1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_RESCTRL_INTERNAL_H
3 #define _ASM_X86_RESCTRL_INTERNAL_H
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 #include <linux/tick.h>
12 #include <asm/resctrl.h>
14 #define L3_QOS_CDP_ENABLE 0x01ULL
16 #define L2_QOS_CDP_ENABLE 0x01ULL
18 #define CQM_LIMBOCHECK_INTERVAL 1000
20 #define MBM_CNTR_WIDTH_BASE 24
21 #define MBM_OVERFLOW_INTERVAL 1000
22 #define MAX_MBA_BW 100u
23 #define MBA_IS_LINEAR 0x4
24 #define MBM_CNTR_WIDTH_OFFSET_AMD 20
26 #define RMID_VAL_ERROR BIT_ULL(63)
27 #define RMID_VAL_UNAVAIL BIT_ULL(62)
29 * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for
30 * data to be returned. The counter width is discovered from the hardware
31 * as an offset from MBM_CNTR_WIDTH_BASE.
33 #define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
35 /* Reads to Local DRAM Memory */
36 #define READS_TO_LOCAL_MEM BIT(0)
38 /* Reads to Remote DRAM Memory */
39 #define READS_TO_REMOTE_MEM BIT(1)
41 /* Non-Temporal Writes to Local Memory */
42 #define NON_TEMP_WRITE_TO_LOCAL_MEM BIT(2)
44 /* Non-Temporal Writes to Remote Memory */
45 #define NON_TEMP_WRITE_TO_REMOTE_MEM BIT(3)
47 /* Reads to Local Memory the system identifies as "Slow Memory" */
48 #define READS_TO_LOCAL_S_MEM BIT(4)
50 /* Reads to Remote Memory the system identifies as "Slow Memory" */
51 #define READS_TO_REMOTE_S_MEM BIT(5)
53 /* Dirty Victims to All Types of Memory */
54 #define DIRTY_VICTIMS_TO_ALL_MEM BIT(6)
56 /* Max event bits supported */
57 #define MAX_EVT_CONFIG_BITS GENMASK(6, 0)
60 * cpumask_any_housekeeping() - Choose any CPU in @mask, preferring those that
61 * aren't marked nohz_full
62 * @mask: The mask to pick a CPU from.
63 * @exclude_cpu:The CPU to avoid picking.
65 * Returns a CPU from @mask, but not @exclude_cpu. If there are housekeeping
66 * CPUs that don't use nohz_full, these are preferred. Pass
67 * RESCTRL_PICK_ANY_CPU to avoid excluding any CPUs.
69 * When a CPU is excluded, returns >= nr_cpu_ids if no CPUs are available.
71 static inline unsigned int
72 cpumask_any_housekeeping(const struct cpumask *mask, int exclude_cpu)
74 unsigned int cpu, hk_cpu;
76 if (exclude_cpu == RESCTRL_PICK_ANY_CPU)
77 cpu = cpumask_any(mask);
79 cpu = cpumask_any_but(mask, exclude_cpu);
81 if (!IS_ENABLED(CONFIG_NO_HZ_FULL))
84 /* If the CPU picked isn't marked nohz_full nothing more needs doing. */
85 if (cpu < nr_cpu_ids && !tick_nohz_full_cpu(cpu))
88 /* Try to find a CPU that isn't nohz_full to use in preference */
89 hk_cpu = cpumask_nth_andnot(0, mask, tick_nohz_full_mask);
90 if (hk_cpu == exclude_cpu)
91 hk_cpu = cpumask_nth_andnot(1, mask, tick_nohz_full_mask);
93 if (hk_cpu < nr_cpu_ids)
99 struct rdt_fs_context {
100 struct kernfs_fs_context kfc;
103 bool enable_mba_mbps;
107 static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
109 struct kernfs_fs_context *kfc = fc->fs_private;
111 return container_of(kfc, struct rdt_fs_context, kfc);
115 * struct mon_evt - Entry in the event list of a resource
117 * @name: name of the event
118 * @configurable: true if the event is configurable
119 * @list: entry in &rdt_resource->evt_list
122 enum resctrl_event_id evtid;
125 struct list_head list;
129 * union mon_data_bits - Monitoring details for each event file
130 * @priv: Used to store monitoring event data in @u
131 * as kernfs private data
132 * @rid: Resource id associated with the event file
133 * @evtid: Event id associated with the event file
134 * @domid: The domain to which the event file belongs
135 * @u: Name of the bit fields struct
137 union mon_data_bits {
140 unsigned int rid : 10;
141 enum resctrl_event_id evtid : 8;
142 unsigned int domid : 14;
147 struct rdtgroup *rgrp;
148 struct rdt_resource *r;
149 struct rdt_domain *d;
150 enum resctrl_event_id evtid;
157 extern unsigned int rdt_mon_features;
158 extern struct list_head resctrl_schema_all;
159 extern bool resctrl_mounted;
161 enum rdt_group_type {
168 * enum rdtgrp_mode - Mode of a RDT resource group
169 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
170 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
171 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
172 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
173 * allowed AND the allocations are Cache Pseudo-Locked
174 * @RDT_NUM_MODES: Total number of modes
176 * The mode of a resource group enables control over the allowed overlap
177 * between allocations associated with different resource groups (classes
178 * of service). User is able to modify the mode of a resource group by
179 * writing to the "mode" resctrl file associated with the resource group.
181 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
182 * writing the appropriate text to the "mode" file. A resource group enters
183 * "pseudo-locked" mode after the schemata is written while the resource
184 * group is in "pseudo-locksetup" mode.
187 RDT_MODE_SHAREABLE = 0,
189 RDT_MODE_PSEUDO_LOCKSETUP,
190 RDT_MODE_PSEUDO_LOCKED,
197 * struct mongroup - store mon group's data in resctrl fs.
198 * @mon_data_kn: kernfs node for the mon_data directory
199 * @parent: parent rdtgrp
200 * @crdtgrp_list: child rdtgroup node list
201 * @rmid: rmid for this rdtgroup
204 struct kernfs_node *mon_data_kn;
205 struct rdtgroup *parent;
206 struct list_head crdtgrp_list;
211 * struct pseudo_lock_region - pseudo-lock region information
212 * @s: Resctrl schema for the resource to which this
213 * pseudo-locked region belongs
214 * @d: RDT domain to which this pseudo-locked region
216 * @cbm: bitmask of the pseudo-locked region
217 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread
219 * @thread_done: variable used by waitqueue to test if pseudo-locking
221 * @cpu: core associated with the cache on which the setup code
223 * @line_size: size of the cache lines
224 * @size: size of pseudo-locked region in bytes
225 * @kmem: the kernel memory associated with pseudo-locked region
226 * @minor: minor number of character device associated with this
228 * @debugfs_dir: pointer to this region's directory in the debugfs
230 * @pm_reqs: Power management QoS requests related to this region
232 struct pseudo_lock_region {
233 struct resctrl_schema *s;
234 struct rdt_domain *d;
236 wait_queue_head_t lock_thread_wq;
239 unsigned int line_size;
243 struct dentry *debugfs_dir;
244 struct list_head pm_reqs;
248 * struct rdtgroup - store rdtgroup's data in resctrl file system.
250 * @rdtgroup_list: linked list for all rdtgroups
251 * @closid: closid for this rdtgroup
252 * @cpu_mask: CPUs assigned to this rdtgroup
253 * @flags: status bits
254 * @waitcount: how many cpus expect to find this
255 * group when they acquire rdtgroup_mutex
256 * @type: indicates type of this rdtgroup - either
257 * monitor only or ctrl_mon group
258 * @mon: mongroup related data
259 * @mode: mode of resource group
260 * @plr: pseudo-locked region
263 struct kernfs_node *kn;
264 struct list_head rdtgroup_list;
266 struct cpumask cpu_mask;
269 enum rdt_group_type type;
271 enum rdtgrp_mode mode;
272 struct pseudo_lock_region *plr;
276 #define RDT_DELETED 1
279 #define RFTYPE_FLAGS_CPUS_LIST 1
282 * Define the file type flags for base and info directories.
284 #define RFTYPE_INFO BIT(0)
285 #define RFTYPE_BASE BIT(1)
286 #define RFTYPE_CTRL BIT(4)
287 #define RFTYPE_MON BIT(5)
288 #define RFTYPE_TOP BIT(6)
289 #define RFTYPE_RES_CACHE BIT(8)
290 #define RFTYPE_RES_MB BIT(9)
291 #define RFTYPE_DEBUG BIT(10)
292 #define RFTYPE_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
293 #define RFTYPE_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
294 #define RFTYPE_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
295 #define RFTYPE_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
296 #define RFTYPE_MON_BASE (RFTYPE_BASE | RFTYPE_MON)
298 /* List of all resource groups */
299 extern struct list_head rdt_all_groups;
301 extern int max_name_width, max_data_width;
303 int __init rdtgroup_init(void);
304 void __exit rdtgroup_exit(void);
307 * struct rftype - describe each file in the resctrl file system
310 * @kf_ops: File operations
311 * @flags: File specific RFTYPE_FLAGS_* flags
312 * @fflags: File specific RFTYPE_* flags
313 * @seq_show: Show content of the file
314 * @write: Write to the file
319 const struct kernfs_ops *kf_ops;
321 unsigned long fflags;
323 int (*seq_show)(struct kernfs_open_file *of,
324 struct seq_file *sf, void *v);
326 * write() is the generic write callback which maps directly to
327 * kernfs write operation and overrides all other operations.
328 * Maximum write size is determined by ->max_write_len.
330 ssize_t (*write)(struct kernfs_open_file *of,
331 char *buf, size_t nbytes, loff_t off);
335 * struct mbm_state - status for each MBM counter in each domain
336 * @prev_bw_bytes: Previous bytes value read for bandwidth calculation
337 * @prev_bw: The most recent bandwidth in MBps
345 * struct arch_mbm_state - values used to compute resctrl_arch_rmid_read()s
347 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
348 * @prev_msr: Value of IA32_QM_CTR last time it was read for the RMID used to
351 struct arch_mbm_state {
357 * struct rdt_hw_domain - Arch private attributes of a set of CPUs that share
359 * @d_resctrl: Properties exposed to the resctrl file system
360 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
361 * @arch_mbm_total: arch private state for MBM total bandwidth
362 * @arch_mbm_local: arch private state for MBM local bandwidth
364 * Members of this structure are accessed via helpers that provide abstraction.
366 struct rdt_hw_domain {
367 struct rdt_domain d_resctrl;
369 struct arch_mbm_state *arch_mbm_total;
370 struct arch_mbm_state *arch_mbm_local;
373 static inline struct rdt_hw_domain *resctrl_to_arch_dom(struct rdt_domain *r)
375 return container_of(r, struct rdt_hw_domain, d_resctrl);
379 * struct msr_param - set a range of MSRs from a domain
380 * @res: The resource to use
381 * @low: Beginning index from base MSR
385 struct rdt_resource *res;
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_hw_resource - arch private attributes of a resctrl resource
423 * @r_resctrl: Attributes of the resource used directly by resctrl.
424 * @num_closid: Maximum number of closid this hardware can support,
425 * regardless of CDP. This is exposed via
426 * resctrl_arch_get_num_closid() to avoid confusion
427 * with struct resctrl_schema's property of the same name,
428 * which has been corrected for features like CDP.
429 * @msr_base: Base MSR address for CBMs
430 * @msr_update: Function pointer to update QOS MSRs
431 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
432 * @mbm_width: Monitor width, to detect and correct for overflow.
433 * @mbm_cfg_mask: Bandwidth sources that can be tracked when Bandwidth
434 * Monitoring Event Configuration (BMEC) is supported.
435 * @cdp_enabled: CDP state of this resource
437 * Members of this structure are either private to the architecture
438 * e.g. mbm_width, or accessed via helpers that provide abstraction. e.g.
439 * msr_update and msr_base.
441 struct rdt_hw_resource {
442 struct rdt_resource r_resctrl;
444 unsigned int msr_base;
445 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
446 struct rdt_resource *r);
447 unsigned int mon_scale;
448 unsigned int mbm_width;
449 unsigned int mbm_cfg_mask;
453 static inline struct rdt_hw_resource *resctrl_to_arch_res(struct rdt_resource *r)
455 return container_of(r, struct rdt_hw_resource, r_resctrl);
458 int parse_cbm(struct rdt_parse_data *data, struct resctrl_schema *s,
459 struct rdt_domain *d);
460 int parse_bw(struct rdt_parse_data *data, struct resctrl_schema *s,
461 struct rdt_domain *d);
463 extern struct mutex rdtgroup_mutex;
465 extern struct rdt_hw_resource rdt_resources_all[];
466 extern struct rdtgroup rdtgroup_default;
467 extern struct dentry *debugfs_resctrl;
469 enum resctrl_res_level {
475 /* Must be the last */
479 static inline struct rdt_resource *resctrl_inc(struct rdt_resource *res)
481 struct rdt_hw_resource *hw_res = resctrl_to_arch_res(res);
484 return &hw_res->r_resctrl;
487 static inline bool resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)
489 return rdt_resources_all[l].cdp_enabled;
492 int resctrl_arch_set_cdp_enabled(enum resctrl_res_level l, bool enable);
495 * To return the common struct rdt_resource, which is contained in struct
496 * rdt_hw_resource, walk the resctrl member of struct rdt_hw_resource.
498 #define for_each_rdt_resource(r) \
499 for (r = &rdt_resources_all[0].r_resctrl; \
500 r <= &rdt_resources_all[RDT_NUM_RESOURCES - 1].r_resctrl; \
503 #define for_each_capable_rdt_resource(r) \
504 for_each_rdt_resource(r) \
505 if (r->alloc_capable || r->mon_capable)
507 #define for_each_alloc_capable_rdt_resource(r) \
508 for_each_rdt_resource(r) \
509 if (r->alloc_capable)
511 #define for_each_mon_capable_rdt_resource(r) \
512 for_each_rdt_resource(r) \
515 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */
516 union cpuid_0x10_1_eax {
518 unsigned int cbm_len:5;
523 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */
524 union cpuid_0x10_3_eax {
526 unsigned int max_delay:12;
531 /* CPUID.(EAX=10H, ECX=ResID).ECX */
532 union cpuid_0x10_x_ecx {
534 unsigned int reserved:3;
535 unsigned int noncont:1;
540 /* CPUID.(EAX=10H, ECX=ResID).EDX */
541 union cpuid_0x10_x_edx {
543 unsigned int cos_max:16;
548 void rdt_last_cmd_clear(void);
549 void rdt_last_cmd_puts(const char *s);
551 void rdt_last_cmd_printf(const char *fmt, ...);
553 void rdt_ctrl_update(void *arg);
554 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
555 void rdtgroup_kn_unlock(struct kernfs_node *kn);
556 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
557 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
559 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
560 struct list_head **pos);
561 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
562 char *buf, size_t nbytes, loff_t off);
563 int rdtgroup_schemata_show(struct kernfs_open_file *of,
564 struct seq_file *s, void *v);
565 bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_domain *d,
566 unsigned long cbm, int closid, bool exclusive);
567 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
569 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
570 int rdtgroup_tasks_assigned(struct rdtgroup *r);
571 int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
572 int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
573 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
574 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
575 int rdt_pseudo_lock_init(void);
576 void rdt_pseudo_lock_release(void);
577 int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
578 void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
579 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
580 int closids_supported(void);
581 void closid_free(int closid);
582 int alloc_rmid(u32 closid);
583 void free_rmid(u32 closid, u32 rmid);
584 int rdt_get_mon_l3_config(struct rdt_resource *r);
585 void __exit rdt_put_mon_l3_config(void);
586 bool __init rdt_cpu_has(int flag);
587 void mon_event_count(void *info);
588 int rdtgroup_mondata_show(struct seq_file *m, void *arg);
589 void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
590 struct rdt_domain *d, struct rdtgroup *rdtgrp,
591 int evtid, int first);
592 void mbm_setup_overflow_handler(struct rdt_domain *dom,
593 unsigned long delay_ms,
595 void mbm_handle_overflow(struct work_struct *work);
596 void __init intel_rdt_mbm_apply_quirk(void);
597 bool is_mba_sc(struct rdt_resource *r);
598 void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms,
600 void cqm_handle_limbo(struct work_struct *work);
601 bool has_busy_rmid(struct rdt_domain *d);
602 void __check_limbo(struct rdt_domain *d, bool force_free);
603 void rdt_domain_reconfigure_cdp(struct rdt_resource *r);
604 void __init thread_throttle_mode_init(void);
605 void __init mbm_config_rftype_init(const char *config);
606 void rdt_staged_configs_clear(void);
607 bool closid_allocated(unsigned int closid);
608 int resctrl_find_cleanest_closid(void);
610 #endif /* _ASM_X86_RESCTRL_INTERNAL_H */