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"
26 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT
!= MPOL_DEFAULT
);
27 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED
!= MPOL_PREFERRED
);
28 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND
!= MPOL_BIND
);
29 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE
!= MPOL_INTERLEAVE
);
33 host_memory_backend_get_name(HostMemoryBackend
*backend
)
35 if (!backend
->use_canonical_path
) {
36 return g_strdup(object_get_canonical_path_component(OBJECT(backend
)));
39 return object_get_canonical_path(OBJECT(backend
));
43 host_memory_backend_get_size(Object
*obj
, Visitor
*v
, const char *name
,
44 void *opaque
, Error
**errp
)
46 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
47 uint64_t value
= backend
->size
;
49 visit_type_size(v
, name
, &value
, errp
);
53 host_memory_backend_set_size(Object
*obj
, Visitor
*v
, const char *name
,
54 void *opaque
, Error
**errp
)
56 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
59 if (host_memory_backend_mr_inited(backend
)) {
60 error_setg(errp
, "cannot change property %s of %s ", name
,
61 object_get_typename(obj
));
65 if (!visit_type_size(v
, name
, &value
, errp
)) {
70 "property '%s' of %s doesn't take value '%" PRIu64
"'",
71 name
, object_get_typename(obj
), value
);
74 backend
->size
= value
;
78 host_memory_backend_get_host_nodes(Object
*obj
, Visitor
*v
, const char *name
,
79 void *opaque
, Error
**errp
)
81 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
82 uint16List
*host_nodes
= NULL
;
83 uint16List
**tail
= &host_nodes
;
86 value
= find_first_bit(backend
->host_nodes
, MAX_NODES
);
87 if (value
== MAX_NODES
) {
91 QAPI_LIST_APPEND(tail
, value
);
94 value
= find_next_bit(backend
->host_nodes
, MAX_NODES
, value
+ 1);
95 if (value
== MAX_NODES
) {
99 QAPI_LIST_APPEND(tail
, value
);
103 visit_type_uint16List(v
, name
, &host_nodes
, errp
);
104 qapi_free_uint16List(host_nodes
);
108 host_memory_backend_set_host_nodes(Object
*obj
, Visitor
*v
, const char *name
,
109 void *opaque
, Error
**errp
)
112 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
113 uint16List
*l
, *host_nodes
= NULL
;
115 visit_type_uint16List(v
, name
, &host_nodes
, errp
);
117 for (l
= host_nodes
; l
; l
= l
->next
) {
118 if (l
->value
>= MAX_NODES
) {
119 error_setg(errp
, "Invalid host-nodes value: %d", l
->value
);
124 for (l
= host_nodes
; l
; l
= l
->next
) {
125 bitmap_set(backend
->host_nodes
, l
->value
, 1);
129 qapi_free_uint16List(host_nodes
);
131 error_setg(errp
, "NUMA node binding are not supported by this QEMU");
136 host_memory_backend_get_policy(Object
*obj
, Error
**errp G_GNUC_UNUSED
)
138 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
139 return backend
->policy
;
143 host_memory_backend_set_policy(Object
*obj
, int policy
, Error
**errp
)
145 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
146 backend
->policy
= policy
;
149 if (policy
!= HOST_MEM_POLICY_DEFAULT
) {
150 error_setg(errp
, "NUMA policies are not supported by this QEMU");
155 static bool host_memory_backend_get_merge(Object
*obj
, Error
**errp
)
157 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
159 return backend
->merge
;
162 static void host_memory_backend_set_merge(Object
*obj
, bool value
, Error
**errp
)
164 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
166 if (!host_memory_backend_mr_inited(backend
)) {
167 backend
->merge
= value
;
171 if (value
!= backend
->merge
) {
172 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
173 uint64_t sz
= memory_region_size(&backend
->mr
);
175 qemu_madvise(ptr
, sz
,
176 value
? QEMU_MADV_MERGEABLE
: QEMU_MADV_UNMERGEABLE
);
177 backend
->merge
= value
;
181 static bool host_memory_backend_get_dump(Object
*obj
, Error
**errp
)
183 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
185 return backend
->dump
;
188 static void host_memory_backend_set_dump(Object
*obj
, bool value
, Error
**errp
)
190 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
192 if (!host_memory_backend_mr_inited(backend
)) {
193 backend
->dump
= value
;
197 if (value
!= backend
->dump
) {
198 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
199 uint64_t sz
= memory_region_size(&backend
->mr
);
201 qemu_madvise(ptr
, sz
,
202 value
? QEMU_MADV_DODUMP
: QEMU_MADV_DONTDUMP
);
203 backend
->dump
= value
;
207 static bool host_memory_backend_get_prealloc(Object
*obj
, Error
**errp
)
209 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
211 return backend
->prealloc
;
214 static void host_memory_backend_set_prealloc(Object
*obj
, bool value
,
217 Error
*local_err
= NULL
;
218 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
220 if (!backend
->reserve
&& value
) {
221 error_setg(errp
, "'prealloc=on' and 'reserve=off' are incompatible");
225 if (!host_memory_backend_mr_inited(backend
)) {
226 backend
->prealloc
= value
;
230 if (value
&& !backend
->prealloc
) {
231 int fd
= memory_region_get_fd(&backend
->mr
);
232 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
233 uint64_t sz
= memory_region_size(&backend
->mr
);
235 qemu_prealloc_mem(fd
, ptr
, sz
, backend
->prealloc_threads
,
236 backend
->prealloc_context
, &local_err
);
238 error_propagate(errp
, local_err
);
241 backend
->prealloc
= true;
245 static void host_memory_backend_get_prealloc_threads(Object
*obj
, Visitor
*v
,
246 const char *name
, void *opaque
, Error
**errp
)
248 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
249 visit_type_uint32(v
, name
, &backend
->prealloc_threads
, errp
);
252 static void host_memory_backend_set_prealloc_threads(Object
*obj
, Visitor
*v
,
253 const char *name
, void *opaque
, Error
**errp
)
255 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
258 if (!visit_type_uint32(v
, name
, &value
, errp
)) {
262 error_setg(errp
, "property '%s' of %s doesn't take value '%d'", name
,
263 object_get_typename(obj
), value
);
266 backend
->prealloc_threads
= value
;
269 static void host_memory_backend_init(Object
*obj
)
271 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
272 MachineState
*machine
= MACHINE(qdev_get_machine());
274 /* TODO: convert access to globals to compat properties */
275 backend
->merge
= machine_mem_merge(machine
);
276 backend
->dump
= machine_dump_guest_core(machine
);
277 backend
->reserve
= true;
278 backend
->prealloc_threads
= machine
->smp
.cpus
;
281 static void host_memory_backend_post_init(Object
*obj
)
283 object_apply_compat_props(obj
);
286 bool host_memory_backend_mr_inited(HostMemoryBackend
*backend
)
289 * NOTE: We forbid zero-length memory backend, so here zero means
290 * "we haven't inited the backend memory region yet".
292 return memory_region_size(&backend
->mr
) != 0;
295 MemoryRegion
*host_memory_backend_get_memory(HostMemoryBackend
*backend
)
297 return host_memory_backend_mr_inited(backend
) ? &backend
->mr
: NULL
;
300 void host_memory_backend_set_mapped(HostMemoryBackend
*backend
, bool mapped
)
302 backend
->is_mapped
= mapped
;
305 bool host_memory_backend_is_mapped(HostMemoryBackend
*backend
)
307 return backend
->is_mapped
;
310 size_t host_memory_backend_pagesize(HostMemoryBackend
*memdev
)
312 size_t pagesize
= qemu_ram_pagesize(memdev
->mr
.ram_block
);
313 g_assert(pagesize
>= qemu_real_host_page_size());
318 host_memory_backend_memory_complete(UserCreatable
*uc
, Error
**errp
)
320 HostMemoryBackend
*backend
= MEMORY_BACKEND(uc
);
321 HostMemoryBackendClass
*bc
= MEMORY_BACKEND_GET_CLASS(uc
);
322 Error
*local_err
= NULL
;
327 bc
->alloc(backend
, &local_err
);
332 ptr
= memory_region_get_ram_ptr(&backend
->mr
);
333 sz
= memory_region_size(&backend
->mr
);
335 if (backend
->merge
) {
336 qemu_madvise(ptr
, sz
, QEMU_MADV_MERGEABLE
);
338 if (!backend
->dump
) {
339 qemu_madvise(ptr
, sz
, QEMU_MADV_DONTDUMP
);
342 unsigned long lastbit
= find_last_bit(backend
->host_nodes
, MAX_NODES
);
343 /* lastbit == MAX_NODES means maxnode = 0 */
344 unsigned long maxnode
= (lastbit
+ 1) % (MAX_NODES
+ 1);
345 /* ensure policy won't be ignored in case memory is preallocated
346 * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
347 * this doesn't catch hugepage case. */
348 unsigned flags
= MPOL_MF_STRICT
| MPOL_MF_MOVE
;
350 /* check for invalid host-nodes and policies and give more verbose
351 * error messages than mbind(). */
352 if (maxnode
&& backend
->policy
== MPOL_DEFAULT
) {
353 error_setg(errp
, "host-nodes must be empty for policy default,"
354 " or you should explicitly specify a policy other"
357 } else if (maxnode
== 0 && backend
->policy
!= MPOL_DEFAULT
) {
358 error_setg(errp
, "host-nodes must be set for policy %s",
359 HostMemPolicy_str(backend
->policy
));
363 /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
364 * as argument to mbind() due to an old Linux bug (feature?) which
365 * cuts off the last specified node. This means backend->host_nodes
366 * must have MAX_NODES+1 bits available.
368 assert(sizeof(backend
->host_nodes
) >=
369 BITS_TO_LONGS(MAX_NODES
+ 1) * sizeof(unsigned long));
370 assert(maxnode
<= MAX_NODES
);
373 mbind(ptr
, sz
, backend
->policy
, backend
->host_nodes
, maxnode
+ 1,
375 if (backend
->policy
!= MPOL_DEFAULT
|| errno
!= ENOSYS
) {
376 error_setg_errno(errp
, errno
,
377 "cannot bind memory to host NUMA nodes");
382 /* Preallocate memory after the NUMA policy has been instantiated.
383 * This is necessary to guarantee memory is allocated with
384 * specified NUMA policy in place.
386 if (backend
->prealloc
) {
387 qemu_prealloc_mem(memory_region_get_fd(&backend
->mr
), ptr
, sz
,
388 backend
->prealloc_threads
,
389 backend
->prealloc_context
, &local_err
);
396 error_propagate(errp
, local_err
);
400 host_memory_backend_can_be_deleted(UserCreatable
*uc
)
402 if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc
))) {
409 static bool host_memory_backend_get_share(Object
*o
, Error
**errp
)
411 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
413 return backend
->share
;
416 static void host_memory_backend_set_share(Object
*o
, bool value
, Error
**errp
)
418 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
420 if (host_memory_backend_mr_inited(backend
)) {
421 error_setg(errp
, "cannot change property value");
424 backend
->share
= value
;
428 static bool host_memory_backend_get_reserve(Object
*o
, Error
**errp
)
430 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
432 return backend
->reserve
;
435 static void host_memory_backend_set_reserve(Object
*o
, bool value
, Error
**errp
)
437 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
439 if (host_memory_backend_mr_inited(backend
)) {
440 error_setg(errp
, "cannot change property value");
443 if (backend
->prealloc
&& !value
) {
444 error_setg(errp
, "'prealloc=on' and 'reserve=off' are incompatible");
447 backend
->reserve
= value
;
449 #endif /* CONFIG_LINUX */
452 host_memory_backend_get_use_canonical_path(Object
*obj
, Error
**errp
)
454 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
456 return backend
->use_canonical_path
;
460 host_memory_backend_set_use_canonical_path(Object
*obj
, bool value
,
463 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
465 backend
->use_canonical_path
= value
;
469 host_memory_backend_class_init(ObjectClass
*oc
, void *data
)
471 UserCreatableClass
*ucc
= USER_CREATABLE_CLASS(oc
);
473 ucc
->complete
= host_memory_backend_memory_complete
;
474 ucc
->can_be_deleted
= host_memory_backend_can_be_deleted
;
476 object_class_property_add_bool(oc
, "merge",
477 host_memory_backend_get_merge
,
478 host_memory_backend_set_merge
);
479 object_class_property_set_description(oc
, "merge",
480 "Mark memory as mergeable");
481 object_class_property_add_bool(oc
, "dump",
482 host_memory_backend_get_dump
,
483 host_memory_backend_set_dump
);
484 object_class_property_set_description(oc
, "dump",
485 "Set to 'off' to exclude from core dump");
486 object_class_property_add_bool(oc
, "prealloc",
487 host_memory_backend_get_prealloc
,
488 host_memory_backend_set_prealloc
);
489 object_class_property_set_description(oc
, "prealloc",
490 "Preallocate memory");
491 object_class_property_add(oc
, "prealloc-threads", "int",
492 host_memory_backend_get_prealloc_threads
,
493 host_memory_backend_set_prealloc_threads
,
495 object_class_property_set_description(oc
, "prealloc-threads",
496 "Number of CPU threads to use for prealloc");
497 object_class_property_add_link(oc
, "prealloc-context",
498 TYPE_THREAD_CONTEXT
, offsetof(HostMemoryBackend
, prealloc_context
),
499 object_property_allow_set_link
, OBJ_PROP_LINK_STRONG
);
500 object_class_property_set_description(oc
, "prealloc-context",
501 "Context to use for creating CPU threads for preallocation");
502 object_class_property_add(oc
, "size", "int",
503 host_memory_backend_get_size
,
504 host_memory_backend_set_size
,
506 object_class_property_set_description(oc
, "size",
507 "Size of the memory region (ex: 500M)");
508 object_class_property_add(oc
, "host-nodes", "int",
509 host_memory_backend_get_host_nodes
,
510 host_memory_backend_set_host_nodes
,
512 object_class_property_set_description(oc
, "host-nodes",
513 "Binds memory to the list of NUMA host nodes");
514 object_class_property_add_enum(oc
, "policy", "HostMemPolicy",
515 &HostMemPolicy_lookup
,
516 host_memory_backend_get_policy
,
517 host_memory_backend_set_policy
);
518 object_class_property_set_description(oc
, "policy",
519 "Set the NUMA policy");
520 object_class_property_add_bool(oc
, "share",
521 host_memory_backend_get_share
, host_memory_backend_set_share
);
522 object_class_property_set_description(oc
, "share",
523 "Mark the memory as private to QEMU or shared");
525 object_class_property_add_bool(oc
, "reserve",
526 host_memory_backend_get_reserve
, host_memory_backend_set_reserve
);
527 object_class_property_set_description(oc
, "reserve",
528 "Reserve swap space (or huge pages) if applicable");
529 #endif /* CONFIG_LINUX */
531 * Do not delete/rename option. This option must be considered stable
532 * (as if it didn't have the 'x-' prefix including deprecation period) as
533 * long as 4.0 and older machine types exists.
534 * Option will be used by upper layers to override (disable) canonical path
535 * for ramblock-id set by compat properties on old machine types ( <= 4.0),
536 * to keep migration working when backend is used for main RAM with
537 * -machine memory-backend= option (main RAM historically used prefix-less
540 object_class_property_add_bool(oc
, "x-use-canonical-path-for-ramblock-id",
541 host_memory_backend_get_use_canonical_path
,
542 host_memory_backend_set_use_canonical_path
);
545 static const TypeInfo host_memory_backend_info
= {
546 .name
= TYPE_MEMORY_BACKEND
,
547 .parent
= TYPE_OBJECT
,
549 .class_size
= sizeof(HostMemoryBackendClass
),
550 .class_init
= host_memory_backend_class_init
,
551 .instance_size
= sizeof(HostMemoryBackend
),
552 .instance_init
= host_memory_backend_init
,
553 .instance_post_init
= host_memory_backend_post_init
,
554 .interfaces
= (InterfaceInfo
[]) {
555 { TYPE_USER_CREATABLE
},
560 static void register_types(void)
562 type_register_static(&host_memory_backend_info
);
565 type_init(register_types
);