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#include <linux/tick.h>
11
12#include <asm/resctrl.h>
13
14#define L3_QOS_CDP_ENABLE 0x01ULL
15
16#define L2_QOS_CDP_ENABLE 0x01ULL
17
18#define CQM_LIMBOCHECK_INTERVAL 1000
19
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
25
26#define RMID_VAL_ERROR BIT_ULL(63)
27#define RMID_VAL_UNAVAIL BIT_ULL(62)
28/*
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.
32 */
33#define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
34
35/* Reads to Local DRAM Memory */
36#define READS_TO_LOCAL_MEM BIT(0)
37
38/* Reads to Remote DRAM Memory */
39#define READS_TO_REMOTE_MEM BIT(1)
40
41/* Non-Temporal Writes to Local Memory */
42#define NON_TEMP_WRITE_TO_LOCAL_MEM BIT(2)
43
44/* Non-Temporal Writes to Remote Memory */
45#define NON_TEMP_WRITE_TO_REMOTE_MEM BIT(3)
46
47/* Reads to Local Memory the system identifies as "Slow Memory" */
48#define READS_TO_LOCAL_S_MEM BIT(4)
49
50/* Reads to Remote Memory the system identifies as "Slow Memory" */
51#define READS_TO_REMOTE_S_MEM BIT(5)
52
53/* Dirty Victims to All Types of Memory */
54#define DIRTY_VICTIMS_TO_ALL_MEM BIT(6)
55
56/* Max event bits supported */
57#define MAX_EVT_CONFIG_BITS GENMASK(6, 0)
58
59/**
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.
64 *
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.
68 *
69 * When a CPU is excluded, returns >= nr_cpu_ids if no CPUs are available.
70 */
71static inline unsigned int
72cpumask_any_housekeeping(const struct cpumask *mask, int exclude_cpu)
73{
74 unsigned int cpu, hk_cpu;
75
76 if (exclude_cpu == RESCTRL_PICK_ANY_CPU)
77 cpu = cpumask_any(mask);
78 else
79 cpu = cpumask_any_but(mask, cpu: exclude_cpu);
80
81 /* Only continue if tick_nohz_full_mask has been initialized. */
82 if (!tick_nohz_full_enabled())
83 return cpu;
84
85 /* If the CPU picked isn't marked nohz_full nothing more needs doing. */
86 if (cpu < nr_cpu_ids && !tick_nohz_full_cpu(cpu))
87 return cpu;
88
89 /* Try to find a CPU that isn't nohz_full to use in preference */
90 hk_cpu = cpumask_nth_andnot(cpu: 0, srcp1: mask, srcp2: tick_nohz_full_mask);
91 if (hk_cpu == exclude_cpu)
92 hk_cpu = cpumask_nth_andnot(cpu: 1, srcp1: mask, srcp2: tick_nohz_full_mask);
93
94 if (hk_cpu < nr_cpu_ids)
95 cpu = hk_cpu;
96
97 return cpu;
98}
99
100struct rdt_fs_context {
101 struct kernfs_fs_context kfc;
102 bool enable_cdpl2;
103 bool enable_cdpl3;
104 bool enable_mba_mbps;
105 bool enable_debug;
106};
107
108static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
109{
110 struct kernfs_fs_context *kfc = fc->fs_private;
111
112 return container_of(kfc, struct rdt_fs_context, kfc);
113}
114
115/**
116 * struct mon_evt - Entry in the event list of a resource
117 * @evtid: event id
118 * @name: name of the event
119 * @configurable: true if the event is configurable
120 * @list: entry in &rdt_resource->evt_list
121 */
122struct mon_evt {
123 enum resctrl_event_id evtid;
124 char *name;
125 bool configurable;
126 struct list_head list;
127};
128
129/**
130 * union mon_data_bits - Monitoring details for each event file
131 * @priv: Used to store monitoring event data in @u
132 * as kernfs private data
133 * @rid: Resource id associated with the event file
134 * @evtid: Event id associated with the event file
135 * @domid: The domain to which the event file belongs
136 * @u: Name of the bit fields struct
137 */
138union mon_data_bits {
139 void *priv;
140 struct {
141 unsigned int rid : 10;
142 enum resctrl_event_id evtid : 8;
143 unsigned int domid : 14;
144 } u;
145};
146
147struct rmid_read {
148 struct rdtgroup *rgrp;
149 struct rdt_resource *r;
150 struct rdt_domain *d;
151 enum resctrl_event_id evtid;
152 bool first;
153 int err;
154 u64 val;
155 void *arch_mon_ctx;
156};
157
158extern unsigned int rdt_mon_features;
159extern struct list_head resctrl_schema_all;
160extern bool resctrl_mounted;
161
162enum rdt_group_type {
163 RDTCTRL_GROUP = 0,
164 RDTMON_GROUP,
165 RDT_NUM_GROUP,
166};
167
168/**
169 * enum rdtgrp_mode - Mode of a RDT resource group
170 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
171 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
172 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
173 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
174 * allowed AND the allocations are Cache Pseudo-Locked
175 * @RDT_NUM_MODES: Total number of modes
176 *
177 * The mode of a resource group enables control over the allowed overlap
178 * between allocations associated with different resource groups (classes
179 * of service). User is able to modify the mode of a resource group by
180 * writing to the "mode" resctrl file associated with the resource group.
181 *
182 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
183 * writing the appropriate text to the "mode" file. A resource group enters
184 * "pseudo-locked" mode after the schemata is written while the resource
185 * group is in "pseudo-locksetup" mode.
186 */
187enum rdtgrp_mode {
188 RDT_MODE_SHAREABLE = 0,
189 RDT_MODE_EXCLUSIVE,
190 RDT_MODE_PSEUDO_LOCKSETUP,
191 RDT_MODE_PSEUDO_LOCKED,
192
193 /* Must be last */
194 RDT_NUM_MODES,
195};
196
197/**
198 * struct mongroup - store mon group's data in resctrl fs.
199 * @mon_data_kn: kernfs node for the mon_data directory
200 * @parent: parent rdtgrp
201 * @crdtgrp_list: child rdtgroup node list
202 * @rmid: rmid for this rdtgroup
203 */
204struct mongroup {
205 struct kernfs_node *mon_data_kn;
206 struct rdtgroup *parent;
207 struct list_head crdtgrp_list;
208 u32 rmid;
209};
210
211/**
212 * struct pseudo_lock_region - pseudo-lock region information
213 * @s: Resctrl schema for the resource to which this
214 * pseudo-locked region belongs
215 * @d: RDT domain to which this pseudo-locked region
216 * belongs
217 * @cbm: bitmask of the pseudo-locked region
218 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread
219 * completion
220 * @thread_done: variable used by waitqueue to test if pseudo-locking
221 * thread completed
222 * @cpu: core associated with the cache on which the setup code
223 * will be run
224 * @line_size: size of the cache lines
225 * @size: size of pseudo-locked region in bytes
226 * @kmem: the kernel memory associated with pseudo-locked region
227 * @minor: minor number of character device associated with this
228 * region
229 * @debugfs_dir: pointer to this region's directory in the debugfs
230 * filesystem
231 * @pm_reqs: Power management QoS requests related to this region
232 */
233struct pseudo_lock_region {
234 struct resctrl_schema *s;
235 struct rdt_domain *d;
236 u32 cbm;
237 wait_queue_head_t lock_thread_wq;
238 int thread_done;
239 int cpu;
240 unsigned int line_size;
241 unsigned int size;
242 void *kmem;
243 unsigned int minor;
244 struct dentry *debugfs_dir;
245 struct list_head pm_reqs;
246};
247
248/**
249 * struct rdtgroup - store rdtgroup's data in resctrl file system.
250 * @kn: kernfs node
251 * @rdtgroup_list: linked list for all rdtgroups
252 * @closid: closid for this rdtgroup
253 * @cpu_mask: CPUs assigned to this rdtgroup
254 * @flags: status bits
255 * @waitcount: how many cpus expect to find this
256 * group when they acquire rdtgroup_mutex
257 * @type: indicates type of this rdtgroup - either
258 * monitor only or ctrl_mon group
259 * @mon: mongroup related data
260 * @mode: mode of resource group
261 * @plr: pseudo-locked region
262 */
263struct rdtgroup {
264 struct kernfs_node *kn;
265 struct list_head rdtgroup_list;
266 u32 closid;
267 struct cpumask cpu_mask;
268 int flags;
269 atomic_t waitcount;
270 enum rdt_group_type type;
271 struct mongroup mon;
272 enum rdtgrp_mode mode;
273 struct pseudo_lock_region *plr;
274};
275
276/* rdtgroup.flags */
277#define RDT_DELETED 1
278
279/* rftype.flags */
280#define RFTYPE_FLAGS_CPUS_LIST 1
281
282/*
283 * Define the file type flags for base and info directories.
284 */
285#define RFTYPE_INFO BIT(0)
286#define RFTYPE_BASE BIT(1)
287#define RFTYPE_CTRL BIT(4)
288#define RFTYPE_MON BIT(5)
289#define RFTYPE_TOP BIT(6)
290#define RFTYPE_RES_CACHE BIT(8)
291#define RFTYPE_RES_MB BIT(9)
292#define RFTYPE_DEBUG BIT(10)
293#define RFTYPE_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
294#define RFTYPE_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
295#define RFTYPE_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
296#define RFTYPE_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
297#define RFTYPE_MON_BASE (RFTYPE_BASE | RFTYPE_MON)
298
299/* List of all resource groups */
300extern struct list_head rdt_all_groups;
301
302extern int max_name_width, max_data_width;
303
304int __init rdtgroup_init(void);
305void __exit rdtgroup_exit(void);
306
307/**
308 * struct rftype - describe each file in the resctrl file system
309 * @name: File name
310 * @mode: Access mode
311 * @kf_ops: File operations
312 * @flags: File specific RFTYPE_FLAGS_* flags
313 * @fflags: File specific RFTYPE_* flags
314 * @seq_show: Show content of the file
315 * @write: Write to the file
316 */
317struct rftype {
318 char *name;
319 umode_t mode;
320 const struct kernfs_ops *kf_ops;
321 unsigned long flags;
322 unsigned long fflags;
323
324 int (*seq_show)(struct kernfs_open_file *of,
325 struct seq_file *sf, void *v);
326 /*
327 * write() is the generic write callback which maps directly to
328 * kernfs write operation and overrides all other operations.
329 * Maximum write size is determined by ->max_write_len.
330 */
331 ssize_t (*write)(struct kernfs_open_file *of,
332 char *buf, size_t nbytes, loff_t off);
333};
334
335/**
336 * struct mbm_state - status for each MBM counter in each domain
337 * @prev_bw_bytes: Previous bytes value read for bandwidth calculation
338 * @prev_bw: The most recent bandwidth in MBps
339 */
340struct mbm_state {
341 u64 prev_bw_bytes;
342 u32 prev_bw;
343};
344
345/**
346 * struct arch_mbm_state - values used to compute resctrl_arch_rmid_read()s
347 * return value.
348 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
349 * @prev_msr: Value of IA32_QM_CTR last time it was read for the RMID used to
350 * find this struct.
351 */
352struct arch_mbm_state {
353 u64 chunks;
354 u64 prev_msr;
355};
356
357/**
358 * struct rdt_hw_domain - Arch private attributes of a set of CPUs that share
359 * a resource
360 * @d_resctrl: Properties exposed to the resctrl file system
361 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
362 * @arch_mbm_total: arch private state for MBM total bandwidth
363 * @arch_mbm_local: arch private state for MBM local bandwidth
364 *
365 * Members of this structure are accessed via helpers that provide abstraction.
366 */
367struct rdt_hw_domain {
368 struct rdt_domain d_resctrl;
369 u32 *ctrl_val;
370 struct arch_mbm_state *arch_mbm_total;
371 struct arch_mbm_state *arch_mbm_local;
372};
373
374static inline struct rdt_hw_domain *resctrl_to_arch_dom(struct rdt_domain *r)
375{
376 return container_of(r, struct rdt_hw_domain, d_resctrl);
377}
378
379/**
380 * struct msr_param - set a range of MSRs from a domain
381 * @res: The resource to use
382 * @low: Beginning index from base MSR
383 * @high: End index
384 */
385struct msr_param {
386 struct rdt_resource *res;
387 u32 low;
388 u32 high;
389};
390
391static inline bool is_llc_occupancy_enabled(void)
392{
393 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
394}
395
396static inline bool is_mbm_total_enabled(void)
397{
398 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
399}
400
401static inline bool is_mbm_local_enabled(void)
402{
403 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
404}
405
406static inline bool is_mbm_enabled(void)
407{
408 return (is_mbm_total_enabled() || is_mbm_local_enabled());
409}
410
411static inline bool is_mbm_event(int e)
412{
413 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
414 e <= QOS_L3_MBM_LOCAL_EVENT_ID);
415}
416
417struct rdt_parse_data {
418 struct rdtgroup *rdtgrp;
419 char *buf;
420};
421
422/**
423 * struct rdt_hw_resource - arch private attributes of a resctrl resource
424 * @r_resctrl: Attributes of the resource used directly by resctrl.
425 * @num_closid: Maximum number of closid this hardware can support,
426 * regardless of CDP. This is exposed via
427 * resctrl_arch_get_num_closid() to avoid confusion
428 * with struct resctrl_schema's property of the same name,
429 * which has been corrected for features like CDP.
430 * @msr_base: Base MSR address for CBMs
431 * @msr_update: Function pointer to update QOS MSRs
432 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
433 * @mbm_width: Monitor width, to detect and correct for overflow.
434 * @mbm_cfg_mask: Bandwidth sources that can be tracked when Bandwidth
435 * Monitoring Event Configuration (BMEC) is supported.
436 * @cdp_enabled: CDP state of this resource
437 *
438 * Members of this structure are either private to the architecture
439 * e.g. mbm_width, or accessed via helpers that provide abstraction. e.g.
440 * msr_update and msr_base.
441 */
442struct rdt_hw_resource {
443 struct rdt_resource r_resctrl;
444 u32 num_closid;
445 unsigned int msr_base;
446 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
447 struct rdt_resource *r);
448 unsigned int mon_scale;
449 unsigned int mbm_width;
450 unsigned int mbm_cfg_mask;
451 bool cdp_enabled;
452};
453
454static inline struct rdt_hw_resource *resctrl_to_arch_res(struct rdt_resource *r)
455{
456 return container_of(r, struct rdt_hw_resource, r_resctrl);
457}
458
459int parse_cbm(struct rdt_parse_data *data, struct resctrl_schema *s,
460 struct rdt_domain *d);
461int parse_bw(struct rdt_parse_data *data, struct resctrl_schema *s,
462 struct rdt_domain *d);
463
464extern struct mutex rdtgroup_mutex;
465
466extern struct rdt_hw_resource rdt_resources_all[];
467extern struct rdtgroup rdtgroup_default;
468extern struct dentry *debugfs_resctrl;
469
470enum resctrl_res_level {
471 RDT_RESOURCE_L3,
472 RDT_RESOURCE_L2,
473 RDT_RESOURCE_MBA,
474 RDT_RESOURCE_SMBA,
475
476 /* Must be the last */
477 RDT_NUM_RESOURCES,
478};
479
480static inline struct rdt_resource *resctrl_inc(struct rdt_resource *res)
481{
482 struct rdt_hw_resource *hw_res = resctrl_to_arch_res(r: res);
483
484 hw_res++;
485 return &hw_res->r_resctrl;
486}
487
488static inline bool resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)
489{
490 return rdt_resources_all[l].cdp_enabled;
491}
492
493int resctrl_arch_set_cdp_enabled(enum resctrl_res_level l, bool enable);
494
495/*
496 * To return the common struct rdt_resource, which is contained in struct
497 * rdt_hw_resource, walk the resctrl member of struct rdt_hw_resource.
498 */
499#define for_each_rdt_resource(r) \
500 for (r = &rdt_resources_all[0].r_resctrl; \
501 r <= &rdt_resources_all[RDT_NUM_RESOURCES - 1].r_resctrl; \
502 r = resctrl_inc(r))
503
504#define for_each_capable_rdt_resource(r) \
505 for_each_rdt_resource(r) \
506 if (r->alloc_capable || r->mon_capable)
507
508#define for_each_alloc_capable_rdt_resource(r) \
509 for_each_rdt_resource(r) \
510 if (r->alloc_capable)
511
512#define for_each_mon_capable_rdt_resource(r) \
513 for_each_rdt_resource(r) \
514 if (r->mon_capable)
515
516/* CPUID.(EAX=10H, ECX=ResID=1).EAX */
517union cpuid_0x10_1_eax {
518 struct {
519 unsigned int cbm_len:5;
520 } split;
521 unsigned int full;
522};
523
524/* CPUID.(EAX=10H, ECX=ResID=3).EAX */
525union cpuid_0x10_3_eax {
526 struct {
527 unsigned int max_delay:12;
528 } split;
529 unsigned int full;
530};
531
532/* CPUID.(EAX=10H, ECX=ResID).ECX */
533union cpuid_0x10_x_ecx {
534 struct {
535 unsigned int reserved:3;
536 unsigned int noncont:1;
537 } split;
538 unsigned int full;
539};
540
541/* CPUID.(EAX=10H, ECX=ResID).EDX */
542union cpuid_0x10_x_edx {
543 struct {
544 unsigned int cos_max:16;
545 } split;
546 unsigned int full;
547};
548
549void rdt_last_cmd_clear(void);
550void rdt_last_cmd_puts(const char *s);
551__printf(1, 2)
552void rdt_last_cmd_printf(const char *fmt, ...);
553
554void rdt_ctrl_update(void *arg);
555struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
556void rdtgroup_kn_unlock(struct kernfs_node *kn);
557int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
558int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
559 umode_t mask);
560struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
561 struct list_head **pos);
562ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
563 char *buf, size_t nbytes, loff_t off);
564int rdtgroup_schemata_show(struct kernfs_open_file *of,
565 struct seq_file *s, void *v);
566bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_domain *d,
567 unsigned long cbm, int closid, bool exclusive);
568unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
569 unsigned long cbm);
570enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
571int rdtgroup_tasks_assigned(struct rdtgroup *r);
572int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
573int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
574bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
575bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
576int rdt_pseudo_lock_init(void);
577void rdt_pseudo_lock_release(void);
578int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
579void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
580struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
581int closids_supported(void);
582void closid_free(int closid);
583int alloc_rmid(u32 closid);
584void free_rmid(u32 closid, u32 rmid);
585int rdt_get_mon_l3_config(struct rdt_resource *r);
586void __exit rdt_put_mon_l3_config(void);
587bool __init rdt_cpu_has(int flag);
588void mon_event_count(void *info);
589int rdtgroup_mondata_show(struct seq_file *m, void *arg);
590void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
591 struct rdt_domain *d, struct rdtgroup *rdtgrp,
592 int evtid, int first);
593void mbm_setup_overflow_handler(struct rdt_domain *dom,
594 unsigned long delay_ms,
595 int exclude_cpu);
596void mbm_handle_overflow(struct work_struct *work);
597void __init intel_rdt_mbm_apply_quirk(void);
598bool is_mba_sc(struct rdt_resource *r);
599void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms,
600 int exclude_cpu);
601void cqm_handle_limbo(struct work_struct *work);
602bool has_busy_rmid(struct rdt_domain *d);
603void __check_limbo(struct rdt_domain *d, bool force_free);
604void rdt_domain_reconfigure_cdp(struct rdt_resource *r);
605void __init thread_throttle_mode_init(void);
606void __init mbm_config_rftype_init(const char *config);
607void rdt_staged_configs_clear(void);
608bool closid_allocated(unsigned int closid);
609int resctrl_find_cleanest_closid(void);
610
611#endif /* _ASM_X86_RESCTRL_INTERNAL_H */
612

source code of linux/arch/x86/kernel/cpu/resctrl/internal.h