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"
27 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT
!= MPOL_DEFAULT
);
29 * HOST_MEM_POLICY_PREFERRED may either translate to MPOL_PREFERRED or
30 * MPOL_PREFERRED_MANY, see comments further below.
32 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED
!= MPOL_PREFERRED
);
33 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND
!= MPOL_BIND
);
34 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE
!= MPOL_INTERLEAVE
);
38 host_memory_backend_get_name(HostMemoryBackend
*backend
)
40 if (!backend
->use_canonical_path
) {
41 return g_strdup(object_get_canonical_path_component(OBJECT(backend
)));
44 return object_get_canonical_path(OBJECT(backend
));
48 host_memory_backend_get_size(Object
*obj
, Visitor
*v
, const char *name
,
49 void *opaque
, Error
**errp
)
51 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
52 uint64_t value
= backend
->size
;
54 visit_type_size(v
, name
, &value
, errp
);
58 host_memory_backend_set_size(Object
*obj
, Visitor
*v
, const char *name
,
59 void *opaque
, Error
**errp
)
61 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
64 if (host_memory_backend_mr_inited(backend
)) {
65 error_setg(errp
, "cannot change property %s of %s ", name
,
66 object_get_typename(obj
));
70 if (!visit_type_size(v
, name
, &value
, errp
)) {
75 "property '%s' of %s doesn't take value '%" PRIu64
"'",
76 name
, object_get_typename(obj
), value
);
79 backend
->size
= value
;
83 host_memory_backend_get_host_nodes(Object
*obj
, Visitor
*v
, const char *name
,
84 void *opaque
, Error
**errp
)
86 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
87 uint16List
*host_nodes
= NULL
;
88 uint16List
**tail
= &host_nodes
;
91 value
= find_first_bit(backend
->host_nodes
, MAX_NODES
);
92 if (value
== MAX_NODES
) {
96 QAPI_LIST_APPEND(tail
, value
);
99 value
= find_next_bit(backend
->host_nodes
, MAX_NODES
, value
+ 1);
100 if (value
== MAX_NODES
) {
104 QAPI_LIST_APPEND(tail
, value
);
108 visit_type_uint16List(v
, name
, &host_nodes
, errp
);
109 qapi_free_uint16List(host_nodes
);
113 host_memory_backend_set_host_nodes(Object
*obj
, Visitor
*v
, const char *name
,
114 void *opaque
, Error
**errp
)
117 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
118 uint16List
*l
, *host_nodes
= NULL
;
120 visit_type_uint16List(v
, name
, &host_nodes
, errp
);
122 for (l
= host_nodes
; l
; l
= l
->next
) {
123 if (l
->value
>= MAX_NODES
) {
124 error_setg(errp
, "Invalid host-nodes value: %d", l
->value
);
129 for (l
= host_nodes
; l
; l
= l
->next
) {
130 bitmap_set(backend
->host_nodes
, l
->value
, 1);
134 qapi_free_uint16List(host_nodes
);
136 error_setg(errp
, "NUMA node binding are not supported by this QEMU");
141 host_memory_backend_get_policy(Object
*obj
, Error
**errp G_GNUC_UNUSED
)
143 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
144 return backend
->policy
;
148 host_memory_backend_set_policy(Object
*obj
, int policy
, Error
**errp
)
150 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
151 backend
->policy
= policy
;
154 if (policy
!= HOST_MEM_POLICY_DEFAULT
) {
155 error_setg(errp
, "NUMA policies are not supported by this QEMU");
160 static bool host_memory_backend_get_merge(Object
*obj
, Error
**errp
)
162 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
164 return backend
->merge
;
167 static void host_memory_backend_set_merge(Object
*obj
, bool value
, Error
**errp
)
169 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
171 if (!host_memory_backend_mr_inited(backend
)) {
172 backend
->merge
= value
;
176 if (value
!= backend
->merge
) {
177 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
178 uint64_t sz
= memory_region_size(&backend
->mr
);
180 qemu_madvise(ptr
, sz
,
181 value
? QEMU_MADV_MERGEABLE
: QEMU_MADV_UNMERGEABLE
);
182 backend
->merge
= value
;
186 static bool host_memory_backend_get_dump(Object
*obj
, Error
**errp
)
188 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
190 return backend
->dump
;
193 static void host_memory_backend_set_dump(Object
*obj
, bool value
, Error
**errp
)
195 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
197 if (!host_memory_backend_mr_inited(backend
)) {
198 backend
->dump
= value
;
202 if (value
!= backend
->dump
) {
203 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
204 uint64_t sz
= memory_region_size(&backend
->mr
);
206 qemu_madvise(ptr
, sz
,
207 value
? QEMU_MADV_DODUMP
: QEMU_MADV_DONTDUMP
);
208 backend
->dump
= value
;
212 static bool host_memory_backend_get_prealloc(Object
*obj
, Error
**errp
)
214 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
216 return backend
->prealloc
;
219 static void host_memory_backend_set_prealloc(Object
*obj
, bool value
,
222 Error
*local_err
= NULL
;
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 qemu_prealloc_mem(fd
, ptr
, sz
, backend
->prealloc_threads
,
241 backend
->prealloc_context
, &local_err
);
243 error_propagate(errp
, local_err
);
246 backend
->prealloc
= true;
250 static void host_memory_backend_get_prealloc_threads(Object
*obj
, Visitor
*v
,
251 const char *name
, void *opaque
, Error
**errp
)
253 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
254 visit_type_uint32(v
, name
, &backend
->prealloc_threads
, errp
);
257 static void host_memory_backend_set_prealloc_threads(Object
*obj
, Visitor
*v
,
258 const char *name
, void *opaque
, Error
**errp
)
260 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
263 if (!visit_type_uint32(v
, name
, &value
, errp
)) {
267 error_setg(errp
, "property '%s' of %s doesn't take value '%d'", name
,
268 object_get_typename(obj
), value
);
271 backend
->prealloc_threads
= value
;
274 static void host_memory_backend_init(Object
*obj
)
276 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
277 MachineState
*machine
= MACHINE(qdev_get_machine());
279 /* TODO: convert access to globals to compat properties */
280 backend
->merge
= machine_mem_merge(machine
);
281 backend
->dump
= machine_dump_guest_core(machine
);
282 backend
->reserve
= true;
283 backend
->prealloc_threads
= machine
->smp
.cpus
;
286 static void host_memory_backend_post_init(Object
*obj
)
288 object_apply_compat_props(obj
);
291 bool host_memory_backend_mr_inited(HostMemoryBackend
*backend
)
294 * NOTE: We forbid zero-length memory backend, so here zero means
295 * "we haven't inited the backend memory region yet".
297 return memory_region_size(&backend
->mr
) != 0;
300 MemoryRegion
*host_memory_backend_get_memory(HostMemoryBackend
*backend
)
302 return host_memory_backend_mr_inited(backend
) ? &backend
->mr
: NULL
;
305 void host_memory_backend_set_mapped(HostMemoryBackend
*backend
, bool mapped
)
307 backend
->is_mapped
= mapped
;
310 bool host_memory_backend_is_mapped(HostMemoryBackend
*backend
)
312 return backend
->is_mapped
;
315 size_t host_memory_backend_pagesize(HostMemoryBackend
*memdev
)
317 size_t pagesize
= qemu_ram_pagesize(memdev
->mr
.ram_block
);
318 g_assert(pagesize
>= qemu_real_host_page_size());
323 host_memory_backend_memory_complete(UserCreatable
*uc
, Error
**errp
)
325 HostMemoryBackend
*backend
= MEMORY_BACKEND(uc
);
326 HostMemoryBackendClass
*bc
= MEMORY_BACKEND_GET_CLASS(uc
);
327 Error
*local_err
= NULL
;
332 bc
->alloc(backend
, &local_err
);
337 ptr
= memory_region_get_ram_ptr(&backend
->mr
);
338 sz
= memory_region_size(&backend
->mr
);
340 if (backend
->merge
) {
341 qemu_madvise(ptr
, sz
, QEMU_MADV_MERGEABLE
);
343 if (!backend
->dump
) {
344 qemu_madvise(ptr
, sz
, QEMU_MADV_DONTDUMP
);
347 unsigned long lastbit
= find_last_bit(backend
->host_nodes
, MAX_NODES
);
348 /* lastbit == MAX_NODES means maxnode = 0 */
349 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. */
353 unsigned flags
= MPOL_MF_STRICT
| MPOL_MF_MOVE
;
354 int mode
= backend
->policy
;
356 /* check for invalid host-nodes and policies and give more verbose
357 * error messages than mbind(). */
358 if (maxnode
&& backend
->policy
== MPOL_DEFAULT
) {
359 error_setg(errp
, "host-nodes must be empty for policy default,"
360 " or you should explicitly specify a policy other"
363 } else if (maxnode
== 0 && backend
->policy
!= MPOL_DEFAULT
) {
364 error_setg(errp
, "host-nodes must be set for policy %s",
365 HostMemPolicy_str(backend
->policy
));
369 /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
370 * as argument to mbind() due to an old Linux bug (feature?) which
371 * cuts off the last specified node. This means backend->host_nodes
372 * must have MAX_NODES+1 bits available.
374 assert(sizeof(backend
->host_nodes
) >=
375 BITS_TO_LONGS(MAX_NODES
+ 1) * sizeof(unsigned long));
376 assert(maxnode
<= MAX_NODES
);
378 #ifdef HAVE_NUMA_HAS_PREFERRED_MANY
379 if (mode
== MPOL_PREFERRED
&& numa_has_preferred_many() > 0) {
381 * Replace with MPOL_PREFERRED_MANY otherwise the mbind() below
382 * silently picks the first node.
384 mode
= MPOL_PREFERRED_MANY
;
389 mbind(ptr
, sz
, mode
, backend
->host_nodes
, maxnode
+ 1, flags
)) {
390 if (backend
->policy
!= MPOL_DEFAULT
|| errno
!= ENOSYS
) {
391 error_setg_errno(errp
, errno
,
392 "cannot bind memory to host NUMA nodes");
397 /* Preallocate memory after the NUMA policy has been instantiated.
398 * This is necessary to guarantee memory is allocated with
399 * specified NUMA policy in place.
401 if (backend
->prealloc
) {
402 qemu_prealloc_mem(memory_region_get_fd(&backend
->mr
), ptr
, sz
,
403 backend
->prealloc_threads
,
404 backend
->prealloc_context
, &local_err
);
411 error_propagate(errp
, local_err
);
415 host_memory_backend_can_be_deleted(UserCreatable
*uc
)
417 if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc
))) {
424 static bool host_memory_backend_get_share(Object
*o
, Error
**errp
)
426 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
428 return backend
->share
;
431 static void host_memory_backend_set_share(Object
*o
, bool value
, Error
**errp
)
433 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
435 if (host_memory_backend_mr_inited(backend
)) {
436 error_setg(errp
, "cannot change property value");
439 backend
->share
= value
;
443 static bool host_memory_backend_get_reserve(Object
*o
, Error
**errp
)
445 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
447 return backend
->reserve
;
450 static void host_memory_backend_set_reserve(Object
*o
, bool value
, Error
**errp
)
452 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
454 if (host_memory_backend_mr_inited(backend
)) {
455 error_setg(errp
, "cannot change property value");
458 if (backend
->prealloc
&& !value
) {
459 error_setg(errp
, "'prealloc=on' and 'reserve=off' are incompatible");
462 backend
->reserve
= value
;
464 #endif /* CONFIG_LINUX */
467 host_memory_backend_get_use_canonical_path(Object
*obj
, Error
**errp
)
469 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
471 return backend
->use_canonical_path
;
475 host_memory_backend_set_use_canonical_path(Object
*obj
, bool value
,
478 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
480 backend
->use_canonical_path
= value
;
484 host_memory_backend_class_init(ObjectClass
*oc
, void *data
)
486 UserCreatableClass
*ucc
= USER_CREATABLE_CLASS(oc
);
488 ucc
->complete
= host_memory_backend_memory_complete
;
489 ucc
->can_be_deleted
= host_memory_backend_can_be_deleted
;
491 object_class_property_add_bool(oc
, "merge",
492 host_memory_backend_get_merge
,
493 host_memory_backend_set_merge
);
494 object_class_property_set_description(oc
, "merge",
495 "Mark memory as mergeable");
496 object_class_property_add_bool(oc
, "dump",
497 host_memory_backend_get_dump
,
498 host_memory_backend_set_dump
);
499 object_class_property_set_description(oc
, "dump",
500 "Set to 'off' to exclude from core dump");
501 object_class_property_add_bool(oc
, "prealloc",
502 host_memory_backend_get_prealloc
,
503 host_memory_backend_set_prealloc
);
504 object_class_property_set_description(oc
, "prealloc",
505 "Preallocate memory");
506 object_class_property_add(oc
, "prealloc-threads", "int",
507 host_memory_backend_get_prealloc_threads
,
508 host_memory_backend_set_prealloc_threads
,
510 object_class_property_set_description(oc
, "prealloc-threads",
511 "Number of CPU threads to use for prealloc");
512 object_class_property_add_link(oc
, "prealloc-context",
513 TYPE_THREAD_CONTEXT
, offsetof(HostMemoryBackend
, prealloc_context
),
514 object_property_allow_set_link
, OBJ_PROP_LINK_STRONG
);
515 object_class_property_set_description(oc
, "prealloc-context",
516 "Context to use for creating CPU threads for preallocation");
517 object_class_property_add(oc
, "size", "int",
518 host_memory_backend_get_size
,
519 host_memory_backend_set_size
,
521 object_class_property_set_description(oc
, "size",
522 "Size of the memory region (ex: 500M)");
523 object_class_property_add(oc
, "host-nodes", "int",
524 host_memory_backend_get_host_nodes
,
525 host_memory_backend_set_host_nodes
,
527 object_class_property_set_description(oc
, "host-nodes",
528 "Binds memory to the list of NUMA host nodes");
529 object_class_property_add_enum(oc
, "policy", "HostMemPolicy",
530 &HostMemPolicy_lookup
,
531 host_memory_backend_get_policy
,
532 host_memory_backend_set_policy
);
533 object_class_property_set_description(oc
, "policy",
534 "Set the NUMA policy");
535 object_class_property_add_bool(oc
, "share",
536 host_memory_backend_get_share
, host_memory_backend_set_share
);
537 object_class_property_set_description(oc
, "share",
538 "Mark the memory as private to QEMU or shared");
540 object_class_property_add_bool(oc
, "reserve",
541 host_memory_backend_get_reserve
, host_memory_backend_set_reserve
);
542 object_class_property_set_description(oc
, "reserve",
543 "Reserve swap space (or huge pages) if applicable");
544 #endif /* CONFIG_LINUX */
546 * Do not delete/rename option. This option must be considered stable
547 * (as if it didn't have the 'x-' prefix including deprecation period) as
548 * long as 4.0 and older machine types exists.
549 * Option will be used by upper layers to override (disable) canonical path
550 * for ramblock-id set by compat properties on old machine types ( <= 4.0),
551 * to keep migration working when backend is used for main RAM with
552 * -machine memory-backend= option (main RAM historically used prefix-less
555 object_class_property_add_bool(oc
, "x-use-canonical-path-for-ramblock-id",
556 host_memory_backend_get_use_canonical_path
,
557 host_memory_backend_set_use_canonical_path
);
560 static const TypeInfo host_memory_backend_info
= {
561 .name
= TYPE_MEMORY_BACKEND
,
562 .parent
= TYPE_OBJECT
,
564 .class_size
= sizeof(HostMemoryBackendClass
),
565 .class_init
= host_memory_backend_class_init
,
566 .instance_size
= sizeof(HostMemoryBackend
),
567 .instance_init
= host_memory_backend_init
,
568 .instance_post_init
= host_memory_backend_post_init
,
569 .interfaces
= (InterfaceInfo
[]) {
570 { TYPE_USER_CREATABLE
},
575 static void register_types(void)
577 type_register_static(&host_memory_backend_info
);
580 type_init(register_types
);