2 * QEMU Host Memory Backend
4 * Copyright (C) 2013-2014 Red Hat Inc
7 * Igor Mammedov <imammedo@redhat.com>
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
13 #include "qemu/osdep.h"
14 #include "sysemu/hostmem.h"
15 #include "hw/boards.h"
16 #include "qapi/error.h"
17 #include "qapi/qapi-builtin-visit.h"
18 #include "qapi/visitor.h"
19 #include "qemu/config-file.h"
20 #include "qom/object_interfaces.h"
21 #include "qemu/mmap-alloc.h"
22 #include "qemu/madvise.h"
23 #include "hw/qdev-core.h"
28 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT
!= MPOL_DEFAULT
);
30 * HOST_MEM_POLICY_PREFERRED may either translate to MPOL_PREFERRED or
31 * MPOL_PREFERRED_MANY, see comments further below.
33 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED
!= MPOL_PREFERRED
);
34 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND
!= MPOL_BIND
);
35 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE
!= MPOL_INTERLEAVE
);
39 host_memory_backend_get_name(HostMemoryBackend
*backend
)
41 if (!backend
->use_canonical_path
) {
42 return g_strdup(object_get_canonical_path_component(OBJECT(backend
)));
45 return object_get_canonical_path(OBJECT(backend
));
49 host_memory_backend_get_size(Object
*obj
, Visitor
*v
, const char *name
,
50 void *opaque
, Error
**errp
)
52 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
53 uint64_t value
= backend
->size
;
55 visit_type_size(v
, name
, &value
, errp
);
59 host_memory_backend_set_size(Object
*obj
, Visitor
*v
, const char *name
,
60 void *opaque
, Error
**errp
)
62 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
65 if (host_memory_backend_mr_inited(backend
)) {
66 error_setg(errp
, "cannot change property %s of %s ", name
,
67 object_get_typename(obj
));
71 if (!visit_type_size(v
, name
, &value
, errp
)) {
76 "property '%s' of %s doesn't take value '%" PRIu64
"'",
77 name
, object_get_typename(obj
), value
);
80 backend
->size
= value
;
84 host_memory_backend_get_host_nodes(Object
*obj
, Visitor
*v
, const char *name
,
85 void *opaque
, Error
**errp
)
87 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
88 uint16List
*host_nodes
= NULL
;
89 uint16List
**tail
= &host_nodes
;
92 value
= find_first_bit(backend
->host_nodes
, MAX_NODES
);
93 if (value
== MAX_NODES
) {
97 QAPI_LIST_APPEND(tail
, value
);
100 value
= find_next_bit(backend
->host_nodes
, MAX_NODES
, value
+ 1);
101 if (value
== MAX_NODES
) {
105 QAPI_LIST_APPEND(tail
, value
);
109 visit_type_uint16List(v
, name
, &host_nodes
, errp
);
110 qapi_free_uint16List(host_nodes
);
114 host_memory_backend_set_host_nodes(Object
*obj
, Visitor
*v
, const char *name
,
115 void *opaque
, Error
**errp
)
118 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
119 uint16List
*l
, *host_nodes
= NULL
;
121 visit_type_uint16List(v
, name
, &host_nodes
, errp
);
123 for (l
= host_nodes
; l
; l
= l
->next
) {
124 if (l
->value
>= MAX_NODES
) {
125 error_setg(errp
, "Invalid host-nodes value: %d", l
->value
);
130 for (l
= host_nodes
; l
; l
= l
->next
) {
131 bitmap_set(backend
->host_nodes
, l
->value
, 1);
135 qapi_free_uint16List(host_nodes
);
137 error_setg(errp
, "NUMA node binding are not supported by this QEMU");
142 host_memory_backend_get_policy(Object
*obj
, Error
**errp G_GNUC_UNUSED
)
144 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
145 return backend
->policy
;
149 host_memory_backend_set_policy(Object
*obj
, int policy
, Error
**errp
)
151 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
152 backend
->policy
= policy
;
155 if (policy
!= HOST_MEM_POLICY_DEFAULT
) {
156 error_setg(errp
, "NUMA policies are not supported by this QEMU");
161 static bool host_memory_backend_get_merge(Object
*obj
, Error
**errp
)
163 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
165 return backend
->merge
;
168 static void host_memory_backend_set_merge(Object
*obj
, bool value
, Error
**errp
)
170 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
172 if (!host_memory_backend_mr_inited(backend
)) {
173 backend
->merge
= value
;
177 if (value
!= backend
->merge
) {
178 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
179 uint64_t sz
= memory_region_size(&backend
->mr
);
181 qemu_madvise(ptr
, sz
,
182 value
? QEMU_MADV_MERGEABLE
: QEMU_MADV_UNMERGEABLE
);
183 backend
->merge
= value
;
187 static bool host_memory_backend_get_dump(Object
*obj
, Error
**errp
)
189 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
191 return backend
->dump
;
194 static void host_memory_backend_set_dump(Object
*obj
, bool value
, Error
**errp
)
196 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
198 if (!host_memory_backend_mr_inited(backend
)) {
199 backend
->dump
= value
;
203 if (value
!= backend
->dump
) {
204 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
205 uint64_t sz
= memory_region_size(&backend
->mr
);
207 qemu_madvise(ptr
, sz
,
208 value
? QEMU_MADV_DODUMP
: QEMU_MADV_DONTDUMP
);
209 backend
->dump
= value
;
213 static bool host_memory_backend_get_prealloc(Object
*obj
, Error
**errp
)
215 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
217 return backend
->prealloc
;
220 static void host_memory_backend_set_prealloc(Object
*obj
, bool value
,
223 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
225 if (!backend
->reserve
&& value
) {
226 error_setg(errp
, "'prealloc=on' and 'reserve=off' are incompatible");
230 if (!host_memory_backend_mr_inited(backend
)) {
231 backend
->prealloc
= value
;
235 if (value
&& !backend
->prealloc
) {
236 int fd
= memory_region_get_fd(&backend
->mr
);
237 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
238 uint64_t sz
= memory_region_size(&backend
->mr
);
240 if (!qemu_prealloc_mem(fd
, ptr
, sz
, backend
->prealloc_threads
,
241 backend
->prealloc_context
, false, errp
)) {
244 backend
->prealloc
= true;
248 static void host_memory_backend_get_prealloc_threads(Object
*obj
, Visitor
*v
,
249 const char *name
, void *opaque
, Error
**errp
)
251 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
252 visit_type_uint32(v
, name
, &backend
->prealloc_threads
, errp
);
255 static void host_memory_backend_set_prealloc_threads(Object
*obj
, Visitor
*v
,
256 const char *name
, void *opaque
, Error
**errp
)
258 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
261 if (!visit_type_uint32(v
, name
, &value
, errp
)) {
265 error_setg(errp
, "property '%s' of %s doesn't take value '%d'", name
,
266 object_get_typename(obj
), value
);
269 backend
->prealloc_threads
= value
;
272 static void host_memory_backend_init(Object
*obj
)
274 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
275 MachineState
*machine
= MACHINE(qdev_get_machine());
277 /* TODO: convert access to globals to compat properties */
278 backend
->merge
= machine_mem_merge(machine
);
279 backend
->dump
= machine_dump_guest_core(machine
);
280 backend
->reserve
= true;
281 backend
->prealloc_threads
= machine
->smp
.cpus
;
284 static void host_memory_backend_post_init(Object
*obj
)
286 object_apply_compat_props(obj
);
289 bool host_memory_backend_mr_inited(HostMemoryBackend
*backend
)
292 * NOTE: We forbid zero-length memory backend, so here zero means
293 * "we haven't inited the backend memory region yet".
295 return memory_region_size(&backend
->mr
) != 0;
298 MemoryRegion
*host_memory_backend_get_memory(HostMemoryBackend
*backend
)
300 return host_memory_backend_mr_inited(backend
) ? &backend
->mr
: NULL
;
303 void host_memory_backend_set_mapped(HostMemoryBackend
*backend
, bool mapped
)
305 backend
->is_mapped
= mapped
;
308 bool host_memory_backend_is_mapped(HostMemoryBackend
*backend
)
310 return backend
->is_mapped
;
313 size_t host_memory_backend_pagesize(HostMemoryBackend
*memdev
)
315 size_t pagesize
= qemu_ram_pagesize(memdev
->mr
.ram_block
);
316 g_assert(pagesize
>= qemu_real_host_page_size());
321 host_memory_backend_memory_complete(UserCreatable
*uc
, Error
**errp
)
323 HostMemoryBackend
*backend
= MEMORY_BACKEND(uc
);
324 HostMemoryBackendClass
*bc
= MEMORY_BACKEND_GET_CLASS(uc
);
327 bool async
= !phase_check(PHASE_LATE_BACKENDS_CREATED
);
332 if (!bc
->alloc(backend
, errp
)) {
336 ptr
= memory_region_get_ram_ptr(&backend
->mr
);
337 sz
= memory_region_size(&backend
->mr
);
339 if (backend
->merge
) {
340 qemu_madvise(ptr
, sz
, QEMU_MADV_MERGEABLE
);
342 if (!backend
->dump
) {
343 qemu_madvise(ptr
, sz
, QEMU_MADV_DONTDUMP
);
346 unsigned long lastbit
= find_last_bit(backend
->host_nodes
, MAX_NODES
);
347 /* lastbit == MAX_NODES means maxnode = 0 */
348 unsigned long maxnode
= (lastbit
+ 1) % (MAX_NODES
+ 1);
350 * Ensure policy won't be ignored in case memory is preallocated
351 * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
352 * this doesn't catch hugepage case.
354 unsigned flags
= MPOL_MF_STRICT
| MPOL_MF_MOVE
;
355 int mode
= backend
->policy
;
357 /* check for invalid host-nodes and policies and give more verbose
358 * error messages than mbind(). */
359 if (maxnode
&& backend
->policy
== MPOL_DEFAULT
) {
360 error_setg(errp
, "host-nodes must be empty for policy default,"
361 " or you should explicitly specify a policy other"
364 } else if (maxnode
== 0 && backend
->policy
!= MPOL_DEFAULT
) {
365 error_setg(errp
, "host-nodes must be set for policy %s",
366 HostMemPolicy_str(backend
->policy
));
371 * We can have up to MAX_NODES nodes, but we need to pass maxnode+1
372 * as argument to mbind() due to an old Linux bug (feature?) which
373 * cuts off the last specified node. This means backend->host_nodes
374 * must have MAX_NODES+1 bits available.
376 assert(sizeof(backend
->host_nodes
) >=
377 BITS_TO_LONGS(MAX_NODES
+ 1) * sizeof(unsigned long));
378 assert(maxnode
<= MAX_NODES
);
380 #ifdef HAVE_NUMA_HAS_PREFERRED_MANY
381 if (mode
== MPOL_PREFERRED
&& numa_has_preferred_many() > 0) {
383 * Replace with MPOL_PREFERRED_MANY otherwise the mbind() below
384 * silently picks the first node.
386 mode
= MPOL_PREFERRED_MANY
;
391 mbind(ptr
, sz
, mode
, backend
->host_nodes
, maxnode
+ 1, flags
)) {
392 if (backend
->policy
!= MPOL_DEFAULT
|| errno
!= ENOSYS
) {
393 error_setg_errno(errp
, errno
,
394 "cannot bind memory to host NUMA nodes");
400 * Preallocate memory after the NUMA policy has been instantiated.
401 * This is necessary to guarantee memory is allocated with
402 * specified NUMA policy in place.
404 if (backend
->prealloc
&& !qemu_prealloc_mem(memory_region_get_fd(&backend
->mr
),
406 backend
->prealloc_threads
,
407 backend
->prealloc_context
,
414 host_memory_backend_can_be_deleted(UserCreatable
*uc
)
416 if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc
))) {
423 static bool host_memory_backend_get_share(Object
*o
, Error
**errp
)
425 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
427 return backend
->share
;
430 static void host_memory_backend_set_share(Object
*o
, bool value
, Error
**errp
)
432 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
434 if (host_memory_backend_mr_inited(backend
)) {
435 error_setg(errp
, "cannot change property value");
438 backend
->share
= value
;
442 static bool host_memory_backend_get_reserve(Object
*o
, Error
**errp
)
444 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
446 return backend
->reserve
;
449 static void host_memory_backend_set_reserve(Object
*o
, bool value
, Error
**errp
)
451 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
453 if (host_memory_backend_mr_inited(backend
)) {
454 error_setg(errp
, "cannot change property value");
457 if (backend
->prealloc
&& !value
) {
458 error_setg(errp
, "'prealloc=on' and 'reserve=off' are incompatible");
461 backend
->reserve
= value
;
463 #endif /* CONFIG_LINUX */
466 host_memory_backend_get_use_canonical_path(Object
*obj
, Error
**errp
)
468 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
470 return backend
->use_canonical_path
;
474 host_memory_backend_set_use_canonical_path(Object
*obj
, bool value
,
477 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
479 backend
->use_canonical_path
= value
;
483 host_memory_backend_class_init(ObjectClass
*oc
, void *data
)
485 UserCreatableClass
*ucc
= USER_CREATABLE_CLASS(oc
);
487 ucc
->complete
= host_memory_backend_memory_complete
;
488 ucc
->can_be_deleted
= host_memory_backend_can_be_deleted
;
490 object_class_property_add_bool(oc
, "merge",
491 host_memory_backend_get_merge
,
492 host_memory_backend_set_merge
);
493 object_class_property_set_description(oc
, "merge",
494 "Mark memory as mergeable");
495 object_class_property_add_bool(oc
, "dump",
496 host_memory_backend_get_dump
,
497 host_memory_backend_set_dump
);
498 object_class_property_set_description(oc
, "dump",
499 "Set to 'off' to exclude from core dump");
500 object_class_property_add_bool(oc
, "prealloc",
501 host_memory_backend_get_prealloc
,
502 host_memory_backend_set_prealloc
);
503 object_class_property_set_description(oc
, "prealloc",
504 "Preallocate memory");
505 object_class_property_add(oc
, "prealloc-threads", "int",
506 host_memory_backend_get_prealloc_threads
,
507 host_memory_backend_set_prealloc_threads
,
509 object_class_property_set_description(oc
, "prealloc-threads",
510 "Number of CPU threads to use for prealloc");
511 object_class_property_add_link(oc
, "prealloc-context",
512 TYPE_THREAD_CONTEXT
, offsetof(HostMemoryBackend
, prealloc_context
),
513 object_property_allow_set_link
, OBJ_PROP_LINK_STRONG
);
514 object_class_property_set_description(oc
, "prealloc-context",
515 "Context to use for creating CPU threads for preallocation");
516 object_class_property_add(oc
, "size", "int",
517 host_memory_backend_get_size
,
518 host_memory_backend_set_size
,
520 object_class_property_set_description(oc
, "size",
521 "Size of the memory region (ex: 500M)");
522 object_class_property_add(oc
, "host-nodes", "int",
523 host_memory_backend_get_host_nodes
,
524 host_memory_backend_set_host_nodes
,
526 object_class_property_set_description(oc
, "host-nodes",
527 "Binds memory to the list of NUMA host nodes");
528 object_class_property_add_enum(oc
, "policy", "HostMemPolicy",
529 &HostMemPolicy_lookup
,
530 host_memory_backend_get_policy
,
531 host_memory_backend_set_policy
);
532 object_class_property_set_description(oc
, "policy",
533 "Set the NUMA policy");
534 object_class_property_add_bool(oc
, "share",
535 host_memory_backend_get_share
, host_memory_backend_set_share
);
536 object_class_property_set_description(oc
, "share",
537 "Mark the memory as private to QEMU or shared");
539 object_class_property_add_bool(oc
, "reserve",
540 host_memory_backend_get_reserve
, host_memory_backend_set_reserve
);
541 object_class_property_set_description(oc
, "reserve",
542 "Reserve swap space (or huge pages) if applicable");
543 #endif /* CONFIG_LINUX */
545 * Do not delete/rename option. This option must be considered stable
546 * (as if it didn't have the 'x-' prefix including deprecation period) as
547 * long as 4.0 and older machine types exists.
548 * Option will be used by upper layers to override (disable) canonical path
549 * for ramblock-id set by compat properties on old machine types ( <= 4.0),
550 * to keep migration working when backend is used for main RAM with
551 * -machine memory-backend= option (main RAM historically used prefix-less
554 object_class_property_add_bool(oc
, "x-use-canonical-path-for-ramblock-id",
555 host_memory_backend_get_use_canonical_path
,
556 host_memory_backend_set_use_canonical_path
);
559 static const TypeInfo host_memory_backend_info
= {
560 .name
= TYPE_MEMORY_BACKEND
,
561 .parent
= TYPE_OBJECT
,
563 .class_size
= sizeof(HostMemoryBackendClass
),
564 .class_init
= host_memory_backend_class_init
,
565 .instance_size
= sizeof(HostMemoryBackend
),
566 .instance_init
= host_memory_backend_init
,
567 .instance_post_init
= host_memory_backend_post_init
,
568 .interfaces
= (InterfaceInfo
[]) {
569 { TYPE_USER_CREATABLE
},
574 static void register_types(void)
576 type_register_static(&host_memory_backend_info
);
579 type_init(register_types
);