1 // SPDX-License-Identifier: GPL-2.0
3 * cacheinfo support - processor cache information via sysfs
5 * Based on arch/x86/kernel/cpu/intel_cacheinfo.c
6 * Author: Sudeep Holla <sudeep.holla@arm.com>
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/acpi.h>
11 #include <linux/bitops.h>
12 #include <linux/cacheinfo.h>
13 #include <linux/compiler.h>
14 #include <linux/cpu.h>
15 #include <linux/device.h>
16 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/slab.h>
20 #include <linux/smp.h>
21 #include <linux/sysfs.h>
23 /* pointer to per cpu cacheinfo */
24 static DEFINE_PER_CPU(struct cpu_cacheinfo
, ci_cpu_cacheinfo
);
25 #define ci_cacheinfo(cpu) (&per_cpu(ci_cpu_cacheinfo, cpu))
26 #define cache_leaves(cpu) (ci_cacheinfo(cpu)->num_leaves)
27 #define per_cpu_cacheinfo(cpu) (ci_cacheinfo(cpu)->info_list)
28 #define per_cpu_cacheinfo_idx(cpu, idx) \
29 (per_cpu_cacheinfo(cpu) + (idx))
31 /* Set if no cache information is found in DT/ACPI. */
32 static bool use_arch_info
;
34 struct cpu_cacheinfo
*get_cpu_cacheinfo(unsigned int cpu
)
36 return ci_cacheinfo(cpu
);
39 static inline bool cache_leaves_are_shared(struct cacheinfo
*this_leaf
,
40 struct cacheinfo
*sib_leaf
)
43 * For non DT/ACPI systems, assume unique level 1 caches,
44 * system-wide shared caches for all other levels.
46 if (!(IS_ENABLED(CONFIG_OF
) || IS_ENABLED(CONFIG_ACPI
)) ||
48 return (this_leaf
->level
!= 1) && (sib_leaf
->level
!= 1);
50 if ((sib_leaf
->attributes
& CACHE_ID
) &&
51 (this_leaf
->attributes
& CACHE_ID
))
52 return sib_leaf
->id
== this_leaf
->id
;
54 return sib_leaf
->fw_token
== this_leaf
->fw_token
;
57 bool last_level_cache_is_valid(unsigned int cpu
)
59 struct cacheinfo
*llc
;
61 if (!cache_leaves(cpu
))
64 llc
= per_cpu_cacheinfo_idx(cpu
, cache_leaves(cpu
) - 1);
66 return (llc
->attributes
& CACHE_ID
) || !!llc
->fw_token
;
70 bool last_level_cache_is_shared(unsigned int cpu_x
, unsigned int cpu_y
)
72 struct cacheinfo
*llc_x
, *llc_y
;
74 if (!last_level_cache_is_valid(cpu_x
) ||
75 !last_level_cache_is_valid(cpu_y
))
78 llc_x
= per_cpu_cacheinfo_idx(cpu_x
, cache_leaves(cpu_x
) - 1);
79 llc_y
= per_cpu_cacheinfo_idx(cpu_y
, cache_leaves(cpu_y
) - 1);
81 return cache_leaves_are_shared(llc_x
, llc_y
);
86 static bool of_check_cache_nodes(struct device_node
*np
);
88 /* OF properties to query for a given cache type */
89 struct cache_type_info
{
90 const char *size_prop
;
91 const char *line_size_props
[2];
92 const char *nr_sets_prop
;
95 static const struct cache_type_info cache_type_info
[] = {
97 .size_prop
= "cache-size",
98 .line_size_props
= { "cache-line-size",
99 "cache-block-size", },
100 .nr_sets_prop
= "cache-sets",
102 .size_prop
= "i-cache-size",
103 .line_size_props
= { "i-cache-line-size",
104 "i-cache-block-size", },
105 .nr_sets_prop
= "i-cache-sets",
107 .size_prop
= "d-cache-size",
108 .line_size_props
= { "d-cache-line-size",
109 "d-cache-block-size", },
110 .nr_sets_prop
= "d-cache-sets",
114 static inline int get_cacheinfo_idx(enum cache_type type
)
116 if (type
== CACHE_TYPE_UNIFIED
)
121 static void cache_size(struct cacheinfo
*this_leaf
, struct device_node
*np
)
123 const char *propname
;
126 ct_idx
= get_cacheinfo_idx(this_leaf
->type
);
127 propname
= cache_type_info
[ct_idx
].size_prop
;
129 of_property_read_u32(np
, propname
, &this_leaf
->size
);
132 /* not cache_line_size() because that's a macro in include/linux/cache.h */
133 static void cache_get_line_size(struct cacheinfo
*this_leaf
,
134 struct device_node
*np
)
138 ct_idx
= get_cacheinfo_idx(this_leaf
->type
);
139 lim
= ARRAY_SIZE(cache_type_info
[ct_idx
].line_size_props
);
141 for (i
= 0; i
< lim
; i
++) {
144 const char *propname
;
146 propname
= cache_type_info
[ct_idx
].line_size_props
[i
];
147 ret
= of_property_read_u32(np
, propname
, &line_size
);
149 this_leaf
->coherency_line_size
= line_size
;
155 static void cache_nr_sets(struct cacheinfo
*this_leaf
, struct device_node
*np
)
157 const char *propname
;
160 ct_idx
= get_cacheinfo_idx(this_leaf
->type
);
161 propname
= cache_type_info
[ct_idx
].nr_sets_prop
;
163 of_property_read_u32(np
, propname
, &this_leaf
->number_of_sets
);
166 static void cache_associativity(struct cacheinfo
*this_leaf
)
168 unsigned int line_size
= this_leaf
->coherency_line_size
;
169 unsigned int nr_sets
= this_leaf
->number_of_sets
;
170 unsigned int size
= this_leaf
->size
;
173 * If the cache is fully associative, there is no need to
174 * check the other properties.
176 if (!(nr_sets
== 1) && (nr_sets
> 0 && size
> 0 && line_size
> 0))
177 this_leaf
->ways_of_associativity
= (size
/ nr_sets
) / line_size
;
180 static bool cache_node_is_unified(struct cacheinfo
*this_leaf
,
181 struct device_node
*np
)
183 return of_property_read_bool(np
, "cache-unified");
186 static void cache_of_set_props(struct cacheinfo
*this_leaf
,
187 struct device_node
*np
)
190 * init_cache_level must setup the cache level correctly
191 * overriding the architecturally specified levels, so
192 * if type is NONE at this stage, it should be unified
194 if (this_leaf
->type
== CACHE_TYPE_NOCACHE
&&
195 cache_node_is_unified(this_leaf
, np
))
196 this_leaf
->type
= CACHE_TYPE_UNIFIED
;
197 cache_size(this_leaf
, np
);
198 cache_get_line_size(this_leaf
, np
);
199 cache_nr_sets(this_leaf
, np
);
200 cache_associativity(this_leaf
);
203 static int cache_setup_of_node(unsigned int cpu
)
205 struct cacheinfo
*this_leaf
;
206 unsigned int index
= 0;
208 struct device_node
*np
__free(device_node
) = of_cpu_device_node_get(cpu
);
210 pr_err("Failed to find cpu%d device node\n", cpu
);
214 if (!of_check_cache_nodes(np
)) {
218 while (index
< cache_leaves(cpu
)) {
219 this_leaf
= per_cpu_cacheinfo_idx(cpu
, index
);
220 if (this_leaf
->level
!= 1) {
221 struct device_node
*prev
__free(device_node
) = np
;
222 np
= of_find_next_cache_node(np
);
226 cache_of_set_props(this_leaf
, np
);
227 this_leaf
->fw_token
= np
;
231 if (index
!= cache_leaves(cpu
)) /* not all OF nodes populated */
237 static bool of_check_cache_nodes(struct device_node
*np
)
239 if (of_property_present(np
, "cache-size") ||
240 of_property_present(np
, "i-cache-size") ||
241 of_property_present(np
, "d-cache-size") ||
242 of_property_present(np
, "cache-unified"))
245 struct device_node
*next
__free(device_node
) = of_find_next_cache_node(np
);
253 static int of_count_cache_leaves(struct device_node
*np
)
255 unsigned int leaves
= 0;
257 if (of_property_present(np
, "cache-size"))
259 if (of_property_present(np
, "i-cache-size"))
261 if (of_property_present(np
, "d-cache-size"))
265 /* The '[i-|d-|]cache-size' property is required, but
266 * if absent, fallback on the 'cache-unified' property.
268 if (of_property_read_bool(np
, "cache-unified"))
277 int init_of_cache_level(unsigned int cpu
)
279 struct cpu_cacheinfo
*this_cpu_ci
= get_cpu_cacheinfo(cpu
);
280 struct device_node
*np
__free(device_node
) = of_cpu_device_node_get(cpu
);
281 unsigned int levels
= 0, leaves
, level
;
283 if (!of_check_cache_nodes(np
)) {
287 leaves
= of_count_cache_leaves(np
);
292 struct device_node
*prev
__free(device_node
) = np
;
293 np
= of_find_next_cache_node(np
);
297 if (!of_device_is_compatible(np
, "cache"))
299 if (of_property_read_u32(np
, "cache-level", &level
))
304 leaves
+= of_count_cache_leaves(np
);
308 this_cpu_ci
->num_levels
= levels
;
309 this_cpu_ci
->num_leaves
= leaves
;
315 static inline int cache_setup_of_node(unsigned int cpu
) { return 0; }
316 int init_of_cache_level(unsigned int cpu
) { return 0; }
319 int __weak
cache_setup_acpi(unsigned int cpu
)
324 unsigned int coherency_max_size
;
326 static int cache_setup_properties(unsigned int cpu
)
330 if (of_have_populated_dt())
331 ret
= cache_setup_of_node(cpu
);
332 else if (!acpi_disabled
)
333 ret
= cache_setup_acpi(cpu
);
335 // Assume there is no cache information available in DT/ACPI from now.
336 if (ret
&& use_arch_cache_info())
337 use_arch_info
= true;
342 static int cache_shared_cpu_map_setup(unsigned int cpu
)
344 struct cpu_cacheinfo
*this_cpu_ci
= get_cpu_cacheinfo(cpu
);
345 struct cacheinfo
*this_leaf
, *sib_leaf
;
346 unsigned int index
, sib_index
;
349 if (this_cpu_ci
->cpu_map_populated
)
353 * skip setting up cache properties if LLC is valid, just need
354 * to update the shared cpu_map if the cache attributes were
355 * populated early before all the cpus are brought online
357 if (!last_level_cache_is_valid(cpu
) && !use_arch_info
) {
358 ret
= cache_setup_properties(cpu
);
363 for (index
= 0; index
< cache_leaves(cpu
); index
++) {
366 this_leaf
= per_cpu_cacheinfo_idx(cpu
, index
);
368 cpumask_set_cpu(cpu
, &this_leaf
->shared_cpu_map
);
369 for_each_online_cpu(i
) {
370 if (i
== cpu
|| !per_cpu_cacheinfo(i
))
371 continue;/* skip if itself or no cacheinfo */
372 for (sib_index
= 0; sib_index
< cache_leaves(i
); sib_index
++) {
373 sib_leaf
= per_cpu_cacheinfo_idx(i
, sib_index
);
376 * Comparing cache IDs only makes sense if the leaves
377 * belong to the same cache level of same type. Skip
378 * the check if level and type do not match.
380 if (sib_leaf
->level
!= this_leaf
->level
||
381 sib_leaf
->type
!= this_leaf
->type
)
384 if (cache_leaves_are_shared(this_leaf
, sib_leaf
)) {
385 cpumask_set_cpu(cpu
, &sib_leaf
->shared_cpu_map
);
386 cpumask_set_cpu(i
, &this_leaf
->shared_cpu_map
);
391 /* record the maximum cache line size */
392 if (this_leaf
->coherency_line_size
> coherency_max_size
)
393 coherency_max_size
= this_leaf
->coherency_line_size
;
396 /* shared_cpu_map is now populated for the cpu */
397 this_cpu_ci
->cpu_map_populated
= true;
401 static void cache_shared_cpu_map_remove(unsigned int cpu
)
403 struct cpu_cacheinfo
*this_cpu_ci
= get_cpu_cacheinfo(cpu
);
404 struct cacheinfo
*this_leaf
, *sib_leaf
;
405 unsigned int sibling
, index
, sib_index
;
407 for (index
= 0; index
< cache_leaves(cpu
); index
++) {
408 this_leaf
= per_cpu_cacheinfo_idx(cpu
, index
);
409 for_each_cpu(sibling
, &this_leaf
->shared_cpu_map
) {
410 if (sibling
== cpu
|| !per_cpu_cacheinfo(sibling
))
411 continue;/* skip if itself or no cacheinfo */
413 for (sib_index
= 0; sib_index
< cache_leaves(sibling
); sib_index
++) {
414 sib_leaf
= per_cpu_cacheinfo_idx(sibling
, sib_index
);
417 * Comparing cache IDs only makes sense if the leaves
418 * belong to the same cache level of same type. Skip
419 * the check if level and type do not match.
421 if (sib_leaf
->level
!= this_leaf
->level
||
422 sib_leaf
->type
!= this_leaf
->type
)
425 if (cache_leaves_are_shared(this_leaf
, sib_leaf
)) {
426 cpumask_clear_cpu(cpu
, &sib_leaf
->shared_cpu_map
);
427 cpumask_clear_cpu(sibling
, &this_leaf
->shared_cpu_map
);
434 /* cpu is no longer populated in the shared map */
435 this_cpu_ci
->cpu_map_populated
= false;
438 static void free_cache_attributes(unsigned int cpu
)
440 if (!per_cpu_cacheinfo(cpu
))
443 cache_shared_cpu_map_remove(cpu
);
446 int __weak
early_cache_level(unsigned int cpu
)
451 int __weak
init_cache_level(unsigned int cpu
)
456 int __weak
populate_cache_leaves(unsigned int cpu
)
462 int allocate_cache_info(int cpu
)
464 per_cpu_cacheinfo(cpu
) = kcalloc(cache_leaves(cpu
),
465 sizeof(struct cacheinfo
), GFP_ATOMIC
);
466 if (!per_cpu_cacheinfo(cpu
)) {
467 cache_leaves(cpu
) = 0;
474 int fetch_cache_info(unsigned int cpu
)
476 struct cpu_cacheinfo
*this_cpu_ci
= get_cpu_cacheinfo(cpu
);
477 unsigned int levels
= 0, split_levels
= 0;
481 ret
= init_of_cache_level(cpu
);
483 ret
= acpi_get_cache_info(cpu
, &levels
, &split_levels
);
485 this_cpu_ci
->num_levels
= levels
;
488 * - there cannot be any split caches (data/instruction)
489 * above a unified cache
490 * - data/instruction caches come by pair
492 this_cpu_ci
->num_leaves
= levels
+ split_levels
;
496 if (ret
|| !cache_leaves(cpu
)) {
497 ret
= early_cache_level(cpu
);
501 if (!cache_leaves(cpu
))
504 this_cpu_ci
->early_ci_levels
= true;
507 return allocate_cache_info(cpu
);
510 static inline int init_level_allocate_ci(unsigned int cpu
)
512 unsigned int early_leaves
= cache_leaves(cpu
);
514 /* Since early initialization/allocation of the cacheinfo is allowed
515 * via fetch_cache_info() and this also gets called as CPU hotplug
516 * callbacks via cacheinfo_cpu_online, the init/alloc can be skipped
517 * as it will happen only once (the cacheinfo memory is never freed).
518 * Just populate the cacheinfo. However, if the cacheinfo has been
519 * allocated early through the arch-specific early_cache_level() call,
520 * there is a chance the info is wrong (this can happen on arm64). In
521 * that case, call init_cache_level() anyway to give the arch-specific
522 * code a chance to make things right.
524 if (per_cpu_cacheinfo(cpu
) && !ci_cacheinfo(cpu
)->early_ci_levels
)
527 if (init_cache_level(cpu
) || !cache_leaves(cpu
))
531 * Now that we have properly initialized the cache level info, make
532 * sure we don't try to do that again the next time we are called
533 * (e.g. as CPU hotplug callbacks).
535 ci_cacheinfo(cpu
)->early_ci_levels
= false;
537 if (cache_leaves(cpu
) <= early_leaves
)
540 kfree(per_cpu_cacheinfo(cpu
));
541 return allocate_cache_info(cpu
);
544 int detect_cache_attributes(unsigned int cpu
)
548 ret
= init_level_allocate_ci(cpu
);
553 * If LLC is valid the cache leaves were already populated so just go to
554 * update the cpu map.
556 if (!last_level_cache_is_valid(cpu
)) {
558 * populate_cache_leaves() may completely setup the cache leaves and
559 * shared_cpu_map or it may leave it partially setup.
561 ret
= populate_cache_leaves(cpu
);
567 * For systems using DT for cache hierarchy, fw_token
568 * and shared_cpu_map will be set up here only if they are
569 * not populated already
571 ret
= cache_shared_cpu_map_setup(cpu
);
573 pr_warn("Unable to detect cache hierarchy for CPU %d\n", cpu
);
580 free_cache_attributes(cpu
);
584 /* pointer to cpuX/cache device */
585 static DEFINE_PER_CPU(struct device
*, ci_cache_dev
);
586 #define per_cpu_cache_dev(cpu) (per_cpu(ci_cache_dev, cpu))
588 static cpumask_t cache_dev_map
;
590 /* pointer to array of devices for cpuX/cache/indexY */
591 static DEFINE_PER_CPU(struct device
**, ci_index_dev
);
592 #define per_cpu_index_dev(cpu) (per_cpu(ci_index_dev, cpu))
593 #define per_cache_index_dev(cpu, idx) ((per_cpu_index_dev(cpu))[idx])
595 #define show_one(file_name, object) \
596 static ssize_t file_name##_show(struct device *dev, \
597 struct device_attribute *attr, char *buf) \
599 struct cacheinfo *this_leaf = dev_get_drvdata(dev); \
600 return sysfs_emit(buf, "%u\n", this_leaf->object); \
604 show_one(level
, level
);
605 show_one(coherency_line_size
, coherency_line_size
);
606 show_one(number_of_sets
, number_of_sets
);
607 show_one(physical_line_partition
, physical_line_partition
);
608 show_one(ways_of_associativity
, ways_of_associativity
);
610 static ssize_t
size_show(struct device
*dev
,
611 struct device_attribute
*attr
, char *buf
)
613 struct cacheinfo
*this_leaf
= dev_get_drvdata(dev
);
615 return sysfs_emit(buf
, "%uK\n", this_leaf
->size
>> 10);
618 static ssize_t
shared_cpu_map_show(struct device
*dev
,
619 struct device_attribute
*attr
, char *buf
)
621 struct cacheinfo
*this_leaf
= dev_get_drvdata(dev
);
622 const struct cpumask
*mask
= &this_leaf
->shared_cpu_map
;
624 return sysfs_emit(buf
, "%*pb\n", nr_cpu_ids
, mask
);
627 static ssize_t
shared_cpu_list_show(struct device
*dev
,
628 struct device_attribute
*attr
, char *buf
)
630 struct cacheinfo
*this_leaf
= dev_get_drvdata(dev
);
631 const struct cpumask
*mask
= &this_leaf
->shared_cpu_map
;
633 return sysfs_emit(buf
, "%*pbl\n", nr_cpu_ids
, mask
);
636 static ssize_t
type_show(struct device
*dev
,
637 struct device_attribute
*attr
, char *buf
)
639 struct cacheinfo
*this_leaf
= dev_get_drvdata(dev
);
642 switch (this_leaf
->type
) {
643 case CACHE_TYPE_DATA
:
646 case CACHE_TYPE_INST
:
647 output
= "Instruction";
649 case CACHE_TYPE_UNIFIED
:
656 return sysfs_emit(buf
, "%s\n", output
);
659 static ssize_t
allocation_policy_show(struct device
*dev
,
660 struct device_attribute
*attr
, char *buf
)
662 struct cacheinfo
*this_leaf
= dev_get_drvdata(dev
);
663 unsigned int ci_attr
= this_leaf
->attributes
;
666 if ((ci_attr
& CACHE_READ_ALLOCATE
) && (ci_attr
& CACHE_WRITE_ALLOCATE
))
667 output
= "ReadWriteAllocate";
668 else if (ci_attr
& CACHE_READ_ALLOCATE
)
669 output
= "ReadAllocate";
670 else if (ci_attr
& CACHE_WRITE_ALLOCATE
)
671 output
= "WriteAllocate";
675 return sysfs_emit(buf
, "%s\n", output
);
678 static ssize_t
write_policy_show(struct device
*dev
,
679 struct device_attribute
*attr
, char *buf
)
681 struct cacheinfo
*this_leaf
= dev_get_drvdata(dev
);
682 unsigned int ci_attr
= this_leaf
->attributes
;
685 if (ci_attr
& CACHE_WRITE_THROUGH
)
686 n
= sysfs_emit(buf
, "WriteThrough\n");
687 else if (ci_attr
& CACHE_WRITE_BACK
)
688 n
= sysfs_emit(buf
, "WriteBack\n");
692 static DEVICE_ATTR_RO(id
);
693 static DEVICE_ATTR_RO(level
);
694 static DEVICE_ATTR_RO(type
);
695 static DEVICE_ATTR_RO(coherency_line_size
);
696 static DEVICE_ATTR_RO(ways_of_associativity
);
697 static DEVICE_ATTR_RO(number_of_sets
);
698 static DEVICE_ATTR_RO(size
);
699 static DEVICE_ATTR_RO(allocation_policy
);
700 static DEVICE_ATTR_RO(write_policy
);
701 static DEVICE_ATTR_RO(shared_cpu_map
);
702 static DEVICE_ATTR_RO(shared_cpu_list
);
703 static DEVICE_ATTR_RO(physical_line_partition
);
705 static struct attribute
*cache_default_attrs
[] = {
708 &dev_attr_level
.attr
,
709 &dev_attr_shared_cpu_map
.attr
,
710 &dev_attr_shared_cpu_list
.attr
,
711 &dev_attr_coherency_line_size
.attr
,
712 &dev_attr_ways_of_associativity
.attr
,
713 &dev_attr_number_of_sets
.attr
,
715 &dev_attr_allocation_policy
.attr
,
716 &dev_attr_write_policy
.attr
,
717 &dev_attr_physical_line_partition
.attr
,
722 cache_default_attrs_is_visible(struct kobject
*kobj
,
723 struct attribute
*attr
, int unused
)
725 struct device
*dev
= kobj_to_dev(kobj
);
726 struct cacheinfo
*this_leaf
= dev_get_drvdata(dev
);
727 const struct cpumask
*mask
= &this_leaf
->shared_cpu_map
;
728 umode_t mode
= attr
->mode
;
730 if ((attr
== &dev_attr_id
.attr
) && (this_leaf
->attributes
& CACHE_ID
))
732 if ((attr
== &dev_attr_type
.attr
) && this_leaf
->type
)
734 if ((attr
== &dev_attr_level
.attr
) && this_leaf
->level
)
736 if ((attr
== &dev_attr_shared_cpu_map
.attr
) && !cpumask_empty(mask
))
738 if ((attr
== &dev_attr_shared_cpu_list
.attr
) && !cpumask_empty(mask
))
740 if ((attr
== &dev_attr_coherency_line_size
.attr
) &&
741 this_leaf
->coherency_line_size
)
743 if ((attr
== &dev_attr_ways_of_associativity
.attr
) &&
744 this_leaf
->size
) /* allow 0 = full associativity */
746 if ((attr
== &dev_attr_number_of_sets
.attr
) &&
747 this_leaf
->number_of_sets
)
749 if ((attr
== &dev_attr_size
.attr
) && this_leaf
->size
)
751 if ((attr
== &dev_attr_write_policy
.attr
) &&
752 (this_leaf
->attributes
& CACHE_WRITE_POLICY_MASK
))
754 if ((attr
== &dev_attr_allocation_policy
.attr
) &&
755 (this_leaf
->attributes
& CACHE_ALLOCATE_POLICY_MASK
))
757 if ((attr
== &dev_attr_physical_line_partition
.attr
) &&
758 this_leaf
->physical_line_partition
)
764 static const struct attribute_group cache_default_group
= {
765 .attrs
= cache_default_attrs
,
766 .is_visible
= cache_default_attrs_is_visible
,
769 static const struct attribute_group
*cache_default_groups
[] = {
770 &cache_default_group
,
774 static const struct attribute_group
*cache_private_groups
[] = {
775 &cache_default_group
,
776 NULL
, /* Place holder for private group */
780 const struct attribute_group
*
781 __weak
cache_get_priv_group(struct cacheinfo
*this_leaf
)
786 static const struct attribute_group
**
787 cache_get_attribute_groups(struct cacheinfo
*this_leaf
)
789 const struct attribute_group
*priv_group
=
790 cache_get_priv_group(this_leaf
);
793 return cache_default_groups
;
795 if (!cache_private_groups
[1])
796 cache_private_groups
[1] = priv_group
;
798 return cache_private_groups
;
801 /* Add/Remove cache interface for CPU device */
802 static void cpu_cache_sysfs_exit(unsigned int cpu
)
805 struct device
*ci_dev
;
807 if (per_cpu_index_dev(cpu
)) {
808 for (i
= 0; i
< cache_leaves(cpu
); i
++) {
809 ci_dev
= per_cache_index_dev(cpu
, i
);
812 device_unregister(ci_dev
);
814 kfree(per_cpu_index_dev(cpu
));
815 per_cpu_index_dev(cpu
) = NULL
;
817 device_unregister(per_cpu_cache_dev(cpu
));
818 per_cpu_cache_dev(cpu
) = NULL
;
821 static int cpu_cache_sysfs_init(unsigned int cpu
)
823 struct device
*dev
= get_cpu_device(cpu
);
825 if (per_cpu_cacheinfo(cpu
) == NULL
)
828 per_cpu_cache_dev(cpu
) = cpu_device_create(dev
, NULL
, NULL
, "cache");
829 if (IS_ERR(per_cpu_cache_dev(cpu
)))
830 return PTR_ERR(per_cpu_cache_dev(cpu
));
832 /* Allocate all required memory */
833 per_cpu_index_dev(cpu
) = kcalloc(cache_leaves(cpu
),
834 sizeof(struct device
*), GFP_KERNEL
);
835 if (unlikely(per_cpu_index_dev(cpu
) == NULL
))
841 cpu_cache_sysfs_exit(cpu
);
845 static int cache_add_dev(unsigned int cpu
)
849 struct device
*ci_dev
, *parent
;
850 struct cacheinfo
*this_leaf
;
851 const struct attribute_group
**cache_groups
;
853 rc
= cpu_cache_sysfs_init(cpu
);
854 if (unlikely(rc
< 0))
857 parent
= per_cpu_cache_dev(cpu
);
858 for (i
= 0; i
< cache_leaves(cpu
); i
++) {
859 this_leaf
= per_cpu_cacheinfo_idx(cpu
, i
);
860 if (this_leaf
->disable_sysfs
)
862 if (this_leaf
->type
== CACHE_TYPE_NOCACHE
)
864 cache_groups
= cache_get_attribute_groups(this_leaf
);
865 ci_dev
= cpu_device_create(parent
, this_leaf
, cache_groups
,
867 if (IS_ERR(ci_dev
)) {
868 rc
= PTR_ERR(ci_dev
);
871 per_cache_index_dev(cpu
, i
) = ci_dev
;
873 cpumask_set_cpu(cpu
, &cache_dev_map
);
877 cpu_cache_sysfs_exit(cpu
);
881 static unsigned int cpu_map_shared_cache(bool online
, unsigned int cpu
,
884 struct cacheinfo
*llc
, *sib_llc
;
885 unsigned int sibling
;
887 if (!last_level_cache_is_valid(cpu
))
890 llc
= per_cpu_cacheinfo_idx(cpu
, cache_leaves(cpu
) - 1);
892 if (llc
->type
!= CACHE_TYPE_DATA
&& llc
->type
!= CACHE_TYPE_UNIFIED
)
896 *map
= &llc
->shared_cpu_map
;
897 return cpumask_weight(*map
);
900 /* shared_cpu_map of offlined CPU will be cleared, so use sibling map */
901 for_each_cpu(sibling
, &llc
->shared_cpu_map
) {
902 if (sibling
== cpu
|| !last_level_cache_is_valid(sibling
))
904 sib_llc
= per_cpu_cacheinfo_idx(sibling
, cache_leaves(sibling
) - 1);
905 *map
= &sib_llc
->shared_cpu_map
;
906 return cpumask_weight(*map
);
913 * Calculate the size of the per-CPU data cache slice. This can be
914 * used to estimate the size of the data cache slice that can be used
915 * by one CPU under ideal circumstances. UNIFIED caches are counted
916 * in addition to DATA caches. So, please consider code cache usage
917 * when use the result.
919 * Because the cache inclusive/non-inclusive information isn't
920 * available, we just use the size of the per-CPU slice of LLC to make
921 * the result more predictable across architectures.
923 static void update_per_cpu_data_slice_size_cpu(unsigned int cpu
)
925 struct cpu_cacheinfo
*ci
;
926 struct cacheinfo
*llc
;
927 unsigned int nr_shared
;
929 if (!last_level_cache_is_valid(cpu
))
932 ci
= ci_cacheinfo(cpu
);
933 llc
= per_cpu_cacheinfo_idx(cpu
, cache_leaves(cpu
) - 1);
935 if (llc
->type
!= CACHE_TYPE_DATA
&& llc
->type
!= CACHE_TYPE_UNIFIED
)
938 nr_shared
= cpumask_weight(&llc
->shared_cpu_map
);
940 ci
->per_cpu_data_slice_size
= llc
->size
/ nr_shared
;
943 static void update_per_cpu_data_slice_size(bool cpu_online
, unsigned int cpu
,
948 for_each_cpu(icpu
, cpu_map
) {
949 if (!cpu_online
&& icpu
== cpu
)
951 update_per_cpu_data_slice_size_cpu(icpu
);
952 setup_pcp_cacheinfo(icpu
);
956 static int cacheinfo_cpu_online(unsigned int cpu
)
958 int rc
= detect_cache_attributes(cpu
);
963 rc
= cache_add_dev(cpu
);
966 if (cpu_map_shared_cache(true, cpu
, &cpu_map
))
967 update_per_cpu_data_slice_size(true, cpu
, cpu_map
);
970 free_cache_attributes(cpu
);
974 static int cacheinfo_cpu_pre_down(unsigned int cpu
)
977 unsigned int nr_shared
;
979 nr_shared
= cpu_map_shared_cache(false, cpu
, &cpu_map
);
980 if (cpumask_test_and_clear_cpu(cpu
, &cache_dev_map
))
981 cpu_cache_sysfs_exit(cpu
);
983 free_cache_attributes(cpu
);
985 update_per_cpu_data_slice_size(false, cpu
, cpu_map
);
989 static int __init
cacheinfo_sysfs_init(void)
991 return cpuhp_setup_state(CPUHP_AP_BASE_CACHEINFO_ONLINE
,
992 "base/cacheinfo:online",
993 cacheinfo_cpu_online
, cacheinfo_cpu_pre_down
);
995 device_initcall(cacheinfo_sysfs_init
);