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 L3_QOS_CDP_ENABLE		0x01ULL
12 
13 #define L2_QOS_CDP_ENABLE		0x01ULL
14 
15 #define CQM_LIMBOCHECK_INTERVAL	1000
16 
17 #define MBM_CNTR_WIDTH_BASE		24
18 #define MBM_OVERFLOW_INTERVAL		1000
19 #define MAX_MBA_BW			100u
20 #define MBA_IS_LINEAR			0x4
21 #define MBM_CNTR_WIDTH_OFFSET_AMD	20
22 
23 #define RMID_VAL_ERROR			BIT_ULL(63)
24 #define RMID_VAL_UNAVAIL		BIT_ULL(62)
25 /*
26  * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for
27  * data to be returned. The counter width is discovered from the hardware
28  * as an offset from MBM_CNTR_WIDTH_BASE.
29  */
30 #define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
31 
32 /* Reads to Local DRAM Memory */
33 #define READS_TO_LOCAL_MEM		BIT(0)
34 
35 /* Reads to Remote DRAM Memory */
36 #define READS_TO_REMOTE_MEM		BIT(1)
37 
38 /* Non-Temporal Writes to Local Memory */
39 #define NON_TEMP_WRITE_TO_LOCAL_MEM	BIT(2)
40 
41 /* Non-Temporal Writes to Remote Memory */
42 #define NON_TEMP_WRITE_TO_REMOTE_MEM	BIT(3)
43 
44 /* Reads to Local Memory the system identifies as "Slow Memory" */
45 #define READS_TO_LOCAL_S_MEM		BIT(4)
46 
47 /* Reads to Remote Memory the system identifies as "Slow Memory" */
48 #define READS_TO_REMOTE_S_MEM		BIT(5)
49 
50 /* Dirty Victims to All Types of Memory */
51 #define DIRTY_VICTIMS_TO_ALL_MEM	BIT(6)
52 
53 /* Max event bits supported */
54 #define MAX_EVT_CONFIG_BITS		GENMASK(6, 0)
55 
56 struct rdt_fs_context {
57 	struct kernfs_fs_context	kfc;
58 	bool				enable_cdpl2;
59 	bool				enable_cdpl3;
60 	bool				enable_mba_mbps;
61 };
62 
rdt_fc2context(struct fs_context * fc)63 static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
64 {
65 	struct kernfs_fs_context *kfc = fc->fs_private;
66 
67 	return container_of(kfc, struct rdt_fs_context, kfc);
68 }
69 
70 DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
71 DECLARE_STATIC_KEY_FALSE(rdt_mon_enable_key);
72 
73 /**
74  * struct mon_evt - Entry in the event list of a resource
75  * @evtid:		event id
76  * @name:		name of the event
77  * @configurable:	true if the event is configurable
78  * @list:		entry in &rdt_resource->evt_list
79  */
80 struct mon_evt {
81 	enum resctrl_event_id	evtid;
82 	char			*name;
83 	bool			configurable;
84 	struct list_head	list;
85 };
86 
87 /**
88  * union mon_data_bits - Monitoring details for each event file
89  * @priv:              Used to store monitoring event data in @u
90  *                     as kernfs private data
91  * @rid:               Resource id associated with the event file
92  * @evtid:             Event id associated with the event file
93  * @domid:             The domain to which the event file belongs
94  * @u:                 Name of the bit fields struct
95  */
96 union mon_data_bits {
97 	void *priv;
98 	struct {
99 		unsigned int rid		: 10;
100 		enum resctrl_event_id evtid	: 8;
101 		unsigned int domid		: 14;
102 	} u;
103 };
104 
105 struct rmid_read {
106 	struct rdtgroup		*rgrp;
107 	struct rdt_resource	*r;
108 	struct rdt_domain	*d;
109 	enum resctrl_event_id	evtid;
110 	bool			first;
111 	int			err;
112 	u64			val;
113 };
114 
115 extern bool rdt_alloc_capable;
116 extern bool rdt_mon_capable;
117 extern unsigned int rdt_mon_features;
118 extern struct list_head resctrl_schema_all;
119 
120 enum rdt_group_type {
121 	RDTCTRL_GROUP = 0,
122 	RDTMON_GROUP,
123 	RDT_NUM_GROUP,
124 };
125 
126 /**
127  * enum rdtgrp_mode - Mode of a RDT resource group
128  * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
129  * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
130  * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
131  * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
132  *                          allowed AND the allocations are Cache Pseudo-Locked
133  * @RDT_NUM_MODES: Total number of modes
134  *
135  * The mode of a resource group enables control over the allowed overlap
136  * between allocations associated with different resource groups (classes
137  * of service). User is able to modify the mode of a resource group by
138  * writing to the "mode" resctrl file associated with the resource group.
139  *
140  * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
141  * writing the appropriate text to the "mode" file. A resource group enters
142  * "pseudo-locked" mode after the schemata is written while the resource
143  * group is in "pseudo-locksetup" mode.
144  */
145 enum rdtgrp_mode {
146 	RDT_MODE_SHAREABLE = 0,
147 	RDT_MODE_EXCLUSIVE,
148 	RDT_MODE_PSEUDO_LOCKSETUP,
149 	RDT_MODE_PSEUDO_LOCKED,
150 
151 	/* Must be last */
152 	RDT_NUM_MODES,
153 };
154 
155 /**
156  * struct mongroup - store mon group's data in resctrl fs.
157  * @mon_data_kn:		kernfs node for the mon_data directory
158  * @parent:			parent rdtgrp
159  * @crdtgrp_list:		child rdtgroup node list
160  * @rmid:			rmid for this rdtgroup
161  */
162 struct mongroup {
163 	struct kernfs_node	*mon_data_kn;
164 	struct rdtgroup		*parent;
165 	struct list_head	crdtgrp_list;
166 	u32			rmid;
167 };
168 
169 /**
170  * struct pseudo_lock_region - pseudo-lock region information
171  * @s:			Resctrl schema for the resource to which this
172  *			pseudo-locked region belongs
173  * @d:			RDT domain to which this pseudo-locked region
174  *			belongs
175  * @cbm:		bitmask of the pseudo-locked region
176  * @lock_thread_wq:	waitqueue used to wait on the pseudo-locking thread
177  *			completion
178  * @thread_done:	variable used by waitqueue to test if pseudo-locking
179  *			thread completed
180  * @cpu:		core associated with the cache on which the setup code
181  *			will be run
182  * @line_size:		size of the cache lines
183  * @size:		size of pseudo-locked region in bytes
184  * @kmem:		the kernel memory associated with pseudo-locked region
185  * @minor:		minor number of character device associated with this
186  *			region
187  * @debugfs_dir:	pointer to this region's directory in the debugfs
188  *			filesystem
189  * @pm_reqs:		Power management QoS requests related to this region
190  */
191 struct pseudo_lock_region {
192 	struct resctrl_schema	*s;
193 	struct rdt_domain	*d;
194 	u32			cbm;
195 	wait_queue_head_t	lock_thread_wq;
196 	int			thread_done;
197 	int			cpu;
198 	unsigned int		line_size;
199 	unsigned int		size;
200 	void			*kmem;
201 	unsigned int		minor;
202 	struct dentry		*debugfs_dir;
203 	struct list_head	pm_reqs;
204 };
205 
206 /**
207  * struct rdtgroup - store rdtgroup's data in resctrl file system.
208  * @kn:				kernfs node
209  * @rdtgroup_list:		linked list for all rdtgroups
210  * @closid:			closid for this rdtgroup
211  * @cpu_mask:			CPUs assigned to this rdtgroup
212  * @flags:			status bits
213  * @waitcount:			how many cpus expect to find this
214  *				group when they acquire rdtgroup_mutex
215  * @type:			indicates type of this rdtgroup - either
216  *				monitor only or ctrl_mon group
217  * @mon:			mongroup related data
218  * @mode:			mode of resource group
219  * @plr:			pseudo-locked region
220  */
221 struct rdtgroup {
222 	struct kernfs_node		*kn;
223 	struct list_head		rdtgroup_list;
224 	u32				closid;
225 	struct cpumask			cpu_mask;
226 	int				flags;
227 	atomic_t			waitcount;
228 	enum rdt_group_type		type;
229 	struct mongroup			mon;
230 	enum rdtgrp_mode		mode;
231 	struct pseudo_lock_region	*plr;
232 };
233 
234 /* rdtgroup.flags */
235 #define	RDT_DELETED		1
236 
237 /* rftype.flags */
238 #define RFTYPE_FLAGS_CPUS_LIST	1
239 
240 /*
241  * Define the file type flags for base and info directories.
242  */
243 #define RFTYPE_INFO			BIT(0)
244 #define RFTYPE_BASE			BIT(1)
245 #define RF_CTRLSHIFT			4
246 #define RF_MONSHIFT			5
247 #define RF_TOPSHIFT			6
248 #define RFTYPE_CTRL			BIT(RF_CTRLSHIFT)
249 #define RFTYPE_MON			BIT(RF_MONSHIFT)
250 #define RFTYPE_TOP			BIT(RF_TOPSHIFT)
251 #define RFTYPE_RES_CACHE		BIT(8)
252 #define RFTYPE_RES_MB			BIT(9)
253 #define RF_CTRL_INFO			(RFTYPE_INFO | RFTYPE_CTRL)
254 #define RF_MON_INFO			(RFTYPE_INFO | RFTYPE_MON)
255 #define RF_TOP_INFO			(RFTYPE_INFO | RFTYPE_TOP)
256 #define RF_CTRL_BASE			(RFTYPE_BASE | RFTYPE_CTRL)
257 
258 /* List of all resource groups */
259 extern struct list_head rdt_all_groups;
260 
261 extern int max_name_width, max_data_width;
262 
263 int __init rdtgroup_init(void);
264 void __exit rdtgroup_exit(void);
265 
266 /**
267  * struct rftype - describe each file in the resctrl file system
268  * @name:	File name
269  * @mode:	Access mode
270  * @kf_ops:	File operations
271  * @flags:	File specific RFTYPE_FLAGS_* flags
272  * @fflags:	File specific RF_* or RFTYPE_* flags
273  * @seq_show:	Show content of the file
274  * @write:	Write to the file
275  */
276 struct rftype {
277 	char			*name;
278 	umode_t			mode;
279 	const struct kernfs_ops	*kf_ops;
280 	unsigned long		flags;
281 	unsigned long		fflags;
282 
283 	int (*seq_show)(struct kernfs_open_file *of,
284 			struct seq_file *sf, void *v);
285 	/*
286 	 * write() is the generic write callback which maps directly to
287 	 * kernfs write operation and overrides all other operations.
288 	 * Maximum write size is determined by ->max_write_len.
289 	 */
290 	ssize_t (*write)(struct kernfs_open_file *of,
291 			 char *buf, size_t nbytes, loff_t off);
292 };
293 
294 /**
295  * struct mbm_state - status for each MBM counter in each domain
296  * @prev_bw_bytes: Previous bytes value read for bandwidth calculation
297  * @prev_bw:	The most recent bandwidth in MBps
298  */
299 struct mbm_state {
300 	u64	prev_bw_bytes;
301 	u32	prev_bw;
302 };
303 
304 /**
305  * struct arch_mbm_state - values used to compute resctrl_arch_rmid_read()s
306  *			   return value.
307  * @chunks:	Total data moved (multiply by rdt_group.mon_scale to get bytes)
308  * @prev_msr:	Value of IA32_QM_CTR last time it was read for the RMID used to
309  *		find this struct.
310  */
311 struct arch_mbm_state {
312 	u64	chunks;
313 	u64	prev_msr;
314 };
315 
316 /**
317  * struct rdt_hw_domain - Arch private attributes of a set of CPUs that share
318  *			  a resource
319  * @d_resctrl:	Properties exposed to the resctrl file system
320  * @ctrl_val:	array of cache or mem ctrl values (indexed by CLOSID)
321  * @arch_mbm_total:	arch private state for MBM total bandwidth
322  * @arch_mbm_local:	arch private state for MBM local bandwidth
323  *
324  * Members of this structure are accessed via helpers that provide abstraction.
325  */
326 struct rdt_hw_domain {
327 	struct rdt_domain		d_resctrl;
328 	u32				*ctrl_val;
329 	struct arch_mbm_state		*arch_mbm_total;
330 	struct arch_mbm_state		*arch_mbm_local;
331 };
332 
resctrl_to_arch_dom(struct rdt_domain * r)333 static inline struct rdt_hw_domain *resctrl_to_arch_dom(struct rdt_domain *r)
334 {
335 	return container_of(r, struct rdt_hw_domain, d_resctrl);
336 }
337 
338 /**
339  * struct msr_param - set a range of MSRs from a domain
340  * @res:       The resource to use
341  * @low:       Beginning index from base MSR
342  * @high:      End index
343  */
344 struct msr_param {
345 	struct rdt_resource	*res;
346 	u32			low;
347 	u32			high;
348 };
349 
is_llc_occupancy_enabled(void)350 static inline bool is_llc_occupancy_enabled(void)
351 {
352 	return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
353 }
354 
is_mbm_total_enabled(void)355 static inline bool is_mbm_total_enabled(void)
356 {
357 	return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
358 }
359 
is_mbm_local_enabled(void)360 static inline bool is_mbm_local_enabled(void)
361 {
362 	return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
363 }
364 
is_mbm_enabled(void)365 static inline bool is_mbm_enabled(void)
366 {
367 	return (is_mbm_total_enabled() || is_mbm_local_enabled());
368 }
369 
is_mbm_event(int e)370 static inline bool is_mbm_event(int e)
371 {
372 	return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
373 		e <= QOS_L3_MBM_LOCAL_EVENT_ID);
374 }
375 
376 struct rdt_parse_data {
377 	struct rdtgroup		*rdtgrp;
378 	char			*buf;
379 };
380 
381 /**
382  * struct rdt_hw_resource - arch private attributes of a resctrl resource
383  * @r_resctrl:		Attributes of the resource used directly by resctrl.
384  * @num_closid:		Maximum number of closid this hardware can support,
385  *			regardless of CDP. This is exposed via
386  *			resctrl_arch_get_num_closid() to avoid confusion
387  *			with struct resctrl_schema's property of the same name,
388  *			which has been corrected for features like CDP.
389  * @msr_base:		Base MSR address for CBMs
390  * @msr_update:		Function pointer to update QOS MSRs
391  * @mon_scale:		cqm counter * mon_scale = occupancy in bytes
392  * @mbm_width:		Monitor width, to detect and correct for overflow.
393  * @mbm_cfg_mask:	Bandwidth sources that can be tracked when Bandwidth
394  *			Monitoring Event Configuration (BMEC) is supported.
395  * @cdp_enabled:	CDP state of this resource
396  *
397  * Members of this structure are either private to the architecture
398  * e.g. mbm_width, or accessed via helpers that provide abstraction. e.g.
399  * msr_update and msr_base.
400  */
401 struct rdt_hw_resource {
402 	struct rdt_resource	r_resctrl;
403 	u32			num_closid;
404 	unsigned int		msr_base;
405 	void (*msr_update)	(struct rdt_domain *d, struct msr_param *m,
406 				 struct rdt_resource *r);
407 	unsigned int		mon_scale;
408 	unsigned int		mbm_width;
409 	unsigned int		mbm_cfg_mask;
410 	bool			cdp_enabled;
411 };
412 
resctrl_to_arch_res(struct rdt_resource * r)413 static inline struct rdt_hw_resource *resctrl_to_arch_res(struct rdt_resource *r)
414 {
415 	return container_of(r, struct rdt_hw_resource, r_resctrl);
416 }
417 
418 int parse_cbm(struct rdt_parse_data *data, struct resctrl_schema *s,
419 	      struct rdt_domain *d);
420 int parse_bw(struct rdt_parse_data *data, struct resctrl_schema *s,
421 	     struct rdt_domain *d);
422 
423 extern struct mutex rdtgroup_mutex;
424 
425 extern struct rdt_hw_resource rdt_resources_all[];
426 extern struct rdtgroup rdtgroup_default;
427 DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
428 
429 extern struct dentry *debugfs_resctrl;
430 
431 enum resctrl_res_level {
432 	RDT_RESOURCE_L3,
433 	RDT_RESOURCE_L2,
434 	RDT_RESOURCE_MBA,
435 	RDT_RESOURCE_SMBA,
436 
437 	/* Must be the last */
438 	RDT_NUM_RESOURCES,
439 };
440 
resctrl_inc(struct rdt_resource * res)441 static inline struct rdt_resource *resctrl_inc(struct rdt_resource *res)
442 {
443 	struct rdt_hw_resource *hw_res = resctrl_to_arch_res(res);
444 
445 	hw_res++;
446 	return &hw_res->r_resctrl;
447 }
448 
resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)449 static inline bool resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)
450 {
451 	return rdt_resources_all[l].cdp_enabled;
452 }
453 
454 int resctrl_arch_set_cdp_enabled(enum resctrl_res_level l, bool enable);
455 
456 /*
457  * To return the common struct rdt_resource, which is contained in struct
458  * rdt_hw_resource, walk the resctrl member of struct rdt_hw_resource.
459  */
460 #define for_each_rdt_resource(r)					      \
461 	for (r = &rdt_resources_all[0].r_resctrl;			      \
462 	     r <= &rdt_resources_all[RDT_NUM_RESOURCES - 1].r_resctrl;	      \
463 	     r = resctrl_inc(r))
464 
465 #define for_each_capable_rdt_resource(r)				      \
466 	for_each_rdt_resource(r)					      \
467 		if (r->alloc_capable || r->mon_capable)
468 
469 #define for_each_alloc_capable_rdt_resource(r)				      \
470 	for_each_rdt_resource(r)					      \
471 		if (r->alloc_capable)
472 
473 #define for_each_mon_capable_rdt_resource(r)				      \
474 	for_each_rdt_resource(r)					      \
475 		if (r->mon_capable)
476 
477 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */
478 union cpuid_0x10_1_eax {
479 	struct {
480 		unsigned int cbm_len:5;
481 	} split;
482 	unsigned int full;
483 };
484 
485 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */
486 union cpuid_0x10_3_eax {
487 	struct {
488 		unsigned int max_delay:12;
489 	} split;
490 	unsigned int full;
491 };
492 
493 /* CPUID.(EAX=10H, ECX=ResID).EDX */
494 union cpuid_0x10_x_edx {
495 	struct {
496 		unsigned int cos_max:16;
497 	} split;
498 	unsigned int full;
499 };
500 
501 void rdt_last_cmd_clear(void);
502 void rdt_last_cmd_puts(const char *s);
503 __printf(1, 2)
504 void rdt_last_cmd_printf(const char *fmt, ...);
505 
506 void rdt_ctrl_update(void *arg);
507 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
508 void rdtgroup_kn_unlock(struct kernfs_node *kn);
509 int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
510 int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
511 			     umode_t mask);
512 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
513 				   struct list_head **pos);
514 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
515 				char *buf, size_t nbytes, loff_t off);
516 int rdtgroup_schemata_show(struct kernfs_open_file *of,
517 			   struct seq_file *s, void *v);
518 bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_domain *d,
519 			   unsigned long cbm, int closid, bool exclusive);
520 unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
521 				  unsigned long cbm);
522 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
523 int rdtgroup_tasks_assigned(struct rdtgroup *r);
524 int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
525 int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
526 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
527 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
528 int rdt_pseudo_lock_init(void);
529 void rdt_pseudo_lock_release(void);
530 int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
531 void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
532 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
533 int closids_supported(void);
534 void closid_free(int closid);
535 int alloc_rmid(void);
536 void free_rmid(u32 rmid);
537 int rdt_get_mon_l3_config(struct rdt_resource *r);
538 bool __init rdt_cpu_has(int flag);
539 void mon_event_count(void *info);
540 int rdtgroup_mondata_show(struct seq_file *m, void *arg);
541 void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
542 		    struct rdt_domain *d, struct rdtgroup *rdtgrp,
543 		    int evtid, int first);
544 void mbm_setup_overflow_handler(struct rdt_domain *dom,
545 				unsigned long delay_ms);
546 void mbm_handle_overflow(struct work_struct *work);
547 void __init intel_rdt_mbm_apply_quirk(void);
548 bool is_mba_sc(struct rdt_resource *r);
549 void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
550 void cqm_handle_limbo(struct work_struct *work);
551 bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
552 void __check_limbo(struct rdt_domain *d, bool force_free);
553 void rdt_domain_reconfigure_cdp(struct rdt_resource *r);
554 void __init thread_throttle_mode_init(void);
555 void __init mbm_config_rftype_init(const char *config);
556 void rdt_staged_configs_clear(void);
557 
558 #endif /* _ASM_X86_RESCTRL_INTERNAL_H */
559