Re: [ACPI] Re: [PATCH] PNPACPI: handle address descriptors in _PRS

[Shaohua Li <sha...@in...>]

On Thu, 2005-09-22 at 17:48 -0600, Bjorn Helgaas wrote:
> On Wednesday 21 September 2005 6:49 pm, Shaohua Li wrote:
> > On Wed, 2005-09-21 at 23:27 +0800, Bjorn Helgaas wrote:
> > > On Tuesday 20 September 2005 7:17 pm, Shaohua Li wrote: 
> > > > On Tue, 2005-09-20 at 16:26 -0600, Bjorn Helgaas wrote: 
> > > > > PNPACPI complained about and ignored devices with ADDRESS16,
> > > ADDRESS32, 
> > > > > or ADDRESS64 descriptors in _PRS.  Those descriptors seem
> > > legitimate, 
> > > > > and HP firmware uses them for built-in serial ports, so this patch
> > > adds 
> > > > > support for them.
> > > 
> > > > You might have missed the similar change in pnpacpi_encode_resources
> > > in 
> > > > your patch. It's used when we assign new resources (_SRS method) to
> > > pnp 
> > > > devices.
> > > 
> > > You're right, I did miss the corresponding encoding change.  I'll 
> > > add that.  I feel a bit uneasy about it, though, because it will be 
> > > a fair amount of code, and I don't know how to test it.  Do we ever 
> > > really use _SRS today?  And there's information lost when decoding 
> > > the ACPI resources, such as the _TRA.
> > We did use _SRS, at least under x86. This only is done when some PNP
> > devices havn't resources, so only very small number of systems need it.
> > Ignoring it might be ok currently, if HP firmware always assigns
> > resources to PNP devices. Yes, we did lose the info of _TRA. I didn't
> > find any system uses it. Maybe some big IA64 systems use.
> 
> HP ia64 systems definitely use _TRA.  I would expect most systems
> with multiple PCI root bridges would use _TRA, because if you
> don't, the amount of 32-bit PCI bus address space you can use
> is severely limited.
> 
> HP firmware does assign resources to the PNP devices (typically
> they're things hardwired onto the motherboard at fixed addresses).
> So we don't actually need the encoding changes at the moment.
> 
> I'm inclined to just leave the encoding out for now.  That
> way, we don't carry around the untested code, and if somebody
> eventually tries to use it, the attempt to set resources will
> fail, and we can write the code and test it at that time.
Ok, lets wait till it actually is used.

Thanks,
Shaohua