]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - arch/x86/kernel/cpu/intel_rdt.h
x86/speculation: Consolidate CPU whitelists
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / kernel / cpu / intel_rdt.h
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_L3_CBM_BASE 0xc90
11 #define IA32_L2_CBM_BASE 0xd10
12 #define IA32_MBA_THRTL_BASE 0xd50
13
14 #define L3_QOS_CDP_ENABLE 0x01ULL
15
16 /*
17 * Event IDs are used to program IA32_QM_EVTSEL before reading event
18 * counter from IA32_QM_CTR
19 */
20 #define QOS_L3_OCCUP_EVENT_ID 0x01
21 #define QOS_L3_MBM_TOTAL_EVENT_ID 0x02
22 #define QOS_L3_MBM_LOCAL_EVENT_ID 0x03
23
24 #define CQM_LIMBOCHECK_INTERVAL 1000
25
26 #define MBM_CNTR_WIDTH 24
27 #define MBM_OVERFLOW_INTERVAL 1000
28
29 #define RMID_VAL_ERROR BIT_ULL(63)
30 #define RMID_VAL_UNAVAIL BIT_ULL(62)
31
32 DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
33
34 /**
35 * struct mon_evt - Entry in the event list of a resource
36 * @evtid: event id
37 * @name: name of the event
38 */
39 struct mon_evt {
40 u32 evtid;
41 char *name;
42 struct list_head list;
43 };
44
45 /**
46 * struct mon_data_bits - Monitoring details for each event file
47 * @rid: Resource id associated with the event file.
48 * @evtid: Event id associated with the event file
49 * @domid: The domain to which the event file belongs
50 */
51 union mon_data_bits {
52 void *priv;
53 struct {
54 unsigned int rid : 10;
55 unsigned int evtid : 8;
56 unsigned int domid : 14;
57 } u;
58 };
59
60 struct rmid_read {
61 struct rdtgroup *rgrp;
62 struct rdt_domain *d;
63 int evtid;
64 bool first;
65 u64 val;
66 };
67
68 extern unsigned int intel_cqm_threshold;
69 extern bool rdt_alloc_capable;
70 extern bool rdt_mon_capable;
71 extern unsigned int rdt_mon_features;
72
73 enum rdt_group_type {
74 RDTCTRL_GROUP = 0,
75 RDTMON_GROUP,
76 RDT_NUM_GROUP,
77 };
78
79 /**
80 * struct mongroup - store mon group's data in resctrl fs.
81 * @mon_data_kn kernlfs node for the mon_data directory
82 * @parent: parent rdtgrp
83 * @crdtgrp_list: child rdtgroup node list
84 * @rmid: rmid for this rdtgroup
85 */
86 struct mongroup {
87 struct kernfs_node *mon_data_kn;
88 struct rdtgroup *parent;
89 struct list_head crdtgrp_list;
90 u32 rmid;
91 };
92
93 /**
94 * struct rdtgroup - store rdtgroup's data in resctrl file system.
95 * @kn: kernfs node
96 * @rdtgroup_list: linked list for all rdtgroups
97 * @closid: closid for this rdtgroup
98 * @cpu_mask: CPUs assigned to this rdtgroup
99 * @flags: status bits
100 * @waitcount: how many cpus expect to find this
101 * group when they acquire rdtgroup_mutex
102 * @type: indicates type of this rdtgroup - either
103 * monitor only or ctrl_mon group
104 * @mon: mongroup related data
105 */
106 struct rdtgroup {
107 struct kernfs_node *kn;
108 struct list_head rdtgroup_list;
109 u32 closid;
110 struct cpumask cpu_mask;
111 int flags;
112 atomic_t waitcount;
113 enum rdt_group_type type;
114 struct mongroup mon;
115 };
116
117 /* rdtgroup.flags */
118 #define RDT_DELETED 1
119
120 /* rftype.flags */
121 #define RFTYPE_FLAGS_CPUS_LIST 1
122
123 /*
124 * Define the file type flags for base and info directories.
125 */
126 #define RFTYPE_INFO BIT(0)
127 #define RFTYPE_BASE BIT(1)
128 #define RF_CTRLSHIFT 4
129 #define RF_MONSHIFT 5
130 #define RF_TOPSHIFT 6
131 #define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
132 #define RFTYPE_MON BIT(RF_MONSHIFT)
133 #define RFTYPE_TOP BIT(RF_TOPSHIFT)
134 #define RFTYPE_RES_CACHE BIT(8)
135 #define RFTYPE_RES_MB BIT(9)
136 #define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
137 #define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
138 #define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
139 #define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
140
141 /* List of all resource groups */
142 extern struct list_head rdt_all_groups;
143
144 extern int max_name_width, max_data_width;
145
146 int __init rdtgroup_init(void);
147
148 /**
149 * struct rftype - describe each file in the resctrl file system
150 * @name: File name
151 * @mode: Access mode
152 * @kf_ops: File operations
153 * @flags: File specific RFTYPE_FLAGS_* flags
154 * @fflags: File specific RF_* or RFTYPE_* flags
155 * @seq_show: Show content of the file
156 * @write: Write to the file
157 */
158 struct rftype {
159 char *name;
160 umode_t mode;
161 struct kernfs_ops *kf_ops;
162 unsigned long flags;
163 unsigned long fflags;
164
165 int (*seq_show)(struct kernfs_open_file *of,
166 struct seq_file *sf, void *v);
167 /*
168 * write() is the generic write callback which maps directly to
169 * kernfs write operation and overrides all other operations.
170 * Maximum write size is determined by ->max_write_len.
171 */
172 ssize_t (*write)(struct kernfs_open_file *of,
173 char *buf, size_t nbytes, loff_t off);
174 };
175
176 /**
177 * struct mbm_state - status for each MBM counter in each domain
178 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
179 * @prev_msr Value of IA32_QM_CTR for this RMID last time we read it
180 */
181 struct mbm_state {
182 u64 chunks;
183 u64 prev_msr;
184 };
185
186 /**
187 * struct rdt_domain - group of cpus sharing an RDT resource
188 * @list: all instances of this resource
189 * @id: unique id for this instance
190 * @cpu_mask: which cpus share this resource
191 * @rmid_busy_llc:
192 * bitmap of which limbo RMIDs are above threshold
193 * @mbm_total: saved state for MBM total bandwidth
194 * @mbm_local: saved state for MBM local bandwidth
195 * @mbm_over: worker to periodically read MBM h/w counters
196 * @cqm_limbo: worker to periodically read CQM h/w counters
197 * @mbm_work_cpu:
198 * worker cpu for MBM h/w counters
199 * @cqm_work_cpu:
200 * worker cpu for CQM h/w counters
201 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
202 * @new_ctrl: new ctrl value to be loaded
203 * @have_new_ctrl: did user provide new_ctrl for this domain
204 */
205 struct rdt_domain {
206 struct list_head list;
207 int id;
208 struct cpumask cpu_mask;
209 unsigned long *rmid_busy_llc;
210 struct mbm_state *mbm_total;
211 struct mbm_state *mbm_local;
212 struct delayed_work mbm_over;
213 struct delayed_work cqm_limbo;
214 int mbm_work_cpu;
215 int cqm_work_cpu;
216 u32 *ctrl_val;
217 u32 new_ctrl;
218 bool have_new_ctrl;
219 };
220
221 /**
222 * struct msr_param - set a range of MSRs from a domain
223 * @res: The resource to use
224 * @low: Beginning index from base MSR
225 * @high: End index
226 */
227 struct msr_param {
228 struct rdt_resource *res;
229 int low;
230 int high;
231 };
232
233 /**
234 * struct rdt_cache - Cache allocation related data
235 * @cbm_len: Length of the cache bit mask
236 * @min_cbm_bits: Minimum number of consecutive bits to be set
237 * @cbm_idx_mult: Multiplier of CBM index
238 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by:
239 * closid * cbm_idx_multi + cbm_idx_offset
240 * in a cache bit mask
241 * @shareable_bits: Bitmask of shareable resource with other
242 * executing entities
243 */
244 struct rdt_cache {
245 unsigned int cbm_len;
246 unsigned int min_cbm_bits;
247 unsigned int cbm_idx_mult;
248 unsigned int cbm_idx_offset;
249 unsigned int shareable_bits;
250 };
251
252 /**
253 * struct rdt_membw - Memory bandwidth allocation related data
254 * @max_delay: Max throttle delay. Delay is the hardware
255 * representation for memory bandwidth.
256 * @min_bw: Minimum memory bandwidth percentage user can request
257 * @bw_gran: Granularity at which the memory bandwidth is allocated
258 * @delay_linear: True if memory B/W delay is in linear scale
259 * @mb_map: Mapping of memory B/W percentage to memory B/W delay
260 */
261 struct rdt_membw {
262 u32 max_delay;
263 u32 min_bw;
264 u32 bw_gran;
265 u32 delay_linear;
266 u32 *mb_map;
267 };
268
269 static inline bool is_llc_occupancy_enabled(void)
270 {
271 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
272 }
273
274 static inline bool is_mbm_total_enabled(void)
275 {
276 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
277 }
278
279 static inline bool is_mbm_local_enabled(void)
280 {
281 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
282 }
283
284 static inline bool is_mbm_enabled(void)
285 {
286 return (is_mbm_total_enabled() || is_mbm_local_enabled());
287 }
288
289 static inline bool is_mbm_event(int e)
290 {
291 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
292 e <= QOS_L3_MBM_LOCAL_EVENT_ID);
293 }
294
295 /**
296 * struct rdt_resource - attributes of an RDT resource
297 * @rid: The index of the resource
298 * @alloc_enabled: Is allocation enabled on this machine
299 * @mon_enabled: Is monitoring enabled for this feature
300 * @alloc_capable: Is allocation available on this machine
301 * @mon_capable: Is monitor feature available on this machine
302 * @name: Name to use in "schemata" file
303 * @num_closid: Number of CLOSIDs available
304 * @cache_level: Which cache level defines scope of this resource
305 * @default_ctrl: Specifies default cache cbm or memory B/W percent.
306 * @msr_base: Base MSR address for CBMs
307 * @msr_update: Function pointer to update QOS MSRs
308 * @data_width: Character width of data when displaying
309 * @domains: All domains for this resource
310 * @cache: Cache allocation related data
311 * @format_str: Per resource format string to show domain value
312 * @parse_ctrlval: Per resource function pointer to parse control values
313 * @evt_list: List of monitoring events
314 * @num_rmid: Number of RMIDs available
315 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
316 * @fflags: flags to choose base and info files
317 */
318 struct rdt_resource {
319 int rid;
320 bool alloc_enabled;
321 bool mon_enabled;
322 bool alloc_capable;
323 bool mon_capable;
324 char *name;
325 int num_closid;
326 int cache_level;
327 u32 default_ctrl;
328 unsigned int msr_base;
329 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
330 struct rdt_resource *r);
331 int data_width;
332 struct list_head domains;
333 struct rdt_cache cache;
334 struct rdt_membw membw;
335 const char *format_str;
336 int (*parse_ctrlval) (char *buf, struct rdt_resource *r,
337 struct rdt_domain *d);
338 struct list_head evt_list;
339 int num_rmid;
340 unsigned int mon_scale;
341 unsigned long fflags;
342 };
343
344 int parse_cbm(char *buf, struct rdt_resource *r, struct rdt_domain *d);
345 int parse_bw(char *buf, struct rdt_resource *r, struct rdt_domain *d);
346
347 extern struct mutex rdtgroup_mutex;
348
349 extern struct rdt_resource rdt_resources_all[];
350 extern struct rdtgroup rdtgroup_default;
351 DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
352
353 int __init rdtgroup_init(void);
354
355 enum {
356 RDT_RESOURCE_L3,
357 RDT_RESOURCE_L3DATA,
358 RDT_RESOURCE_L3CODE,
359 RDT_RESOURCE_L2,
360 RDT_RESOURCE_MBA,
361
362 /* Must be the last */
363 RDT_NUM_RESOURCES,
364 };
365
366 #define for_each_capable_rdt_resource(r) \
367 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
368 r++) \
369 if (r->alloc_capable || r->mon_capable)
370
371 #define for_each_alloc_capable_rdt_resource(r) \
372 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
373 r++) \
374 if (r->alloc_capable)
375
376 #define for_each_mon_capable_rdt_resource(r) \
377 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
378 r++) \
379 if (r->mon_capable)
380
381 #define for_each_alloc_enabled_rdt_resource(r) \
382 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
383 r++) \
384 if (r->alloc_enabled)
385
386 #define for_each_mon_enabled_rdt_resource(r) \
387 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
388 r++) \
389 if (r->mon_enabled)
390
391 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */
392 union cpuid_0x10_1_eax {
393 struct {
394 unsigned int cbm_len:5;
395 } split;
396 unsigned int full;
397 };
398
399 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */
400 union cpuid_0x10_3_eax {
401 struct {
402 unsigned int max_delay:12;
403 } split;
404 unsigned int full;
405 };
406
407 /* CPUID.(EAX=10H, ECX=ResID).EDX */
408 union cpuid_0x10_x_edx {
409 struct {
410 unsigned int cos_max:16;
411 } split;
412 unsigned int full;
413 };
414
415 void rdt_last_cmd_clear(void);
416 void rdt_last_cmd_puts(const char *s);
417 void rdt_last_cmd_printf(const char *fmt, ...);
418
419 void rdt_ctrl_update(void *arg);
420 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
421 void rdtgroup_kn_unlock(struct kernfs_node *kn);
422 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
423 struct list_head **pos);
424 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
425 char *buf, size_t nbytes, loff_t off);
426 int rdtgroup_schemata_show(struct kernfs_open_file *of,
427 struct seq_file *s, void *v);
428 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
429 int alloc_rmid(void);
430 void free_rmid(u32 rmid);
431 int rdt_get_mon_l3_config(struct rdt_resource *r);
432 void mon_event_count(void *info);
433 int rdtgroup_mondata_show(struct seq_file *m, void *arg);
434 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
435 unsigned int dom_id);
436 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
437 struct rdt_domain *d);
438 void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
439 struct rdtgroup *rdtgrp, int evtid, int first);
440 void mbm_setup_overflow_handler(struct rdt_domain *dom,
441 unsigned long delay_ms);
442 void mbm_handle_overflow(struct work_struct *work);
443 void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
444 void cqm_handle_limbo(struct work_struct *work);
445 bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
446 void __check_limbo(struct rdt_domain *d, bool force_free);
447
448 #endif /* _ASM_X86_INTEL_RDT_H */