Re: [PATCH 3/5] Float the pci bus number on MPC8641HPCN board.

From: Matt Sealey <matt@genesi-usa•com>
To: Benjamin Herrenschmidt <benh@kernel•crashing.org>
Cc: linuxppc-dev@ozlabs•org, Paul Mackerras <paulus@samba•org>,
	Zhang Wei-r63237 <Wei.Zhang@freescale•com>
Subject: Re: [PATCH 3/5] Float the pci bus number on MPC8641HPCN board.
Date: Fri, 01 Jun 2007 17:35:44 +0100	[thread overview]
Message-ID: <46604AE0.5040700@genesi-usa.com> (raw)
In-Reply-To: <1180683854.19517.419.camel@localhost.localdomain>

Re enabling proper domain support on 32-bit... now, now, now please!

X guys won't merge in code, we've been punished by both sides on Pegasos
for nearly 12 months now after there's been a bunch of misguided attempts
to 'fix' domain support in X. The real fix is in the kernel to make it
very clear that domains and proper, per-domain bus numbering (not
global bus numbering) and suchlike is the standard like it is on SPARC
and PPC64 and IA64.

If ppc32 suddenly goes the way of these, X guys will fix it, there will
be patches to work against this kernel, even if they have to detect the
kernel version to do it or look at a procfs or sysfs entry to fix up
their dumb pci scanning code.

I don't think stalling on it "because of X" is right. Fix Linux, X will
follow, because it really has to.

-- 
Matt Sealey <matt@genesi-usa•com>
Genesi, Manager, Developer Relations

Benjamin Herrenschmidt wrote:
> On Fri, 2007-06-01 at 14:45 +0800, Zhang Wei-r63237 wrote:
>> There are lots of lagacy PCI codes in kernel, such as
>> pci_bus_to_hose(busnr). If two hose's bus number have overlaps,
>> pci_bus_to_hose() can not return the correct hose. And there will
>> effect
>> /proc/pci and lspci tools.
> 
> More specifically...
> 
> Overlapping bus numbers on separate domains work... on 64 bits :-) The
> 32 bits PCI code assumes non-overlapping numbers. The main reason for
> that historically was that back then, domains didn't work well.
> 
> The second reasons why I didn't actually fix it is that turning on full
> domain support had (and probably still has) the nasty habit of breaking
> the X server :-( X is full of most horrible hacks for PCI and it will be
> terminally broken until they finally merge the pci-rework branch that
> Ian Romanick's been working on for some time now.
> 
> One of these days, one of us will finally tackle merging 32 and 64 bits
> PCI code (and fix both of them while at it to better use the generic
> code, which involves fixing issues in the said generic code too :-) at
> which point I'll enable full domain support for 32 bits as well, with an
> option to degrade it to bus number remapping.
> 
> Ben.
> 
> 
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@ozlabs•org
> https://ozlabs.org/mailman/listinfo/linuxppc-dev