On Mon, Feb 24, 2020 at 01:46:22AM -0500, Ryan Moeller wrote:
Handle PCI passthrough and virtio-scsi using hostdev devices.
Example PCI passthrough: domain xml snippet ``` <memoryBacking> <locked/> </memoryBacking> <devices> <hostdev mode='subsystem' type='pci'> <source> <address domain='0x0000' bus='0x06' slot='0x02' function='0x00'/> </source> <address type='pci' domain='0x0000' bus='0x00' slot='0x02' function='0x00'/> </hostdev> </devices> ``` loader.conf snippet ``` vmm_load="YES" pptdevs="6/2/0" ```
Example SCSI passthrough: domain xml snippet ``` <hostdev mode='subsystem' type='scsi_ctl' model='virtio'> <source protocol='ioctl' pp='5' vp='0'/> <address type='pci' domain='0x0000' bus='0x00' slot='0x02' function='0x00'/> </hostdev> ``` ctl.conf snippet ``` portal-group "pg0" { discovery-auth-group "no-authentication" listen "127.0.0.1" }
target iqn.2020-01.com.example:target0 { auth-group "no-authentication" portal-group "pg0" port ioctl/5/0
lun 0 { path "/dev/zvol/storage/lun0" } lun 1 { path "/dev/zvol/storage/lun1" } lun 2 { path "/dev/zvol/storage/lun2" } lun 3 { path "/dev/zvol/storage/lun3" } } ```
Can you split this patch up into a few pieces. We generally want the changes to the XML parser/formatter to be separate from any driver code, as the first part. Then I'd suggest a separate patch for the PCI and the SCSI hostdev support in bhyve.
Signed-off-by: Ryan Moeller <ryan@iXsystems.com> --- docs/schemas/domaincommon.rng | 30 ++++ src/bhyve/bhyve_capabilities.c | 14 ++ src/bhyve/bhyve_capabilities.h | 1 + src/bhyve/bhyve_command.c | 121 ++++++++++++++++ src/bhyve/bhyve_parse_command.c | 90 ++++++++++++ src/conf/domain_audit.c | 5 + src/conf/domain_conf.c | 131 ++++++++++++++++++ src/conf/domain_conf.h | 29 +++-
For the conf stuff we'll need a docs update in docs/formatdomain.html.in at the same time as this parser additions. If you can provide a little detail in the commit message on why the current SCSI hostdev stuff doesn't work for FreeBSD that'd be useful too.
src/conf/virconftypes.h | 3 + src/qemu/qemu_command.c | 2 + src/qemu/qemu_domain.c | 5 + src/qemu/qemu_hostdev.c | 1 + src/qemu/qemu_hotplug.c | 2 + src/qemu/qemu_migration.c | 1 + src/security/security_apparmor.c | 1 + src/security/security_dac.c | 28 ++++ src/security/security_selinux.c | 8 ++ .../bhyveargv2xml-passthru.args | 8 ++ .../bhyveargv2xml-passthru.xml | 26 ++++ .../bhyveargv2xml-virtio-scsi.args | 9 ++ .../bhyveargv2xml-virtio-scsi.xml | 20 +++ tests/bhyveargv2xmltest.c | 2 + [> .../bhyvexml2argv-passthru.args | 11 ++ .../bhyvexml2argv-passthru.ldargs | 1 + .../bhyvexml2argv-passthru.xml | 22 +++ .../bhyvexml2argv-virtio-scsi.args | 9 ++ .../bhyvexml2argv-virtio-scsi.ldargs | 1 + .../bhyvexml2argv-virtio-scsi.xml | 21 +++ tests/bhyvexml2argvtest.c | 4 +- .../bhyvexml2xmlout-passthru.xml | 29 ++++ .../bhyvexml2xmlout-virtio-scsi.xml | 23 +++ tests/bhyvexml2xmltest.c | 2 + 32 files changed, 658 insertions(+), 2 deletions(-) create mode 100644 tests/bhyveargv2xmldata/bhyveargv2xml-passthru.args create mode 100644 tests/bhyveargv2xmldata/bhyveargv2xml-passthru.xml create mode 100644 tests/bhyveargv2xmldata/bhyveargv2xml-virtio-scsi.args create mode 100644 tests/bhyveargv2xmldata/bhyveargv2xml-virtio-scsi.xml create mode 100644 tests/bhyvexml2argvdata/bhyvexml2argv-passthru.args create mode 100644 tests/bhyvexml2argvdata/bhyvexml2argv-passthru.ldargs create mode 100644 tests/bhyvexml2argvdata/bhyvexml2argv-passthru.xml create mode 100644 tests/bhyvexml2argvdata/bhyvexml2argv-virtio-scsi.args create mode 100644 tests/bhyvexml2argvdata/bhyvexml2argv-virtio-scsi.ldargs create mode 100644 tests/bhyvexml2argvdata/bhyvexml2argv-virtio-scsi.xml create mode 100644 tests/bhyvexml2xmloutdata/bhyvexml2xmlout-passthru.xml create mode 100644 tests/bhyvexml2xmloutdata/bhyvexml2xmlout-virtio-scsi.xml
I didn't see anything in the code is keeping track of in-use PCI devices. Does something else in FreeBSD guarantee that you won't have bad stuff happening if the PCI device is attempted to be assigned to 2 guests ? Also is it required to manually detach the host OS driver first, or is that automatic ? This ties into which 'managed=no|yes' attribute choices you should permit for the hostdev. Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|