Linux Audio

Check our new training course

Loading...
v5.4
  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/sched.h>
  6#include <linux/kernfs.h>
  7#include <linux/fs_context.h>
  8#include <linux/jump_label.h>
  9
 10#define MSR_IA32_L3_QOS_CFG		0xc81
 11#define MSR_IA32_L2_QOS_CFG		0xc82
 12#define MSR_IA32_L3_CBM_BASE		0xc90
 13#define MSR_IA32_L2_CBM_BASE		0xd10
 14#define MSR_IA32_MBA_THRTL_BASE		0xd50
 15#define MSR_IA32_MBA_BW_BASE		0xc0000200
 16
 17#define MSR_IA32_QM_CTR			0x0c8e
 18#define MSR_IA32_QM_EVTSEL		0x0c8d
 19
 20#define L3_QOS_CDP_ENABLE		0x01ULL
 21
 22#define L2_QOS_CDP_ENABLE		0x01ULL
 23
 24/*
 25 * Event IDs are used to program IA32_QM_EVTSEL before reading event
 26 * counter from IA32_QM_CTR
 27 */
 28#define QOS_L3_OCCUP_EVENT_ID		0x01
 29#define QOS_L3_MBM_TOTAL_EVENT_ID	0x02
 30#define QOS_L3_MBM_LOCAL_EVENT_ID	0x03
 31
 32#define CQM_LIMBOCHECK_INTERVAL	1000
 33
 34#define MBM_CNTR_WIDTH			24
 35#define MBM_OVERFLOW_INTERVAL		1000
 36#define MAX_MBA_BW			100u
 37#define MBA_IS_LINEAR			0x4
 38#define MBA_MAX_MBPS			U32_MAX
 39#define MAX_MBA_BW_AMD			0x800
 
 40
 41#define RMID_VAL_ERROR			BIT_ULL(63)
 42#define RMID_VAL_UNAVAIL		BIT_ULL(62)
 
 
 
 
 
 
 43
 44
 45struct rdt_fs_context {
 46	struct kernfs_fs_context	kfc;
 47	bool				enable_cdpl2;
 48	bool				enable_cdpl3;
 49	bool				enable_mba_mbps;
 50};
 51
 52static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
 53{
 54	struct kernfs_fs_context *kfc = fc->fs_private;
 55
 56	return container_of(kfc, struct rdt_fs_context, kfc);
 57}
 58
 59DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
 
 60
 61/**
 62 * struct mon_evt - Entry in the event list of a resource
 63 * @evtid:		event id
 64 * @name:		name of the event
 
 65 */
 66struct mon_evt {
 67	u32			evtid;
 68	char			*name;
 69	struct list_head	list;
 70};
 71
 72/**
 73 * struct mon_data_bits - Monitoring details for each event file
 74 * @rid:               Resource id associated with the event file.
 
 
 75 * @evtid:             Event id associated with the event file
 76 * @domid:             The domain to which the event file belongs
 
 77 */
 78union mon_data_bits {
 79	void *priv;
 80	struct {
 81		unsigned int rid	: 10;
 82		unsigned int evtid	: 8;
 83		unsigned int domid	: 14;
 84	} u;
 85};
 86
 87struct rmid_read {
 88	struct rdtgroup		*rgrp;
 
 89	struct rdt_domain	*d;
 90	int			evtid;
 91	bool			first;
 
 92	u64			val;
 93};
 94
 95extern unsigned int resctrl_cqm_threshold;
 96extern bool rdt_alloc_capable;
 97extern bool rdt_mon_capable;
 98extern unsigned int rdt_mon_features;
 
 99
100enum rdt_group_type {
101	RDTCTRL_GROUP = 0,
102	RDTMON_GROUP,
103	RDT_NUM_GROUP,
104};
105
106/**
107 * enum rdtgrp_mode - Mode of a RDT resource group
108 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
109 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
110 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
111 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
112 *                          allowed AND the allocations are Cache Pseudo-Locked
 
113 *
114 * The mode of a resource group enables control over the allowed overlap
115 * between allocations associated with different resource groups (classes
116 * of service). User is able to modify the mode of a resource group by
117 * writing to the "mode" resctrl file associated with the resource group.
118 *
119 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
120 * writing the appropriate text to the "mode" file. A resource group enters
121 * "pseudo-locked" mode after the schemata is written while the resource
122 * group is in "pseudo-locksetup" mode.
123 */
124enum rdtgrp_mode {
125	RDT_MODE_SHAREABLE = 0,
126	RDT_MODE_EXCLUSIVE,
127	RDT_MODE_PSEUDO_LOCKSETUP,
128	RDT_MODE_PSEUDO_LOCKED,
129
130	/* Must be last */
131	RDT_NUM_MODES,
132};
133
134/**
135 * struct mongroup - store mon group's data in resctrl fs.
136 * @mon_data_kn		kernlfs node for the mon_data directory
137 * @parent:			parent rdtgrp
138 * @crdtgrp_list:		child rdtgroup node list
139 * @rmid:			rmid for this rdtgroup
140 */
141struct mongroup {
142	struct kernfs_node	*mon_data_kn;
143	struct rdtgroup		*parent;
144	struct list_head	crdtgrp_list;
145	u32			rmid;
146};
147
148/**
149 * struct pseudo_lock_region - pseudo-lock region information
150 * @r:			RDT resource to which this pseudo-locked region
151 *			belongs
152 * @d:			RDT domain to which this pseudo-locked region
153 *			belongs
154 * @cbm:		bitmask of the pseudo-locked region
155 * @lock_thread_wq:	waitqueue used to wait on the pseudo-locking thread
156 *			completion
157 * @thread_done:	variable used by waitqueue to test if pseudo-locking
158 *			thread completed
159 * @cpu:		core associated with the cache on which the setup code
160 *			will be run
161 * @line_size:		size of the cache lines
162 * @size:		size of pseudo-locked region in bytes
163 * @kmem:		the kernel memory associated with pseudo-locked region
164 * @minor:		minor number of character device associated with this
165 *			region
166 * @debugfs_dir:	pointer to this region's directory in the debugfs
167 *			filesystem
168 * @pm_reqs:		Power management QoS requests related to this region
169 */
170struct pseudo_lock_region {
171	struct rdt_resource	*r;
172	struct rdt_domain	*d;
173	u32			cbm;
174	wait_queue_head_t	lock_thread_wq;
175	int			thread_done;
176	int			cpu;
177	unsigned int		line_size;
178	unsigned int		size;
179	void			*kmem;
180	unsigned int		minor;
181	struct dentry		*debugfs_dir;
182	struct list_head	pm_reqs;
183};
184
185/**
186 * struct rdtgroup - store rdtgroup's data in resctrl file system.
187 * @kn:				kernfs node
188 * @rdtgroup_list:		linked list for all rdtgroups
189 * @closid:			closid for this rdtgroup
190 * @cpu_mask:			CPUs assigned to this rdtgroup
191 * @flags:			status bits
192 * @waitcount:			how many cpus expect to find this
193 *				group when they acquire rdtgroup_mutex
194 * @type:			indicates type of this rdtgroup - either
195 *				monitor only or ctrl_mon group
196 * @mon:			mongroup related data
197 * @mode:			mode of resource group
198 * @plr:			pseudo-locked region
199 */
200struct rdtgroup {
201	struct kernfs_node		*kn;
202	struct list_head		rdtgroup_list;
203	u32				closid;
204	struct cpumask			cpu_mask;
205	int				flags;
206	atomic_t			waitcount;
207	enum rdt_group_type		type;
208	struct mongroup			mon;
209	enum rdtgrp_mode		mode;
210	struct pseudo_lock_region	*plr;
211};
212
213/* rdtgroup.flags */
214#define	RDT_DELETED		1
215
216/* rftype.flags */
217#define RFTYPE_FLAGS_CPUS_LIST	1
218
219/*
220 * Define the file type flags for base and info directories.
221 */
222#define RFTYPE_INFO			BIT(0)
223#define RFTYPE_BASE			BIT(1)
224#define RF_CTRLSHIFT			4
225#define RF_MONSHIFT			5
226#define RF_TOPSHIFT			6
227#define RFTYPE_CTRL			BIT(RF_CTRLSHIFT)
228#define RFTYPE_MON			BIT(RF_MONSHIFT)
229#define RFTYPE_TOP			BIT(RF_TOPSHIFT)
230#define RFTYPE_RES_CACHE		BIT(8)
231#define RFTYPE_RES_MB			BIT(9)
232#define RF_CTRL_INFO			(RFTYPE_INFO | RFTYPE_CTRL)
233#define RF_MON_INFO			(RFTYPE_INFO | RFTYPE_MON)
234#define RF_TOP_INFO			(RFTYPE_INFO | RFTYPE_TOP)
235#define RF_CTRL_BASE			(RFTYPE_BASE | RFTYPE_CTRL)
236
237/* List of all resource groups */
238extern struct list_head rdt_all_groups;
239
240extern int max_name_width, max_data_width;
241
242int __init rdtgroup_init(void);
243void __exit rdtgroup_exit(void);
244
245/**
246 * struct rftype - describe each file in the resctrl file system
247 * @name:	File name
248 * @mode:	Access mode
249 * @kf_ops:	File operations
250 * @flags:	File specific RFTYPE_FLAGS_* flags
251 * @fflags:	File specific RF_* or RFTYPE_* flags
252 * @seq_show:	Show content of the file
253 * @write:	Write to the file
254 */
255struct rftype {
256	char			*name;
257	umode_t			mode;
258	struct kernfs_ops	*kf_ops;
259	unsigned long		flags;
260	unsigned long		fflags;
261
262	int (*seq_show)(struct kernfs_open_file *of,
263			struct seq_file *sf, void *v);
264	/*
265	 * write() is the generic write callback which maps directly to
266	 * kernfs write operation and overrides all other operations.
267	 * Maximum write size is determined by ->max_write_len.
268	 */
269	ssize_t (*write)(struct kernfs_open_file *of,
270			 char *buf, size_t nbytes, loff_t off);
271};
272
273/**
274 * struct mbm_state - status for each MBM counter in each domain
275 * @chunks:	Total data moved (multiply by rdt_group.mon_scale to get bytes)
276 * @prev_msr	Value of IA32_QM_CTR for this RMID last time we read it
277 * @chunks_bw	Total local data moved. Used for bandwidth calculation
278 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting
279 * @prev_bw	The most recent bandwidth in MBps
280 * @delta_bw	Difference between the current and previous bandwidth
281 * @delta_comp	Indicates whether to compute the delta_bw
282 */
283struct mbm_state {
284	u64	chunks;
285	u64	prev_msr;
286	u64	chunks_bw;
287	u64	prev_bw_msr;
288	u32	prev_bw;
289	u32	delta_bw;
290	bool	delta_comp;
291};
292
293/**
294 * struct rdt_domain - group of cpus sharing an RDT resource
295 * @list:	all instances of this resource
296 * @id:		unique id for this instance
297 * @cpu_mask:	which cpus share this resource
298 * @rmid_busy_llc:
299 *		bitmap of which limbo RMIDs are above threshold
300 * @mbm_total:	saved state for MBM total bandwidth
301 * @mbm_local:	saved state for MBM local bandwidth
302 * @mbm_over:	worker to periodically read MBM h/w counters
303 * @cqm_limbo:	worker to periodically read CQM h/w counters
304 * @mbm_work_cpu:
305 *		worker cpu for MBM h/w counters
306 * @cqm_work_cpu:
307 *		worker cpu for CQM h/w counters
 
308 * @ctrl_val:	array of cache or mem ctrl values (indexed by CLOSID)
309 * @mbps_val:	When mba_sc is enabled, this holds the bandwidth in MBps
310 * @new_ctrl:	new ctrl value to be loaded
311 * @have_new_ctrl: did user provide new_ctrl for this domain
312 * @plr:	pseudo-locked region (if any) associated with domain
313 */
314struct rdt_domain {
315	struct list_head		list;
316	int				id;
317	struct cpumask			cpu_mask;
318	unsigned long			*rmid_busy_llc;
319	struct mbm_state		*mbm_total;
320	struct mbm_state		*mbm_local;
321	struct delayed_work		mbm_over;
322	struct delayed_work		cqm_limbo;
323	int				mbm_work_cpu;
324	int				cqm_work_cpu;
325	u32				*ctrl_val;
326	u32				*mbps_val;
327	u32				new_ctrl;
328	bool				have_new_ctrl;
329	struct pseudo_lock_region	*plr;
330};
331
 
 
 
 
 
332/**
333 * struct msr_param - set a range of MSRs from a domain
334 * @res:       The resource to use
335 * @low:       Beginning index from base MSR
336 * @high:      End index
337 */
338struct msr_param {
339	struct rdt_resource	*res;
340	int			low;
341	int			high;
342};
343
344/**
345 * struct rdt_cache - Cache allocation related data
346 * @cbm_len:		Length of the cache bit mask
347 * @min_cbm_bits:	Minimum number of consecutive bits to be set
348 * @cbm_idx_mult:	Multiplier of CBM index
349 * @cbm_idx_offset:	Offset of CBM index. CBM index is computed by:
350 *			closid * cbm_idx_multi + cbm_idx_offset
351 *			in a cache bit mask
352 * @shareable_bits:	Bitmask of shareable resource with other
353 *			executing entities
354 */
355struct rdt_cache {
356	unsigned int	cbm_len;
357	unsigned int	min_cbm_bits;
358	unsigned int	cbm_idx_mult;
359	unsigned int	cbm_idx_offset;
360	unsigned int	shareable_bits;
361};
362
363/**
364 * struct rdt_membw - Memory bandwidth allocation related data
365 * @max_delay:		Max throttle delay. Delay is the hardware
366 *			representation for memory bandwidth.
367 * @min_bw:		Minimum memory bandwidth percentage user can request
368 * @bw_gran:		Granularity at which the memory bandwidth is allocated
369 * @delay_linear:	True if memory B/W delay is in linear scale
370 * @mba_sc:		True if MBA software controller(mba_sc) is enabled
371 * @mb_map:		Mapping of memory B/W percentage to memory B/W delay
372 */
373struct rdt_membw {
374	u32		max_delay;
375	u32		min_bw;
376	u32		bw_gran;
377	u32		delay_linear;
378	bool		mba_sc;
379	u32		*mb_map;
380};
381
382static inline bool is_llc_occupancy_enabled(void)
383{
384	return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
385}
386
387static inline bool is_mbm_total_enabled(void)
388{
389	return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
390}
391
392static inline bool is_mbm_local_enabled(void)
393{
394	return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
395}
396
397static inline bool is_mbm_enabled(void)
398{
399	return (is_mbm_total_enabled() || is_mbm_local_enabled());
400}
401
402static inline bool is_mbm_event(int e)
403{
404	return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
405		e <= QOS_L3_MBM_LOCAL_EVENT_ID);
406}
407
408struct rdt_parse_data {
409	struct rdtgroup		*rdtgrp;
410	char			*buf;
411};
412
413/**
414 * struct rdt_resource - attributes of an RDT resource
415 * @rid:		The index of the resource
416 * @alloc_enabled:	Is allocation enabled on this machine
417 * @mon_enabled:	Is monitoring enabled for this feature
418 * @alloc_capable:	Is allocation available on this machine
419 * @mon_capable:	Is monitor feature available on this machine
420 * @name:		Name to use in "schemata" file
421 * @num_closid:		Number of CLOSIDs available
422 * @cache_level:	Which cache level defines scope of this resource
423 * @default_ctrl:	Specifies default cache cbm or memory B/W percent.
424 * @msr_base:		Base MSR address for CBMs
425 * @msr_update:		Function pointer to update QOS MSRs
426 * @data_width:		Character width of data when displaying
427 * @domains:		All domains for this resource
428 * @cache:		Cache allocation related data
429 * @format_str:		Per resource format string to show domain value
430 * @parse_ctrlval:	Per resource function pointer to parse control values
431 * @cbm_validate	Cache bitmask validate function
432 * @evt_list:		List of monitoring events
433 * @num_rmid:		Number of RMIDs available
434 * @mon_scale:		cqm counter * mon_scale = occupancy in bytes
435 * @fflags:		flags to choose base and info files
436 */
437struct rdt_resource {
438	int			rid;
439	bool			alloc_enabled;
440	bool			mon_enabled;
441	bool			alloc_capable;
442	bool			mon_capable;
443	char			*name;
444	int			num_closid;
445	int			cache_level;
446	u32			default_ctrl;
447	unsigned int		msr_base;
448	void (*msr_update)	(struct rdt_domain *d, struct msr_param *m,
449				 struct rdt_resource *r);
450	int			data_width;
451	struct list_head	domains;
452	struct rdt_cache	cache;
453	struct rdt_membw	membw;
454	const char		*format_str;
455	int (*parse_ctrlval)(struct rdt_parse_data *data,
456			     struct rdt_resource *r,
457			     struct rdt_domain *d);
458	bool (*cbm_validate)(char *buf, u32 *data, struct rdt_resource *r);
459	struct list_head	evt_list;
460	int			num_rmid;
461	unsigned int		mon_scale;
462	unsigned long		fflags;
 
463};
464
465int parse_cbm(struct rdt_parse_data *data, struct rdt_resource *r,
 
 
 
 
 
466	      struct rdt_domain *d);
467int parse_bw_intel(struct rdt_parse_data *data, struct rdt_resource *r,
468		   struct rdt_domain *d);
469int parse_bw_amd(struct rdt_parse_data *data, struct rdt_resource *r,
470		 struct rdt_domain *d);
471
472extern struct mutex rdtgroup_mutex;
473
474extern struct rdt_resource rdt_resources_all[];
475extern struct rdtgroup rdtgroup_default;
476DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
477
478extern struct dentry *debugfs_resctrl;
479
480enum {
481	RDT_RESOURCE_L3,
482	RDT_RESOURCE_L3DATA,
483	RDT_RESOURCE_L3CODE,
484	RDT_RESOURCE_L2,
485	RDT_RESOURCE_L2DATA,
486	RDT_RESOURCE_L2CODE,
487	RDT_RESOURCE_MBA,
488
489	/* Must be the last */
490	RDT_NUM_RESOURCES,
491};
492
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
493#define for_each_rdt_resource(r)					      \
494	for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
495	     r++)
 
496
497#define for_each_capable_rdt_resource(r)				      \
498	for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
499	     r++)							      \
500		if (r->alloc_capable || r->mon_capable)
501
502#define for_each_alloc_capable_rdt_resource(r)				      \
503	for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
504	     r++)							      \
505		if (r->alloc_capable)
506
507#define for_each_mon_capable_rdt_resource(r)				      \
508	for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
509	     r++)							      \
510		if (r->mon_capable)
511
512#define for_each_alloc_enabled_rdt_resource(r)				      \
513	for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
514	     r++)							      \
515		if (r->alloc_enabled)
516
517#define for_each_mon_enabled_rdt_resource(r)				      \
518	for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
519	     r++)							      \
520		if (r->mon_enabled)
521
522/* CPUID.(EAX=10H, ECX=ResID=1).EAX */
523union cpuid_0x10_1_eax {
524	struct {
525		unsigned int cbm_len:5;
526	} split;
527	unsigned int full;
528};
529
530/* CPUID.(EAX=10H, ECX=ResID=3).EAX */
531union cpuid_0x10_3_eax {
532	struct {
533		unsigned int max_delay:12;
534	} split;
535	unsigned int full;
536};
537
538/* CPUID.(EAX=10H, ECX=ResID).EDX */
539union cpuid_0x10_x_edx {
540	struct {
541		unsigned int cos_max:16;
542	} split;
543	unsigned int full;
544};
545
546void rdt_last_cmd_clear(void);
547void rdt_last_cmd_puts(const char *s);
 
548void rdt_last_cmd_printf(const char *fmt, ...);
549
550void rdt_ctrl_update(void *arg);
551struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
552void rdtgroup_kn_unlock(struct kernfs_node *kn);
553int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
554int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
555			     umode_t mask);
556struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
557				   struct list_head **pos);
558ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
559				char *buf, size_t nbytes, loff_t off);
560int rdtgroup_schemata_show(struct kernfs_open_file *of,
561			   struct seq_file *s, void *v);
562bool rdtgroup_cbm_overlaps(struct rdt_resource *r, struct rdt_domain *d,
563			   unsigned long cbm, int closid, bool exclusive);
564unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
565				  unsigned long cbm);
566enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
567int rdtgroup_tasks_assigned(struct rdtgroup *r);
568int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
569int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
570bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
571bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
572int rdt_pseudo_lock_init(void);
573void rdt_pseudo_lock_release(void);
574int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
575void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
576struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
577int update_domains(struct rdt_resource *r, int closid);
578int closids_supported(void);
579void closid_free(int closid);
580int alloc_rmid(void);
581void free_rmid(u32 rmid);
582int rdt_get_mon_l3_config(struct rdt_resource *r);
583void mon_event_count(void *info);
584int rdtgroup_mondata_show(struct seq_file *m, void *arg);
585void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
586				    unsigned int dom_id);
587void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
588				    struct rdt_domain *d);
589void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
590		    struct rdtgroup *rdtgrp, int evtid, int first);
591void mbm_setup_overflow_handler(struct rdt_domain *dom,
592				unsigned long delay_ms);
593void mbm_handle_overflow(struct work_struct *work);
 
594bool is_mba_sc(struct rdt_resource *r);
595void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm);
596u32 delay_bw_map(unsigned long bw, struct rdt_resource *r);
597void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
598void cqm_handle_limbo(struct work_struct *work);
599bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
600void __check_limbo(struct rdt_domain *d, bool force_free);
601bool cbm_validate_intel(char *buf, u32 *data, struct rdt_resource *r);
602bool cbm_validate_amd(char *buf, u32 *data, struct rdt_resource *r);
603
604#endif /* _ASM_X86_RESCTRL_INTERNAL_H */
v6.2
  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 MAX_MBA_BW_AMD			0x800
 22#define MBM_CNTR_WIDTH_OFFSET_AMD	20
 23
 24#define RMID_VAL_ERROR			BIT_ULL(63)
 25#define RMID_VAL_UNAVAIL		BIT_ULL(62)
 26/*
 27 * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for
 28 * data to be returned. The counter width is discovered from the hardware
 29 * as an offset from MBM_CNTR_WIDTH_BASE.
 30 */
 31#define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
 32
 33
 34struct rdt_fs_context {
 35	struct kernfs_fs_context	kfc;
 36	bool				enable_cdpl2;
 37	bool				enable_cdpl3;
 38	bool				enable_mba_mbps;
 39};
 40
 41static inline struct rdt_fs_context *rdt_fc2context(struct fs_context *fc)
 42{
 43	struct kernfs_fs_context *kfc = fc->fs_private;
 44
 45	return container_of(kfc, struct rdt_fs_context, kfc);
 46}
 47
 48DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
 49DECLARE_STATIC_KEY_FALSE(rdt_mon_enable_key);
 50
 51/**
 52 * struct mon_evt - Entry in the event list of a resource
 53 * @evtid:		event id
 54 * @name:		name of the event
 55 * @list:		entry in &rdt_resource->evt_list
 56 */
 57struct mon_evt {
 58	enum resctrl_event_id	evtid;
 59	char			*name;
 60	struct list_head	list;
 61};
 62
 63/**
 64 * union mon_data_bits - Monitoring details for each event file
 65 * @priv:              Used to store monitoring event data in @u
 66 *                     as kernfs private data
 67 * @rid:               Resource id associated with the event file
 68 * @evtid:             Event id associated with the event file
 69 * @domid:             The domain to which the event file belongs
 70 * @u:                 Name of the bit fields struct
 71 */
 72union mon_data_bits {
 73	void *priv;
 74	struct {
 75		unsigned int rid		: 10;
 76		enum resctrl_event_id evtid	: 8;
 77		unsigned int domid		: 14;
 78	} u;
 79};
 80
 81struct rmid_read {
 82	struct rdtgroup		*rgrp;
 83	struct rdt_resource	*r;
 84	struct rdt_domain	*d;
 85	enum resctrl_event_id	evtid;
 86	bool			first;
 87	int			err;
 88	u64			val;
 89};
 90
 
 91extern bool rdt_alloc_capable;
 92extern bool rdt_mon_capable;
 93extern unsigned int rdt_mon_features;
 94extern struct list_head resctrl_schema_all;
 95
 96enum rdt_group_type {
 97	RDTCTRL_GROUP = 0,
 98	RDTMON_GROUP,
 99	RDT_NUM_GROUP,
100};
101
102/**
103 * enum rdtgrp_mode - Mode of a RDT resource group
104 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
105 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
106 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
107 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
108 *                          allowed AND the allocations are Cache Pseudo-Locked
109 * @RDT_NUM_MODES: Total number of modes
110 *
111 * The mode of a resource group enables control over the allowed overlap
112 * between allocations associated with different resource groups (classes
113 * of service). User is able to modify the mode of a resource group by
114 * writing to the "mode" resctrl file associated with the resource group.
115 *
116 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
117 * writing the appropriate text to the "mode" file. A resource group enters
118 * "pseudo-locked" mode after the schemata is written while the resource
119 * group is in "pseudo-locksetup" mode.
120 */
121enum rdtgrp_mode {
122	RDT_MODE_SHAREABLE = 0,
123	RDT_MODE_EXCLUSIVE,
124	RDT_MODE_PSEUDO_LOCKSETUP,
125	RDT_MODE_PSEUDO_LOCKED,
126
127	/* Must be last */
128	RDT_NUM_MODES,
129};
130
131/**
132 * struct mongroup - store mon group's data in resctrl fs.
133 * @mon_data_kn:		kernfs node for the mon_data directory
134 * @parent:			parent rdtgrp
135 * @crdtgrp_list:		child rdtgroup node list
136 * @rmid:			rmid for this rdtgroup
137 */
138struct mongroup {
139	struct kernfs_node	*mon_data_kn;
140	struct rdtgroup		*parent;
141	struct list_head	crdtgrp_list;
142	u32			rmid;
143};
144
145/**
146 * struct pseudo_lock_region - pseudo-lock region information
147 * @s:			Resctrl schema for the resource to which this
148 *			pseudo-locked region belongs
149 * @d:			RDT domain to which this pseudo-locked region
150 *			belongs
151 * @cbm:		bitmask of the pseudo-locked region
152 * @lock_thread_wq:	waitqueue used to wait on the pseudo-locking thread
153 *			completion
154 * @thread_done:	variable used by waitqueue to test if pseudo-locking
155 *			thread completed
156 * @cpu:		core associated with the cache on which the setup code
157 *			will be run
158 * @line_size:		size of the cache lines
159 * @size:		size of pseudo-locked region in bytes
160 * @kmem:		the kernel memory associated with pseudo-locked region
161 * @minor:		minor number of character device associated with this
162 *			region
163 * @debugfs_dir:	pointer to this region's directory in the debugfs
164 *			filesystem
165 * @pm_reqs:		Power management QoS requests related to this region
166 */
167struct pseudo_lock_region {
168	struct resctrl_schema	*s;
169	struct rdt_domain	*d;
170	u32			cbm;
171	wait_queue_head_t	lock_thread_wq;
172	int			thread_done;
173	int			cpu;
174	unsigned int		line_size;
175	unsigned int		size;
176	void			*kmem;
177	unsigned int		minor;
178	struct dentry		*debugfs_dir;
179	struct list_head	pm_reqs;
180};
181
182/**
183 * struct rdtgroup - store rdtgroup's data in resctrl file system.
184 * @kn:				kernfs node
185 * @rdtgroup_list:		linked list for all rdtgroups
186 * @closid:			closid for this rdtgroup
187 * @cpu_mask:			CPUs assigned to this rdtgroup
188 * @flags:			status bits
189 * @waitcount:			how many cpus expect to find this
190 *				group when they acquire rdtgroup_mutex
191 * @type:			indicates type of this rdtgroup - either
192 *				monitor only or ctrl_mon group
193 * @mon:			mongroup related data
194 * @mode:			mode of resource group
195 * @plr:			pseudo-locked region
196 */
197struct rdtgroup {
198	struct kernfs_node		*kn;
199	struct list_head		rdtgroup_list;
200	u32				closid;
201	struct cpumask			cpu_mask;
202	int				flags;
203	atomic_t			waitcount;
204	enum rdt_group_type		type;
205	struct mongroup			mon;
206	enum rdtgrp_mode		mode;
207	struct pseudo_lock_region	*plr;
208};
209
210/* rdtgroup.flags */
211#define	RDT_DELETED		1
212
213/* rftype.flags */
214#define RFTYPE_FLAGS_CPUS_LIST	1
215
216/*
217 * Define the file type flags for base and info directories.
218 */
219#define RFTYPE_INFO			BIT(0)
220#define RFTYPE_BASE			BIT(1)
221#define RF_CTRLSHIFT			4
222#define RF_MONSHIFT			5
223#define RF_TOPSHIFT			6
224#define RFTYPE_CTRL			BIT(RF_CTRLSHIFT)
225#define RFTYPE_MON			BIT(RF_MONSHIFT)
226#define RFTYPE_TOP			BIT(RF_TOPSHIFT)
227#define RFTYPE_RES_CACHE		BIT(8)
228#define RFTYPE_RES_MB			BIT(9)
229#define RF_CTRL_INFO			(RFTYPE_INFO | RFTYPE_CTRL)
230#define RF_MON_INFO			(RFTYPE_INFO | RFTYPE_MON)
231#define RF_TOP_INFO			(RFTYPE_INFO | RFTYPE_TOP)
232#define RF_CTRL_BASE			(RFTYPE_BASE | RFTYPE_CTRL)
233
234/* List of all resource groups */
235extern struct list_head rdt_all_groups;
236
237extern int max_name_width, max_data_width;
238
239int __init rdtgroup_init(void);
240void __exit rdtgroup_exit(void);
241
242/**
243 * struct rftype - describe each file in the resctrl file system
244 * @name:	File name
245 * @mode:	Access mode
246 * @kf_ops:	File operations
247 * @flags:	File specific RFTYPE_FLAGS_* flags
248 * @fflags:	File specific RF_* or RFTYPE_* flags
249 * @seq_show:	Show content of the file
250 * @write:	Write to the file
251 */
252struct rftype {
253	char			*name;
254	umode_t			mode;
255	const struct kernfs_ops	*kf_ops;
256	unsigned long		flags;
257	unsigned long		fflags;
258
259	int (*seq_show)(struct kernfs_open_file *of,
260			struct seq_file *sf, void *v);
261	/*
262	 * write() is the generic write callback which maps directly to
263	 * kernfs write operation and overrides all other operations.
264	 * Maximum write size is determined by ->max_write_len.
265	 */
266	ssize_t (*write)(struct kernfs_open_file *of,
267			 char *buf, size_t nbytes, loff_t off);
268};
269
270/**
271 * struct mbm_state - status for each MBM counter in each domain
272 * @prev_bw_bytes: Previous bytes value read for bandwidth calculation
273 * @prev_bw:	The most recent bandwidth in MBps
274 * @delta_bw:	Difference between the current and previous bandwidth
275 * @delta_comp:	Indicates whether to compute the delta_bw
 
 
 
276 */
277struct mbm_state {
278	u64	prev_bw_bytes;
 
 
 
279	u32	prev_bw;
280	u32	delta_bw;
281	bool	delta_comp;
282};
283
284/**
285 * struct arch_mbm_state - values used to compute resctrl_arch_rmid_read()s
286 *			   return value.
287 * @chunks:	Total data moved (multiply by rdt_group.mon_scale to get bytes)
288 * @prev_msr:	Value of IA32_QM_CTR last time it was read for the RMID used to
289 *		find this struct.
290 */
291struct arch_mbm_state {
292	u64	chunks;
293	u64	prev_msr;
294};
295
296/**
297 * struct rdt_hw_domain - Arch private attributes of a set of CPUs that share
298 *			  a resource
299 * @d_resctrl:	Properties exposed to the resctrl file system
300 * @ctrl_val:	array of cache or mem ctrl values (indexed by CLOSID)
301 * @arch_mbm_total:	arch private state for MBM total bandwidth
302 * @arch_mbm_local:	arch private state for MBM local bandwidth
303 *
304 * Members of this structure are accessed via helpers that provide abstraction.
305 */
306struct rdt_hw_domain {
307	struct rdt_domain		d_resctrl;
 
 
 
 
 
 
 
 
 
308	u32				*ctrl_val;
309	struct arch_mbm_state		*arch_mbm_total;
310	struct arch_mbm_state		*arch_mbm_local;
 
 
311};
312
313static inline struct rdt_hw_domain *resctrl_to_arch_dom(struct rdt_domain *r)
314{
315	return container_of(r, struct rdt_hw_domain, d_resctrl);
316}
317
318/**
319 * struct msr_param - set a range of MSRs from a domain
320 * @res:       The resource to use
321 * @low:       Beginning index from base MSR
322 * @high:      End index
323 */
324struct msr_param {
325	struct rdt_resource	*res;
326	u32			low;
327	u32			high;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
328};
329
330static inline bool is_llc_occupancy_enabled(void)
331{
332	return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
333}
334
335static inline bool is_mbm_total_enabled(void)
336{
337	return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
338}
339
340static inline bool is_mbm_local_enabled(void)
341{
342	return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
343}
344
345static inline bool is_mbm_enabled(void)
346{
347	return (is_mbm_total_enabled() || is_mbm_local_enabled());
348}
349
350static inline bool is_mbm_event(int e)
351{
352	return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
353		e <= QOS_L3_MBM_LOCAL_EVENT_ID);
354}
355
356struct rdt_parse_data {
357	struct rdtgroup		*rdtgrp;
358	char			*buf;
359};
360
361/**
362 * struct rdt_hw_resource - arch private attributes of a resctrl resource
363 * @r_resctrl:		Attributes of the resource used directly by resctrl.
364 * @num_closid:		Maximum number of closid this hardware can support,
365 *			regardless of CDP. This is exposed via
366 *			resctrl_arch_get_num_closid() to avoid confusion
367 *			with struct resctrl_schema's property of the same name,
368 *			which has been corrected for features like CDP.
 
 
 
369 * @msr_base:		Base MSR address for CBMs
370 * @msr_update:		Function pointer to update QOS MSRs
 
 
 
 
 
 
 
 
371 * @mon_scale:		cqm counter * mon_scale = occupancy in bytes
372 * @mbm_width:		Monitor width, to detect and correct for overflow.
373 * @cdp_enabled:	CDP state of this resource
374 *
375 * Members of this structure are either private to the architecture
376 * e.g. mbm_width, or accessed via helpers that provide abstraction. e.g.
377 * msr_update and msr_base.
378 */
379struct rdt_hw_resource {
380	struct rdt_resource	r_resctrl;
381	u32			num_closid;
 
 
382	unsigned int		msr_base;
383	void (*msr_update)	(struct rdt_domain *d, struct msr_param *m,
384				 struct rdt_resource *r);
 
 
 
 
 
 
 
 
 
 
 
385	unsigned int		mon_scale;
386	unsigned int		mbm_width;
387	bool			cdp_enabled;
388};
389
390static inline struct rdt_hw_resource *resctrl_to_arch_res(struct rdt_resource *r)
391{
392	return container_of(r, struct rdt_hw_resource, r_resctrl);
393}
394
395int parse_cbm(struct rdt_parse_data *data, struct resctrl_schema *s,
396	      struct rdt_domain *d);
397int parse_bw(struct rdt_parse_data *data, struct resctrl_schema *s,
398	     struct rdt_domain *d);
 
 
399
400extern struct mutex rdtgroup_mutex;
401
402extern struct rdt_hw_resource rdt_resources_all[];
403extern struct rdtgroup rdtgroup_default;
404DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
405
406extern struct dentry *debugfs_resctrl;
407
408enum resctrl_res_level {
409	RDT_RESOURCE_L3,
 
 
410	RDT_RESOURCE_L2,
 
 
411	RDT_RESOURCE_MBA,
412
413	/* Must be the last */
414	RDT_NUM_RESOURCES,
415};
416
417static inline struct rdt_resource *resctrl_inc(struct rdt_resource *res)
418{
419	struct rdt_hw_resource *hw_res = resctrl_to_arch_res(res);
420
421	hw_res++;
422	return &hw_res->r_resctrl;
423}
424
425static inline bool resctrl_arch_get_cdp_enabled(enum resctrl_res_level l)
426{
427	return rdt_resources_all[l].cdp_enabled;
428}
429
430int resctrl_arch_set_cdp_enabled(enum resctrl_res_level l, bool enable);
431
432/*
433 * To return the common struct rdt_resource, which is contained in struct
434 * rdt_hw_resource, walk the resctrl member of struct rdt_hw_resource.
435 */
436#define for_each_rdt_resource(r)					      \
437	for (r = &rdt_resources_all[0].r_resctrl;			      \
438	     r <= &rdt_resources_all[RDT_NUM_RESOURCES - 1].r_resctrl;	      \
439	     r = resctrl_inc(r))
440
441#define for_each_capable_rdt_resource(r)				      \
442	for_each_rdt_resource(r)					      \
 
443		if (r->alloc_capable || r->mon_capable)
444
445#define for_each_alloc_capable_rdt_resource(r)				      \
446	for_each_rdt_resource(r)					      \
 
447		if (r->alloc_capable)
448
449#define for_each_mon_capable_rdt_resource(r)				      \
450	for_each_rdt_resource(r)					      \
 
451		if (r->mon_capable)
452
 
 
 
 
 
 
 
 
 
 
453/* CPUID.(EAX=10H, ECX=ResID=1).EAX */
454union cpuid_0x10_1_eax {
455	struct {
456		unsigned int cbm_len:5;
457	} split;
458	unsigned int full;
459};
460
461/* CPUID.(EAX=10H, ECX=ResID=3).EAX */
462union cpuid_0x10_3_eax {
463	struct {
464		unsigned int max_delay:12;
465	} split;
466	unsigned int full;
467};
468
469/* CPUID.(EAX=10H, ECX=ResID).EDX */
470union cpuid_0x10_x_edx {
471	struct {
472		unsigned int cos_max:16;
473	} split;
474	unsigned int full;
475};
476
477void rdt_last_cmd_clear(void);
478void rdt_last_cmd_puts(const char *s);
479__printf(1, 2)
480void rdt_last_cmd_printf(const char *fmt, ...);
481
482void rdt_ctrl_update(void *arg);
483struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
484void rdtgroup_kn_unlock(struct kernfs_node *kn);
485int rdtgroup_kn_mode_restrict(struct rdtgroup *r, const char *name);
486int rdtgroup_kn_mode_restore(struct rdtgroup *r, const char *name,
487			     umode_t mask);
488struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
489				   struct list_head **pos);
490ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
491				char *buf, size_t nbytes, loff_t off);
492int rdtgroup_schemata_show(struct kernfs_open_file *of,
493			   struct seq_file *s, void *v);
494bool rdtgroup_cbm_overlaps(struct resctrl_schema *s, struct rdt_domain *d,
495			   unsigned long cbm, int closid, bool exclusive);
496unsigned int rdtgroup_cbm_to_size(struct rdt_resource *r, struct rdt_domain *d,
497				  unsigned long cbm);
498enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
499int rdtgroup_tasks_assigned(struct rdtgroup *r);
500int rdtgroup_locksetup_enter(struct rdtgroup *rdtgrp);
501int rdtgroup_locksetup_exit(struct rdtgroup *rdtgrp);
502bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain *d, unsigned long cbm);
503bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain *d);
504int rdt_pseudo_lock_init(void);
505void rdt_pseudo_lock_release(void);
506int rdtgroup_pseudo_lock_create(struct rdtgroup *rdtgrp);
507void rdtgroup_pseudo_lock_remove(struct rdtgroup *rdtgrp);
508struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
 
509int closids_supported(void);
510void closid_free(int closid);
511int alloc_rmid(void);
512void free_rmid(u32 rmid);
513int rdt_get_mon_l3_config(struct rdt_resource *r);
514void mon_event_count(void *info);
515int rdtgroup_mondata_show(struct seq_file *m, void *arg);
516void mon_event_read(struct rmid_read *rr, struct rdt_resource *r,
517		    struct rdt_domain *d, struct rdtgroup *rdtgrp,
518		    int evtid, int first);
 
 
 
519void mbm_setup_overflow_handler(struct rdt_domain *dom,
520				unsigned long delay_ms);
521void mbm_handle_overflow(struct work_struct *work);
522void __init intel_rdt_mbm_apply_quirk(void);
523bool is_mba_sc(struct rdt_resource *r);
 
 
524void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
525void cqm_handle_limbo(struct work_struct *work);
526bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
527void __check_limbo(struct rdt_domain *d, bool force_free);
528void rdt_domain_reconfigure_cdp(struct rdt_resource *r);
529void __init thread_throttle_mode_init(void);
530
531#endif /* _ASM_X86_RESCTRL_INTERNAL_H */