[Qemu-devel] Re: [PATCH] qemu PPC Linux 2.6 IDE

[J. Mayer]

On Wed, 2005-02-23 at 23:28, Thayne Harbaugh wrote:
> I have qemu PPC booting Linux 2.6.  Here's a patch for IDE.
> 
> IDE detection of macio has changed significantly.  2.4
> drivers/ide/ppc/pmac.c:pmac_ide_probe() simply looked for the IDE/ATA
> controllers in OF.  2.6 pmac.c:pmace_ide_probe() registers macio device
> filters for IDE/ATA and relies on macio device enumeration to connect to
> the IDE driver.  Where things are broken is that
> drivers/macintosh/macio_asic.c:macio_pci_probe() fails to match the
> macio device due to the bus:devfn not being set as a "reg" property in
> OHW.

OK, your patch is as false as what I've done for many reasons (see
below)

> 
> There are two questions about this patch:
> 
> * What is the complete information expected in the "reg" property of a
> PCI device?

"reg" is supposed to contain information about the configuration space,
I/O regions and memory regions for the PCI device.

> * Where should the current macio "reg" information really be stored, if
> at all?

Current MacIO "reg" informations are to be stored in "reg" property, but
there are false, as you've seen. Those "reg" properties where taken from
a real mac...
I did not notice that I should have changed the preg.hi cell to match
the Qemu PCI device informations, but the real fix is to generate the
whole information...

I made a quick fix, following OF bindings for PCI specifications.
Note that I already had a fix for OF_property_set with no value, less
intrusive than yours.
I also did add a dprintf (as debug printf) function to make early debug
easier and there's also a needed fix in CHRP boot file loader.

Please consider this patch against 0.4-pre1d, it should fix the macIO
problem the right way.
There's still a lack of informations about configuration space in reg
properties for PCI devices, but this should not hurt.

-- 
J. Mayer <address@hidden>
Never organized

current.diff
Description: Text Data