On Thu, May 13, 2021 at 11:45:25 +0800, Zhiyong Ye wrote: > Implements QEMU support for NVMe device emulation, which was added in > qemu 1.6.0: > > https://git.qemu.org/?p=qemu.git;a=commit;h=f3c507adcd7 > > It can be used by adding elements in XML like this: > <devices> > ... > <disk type='file' device='disk'> > <driver name='qemu' type='raw'/> > <source file='/var/lib/libvirt/images/nvme.img'/> > <target dev='nvmea' bus='nvme'/> > <serial>QEMU-WMAP9A966149</serial> > </disk> > ... > </devices> > > Signed-off-by: Zhiyong Ye <[email protected]> > Signed-off-by: zhenwei pi <[email protected]> > --- > docs/formatdomain.rst | 11 ++++++-- > docs/schemas/domaincommon.rng | 3 ++- > src/bhyve/bhyve_command.c | 1 + > src/bhyve/bhyve_domain.c | 1 + > src/conf/domain_conf.c | 5 ++++ > src/conf/domain_conf.h | 1 + > src/conf/domain_validate.c | 4 ++- > src/hyperv/hyperv_driver.c | 2 ++ > src/qemu/qemu_alias.c | 1 + > src/qemu/qemu_command.c | 8 ++++++ > src/qemu/qemu_domain_address.c | 11 +++++--- > src/qemu/qemu_hotplug.c | 2 ++ > src/qemu/qemu_validate.c | 13 ++++++++++ > src/util/virutil.c | 3 ++- > src/vbox/vbox_common.c | 1 + > src/vz/vz_sdk.c | 2 ++ > src/vz/vz_utils.c | 1 + > tests/qemuxml2argvdata/disk-nvme-device.args | 28 +++++++++++++++++++++ > tests/qemuxml2argvdata/disk-nvme-device.xml | 30 ++++++++++++++++++++++ > tests/qemuxml2argvtest.c | 1 + > tests/qemuxml2xmloutdata/disk-nvme-device.xml | 36 > +++++++++++++++++++++++++++ > tests/qemuxml2xmltest.c | 1 + > 22 files changed, 157 insertions(+), 9 deletions(-) > create mode 100644 tests/qemuxml2argvdata/disk-nvme-device.args > create mode 100644 tests/qemuxml2argvdata/disk-nvme-device.xml > create mode 100644 tests/qemuxml2xmloutdata/disk-nvme-device.xml
Hi, firstly it's customary in libvirt to split patches at least into the following categories: conf/ and XML additions qemu/qemu_capabilities as separate change and then required changes to the qemu driver possibly also split if there are appropriate points. Please note that after every commit the tree MUST compile cleanly although obviously the new feature is not required to work properly until the end of the series. Additionally there was a discussion some time ago where we've discussed some of the technicalies: https://listman.redhat.com/archives/libvir-list/2020-November/msg00415.html We've discussed there that the NVMe device should be more like a '<controller>' and '<disk>' would map to a signle namespace on the controller. According to that thread NVMe actually supports that.
