x86/intel_rdt/mba_sc: Prepare for feedback loop
[linux-2.6-block.git] / arch / x86 / kernel / cpu / intel_rdt.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
113c6097
FY
2#ifndef _ASM_X86_INTEL_RDT_H
3#define _ASM_X86_INTEL_RDT_H
4
5b825c3a 5#include <linux/sched.h>
8ff42c02 6#include <linux/kernfs.h>
5ff193fb
FY
7#include <linux/jump_label.h>
8
9#define IA32_L3_QOS_CFG 0xc81
99adde9b 10#define IA32_L2_QOS_CFG 0xc82
113c6097 11#define IA32_L3_CBM_BASE 0xc90
c1c7c3f9 12#define IA32_L2_CBM_BASE 0xd10
05b93417 13#define IA32_MBA_THRTL_BASE 0xd50
113c6097 14
5ff193fb
FY
15#define L3_QOS_CDP_ENABLE 0x01ULL
16
99adde9b
FY
17#define L2_QOS_CDP_ENABLE 0x01ULL
18
6a445edc
VS
19/*
20 * Event IDs are used to program IA32_QM_EVTSEL before reading event
21 * counter from IA32_QM_CTR
22 */
23#define QOS_L3_OCCUP_EVENT_ID 0x01
24#define QOS_L3_MBM_TOTAL_EVENT_ID 0x02
25#define QOS_L3_MBM_LOCAL_EVENT_ID 0x03
9f52425b 26
24247aee
VS
27#define CQM_LIMBOCHECK_INTERVAL 1000
28
9f52425b 29#define MBM_CNTR_WIDTH 24
e3302683 30#define MBM_OVERFLOW_INTERVAL 1000
9f52425b 31
edf6fa1c
VS
32#define RMID_VAL_ERROR BIT_ULL(63)
33#define RMID_VAL_UNAVAIL BIT_ULL(62)
6a445edc 34
e3302683
VS
35DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
36
6a445edc
VS
37/**
38 * struct mon_evt - Entry in the event list of a resource
39 * @evtid: event id
40 * @name: name of the event
41 */
42struct mon_evt {
43 u32 evtid;
44 char *name;
45 struct list_head list;
46};
47
d89b7379
VS
48/**
49 * struct mon_data_bits - Monitoring details for each event file
50 * @rid: Resource id associated with the event file.
51 * @evtid: Event id associated with the event file
52 * @domid: The domain to which the event file belongs
53 */
54union mon_data_bits {
55 void *priv;
56 struct {
57 unsigned int rid : 10;
58 unsigned int evtid : 8;
59 unsigned int domid : 14;
60 } u;
61};
62
63struct rmid_read {
64 struct rdtgroup *rgrp;
9f52425b 65 struct rdt_domain *d;
d89b7379 66 int evtid;
a4de1dfd 67 bool first;
d89b7379
VS
68 u64 val;
69};
70
6a445edc
VS
71extern unsigned int intel_cqm_threshold;
72extern bool rdt_alloc_capable;
73extern bool rdt_mon_capable;
74extern unsigned int rdt_mon_features;
c7d9aac6
VS
75
76enum rdt_group_type {
77 RDTCTRL_GROUP = 0,
78 RDTMON_GROUP,
79 RDT_NUM_GROUP,
80};
81
82/**
83 * struct mongroup - store mon group's data in resctrl fs.
d89b7379 84 * @mon_data_kn kernlfs node for the mon_data directory
c7d9aac6
VS
85 * @parent: parent rdtgrp
86 * @crdtgrp_list: child rdtgroup node list
87 * @rmid: rmid for this rdtgroup
88 */
89struct mongroup {
d89b7379 90 struct kernfs_node *mon_data_kn;
c7d9aac6
VS
91 struct rdtgroup *parent;
92 struct list_head crdtgrp_list;
93 u32 rmid;
94};
95
5ff193fb
FY
96/**
97 * struct rdtgroup - store rdtgroup's data in resctrl file system.
98 * @kn: kernfs node
99 * @rdtgroup_list: linked list for all rdtgroups
100 * @closid: closid for this rdtgroup
12e0110c 101 * @cpu_mask: CPUs assigned to this rdtgroup
60cf5e10
FY
102 * @flags: status bits
103 * @waitcount: how many cpus expect to find this
12e0110c 104 * group when they acquire rdtgroup_mutex
c7d9aac6
VS
105 * @type: indicates type of this rdtgroup - either
106 * monitor only or ctrl_mon group
107 * @mon: mongroup related data
5ff193fb
FY
108 */
109struct rdtgroup {
110 struct kernfs_node *kn;
111 struct list_head rdtgroup_list;
0734ded1 112 u32 closid;
12e0110c 113 struct cpumask cpu_mask;
60cf5e10
FY
114 int flags;
115 atomic_t waitcount;
c7d9aac6
VS
116 enum rdt_group_type type;
117 struct mongroup mon;
5ff193fb
FY
118};
119
60cf5e10
FY
120/* rdtgroup.flags */
121#define RDT_DELETED 1
122
4ffa3c97
JO
123/* rftype.flags */
124#define RFTYPE_FLAGS_CPUS_LIST 1
125
5dc1d5c6
T
126/*
127 * Define the file type flags for base and info directories.
128 */
129#define RFTYPE_INFO BIT(0)
130#define RFTYPE_BASE BIT(1)
131#define RF_CTRLSHIFT 4
d4ab3320 132#define RF_MONSHIFT 5
9b3a7fd0 133#define RF_TOPSHIFT 6
5dc1d5c6 134#define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
d4ab3320 135#define RFTYPE_MON BIT(RF_MONSHIFT)
9b3a7fd0 136#define RFTYPE_TOP BIT(RF_TOPSHIFT)
5dc1d5c6
T
137#define RFTYPE_RES_CACHE BIT(8)
138#define RFTYPE_RES_MB BIT(9)
139#define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
d4ab3320 140#define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
9b3a7fd0 141#define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
5dc1d5c6
T
142#define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
143
5ff193fb
FY
144/* List of all resource groups */
145extern struct list_head rdt_all_groups;
146
de016df8
VS
147extern int max_name_width, max_data_width;
148
5ff193fb
FY
149int __init rdtgroup_init(void);
150
4e978d06
FY
151/**
152 * struct rftype - describe each file in the resctrl file system
17f8ba1d
TG
153 * @name: File name
154 * @mode: Access mode
155 * @kf_ops: File operations
4ffa3c97 156 * @flags: File specific RFTYPE_FLAGS_* flags
5dc1d5c6 157 * @fflags: File specific RF_* or RFTYPE_* flags
17f8ba1d
TG
158 * @seq_show: Show content of the file
159 * @write: Write to the file
4e978d06
FY
160 */
161struct rftype {
162 char *name;
163 umode_t mode;
164 struct kernfs_ops *kf_ops;
4ffa3c97 165 unsigned long flags;
5dc1d5c6 166 unsigned long fflags;
4e978d06
FY
167
168 int (*seq_show)(struct kernfs_open_file *of,
169 struct seq_file *sf, void *v);
170 /*
171 * write() is the generic write callback which maps directly to
172 * kernfs write operation and overrides all other operations.
173 * Maximum write size is determined by ->max_write_len.
174 */
175 ssize_t (*write)(struct kernfs_open_file *of,
176 char *buf, size_t nbytes, loff_t off);
177};
178
9f52425b
TL
179/**
180 * struct mbm_state - status for each MBM counter in each domain
181 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
182 * @prev_msr Value of IA32_QM_CTR for this RMID last time we read it
ba0f26d8
VS
183 * @chunks_bw Total local data moved. Used for bandwidth calculation
184 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting
185 * @prev_bw The most recent bandwidth in MBps
186 * @delta_bw Difference between the current and previous bandwidth
187 * @delta_comp Indicates whether to compute the delta_bw
9f52425b
TL
188 */
189struct mbm_state {
190 u64 chunks;
191 u64 prev_msr;
ba0f26d8
VS
192 u64 chunks_bw;
193 u64 prev_bw_msr;
194 u32 prev_bw;
195 u32 delta_bw;
196 bool delta_comp;
9f52425b
TL
197};
198
0921c547
TG
199/**
200 * struct rdt_domain - group of cpus sharing an RDT resource
201 * @list: all instances of this resource
202 * @id: unique id for this instance
203 * @cpu_mask: which cpus share this resource
edf6fa1c
VS
204 * @rmid_busy_llc:
205 * bitmap of which limbo RMIDs are above threshold
9f52425b
TL
206 * @mbm_total: saved state for MBM total bandwidth
207 * @mbm_local: saved state for MBM local bandwidth
e3302683 208 * @mbm_over: worker to periodically read MBM h/w counters
24247aee 209 * @cqm_limbo: worker to periodically read CQM h/w counters
e3302683
VS
210 * @mbm_work_cpu:
211 * worker cpu for MBM h/w counters
24247aee
VS
212 * @cqm_work_cpu:
213 * worker cpu for CQM h/w counters
0921c547 214 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
1bd2a63b 215 * @mbps_val: When mba_sc is enabled, this holds the bandwidth in MBps
0921c547
TG
216 * @new_ctrl: new ctrl value to be loaded
217 * @have_new_ctrl: did user provide new_ctrl for this domain
218 */
219struct rdt_domain {
220 struct list_head list;
221 int id;
222 struct cpumask cpu_mask;
edf6fa1c 223 unsigned long *rmid_busy_llc;
9f52425b
TL
224 struct mbm_state *mbm_total;
225 struct mbm_state *mbm_local;
e3302683 226 struct delayed_work mbm_over;
24247aee 227 struct delayed_work cqm_limbo;
e3302683 228 int mbm_work_cpu;
24247aee 229 int cqm_work_cpu;
0921c547 230 u32 *ctrl_val;
1bd2a63b 231 u32 *mbps_val;
0921c547
TG
232 u32 new_ctrl;
233 bool have_new_ctrl;
234};
235
236/**
237 * struct msr_param - set a range of MSRs from a domain
238 * @res: The resource to use
239 * @low: Beginning index from base MSR
240 * @high: End index
241 */
242struct msr_param {
243 struct rdt_resource *res;
244 int low;
245 int high;
246};
247
d3e11b4d
TG
248/**
249 * struct rdt_cache - Cache allocation related data
250 * @cbm_len: Length of the cache bit mask
251 * @min_cbm_bits: Minimum number of consecutive bits to be set
252 * @cbm_idx_mult: Multiplier of CBM index
253 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by:
254 * closid * cbm_idx_multi + cbm_idx_offset
255 * in a cache bit mask
0dd2d749
FY
256 * @shareable_bits: Bitmask of shareable resource with other
257 * executing entities
d3e11b4d
TG
258 */
259struct rdt_cache {
260 unsigned int cbm_len;
261 unsigned int min_cbm_bits;
262 unsigned int cbm_idx_mult;
263 unsigned int cbm_idx_offset;
0dd2d749 264 unsigned int shareable_bits;
d3e11b4d
TG
265};
266
05b93417
VS
267/**
268 * struct rdt_membw - Memory bandwidth allocation related data
269 * @max_delay: Max throttle delay. Delay is the hardware
270 * representation for memory bandwidth.
271 * @min_bw: Minimum memory bandwidth percentage user can request
272 * @bw_gran: Granularity at which the memory bandwidth is allocated
273 * @delay_linear: True if memory B/W delay is in linear scale
19c635ab 274 * @mba_sc: True if MBA software controller(mba_sc) is enabled
05b93417
VS
275 * @mb_map: Mapping of memory B/W percentage to memory B/W delay
276 */
277struct rdt_membw {
278 u32 max_delay;
279 u32 min_bw;
280 u32 bw_gran;
281 u32 delay_linear;
19c635ab 282 bool mba_sc;
05b93417
VS
283 u32 *mb_map;
284};
285
6a445edc
VS
286static inline bool is_llc_occupancy_enabled(void)
287{
288 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
289}
290
9f52425b
TL
291static inline bool is_mbm_total_enabled(void)
292{
293 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
294}
295
296static inline bool is_mbm_local_enabled(void)
297{
298 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
299}
300
301static inline bool is_mbm_enabled(void)
302{
303 return (is_mbm_total_enabled() || is_mbm_local_enabled());
304}
305
a4de1dfd
VS
306static inline bool is_mbm_event(int e)
307{
308 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
309 e <= QOS_L3_MBM_LOCAL_EVENT_ID);
310}
311
c1c7c3f9
FY
312/**
313 * struct rdt_resource - attributes of an RDT resource
d89b7379 314 * @rid: The index of the resource
1b5c0b75 315 * @alloc_enabled: Is allocation enabled on this machine
6a445edc 316 * @mon_enabled: Is monitoring enabled for this feature
1b5c0b75 317 * @alloc_capable: Is allocation available on this machine
6a445edc 318 * @mon_capable: Is monitor feature available on this machine
d3e11b4d
TG
319 * @name: Name to use in "schemata" file
320 * @num_closid: Number of CLOSIDs available
321 * @cache_level: Which cache level defines scope of this resource
322 * @default_ctrl: Specifies default cache cbm or memory B/W percent.
323 * @msr_base: Base MSR address for CBMs
0921c547 324 * @msr_update: Function pointer to update QOS MSRs
d3e11b4d
TG
325 * @data_width: Character width of data when displaying
326 * @domains: All domains for this resource
327 * @cache: Cache allocation related data
c6ea67de
VS
328 * @format_str: Per resource format string to show domain value
329 * @parse_ctrlval: Per resource function pointer to parse control values
6a445edc
VS
330 * @evt_list: List of monitoring events
331 * @num_rmid: Number of RMIDs available
332 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
5dc1d5c6 333 * @fflags: flags to choose base and info files
c1c7c3f9
FY
334 */
335struct rdt_resource {
d89b7379 336 int rid;
1b5c0b75 337 bool alloc_enabled;
6a445edc 338 bool mon_enabled;
1b5c0b75 339 bool alloc_capable;
6a445edc 340 bool mon_capable;
c1c7c3f9
FY
341 char *name;
342 int num_closid;
d3e11b4d 343 int cache_level;
2545e9f5 344 u32 default_ctrl;
d3e11b4d 345 unsigned int msr_base;
0921c547
TG
346 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
347 struct rdt_resource *r);
de016df8 348 int data_width;
c1c7c3f9 349 struct list_head domains;
a83827d0
TG
350 struct rdt_cache cache;
351 struct rdt_membw membw;
c6ea67de
VS
352 const char *format_str;
353 int (*parse_ctrlval) (char *buf, struct rdt_resource *r,
354 struct rdt_domain *d);
6a445edc
VS
355 struct list_head evt_list;
356 int num_rmid;
357 unsigned int mon_scale;
5dc1d5c6 358 unsigned long fflags;
c1c7c3f9
FY
359};
360
c6ea67de 361int parse_cbm(char *buf, struct rdt_resource *r, struct rdt_domain *d);
64e8ed3d 362int parse_bw(char *buf, struct rdt_resource *r, struct rdt_domain *d);
6a507a6a 363
2264d9c7
TL
364extern struct mutex rdtgroup_mutex;
365
c1c7c3f9 366extern struct rdt_resource rdt_resources_all[];
5ff193fb 367extern struct rdtgroup rdtgroup_default;
1b5c0b75 368DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
5ff193fb
FY
369
370int __init rdtgroup_init(void);
c1c7c3f9
FY
371
372enum {
373 RDT_RESOURCE_L3,
374 RDT_RESOURCE_L3DATA,
375 RDT_RESOURCE_L3CODE,
376 RDT_RESOURCE_L2,
def10853
FY
377 RDT_RESOURCE_L2DATA,
378 RDT_RESOURCE_L2CODE,
05b93417 379 RDT_RESOURCE_MBA,
c1c7c3f9
FY
380
381 /* Must be the last */
382 RDT_NUM_RESOURCES,
383};
384
895c663e
VS
385#define for_each_capable_rdt_resource(r) \
386 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
387 r++) \
388 if (r->alloc_capable || r->mon_capable)
389
1b5c0b75 390#define for_each_alloc_capable_rdt_resource(r) \
c1c7c3f9 391 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
17f8ba1d 392 r++) \
1b5c0b75 393 if (r->alloc_capable)
c1c7c3f9 394
6a445edc
VS
395#define for_each_mon_capable_rdt_resource(r) \
396 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
397 r++) \
398 if (r->mon_capable)
399
1b5c0b75 400#define for_each_alloc_enabled_rdt_resource(r) \
2264d9c7
TL
401 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
402 r++) \
1b5c0b75 403 if (r->alloc_enabled)
2264d9c7 404
d4ab3320
VS
405#define for_each_mon_enabled_rdt_resource(r) \
406 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
407 r++) \
408 if (r->mon_enabled)
409
c1c7c3f9
FY
410/* CPUID.(EAX=10H, ECX=ResID=1).EAX */
411union cpuid_0x10_1_eax {
412 struct {
413 unsigned int cbm_len:5;
414 } split;
415 unsigned int full;
416};
417
ab66a33b
VS
418/* CPUID.(EAX=10H, ECX=ResID=3).EAX */
419union cpuid_0x10_3_eax {
420 struct {
421 unsigned int max_delay:12;
422 } split;
423 unsigned int full;
424};
425
2545e9f5
VS
426/* CPUID.(EAX=10H, ECX=ResID).EDX */
427union cpuid_0x10_x_edx {
c1c7c3f9
FY
428 struct {
429 unsigned int cos_max:16;
430 } split;
431 unsigned int full;
432};
2264d9c7 433
9b3a7fd0
TL
434void rdt_last_cmd_clear(void);
435void rdt_last_cmd_puts(const char *s);
436void rdt_last_cmd_printf(const char *fmt, ...);
437
2545e9f5 438void rdt_ctrl_update(void *arg);
60cf5e10
FY
439struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
440void rdtgroup_kn_unlock(struct kernfs_node *kn);
d89b7379
VS
441struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
442 struct list_head **pos);
60ec2440
TL
443ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
444 char *buf, size_t nbytes, loff_t off);
445int rdtgroup_schemata_show(struct kernfs_open_file *of,
446 struct seq_file *s, void *v);
edf6fa1c 447struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
c7d9aac6
VS
448int alloc_rmid(void);
449void free_rmid(u32 rmid);
6a445edc 450int rdt_get_mon_l3_config(struct rdt_resource *r);
d89b7379
VS
451void mon_event_count(void *info);
452int rdtgroup_mondata_show(struct seq_file *m, void *arg);
895c663e
VS
453void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
454 unsigned int dom_id);
455void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
456 struct rdt_domain *d);
a4de1dfd
VS
457void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
458 struct rdtgroup *rdtgrp, int evtid, int first);
24247aee
VS
459void mbm_setup_overflow_handler(struct rdt_domain *dom,
460 unsigned long delay_ms);
e3302683 461void mbm_handle_overflow(struct work_struct *work);
19c635ab 462bool is_mba_sc(struct rdt_resource *r);
1bd2a63b 463void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm);
24247aee
VS
464void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
465void cqm_handle_limbo(struct work_struct *work);
466bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
467void __check_limbo(struct rdt_domain *d, bool force_free);
4f341a5e 468
113c6097 469#endif /* _ASM_X86_INTEL_RDT_H */