pci32 code - early_*config*

pci32 code - early_*config*

[Kumar Gala]

Ben,

I was looking at what I need to get some additional bits of PCI code  
building w/your ppc64 booke patches.  One thing it looks like we need  
is the early config cycle code.  The question I have is do we think we  
still need the null_ops support?  Also do we think we every get called  
with a NULL hose?

This is fake_pci_bus() in arch/powerpc/kernel/pci_32.c

- k

Re: pci32 code - early_*config*

[Benjamin Herrenschmidt]

On Wed, 2009-04-15 at 01:17 -0500, Kumar Gala wrote:
> Ben,
> 
> I was looking at what I need to get some additional bits of PCI code  
> building w/your ppc64 booke patches.  One thing it looks like we need  
> is the early config cycle code.  The question I have is do we think we  
> still need the null_ops support?  Also do we think we every get called  
> with a NULL hose?
> 
> This is fake_pci_bus() in arch/powerpc/kernel/pci_32.c

Hrm.

I'm not fan of the early config stuff, remind me why we need it ?

Ben.

Re: pci32 code - early_*config*

[Roderick Colenbrander]

On Fri, Apr 17, 2009 at 9:25 AM, Benjamin Herrenschmidt
<benh@kernel.crashing.org> wrote:
> On Wed, 2009-04-15 at 01:17 -0500, Kumar Gala wrote:
>> Ben,
>>
>> I was looking at what I need to get some additional bits of PCI code
>> building w/your ppc64 booke patches. =A0One thing it looks like we need
>> is the early config cycle code. =A0The question I have is do we think we
>> still need the null_ops support? =A0Also do we think we every get called
>> with a NULL hose?
>>
>> This is fake_pci_bus() in arch/powerpc/kernel/pci_32.c
>
> Hrm.
>
> I'm not fan of the early config stuff, remind me why we need it ?
>
> Ben.
>
>
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-dev
>

There are cases when it is really needed (or you would have to access
the address and data bus by hand). In my case I'm using a Xilinx PLB
soft-core and if certain options in the configuration header aren't
set (which aren't set by default) the soft-core is basically disabled.
Cases like this look like valid cases in which early config should be
used.

Roderick Colenbrander

Re: pci32 code - early_*config*

[Benjamin Herrenschmidt]

On Fri, 2009-04-17 at 10:03 +0200, Roderick Colenbrander wrote:
> 
> There are cases when it is really needed (or you would have to access
> the address and data bus by hand). In my case I'm using a Xilinx PLB
> soft-core and if certain options in the configuration header aren't
> set (which aren't set by default) the soft-core is basically disabled.
> Cases like this look like valid cases in which early config should be
> used.

I still wonder whether we could just setup the pci_controller data
structure and use "normal" PCI config access routines...

Cheers,
Ben.

Re: pci32 code - early_*config*

[Kumar Gala]

On Apr 17, 2009, at 3:33 AM, Benjamin Herrenschmidt wrote:

> On Fri, 2009-04-17 at 10:03 +0200, Roderick Colenbrander wrote:
>>
>> There are cases when it is really needed (or you would have to access
>> the address and data bus by hand). In my case I'm using a Xilinx PLB
>> soft-core and if certain options in the configuration header aren't
>> set (which aren't set by default) the soft-core is basically  
>> disabled.
>> Cases like this look like valid cases in which early config should be
>> used.
>
> I still wonder whether we could just setup the pci_controller data
> structure and use "normal" PCI config access routines...

The problem is that the "normal" PCI config access routines need a  
pci_bus.  However we don't have one yet and we need to do PCI config  
cycles to the PHB before we call the remainder of the setup code to  
get one.

We could all call our own ops directly and provide a fake bus but this  
is pretty much what the early routines do for us in a common way.

- k

Re: pci32 code - early_*config*

[Benjamin Herrenschmidt]

On Fri, 2009-04-17 at 09:08 -0500, Kumar Gala wrote:
> The problem is that the "normal" PCI config access routines need a  
> pci_bus.  However we don't have one yet and we need to do PCI config  
> cycles to the PHB before we call the remainder of the setup code to  
> get one.
> 
> We could all call our own ops directly and provide a fake bus but
> this  
> is pretty much what the early routines do for us in a common way.

Ah right, slipped out of my mind. Oh well, easy to make the code
common, move it to pci-common.c

Cheers,
Ben.

Re: pci32 code - early_*config*

[Kumar Gala]

On Apr 17, 2009, at 12:05 PM, Benjamin Herrenschmidt wrote:

> On Fri, 2009-04-17 at 09:08 -0500, Kumar Gala wrote:
>> The problem is that the "normal" PCI config access routines need a
>> pci_bus.  However we don't have one yet and we need to do PCI config
>> cycles to the PHB before we call the remainder of the setup code to
>> get one.
>>
>> We could all call our own ops directly and provide a fake bus but
>> this
>> is pretty much what the early routines do for us in a common way.
>
> Ah right, slipped out of my mind. Oh well, easy to make the code
> common, move it to pci-common.c

Right, I can do that... In doing so I was wondering two things w/ 
regards to fake_pci_bus:

1. can we drop the hose == 0 check, it looks like there are 3 users of  
the early pci ops code (fsl, 4xx, pmac).  The FSL one I can say passes  
a valid hose in and it looks like 4xx does as well.  The pmac code is  
always a bit more trick so I figured you might know

2. Do we believe those hose->ops is always valid -- again FSL & 4xx it  
is, pmac??

static struct pci_bus *
fake_pci_bus(struct pci_controller *hose, int busnr)
{
         static struct pci_bus bus;

         if (hose == 0) {
                 hose = pci_bus_to_hose(busnr);
                 if (hose == 0)
                         printk(KERN_ERR "Can't find hose for PCI bus  
%d!\n", busnr);
         }
         bus.number = busnr;
         bus.sysdata = hose;
         bus.ops = hose? hose->ops: &null_pci_ops;
         return &bus;
}