Re: Patch to support drivers compiled into the kernel

[Dan Williams <dcbw redhat com>]

On Tue, 2009-03-03 at 13:45 -0500, Bill C Riemers wrote:
> Hmm.  2.6.22 is not very old.   It is the most recent "stable" coLinux
> version.  I certainly hope we don't decide this is a driver problem...,
> as that will be much more of a pain to try and get fixed.
> I wouldn't expect to find to find a virtual driver using the PCI bus. 
> In fact, since Windows has control of the PCI bus, I would not be
> surprised if there are no PCI bus drivers for coLinux.
> 
> [root localhost ~]# ls -la /sys/bus/pci/drivers
> ls: cannot access /sys/bus/pci/drivers: No such file or directory
> [root localhost ~]# uname -a
> Linux localhost.localdomain 2.6.22.18-co-0.7.3 #1 PREEMPT Sat May 24
> 22:27:30 UTC 2008 i686 i686 i386 GNU/Linux
> 
> As near as I can tell NetworkManager never actually uses the driver
> information, except in print statements.  So setting a value like
> "<unknown>" or "<kernel>" is probably better than ignoring the network
> device.

The driver gets used in a number of places for quirks.  It's also a
mechanism to filter crappy drivers that don't set up sysfs correctly,
and to ensure that the device's information can be properly obtained by
applets, which need things like PCI/USB/PCMCIA device and vendor IDs.
If the driver isn't properly setting up the device link in sysfs, then
the driver needs to be fixed because it's a bug in the driver.  We don't
improve the whole situation by working around the bugs, we fix the bugs
at their source...

Dan

> But then again, I haven't looked at the 0.7.1 code.  Maybe it is more
> important to know the correct driver in the newer branch of the code?
> 
> Note: I've also seen this same symptoms qemu, but only with some
> configurations.  I don't have that environment setup right now to see if
> this same solution works.
> 
> Bill
> 
> 
> Kay Sievers wrote:
> > On Tue, Mar 3, 2009 at 19:20, Bill C Riemers <briemers redhat com> wrote:
> >   
> >> I'm sorry, it seems the list manager did not deliver all the replies Kay
> >> Sievers quoted...
> >>     
> >
> >   
> >> 13: udi = '/org/freedesktop/Hal/devices/net_00_ff_d3_da_b0_00'
> >>  net.interface = 'eth0'  (string)
> >>  linux.sysfs_path = '/sys/class/net/eth0'  (string)
> >>  net.address = '00:ff:d3:da:b0:00'  (string)
> >>  info.parent = '/org/freedesktop/Hal/devices/computer'  (string)
> >>     
> >
> > Seems the driver does not set its parent device properly in the
> > kernel. That nothing to fix in HAL or NM, it's a rather trivial kernel
> > driver fix. What's the driver that creates your eth0 interface? It's
> > usually one of that list:
> >   ls -l /sys/bus/pci/drivers/
> >
> > Also note, no recent kernel should set CONFIG_SYSFS_DEPRECATED=y.
> > Recent udev/HAL does not supports that old sysfs format properly
> > anymore. You should not see any "/sys/class/..." path in HAL, they
> > should all start with "/sys/devices/...".
> >
> > Kay
> >   
>