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);
111 int numa_init_func(QemuOpts *opts, void *opaque)
113 NumaOptions *object = NULL;
117 OptsVisitor *ov = opts_visitor_new(opts);
118 visit_type_NumaOptions(opts_get_visitor(ov), &object, NULL, &err);
119 opts_visitor_cleanup(ov);
126 switch (object->kind) {
127 case NUMA_OPTIONS_KIND_NODE:
128 numa_node_parse(object->node, opts, &err);
141 qerror_report_err(err);
145 QapiDeallocVisitor *dv = qapi_dealloc_visitor_new();
146 visit_type_NumaOptions(qapi_dealloc_get_visitor(dv),
147 &object, NULL, NULL);
148 qapi_dealloc_visitor_cleanup(dv);
154 void set_numa_nodes(void)
156 if (nb_numa_nodes > 0) {
160 if (nb_numa_nodes > MAX_NODES) {
161 nb_numa_nodes = MAX_NODES;
164 /* If no memory size if given for any node, assume the default case
165 * and distribute the available memory equally across all nodes
167 for (i = 0; i < nb_numa_nodes; i++) {
168 if (numa_info[i].node_mem != 0) {
172 if (i == nb_numa_nodes) {
173 uint64_t usedmem = 0;
175 /* On Linux, the each node's border has to be 8MB aligned,
176 * the final node gets the rest.
178 for (i = 0; i < nb_numa_nodes - 1; i++) {
179 numa_info[i].node_mem = (ram_size / nb_numa_nodes) &
181 usedmem += numa_info[i].node_mem;
183 numa_info[i].node_mem = ram_size - usedmem;
187 for (i = 0; i < nb_numa_nodes; i++) {
188 numa_total += numa_info[i].node_mem;
190 if (numa_total != ram_size) {
191 error_report("total memory for NUMA nodes (%" PRIu64 ")"
192 " should equal RAM size (" RAM_ADDR_FMT ")",
193 numa_total, ram_size);
197 for (i = 0; i < nb_numa_nodes; i++) {
198 if (!bitmap_empty(numa_info[i].node_cpu, MAX_CPUMASK_BITS)) {
202 /* assigning the VCPUs round-robin is easier to implement, guest OSes
203 * must cope with this anyway, because there are BIOSes out there in
204 * real machines which also use this scheme.
206 if (i == nb_numa_nodes) {
207 for (i = 0; i < max_cpus; i++) {
208 set_bit(i, numa_info[i % nb_numa_nodes].node_cpu);
214 void set_numa_modes(void)
220 for (i = 0; i < nb_numa_nodes; i++) {
221 if (test_bit(cpu->cpu_index, numa_info[i].node_cpu)) {
228 static void allocate_system_memory_nonnuma(MemoryRegion *mr, Object *owner,
235 memory_region_init_ram_from_file(mr, owner, name, ram_size, false,
238 /* Legacy behavior: if allocation failed, fall back to
239 * regular RAM allocation.
242 qerror_report_err(err);
244 memory_region_init_ram(mr, owner, name, ram_size);
247 fprintf(stderr, "-mem-path not supported on this host\n");
251 memory_region_init_ram(mr, owner, name, ram_size);
253 vmstate_register_ram_global(mr);
256 void memory_region_allocate_system_memory(MemoryRegion *mr, Object *owner,
263 if (nb_numa_nodes == 0 || !have_memdevs) {
264 allocate_system_memory_nonnuma(mr, owner, name, ram_size);
268 memory_region_init(mr, owner, name, ram_size);
269 for (i = 0; i < MAX_NODES; i++) {
270 Error *local_err = NULL;
271 uint64_t size = numa_info[i].node_mem;
272 HostMemoryBackend *backend = numa_info[i].node_memdev;
276 MemoryRegion *seg = host_memory_backend_get_memory(backend, &local_err);
278 qerror_report_err(local_err);
282 memory_region_add_subregion(mr, addr, seg);
283 vmstate_register_ram_global(seg);
288 static int query_memdev(Object *obj, void *opaque)
290 MemdevList **list = opaque;
293 if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) {
294 MemdevList *m = g_malloc0(sizeof(*m));
296 m->value = g_malloc0(sizeof(*m->value));
298 m->value->size = object_property_get_int(obj, "size",
304 m->value->merge = object_property_get_bool(obj, "merge",
310 m->value->dump = object_property_get_bool(obj, "dump",
316 m->value->prealloc = object_property_get_bool(obj,
322 m->value->policy = object_property_get_enum(obj,
324 HostMemPolicy_lookup,
330 object_property_get_uint16List(obj, "host-nodes",
331 &m->value->host_nodes, &err);
345 MemdevList *qmp_query_memdev(Error **errp)
348 MemdevList *list = NULL, *m;
350 obj = object_resolve_path("/objects", NULL);
355 if (object_child_foreach(obj, query_memdev, &list) != 0) {