Re: [PATCH] Re: 2.6.17-rc5-git1: regression: resume from suspend(RAM) fails: libata issue

[Tejun Heo <htejun@gmail.com>]

Jeff Garzik wrote:
> Jens Axboe wrote:
>> On Sat, May 27 2006, Jeff Garzik wrote:
>>>> +    msleep(500);
>>>>     pci_set_power_state(pdev, PCI_D0);
>>>>     pci_restore_state(pdev);
>>>>     pci_enable_device(pdev);
>>> Does it work if you move msleep() below pci_restore_state()?
>>> "    "     "    "      below pci_enable_device()?
>>
>> Placement makes no difference here, it works after pci_enable_device()
>> as well. What matters is the delay coming before the ata_busy_wait().
>> You could just stuff it in front of that, if you wanted.
> 
> That would increase the wait needlessly, since it would be then done 
> per-device, even though the devices spin up in parallel on ata_piix.
> 

This sounds a lot like hotplug spinup-wait case.  Are those harddisks 
PATA or SATA?  Also, I think this can be handled better by EH.  ie. 
register it as hotplug event -> wait for spin up if necessary -> reset & 
revalidate.

-- 
tejun