11:53 a.m.
Halil Pasic (1):
Reflect in virtiofs.rst that virtiofs can be used without NUMA
Marc Hartmayer (1):
qemu: virtiofs can be used without NUMA nodes
docs/kbase/virtiofs.rst | 17 ++++++++++++-----
src/qemu/qemu_validate.c | 13 +++++++++----
2 files changed, 21 insertions(+), 9 deletions(-)
--
2.25.4
11:53 a.m.
New subject: [PATCH 1/2] qemu: virtiofs can be used without NUMA nodes
...if a machine memory-backend using shared memory is configured for
the guest. This is especially important for QEMU machine types that
don't have NUMA but virtiofs support.
An example snippet:
<domain type='kvm'>
<name>test</name>
<memory unit='KiB'>2097152</memory>
<memoryBacking>
<access mode='shared'/>
</memoryBacking>
<devices>
<filesystem type='mount' accessmode='passthrough'>
<driver type='virtiofs'/>
<source dir='/tmp/test'/>
<target dir='coffee'/>
</filesystem>
...
</devices>
...
</domain>
and the corresponding QEMU command line:
/usr/bin/qemu-system-s390x \
-machine s390-ccw-virtio-5.2,memory-backend=s390.ram \
-m 2048 \
-object
memory-backend-file,id=s390.ram,mem-path=/var/lib/libvirt/qemu/ram/46-test/s390.ram,share=yes,size=2147483648
\
...
Signed-off-by: Marc Hartmayer <mhartmay(a)linux.ibm.com>
---
src/qemu/qemu_validate.c | 13 +++++++++----
1 file changed, 9 insertions(+), 4 deletions(-)
diff --git a/src/qemu/qemu_validate.c b/src/qemu/qemu_validate.c
index 27e10d59fd25..bc3043bb3f0d 100644
--- a/src/qemu/qemu_validate.c
+++ b/src/qemu/qemu_validate.c
@@ -3470,14 +3470,19 @@ qemuValidateDomainDeviceDefGraphics(const virDomainGraphicsDef
*graphics,
static int
-qemuValidateDomainDefVirtioFSSharedMemory(const virDomainDef *def)
+qemuValidateDomainDefVirtioFSSharedMemory(const virDomainDef *def,
+ virQEMUCapsPtr qemuCaps)
{
+ const char *defaultRAMId = virQEMUCapsGetMachineDefaultRAMid(qemuCaps,
+ def->virtType,
+ def->os.machine);
size_t numa_nodes = virDomainNumaGetNodeCount(def->numa);
size_t i;
- if (numa_nodes == 0) {
+ if (numa_nodes == 0 &&
+ !(defaultRAMId && def->mem.access == VIR_DOMAIN_MEMORY_ACCESS_SHARED))
{
virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
- _("virtiofs requires one or more NUMA nodes"));
+ _("virtiofs requires shared memory"));
return -1;
}
@@ -3591,7 +3596,7 @@ qemuValidateDomainDeviceDefFS(virDomainFSDefPtr fs,
_("virtiofs does not support multidevs"));
return -1;
}
- if (qemuValidateDomainDefVirtioFSSharedMemory(def) < 0)
+ if (qemuValidateDomainDefVirtioFSSharedMemory(def, qemuCaps) < 0)
return -1;
break;
--
2.25.4
11:53 a.m.
New subject: [PATCH 2/2] Reflect in virtiofs.rst that virtiofs can be used without NUMA
From: Halil Pasic <pasic(a)linux.ibm.com>
Reflect in the virtiofs documentation that virtiofs can now be used
even without NUMA. While at it, be more precise where and why shared
memory is required.
Signed-off-by: Halil Pasic <pasic(a)linux.ibm.com>
Signed-off-by: Marc Hartmayer <mhartmay(a)linux.ibm.com>
---
docs/kbase/virtiofs.rst | 17 ++++++++++++-----
1 file changed, 12 insertions(+), 5 deletions(-)
diff --git a/docs/kbase/virtiofs.rst b/docs/kbase/virtiofs.rst
index dea8e79f833c..01440420d76d 100644
--- a/docs/kbase/virtiofs.rst
+++ b/docs/kbase/virtiofs.rst
@@ -16,10 +16,17 @@ See https://virtio-fs.gitlab.io/
Host setup
==========
-The host-side virtiofsd daemon, like other vhost-user backed devices,
-requires shared memory between the host and the guest. As of QEMU 4.2, this
-requires specifying a NUMA topology for the guest and explicitly specifying
-a memory backend. Multiple options are available:
+Almost all virtio devices (all that use virtqueues) require access to
+at least certain portions of guest RAM (possibly policed by DMA). In
+case of virtiofsd, much like in case of other vhost-user (see
+https://www.qemu.org/docs/master/interop/vhost-user.html) virtio
+devices that are realized by an userspace process, this in practice
+means that QEMU needs to allocate the backing memory for all the guest
+RAM as shared memory. As of QEMU 4.2, it is possible to explicitly
+specify a memory backend when specifying the NUMA topology. This
+method is however only viable for machine types that do support
+NUMA. As of QEMU 5.0.0, it is possible to specify the memory backend
+without NUMA (using the so called memobject interface).
Either of the following:
@@ -46,7 +53,7 @@ Either of the following:
Guest setup
===========
-#. Specify the NUMA topology
+#. Specify the NUMA topology (this step is only required for the NUMA case)
in the domain XML of the guest.
For the simplest one-node topology for a guest with 2GiB of RAM and 8 vCPUs:
--
2.25.4
12:10 p.m.
On 10/13/20 6:53 PM, Marc Hartmayer wrote:
Halil Pasic (1):
Reflect in virtiofs.rst that virtiofs can be used without NUMA
Marc Hartmayer (1):
qemu: virtiofs can be used without NUMA nodes
docs/kbase/virtiofs.rst | 17 ++++++++++++-----
src/qemu/qemu_validate.c | 13 +++++++++----
2 files changed, 21 insertions(+), 9 deletions(-)
Reviewed-by: Michal Privoznik <mprivozn(a)redhat.com>
and pushed.
Michal
12:20 p.m.
On Tue, Oct 13, 2020 at 07:10 PM +0200, Michal Privoznik <mprivozn(a)redhat.com>
wrote:
On 10/13/20 6:53 PM, Marc Hartmayer wrote:
> Halil Pasic (1):
> Reflect in virtiofs.rst that virtiofs can be used without NUMA
>
> Marc Hartmayer (1):
> qemu: virtiofs can be used without NUMA nodes
>
> docs/kbase/virtiofs.rst | 17 ++++++++++++-----
> src/qemu/qemu_validate.c | 13 +++++++++----
> 2 files changed, 21 insertions(+), 9 deletions(-)
>
Reviewed-by: Michal Privoznik <mprivozn(a)redhat.com>
and pushed.
Michal
Thanks.
--
Kind regards / Beste Grüße
Marc Hartmayer
IBM Deutschland Research & Development GmbH
Vorsitzender des Aufsichtsrats: Gregor Pillen
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen
Registergericht: Amtsgericht Stuttgart, HRB 243294
1533
days inactive
1533
days old
4 comments
2 participants
participants (2)
-
Marc Hartmayer -
Michal Privoznik