Re: [git-patch] hotplug fix patches added

[Jeff Garzik <jgarzik@pobox.com>]

Tejun Heo wrote:
> Jeff Garzik wrote:
>> Tejun Heo wrote:
>>> Hello,
>>>
>>> Two hotplug fix patches have been added to libata-tj #for-jeff.
>>>
>>> * libata: add missing finish_wait() call in ata_port_wait_eh()
>>> * libata: cosmetic change in struct ata_port
>>>
>>> The following two patches have been dropped as it hasn't been acked yet.
>>>
>>> * ata_piix: fix ghost device probing by honoring PCS present bits[1]
>>
>> Seems vaguely OK to me...
>>
>>
>>> * libata: add ata_port->private_flags[2]
>>
>> What's the justification for this?  Running out of room in flags?
> 
> Yeap, ran out of bits while implementing power management.
> 
>> If we are going to do this, I would move any such flags to the 
>> LLDD-allocated hpriv structure completely.  I wouldn't add hpriv_flags 
>> to struct ata_port.
> 
> Several drivers only need flags.  IMHO, having to allocate hpriv just 
> for flags is a bit annoying.  I also thought about separating capability 
> flags and dynamic flags but couldn't think of proper field name. 
> ap->dyn_flags?  Also, it would be easy to screw up and test the wrong 
> field.  Another sucky option was using u64 for flags.

I tend to think that most drivers fall into two categories:

* bmdma (PCI IDE-like) drivers
* drivers which already do their own hpriv allocation and management

Therefore, I would be OK with adding a bmdma_flags member, to be used 
only by bmdma drivers [and eventually separated from the high level 
libata API, as discussed].  Otherwise, I would go ahead and add code to 
allocate an hpriv structure in the driver.

	Jeff