On Sat, Sep 20, 2003 at 07:23:26PM +0200, Manuel Bouyer wrote:
> On Sat, Sep 20, 2003 at 07:40:33AM -0700, Jason Thorpe wrote:
> > 
> > On Saturday, September 20, 2003, at 03:30  AM, Manuel Bouyer wrote:
> > 
> > >Yes, and the kernel thread will help for this too. But I have no ETA 
> > >for this
> > >yet. I'd like to get hot-plug before.
> > 
> > OK.  Hot-plug for SATA is something I've been thinking about.  You need 
> > some chip-specific code to detect the PHY status change.
> 
> Yes. This shouldn't be the hard part.
> There is also hot-swap hardware available for parallel ATA.
> I think I'll work on that first; SATA is hard to find and expensive here
> for now. In 6 months it should be better.
> 
> > 
> > Regarding tagged queueing, the Intel SATA chip (31244) has a mode 
> > called "DPA mode", which basically presents 4 logical controllers, one 
> > for each SATA port, with slightly different register offsets than the 
> > PCI IDE-compatible mode ("legacy mode").  In DPA mode, the 31244 
> > supports tagged queueing, and the performance is quite good when it is 
> > used.
> 
> Some parallel ATA (don't remember if it's HPT or Promise) also have some
> support for it.
> But I still have problems getting docs for IDE controllers :(

Well, I haven't used it in ages, but I'm willing to bet that the problem
(that was never tracked down) with the on-board HPT366 on the Abit BE6-II
motherboard still exists. (when you have a disk on each channel and do a
bunch of IO that uses both disks, the system will deadlock, happened in
FreeBSD too, but not in windows with official highpoint drivers.)

-Sean

-- 
/~\ The ASCII
\ / Ribbon Campaign                   Sean Davis
 X  Against HTML                       aka dive
/ \ Email!