Re: [Qemu-devel] [PATCH v3 1/2] qapi: move to QOM path for x-block-latency-histogram-set

[Kevin Wolf]

Am 11.02.2019 um 18:39 hat Vladimir Sementsov-Ogievskiy geschrieben:
> 08.01.2019 16:20, Markus Armbruster wrote:
> > Vladimir Sementsov-Ogievskiy <address@hidden> writes:
> > 
> >> Move to way of device selecting, however fall back to device name if
> >> path is not found.
> >>
> >> Signed-off-by: Vladimir Sementsov-Ogievskiy <address@hidden>
> >> ---
> >>   qapi/block-core.json |  4 ++--
> >>   blockdev.c           | 22 +++++++++++++++-------
> >>   2 files changed, 17 insertions(+), 9 deletions(-)
> >>
> >> diff --git a/qapi/block-core.json b/qapi/block-core.json
> >> index 762000f31f..bb70c51a57 100644
> >> --- a/qapi/block-core.json
> >> +++ b/qapi/block-core.json
> >> @@ -489,7 +489,7 @@
> >>   # If only @device parameter is specified, remove all present latency 
> >> histograms
> >>   # for the device. Otherwise, add/reset some of (or all) latency 
> >> histograms.
> >>   #
> >> -# @device: device name to set latency histogram for.
> >> +# @id: The QOM path or name of the guest device.
> >>   #
> >>   # @boundaries: list of interval boundary values (see description in
> >>   #              BlockLatencyHistogramInfo definition). If specified, all
> > 
> > Is such overloaded semantics what we want in new interfaces?
> > 
> > Hmm, looks like there's ample precedence for it.  Escaped my grep at
> > first because its commonly documented as "The name or QOM path of the
> > guest device".  Suggest to stick to that for consistency.
> 
> 
> Interesting, that in cases you mean, documentation seems wrong:
> it goes through qmp_get_blk, which works like @id may be only QOM path, not 
> name,
> so for the it should be @id: The QOM path.

It's really a QOM path relative to /machine/peripheral (see
find_device_state()), which is where named devices live, accessible
through their id. So relative paths are both QOM paths and names of
guest devices. (Relative paths aren't a QOM concept, though, which
provides only absolute and partial paths. The relative paths have a
one-off implementation here.)

So in the end, I think the description is actually correct, just with a
higher level perspective, ignoring all the low-level details.

Kevin