Igor Mammedov <imammedo(a)redhat.com&gt; writes: Suggest s/to MachineInfo description in QAPI schema/to output of query-machines/, because query-machines is the external interface people know. This is correct when the TYPE_VIRT_MACHINE, TYPE_PC_MACHINE and TYPE_SPAPR_MACHINE are exactly the machines supporting NUMA. How could I check that? "by the machine type", for consistency with @cpu-max. Also, rather long line.

Legacy '-numa node,mem' option has a number of issues and mgmt often defaults to it. Unfortunately it's no possible to replace it with an alternative '-numa memdev' without breaking migration compatibility. What's possible though is to deprecate it, keeping option working with old machine types only. In order to help users to find out if being deprecated CLI option '-numa node,mem' is still supported by particular machine type, add new "numa-mem-supported" property to output of query-machines. "numa-mem-supported" is set to 'true' for machines that currently support NUMA, but it will be flipped to 'false' later on, once deprecation period expires and kept 'true' only for old machine types that used to support the legacy option so it won't break existing configuration that are using it. Signed-off-by: Igor Mammedov <imammedo(a)redhat.com&gt; --- v5: (Markus Armbruster <armbru(a)redhat.com>) * s/by machine type/by the machine type/ * ammend commit message s/to MachineInfo description in QAPI schema/to output of query-machines/ v4: * drop idea to use "qom-list-properties" and use MachineInfo instead which could be inspected with 'query-machines' include/hw/boards.h | 3 +++ hw/arm/virt.c | 1 + hw/i386/pc.c | 1 + hw/ppc/spapr.c | 1 + qapi/misc.json | 5 ++++- vl.c | 1 + 6 files changed, 11 insertions(+), 1 deletion(-) diff --git a/include/hw/boards.h b/include/hw/boards.h index 6ff02bf..ab6badc 100644 --- a/include/hw/boards.h +++ b/include/hw/boards.h @@ -158,6 +158,8 @@ typedef struct { * @kvm_type: * Return the type of KVM corresponding to the kvm-type string option or * computed based on other criteria such as the host kernel capabilities. + * @numa_mem_supported: + * true if '--numa node.mem' option is supported and false otherwise */ struct MachineClass { /*< private >*/ @@ -210,6 +212,7 @@ struct MachineClass { bool ignore_boot_device_suffixes; bool smbus_no_migration_support; bool nvdimm_supported; + bool numa_mem_supported; HotplugHandler *(*get_hotplug_handler)(MachineState *machine, DeviceState *dev); diff --git a/hw/arm/virt.c b/hw/arm/virt.c index bf54f10..481a603 100644 --- a/hw/arm/virt.c +++ b/hw/arm/virt.c @@ -1943,6 +1943,7 @@ static void virt_machine_class_init(ObjectClass *oc, void *data) assert(!mc->get_hotplug_handler); mc->get_hotplug_handler = virt_machine_get_hotplug_handler; hc->plug = virt_machine_device_plug_cb; + mc->numa_mem_supported = true; } static void virt_instance_init(Object *obj) diff --git a/hw/i386/pc.c b/hw/i386/pc.c index edc240b..25146d7 100644 --- a/hw/i386/pc.c +++ b/hw/i386/pc.c @@ -2750,6 +2750,7 @@ static void pc_machine_class_init(ObjectClass *oc, void *data) nc->nmi_monitor_handler = x86_nmi; mc->default_cpu_type = TARGET_DEFAULT_CPU_TYPE; mc->nvdimm_supported = true; + mc->numa_mem_supported = true; object_class_property_add(oc, PC_MACHINE_DEVMEM_REGION_SIZE, "int", pc_machine_get_device_memory_region_size, NULL, diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c index e2b33e5..89d5814 100644 --- a/hw/ppc/spapr.c +++ b/hw/ppc/spapr.c @@ -4340,6 +4340,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data) * in which LMBs are represented and hot-added */ mc->numa_mem_align_shift = 28; + mc->numa_mem_supported = true; smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF; smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON; diff --git a/qapi/misc.json b/qapi/misc.json index 8b3ca4f..2dbfdf0 100644 --- a/qapi/misc.json +++ b/qapi/misc.json @@ -2018,12 +2018,15 @@ # # @hotpluggable-cpus: cpu hotplug via -device is supported (since 2.7.0) # +# @numa-mem-supported: true if '-numa node,mem' option is supported by +# the machine type and false otherwise (since 4.1) +# # Since: 1.2.0 ## { 'struct': 'MachineInfo', 'data': { 'name': 'str', '*alias': 'str', '*is-default': 'bool', 'cpu-max': 'int', - 'hotpluggable-cpus': 'bool'} } + 'hotpluggable-cpus': 'bool', 'numa-mem-supported': 'bool'} } ## # @query-machines: diff --git a/vl.c b/vl.c index cd1fbc4..f5b083f 100644 --- a/vl.c +++ b/vl.c @@ -1428,6 +1428,7 @@ MachineInfoList *qmp_query_machines(Error **errp) info->name = g_strdup(mc->name); info->cpu_max = !mc->max_cpus ? 1 : mc->max_cpus; info->hotpluggable_cpus = mc->has_hotpluggable_cpus; + info->numa_mem_supported = mc->numa_mem_supported; entry = g_malloc0(sizeof(*entry)); entry->value = info; -- 2.7.4

The parameter allows to configure fake NUMA topology where guest VM simulates NUMA topology but not actually getting performance benefits from it. The same or better results could be achieved using 'memdev' parameter. Beside of unpredictable performance, '-numa node.mem' option has other issues when it's used with combination of -mem-path + + -mem-prealloc + memdev backends (pc-dimm), breaking binding of memdev backends since mem-path/mem-prealloc are global and affect the most of RAM allocations. It's possible to make memdevs and global -mem-path/mem-prealloc to play nicely together but that will just complicate already complicated code and add unobious ways it could break on 2 different memmory allocation pathes and their combinations. Instead of it, consolidate all guest RAM allocation over memdev which still allows to create fake NUMA configurations if desired and leaves one simplifyed code path to consider when it comes to guest RAM allocation. To achieve desired simplification deprecate 'mem' parameter as its ad-hoc partitioning of initial RAM MemoryRegion can't be translated to memdev based backend transparently to users and in compatible manner (migration wise). Later down the road that will allow to consolidate means of how guest RAM is allocated and would permit us to clean up quite a bit memory allocations and numa code, leaving only 'memdev' implementation in place. Signed-off-by: Igor Mammedov <imammedo(a)redhat.com&gt; --- Notes: v4: * fix up documentation to mention where users should look to check if -numa node.mem is supported numa.c | 2 ++ qemu-deprecated.texi | 16 ++++++++++++++++ 2 files changed, 18 insertions(+) diff --git a/numa.c b/numa.c index 3875e1e..2205773 100644 --- a/numa.c +++ b/numa.c @@ -121,6 +121,8 @@ static void parse_numa_node(MachineState *ms, NumaNodeOptions *node, if (node->has_mem) { numa_info[nodenr].node_mem = node->mem; + warn_report("Parameter -numa node,mem is deprecated," + " use -numa node,memdev instead"); } if (node->has_memdev) { Object *o; diff --git a/qemu-deprecated.texi b/qemu-deprecated.texi index 50292d8..eb347f5 100644 --- a/qemu-deprecated.texi +++ b/qemu-deprecated.texi @@ -82,6 +82,22 @@ The @code{-realtime mlock=on|off} argument has been replaced by the The ``-virtfs_synth'' argument is now deprecated. Please use ``-fsdev synth'' and ``-device virtio-9p-...'' instead. +@subsection -numa node,mem=@var{size} (since 4.1) + +The parameter @option{mem} of @option{-numa node} is used to assign a part of +guest RAM to a NUMA node. But when using it, it's impossible to manage specified +RAM chunk on the host side (like bind it to a host node, setting bind policy, ...), +so guest end-ups with the fake NUMA configuration with suboptiomal performance. +However since 2014 there is an alternative way to assign RAM to a NUMA node +using parameter @option{memdev}, which does the same as @option{mem} and adds +means to actualy manage node RAM on the host side. Use parameter @option{memdev} +with @var{memory-backend-ram} backend as an replacement for parameter @option{mem} +to achieve the same fake NUMA effect or a properly configured +@var{memory-backend-file} backend to actually benefit from NUMA configuration. +In future new machine versions will not accept the option but it will still +work with old machine types. User can check QAPI schema to see if the legacy +option is supported by looking at MachineInfo::numa-mem-supported property. + @section QEMU Machine Protocol (QMP) commands @subsection block-dirty-bitmap-add "autoload" parameter (since 2.12.0) -- 2.7.4