qxl: check release info object
[qemu.git] / numa.c
1 /*
2 * NUMA parameter parsing routines
3 *
4 * Copyright (c) 2014 Fujitsu Ltd.
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "sysemu/numa.h"
27 #include "exec/cpu-common.h"
28 #include "exec/ramlist.h"
29 #include "qemu/bitmap.h"
30 #include "qom/cpu.h"
31 #include "qemu/error-report.h"
32 #include "qapi/error.h"
33 #include "qapi/opts-visitor.h"
34 #include "qapi/qapi-commands-misc.h"
35 #include "qapi/qapi-visit-misc.h"
36 #include "hw/boards.h"
37 #include "sysemu/hostmem.h"
38 #include "hw/mem/pc-dimm.h"
39 #include "hw/mem/memory-device.h"
40 #include "qemu/option.h"
41 #include "qemu/config-file.h"
42 #include "qemu/cutils.h"
43
44 QemuOptsList qemu_numa_opts = {
45 .name = "numa",
46 .implied_opt_name = "type",
47 .head = QTAILQ_HEAD_INITIALIZER(qemu_numa_opts.head),
48 .desc = { { 0 } } /* validated with OptsVisitor */
49 };
50
51 static int have_memdevs = -1;
52 static int max_numa_nodeid; /* Highest specified NUMA node ID, plus one.
53 * For all nodes, nodeid < max_numa_nodeid
54 */
55 int nb_numa_nodes;
56 bool have_numa_distance;
57 NodeInfo numa_info[MAX_NODES];
58
59
60 static void parse_numa_node(MachineState *ms, NumaNodeOptions *node,
61 Error **errp)
62 {
63 Error *err = NULL;
64 uint16_t nodenr;
65 uint16List *cpus = NULL;
66 MachineClass *mc = MACHINE_GET_CLASS(ms);
67
68 if (node->has_nodeid) {
69 nodenr = node->nodeid;
70 } else {
71 nodenr = nb_numa_nodes;
72 }
73
74 if (nodenr >= MAX_NODES) {
75 error_setg(errp, "Max number of NUMA nodes reached: %"
76 PRIu16 "", nodenr);
77 return;
78 }
79
80 if (numa_info[nodenr].present) {
81 error_setg(errp, "Duplicate NUMA nodeid: %" PRIu16, nodenr);
82 return;
83 }
84
85 if (!mc->cpu_index_to_instance_props || !mc->get_default_cpu_node_id) {
86 error_setg(errp, "NUMA is not supported by this machine-type");
87 return;
88 }
89 for (cpus = node->cpus; cpus; cpus = cpus->next) {
90 CpuInstanceProperties props;
91 if (cpus->value >= max_cpus) {
92 error_setg(errp,
93 "CPU index (%" PRIu16 ")"
94 " should be smaller than maxcpus (%d)",
95 cpus->value, max_cpus);
96 return;
97 }
98 props = mc->cpu_index_to_instance_props(ms, cpus->value);
99 props.node_id = nodenr;
100 props.has_node_id = true;
101 machine_set_cpu_numa_node(ms, &props, &err);
102 if (err) {
103 error_propagate(errp, err);
104 return;
105 }
106 }
107
108 if (node->has_mem && node->has_memdev) {
109 error_setg(errp, "cannot specify both mem= and memdev=");
110 return;
111 }
112
113 if (have_memdevs == -1) {
114 have_memdevs = node->has_memdev;
115 }
116 if (node->has_memdev != have_memdevs) {
117 error_setg(errp, "memdev option must be specified for either "
118 "all or no nodes");
119 return;
120 }
121
122 if (node->has_mem) {
123 numa_info[nodenr].node_mem = node->mem;
124 }
125 if (node->has_memdev) {
126 Object *o;
127 o = object_resolve_path_type(node->memdev, TYPE_MEMORY_BACKEND, NULL);
128 if (!o) {
129 error_setg(errp, "memdev=%s is ambiguous", node->memdev);
130 return;
131 }
132
133 object_ref(o);
134 numa_info[nodenr].node_mem = object_property_get_uint(o, "size", NULL);
135 numa_info[nodenr].node_memdev = MEMORY_BACKEND(o);
136 }
137 numa_info[nodenr].present = true;
138 max_numa_nodeid = MAX(max_numa_nodeid, nodenr + 1);
139 nb_numa_nodes++;
140 }
141
142 static void parse_numa_distance(NumaDistOptions *dist, Error **errp)
143 {
144 uint16_t src = dist->src;
145 uint16_t dst = dist->dst;
146 uint8_t val = dist->val;
147
148 if (src >= MAX_NODES || dst >= MAX_NODES) {
149 error_setg(errp, "Parameter '%s' expects an integer between 0 and %d",
150 src >= MAX_NODES ? "src" : "dst", MAX_NODES - 1);
151 return;
152 }
153
154 if (!numa_info[src].present || !numa_info[dst].present) {
155 error_setg(errp, "Source/Destination NUMA node is missing. "
156 "Please use '-numa node' option to declare it first.");
157 return;
158 }
159
160 if (val < NUMA_DISTANCE_MIN) {
161 error_setg(errp, "NUMA distance (%" PRIu8 ") is invalid, "
162 "it shouldn't be less than %d.",
163 val, NUMA_DISTANCE_MIN);
164 return;
165 }
166
167 if (src == dst && val != NUMA_DISTANCE_MIN) {
168 error_setg(errp, "Local distance of node %d should be %d.",
169 src, NUMA_DISTANCE_MIN);
170 return;
171 }
172
173 numa_info[src].distance[dst] = val;
174 have_numa_distance = true;
175 }
176
177 static
178 void set_numa_options(MachineState *ms, NumaOptions *object, Error **errp)
179 {
180 Error *err = NULL;
181
182 switch (object->type) {
183 case NUMA_OPTIONS_TYPE_NODE:
184 parse_numa_node(ms, &object->u.node, &err);
185 if (err) {
186 goto end;
187 }
188 break;
189 case NUMA_OPTIONS_TYPE_DIST:
190 parse_numa_distance(&object->u.dist, &err);
191 if (err) {
192 goto end;
193 }
194 break;
195 case NUMA_OPTIONS_TYPE_CPU:
196 if (!object->u.cpu.has_node_id) {
197 error_setg(&err, "Missing mandatory node-id property");
198 goto end;
199 }
200 if (!numa_info[object->u.cpu.node_id].present) {
201 error_setg(&err, "Invalid node-id=%" PRId64 ", NUMA node must be "
202 "defined with -numa node,nodeid=ID before it's used with "
203 "-numa cpu,node-id=ID", object->u.cpu.node_id);
204 goto end;
205 }
206
207 machine_set_cpu_numa_node(ms, qapi_NumaCpuOptions_base(&object->u.cpu),
208 &err);
209 break;
210 default:
211 abort();
212 }
213
214 end:
215 error_propagate(errp, err);
216 }
217
218 static int parse_numa(void *opaque, QemuOpts *opts, Error **errp)
219 {
220 NumaOptions *object = NULL;
221 MachineState *ms = MACHINE(opaque);
222 Error *err = NULL;
223 Visitor *v = opts_visitor_new(opts);
224
225 visit_type_NumaOptions(v, NULL, &object, &err);
226 visit_free(v);
227 if (err) {
228 goto end;
229 }
230
231 /* Fix up legacy suffix-less format */
232 if ((object->type == NUMA_OPTIONS_TYPE_NODE) && object->u.node.has_mem) {
233 const char *mem_str = qemu_opt_get(opts, "mem");
234 qemu_strtosz_MiB(mem_str, NULL, &object->u.node.mem);
235 }
236
237 set_numa_options(ms, object, &err);
238
239 end:
240 qapi_free_NumaOptions(object);
241 if (err) {
242 error_propagate(errp, err);
243 return -1;
244 }
245
246 return 0;
247 }
248
249 /* If all node pair distances are symmetric, then only distances
250 * in one direction are enough. If there is even one asymmetric
251 * pair, though, then all distances must be provided. The
252 * distance from a node to itself is always NUMA_DISTANCE_MIN,
253 * so providing it is never necessary.
254 */
255 static void validate_numa_distance(void)
256 {
257 int src, dst;
258 bool is_asymmetrical = false;
259
260 for (src = 0; src < nb_numa_nodes; src++) {
261 for (dst = src; dst < nb_numa_nodes; dst++) {
262 if (numa_info[src].distance[dst] == 0 &&
263 numa_info[dst].distance[src] == 0) {
264 if (src != dst) {
265 error_report("The distance between node %d and %d is "
266 "missing, at least one distance value "
267 "between each nodes should be provided.",
268 src, dst);
269 exit(EXIT_FAILURE);
270 }
271 }
272
273 if (numa_info[src].distance[dst] != 0 &&
274 numa_info[dst].distance[src] != 0 &&
275 numa_info[src].distance[dst] !=
276 numa_info[dst].distance[src]) {
277 is_asymmetrical = true;
278 }
279 }
280 }
281
282 if (is_asymmetrical) {
283 for (src = 0; src < nb_numa_nodes; src++) {
284 for (dst = 0; dst < nb_numa_nodes; dst++) {
285 if (src != dst && numa_info[src].distance[dst] == 0) {
286 error_report("At least one asymmetrical pair of "
287 "distances is given, please provide distances "
288 "for both directions of all node pairs.");
289 exit(EXIT_FAILURE);
290 }
291 }
292 }
293 }
294 }
295
296 static void complete_init_numa_distance(void)
297 {
298 int src, dst;
299
300 /* Fixup NUMA distance by symmetric policy because if it is an
301 * asymmetric distance table, it should be a complete table and
302 * there would not be any missing distance except local node, which
303 * is verified by validate_numa_distance above.
304 */
305 for (src = 0; src < nb_numa_nodes; src++) {
306 for (dst = 0; dst < nb_numa_nodes; dst++) {
307 if (numa_info[src].distance[dst] == 0) {
308 if (src == dst) {
309 numa_info[src].distance[dst] = NUMA_DISTANCE_MIN;
310 } else {
311 numa_info[src].distance[dst] = numa_info[dst].distance[src];
312 }
313 }
314 }
315 }
316 }
317
318 void numa_legacy_auto_assign_ram(MachineClass *mc, NodeInfo *nodes,
319 int nb_nodes, ram_addr_t size)
320 {
321 int i;
322 uint64_t usedmem = 0;
323
324 /* Align each node according to the alignment
325 * requirements of the machine class
326 */
327
328 for (i = 0; i < nb_nodes - 1; i++) {
329 nodes[i].node_mem = (size / nb_nodes) &
330 ~((1 << mc->numa_mem_align_shift) - 1);
331 usedmem += nodes[i].node_mem;
332 }
333 nodes[i].node_mem = size - usedmem;
334 }
335
336 void numa_default_auto_assign_ram(MachineClass *mc, NodeInfo *nodes,
337 int nb_nodes, ram_addr_t size)
338 {
339 int i;
340 uint64_t usedmem = 0, node_mem;
341 uint64_t granularity = size / nb_nodes;
342 uint64_t propagate = 0;
343
344 for (i = 0; i < nb_nodes - 1; i++) {
345 node_mem = (granularity + propagate) &
346 ~((1 << mc->numa_mem_align_shift) - 1);
347 propagate = granularity + propagate - node_mem;
348 nodes[i].node_mem = node_mem;
349 usedmem += node_mem;
350 }
351 nodes[i].node_mem = size - usedmem;
352 }
353
354 void numa_complete_configuration(MachineState *ms)
355 {
356 int i;
357 MachineClass *mc = MACHINE_GET_CLASS(ms);
358
359 /*
360 * If memory hotplug is enabled (slots > 0) but without '-numa'
361 * options explicitly on CLI, guestes will break.
362 *
363 * Windows: won't enable memory hotplug without SRAT table at all
364 *
365 * Linux: if QEMU is started with initial memory all below 4Gb
366 * and no SRAT table present, guest kernel will use nommu DMA ops,
367 * which breaks 32bit hw drivers when memory is hotplugged and
368 * guest tries to use it with that drivers.
369 *
370 * Enable NUMA implicitly by adding a new NUMA node automatically.
371 */
372 if (ms->ram_slots > 0 && nb_numa_nodes == 0 &&
373 mc->auto_enable_numa_with_memhp) {
374 NumaNodeOptions node = { };
375 parse_numa_node(ms, &node, &error_abort);
376 }
377
378 assert(max_numa_nodeid <= MAX_NODES);
379
380 /* No support for sparse NUMA node IDs yet: */
381 for (i = max_numa_nodeid - 1; i >= 0; i--) {
382 /* Report large node IDs first, to make mistakes easier to spot */
383 if (!numa_info[i].present) {
384 error_report("numa: Node ID missing: %d", i);
385 exit(1);
386 }
387 }
388
389 /* This must be always true if all nodes are present: */
390 assert(nb_numa_nodes == max_numa_nodeid);
391
392 if (nb_numa_nodes > 0) {
393 uint64_t numa_total;
394
395 if (nb_numa_nodes > MAX_NODES) {
396 nb_numa_nodes = MAX_NODES;
397 }
398
399 /* If no memory size is given for any node, assume the default case
400 * and distribute the available memory equally across all nodes
401 */
402 for (i = 0; i < nb_numa_nodes; i++) {
403 if (numa_info[i].node_mem != 0) {
404 break;
405 }
406 }
407 if (i == nb_numa_nodes) {
408 assert(mc->numa_auto_assign_ram);
409 mc->numa_auto_assign_ram(mc, numa_info, nb_numa_nodes, ram_size);
410 }
411
412 numa_total = 0;
413 for (i = 0; i < nb_numa_nodes; i++) {
414 numa_total += numa_info[i].node_mem;
415 }
416 if (numa_total != ram_size) {
417 error_report("total memory for NUMA nodes (0x%" PRIx64 ")"
418 " should equal RAM size (0x" RAM_ADDR_FMT ")",
419 numa_total, ram_size);
420 exit(1);
421 }
422
423 /* QEMU needs at least all unique node pair distances to build
424 * the whole NUMA distance table. QEMU treats the distance table
425 * as symmetric by default, i.e. distance A->B == distance B->A.
426 * Thus, QEMU is able to complete the distance table
427 * initialization even though only distance A->B is provided and
428 * distance B->A is not. QEMU knows the distance of a node to
429 * itself is always 10, so A->A distances may be omitted. When
430 * the distances of two nodes of a pair differ, i.e. distance
431 * A->B != distance B->A, then that means the distance table is
432 * asymmetric. In this case, the distances for both directions
433 * of all node pairs are required.
434 */
435 if (have_numa_distance) {
436 /* Validate enough NUMA distance information was provided. */
437 validate_numa_distance();
438
439 /* Validation succeeded, now fill in any missing distances. */
440 complete_init_numa_distance();
441 }
442 }
443 }
444
445 void parse_numa_opts(MachineState *ms)
446 {
447 qemu_opts_foreach(qemu_find_opts("numa"), parse_numa, ms, &error_fatal);
448 }
449
450 void qmp_set_numa_node(NumaOptions *cmd, Error **errp)
451 {
452 if (!runstate_check(RUN_STATE_PRECONFIG)) {
453 error_setg(errp, "The command is permitted only in '%s' state",
454 RunState_str(RUN_STATE_PRECONFIG));
455 return;
456 }
457
458 set_numa_options(MACHINE(qdev_get_machine()), cmd, errp);
459 }
460
461 void numa_cpu_pre_plug(const CPUArchId *slot, DeviceState *dev, Error **errp)
462 {
463 int node_id = object_property_get_int(OBJECT(dev), "node-id", &error_abort);
464
465 if (node_id == CPU_UNSET_NUMA_NODE_ID) {
466 /* due to bug in libvirt, it doesn't pass node-id from props on
467 * device_add as expected, so we have to fix it up here */
468 if (slot->props.has_node_id) {
469 object_property_set_int(OBJECT(dev), slot->props.node_id,
470 "node-id", errp);
471 }
472 } else if (node_id != slot->props.node_id) {
473 error_setg(errp, "node-id=%d must match numa node specified "
474 "with -numa option", node_id);
475 }
476 }
477
478 static void allocate_system_memory_nonnuma(MemoryRegion *mr, Object *owner,
479 const char *name,
480 uint64_t ram_size)
481 {
482 if (mem_path) {
483 #ifdef __linux__
484 Error *err = NULL;
485 memory_region_init_ram_from_file(mr, owner, name, ram_size, 0, 0,
486 mem_path, &err);
487 if (err) {
488 error_report_err(err);
489 if (mem_prealloc) {
490 exit(1);
491 }
492 error_report("falling back to regular RAM allocation.");
493
494 /* Legacy behavior: if allocation failed, fall back to
495 * regular RAM allocation.
496 */
497 mem_path = NULL;
498 memory_region_init_ram_nomigrate(mr, owner, name, ram_size, &error_fatal);
499 }
500 #else
501 fprintf(stderr, "-mem-path not supported on this host\n");
502 exit(1);
503 #endif
504 } else {
505 memory_region_init_ram_nomigrate(mr, owner, name, ram_size, &error_fatal);
506 }
507 vmstate_register_ram_global(mr);
508 }
509
510 void memory_region_allocate_system_memory(MemoryRegion *mr, Object *owner,
511 const char *name,
512 uint64_t ram_size)
513 {
514 uint64_t addr = 0;
515 int i;
516
517 if (nb_numa_nodes == 0 || !have_memdevs) {
518 allocate_system_memory_nonnuma(mr, owner, name, ram_size);
519 return;
520 }
521
522 memory_region_init(mr, owner, name, ram_size);
523 for (i = 0; i < nb_numa_nodes; i++) {
524 uint64_t size = numa_info[i].node_mem;
525 HostMemoryBackend *backend = numa_info[i].node_memdev;
526 if (!backend) {
527 continue;
528 }
529 MemoryRegion *seg = host_memory_backend_get_memory(backend);
530
531 if (memory_region_is_mapped(seg)) {
532 char *path = object_get_canonical_path_component(OBJECT(backend));
533 error_report("memory backend %s is used multiple times. Each "
534 "-numa option must use a different memdev value.",
535 path);
536 g_free(path);
537 exit(1);
538 }
539
540 host_memory_backend_set_mapped(backend, true);
541 memory_region_add_subregion(mr, addr, seg);
542 vmstate_register_ram_global(seg);
543 addr += size;
544 }
545 }
546
547 static void numa_stat_memory_devices(NumaNodeMem node_mem[])
548 {
549 MemoryDeviceInfoList *info_list = qmp_memory_device_list();
550 MemoryDeviceInfoList *info;
551 PCDIMMDeviceInfo *pcdimm_info;
552
553 for (info = info_list; info; info = info->next) {
554 MemoryDeviceInfo *value = info->value;
555
556 if (value) {
557 switch (value->type) {
558 case MEMORY_DEVICE_INFO_KIND_DIMM:
559 pcdimm_info = value->u.dimm.data;
560 break;
561
562 case MEMORY_DEVICE_INFO_KIND_NVDIMM:
563 pcdimm_info = value->u.nvdimm.data;
564 break;
565
566 default:
567 pcdimm_info = NULL;
568 break;
569 }
570
571 if (pcdimm_info) {
572 node_mem[pcdimm_info->node].node_mem += pcdimm_info->size;
573 node_mem[pcdimm_info->node].node_plugged_mem +=
574 pcdimm_info->size;
575 }
576 }
577 }
578 qapi_free_MemoryDeviceInfoList(info_list);
579 }
580
581 void query_numa_node_mem(NumaNodeMem node_mem[])
582 {
583 int i;
584
585 if (nb_numa_nodes <= 0) {
586 return;
587 }
588
589 numa_stat_memory_devices(node_mem);
590 for (i = 0; i < nb_numa_nodes; i++) {
591 node_mem[i].node_mem += numa_info[i].node_mem;
592 }
593 }
594
595 static int query_memdev(Object *obj, void *opaque)
596 {
597 MemdevList **list = opaque;
598 MemdevList *m = NULL;
599
600 if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) {
601 m = g_malloc0(sizeof(*m));
602
603 m->value = g_malloc0(sizeof(*m->value));
604
605 m->value->id = object_get_canonical_path_component(obj);
606 m->value->has_id = !!m->value->id;
607
608 m->value->size = object_property_get_uint(obj, "size",
609 &error_abort);
610 m->value->merge = object_property_get_bool(obj, "merge",
611 &error_abort);
612 m->value->dump = object_property_get_bool(obj, "dump",
613 &error_abort);
614 m->value->prealloc = object_property_get_bool(obj,
615 "prealloc",
616 &error_abort);
617 m->value->policy = object_property_get_enum(obj,
618 "policy",
619 "HostMemPolicy",
620 &error_abort);
621 object_property_get_uint16List(obj, "host-nodes",
622 &m->value->host_nodes,
623 &error_abort);
624
625 m->next = *list;
626 *list = m;
627 }
628
629 return 0;
630 }
631
632 MemdevList *qmp_query_memdev(Error **errp)
633 {
634 Object *obj = object_get_objects_root();
635 MemdevList *list = NULL;
636
637 object_child_foreach(obj, query_memdev, &list);
638 return list;
639 }
640
641 void ram_block_notifier_add(RAMBlockNotifier *n)
642 {
643 QLIST_INSERT_HEAD(&ram_list.ramblock_notifiers, n, next);
644 }
645
646 void ram_block_notifier_remove(RAMBlockNotifier *n)
647 {
648 QLIST_REMOVE(n, next);
649 }
650
651 void ram_block_notify_add(void *host, size_t size)
652 {
653 RAMBlockNotifier *notifier;
654
655 QLIST_FOREACH(notifier, &ram_list.ramblock_notifiers, next) {
656 notifier->ram_block_added(notifier, host, size);
657 }
658 }
659
660 void ram_block_notify_remove(void *host, size_t size)
661 {
662 RAMBlockNotifier *notifier;
663
664 QLIST_FOREACH(notifier, &ram_list.ramblock_notifiers, next) {
665 notifier->ram_block_removed(notifier, host, size);
666 }
667 }