regarding ACPI support in 2.6.21

regarding ACPI support in 2.6.21

[Tejun Heo]

Hello, all.

Currently, libata-acpi has the following problems.

1. Matching controller/device to ACPI node.  It currently uses ap->cbl
== ATA_CBL_SATA test to choose between two formats - the traditional ATA
nodes with master/slave devices and new native SATA nodes.  This is
incorrect as ata_piix even if it's SATA should map to traditional ATA
nodes while in ahci 00 _ADR in traditional format shouldn't trigger
"don't know how to handle SATA port multiplier".

2. Although this isn't very clear in the spec, but some methods seem to
be designed to be used only over suspend/resume cycle.  SATA aware
methods (w/ _SDD) should probably called after every PHY event (device
might have been hardreset), but _GTM/_STM/_GTF combination does look
like they should be used only over suspend/resume cycles.  This is caus

I think it might be better to give up ACPI support in 2.6.21 and target
2.6.22.  What do you think?

Thanks.

-- 
tejun

Re: regarding ACPI support in 2.6.21

[Alan Cox]

> I think it might be better to give up ACPI support in 2.6.21 and target
> 2.6.22.  What do you think?

I removed it from my tree already so that I can actually use libata and
do real work. The "crash every non PCI controller" feature in the current
ACPI hacks means PCMCIA and ISAPnP do not work any more with libata.

In its current state it should be disabled, its broken, its wrong.

Alan

Re: regarding ACPI support in 2.6.21

[Jeff Garzik]

Alan Cox wrote:
>> I think it might be better to give up ACPI support in 2.6.21 and target
>> 2.6.22.  What do you think?
> 
> I removed it from my tree already so that I can actually use libata and
> do real work. The "crash every non PCI controller" feature in the current
> ACPI hacks means PCMCIA and ISAPnP do not work any more with libata.

It sounds like your tree is out-of-date.  Your patch to fix that went in 
days ago, applied by Linus directly:

commit ca4266359d0c1199af088447f209ab5bcc32a989
Author: Alan Cox <alan@lxorguk.ukuu.org.uk>
Date:   Thu Mar 8 23:13:50 2007 +0000

      [PATCH] libata-acpi: Try and stop all the non PCI devices crashing


> In its current state it should be disabled, its broken, its wrong.

Does the above change your opinion any?

I lean towards disabling it by default in 2.6.21 anyway, but am 
interested in hearing an updated opinion on what's actually in the 
public tree.

	Jeff

Re: regarding ACPI support in 2.6.21

[Alan Cox]

> It sounds like your tree is out-of-date.  Your patch to fix that went in 
> days ago, applied by Linus directly:

Its purposefully not tracking every Linus update so I can build a
replicable environment. I missed the list mail it went in tho.

> I lean towards disabling it by default in 2.6.21 anyway, but am 
> interested in hearing an updated opinion on what's actually in the 
> public tree.

For PATA disable it, for SATA no opinion.