Problem with ide=nodma

Problem with ide=nodma

[Ken Moffat]

Hi,

 I'm trying a sii 0680 disk controller at the moment, as a possible
workaround for some via southbridge problems (this is on a ppc which
isn't yet supported by the official kernels, but it has been stable here
since 2.6.7 and looks nearly ready for a first review).  Unfortunately,
DMA is a big no go at the moment so I have to pass ide=nodma in the
bootargs.

 I've got the drives plugged into the sii card, and ide=reverse is doing
its job.  But although dmesg shows that dma has been turned off,
/proc/ide/hda/settings and hdparm show that dma is in use.  This is in
2.6.9-rc3.

 Doesn't ide=nodma work for off-board chipsets ?

Ken
-- 
 das eine Mal als Tragödie, das andere Mal als Farce

Re: Problem with ide=nodma

[Bartlomiej Zolnierkiewicz]

On Sat, 9 Oct 2004 00:32:01 +0100 (BST), Ken Moffat
<ken@kenmoffat.uklinux.net> wrote:
> Hi,
> 
>  I'm trying a sii 0680 disk controller at the moment, as a possible
> workaround for some via southbridge problems (this is on a ppc which
> isn't yet supported by the official kernels, but it has been stable here
> since 2.6.7 and looks nearly ready for a first review).  Unfortunately,
> DMA is a big no go at the moment so I have to pass ide=nodma in the
> bootargs.
> 
>  I've got the drives plugged into the sii card, and ide=reverse is doing
> its job.  But although dmesg shows that dma has been turned off,

Is it possible that you are reading it wrong?

> /proc/ide/hda/settings and hdparm show that dma is in use.  This is in
> 2.6.9-rc3.
> 
>  Doesn't ide=nodma work for off-board chipsets ?

siimage host driver doesn't respect "ide=nodma".
You can hack siimage.c and comment out "hwif->autodma = 1".

Re: Problem with ide=nodma

[Ken Moffat]

On Sat, 9 Oct 2004, Bartlomiej Zolnierkiewicz wrote:

> On Sat, 9 Oct 2004 00:32:01 +0100 (BST), Ken Moffat
> <ken@kenmoffat.uklinux.net> wrote:
> > Hi,
> >
> >  I'm trying a sii 0680 disk controller at the moment, as a possible
> > workaround for some via southbridge problems (this is on a ppc which
> > isn't yet supported by the official kernels, but it has been stable here
> > since 2.6.7 and looks nearly ready for a first review).  Unfortunately,
> > DMA is a big no go at the moment so I have to pass ide=nodma in the
> > bootargs.
> >
> >  I've got the drives plugged into the sii card, and ide=reverse is doing
> > its job.  But although dmesg shows that dma has been turned off,
>
> Is it possible that you are reading it wrong?

 I don't think so, and the box is a lot more responsive.  dmesg shows

ide_setup: ide=nodmaIDE: Prevented DMA

>
> > /proc/ide/hda/settings and hdparm show that dma is in use.  This is in
> > 2.6.9-rc3.
> >
> >  Doesn't ide=nodma work for off-board chipsets ?
>
> siimage host driver doesn't respect "ide=nodma".
> You can hack siimage.c and comment out "hwif->autodma = 1".
>

Ok, thanks.  I'll give that a try later.

Ken
-- 
 das eine Mal als Tragödie, das andere Mal als Farce

Re: Problem with ide=nodma

[Bartlomiej Zolnierkiewicz]

On Sat, 9 Oct 2004 01:44:57 +0100 (BST), Ken Moffat
<ken@kenmoffat.uklinux.net> wrote:
> On Sat, 9 Oct 2004, Bartlomiej Zolnierkiewicz wrote:
> 
> > On Sat, 9 Oct 2004 00:32:01 +0100 (BST), Ken Moffat
> > <ken@kenmoffat.uklinux.net> wrote:
> > > Hi,
> > >
> > >  I'm trying a sii 0680 disk controller at the moment, as a possible
> > > workaround for some via southbridge problems (this is on a ppc which
> > > isn't yet supported by the official kernels, but it has been stable here
> > > since 2.6.7 and looks nearly ready for a first review).  Unfortunately,
> > > DMA is a big no go at the moment so I have to pass ide=nodma in the
> > > bootargs.
> > >
> > >  I've got the drives plugged into the sii card, and ide=reverse is doing
> > > its job.  But although dmesg shows that dma has been turned off,
> >
> > Is it possible that you are reading it wrong?
> 
>  I don't think so, and the box is a lot more responsive.  dmesg shows
> 
> ide_setup: ide=nodmaIDE: Prevented DMA

This is misleading as drivers are free to override this setting.

Re: Problem with ide=nodma

[Ken Moffat]

On Sat, 9 Oct 2004, Bartlomiej Zolnierkiewicz wrote:

> > > Is it possible that you are reading it wrong?
> >
> >  I don't think so, and the box is a lot more responsive.  dmesg shows
> >
> > ide_setup: ide=nodmaIDE: Prevented DMA
>
> This is misleading as drivers are free to override this setting.
>

 Bart, thanks for putting me straight.  Indeed, siimage.c doesn't have
'if (!autodma)' to wrap 'hwif->autodma = 1;' like some of the others
(e.g. hpt366).  Easy enough for me to add it, but for reasons I'm too
dumb to understand that turns off dma even without ide=nodma in the
bootargs, and it then does a series of time out / resets if I try to
enable dma with hdparm :-(

 Obviously, the sensible thing for me to do is to not touch siimage.c
and to turn dma off with hdparm until I'm ready to do extended dma
tests.  It's not as if any sane platforms really want to add extra ide
controllers and then cripple them.

 Thanks anyway.

Ken
-- 
 das eine Mal als Tragödie, das andere Mal als Farce