[LinuxBIOS] lpci_set_subsystem question

[Marc Jones]

I see a couple of different versions of lpci_set_subsystem function and 
I wonder why.

For example:
src\southbridge\broadcom\bcm5780\bcm5780_nic.c:
static void lpci_set_subsystem(device_t dev, unsigned vendor, unsigned 
device)
{
     pci_write_config32(dev, 0x40,
         ((device & 0xffff) << 16) | (vendor & 0xffff));
}

src\southbridge\amd\amd8111\amd8111_ide.c:
static void lpci_set_subsystem(device_t dev, unsigned vendor, unsigned 
device)
{
     pci_write_config32(dev, 0x70,
         ((device & 0xffff) << 16) | (vendor & 0xffff));
}

src\southbridge\broadcom\bcm5780\bcm5780_sb_pci_main.c:
static void lpci_set_subsystem(device_t dev, unsigned vendor, unsigned 
device)
{
         pci_write_config32(dev, 0x2c,
                 ((device & 0xffff) << 16) | (vendor & 0xffff));
}

Shouldn't the PCI config register always be 0x2c (subsystem ID register) 
like in the last example (bcm5780_sb_pci_main.c)? Should these functions 
even be there since there is a stock pci_dev_set_subsystem() that seems 
to be correct?
Am I mis-understanding this function?

Thanks,
Marc

[ron minnich]

On 8/15/07, Marc Jones  wrote:
> Shouldn't the PCI config register always be 0x2c (subsystem ID register)
> like in the last example (bcm5780_sb_pci_main.c)? Should these functions
> even be there since there is a stock pci_dev_set_subsystem() that seems
> to be correct?

IIRC it should be, but is not always. A lot of chips don't quite
follow the spec, and this weird out-of-band setting of registers crops
up. That's my take on it anyway.

ron

[Segher Boessenkool]

> > Shouldn't the PCI config register always be 0x2c (subsystem ID 
> > register)
> > like in the last example (bcm5780_sb_pci_main.c)? Should these 
> > functions
> > even be there since there is a stock pci_dev_set_subsystem() that 
> > seems
> > to be correct?
>
> IIRC it should be, but is not always. A lot of chips don't quite
> follow the spec, and this weird out-of-band setting of registers crops
> up. That's my take on it anyway.

The PCI spec only defines the "read" behaviour of the subsystem
config registers as far as I know.  Anything writing those regs
is just as non-standard (or moreso!) as something using special
regs is.

Segher

[Stefan Reinauer]

* Marc Jones  [070815 23:01]:
> Shouldn't the PCI config register always be 0x2c (subsystem ID register) 
> like in the last example (bcm5780_sb_pci_main.c)? Should these functions 
> even be there since there is a stock pci_dev_set_subsystem() that seems 
> to be correct?

I think on some systems 2c is always read-only, while on others it is
read/write or read-once. Those with read-only 2c have a writable
register that mirrors the value to 2c.

Not sure if that's the case here.

[Marc Jones]

Stefan Reinauer wrote:
> Marc Jones  [070815 23:01]:
> > Shouldn't the PCI config register always be 0x2c (subsystem ID register) 
> > like in the last example (bcm5780_sb_pci_main.c)? Should these functions 
> > even be there since there is a stock pci_dev_set_subsystem() that seems 
> > to be correct?
>
> I think on some systems 2c is always read-only, while on others it is
> read/write or read-once. Those with read-only 2c have a writable
> register that mirrors the value to 2c.
>
> Not sure if that's the case here.

Ah, That could be it. I will see if I have some specs for those devices 
that look odd.

Thanks,
Marc