Port-Multiplier hotplug support: ATA_FLAG_AN

Port-Multiplier hotplug support: ATA_FLAG_AN

[Mark Lord]

Tejun,

Apparently the newer (GenIIe) sata_mv chips *can* support asynchronous
notification without problems.  Some/all of the older ones cannot (errata).

So, I've done the one-liner change in sata_mv, to add ATA_FLAG_AN to
the 7042 and 6042 chips, and it all seems to work rather nicely.

Am I missing something?  Is there any other local code to sata_mv
that should be provided before setting ATA_FLAG_AN ?

Like I said, it works, but I just want to check whether any other
pieces are expected to be present? 

Cheers

Mark

Re: Port-Multiplier hotplug support: ATA_FLAG_AN

[Jeff Garzik]

Mark Lord wrote:
> Tejun,
> 
> Apparently the newer (GenIIe) sata_mv chips *can* support asynchronous
> notification without problems.  Some/all of the older ones cannot (errata).
> 
> So, I've done the one-liner change in sata_mv, to add ATA_FLAG_AN to
> the 7042 and 6042 chips, and it all seems to work rather nicely.
> 
> Am I missing something?  Is there any other local code to sata_mv
> that should be provided before setting ATA_FLAG_AN ?
> 
> Like I said, it works, but I just want to check whether any other
> pieces are expected to be present?

Did you check ATAPI or just PMP AN?

	Jeff

Re: Port-Multiplier hotplug support: ATA_FLAG_AN

[Mark Lord]

Jeff Garzik wrote:
> Mark Lord wrote:
>> Tejun,
>>
>> Apparently the newer (GenIIe) sata_mv chips *can* support asynchronous
>> notification without problems.  Some/all of the older ones cannot 
>> (errata).
>>
>> So, I've done the one-liner change in sata_mv, to add ATA_FLAG_AN to
>> the 7042 and 6042 chips, and it all seems to work rather nicely.
>>
>> Am I missing something?  Is there any other local code to sata_mv
>> that should be provided before setting ATA_FLAG_AN ?
>>
>> Like I said, it works, but I just want to check whether any other
>> pieces are expected to be present?
> 
> Did you check ATAPI or just PMP AN?
..

sata_mv does not (yet) support ATAPI for anything.

-ml

Re: Port-Multiplier hotplug support: ATA_FLAG_AN

[Tejun Heo]

Mark Lord wrote:
> Tejun,
> 
> Apparently the newer (GenIIe) sata_mv chips *can* support asynchronous
> notification without problems.  Some/all of the older ones cannot (errata).
> 
> So, I've done the one-liner change in sata_mv, to add ATA_FLAG_AN to
> the 7042 and 6042 chips, and it all seems to work rather nicely.
> 
> Am I missing something?  Is there any other local code to sata_mv
> that should be provided before setting ATA_FLAG_AN ?
> 
> Like I said, it works, but I just want to check whether any other
> pieces are expected to be present?

You need to call sata_async_notification() from interrupt handler after 
AN FIS is received.  sata_async_notification() is responsible for 
discerning between ATAPI and PMP ANs using SNotification and schedule 
appropriate actions.  If SNotification is not available && PMP is 
attached, it just calls ata_port_schedule_eh().  That's probably why it 
just worked by setting ATA_FLAG_AN.

Thanks.

-- 
tejun

Re: Port-Multiplier hotplug support: ATA_FLAG_AN

[Mark Lord]

Tejun Heo wrote:
> Mark Lord wrote:
>> Tejun,
>>
>> Apparently the newer (GenIIe) sata_mv chips *can* support asynchronous
>> notification without problems.  Some/all of the older ones cannot 
>> (errata).
>>
>> So, I've done the one-liner change in sata_mv, to add ATA_FLAG_AN to
>> the 7042 and 6042 chips, and it all seems to work rather nicely.
>>
>> Am I missing something?  Is there any other local code to sata_mv
>> that should be provided before setting ATA_FLAG_AN ?
>>
>> Like I said, it works, but I just want to check whether any other
>> pieces are expected to be present?
> 
> You need to call sata_async_notification() from interrupt handler after 
> AN FIS is received.  sata_async_notification() is responsible for 
> discerning between ATAPI and PMP ANs using SNotification and schedule 
> appropriate actions.  If SNotification is not available && PMP is 
> attached, it just calls ata_port_schedule_eh().  That's probably why it 
> just worked by setting ATA_FLAG_AN.
..

Okay, that's cool.  So mv_err_intr() now just calls sata_async_notification(),
and then bails out without further action (if no other errors coincided),
and the EH then runs for just that one PM link, rather than the whole port.

(tested, working).

Cool!