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/resctrl.h> 6 #include <linux/sched.h> 7 #include <linux/kernfs.h> 8 #include <linux/fs_context.h> 9 #include <linux/jump_label.h> 10 11 #define L3_QOS_CDP_ENABLE 0x01ULL 12 13 #define L2_QOS_CDP_ENABLE 0x01ULL 14 15 #define CQM_LIMBOCHECK_INTERVAL 1000 16 17 #define MBM_CNTR_WIDTH_BASE 24 18 #define MBM_OVERFLOW_INTERVAL 1000 19 #define MAX_MBA_BW 100u 20 #define MBA_IS_LINEAR 0x4 21 #define MAX_MBA_BW_AMD 0x800 22 #define MBM_CNTR_WIDTH_OFFSET_AMD 20 23 24 #define RMID_VAL_ERROR BIT_ULL(63) 25 #define RMID_VAL_UNAVAIL BIT_ULL(62) 26 /* 27 * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for 28 * data to be returned. The counter width is discovered from the hardware 29 * as an offset from MBM_CNTR_WIDTH_BASE. 30 */ 31 #define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE) 32 33 /* Reads to Local DRAM Memory */ 34 #define READS_TO_LOCAL_MEM BIT(0) 35 36 /* Reads to Remote DRAM Memory */ 37 #define READS_TO_REMOTE_MEM BIT(1) 38 39 /* Non-Temporal Writes to Local Memory */ 40 #define NON_TEMP_WRITE_TO_LOCAL_MEM BIT(2) 41 42 /* Non-Temporal Writes to Remote Memory */ 43 #define NON_TEMP_WRITE_TO_REMOTE_MEM BIT(3) 44 45 /* Reads to Local Memory the system identifies as "Slow Memory" */ 46 #define READS_TO_LOCAL_S_MEM BIT(4) 47 48 /* Reads to Remote Memory the system identifies as "Slow Memory" */ 49 #define READS_TO_REMOTE_S_MEM BIT(5) 50 51 /* Dirty Victims to All Types of Memory */ 52 #define DIRTY_VICTIMS_TO_ALL_MEM BIT(6) 53 54 /* Max event bits supported */ 55 #define MAX_EVT_CONFIG_BITS GENMASK(6, 0) 56 57 struct rdt_fs_context { 58 struct kernfs_fs_context kfc; 59 bool enable_cdpl2; 60 bool enable_cdpl3; 61 bool enable_mba_mbps; 62 }; 63 64 static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc) 65 { 66 struct kernfs_fs_context *kfc = fc->fs_private; 67 68 return container_of(kfc, struct rdt_fs_context, kfc); 69 } 70 71 DECLARE_STATIC_KEY_FALSE(rdt_enable_key); 72 DECLARE_STATIC_KEY_FALSE(rdt_mon_enable_key); 73 74 /** 75 * struct mon_evt - Entry in the event list of a resource 76 * @evtid: event id 77 * @name: name of the event 78 * @configurable: true if the event is configurable 79 * @list: entry in &rdt_resource->evt_list 80 */ 81 struct mon_evt { 82 enum resctrl_event_id evtid; 83 char *name; 84 bool configurable; 85 struct list_head list; 86 }; 87 88 /** 89 * union mon_data_bits - Monitoring details for each event file 90 * @priv: Used to store monitoring event data in @u 91 * as kernfs private data 92 * @rid: Resource id associated with the event file 93 * @evtid: Event id associated with the event file 94 * @domid: The domain to which the event file belongs 95 * @u: Name of the bit fields struct 96 */ 97 union mon_data_bits { 98 void *priv; 99 struct { 100 unsigned int rid : 10; 101 enum resctrl_event_id evtid : 8; 102 unsigned int domid : 14; 103 } u; 104 }; 105 106 struct rmid_read { 107 struct rdtgroup *rgrp; 108 struct rdt_resource *r; 109 struct rdt_domain *d; 110 enum resctrl_event_id evtid; 111 bool first; 112 int err; 113 u64 val; 114 }; 115 116 extern bool rdt_alloc_capable; 117 extern bool rdt_mon_capable; 118 extern unsigned int rdt_mon_features; 119 extern struct list_head resctrl_schema_all; 120 121 enum rdt_group_type { 122 RDTCTRL_GROUP = 0, 123 RDTMON_GROUP, 124 RDT_NUM_GROUP, 125 }; 126 127 /** 128 * enum rdtgrp_mode - Mode of a RDT resource group 129 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations 130 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed 131 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking 132 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations 133 * allowed AND the allocations are Cache Pseudo-Locked 134 * @RDT_NUM_MODES: Total number of modes 135 * 136 * The mode of a resource group enables control over the allowed overlap 137 * between allocations associated with different resource groups (classes 138 * of service). User is able to modify the mode of a resource group by 139 * writing to the "mode" resctrl file associated with the resource group. 140 * 141 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by 142 * writing the appropriate text to the "mode" file. A resource group enters 143 * "pseudo-locked" mode after the schemata is written while the resource 144 * group is in "pseudo-locksetup" mode. 145 */ 146 enum rdtgrp_mode { 147 RDT_MODE_SHAREABLE = 0, 148 RDT_MODE_EXCLUSIVE, 149 RDT_MODE_PSEUDO_LOCKSETUP, 150 RDT_MODE_PSEUDO_LOCKED, 151 152 /* Must be last */ 153 RDT_NUM_MODES, 154 }; 155 156 /** 157 * struct mongroup - store mon group's data in resctrl fs. 158 * @mon_data_kn: kernfs node for the mon_data directory 159 * @parent: parent rdtgrp 160 * @crdtgrp_list: child rdtgroup node list 161 * @rmid: rmid for this rdtgroup 162 */ 163 struct mongroup { 164 struct kernfs_node *mon_data_kn; 165 struct rdtgroup *parent; 166 struct list_head crdtgrp_list; 167 u32 rmid; 168 }; 169 170 /** 171 * struct pseudo_lock_region - pseudo-lock region information 172 * @s: Resctrl schema for the resource to which this 173 * pseudo-locked region belongs 174 * @d: RDT domain to which this pseudo-locked region 175 * belongs 176 * @cbm: bitmask of the pseudo-locked region 177 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread 178 * completion 179 * @thread_done: variable used by waitqueue to test if pseudo-locking 180 * thread completed 181 * @cpu: core associated with the cache on which the setup code 182 * will be run 183 * @line_size: size of the cache lines 184 * @size: size of pseudo-locked region in bytes 185 * @kmem: the kernel memory associated with pseudo-locked region 186 * @minor: minor number of character device associated with this 187 * region 188 * @debugfs_dir: pointer to this region's directory in the debugfs 189 * filesystem 190 * @pm_reqs: Power management QoS requests related to this region 191 */ 192 struct pseudo_lock_region { 193 struct resctrl_schema *s; 194 struct rdt_domain *d; 195 u32 cbm; 196 wait_queue_head_t lock_thread_wq; 197 int thread_done; 198 int cpu; 199 unsigned int line_size; 200 unsigned int size; 201 void *kmem; 202 unsigned int minor; 203 struct dentry *debugfs_dir; 204 struct list_head pm_reqs; 205 }; 206 207 /** 208 * struct rdtgroup - store rdtgroup's data in resctrl file system. 209 * @kn: kernfs node 210 * @rdtgroup_list: linked list for all rdtgroups 211 * @closid: closid for this rdtgroup 212 * @cpu_mask: CPUs assigned to this rdtgroup 213 * @flags: status bits 214 * @waitcount: how many cpus expect to find this 215 * group when they acquire rdtgroup_mutex 216 * @type: indicates type of this rdtgroup - either 217 * monitor only or ctrl_mon group 218 * @mon: mongroup related data 219 * @mode: mode of resource group 220 * @plr: pseudo-locked region 221 */ 222 struct rdtgroup { 223 struct kernfs_node *kn; 224 struct list_head rdtgroup_list; 225 u32 closid; 226 struct cpumask cpu_mask; 227 int flags; 228 atomic_t waitcount; 229 enum rdt_group_type type; 230 struct mongroup mon; 231 enum rdtgrp_mode mode; 232 struct pseudo_lock_region *plr; 233 }; 234 235 /* rdtgroup.flags */ 236 #define RDT_DELETED 1 237 238 /* rftype.flags */ 239 #define RFTYPE_FLAGS_CPUS_LIST 1 240 241 /* 242 * Define the file type flags for base and info directories. 243 */ 244 #define RFTYPE_INFO BIT(0) 245 #define RFTYPE_BASE BIT(1) 246 #define RF_CTRLSHIFT 4 247 #define RF_MONSHIFT 5 248 #define RF_TOPSHIFT 6 249 #define RFTYPE_CTRL BIT(RF_CTRLSHIFT) 250 #define RFTYPE_MON BIT(RF_MONSHIFT) 251 #define RFTYPE_TOP BIT(RF_TOPSHIFT) 252 #define RFTYPE_RES_CACHE BIT(8) 253 #define RFTYPE_RES_MB BIT(9) 254 #define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL) 255 #define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON) 256 #define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP) 257 #define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL) 258 259 /* List of all resource groups */ 260 extern struct list_head rdt_all_groups; 261 262 extern int max_name_width, max_data_width; 263 264 int __init rdtgroup_init(void); 265 void __exit rdtgroup_exit(void); 266 267 /** 268 * struct rftype - describe each file in the resctrl file system 269 * @name: File name 270 * @mode: Access mode 271 * @kf_ops: File operations 272 * @flags: File specific RFTYPE_FLAGS_* flags 273 * @fflags: File specific RF_* or RFTYPE_* flags 274 * @seq_show: Show content of the file 275 * @write: Write to the file 276 */ 277 struct rftype { 278 char *name; 279 umode_t mode; 280 const struct kernfs_ops *kf_ops; 281 unsigned long flags; 282 unsigned long fflags; 283 284 int (*seq_show)(struct kernfs_open_file *of, 285 struct seq_file *sf, void *v); 286 /* 287 * write() is the generic write callback which maps directly to 288 * kernfs write operation and overrides all other operations. 289 * Maximum write size is determined by ->max_write_len. 290 */ 291 ssize_t (*write)(struct kernfs_open_file *of, 292 char *buf, size_t nbytes, loff_t off); 293 }; 294 295 /** 296 * struct mbm_state - status for each MBM counter in each domain 297 * @prev_bw_bytes: Previous bytes value read for bandwidth calculation 298 * @prev_bw: The most recent bandwidth in MBps 299 * @delta_bw: Difference between the current and previous bandwidth 300 * @delta_comp: Indicates whether to compute the delta_bw 301 */ 302 struct mbm_state { 303 u64 prev_bw_bytes; 304 u32 prev_bw; 305 u32 delta_bw; 306 bool delta_comp; 307 }; 308 309 /** 310 * struct arch_mbm_state - values used to compute resctrl_arch_rmid_read()s 311 * return value. 312 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes) 313 * @prev_msr: Value of IA32_QM_CTR last time it was read for the RMID used to 314 * find this struct. 315 */ 316 struct arch_mbm_state { 317 u64 chunks; 318 u64 prev_msr; 319 }; 320 321 /** 322 * struct rdt_hw_domain - Arch private attributes of a set of CPUs that share 323 * a resource 324 * @d_resctrl: Properties exposed to the resctrl file system 325 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID) 326 * @arch_mbm_total: arch private state for MBM total bandwidth 327 * @arch_mbm_local: arch private state for MBM local bandwidth 328 * 329 * Members of this structure are accessed via helpers that provide abstraction. 330 */ 331 struct rdt_hw_domain { 332 struct rdt_domain d_resctrl; 333 u32 *ctrl_val; 334 struct arch_mbm_state *arch_mbm_total; 335 struct arch_mbm_state *arch_mbm_local; 336 }; 337 338 static inline struct rdt_hw_domain *resctrl_to_arch_dom(struct rdt_domain *r) 339 { 340 return container_of(r, struct rdt_hw_domain, d_resctrl); 341 } 342 343 /** 344 * struct msr_param - set a range of MSRs from a domain 345 * @res: The resource to use 346 * @low: Beginning index from base MSR 347 * @high: End index 348 */ 349 struct msr_param { 350 struct rdt_resource *res; 351 u32 low; 352 u32 high; 353 }; 354 355 static inline bool is_llc_occupancy_enabled(void) 356 { 357 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID)); 358 } 359 360 static inline bool is_mbm_total_enabled(void) 361 { 362 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID)); 363 } 364 365 static inline bool is_mbm_local_enabled(void) 366 { 367 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID)); 368 } 369 370 static inline bool is_mbm_enabled(void) 371 { 372 return (is_mbm_total_enabled() || is_mbm_local_enabled()); 373 } 374 375 static inline bool is_mbm_event(int e) 376 { 377 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID && 378 e <= QOS_L3_MBM_LOCAL_EVENT_ID); 379 } 380 381 struct rdt_parse_data { 382 struct rdtgroup *rdtgrp; 383 char *buf; 384 }; 385 386 /** 387 * struct rdt_hw_resource - arch private attributes of a resctrl resource 388 * @r_resctrl: Attributes of the resource used directly by resctrl. 389 * @num_closid: Maximum number of closid this hardware can support, 390 * regardless of CDP. This is exposed via 391 * resctrl_arch_get_num_closid() to avoid confusion 392 * with struct resctrl_schema's property of the same name, 393 * which has been corrected for features like CDP. 394 * @msr_base: Base MSR address for CBMs 395 * @msr_update: Function pointer to update QOS MSRs 396 * @mon_scale: cqm counter * mon_scale = occupancy in bytes 397 * @mbm_width: Monitor width, to detect and correct for overflow. 398 * @cdp_enabled: CDP state of this resource 399 * 400 * Members of this structure are either private to the architecture 401 * e.g. mbm_width, or accessed via helpers that provide abstraction. e.g. 402 * msr_update and msr_base. 403 */ 404 struct rdt_hw_resource { 405 struct rdt_resource r_resctrl; 406 u32 num_closid; 407 unsigned int msr_base; 408 void (*msr_update) (struct rdt_domain *d, struct msr_param *m, 409 struct rdt_resource *r); 410 unsigned int mon_scale; 411 unsigned int mbm_width; 412 bool cdp_enabled; 413 }; 414 415 static inline struct rdt_hw_resource *resctrl_to_arch_res(struct rdt_resource *r) 416 { 417 return container_of(r, struct rdt_hw_resource, r_resctrl); 418 } 419 420 int parse_cbm(struct rdt_parse_data *data, struct resctrl_schema *s, 421 struct rdt_domain *d); 422 int parse_bw(struct rdt_parse_data *data, struct resctrl_schema *s, 423 struct rdt_domain *d); 424 425 extern struct mutex rdtgroup_mutex; 426 427 extern struct rdt_hw_resource rdt_resources_all[]; 428 extern struct rdtgroup rdtgroup_default; 429 DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key); 430 431 extern struct dentry *debugfs_resctrl; 432 433 enum resctrl_res_level { 434 RDT_RESOURCE_L3, 435 RDT_RESOURCE_L2, 436 RDT_RESOURCE_MBA, 437 RDT_RESOURCE_SMBA, 438 439 /* Must be the last */ 440 RDT_NUM_RESOURCES, 441 }; 442 443 static inline struct rdt_resource *resctrl_inc(struct rdt_resource *res) 444 { 445 struct rdt_hw_resource *hw_res = resctrl_to_arch_res(res); 446 447 hw_res++; 448 return &hw_res->r_resctrl; 449 } 450 451 static inline bool resctrl_arch_get_cdp_enabled(enum resctrl_res_level l) 452 { 453 return rdt_resources_all[l].cdp_enabled; 454 } 455 456 int resctrl_arch_set_cdp_enabled(enum resctrl_res_level l, bool enable); 457 458 /* 459 * To return the common struct rdt_resource, which is contained in struct 460 * rdt_hw_resource, walk the resctrl member of struct rdt_hw_resource. 461 */ 462 #define for_each_rdt_resource(r) \ 463 for (r = &rdt_resources_all[0].r_resctrl; \ 464 r <= &rdt_resources_all[RDT_NUM_RESOURCES - 1].r_resctrl; \ 465 r = resctrl_inc(r)) 466 467 #define for_each_capable_rdt_resource(r) \ 468 for_each_rdt_resource(r) \ 469 if (r->alloc_capable || r->mon_capable) 470 471 #define for_each_alloc_capable_rdt_resource(r) \ 472 for_each_rdt_resource(r) \ 473 if (r->alloc_capable) 474 475 #define for_each_mon_capable_rdt_resource(r) \ 476 for_each_rdt_resource(r) \ 477 if (r->mon_capable) 478 479 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */ 480 union cpuid_0x10_1_eax { 481 struct { 482 unsigned int cbm_len:5; 483 } split; 484 unsigned int full; 485 }; 486 487 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */ 488 union cpuid_0x10_3_eax { 489 struct { 490 unsigned int max_delay:12; 491 } split; 492 unsigned int full; 493 }; 494 495 /* CPUID.(EAX=10H, ECX=ResID).EDX */ 496 union cpuid_0x10_x_edx { 497 struct { 498 unsigned int cos_max:16; 499 } split; 500 unsigned int full; 501 }; 502 503 void rdt_last_cmd_clear(void); 504 void rdt_last_cmd_puts(const char *s); 505 __printf(1, 2) 506 void rdt_last_cmd_printf(const char *fmt, ...); 507 508 void rdt_ctrl_update(void *arg); 509 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn); 510 void rdtgroup_kn_unlock(struct kernfs_node *kn); 511 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name); 512 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name, 513 umode_t mask); 514 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id, 515 struct list_head **pos); 516 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of, 517 char *buf, size_t nbytes, loff_t off); 518 int rdtgroup_schemata_show(struct kernfs_open_file *of, 519 struct seq_file *s, void *v); 520 bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_domain *d, 521 unsigned long cbm, int closid, bool exclusive); 522 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d, 523 unsigned long cbm); 524 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid); 525 int rdtgroup_tasks_assigned(struct rdtgroup *r); 526 int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp); 527 int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp); 528 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm); 529 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d); 530 int rdt_pseudo_lock_init(void); 531 void rdt_pseudo_lock_release(void); 532 int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp); 533 void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp); 534 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r); 535 int closids_supported(void); 536 void closid_free(int closid); 537 int alloc_rmid(void); 538 void free_rmid(u32 rmid); 539 int rdt_get_mon_l3_config(struct rdt_resource *r); 540 bool __init rdt_cpu_has(int flag); 541 void mon_event_count(void *info); 542 int rdtgroup_mondata_show(struct seq_file *m, void *arg); 543 void mon_event_read(struct rmid_read *rr, struct rdt_resource *r, 544 struct rdt_domain *d, struct rdtgroup *rdtgrp, 545 int evtid, int first); 546 void mbm_setup_overflow_handler(struct rdt_domain *dom, 547 unsigned long delay_ms); 548 void mbm_handle_overflow(struct work_struct *work); 549 void __init intel_rdt_mbm_apply_quirk(void); 550 bool is_mba_sc(struct rdt_resource *r); 551 void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms); 552 void cqm_handle_limbo(struct work_struct *work); 553 bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d); 554 void __check_limbo(struct rdt_domain *d, bool force_free); 555 void rdt_domain_reconfigure_cdp(struct rdt_resource *r); 556 void __init thread_throttle_mode_init(void); 557 void __init mbm_config_rftype_init(const char *config); 558 void rdt_staged_configs_clear(void); 559 560 #endif /* _ASM_X86_RESCTRL_INTERNAL_H */ 561