[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [RFC] block: Is name of BlockBackend deprecated with -b
From: |
Kevin Wolf |
Subject: |
Re: [Qemu-devel] [RFC] block: Is name of BlockBackend deprecated with -blockdev parameter? |
Date: |
Tue, 11 Dec 2018 10:47:09 +0100 |
User-agent: |
Mutt/1.10.1 (2018-07-13) |
Am 11.12.2018 um 08:28 hat Markus Armbruster geschrieben:
> I figure Kevin knows, but you typoed his e-mail address. I fixed it for
> you.
>
> Anton Kuchin <address@hidden> writes:
>
> > Hello,
> >
> > I'm trying to switch from -drive parameter to -blockdev + -device and
> > having problems. Looks like with this option I have no way to set the
> > name of created BlockBackend, but some QMP and HMP commands are
> > trying to find blk with blk_by_name() and fail to locate my device
> > (e.g. hmp_commit, qmp_x_bloc_latency_histogram_set ...). Was it
> > intentional and BB names are going to be deprecated?
> >
> > This also seems to be a the case for all block devices hotplugged with
> > QMP as they use the same code path.
> >
> > As far as I understand all named backends are stored in
> > monitor_block_backends list, but I can't get what is the point of
> > having this list, and why parse_drive() function doesn't call
> > monitor_add_blk() like blockdev_init() does with -drive option. Can
> > someone give me a hint on this?
> >
> > I also noticed that some commands fallback to search by qdev_id or
> > BDS->node_name, but at the creation time (both in
> > bdrv_assing_node_name and monitor_add_blk) it is already checked that
> > names are unique across these namespaces so may be it would be useful
> > to introduce generic search function?
Yes, BlockBackend names are not supposed to be used in the external
interfaces any more. If the QMP command is about the backend, it will
take a node name, and if it's about the guest device (which may not even
have a node attached with removable media), it will take a qdev ID.
The implementation of qmp_x_block_latency_histogram_set() is incorrect,
it shouldn't use blk_by_name(). I wonder where it got that pattern from
because it was added long after all other QMP commands had switched to
qmp_get_blk() or bdrv_lookup_bs().
hmp_commit() should indeed use bdrv_lookup_bs() to also accept node
names. I think we reviewed QMP to make sure we converted everything
and forgot about HMP commands that aren't mapped to QMP.
Kevin