-------- Original Message --------
Subject: Re: [PATCH 2/5]PCI: x86 MMCONFIG: add legacy pci conf functions
Date: Wed, 19 Dec 2007 19:07:27 -0500
From: Tony Camuso <[email protected]>
Reply-To: [email protected]
To: Greg KH <[email protected]>
References:
<20071219221746.20362.39243.sendpatchset@dhcp83-188.boston.redhat.com>
<20071219221756.20362.84805.sendpatchset@dhcp83-188.boston.redhat.com>
<[email protected]>
Greg KH wrote:
>> +struct pci_ops pci_legacy_ops = {
>> + .read = pci_ops_legacy_read,
>> + .write = pci_ops_legacy_write,
>> +};
>
> This whole structure is never used in this patch, why add it now?
It gets used by the routine pcibios_fix_bus_scan_quirk in
arch/x86/pci/common.c
>
>> /*
>> * Before we decide to use direct hardware access mechanisms, we try to do some
>> @@ -258,18 +284,28 @@ void __init pci_direct_init(int type)
>> {
>> if (type == 0)
>> return;
>> - printk(KERN_INFO "PCI: Using configuration type %d\n", type);
>> - if (type == 1)
>> + if (type == 1) {
>> raw_pci_ops = &pci_direct_conf1;
>> - else
>> + pci_legacy_conf = &pci_direct_conf1;
>> + } else {
>> raw_pci_ops = &pci_direct_conf2;
>> + pci_legacy_conf = &pci_direct_conf2;
>
> Why have two pointers to the same thing? What is pci_legacy_conf going
> to be used for? Why not just use raw_pci_ops?
>
> thanks,
>
> greg k-h
The way the code works today, when raw_pci_ops is set to use mmconf,
it is platform-wide. Legacy pci config reads are not available.
I would have simply used the pci_direct_conf1, but because there can
also be a pci_direct_conf2, I thought it best to abstract it as
pci_legacy_conf.
A pointer to struct pci_ops pci_legacy_ops data is written into the
pci_bus.ops field for buses that must use legacy pci config access.
The other busses can continue to use the pointer to the mmconf pci_ops.