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 #include <linux/tick.h>
11
12 #include <asm/resctrl.h>
13
14 #define L3_QOS_CDP_ENABLE 0x01ULL
15
16 #define L2_QOS_CDP_ENABLE 0x01ULL
17
18 #define CQM_LIMBOCHECK_INTERVAL 1000
19
20 #define MBM_CNTR_WIDTH_BASE 24
21 #define MBM_OVERFLOW_INTERVAL 1000
22 #define MAX_MBA_BW 100u
23 #define MBA_IS_LINEAR 0x4
24 #define MBM_CNTR_WIDTH_OFFSET_AMD 20
25
26 #define RMID_VAL_ERROR BIT_ULL(63)
27 #define RMID_VAL_UNAVAIL BIT_ULL(62)
28 /*
29 * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for
30 * data to be returned. The counter width is discovered from the hardware
31 * as an offset from MBM_CNTR_WIDTH_BASE.
32 */
33 #define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
34
35 /**
36 * cpumask_any_housekeeping() - Choose any CPU in @mask, preferring those that
37 * aren't marked nohz_full
38 * @mask: The mask to pick a CPU from.
39 * @exclude_cpu:The CPU to avoid picking.
40 *
41 * Returns a CPU from @mask, but not @exclude_cpu. If there are housekeeping
42 * CPUs that don't use nohz_full, these are preferred. Pass
43 * RESCTRL_PICK_ANY_CPU to avoid excluding any CPUs.
44 *
45 * When a CPU is excluded, returns >= nr_cpu_ids if no CPUs are available.
46 */
47 static inline unsigned int
cpumask_any_housekeeping(const struct cpumask * mask,int exclude_cpu)48 cpumask_any_housekeeping(const struct cpumask *mask, int exclude_cpu)
49 {
50 unsigned int cpu, hk_cpu;
51
52 if (exclude_cpu == RESCTRL_PICK_ANY_CPU)
53 cpu = cpumask_any(mask);
54 else
55 cpu = cpumask_any_but(mask, exclude_cpu);
56
57 /* Only continue if tick_nohz_full_mask has been initialized. */
58 if (!tick_nohz_full_enabled())
59 return cpu;
60
61 /* If the CPU picked isn't marked nohz_full nothing more needs doing. */
62 if (cpu < nr_cpu_ids && !tick_nohz_full_cpu(cpu))
63 return cpu;
64
65 /* Try to find a CPU that isn't nohz_full to use in preference */
66 hk_cpu = cpumask_nth_andnot(0, mask, tick_nohz_full_mask);
67 if (hk_cpu == exclude_cpu)
68 hk_cpu = cpumask_nth_andnot(1, mask, tick_nohz_full_mask);
69
70 if (hk_cpu < nr_cpu_ids)
71 cpu = hk_cpu;
72
73 return cpu;
74 }
75
76 struct rdt_fs_context {
77 struct kernfs_fs_context kfc;
78 bool enable_cdpl2;
79 bool enable_cdpl3;
80 bool enable_mba_mbps;
81 bool enable_debug;
82 };
83
rdt_fc2context(struct fs_context * fc)84 static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
85 {
86 struct kernfs_fs_context *kfc = fc->fs_private;
87
88 return container_of(kfc, struct rdt_fs_context, kfc);
89 }
90
91 /**
92 * struct mon_evt - Entry in the event list of a resource
93 * @evtid: event id
94 * @name: name of the event
95 * @configurable: true if the event is configurable
96 * @list: entry in &rdt_resource->evt_list
97 */
98 struct mon_evt {
99 enum resctrl_event_id evtid;
100 char *name;
101 bool configurable;
102 struct list_head list;
103 };
104
105 /**
106 * union mon_data_bits - Monitoring details for each event file.
107 * @priv: Used to store monitoring event data in @u
108 * as kernfs private data.
109 * @u.rid: Resource id associated with the event file.
110 * @u.evtid: Event id associated with the event file.
111 * @u.sum: Set when event must be summed across multiple
112 * domains.
113 * @u.domid: When @u.sum is zero this is the domain to which
114 * the event file belongs. When @sum is one this
115 * is the id of the L3 cache that all domains to be
116 * summed share.
117 * @u: Name of the bit fields struct.
118 */
119 union mon_data_bits {
120 void *priv;
121 struct {
122 unsigned int rid : 10;
123 enum resctrl_event_id evtid : 7;
124 unsigned int sum : 1;
125 unsigned int domid : 14;
126 } u;
127 };
128
129 /**
130 * struct rmid_read - Data passed across smp_call*() to read event count.
131 * @rgrp: Resource group for which the counter is being read. If it is a parent
132 * resource group then its event count is summed with the count from all
133 * its child resource groups.
134 * @r: Resource describing the properties of the event being read.
135 * @d: Domain that the counter should be read from. If NULL then sum all
136 * domains in @r sharing L3 @ci.id
137 * @evtid: Which monitor event to read.
138 * @first: Initialize MBM counter when true.
139 * @ci: Cacheinfo for L3. Only set when @d is NULL. Used when summing domains.
140 * @err: Error encountered when reading counter.
141 * @val: Returned value of event counter. If @rgrp is a parent resource group,
142 * @val includes the sum of event counts from its child resource groups.
143 * If @d is NULL, @val includes the sum of all domains in @r sharing @ci.id,
144 * (summed across child resource groups if @rgrp is a parent resource group).
145 * @arch_mon_ctx: Hardware monitor allocated for this read request (MPAM only).
146 */
147 struct rmid_read {
148 struct rdtgroup *rgrp;
149 struct rdt_resource *r;
150 struct rdt_mon_domain *d;
151 enum resctrl_event_id evtid;
152 bool first;
153 struct cacheinfo *ci;
154 int err;
155 u64 val;
156 void *arch_mon_ctx;
157 };
158
159 extern struct list_head resctrl_schema_all;
160 extern bool resctrl_mounted;
161
162 enum rdt_group_type {
163 RDTCTRL_GROUP = 0,
164 RDTMON_GROUP,
165 RDT_NUM_GROUP,
166 };
167
168 /**
169 * enum rdtgrp_mode - Mode of a RDT resource group
170 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
171 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
172 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
173 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
174 * allowed AND the allocations are Cache Pseudo-Locked
175 * @RDT_NUM_MODES: Total number of modes
176 *
177 * The mode of a resource group enables control over the allowed overlap
178 * between allocations associated with different resource groups (classes
179 * of service). User is able to modify the mode of a resource group by
180 * writing to the "mode" resctrl file associated with the resource group.
181 *
182 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
183 * writing the appropriate text to the "mode" file. A resource group enters
184 * "pseudo-locked" mode after the schemata is written while the resource
185 * group is in "pseudo-locksetup" mode.
186 */
187 enum rdtgrp_mode {
188 RDT_MODE_SHAREABLE = 0,
189 RDT_MODE_EXCLUSIVE,
190 RDT_MODE_PSEUDO_LOCKSETUP,
191 RDT_MODE_PSEUDO_LOCKED,
192
193 /* Must be last */
194 RDT_NUM_MODES,
195 };
196
197 /**
198 * struct mongroup - store mon group's data in resctrl fs.
199 * @mon_data_kn: kernfs node for the mon_data directory
200 * @parent: parent rdtgrp
201 * @crdtgrp_list: child rdtgroup node list
202 * @rmid: rmid for this rdtgroup
203 */
204 struct mongroup {
205 struct kernfs_node *mon_data_kn;
206 struct rdtgroup *parent;
207 struct list_head crdtgrp_list;
208 u32 rmid;
209 };
210
211 /**
212 * struct rdtgroup - store rdtgroup's data in resctrl file system.
213 * @kn: kernfs node
214 * @rdtgroup_list: linked list for all rdtgroups
215 * @closid: closid for this rdtgroup
216 * @cpu_mask: CPUs assigned to this rdtgroup
217 * @flags: status bits
218 * @waitcount: how many cpus expect to find this
219 * group when they acquire rdtgroup_mutex
220 * @type: indicates type of this rdtgroup - either
221 * monitor only or ctrl_mon group
222 * @mon: mongroup related data
223 * @mode: mode of resource group
224 * @mba_mbps_event: input monitoring event id when mba_sc is enabled
225 * @plr: pseudo-locked region
226 */
227 struct rdtgroup {
228 struct kernfs_node *kn;
229 struct list_head rdtgroup_list;
230 u32 closid;
231 struct cpumask cpu_mask;
232 int flags;
233 atomic_t waitcount;
234 enum rdt_group_type type;
235 struct mongroup mon;
236 enum rdtgrp_mode mode;
237 enum resctrl_event_id mba_mbps_event;
238 struct pseudo_lock_region *plr;
239 };
240
241 /* rdtgroup.flags */
242 #define RDT_DELETED 1
243
244 /* rftype.flags */
245 #define RFTYPE_FLAGS_CPUS_LIST 1
246
247 /*
248 * Define the file type flags for base and info directories.
249 */
250 #define RFTYPE_INFO BIT(0)
251 #define RFTYPE_BASE BIT(1)
252 #define RFTYPE_CTRL BIT(4)
253 #define RFTYPE_MON BIT(5)
254 #define RFTYPE_TOP BIT(6)
255 #define RFTYPE_RES_CACHE BIT(8)
256 #define RFTYPE_RES_MB BIT(9)
257 #define RFTYPE_DEBUG BIT(10)
258 #define RFTYPE_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
259 #define RFTYPE_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
260 #define RFTYPE_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
261 #define RFTYPE_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
262 #define RFTYPE_MON_BASE (RFTYPE_BASE | RFTYPE_MON)
263
264 /* List of all resource groups */
265 extern struct list_head rdt_all_groups;
266
267 extern int max_name_width;
268
269 /**
270 * struct rftype - describe each file in the resctrl file system
271 * @name: File name
272 * @mode: Access mode
273 * @kf_ops: File operations
274 * @flags: File specific RFTYPE_FLAGS_* flags
275 * @fflags: File specific RFTYPE_* flags
276 * @seq_show: Show content of the file
277 * @write: Write to the file
278 */
279 struct rftype {
280 char *name;
281 umode_t mode;
282 const struct kernfs_ops *kf_ops;
283 unsigned long flags;
284 unsigned long fflags;
285
286 int (*seq_show)(struct kernfs_open_file *of,
287 struct seq_file *sf, void *v);
288 /*
289 * write() is the generic write callback which maps directly to
290 * kernfs write operation and overrides all other operations.
291 * Maximum write size is determined by ->max_write_len.
292 */
293 ssize_t (*write)(struct kernfs_open_file *of,
294 char *buf, size_t nbytes, loff_t off);
295 };
296
297 /**
298 * struct mbm_state - status for each MBM counter in each domain
299 * @prev_bw_bytes: Previous bytes value read for bandwidth calculation
300 * @prev_bw: The most recent bandwidth in MBps
301 */
302 struct mbm_state {
303 u64 prev_bw_bytes;
304 u32 prev_bw;
305 };
306
307 /**
308 * struct arch_mbm_state - values used to compute resctrl_arch_rmid_read()s
309 * return value.
310 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
311 * @prev_msr: Value of IA32_QM_CTR last time it was read for the RMID used to
312 * find this struct.
313 */
314 struct arch_mbm_state {
315 u64 chunks;
316 u64 prev_msr;
317 };
318
319 /**
320 * struct rdt_hw_ctrl_domain - Arch private attributes of a set of CPUs that share
321 * a resource for a control function
322 * @d_resctrl: Properties exposed to the resctrl file system
323 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
324 *
325 * Members of this structure are accessed via helpers that provide abstraction.
326 */
327 struct rdt_hw_ctrl_domain {
328 struct rdt_ctrl_domain d_resctrl;
329 u32 *ctrl_val;
330 };
331
332 /**
333 * struct rdt_hw_mon_domain - Arch private attributes of a set of CPUs that share
334 * a resource for a monitor function
335 * @d_resctrl: Properties exposed to the resctrl file system
336 * @arch_mbm_total: arch private state for MBM total bandwidth
337 * @arch_mbm_local: arch private state for MBM local bandwidth
338 *
339 * Members of this structure are accessed via helpers that provide abstraction.
340 */
341 struct rdt_hw_mon_domain {
342 struct rdt_mon_domain d_resctrl;
343 struct arch_mbm_state *arch_mbm_total;
344 struct arch_mbm_state *arch_mbm_local;
345 };
346
resctrl_to_arch_ctrl_dom(struct rdt_ctrl_domain * r)347 static inline struct rdt_hw_ctrl_domain *resctrl_to_arch_ctrl_dom(struct rdt_ctrl_domain *r)
348 {
349 return container_of(r, struct rdt_hw_ctrl_domain, d_resctrl);
350 }
351
resctrl_to_arch_mon_dom(struct rdt_mon_domain * r)352 static inline struct rdt_hw_mon_domain *resctrl_to_arch_mon_dom(struct rdt_mon_domain *r)
353 {
354 return container_of(r, struct rdt_hw_mon_domain, d_resctrl);
355 }
356
357 /**
358 * struct msr_param - set a range of MSRs from a domain
359 * @res: The resource to use
360 * @dom: The domain to update
361 * @low: Beginning index from base MSR
362 * @high: End index
363 */
364 struct msr_param {
365 struct rdt_resource *res;
366 struct rdt_ctrl_domain *dom;
367 u32 low;
368 u32 high;
369 };
370
371 /**
372 * struct rdt_hw_resource - arch private attributes of a resctrl resource
373 * @r_resctrl: Attributes of the resource used directly by resctrl.
374 * @num_closid: Maximum number of closid this hardware can support,
375 * regardless of CDP. This is exposed via
376 * resctrl_arch_get_num_closid() to avoid confusion
377 * with struct resctrl_schema's property of the same name,
378 * which has been corrected for features like CDP.
379 * @msr_base: Base MSR address for CBMs
380 * @msr_update: Function pointer to update QOS MSRs
381 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
382 * @mbm_width: Monitor width, to detect and correct for overflow.
383 * @cdp_enabled: CDP state of this resource
384 *
385 * Members of this structure are either private to the architecture
386 * e.g. mbm_width, or accessed via helpers that provide abstraction. e.g.
387 * msr_update and msr_base.
388 */
389 struct rdt_hw_resource {
390 struct rdt_resource r_resctrl;
391 u32 num_closid;
392 unsigned int msr_base;
393 void (*msr_update)(struct msr_param *m);
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 extern struct mutex rdtgroup_mutex;
405
rdt_kn_name(const struct kernfs_node * kn)406 static inline const char *rdt_kn_name(const struct kernfs_node *kn)
407 {
408 return rcu_dereference_check(kn->name, lockdep_is_held(&rdtgroup_mutex));
409 }
410
411 extern struct rdt_hw_resource rdt_resources_all[];
412 extern struct rdtgroup rdtgroup_default;
413 extern struct dentry *debugfs_resctrl;
414 extern enum resctrl_event_id mba_mbps_default_event;
415
resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)416 static inline bool resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)
417 {
418 return rdt_resources_all[l].cdp_enabled;
419 }
420
421 int resctrl_arch_set_cdp_enabled(enum resctrl_res_level l, bool enable);
422
423 void arch_mon_domain_online(struct rdt_resource *r, struct rdt_mon_domain *d);
424
425 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */
426 union cpuid_0x10_1_eax {
427 struct {
428 unsigned int cbm_len:5;
429 } split;
430 unsigned int full;
431 };
432
433 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */
434 union cpuid_0x10_3_eax {
435 struct {
436 unsigned int max_delay:12;
437 } split;
438 unsigned int full;
439 };
440
441 /* CPUID.(EAX=10H, ECX=ResID).ECX */
442 union cpuid_0x10_x_ecx {
443 struct {
444 unsigned int reserved:3;
445 unsigned int noncont:1;
446 } split;
447 unsigned int full;
448 };
449
450 /* CPUID.(EAX=10H, ECX=ResID).EDX */
451 union cpuid_0x10_x_edx {
452 struct {
453 unsigned int cos_max:16;
454 } split;
455 unsigned int full;
456 };
457
458 void rdt_last_cmd_clear(void);
459 void rdt_last_cmd_puts(const char *s);
460 __printf(1, 2)
461 void rdt_last_cmd_printf(const char *fmt, ...);
462
463 void rdt_ctrl_update(void *arg);
464 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
465 void rdtgroup_kn_unlock(struct kernfs_node *kn);
466 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
467 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
468 umode_t mask);
469 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
470 char *buf, size_t nbytes, loff_t off);
471 int rdtgroup_schemata_show(struct kernfs_open_file *of,
472 struct seq_file *s, void *v);
473 ssize_t rdtgroup_mba_mbps_event_write(struct kernfs_open_file *of,
474 char *buf, size_t nbytes, loff_t off);
475 int rdtgroup_mba_mbps_event_show(struct kernfs_open_file *of,
476 struct seq_file *s, void *v);
477 bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_ctrl_domain *d,
478 unsigned long cbm, int closid, bool exclusive);
479 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_ctrl_domain *d,
480 unsigned long cbm);
481 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
482 int rdtgroup_tasks_assigned(struct rdtgroup *r);
483 int closids_supported(void);
484 void closid_free(int closid);
485 int alloc_rmid(u32 closid);
486 void free_rmid(u32 closid, u32 rmid);
487 int rdt_get_mon_l3_config(struct rdt_resource *r);
488 void resctrl_mon_resource_exit(void);
489 bool __init rdt_cpu_has(int flag);
490 void mon_event_count(void *info);
491 int rdtgroup_mondata_show(struct seq_file *m, void *arg);
492 void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
493 struct rdt_mon_domain *d, struct rdtgroup *rdtgrp,
494 cpumask_t *cpumask, int evtid, int first);
495 int __init resctrl_mon_resource_init(void);
496 void mbm_setup_overflow_handler(struct rdt_mon_domain *dom,
497 unsigned long delay_ms,
498 int exclude_cpu);
499 void mbm_handle_overflow(struct work_struct *work);
500 void __init intel_rdt_mbm_apply_quirk(void);
501 bool is_mba_sc(struct rdt_resource *r);
502 void cqm_setup_limbo_handler(struct rdt_mon_domain *dom, unsigned long delay_ms,
503 int exclude_cpu);
504 void cqm_handle_limbo(struct work_struct *work);
505 bool has_busy_rmid(struct rdt_mon_domain *d);
506 void __check_limbo(struct rdt_mon_domain *d, bool force_free);
507 void rdt_domain_reconfigure_cdp(struct rdt_resource *r);
508 void resctrl_file_fflags_init(const char *config, unsigned long fflags);
509 void rdt_staged_configs_clear(void);
510 bool closid_allocated(unsigned int closid);
511 int resctrl_find_cleanest_closid(void);
512
513 #ifdef CONFIG_RESCTRL_FS_PSEUDO_LOCK
514 int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
515 int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
516 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_ctrl_domain *d, unsigned long cbm);
517 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_ctrl_domain *d);
518 int rdt_pseudo_lock_init(void);
519 void rdt_pseudo_lock_release(void);
520 int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
521 void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
522 #else
rdtgroup_locksetup_enter(struct rdtgroup * rdtgrp)523 static inline int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp)
524 {
525 return -EOPNOTSUPP;
526 }
527
rdtgroup_locksetup_exit(struct rdtgroup * rdtgrp)528 static inline int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp)
529 {
530 return -EOPNOTSUPP;
531 }
532
rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_ctrl_domain * d,unsigned long cbm)533 static inline bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_ctrl_domain *d, unsigned long cbm)
534 {
535 return false;
536 }
537
rdtgroup_pseudo_locked_in_hierarchy(struct rdt_ctrl_domain * d)538 static inline bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_ctrl_domain *d)
539 {
540 return false;
541 }
542
rdt_pseudo_lock_init(void)543 static inline int rdt_pseudo_lock_init(void) { return 0; }
rdt_pseudo_lock_release(void)544 static inline void rdt_pseudo_lock_release(void) { }
rdtgroup_pseudo_lock_create(struct rdtgroup * rdtgrp)545 static inline int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp)
546 {
547 return -EOPNOTSUPP;
548 }
549
rdtgroup_pseudo_lock_remove(struct rdtgroup * rdtgrp)550 static inline void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp) { }
551 #endif /* CONFIG_RESCTRL_FS_PSEUDO_LOCK */
552
553 #endif /* _ASM_X86_RESCTRL_INTERNAL_H */
554