On 2015/06/16 21:27, Eric Blake wrote: > On 05/28/2015 10:42 PM, Tetsuya Mukawa wrote: >> This patch adds below '-net' options to let QEMU know which features >> vhost-user backend will support. > [meta-comment: when posting a new revision of a series, do it as a new > top-level thread rather than buried in-reply-to an earlier version]
Hi Eric, Thanks for comments. Yes, I will follow like above in next patches. >> If above features are specified, QEMU assumes vhost-user backend supports >> the features, then QEMU can start without vhost-user backend connection. >> (While no connection, link status of virtio-net device will be down) >> >> When connection between QEMU and the backend is established, QEMU checkes >> feature > s/checkes/checks/ > >> values of the backend to make sure the expected features are provided. >> If it doesn't, the connection will be closed by QEMU. >> >> Signed-off-by: Tetsuya Mukawa <muk...@igel.co.jp> >> --- >> +++ b/qapi-schema.json >> @@ -2241,25 +2241,29 @@ >> # >> # @vhostforce: #optional vhost on for non-MSIX virtio guests >> # >> +# @backend_features: #optional feature flag to support vhost user backend >> +# (Since 2.4) >> +# > s/backend_features/backend-features/ > > (we document that qapi should prefer - over _ unless there is a > consistency issue with _ already used in the struct) > > >> @@ -2444,12 +2448,92 @@ >> # >> # @vhostforce: #optional vhost on for non-MSIX virtio guests (default: >> false). >> # >> +# @backend_csum: #optional feature backend supports (default: false). >> +# (Since 2.4) > Again, s/_/-/ throughout this struct. I will use '-' instead of '_'. Also will fix typo. >> +# >> +# @backend_mq: #optional feature backend supports (default: false). >> +# (Since 2.4) > A lot of identical descriptions. Might be worth more specific text to > each option; for example, > @backend_mq: #optional: True if backend supports multiqueue feature > (default: false). (Since 2.4) > Sure I will add short descriptions, if we use this kind of command options. >> +++ b/qemu-options.hx >> @@ -1467,6 +1467,15 @@ DEF("netdev", HAS_ARG, QEMU_OPTION_netdev, >> "-netdev >> tap,id=str[,fd=h][,fds=x:y:...:z][,ifname=name][,script=file][,downscript=dfile]\n" >> " >> [,helper=helper][,sndbuf=nbytes][,vnet_hdr=on|off][,vhost=on|off]\n" >> " >> [,vhostfd=h][,vhostfds=x:y:...:z][,vhostforce=on|off][,queues=n]\n" >> + " >> [,backend_csum=on|off][,backend_guest_csum=on|off][,backend_gso=on|off]\n" >> + " [,backend_guest_tso4=on|off][,backend_guest_tso6=on|off]\n" >> + " [,backend_guest_ecn=on|off][,backend_guest_ufo=on|off]\n" >> + " [,backend_guest_announce=on|off][,backend_host_tso4=on|off]\n" >> + " [,backend_host_tso6=on|off][,backend_host_ecn=on|off]\n" >> + " >> [,backend_host_ufo=on|off][,backend_mrg_rxbuf=on|off][,backend_status=on|off]\n" >> + " >> [,backend_ctrl_vq=on|off][,backend_ctrl_vx=on|off][,backend_ctrl_vlan=on|off]\n" >> + " >> [,backend_ctrl_rx_extra=on|off][,backend_ctrl_mac_addr=on|off]\n" >> + " [,backend_ctrl_guest_offloads=on|off][,backend_mq=on|off]\n" >> " configure a host TAP network backend with ID 'str'\n" >> " use network scripts 'file' (default=" >> DEFAULT_NETWORK_SCRIPT ")\n" >> " to configure it and 'dfile' (default=" >> DEFAULT_NETWORK_DOWN_SCRIPT ")\n" >> @@ -1486,6 +1495,7 @@ DEF("netdev", HAS_ARG, QEMU_OPTION_netdev, >> " use 'vhostfd=h' to connect to an already opened vhost >> net device\n" >> " use 'vhostfds=x:y:...:z to connect to multiple already >> opened vhost net devices\n" >> " use 'queues=n' to specify the number of queues to be >> created for multiqueue TAP\n" >> + " use 'backend_*=on' to specify virtio-net feature that >> vhost-user backend supports\n" > Bikeshedding: Rather than having a lot of booleans, I wonder if it would > be better to have an array of flag names. That is, in QMP form, this > would select three features (and leave the others off) > > 'backend-features':[ 'csum', 'guest-csum', gso' ] > > although I'm not sure on how that would best translate to the command line. > One more suggestion is here. https://lists.gnu.org/archive/html/qemu-devel/2015-06/msg03807.html I am also not sure what is best. But I guess above explanation is reasonable at this time. When we find a case that user needs to describe or change backend-features frequently by hand, expand backend-features and add these human readable options. Regards, Tetsuya