xref: /linux-6.15/include/linux/resctrl.h (revision 9be68b14)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _RESCTRL_H
3 #define _RESCTRL_H
4 
5 #include <linux/cacheinfo.h>
6 #include <linux/kernel.h>
7 #include <linux/list.h>
8 #include <linux/pid.h>
9 #include <linux/resctrl_types.h>
10 
11 /* CLOSID, RMID value used by the default control group */
12 #define RESCTRL_RESERVED_CLOSID		0
13 #define RESCTRL_RESERVED_RMID		0
14 
15 #define RESCTRL_PICK_ANY_CPU		-1
16 
17 #ifdef CONFIG_PROC_CPU_RESCTRL
18 
19 int proc_resctrl_show(struct seq_file *m,
20 		      struct pid_namespace *ns,
21 		      struct pid *pid,
22 		      struct task_struct *tsk);
23 
24 #endif
25 
26 /* max value for struct rdt_domain's mbps_val */
27 #define MBA_MAX_MBPS   U32_MAX
28 
29 /**
30  * enum resctrl_conf_type - The type of configuration.
31  * @CDP_NONE:	No prioritisation, both code and data are controlled or monitored.
32  * @CDP_CODE:	Configuration applies to instruction fetches.
33  * @CDP_DATA:	Configuration applies to reads and writes.
34  */
35 enum resctrl_conf_type {
36 	CDP_NONE,
37 	CDP_CODE,
38 	CDP_DATA,
39 };
40 
41 #define CDP_NUM_TYPES	(CDP_DATA + 1)
42 
43 /**
44  * struct resctrl_staged_config - parsed configuration to be applied
45  * @new_ctrl:		new ctrl value to be loaded
46  * @have_new_ctrl:	whether the user provided new_ctrl is valid
47  */
48 struct resctrl_staged_config {
49 	u32			new_ctrl;
50 	bool			have_new_ctrl;
51 };
52 
53 enum resctrl_domain_type {
54 	RESCTRL_CTRL_DOMAIN,
55 	RESCTRL_MON_DOMAIN,
56 };
57 
58 /**
59  * struct rdt_domain_hdr - common header for different domain types
60  * @list:		all instances of this resource
61  * @id:			unique id for this instance
62  * @type:		type of this instance
63  * @cpu_mask:		which CPUs share this resource
64  */
65 struct rdt_domain_hdr {
66 	struct list_head		list;
67 	int				id;
68 	enum resctrl_domain_type	type;
69 	struct cpumask			cpu_mask;
70 };
71 
72 /**
73  * struct rdt_ctrl_domain - group of CPUs sharing a resctrl control resource
74  * @hdr:		common header for different domain types
75  * @plr:		pseudo-locked region (if any) associated with domain
76  * @staged_config:	parsed configuration to be applied
77  * @mbps_val:		When mba_sc is enabled, this holds the array of user
78  *			specified control values for mba_sc in MBps, indexed
79  *			by closid
80  */
81 struct rdt_ctrl_domain {
82 	struct rdt_domain_hdr		hdr;
83 	struct pseudo_lock_region	*plr;
84 	struct resctrl_staged_config	staged_config[CDP_NUM_TYPES];
85 	u32				*mbps_val;
86 };
87 
88 /**
89  * struct rdt_mon_domain - group of CPUs sharing a resctrl monitor resource
90  * @hdr:		common header for different domain types
91  * @ci:			cache info for this domain
92  * @rmid_busy_llc:	bitmap of which limbo RMIDs are above threshold
93  * @mbm_total:		saved state for MBM total bandwidth
94  * @mbm_local:		saved state for MBM local bandwidth
95  * @mbm_over:		worker to periodically read MBM h/w counters
96  * @cqm_limbo:		worker to periodically read CQM h/w counters
97  * @mbm_work_cpu:	worker CPU for MBM h/w counters
98  * @cqm_work_cpu:	worker CPU for CQM h/w counters
99  */
100 struct rdt_mon_domain {
101 	struct rdt_domain_hdr		hdr;
102 	struct cacheinfo		*ci;
103 	unsigned long			*rmid_busy_llc;
104 	struct mbm_state		*mbm_total;
105 	struct mbm_state		*mbm_local;
106 	struct delayed_work		mbm_over;
107 	struct delayed_work		cqm_limbo;
108 	int				mbm_work_cpu;
109 	int				cqm_work_cpu;
110 };
111 
112 /**
113  * struct resctrl_cache - Cache allocation related data
114  * @cbm_len:		Length of the cache bit mask
115  * @min_cbm_bits:	Minimum number of consecutive bits to be set.
116  *			The value 0 means the architecture can support
117  *			zero CBM.
118  * @shareable_bits:	Bitmask of shareable resource with other
119  *			executing entities
120  * @arch_has_sparse_bitmasks:	True if a bitmask like f00f is valid.
121  * @arch_has_per_cpu_cfg:	True if QOS_CFG register for this cache
122  *				level has CPU scope.
123  */
124 struct resctrl_cache {
125 	unsigned int	cbm_len;
126 	unsigned int	min_cbm_bits;
127 	unsigned int	shareable_bits;
128 	bool		arch_has_sparse_bitmasks;
129 	bool		arch_has_per_cpu_cfg;
130 };
131 
132 /**
133  * enum membw_throttle_mode - System's memory bandwidth throttling mode
134  * @THREAD_THROTTLE_UNDEFINED:	Not relevant to the system
135  * @THREAD_THROTTLE_MAX:	Memory bandwidth is throttled at the core
136  *				always using smallest bandwidth percentage
137  *				assigned to threads, aka "max throttling"
138  * @THREAD_THROTTLE_PER_THREAD:	Memory bandwidth is throttled at the thread
139  */
140 enum membw_throttle_mode {
141 	THREAD_THROTTLE_UNDEFINED = 0,
142 	THREAD_THROTTLE_MAX,
143 	THREAD_THROTTLE_PER_THREAD,
144 };
145 
146 /**
147  * struct resctrl_membw - Memory bandwidth allocation related data
148  * @min_bw:		Minimum memory bandwidth percentage user can request
149  * @max_bw:		Maximum memory bandwidth value, used as the reset value
150  * @bw_gran:		Granularity at which the memory bandwidth is allocated
151  * @delay_linear:	True if memory B/W delay is in linear scale
152  * @arch_needs_linear:	True if we can't configure non-linear resources
153  * @throttle_mode:	Bandwidth throttling mode when threads request
154  *			different memory bandwidths
155  * @mba_sc:		True if MBA software controller(mba_sc) is enabled
156  * @mb_map:		Mapping of memory B/W percentage to memory B/W delay
157  */
158 struct resctrl_membw {
159 	u32				min_bw;
160 	u32				max_bw;
161 	u32				bw_gran;
162 	u32				delay_linear;
163 	bool				arch_needs_linear;
164 	enum membw_throttle_mode	throttle_mode;
165 	bool				mba_sc;
166 	u32				*mb_map;
167 };
168 
169 struct resctrl_schema;
170 
171 enum resctrl_scope {
172 	RESCTRL_L2_CACHE = 2,
173 	RESCTRL_L3_CACHE = 3,
174 	RESCTRL_L3_NODE,
175 };
176 
177 /**
178  * enum resctrl_schema_fmt - The format user-space provides for a schema.
179  * @RESCTRL_SCHEMA_BITMAP:	The schema is a bitmap in hex.
180  * @RESCTRL_SCHEMA_RANGE:	The schema is a decimal number.
181  */
182 enum resctrl_schema_fmt {
183 	RESCTRL_SCHEMA_BITMAP,
184 	RESCTRL_SCHEMA_RANGE,
185 };
186 
187 /**
188  * struct rdt_resource - attributes of a resctrl resource
189  * @rid:		The index of the resource
190  * @alloc_capable:	Is allocation available on this machine
191  * @mon_capable:	Is monitor feature available on this machine
192  * @num_rmid:		Number of RMIDs available
193  * @ctrl_scope:		Scope of this resource for control functions
194  * @mon_scope:		Scope of this resource for monitor functions
195  * @cache:		Cache allocation related data
196  * @membw:		If the component has bandwidth controls, their properties.
197  * @ctrl_domains:	RCU list of all control domains for this resource
198  * @mon_domains:	RCU list of all monitor domains for this resource
199  * @name:		Name to use in "schemata" file.
200  * @schema_fmt:		Which format string and parser is used for this schema.
201  * @evt_list:		List of monitoring events
202  * @cdp_capable:	Is the CDP feature available on this resource
203  */
204 struct rdt_resource {
205 	int			rid;
206 	bool			alloc_capable;
207 	bool			mon_capable;
208 	int			num_rmid;
209 	enum resctrl_scope	ctrl_scope;
210 	enum resctrl_scope	mon_scope;
211 	struct resctrl_cache	cache;
212 	struct resctrl_membw	membw;
213 	struct list_head	ctrl_domains;
214 	struct list_head	mon_domains;
215 	char			*name;
216 	enum resctrl_schema_fmt	schema_fmt;
217 	struct list_head	evt_list;
218 	bool			cdp_capable;
219 };
220 
221 /*
222  * Get the resource that exists at this level. If the level is not supported
223  * a dummy/not-capable resource can be returned. Levels >= RDT_NUM_RESOURCES
224  * will return NULL.
225  */
226 struct rdt_resource *resctrl_arch_get_resource(enum resctrl_res_level l);
227 
228 /**
229  * struct resctrl_schema - configuration abilities of a resource presented to
230  *			   user-space
231  * @list:	Member of resctrl_schema_all.
232  * @name:	The name to use in the "schemata" file.
233  * @fmt_str:	Format string to show domain value.
234  * @conf_type:	Whether this schema is specific to code/data.
235  * @res:	The resource structure exported by the architecture to describe
236  *		the hardware that is configured by this schema.
237  * @num_closid:	The number of closid that can be used with this schema. When
238  *		features like CDP are enabled, this will be lower than the
239  *		hardware supports for the resource.
240  */
241 struct resctrl_schema {
242 	struct list_head		list;
243 	char				name[8];
244 	const char			*fmt_str;
245 	enum resctrl_conf_type		conf_type;
246 	struct rdt_resource		*res;
247 	u32				num_closid;
248 };
249 
250 struct resctrl_cpu_defaults {
251 	u32 closid;
252 	u32 rmid;
253 };
254 
255 /**
256  * resctrl_arch_sync_cpu_closid_rmid() - Refresh this CPU's CLOSID and RMID.
257  *					 Call via IPI.
258  * @info:	If non-NULL, a pointer to a struct resctrl_cpu_defaults
259  *		specifying the new CLOSID and RMID for tasks in the default
260  *		resctrl ctrl and mon group when running on this CPU.  If NULL,
261  *		this CPU is not re-assigned to a different default group.
262  *
263  * Propagates reassignment of CPUs and/or tasks to different resctrl groups
264  * when requested by the resctrl core code.
265  *
266  * This function records the per-cpu defaults specified by @info (if any),
267  * and then reconfigures the CPU's hardware CLOSID and RMID for subsequent
268  * execution based on @current, in the same way as during a task switch.
269  */
270 void resctrl_arch_sync_cpu_closid_rmid(void *info);
271 
272 /**
273  * resctrl_get_default_ctrl() - Return the default control value for this
274  *                              resource.
275  * @r:		The resource whose default control type is queried.
276  */
277 static inline u32 resctrl_get_default_ctrl(struct rdt_resource *r)
278 {
279 	switch (r->schema_fmt) {
280 	case RESCTRL_SCHEMA_BITMAP:
281 		return BIT_MASK(r->cache.cbm_len) - 1;
282 	case RESCTRL_SCHEMA_RANGE:
283 		return r->membw.max_bw;
284 	}
285 
286 	return WARN_ON_ONCE(1);
287 }
288 
289 /* The number of closid supported by this resource regardless of CDP */
290 u32 resctrl_arch_get_num_closid(struct rdt_resource *r);
291 u32 resctrl_arch_system_num_rmid_idx(void);
292 int resctrl_arch_update_domains(struct rdt_resource *r, u32 closid);
293 
294 /*
295  * Update the ctrl_val and apply this config right now.
296  * Must be called on one of the domain's CPUs.
297  */
298 int resctrl_arch_update_one(struct rdt_resource *r, struct rdt_ctrl_domain *d,
299 			    u32 closid, enum resctrl_conf_type t, u32 cfg_val);
300 
301 u32 resctrl_arch_get_config(struct rdt_resource *r, struct rdt_ctrl_domain *d,
302 			    u32 closid, enum resctrl_conf_type type);
303 int resctrl_online_ctrl_domain(struct rdt_resource *r, struct rdt_ctrl_domain *d);
304 int resctrl_online_mon_domain(struct rdt_resource *r, struct rdt_mon_domain *d);
305 void resctrl_offline_ctrl_domain(struct rdt_resource *r, struct rdt_ctrl_domain *d);
306 void resctrl_offline_mon_domain(struct rdt_resource *r, struct rdt_mon_domain *d);
307 void resctrl_online_cpu(unsigned int cpu);
308 void resctrl_offline_cpu(unsigned int cpu);
309 
310 /**
311  * resctrl_arch_rmid_read() - Read the eventid counter corresponding to rmid
312  *			      for this resource and domain.
313  * @r:			resource that the counter should be read from.
314  * @d:			domain that the counter should be read from.
315  * @closid:		closid that matches the rmid. Depending on the architecture, the
316  *			counter may match traffic of both @closid and @rmid, or @rmid
317  *			only.
318  * @rmid:		rmid of the counter to read.
319  * @eventid:		eventid to read, e.g. L3 occupancy.
320  * @val:		result of the counter read in bytes.
321  * @arch_mon_ctx:	An architecture specific value from
322  *			resctrl_arch_mon_ctx_alloc(), for MPAM this identifies
323  *			the hardware monitor allocated for this read request.
324  *
325  * Some architectures need to sleep when first programming some of the counters.
326  * (specifically: arm64's MPAM cache occupancy counters can return 'not ready'
327  *  for a short period of time). Call from a non-migrateable process context on
328  * a CPU that belongs to domain @d. e.g. use smp_call_on_cpu() or
329  * schedule_work_on(). This function can be called with interrupts masked,
330  * e.g. using smp_call_function_any(), but may consistently return an error.
331  *
332  * Return:
333  * 0 on success, or -EIO, -EINVAL etc on error.
334  */
335 int resctrl_arch_rmid_read(struct rdt_resource *r, struct rdt_mon_domain *d,
336 			   u32 closid, u32 rmid, enum resctrl_event_id eventid,
337 			   u64 *val, void *arch_mon_ctx);
338 
339 /**
340  * resctrl_arch_rmid_read_context_check()  - warn about invalid contexts
341  *
342  * When built with CONFIG_DEBUG_ATOMIC_SLEEP generate a warning when
343  * resctrl_arch_rmid_read() is called with preemption disabled.
344  *
345  * The contract with resctrl_arch_rmid_read() is that if interrupts
346  * are unmasked, it can sleep. This allows NOHZ_FULL systems to use an
347  * IPI, (and fail if the call needed to sleep), while most of the time
348  * the work is scheduled, allowing the call to sleep.
349  */
350 static inline void resctrl_arch_rmid_read_context_check(void)
351 {
352 	if (!irqs_disabled())
353 		might_sleep();
354 }
355 
356 /**
357  * resctrl_find_domain() - Search for a domain id in a resource domain list.
358  * @h:		The domain list to search.
359  * @id:		The domain id to search for.
360  * @pos:	A pointer to position in the list id should be inserted.
361  *
362  * Search the domain list to find the domain id. If the domain id is
363  * found, return the domain. NULL otherwise.  If the domain id is not
364  * found (and NULL returned) then the first domain with id bigger than
365  * the input id can be returned to the caller via @pos.
366  */
367 struct rdt_domain_hdr *resctrl_find_domain(struct list_head *h, int id,
368 					   struct list_head **pos);
369 
370 /**
371  * resctrl_arch_reset_rmid() - Reset any private state associated with rmid
372  *			       and eventid.
373  * @r:		The domain's resource.
374  * @d:		The rmid's domain.
375  * @closid:	closid that matches the rmid. Depending on the architecture, the
376  *		counter may match traffic of both @closid and @rmid, or @rmid only.
377  * @rmid:	The rmid whose counter values should be reset.
378  * @eventid:	The eventid whose counter values should be reset.
379  *
380  * This can be called from any CPU.
381  */
382 void resctrl_arch_reset_rmid(struct rdt_resource *r, struct rdt_mon_domain *d,
383 			     u32 closid, u32 rmid,
384 			     enum resctrl_event_id eventid);
385 
386 /**
387  * resctrl_arch_reset_rmid_all() - Reset all private state associated with
388  *				   all rmids and eventids.
389  * @r:		The resctrl resource.
390  * @d:		The domain for which all architectural counter state will
391  *		be cleared.
392  *
393  * This can be called from any CPU.
394  */
395 void resctrl_arch_reset_rmid_all(struct rdt_resource *r, struct rdt_mon_domain *d);
396 
397 /**
398  * resctrl_arch_reset_all_ctrls() - Reset the control for each CLOSID to its
399  *				    default.
400  * @r:		The resctrl resource to reset.
401  *
402  * This can be called from any CPU.
403  */
404 void resctrl_arch_reset_all_ctrls(struct rdt_resource *r);
405 
406 extern unsigned int resctrl_rmid_realloc_threshold;
407 extern unsigned int resctrl_rmid_realloc_limit;
408 
409 int __init resctrl_init(void);
410 void __exit resctrl_exit(void);
411 
412 #endif /* _RESCTRL_H */
413