1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_RESCTRL_INTERNAL_H 3 #define _ASM_X86_RESCTRL_INTERNAL_H 4 5 #include <linux/sched.h> 6 #include <linux/kernfs.h> 7 #include <linux/fs_context.h> 8 #include <linux/jump_label.h> 9 10 #define MSR_IA32_L3_QOS_CFG 0xc81 11 #define MSR_IA32_L2_QOS_CFG 0xc82 12 #define MSR_IA32_L3_CBM_BASE 0xc90 13 #define MSR_IA32_L2_CBM_BASE 0xd10 14 #define MSR_IA32_MBA_THRTL_BASE 0xd50 15 #define MSR_IA32_MBA_BW_BASE 0xc0000200 16 17 #define MSR_IA32_QM_CTR 0x0c8e 18 #define MSR_IA32_QM_EVTSEL 0x0c8d 19 20 #define L3_QOS_CDP_ENABLE 0x01ULL 21 22 #define L2_QOS_CDP_ENABLE 0x01ULL 23 24 /* 25 * Event IDs are used to program IA32_QM_EVTSEL before reading event 26 * counter from IA32_QM_CTR 27 */ 28 #define QOS_L3_OCCUP_EVENT_ID 0x01 29 #define QOS_L3_MBM_TOTAL_EVENT_ID 0x02 30 #define QOS_L3_MBM_LOCAL_EVENT_ID 0x03 31 32 #define CQM_LIMBOCHECK_INTERVAL 1000 33 34 #define MBM_CNTR_WIDTH 24 35 #define MBM_OVERFLOW_INTERVAL 1000 36 #define MAX_MBA_BW 100u 37 #define MBA_IS_LINEAR 0x4 38 #define MBA_MAX_MBPS U32_MAX 39 #define MAX_MBA_BW_AMD 0x800 40 41 #define RMID_VAL_ERROR BIT_ULL(63) 42 #define RMID_VAL_UNAVAIL BIT_ULL(62) 43 44 45 struct rdt_fs_context { 46 struct kernfs_fs_context kfc; 47 bool enable_cdpl2; 48 bool enable_cdpl3; 49 bool enable_mba_mbps; 50 }; 51 52 static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc) 53 { 54 struct kernfs_fs_context *kfc = fc->fs_private; 55 56 return container_of(kfc, struct rdt_fs_context, kfc); 57 } 58 59 DECLARE_STATIC_KEY_FALSE(rdt_enable_key); 60 DECLARE_STATIC_KEY_FALSE(rdt_mon_enable_key); 61 62 /** 63 * struct mon_evt - Entry in the event list of a resource 64 * @evtid: event id 65 * @name: name of the event 66 */ 67 struct mon_evt { 68 u32 evtid; 69 char *name; 70 struct list_head list; 71 }; 72 73 /** 74 * struct mon_data_bits - Monitoring details for each event file 75 * @rid: Resource id associated with the event file. 76 * @evtid: Event id associated with the event file 77 * @domid: The domain to which the event file belongs 78 */ 79 union mon_data_bits { 80 void *priv; 81 struct { 82 unsigned int rid : 10; 83 unsigned int evtid : 8; 84 unsigned int domid : 14; 85 } u; 86 }; 87 88 struct rmid_read { 89 struct rdtgroup *rgrp; 90 struct rdt_domain *d; 91 int evtid; 92 bool first; 93 u64 val; 94 }; 95 96 extern unsigned int resctrl_cqm_threshold; 97 extern bool rdt_alloc_capable; 98 extern bool rdt_mon_capable; 99 extern unsigned int rdt_mon_features; 100 101 enum rdt_group_type { 102 RDTCTRL_GROUP = 0, 103 RDTMON_GROUP, 104 RDT_NUM_GROUP, 105 }; 106 107 /** 108 * enum rdtgrp_mode - Mode of a RDT resource group 109 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations 110 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed 111 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking 112 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations 113 * allowed AND the allocations are Cache Pseudo-Locked 114 * 115 * The mode of a resource group enables control over the allowed overlap 116 * between allocations associated with different resource groups (classes 117 * of service). User is able to modify the mode of a resource group by 118 * writing to the "mode" resctrl file associated with the resource group. 119 * 120 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by 121 * writing the appropriate text to the "mode" file. A resource group enters 122 * "pseudo-locked" mode after the schemata is written while the resource 123 * group is in "pseudo-locksetup" mode. 124 */ 125 enum rdtgrp_mode { 126 RDT_MODE_SHAREABLE = 0, 127 RDT_MODE_EXCLUSIVE, 128 RDT_MODE_PSEUDO_LOCKSETUP, 129 RDT_MODE_PSEUDO_LOCKED, 130 131 /* Must be last */ 132 RDT_NUM_MODES, 133 }; 134 135 /** 136 * struct mongroup - store mon group's data in resctrl fs. 137 * @mon_data_kn kernlfs node for the mon_data directory 138 * @parent: parent rdtgrp 139 * @crdtgrp_list: child rdtgroup node list 140 * @rmid: rmid for this rdtgroup 141 */ 142 struct mongroup { 143 struct kernfs_node *mon_data_kn; 144 struct rdtgroup *parent; 145 struct list_head crdtgrp_list; 146 u32 rmid; 147 }; 148 149 /** 150 * struct pseudo_lock_region - pseudo-lock region information 151 * @r: RDT resource to which this pseudo-locked region 152 * belongs 153 * @d: RDT domain to which this pseudo-locked region 154 * belongs 155 * @cbm: bitmask of the pseudo-locked region 156 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread 157 * completion 158 * @thread_done: variable used by waitqueue to test if pseudo-locking 159 * thread completed 160 * @cpu: core associated with the cache on which the setup code 161 * will be run 162 * @line_size: size of the cache lines 163 * @size: size of pseudo-locked region in bytes 164 * @kmem: the kernel memory associated with pseudo-locked region 165 * @minor: minor number of character device associated with this 166 * region 167 * @debugfs_dir: pointer to this region's directory in the debugfs 168 * filesystem 169 * @pm_reqs: Power management QoS requests related to this region 170 */ 171 struct pseudo_lock_region { 172 struct rdt_resource *r; 173 struct rdt_domain *d; 174 u32 cbm; 175 wait_queue_head_t lock_thread_wq; 176 int thread_done; 177 int cpu; 178 unsigned int line_size; 179 unsigned int size; 180 void *kmem; 181 unsigned int minor; 182 struct dentry *debugfs_dir; 183 struct list_head pm_reqs; 184 }; 185 186 /** 187 * struct rdtgroup - store rdtgroup's data in resctrl file system. 188 * @kn: kernfs node 189 * @rdtgroup_list: linked list for all rdtgroups 190 * @closid: closid for this rdtgroup 191 * @cpu_mask: CPUs assigned to this rdtgroup 192 * @flags: status bits 193 * @waitcount: how many cpus expect to find this 194 * group when they acquire rdtgroup_mutex 195 * @type: indicates type of this rdtgroup - either 196 * monitor only or ctrl_mon group 197 * @mon: mongroup related data 198 * @mode: mode of resource group 199 * @plr: pseudo-locked region 200 */ 201 struct rdtgroup { 202 struct kernfs_node *kn; 203 struct list_head rdtgroup_list; 204 u32 closid; 205 struct cpumask cpu_mask; 206 int flags; 207 atomic_t waitcount; 208 enum rdt_group_type type; 209 struct mongroup mon; 210 enum rdtgrp_mode mode; 211 struct pseudo_lock_region *plr; 212 }; 213 214 /* rdtgroup.flags */ 215 #define RDT_DELETED 1 216 217 /* rftype.flags */ 218 #define RFTYPE_FLAGS_CPUS_LIST 1 219 220 /* 221 * Define the file type flags for base and info directories. 222 */ 223 #define RFTYPE_INFO BIT(0) 224 #define RFTYPE_BASE BIT(1) 225 #define RF_CTRLSHIFT 4 226 #define RF_MONSHIFT 5 227 #define RF_TOPSHIFT 6 228 #define RFTYPE_CTRL BIT(RF_CTRLSHIFT) 229 #define RFTYPE_MON BIT(RF_MONSHIFT) 230 #define RFTYPE_TOP BIT(RF_TOPSHIFT) 231 #define RFTYPE_RES_CACHE BIT(8) 232 #define RFTYPE_RES_MB BIT(9) 233 #define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL) 234 #define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON) 235 #define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP) 236 #define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL) 237 238 /* List of all resource groups */ 239 extern struct list_head rdt_all_groups; 240 241 extern int max_name_width, max_data_width; 242 243 int __init rdtgroup_init(void); 244 void __exit rdtgroup_exit(void); 245 246 /** 247 * struct rftype - describe each file in the resctrl file system 248 * @name: File name 249 * @mode: Access mode 250 * @kf_ops: File operations 251 * @flags: File specific RFTYPE_FLAGS_* flags 252 * @fflags: File specific RF_* or RFTYPE_* flags 253 * @seq_show: Show content of the file 254 * @write: Write to the file 255 */ 256 struct rftype { 257 char *name; 258 umode_t mode; 259 struct kernfs_ops *kf_ops; 260 unsigned long flags; 261 unsigned long fflags; 262 263 int (*seq_show)(struct kernfs_open_file *of, 264 struct seq_file *sf, void *v); 265 /* 266 * write() is the generic write callback which maps directly to 267 * kernfs write operation and overrides all other operations. 268 * Maximum write size is determined by ->max_write_len. 269 */ 270 ssize_t (*write)(struct kernfs_open_file *of, 271 char *buf, size_t nbytes, loff_t off); 272 }; 273 274 /** 275 * struct mbm_state - status for each MBM counter in each domain 276 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes) 277 * @prev_msr Value of IA32_QM_CTR for this RMID last time we read it 278 * @chunks_bw Total local data moved. Used for bandwidth calculation 279 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting 280 * @prev_bw The most recent bandwidth in MBps 281 * @delta_bw Difference between the current and previous bandwidth 282 * @delta_comp Indicates whether to compute the delta_bw 283 */ 284 struct mbm_state { 285 u64 chunks; 286 u64 prev_msr; 287 u64 chunks_bw; 288 u64 prev_bw_msr; 289 u32 prev_bw; 290 u32 delta_bw; 291 bool delta_comp; 292 }; 293 294 /** 295 * struct rdt_domain - group of cpus sharing an RDT resource 296 * @list: all instances of this resource 297 * @id: unique id for this instance 298 * @cpu_mask: which cpus share this resource 299 * @rmid_busy_llc: 300 * bitmap of which limbo RMIDs are above threshold 301 * @mbm_total: saved state for MBM total bandwidth 302 * @mbm_local: saved state for MBM local bandwidth 303 * @mbm_over: worker to periodically read MBM h/w counters 304 * @cqm_limbo: worker to periodically read CQM h/w counters 305 * @mbm_work_cpu: 306 * worker cpu for MBM h/w counters 307 * @cqm_work_cpu: 308 * worker cpu for CQM h/w counters 309 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID) 310 * @mbps_val: When mba_sc is enabled, this holds the bandwidth in MBps 311 * @new_ctrl: new ctrl value to be loaded 312 * @have_new_ctrl: did user provide new_ctrl for this domain 313 * @plr: pseudo-locked region (if any) associated with domain 314 */ 315 struct rdt_domain { 316 struct list_head list; 317 int id; 318 struct cpumask cpu_mask; 319 unsigned long *rmid_busy_llc; 320 struct mbm_state *mbm_total; 321 struct mbm_state *mbm_local; 322 struct delayed_work mbm_over; 323 struct delayed_work cqm_limbo; 324 int mbm_work_cpu; 325 int cqm_work_cpu; 326 u32 *ctrl_val; 327 u32 *mbps_val; 328 u32 new_ctrl; 329 bool have_new_ctrl; 330 struct pseudo_lock_region *plr; 331 }; 332 333 /** 334 * struct msr_param - set a range of MSRs from a domain 335 * @res: The resource to use 336 * @low: Beginning index from base MSR 337 * @high: End index 338 */ 339 struct msr_param { 340 struct rdt_resource *res; 341 int low; 342 int high; 343 }; 344 345 /** 346 * struct rdt_cache - Cache allocation related data 347 * @cbm_len: Length of the cache bit mask 348 * @min_cbm_bits: Minimum number of consecutive bits to be set 349 * @cbm_idx_mult: Multiplier of CBM index 350 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by: 351 * closid * cbm_idx_multi + cbm_idx_offset 352 * in a cache bit mask 353 * @shareable_bits: Bitmask of shareable resource with other 354 * executing entities 355 */ 356 struct rdt_cache { 357 unsigned int cbm_len; 358 unsigned int min_cbm_bits; 359 unsigned int cbm_idx_mult; 360 unsigned int cbm_idx_offset; 361 unsigned int shareable_bits; 362 }; 363 364 /** 365 * struct rdt_membw - Memory bandwidth allocation related data 366 * @max_delay: Max throttle delay. Delay is the hardware 367 * representation for memory bandwidth. 368 * @min_bw: Minimum memory bandwidth percentage user can request 369 * @bw_gran: Granularity at which the memory bandwidth is allocated 370 * @delay_linear: True if memory B/W delay is in linear scale 371 * @mba_sc: True if MBA software controller(mba_sc) is enabled 372 * @mb_map: Mapping of memory B/W percentage to memory B/W delay 373 */ 374 struct rdt_membw { 375 u32 max_delay; 376 u32 min_bw; 377 u32 bw_gran; 378 u32 delay_linear; 379 bool mba_sc; 380 u32 *mb_map; 381 }; 382 383 static inline bool is_llc_occupancy_enabled(void) 384 { 385 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID)); 386 } 387 388 static inline bool is_mbm_total_enabled(void) 389 { 390 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID)); 391 } 392 393 static inline bool is_mbm_local_enabled(void) 394 { 395 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID)); 396 } 397 398 static inline bool is_mbm_enabled(void) 399 { 400 return (is_mbm_total_enabled() || is_mbm_local_enabled()); 401 } 402 403 static inline bool is_mbm_event(int e) 404 { 405 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID && 406 e <= QOS_L3_MBM_LOCAL_EVENT_ID); 407 } 408 409 struct rdt_parse_data { 410 struct rdtgroup *rdtgrp; 411 char *buf; 412 }; 413 414 /** 415 * struct rdt_resource - attributes of an RDT resource 416 * @rid: The index of the resource 417 * @alloc_enabled: Is allocation enabled on this machine 418 * @mon_enabled: Is monitoring enabled for this feature 419 * @alloc_capable: Is allocation available on this machine 420 * @mon_capable: Is monitor feature available on this machine 421 * @name: Name to use in "schemata" file 422 * @num_closid: Number of CLOSIDs available 423 * @cache_level: Which cache level defines scope of this resource 424 * @default_ctrl: Specifies default cache cbm or memory B/W percent. 425 * @msr_base: Base MSR address for CBMs 426 * @msr_update: Function pointer to update QOS MSRs 427 * @data_width: Character width of data when displaying 428 * @domains: All domains for this resource 429 * @cache: Cache allocation related data 430 * @format_str: Per resource format string to show domain value 431 * @parse_ctrlval: Per resource function pointer to parse control values 432 * @cbm_validate Cache bitmask validate function 433 * @evt_list: List of monitoring events 434 * @num_rmid: Number of RMIDs available 435 * @mon_scale: cqm counter * mon_scale = occupancy in bytes 436 * @fflags: flags to choose base and info files 437 */ 438 struct rdt_resource { 439 int rid; 440 bool alloc_enabled; 441 bool mon_enabled; 442 bool alloc_capable; 443 bool mon_capable; 444 char *name; 445 int num_closid; 446 int cache_level; 447 u32 default_ctrl; 448 unsigned int msr_base; 449 void (*msr_update) (struct rdt_domain *d, struct msr_param *m, 450 struct rdt_resource *r); 451 int data_width; 452 struct list_head domains; 453 struct rdt_cache cache; 454 struct rdt_membw membw; 455 const char *format_str; 456 int (*parse_ctrlval)(struct rdt_parse_data *data, 457 struct rdt_resource *r, 458 struct rdt_domain *d); 459 bool (*cbm_validate)(char *buf, u32 *data, struct rdt_resource *r); 460 struct list_head evt_list; 461 int num_rmid; 462 unsigned int mon_scale; 463 unsigned long fflags; 464 }; 465 466 int parse_cbm(struct rdt_parse_data *data, struct rdt_resource *r, 467 struct rdt_domain *d); 468 int parse_bw_intel(struct rdt_parse_data *data, struct rdt_resource *r, 469 struct rdt_domain *d); 470 int parse_bw_amd(struct rdt_parse_data *data, struct rdt_resource *r, 471 struct rdt_domain *d); 472 473 extern struct mutex rdtgroup_mutex; 474 475 extern struct rdt_resource rdt_resources_all[]; 476 extern struct rdtgroup rdtgroup_default; 477 DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key); 478 479 extern struct dentry *debugfs_resctrl; 480 481 enum { 482 RDT_RESOURCE_L3, 483 RDT_RESOURCE_L3DATA, 484 RDT_RESOURCE_L3CODE, 485 RDT_RESOURCE_L2, 486 RDT_RESOURCE_L2DATA, 487 RDT_RESOURCE_L2CODE, 488 RDT_RESOURCE_MBA, 489 490 /* Must be the last */ 491 RDT_NUM_RESOURCES, 492 }; 493 494 #define for_each_rdt_resource(r) \ 495 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\ 496 r++) 497 498 #define for_each_capable_rdt_resource(r) \ 499 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\ 500 r++) \ 501 if (r->alloc_capable || r->mon_capable) 502 503 #define for_each_alloc_capable_rdt_resource(r) \ 504 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\ 505 r++) \ 506 if (r->alloc_capable) 507 508 #define for_each_mon_capable_rdt_resource(r) \ 509 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\ 510 r++) \ 511 if (r->mon_capable) 512 513 #define for_each_alloc_enabled_rdt_resource(r) \ 514 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\ 515 r++) \ 516 if (r->alloc_enabled) 517 518 #define for_each_mon_enabled_rdt_resource(r) \ 519 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\ 520 r++) \ 521 if (r->mon_enabled) 522 523 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */ 524 union cpuid_0x10_1_eax { 525 struct { 526 unsigned int cbm_len:5; 527 } split; 528 unsigned int full; 529 }; 530 531 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */ 532 union cpuid_0x10_3_eax { 533 struct { 534 unsigned int max_delay:12; 535 } split; 536 unsigned int full; 537 }; 538 539 /* CPUID.(EAX=10H, ECX=ResID).EDX */ 540 union cpuid_0x10_x_edx { 541 struct { 542 unsigned int cos_max:16; 543 } split; 544 unsigned int full; 545 }; 546 547 void rdt_last_cmd_clear(void); 548 void rdt_last_cmd_puts(const char *s); 549 void rdt_last_cmd_printf(const char *fmt, ...); 550 551 void rdt_ctrl_update(void *arg); 552 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn); 553 void rdtgroup_kn_unlock(struct kernfs_node *kn); 554 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name); 555 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name, 556 umode_t mask); 557 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id, 558 struct list_head **pos); 559 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of, 560 char *buf, size_t nbytes, loff_t off); 561 int rdtgroup_schemata_show(struct kernfs_open_file *of, 562 struct seq_file *s, void *v); 563 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d, 564 unsigned long cbm, int closid, bool exclusive); 565 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d, 566 unsigned long cbm); 567 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid); 568 int rdtgroup_tasks_assigned(struct rdtgroup *r); 569 int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp); 570 int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp); 571 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm); 572 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d); 573 int rdt_pseudo_lock_init(void); 574 void rdt_pseudo_lock_release(void); 575 int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp); 576 void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp); 577 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r); 578 int update_domains(struct rdt_resource *r, int closid); 579 int closids_supported(void); 580 void closid_free(int closid); 581 int alloc_rmid(void); 582 void free_rmid(u32 rmid); 583 int rdt_get_mon_l3_config(struct rdt_resource *r); 584 void mon_event_count(void *info); 585 int rdtgroup_mondata_show(struct seq_file *m, void *arg); 586 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, 587 unsigned int dom_id); 588 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, 589 struct rdt_domain *d); 590 void mon_event_read(struct rmid_read *rr, struct rdt_domain *d, 591 struct rdtgroup *rdtgrp, int evtid, int first); 592 void mbm_setup_overflow_handler(struct rdt_domain *dom, 593 unsigned long delay_ms); 594 void mbm_handle_overflow(struct work_struct *work); 595 bool is_mba_sc(struct rdt_resource *r); 596 void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm); 597 u32 delay_bw_map(unsigned long bw, struct rdt_resource *r); 598 void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms); 599 void cqm_handle_limbo(struct work_struct *work); 600 bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d); 601 void __check_limbo(struct rdt_domain *d, bool force_free); 602 bool cbm_validate_intel(char *buf, u32 *data, struct rdt_resource *r); 603 bool cbm_validate_amd(char *buf, u32 *data, struct rdt_resource *r); 604 605 #endif /* _ASM_X86_RESCTRL_INTERNAL_H */ 606