1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_INTEL_RDT_H
3 #define _ASM_X86_INTEL_RDT_H
5 #include <linux/sched.h>
6 #include <linux/kernfs.h>
7 #include <linux/jump_label.h>
9 #define IA32_L3_QOS_CFG 0xc81
10 #define IA32_L2_QOS_CFG 0xc82
11 #define IA32_L3_CBM_BASE 0xc90
12 #define IA32_L2_CBM_BASE 0xd10
13 #define IA32_MBA_THRTL_BASE 0xd50
15 #define L3_QOS_CDP_ENABLE 0x01ULL
17 #define L2_QOS_CDP_ENABLE 0x01ULL
20 * Event IDs are used to program IA32_QM_EVTSEL before reading event
21 * counter from IA32_QM_CTR
23 #define QOS_L3_OCCUP_EVENT_ID 0x01
24 #define QOS_L3_MBM_TOTAL_EVENT_ID 0x02
25 #define QOS_L3_MBM_LOCAL_EVENT_ID 0x03
27 #define CQM_LIMBOCHECK_INTERVAL 1000
29 #define MBM_CNTR_WIDTH 24
30 #define MBM_OVERFLOW_INTERVAL 1000
31 #define MAX_MBA_BW 100u
33 #define RMID_VAL_ERROR BIT_ULL(63)
34 #define RMID_VAL_UNAVAIL BIT_ULL(62)
36 DECLARE_STATIC_KEY_FALSE(rdt_enable_key);
39 * struct mon_evt - Entry in the event list of a resource
41 * @name: name of the event
46 struct list_head list;
50 * struct mon_data_bits - Monitoring details for each event file
51 * @rid: Resource id associated with the event file.
52 * @evtid: Event id associated with the event file
53 * @domid: The domain to which the event file belongs
58 unsigned int rid : 10;
59 unsigned int evtid : 8;
60 unsigned int domid : 14;
65 struct rdtgroup *rgrp;
72 extern unsigned int intel_cqm_threshold;
73 extern bool rdt_alloc_capable;
74 extern bool rdt_mon_capable;
75 extern unsigned int rdt_mon_features;
84 * enum rdtgrp_mode - Mode of a RDT resource group
85 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
86 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
88 * The mode of a resource group enables control over the allowed overlap
89 * between allocations associated with different resource groups (classes
90 * of service). User is able to modify the mode of a resource group by
91 * writing to the "mode" resctrl file associated with the resource group.
94 RDT_MODE_SHAREABLE = 0,
102 * struct mongroup - store mon group's data in resctrl fs.
103 * @mon_data_kn kernlfs node for the mon_data directory
104 * @parent: parent rdtgrp
105 * @crdtgrp_list: child rdtgroup node list
106 * @rmid: rmid for this rdtgroup
109 struct kernfs_node *mon_data_kn;
110 struct rdtgroup *parent;
111 struct list_head crdtgrp_list;
116 * struct rdtgroup - store rdtgroup's data in resctrl file system.
118 * @rdtgroup_list: linked list for all rdtgroups
119 * @closid: closid for this rdtgroup
120 * @cpu_mask: CPUs assigned to this rdtgroup
121 * @flags: status bits
122 * @waitcount: how many cpus expect to find this
123 * group when they acquire rdtgroup_mutex
124 * @type: indicates type of this rdtgroup - either
125 * monitor only or ctrl_mon group
126 * @mon: mongroup related data
127 * @mode: mode of resource group
130 struct kernfs_node *kn;
131 struct list_head rdtgroup_list;
133 struct cpumask cpu_mask;
136 enum rdt_group_type type;
138 enum rdtgrp_mode mode;
142 #define RDT_DELETED 1
145 #define RFTYPE_FLAGS_CPUS_LIST 1
148 * Define the file type flags for base and info directories.
150 #define RFTYPE_INFO BIT(0)
151 #define RFTYPE_BASE BIT(1)
152 #define RF_CTRLSHIFT 4
153 #define RF_MONSHIFT 5
154 #define RF_TOPSHIFT 6
155 #define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
156 #define RFTYPE_MON BIT(RF_MONSHIFT)
157 #define RFTYPE_TOP BIT(RF_TOPSHIFT)
158 #define RFTYPE_RES_CACHE BIT(8)
159 #define RFTYPE_RES_MB BIT(9)
160 #define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
161 #define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
162 #define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
163 #define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
165 /* List of all resource groups */
166 extern struct list_head rdt_all_groups;
168 extern int max_name_width, max_data_width;
170 int __init rdtgroup_init(void);
173 * struct rftype - describe each file in the resctrl file system
176 * @kf_ops: File operations
177 * @flags: File specific RFTYPE_FLAGS_* flags
178 * @fflags: File specific RF_* or RFTYPE_* flags
179 * @seq_show: Show content of the file
180 * @write: Write to the file
185 struct kernfs_ops *kf_ops;
187 unsigned long fflags;
189 int (*seq_show)(struct kernfs_open_file *of,
190 struct seq_file *sf, void *v);
192 * write() is the generic write callback which maps directly to
193 * kernfs write operation and overrides all other operations.
194 * Maximum write size is determined by ->max_write_len.
196 ssize_t (*write)(struct kernfs_open_file *of,
197 char *buf, size_t nbytes, loff_t off);
201 * struct mbm_state - status for each MBM counter in each domain
202 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
203 * @prev_msr Value of IA32_QM_CTR for this RMID last time we read it
204 * @chunks_bw Total local data moved. Used for bandwidth calculation
205 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting
206 * @prev_bw The most recent bandwidth in MBps
207 * @delta_bw Difference between the current and previous bandwidth
208 * @delta_comp Indicates whether to compute the delta_bw
221 * struct rdt_domain - group of cpus sharing an RDT resource
222 * @list: all instances of this resource
223 * @id: unique id for this instance
224 * @cpu_mask: which cpus share this resource
226 * bitmap of which limbo RMIDs are above threshold
227 * @mbm_total: saved state for MBM total bandwidth
228 * @mbm_local: saved state for MBM local bandwidth
229 * @mbm_over: worker to periodically read MBM h/w counters
230 * @cqm_limbo: worker to periodically read CQM h/w counters
232 * worker cpu for MBM h/w counters
234 * worker cpu for CQM h/w counters
235 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
236 * @mbps_val: When mba_sc is enabled, this holds the bandwidth in MBps
237 * @new_ctrl: new ctrl value to be loaded
238 * @have_new_ctrl: did user provide new_ctrl for this domain
241 struct list_head list;
243 struct cpumask cpu_mask;
244 unsigned long *rmid_busy_llc;
245 struct mbm_state *mbm_total;
246 struct mbm_state *mbm_local;
247 struct delayed_work mbm_over;
248 struct delayed_work cqm_limbo;
258 * struct msr_param - set a range of MSRs from a domain
259 * @res: The resource to use
260 * @low: Beginning index from base MSR
264 struct rdt_resource *res;
270 * struct rdt_cache - Cache allocation related data
271 * @cbm_len: Length of the cache bit mask
272 * @min_cbm_bits: Minimum number of consecutive bits to be set
273 * @cbm_idx_mult: Multiplier of CBM index
274 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by:
275 * closid * cbm_idx_multi + cbm_idx_offset
276 * in a cache bit mask
277 * @shareable_bits: Bitmask of shareable resource with other
281 unsigned int cbm_len;
282 unsigned int min_cbm_bits;
283 unsigned int cbm_idx_mult;
284 unsigned int cbm_idx_offset;
285 unsigned int shareable_bits;
289 * struct rdt_membw - Memory bandwidth allocation related data
290 * @max_delay: Max throttle delay. Delay is the hardware
291 * representation for memory bandwidth.
292 * @min_bw: Minimum memory bandwidth percentage user can request
293 * @bw_gran: Granularity at which the memory bandwidth is allocated
294 * @delay_linear: True if memory B/W delay is in linear scale
295 * @mba_sc: True if MBA software controller(mba_sc) is enabled
296 * @mb_map: Mapping of memory B/W percentage to memory B/W delay
307 static inline bool is_llc_occupancy_enabled(void)
309 return (rdt_mon_features & (1 << QOS_L3_OCCUP_EVENT_ID));
312 static inline bool is_mbm_total_enabled(void)
314 return (rdt_mon_features & (1 << QOS_L3_MBM_TOTAL_EVENT_ID));
317 static inline bool is_mbm_local_enabled(void)
319 return (rdt_mon_features & (1 << QOS_L3_MBM_LOCAL_EVENT_ID));
322 static inline bool is_mbm_enabled(void)
324 return (is_mbm_total_enabled() || is_mbm_local_enabled());
327 static inline bool is_mbm_event(int e)
329 return (e >= QOS_L3_MBM_TOTAL_EVENT_ID &&
330 e <= QOS_L3_MBM_LOCAL_EVENT_ID);
334 * struct rdt_resource - attributes of an RDT resource
335 * @rid: The index of the resource
336 * @alloc_enabled: Is allocation enabled on this machine
337 * @mon_enabled: Is monitoring enabled for this feature
338 * @alloc_capable: Is allocation available on this machine
339 * @mon_capable: Is monitor feature available on this machine
340 * @name: Name to use in "schemata" file
341 * @num_closid: Number of CLOSIDs available
342 * @cache_level: Which cache level defines scope of this resource
343 * @default_ctrl: Specifies default cache cbm or memory B/W percent.
344 * @msr_base: Base MSR address for CBMs
345 * @msr_update: Function pointer to update QOS MSRs
346 * @data_width: Character width of data when displaying
347 * @domains: All domains for this resource
348 * @cache: Cache allocation related data
349 * @format_str: Per resource format string to show domain value
350 * @parse_ctrlval: Per resource function pointer to parse control values
351 * @evt_list: List of monitoring events
352 * @num_rmid: Number of RMIDs available
353 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
354 * @fflags: flags to choose base and info files
356 struct rdt_resource {
366 unsigned int msr_base;
367 void (*msr_update) (struct rdt_domain *d, struct msr_param *m,
368 struct rdt_resource *r);
370 struct list_head domains;
371 struct rdt_cache cache;
372 struct rdt_membw membw;
373 const char *format_str;
374 int (*parse_ctrlval) (void *data, struct rdt_resource *r,
375 struct rdt_domain *d);
376 struct list_head evt_list;
378 unsigned int mon_scale;
379 unsigned long fflags;
382 int parse_cbm(void *_data, struct rdt_resource *r, struct rdt_domain *d);
383 int parse_bw(void *_buf, struct rdt_resource *r, struct rdt_domain *d);
385 extern struct mutex rdtgroup_mutex;
387 extern struct rdt_resource rdt_resources_all[];
388 extern struct rdtgroup rdtgroup_default;
389 DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key);
391 int __init rdtgroup_init(void);
402 /* Must be the last */
406 #define for_each_capable_rdt_resource(r) \
407 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
409 if (r->alloc_capable || r->mon_capable)
411 #define for_each_alloc_capable_rdt_resource(r) \
412 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
414 if (r->alloc_capable)
416 #define for_each_mon_capable_rdt_resource(r) \
417 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
421 #define for_each_alloc_enabled_rdt_resource(r) \
422 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
424 if (r->alloc_enabled)
426 #define for_each_mon_enabled_rdt_resource(r) \
427 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
431 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */
432 union cpuid_0x10_1_eax {
434 unsigned int cbm_len:5;
439 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */
440 union cpuid_0x10_3_eax {
442 unsigned int max_delay:12;
447 /* CPUID.(EAX=10H, ECX=ResID).EDX */
448 union cpuid_0x10_x_edx {
450 unsigned int cos_max:16;
455 void rdt_last_cmd_clear(void);
456 void rdt_last_cmd_puts(const char *s);
457 void rdt_last_cmd_printf(const char *fmt, ...);
459 void rdt_ctrl_update(void *arg);
460 struct rdtgroup *rdtgroup_kn_lock_live(struct kernfs_node *kn);
461 void rdtgroup_kn_unlock(struct kernfs_node *kn);
462 struct rdt_domain *rdt_find_domain(struct rdt_resource *r, int id,
463 struct list_head **pos);
464 ssize_t rdtgroup_schemata_write(struct kernfs_open_file *of,
465 char *buf, size_t nbytes, loff_t off);
466 int rdtgroup_schemata_show(struct kernfs_open_file *of,
467 struct seq_file *s, void *v);
468 enum rdtgrp_mode rdtgroup_mode_by_closid(int closid);
469 struct rdt_domain *get_domain_from_cpu(int cpu, struct rdt_resource *r);
470 int update_domains(struct rdt_resource *r, int closid);
471 void closid_free(int closid);
472 int alloc_rmid(void);
473 void free_rmid(u32 rmid);
474 int rdt_get_mon_l3_config(struct rdt_resource *r);
475 void mon_event_count(void *info);
476 int rdtgroup_mondata_show(struct seq_file *m, void *arg);
477 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
478 unsigned int dom_id);
479 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
480 struct rdt_domain *d);
481 void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
482 struct rdtgroup *rdtgrp, int evtid, int first);
483 void mbm_setup_overflow_handler(struct rdt_domain *dom,
484 unsigned long delay_ms);
485 void mbm_handle_overflow(struct work_struct *work);
486 bool is_mba_sc(struct rdt_resource *r);
487 void setup_default_ctrlval(struct rdt_resource *r, u32 *dc, u32 *dm);
488 u32 delay_bw_map(unsigned long bw, struct rdt_resource *r);
489 void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
490 void cqm_handle_limbo(struct work_struct *work);
491 bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
492 void __check_limbo(struct rdt_domain *d, bool force_free);
495 * Define the hooks for Cache Pseudo-Locking to use within rdt_mount().
496 * These are no-ops provided for the new kernfs changes to use as a
497 * baseline in preparation for a conflict-free merge between it
498 * (kernfs changes) and the Cache Pseudo-Locking enabling.
500 static inline int rdt_pseudo_lock_init(void) { return 0; }
501 static inline void rdt_pseudo_lock_release(void) { }
503 #endif /* _ASM_X86_INTEL_RDT_H */