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
11 #define MSR_IA32_L3_QOS_CFG 0xc81
12 #define MSR_IA32_L2_QOS_CFG 0xc82
13 #define MSR_IA32_L3_CBM_BASE 0xc90
14 #define MSR_IA32_L2_CBM_BASE 0xd10
15 #define MSR_IA32_MBA_THRTL_BASE 0xd50
16 #define MSR_IA32_MBA_BW_BASE 0xc0000200
17
18 #define MSR_IA32_QM_CTR 0x0c8e
19 #define MSR_IA32_QM_EVTSEL 0x0c8d
20
21 #define L3_QOS_CDP_ENABLE 0x01ULL
22
23 #define L2_QOS_CDP_ENABLE 0x01ULL
24
25 /*
26 * Event IDs are used to program IA32_QM_EVTSEL before reading event
27 * counter from IA32_QM_CTR
28 */
29 #define QOS_L3_OCCUP_EVENT_ID 0x01
30 #define QOS_L3_MBM_TOTAL_EVENT_ID 0x02
31 #define QOS_L3_MBM_LOCAL_EVENT_ID 0x03
32
33 #define CQM_LIMBOCHECK_INTERVAL 1000
34
35 #define MBM_CNTR_WIDTH_BASE 24
36 #define MBM_OVERFLOW_INTERVAL 1000
37 #define MAX_MBA_BW 100u
38 #define MBA_IS_LINEAR 0x4
39 #define MBA_MAX_MBPS U32_MAX
40 #define MAX_MBA_BW_AMD 0x800
41 #define MBM_CNTR_WIDTH_OFFSET_AMD 20
42
43 #define RMID_VAL_ERROR BIT_ULL(63)
44 #define RMID_VAL_UNAVAIL BIT_ULL(62)
45 /*
46 * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for
47 * data to be returned. The counter width is discovered from the hardware
48 * as an offset from MBM_CNTR_WIDTH_BASE.
49 */
50 #define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
51
52
53 struct rdt_fs_context {
54 struct kernfs_fs_context kfc;
55 bool enable_cdpl2;
56 bool enable_cdpl3;
57 bool enable_mba_mbps;
58 };
59
rdt_fc2context(struct fs_context * fc)60 static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
61 {
62 struct kernfs_fs_context *kfc = fc->fs_private;
63
64 return container_of(kfc, struct rdt_fs_context, kfc);
65 }
66
67 DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
68 DECLARE_STATIC_KEY_FALSE(rdt_mon_enable_key);
69
70 /**
71 * struct mon_evt - Entry in the event list of a resource
72 * @evtid: event id
73 * @name: name of the event
74 * @list: entry in &rdt_resource->evt_list
75 */
76 struct mon_evt {
77 u32 evtid;
78 char *name;
79 struct list_head list;
80 };
81
82 /**
83 * union mon_data_bits - Monitoring details for each event file
84 * @priv: Used to store monitoring event data in @u
85 * as kernfs private data
86 * @rid: Resource id associated with the event file
87 * @evtid: Event id associated with the event file
88 * @domid: The domain to which the event file belongs
89 * @u: Name of the bit fields struct
90 */
91 union mon_data_bits {
92 void *priv;
93 struct {
94 unsigned int rid : 10;
95 unsigned int evtid : 8;
96 unsigned int domid : 14;
97 } u;
98 };
99
100 struct rmid_read {
101 struct rdtgroup *rgrp;
102 struct rdt_resource *r;
103 struct rdt_domain *d;
104 int evtid;
105 bool first;
106 u64 val;
107 };
108
109 extern unsigned int resctrl_cqm_threshold;
110 extern bool rdt_alloc_capable;
111 extern bool rdt_mon_capable;
112 extern unsigned int rdt_mon_features;
113 extern struct list_head resctrl_schema_all;
114
115 enum rdt_group_type {
116 RDTCTRL_GROUP = 0,
117 RDTMON_GROUP,
118 RDT_NUM_GROUP,
119 };
120
121 /**
122 * enum rdtgrp_mode - Mode of a RDT resource group
123 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
124 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
125 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
126 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
127 * allowed AND the allocations are Cache Pseudo-Locked
128 * @RDT_NUM_MODES: Total number of modes
129 *
130 * The mode of a resource group enables control over the allowed overlap
131 * between allocations associated with different resource groups (classes
132 * of service). User is able to modify the mode of a resource group by
133 * writing to the "mode" resctrl file associated with the resource group.
134 *
135 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
136 * writing the appropriate text to the "mode" file. A resource group enters
137 * "pseudo-locked" mode after the schemata is written while the resource
138 * group is in "pseudo-locksetup" mode.
139 */
140 enum rdtgrp_mode {
141 RDT_MODE_SHAREABLE = 0,
142 RDT_MODE_EXCLUSIVE,
143 RDT_MODE_PSEUDO_LOCKSETUP,
144 RDT_MODE_PSEUDO_LOCKED,
145
146 /* Must be last */
147 RDT_NUM_MODES,
148 };
149
150 /**
151 * struct mongroup - store mon group's data in resctrl fs.
152 * @mon_data_kn: kernfs node for the mon_data directory
153 * @parent: parent rdtgrp
154 * @crdtgrp_list: child rdtgroup node list
155 * @rmid: rmid for this rdtgroup
156 */
157 struct mongroup {
158 struct kernfs_node *mon_data_kn;
159 struct rdtgroup *parent;
160 struct list_head crdtgrp_list;
161 u32 rmid;
162 };
163
164 /**
165 * struct pseudo_lock_region - pseudo-lock region information
166 * @s: Resctrl schema for the resource to which this
167 * pseudo-locked region belongs
168 * @d: RDT domain to which this pseudo-locked region
169 * belongs
170 * @cbm: bitmask of the pseudo-locked region
171 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread
172 * completion
173 * @thread_done: variable used by waitqueue to test if pseudo-locking
174 * thread completed
175 * @cpu: core associated with the cache on which the setup code
176 * will be run
177 * @line_size: size of the cache lines
178 * @size: size of pseudo-locked region in bytes
179 * @kmem: the kernel memory associated with pseudo-locked region
180 * @minor: minor number of character device associated with this
181 * region
182 * @debugfs_dir: pointer to this region's directory in the debugfs
183 * filesystem
184 * @pm_reqs: Power management QoS requests related to this region
185 */
186 struct pseudo_lock_region {
187 struct resctrl_schema *s;
188 struct rdt_domain *d;
189 u32 cbm;
190 wait_queue_head_t lock_thread_wq;
191 int thread_done;
192 int cpu;
193 unsigned int line_size;
194 unsigned int size;
195 void *kmem;
196 unsigned int minor;
197 struct dentry *debugfs_dir;
198 struct list_head pm_reqs;
199 };
200
201 /**
202 * struct rdtgroup - store rdtgroup's data in resctrl file system.
203 * @kn: kernfs node
204 * @rdtgroup_list: linked list for all rdtgroups
205 * @closid: closid for this rdtgroup
206 * @cpu_mask: CPUs assigned to this rdtgroup
207 * @flags: status bits
208 * @waitcount: how many cpus expect to find this
209 * group when they acquire rdtgroup_mutex
210 * @type: indicates type of this rdtgroup - either
211 * monitor only or ctrl_mon group
212 * @mon: mongroup related data
213 * @mode: mode of resource group
214 * @plr: pseudo-locked region
215 */
216 struct rdtgroup {
217 struct kernfs_node *kn;
218 struct list_head rdtgroup_list;
219 u32 closid;
220 struct cpumask cpu_mask;
221 int flags;
222 atomic_t waitcount;
223 enum rdt_group_type type;
224 struct mongroup mon;
225 enum rdtgrp_mode mode;
226 struct pseudo_lock_region *plr;
227 };
228
229 /* rdtgroup.flags */
230 #define RDT_DELETED 1
231
232 /* rftype.flags */
233 #define RFTYPE_FLAGS_CPUS_LIST 1
234
235 /*
236 * Define the file type flags for base and info directories.
237 */
238 #define RFTYPE_INFO BIT(0)
239 #define RFTYPE_BASE BIT(1)
240 #define RF_CTRLSHIFT 4
241 #define RF_MONSHIFT 5
242 #define RF_TOPSHIFT 6
243 #define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
244 #define RFTYPE_MON BIT(RF_MONSHIFT)
245 #define RFTYPE_TOP BIT(RF_TOPSHIFT)
246 #define RFTYPE_RES_CACHE BIT(8)
247 #define RFTYPE_RES_MB BIT(9)
248 #define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
249 #define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
250 #define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
251 #define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
252
253 /* List of all resource groups */
254 extern struct list_head rdt_all_groups;
255
256 extern int max_name_width, max_data_width;
257
258 int __init rdtgroup_init(void);
259 void __exit rdtgroup_exit(void);
260
261 /**
262 * struct rftype - describe each file in the resctrl file system
263 * @name: File name
264 * @mode: Access mode
265 * @kf_ops: File operations
266 * @flags: File specific RFTYPE_FLAGS_* flags
267 * @fflags: File specific RF_* or RFTYPE_* flags
268 * @seq_show: Show content of the file
269 * @write: Write to the file
270 */
271 struct rftype {
272 char *name;
273 umode_t mode;
274 const struct kernfs_ops *kf_ops;
275 unsigned long flags;
276 unsigned long fflags;
277
278 int (*seq_show)(struct kernfs_open_file *of,
279 struct seq_file *sf, void *v);
280 /*
281 * write() is the generic write callback which maps directly to
282 * kernfs write operation and overrides all other operations.
283 * Maximum write size is determined by ->max_write_len.
284 */
285 ssize_t (*write)(struct kernfs_open_file *of,
286 char *buf, size_t nbytes, loff_t off);
287 };
288
289 /**
290 * struct mbm_state - status for each MBM counter in each domain
291 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
292 * @prev_msr: Value of IA32_QM_CTR for this RMID last time we read it
293 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting
294 * @prev_bw: The most recent bandwidth in MBps
295 * @delta_bw: Difference between the current and previous bandwidth
296 * @delta_comp: Indicates whether to compute the delta_bw
297 */
298 struct mbm_state {
299 u64 chunks;
300 u64 prev_msr;
301 u64 prev_bw_msr;
302 u32 prev_bw;
303 u32 delta_bw;
304 bool delta_comp;
305 };
306
307 /**
308 * struct rdt_hw_domain - Arch private attributes of a set of CPUs that share
309 * a resource
310 * @d_resctrl: Properties exposed to the resctrl file system
311 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
312 * @mbps_val: When mba_sc is enabled, this holds the bandwidth in MBps
313 *
314 * Members of this structure are accessed via helpers that provide abstraction.
315 */
316 struct rdt_hw_domain {
317 struct rdt_domain d_resctrl;
318 u32 *ctrl_val;
319 u32 *mbps_val;
320 };
321
resctrl_to_arch_dom(struct rdt_domain * r)322 static inline struct rdt_hw_domain *resctrl_to_arch_dom(struct rdt_domain *r)
323 {
324 return container_of(r, struct rdt_hw_domain, d_resctrl);
325 }
326
327 /**
328 * struct msr_param - set a range of MSRs from a domain
329 * @res: The resource to use
330 * @low: Beginning index from base MSR
331 * @high: End index
332 */
333 struct msr_param {
334 struct rdt_resource *res;
335 u32 low;
336 u32 high;
337 };
338
is_llc_occupancy_enabled(void)339 static inline bool is_llc_occupancy_enabled(void)
340 {
341 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
342 }
343
is_mbm_total_enabled(void)344 static inline bool is_mbm_total_enabled(void)
345 {
346 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
347 }
348
is_mbm_local_enabled(void)349 static inline bool is_mbm_local_enabled(void)
350 {
351 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
352 }
353
is_mbm_enabled(void)354 static inline bool is_mbm_enabled(void)
355 {
356 return (is_mbm_total_enabled() || is_mbm_local_enabled());
357 }
358
is_mbm_event(int e)359 static inline bool is_mbm_event(int e)
360 {
361 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
362 e <= QOS_L3_MBM_LOCAL_EVENT_ID);
363 }
364
365 struct rdt_parse_data {
366 struct rdtgroup *rdtgrp;
367 char *buf;
368 };
369
370 /**
371 * struct rdt_hw_resource - arch private attributes of a resctrl resource
372 * @r_resctrl: Attributes of the resource used directly by resctrl.
373 * @num_closid: Maximum number of closid this hardware can support,
374 * regardless of CDP. This is exposed via
375 * resctrl_arch_get_num_closid() to avoid confusion
376 * with struct resctrl_schema's property of the same name,
377 * which has been corrected for features like CDP.
378 * @msr_base: Base MSR address for CBMs
379 * @msr_update: Function pointer to update QOS MSRs
380 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
381 * @mbm_width: Monitor width, to detect and correct for overflow.
382 * @cdp_enabled: CDP state of this resource
383 *
384 * Members of this structure are either private to the architecture
385 * e.g. mbm_width, or accessed via helpers that provide abstraction. e.g.
386 * msr_update and msr_base.
387 */
388 struct rdt_hw_resource {
389 struct rdt_resource r_resctrl;
390 u32 num_closid;
391 unsigned int msr_base;
392 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
393 struct rdt_resource *r);
394 unsigned int mon_scale;
395 unsigned int mbm_width;
396 bool cdp_enabled;
397 };
398
resctrl_to_arch_res(struct rdt_resource * r)399 static inline struct rdt_hw_resource *resctrl_to_arch_res(struct rdt_resource *r)
400 {
401 return container_of(r, struct rdt_hw_resource, r_resctrl);
402 }
403
404 int parse_cbm(struct rdt_parse_data *data, struct resctrl_schema *s,
405 struct rdt_domain *d);
406 int parse_bw(struct rdt_parse_data *data, struct resctrl_schema *s,
407 struct rdt_domain *d);
408
409 extern struct mutex rdtgroup_mutex;
410
411 extern struct rdt_hw_resource rdt_resources_all[];
412 extern struct rdtgroup rdtgroup_default;
413 DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
414
415 extern struct dentry *debugfs_resctrl;
416
417 enum resctrl_res_level {
418 RDT_RESOURCE_L3,
419 RDT_RESOURCE_L2,
420 RDT_RESOURCE_MBA,
421
422 /* Must be the last */
423 RDT_NUM_RESOURCES,
424 };
425
resctrl_inc(struct rdt_resource * res)426 static inline struct rdt_resource *resctrl_inc(struct rdt_resource *res)
427 {
428 struct rdt_hw_resource *hw_res = resctrl_to_arch_res(res);
429
430 hw_res++;
431 return &hw_res->r_resctrl;
432 }
433
resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)434 static inline bool resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)
435 {
436 return rdt_resources_all[l].cdp_enabled;
437 }
438
439 int resctrl_arch_set_cdp_enabled(enum resctrl_res_level l, bool enable);
440
441 /*
442 * To return the common struct rdt_resource, which is contained in struct
443 * rdt_hw_resource, walk the resctrl member of struct rdt_hw_resource.
444 */
445 #define for_each_rdt_resource(r) \
446 for (r = &rdt_resources_all[0].r_resctrl; \
447 r <= &rdt_resources_all[RDT_NUM_RESOURCES - 1].r_resctrl; \
448 r = resctrl_inc(r))
449
450 #define for_each_capable_rdt_resource(r) \
451 for_each_rdt_resource(r) \
452 if (r->alloc_capable || r->mon_capable)
453
454 #define for_each_alloc_capable_rdt_resource(r) \
455 for_each_rdt_resource(r) \
456 if (r->alloc_capable)
457
458 #define for_each_mon_capable_rdt_resource(r) \
459 for_each_rdt_resource(r) \
460 if (r->mon_capable)
461
462 #define for_each_alloc_enabled_rdt_resource(r) \
463 for_each_rdt_resource(r) \
464 if (r->alloc_enabled)
465
466 #define for_each_mon_enabled_rdt_resource(r) \
467 for_each_rdt_resource(r) \
468 if (r->mon_enabled)
469
470 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */
471 union cpuid_0x10_1_eax {
472 struct {
473 unsigned int cbm_len:5;
474 } split;
475 unsigned int full;
476 };
477
478 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */
479 union cpuid_0x10_3_eax {
480 struct {
481 unsigned int max_delay:12;
482 } split;
483 unsigned int full;
484 };
485
486 /* CPUID.(EAX=10H, ECX=ResID).EDX */
487 union cpuid_0x10_x_edx {
488 struct {
489 unsigned int cos_max:16;
490 } split;
491 unsigned int full;
492 };
493
494 void rdt_last_cmd_clear(void);
495 void rdt_last_cmd_puts(const char *s);
496 __printf(1, 2)
497 void rdt_last_cmd_printf(const char *fmt, ...);
498
499 void rdt_ctrl_update(void *arg);
500 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
501 void rdtgroup_kn_unlock(struct kernfs_node *kn);
502 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
503 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
504 umode_t mask);
505 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
506 struct list_head **pos);
507 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
508 char *buf, size_t nbytes, loff_t off);
509 int rdtgroup_schemata_show(struct kernfs_open_file *of,
510 struct seq_file *s, void *v);
511 bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_domain *d,
512 unsigned long cbm, int closid, bool exclusive);
513 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
514 unsigned long cbm);
515 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
516 int rdtgroup_tasks_assigned(struct rdtgroup *r);
517 int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
518 int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
519 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
520 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
521 int rdt_pseudo_lock_init(void);
522 void rdt_pseudo_lock_release(void);
523 int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
524 void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
525 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
526 int closids_supported(void);
527 void closid_free(int closid);
528 int alloc_rmid(void);
529 void free_rmid(u32 rmid);
530 int rdt_get_mon_l3_config(struct rdt_resource *r);
531 void mon_event_count(void *info);
532 int rdtgroup_mondata_show(struct seq_file *m, void *arg);
533 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
534 unsigned int dom_id);
535 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
536 struct rdt_domain *d);
537 void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
538 struct rdt_domain *d, struct rdtgroup *rdtgrp,
539 int evtid, int first);
540 void mbm_setup_overflow_handler(struct rdt_domain *dom,
541 unsigned long delay_ms);
542 void mbm_handle_overflow(struct work_struct *work);
543 void __init intel_rdt_mbm_apply_quirk(void);
544 bool is_mba_sc(struct rdt_resource *r);
545 void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm);
546 u32 delay_bw_map(unsigned long bw, struct rdt_resource *r);
547 void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
548 void cqm_handle_limbo(struct work_struct *work);
549 bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
550 void __check_limbo(struct rdt_domain *d, bool force_free);
551 void rdt_domain_reconfigure_cdp(struct rdt_resource *r);
552 void __init thread_throttle_mode_init(void);
553
554 #endif /* _ASM_X86_RESCTRL_INTERNAL_H */
555