Re: [libvirt] [PATCH 0/3] Add vhost-user-scsi/blk support
On 9/30/19 8:54 AM, Li Feng wrote:
The latest qemu has support the vhost-user-scsi and vhost-user-blk
backend for
better performance.
This patch let the libvirt could manage the vhost-user disks.
Usage in xml like this:
<vhost-user-blk-pci type='unix'>
<source type='bind' path='/tmp/vhost-blk.sock'>
<reconnect enabled='yes' timeout='5' />
</source>
<queue num='4'/>
</vhost-user-blk-pci>
This is not so nice XML. We tend to hide these implementation details
from users. We already have <drive/> element which has <driver
name=''/>
sub-element. Maybe we can use that somehow? It's better to create the
abstraction at libvirt layer instead of blindly exposing qemu's
features. For instance, recently vhost-user-vga and vhost-user-gpu-pci
was introduced into libvirt and it's hidden under <video/> element:
fc91a182d7cc78fe5ce0e03ef709c834035e0f33 and friends
Li Feng (3):
qemu: Add vhost-user-scsi/blk support
vhost-user-scsi/blk: add xml validation check
qemu: Add multiqueue support for vhost-user-scsi/blk
docs/schemas/domaincommon.rng | 48 ++++++++++
And you want to document every new feature that's added (and also
mention it in docs/news.xml).
src/conf/device_conf.h | 5 +
src/conf/domain_conf.c | 93 +++++++++++++++++-
src/conf/domain_conf.h | 9 ++
src/qemu/qemu_alias.c | 16 ++++
src/qemu/qemu_command.c | 105 +++++++++++++++++++++
src/qemu/qemu_domain.c | 2 +
.../vhost-user-blk.x86_64-latest.args | 37 ++++++++
tests/qemuxml2argvdata/vhost-user-blk.xml | 38 ++++++++
.../vhost-user-scsi.x86_64-latest.args | 37 ++++++++
tests/qemuxml2argvdata/vhost-user-scsi.xml | 37 ++++++++
tests/qemuxml2argvtest.c | 3 +
12 files changed, 429 insertions(+), 1 deletion(-)
create mode 100644 tests/qemuxml2argvdata/vhost-user-blk.x86_64-latest.args
create mode 100644 tests/qemuxml2argvdata/vhost-user-blk.xml
create mode 100644 tests/qemuxml2argvdata/vhost-user-scsi.x86_64-latest.args
create mode 100644 tests/qemuxml2argvdata/vhost-user-scsi.xml
I haven't looked at any code, because we need to figure out XML design
first.
Michal