1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_RESCTRL_INTERNAL_H
3 #define _ASM_X86_RESCTRL_INTERNAL_H
5 #include <linux/sched.h>
6 #include <linux/kernfs.h>
7 #include <linux/fs_context.h>
8 #include <linux/jump_label.h>
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
17 #define MSR_IA32_QM_CTR 0x0c8e
18 #define MSR_IA32_QM_EVTSEL 0x0c8d
20 #define L3_QOS_CDP_ENABLE 0x01ULL
22 #define L2_QOS_CDP_ENABLE 0x01ULL
25 * Event IDs are used to program IA32_QM_EVTSEL before reading event
26 * counter from IA32_QM_CTR
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
32 #define CQM_LIMBOCHECK_INTERVAL 1000
34 #define MBM_CNTR_WIDTH_BASE 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 #define MBM_CNTR_WIDTH_OFFSET_AMD 20
42 #define RMID_VAL_ERROR BIT_ULL(63)
43 #define RMID_VAL_UNAVAIL BIT_ULL(62)
45 * With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for
46 * data to be returned. The counter width is discovered from the hardware
47 * as an offset from MBM_CNTR_WIDTH_BASE.
49 #define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)
52 struct rdt_fs_context
{
53 struct kernfs_fs_context kfc
;
59 static inline struct rdt_fs_context
*rdt_fc2context(struct fs_context
*fc
)
61 struct kernfs_fs_context
*kfc
= fc
->fs_private
;
63 return container_of(kfc
, struct rdt_fs_context
, kfc
);
66 DECLARE_STATIC_KEY_FALSE(rdt_enable_key
);
67 DECLARE_STATIC_KEY_FALSE(rdt_mon_enable_key
);
70 * struct mon_evt - Entry in the event list of a resource
72 * @name: name of the event
77 struct list_head list
;
81 * struct mon_data_bits - Monitoring details for each event file
82 * @rid: Resource id associated with the event file.
83 * @evtid: Event id associated with the event file
84 * @domid: The domain to which the event file belongs
89 unsigned int rid
: 10;
90 unsigned int evtid
: 8;
91 unsigned int domid
: 14;
96 struct rdtgroup
*rgrp
;
97 struct rdt_resource
*r
;
104 extern unsigned int resctrl_cqm_threshold
;
105 extern bool rdt_alloc_capable
;
106 extern bool rdt_mon_capable
;
107 extern unsigned int rdt_mon_features
;
109 enum rdt_group_type
{
116 * enum rdtgrp_mode - Mode of a RDT resource group
117 * @RDT_MODE_SHAREABLE: This resource group allows sharing of its allocations
118 * @RDT_MODE_EXCLUSIVE: No sharing of this resource group's allocations allowed
119 * @RDT_MODE_PSEUDO_LOCKSETUP: Resource group will be used for Pseudo-Locking
120 * @RDT_MODE_PSEUDO_LOCKED: No sharing of this resource group's allocations
121 * allowed AND the allocations are Cache Pseudo-Locked
123 * The mode of a resource group enables control over the allowed overlap
124 * between allocations associated with different resource groups (classes
125 * of service). User is able to modify the mode of a resource group by
126 * writing to the "mode" resctrl file associated with the resource group.
128 * The "shareable", "exclusive", and "pseudo-locksetup" modes are set by
129 * writing the appropriate text to the "mode" file. A resource group enters
130 * "pseudo-locked" mode after the schemata is written while the resource
131 * group is in "pseudo-locksetup" mode.
134 RDT_MODE_SHAREABLE
= 0,
136 RDT_MODE_PSEUDO_LOCKSETUP
,
137 RDT_MODE_PSEUDO_LOCKED
,
144 * struct mongroup - store mon group's data in resctrl fs.
145 * @mon_data_kn kernlfs node for the mon_data directory
146 * @parent: parent rdtgrp
147 * @crdtgrp_list: child rdtgroup node list
148 * @rmid: rmid for this rdtgroup
151 struct kernfs_node
*mon_data_kn
;
152 struct rdtgroup
*parent
;
153 struct list_head crdtgrp_list
;
158 * struct pseudo_lock_region - pseudo-lock region information
159 * @r: RDT resource to which this pseudo-locked region
161 * @d: RDT domain to which this pseudo-locked region
163 * @cbm: bitmask of the pseudo-locked region
164 * @lock_thread_wq: waitqueue used to wait on the pseudo-locking thread
166 * @thread_done: variable used by waitqueue to test if pseudo-locking
168 * @cpu: core associated with the cache on which the setup code
170 * @line_size: size of the cache lines
171 * @size: size of pseudo-locked region in bytes
172 * @kmem: the kernel memory associated with pseudo-locked region
173 * @minor: minor number of character device associated with this
175 * @debugfs_dir: pointer to this region's directory in the debugfs
177 * @pm_reqs: Power management QoS requests related to this region
179 struct pseudo_lock_region
{
180 struct rdt_resource
*r
;
181 struct rdt_domain
*d
;
183 wait_queue_head_t lock_thread_wq
;
186 unsigned int line_size
;
190 struct dentry
*debugfs_dir
;
191 struct list_head pm_reqs
;
195 * struct rdtgroup - store rdtgroup's data in resctrl file system.
197 * @rdtgroup_list: linked list for all rdtgroups
198 * @closid: closid for this rdtgroup
199 * @cpu_mask: CPUs assigned to this rdtgroup
200 * @flags: status bits
201 * @waitcount: how many cpus expect to find this
202 * group when they acquire rdtgroup_mutex
203 * @type: indicates type of this rdtgroup - either
204 * monitor only or ctrl_mon group
205 * @mon: mongroup related data
206 * @mode: mode of resource group
207 * @plr: pseudo-locked region
210 struct kernfs_node
*kn
;
211 struct list_head rdtgroup_list
;
213 struct cpumask cpu_mask
;
216 enum rdt_group_type type
;
218 enum rdtgrp_mode mode
;
219 struct pseudo_lock_region
*plr
;
223 #define RDT_DELETED 1
226 #define RFTYPE_FLAGS_CPUS_LIST 1
229 * Define the file type flags for base and info directories.
231 #define RFTYPE_INFO BIT(0)
232 #define RFTYPE_BASE BIT(1)
233 #define RF_CTRLSHIFT 4
234 #define RF_MONSHIFT 5
235 #define RF_TOPSHIFT 6
236 #define RFTYPE_CTRL BIT(RF_CTRLSHIFT)
237 #define RFTYPE_MON BIT(RF_MONSHIFT)
238 #define RFTYPE_TOP BIT(RF_TOPSHIFT)
239 #define RFTYPE_RES_CACHE BIT(8)
240 #define RFTYPE_RES_MB BIT(9)
241 #define RF_CTRL_INFO (RFTYPE_INFO | RFTYPE_CTRL)
242 #define RF_MON_INFO (RFTYPE_INFO | RFTYPE_MON)
243 #define RF_TOP_INFO (RFTYPE_INFO | RFTYPE_TOP)
244 #define RF_CTRL_BASE (RFTYPE_BASE | RFTYPE_CTRL)
246 /* List of all resource groups */
247 extern struct list_head rdt_all_groups
;
249 extern int max_name_width
, max_data_width
;
251 int __init
rdtgroup_init(void);
252 void __exit
rdtgroup_exit(void);
255 * struct rftype - describe each file in the resctrl file system
258 * @kf_ops: File operations
259 * @flags: File specific RFTYPE_FLAGS_* flags
260 * @fflags: File specific RF_* or RFTYPE_* flags
261 * @seq_show: Show content of the file
262 * @write: Write to the file
267 struct kernfs_ops
*kf_ops
;
269 unsigned long fflags
;
271 int (*seq_show
)(struct kernfs_open_file
*of
,
272 struct seq_file
*sf
, void *v
);
274 * write() is the generic write callback which maps directly to
275 * kernfs write operation and overrides all other operations.
276 * Maximum write size is determined by ->max_write_len.
278 ssize_t (*write
)(struct kernfs_open_file
*of
,
279 char *buf
, size_t nbytes
, loff_t off
);
283 * struct mbm_state - status for each MBM counter in each domain
284 * @chunks: Total data moved (multiply by rdt_group.mon_scale to get bytes)
285 * @prev_msr Value of IA32_QM_CTR for this RMID last time we read it
286 * @prev_bw_msr:Value of previous IA32_QM_CTR for bandwidth counting
287 * @prev_bw The most recent bandwidth in MBps
288 * @delta_bw Difference between the current and previous bandwidth
289 * @delta_comp Indicates whether to compute the delta_bw
301 * struct rdt_domain - group of cpus sharing an RDT resource
302 * @list: all instances of this resource
303 * @id: unique id for this instance
304 * @cpu_mask: which cpus share this resource
306 * bitmap of which limbo RMIDs are above threshold
307 * @mbm_total: saved state for MBM total bandwidth
308 * @mbm_local: saved state for MBM local bandwidth
309 * @mbm_over: worker to periodically read MBM h/w counters
310 * @cqm_limbo: worker to periodically read CQM h/w counters
312 * worker cpu for MBM h/w counters
314 * worker cpu for CQM h/w counters
315 * @ctrl_val: array of cache or mem ctrl values (indexed by CLOSID)
316 * @mbps_val: When mba_sc is enabled, this holds the bandwidth in MBps
317 * @new_ctrl: new ctrl value to be loaded
318 * @have_new_ctrl: did user provide new_ctrl for this domain
319 * @plr: pseudo-locked region (if any) associated with domain
322 struct list_head list
;
324 struct cpumask cpu_mask
;
325 unsigned long *rmid_busy_llc
;
326 struct mbm_state
*mbm_total
;
327 struct mbm_state
*mbm_local
;
328 struct delayed_work mbm_over
;
329 struct delayed_work cqm_limbo
;
336 struct pseudo_lock_region
*plr
;
340 * struct msr_param - set a range of MSRs from a domain
341 * @res: The resource to use
342 * @low: Beginning index from base MSR
346 struct rdt_resource
*res
;
352 * struct rdt_cache - Cache allocation related data
353 * @cbm_len: Length of the cache bit mask
354 * @min_cbm_bits: Minimum number of consecutive bits to be set
355 * @cbm_idx_mult: Multiplier of CBM index
356 * @cbm_idx_offset: Offset of CBM index. CBM index is computed by:
357 * closid * cbm_idx_multi + cbm_idx_offset
358 * in a cache bit mask
359 * @shareable_bits: Bitmask of shareable resource with other
363 unsigned int cbm_len
;
364 unsigned int min_cbm_bits
;
365 unsigned int cbm_idx_mult
;
366 unsigned int cbm_idx_offset
;
367 unsigned int shareable_bits
;
371 * struct rdt_membw - Memory bandwidth allocation related data
372 * @min_bw: Minimum memory bandwidth percentage user can request
373 * @bw_gran: Granularity at which the memory bandwidth is allocated
374 * @delay_linear: True if memory B/W delay is in linear scale
375 * @mba_sc: True if MBA software controller(mba_sc) is enabled
376 * @mb_map: Mapping of memory B/W percentage to memory B/W delay
386 static inline bool is_llc_occupancy_enabled(void)
388 return (rdt_mon_features
& (1 << QOS_L3_OCCUP_EVENT_ID
));
391 static inline bool is_mbm_total_enabled(void)
393 return (rdt_mon_features
& (1 << QOS_L3_MBM_TOTAL_EVENT_ID
));
396 static inline bool is_mbm_local_enabled(void)
398 return (rdt_mon_features
& (1 << QOS_L3_MBM_LOCAL_EVENT_ID
));
401 static inline bool is_mbm_enabled(void)
403 return (is_mbm_total_enabled() || is_mbm_local_enabled());
406 static inline bool is_mbm_event(int e
)
408 return (e
>= QOS_L3_MBM_TOTAL_EVENT_ID
&&
409 e
<= QOS_L3_MBM_LOCAL_EVENT_ID
);
412 struct rdt_parse_data
{
413 struct rdtgroup
*rdtgrp
;
418 * struct rdt_resource - attributes of an RDT resource
419 * @rid: The index of the resource
420 * @alloc_enabled: Is allocation enabled on this machine
421 * @mon_enabled: Is monitoring enabled for this feature
422 * @alloc_capable: Is allocation available on this machine
423 * @mon_capable: Is monitor feature available on this machine
424 * @name: Name to use in "schemata" file
425 * @num_closid: Number of CLOSIDs available
426 * @cache_level: Which cache level defines scope of this resource
427 * @default_ctrl: Specifies default cache cbm or memory B/W percent.
428 * @msr_base: Base MSR address for CBMs
429 * @msr_update: Function pointer to update QOS MSRs
430 * @data_width: Character width of data when displaying
431 * @domains: All domains for this resource
432 * @cache: Cache allocation related data
433 * @format_str: Per resource format string to show domain value
434 * @parse_ctrlval: Per resource function pointer to parse control values
435 * @cbm_validate Cache bitmask validate function
436 * @evt_list: List of monitoring events
437 * @num_rmid: Number of RMIDs available
438 * @mon_scale: cqm counter * mon_scale = occupancy in bytes
439 * @fflags: flags to choose base and info files
441 struct rdt_resource
{
451 unsigned int msr_base
;
452 void (*msr_update
) (struct rdt_domain
*d
, struct msr_param
*m
,
453 struct rdt_resource
*r
);
455 struct list_head domains
;
456 struct rdt_cache cache
;
457 struct rdt_membw membw
;
458 const char *format_str
;
459 int (*parse_ctrlval
)(struct rdt_parse_data
*data
,
460 struct rdt_resource
*r
,
461 struct rdt_domain
*d
);
462 bool (*cbm_validate
)(char *buf
, u32
*data
, struct rdt_resource
*r
);
463 struct list_head evt_list
;
465 unsigned int mon_scale
;
466 unsigned int mbm_width
;
467 unsigned long fflags
;
470 int parse_cbm(struct rdt_parse_data
*data
, struct rdt_resource
*r
,
471 struct rdt_domain
*d
);
472 int parse_bw_intel(struct rdt_parse_data
*data
, struct rdt_resource
*r
,
473 struct rdt_domain
*d
);
474 int parse_bw_amd(struct rdt_parse_data
*data
, struct rdt_resource
*r
,
475 struct rdt_domain
*d
);
477 extern struct mutex rdtgroup_mutex
;
479 extern struct rdt_resource rdt_resources_all
[];
480 extern struct rdtgroup rdtgroup_default
;
481 DECLARE_STATIC_KEY_FALSE(rdt_alloc_enable_key
);
483 extern struct dentry
*debugfs_resctrl
;
494 /* Must be the last */
498 #define for_each_rdt_resource(r) \
499 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
502 #define for_each_capable_rdt_resource(r) \
503 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
505 if (r->alloc_capable || r->mon_capable)
507 #define for_each_alloc_capable_rdt_resource(r) \
508 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
510 if (r->alloc_capable)
512 #define for_each_mon_capable_rdt_resource(r) \
513 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
517 #define for_each_alloc_enabled_rdt_resource(r) \
518 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
520 if (r->alloc_enabled)
522 #define for_each_mon_enabled_rdt_resource(r) \
523 for (r = rdt_resources_all; r < rdt_resources_all + RDT_NUM_RESOURCES;\
527 /* CPUID.(EAX=10H, ECX=ResID=1).EAX */
528 union cpuid_0x10_1_eax
{
530 unsigned int cbm_len
:5;
535 /* CPUID.(EAX=10H, ECX=ResID=3).EAX */
536 union cpuid_0x10_3_eax
{
538 unsigned int max_delay
:12;
543 /* CPUID.(EAX=10H, ECX=ResID).EDX */
544 union cpuid_0x10_x_edx
{
546 unsigned int cos_max
:16;
551 void rdt_last_cmd_clear(void);
552 void rdt_last_cmd_puts(const char *s
);
553 void rdt_last_cmd_printf(const char *fmt
, ...);
555 void rdt_ctrl_update(void *arg
);
556 struct rdtgroup
*rdtgroup_kn_lock_live(struct kernfs_node
*kn
);
557 void rdtgroup_kn_unlock(struct kernfs_node
*kn
);
558 int rdtgroup_kn_mode_restrict(struct rdtgroup
*r
, const char *name
);
559 int rdtgroup_kn_mode_restore(struct rdtgroup
*r
, const char *name
,
561 struct rdt_domain
*rdt_find_domain(struct rdt_resource
*r
, int id
,
562 struct list_head
**pos
);
563 ssize_t
rdtgroup_schemata_write(struct kernfs_open_file
*of
,
564 char *buf
, size_t nbytes
, loff_t off
);
565 int rdtgroup_schemata_show(struct kernfs_open_file
*of
,
566 struct seq_file
*s
, void *v
);
567 bool rdtgroup_cbm_overlaps(struct rdt_resource
*r
, struct rdt_domain
*d
,
568 unsigned long cbm
, int closid
, bool exclusive
);
569 unsigned int rdtgroup_cbm_to_size(struct rdt_resource
*r
, struct rdt_domain
*d
,
571 enum rdtgrp_mode
rdtgroup_mode_by_closid(int closid
);
572 int rdtgroup_tasks_assigned(struct rdtgroup
*r
);
573 int rdtgroup_locksetup_enter(struct rdtgroup
*rdtgrp
);
574 int rdtgroup_locksetup_exit(struct rdtgroup
*rdtgrp
);
575 bool rdtgroup_cbm_overlaps_pseudo_locked(struct rdt_domain
*d
, unsigned long cbm
);
576 bool rdtgroup_pseudo_locked_in_hierarchy(struct rdt_domain
*d
);
577 int rdt_pseudo_lock_init(void);
578 void rdt_pseudo_lock_release(void);
579 int rdtgroup_pseudo_lock_create(struct rdtgroup
*rdtgrp
);
580 void rdtgroup_pseudo_lock_remove(struct rdtgroup
*rdtgrp
);
581 struct rdt_domain
*get_domain_from_cpu(int cpu
, struct rdt_resource
*r
);
582 int update_domains(struct rdt_resource
*r
, int closid
);
583 int closids_supported(void);
584 void closid_free(int closid
);
585 int alloc_rmid(void);
586 void free_rmid(u32 rmid
);
587 int rdt_get_mon_l3_config(struct rdt_resource
*r
);
588 void mon_event_count(void *info
);
589 int rdtgroup_mondata_show(struct seq_file
*m
, void *arg
);
590 void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource
*r
,
591 unsigned int dom_id
);
592 void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource
*r
,
593 struct rdt_domain
*d
);
594 void mon_event_read(struct rmid_read
*rr
, struct rdt_resource
*r
,
595 struct rdt_domain
*d
, struct rdtgroup
*rdtgrp
,
596 int evtid
, int first
);
597 void mbm_setup_overflow_handler(struct rdt_domain
*dom
,
598 unsigned long delay_ms
);
599 void mbm_handle_overflow(struct work_struct
*work
);
600 bool is_mba_sc(struct rdt_resource
*r
);
601 void setup_default_ctrlval(struct rdt_resource
*r
, u32
*dc
, u32
*dm
);
602 u32
delay_bw_map(unsigned long bw
, struct rdt_resource
*r
);
603 void cqm_setup_limbo_handler(struct rdt_domain
*dom
, unsigned long delay_ms
);
604 void cqm_handle_limbo(struct work_struct
*work
);
605 bool has_busy_rmid(struct rdt_resource
*r
, struct rdt_domain
*d
);
606 void __check_limbo(struct rdt_domain
*d
, bool force_free
);
607 bool cbm_validate_intel(char *buf
, u32
*data
, struct rdt_resource
*r
);
608 bool cbm_validate_amd(char *buf
, u32
*data
, struct rdt_resource
*r
);
609 void rdt_domain_reconfigure_cdp(struct rdt_resource
*r
);
611 #endif /* _ASM_X86_RESCTRL_INTERNAL_H */