Skip to contentSkip to conversation

Libata for parallel ATA controllers

DISCUSSION Alan Cox 2 participants 3 messages Sep 22, 2005 – Sep 23, 2005

lore
Share & tools
lore

Alan Cox

author
#0 lore raw

I mentioned a while ago I was hacking on libata and PATA drivers. I've also fed a few bits of the needed support code to Jeff Garzik as I went.

Some initial patches are now ready for wider testing although strictly suicide squad material at this point. I'm now at the point I'm running a full Fedora Core 4 with CONFIG_IDE=n using four disk raid on SIL680 controllers.

A lot of hardware isn't yet covered - I'm working on adding more support but I wanted to start with weirder devices first to better understand what was needed in libata.

Status info and patches are at

http://zeniv.linux.org.uk/~alan/IDE

Enjoy but remember this is very early code and don't use it for production!

#1 lore raw
Quoted 7 lines · Alan Cox

Status info and patches are at

http://zeniv.linux.org.uk/~alan/IDE

Enjoy but remember this is very early code and don't use it for production!

And especially keep in mind, that libata has practically *no* built-in error-handling or recovery mechanisms yet. If a drive gets into a "reset me to recover" state, then libata just might require a reboot to recover, whereas the IDE subsystem will usually try a reset operation at some point.

Not a problem with modern, mostly bug-free hardware (eg. most SATA), but this could be an issue for some PATA interfaces.

Cheers

author
#2 lore raw
Quoted 4 lines · Mark Lord

built-in error-handling or recovery mechanisms yet. If a drive gets into a "reset me to recover" state, then libata just might require a reboot to recover, whereas the IDE subsystem will usually try a reset operation at some point.

Or crash.

Quoted 2 lines

Not a problem with modern, mostly bug-free hardware (eg. most SATA), but this could be an issue for some PATA interfaces.

The basic error handling in the libata code seems to work as well when I tested it, if not better because the old PATA code hangs the box on SMP or pre-empt if you get a DMA timeout and cable changedown due to locking flaws and also issues an immediate idle in error recovery which seems to crash some drives for good.

What doesn't work at all is failed cable detect - the speed change down support simply isn't in libata yet and that turns a downspeed change for poor cables or cable misdetect into a hang.