[Spice-devel] [RFC linux vdagent] Add POC for getting xrandr output from monitor ID

[Jonathon Jongsma]

On Tue, 2018-11-13 at 11:21 +0100, Lukáš Hrázký wrote:
> Hi,
> 
> On Tue, 2018-11-13 at 09:04 +0100, Gerd Hoffmann wrote:
> >   Hi,
> > 
> > > To look up a PCI device by address, I could use the /sys
> > > filesystem. So
> > > for pci/0000/02.0, I would translate the domain 0000 into the
> > > filesystem path '/sys/devices/pci0000:00/' assuming bus
> > > '00'?  And then
> > > look inside that directory for the the path '0000:00:02.0'. And
> > > then
> > > look for a path named 'drm' within that directory to determine
> > > whether
> > > it's a drm card or not? Is the existence of the 'drm' directory
> > > the
> > > thing that we should use to determine whether this is a drm
> > > card? 
> > 
> > Typically yes, except virtio-gpu which has an additional virtio
> > level,
> > i.e. /sys/devices/pci0000:00/0000:00:02.0/virtio<n>/drm.
> 
> Why the exception? Doesn't make this check look that great and
> reliable...
> 
> > > Or a slightly more complicated example (based on my laptop):
> > > Let's look
> > > for this device:
> > > 
> > > 02:00.0 3D controller: NVIDIA Corporation GM108M [GeForce 940MX]
> > > (rev
> > > a2)
> > > 
> > > So it's device 00.0 on bus 02. Here's the first part of lspci -t:
> > > -[0000:00]-+-00.0
> > >            +-01.0-[01]--
> > >            +-01.2-[02]----00.0
> > > 
> > > So it's attached to the PCI bridge at 01.2. Since the bus number
> > > ('02')
> > > is not a part of our device address format, we would only be
> > > provided
> > > the following address: pci/0000/01.2/00.0
> > > 
> > > From this we can look for the PCI bridge at path
> > > /sys/devices/pci0000:00/0000:00:01.2. Within this directory, we
> > > can
> > > look for a path of ... it's unclear because we're not provided
> > > the bus
> > > number, so we can't necessarily construct the path '0000:02:00.0'
> > > that
> > > represents this device. We could simply look for files within the
> > > directory that end with '00.0' and just assume that the one we
> > > find is
> > > the correct one. But that doesn't seem cleaner and more elegant
> > > to me.
> > 
> > You can translate the path into a sysfs shell globbing pattern,
> > then
> > feed that into glob (see "man glob").
> > 
> > For "pci/0000/01.2/00.0" the glob pattern would be
> > "/sys/devices/pci0000:*/*:01.2/*:00.0"
> 
> That's what I had in mind.
> 
> Cheers,
> Lukas

And you think that's a cleaner solution? Seems less elegant to me, to
be honest.

Jonathon