Re: [PATCH 8/20] libata: reimplement ata_pci_init_one() using new init helpers

[Tejun Heo <htejun@gmail.com>]

Agreed on all your comments for 6 & 7.

Jeff Garzik wrote:
> Tejun Heo wrote:
> 
> Nevermind previous question about legacy resources.
> 
> It does seem like the change is split across these two patches though... 
>  so the previous patch is incorrect without this one.

Yeap, that was more of logical devision and the code isn't yet activated.

>> +int ata_pci_init_one(struct pci_dev *pdev, struct ata_port_info 
>> **pinfo_ar,
>> +             unsigned int n_ports)
>> +
>> +{
>> +    unsigned int mask = ATA_PORT_PRIMARY | ATA_PORT_SECONDARY;
>> +    unsigned int legacy_mask = 0;
>> +    struct ata_host *host;
>> +    int rc;
>> +
>> +    DPRINTK("ENTER\n");
>> +
>> +    /* legacy mode? */
>> +    if (!(pinfo_ar[0]->flags & ATA_FLAG_NO_LEGACY))
>> +        legacy_mask = ata_pci_legacy_mask(pdev);
>> +
> 
> NAK -- you can't do anything with the PCI device before calling 
> pci_enable_device()

This is what we're currently doing in ata_pci_init_one().  I thought the 
assumption wass that any device which can act as a legacy device must 
have been initialized by BIOS and thus is safe to access.  If not, 
moving this backward isn't difficult.  FWIW, ata_pci_legacy_mask() only 
performs PCI-configuration access.

Thanks.

-- 
tejun