Loading...
Note: File does not exist in v3.1.
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _ASM_X86_INTEL_RDT_H
3#define _ASM_X86_INTEL_RDT_H
4
5#include <linux/sched.h>
6#include <linux/kernfs.h>
7#include <linux/jump_label.h>
8
9#define IA32_L3_QOS_CFG 0xc81
10#define IA32_L2_QOS_CFG 0xc82
11#define IA32_L3_CBM_BASE 0xc90
12#define IA32_L2_CBM_BASE 0xd10
13#define IA32_MBA_THRTL_BASE 0xd50
14
15#define L3_QOS_CDP_ENABLE 0x01ULL
16
17#define L2_QOS_CDP_ENABLE 0x01ULL
18
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
26
27#define CQM_LIMBOCHECK_INTERVAL 1000
28
29#define MBM_CNTR_WIDTH 24
30#define MBM_OVERFLOW_INTERVAL 1000
31
32#define RMID_VAL_ERROR BIT_ULL(63)
33#define RMID_VAL_UNAVAIL BIT_ULL(62)
34
35DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
36
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
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;
65 struct rdt_domain *d;
66 int evtid;
67 bool first;
68 u64 val;
69};
70
71extern unsigned int intel_cqm_threshold;
72extern bool rdt_alloc_capable;
73extern bool rdt_mon_capable;
74extern unsigned int rdt_mon_features;
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.
84 * @mon_data_kn kernlfs node for the mon_data directory
85 * @parent: parent rdtgrp
86 * @crdtgrp_list: child rdtgroup node list
87 * @rmid: rmid for this rdtgroup
88 */
89struct mongroup {
90 struct kernfs_node *mon_data_kn;
91 struct rdtgroup *parent;
92 struct list_head crdtgrp_list;
93 u32 rmid;
94};
95
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
101 * @cpu_mask: CPUs assigned to this rdtgroup
102 * @flags: status bits
103 * @waitcount: how many cpus expect to find this
104 * group when they acquire rdtgroup_mutex
105 * @type: indicates type of this rdtgroup - either
106 * monitor only or ctrl_mon group
107 * @mon: mongroup related data
108 */
109struct rdtgroup {
110 struct kernfs_node *kn;
111 struct list_head rdtgroup_list;
112 u32 closid;
113 struct cpumask cpu_mask;
114 int flags;
115 atomic_t waitcount;
116 enum rdt_group_type type;
117 struct mongroup mon;
118};
119
120/* rdtgroup.flags */
121#define RDT_DELETED 1
122
123/* rftype.flags */
124#define RFTYPE_FLAGS_CPUS_LIST 1
125
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
132#define RF_MONSHIFT 5
133#define RF_TOPSHIFT 6
134#define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
135#define RFTYPE_MON BIT(RF_MONSHIFT)
136#define RFTYPE_TOP BIT(RF_TOPSHIFT)
137#define RFTYPE_RES_CACHE BIT(8)
138#define RFTYPE_RES_MB BIT(9)
139#define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
140#define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
141#define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
142#define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
143
144/* List of all resource groups */
145extern struct list_head rdt_all_groups;
146
147extern int max_name_width, max_data_width;
148
149int __init rdtgroup_init(void);
150
151/**
152 * struct rftype - describe each file in the resctrl file system
153 * @name: File name
154 * @mode: Access mode
155 * @kf_ops: File operations
156 * @flags: File specific RFTYPE_FLAGS_* flags
157 * @fflags: File specific RF_* or RFTYPE_* flags
158 * @seq_show: Show content of the file
159 * @write: Write to the file
160 */
161struct rftype {
162 char *name;
163 umode_t mode;
164 struct kernfs_ops *kf_ops;
165 unsigned long flags;
166 unsigned long fflags;
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
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
183 */
184struct mbm_state {
185 u64 chunks;
186 u64 prev_msr;
187};
188
189/**
190 * struct rdt_domain - group of cpus sharing an RDT resource
191 * @list: all instances of this resource
192 * @id: unique id for this instance
193 * @cpu_mask: which cpus share this resource
194 * @rmid_busy_llc:
195 * bitmap of which limbo RMIDs are above threshold
196 * @mbm_total: saved state for MBM total bandwidth
197 * @mbm_local: saved state for MBM local bandwidth
198 * @mbm_over: worker to periodically read MBM h/w counters
199 * @cqm_limbo: worker to periodically read CQM h/w counters
200 * @mbm_work_cpu:
201 * worker cpu for MBM h/w counters
202 * @cqm_work_cpu:
203 * worker cpu for CQM h/w counters
204 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
205 * @new_ctrl: new ctrl value to be loaded
206 * @have_new_ctrl: did user provide new_ctrl for this domain
207 */
208struct rdt_domain {
209 struct list_head list;
210 int id;
211 struct cpumask cpu_mask;
212 unsigned long *rmid_busy_llc;
213 struct mbm_state *mbm_total;
214 struct mbm_state *mbm_local;
215 struct delayed_work mbm_over;
216 struct delayed_work cqm_limbo;
217 int mbm_work_cpu;
218 int cqm_work_cpu;
219 u32 *ctrl_val;
220 u32 new_ctrl;
221 bool have_new_ctrl;
222};
223
224/**
225 * struct msr_param - set a range of MSRs from a domain
226 * @res: The resource to use
227 * @low: Beginning index from base MSR
228 * @high: End index
229 */
230struct msr_param {
231 struct rdt_resource *res;
232 int low;
233 int high;
234};
235
236/**
237 * struct rdt_cache - Cache allocation related data
238 * @cbm_len: Length of the cache bit mask
239 * @min_cbm_bits: Minimum number of consecutive bits to be set
240 * @cbm_idx_mult: Multiplier of CBM index
241 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by:
242 * closid * cbm_idx_multi + cbm_idx_offset
243 * in a cache bit mask
244 * @shareable_bits: Bitmask of shareable resource with other
245 * executing entities
246 */
247struct rdt_cache {
248 unsigned int cbm_len;
249 unsigned int min_cbm_bits;
250 unsigned int cbm_idx_mult;
251 unsigned int cbm_idx_offset;
252 unsigned int shareable_bits;
253};
254
255/**
256 * struct rdt_membw - Memory bandwidth allocation related data
257 * @max_delay: Max throttle delay. Delay is the hardware
258 * representation for memory bandwidth.
259 * @min_bw: Minimum memory bandwidth percentage user can request
260 * @bw_gran: Granularity at which the memory bandwidth is allocated
261 * @delay_linear: True if memory B/W delay is in linear scale
262 * @mb_map: Mapping of memory B/W percentage to memory B/W delay
263 */
264struct rdt_membw {
265 u32 max_delay;
266 u32 min_bw;
267 u32 bw_gran;
268 u32 delay_linear;
269 u32 *mb_map;
270};
271
272static inline bool is_llc_occupancy_enabled(void)
273{
274 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
275}
276
277static inline bool is_mbm_total_enabled(void)
278{
279 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
280}
281
282static inline bool is_mbm_local_enabled(void)
283{
284 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
285}
286
287static inline bool is_mbm_enabled(void)
288{
289 return (is_mbm_total_enabled() || is_mbm_local_enabled());
290}
291
292static inline bool is_mbm_event(int e)
293{
294 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
295 e <= QOS_L3_MBM_LOCAL_EVENT_ID);
296}
297
298/**
299 * struct rdt_resource - attributes of an RDT resource
300 * @rid: The index of the resource
301 * @alloc_enabled: Is allocation enabled on this machine
302 * @mon_enabled: Is monitoring enabled for this feature
303 * @alloc_capable: Is allocation available on this machine
304 * @mon_capable: Is monitor feature available on this machine
305 * @name: Name to use in "schemata" file
306 * @num_closid: Number of CLOSIDs available
307 * @cache_level: Which cache level defines scope of this resource
308 * @default_ctrl: Specifies default cache cbm or memory B/W percent.
309 * @msr_base: Base MSR address for CBMs
310 * @msr_update: Function pointer to update QOS MSRs
311 * @data_width: Character width of data when displaying
312 * @domains: All domains for this resource
313 * @cache: Cache allocation related data
314 * @format_str: Per resource format string to show domain value
315 * @parse_ctrlval: Per resource function pointer to parse control values
316 * @evt_list: List of monitoring events
317 * @num_rmid: Number of RMIDs available
318 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
319 * @fflags: flags to choose base and info files
320 */
321struct rdt_resource {
322 int rid;
323 bool alloc_enabled;
324 bool mon_enabled;
325 bool alloc_capable;
326 bool mon_capable;
327 char *name;
328 int num_closid;
329 int cache_level;
330 u32 default_ctrl;
331 unsigned int msr_base;
332 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
333 struct rdt_resource *r);
334 int data_width;
335 struct list_head domains;
336 struct rdt_cache cache;
337 struct rdt_membw membw;
338 const char *format_str;
339 int (*parse_ctrlval) (char *buf, struct rdt_resource *r,
340 struct rdt_domain *d);
341 struct list_head evt_list;
342 int num_rmid;
343 unsigned int mon_scale;
344 unsigned long fflags;
345};
346
347int parse_cbm(char *buf, struct rdt_resource *r, struct rdt_domain *d);
348int parse_bw(char *buf, struct rdt_resource *r, struct rdt_domain *d);
349
350extern struct mutex rdtgroup_mutex;
351
352extern struct rdt_resource rdt_resources_all[];
353extern struct rdtgroup rdtgroup_default;
354DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
355
356int __init rdtgroup_init(void);
357
358enum {
359 RDT_RESOURCE_L3,
360 RDT_RESOURCE_L3DATA,
361 RDT_RESOURCE_L3CODE,
362 RDT_RESOURCE_L2,
363 RDT_RESOURCE_L2DATA,
364 RDT_RESOURCE_L2CODE,
365 RDT_RESOURCE_MBA,
366
367 /* Must be the last */
368 RDT_NUM_RESOURCES,
369};
370
371#define for_each_capable_rdt_resource(r) \
372 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
373 r++) \
374 if (r->alloc_capable || r->mon_capable)
375
376#define for_each_alloc_capable_rdt_resource(r) \
377 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
378 r++) \
379 if (r->alloc_capable)
380
381#define for_each_mon_capable_rdt_resource(r) \
382 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
383 r++) \
384 if (r->mon_capable)
385
386#define for_each_alloc_enabled_rdt_resource(r) \
387 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
388 r++) \
389 if (r->alloc_enabled)
390
391#define for_each_mon_enabled_rdt_resource(r) \
392 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
393 r++) \
394 if (r->mon_enabled)
395
396/* CPUID.(EAX=10H, ECX=ResID=1).EAX */
397union cpuid_0x10_1_eax {
398 struct {
399 unsigned int cbm_len:5;
400 } split;
401 unsigned int full;
402};
403
404/* CPUID.(EAX=10H, ECX=ResID=3).EAX */
405union cpuid_0x10_3_eax {
406 struct {
407 unsigned int max_delay:12;
408 } split;
409 unsigned int full;
410};
411
412/* CPUID.(EAX=10H, ECX=ResID).EDX */
413union cpuid_0x10_x_edx {
414 struct {
415 unsigned int cos_max:16;
416 } split;
417 unsigned int full;
418};
419
420void rdt_last_cmd_clear(void);
421void rdt_last_cmd_puts(const char *s);
422void rdt_last_cmd_printf(const char *fmt, ...);
423
424void rdt_ctrl_update(void *arg);
425struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
426void rdtgroup_kn_unlock(struct kernfs_node *kn);
427struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
428 struct list_head **pos);
429ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
430 char *buf, size_t nbytes, loff_t off);
431int rdtgroup_schemata_show(struct kernfs_open_file *of,
432 struct seq_file *s, void *v);
433struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
434int alloc_rmid(void);
435void free_rmid(u32 rmid);
436int rdt_get_mon_l3_config(struct rdt_resource *r);
437void mon_event_count(void *info);
438int rdtgroup_mondata_show(struct seq_file *m, void *arg);
439void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
440 unsigned int dom_id);
441void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
442 struct rdt_domain *d);
443void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
444 struct rdtgroup *rdtgrp, int evtid, int first);
445void mbm_setup_overflow_handler(struct rdt_domain *dom,
446 unsigned long delay_ms);
447void mbm_handle_overflow(struct work_struct *work);
448void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
449void cqm_handle_limbo(struct work_struct *work);
450bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
451void __check_limbo(struct rdt_domain *d, bool force_free);
452
453#endif /* _ASM_X86_INTEL_RDT_H */