The aim of thread-context object is to set affinity on threads
that allocate memory for a memory-backend-* object. For instance:

-object '{"qom-type":"thread-context","id":"tc-ram-node0","node-affinity":[3]}' \
-object '{"qom-type":"memory-backend-memfd","id":"ram-node0","hugetlb":true,\
          "hugetlbsize":2097152,"share":true,"prealloc":true,"prealloc-threads":8,\
          "size":15032385536,"host-nodes":[3],"policy":"preferred",\
          "prealloc-context":"tc-ram-node0"}' \

allocates 14GiB worth of memory, backed by 2MiB hugepages from
host NUMA node 3, using 8 threads. If it weren't for
thread-context these threads wouldn't have any affinity and thus
theoretically could be scheduled to run on CPUs of different NUMA
node (which is what I saw occasionally).

Therefore, whenever we are pinning memory (IOW setting host-nodes
attribute), we can generate thread-context object with the same
affinity.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
---
 src/qemu/qemu_command.c | 48 +++++++++++++++++++++++++++++++++++++++++
 src/qemu/qemu_command.h |  5 +++++
 2 files changed, 53 insertions(+)

diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c
index 41b9f7cb52..392b248628 100644
--- a/src/qemu/qemu_command.c
+++ b/src/qemu/qemu_command.c
@@ -3603,6 +3603,54 @@ qemuBuildMemoryDeviceProps(virQEMUDriverConfig *cfg,
 }
 
 
+int
+qemuBuildThreadContextProps(virJSONValue **tcProps,
+                            virJSONValue **memProps,
+                            qemuDomainObjPrivate *priv)
+{
+    g_autoptr(virJSONValue) props = NULL;
+    virJSONValue *nodemask = NULL;
+    g_autoptr(virJSONValue) nodemaskCopy = NULL;
+    g_autofree char *tcAlias = NULL;
+    const char *memalias = NULL;
+
+    if (tcProps)
+        *tcProps = NULL;
+
+    if (!virQEMUCapsGet(priv->qemuCaps, QEMU_CAPS_THREAD_CONTEXT))
+        return 0;
+
+    nodemask = virJSONValueObjectGetArray(*memProps, "host-nodes");
+    if (!nodemask)
+        return 0;
+
+    memalias = virJSONValueObjectGetString(*memProps, "id");
+    if (!memalias) {
+        virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
+                       _("memory device alias is not assigned"));
+        return -1;
+    }
+
+    tcAlias = g_strdup_printf("tc-%s", memalias);
+    nodemaskCopy = virJSONValueCopy(nodemask);
+
+    if (virJSONValueObjectAdd(&props,
+                              "s:qom-type", "thread-context",
+                              "s:id", tcAlias,
+                              "a:node-affinity", &nodemaskCopy,
+                              NULL) < 0)
+        return -1;
+
+    if (virJSONValueObjectAdd(memProps,
+                              "s:prealloc-context", tcAlias,
+                              NULL) < 0)
+        return -1;
+
+    *tcProps = g_steal_pointer(&props);
+    return 0;
+}
+
+
 static char *
 qemuBuildLegacyNicStr(virDomainNetDef *net)
 {
diff --git a/src/qemu/qemu_command.h b/src/qemu/qemu_command.h
index 2578367df9..761cc5d865 100644
--- a/src/qemu/qemu_command.h
+++ b/src/qemu/qemu_command.h
@@ -147,6 +147,11 @@ qemuBuildMemoryDeviceProps(virQEMUDriverConfig *cfg,
                            const virDomainDef *def,
                            const virDomainMemoryDef *mem);
 
+int
+qemuBuildThreadContextProps(virJSONValue **tcProps,
+                            virJSONValue **memProps,
+                            qemuDomainObjPrivate *priv);
+
 /* Current, best practice */
 virJSONValue *
 qemuBuildPCIHostdevDevProps(const virDomainDef *def,
-- 
2.37.4

On Mon, Nov 14, 2022 at 04:35:30PM +0100, Michal Privoznik wrote:
>The aim of thread-context object is to set affinity on threads
>that allocate memory for a memory-backend-* object. For instance:
>
>-object '{"qom-type":"thread-context","id":"tc-ram-node0","node-affinity":[3]}' \
>-object '{"qom-type":"memory-backend-memfd","id":"ram-node0","hugetlb":true,\
>          "hugetlbsize":2097152,"share":true,"prealloc":true,"prealloc-threads":8,\
>          "size":15032385536,"host-nodes":[3],"policy":"preferred",\
>          "prealloc-context":"tc-ram-node0"}' \
>
>allocates 14GiB worth of memory, backed by 2MiB hugepages from
>host NUMA node 3, using 8 threads. If it weren't for
>thread-context these threads wouldn't have any affinity and thus
>theoretically could be scheduled to run on CPUs of different NUMA
>node (which is what I saw occasionally).
>
>Therefore, whenever we are pinning memory (IOW setting host-nodes
>attribute), we can generate thread-context object with the same
>affinity.
>
>Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
>---
> src/qemu/qemu_command.c | 48 +++++++++++++++++++++++++++++++++++++++++
> src/qemu/qemu_command.h |  5 +++++
> 2 files changed, 53 insertions(+)
>
>diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c
>index 41b9f7cb52..392b248628 100644
>--- a/src/qemu/qemu_command.c
>+++ b/src/qemu/qemu_command.c
>@@ -3603,6 +3603,54 @@ qemuBuildMemoryDeviceProps(virQEMUDriverConfig *cfg,
> }
>
>
>+int
>+qemuBuildThreadContextProps(virJSONValue **tcProps,
>+                            virJSONValue **memProps,
>+                            qemuDomainObjPrivate *priv)
>+{
>+    g_autoptr(virJSONValue) props = NULL;
>+    virJSONValue *nodemask = NULL;
>+    g_autoptr(virJSONValue) nodemaskCopy = NULL;
>+    g_autofree char *tcAlias = NULL;
>+    const char *memalias = NULL;
>+
>+    if (tcProps)
>+        *tcProps = NULL;
>+

What's the reason for this function to support tcProps == NULL?

>+    if (!virQEMUCapsGet(priv->qemuCaps, QEMU_CAPS_THREAD_CONTEXT))
>+        return 0;
>+
>+    nodemask = virJSONValueObjectGetArray(*memProps, "host-nodes");
>+    if (!nodemask)
>+        return 0;
>+
>+    memalias = virJSONValueObjectGetString(*memProps, "id");
>+    if (!memalias) {
>+        virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
>+                       _("memory device alias is not assigned"));
>+        return -1;
>+    }
>+
>+    tcAlias = g_strdup_printf("tc-%s", memalias);
>+    nodemaskCopy = virJSONValueCopy(nodemask);
>+
>+    if (virJSONValueObjectAdd(&props,
>+                              "s:qom-type", "thread-context",
>+                              "s:id", tcAlias,
>+                              "a:node-affinity", &nodemaskCopy,
>+                              NULL) < 0)
>+        return -1;
>+
>+    if (virJSONValueObjectAdd(memProps,
>+                              "s:prealloc-context", tcAlias,
>+                              NULL) < 0)
>+        return -1;
>+
>+    *tcProps = g_steal_pointer(&props);

It would crash here anyway if tcProps == NULL; otherwise

Reviewed-by: Martin Kletzander <mkletzan@redhat.com>

>+    return 0;
>+}
>+
>+
> static char *
> qemuBuildLegacyNicStr(virDomainNetDef *net)
> {
>diff --git a/src/qemu/qemu_command.h b/src/qemu/qemu_command.h
>index 2578367df9..761cc5d865 100644
>--- a/src/qemu/qemu_command.h
>+++ b/src/qemu/qemu_command.h
>@@ -147,6 +147,11 @@ qemuBuildMemoryDeviceProps(virQEMUDriverConfig *cfg,
>                            const virDomainDef *def,
>                            const virDomainMemoryDef *mem);
>
>+int
>+qemuBuildThreadContextProps(virJSONValue **tcProps,
>+                            virJSONValue **memProps,
>+                            qemuDomainObjPrivate *priv);
>+
> /* Current, best practice */
> virJSONValue *
> qemuBuildPCIHostdevDevProps(const virDomainDef *def,
>-- 
>2.37.4
>