2 * Copyright (c) 2012 The DragonFly Project. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in
12 * the documentation and/or other materials provided with the
14 * 3. Neither the name of The DragonFly Project nor the names of its
15 * contributors may be used to endorse or promote products derived
16 * from this software without specific, prior written permission.
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
21 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
22 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
24 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
26 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
27 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
28 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/sysctl.h>
38 #include <sys/cpu_topology.h>
40 #include <machine/smp.h>
46 #define INDENT_BUF_SIZE LEVEL_NO*3
49 /* Per-cpu sysctl nodes and info */
50 struct per_cpu_sysctl_info
{
51 struct sysctl_ctx_list sysctl_ctx
;
52 struct sysctl_oid
*sysctl_tree
;
56 char physical_siblings
[8*MAXCPU
];
57 char core_siblings
[8*MAXCPU
];
59 typedef struct per_cpu_sysctl_info per_cpu_sysctl_info_t
;
61 static cpu_node_t cpu_topology_nodes
[MAXCPU
]; /* Memory for topology */
62 static cpu_node_t
*cpu_root_node
; /* Root node pointer */
64 static struct sysctl_ctx_list cpu_topology_sysctl_ctx
;
65 static struct sysctl_oid
*cpu_topology_sysctl_tree
;
66 static char cpu_topology_members
[8*MAXCPU
];
67 static per_cpu_sysctl_info_t
*pcpu_sysctl
;
68 static void sbuf_print_cpuset(struct sbuf
*sb
, cpumask_t
*mask
);
70 int cpu_topology_levels_number
= 1;
71 cpu_node_t
*root_cpu_node
;
73 MALLOC_DEFINE(M_PCPUSYS
, "pcpusys", "pcpu sysctl topology");
76 /* Get the next valid apicid starting
77 * from current apicid (curr_apicid
80 get_next_valid_apicid(int curr_apicid
)
82 int next_apicid
= curr_apicid
;
86 while(get_cpuid_from_apicid(next_apicid
) == -1 &&
87 next_apicid
< NAPICID
);
88 if (next_apicid
== NAPICID
) {
89 kprintf("Warning: No next valid APICID found. Returning -1\n");
95 /* Generic topology tree. The parameters have the following meaning:
96 * - children_no_per_level : the number of children on each level
97 * - level_types : the type of the level (THREAD, CORE, CHIP, etc)
98 * - cur_level : the current level of the tree
99 * - node : the current node
100 * - last_free_node : the last free node in the global array.
101 * - cpuid : basicly this are the ids of the leafs
104 build_topology_tree(int *children_no_per_level
,
105 uint8_t *level_types
,
108 cpu_node_t
**last_free_node
,
113 node
->child_no
= children_no_per_level
[cur_level
];
114 node
->type
= level_types
[cur_level
];
115 CPUMASK_ASSZERO(node
->members
);
116 node
->compute_unit_id
= -1;
118 if (node
->child_no
== 0) {
119 *apicid
= get_next_valid_apicid(*apicid
);
120 CPUMASK_ASSBIT(node
->members
, get_cpuid_from_apicid(*apicid
));
124 if (node
->parent_node
== NULL
)
125 root_cpu_node
= node
;
127 for (i
= 0; i
< node
->child_no
; i
++) {
128 node
->child_node
[i
] = *last_free_node
;
131 node
->child_node
[i
]->parent_node
= node
;
133 build_topology_tree(children_no_per_level
,
140 CPUMASK_ORMASK(node
->members
, node
->child_node
[i
]->members
);
144 #if defined(__x86_64__) && !defined(_KERNEL_VIRTUAL)
146 migrate_elements(cpu_node_t
**a
, int n
, int pos
)
150 for (i
= pos
; i
< n
- 1 ; i
++) {
157 /* Build CPU topology. The detection is made by comparing the
158 * chip, core and logical IDs of each CPU with the IDs of the
159 * BSP. When we found a match, at that level the CPUs are siblings.
162 build_cpu_topology(void)
164 detect_cpu_topology();
167 int threads_per_core
= 0;
168 int cores_per_chip
= 0;
169 int chips_per_package
= 0;
170 int children_no_per_level
[LEVEL_NO
];
171 uint8_t level_types
[LEVEL_NO
];
174 cpu_node_t
*root
= &cpu_topology_nodes
[0];
175 cpu_node_t
*last_free_node
= root
+ 1;
177 /* Assume that the topology is uniform.
178 * Find the number of siblings within chip
179 * and witin core to build up the topology
181 for (i
= 0; i
< ncpus
; i
++) {
184 CPUMASK_ASSBIT(mask
, i
);
186 if (CPUMASK_TESTMASK(mask
, smp_active_mask
) == 0)
189 if (get_chip_ID(BSPID
) == get_chip_ID(i
))
194 if (get_core_number_within_chip(BSPID
) ==
195 get_core_number_within_chip(i
))
199 cores_per_chip
/= threads_per_core
;
200 chips_per_package
= ncpus
/ (cores_per_chip
* threads_per_core
);
203 kprintf("CPU Topology: cores_per_chip: %d; threads_per_core: %d; chips_per_package: %d;\n",
204 cores_per_chip
, threads_per_core
, chips_per_package
);
206 if (threads_per_core
> 1) { /* HT available - 4 levels */
208 children_no_per_level
[0] = chips_per_package
;
209 children_no_per_level
[1] = cores_per_chip
;
210 children_no_per_level
[2] = threads_per_core
;
211 children_no_per_level
[3] = 0;
213 level_types
[0] = PACKAGE_LEVEL
;
214 level_types
[1] = CHIP_LEVEL
;
215 level_types
[2] = CORE_LEVEL
;
216 level_types
[3] = THREAD_LEVEL
;
218 build_topology_tree(children_no_per_level
,
225 cpu_topology_levels_number
= 4;
227 } else if (cores_per_chip
> 1) { /* No HT available - 3 levels */
229 children_no_per_level
[0] = chips_per_package
;
230 children_no_per_level
[1] = cores_per_chip
;
231 children_no_per_level
[2] = 0;
233 level_types
[0] = PACKAGE_LEVEL
;
234 level_types
[1] = CHIP_LEVEL
;
235 level_types
[2] = CORE_LEVEL
;
237 build_topology_tree(children_no_per_level
,
244 cpu_topology_levels_number
= 3;
246 } else { /* No HT and no Multi-Core - 2 levels */
248 children_no_per_level
[0] = chips_per_package
;
249 children_no_per_level
[1] = 0;
251 level_types
[0] = PACKAGE_LEVEL
;
252 level_types
[1] = CHIP_LEVEL
;
254 build_topology_tree(children_no_per_level
,
261 cpu_topology_levels_number
= 2;
265 cpu_root_node
= root
;
268 #if defined(__x86_64__) && !defined(_KERNEL_VIRTUAL)
269 if (fix_amd_topology() == 0) {
270 int visited
[MAXCPU
], i
, j
, pos
, cpuid
;
271 cpu_node_t
*leaf
, *parent
;
273 bzero(visited
, MAXCPU
* sizeof(int));
275 for (i
= 0; i
< ncpus
; i
++) {
276 if (visited
[i
] == 0) {
279 leaf
= get_cpu_node_by_cpuid(i
);
281 if (leaf
->type
== CORE_LEVEL
) {
282 parent
= leaf
->parent_node
;
284 last_free_node
->child_node
[0] = leaf
;
285 last_free_node
->child_no
= 1;
286 last_free_node
->members
= leaf
->members
;
287 last_free_node
->compute_unit_id
= leaf
->compute_unit_id
;
288 last_free_node
->parent_node
= parent
;
289 last_free_node
->type
= CORE_LEVEL
;
292 for (j
= 0; j
< parent
->child_no
; j
++) {
293 if (parent
->child_node
[j
] != leaf
) {
295 cpuid
= BSFCPUMASK(parent
->child_node
[j
]->members
);
296 if (visited
[cpuid
] == 0 &&
297 parent
->child_node
[j
]->compute_unit_id
== leaf
->compute_unit_id
) {
299 last_free_node
->child_node
[last_free_node
->child_no
] = parent
->child_node
[j
];
300 last_free_node
->child_no
++;
301 CPUMASK_ORMASK(last_free_node
->members
, parent
->child_node
[j
]->members
);
303 parent
->child_node
[j
]->type
= THREAD_LEVEL
;
304 parent
->child_node
[j
]->parent_node
= last_free_node
;
307 migrate_elements(parent
->child_node
, parent
->child_no
, j
);
315 if (last_free_node
->child_no
> 1) {
316 parent
->child_node
[pos
] = last_free_node
;
317 leaf
->type
= THREAD_LEVEL
;
318 leaf
->parent_node
= last_free_node
;
328 /* Recursive function helper to print the CPU topology tree */
330 print_cpu_topology_tree_sysctl_helper(cpu_node_t
*node
,
339 sbuf_bcat(sb
, buf
, buf_len
);
341 sbuf_printf(sb
, "\\-");
342 buf
[buf_len
] = ' ';buf_len
++;
343 buf
[buf_len
] = ' ';buf_len
++;
345 sbuf_printf(sb
, "|-");
346 buf
[buf_len
] = '|';buf_len
++;
347 buf
[buf_len
] = ' ';buf_len
++;
350 bsr_member
= BSRCPUMASK(node
->members
);
352 if (node
->type
== PACKAGE_LEVEL
) {
353 sbuf_printf(sb
,"PACKAGE MEMBERS: ");
354 } else if (node
->type
== CHIP_LEVEL
) {
355 sbuf_printf(sb
,"CHIP ID %d: ",
356 get_chip_ID(bsr_member
));
357 } else if (node
->type
== CORE_LEVEL
) {
358 if (node
->compute_unit_id
!= (uint8_t)-1) {
359 sbuf_printf(sb
,"Compute Unit ID %d: ",
360 node
->compute_unit_id
);
362 sbuf_printf(sb
,"CORE ID %d: ",
363 get_core_number_within_chip(bsr_member
));
365 } else if (node
->type
== THREAD_LEVEL
) {
366 if (node
->compute_unit_id
!= (uint8_t)-1) {
367 sbuf_printf(sb
,"CORE ID %d: ",
368 get_core_number_within_chip(bsr_member
));
370 sbuf_printf(sb
,"THREAD ID %d: ",
371 get_logical_CPU_number_within_core(bsr_member
));
374 sbuf_printf(sb
,"UNKNOWN: ");
376 sbuf_print_cpuset(sb
, &node
->members
);
377 sbuf_printf(sb
,"\n");
379 for (i
= 0; i
< node
->child_no
; i
++) {
380 print_cpu_topology_tree_sysctl_helper(node
->child_node
[i
],
381 sb
, buf
, buf_len
, i
== (node
->child_no
-1));
385 /* SYSCTL PROCEDURE for printing the CPU Topology tree */
387 print_cpu_topology_tree_sysctl(SYSCTL_HANDLER_ARGS
)
391 char buf
[INDENT_BUF_SIZE
];
393 KASSERT(cpu_root_node
!= NULL
, ("cpu_root_node isn't initialized"));
395 sb
= sbuf_new(NULL
, NULL
, 500, SBUF_AUTOEXTEND
);
399 sbuf_printf(sb
,"\n");
400 print_cpu_topology_tree_sysctl_helper(cpu_root_node
, sb
, buf
, 0, 1);
404 ret
= SYSCTL_OUT(req
, sbuf_data(sb
), sbuf_len(sb
));
411 /* SYSCTL PROCEDURE for printing the CPU Topology level description */
413 print_cpu_topology_level_description_sysctl(SYSCTL_HANDLER_ARGS
)
418 sb
= sbuf_new(NULL
, NULL
, 500, SBUF_AUTOEXTEND
);
422 if (cpu_topology_levels_number
== 4) /* HT available */
423 sbuf_printf(sb
, "0 - thread; 1 - core; 2 - socket; 3 - anything");
424 else if (cpu_topology_levels_number
== 3) /* No HT available */
425 sbuf_printf(sb
, "0 - core; 1 - socket; 2 - anything");
426 else if (cpu_topology_levels_number
== 2) /* No HT and no Multi-Core */
427 sbuf_printf(sb
, "0 - socket; 1 - anything");
429 sbuf_printf(sb
, "Unknown");
433 ret
= SYSCTL_OUT(req
, sbuf_data(sb
), sbuf_len(sb
));
440 /* Find a cpu_node_t by a mask */
442 get_cpu_node_by_cpumask(cpu_node_t
* node
,
445 cpu_node_t
* found
= NULL
;
448 if (CPUMASK_CMPMASKEQ(node
->members
, mask
))
451 for (i
= 0; i
< node
->child_no
; i
++) {
452 found
= get_cpu_node_by_cpumask(node
->child_node
[i
], mask
);
461 get_cpu_node_by_cpuid(int cpuid
) {
464 CPUMASK_ASSBIT(mask
, cpuid
);
466 KASSERT(cpu_root_node
!= NULL
, ("cpu_root_node isn't initialized"));
468 return get_cpu_node_by_cpumask(cpu_root_node
, mask
);
471 /* Get the mask of siblings for level_type of a cpuid */
473 get_cpumask_from_level(int cpuid
,
479 CPUMASK_ASSBIT(mask
, cpuid
);
481 KASSERT(cpu_root_node
!= NULL
, ("cpu_root_node isn't initialized"));
483 node
= get_cpu_node_by_cpumask(cpu_root_node
, mask
);
486 CPUMASK_ASSZERO(mask
);
490 while (node
!= NULL
) {
491 if (node
->type
== level_type
) {
492 return node
->members
;
494 node
= node
->parent_node
;
496 CPUMASK_ASSZERO(mask
);
501 static const cpu_node_t
*
502 get_cpu_node_by_chipid2(const cpu_node_t
*node
, int chip_id
)
506 if (node
->type
!= CHIP_LEVEL
) {
507 const cpu_node_t
*ret
= NULL
;
510 for (i
= 0; i
< node
->child_no
; ++i
) {
511 ret
= get_cpu_node_by_chipid2(node
->child_node
[i
],
519 cpuid
= BSRCPUMASK(node
->members
);
520 if (get_chip_ID(cpuid
) == chip_id
)
526 get_cpu_node_by_chipid(int chip_id
)
528 KASSERT(cpu_root_node
!= NULL
, ("cpu_root_node isn't initialized"));
529 return get_cpu_node_by_chipid2(cpu_root_node
, chip_id
);
532 /* init pcpu_sysctl structure info */
534 init_pcpu_topology_sysctl(void)
540 pcpu_sysctl
= kmalloc(sizeof(*pcpu_sysctl
) * MAXCPU
, M_PCPUSYS
,
543 for (i
= 0; i
< ncpus
; i
++) {
544 sbuf_new(&sb
, pcpu_sysctl
[i
].cpu_name
,
545 sizeof(pcpu_sysctl
[i
].cpu_name
), SBUF_FIXEDLEN
);
546 sbuf_printf(&sb
,"cpu%d", i
);
550 /* Get physical siblings */
551 mask
= get_cpumask_from_level(i
, CHIP_LEVEL
);
552 if (CPUMASK_TESTZERO(mask
)) {
553 pcpu_sysctl
[i
].physical_id
= INVALID_ID
;
557 sbuf_new(&sb
, pcpu_sysctl
[i
].physical_siblings
,
558 sizeof(pcpu_sysctl
[i
].physical_siblings
), SBUF_FIXEDLEN
);
559 sbuf_print_cpuset(&sb
, &mask
);
563 pcpu_sysctl
[i
].physical_id
= get_chip_ID(i
);
565 /* Get core siblings */
566 mask
= get_cpumask_from_level(i
, CORE_LEVEL
);
567 if (CPUMASK_TESTZERO(mask
)) {
568 pcpu_sysctl
[i
].core_id
= INVALID_ID
;
572 sbuf_new(&sb
, pcpu_sysctl
[i
].core_siblings
,
573 sizeof(pcpu_sysctl
[i
].core_siblings
), SBUF_FIXEDLEN
);
574 sbuf_print_cpuset(&sb
, &mask
);
578 pcpu_sysctl
[i
].core_id
= get_core_number_within_chip(i
);
583 /* Build SYSCTL structure for revealing
584 * the CPU Topology to user-space.
587 build_sysctl_cpu_topology(void)
592 /* SYSCTL new leaf for "cpu_topology" */
593 sysctl_ctx_init(&cpu_topology_sysctl_ctx
);
594 cpu_topology_sysctl_tree
= SYSCTL_ADD_NODE(&cpu_topology_sysctl_ctx
,
595 SYSCTL_STATIC_CHILDREN(_hw
),
600 /* SYSCTL cpu_topology "tree" entry */
601 SYSCTL_ADD_PROC(&cpu_topology_sysctl_ctx
,
602 SYSCTL_CHILDREN(cpu_topology_sysctl_tree
),
603 OID_AUTO
, "tree", CTLTYPE_STRING
| CTLFLAG_RD
,
604 NULL
, 0, print_cpu_topology_tree_sysctl
, "A",
605 "Tree print of CPU topology");
607 /* SYSCTL cpu_topology "level_description" entry */
608 SYSCTL_ADD_PROC(&cpu_topology_sysctl_ctx
,
609 SYSCTL_CHILDREN(cpu_topology_sysctl_tree
),
610 OID_AUTO
, "level_description", CTLTYPE_STRING
| CTLFLAG_RD
,
611 NULL
, 0, print_cpu_topology_level_description_sysctl
, "A",
612 "Level description of CPU topology");
614 /* SYSCTL cpu_topology "members" entry */
615 sbuf_new(&sb
, cpu_topology_members
,
616 sizeof(cpu_topology_members
), SBUF_FIXEDLEN
);
617 sbuf_print_cpuset(&sb
, &cpu_root_node
->members
);
620 SYSCTL_ADD_STRING(&cpu_topology_sysctl_ctx
,
621 SYSCTL_CHILDREN(cpu_topology_sysctl_tree
),
622 OID_AUTO
, "members", CTLFLAG_RD
,
623 cpu_topology_members
, 0,
624 "Members of the CPU Topology");
626 /* SYSCTL per_cpu info */
627 for (i
= 0; i
< ncpus
; i
++) {
628 /* New leaf : hw.cpu_topology.cpux */
629 sysctl_ctx_init(&pcpu_sysctl
[i
].sysctl_ctx
);
630 pcpu_sysctl
[i
].sysctl_tree
= SYSCTL_ADD_NODE(&pcpu_sysctl
[i
].sysctl_ctx
,
631 SYSCTL_CHILDREN(cpu_topology_sysctl_tree
),
633 pcpu_sysctl
[i
].cpu_name
,
636 /* Check if the physical_id found is valid */
637 if (pcpu_sysctl
[i
].physical_id
== INVALID_ID
) {
641 /* Add physical id info */
642 SYSCTL_ADD_INT(&pcpu_sysctl
[i
].sysctl_ctx
,
643 SYSCTL_CHILDREN(pcpu_sysctl
[i
].sysctl_tree
),
644 OID_AUTO
, "physical_id", CTLFLAG_RD
,
645 &pcpu_sysctl
[i
].physical_id
, 0,
648 /* Add physical siblings */
649 SYSCTL_ADD_STRING(&pcpu_sysctl
[i
].sysctl_ctx
,
650 SYSCTL_CHILDREN(pcpu_sysctl
[i
].sysctl_tree
),
651 OID_AUTO
, "physical_siblings", CTLFLAG_RD
,
652 pcpu_sysctl
[i
].physical_siblings
, 0,
653 "Physical siblings");
655 /* Check if the core_id found is valid */
656 if (pcpu_sysctl
[i
].core_id
== INVALID_ID
) {
660 /* Add core id info */
661 SYSCTL_ADD_INT(&pcpu_sysctl
[i
].sysctl_ctx
,
662 SYSCTL_CHILDREN(pcpu_sysctl
[i
].sysctl_tree
),
663 OID_AUTO
, "core_id", CTLFLAG_RD
,
664 &pcpu_sysctl
[i
].core_id
, 0,
667 /*Add core siblings */
668 SYSCTL_ADD_STRING(&pcpu_sysctl
[i
].sysctl_ctx
,
669 SYSCTL_CHILDREN(pcpu_sysctl
[i
].sysctl_tree
),
670 OID_AUTO
, "core_siblings", CTLFLAG_RD
,
671 pcpu_sysctl
[i
].core_siblings
, 0,
678 sbuf_print_cpuset(struct sbuf
*sb
, cpumask_t
*mask
)
685 sbuf_printf(sb
, "cpus(");
686 CPUSET_FOREACH(i
, *mask
) {
697 sbuf_printf(sb
, ", ");
699 sbuf_printf(sb
, "%d", b
);
701 sbuf_printf(sb
, "%d-%d", b
, e
- 1);
708 sbuf_printf(sb
, ", ");
711 sbuf_printf(sb
, "%d", b
);
713 sbuf_printf(sb
, "%d-%d", b
, e
- 1);
716 sbuf_printf(sb
, ") ");
719 /* Build the CPU Topology and SYSCTL Topology tree */
721 init_cpu_topology(void)
723 build_cpu_topology();
725 init_pcpu_topology_sysctl();
726 build_sysctl_cpu_topology();
728 SYSINIT(cpu_topology
, SI_BOOT2_CPU_TOPOLOGY
, SI_ORDER_FIRST
,
729 init_cpu_topology
, NULL
);