On 03/20/2015 08:19 AM, Markus Armbruster wrote:

...

If (a working version of) this makes it in 2.3, libvirt WILL use it in the next release. It will take me less than 5 minutes to write up the libvirt patch, as long as the new option is advertised via query-command-line-options (which means that QMP introspection of the new option is a must for v2 :)

The real libvirt requirement is that the new option is accessible via _some_ QMP command (doesn't have to be query-command-line-options, but that is the most obvious).

query-command-line-options covers only options using QemuOpts. Fixing that defect isn't in the cards for 2.3, which means I'll have to implement it with QemuOpts.

Makes sense, and is certainly easier than figuring out an alternative QMP probing mechanism.

Ways to do that:

* Stick it into an existing QemuOpts option. Is there one that fits? --machine doesn't really fit.

Mostly agree; --machine is more for what the guest sees, while this has no impact on the guest ABI. On the other hand, things like -machine accel=[tcg|kvm] aren't necessarily guest visible either, so -machine has tended to be a catch-all for other things. But that doesn't mean it should continue to be one.

* Create a new QemuOpts option for miscellaneous settings. Would --misc format-probing=off be too ugly? Got a better name than --misc?

The only clients using the new option will be machine generated, so I don't think ugly appearance is a show-stopper. And -misc really does sound like it is the most amenable to adding other random flags in the future, so I can live with that name. Unless anyone else has a better idea, '-misc format-probing=off' has my vote and works for libvirt; it is only a two-line change to my libvirt RFC patch: diff --git i/src/qemu/qemu_capabilities.c w/src/qemu/qemu_capabilities.c index b452a75..f7a46c1 100644 --- i/src/qemu/qemu_capabilities.c +++ w/src/qemu/qemu_capabilities.c @@ -2517,7 +2517,7 @@ static struct virQEMUCapsCommandLineProps virQEMUCapsCommandLine[] = { { "msg", "timestamp", QEMU_CAPS_MSG_TIMESTAMP }, { "numa", NULL, QEMU_CAPS_NUMA }, { "drive", "throttling.bps-total-max", QEMU_CAPS_DRIVE_IOTUNE_MAX }, - { "no-format-probing", NULL, QEMU_CAPS_NO_FORMAT_PROBING }, + { "misc", "format-probing", QEMU_CAPS_NO_FORMAT_PROBING }, }; static int diff --git i/src/qemu/qemu_command.c w/src/qemu/qemu_command.c index 1085639..08f6560 100644 --- i/src/qemu/qemu_command.c +++ w/src/qemu/qemu_command.c @@ -8674,7 +8674,7 @@ qemuBuildCommandLine(virConnectPtr conn, } if (virQEMUCapsGet(qemuCaps, QEMU_CAPS_NO_FORMAT_PROBING) && !cfg->allowDiskFormatProbing) - virCommandAddArg(cmd, "-no-format-probing"); + virCommandAddArgList(cmd, "-misc", "format-probing=off", NULL); /* Serial graphics adapter */ if (def->os.bios.useserial == VIR_TRISTATE_BOOL_YES) {

Existing miscellaneous non-QemeOpts options could then (in 2.4!) become sugar for something in this option group, thus become available with -readconfig.

Nice way to plan for future cleanups. I also agree that while this one option about format-probing is reasonable to get into 2.3, that any other sugar cleanups are better deferred until after the freeze. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org