2 * NUMA parameter parsing routines
4 * Copyright (c) 2014 Fujitsu Ltd.
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "sysemu/numa.h"
26 #include "exec/cpu-common.h"
27 #include "qemu/bitmap.h"
29 #include "qemu/error-report.h"
30 #include "include/exec/cpu-common.h" /* for RAM_ADDR_FMT */
31 #include "qapi-visit.h"
32 #include "qapi/opts-visitor.h"
33 #include "qapi/dealloc-visitor.h"
34 #include "qapi/qmp/qerror.h"
35 #include "hw/boards.h"
36 #include "sysemu/hostmem.h"
37 #include "qmp-commands.h"
38 #include "hw/mem/pc-dimm.h"
40 QemuOptsList qemu_numa_opts
= {
42 .implied_opt_name
= "type",
43 .head
= QTAILQ_HEAD_INITIALIZER(qemu_numa_opts
.head
),
44 .desc
= { { 0 } } /* validated with OptsVisitor */
47 static int have_memdevs
= -1;
48 static int max_numa_nodeid
; /* Highest specified NUMA node ID, plus one.
49 * For all nodes, nodeid < max_numa_nodeid
52 NodeInfo numa_info
[MAX_NODES
];
54 static void numa_node_parse(NumaNodeOptions
*node
, QemuOpts
*opts
, Error
**errp
)
57 uint16List
*cpus
= NULL
;
59 if (node
->has_nodeid
) {
60 nodenr
= node
->nodeid
;
62 nodenr
= nb_numa_nodes
;
65 if (nodenr
>= MAX_NODES
) {
66 error_setg(errp
, "Max number of NUMA nodes reached: %"
71 if (numa_info
[nodenr
].present
) {
72 error_setg(errp
, "Duplicate NUMA nodeid: %" PRIu16
, nodenr
);
76 for (cpus
= node
->cpus
; cpus
; cpus
= cpus
->next
) {
77 if (cpus
->value
> MAX_CPUMASK_BITS
) {
78 error_setg(errp
, "CPU number %" PRIu16
" is bigger than %d",
79 cpus
->value
, MAX_CPUMASK_BITS
);
82 bitmap_set(numa_info
[nodenr
].node_cpu
, cpus
->value
, 1);
85 if (node
->has_mem
&& node
->has_memdev
) {
86 error_setg(errp
, "qemu: cannot specify both mem= and memdev=\n");
90 if (have_memdevs
== -1) {
91 have_memdevs
= node
->has_memdev
;
93 if (node
->has_memdev
!= have_memdevs
) {
94 error_setg(errp
, "qemu: memdev option must be specified for either "
100 uint64_t mem_size
= node
->mem
;
101 const char *mem_str
= qemu_opt_get(opts
, "mem");
102 /* Fix up legacy suffix-less format */
103 if (g_ascii_isdigit(mem_str
[strlen(mem_str
) - 1])) {
106 numa_info
[nodenr
].node_mem
= mem_size
;
108 if (node
->has_memdev
) {
110 o
= object_resolve_path_type(node
->memdev
, TYPE_MEMORY_BACKEND
, NULL
);
112 error_setg(errp
, "memdev=%s is ambiguous", node
->memdev
);
117 numa_info
[nodenr
].node_mem
= object_property_get_int(o
, "size", NULL
);
118 numa_info
[nodenr
].node_memdev
= MEMORY_BACKEND(o
);
120 numa_info
[nodenr
].present
= true;
121 max_numa_nodeid
= MAX(max_numa_nodeid
, nodenr
+ 1);
124 int numa_init_func(QemuOpts
*opts
, void *opaque
)
126 NumaOptions
*object
= NULL
;
130 OptsVisitor
*ov
= opts_visitor_new(opts
);
131 visit_type_NumaOptions(opts_get_visitor(ov
), &object
, NULL
, &err
);
132 opts_visitor_cleanup(ov
);
139 switch (object
->kind
) {
140 case NUMA_OPTIONS_KIND_NODE
:
141 numa_node_parse(object
->node
, opts
, &err
);
154 qerror_report_err(err
);
158 QapiDeallocVisitor
*dv
= qapi_dealloc_visitor_new();
159 visit_type_NumaOptions(qapi_dealloc_get_visitor(dv
),
160 &object
, NULL
, NULL
);
161 qapi_dealloc_visitor_cleanup(dv
);
167 void set_numa_nodes(void)
171 assert(max_numa_nodeid
<= MAX_NODES
);
173 /* No support for sparse NUMA node IDs yet: */
174 for (i
= max_numa_nodeid
- 1; i
>= 0; i
--) {
175 /* Report large node IDs first, to make mistakes easier to spot */
176 if (!numa_info
[i
].present
) {
177 error_report("numa: Node ID missing: %d", i
);
182 /* This must be always true if all nodes are present: */
183 assert(nb_numa_nodes
== max_numa_nodeid
);
185 if (nb_numa_nodes
> 0) {
188 if (nb_numa_nodes
> MAX_NODES
) {
189 nb_numa_nodes
= MAX_NODES
;
192 /* If no memory size is given for any node, assume the default case
193 * and distribute the available memory equally across all nodes
195 for (i
= 0; i
< nb_numa_nodes
; i
++) {
196 if (numa_info
[i
].node_mem
!= 0) {
200 if (i
== nb_numa_nodes
) {
201 uint64_t usedmem
= 0;
203 /* On Linux, each node's border has to be 8MB aligned,
204 * the final node gets the rest.
206 for (i
= 0; i
< nb_numa_nodes
- 1; i
++) {
207 numa_info
[i
].node_mem
= (ram_size
/ nb_numa_nodes
) &
209 usedmem
+= numa_info
[i
].node_mem
;
211 numa_info
[i
].node_mem
= ram_size
- usedmem
;
215 for (i
= 0; i
< nb_numa_nodes
; i
++) {
216 numa_total
+= numa_info
[i
].node_mem
;
218 if (numa_total
!= ram_size
) {
219 error_report("total memory for NUMA nodes (0x%" PRIx64
")"
220 " should equal RAM size (0x" RAM_ADDR_FMT
")",
221 numa_total
, ram_size
);
225 for (i
= 0; i
< nb_numa_nodes
; i
++) {
226 if (!bitmap_empty(numa_info
[i
].node_cpu
, MAX_CPUMASK_BITS
)) {
230 /* assigning the VCPUs round-robin is easier to implement, guest OSes
231 * must cope with this anyway, because there are BIOSes out there in
232 * real machines which also use this scheme.
234 if (i
== nb_numa_nodes
) {
235 for (i
= 0; i
< max_cpus
; i
++) {
236 set_bit(i
, numa_info
[i
% nb_numa_nodes
].node_cpu
);
242 void set_numa_modes(void)
248 for (i
= 0; i
< nb_numa_nodes
; i
++) {
249 if (test_bit(cpu
->cpu_index
, numa_info
[i
].node_cpu
)) {
256 static void allocate_system_memory_nonnuma(MemoryRegion
*mr
, Object
*owner
,
263 memory_region_init_ram_from_file(mr
, owner
, name
, ram_size
, false,
266 /* Legacy behavior: if allocation failed, fall back to
267 * regular RAM allocation.
270 qerror_report_err(err
);
272 memory_region_init_ram(mr
, owner
, name
, ram_size
, &error_abort
);
275 fprintf(stderr
, "-mem-path not supported on this host\n");
279 memory_region_init_ram(mr
, owner
, name
, ram_size
, &error_abort
);
281 vmstate_register_ram_global(mr
);
284 void memory_region_allocate_system_memory(MemoryRegion
*mr
, Object
*owner
,
291 if (nb_numa_nodes
== 0 || !have_memdevs
) {
292 allocate_system_memory_nonnuma(mr
, owner
, name
, ram_size
);
296 memory_region_init(mr
, owner
, name
, ram_size
);
297 for (i
= 0; i
< MAX_NODES
; i
++) {
298 Error
*local_err
= NULL
;
299 uint64_t size
= numa_info
[i
].node_mem
;
300 HostMemoryBackend
*backend
= numa_info
[i
].node_memdev
;
304 MemoryRegion
*seg
= host_memory_backend_get_memory(backend
, &local_err
);
306 qerror_report_err(local_err
);
310 if (memory_region_is_mapped(seg
)) {
311 char *path
= object_get_canonical_path_component(OBJECT(backend
));
312 error_report("memory backend %s is used multiple times. Each "
313 "-numa option must use a different memdev value.",
318 memory_region_add_subregion(mr
, addr
, seg
);
319 vmstate_register_ram_global(seg
);
324 static void numa_stat_memory_devices(uint64_t node_mem
[])
326 MemoryDeviceInfoList
*info_list
= NULL
;
327 MemoryDeviceInfoList
**prev
= &info_list
;
328 MemoryDeviceInfoList
*info
;
330 qmp_pc_dimm_device_list(qdev_get_machine(), &prev
);
331 for (info
= info_list
; info
; info
= info
->next
) {
332 MemoryDeviceInfo
*value
= info
->value
;
335 switch (value
->kind
) {
336 case MEMORY_DEVICE_INFO_KIND_DIMM
:
337 node_mem
[value
->dimm
->node
] += value
->dimm
->size
;
344 qapi_free_MemoryDeviceInfoList(info_list
);
347 void query_numa_node_mem(uint64_t node_mem
[])
351 if (nb_numa_nodes
<= 0) {
355 numa_stat_memory_devices(node_mem
);
356 for (i
= 0; i
< nb_numa_nodes
; i
++) {
357 node_mem
[i
] += numa_info
[i
].node_mem
;
361 static int query_memdev(Object
*obj
, void *opaque
)
363 MemdevList
**list
= opaque
;
364 MemdevList
*m
= NULL
;
367 if (object_dynamic_cast(obj
, TYPE_MEMORY_BACKEND
)) {
368 m
= g_malloc0(sizeof(*m
));
370 m
->value
= g_malloc0(sizeof(*m
->value
));
372 m
->value
->size
= object_property_get_int(obj
, "size",
378 m
->value
->merge
= object_property_get_bool(obj
, "merge",
384 m
->value
->dump
= object_property_get_bool(obj
, "dump",
390 m
->value
->prealloc
= object_property_get_bool(obj
,
396 m
->value
->policy
= object_property_get_enum(obj
,
398 HostMemPolicy_lookup
,
404 object_property_get_uint16List(obj
, "host-nodes",
405 &m
->value
->host_nodes
, &err
);
422 MemdevList
*qmp_query_memdev(Error
**errp
)
425 MemdevList
*list
= NULL
;
427 obj
= object_resolve_path("/objects", NULL
);
432 if (object_child_foreach(obj
, query_memdev
, &list
) != 0) {
439 qapi_free_MemdevList(list
);