| Message ID | 20220928164542.117952-5-david@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | hostmem: NUMA-aware memory preallocation using ThreadContext | expand |
David Hildenbrand <david@redhat.com> writes: > Let's make it easier to pin threads created via a ThreadContext to > all current CPUs belonging to given NUMA nodes. > > As "node-affinity" is simply a shortcut for setting "cpu-affinity", that > property cannot be read and if the CPUs for a node change due do CPU > hotplug, the CPU affinity will not get updated. Color me confused. > Reviewed-by: Michal Privoznik <mprivozn@redhat.com> > Signed-off-by: David Hildenbrand <david@redhat.com> > --- > qapi/qom.json | 7 +++- > util/meson.build | 2 +- > util/thread-context.c | 84 +++++++++++++++++++++++++++++++++++++++++++ > 3 files changed, 91 insertions(+), 2 deletions(-) > > diff --git a/qapi/qom.json b/qapi/qom.json > index 4775a333ed..d36bf3355f 100644 > --- a/qapi/qom.json > +++ b/qapi/qom.json > @@ -838,10 +838,15 @@ > # @cpu-affinity: the CPU affinity for all threads created in the thread > # context (default: QEMU main thread affinity) > # > +# @node-affinity: shortcut for looking up the current CPUs for the given nodes > +# and setting @cpu-affinity (default: QEMU main thread > +# affinity) > +# Still confused. Explain for dummies? With an example, perhaps? > # Since: 7.2 > ## > { 'struct': 'ThreadContextProperties', > - 'data': { '*cpu-affinity': ['uint16'] } } > + 'data': { '*cpu-affinity': ['uint16'], > + '*node-affinity': ['uint16'] } } > > > ## [...]
On 29.09.22 13:13, Markus Armbruster wrote: > David Hildenbrand <david@redhat.com> writes: > >> Let's make it easier to pin threads created via a ThreadContext to >> all current CPUs belonging to given NUMA nodes. >> >> As "node-affinity" is simply a shortcut for setting "cpu-affinity", that >> property cannot be read and if the CPUs for a node change due do CPU >> hotplug, the CPU affinity will not get updated. > > Color me confused. I adjusted that to: util: Add write-only "node-affinity" property for ThreadContext Let's make it easier to pin threads created via a ThreadContext to all CPUs currently belonging to a given set of NUMA nodes -- which is the common case. "node-affinity" is simply a shortcut for setting "cpu-affinity" manually to the list of CPUs belonging to the set of nodes. This property can only be written. A simple QEMU example to set the CPU affinity to Node 1 on a system with two NUMA nodes, 24 CPUs each: qemu-system-x86_64 -S \ -object thread-context,id=tc1,node-affinity=1 And we can query the cpu-affinity via HMP/QMP: (qemu) qom-get tc1 cpu-affinity [ 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31, 33, 35, 37, 39, 41, 43, 45, 47 ] We cannot query the node-affinity: (qemu) qom-get tc1 node-affinity Error: Insufficient permission to perform this operation But note that due to dynamic library loading this example will not work before we actually make use of thread_context_create_thread() in QEMU code, because the type will otherwise not get registered. Note that if the CPUs for a node change due do physical CPU hotplug or hotunplug (e.g., lscpu output changes) after the ThreadContext was started, the CPU affinity will not get updated. > >> Reviewed-by: Michal Privoznik <mprivozn@redhat.com> >> Signed-off-by: David Hildenbrand <david@redhat.com> >> --- >> qapi/qom.json | 7 +++- >> util/meson.build | 2 +- >> util/thread-context.c | 84 +++++++++++++++++++++++++++++++++++++++++++ >> 3 files changed, 91 insertions(+), 2 deletions(-) >> >> diff --git a/qapi/qom.json b/qapi/qom.json >> index 4775a333ed..d36bf3355f 100644 >> --- a/qapi/qom.json >> +++ b/qapi/qom.json >> @@ -838,10 +838,15 @@ >> # @cpu-affinity: the CPU affinity for all threads created in the thread >> # context (default: QEMU main thread affinity) >> # >> +# @node-affinity: shortcut for looking up the current CPUs for the given nodes >> +# and setting @cpu-affinity (default: QEMU main thread >> +# affinity) >> +# > > Still confused. Explain for dummies? With an example, perhaps? > I adjusted that to: +# @node-affinity: the list of node numbers that will be resolved to a list +# of CPU numbers used as CPU affinity. This is a shortcut for +# specifying the list of CPU numbers belonging to the nodes +# manually by setting @cpu-affinity. (default: QEMU main thread +# affinity) +# Thanks!
diff --git a/qapi/qom.json b/qapi/qom.json index 4775a333ed..d36bf3355f 100644 --- a/qapi/qom.json +++ b/qapi/qom.json @@ -838,10 +838,15 @@ # @cpu-affinity: the CPU affinity for all threads created in the thread # context (default: QEMU main thread affinity) # +# @node-affinity: shortcut for looking up the current CPUs for the given nodes +# and setting @cpu-affinity (default: QEMU main thread +# affinity) +# # Since: 7.2 ## { 'struct': 'ThreadContextProperties', - 'data': { '*cpu-affinity': ['uint16'] } } + 'data': { '*cpu-affinity': ['uint16'], + '*node-affinity': ['uint16'] } } ## diff --git a/util/meson.build b/util/meson.build index e97cd2d779..c0a7bc54d4 100644 --- a/util/meson.build +++ b/util/meson.build @@ -1,5 +1,5 @@ util_ss.add(files('osdep.c', 'cutils.c', 'unicode.c', 'qemu-timer-common.c')) -util_ss.add(files('thread-context.c')) +util_ss.add(files('thread-context.c'), numa) if not config_host_data.get('CONFIG_ATOMIC64') util_ss.add(files('atomic64.c')) endif diff --git a/util/thread-context.c b/util/thread-context.c index dcd607c532..880f0441be 100644 --- a/util/thread-context.c +++ b/util/thread-context.c @@ -22,6 +22,10 @@ #include "qemu/module.h" #include "qemu/bitmap.h" +#ifdef CONFIG_NUMA +#include <numa.h> +#endif + enum { TC_CMD_NONE = 0, TC_CMD_STOP, @@ -89,6 +93,11 @@ static void thread_context_set_cpu_affinity(Object *obj, Visitor *v, int nbits = 0, ret; Error *err = NULL; + if (tc->init_cpu_bitmap) { + error_setg(errp, "Mixing CPU and node affinity not supported"); + return; + } + visit_type_uint16List(v, name, &host_cpus, &err); if (err) { error_propagate(errp, err); @@ -160,6 +169,79 @@ static void thread_context_get_cpu_affinity(Object *obj, Visitor *v, qapi_free_uint16List(host_cpus); } +static void thread_context_set_node_affinity(Object *obj, Visitor *v, + const char *name, void *opaque, + Error **errp) +{ +#ifdef CONFIG_NUMA + const int nbits = numa_num_possible_cpus(); + ThreadContext *tc = THREAD_CONTEXT(obj); + uint16List *l, *host_nodes = NULL; + unsigned long *bitmap = NULL; + struct bitmask *tmp_cpus; + Error *err = NULL; + int ret, i; + + if (tc->init_cpu_bitmap) { + error_setg(errp, "Mixing CPU and node affinity not supported"); + return; + } + + visit_type_uint16List(v, name, &host_nodes, &err); + if (err) { + error_propagate(errp, err); + return; + } + + if (!host_nodes) { + error_setg(errp, "Node list is empty"); + goto out; + } + + bitmap = bitmap_new(nbits); + tmp_cpus = numa_allocate_cpumask(); + for (l = host_nodes; l; l = l->next) { + numa_bitmask_clearall(tmp_cpus); + ret = numa_node_to_cpus(l->value, tmp_cpus); + if (ret) { + /* We ignore any errors, such as impossible nodes. */ + continue; + } + for (i = 0; i < nbits; i++) { + if (numa_bitmask_isbitset(tmp_cpus, i)) { + set_bit(i, bitmap); + } + } + } + numa_free_cpumask(tmp_cpus); + + if (bitmap_empty(bitmap, nbits)) { + error_setg(errp, "The nodes select no CPUs"); + goto out; + } + + if (tc->thread_id != -1) { + /* + * Note: we won't be adjusting the affinity of any thread that is still + * around for now, but only the affinity of the context thread. + */ + ret = qemu_thread_set_affinity(&tc->thread, bitmap, nbits); + if (ret) { + error_setg(errp, "Setting CPU affinity failed: %s", strerror(ret)); + } + } else { + tc->init_cpu_bitmap = bitmap; + bitmap = NULL; + tc->init_cpu_nbits = nbits; + } +out: + g_free(bitmap); + qapi_free_uint16List(host_nodes); +#else + error_setg(errp, "NUMA node affinity is not supported by this QEMU"); +#endif +} + static void thread_context_get_thread_id(Object *obj, Visitor *v, const char *name, void *opaque, Error **errp) @@ -209,6 +291,8 @@ static void thread_context_class_init(ObjectClass *oc, void *data) object_class_property_add(oc, "cpu-affinity", "int", thread_context_get_cpu_affinity, thread_context_set_cpu_affinity, NULL, NULL); + object_class_property_add(oc, "node-affinity", "int", NULL, + thread_context_set_node_affinity, NULL, NULL); } static void thread_context_instance_init(Object *obj)