1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_RESCTRL_INTERNAL_H
3 #define _ASM_X86_RESCTRL_INTERNAL_H
5 #include <linux/sched.h>
6 #include <linux/kernfs.h>
7 #include <linux/fs_context.h>
8 #include <linux/jump_label.h>
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
17 #define MSR_IA32_QM_CTR 0x0c8e
18 #define MSR_IA32_QM_EVTSEL 0x0c8d
20 #define L3_QOS_CDP_ENABLE 0x01ULL
22 #define L2_QOS_CDP_ENABLE 0x01ULL
25 * Event IDs are used to program IA32_QM_EVTSEL before reading event
26 * counter from IA32_QM_CTR
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
32 #define CQM_LIMBOCHECK_INTERVAL 1000
34 #define MBM_CNTR_WIDTH_BASE 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
41 #define RMID_VAL_ERROR BIT_ULL(63)
42 #define RMID_VAL_UNAVAIL BIT_ULL(62)
44 * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for
45 * data to be returned. The counter width is discovered from the hardware
46 * as an offset from MBM_CNTR_WIDTH_BASE.
48 #define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
51 struct rdt_fs_context {
52 struct kernfs_fs_context kfc;
58 static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
60 struct kernfs_fs_context *kfc = fc->fs_private;
62 return container_of(kfc, struct rdt_fs_context, kfc);
65 DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
66 DECLARE_STATIC_KEY_FALSE(rdt_mon_enable_key);
69 * struct mon_evt - Entry in the event list of a resource
71 * @name: name of the event
76 struct list_head list;
80 * struct mon_data_bits - Monitoring details for each event file
81 * @rid: Resource id associated with the event file.
82 * @evtid: Event id associated with the event file
83 * @domid: The domain to which the event file belongs
88 unsigned int rid : 10;
89 unsigned int evtid : 8;
90 unsigned int domid : 14;
95 struct rdtgroup *rgrp;
96 struct rdt_resource *r;
103 extern unsigned int resctrl_cqm_threshold;
104 extern bool rdt_alloc_capable;
105 extern bool rdt_mon_capable;
106 extern unsigned int rdt_mon_features;
108 enum rdt_group_type {
115 * enum rdtgrp_mode - Mode of a RDT resource group
116 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
117 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
118 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
119 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
120 * allowed AND the allocations are Cache Pseudo-Locked
122 * The mode of a resource group enables control over the allowed overlap
123 * between allocations associated with different resource groups (classes
124 * of service). User is able to modify the mode of a resource group by
125 * writing to the "mode" resctrl file associated with the resource group.
127 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
128 * writing the appropriate text to the "mode" file. A resource group enters
129 * "pseudo-locked" mode after the schemata is written while the resource
130 * group is in "pseudo-locksetup" mode.
133 RDT_MODE_SHAREABLE = 0,
135 RDT_MODE_PSEUDO_LOCKSETUP,
136 RDT_MODE_PSEUDO_LOCKED,
143 * struct mongroup - store mon group's data in resctrl fs.
144 * @mon_data_kn kernlfs node for the mon_data directory
145 * @parent: parent rdtgrp
146 * @crdtgrp_list: child rdtgroup node list
147 * @rmid: rmid for this rdtgroup
150 struct kernfs_node *mon_data_kn;
151 struct rdtgroup *parent;
152 struct list_head crdtgrp_list;
157 * struct pseudo_lock_region - pseudo-lock region information
158 * @r: RDT resource to which this pseudo-locked region
160 * @d: RDT domain to which this pseudo-locked region
162 * @cbm: bitmask of the pseudo-locked region
163 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread
165 * @thread_done: variable used by waitqueue to test if pseudo-locking
167 * @cpu: core associated with the cache on which the setup code
169 * @line_size: size of the cache lines
170 * @size: size of pseudo-locked region in bytes
171 * @kmem: the kernel memory associated with pseudo-locked region
172 * @minor: minor number of character device associated with this
174 * @debugfs_dir: pointer to this region's directory in the debugfs
176 * @pm_reqs: Power management QoS requests related to this region
178 struct pseudo_lock_region {
179 struct rdt_resource *r;
180 struct rdt_domain *d;
182 wait_queue_head_t lock_thread_wq;
185 unsigned int line_size;
189 struct dentry *debugfs_dir;
190 struct list_head pm_reqs;
194 * struct rdtgroup - store rdtgroup's data in resctrl file system.
196 * @rdtgroup_list: linked list for all rdtgroups
197 * @closid: closid for this rdtgroup
198 * @cpu_mask: CPUs assigned to this rdtgroup
199 * @flags: status bits
200 * @waitcount: how many cpus expect to find this
201 * group when they acquire rdtgroup_mutex
202 * @type: indicates type of this rdtgroup - either
203 * monitor only or ctrl_mon group
204 * @mon: mongroup related data
205 * @mode: mode of resource group
206 * @plr: pseudo-locked region
209 struct kernfs_node *kn;
210 struct list_head rdtgroup_list;
212 struct cpumask cpu_mask;
215 enum rdt_group_type type;
217 enum rdtgrp_mode mode;
218 struct pseudo_lock_region *plr;
222 #define RDT_DELETED 1
225 #define RFTYPE_FLAGS_CPUS_LIST 1
228 * Define the file type flags for base and info directories.
230 #define RFTYPE_INFO BIT(0)
231 #define RFTYPE_BASE BIT(1)
232 #define RF_CTRLSHIFT 4
233 #define RF_MONSHIFT 5
234 #define RF_TOPSHIFT 6
235 #define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
236 #define RFTYPE_MON BIT(RF_MONSHIFT)
237 #define RFTYPE_TOP BIT(RF_TOPSHIFT)
238 #define RFTYPE_RES_CACHE BIT(8)
239 #define RFTYPE_RES_MB BIT(9)
240 #define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
241 #define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
242 #define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
243 #define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
245 /* List of all resource groups */
246 extern struct list_head rdt_all_groups;
248 extern int max_name_width, max_data_width;
250 int __init rdtgroup_init(void);
251 void __exit rdtgroup_exit(void);
254 * struct rftype - describe each file in the resctrl file system
257 * @kf_ops: File operations
258 * @flags: File specific RFTYPE_FLAGS_* flags
259 * @fflags: File specific RF_* or RFTYPE_* flags
260 * @seq_show: Show content of the file
261 * @write: Write to the file
266 struct kernfs_ops *kf_ops;
268 unsigned long fflags;
270 int (*seq_show)(struct kernfs_open_file *of,
271 struct seq_file *sf, void *v);
273 * write() is the generic write callback which maps directly to
274 * kernfs write operation and overrides all other operations.
275 * Maximum write size is determined by ->max_write_len.
277 ssize_t (*write)(struct kernfs_open_file *of,
278 char *buf, size_t nbytes, loff_t off);
282 * struct mbm_state - status for each MBM counter in each domain
283 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
284 * @prev_msr Value of IA32_QM_CTR for this RMID last time we read it
285 * @chunks_bw Total local data moved. Used for bandwidth calculation
286 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting
287 * @prev_bw The most recent bandwidth in MBps
288 * @delta_bw Difference between the current and previous bandwidth
289 * @delta_comp Indicates whether to compute the delta_bw
302 * struct rdt_domain - group of cpus sharing an RDT resource
303 * @list: all instances of this resource
304 * @id: unique id for this instance
305 * @cpu_mask: which cpus share this resource
307 * bitmap of which limbo RMIDs are above threshold
308 * @mbm_total: saved state for MBM total bandwidth
309 * @mbm_local: saved state for MBM local bandwidth
310 * @mbm_over: worker to periodically read MBM h/w counters
311 * @cqm_limbo: worker to periodically read CQM h/w counters
313 * worker cpu for MBM h/w counters
315 * worker cpu for CQM h/w counters
316 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
317 * @mbps_val: When mba_sc is enabled, this holds the bandwidth in MBps
318 * @new_ctrl: new ctrl value to be loaded
319 * @have_new_ctrl: did user provide new_ctrl for this domain
320 * @plr: pseudo-locked region (if any) associated with domain
323 struct list_head list;
325 struct cpumask cpu_mask;
326 unsigned long *rmid_busy_llc;
327 struct mbm_state *mbm_total;
328 struct mbm_state *mbm_local;
329 struct delayed_work mbm_over;
330 struct delayed_work cqm_limbo;
337 struct pseudo_lock_region *plr;
341 * struct msr_param - set a range of MSRs from a domain
342 * @res: The resource to use
343 * @low: Beginning index from base MSR
347 struct rdt_resource *res;
353 * struct rdt_cache - Cache allocation related data
354 * @cbm_len: Length of the cache bit mask
355 * @min_cbm_bits: Minimum number of consecutive bits to be set
356 * @cbm_idx_mult: Multiplier of CBM index
357 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by:
358 * closid * cbm_idx_multi + cbm_idx_offset
359 * in a cache bit mask
360 * @shareable_bits: Bitmask of shareable resource with other
364 unsigned int cbm_len;
365 unsigned int min_cbm_bits;
366 unsigned int cbm_idx_mult;
367 unsigned int cbm_idx_offset;
368 unsigned int shareable_bits;
372 * struct rdt_membw - Memory bandwidth allocation related data
373 * @max_delay: Max throttle delay. Delay is the hardware
374 * representation for memory bandwidth.
375 * @min_bw: Minimum memory bandwidth percentage user can request
376 * @bw_gran: Granularity at which the memory bandwidth is allocated
377 * @delay_linear: True if memory B/W delay is in linear scale
378 * @mba_sc: True if MBA software controller(mba_sc) is enabled
379 * @mb_map: Mapping of memory B/W percentage to memory B/W delay
390 static inline bool is_llc_occupancy_enabled(void)
392 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
395 static inline bool is_mbm_total_enabled(void)
397 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
400 static inline bool is_mbm_local_enabled(void)
402 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
405 static inline bool is_mbm_enabled(void)
407 return (is_mbm_total_enabled() || is_mbm_local_enabled());
410 static inline bool is_mbm_event(int e)
412 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
413 e <= QOS_L3_MBM_LOCAL_EVENT_ID);
416 struct rdt_parse_data {
417 struct rdtgroup *rdtgrp;
422 * struct rdt_resource - attributes of an RDT resource
423 * @rid: The index of the resource
424 * @alloc_enabled: Is allocation enabled on this machine
425 * @mon_enabled: Is monitoring enabled for this feature
426 * @alloc_capable: Is allocation available on this machine
427 * @mon_capable: Is monitor feature available on this machine
428 * @name: Name to use in "schemata" file
429 * @num_closid: Number of CLOSIDs available
430 * @cache_level: Which cache level defines scope of this resource
431 * @default_ctrl: Specifies default cache cbm or memory B/W percent.
432 * @msr_base: Base MSR address for CBMs
433 * @msr_update: Function pointer to update QOS MSRs
434 * @data_width: Character width of data when displaying
435 * @domains: All domains for this resource
436 * @cache: Cache allocation related data
437 * @format_str: Per resource format string to show domain value
438 * @parse_ctrlval: Per resource function pointer to parse control values
439 * @cbm_validate Cache bitmask validate function
440 * @evt_list: List of monitoring events
441 * @num_rmid: Number of RMIDs available
442 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
443 * @fflags: flags to choose base and info files
445 struct rdt_resource {
455 unsigned int msr_base;
456 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
457 struct rdt_resource *r);
459 struct list_head domains;
460 struct rdt_cache cache;
461 struct rdt_membw membw;
462 const char *format_str;
463 int (*parse_ctrlval)(struct rdt_parse_data *data,
464 struct rdt_resource *r,
465 struct rdt_domain *d);
466 bool (*cbm_validate)(char *buf, u32 *data, struct rdt_resource *r);
467 struct list_head evt_list;
469 unsigned int mon_scale;
470 unsigned int mbm_width;
471 unsigned long fflags;
474 int parse_cbm(struct rdt_parse_data *data, struct rdt_resource *r,
475 struct rdt_domain *d);
476 int parse_bw_intel(struct rdt_parse_data *data, struct rdt_resource *r,
477 struct rdt_domain *d);
478 int parse_bw_amd(struct rdt_parse_data *data, struct rdt_resource *r,
479 struct rdt_domain *d);
481 extern struct mutex rdtgroup_mutex;
483 extern struct rdt_resource rdt_resources_all[];
484 extern struct rdtgroup rdtgroup_default;
485 DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
487 extern struct dentry *debugfs_resctrl;
498 /* Must be the last */
502 #define for_each_rdt_resource(r) \
503 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
506 #define for_each_capable_rdt_resource(r) \
507 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
509 if (r->alloc_capable || r->mon_capable)
511 #define for_each_alloc_capable_rdt_resource(r) \
512 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
514 if (r->alloc_capable)
516 #define for_each_mon_capable_rdt_resource(r) \
517 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
521 #define for_each_alloc_enabled_rdt_resource(r) \
522 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
524 if (r->alloc_enabled)
526 #define for_each_mon_enabled_rdt_resource(r) \
527 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
531 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */
532 union cpuid_0x10_1_eax {
534 unsigned int cbm_len:5;
539 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */
540 union cpuid_0x10_3_eax {
542 unsigned int max_delay:12;
547 /* CPUID.(EAX=10H, ECX=ResID).EDX */
548 union cpuid_0x10_x_edx {
550 unsigned int cos_max:16;
555 void rdt_last_cmd_clear(void);
556 void rdt_last_cmd_puts(const char *s);
557 void rdt_last_cmd_printf(const char *fmt, ...);
559 void rdt_ctrl_update(void *arg);
560 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
561 void rdtgroup_kn_unlock(struct kernfs_node *kn);
562 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
563 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
565 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
566 struct list_head **pos);
567 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
568 char *buf, size_t nbytes, loff_t off);
569 int rdtgroup_schemata_show(struct kernfs_open_file *of,
570 struct seq_file *s, void *v);
571 bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
572 unsigned long cbm, int closid, bool exclusive);
573 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
575 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
576 int rdtgroup_tasks_assigned(struct rdtgroup *r);
577 int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
578 int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
579 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
580 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
581 int rdt_pseudo_lock_init(void);
582 void rdt_pseudo_lock_release(void);
583 int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
584 void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
585 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
586 int update_domains(struct rdt_resource *r, int closid);
587 int closids_supported(void);
588 void closid_free(int closid);
589 int alloc_rmid(void);
590 void free_rmid(u32 rmid);
591 int rdt_get_mon_l3_config(struct rdt_resource *r);
592 void mon_event_count(void *info);
593 int rdtgroup_mondata_show(struct seq_file *m, void *arg);
594 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
595 unsigned int dom_id);
596 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
597 struct rdt_domain *d);
598 void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
599 struct rdt_domain *d, struct rdtgroup *rdtgrp,
600 int evtid, int first);
601 void mbm_setup_overflow_handler(struct rdt_domain *dom,
602 unsigned long delay_ms);
603 void mbm_handle_overflow(struct work_struct *work);
604 bool is_mba_sc(struct rdt_resource *r);
605 void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm);
606 u32 delay_bw_map(unsigned long bw, struct rdt_resource *r);
607 void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
608 void cqm_handle_limbo(struct work_struct *work);
609 bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
610 void __check_limbo(struct rdt_domain *d, bool force_free);
611 bool cbm_validate_intel(char *buf, u32 *data, struct rdt_resource *r);
612 bool cbm_validate_amd(char *buf, u32 *data, struct rdt_resource *r);
613 void rdt_domain_reconfigure_cdp(struct rdt_resource *r);
615 #endif /* _ASM_X86_RESCTRL_INTERNAL_H */