xref: /linux/fs/resctrl/internal.h (revision ab0308aee3819a3eccde42f9eb5bb01d6733be38)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _FS_RESCTRL_INTERNAL_H
3 #define _FS_RESCTRL_INTERNAL_H
4 
5 #include <linux/resctrl.h>
6 #include <linux/kernfs.h>
7 #include <linux/fs_context.h>
8 #include <linux/tick.h>
9 
10 #define CQM_LIMBOCHECK_INTERVAL	1000
11 
12 /**
13  * cpumask_any_housekeeping() - Choose any CPU in @mask, preferring those that
14  *			        aren't marked nohz_full
15  * @mask:	The mask to pick a CPU from.
16  * @exclude_cpu:The CPU to avoid picking.
17  *
18  * Returns a CPU from @mask, but not @exclude_cpu. If there are housekeeping
19  * CPUs that don't use nohz_full, these are preferred. Pass
20  * RESCTRL_PICK_ANY_CPU to avoid excluding any CPUs.
21  *
22  * When a CPU is excluded, returns >= nr_cpu_ids if no CPUs are available.
23  */
24 static inline unsigned int
25 cpumask_any_housekeeping(const struct cpumask *mask, int exclude_cpu)
26 {
27 	unsigned int cpu;
28 
29 	/* Try to find a CPU that isn't nohz_full to use in preference */
30 	if (tick_nohz_full_enabled()) {
31 		cpu = cpumask_any_andnot_but(mask, tick_nohz_full_mask, exclude_cpu);
32 		if (cpu < nr_cpu_ids)
33 			return cpu;
34 	}
35 
36 	return cpumask_any_but(mask, exclude_cpu);
37 }
38 
39 struct rdt_fs_context {
40 	struct kernfs_fs_context	kfc;
41 	bool				enable_cdpl2;
42 	bool				enable_cdpl3;
43 	bool				enable_mba_mbps;
44 	bool				enable_debug;
45 };
46 
47 static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
48 {
49 	struct kernfs_fs_context *kfc = fc->fs_private;
50 
51 	return container_of(kfc, struct rdt_fs_context, kfc);
52 }
53 
54 /**
55  * struct mon_evt - Properties of a monitor event
56  * @evtid:		event id
57  * @rid:		resource id for this event
58  * @name:		name of the event
59  * @evt_cfg:		Event configuration value that represents the
60  *			memory transactions (e.g., READS_TO_LOCAL_MEM,
61  *			READS_TO_REMOTE_MEM) being tracked by @evtid.
62  *			Only valid if @evtid is an MBM event.
63  * @configurable:	true if the event is configurable
64  * @any_cpu:		true if the event can be read from any CPU
65  * @enabled:		true if the event is enabled
66  */
67 struct mon_evt {
68 	enum resctrl_event_id	evtid;
69 	enum resctrl_res_level	rid;
70 	char			*name;
71 	u32			evt_cfg;
72 	bool			configurable;
73 	bool			any_cpu;
74 	bool			enabled;
75 };
76 
77 extern struct mon_evt mon_event_all[QOS_NUM_EVENTS];
78 
79 #define for_each_mon_event(mevt) for (mevt = &mon_event_all[QOS_FIRST_EVENT];	\
80 				      mevt < &mon_event_all[QOS_NUM_EVENTS]; mevt++)
81 
82 /**
83  * struct mon_data - Monitoring details for each event file.
84  * @list:            Member of the global @mon_data_kn_priv_list list.
85  * @rid:             Resource id associated with the event file.
86  * @evt:             Event structure associated with the event file.
87  * @sum:             Set when event must be summed across multiple
88  *                   domains.
89  * @domid:           When @sum is zero this is the domain to which
90  *                   the event file belongs. When @sum is one this
91  *                   is the id of the L3 cache that all domains to be
92  *                   summed share.
93  *
94  * Pointed to by the kernfs kn->priv field of monitoring event files.
95  * Readers and writers must hold rdtgroup_mutex.
96  */
97 struct mon_data {
98 	struct list_head	list;
99 	enum resctrl_res_level	rid;
100 	struct mon_evt		*evt;
101 	int			domid;
102 	bool			sum;
103 };
104 
105 /**
106  * struct rmid_read - Data passed across smp_call*() to read event count.
107  * @rgrp:  Resource group for which the counter is being read. If it is a parent
108  *	   resource group then its event count is summed with the count from all
109  *	   its child resource groups.
110  * @r:	   Resource describing the properties of the event being read.
111  * @hdr:   Header of domain that the counter should be read from. If NULL then
112  *	   sum all domains in @r sharing L3 @ci.id
113  * @evt:   Which monitor event to read.
114  * @first: Initialize MBM counter when true.
115  * @ci:    Cacheinfo for L3. Only set when @hdr is NULL. Used when summing
116  *	   domains.
117  * @is_mbm_cntr: true if "mbm_event" counter assignment mode is enabled and it
118  *	   is an MBM event.
119  * @err:   Error encountered when reading counter.
120  * @val:   Returned value of event counter. If @rgrp is a parent resource
121  *	   group, @val includes the sum of event counts from its child
122  *	   resource groups.  If @hdr is NULL, @val includes the sum of all
123  *	   domains in @r sharing @ci.id, (summed across child resource groups
124  *	   if @rgrp is a parent resource group).
125  * @arch_mon_ctx: Hardware monitor allocated for this read request (MPAM only).
126  */
127 struct rmid_read {
128 	struct rdtgroup		*rgrp;
129 	struct rdt_resource	*r;
130 	struct rdt_domain_hdr	*hdr;
131 	struct mon_evt		*evt;
132 	bool			first;
133 	struct cacheinfo	*ci;
134 	bool			is_mbm_cntr;
135 	int			err;
136 	u64			val;
137 	void			*arch_mon_ctx;
138 };
139 
140 extern struct list_head resctrl_schema_all;
141 
142 extern bool resctrl_mounted;
143 
144 enum rdt_group_type {
145 	RDTCTRL_GROUP = 0,
146 	RDTMON_GROUP,
147 	RDT_NUM_GROUP,
148 };
149 
150 /**
151  * enum rdtgrp_mode - Mode of a RDT resource group
152  * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
153  * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
154  * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
155  * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
156  *                          allowed AND the allocations are Cache Pseudo-Locked
157  * @RDT_NUM_MODES: Total number of modes
158  *
159  * The mode of a resource group enables control over the allowed overlap
160  * between allocations associated with different resource groups (classes
161  * of service). User is able to modify the mode of a resource group by
162  * writing to the "mode" resctrl file associated with the resource group.
163  *
164  * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
165  * writing the appropriate text to the "mode" file. A resource group enters
166  * "pseudo-locked" mode after the schemata is written while the resource
167  * group is in "pseudo-locksetup" mode.
168  */
169 enum rdtgrp_mode {
170 	RDT_MODE_SHAREABLE = 0,
171 	RDT_MODE_EXCLUSIVE,
172 	RDT_MODE_PSEUDO_LOCKSETUP,
173 	RDT_MODE_PSEUDO_LOCKED,
174 
175 	/* Must be last */
176 	RDT_NUM_MODES,
177 };
178 
179 /**
180  * struct mongroup - store mon group's data in resctrl fs.
181  * @mon_data_kn:		kernfs node for the mon_data directory
182  * @parent:			parent rdtgrp
183  * @crdtgrp_list:		child rdtgroup node list
184  * @rmid:			rmid for this rdtgroup
185  */
186 struct mongroup {
187 	struct kernfs_node	*mon_data_kn;
188 	struct rdtgroup		*parent;
189 	struct list_head	crdtgrp_list;
190 	u32			rmid;
191 };
192 
193 /**
194  * struct rdtgroup - store rdtgroup's data in resctrl file system.
195  * @kn:				kernfs node
196  * @rdtgroup_list:		linked list for all rdtgroups
197  * @closid:			closid for this rdtgroup
198  * @cpu_mask:			CPUs assigned to this rdtgroup
199  * @flags:			status bits
200  * @waitcount:			how many cpus expect to find this
201  *				group when they acquire rdtgroup_mutex
202  * @type:			indicates type of this rdtgroup - either
203  *				monitor only or ctrl_mon group
204  * @mon:			mongroup related data
205  * @mode:			mode of resource group
206  * @mba_mbps_event:		input monitoring event id when mba_sc is enabled
207  * @plr:			pseudo-locked region
208  */
209 struct rdtgroup {
210 	struct kernfs_node		*kn;
211 	struct list_head		rdtgroup_list;
212 	u32				closid;
213 	struct cpumask			cpu_mask;
214 	int				flags;
215 	atomic_t			waitcount;
216 	enum rdt_group_type		type;
217 	struct mongroup			mon;
218 	enum rdtgrp_mode		mode;
219 	enum resctrl_event_id		mba_mbps_event;
220 	struct pseudo_lock_region	*plr;
221 };
222 
223 /* rdtgroup.flags */
224 #define	RDT_DELETED		1
225 
226 /* rftype.flags */
227 #define RFTYPE_FLAGS_CPUS_LIST	1
228 
229 /*
230  * Define the file type flags for base and info directories.
231  */
232 #define RFTYPE_INFO			BIT(0)
233 
234 #define RFTYPE_BASE			BIT(1)
235 
236 #define RFTYPE_CTRL			BIT(4)
237 
238 #define RFTYPE_MON			BIT(5)
239 
240 #define RFTYPE_TOP			BIT(6)
241 
242 #define RFTYPE_RES_CACHE		BIT(8)
243 
244 #define RFTYPE_RES_MB			BIT(9)
245 
246 #define RFTYPE_DEBUG			BIT(10)
247 
248 #define RFTYPE_ASSIGN_CONFIG		BIT(11)
249 
250 #define RFTYPE_CTRL_INFO		(RFTYPE_INFO | RFTYPE_CTRL)
251 
252 #define RFTYPE_MON_INFO			(RFTYPE_INFO | RFTYPE_MON)
253 
254 #define RFTYPE_TOP_INFO			(RFTYPE_INFO | RFTYPE_TOP)
255 
256 #define RFTYPE_CTRL_BASE		(RFTYPE_BASE | RFTYPE_CTRL)
257 
258 #define RFTYPE_MON_BASE			(RFTYPE_BASE | RFTYPE_MON)
259 
260 /* List of all resource groups */
261 extern struct list_head rdt_all_groups;
262 
263 extern int max_name_width;
264 
265 /**
266  * struct rftype - describe each file in the resctrl file system
267  * @name:	File name
268  * @mode:	Access mode
269  * @kf_ops:	File operations
270  * @flags:	File specific RFTYPE_FLAGS_* flags
271  * @fflags:	File specific RFTYPE_* flags
272  * @seq_show:	Show content of the file
273  * @write:	Write to the file
274  */
275 struct rftype {
276 	char			*name;
277 	umode_t			mode;
278 	const struct kernfs_ops	*kf_ops;
279 	unsigned long		flags;
280 	unsigned long		fflags;
281 
282 	int (*seq_show)(struct kernfs_open_file *of,
283 			struct seq_file *sf, void *v);
284 	/*
285 	 * write() is the generic write callback which maps directly to
286 	 * kernfs write operation and overrides all other operations.
287 	 * Maximum write size is determined by ->max_write_len.
288 	 */
289 	ssize_t (*write)(struct kernfs_open_file *of,
290 			 char *buf, size_t nbytes, loff_t off);
291 };
292 
293 /**
294  * struct mbm_state - status for each MBM counter in each domain
295  * @prev_bw_bytes: Previous bytes value read for bandwidth calculation
296  * @prev_bw:	The most recent bandwidth in MBps
297  */
298 struct mbm_state {
299 	u64	prev_bw_bytes;
300 	u32	prev_bw;
301 };
302 
303 extern struct mutex rdtgroup_mutex;
304 
305 static inline const char *rdt_kn_name(const struct kernfs_node *kn)
306 {
307 	return rcu_dereference_check(kn->name, lockdep_is_held(&rdtgroup_mutex));
308 }
309 
310 extern struct rdtgroup rdtgroup_default;
311 
312 extern struct dentry *debugfs_resctrl;
313 
314 extern enum resctrl_event_id mba_mbps_default_event;
315 
316 void rdt_last_cmd_clear(void);
317 
318 void rdt_last_cmd_puts(const char *s);
319 
320 __printf(1, 2)
321 void rdt_last_cmd_printf(const char *fmt, ...);
322 
323 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
324 
325 void rdtgroup_kn_unlock(struct kernfs_node *kn);
326 
327 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
328 
329 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
330 			     umode_t mask);
331 
332 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
333 				char *buf, size_t nbytes, loff_t off);
334 
335 int rdtgroup_schemata_show(struct kernfs_open_file *of,
336 			   struct seq_file *s, void *v);
337 
338 ssize_t rdtgroup_mba_mbps_event_write(struct kernfs_open_file *of,
339 				      char *buf, size_t nbytes, loff_t off);
340 
341 int rdtgroup_mba_mbps_event_show(struct kernfs_open_file *of,
342 				 struct seq_file *s, void *v);
343 
344 bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_ctrl_domain *d,
345 			   unsigned long cbm, int closid, bool exclusive);
346 
347 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_ctrl_domain *d,
348 				  unsigned long cbm);
349 
350 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
351 
352 int rdtgroup_tasks_assigned(struct rdtgroup *r);
353 
354 int closids_supported(void);
355 
356 void closid_free(int closid);
357 
358 int alloc_rmid(u32 closid);
359 
360 void free_rmid(u32 closid, u32 rmid);
361 
362 int resctrl_l3_mon_resource_init(void);
363 
364 void resctrl_l3_mon_resource_exit(void);
365 
366 void mon_event_count(void *info);
367 
368 int rdtgroup_mondata_show(struct seq_file *m, void *arg);
369 
370 void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
371 		    struct rdt_domain_hdr *hdr, struct rdtgroup *rdtgrp,
372 		    cpumask_t *cpumask, struct mon_evt *evt, int first);
373 
374 void mbm_setup_overflow_handler(struct rdt_l3_mon_domain *dom,
375 				unsigned long delay_ms,
376 				int exclude_cpu);
377 
378 void mbm_handle_overflow(struct work_struct *work);
379 
380 bool is_mba_sc(struct rdt_resource *r);
381 
382 void cqm_setup_limbo_handler(struct rdt_l3_mon_domain *dom, unsigned long delay_ms,
383 			     int exclude_cpu);
384 
385 void cqm_handle_limbo(struct work_struct *work);
386 
387 bool has_busy_rmid(struct rdt_l3_mon_domain *d);
388 
389 void __check_limbo(struct rdt_l3_mon_domain *d, bool force_free);
390 
391 void resctrl_file_fflags_init(const char *config, unsigned long fflags);
392 
393 void rdt_staged_configs_clear(void);
394 
395 bool closid_allocated(unsigned int closid);
396 
397 bool closid_alloc_fixed(u32 closid);
398 
399 int resctrl_find_cleanest_closid(void);
400 
401 void *rdt_kn_parent_priv(struct kernfs_node *kn);
402 
403 int resctrl_mbm_assign_mode_show(struct kernfs_open_file *of, struct seq_file *s, void *v);
404 
405 ssize_t resctrl_mbm_assign_mode_write(struct kernfs_open_file *of, char *buf,
406 				      size_t nbytes, loff_t off);
407 
408 void resctrl_bmec_files_show(struct rdt_resource *r, struct kernfs_node *l3_mon_kn,
409 			     bool show);
410 
411 int resctrl_num_mbm_cntrs_show(struct kernfs_open_file *of, struct seq_file *s, void *v);
412 
413 int resctrl_available_mbm_cntrs_show(struct kernfs_open_file *of, struct seq_file *s,
414 				     void *v);
415 
416 void rdtgroup_assign_cntrs(struct rdtgroup *rdtgrp);
417 
418 void rdtgroup_unassign_cntrs(struct rdtgroup *rdtgrp);
419 
420 int event_filter_show(struct kernfs_open_file *of, struct seq_file *seq, void *v);
421 
422 ssize_t event_filter_write(struct kernfs_open_file *of, char *buf, size_t nbytes,
423 			   loff_t off);
424 
425 int resctrl_mbm_assign_on_mkdir_show(struct kernfs_open_file *of,
426 				     struct seq_file *s, void *v);
427 
428 ssize_t resctrl_mbm_assign_on_mkdir_write(struct kernfs_open_file *of, char *buf,
429 					  size_t nbytes, loff_t off);
430 
431 int mbm_L3_assignments_show(struct kernfs_open_file *of, struct seq_file *s, void *v);
432 
433 ssize_t mbm_L3_assignments_write(struct kernfs_open_file *of, char *buf, size_t nbytes,
434 				 loff_t off);
435 int resctrl_io_alloc_show(struct kernfs_open_file *of, struct seq_file *seq, void *v);
436 
437 int rdtgroup_init_cat(struct resctrl_schema *s, u32 closid);
438 
439 enum resctrl_conf_type resctrl_peer_type(enum resctrl_conf_type my_type);
440 
441 ssize_t resctrl_io_alloc_write(struct kernfs_open_file *of, char *buf,
442 			       size_t nbytes, loff_t off);
443 
444 const char *rdtgroup_name_by_closid(u32 closid);
445 int resctrl_io_alloc_cbm_show(struct kernfs_open_file *of, struct seq_file *seq,
446 			      void *v);
447 ssize_t resctrl_io_alloc_cbm_write(struct kernfs_open_file *of, char *buf,
448 				   size_t nbytes, loff_t off);
449 u32 resctrl_io_alloc_closid(struct rdt_resource *r);
450 
451 #ifdef CONFIG_RESCTRL_FS_PSEUDO_LOCK
452 int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
453 
454 int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
455 
456 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_ctrl_domain *d, unsigned long cbm);
457 
458 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_ctrl_domain *d);
459 
460 int rdt_pseudo_lock_init(void);
461 
462 void rdt_pseudo_lock_release(void);
463 
464 int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
465 
466 void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
467 
468 #else
469 static inline int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp)
470 {
471 	return -EOPNOTSUPP;
472 }
473 
474 static inline int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp)
475 {
476 	return -EOPNOTSUPP;
477 }
478 
479 static inline bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_ctrl_domain *d, unsigned long cbm)
480 {
481 	return false;
482 }
483 
484 static inline bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_ctrl_domain *d)
485 {
486 	return false;
487 }
488 
489 static inline int rdt_pseudo_lock_init(void) { return 0; }
490 static inline void rdt_pseudo_lock_release(void) { }
491 static inline int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp)
492 {
493 	return -EOPNOTSUPP;
494 }
495 
496 static inline void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp) { }
497 #endif /* CONFIG_RESCTRL_FS_PSEUDO_LOCK */
498 
499 #endif /* _FS_RESCTRL_INTERNAL_H */
500