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

fixed! Re: [Xen-devel] PCI bus scan failure on 3.0.0 testing

On Wed, Jan 25, 2006 at 11:15:08AM -0500, Ryan wrote:
> On Sat, 2006-01-21 at 20:45 +0000, Keir Fraser wrote:
> > Yep, if you can find your way around the Linux PCI code a bit, it is 
> > worth adding tracing to find out how it is finding buses 1 and 2. If 
> > you can do that then it should be possible to backtrack and find out 
> > where xenlinux is going wrong.
> > 
> >   -- Keir
> I think the reason Linux is finding PCI buses 1 and 2 is because Steve
> is using Xen 2.0 (which passes the bus numbers with PCI devices on them
> explicitly to the Linux kernel via a hypercall since Xen hides the
> bridges).
> Steve, I'm not sure there's a way to know from your output if Linux
> found the device you need for your hard drive or not (at least not
> without compiling in some debug output), but I'd try double checking
> that you've compiled in SCSI and the appropriate SCSI low-level driver
> for your hardware (I didn't see a "SCSI subsystem initialized" line in
> the log for the kernel that didn't work).

Ryan gets the prize for this one.  (For the record, and to help search
engines, this again is for Xen 3.0 and linux 2.6.12 on a Network
Engines machine, equivalent to an IBM Netfinity 4000R).  It turned out
I had two problems:

- I needed 'noapic' on the xen boot command line.  I had already tried
  this, but then:
- I also needed the SCSI aic7xxx driver.  Doh!  Bone-headed mistake,
  excused in part because, while xen0 builds the driver in, the
  full-sized 'make KERNELS=linux-2.6-xen world' kernel builds it as a
  module instead, and because I was distracted by boot options I
  hadn't noticed the missing driver nor built an initrd. 
I finally figured this out when I went back to using a 'make dist'
xen0 kernel for 3.0 the way I always used to do for xen 2 -- suddenly
things started working, mostly, and stabilized when I added 'noapic'.
The xen0 kernels work great and build in 20 minutes on these machines;
I should have just stuck to that in the first place (the 'world'
kernel takes 4 hours to build).

Problems I *didn't* have were:

- Anything having to do with PCI bus scanning, nosmp, or acpi.  As
  Ryan says, the PCI scanning output differences are just an artifact
  of normal differences between Xen 2 and Xen 3.  These aren't the
  droids you're looking for, move along...

Thanks All,

Stephen G. Traugott  (KG6HDQ)
UNIX/Linux Infrastructure Architect, TerraLuna LLC
http://www.stevegt.com -- http://Infrastructures.Org

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.