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/sysemu.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"
39 QemuOptsList qemu_numa_opts
= {
41 .implied_opt_name
= "type",
42 .head
= QTAILQ_HEAD_INITIALIZER(qemu_numa_opts
.head
),
43 .desc
= { { 0 } } /* validated with OptsVisitor */
46 static int have_memdevs
= -1;
48 static void numa_node_parse(NumaNodeOptions
*node
, QemuOpts
*opts
, Error
**errp
)
51 uint16List
*cpus
= NULL
;
53 if (node
->has_nodeid
) {
54 nodenr
= node
->nodeid
;
56 nodenr
= nb_numa_nodes
;
59 if (nodenr
>= MAX_NODES
) {
60 error_setg(errp
, "Max number of NUMA nodes reached: %"
65 for (cpus
= node
->cpus
; cpus
; cpus
= cpus
->next
) {
66 if (cpus
->value
> MAX_CPUMASK_BITS
) {
67 error_setg(errp
, "CPU number %" PRIu16
" is bigger than %d",
68 cpus
->value
, MAX_CPUMASK_BITS
);
71 bitmap_set(numa_info
[nodenr
].node_cpu
, cpus
->value
, 1);
74 if (node
->has_mem
&& node
->has_memdev
) {
75 error_setg(errp
, "qemu: cannot specify both mem= and memdev=\n");
79 if (have_memdevs
== -1) {
80 have_memdevs
= node
->has_memdev
;
82 if (node
->has_memdev
!= have_memdevs
) {
83 error_setg(errp
, "qemu: memdev option must be specified for either "
89 uint64_t mem_size
= node
->mem
;
90 const char *mem_str
= qemu_opt_get(opts
, "mem");
91 /* Fix up legacy suffix-less format */
92 if (g_ascii_isdigit(mem_str
[strlen(mem_str
) - 1])) {
95 numa_info
[nodenr
].node_mem
= mem_size
;
97 if (node
->has_memdev
) {
99 o
= object_resolve_path_type(node
->memdev
, TYPE_MEMORY_BACKEND
, NULL
);
101 error_setg(errp
, "memdev=%s is ambiguous", node
->memdev
);
106 numa_info
[nodenr
].node_mem
= object_property_get_int(o
, "size", NULL
);
107 numa_info
[nodenr
].node_memdev
= MEMORY_BACKEND(o
);
109 numa_info
[nodenr
].present
= true;
110 max_numa_nodeid
= MAX(max_numa_nodeid
, nodenr
+ 1);
113 int numa_init_func(QemuOpts
*opts
, void *opaque
)
115 NumaOptions
*object
= NULL
;
119 OptsVisitor
*ov
= opts_visitor_new(opts
);
120 visit_type_NumaOptions(opts_get_visitor(ov
), &object
, NULL
, &err
);
121 opts_visitor_cleanup(ov
);
128 switch (object
->kind
) {
129 case NUMA_OPTIONS_KIND_NODE
:
130 numa_node_parse(object
->node
, opts
, &err
);
143 qerror_report_err(err
);
147 QapiDeallocVisitor
*dv
= qapi_dealloc_visitor_new();
148 visit_type_NumaOptions(qapi_dealloc_get_visitor(dv
),
149 &object
, NULL
, NULL
);
150 qapi_dealloc_visitor_cleanup(dv
);
156 void set_numa_nodes(void)
158 if (nb_numa_nodes
> 0) {
162 if (nb_numa_nodes
> MAX_NODES
) {
163 nb_numa_nodes
= MAX_NODES
;
166 /* If no memory size is given for any node, assume the default case
167 * and distribute the available memory equally across all nodes
169 for (i
= 0; i
< nb_numa_nodes
; i
++) {
170 if (numa_info
[i
].node_mem
!= 0) {
174 if (i
== nb_numa_nodes
) {
175 uint64_t usedmem
= 0;
177 /* On Linux, each node's border has to be 8MB aligned,
178 * the final node gets the rest.
180 for (i
= 0; i
< nb_numa_nodes
- 1; i
++) {
181 numa_info
[i
].node_mem
= (ram_size
/ nb_numa_nodes
) &
183 usedmem
+= numa_info
[i
].node_mem
;
185 numa_info
[i
].node_mem
= ram_size
- usedmem
;
189 for (i
= 0; i
< nb_numa_nodes
; i
++) {
190 numa_total
+= numa_info
[i
].node_mem
;
192 if (numa_total
!= ram_size
) {
193 error_report("total memory for NUMA nodes (%" PRIu64
")"
194 " should equal RAM size (" RAM_ADDR_FMT
")",
195 numa_total
, ram_size
);
199 for (i
= 0; i
< nb_numa_nodes
; i
++) {
200 if (!bitmap_empty(numa_info
[i
].node_cpu
, MAX_CPUMASK_BITS
)) {
204 /* assigning the VCPUs round-robin is easier to implement, guest OSes
205 * must cope with this anyway, because there are BIOSes out there in
206 * real machines which also use this scheme.
208 if (i
== nb_numa_nodes
) {
209 for (i
= 0; i
< max_cpus
; i
++) {
210 set_bit(i
, numa_info
[i
% nb_numa_nodes
].node_cpu
);
216 void set_numa_modes(void)
222 for (i
= 0; i
< nb_numa_nodes
; i
++) {
223 if (test_bit(cpu
->cpu_index
, numa_info
[i
].node_cpu
)) {
230 static void allocate_system_memory_nonnuma(MemoryRegion
*mr
, Object
*owner
,
237 memory_region_init_ram_from_file(mr
, owner
, name
, ram_size
, false,
240 /* Legacy behavior: if allocation failed, fall back to
241 * regular RAM allocation.
244 qerror_report_err(err
);
246 memory_region_init_ram(mr
, owner
, name
, ram_size
);
249 fprintf(stderr
, "-mem-path not supported on this host\n");
253 memory_region_init_ram(mr
, owner
, name
, ram_size
);
255 vmstate_register_ram_global(mr
);
258 void memory_region_allocate_system_memory(MemoryRegion
*mr
, Object
*owner
,
265 if (nb_numa_nodes
== 0 || !have_memdevs
) {
266 allocate_system_memory_nonnuma(mr
, owner
, name
, ram_size
);
270 memory_region_init(mr
, owner
, name
, ram_size
);
271 for (i
= 0; i
< MAX_NODES
; i
++) {
272 Error
*local_err
= NULL
;
273 uint64_t size
= numa_info
[i
].node_mem
;
274 HostMemoryBackend
*backend
= numa_info
[i
].node_memdev
;
278 MemoryRegion
*seg
= host_memory_backend_get_memory(backend
, &local_err
);
280 qerror_report_err(local_err
);
284 memory_region_add_subregion(mr
, addr
, seg
);
285 vmstate_register_ram_global(seg
);
290 static int query_memdev(Object
*obj
, void *opaque
)
292 MemdevList
**list
= opaque
;
295 if (object_dynamic_cast(obj
, TYPE_MEMORY_BACKEND
)) {
296 MemdevList
*m
= g_malloc0(sizeof(*m
));
298 m
->value
= g_malloc0(sizeof(*m
->value
));
300 m
->value
->size
= object_property_get_int(obj
, "size",
306 m
->value
->merge
= object_property_get_bool(obj
, "merge",
312 m
->value
->dump
= object_property_get_bool(obj
, "dump",
318 m
->value
->prealloc
= object_property_get_bool(obj
,
324 m
->value
->policy
= object_property_get_enum(obj
,
326 HostMemPolicy_lookup
,
332 object_property_get_uint16List(obj
, "host-nodes",
333 &m
->value
->host_nodes
, &err
);
347 MemdevList
*qmp_query_memdev(Error
**errp
)
350 MemdevList
*list
= NULL
, *m
;
352 obj
= object_resolve_path("/objects", NULL
);
357 if (object_child_foreach(obj
, query_memdev
, &list
) != 0) {