Re: [PATCH] libata: HPA support

[ROBERT HANCOCK <hancockr@shaw.ca>]

----- Original Message -----
From: Kyle McMartin <kyle@canonical.com>
Date: Friday, April 13, 2007 10:47 am
Subject: Re: [PATCH] libata: HPA support

> [Adding Robert to the CC incase he doesn't follow linux-ide]
> 
> On Fri, Apr 13, 2007 at 12:33:41PM -0400, Kyle McMartin wrote:
> > On Fri, Apr 13, 2007 at 12:24:34PM -0400, Jeff Garzik wrote:
> > > Kyle McMartin wrote:
> > > >Oddly, the command at least executes and doesn't MCE (but 
> it's not at all
> > > >happy either) if I use ATA_PROT_PIO. I wonder if 
> ATA_PROT_NODATA is 
> > > >buggered
> > > >on this sata_nv chip (Asus A8N-E).
> > > >
> > > >Weird...
> > > 
> > > 
> > > Try turning off ADMA using the module parameter, and see if 
> > > ATA_PROT_NODATA magically works.
> > > 
> > > ADMA is an advanced command execution mode, and it may not be 
> > > appropriate for certain non-data commands.
> > > 
> > 
> > Thanks so much, Jeff! This did it. Think we should drop ADMA by 
> default?> Do you know off-hand if there's any other drivers this 
> might bite us on?
> > 
> 
> Seems to have been commit 382a6652e91b34d5480cfc0ed840c196650493d4 
> thatcaused it (submitting NODATA commands using ADMA.)
> 
> Reverting that commit (or booting with sata_nv.adma=0) fixes HPA 
> for me
> here... Robert, is reverting that commit going to crush my little 
> world, or
> is it a safe course of action? I'd rather not disable ADMA (which 
> turns off
> NCQ, right?) wholesale, as the whizbang-gentoo crowd will hang me.

There is already a patch in libata-dev that will fix this, assuming those commands are marked as requiring a result taskfile in the command flags, which they should be:

http://git.kernel.org/?p=linux/kernel/git/jgarzik/libata-dev.git;a=commit;h=eb20a5742d230c67b9af4efd71b8b6b680ca3a09

ADMA should only be used for NODATA commands which don't require any result taskfile, such as cache flushes.