Re: TASKFILE ioctl for libata?

[Tejun Heo <htejun@gmail.com>]

Jeff Garzik wrote:
> Tejun Heo wrote:
> 
>> Just a side note.  Taskfile has finer granuality regarding which 
>> registers are written and read back than current libata does and IDE 
>> taskfile implementation is somewhat broken/weird in a few delicate fun 
>> ways, so... be careful.  Whoever tries it.

Hello, Jeff.

> Yes -- it opens the question about whether we care enough to fully 
> support flagged taskfiles, and if not, how to best emulate that support 
> under libata.

Yeah, and another difficult question is - Are we going to exactly 
replicate the behavior of IDE TASKFILE or try to fix / clean up?  I have 
no idea what should be done.  The thing with TASKFILE is that it doesn't 
have too many users but has enough annoying inconsistencies/bugs.  So, 
neither of rigorous replication or behavior cleanup feels right.

> I'm told that flagging individual ATA shadow registers for modification 
> (or not) is required for issuing certain specialized PATA 
> vendor-specific commands.  SATA, OTOH, transmits all ATA shadow 
> registers in a FIS, so flagged taskfiles are useless.
> 
> So, I'm now thinking the best route is to leave the code as it is ;-) 
> Rather than imperfectly implementing the flagged taskfile ioctl, punt 
> the remaining userland users to SG_IO.
> 
> I don't see lack of full flagged taskfile support as a big stumbling 
> block to libata use.

I agree.  HDIO_DRIVE_CMD and HDIO_DRIVE_TASK are orders simpler compared 
to HDIO_DRIVE_TASKFILE and we don't even know what's TASKFILE's correct 
behavior is.  So, I think we should just leave TASKFILE out there in the 
wilderness and watch it with sorrow as it dies slowly.

Also, as you pointed out, it's hardware-wise impossible to do 
register-level granuality with SATA and I'll be surprised if any more 
need for register-level granuality arises than there currently is.  So, 
I say we can forget about it without too much trouble.  And if such need 
EVER arises, let's hope SAT guys come up with something reasonable.

Thanks.

-- 
tejun