More device cleanup

[Myles Watson]

This patch continues the device code cleanup.

The largest changes are to get_pci_bridge_ops, and related changes to make
it
compile and use correct declarations.  I'd appreciate someone else checking
my const usage for sanity.

While I was doing that I moved the checks for CONFIG_<BUS>_PLUGIN_SUPPORT to
the Makefile.

The only functional difference is a possible NULL dereference in a debug
statement.

Signed-off-by: Myles Watson

Thanks,
Myles

[Myles Watson]

On Wed, Nov 5, 2008 at 12:27 PM, Myles Watson  wrote:
> This patch continues the device code cleanup.

I forgot the include/device/ changes in the first patch.

Thanks,
Myles
> The largest changes are to get_pci_bridge_ops, and related changes to make
> it
> compile and use correct declarations.  I'd appreciate someone else checking
> my const usage for sanity.
>
> While I was doing that I moved the checks for CONFIG_<BUS>_PLUGIN_SUPPORT
> to
> the Makefile.
>
> The only functional difference is a possible NULL dereference in a debug
> statement.
>
> Signed-off-by: Myles Watson
>
> Thanks,
> Myles

[ron minnich]

Are you testing these on qemu I assume?

ron

[Myles Watson]

> Are you testing these on qemu I assume?

I can.  I don't think I've made any functional changes, but it's probably
good to check.

Most of this patch isn't exercised by qemu, though.  There are no bridges
which are not in the dts.

Thanks,
Myles

[Myles Watson]

On Wed, Nov 5, 2008 at 1:24 PM, Myles Watson  wrote:
> > Are you testing these on qemu I assume?
>
> I can.  I don't think I've made any functional changes, but it's probably
> good to check.
>
> Most of this patch isn't exercised by qemu, though.  There are no bridges
> which are not in the dts.

It turns out that I forgot the importance of make clean for build testing.
I wish I knew better how to fix the dependencies.

This breaks the build now that I made the change for the pci_ops structs to
be const.
> From pci_ops_auto.c:

const struct pci_bus_operations *pci_check_direct(void)
{
    unsigned int tmp;

/*
     * Check if configuration type 1 works.
     */
    {
        outb(0x01, 0xCFB);
        tmp = inl(0xCF8);
        outl(0x80000000, 0xCF8);
        if ((inl(0xCF8) == 0x80000000) &&
            pci_sanity_check(&pci_cf8_conf1))
        {
            outl(tmp, 0xCF8);
            printk(BIOS_DEBUG, "PCI: Using configuration type 1\n");
            return &pci_cf8_conf1;
        }
        outl(tmp, 0xCF8);
    }

die("pci_check_direct failed\n");
    return NULL;
}

/** Set the method to be used for PCI, type I or type II
 */
void pci_set_method(struct device * dev)
{
    printk(BIOS_INFO, "Finding PCI configuration type.\n");
    dev->ops->ops_pci_bus = pci_check_direct();
    post_code(POST_STAGE2_PHASE2_PCI_SET_METHOD);
}

Here are the interesting things:
1. pci_check_direct either returns pci_cfg8_conf1 or dies
2. This code is only used by geodelx

I think the const struct is the right way to do this.  Can we change these
functions?

Two solutions:
1. Set the entire ops
2. Initialize it correctly and die if the check fails

Comments?

Thanks,
Myles

[ron minnich]

> /** Set the method to be used for PCI, type I or type II
>  */
> void pci_set_method(struct device * dev)
> {
>     printk(BIOS_INFO, "Finding PCI configuration type.\n");
>     dev->ops->ops_pci_bus = pci_check_direct();
>     post_code(POST_STAGE2_PHASE2_PCI_SET_METHOD);
> }

yeah, just set it. This is historical from when we supported type i or
type ii, but nobody uses type ii

ron

[Myles Watson]

On Wed, Nov 5, 2008 at 3:42 PM, ron minnich  wrote:
> > /** Set the method to be used for PCI, type I or type II
> >  */
> > void pci_set_method(struct device * dev)
> > {
> >     printk(BIOS_INFO, "Finding PCI configuration type.\n");
> >     dev->ops->ops_pci_bus = pci_check_direct();
> >     post_code(POST_STAGE2_PHASE2_PCI_SET_METHOD);
> > }
>
> yeah, just set it. This is historical from when we supported type i or
> type ii, but nobody uses type ii

I left the infrastructure there so that it could be easily used for
memory-mapped checking.  Otherwise the whole file needs to die.

I also changed the function call, since it was only using the device pointer
to set the operations.  Now it uses the device pointer to tell it which ops
to check.

Signed-off-by: Myles Watson

With this change, my other patch is build tested for qemu, norwich, &
serengeti.

Run-tested on qemu & serengeti.

Thanks,
Myles

[ron minnich]

Acked-by: Ronald G. Minnich

[Myles Watson]

On Wed, Nov 5, 2008 at 4:14 PM, ron minnich  wrote:
> Acked-by: Ronald G. Minnich

Rev 982.

Thanks,
Myles

[ron minnich]

looks good, not build or boot tested.

Acked-by: Ronald G. Minnich

[Myles Watson]

On Wed, Nov 5, 2008 at 1:23 PM, ron minnich  wrote:
> looks good, not build or boot tested.
>
> Acked-by: Ronald G. Minnich

Rev 983.

Thanks,
Myles

[ron minnich]

Challenge: get via or k8 working by rev 1000. I used to think we'd make it :-)

ron