xref: /linux-6.15/include/linux/resctrl.h (revision c103d4d4)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _RESCTRL_H
3 #define _RESCTRL_H
4 
5 #include <linux/kernel.h>
6 #include <linux/list.h>
7 #include <linux/pid.h>
8 
9 /* CLOSID, RMID value used by the default control group */
10 #define RESCTRL_RESERVED_CLOSID		0
11 #define RESCTRL_RESERVED_RMID		0
12 
13 #define RESCTRL_PICK_ANY_CPU		-1
14 
15 #ifdef CONFIG_PROC_CPU_RESCTRL
16 
17 int proc_resctrl_show(struct seq_file *m,
18 		      struct pid_namespace *ns,
19 		      struct pid *pid,
20 		      struct task_struct *tsk);
21 
22 #endif
23 
24 /* max value for struct rdt_domain's mbps_val */
25 #define MBA_MAX_MBPS   U32_MAX
26 
27 /**
28  * enum resctrl_conf_type - The type of configuration.
29  * @CDP_NONE:	No prioritisation, both code and data are controlled or monitored.
30  * @CDP_CODE:	Configuration applies to instruction fetches.
31  * @CDP_DATA:	Configuration applies to reads and writes.
32  */
33 enum resctrl_conf_type {
34 	CDP_NONE,
35 	CDP_CODE,
36 	CDP_DATA,
37 };
38 
39 #define CDP_NUM_TYPES	(CDP_DATA + 1)
40 
41 /*
42  * Event IDs, the values match those used to program IA32_QM_EVTSEL before
43  * reading IA32_QM_CTR on RDT systems.
44  */
45 enum resctrl_event_id {
46 	QOS_L3_OCCUP_EVENT_ID		= 0x01,
47 	QOS_L3_MBM_TOTAL_EVENT_ID	= 0x02,
48 	QOS_L3_MBM_LOCAL_EVENT_ID	= 0x03,
49 };
50 
51 /**
52  * struct resctrl_staged_config - parsed configuration to be applied
53  * @new_ctrl:		new ctrl value to be loaded
54  * @have_new_ctrl:	whether the user provided new_ctrl is valid
55  */
56 struct resctrl_staged_config {
57 	u32			new_ctrl;
58 	bool			have_new_ctrl;
59 };
60 
61 /**
62  * struct rdt_domain_hdr - common header for different domain types
63  * @list:		all instances of this resource
64  * @id:			unique id for this instance
65  * @cpu_mask:		which CPUs share this resource
66  */
67 struct rdt_domain_hdr {
68 	struct list_head		list;
69 	int				id;
70 	struct cpumask			cpu_mask;
71 };
72 
73 /**
74  * struct rdt_domain - group of CPUs sharing a resctrl resource
75  * @hdr:		common header for different domain types
76  * @rmid_busy_llc:	bitmap of which limbo RMIDs are above threshold
77  * @mbm_total:		saved state for MBM total bandwidth
78  * @mbm_local:		saved state for MBM local bandwidth
79  * @mbm_over:		worker to periodically read MBM h/w counters
80  * @cqm_limbo:		worker to periodically read CQM h/w counters
81  * @mbm_work_cpu:	worker CPU for MBM h/w counters
82  * @cqm_work_cpu:	worker CPU for CQM h/w counters
83  * @plr:		pseudo-locked region (if any) associated with domain
84  * @staged_config:	parsed configuration to be applied
85  * @mbps_val:		When mba_sc is enabled, this holds the array of user
86  *			specified control values for mba_sc in MBps, indexed
87  *			by closid
88  */
89 struct rdt_domain {
90 	struct rdt_domain_hdr		hdr;
91 	unsigned long			*rmid_busy_llc;
92 	struct mbm_state		*mbm_total;
93 	struct mbm_state		*mbm_local;
94 	struct delayed_work		mbm_over;
95 	struct delayed_work		cqm_limbo;
96 	int				mbm_work_cpu;
97 	int				cqm_work_cpu;
98 	struct pseudo_lock_region	*plr;
99 	struct resctrl_staged_config	staged_config[CDP_NUM_TYPES];
100 	u32				*mbps_val;
101 };
102 
103 /**
104  * struct resctrl_cache - Cache allocation related data
105  * @cbm_len:		Length of the cache bit mask
106  * @min_cbm_bits:	Minimum number of consecutive bits to be set.
107  *			The value 0 means the architecture can support
108  *			zero CBM.
109  * @shareable_bits:	Bitmask of shareable resource with other
110  *			executing entities
111  * @arch_has_sparse_bitmasks:	True if a bitmask like f00f is valid.
112  * @arch_has_per_cpu_cfg:	True if QOS_CFG register for this cache
113  *				level has CPU scope.
114  */
115 struct resctrl_cache {
116 	unsigned int	cbm_len;
117 	unsigned int	min_cbm_bits;
118 	unsigned int	shareable_bits;
119 	bool		arch_has_sparse_bitmasks;
120 	bool		arch_has_per_cpu_cfg;
121 };
122 
123 /**
124  * enum membw_throttle_mode - System's memory bandwidth throttling mode
125  * @THREAD_THROTTLE_UNDEFINED:	Not relevant to the system
126  * @THREAD_THROTTLE_MAX:	Memory bandwidth is throttled at the core
127  *				always using smallest bandwidth percentage
128  *				assigned to threads, aka "max throttling"
129  * @THREAD_THROTTLE_PER_THREAD:	Memory bandwidth is throttled at the thread
130  */
131 enum membw_throttle_mode {
132 	THREAD_THROTTLE_UNDEFINED = 0,
133 	THREAD_THROTTLE_MAX,
134 	THREAD_THROTTLE_PER_THREAD,
135 };
136 
137 /**
138  * struct resctrl_membw - Memory bandwidth allocation related data
139  * @min_bw:		Minimum memory bandwidth percentage user can request
140  * @bw_gran:		Granularity at which the memory bandwidth is allocated
141  * @delay_linear:	True if memory B/W delay is in linear scale
142  * @arch_needs_linear:	True if we can't configure non-linear resources
143  * @throttle_mode:	Bandwidth throttling mode when threads request
144  *			different memory bandwidths
145  * @mba_sc:		True if MBA software controller(mba_sc) is enabled
146  * @mb_map:		Mapping of memory B/W percentage to memory B/W delay
147  */
148 struct resctrl_membw {
149 	u32				min_bw;
150 	u32				bw_gran;
151 	u32				delay_linear;
152 	bool				arch_needs_linear;
153 	enum membw_throttle_mode	throttle_mode;
154 	bool				mba_sc;
155 	u32				*mb_map;
156 };
157 
158 struct rdt_parse_data;
159 struct resctrl_schema;
160 
161 enum resctrl_scope {
162 	RESCTRL_L2_CACHE = 2,
163 	RESCTRL_L3_CACHE = 3,
164 };
165 
166 /**
167  * struct rdt_resource - attributes of a resctrl resource
168  * @rid:		The index of the resource
169  * @alloc_capable:	Is allocation available on this machine
170  * @mon_capable:	Is monitor feature available on this machine
171  * @num_rmid:		Number of RMIDs available
172  * @scope:		Scope of this resource
173  * @cache:		Cache allocation related data
174  * @membw:		If the component has bandwidth controls, their properties.
175  * @domains:		RCU list of all domains for this resource
176  * @name:		Name to use in "schemata" file.
177  * @data_width:		Character width of data when displaying
178  * @default_ctrl:	Specifies default cache cbm or memory B/W percent.
179  * @format_str:		Per resource format string to show domain value
180  * @parse_ctrlval:	Per resource function pointer to parse control values
181  * @evt_list:		List of monitoring events
182  * @fflags:		flags to choose base and info files
183  * @cdp_capable:	Is the CDP feature available on this resource
184  */
185 struct rdt_resource {
186 	int			rid;
187 	bool			alloc_capable;
188 	bool			mon_capable;
189 	int			num_rmid;
190 	enum resctrl_scope	scope;
191 	struct resctrl_cache	cache;
192 	struct resctrl_membw	membw;
193 	struct list_head	domains;
194 	char			*name;
195 	int			data_width;
196 	u32			default_ctrl;
197 	const char		*format_str;
198 	int			(*parse_ctrlval)(struct rdt_parse_data *data,
199 						 struct resctrl_schema *s,
200 						 struct rdt_domain *d);
201 	struct list_head	evt_list;
202 	unsigned long		fflags;
203 	bool			cdp_capable;
204 };
205 
206 /**
207  * struct resctrl_schema - configuration abilities of a resource presented to
208  *			   user-space
209  * @list:	Member of resctrl_schema_all.
210  * @name:	The name to use in the "schemata" file.
211  * @conf_type:	Whether this schema is specific to code/data.
212  * @res:	The resource structure exported by the architecture to describe
213  *		the hardware that is configured by this schema.
214  * @num_closid:	The number of closid that can be used with this schema. When
215  *		features like CDP are enabled, this will be lower than the
216  *		hardware supports for the resource.
217  */
218 struct resctrl_schema {
219 	struct list_head		list;
220 	char				name[8];
221 	enum resctrl_conf_type		conf_type;
222 	struct rdt_resource		*res;
223 	u32				num_closid;
224 };
225 
226 /* The number of closid supported by this resource regardless of CDP */
227 u32 resctrl_arch_get_num_closid(struct rdt_resource *r);
228 int resctrl_arch_update_domains(struct rdt_resource *r, u32 closid);
229 
230 /*
231  * Update the ctrl_val and apply this config right now.
232  * Must be called on one of the domain's CPUs.
233  */
234 int resctrl_arch_update_one(struct rdt_resource *r, struct rdt_domain *d,
235 			    u32 closid, enum resctrl_conf_type t, u32 cfg_val);
236 
237 u32 resctrl_arch_get_config(struct rdt_resource *r, struct rdt_domain *d,
238 			    u32 closid, enum resctrl_conf_type type);
239 int resctrl_online_domain(struct rdt_resource *r, struct rdt_domain *d);
240 void resctrl_offline_domain(struct rdt_resource *r, struct rdt_domain *d);
241 void resctrl_online_cpu(unsigned int cpu);
242 void resctrl_offline_cpu(unsigned int cpu);
243 
244 /**
245  * resctrl_arch_rmid_read() - Read the eventid counter corresponding to rmid
246  *			      for this resource and domain.
247  * @r:			resource that the counter should be read from.
248  * @d:			domain that the counter should be read from.
249  * @closid:		closid that matches the rmid. Depending on the architecture, the
250  *			counter may match traffic of both @closid and @rmid, or @rmid
251  *			only.
252  * @rmid:		rmid of the counter to read.
253  * @eventid:		eventid to read, e.g. L3 occupancy.
254  * @val:		result of the counter read in bytes.
255  * @arch_mon_ctx:	An architecture specific value from
256  *			resctrl_arch_mon_ctx_alloc(), for MPAM this identifies
257  *			the hardware monitor allocated for this read request.
258  *
259  * Some architectures need to sleep when first programming some of the counters.
260  * (specifically: arm64's MPAM cache occupancy counters can return 'not ready'
261  *  for a short period of time). Call from a non-migrateable process context on
262  * a CPU that belongs to domain @d. e.g. use smp_call_on_cpu() or
263  * schedule_work_on(). This function can be called with interrupts masked,
264  * e.g. using smp_call_function_any(), but may consistently return an error.
265  *
266  * Return:
267  * 0 on success, or -EIO, -EINVAL etc on error.
268  */
269 int resctrl_arch_rmid_read(struct rdt_resource *r, struct rdt_domain *d,
270 			   u32 closid, u32 rmid, enum resctrl_event_id eventid,
271 			   u64 *val, void *arch_mon_ctx);
272 
273 /**
274  * resctrl_arch_rmid_read_context_check()  - warn about invalid contexts
275  *
276  * When built with CONFIG_DEBUG_ATOMIC_SLEEP generate a warning when
277  * resctrl_arch_rmid_read() is called with preemption disabled.
278  *
279  * The contract with resctrl_arch_rmid_read() is that if interrupts
280  * are unmasked, it can sleep. This allows NOHZ_FULL systems to use an
281  * IPI, (and fail if the call needed to sleep), while most of the time
282  * the work is scheduled, allowing the call to sleep.
283  */
284 static inline void resctrl_arch_rmid_read_context_check(void)
285 {
286 	if (!irqs_disabled())
287 		might_sleep();
288 }
289 
290 /**
291  * resctrl_arch_reset_rmid() - Reset any private state associated with rmid
292  *			       and eventid.
293  * @r:		The domain's resource.
294  * @d:		The rmid's domain.
295  * @closid:	closid that matches the rmid. Depending on the architecture, the
296  *		counter may match traffic of both @closid and @rmid, or @rmid only.
297  * @rmid:	The rmid whose counter values should be reset.
298  * @eventid:	The eventid whose counter values should be reset.
299  *
300  * This can be called from any CPU.
301  */
302 void resctrl_arch_reset_rmid(struct rdt_resource *r, struct rdt_domain *d,
303 			     u32 closid, u32 rmid,
304 			     enum resctrl_event_id eventid);
305 
306 /**
307  * resctrl_arch_reset_rmid_all() - Reset all private state associated with
308  *				   all rmids and eventids.
309  * @r:		The resctrl resource.
310  * @d:		The domain for which all architectural counter state will
311  *		be cleared.
312  *
313  * This can be called from any CPU.
314  */
315 void resctrl_arch_reset_rmid_all(struct rdt_resource *r, struct rdt_domain *d);
316 
317 extern unsigned int resctrl_rmid_realloc_threshold;
318 extern unsigned int resctrl_rmid_realloc_limit;
319 
320 #endif /* _RESCTRL_H */
321