[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: VIO on sparc / udev support



On Mon, May 13, 2013 at 04:17:40PM +0200, Michael Biebl wrote:
> Am 13.05.2013 13:45, schrieb Julien Cristau:
> > Michael,
> > 
> > On Mon, May 13, 2013 at 02:51:51 +0200, Michael Biebl wrote:
> > 
> >>> Since no-one in the systemd/udev team is using SPARC hardware, I was
> >>> wondering, if vio_type would have a better home in e.g. the sparc-utils
> >>> [2] package, where it would be maintained and tested by people with
> >>> knowledge of that architecture.
> > 
> > I don't think that would make much sense, and this belongs with the
> > related utils in udev IMO.
> 
> Care to elaborate? That helper is highly sparc specific and it's an odd
> one, which doesn't really fit into udev. E.g. it's built on all
> architectures, then thrown away later during install. It's just weird.
> 
> Take s390 as an example: The s390 specific helpers and udev rules are
> shipped in the s390-tools package and I think it would make a lot of
> sense to do the same for sparc. Even if it would only be for
> consistencies sake.
> 
> >>> Seeing a bug report like [3], it's not actually clear to me, if vio_type
> >>> is still functional with newer kernel releases and for which type of
> >>> hardware this is actually required. I poked Marco about this, but he
> >>> doesn't remember the details anymore.
> >>>
> > 
> > Your [3] points at an udev packaging bug.  How does it relate to "newer
> > kernel releases"?  The hardware for which this is required was described
> > in #526621, so I don't really get that question either...
> 
> Well, maybe the kernel got autoloading for that those types of devices,
> who knows. Something similar happened for e.g. cpufreq drivers.
> So I think this question is justified, if this helper is still required
> today and for what setups.

Michael, sorry for not replying sooner.

For context, these are device drivers which support Debian 
installation into Solaris logical domains (LDOMs), which are a kind of 
virtual machine. LDOMs export an API for network and disk access, and 
host operating system should have appropriate drivers for them to 
work, sunvnet and sunvdc kernel modules implement this functionality.

Back in 2012 I got access to a Solaris box which allowed me to 
experiment with Debian installation into an LDOM. I found that 
this setup is pretty much unusable, as I ended up with disk corruption 
in the LDOM every time I completed the installation. I reported it 
upstream [0] and David Miller replied [1]:

"I haven't tested LDOM in at least 2 years, it's very likely bugs have 
crept in."

I haven't heard from anyone attempting to install into LDOMs for 
about that long, not counting my attempts. Given that it's an 
extremely rare use case and I don't think that the drivers are going 
to be fixed any time soon (if ever), my suggestion is not to spend too 
much time trying to support their autoloading. If they are causing 
additional maintenance burden, under current circumstances I would 
consider it reasonable to simply drop vio_type and related 
infrastructure from udev. After all, these drivers are never required 
for booting the installer, so they can be loaded manually from 
installer interface if needed.

[0] http://marc.info/?l=linux-sparc&m=134839643929797&w=2
[1] http://marc.info/?l=linux-sparc&m=134841688520005&w=2

Best regards,
-- 
Jurij Smakov                                           jurij@wooyd.org
Key: http://www.wooyd.org/pgpkey/                      KeyID: C99E03CC


Reply to: