Re: [PATCH 3/3] Add disk hotswap support to libata RESEND #2

[Mark Lord <liml@rtr.ca>]

Lukasz Kosewski wrote:
> On 9/6/05, Jim Ramsay <jim.ramsay@gmail.com> wrote:
> 
>>However, I have seen the occasion where a single IRQ is used to signal
>>both a DMA completion AND a hotplug event.  Of course in this case the
>>hotplug event itself would be ignored completely.
>>
>>So I would recommend getting rid of that check entirely.
> 
> 
> Hey Jim,
> 
> Not that I disbelieve you, but do you have an example of a controller
> where this happens?  I've done a lot of testing and never seen this...

I missed the beginning of this discussion,
but here's a data point:

The QStor SATA/RAID controller hardware fully supports hotplug
(and NCQ, TCQ, Host-Queuing, RAID 0/1/10, PM, etc..).

It uses a single interrupt for all onboard events from the four channels.
An internal "status FIFO" provides a readout for the interrupt handler
of recent happenings, in sequence, mixing together DMA-completions
with hotplug-events (insert, removal) and various fault-conditions.

All of this is supported in the out-of-tree qstor driver,
but only simple single-IO is supported by sata_qstor at present.

Dunno if that info is of any use to you in hotplug considerations.
Once the libata infrastructure for hotplug is in place,
I *may* experiment with adding that functionality to sata_qstor.

Cheers