Loading...
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
45struct rdt_fs_context {
46 struct kernfs_fs_context kfc;
47 bool enable_cdpl2;
48 bool enable_cdpl3;
49 bool enable_mba_mbps;
50};
51
52static 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
59DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
60
61/**
62 * struct mon_evt - Entry in the event list of a resource
63 * @evtid: event id
64 * @name: name of the event
65 */
66struct mon_evt {
67 u32 evtid;
68 char *name;
69 struct list_head list;
70};
71
72/**
73 * struct mon_data_bits - Monitoring details for each event file
74 * @rid: Resource id associated with the event file.
75 * @evtid: Event id associated with the event file
76 * @domid: The domain to which the event file belongs
77 */
78union mon_data_bits {
79 void *priv;
80 struct {
81 unsigned int rid : 10;
82 unsigned int evtid : 8;
83 unsigned int domid : 14;
84 } u;
85};
86
87struct rmid_read {
88 struct rdtgroup *rgrp;
89 struct rdt_domain *d;
90 int evtid;
91 bool first;
92 u64 val;
93};
94
95extern unsigned int resctrl_cqm_threshold;
96extern bool rdt_alloc_capable;
97extern bool rdt_mon_capable;
98extern unsigned int rdt_mon_features;
99
100enum rdt_group_type {
101 RDTCTRL_GROUP = 0,
102 RDTMON_GROUP,
103 RDT_NUM_GROUP,
104};
105
106/**
107 * enum rdtgrp_mode - Mode of a RDT resource group
108 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
109 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
110 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
111 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
112 * allowed AND the allocations are Cache Pseudo-Locked
113 *
114 * The mode of a resource group enables control over the allowed overlap
115 * between allocations associated with different resource groups (classes
116 * of service). User is able to modify the mode of a resource group by
117 * writing to the "mode" resctrl file associated with the resource group.
118 *
119 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
120 * writing the appropriate text to the "mode" file. A resource group enters
121 * "pseudo-locked" mode after the schemata is written while the resource
122 * group is in "pseudo-locksetup" mode.
123 */
124enum rdtgrp_mode {
125 RDT_MODE_SHAREABLE = 0,
126 RDT_MODE_EXCLUSIVE,
127 RDT_MODE_PSEUDO_LOCKSETUP,
128 RDT_MODE_PSEUDO_LOCKED,
129
130 /* Must be last */
131 RDT_NUM_MODES,
132};
133
134/**
135 * struct mongroup - store mon group's data in resctrl fs.
136 * @mon_data_kn kernlfs node for the mon_data directory
137 * @parent: parent rdtgrp
138 * @crdtgrp_list: child rdtgroup node list
139 * @rmid: rmid for this rdtgroup
140 */
141struct mongroup {
142 struct kernfs_node *mon_data_kn;
143 struct rdtgroup *parent;
144 struct list_head crdtgrp_list;
145 u32 rmid;
146};
147
148/**
149 * struct pseudo_lock_region - pseudo-lock region information
150 * @r: RDT resource to which this pseudo-locked region
151 * belongs
152 * @d: RDT domain to which this pseudo-locked region
153 * belongs
154 * @cbm: bitmask of the pseudo-locked region
155 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread
156 * completion
157 * @thread_done: variable used by waitqueue to test if pseudo-locking
158 * thread completed
159 * @cpu: core associated with the cache on which the setup code
160 * will be run
161 * @line_size: size of the cache lines
162 * @size: size of pseudo-locked region in bytes
163 * @kmem: the kernel memory associated with pseudo-locked region
164 * @minor: minor number of character device associated with this
165 * region
166 * @debugfs_dir: pointer to this region's directory in the debugfs
167 * filesystem
168 * @pm_reqs: Power management QoS requests related to this region
169 */
170struct pseudo_lock_region {
171 struct rdt_resource *r;
172 struct rdt_domain *d;
173 u32 cbm;
174 wait_queue_head_t lock_thread_wq;
175 int thread_done;
176 int cpu;
177 unsigned int line_size;
178 unsigned int size;
179 void *kmem;
180 unsigned int minor;
181 struct dentry *debugfs_dir;
182 struct list_head pm_reqs;
183};
184
185/**
186 * struct rdtgroup - store rdtgroup's data in resctrl file system.
187 * @kn: kernfs node
188 * @rdtgroup_list: linked list for all rdtgroups
189 * @closid: closid for this rdtgroup
190 * @cpu_mask: CPUs assigned to this rdtgroup
191 * @flags: status bits
192 * @waitcount: how many cpus expect to find this
193 * group when they acquire rdtgroup_mutex
194 * @type: indicates type of this rdtgroup - either
195 * monitor only or ctrl_mon group
196 * @mon: mongroup related data
197 * @mode: mode of resource group
198 * @plr: pseudo-locked region
199 */
200struct rdtgroup {
201 struct kernfs_node *kn;
202 struct list_head rdtgroup_list;
203 u32 closid;
204 struct cpumask cpu_mask;
205 int flags;
206 atomic_t waitcount;
207 enum rdt_group_type type;
208 struct mongroup mon;
209 enum rdtgrp_mode mode;
210 struct pseudo_lock_region *plr;
211};
212
213/* rdtgroup.flags */
214#define RDT_DELETED 1
215
216/* rftype.flags */
217#define RFTYPE_FLAGS_CPUS_LIST 1
218
219/*
220 * Define the file type flags for base and info directories.
221 */
222#define RFTYPE_INFO BIT(0)
223#define RFTYPE_BASE BIT(1)
224#define RF_CTRLSHIFT 4
225#define RF_MONSHIFT 5
226#define RF_TOPSHIFT 6
227#define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
228#define RFTYPE_MON BIT(RF_MONSHIFT)
229#define RFTYPE_TOP BIT(RF_TOPSHIFT)
230#define RFTYPE_RES_CACHE BIT(8)
231#define RFTYPE_RES_MB BIT(9)
232#define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
233#define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
234#define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
235#define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
236
237/* List of all resource groups */
238extern struct list_head rdt_all_groups;
239
240extern int max_name_width, max_data_width;
241
242int __init rdtgroup_init(void);
243void __exit rdtgroup_exit(void);
244
245/**
246 * struct rftype - describe each file in the resctrl file system
247 * @name: File name
248 * @mode: Access mode
249 * @kf_ops: File operations
250 * @flags: File specific RFTYPE_FLAGS_* flags
251 * @fflags: File specific RF_* or RFTYPE_* flags
252 * @seq_show: Show content of the file
253 * @write: Write to the file
254 */
255struct rftype {
256 char *name;
257 umode_t mode;
258 struct kernfs_ops *kf_ops;
259 unsigned long flags;
260 unsigned long fflags;
261
262 int (*seq_show)(struct kernfs_open_file *of,
263 struct seq_file *sf, void *v);
264 /*
265 * write() is the generic write callback which maps directly to
266 * kernfs write operation and overrides all other operations.
267 * Maximum write size is determined by ->max_write_len.
268 */
269 ssize_t (*write)(struct kernfs_open_file *of,
270 char *buf, size_t nbytes, loff_t off);
271};
272
273/**
274 * struct mbm_state - status for each MBM counter in each domain
275 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
276 * @prev_msr Value of IA32_QM_CTR for this RMID last time we read it
277 * @chunks_bw Total local data moved. Used for bandwidth calculation
278 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting
279 * @prev_bw The most recent bandwidth in MBps
280 * @delta_bw Difference between the current and previous bandwidth
281 * @delta_comp Indicates whether to compute the delta_bw
282 */
283struct mbm_state {
284 u64 chunks;
285 u64 prev_msr;
286 u64 chunks_bw;
287 u64 prev_bw_msr;
288 u32 prev_bw;
289 u32 delta_bw;
290 bool delta_comp;
291};
292
293/**
294 * struct rdt_domain - group of cpus sharing an RDT resource
295 * @list: all instances of this resource
296 * @id: unique id for this instance
297 * @cpu_mask: which cpus share this resource
298 * @rmid_busy_llc:
299 * bitmap of which limbo RMIDs are above threshold
300 * @mbm_total: saved state for MBM total bandwidth
301 * @mbm_local: saved state for MBM local bandwidth
302 * @mbm_over: worker to periodically read MBM h/w counters
303 * @cqm_limbo: worker to periodically read CQM h/w counters
304 * @mbm_work_cpu:
305 * worker cpu for MBM h/w counters
306 * @cqm_work_cpu:
307 * worker cpu for CQM h/w counters
308 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
309 * @mbps_val: When mba_sc is enabled, this holds the bandwidth in MBps
310 * @new_ctrl: new ctrl value to be loaded
311 * @have_new_ctrl: did user provide new_ctrl for this domain
312 * @plr: pseudo-locked region (if any) associated with domain
313 */
314struct rdt_domain {
315 struct list_head list;
316 int id;
317 struct cpumask cpu_mask;
318 unsigned long *rmid_busy_llc;
319 struct mbm_state *mbm_total;
320 struct mbm_state *mbm_local;
321 struct delayed_work mbm_over;
322 struct delayed_work cqm_limbo;
323 int mbm_work_cpu;
324 int cqm_work_cpu;
325 u32 *ctrl_val;
326 u32 *mbps_val;
327 u32 new_ctrl;
328 bool have_new_ctrl;
329 struct pseudo_lock_region *plr;
330};
331
332/**
333 * struct msr_param - set a range of MSRs from a domain
334 * @res: The resource to use
335 * @low: Beginning index from base MSR
336 * @high: End index
337 */
338struct msr_param {
339 struct rdt_resource *res;
340 int low;
341 int high;
342};
343
344/**
345 * struct rdt_cache - Cache allocation related data
346 * @cbm_len: Length of the cache bit mask
347 * @min_cbm_bits: Minimum number of consecutive bits to be set
348 * @cbm_idx_mult: Multiplier of CBM index
349 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by:
350 * closid * cbm_idx_multi + cbm_idx_offset
351 * in a cache bit mask
352 * @shareable_bits: Bitmask of shareable resource with other
353 * executing entities
354 */
355struct rdt_cache {
356 unsigned int cbm_len;
357 unsigned int min_cbm_bits;
358 unsigned int cbm_idx_mult;
359 unsigned int cbm_idx_offset;
360 unsigned int shareable_bits;
361};
362
363/**
364 * struct rdt_membw - Memory bandwidth allocation related data
365 * @max_delay: Max throttle delay. Delay is the hardware
366 * representation for memory bandwidth.
367 * @min_bw: Minimum memory bandwidth percentage user can request
368 * @bw_gran: Granularity at which the memory bandwidth is allocated
369 * @delay_linear: True if memory B/W delay is in linear scale
370 * @mba_sc: True if MBA software controller(mba_sc) is enabled
371 * @mb_map: Mapping of memory B/W percentage to memory B/W delay
372 */
373struct rdt_membw {
374 u32 max_delay;
375 u32 min_bw;
376 u32 bw_gran;
377 u32 delay_linear;
378 bool mba_sc;
379 u32 *mb_map;
380};
381
382static inline bool is_llc_occupancy_enabled(void)
383{
384 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
385}
386
387static inline bool is_mbm_total_enabled(void)
388{
389 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
390}
391
392static inline bool is_mbm_local_enabled(void)
393{
394 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
395}
396
397static inline bool is_mbm_enabled(void)
398{
399 return (is_mbm_total_enabled() || is_mbm_local_enabled());
400}
401
402static inline bool is_mbm_event(int e)
403{
404 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
405 e <= QOS_L3_MBM_LOCAL_EVENT_ID);
406}
407
408struct rdt_parse_data {
409 struct rdtgroup *rdtgrp;
410 char *buf;
411};
412
413/**
414 * struct rdt_resource - attributes of an RDT resource
415 * @rid: The index of the resource
416 * @alloc_enabled: Is allocation enabled on this machine
417 * @mon_enabled: Is monitoring enabled for this feature
418 * @alloc_capable: Is allocation available on this machine
419 * @mon_capable: Is monitor feature available on this machine
420 * @name: Name to use in "schemata" file
421 * @num_closid: Number of CLOSIDs available
422 * @cache_level: Which cache level defines scope of this resource
423 * @default_ctrl: Specifies default cache cbm or memory B/W percent.
424 * @msr_base: Base MSR address for CBMs
425 * @msr_update: Function pointer to update QOS MSRs
426 * @data_width: Character width of data when displaying
427 * @domains: All domains for this resource
428 * @cache: Cache allocation related data
429 * @format_str: Per resource format string to show domain value
430 * @parse_ctrlval: Per resource function pointer to parse control values
431 * @cbm_validate Cache bitmask validate function
432 * @evt_list: List of monitoring events
433 * @num_rmid: Number of RMIDs available
434 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
435 * @fflags: flags to choose base and info files
436 */
437struct rdt_resource {
438 int rid;
439 bool alloc_enabled;
440 bool mon_enabled;
441 bool alloc_capable;
442 bool mon_capable;
443 char *name;
444 int num_closid;
445 int cache_level;
446 u32 default_ctrl;
447 unsigned int msr_base;
448 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
449 struct rdt_resource *r);
450 int data_width;
451 struct list_head domains;
452 struct rdt_cache cache;
453 struct rdt_membw membw;
454 const char *format_str;
455 int (*parse_ctrlval)(struct rdt_parse_data *data,
456 struct rdt_resource *r,
457 struct rdt_domain *d);
458 bool (*cbm_validate)(char *buf, u32 *data, struct rdt_resource *r);
459 struct list_head evt_list;
460 int num_rmid;
461 unsigned int mon_scale;
462 unsigned long fflags;
463};
464
465int parse_cbm(struct rdt_parse_data *data, struct rdt_resource *r,
466 struct rdt_domain *d);
467int parse_bw_intel(struct rdt_parse_data *data, struct rdt_resource *r,
468 struct rdt_domain *d);
469int parse_bw_amd(struct rdt_parse_data *data, struct rdt_resource *r,
470 struct rdt_domain *d);
471
472extern struct mutex rdtgroup_mutex;
473
474extern struct rdt_resource rdt_resources_all[];
475extern struct rdtgroup rdtgroup_default;
476DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
477
478extern struct dentry *debugfs_resctrl;
479
480enum {
481 RDT_RESOURCE_L3,
482 RDT_RESOURCE_L3DATA,
483 RDT_RESOURCE_L3CODE,
484 RDT_RESOURCE_L2,
485 RDT_RESOURCE_L2DATA,
486 RDT_RESOURCE_L2CODE,
487 RDT_RESOURCE_MBA,
488
489 /* Must be the last */
490 RDT_NUM_RESOURCES,
491};
492
493#define for_each_rdt_resource(r) \
494 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
495 r++)
496
497#define for_each_capable_rdt_resource(r) \
498 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
499 r++) \
500 if (r->alloc_capable || r->mon_capable)
501
502#define for_each_alloc_capable_rdt_resource(r) \
503 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
504 r++) \
505 if (r->alloc_capable)
506
507#define for_each_mon_capable_rdt_resource(r) \
508 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
509 r++) \
510 if (r->mon_capable)
511
512#define for_each_alloc_enabled_rdt_resource(r) \
513 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
514 r++) \
515 if (r->alloc_enabled)
516
517#define for_each_mon_enabled_rdt_resource(r) \
518 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
519 r++) \
520 if (r->mon_enabled)
521
522/* CPUID.(EAX=10H, ECX=ResID=1).EAX */
523union cpuid_0x10_1_eax {
524 struct {
525 unsigned int cbm_len:5;
526 } split;
527 unsigned int full;
528};
529
530/* CPUID.(EAX=10H, ECX=ResID=3).EAX */
531union cpuid_0x10_3_eax {
532 struct {
533 unsigned int max_delay:12;
534 } split;
535 unsigned int full;
536};
537
538/* CPUID.(EAX=10H, ECX=ResID).EDX */
539union cpuid_0x10_x_edx {
540 struct {
541 unsigned int cos_max:16;
542 } split;
543 unsigned int full;
544};
545
546void rdt_last_cmd_clear(void);
547void rdt_last_cmd_puts(const char *s);
548void rdt_last_cmd_printf(const char *fmt, ...);
549
550void rdt_ctrl_update(void *arg);
551struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
552void rdtgroup_kn_unlock(struct kernfs_node *kn);
553int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
554int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
555 umode_t mask);
556struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
557 struct list_head **pos);
558ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
559 char *buf, size_t nbytes, loff_t off);
560int rdtgroup_schemata_show(struct kernfs_open_file *of,
561 struct seq_file *s, void *v);
562bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
563 unsigned long cbm, int closid, bool exclusive);
564unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
565 unsigned long cbm);
566enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
567int rdtgroup_tasks_assigned(struct rdtgroup *r);
568int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
569int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
570bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
571bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
572int rdt_pseudo_lock_init(void);
573void rdt_pseudo_lock_release(void);
574int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
575void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
576struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
577int update_domains(struct rdt_resource *r, int closid);
578int closids_supported(void);
579void closid_free(int closid);
580int alloc_rmid(void);
581void free_rmid(u32 rmid);
582int rdt_get_mon_l3_config(struct rdt_resource *r);
583void mon_event_count(void *info);
584int rdtgroup_mondata_show(struct seq_file *m, void *arg);
585void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
586 unsigned int dom_id);
587void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
588 struct rdt_domain *d);
589void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
590 struct rdtgroup *rdtgrp, int evtid, int first);
591void mbm_setup_overflow_handler(struct rdt_domain *dom,
592 unsigned long delay_ms);
593void mbm_handle_overflow(struct work_struct *work);
594bool is_mba_sc(struct rdt_resource *r);
595void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm);
596u32 delay_bw_map(unsigned long bw, struct rdt_resource *r);
597void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
598void cqm_handle_limbo(struct work_struct *work);
599bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
600void __check_limbo(struct rdt_domain *d, bool force_free);
601bool cbm_validate_intel(char *buf, u32 *data, struct rdt_resource *r);
602bool cbm_validate_amd(char *buf, u32 *data, struct rdt_resource *r);
603
604#endif /* _ASM_X86_RESCTRL_INTERNAL_H */
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_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
40#define MBM_CNTR_WIDTH_OFFSET_AMD 20
41
42#define RMID_VAL_ERROR BIT_ULL(63)
43#define RMID_VAL_UNAVAIL BIT_ULL(62)
44/*
45 * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for
46 * data to be returned. The counter width is discovered from the hardware
47 * as an offset from MBM_CNTR_WIDTH_BASE.
48 */
49#define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
50
51
52struct rdt_fs_context {
53 struct kernfs_fs_context kfc;
54 bool enable_cdpl2;
55 bool enable_cdpl3;
56 bool enable_mba_mbps;
57};
58
59static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
60{
61 struct kernfs_fs_context *kfc = fc->fs_private;
62
63 return container_of(kfc, struct rdt_fs_context, kfc);
64}
65
66DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
67DECLARE_STATIC_KEY_FALSE(rdt_mon_enable_key);
68
69/**
70 * struct mon_evt - Entry in the event list of a resource
71 * @evtid: event id
72 * @name: name of the event
73 */
74struct mon_evt {
75 u32 evtid;
76 char *name;
77 struct list_head list;
78};
79
80/**
81 * struct mon_data_bits - Monitoring details for each event file
82 * @rid: Resource id associated with the event file.
83 * @evtid: Event id associated with the event file
84 * @domid: The domain to which the event file belongs
85 */
86union mon_data_bits {
87 void *priv;
88 struct {
89 unsigned int rid : 10;
90 unsigned int evtid : 8;
91 unsigned int domid : 14;
92 } u;
93};
94
95struct rmid_read {
96 struct rdtgroup *rgrp;
97 struct rdt_resource *r;
98 struct rdt_domain *d;
99 int evtid;
100 bool first;
101 u64 val;
102};
103
104extern unsigned int resctrl_cqm_threshold;
105extern bool rdt_alloc_capable;
106extern bool rdt_mon_capable;
107extern unsigned int rdt_mon_features;
108
109enum rdt_group_type {
110 RDTCTRL_GROUP = 0,
111 RDTMON_GROUP,
112 RDT_NUM_GROUP,
113};
114
115/**
116 * enum rdtgrp_mode - Mode of a RDT resource group
117 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
118 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
119 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
120 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
121 * allowed AND the allocations are Cache Pseudo-Locked
122 *
123 * The mode of a resource group enables control over the allowed overlap
124 * between allocations associated with different resource groups (classes
125 * of service). User is able to modify the mode of a resource group by
126 * writing to the "mode" resctrl file associated with the resource group.
127 *
128 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
129 * writing the appropriate text to the "mode" file. A resource group enters
130 * "pseudo-locked" mode after the schemata is written while the resource
131 * group is in "pseudo-locksetup" mode.
132 */
133enum rdtgrp_mode {
134 RDT_MODE_SHAREABLE = 0,
135 RDT_MODE_EXCLUSIVE,
136 RDT_MODE_PSEUDO_LOCKSETUP,
137 RDT_MODE_PSEUDO_LOCKED,
138
139 /* Must be last */
140 RDT_NUM_MODES,
141};
142
143/**
144 * struct mongroup - store mon group's data in resctrl fs.
145 * @mon_data_kn kernlfs node for the mon_data directory
146 * @parent: parent rdtgrp
147 * @crdtgrp_list: child rdtgroup node list
148 * @rmid: rmid for this rdtgroup
149 */
150struct mongroup {
151 struct kernfs_node *mon_data_kn;
152 struct rdtgroup *parent;
153 struct list_head crdtgrp_list;
154 u32 rmid;
155};
156
157/**
158 * struct pseudo_lock_region - pseudo-lock region information
159 * @r: RDT resource to which this pseudo-locked region
160 * belongs
161 * @d: RDT domain to which this pseudo-locked region
162 * belongs
163 * @cbm: bitmask of the pseudo-locked region
164 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread
165 * completion
166 * @thread_done: variable used by waitqueue to test if pseudo-locking
167 * thread completed
168 * @cpu: core associated with the cache on which the setup code
169 * will be run
170 * @line_size: size of the cache lines
171 * @size: size of pseudo-locked region in bytes
172 * @kmem: the kernel memory associated with pseudo-locked region
173 * @minor: minor number of character device associated with this
174 * region
175 * @debugfs_dir: pointer to this region's directory in the debugfs
176 * filesystem
177 * @pm_reqs: Power management QoS requests related to this region
178 */
179struct pseudo_lock_region {
180 struct rdt_resource *r;
181 struct rdt_domain *d;
182 u32 cbm;
183 wait_queue_head_t lock_thread_wq;
184 int thread_done;
185 int cpu;
186 unsigned int line_size;
187 unsigned int size;
188 void *kmem;
189 unsigned int minor;
190 struct dentry *debugfs_dir;
191 struct list_head pm_reqs;
192};
193
194/**
195 * struct rdtgroup - store rdtgroup's data in resctrl file system.
196 * @kn: kernfs node
197 * @rdtgroup_list: linked list for all rdtgroups
198 * @closid: closid for this rdtgroup
199 * @cpu_mask: CPUs assigned to this rdtgroup
200 * @flags: status bits
201 * @waitcount: how many cpus expect to find this
202 * group when they acquire rdtgroup_mutex
203 * @type: indicates type of this rdtgroup - either
204 * monitor only or ctrl_mon group
205 * @mon: mongroup related data
206 * @mode: mode of resource group
207 * @plr: pseudo-locked region
208 */
209struct rdtgroup {
210 struct kernfs_node *kn;
211 struct list_head rdtgroup_list;
212 u32 closid;
213 struct cpumask cpu_mask;
214 int flags;
215 atomic_t waitcount;
216 enum rdt_group_type type;
217 struct mongroup mon;
218 enum rdtgrp_mode mode;
219 struct pseudo_lock_region *plr;
220};
221
222/* rdtgroup.flags */
223#define RDT_DELETED 1
224
225/* rftype.flags */
226#define RFTYPE_FLAGS_CPUS_LIST 1
227
228/*
229 * Define the file type flags for base and info directories.
230 */
231#define RFTYPE_INFO BIT(0)
232#define RFTYPE_BASE BIT(1)
233#define RF_CTRLSHIFT 4
234#define RF_MONSHIFT 5
235#define RF_TOPSHIFT 6
236#define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
237#define RFTYPE_MON BIT(RF_MONSHIFT)
238#define RFTYPE_TOP BIT(RF_TOPSHIFT)
239#define RFTYPE_RES_CACHE BIT(8)
240#define RFTYPE_RES_MB BIT(9)
241#define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
242#define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
243#define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
244#define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
245
246/* List of all resource groups */
247extern struct list_head rdt_all_groups;
248
249extern int max_name_width, max_data_width;
250
251int __init rdtgroup_init(void);
252void __exit rdtgroup_exit(void);
253
254/**
255 * struct rftype - describe each file in the resctrl file system
256 * @name: File name
257 * @mode: Access mode
258 * @kf_ops: File operations
259 * @flags: File specific RFTYPE_FLAGS_* flags
260 * @fflags: File specific RF_* or RFTYPE_* flags
261 * @seq_show: Show content of the file
262 * @write: Write to the file
263 */
264struct rftype {
265 char *name;
266 umode_t mode;
267 struct kernfs_ops *kf_ops;
268 unsigned long flags;
269 unsigned long fflags;
270
271 int (*seq_show)(struct kernfs_open_file *of,
272 struct seq_file *sf, void *v);
273 /*
274 * write() is the generic write callback which maps directly to
275 * kernfs write operation and overrides all other operations.
276 * Maximum write size is determined by ->max_write_len.
277 */
278 ssize_t (*write)(struct kernfs_open_file *of,
279 char *buf, size_t nbytes, loff_t off);
280};
281
282/**
283 * struct mbm_state - status for each MBM counter in each domain
284 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
285 * @prev_msr Value of IA32_QM_CTR for this RMID last time we read it
286 * @chunks_bw Total local data moved. Used for bandwidth calculation
287 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting
288 * @prev_bw The most recent bandwidth in MBps
289 * @delta_bw Difference between the current and previous bandwidth
290 * @delta_comp Indicates whether to compute the delta_bw
291 */
292struct mbm_state {
293 u64 chunks;
294 u64 prev_msr;
295 u64 chunks_bw;
296 u64 prev_bw_msr;
297 u32 prev_bw;
298 u32 delta_bw;
299 bool delta_comp;
300};
301
302/**
303 * struct rdt_domain - group of cpus sharing an RDT resource
304 * @list: all instances of this resource
305 * @id: unique id for this instance
306 * @cpu_mask: which cpus share this resource
307 * @rmid_busy_llc:
308 * bitmap of which limbo RMIDs are above threshold
309 * @mbm_total: saved state for MBM total bandwidth
310 * @mbm_local: saved state for MBM local bandwidth
311 * @mbm_over: worker to periodically read MBM h/w counters
312 * @cqm_limbo: worker to periodically read CQM h/w counters
313 * @mbm_work_cpu:
314 * worker cpu for MBM h/w counters
315 * @cqm_work_cpu:
316 * worker cpu for CQM h/w counters
317 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
318 * @mbps_val: When mba_sc is enabled, this holds the bandwidth in MBps
319 * @new_ctrl: new ctrl value to be loaded
320 * @have_new_ctrl: did user provide new_ctrl for this domain
321 * @plr: pseudo-locked region (if any) associated with domain
322 */
323struct rdt_domain {
324 struct list_head list;
325 int id;
326 struct cpumask cpu_mask;
327 unsigned long *rmid_busy_llc;
328 struct mbm_state *mbm_total;
329 struct mbm_state *mbm_local;
330 struct delayed_work mbm_over;
331 struct delayed_work cqm_limbo;
332 int mbm_work_cpu;
333 int cqm_work_cpu;
334 u32 *ctrl_val;
335 u32 *mbps_val;
336 u32 new_ctrl;
337 bool have_new_ctrl;
338 struct pseudo_lock_region *plr;
339};
340
341/**
342 * struct msr_param - set a range of MSRs from a domain
343 * @res: The resource to use
344 * @low: Beginning index from base MSR
345 * @high: End index
346 */
347struct msr_param {
348 struct rdt_resource *res;
349 int low;
350 int high;
351};
352
353/**
354 * struct rdt_cache - Cache allocation related data
355 * @cbm_len: Length of the cache bit mask
356 * @min_cbm_bits: Minimum number of consecutive bits to be set
357 * @cbm_idx_mult: Multiplier of CBM index
358 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by:
359 * closid * cbm_idx_multi + cbm_idx_offset
360 * in a cache bit mask
361 * @shareable_bits: Bitmask of shareable resource with other
362 * executing entities
363 */
364struct rdt_cache {
365 unsigned int cbm_len;
366 unsigned int min_cbm_bits;
367 unsigned int cbm_idx_mult;
368 unsigned int cbm_idx_offset;
369 unsigned int shareable_bits;
370};
371
372/**
373 * struct rdt_membw - Memory bandwidth allocation related data
374 * @max_delay: Max throttle delay. Delay is the hardware
375 * representation for memory bandwidth.
376 * @min_bw: Minimum memory bandwidth percentage user can request
377 * @bw_gran: Granularity at which the memory bandwidth is allocated
378 * @delay_linear: True if memory B/W delay is in linear scale
379 * @mba_sc: True if MBA software controller(mba_sc) is enabled
380 * @mb_map: Mapping of memory B/W percentage to memory B/W delay
381 */
382struct rdt_membw {
383 u32 max_delay;
384 u32 min_bw;
385 u32 bw_gran;
386 u32 delay_linear;
387 bool mba_sc;
388 u32 *mb_map;
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_resource - attributes of an RDT resource
424 * @rid: The index of the resource
425 * @alloc_enabled: Is allocation enabled on this machine
426 * @mon_enabled: Is monitoring enabled for this feature
427 * @alloc_capable: Is allocation available on this machine
428 * @mon_capable: Is monitor feature available on this machine
429 * @name: Name to use in "schemata" file
430 * @num_closid: Number of CLOSIDs available
431 * @cache_level: Which cache level defines scope of this resource
432 * @default_ctrl: Specifies default cache cbm or memory B/W percent.
433 * @msr_base: Base MSR address for CBMs
434 * @msr_update: Function pointer to update QOS MSRs
435 * @data_width: Character width of data when displaying
436 * @domains: All domains for this resource
437 * @cache: Cache allocation related data
438 * @format_str: Per resource format string to show domain value
439 * @parse_ctrlval: Per resource function pointer to parse control values
440 * @cbm_validate Cache bitmask validate function
441 * @evt_list: List of monitoring events
442 * @num_rmid: Number of RMIDs available
443 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
444 * @fflags: flags to choose base and info files
445 */
446struct rdt_resource {
447 int rid;
448 bool alloc_enabled;
449 bool mon_enabled;
450 bool alloc_capable;
451 bool mon_capable;
452 char *name;
453 int num_closid;
454 int cache_level;
455 u32 default_ctrl;
456 unsigned int msr_base;
457 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
458 struct rdt_resource *r);
459 int data_width;
460 struct list_head domains;
461 struct rdt_cache cache;
462 struct rdt_membw membw;
463 const char *format_str;
464 int (*parse_ctrlval)(struct rdt_parse_data *data,
465 struct rdt_resource *r,
466 struct rdt_domain *d);
467 bool (*cbm_validate)(char *buf, u32 *data, struct rdt_resource *r);
468 struct list_head evt_list;
469 int num_rmid;
470 unsigned int mon_scale;
471 unsigned int mbm_width;
472 unsigned long fflags;
473};
474
475int parse_cbm(struct rdt_parse_data *data, struct rdt_resource *r,
476 struct rdt_domain *d);
477int parse_bw_intel(struct rdt_parse_data *data, struct rdt_resource *r,
478 struct rdt_domain *d);
479int parse_bw_amd(struct rdt_parse_data *data, struct rdt_resource *r,
480 struct rdt_domain *d);
481
482extern struct mutex rdtgroup_mutex;
483
484extern struct rdt_resource rdt_resources_all[];
485extern struct rdtgroup rdtgroup_default;
486DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
487
488extern struct dentry *debugfs_resctrl;
489
490enum {
491 RDT_RESOURCE_L3,
492 RDT_RESOURCE_L3DATA,
493 RDT_RESOURCE_L3CODE,
494 RDT_RESOURCE_L2,
495 RDT_RESOURCE_L2DATA,
496 RDT_RESOURCE_L2CODE,
497 RDT_RESOURCE_MBA,
498
499 /* Must be the last */
500 RDT_NUM_RESOURCES,
501};
502
503#define for_each_rdt_resource(r) \
504 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
505 r++)
506
507#define for_each_capable_rdt_resource(r) \
508 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
509 r++) \
510 if (r->alloc_capable || r->mon_capable)
511
512#define for_each_alloc_capable_rdt_resource(r) \
513 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
514 r++) \
515 if (r->alloc_capable)
516
517#define for_each_mon_capable_rdt_resource(r) \
518 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
519 r++) \
520 if (r->mon_capable)
521
522#define for_each_alloc_enabled_rdt_resource(r) \
523 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
524 r++) \
525 if (r->alloc_enabled)
526
527#define for_each_mon_enabled_rdt_resource(r) \
528 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
529 r++) \
530 if (r->mon_enabled)
531
532/* CPUID.(EAX=10H, ECX=ResID=1).EAX */
533union cpuid_0x10_1_eax {
534 struct {
535 unsigned int cbm_len:5;
536 } split;
537 unsigned int full;
538};
539
540/* CPUID.(EAX=10H, ECX=ResID=3).EAX */
541union cpuid_0x10_3_eax {
542 struct {
543 unsigned int max_delay:12;
544 } split;
545 unsigned int full;
546};
547
548/* CPUID.(EAX=10H, ECX=ResID).EDX */
549union cpuid_0x10_x_edx {
550 struct {
551 unsigned int cos_max:16;
552 } split;
553 unsigned int full;
554};
555
556void rdt_last_cmd_clear(void);
557void rdt_last_cmd_puts(const char *s);
558void rdt_last_cmd_printf(const char *fmt, ...);
559
560void rdt_ctrl_update(void *arg);
561struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
562void rdtgroup_kn_unlock(struct kernfs_node *kn);
563int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
564int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
565 umode_t mask);
566struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
567 struct list_head **pos);
568ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
569 char *buf, size_t nbytes, loff_t off);
570int rdtgroup_schemata_show(struct kernfs_open_file *of,
571 struct seq_file *s, void *v);
572bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
573 unsigned long cbm, int closid, bool exclusive);
574unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
575 unsigned long cbm);
576enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
577int rdtgroup_tasks_assigned(struct rdtgroup *r);
578int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
579int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
580bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
581bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
582int rdt_pseudo_lock_init(void);
583void rdt_pseudo_lock_release(void);
584int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
585void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
586struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
587int update_domains(struct rdt_resource *r, int closid);
588int closids_supported(void);
589void closid_free(int closid);
590int alloc_rmid(void);
591void free_rmid(u32 rmid);
592int rdt_get_mon_l3_config(struct rdt_resource *r);
593void mon_event_count(void *info);
594int rdtgroup_mondata_show(struct seq_file *m, void *arg);
595void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
596 unsigned int dom_id);
597void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
598 struct rdt_domain *d);
599void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
600 struct rdt_domain *d, struct rdtgroup *rdtgrp,
601 int evtid, int first);
602void mbm_setup_overflow_handler(struct rdt_domain *dom,
603 unsigned long delay_ms);
604void mbm_handle_overflow(struct work_struct *work);
605bool is_mba_sc(struct rdt_resource *r);
606void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm);
607u32 delay_bw_map(unsigned long bw, struct rdt_resource *r);
608void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
609void cqm_handle_limbo(struct work_struct *work);
610bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
611void __check_limbo(struct rdt_domain *d, bool force_free);
612bool cbm_validate_intel(char *buf, u32 *data, struct rdt_resource *r);
613bool cbm_validate_amd(char *buf, u32 *data, struct rdt_resource *r);
614void rdt_domain_reconfigure_cdp(struct rdt_resource *r);
615
616#endif /* _ASM_X86_RESCTRL_INTERNAL_H */