = RFTYPE_MON_BASE | RFTYPE_DEBUG, |
| 2074 |
}, |
| 2075 |
{ |
| 2076 |
.name = "schemata", |
| 2077 |
.mode = 0644, |
| 2078 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2079 |
.write = rdtgroup_schemata_write, |
| 2080 |
.seq_show = rdtgroup_schemata_show, |
| 2081 |
.fflags = RFTYPE_CTRL_BASE, |
| 2082 |
}, |
| 2083 |
{ |
| 2084 |
.name = "mba_MBps_event", |
| 2085 |
.mode = 0644, |
| 2086 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2087 |
.write = rdtgroup_mba_mbps_event_write, |
| 2088 |
.seq_show = rdtgroup_mba_mbps_event_show, |
| 2089 |
}, |
| 2090 |
{ |
| 2091 |
.name = "mode", |
| 2092 |
.mode = 0644, |
| 2093 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2094 |
.write = rdtgroup_mode_write, |
| 2095 |
.seq_show = rdtgroup_mode_show, |
| 2096 |
.fflags = RFTYPE_CTRL_BASE, |
| 2097 |
}, |
| 2098 |
{ |
| 2099 |
.name = "size", |
| 2100 |
.mode = 0444, |
| 2101 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2102 |
.seq_show = rdtgroup_size_show, |
| 2103 |
.fflags = RFTYPE_CTRL_BASE, |
| 2104 |
}, |
| 2105 |
{ |
| 2106 |
.name = "sparse_masks", |
| 2107 |
.mode = 0444, |
| 2108 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2109 |
.seq_show = rdt_has_sparse_bitmasks_show, |
| 2110 |
.fflags = RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE, |
| 2111 |
}, |
| 2112 |
{ |
| 2113 |
.name = "ctrl_hw_id", |
| 2114 |
.mode = 0444, |
| 2115 |
.kf_ops = &rdtgroup_kf_single_ops, |
| 2116 |
.seq_show = rdtgroup_closid_show, |
| 2117 |
.fflags = RFTYPE_CTRL_BASE | RFTYPE_DEBUG, |
| 2118 |
}, |
| 2119 |
}; |
| 2120 |
|
| 2121 |
static int rdtgroup_add_files(struct kernfs_node *kn, unsigned long fflags) |
| 2122 |
{ |
| 2123 |
struct rftype *rfts, *rft; |
| 2124 |
int ret, len; |
| 2125 |
|
| 2126 |
rfts = res_common_files; |
| 2127 |
len = ARRAY_SIZE(res_common_files); |
| 2128 |
|
| 2129 |
lockdep_assert_held(&rdtgroup_mutex); |
| 2130 |
|
| 2131 |
if (resctrl_debug) |
| 2132 |
fflags |= RFTYPE_DEBUG; |
| 2133 |
|
| 2134 |
for (rft = rfts; rft < rfts + len; rft++) { |
| 2135 |
if (rft->fflags && ((fflags & rft->fflags) == rft->fflags)) { |
| 2136 |
ret = rdtgroup_add_file(kn, rft); |
| 2137 |
if (ret) |
| 2138 |
goto error; |
| 2139 |
} |
| 2140 |
} |
| 2141 |
|
| 2142 |
return 0; |
| 2143 |
error: |
| 2144 |
pr_warn("Failed to add %s, err=%d\n", rft->name, ret); |
| 2145 |
while (--rft >= rfts) { |
| 2146 |
if ((fflags & rft->fflags) == rft->fflags) |
| 2147 |
kernfs_remove_by_name(kn, rft->name); |
| 2148 |
} |
| 2149 |
return ret; |
| 2150 |
} |
| 2151 |
|
| 2152 |
static struct rftype *rdtgroup_get_rftype_by_name(const char *name) |
| 2153 |
{ |
| 2154 |
struct rftype *rfts, *rft; |
| 2155 |
int len; |
| 2156 |
|
| 2157 |
rfts = res_common_files; |
| 2158 |
len = ARRAY_SIZE(res_common_files); |
| 2159 |
|
| 2160 |
for (rft = rfts; rft < rfts + len; rft++) { |
| 2161 |
if (!strcmp(rft->name, name)) |
| 2162 |
return rft; |
| 2163 |
} |
| 2164 |
|
| 2165 |
return NULL; |
| 2166 |
} |
| 2167 |
|
| 2168 |
static void thread_throttle_mode_init(void) |
| 2169 |
{ |
| 2170 |
enum membw_throttle_mode throttle_mode = THREAD_THROTTLE_UNDEFINED; |
| 2171 |
struct rdt_resource *r_mba, *r_smba; |
| 2172 |
|
| 2173 |
r_mba = resctrl_arch_get_resource(RDT_RESOURCE_MBA); |
| 2174 |
if (r_mba->alloc_capable && |
| 2175 |
r_mba->membw.throttle_mode != THREAD_THROTTLE_UNDEFINED) |
| 2176 |
throttle_mode = r_mba->membw.throttle_mode; |
| 2177 |
|
| 2178 |
r_smba = resctrl_arch_get_resource(RDT_RESOURCE_SMBA); |
| 2179 |
if (r_smba->alloc_capable && |
| 2180 |
r_smba->membw.throttle_mode != THREAD_THROTTLE_UNDEFINED) |
| 2181 |
throttle_mode = r_smba->membw.throttle_mode; |
| 2182 |
|
| 2183 |
if (throttle_mode == THREAD_THROTTLE_UNDEFINED) |
| 2184 |
return; |
| 2185 |
|
| 2186 |
resctrl_file_fflags_init("thread_throttle_mode", |
| 2187 |
RFTYPE_CTRL_INFO | RFTYPE_RES_MB); |
| 2188 |
} |
| 2189 |
|
| 2190 |
/* |
| 2191 |
* The resctrl file "io_alloc" is added using L3 resource. However, it results |
| 2192 |
* in this file being visible for *all* cache resources (eg. L2 cache), |
| 2193 |
* whether it supports "io_alloc" or not. |
| 2194 |
*/ |
| 2195 |
static void io_alloc_init(void) |
| 2196 |
{ |
| 2197 |
struct rdt_resource *r = resctrl_arch_get_resource(RDT_RESOURCE_L3); |
| 2198 |
|
| 2199 |
if (r->cache.io_alloc_capable) { |
| 2200 |
resctrl_file_fflags_init("io_alloc", RFTYPE_CTRL_INFO | |
| 2201 |
RFTYPE_RES_CACHE); |
| 2202 |
resctrl_file_fflags_init("io_alloc_cbm", |
| 2203 |
RFTYPE_CTRL_INFO | RFTYPE_RES_CACHE); |
| 2204 |
} |
| 2205 |
} |
| 2206 |
|
| 2207 |
void resctrl_file_fflags_init(const char *config, unsigned long fflags) |
| 2208 |
{ |
| 2209 |
struct rftype *rft; |
| 2210 |
|
| 2211 |
rft = rdtgroup_get_rftype_by_name(config); |
| 2212 |
if (rft) |
| 2213 |
rft->fflags = fflags; |
| 2214 |
} |
| 2215 |
|
| 2216 |
/** |
| 2217 |
* rdtgroup_kn_mode_restrict - Restrict user access to named resctrl file |
| 2218 |
* @r: The resource group with which the file is associated. |
| 2219 |
* @name: Name of the file |
| 2220 |
* |
| 2221 |
* The permissions of named resctrl file, directory, or link are modified |
| 2222 |
* to not allow read, write, or execute by any user. |
| 2223 |
* |
| 2224 |
* WARNING: This function is intended to communicate to the user that the |
| 2225 |
* resctrl file has been locked down - that it is not relevant to the |
| 2226 |
* particular state the system finds itself in. It should not be relied |
| 2227 |
* on to protect from user access because after the file's permissions |
| 2228 |
* are restricted the user can still change the permissions using chmod |
| 2229 |
* from the command line. |
| 2230 |
* |
| 2231 |
* Return: 0 on success, <0 on failure. |
| 2232 |
*/ |
| 2233 |
int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name) |
| 2234 |
{ |
| 2235 |
struct iattr iattr = {.ia_valid = ATTR_MODE,}; |
| 2236 |
struct kernfs_node *kn; |
| 2237 |
int ret = 0; |
| 2238 |
|
| 2239 |
kn = kernfs_find_and_get_ns(r->kn, name, NULL); |
| 2240 |
if (!kn) |
| 2241 |
return -ENOENT; |
| 2242 |
|
| 2243 |
switch (kernfs_type(kn)) { |
| 2244 |
case KERNFS_DIR: |
| 2245 |
iattr.ia_mode = S_IFDIR; |
| 2246 |
break; |
| 2247 |
case KERNFS_FILE: |
| 2248 |
iattr.ia_mode = S_IFREG; |
| 2249 |
break; |
| 2250 |
case KERNFS_LINK: |
| 2251 |
iattr.ia_mode = S_IFLNK; |
| 2252 |
break; |
| 2253 |
} |
| 2254 |
|
| 2255 |
ret = kernfs_setattr(kn, &iattr); |
| 2256 |
kernfs_put(kn); |
| 2257 |
return ret; |
| 2258 |
} |
| 2259 |
|
| 2260 |
/** |
| 2261 |
* rdtgroup_kn_mode_restore - Restore user access to named resctrl file |
| 2262 |
* @r: The resource group with which the file is associated. |
| 2263 |
* @name: Name of the file |
| 2264 |
* @mask: Mask of permissions that should be restored |
| 2265 |
* |
| 2266 |
* Restore the permissions of the named file. If @name is a directory the |
| 2267 |
* permissions of its parent will be used. |
| 2268 |
* |
| 2269 |
* Return: 0 on success, <0 on failure. |
| 2270 |
*/ |
| 2271 |
int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name, |
| 2272 |
umode_t mask) |
| 2273 |
{ |
| 2274 |
struct iattr iattr = {.ia_valid = ATTR_MODE,}; |
| 2275 |
struct kernfs_node *kn, *parent; |
| 2276 |
struct rftype *rfts, *rft; |
| 2277 |
int ret, len; |
| 2278 |
|
| 2279 |
rfts = res_common_files; |
| 2280 |
len = ARRAY_SIZE(res_common_files); |
| 2281 |
|
| 2282 |
for (rft = rfts; rft < rfts + len; rft++) { |
| 2283 |
if (!strcmp(rft->name, name)) |
| 2284 |
iattr.ia_mode = rft->mode & mask; |
| 2285 |
} |
| 2286 |
|
| 2287 |
kn = kernfs_find_and_get_ns(r->kn, name, NULL); |
| 2288 |
if (!kn) |
| 2289 |
return -ENOENT; |
| 2290 |
|
| 2291 |
switch (kernfs_type(kn)) { |
| 2292 |
case KERNFS_DIR: |
| 2293 |
parent = kernfs_get_parent(kn); |
| 2294 |
if (parent) { |
| 2295 |
iattr.ia_mode |= parent->mode; |
| 2296 |
kernfs_put(parent); |
| 2297 |
} |
| 2298 |
iattr.ia_mode |= S_IFDIR; |
| 2299 |
break; |
| 2300 |
case KERNFS_FILE: |
| 2301 |
iattr.ia_mode |= S_IFREG; |
| 2302 |
break; |
| 2303 |
case KERNFS_LINK: |
| 2304 |
iattr.ia_mode |= S_IFLNK; |
| 2305 |
break; |
| 2306 |
} |
| 2307 |
|
| 2308 |
ret = kernfs_setattr(kn, &iattr); |
| 2309 |
kernfs_put(kn); |
| 2310 |
return ret; |
| 2311 |
} |
| 2312 |
|
| 2313 |
static int resctrl_mkdir_event_configs(struct rdt_resource *r, struct kernfs_node *l3_mon_kn) |
| 2314 |
{ |
| 2315 |
struct kernfs_node *kn_subdir, *kn_subdir2; |
| 2316 |
struct mon_evt *mevt; |
| 2317 |
int ret; |
| 2318 |
|
| 2319 |
kn_subdir = kernfs_create_dir(l3_mon_kn, "event_configs", l3_mon_kn->mode, NULL); |
| 2320 |
if (IS_ERR(kn_subdir)) |
| 2321 |
return PTR_ERR(kn_subdir); |
| 2322 |
|
| 2323 |
ret = rdtgroup_kn_set_ugid(kn_subdir); |
| 2324 |
if (ret) |
| 2325 |
return ret; |
| 2326 |
|
| 2327 |
for_each_mon_event(mevt) { |
| 2328 |
if (mevt->rid != r->rid || !mevt->enabled || !resctrl_is_mbm_event(mevt->evtid)) |
| 2329 |
continue; |
| 2330 |
|
| 2331 |
kn_subdir2 = kernfs_create_dir(kn_subdir, mevt->name, kn_subdir->mode, mevt); |
| 2332 |
if (IS_ERR(kn_subdir2)) { |
| 2333 |
ret = PTR_ERR(kn_subdir2); |
| 2334 |
goto out; |
| 2335 |
} |
| 2336 |
|
| 2337 |
ret = rdtgroup_kn_set_ugid(kn_subdir2); |
| 2338 |
if (ret) |
| 2339 |
goto out; |
| 2340 |
|
| 2341 |
ret = rdtgroup_add_files(kn_subdir2, RFTYPE_ASSIGN_CONFIG); |
| 2342 |
if (ret) |
| 2343 |
break; |
| 2344 |
} |
| 2345 |
|
| 2346 |
out: |
| 2347 |
return ret; |
| 2348 |
} |
| 2349 |
|
| 2350 |
static int rdtgroup_mkdir_info_resdir(void *priv, char *name, |
| 2351 |
unsigned long fflags) |
| 2352 |
{ |
| 2353 |
struct kernfs_node *kn_subdir; |
| 2354 |
struct rdt_resource *r; |
| 2355 |
int ret; |
| 2356 |
|
| 2357 |
kn_subdir = kernfs_create_dir(kn_info, name, |
| 2358 |
kn_info->mode, priv); |
| 2359 |
if (IS_ERR(kn_subdir)) |
| 2360 |
return PTR_ERR(kn_subdir); |
| 2361 |
|
| 2362 |
ret = rdtgroup_kn_set_ugid(kn_subdir); |
| 2363 |
if (ret) |
| 2364 |
return ret; |
| 2365 |
|
| 2366 |
ret = rdtgroup_add_files(kn_subdir, fflags); |
| 2367 |
if (ret) |
| 2368 |
return ret; |
| 2369 |
|
| 2370 |
if ((fflags & RFTYPE_MON_INFO) == RFTYPE_MON_INFO) { |
| 2371 |
r = priv; |
| 2372 |
if (r->mon.mbm_cntr_assignable) { |
| 2373 |
ret = resctrl_mkdir_event_configs(r, kn_subdir); |
| 2374 |
if (ret) |
| 2375 |
return ret; |
| 2376 |
/* |
| 2377 |
* Hide BMEC related files if mbm_event mode |
| 2378 |
* is enabled. |
| 2379 |
*/ |
| 2380 |
if (resctrl_arch_mbm_cntr_assign_enabled(r)) |
| 2381 |
resctrl_bmec_files_show(r, kn_subdir, false); |
| 2382 |
} |
| 2383 |
} |
| 2384 |
|
| 2385 |
kernfs_activate(kn_subdir); |
| 2386 |
|
| 2387 |
return ret; |
| 2388 |
} |
| 2389 |
|
| 2390 |
static unsigned long fflags_from_resource(struct rdt_resource *r) |
| 2391 |
{ |
| 2392 |
switch (r->rid) { |
| 2393 |
case RDT_RESOURCE_L3: |
| 2394 |
case RDT_RESOURCE_L2: |
| 2395 |
return RFTYPE_RES_CACHE; |
| 2396 |
case RDT_RESOURCE_MBA: |
| 2397 |
case RDT_RESOURCE_SMBA: |
| 2398 |
return RFTYPE_RES_MB; |
| 2399 |
case RDT_RESOURCE_PERF_PKG: |
| 2400 |
return RFTYPE_RES_PERF_PKG; |
| 2401 |
} |
| 2402 |
|
| 2403 |
return WARN_ON_ONCE(1); |
| 2404 |
} |
| 2405 |
|
| 2406 |
static int rdtgroup_create_info_dir(struct kernfs_node *parent_kn) |
| 2407 |
{ |
| 2408 |
struct resctrl_schema *s; |
| 2409 |
struct rdt_resource *r; |
| 2410 |
unsigned long fflags; |
| 2411 |
char name[32]; |
| 2412 |
int ret; |
| 2413 |
|
| 2414 |
/* create the directory */ |
| 2415 |
kn_info = kernfs_create_dir(parent_kn, "info", parent_kn->mode, NULL); |
| 2416 |
if (IS_ERR(kn_info)) |
| 2417 |
return PTR_ERR(kn_info); |
| 2418 |
|
| 2419 |
ret = rdtgroup_add_files(kn_info, RFTYPE_TOP_INFO); |
| 2420 |
if (ret) |
| 2421 |
goto out_destroy; |
| 2422 |
|
| 2423 |
/* loop over enabled controls, these are all alloc_capable */ |
| 2424 |
list_for_each_entry(s, &resctrl_schema_all, list) { |
| 2425 |
r = s->res; |
| 2426 |
fflags = fflags_from_resource(r) | RFTYPE_CTRL_INFO; |
| 2427 |
ret = rdtgroup_mkdir_info_resdir(s, s->name, fflags); |
| 2428 |
if (ret) |
| 2429 |
goto out_destroy; |
| 2430 |
} |
| 2431 |
|
| 2432 |
for_each_mon_capable_rdt_resource(r) { |
| 2433 |
fflags = fflags_from_resource(r) | RFTYPE_MON_INFO; |
| 2434 |
sprintf(name, "%s_MON", r->name); |
| 2435 |
ret = rdtgroup_mkdir_info_resdir(r, name, fflags); |
| 2436 |
if (ret) |
| 2437 |
goto out_destroy; |
| 2438 |
} |
| 2439 |
|
| 2440 |
ret = rdtgroup_kn_set_ugid(kn_info); |
| 2441 |
if (ret) |
| 2442 |
goto out_destroy; |
| 2443 |
|
| 2444 |
kernfs_activate(kn_info); |
| 2445 |
|
| 2446 |
return 0; |
| 2447 |
|
| 2448 |
out_destroy: |
| 2449 |
kernfs_remove(kn_info); |
| 2450 |
return ret; |
| 2451 |
} |
| 2452 |
|
| 2453 |
static int |
| 2454 |
mongroup_create_dir(struct kernfs_node *parent_kn, struct rdtgroup *prgrp, |
| 2455 |
char *name, struct kernfs_node **dest_kn) |
| 2456 |
{ |
| 2457 |
struct kernfs_node *kn; |
| 2458 |
int ret; |
| 2459 |
|
| 2460 |
/* create the directory */ |
| 2461 |
kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); |
| 2462 |
if (IS_ERR(kn)) |
| 2463 |
return PTR_ERR(kn); |
| 2464 |
|
| 2465 |
if (dest_kn) |
| 2466 |
*dest_kn = kn; |
| 2467 |
|
| 2468 |
ret = rdtgroup_kn_set_ugid(kn); |
| 2469 |
if (ret) |
| 2470 |
goto out_destroy; |
| 2471 |
|
| 2472 |
kernfs_activate(kn); |
| 2473 |
|
| 2474 |
return 0; |
| 2475 |
|
| 2476 |
out_destroy: |
| 2477 |
kernfs_remove(kn); |
| 2478 |
return ret; |
| 2479 |
} |
| 2480 |
|
| 2481 |
static inline bool is_mba_linear(void) |
| 2482 |
{ |
| 2483 |
return resctrl_arch_get_resource(RDT_RESOURCE_MBA)->membw.delay_linear; |
| 2484 |
} |
| 2485 |
|
| 2486 |
static int mba_sc_domain_allocate(struct rdt_resource *r, struct rdt_ctrl_domain *d) |
| 2487 |
{ |
| 2488 |
u32 num_closid = resctrl_arch_get_num_closid(r); |
| 2489 |
int cpu = cpumask_any(&d->hdr.cpu_mask); |
| 2490 |
int i; |
| 2491 |
|
| 2492 |
d->mbps_val = kcalloc_node(num_closid, sizeof(*d->mbps_val), |
| 2493 |
GFP_KERNEL, cpu_to_node(cpu)); |
| 2494 |
if (!d->mbps_val) |
| 2495 |
return -ENOMEM; |
| 2496 |
|
| 2497 |
for (i = 0; i < num_closid; i++) |
| 2498 |
d->mbps_val[i] = MBA_MAX_MBPS; |
| 2499 |
|
| 2500 |
return 0; |
| 2501 |
} |
| 2502 |
|
| 2503 |
static void mba_sc_domain_destroy(struct rdt_resource *r, |
| 2504 |
struct rdt_ctrl_domain *d) |
| 2505 |
{ |
| 2506 |
kfree(d->mbps_val); |
| 2507 |
d->mbps_val = NULL; |
| 2508 |
} |
| 2509 |
|
| 2510 |
/* |
| 2511 |
* MBA software controller is supported only if |
| 2512 |
* MBM is supported and MBA is in linear scale, |
| 2513 |
* and the MBM monitor scope is the same as MBA |
| 2514 |
* control scope. |
| 2515 |
*/ |
| 2516 |
static bool supports_mba_mbps(void) |
| 2517 |
{ |
| 2518 |
struct rdt_resource *rmbm = resctrl_arch_get_resource(RDT_RESOURCE_L3); |
| 2519 |
struct rdt_resource *r = resctrl_arch_get_resource(RDT_RESOURCE_MBA); |
| 2520 |
|
| 2521 |
return (resctrl_is_mbm_enabled() && |
| 2522 |
r->alloc_capable && is_mba_linear() && |
| 2523 |
r->ctrl_scope == rmbm->mon_scope); |
| 2524 |
} |
| 2525 |
|
| 2526 |
/* |
| 2527 |
* Enable or disable the MBA software controller |
| 2528 |
* which helps user specify bandwidth in MBps. |
| 2529 |
*/ |
| 2530 |
static int set_mba_sc(bool mba_sc) |
| 2531 |
{ |
| 2532 |
struct rdt_resource *r = resctrl_arch_get_resource(RDT_RESOURCE_MBA); |
| 2533 |
u32 num_closid = resctrl_arch_get_num_closid(r); |
| 2534 |
struct rdt_ctrl_domain *d; |
| 2535 |
unsigned long fflags; |
| 2536 |
int i; |
| 2537 |
|
| 2538 |
if (!supports_mba_mbps() || mba_sc == is_mba_sc(r)) |
| 2539 |
return -EINVAL; |
| 2540 |
|
| 2541 |
r->membw.mba_sc = mba_sc; |
| 2542 |
|
| 2543 |
rdtgroup_default.mba_mbps_event = mba_mbps_default_event; |
| 2544 |
|
| 2545 |
list_for_each_entry(d, &r->ctrl_domains, hdr.list) { |
| 2546 |
for (i = 0; i < num_closid; i++) |
| 2547 |
d->mbps_val[i] = MBA_MAX_MBPS; |
| 2548 |
} |
| 2549 |
|
| 2550 |
fflags = mba_sc ? RFTYPE_CTRL_BASE | RFTYPE_MON_BASE : 0; |
| 2551 |
resctrl_file_fflags_init("mba_MBps_event", fflags); |
| 2552 |
|
| 2553 |
return 0; |
| 2554 |
} |
| 2555 |
|
| 2556 |
/* |
| 2557 |
* We don't allow rdtgroup directories to be created anywhere |
| 2558 |
* except the root directory. Thus when looking for the rdtgroup |
| 2559 |
* structure for a kernfs node we are either looking at a directory, |
| 2560 |
* in which case the rdtgroup structure is pointed at by the "priv" |
| 2561 |
* field, otherwise we have a file, and need only look to the parent |
| 2562 |
* to find the rdtgroup. |
| 2563 |
*/ |
| 2564 |
static struct rdtgroup *kernfs_to_rdtgroup(struct kernfs_node *kn) |
| 2565 |
{ |
| 2566 |
if (kernfs_type(kn) == KERNFS_DIR) { |
| 2567 |
/* |
| 2568 |
* All the resource directories use "kn->priv" |
| 2569 |
* to point to the "struct rdtgroup" for the |
| 2570 |
* resource. "info" and its subdirectories don't |
| 2571 |
* have rdtgroup structures, so return NULL here. |
| 2572 |
*/ |
| 2573 |
if (kn == kn_info || |
| 2574 |
rcu_access_pointer(kn->__parent) == kn_info) |
| 2575 |
return NULL; |
| 2576 |
else |
| 2577 |
return kn->priv; |
| 2578 |
} else { |
| 2579 |
return rdt_kn_parent_priv(kn); |
| 2580 |
} |
| 2581 |
} |
| 2582 |
|
| 2583 |
static void rdtgroup_kn_get(struct rdtgroup *rdtgrp, struct kernfs_node *kn) |
| 2584 |
{ |
| 2585 |
atomic_inc(&rdtgrp->waitcount); |
| 2586 |
kernfs_break_active_protection(kn); |
| 2587 |
} |
| 2588 |
|
| 2589 |
static void rdtgroup_kn_put(struct rdtgroup *rdtgrp, struct kernfs_node *kn) |
| 2590 |
{ |
| 2591 |
if (atomic_dec_and_test(&rdtgrp->waitcount) && |
| 2592 |
(rdtgrp->flags & RDT_DELETED)) { |
| 2593 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
| 2594 |
rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) |
| 2595 |
rdtgroup_pseudo_lock_remove(rdtgrp); |
| 2596 |
kernfs_unbreak_active_protection(kn); |
| 2597 |
rdtgroup_remove(rdtgrp); |
| 2598 |
} else { |
| 2599 |
kernfs_unbreak_active_protection(kn); |
| 2600 |
} |
| 2601 |
} |
| 2602 |
|
| 2603 |
struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn) |
| 2604 |
{ |
| 2605 |
struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); |
| 2606 |
|
| 2607 |
if (!rdtgrp) |
| 2608 |
return NULL; |
| 2609 |
|
| 2610 |
rdtgroup_kn_get(rdtgrp, kn); |
| 2611 |
|
| 2612 |
cpus_read_lock(); |
| 2613 |
mutex_lock(&rdtgroup_mutex); |
| 2614 |
|
| 2615 |
/* Was this group deleted while we waited? */ |
| 2616 |
if (rdtgrp->flags & RDT_DELETED) |
| 2617 |
return NULL; |
| 2618 |
|
| 2619 |
return rdtgrp; |
| 2620 |
} |
| 2621 |
|
| 2622 |
void rdtgroup_kn_unlock(struct kernfs_node *kn) |
| 2623 |
{ |
| 2624 |
struct rdtgroup *rdtgrp = kernfs_to_rdtgroup(kn); |
| 2625 |
|
| 2626 |
if (!rdtgrp) |
| 2627 |
return; |
| 2628 |
|
| 2629 |
mutex_unlock(&rdtgroup_mutex); |
| 2630 |
cpus_read_unlock(); |
| 2631 |
|
| 2632 |
rdtgroup_kn_put(rdtgrp, kn); |
| 2633 |
} |
| 2634 |
|
| 2635 |
static int mkdir_mondata_all(struct kernfs_node *parent_kn, |
| 2636 |
struct rdtgroup *prgrp, |
| 2637 |
struct kernfs_node **mon_data_kn); |
| 2638 |
|
| 2639 |
static void rdt_disable_ctx(void) |
| 2640 |
{ |
| 2641 |
resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L3, false); |
| 2642 |
resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L2, false); |
| 2643 |
set_mba_sc(false); |
| 2644 |
|
| 2645 |
resctrl_debug = false; |
| 2646 |
} |
| 2647 |
|
| 2648 |
static int rdt_enable_ctx(struct rdt_fs_context *ctx) |
| 2649 |
{ |
| 2650 |
int ret = 0; |
| 2651 |
|
| 2652 |
if (ctx->enable_cdpl2) { |
| 2653 |
ret = resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L2, true); |
| 2654 |
if (ret) |
| 2655 |
goto out_done; |
| 2656 |
} |
| 2657 |
|
| 2658 |
if (ctx->enable_cdpl3) { |
| 2659 |
ret = resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L3, true); |
| 2660 |
if (ret) |
| 2661 |
goto out_cdpl2; |
| 2662 |
} |
| 2663 |
|
| 2664 |
if (ctx->enable_mba_mbps) { |
| 2665 |
ret = set_mba_sc(true); |
| 2666 |
if (ret) |
| 2667 |
goto out_cdpl3; |
| 2668 |
} |
| 2669 |
|
| 2670 |
if (ctx->enable_debug) |
| 2671 |
resctrl_debug = true; |
| 2672 |
|
| 2673 |
return 0; |
| 2674 |
|
| 2675 |
out_cdpl3: |
| 2676 |
resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L3, false); |
| 2677 |
out_cdpl2: |
| 2678 |
resctrl_arch_set_cdp_enabled(RDT_RESOURCE_L2, false); |
| 2679 |
out_done: |
| 2680 |
return ret; |
| 2681 |
} |
| 2682 |
|
| 2683 |
static int schemata_list_add(struct rdt_resource *r, enum resctrl_conf_type type) |
| 2684 |
{ |
| 2685 |
struct resctrl_schema *s; |
| 2686 |
const char *suffix = ""; |
| 2687 |
int ret, cl; |
| 2688 |
|
| 2689 |
s = kzalloc_obj(*s); |
| 2690 |
if (!s) |
| 2691 |
return -ENOMEM; |
| 2692 |
|
| 2693 |
s->res = r; |
| 2694 |
s->num_closid = resctrl_arch_get_num_closid(r); |
| 2695 |
if (resctrl_arch_get_cdp_enabled(r->rid)) |
| 2696 |
s->num_closid /= 2; |
| 2697 |
|
| 2698 |
s->conf_type = type; |
| 2699 |
switch (type) { |
| 2700 |
case CDP_CODE: |
| 2701 |
suffix = "CODE"; |
| 2702 |
break; |
| 2703 |
case CDP_DATA: |
| 2704 |
suffix = "DATA"; |
| 2705 |
break; |
| 2706 |
case CDP_NONE: |
| 2707 |
suffix = ""; |
| 2708 |
break; |
| 2709 |
} |
| 2710 |
|
| 2711 |
ret = snprintf(s->name, sizeof(s->name), "%s%s", r->name, suffix); |
| 2712 |
if (ret >= sizeof(s->name)) { |
| 2713 |
kfree(s); |
| 2714 |
return -EINVAL; |
| 2715 |
} |
| 2716 |
|
| 2717 |
cl = strlen(s->name); |
| 2718 |
|
| 2719 |
/* |
| 2720 |
* If CDP is supported by this resource, but not enabled, |
| 2721 |
* include the suffix. This ensures the tabular format of the |
| 2722 |
* schemata file does not change between mounts of the filesystem. |
| 2723 |
*/ |
| 2724 |
if (r->cdp_capable && !resctrl_arch_get_cdp_enabled(r->rid)) |
| 2725 |
cl += 4; |
| 2726 |
|
| 2727 |
if (cl > max_name_width) |
| 2728 |
max_name_width = cl; |
| 2729 |
|
| 2730 |
switch (r->schema_fmt) { |
| 2731 |
case RESCTRL_SCHEMA_BITMAP: |
| 2732 |
s->fmt_str = "%d=%x"; |
| 2733 |
break; |
| 2734 |
case RESCTRL_SCHEMA_RANGE: |
| 2735 |
s->fmt_str = "%d=%u"; |
| 2736 |
break; |
| 2737 |
} |
| 2738 |
|
| 2739 |
if (WARN_ON_ONCE(!s->fmt_str)) { |
| 2740 |
kfree(s); |
| 2741 |
return -EINVAL; |
| 2742 |
} |
| 2743 |
|
| 2744 |
INIT_LIST_HEAD(&s->list); |
| 2745 |
list_add(&s->list, &resctrl_schema_all); |
| 2746 |
|
| 2747 |
return 0; |
| 2748 |
} |
| 2749 |
|
| 2750 |
static int schemata_list_create(void) |
| 2751 |
{ |
| 2752 |
struct rdt_resource *r; |
| 2753 |
int ret = 0; |
| 2754 |
|
| 2755 |
for_each_alloc_capable_rdt_resource(r) { |
| 2756 |
if (resctrl_arch_get_cdp_enabled(r->rid)) { |
| 2757 |
ret = schemata_list_add(r, CDP_CODE); |
| 2758 |
if (ret) |
| 2759 |
break; |
| 2760 |
|
| 2761 |
ret = schemata_list_add(r, CDP_DATA); |
| 2762 |
} else { |
| 2763 |
ret = schemata_list_add(r, CDP_NONE); |
| 2764 |
} |
| 2765 |
|
| 2766 |
if (ret) |
| 2767 |
break; |
| 2768 |
} |
| 2769 |
|
| 2770 |
return ret; |
| 2771 |
} |
| 2772 |
|
| 2773 |
static void schemata_list_destroy(void) |
| 2774 |
{ |
| 2775 |
struct resctrl_schema *s, *tmp; |
| 2776 |
|
| 2777 |
list_for_each_entry_safe(s, tmp, &resctrl_schema_all, list) { |
| 2778 |
list_del(&s->list); |
| 2779 |
kfree(s); |
| 2780 |
} |
| 2781 |
} |
| 2782 |
|
| 2783 |
static int rdt_get_tree(struct fs_context *fc) |
| 2784 |
{ |
| 2785 |
struct rdt_fs_context *ctx = rdt_fc2context(fc); |
| 2786 |
unsigned long flags = RFTYPE_CTRL_BASE; |
| 2787 |
struct rdt_l3_mon_domain *dom; |
| 2788 |
struct rdt_resource *r; |
| 2789 |
int ret; |
| 2790 |
|
| 2791 |
DO_ONCE_SLEEPABLE(resctrl_arch_pre_mount); |
| 2792 |
|
| 2793 |
cpus_read_lock(); |
| 2794 |
mutex_lock(&rdtgroup_mutex); |
| 2795 |
/* |
| 2796 |
* resctrl file system can only be mounted once. |
| 2797 |
*/ |
| 2798 |
if (resctrl_mounted) { |
| 2799 |
ret = -EBUSY; |
| 2800 |
goto out; |
| 2801 |
} |
| 2802 |
|
| 2803 |
ret = setup_rmid_lru_list(); |
| 2804 |
if (ret) |
| 2805 |
goto out; |
| 2806 |
|
| 2807 |
ret = rdtgroup_setup_root(ctx); |
| 2808 |
if (ret) |
| 2809 |
goto out; |
| 2810 |
|
| 2811 |
ret = rdt_enable_ctx(ctx); |
| 2812 |
if (ret) |
| 2813 |
goto out_root; |
| 2814 |
|
| 2815 |
ret = schemata_list_create(); |
| 2816 |
if (ret) |
| 2817 |
goto out_schemata_free; |
| 2818 |
|
| 2819 |
ret = closid_init(); |
| 2820 |
if (ret) |
| 2821 |
goto out_schemata_free; |
| 2822 |
|
| 2823 |
if (resctrl_arch_mon_capable()) |
| 2824 |
flags |= RFTYPE_MON; |
| 2825 |
|
| 2826 |
ret = rdtgroup_add_files(rdtgroup_default.kn, flags); |
| 2827 |
if (ret) |
| 2828 |
goto out_closid_exit; |
| 2829 |
|
| 2830 |
kernfs_activate(rdtgroup_default.kn); |
| 2831 |
|
| 2832 |
ret = rdtgroup_create_info_dir(rdtgroup_default.kn); |
| 2833 |
if (ret < 0) |
| 2834 |
goto out_closid_exit; |
| 2835 |
|
| 2836 |
if (resctrl_arch_mon_capable()) { |
| 2837 |
ret = mongroup_create_dir(rdtgroup_default.kn, |
| 2838 |
&rdtgroup_default, "mon_groups", |
| 2839 |
&kn_mongrp); |
| 2840 |
if (ret < 0) |
| 2841 |
goto out_info; |
| 2842 |
|
| 2843 |
rdtgroup_assign_cntrs(&rdtgroup_default); |
| 2844 |
|
| 2845 |
ret = mkdir_mondata_all(rdtgroup_default.kn, |
| 2846 |
&rdtgroup_default, &kn_mondata); |
| 2847 |
if (ret < 0) |
| 2848 |
goto out_mongrp; |
| 2849 |
rdtgroup_default.mon.mon_data_kn = kn_mondata; |
| 2850 |
} |
| 2851 |
|
| 2852 |
ret = rdt_pseudo_lock_init(); |
| 2853 |
if (ret) |
| 2854 |
goto out_mondata; |
| 2855 |
|
| 2856 |
ret = kernfs_get_tree(fc); |
| 2857 |
if (ret < 0) |
| 2858 |
goto out_psl; |
| 2859 |
|
| 2860 |
if (resctrl_arch_alloc_capable()) |
| 2861 |
resctrl_arch_enable_alloc(); |
| 2862 |
if (resctrl_arch_mon_capable()) |
| 2863 |
resctrl_arch_enable_mon(); |
| 2864 |
|
| 2865 |
if (resctrl_arch_alloc_capable() || resctrl_arch_mon_capable()) |
| 2866 |
resctrl_mounted = true; |
| 2867 |
|
| 2868 |
if (resctrl_is_mbm_enabled()) { |
| 2869 |
r = resctrl_arch_get_resource(RDT_RESOURCE_L3); |
| 2870 |
list_for_each_entry(dom, &r->mon_domains, hdr.list) |
| 2871 |
mbm_setup_overflow_handler(dom, MBM_OVERFLOW_INTERVAL, |
| 2872 |
RESCTRL_PICK_ANY_CPU); |
| 2873 |
} |
| 2874 |
|
| 2875 |
goto out; |
| 2876 |
|
| 2877 |
out_psl: |
| 2878 |
rdt_pseudo_lock_release(); |
| 2879 |
out_mondata: |
| 2880 |
if (resctrl_arch_mon_capable()) |
| 2881 |
kernfs_remove(kn_mondata); |
| 2882 |
out_mongrp: |
| 2883 |
if (resctrl_arch_mon_capable()) { |
| 2884 |
rdtgroup_unassign_cntrs(&rdtgroup_default); |
| 2885 |
kernfs_remove(kn_mongrp); |
| 2886 |
} |
| 2887 |
out_info: |
| 2888 |
kernfs_remove(kn_info); |
| 2889 |
out_closid_exit: |
| 2890 |
closid_exit(); |
| 2891 |
out_schemata_free: |
| 2892 |
schemata_list_destroy(); |
| 2893 |
rdt_disable_ctx(); |
| 2894 |
out_root: |
| 2895 |
rdtgroup_destroy_root(); |
| 2896 |
out: |
| 2897 |
rdt_last_cmd_clear(); |
| 2898 |
mutex_unlock(&rdtgroup_mutex); |
| 2899 |
cpus_read_unlock(); |
| 2900 |
return ret; |
| 2901 |
} |
| 2902 |
|
| 2903 |
enum rdt_param { |
| 2904 |
Opt_cdp, |
| 2905 |
Opt_cdpl2, |
| 2906 |
Opt_mba_mbps, |
| 2907 |
Opt_debug, |
| 2908 |
nr__rdt_params |
| 2909 |
}; |
| 2910 |
|
| 2911 |
static const struct fs_parameter_spec rdt_fs_parameters[] = { |
| 2912 |
fsparam_flag("cdp", Opt_cdp), |
| 2913 |
fsparam_flag("cdpl2", Opt_cdpl2), |
| 2914 |
fsparam_flag("mba_MBps", Opt_mba_mbps), |
| 2915 |
fsparam_flag("debug", Opt_debug), |
| 2916 |
{} |
| 2917 |
}; |
| 2918 |
|
| 2919 |
static int rdt_parse_param(struct fs_context *fc, struct fs_parameter *param) |
| 2920 |
{ |
| 2921 |
struct rdt_fs_context *ctx = rdt_fc2context(fc); |
| 2922 |
struct fs_parse_result result; |
| 2923 |
const char *msg; |
| 2924 |
int opt; |
| 2925 |
|
| 2926 |
opt = fs_parse(fc, rdt_fs_parameters, param, &result); |
| 2927 |
if (opt < 0) |
| 2928 |
return opt; |
| 2929 |
|
| 2930 |
switch (opt) { |
| 2931 |
case Opt_cdp: |
| 2932 |
ctx->enable_cdpl3 = true; |
| 2933 |
return 0; |
| 2934 |
case Opt_cdpl2: |
| 2935 |
ctx->enable_cdpl2 = true; |
| 2936 |
return 0; |
| 2937 |
case Opt_mba_mbps: |
| 2938 |
msg = "mba_MBps requires MBM and linear scale MBA at L3 scope"; |
| 2939 |
if (!supports_mba_mbps()) |
| 2940 |
return invalfc(fc, msg); |
| 2941 |
ctx->enable_mba_mbps = true; |
| 2942 |
return 0; |
| 2943 |
case Opt_debug: |
| 2944 |
ctx->enable_debug = true; |
| 2945 |
return 0; |
| 2946 |
} |
| 2947 |
|
| 2948 |
return -EINVAL; |
| 2949 |
} |
| 2950 |
|
| 2951 |
static void rdt_fs_context_free(struct fs_context *fc) |
| 2952 |
{ |
| 2953 |
struct rdt_fs_context *ctx = rdt_fc2context(fc); |
| 2954 |
|
| 2955 |
kernfs_free_fs_context(fc); |
| 2956 |
kfree(ctx); |
| 2957 |
} |
| 2958 |
|
| 2959 |
static const struct fs_context_operations rdt_fs_context_ops = { |
| 2960 |
.free = rdt_fs_context_free, |
| 2961 |
.parse_param = rdt_parse_param, |
| 2962 |
.get_tree = rdt_get_tree, |
| 2963 |
}; |
| 2964 |
|
| 2965 |
static int rdt_init_fs_context(struct fs_context *fc) |
| 2966 |
{ |
| 2967 |
struct rdt_fs_context *ctx; |
| 2968 |
|
| 2969 |
ctx = kzalloc_obj(*ctx); |
| 2970 |
if (!ctx) |
| 2971 |
return -ENOMEM; |
| 2972 |
|
| 2973 |
ctx->kfc.magic = RDTGROUP_SUPER_MAGIC; |
| 2974 |
fc->fs_private = &ctx->kfc; |
| 2975 |
fc->ops = &rdt_fs_context_ops; |
| 2976 |
put_user_ns(fc->user_ns); |
| 2977 |
fc->user_ns = get_user_ns(&init_user_ns); |
| 2978 |
fc->global = true; |
| 2979 |
return 0; |
| 2980 |
} |
| 2981 |
|
| 2982 |
/* |
| 2983 |
* Move tasks from one to the other group. If @from is NULL, then all tasks |
| 2984 |
* in the systems are moved unconditionally (used for teardown). |
| 2985 |
* |
| 2986 |
* If @mask is not NULL the cpus on which moved tasks are running are set |
| 2987 |
* in that mask so the update smp function call is restricted to affected |
| 2988 |
* cpus. |
| 2989 |
*/ |
| 2990 |
static void rdt_move_group_tasks(struct rdtgroup *from, struct rdtgroup *to, |
| 2991 |
struct cpumask *mask) |
| 2992 |
{ |
| 2993 |
struct task_struct *p, *t; |
| 2994 |
|
| 2995 |
read_lock(&tasklist_lock); |
| 2996 |
for_each_process_thread(p, t) { |
| 2997 |
if (!from || is_closid_match(t, from) || |
| 2998 |
is_rmid_match(t, from)) { |
| 2999 |
resctrl_arch_set_closid_rmid(t, to->closid, |
| 3000 |
to->mon.rmid); |
| 3001 |
|
| 3002 |
/* |
| 3003 |
* Order the closid/rmid stores above before the loads |
| 3004 |
* in task_curr(). This pairs with the full barrier |
| 3005 |
* between the rq->curr update and |
| 3006 |
* resctrl_arch_sched_in() during context switch. |
| 3007 |
*/ |
| 3008 |
smp_mb(); |
| 3009 |
|
| 3010 |
/* |
| 3011 |
* If the task is on a CPU, set the CPU in the mask. |
| 3012 |
* The detection is inaccurate as tasks might move or |
| 3013 |
* schedule before the smp function call takes place. |
| 3014 |
* In such a case the function call is pointless, but |
| 3015 |
* there is no other side effect. |
| 3016 |
*/ |
| 3017 |
if (IS_ENABLED(CONFIG_SMP) && mask && task_curr(t)) |
| 3018 |
cpumask_set_cpu(task_cpu(t), mask); |
| 3019 |
} |
| 3020 |
} |
| 3021 |
read_unlock(&tasklist_lock); |
| 3022 |
} |
| 3023 |
|
| 3024 |
static void free_all_child_rdtgrp(struct rdtgroup *rdtgrp) |
| 3025 |
{ |
| 3026 |
struct rdtgroup *sentry, *stmp; |
| 3027 |
struct list_head *head; |
| 3028 |
|
| 3029 |
head = &rdtgrp->mon.crdtgrp_list; |
| 3030 |
list_for_each_entry_safe(sentry, stmp, head, mon.crdtgrp_list) { |
| 3031 |
rdtgroup_unassign_cntrs(sentry); |
| 3032 |
free_rmid(sentry->closid, sentry->mon.rmid); |
| 3033 |
list_del(&sentry->mon.crdtgrp_list); |
| 3034 |
|
| 3035 |
if (atomic_read(&sentry->waitcount) != 0) |
| 3036 |
sentry->flags = RDT_DELETED; |
| 3037 |
else |
| 3038 |
rdtgroup_remove(sentry); |
| 3039 |
} |
| 3040 |
} |
| 3041 |
|
| 3042 |
/* |
| 3043 |
* Forcibly remove all of subdirectories under root. |
| 3044 |
*/ |
| 3045 |
static void rmdir_all_sub(void) |
| 3046 |
{ |
| 3047 |
struct rdtgroup *rdtgrp, *tmp; |
| 3048 |
|
| 3049 |
/* Move all tasks to the default resource group */ |
| 3050 |
rdt_move_group_tasks(NULL, &rdtgroup_default, NULL); |
| 3051 |
|
| 3052 |
list_for_each_entry_safe(rdtgrp, tmp, &rdt_all_groups, rdtgroup_list) { |
| 3053 |
/* Free any child rmids */ |
| 3054 |
free_all_child_rdtgrp(rdtgrp); |
| 3055 |
|
| 3056 |
/* Remove each rdtgroup other than root */ |
| 3057 |
if (rdtgrp == &rdtgroup_default) |
| 3058 |
continue; |
| 3059 |
|
| 3060 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
| 3061 |
rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) |
| 3062 |
rdtgroup_pseudo_lock_remove(rdtgrp); |
| 3063 |
|
| 3064 |
/* |
| 3065 |
* Give any CPUs back to the default group. We cannot copy |
| 3066 |
* cpu_online_mask because a CPU might have executed the |
| 3067 |
* offline callback already, but is still marked online. |
| 3068 |
*/ |
| 3069 |
cpumask_or(&rdtgroup_default.cpu_mask, |
| 3070 |
&rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); |
| 3071 |
|
| 3072 |
rdtgroup_unassign_cntrs(rdtgrp); |
| 3073 |
|
| 3074 |
free_rmid(rdtgrp->closid, rdtgrp->mon.rmid); |
| 3075 |
|
| 3076 |
kernfs_remove(rdtgrp->kn); |
| 3077 |
list_del(&rdtgrp->rdtgroup_list); |
| 3078 |
|
| 3079 |
if (atomic_read(&rdtgrp->waitcount) != 0) |
| 3080 |
rdtgrp->flags = RDT_DELETED; |
| 3081 |
else |
| 3082 |
rdtgroup_remove(rdtgrp); |
| 3083 |
} |
| 3084 |
/* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */ |
| 3085 |
update_closid_rmid(cpu_online_mask, &rdtgroup_default); |
| 3086 |
|
| 3087 |
kernfs_remove(kn_info); |
| 3088 |
kernfs_remove(kn_mongrp); |
| 3089 |
kernfs_remove(kn_mondata); |
| 3090 |
} |
| 3091 |
|
| 3092 |
/** |
| 3093 |
* mon_get_kn_priv() - Get the mon_data priv data for this event. |
| 3094 |
* |
| 3095 |
* The same values are used across the mon_data directories of all control and |
| 3096 |
* monitor groups for the same event in the same domain. Keep a list of |
| 3097 |
* allocated structures and re-use an existing one with the same values for |
| 3098 |
* @rid, @domid, etc. |
| 3099 |
* |
| 3100 |
* @rid: The resource id for the event file being created. |
| 3101 |
* @domid: The domain id for the event file being created. |
| 3102 |
* @mevt: The type of event file being created. |
| 3103 |
* @do_sum: Whether SNC summing monitors are being created. Only set |
| 3104 |
* when @rid == RDT_RESOURCE_L3. |
| 3105 |
*/ |
| 3106 |
static struct mon_data *mon_get_kn_priv(enum resctrl_res_level rid, int domid, |
| 3107 |
struct mon_evt *mevt, |
| 3108 |
bool do_sum) |
| 3109 |
{ |
| 3110 |
struct mon_data *priv; |
| 3111 |
|
| 3112 |
lockdep_assert_held(&rdtgroup_mutex); |
| 3113 |
|
| 3114 |
list_for_each_entry(priv, &mon_data_kn_priv_list, list) { |
| 3115 |
if (priv->rid == rid && priv->domid == domid && |
| 3116 |
priv->sum == do_sum && priv->evt == mevt) |
| 3117 |
return priv; |
| 3118 |
} |
| 3119 |
|
| 3120 |
priv = kzalloc_obj(*priv); |
| 3121 |
if (!priv) |
| 3122 |
return NULL; |
| 3123 |
|
| 3124 |
priv->rid = rid; |
| 3125 |
priv->domid = domid; |
| 3126 |
priv->sum = do_sum; |
| 3127 |
priv->evt = mevt; |
| 3128 |
list_add_tail(&priv->list, &mon_data_kn_priv_list); |
| 3129 |
|
| 3130 |
return priv; |
| 3131 |
} |
| 3132 |
|
| 3133 |
/** |
| 3134 |
* mon_put_kn_priv() - Free all allocated mon_data structures. |
| 3135 |
* |
| 3136 |
* Called when resctrl file system is unmounted. |
| 3137 |
*/ |
| 3138 |
static void mon_put_kn_priv(void) |
| 3139 |
{ |
| 3140 |
struct mon_data *priv, *tmp; |
| 3141 |
|
| 3142 |
lockdep_assert_held(&rdtgroup_mutex); |
| 3143 |
|
| 3144 |
list_for_each_entry_safe(priv, tmp, &mon_data_kn_priv_list, list) { |
| 3145 |
list_del(&priv->list); |
| 3146 |
kfree(priv); |
| 3147 |
} |
| 3148 |
} |
| 3149 |
|
| 3150 |
static void resctrl_fs_teardown(void) |
| 3151 |
{ |
| 3152 |
lockdep_assert_held(&rdtgroup_mutex); |
| 3153 |
|
| 3154 |
/* Cleared by rdtgroup_destroy_root() */ |
| 3155 |
if (!rdtgroup_default.kn) |
| 3156 |
return; |
| 3157 |
|
| 3158 |
rmdir_all_sub(); |
| 3159 |
rdtgroup_unassign_cntrs(&rdtgroup_default); |
| 3160 |
mon_put_kn_priv(); |
| 3161 |
rdt_pseudo_lock_release(); |
| 3162 |
rdtgroup_default.mode = RDT_MODE_SHAREABLE; |
| 3163 |
closid_exit(); |
| 3164 |
schemata_list_destroy(); |
| 3165 |
rdtgroup_destroy_root(); |
| 3166 |
} |
| 3167 |
|
| 3168 |
static void rdt_kill_sb(struct super_block *sb) |
| 3169 |
{ |
| 3170 |
struct rdt_resource *r; |
| 3171 |
|
| 3172 |
cpus_read_lock(); |
| 3173 |
mutex_lock(&rdtgroup_mutex); |
| 3174 |
|
| 3175 |
rdt_disable_ctx(); |
| 3176 |
|
| 3177 |
/* Put everything back to default values. */ |
| 3178 |
for_each_alloc_capable_rdt_resource(r) |
| 3179 |
resctrl_arch_reset_all_ctrls(r); |
| 3180 |
|
| 3181 |
resctrl_fs_teardown(); |
| 3182 |
if (resctrl_arch_alloc_capable()) |
| 3183 |
resctrl_arch_disable_alloc(); |
| 3184 |
if (resctrl_arch_mon_capable()) |
| 3185 |
resctrl_arch_disable_mon(); |
| 3186 |
resctrl_mounted = false; |
| 3187 |
kernfs_kill_sb(sb); |
| 3188 |
mutex_unlock(&rdtgroup_mutex); |
| 3189 |
cpus_read_unlock(); |
| 3190 |
} |
| 3191 |
|
| 3192 |
static struct file_system_type rdt_fs_type = { |
| 3193 |
.name = "resctrl", |
| 3194 |
.init_fs_context = rdt_init_fs_context, |
| 3195 |
.parameters = rdt_fs_parameters, |
| 3196 |
.kill_sb = rdt_kill_sb, |
| 3197 |
}; |
| 3198 |
|
| 3199 |
static int mon_addfile(struct kernfs_node *parent_kn, const char *name, |
| 3200 |
void *priv) |
| 3201 |
{ |
| 3202 |
struct kernfs_node *kn; |
| 3203 |
int ret = 0; |
| 3204 |
|
| 3205 |
kn = __kernfs_create_file(parent_kn, name, 0444, |
| 3206 |
GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, 0, |
| 3207 |
&kf_mondata_ops, priv, NULL, NULL); |
| 3208 |
if (IS_ERR(kn)) |
| 3209 |
return PTR_ERR(kn); |
| 3210 |
|
| 3211 |
ret = rdtgroup_kn_set_ugid(kn); |
| 3212 |
if (ret) { |
| 3213 |
kernfs_remove(kn); |
| 3214 |
return ret; |
| 3215 |
} |
| 3216 |
|
| 3217 |
return ret; |
| 3218 |
} |
| 3219 |
|
| 3220 |
static void mon_rmdir_one_subdir(struct kernfs_node *pkn, char *name, char *subname) |
| 3221 |
{ |
| 3222 |
struct kernfs_node *kn; |
| 3223 |
|
| 3224 |
kn = kernfs_find_and_get(pkn, name); |
| 3225 |
if (!kn) |
| 3226 |
return; |
| 3227 |
kernfs_put(kn); |
| 3228 |
|
| 3229 |
if (kn->dir.subdirs <= 1) |
| 3230 |
kernfs_remove(kn); |
| 3231 |
else |
| 3232 |
kernfs_remove_by_name(kn, subname); |
| 3233 |
} |
| 3234 |
|
| 3235 |
/* |
| 3236 |
* Remove files and directories for one SNC node. If it is the last node |
| 3237 |
* sharing an L3 cache, then remove the upper level directory containing |
| 3238 |
* the "sum" files too. |
| 3239 |
*/ |
| 3240 |
static void rmdir_mondata_subdir_allrdtgrp_snc(struct rdt_resource *r, |
| 3241 |
struct rdt_domain_hdr *hdr) |
| 3242 |
{ |
| 3243 |
struct rdtgroup *prgrp, *crgrp; |
| 3244 |
struct rdt_l3_mon_domain *d; |
| 3245 |
char subname[32]; |
| 3246 |
char name[32]; |
| 3247 |
|
| 3248 |
if (!domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3)) |
| 3249 |
return; |
| 3250 |
|
| 3251 |
d = container_of(hdr, struct rdt_l3_mon_domain, hdr); |
| 3252 |
sprintf(name, "mon_%s_%02d", r->name, d->ci_id); |
| 3253 |
sprintf(subname, "mon_sub_%s_%02d", r->name, hdr->id); |
| 3254 |
|
| 3255 |
list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { |
| 3256 |
mon_rmdir_one_subdir(prgrp->mon.mon_data_kn, name, subname); |
| 3257 |
|
| 3258 |
list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list) |
| 3259 |
mon_rmdir_one_subdir(crgrp->mon.mon_data_kn, name, subname); |
| 3260 |
} |
| 3261 |
} |
| 3262 |
|
| 3263 |
/* |
| 3264 |
* Remove all subdirectories of mon_data of ctrl_mon groups |
| 3265 |
* and monitor groups for the given domain. |
| 3266 |
*/ |
| 3267 |
static void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, |
| 3268 |
struct rdt_domain_hdr *hdr) |
| 3269 |
{ |
| 3270 |
struct rdtgroup *prgrp, *crgrp; |
| 3271 |
char name[32]; |
| 3272 |
|
| 3273 |
if (r->rid == RDT_RESOURCE_L3 && r->mon_scope == RESCTRL_L3_NODE) { |
| 3274 |
rmdir_mondata_subdir_allrdtgrp_snc(r, hdr); |
| 3275 |
return; |
| 3276 |
} |
| 3277 |
|
| 3278 |
sprintf(name, "mon_%s_%02d", r->name, hdr->id); |
| 3279 |
list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { |
| 3280 |
kernfs_remove_by_name(prgrp->mon.mon_data_kn, name); |
| 3281 |
|
| 3282 |
list_for_each_entry(crgrp, &prgrp->mon.crdtgrp_list, mon.crdtgrp_list) |
| 3283 |
kernfs_remove_by_name(crgrp->mon.mon_data_kn, name); |
| 3284 |
} |
| 3285 |
} |
| 3286 |
|
| 3287 |
/* |
| 3288 |
* Create a directory for a domain and populate it with monitor files. Create |
| 3289 |
* summing monitors when @hdr is NULL. No need to initialize summing monitors. |
| 3290 |
*/ |
| 3291 |
static struct kernfs_node *_mkdir_mondata_subdir(struct kernfs_node *parent_kn, char *name, |
| 3292 |
struct rdt_domain_hdr *hdr, |
| 3293 |
struct rdt_resource *r, |
| 3294 |
struct rdtgroup *prgrp, int domid) |
| 3295 |
{ |
| 3296 |
struct rmid_read rr = {0}; |
| 3297 |
struct kernfs_node *kn; |
| 3298 |
struct mon_data *priv; |
| 3299 |
struct mon_evt *mevt; |
| 3300 |
int ret; |
| 3301 |
|
| 3302 |
kn = kernfs_create_dir(parent_kn, name, parent_kn->mode, prgrp); |
| 3303 |
if (IS_ERR(kn)) |
| 3304 |
return kn; |
| 3305 |
|
| 3306 |
ret = rdtgroup_kn_set_ugid(kn); |
| 3307 |
if (ret) |
| 3308 |
goto out_destroy; |
| 3309 |
|
| 3310 |
for_each_mon_event(mevt) { |
| 3311 |
if (mevt->rid != r->rid || !mevt->enabled) |
| 3312 |
continue; |
| 3313 |
priv = mon_get_kn_priv(r->rid, domid, mevt, !hdr); |
| 3314 |
if (WARN_ON_ONCE(!priv)) { |
| 3315 |
ret = -EINVAL; |
| 3316 |
goto out_destroy; |
| 3317 |
} |
| 3318 |
|
| 3319 |
ret = mon_addfile(kn, mevt->name, priv); |
| 3320 |
if (ret) |
| 3321 |
goto out_destroy; |
| 3322 |
|
| 3323 |
if (hdr && resctrl_is_mbm_event(mevt->evtid)) |
| 3324 |
mon_event_read(&rr, r, hdr, prgrp, &hdr->cpu_mask, mevt, true); |
| 3325 |
} |
| 3326 |
|
| 3327 |
return kn; |
| 3328 |
out_destroy: |
| 3329 |
kernfs_remove(kn); |
| 3330 |
return ERR_PTR(ret); |
| 3331 |
} |
| 3332 |
|
| 3333 |
static int mkdir_mondata_subdir_snc(struct kernfs_node *parent_kn, |
| 3334 |
struct rdt_domain_hdr *hdr, |
| 3335 |
struct rdt_resource *r, struct rdtgroup *prgrp) |
| 3336 |
{ |
| 3337 |
struct kernfs_node *ckn, *kn; |
| 3338 |
struct rdt_l3_mon_domain *d; |
| 3339 |
char name[32]; |
| 3340 |
|
| 3341 |
if (!domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3)) |
| 3342 |
return -EINVAL; |
| 3343 |
|
| 3344 |
d = container_of(hdr, struct rdt_l3_mon_domain, hdr); |
| 3345 |
sprintf(name, "mon_%s_%02d", r->name, d->ci_id); |
| 3346 |
kn = kernfs_find_and_get(parent_kn, name); |
| 3347 |
if (kn) { |
| 3348 |
/* |
| 3349 |
* rdtgroup_mutex will prevent this directory from being |
| 3350 |
* removed. No need to keep this hold. |
| 3351 |
*/ |
| 3352 |
kernfs_put(kn); |
| 3353 |
} else { |
| 3354 |
kn = _mkdir_mondata_subdir(parent_kn, name, NULL, r, prgrp, d->ci_id); |
| 3355 |
if (IS_ERR(kn)) |
| 3356 |
return PTR_ERR(kn); |
| 3357 |
} |
| 3358 |
|
| 3359 |
sprintf(name, "mon_sub_%s_%02d", r->name, hdr->id); |
| 3360 |
ckn = _mkdir_mondata_subdir(kn, name, hdr, r, prgrp, hdr->id); |
| 3361 |
if (IS_ERR(ckn)) { |
| 3362 |
kernfs_remove(kn); |
| 3363 |
return PTR_ERR(ckn); |
| 3364 |
} |
| 3365 |
|
| 3366 |
kernfs_activate(kn); |
| 3367 |
return 0; |
| 3368 |
} |
| 3369 |
|
| 3370 |
static int mkdir_mondata_subdir(struct kernfs_node *parent_kn, |
| 3371 |
struct rdt_domain_hdr *hdr, |
| 3372 |
struct rdt_resource *r, struct rdtgroup *prgrp) |
| 3373 |
{ |
| 3374 |
struct kernfs_node *kn; |
| 3375 |
char name[32]; |
| 3376 |
|
| 3377 |
lockdep_assert_held(&rdtgroup_mutex); |
| 3378 |
|
| 3379 |
if (r->rid == RDT_RESOURCE_L3 && r->mon_scope == RESCTRL_L3_NODE) |
| 3380 |
return mkdir_mondata_subdir_snc(parent_kn, hdr, r, prgrp); |
| 3381 |
|
| 3382 |
sprintf(name, "mon_%s_%02d", r->name, hdr->id); |
| 3383 |
kn = _mkdir_mondata_subdir(parent_kn, name, hdr, r, prgrp, hdr->id); |
| 3384 |
if (IS_ERR(kn)) |
| 3385 |
return PTR_ERR(kn); |
| 3386 |
|
| 3387 |
kernfs_activate(kn); |
| 3388 |
return 0; |
| 3389 |
} |
| 3390 |
|
| 3391 |
/* |
| 3392 |
* Add all subdirectories of mon_data for "ctrl_mon" groups |
| 3393 |
* and "monitor" groups with given domain id. |
| 3394 |
*/ |
| 3395 |
static void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, |
| 3396 |
struct rdt_domain_hdr *hdr) |
| 3397 |
{ |
| 3398 |
struct kernfs_node *parent_kn; |
| 3399 |
struct rdtgroup *prgrp, *crgrp; |
| 3400 |
struct list_head *head; |
| 3401 |
|
| 3402 |
list_for_each_entry(prgrp, &rdt_all_groups, rdtgroup_list) { |
| 3403 |
parent_kn = prgrp->mon.mon_data_kn; |
| 3404 |
mkdir_mondata_subdir(parent_kn, hdr, r, prgrp); |
| 3405 |
|
| 3406 |
head = &prgrp->mon.crdtgrp_list; |
| 3407 |
list_for_each_entry(crgrp, head, mon.crdtgrp_list) { |
| 3408 |
parent_kn = crgrp->mon.mon_data_kn; |
| 3409 |
mkdir_mondata_subdir(parent_kn, hdr, r, crgrp); |
| 3410 |
} |
| 3411 |
} |
| 3412 |
} |
| 3413 |
|
| 3414 |
static int mkdir_mondata_subdir_alldom(struct kernfs_node *parent_kn, |
| 3415 |
struct rdt_resource *r, |
| 3416 |
struct rdtgroup *prgrp) |
| 3417 |
{ |
| 3418 |
struct rdt_domain_hdr *hdr; |
| 3419 |
int ret; |
| 3420 |
|
| 3421 |
/* Walking r->domains, ensure it can't race with cpuhp */ |
| 3422 |
lockdep_assert_cpus_held(); |
| 3423 |
|
| 3424 |
list_for_each_entry(hdr, &r->mon_domains, list) { |
| 3425 |
ret = mkdir_mondata_subdir(parent_kn, hdr, r, prgrp); |
| 3426 |
if (ret) |
| 3427 |
return ret; |
| 3428 |
} |
| 3429 |
|
| 3430 |
return 0; |
| 3431 |
} |
| 3432 |
|
| 3433 |
/* |
| 3434 |
* This creates a directory mon_data which contains the monitored data. |
| 3435 |
* |
| 3436 |
* mon_data has one directory for each domain which are named |
| 3437 |
* in the format mon_<domain_name>_<domain_id>. For ex: A mon_data |
| 3438 |
* with L3 domain looks as below: |
| 3439 |
* ./mon_data: |
| 3440 |
* mon_L3_00 |
| 3441 |
* mon_L3_01 |
| 3442 |
* mon_L3_02 |
| 3443 |
* ... |
| 3444 |
* |
| 3445 |
* Each domain directory has one file per event: |
| 3446 |
* ./mon_L3_00/: |
| 3447 |
* llc_occupancy |
| 3448 |
* |
| 3449 |
*/ |
| 3450 |
static int mkdir_mondata_all(struct kernfs_node *parent_kn, |
| 3451 |
struct rdtgroup *prgrp, |
| 3452 |
struct kernfs_node **dest_kn) |
| 3453 |
{ |
| 3454 |
struct rdt_resource *r; |
| 3455 |
struct kernfs_node *kn; |
| 3456 |
int ret; |
| 3457 |
|
| 3458 |
/* |
| 3459 |
* Create the mon_data directory first. |
| 3460 |
*/ |
| 3461 |
ret = mongroup_create_dir(parent_kn, prgrp, "mon_data", &kn); |
| 3462 |
if (ret) |
| 3463 |
return ret; |
| 3464 |
|
| 3465 |
if (dest_kn) |
| 3466 |
*dest_kn = kn; |
| 3467 |
|
| 3468 |
/* |
| 3469 |
* Create the subdirectories for each domain. Note that all events |
| 3470 |
* in a domain like L3 are grouped into a resource whose domain is L3 |
| 3471 |
*/ |
| 3472 |
for_each_mon_capable_rdt_resource(r) { |
| 3473 |
ret = mkdir_mondata_subdir_alldom(kn, r, prgrp); |
| 3474 |
if (ret) |
| 3475 |
goto out_destroy; |
| 3476 |
} |
| 3477 |
|
| 3478 |
return 0; |
| 3479 |
|
| 3480 |
out_destroy: |
| 3481 |
kernfs_remove(kn); |
| 3482 |
return ret; |
| 3483 |
} |
| 3484 |
|
| 3485 |
/** |
| 3486 |
* cbm_ensure_valid - Enforce validity on provided CBM |
| 3487 |
* @_val: Candidate CBM |
| 3488 |
* @r: RDT resource to which the CBM belongs |
| 3489 |
* |
| 3490 |
* The provided CBM represents all cache portions available for use. This |
| 3491 |
* may be represented by a bitmap that does not consist of contiguous ones |
| 3492 |
* and thus be an invalid CBM. |
| 3493 |
* Here the provided CBM is forced to be a valid CBM by only considering |
| 3494 |
* the first set of contiguous bits as valid and clearing all bits. |
| 3495 |
* The intention here is to provide a valid default CBM with which a new |
| 3496 |
* resource group is initialized. The user can follow this with a |
| 3497 |
* modification to the CBM if the default does not satisfy the |
| 3498 |
* requirements. |
| 3499 |
*/ |
| 3500 |
static u32 cbm_ensure_valid(u32 _val, struct rdt_resource *r) |
| 3501 |
{ |
| 3502 |
unsigned int cbm_len = r->cache.cbm_len; |
| 3503 |
unsigned long first_bit, zero_bit; |
| 3504 |
unsigned long val; |
| 3505 |
|
| 3506 |
if (!_val || r->cache.arch_has_sparse_bitmasks) |
| 3507 |
return _val; |
| 3508 |
|
| 3509 |
val = _val; |
| 3510 |
first_bit = find_first_bit(&val, cbm_len); |
| 3511 |
zero_bit = find_next_zero_bit(&val, cbm_len, first_bit); |
| 3512 |
|
| 3513 |
/* Clear any remaining bits to ensure contiguous region */ |
| 3514 |
bitmap_clear(&val, zero_bit, cbm_len - zero_bit); |
| 3515 |
return (u32)val; |
| 3516 |
} |
| 3517 |
|
| 3518 |
/* |
| 3519 |
* Initialize cache resources per RDT domain |
| 3520 |
* |
| 3521 |
* Set the RDT domain up to start off with all usable allocations. That is, |
| 3522 |
* all shareable and unused bits. All-zero CBM is invalid. |
| 3523 |
*/ |
| 3524 |
static int __init_one_rdt_domain(struct rdt_ctrl_domain *d, struct resctrl_schema *s, |
| 3525 |
u32 closid) |
| 3526 |
{ |
| 3527 |
enum resctrl_conf_type peer_type = resctrl_peer_type(s->conf_type); |
| 3528 |
enum resctrl_conf_type t = s->conf_type; |
| 3529 |
struct resctrl_staged_config *cfg; |
| 3530 |
struct rdt_resource *r = s->res; |
| 3531 |
u32 used_b = 0, unused_b = 0; |
| 3532 |
unsigned long tmp_cbm; |
| 3533 |
enum rdtgrp_mode mode; |
| 3534 |
u32 peer_ctl, ctrl_val; |
| 3535 |
int i; |
| 3536 |
|
| 3537 |
cfg = &d->staged_config[t]; |
| 3538 |
cfg->have_new_ctrl = false; |
| 3539 |
cfg->new_ctrl = r->cache.shareable_bits; |
| 3540 |
used_b = r->cache.shareable_bits; |
| 3541 |
for (i = 0; i < closids_supported(); i++) { |
| 3542 |
if (closid_allocated(i) && i != closid) { |
| 3543 |
mode = rdtgroup_mode_by_closid(i); |
| 3544 |
if (mode == RDT_MODE_PSEUDO_LOCKSETUP) |
| 3545 |
/* |
| 3546 |
* ctrl values for locksetup aren't relevant |
| 3547 |
* until the schemata is written, and the mode |
| 3548 |
* becomes RDT_MODE_PSEUDO_LOCKED. |
| 3549 |
*/ |
| 3550 |
continue; |
| 3551 |
/* |
| 3552 |
* If CDP is active include peer domain's |
| 3553 |
* usage to ensure there is no overlap |
| 3554 |
* with an exclusive group. |
| 3555 |
*/ |
| 3556 |
if (resctrl_arch_get_cdp_enabled(r->rid)) |
| 3557 |
peer_ctl = resctrl_arch_get_config(r, d, i, |
| 3558 |
peer_type); |
| 3559 |
else |
| 3560 |
peer_ctl = 0; |
| 3561 |
ctrl_val = resctrl_arch_get_config(r, d, i, |
| 3562 |
s->conf_type); |
| 3563 |
used_b |= ctrl_val | peer_ctl; |
| 3564 |
if (mode == RDT_MODE_SHAREABLE) |
| 3565 |
cfg->new_ctrl |= ctrl_val | peer_ctl; |
| 3566 |
} |
| 3567 |
} |
| 3568 |
if (d->plr && d->plr->cbm > 0) |
| 3569 |
used_b |= d->plr->cbm; |
| 3570 |
unused_b = used_b ^ (BIT_MASK(r->cache.cbm_len) - 1); |
| 3571 |
unused_b &= BIT_MASK(r->cache.cbm_len) - 1; |
| 3572 |
cfg->new_ctrl |= unused_b; |
| 3573 |
/* |
| 3574 |
* Force the initial CBM to be valid, user can |
| 3575 |
* modify the CBM based on system availability. |
| 3576 |
*/ |
| 3577 |
cfg->new_ctrl = cbm_ensure_valid(cfg->new_ctrl, r); |
| 3578 |
/* |
| 3579 |
* Assign the u32 CBM to an unsigned long to ensure that |
| 3580 |
* bitmap_weight() does not access out-of-bound memory. |
| 3581 |
*/ |
| 3582 |
tmp_cbm = cfg->new_ctrl; |
| 3583 |
if (bitmap_weight(&tmp_cbm, r->cache.cbm_len) < r->cache.min_cbm_bits) { |
| 3584 |
rdt_last_cmd_printf("No space on %s:%d\n", s->name, d->hdr.id); |
| 3585 |
return -ENOSPC; |
| 3586 |
} |
| 3587 |
cfg->have_new_ctrl = true; |
| 3588 |
|
| 3589 |
return 0; |
| 3590 |
} |
| 3591 |
|
| 3592 |
/* |
| 3593 |
* Initialize cache resources with default values. |
| 3594 |
* |
| 3595 |
* A new RDT group is being created on an allocation capable (CAT) |
| 3596 |
* supporting system. Set this group up to start off with all usable |
| 3597 |
* allocations. |
| 3598 |
* |
| 3599 |
* If there are no more shareable bits available on any domain then |
| 3600 |
* the entire allocation will fail. |
| 3601 |
*/ |
| 3602 |
int rdtgroup_init_cat(struct resctrl_schema *s, u32 closid) |
| 3603 |
{ |
| 3604 |
struct rdt_ctrl_domain *d; |
| 3605 |
int ret; |
| 3606 |
|
| 3607 |
list_for_each_entry(d, &s->res->ctrl_domains, hdr.list) { |
| 3608 |
ret = __init_one_rdt_domain(d, s, closid); |
| 3609 |
if (ret < 0) |
| 3610 |
return ret; |
| 3611 |
} |
| 3612 |
|
| 3613 |
return 0; |
| 3614 |
} |
| 3615 |
|
| 3616 |
/* Initialize MBA resource with default values. */ |
| 3617 |
static void rdtgroup_init_mba(struct rdt_resource *r, u32 closid) |
| 3618 |
{ |
| 3619 |
struct resctrl_staged_config *cfg; |
| 3620 |
struct rdt_ctrl_domain *d; |
| 3621 |
|
| 3622 |
list_for_each_entry(d, &r->ctrl_domains, hdr.list) { |
| 3623 |
if (is_mba_sc(r)) { |
| 3624 |
d->mbps_val[closid] = MBA_MAX_MBPS; |
| 3625 |
continue; |
| 3626 |
} |
| 3627 |
|
| 3628 |
cfg = &d->staged_config[CDP_NONE]; |
| 3629 |
cfg->new_ctrl = resctrl_get_default_ctrl(r); |
| 3630 |
cfg->have_new_ctrl = true; |
| 3631 |
} |
| 3632 |
} |
| 3633 |
|
| 3634 |
/* Initialize the RDT group's allocations. */ |
| 3635 |
static int rdtgroup_init_alloc(struct rdtgroup *rdtgrp) |
| 3636 |
{ |
| 3637 |
struct resctrl_schema *s; |
| 3638 |
struct rdt_resource *r; |
| 3639 |
int ret = 0; |
| 3640 |
|
| 3641 |
rdt_staged_configs_clear(); |
| 3642 |
|
| 3643 |
list_for_each_entry(s, &resctrl_schema_all, list) { |
| 3644 |
r = s->res; |
| 3645 |
if (r->rid == RDT_RESOURCE_MBA || |
| 3646 |
r->rid == RDT_RESOURCE_SMBA) { |
| 3647 |
rdtgroup_init_mba(r, rdtgrp->closid); |
| 3648 |
if (is_mba_sc(r)) |
| 3649 |
continue; |
| 3650 |
} else { |
| 3651 |
ret = rdtgroup_init_cat(s, rdtgrp->closid); |
| 3652 |
if (ret < 0) |
| 3653 |
goto out; |
| 3654 |
} |
| 3655 |
|
| 3656 |
ret = resctrl_arch_update_domains(r, rdtgrp->closid); |
| 3657 |
if (ret < 0) { |
| 3658 |
rdt_last_cmd_puts("Failed to initialize allocations\n"); |
| 3659 |
goto out; |
| 3660 |
} |
| 3661 |
} |
| 3662 |
|
| 3663 |
rdtgrp->mode = RDT_MODE_SHAREABLE; |
| 3664 |
|
| 3665 |
out: |
| 3666 |
rdt_staged_configs_clear(); |
| 3667 |
return ret; |
| 3668 |
} |
| 3669 |
|
| 3670 |
static int mkdir_rdt_prepare_rmid_alloc(struct rdtgroup *rdtgrp) |
| 3671 |
{ |
| 3672 |
int ret; |
| 3673 |
|
| 3674 |
if (!resctrl_arch_mon_capable()) |
| 3675 |
return 0; |
| 3676 |
|
| 3677 |
ret = alloc_rmid(rdtgrp->closid); |
| 3678 |
if (ret < 0) { |
| 3679 |
rdt_last_cmd_puts("Out of RMIDs\n"); |
| 3680 |
return ret; |
| 3681 |
} |
| 3682 |
rdtgrp->mon.rmid = ret; |
| 3683 |
|
| 3684 |
rdtgroup_assign_cntrs(rdtgrp); |
| 3685 |
|
| 3686 |
ret = mkdir_mondata_all(rdtgrp->kn, rdtgrp, &rdtgrp->mon.mon_data_kn); |
| 3687 |
if (ret) { |
| 3688 |
rdt_last_cmd_puts("kernfs subdir error\n"); |
| 3689 |
rdtgroup_unassign_cntrs(rdtgrp); |
| 3690 |
free_rmid(rdtgrp->closid, rdtgrp->mon.rmid); |
| 3691 |
return ret; |
| 3692 |
} |
| 3693 |
|
| 3694 |
return 0; |
| 3695 |
} |
| 3696 |
|
| 3697 |
static void mkdir_rdt_prepare_rmid_free(struct rdtgroup *rgrp) |
| 3698 |
{ |
| 3699 |
if (resctrl_arch_mon_capable()) { |
| 3700 |
rdtgroup_unassign_cntrs(rgrp); |
| 3701 |
free_rmid(rgrp->closid, rgrp->mon.rmid); |
| 3702 |
} |
| 3703 |
} |
| 3704 |
|
| 3705 |
/* |
| 3706 |
* We allow creating mon groups only with in a directory called "mon_groups" |
| 3707 |
* which is present in every ctrl_mon group. Check if this is a valid |
| 3708 |
* "mon_groups" directory. |
| 3709 |
* |
| 3710 |
* 1. The directory should be named "mon_groups". |
| 3711 |
* 2. The mon group itself should "not" be named "mon_groups". |
| 3712 |
* This makes sure "mon_groups" directory always has a ctrl_mon group |
| 3713 |
* as parent. |
| 3714 |
*/ |
| 3715 |
static bool is_mon_groups(struct kernfs_node *kn, const char *name) |
| 3716 |
{ |
| 3717 |
return (!strcmp(rdt_kn_name(kn), "mon_groups") && |
| 3718 |
strcmp(name, "mon_groups")); |
| 3719 |
} |
| 3720 |
|
| 3721 |
static int mkdir_rdt_prepare(struct kernfs_node *parent_kn, |
| 3722 |
const char *name, umode_t mode, |
| 3723 |
enum rdt_group_type rtype, struct rdtgroup **r) |
| 3724 |
{ |
| 3725 |
struct rdtgroup *prdtgrp, *rdtgrp; |
| 3726 |
unsigned long files = 0; |
| 3727 |
struct kernfs_node *kn; |
| 3728 |
int ret; |
| 3729 |
|
| 3730 |
prdtgrp = rdtgroup_kn_lock_live(parent_kn); |
| 3731 |
if (!prdtgrp) { |
| 3732 |
ret = -ENODEV; |
| 3733 |
goto out_unlock; |
| 3734 |
} |
| 3735 |
|
| 3736 |
rdt_last_cmd_clear(); |
| 3737 |
|
| 3738 |
/* |
| 3739 |
* Check that the parent directory for a monitor group is a "mon_groups" |
| 3740 |
* directory. |
| 3741 |
*/ |
| 3742 |
if (rtype == RDTMON_GROUP && !is_mon_groups(parent_kn, name)) { |
| 3743 |
ret = -EPERM; |
| 3744 |
goto out_unlock; |
| 3745 |
} |
| 3746 |
|
| 3747 |
if (rtype == RDTMON_GROUP && |
| 3748 |
(prdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
| 3749 |
prdtgrp->mode == RDT_MODE_PSEUDO_LOCKED)) { |
| 3750 |
ret = -EINVAL; |
| 3751 |
rdt_last_cmd_puts("Pseudo-locking in progress\n"); |
| 3752 |
goto out_unlock; |
| 3753 |
} |
| 3754 |
|
| 3755 |
/* allocate the rdtgroup. */ |
| 3756 |
rdtgrp = kzalloc_obj(*rdtgrp); |
| 3757 |
if (!rdtgrp) { |
| 3758 |
ret = -ENOSPC; |
| 3759 |
rdt_last_cmd_puts("Kernel out of memory\n"); |
| 3760 |
goto out_unlock; |
| 3761 |
} |
| 3762 |
*r = rdtgrp; |
| 3763 |
rdtgrp->mon.parent = prdtgrp; |
| 3764 |
rdtgrp->type = rtype; |
| 3765 |
INIT_LIST_HEAD(&rdtgrp->mon.crdtgrp_list); |
| 3766 |
|
| 3767 |
/* kernfs creates the directory for rdtgrp */ |
| 3768 |
kn = kernfs_create_dir(parent_kn, name, mode, rdtgrp); |
| 3769 |
if (IS_ERR(kn)) { |
| 3770 |
ret = PTR_ERR(kn); |
| 3771 |
rdt_last_cmd_puts("kernfs create error\n"); |
| 3772 |
goto out_free_rgrp; |
| 3773 |
} |
| 3774 |
rdtgrp->kn = kn; |
| 3775 |
|
| 3776 |
/* |
| 3777 |
* kernfs_remove() will drop the reference count on "kn" which |
| 3778 |
* will free it. But we still need it to stick around for the |
| 3779 |
* rdtgroup_kn_unlock(kn) call. Take one extra reference here, |
| 3780 |
* which will be dropped by kernfs_put() in rdtgroup_remove(). |
| 3781 |
*/ |
| 3782 |
kernfs_get(kn); |
| 3783 |
|
| 3784 |
ret = rdtgroup_kn_set_ugid(kn); |
| 3785 |
if (ret) { |
| 3786 |
rdt_last_cmd_puts("kernfs perm error\n"); |
| 3787 |
goto out_destroy; |
| 3788 |
} |
| 3789 |
|
| 3790 |
if (rtype == RDTCTRL_GROUP) { |
| 3791 |
files = RFTYPE_BASE | RFTYPE_CTRL; |
| 3792 |
if (resctrl_arch_mon_capable()) |
| 3793 |
files |= RFTYPE_MON; |
| 3794 |
} else { |
| 3795 |
files = RFTYPE_BASE | RFTYPE_MON; |
| 3796 |
} |
| 3797 |
|
| 3798 |
ret = rdtgroup_add_files(kn, files); |
| 3799 |
if (ret) { |
| 3800 |
rdt_last_cmd_puts("kernfs fill error\n"); |
| 3801 |
goto out_destroy; |
| 3802 |
} |
| 3803 |
|
| 3804 |
/* |
| 3805 |
* The caller unlocks the parent_kn upon success. |
| 3806 |
*/ |
| 3807 |
return 0; |
| 3808 |
|
| 3809 |
out_destroy: |
| 3810 |
kernfs_put(rdtgrp->kn); |
| 3811 |
kernfs_remove(rdtgrp->kn); |
| 3812 |
out_free_rgrp: |
| 3813 |
kfree(rdtgrp); |
| 3814 |
out_unlock: |
| 3815 |
rdtgroup_kn_unlock(parent_kn); |
| 3816 |
return ret; |
| 3817 |
} |
| 3818 |
|
| 3819 |
static void mkdir_rdt_prepare_clean(struct rdtgroup *rgrp) |
| 3820 |
{ |
| 3821 |
kernfs_remove(rgrp->kn); |
| 3822 |
rdtgroup_remove(rgrp); |
| 3823 |
} |
| 3824 |
|
| 3825 |
/* |
| 3826 |
* Create a monitor group under "mon_groups" directory of a control |
| 3827 |
* and monitor group(ctrl_mon). This is a resource group |
| 3828 |
* to monitor a subset of tasks and cpus in its parent ctrl_mon group. |
| 3829 |
*/ |
| 3830 |
static int rdtgroup_mkdir_mon(struct kernfs_node *parent_kn, |
| 3831 |
const char *name, umode_t mode) |
| 3832 |
{ |
| 3833 |
struct rdtgroup *rdtgrp, *prgrp; |
| 3834 |
int ret; |
| 3835 |
|
| 3836 |
ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTMON_GROUP, &rdtgrp); |
| 3837 |
if (ret) |
| 3838 |
return ret; |
| 3839 |
|
| 3840 |
prgrp = rdtgrp->mon.parent; |
| 3841 |
rdtgrp->closid = prgrp->closid; |
| 3842 |
|
| 3843 |
ret = mkdir_rdt_prepare_rmid_alloc(rdtgrp); |
| 3844 |
if (ret) { |
| 3845 |
mkdir_rdt_prepare_clean(rdtgrp); |
| 3846 |
goto out_unlock; |
| 3847 |
} |
| 3848 |
|
| 3849 |
kernfs_activate(rdtgrp->kn); |
| 3850 |
|
| 3851 |
/* |
| 3852 |
* Add the rdtgrp to the list of rdtgrps the parent |
| 3853 |
* ctrl_mon group has to track. |
| 3854 |
*/ |
| 3855 |
list_add_tail(&rdtgrp->mon.crdtgrp_list, &prgrp->mon.crdtgrp_list); |
| 3856 |
|
| 3857 |
out_unlock: |
| 3858 |
rdtgroup_kn_unlock(parent_kn); |
| 3859 |
return ret; |
| 3860 |
} |
| 3861 |
|
| 3862 |
/* |
| 3863 |
* These are rdtgroups created under the root directory. Can be used |
| 3864 |
* to allocate and monitor resources. |
| 3865 |
*/ |
| 3866 |
static int rdtgroup_mkdir_ctrl_mon(struct kernfs_node *parent_kn, |
| 3867 |
const char *name, umode_t mode) |
| 3868 |
{ |
| 3869 |
struct rdtgroup *rdtgrp; |
| 3870 |
struct kernfs_node *kn; |
| 3871 |
u32 closid; |
| 3872 |
int ret; |
| 3873 |
|
| 3874 |
ret = mkdir_rdt_prepare(parent_kn, name, mode, RDTCTRL_GROUP, &rdtgrp); |
| 3875 |
if (ret) |
| 3876 |
return ret; |
| 3877 |
|
| 3878 |
kn = rdtgrp->kn; |
| 3879 |
ret = closid_alloc(); |
| 3880 |
if (ret < 0) { |
| 3881 |
rdt_last_cmd_puts("Out of CLOSIDs\n"); |
| 3882 |
goto out_common_fail; |
| 3883 |
} |
| 3884 |
closid = ret; |
| 3885 |
ret = 0; |
| 3886 |
|
| 3887 |
rdtgrp->closid = closid; |
| 3888 |
|
| 3889 |
ret = mkdir_rdt_prepare_rmid_alloc(rdtgrp); |
| 3890 |
if (ret) |
| 3891 |
goto out_closid_free; |
| 3892 |
|
| 3893 |
kernfs_activate(rdtgrp->kn); |
| 3894 |
|
| 3895 |
ret = rdtgroup_init_alloc(rdtgrp); |
| 3896 |
if (ret < 0) |
| 3897 |
goto out_rmid_free; |
| 3898 |
|
| 3899 |
list_add(&rdtgrp->rdtgroup_list, &rdt_all_groups); |
| 3900 |
|
| 3901 |
if (resctrl_arch_mon_capable()) { |
| 3902 |
/* |
| 3903 |
* Create an empty mon_groups directory to hold the subset |
| 3904 |
* of tasks and cpus to monitor. |
| 3905 |
*/ |
| 3906 |
ret = mongroup_create_dir(kn, rdtgrp, "mon_groups", NULL); |
| 3907 |
if (ret) { |
| 3908 |
rdt_last_cmd_puts("kernfs subdir error\n"); |
| 3909 |
goto out_del_list; |
| 3910 |
} |
| 3911 |
if (is_mba_sc(NULL)) |
| 3912 |
rdtgrp->mba_mbps_event = mba_mbps_default_event; |
| 3913 |
} |
| 3914 |
|
| 3915 |
goto out_unlock; |
| 3916 |
|
| 3917 |
out_del_list: |
| 3918 |
list_del(&rdtgrp->rdtgroup_list); |
| 3919 |
out_rmid_free: |
| 3920 |
mkdir_rdt_prepare_rmid_free(rdtgrp); |
| 3921 |
out_closid_free: |
| 3922 |
closid_free(closid); |
| 3923 |
out_common_fail: |
| 3924 |
mkdir_rdt_prepare_clean(rdtgrp); |
| 3925 |
out_unlock: |
| 3926 |
rdtgroup_kn_unlock(parent_kn); |
| 3927 |
return ret; |
| 3928 |
} |
| 3929 |
|
| 3930 |
static int rdtgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
| 3931 |
umode_t mode) |
| 3932 |
{ |
| 3933 |
/* Do not accept '\n' to avoid unparsable situation. */ |
| 3934 |
if (strchr(name, '\n')) |
| 3935 |
return -EINVAL; |
| 3936 |
|
| 3937 |
/* |
| 3938 |
* If the parent directory is the root directory and RDT |
| 3939 |
* allocation is supported, add a control and monitoring |
| 3940 |
* subdirectory |
| 3941 |
*/ |
| 3942 |
if (resctrl_arch_alloc_capable() && parent_kn == rdtgroup_default.kn) |
| 3943 |
return rdtgroup_mkdir_ctrl_mon(parent_kn, name, mode); |
| 3944 |
|
| 3945 |
/* Else, attempt to add a monitoring subdirectory. */ |
| 3946 |
if (resctrl_arch_mon_capable()) |
| 3947 |
return rdtgroup_mkdir_mon(parent_kn, name, mode); |
| 3948 |
|
| 3949 |
return -EPERM; |
| 3950 |
} |
| 3951 |
|
| 3952 |
static int rdtgroup_rmdir_mon(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask) |
| 3953 |
{ |
| 3954 |
struct rdtgroup *prdtgrp = rdtgrp->mon.parent; |
| 3955 |
u32 closid, rmid; |
| 3956 |
int cpu; |
| 3957 |
|
| 3958 |
/* Give any tasks back to the parent group */ |
| 3959 |
rdt_move_group_tasks(rdtgrp, prdtgrp, tmpmask); |
| 3960 |
|
| 3961 |
/* |
| 3962 |
* Update per cpu closid/rmid of the moved CPUs first. |
| 3963 |
* Note: the closid will not change, but the arch code still needs it. |
| 3964 |
*/ |
| 3965 |
closid = prdtgrp->closid; |
| 3966 |
rmid = prdtgrp->mon.rmid; |
| 3967 |
for_each_cpu(cpu, &rdtgrp->cpu_mask) |
| 3968 |
resctrl_arch_set_cpu_default_closid_rmid(cpu, closid, rmid); |
| 3969 |
|
| 3970 |
/* |
| 3971 |
* Update the MSR on moved CPUs and CPUs which have moved |
| 3972 |
* task running on them. |
| 3973 |
*/ |
| 3974 |
cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); |
| 3975 |
update_closid_rmid(tmpmask, NULL); |
| 3976 |
|
| 3977 |
rdtgrp->flags = RDT_DELETED; |
| 3978 |
|
| 3979 |
rdtgroup_unassign_cntrs(rdtgrp); |
| 3980 |
|
| 3981 |
free_rmid(rdtgrp->closid, rdtgrp->mon.rmid); |
| 3982 |
|
| 3983 |
/* |
| 3984 |
* Remove the rdtgrp from the parent ctrl_mon group's list |
| 3985 |
*/ |
| 3986 |
WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list)); |
| 3987 |
list_del(&rdtgrp->mon.crdtgrp_list); |
| 3988 |
|
| 3989 |
kernfs_remove(rdtgrp->kn); |
| 3990 |
|
| 3991 |
return 0; |
| 3992 |
} |
| 3993 |
|
| 3994 |
static int rdtgroup_ctrl_remove(struct rdtgroup *rdtgrp) |
| 3995 |
{ |
| 3996 |
rdtgrp->flags = RDT_DELETED; |
| 3997 |
list_del(&rdtgrp->rdtgroup_list); |
| 3998 |
|
| 3999 |
kernfs_remove(rdtgrp->kn); |
| 4000 |
return 0; |
| 4001 |
} |
| 4002 |
|
| 4003 |
static int rdtgroup_rmdir_ctrl(struct rdtgroup *rdtgrp, cpumask_var_t tmpmask) |
| 4004 |
{ |
| 4005 |
u32 closid, rmid; |
| 4006 |
int cpu; |
| 4007 |
|
| 4008 |
/* Give any tasks back to the default group */ |
| 4009 |
rdt_move_group_tasks(rdtgrp, &rdtgroup_default, tmpmask); |
| 4010 |
|
| 4011 |
/* Give any CPUs back to the default group */ |
| 4012 |
cpumask_or(&rdtgroup_default.cpu_mask, |
| 4013 |
&rdtgroup_default.cpu_mask, &rdtgrp->cpu_mask); |
| 4014 |
|
| 4015 |
/* Update per cpu closid and rmid of the moved CPUs first */ |
| 4016 |
closid = rdtgroup_default.closid; |
| 4017 |
rmid = rdtgroup_default.mon.rmid; |
| 4018 |
for_each_cpu(cpu, &rdtgrp->cpu_mask) |
| 4019 |
resctrl_arch_set_cpu_default_closid_rmid(cpu, closid, rmid); |
| 4020 |
|
| 4021 |
/* |
| 4022 |
* Update the MSR on moved CPUs and CPUs which have moved |
| 4023 |
* task running on them. |
| 4024 |
*/ |
| 4025 |
cpumask_or(tmpmask, tmpmask, &rdtgrp->cpu_mask); |
| 4026 |
update_closid_rmid(tmpmask, NULL); |
| 4027 |
|
| 4028 |
rdtgroup_unassign_cntrs(rdtgrp); |
| 4029 |
|
| 4030 |
free_rmid(rdtgrp->closid, rdtgrp->mon.rmid); |
| 4031 |
closid_free(rdtgrp->closid); |
| 4032 |
|
| 4033 |
rdtgroup_ctrl_remove(rdtgrp); |
| 4034 |
|
| 4035 |
/* |
| 4036 |
* Free all the child monitor group rmids. |
| 4037 |
*/ |
| 4038 |
free_all_child_rdtgrp(rdtgrp); |
| 4039 |
|
| 4040 |
return 0; |
| 4041 |
} |
| 4042 |
|
| 4043 |
static struct kernfs_node *rdt_kn_parent(struct kernfs_node *kn) |
| 4044 |
{ |
| 4045 |
/* |
| 4046 |
* Valid within the RCU section it was obtained or while rdtgroup_mutex |
| 4047 |
* is held. |
| 4048 |
*/ |
| 4049 |
return rcu_dereference_check(kn->__parent, lockdep_is_held(&rdtgroup_mutex)); |
| 4050 |
} |
| 4051 |
|
| 4052 |
static int rdtgroup_rmdir(struct kernfs_node *kn) |
| 4053 |
{ |
| 4054 |
struct kernfs_node *parent_kn; |
| 4055 |
struct rdtgroup *rdtgrp; |
| 4056 |
cpumask_var_t tmpmask; |
| 4057 |
int ret = 0; |
| 4058 |
|
| 4059 |
if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) |
| 4060 |
return -ENOMEM; |
| 4061 |
|
| 4062 |
rdtgrp = rdtgroup_kn_lock_live(kn); |
| 4063 |
if (!rdtgrp) { |
| 4064 |
ret = -EPERM; |
| 4065 |
goto out; |
| 4066 |
} |
| 4067 |
parent_kn = rdt_kn_parent(kn); |
| 4068 |
|
| 4069 |
/* |
| 4070 |
* If the rdtgroup is a ctrl_mon group and parent directory |
| 4071 |
* is the root directory, remove the ctrl_mon group. |
| 4072 |
* |
| 4073 |
* If the rdtgroup is a mon group and parent directory |
| 4074 |
* is a valid "mon_groups" directory, remove the mon group. |
| 4075 |
*/ |
| 4076 |
if (rdtgrp->type == RDTCTRL_GROUP && parent_kn == rdtgroup_default.kn && |
| 4077 |
rdtgrp != &rdtgroup_default) { |
| 4078 |
if (rdtgrp->mode == RDT_MODE_PSEUDO_LOCKSETUP || |
| 4079 |
rdtgrp->mode == RDT_MODE_PSEUDO_LOCKED) { |
| 4080 |
ret = rdtgroup_ctrl_remove(rdtgrp); |
| 4081 |
} else { |
| 4082 |
ret = rdtgroup_rmdir_ctrl(rdtgrp, tmpmask); |
| 4083 |
} |
| 4084 |
} else if (rdtgrp->type == RDTMON_GROUP && |
| 4085 |
is_mon_groups(parent_kn, rdt_kn_name(kn))) { |
| 4086 |
ret = rdtgroup_rmdir_mon(rdtgrp, tmpmask); |
| 4087 |
} else { |
| 4088 |
ret = -EPERM; |
| 4089 |
} |
| 4090 |
|
| 4091 |
out: |
| 4092 |
rdtgroup_kn_unlock(kn); |
| 4093 |
free_cpumask_var(tmpmask); |
| 4094 |
return ret; |
| 4095 |
} |
| 4096 |
|
| 4097 |
/** |
| 4098 |
* mongrp_reparent() - replace parent CTRL_MON group of a MON group |
| 4099 |
* @rdtgrp: the MON group whose parent should be replaced |
| 4100 |
* @new_prdtgrp: replacement parent CTRL_MON group for @rdtgrp |
| 4101 |
* @cpus: cpumask provided by the caller for use during this call |
| 4102 |
* |
| 4103 |
* Replaces the parent CTRL_MON group for a MON group, resulting in all member |
| 4104 |
* tasks' CLOSID immediately changing to that of the new parent group. |
| 4105 |
* Monitoring data for the group is unaffected by this operation. |
| 4106 |
*/ |
| 4107 |
static void mongrp_reparent(struct rdtgroup *rdtgrp, |
| 4108 |
struct rdtgroup *new_prdtgrp, |
| 4109 |
cpumask_var_t cpus) |
| 4110 |
{ |
| 4111 |
struct rdtgroup *prdtgrp = rdtgrp->mon.parent; |
| 4112 |
|
| 4113 |
WARN_ON(rdtgrp->type != RDTMON_GROUP); |
| 4114 |
WARN_ON(new_prdtgrp->type != RDTCTRL_GROUP); |
| 4115 |
|
| 4116 |
/* Nothing to do when simply renaming a MON group. */ |
| 4117 |
if (prdtgrp == new_prdtgrp) |
| 4118 |
return; |
| 4119 |
|
| 4120 |
WARN_ON(list_empty(&prdtgrp->mon.crdtgrp_list)); |
| 4121 |
list_move_tail(&rdtgrp->mon.crdtgrp_list, |
| 4122 |
&new_prdtgrp->mon.crdtgrp_list); |
| 4123 |
|
| 4124 |
rdtgrp->mon.parent = new_prdtgrp; |
| 4125 |
rdtgrp->closid = new_prdtgrp->closid; |
| 4126 |
|
| 4127 |
/* Propagate updated closid to all tasks in this group. */ |
| 4128 |
rdt_move_group_tasks(rdtgrp, rdtgrp, cpus); |
| 4129 |
|
| 4130 |
update_closid_rmid(cpus, NULL); |
| 4131 |
} |
| 4132 |
|
| 4133 |
static int rdtgroup_rename(struct kernfs_node *kn, |
| 4134 |
struct kernfs_node *new_parent, const char *new_name) |
| 4135 |
{ |
| 4136 |
struct kernfs_node *kn_parent; |
| 4137 |
struct rdtgroup *new_prdtgrp; |
| 4138 |
struct rdtgroup *rdtgrp; |
| 4139 |
cpumask_var_t tmpmask; |
| 4140 |
int ret; |
| 4141 |
|
| 4142 |
rdtgrp = kernfs_to_rdtgroup(kn); |
| 4143 |
new_prdtgrp = kernfs_to_rdtgroup(new_parent); |
| 4144 |
if (!rdtgrp || !new_prdtgrp) |
| 4145 |
return -ENOENT; |
| 4146 |
|
| 4147 |
/* Release both kernfs active_refs before obtaining rdtgroup mutex. */ |
| 4148 |
rdtgroup_kn_get(rdtgrp, kn); |
| 4149 |
rdtgroup_kn_get(new_prdtgrp, new_parent); |
| 4150 |
|
| 4151 |
mutex_lock(&rdtgroup_mutex); |
| 4152 |
|
| 4153 |
rdt_last_cmd_clear(); |
| 4154 |
|
| 4155 |
/* |
| 4156 |
* Don't allow kernfs_to_rdtgroup() to return a parent rdtgroup if |
| 4157 |
* either kernfs_node is a file. |
| 4158 |
*/ |
| 4159 |
if (kernfs_type(kn) != KERNFS_DIR || |
| 4160 |
kernfs_type(new_parent) != KERNFS_DIR) { |
| 4161 |
rdt_last_cmd_puts("Source and destination must be directories"); |
| 4162 |
ret = -EPERM; |
| 4163 |
goto out; |
| 4164 |
} |
| 4165 |
|
| 4166 |
if ((rdtgrp->flags & RDT_DELETED) || (new_prdtgrp->flags & RDT_DELETED)) { |
| 4167 |
ret = -ENOENT; |
| 4168 |
goto out; |
| 4169 |
} |
| 4170 |
|
| 4171 |
kn_parent = rdt_kn_parent(kn); |
| 4172 |
if (rdtgrp->type != RDTMON_GROUP || !kn_parent || |
| 4173 |
!is_mon_groups(kn_parent, rdt_kn_name(kn))) { |
| 4174 |
rdt_last_cmd_puts("Source must be a MON group\n"); |
| 4175 |
ret = -EPERM; |
| 4176 |
goto out; |
| 4177 |
} |
| 4178 |
|
| 4179 |
if (!is_mon_groups(new_parent, new_name)) { |
| 4180 |
rdt_last_cmd_puts("Destination must be a mon_groups subdirectory\n"); |
| 4181 |
ret = -EPERM; |
| 4182 |
goto out; |
| 4183 |
} |
| 4184 |
|
| 4185 |
/* |
| 4186 |
* If the MON group is monitoring CPUs, the CPUs must be assigned to the |
| 4187 |
* current parent CTRL_MON group and therefore cannot be assigned to |
| 4188 |
* the new parent, making the move illegal. |
| 4189 |
*/ |
| 4190 |
if (!cpumask_empty(&rdtgrp->cpu_mask) && |
| 4191 |
rdtgrp->mon.parent != new_prdtgrp) { |
| 4192 |
rdt_last_cmd_puts("Cannot move a MON group that monitors CPUs\n"); |
| 4193 |
ret = -EPERM; |
| 4194 |
goto out; |
| 4195 |
} |
| 4196 |
|
| 4197 |
/* |
| 4198 |
* Allocate the cpumask for use in mongrp_reparent() to avoid the |
| 4199 |
* possibility of failing to allocate it after kernfs_rename() has |
| 4200 |
* succeeded. |
| 4201 |
*/ |
| 4202 |
if (!zalloc_cpumask_var(&tmpmask, GFP_KERNEL)) { |
| 4203 |
ret = -ENOMEM; |
| 4204 |
goto out; |
| 4205 |
} |
| 4206 |
|
| 4207 |
/* |
| 4208 |
* Perform all input validation and allocations needed to ensure |
| 4209 |
* mongrp_reparent() will succeed before calling kernfs_rename(), |
| 4210 |
* otherwise it would be necessary to revert this call if |
| 4211 |
* mongrp_reparent() failed. |
| 4212 |
*/ |
| 4213 |
ret = kernfs_rename(kn, new_parent, new_name); |
| 4214 |
if (!ret) |
| 4215 |
mongrp_reparent(rdtgrp, new_prdtgrp, tmpmask); |
| 4216 |
|
| 4217 |
free_cpumask_var(tmpmask); |
| 4218 |
|
| 4219 |
out: |
| 4220 |
mutex_unlock(&rdtgroup_mutex); |
| 4221 |
rdtgroup_kn_put(rdtgrp, kn); |
| 4222 |
rdtgroup_kn_put(new_prdtgrp, new_parent); |
| 4223 |
return ret; |
| 4224 |
} |
| 4225 |
|
| 4226 |
static int rdtgroup_show_options(struct seq_file *seq, struct kernfs_root *kf) |
| 4227 |
{ |
| 4228 |
if (resctrl_arch_get_cdp_enabled(RDT_RESOURCE_L3)) |
| 4229 |
seq_puts(seq, ",cdp"); |
| 4230 |
|
| 4231 |
if (resctrl_arch_get_cdp_enabled(RDT_RESOURCE_L2)) |
| 4232 |
seq_puts(seq, ",cdpl2"); |
| 4233 |
|
| 4234 |
if (is_mba_sc(resctrl_arch_get_resource(RDT_RESOURCE_MBA))) |
| 4235 |
seq_puts(seq, ",mba_MBps"); |
| 4236 |
|
| 4237 |
if (resctrl_debug) |
| 4238 |
seq_puts(seq, ",debug"); |
| 4239 |
|
| 4240 |
return 0; |
| 4241 |
} |
| 4242 |
|
| 4243 |
static struct kernfs_syscall_ops rdtgroup_kf_syscall_ops = { |
| 4244 |
.mkdir = rdtgroup_mkdir, |
| 4245 |
.rmdir = rdtgroup_rmdir, |
| 4246 |
.rename = rdtgroup_rename, |
| 4247 |
.show_options = rdtgroup_show_options, |
| 4248 |
}; |
| 4249 |
|
| 4250 |
static int rdtgroup_setup_root(struct rdt_fs_context *ctx) |
| 4251 |
{ |
| 4252 |
rdt_root = kernfs_create_root(&rdtgroup_kf_syscall_ops, |
| 4253 |
KERNFS_ROOT_CREATE_DEACTIVATED | |
| 4254 |
KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK, |
| 4255 |
&rdtgroup_default); |
| 4256 |
if (IS_ERR(rdt_root)) |
| 4257 |
return PTR_ERR(rdt_root); |
| 4258 |
|
| 4259 |
ctx->kfc.root = rdt_root; |
| 4260 |
rdtgroup_default.kn = kernfs_root_to_node(rdt_root); |
| 4261 |
|
| 4262 |
return 0; |
| 4263 |
} |
| 4264 |
|
| 4265 |
static void rdtgroup_destroy_root(void) |
| 4266 |
{ |
| 4267 |
lockdep_assert_held(&rdtgroup_mutex); |
| 4268 |
|
| 4269 |
kernfs_destroy_root(rdt_root); |
| 4270 |
rdtgroup_default.kn = NULL; |
| 4271 |
} |
| 4272 |
|
| 4273 |
static void rdtgroup_setup_default(void) |
| 4274 |
{ |
| 4275 |
mutex_lock(&rdtgroup_mutex); |
| 4276 |
|
| 4277 |
rdtgroup_default.closid = RESCTRL_RESERVED_CLOSID; |
| 4278 |
rdtgroup_default.mon.rmid = RESCTRL_RESERVED_RMID; |
| 4279 |
rdtgroup_default.type = RDTCTRL_GROUP; |
| 4280 |
INIT_LIST_HEAD(&rdtgroup_default.mon.crdtgrp_list); |
| 4281 |
|
| 4282 |
list_add(&rdtgroup_default.rdtgroup_list, &rdt_all_groups); |
| 4283 |
|
| 4284 |
mutex_unlock(&rdtgroup_mutex); |
| 4285 |
} |
| 4286 |
|
| 4287 |
static void domain_destroy_l3_mon_state(struct rdt_l3_mon_domain *d) |
| 4288 |
{ |
| 4289 |
int idx; |
| 4290 |
|
| 4291 |
kfree(d->cntr_cfg); |
| 4292 |
bitmap_free(d->rmid_busy_llc); |
| 4293 |
for_each_mbm_idx(idx) { |
| 4294 |
kfree(d->mbm_states[idx]); |
| 4295 |
d->mbm_states[idx] = NULL; |
| 4296 |
} |
| 4297 |
} |
| 4298 |
|
| 4299 |
void resctrl_offline_ctrl_domain(struct rdt_resource *r, struct rdt_ctrl_domain *d) |
| 4300 |
{ |
| 4301 |
mutex_lock(&rdtgroup_mutex); |
| 4302 |
|
| 4303 |
if (supports_mba_mbps() && r->rid == RDT_RESOURCE_MBA) |
| 4304 |
mba_sc_domain_destroy(r, d); |
| 4305 |
|
| 4306 |
mutex_unlock(&rdtgroup_mutex); |
| 4307 |
} |
| 4308 |
|
| 4309 |
void resctrl_offline_mon_domain(struct rdt_resource *r, struct rdt_domain_hdr *hdr) |
| 4310 |
{ |
| 4311 |
struct rdt_l3_mon_domain *d; |
| 4312 |
|
| 4313 |
mutex_lock(&rdtgroup_mutex); |
| 4314 |
|
| 4315 |
/* |
| 4316 |
* If resctrl is mounted, remove all the |
| 4317 |
* per domain monitor data directories. |
| 4318 |
*/ |
| 4319 |
if (resctrl_mounted && resctrl_arch_mon_capable()) |
| 4320 |
rmdir_mondata_subdir_allrdtgrp(r, hdr); |
| 4321 |
|
| 4322 |
if (r->rid != RDT_RESOURCE_L3) |
| 4323 |
goto out_unlock; |
| 4324 |
|
| 4325 |
if (!domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3)) |
| 4326 |
goto out_unlock; |
| 4327 |
|
| 4328 |
d = container_of(hdr, struct rdt_l3_mon_domain, hdr); |
| 4329 |
if (resctrl_is_mbm_enabled()) |
| 4330 |
cancel_delayed_work(&d->mbm_over); |
| 4331 |
if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID) && has_busy_rmid(d)) { |
| 4332 |
/* |
| 4333 |
* When a package is going down, forcefully |
| 4334 |
* decrement rmid->ebusy. There is no way to know |
| 4335 |
* that the L3 was flushed and hence may lead to |
| 4336 |
* incorrect counts in rare scenarios, but leaving |
| 4337 |
* the RMID as busy creates RMID leaks if the |
| 4338 |
* package never comes back. |
| 4339 |
*/ |
| 4340 |
__check_limbo(d, true); |
| 4341 |
cancel_delayed_work(&d->cqm_limbo); |
| 4342 |
} |
| 4343 |
|
| 4344 |
domain_destroy_l3_mon_state(d); |
| 4345 |
out_unlock: |
| 4346 |
mutex_unlock(&rdtgroup_mutex); |
| 4347 |
} |
| 4348 |
|
| 4349 |
/** |
| 4350 |
* domain_setup_l3_mon_state() - Initialise domain monitoring structures. |
| 4351 |
* @r: The resource for the newly online domain. |
| 4352 |
* @d: The newly online domain. |
| 4353 |
* |
| 4354 |
* Allocate monitor resources that belong to this domain. |
| 4355 |
* Called when the first CPU of a domain comes online, regardless of whether |
| 4356 |
* the filesystem is mounted. |
| 4357 |
* During boot this may be called before global allocations have been made by |
| 4358 |
* resctrl_l3_mon_resource_init(). |
| 4359 |
* |
| 4360 |
* Called during CPU online that may run as soon as CPU online callbacks |
| 4361 |
* are set up during resctrl initialization. The number of supported RMIDs |
| 4362 |
* may be reduced if additional mon_capable resources are enumerated |
| 4363 |
* at mount time. This means the rdt_l3_mon_domain::mbm_states[] and |
| 4364 |
* rdt_l3_mon_domain::rmid_busy_llc allocations may be larger than needed. |
| 4365 |
* |
| 4366 |
* Return: 0 for success, or -ENOMEM. |
| 4367 |
*/ |
| 4368 |
static int domain_setup_l3_mon_state(struct rdt_resource *r, struct rdt_l3_mon_domain *d) |
| 4369 |
{ |
| 4370 |
u32 idx_limit = resctrl_arch_system_num_rmid_idx(); |
| 4371 |
size_t tsize = sizeof(*d->mbm_states[0]); |
| 4372 |
enum resctrl_event_id eventid; |
| 4373 |
int idx; |
| 4374 |
|
| 4375 |
if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID)) { |
| 4376 |
d->rmid_busy_llc = bitmap_zalloc(idx_limit, GFP_KERNEL); |
| 4377 |
if (!d->rmid_busy_llc) |
| 4378 |
return -ENOMEM; |
| 4379 |
} |
| 4380 |
|
| 4381 |
for_each_mbm_event_id(eventid) { |
| 4382 |
if (!resctrl_is_mon_event_enabled(eventid)) |
| 4383 |
continue; |
| 4384 |
idx = MBM_STATE_IDX(eventid); |
| 4385 |
d->mbm_states[idx] = kcalloc(idx_limit, tsize, GFP_KERNEL); |
| 4386 |
if (!d->mbm_states[idx]) |
| 4387 |
goto cleanup; |
| 4388 |
} |
| 4389 |
|
| 4390 |
if (resctrl_is_mbm_enabled() && r->mon.mbm_cntr_assignable) { |
| 4391 |
tsize = sizeof(*d->cntr_cfg); |
| 4392 |
d->cntr_cfg = kcalloc(r->mon.num_mbm_cntrs, tsize, GFP_KERNEL); |
| 4393 |
if (!d->cntr_cfg) |
| 4394 |
goto cleanup; |
| 4395 |
} |
| 4396 |
|
| 4397 |
return 0; |
| 4398 |
cleanup: |
| 4399 |
bitmap_free(d->rmid_busy_llc); |
| 4400 |
for_each_mbm_idx(idx) { |
| 4401 |
kfree(d->mbm_states[idx]); |
| 4402 |
d->mbm_states[idx] = NULL; |
| 4403 |
} |
| 4404 |
|
| 4405 |
return -ENOMEM; |
| 4406 |
} |
| 4407 |
|
| 4408 |
int resctrl_online_ctrl_domain(struct rdt_resource *r, struct rdt_ctrl_domain *d) |
| 4409 |
{ |
| 4410 |
int err = 0; |
| 4411 |
|
| 4412 |
mutex_lock(&rdtgroup_mutex); |
| 4413 |
|
| 4414 |
if (supports_mba_mbps() && r->rid == RDT_RESOURCE_MBA) { |
| 4415 |
/* RDT_RESOURCE_MBA is never mon_capable */ |
| 4416 |
err = mba_sc_domain_allocate(r, d); |
| 4417 |
} |
| 4418 |
|
| 4419 |
mutex_unlock(&rdtgroup_mutex); |
| 4420 |
|
| 4421 |
return err; |
| 4422 |
} |
| 4423 |
|
| 4424 |
int resctrl_online_mon_domain(struct rdt_resource *r, struct rdt_domain_hdr *hdr) |
| 4425 |
{ |
| 4426 |
struct rdt_l3_mon_domain *d; |
| 4427 |
int err = -EINVAL; |
| 4428 |
|
| 4429 |
mutex_lock(&rdtgroup_mutex); |
| 4430 |
|
| 4431 |
if (r->rid != RDT_RESOURCE_L3) |
| 4432 |
goto mkdir; |
| 4433 |
|
| 4434 |
if (!domain_header_is_valid(hdr, RESCTRL_MON_DOMAIN, RDT_RESOURCE_L3)) |
| 4435 |
goto out_unlock; |
| 4436 |
|
| 4437 |
d = container_of(hdr, struct rdt_l3_mon_domain, hdr); |
| 4438 |
err = domain_setup_l3_mon_state(r, d); |
| 4439 |
if (err) |
| 4440 |
goto out_unlock; |
| 4441 |
|
| 4442 |
if (resctrl_is_mbm_enabled()) { |
| 4443 |
INIT_DELAYED_WORK(&d->mbm_over, mbm_handle_overflow); |
| 4444 |
mbm_setup_overflow_handler(d, MBM_OVERFLOW_INTERVAL, |
| 4445 |
RESCTRL_PICK_ANY_CPU); |
| 4446 |
} |
| 4447 |
|
| 4448 |
if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID)) |
| 4449 |
INIT_DELAYED_WORK(&d->cqm_limbo, cqm_handle_limbo); |
| 4450 |
|
| 4451 |
mkdir: |
| 4452 |
err = 0; |
| 4453 |
/* |
| 4454 |
* If the filesystem is not mounted then only the default resource group |
| 4455 |
* exists. Creation of its directories is deferred until mount time |
| 4456 |
* by rdt_get_tree() calling mkdir_mondata_all(). |
| 4457 |
* If resctrl is mounted, add per domain monitor data directories. |
| 4458 |
*/ |
| 4459 |
if (resctrl_mounted && resctrl_arch_mon_capable()) |
| 4460 |
mkdir_mondata_subdir_allrdtgrp(r, hdr); |
| 4461 |
|
| 4462 |
out_unlock: |
| 4463 |
mutex_unlock(&rdtgroup_mutex); |
| 4464 |
|
| 4465 |
return err; |
| 4466 |
} |
| 4467 |
|
| 4468 |
void resctrl_online_cpu(unsigned int cpu) |
| 4469 |
{ |
| 4470 |
mutex_lock(&rdtgroup_mutex); |
| 4471 |
/* The CPU is set in default rdtgroup after online. */ |
| 4472 |
cpumask_set_cpu(cpu, &rdtgroup_default.cpu_mask); |
| 4473 |
mutex_unlock(&rdtgroup_mutex); |
| 4474 |
} |
| 4475 |
|
| 4476 |
static void clear_childcpus(struct rdtgroup *r, unsigned int cpu) |
| 4477 |
{ |
| 4478 |
struct rdtgroup *cr; |
| 4479 |
|
| 4480 |
list_for_each_entry(cr, &r->mon.crdtgrp_list, mon.crdtgrp_list) { |
| 4481 |
if (cpumask_test_and_clear_cpu(cpu, &cr->cpu_mask)) |
| 4482 |
break; |
| 4483 |
} |
| 4484 |
} |
| 4485 |
|
| 4486 |
static struct rdt_l3_mon_domain *get_mon_domain_from_cpu(int cpu, |
| 4487 |
struct rdt_resource *r) |
| 4488 |
{ |
| 4489 |
struct rdt_l3_mon_domain *d; |
| 4490 |
|
| 4491 |
lockdep_assert_cpus_held(); |
| 4492 |
|
| 4493 |
list_for_each_entry(d, &r->mon_domains, hdr.list) { |
| 4494 |
/* Find the domain that contains this CPU */ |
| 4495 |
if (cpumask_test_cpu(cpu, &d->hdr.cpu_mask)) |
| 4496 |
return d; |
| 4497 |
} |
| 4498 |
|
| 4499 |
return NULL; |
| 4500 |
} |
| 4501 |
|
| 4502 |
void resctrl_offline_cpu(unsigned int cpu) |
| 4503 |
{ |
| 4504 |
struct rdt_resource *l3 = resctrl_arch_get_resource(RDT_RESOURCE_L3); |
| 4505 |
struct rdt_l3_mon_domain *d; |
| 4506 |
struct rdtgroup *rdtgrp; |
| 4507 |
|
| 4508 |
mutex_lock(&rdtgroup_mutex); |
| 4509 |
list_for_each_entry(rdtgrp, &rdt_all_groups, rdtgroup_list) { |
| 4510 |
if (cpumask_test_and_clear_cpu(cpu, &rdtgrp->cpu_mask)) { |
| 4511 |
clear_childcpus(rdtgrp, cpu); |
| 4512 |
break; |
| 4513 |
} |
| 4514 |
} |
| 4515 |
|
| 4516 |
if (!l3->mon_capable) |
| 4517 |
goto out_unlock; |
| 4518 |
|
| 4519 |
d = get_mon_domain_from_cpu(cpu, l3); |
| 4520 |
if (d) { |
| 4521 |
if (resctrl_is_mbm_enabled() && cpu == d->mbm_work_cpu) { |
| 4522 |
cancel_delayed_work(&d->mbm_over); |
| 4523 |
mbm_setup_overflow_handler(d, 0, cpu); |
| 4524 |
} |
| 4525 |
if (resctrl_is_mon_event_enabled(QOS_L3_OCCUP_EVENT_ID) && |
| 4526 |
cpu == d->cqm_work_cpu && has_busy_rmid(d)) { |
| 4527 |
cancel_delayed_work(&d->cqm_limbo); |
| 4528 |
cqm_setup_limbo_handler(d, 0, cpu); |
| 4529 |
} |
| 4530 |
} |
| 4531 |
|
| 4532 |
out_unlock: |
| 4533 |
mutex_unlock(&rdtgroup_mutex); |
| 4534 |
} |
| 4535 |
|
| 4536 |
/* |
| 4537 |
* resctrl_init - resctrl filesystem initialization |
| 4538 |
* |
| 4539 |
* Setup resctrl file system including set up root, create mount point, |
| 4540 |
* register resctrl filesystem, and initialize files under root directory. |
| 4541 |
* |
| 4542 |
* Return: 0 on success or -errno |
| 4543 |
*/ |
| 4544 |
int resctrl_init(void) |
| 4545 |
{ |
| 4546 |
int ret = 0; |
| 4547 |
|
| 4548 |
seq_buf_init(&last_cmd_status, last_cmd_status_buf, |
| 4549 |
sizeof(last_cmd_status_buf)); |
| 4550 |
|
| 4551 |
rdtgroup_setup_default(); |
| 4552 |
|
| 4553 |
thread_throttle_mode_init(); |
| 4554 |
|
| 4555 |
io_alloc_init(); |
| 4556 |
|
| 4557 |
ret = resctrl_l3_mon_resource_init(); |
| 4558 |
if (ret) |
| 4559 |
return ret; |
| 4560 |
|
| 4561 |
ret = sysfs_create_mount_point(fs_kobj, "resctrl"); |
| 4562 |
if (ret) { |
| 4563 |
resctrl_l3_mon_resource_exit(); |
| 4564 |
return ret; |
| 4565 |
} |
| 4566 |
|
| 4567 |
ret = register_filesystem(&rdt_fs_type); |
| 4568 |
if (ret) |
| 4569 |
goto cleanup_mountpoint; |
| 4570 |
|
| 4571 |
/* |
| 4572 |
* Adding the resctrl debugfs directory here may not be ideal since |
| 4573 |
* it would let the resctrl debugfs directory appear on the debugfs |
| 4574 |
* filesystem before the resctrl filesystem is mounted. |
| 4575 |
* It may also be ok since that would enable debugging of RDT before |
| 4576 |
* resctrl is mounted. |
| 4577 |
* The reason why the debugfs directory is created here and not in |
| 4578 |
* rdt_get_tree() is because rdt_get_tree() takes rdtgroup_mutex and |
| 4579 |
* during the debugfs directory creation also &sb->s_type->i_mutex_key |
| 4580 |
* (the lockdep class of inode->i_rwsem). Other filesystem |
| 4581 |
* interactions (eg. SyS_getdents) have the lock ordering: |
| 4582 |
* &sb->s_type->i_mutex_key --> &mm->mmap_lock |
| 4583 |
* During mmap(), called with &mm->mmap_lock, the rdtgroup_mutex |
| 4584 |
* is taken, thus creating dependency: |
| 4585 |
* &mm->mmap_lock --> rdtgroup_mutex for the latter that can cause |
| 4586 |
* issues considering the other two lock dependencies. |
| 4587 |
* By creating the debugfs directory here we avoid a dependency |
| 4588 |
* that may cause deadlock (even though file operations cannot |
| 4589 |
* occur until the filesystem is mounted, but I do not know how to |
| 4590 |
* tell lockdep that). |
| 4591 |
*/ |
| 4592 |
debugfs_resctrl = debugfs_create_dir("resctrl", NULL); |
| 4593 |
|
| 4594 |
return 0; |
| 4595 |
|
| 4596 |
cleanup_mountpoint: |
| 4597 |
sysfs_remove_mount_point(fs_kobj, "resctrl"); |
| 4598 |
resctrl_l3_mon_resource_exit(); |
| 4599 |
|
| 4600 |
return ret; |
| 4601 |
} |
| 4602 |
|
| 4603 |
static bool resctrl_online_domains_exist(void) |
| 4604 |
{ |
| 4605 |
struct rdt_resource *r; |
| 4606 |
|
| 4607 |
/* |
| 4608 |
* Only walk capable resources to allow resctrl_arch_get_resource() |
| 4609 |
* to return dummy 'not capable' resources. |
| 4610 |
*/ |
| 4611 |
for_each_alloc_capable_rdt_resource(r) { |
| 4612 |
if (!list_empty(&r->ctrl_domains)) |
| 4613 |
return true; |
| 4614 |
} |
| 4615 |
|
| 4616 |
for_each_mon_capable_rdt_resource(r) { |
| 4617 |
if (!list_empty(&r->mon_domains)) |
| 4618 |
return true; |
| 4619 |
} |
| 4620 |
|
| 4621 |
return false; |
| 4622 |
} |
| 4623 |
|
| 4624 |
/** |
| 4625 |
* resctrl_exit() - Remove the resctrl filesystem and free resources. |
| 4626 |
* |
| 4627 |
* Called by the architecture code in response to a fatal error. |
| 4628 |
* Removes resctrl files and structures from kernfs to prevent further |
| 4629 |
* configuration. |
| 4630 |
* |
| 4631 |
* When called by the architecture code, all CPUs and resctrl domains must be |
| 4632 |
* offline. This ensures the limbo and overflow handlers are not scheduled to |
| 4633 |
* run, meaning the data structures they access can be freed by |
| 4634 |
* resctrl_l3_mon_resource_exit(). |
| 4635 |
* |
| 4636 |
* After resctrl_exit() returns, the architecture code should return an |
| 4637 |
* error from all resctrl_arch_ functions that can do this. |
| 4638 |
* resctrl_arch_get_resource() must continue to return struct rdt_resources |
| 4639 |
* with the correct rid field to ensure the filesystem can be unmounted. |
| 4640 |
*/ |
| 4641 |
void resctrl_exit(void) |
| 4642 |
{ |
| 4643 |
cpus_read_lock(); |
| 4644 |
WARN_ON_ONCE(resctrl_online_domains_exist()); |
| 4645 |
|
| 4646 |
mutex_lock(&rdtgroup_mutex); |
| 4647 |
resctrl_fs_teardown(); |
| 4648 |
mutex_unlock(&rdtgroup_mutex); |
| 4649 |
|
| 4650 |
cpus_read_unlock(); |
| 4651 |
|
| 4652 |
debugfs_remove_recursive(debugfs_resctrl); |
| 4653 |
debugfs_resctrl = NULL; |
| 4654 |
unregister_filesystem(&rdt_fs_type); |
| 4655 |
|
| 4656 |
/* |
| 4657 |
* Do not remove the sysfs mount point added by resctrl_init() so that |
| 4658 |
* it can be used to umount resctrl. |
| 4659 |
*/ |
| 4660 |
|
| 4661 |
resctrl_l3_mon_resource_exit(); |
| 4662 |
free_rmid_lru_list(); |
| 4663 |
} |