Re: [PATCH] Add support for asynchronous scans to libata

[Jeff Garzik <jgarzik@pobox.com>]

Matthew Wilcox wrote:
> On Mon, Dec 11, 2006 at 10:58:06AM -0500, Jeff Garzik wrote:
>> The time-consuming portion already takes place in a thread.  Do you mean 
>> multiple threads?  Or, ATA's scan is in one thread, while work continues 
>> in other threads?
>>
>> Patch seems sane, provided that I am educated a bit :)
> 
> Each host will be scanned in its own thread.  So boot-up and
> initialisation of other devices (eg, USB, networking, PS/2, etc) will
> continue.  Drive ordering and numbering is maintained, and the default
> configuration right now is to not enable this feature.
> 
> We have infrastructure in place to make sure that all discs are found
> before we try to start init or autorun RAID.  If there's anywhere we're
> missing a call to scsi_complete_async_scans(), it'll affect SCSI as much
> as it will IDE, and I'm extremely interested in fixing it.

oh, err, hum.

libata uses one scsihost per port, not one-per-controller, so it won't 
work like you think.

This will conflict with AHCI's staggered spin-up, at least.

	Jeff