This option provides the infrastructure for binding guest NUMA nodes to host NUMA nodes. For example:
-object memory-ram,size=1024M,policy=membind,host-nodes=0,id=ram-node0 \ -numa node,nodeid=0,cpus=0,memdev=ram-node0 \ -object memory-ram,size=1024M,policy=interleave,host-nodes=1-3,id=ram-node1 \ -numa node,nodeid=1,cpus=1,memdev=ram-node1 The option replaces "-numa node,mem=". Signed-off-by: Paolo Bonzini <pbonz...@redhat.com> --- include/sysemu/sysemu.h | 1 + numa.c | 63 +++++++++++++++++++++++++++++++++++++++++++++++-- qapi-schema.json | 8 ++++++- qemu-options.hx | 12 ++++++---- 4 files changed, 77 insertions(+), 7 deletions(-) diff --git a/include/sysemu/sysemu.h b/include/sysemu/sysemu.h index 54a6f28..4870129 100644 --- a/include/sysemu/sysemu.h +++ b/include/sysemu/sysemu.h @@ -139,6 +139,7 @@ extern int nb_numa_nodes; typedef struct node_info { uint64_t node_mem; DECLARE_BITMAP(node_cpu, MAX_CPUMASK_BITS); + struct HostMemoryBackend *node_memdev; } NodeInfo; extern NodeInfo numa_info[MAX_NODES]; void set_numa_nodes(void); diff --git a/numa.c b/numa.c index 930f49d..b00ef90 100644 --- a/numa.c +++ b/numa.c @@ -32,6 +32,7 @@ #include "qapi/dealloc-visitor.h" #include "qapi/qmp/qerror.h" #include "hw/boards.h" +#include "sysemu/hostmem.h" QemuOptsList qemu_numa_opts = { .name = "numa", @@ -40,6 +41,8 @@ QemuOptsList qemu_numa_opts = { .desc = { { 0 } } /* validated with OptsVisitor */ }; +static int have_memdevs = -1; + static void numa_node_parse(NumaNodeOptions *node, QemuOpts *opts, Error **errp) { uint16_t nodenr; @@ -66,6 +69,20 @@ static void numa_node_parse(NumaNodeOptions *node, QemuOpts *opts, Error **errp) bitmap_set(numa_info[nodenr].node_cpu, cpus->value, 1); } + if (node->has_mem && node->has_memdev) { + error_setg(errp, "qemu: cannot specify both mem= and memdev=\n"); + return; + } + + if (have_memdevs == -1) { + have_memdevs = node->has_memdev; + } + if (node->has_memdev != have_memdevs) { + error_setg(errp, "qemu: memdev option must be specified for either " + "all or no nodes\n"); + return; + } + if (node->has_mem) { uint64_t mem_size = node->mem; const char *mem_str = qemu_opt_get(opts, "mem"); @@ -75,6 +92,18 @@ static void numa_node_parse(NumaNodeOptions *node, QemuOpts *opts, Error **errp) } numa_info[nodenr].node_mem = mem_size; } + if (node->has_memdev) { + Object *o; + o = object_resolve_path_type(node->memdev, TYPE_MEMORY_BACKEND, NULL); + if (!o) { + error_setg(errp, "memdev=%s is ambiguous", node->memdev); + return; + } + + object_ref(o); + numa_info[nodenr].node_mem = object_property_get_int(o, "size", NULL); + numa_info[nodenr].node_memdev = MEMORY_BACKEND(o); + } } int numa_init_func(QemuOpts *opts, void *opaque) @@ -193,12 +222,42 @@ void set_numa_modes(void) } } +static void allocate_system_memory_nonnuma(MemoryRegion *mr, Object *owner, + const char *name, + QEMUMachineInitArgs *args) +{ + uint64_t ram_size = args->ram_size; + + memory_region_init_ram(mr, owner, name, ram_size); + vmstate_register_ram_global(mr); +} + void memory_region_allocate_system_memory(MemoryRegion *mr, Object *owner, const char *name, QEMUMachineInitArgs *args) { uint64_t ram_size = args->ram_size; + uint64_t addr = 0; + int i; - memory_region_init_ram(mr, owner, name, ram_size); - vmstate_register_ram_global(mr); + if (nb_numa_nodes == 0 || !have_memdevs) { + allocate_system_memory_nonnuma(mr, owner, name, args); + return; + } + + memory_region_init(mr, owner, name, ram_size); + for (i = 0; i < nb_numa_nodes; i++) { + Error *local_err = NULL; + uint64_t size = numa_info[i].node_mem; + HostMemoryBackend *backend = numa_info[i].node_memdev; + MemoryRegion *seg = host_memory_backend_get_memory(backend, &local_err); + if (local_err) { + qerror_report_err(local_err); + exit(1); + } + + memory_region_add_subregion(mr, addr, seg); + vmstate_register_ram_global(seg); + addr += size; + } } diff --git a/qapi-schema.json b/qapi-schema.json index ca8c11e..8bd84da 100644 --- a/qapi-schema.json +++ b/qapi-schema.json @@ -4543,10 +4543,16 @@ # @mem: #optional memory size of this node (equally divide total memory among # nodes if omitted) # +# @memdev: #optional memory backend object. If specified for one node, +# it must be specified for all nodes. +# +# @mem: #optional memory size of this node; mutually exclusive with @memdev. +# # Since: 2.1 ## { 'type': 'NumaNodeOptions', 'data': { '*nodeid': 'uint16', '*cpus': ['uint16'], - '*mem': 'size' }} + '*mem': 'size', + '*memdev': 'str' }} diff --git a/qemu-options.hx b/qemu-options.hx index 98e78ca..f3bf291 100644 --- a/qemu-options.hx +++ b/qemu-options.hx @@ -95,16 +95,20 @@ specifies the maximum number of hotpluggable CPUs. ETEXI DEF("numa", HAS_ARG, QEMU_OPTION_numa, - "-numa node[,mem=size][,cpus=cpu[-cpu]][,nodeid=node]\n", QEMU_ARCH_ALL) + "-numa node[,mem=size][,memdev=id][,cpus=cpu[-cpu]][,nodeid=node]\n", QEMU_ARCH_ALL) STEXI -@item -numa node[,mem=@var{size}][,cpus=@var{cpu[-cpu]}][,nodeid=@var{node}] +@item -numa node[,mem=@var{size}][,memdev=@var{id}][,cpus=@var{cpu[-cpu]}][,nodeid=@var{node}] @findex -numa -Simulate a multi node NUMA system. If @samp{mem} +Simulate a multi node NUMA system. If @samp{mem}, @samp{memdev} and @samp{cpus} are omitted, resources are split equally. Also, note that the -@option{numa} option doesn't allocate any of the specified resources. That is, it just assigns existing resources to NUMA nodes. This means that one still has to use the @option{-m}, @option{-smp} options -to respectively allocate RAM and vCPUs. +to respectively allocate RAM and vCPUs, and possibly @option{-object} +to specify the memory backend for the @samp{memdev} suboption. + +@samp{mem} and @samp{memdev} are mutually exclusive. Furthermore, if one +node uses @samp{memdev}, all of them have to use it. ETEXI DEF("add-fd", HAS_ARG, QEMU_OPTION_add_fd, -- 1.8.5.3