2003-06-05 07:42:18

by Warren Togami

[permalink] [raw]
Subject: 2.4.21-rc7 AMD64 dpt_i2o fails compile

On Wed, 2003-06-04 at 19:08, Bill Nottingham wrote:
> Warren Togami ([email protected]) said:
> > My two Tyan S2880GNR dual Opteron servers came in today. Sweet!
> >
> > Unfortunately the Adaptec 2110S SCSI RAID controllers within both
> > servers are unable to work with any of the drivers in GinGin64's
> > Anaconda. Searching the net it seems that this controller needs the
> > dpt_i2o module which exists in Shrike, but not in GinGin64's kernel.
> > Were there problems with that module in AMD64?
> >
> > http://www.togami.com/~warren/archive/2003/tyan_opteron.txt
> > lspci, lspci -n, lspci -vvv
> >
> > I'm installing onto an IDE hard drive now and plan on building custom
> > kernels from there.
> >
> > Any recommendations of things to try?
>
> Rebuild the kernel with the driver, see what happens. I don't think
> we've tested that driver here to confirm whether it does or doesn't
> have problems.
>

http://www.togami.com/~warren/archive/2003/dpt_failure.txt
2.4.20-9.2 (GinGin64)
Build failure when dpt is enabled as a module. This is probably why
this and many other kernel modules were not included in the GinGin64
preview release.

Unfortunately it fails compilation in the same place for 2.4.21-rc7.
I'm testing 2.5.70-bk* next.

LKML, any existing patches for this dpt_i2o module AMD64 compilation
issue? Please CC me because not currently subscribed to lkml.

Thanks,
Warren Togami
[email protected]


2003-06-05 10:10:26

by Alan

[permalink] [raw]
Subject: Re: 2.4.21-rc7 AMD64 dpt_i2o fails compile

> http://www.togami.com/~warren/archive/2003/dpt_failure.txt
> 2.4.20-9.2 (GinGin64)
> Build failure when dpt is enabled as a module. This is probably why
> this and many other kernel modules were not included in the GinGin64
> preview release.
>
> Unfortunately it fails compilation in the same place for 2.4.21-rc7.
> I'm testing 2.5.70-bk* next.
>
> LKML, any existing patches for this dpt_i2o module AMD64 compilation
> issue? Please CC me because not currently subscribed to lkml.

Fixing up dpt_i2o for the new DMA stuff is a major job. Fixing it for
64bit cleanness even more so.

2003-06-05 12:02:01

by Andi Kleen

[permalink] [raw]
Subject: Re: 2.4.21-rc7 AMD64 dpt_i2o fails compile

Alan Cox <[email protected]> writes:

> Fixing up dpt_i2o for the new DMA stuff is a major job. Fixing it for
> 64bit cleanness even more so.

If the hardware/firmware supports 64bit pointers then at least AMD64
can work without the PCI DMA API.

64bit cleanness is needed however.

-Andi

2003-06-05 19:34:42

by Alan

[permalink] [raw]
Subject: Re: 2.4.21-rc7 AMD64 dpt_i2o fails compile

On Iau, 2003-06-05 at 13:15, Andi Kleen wrote:
> Alan Cox <[email protected]> writes:
>
> > Fixing up dpt_i2o for the new DMA stuff is a major job. Fixing it for
> > 64bit cleanness even more so.
>
> If the hardware/firmware supports 64bit pointers then at least AMD64
> can work without the PCI DMA API.

32bit all around as far I as I can tell

2003-06-05 19:38:33

by Andi Kleen

[permalink] [raw]
Subject: Re: 2.4.21-rc7 AMD64 dpt_i2o fails compile

On Thu, Jun 05, 2003 at 08:45:44PM +0100, Alan Cox wrote:
> On Iau, 2003-06-05 at 13:15, Andi Kleen wrote:
> > Alan Cox <[email protected]> writes:
> >
> > > Fixing up dpt_i2o for the new DMA stuff is a major job. Fixing it for
> > > 64bit cleanness even more so.
> >
> > If the hardware/firmware supports 64bit pointers then at least AMD64
> > can work without the PCI DMA API.
>
> 32bit all around as far I as I can tell

Then fixing it to use the DMA code is best - you can use 32bit pointers
fine then. Just the code in the driver itself needs to be able to tolerate
a 64bit host.

-Andi

2003-06-06 05:20:30

by Warren Togami

[permalink] [raw]
Subject: Re: 2.4.21-rc7 AMD64 dpt_i2o fails compile

On Thu, 2003-06-05 at 09:45, Alan Cox wrote:
> On Iau, 2003-06-05 at 13:15, Andi Kleen wrote:
> > Alan Cox <[email protected]> writes:
> >
> > > Fixing up dpt_i2o for the new DMA stuff is a major job. Fixing it
> for
> > > 64bit cleanness even more so.
> >
> > If the hardware/firmware supports 64bit pointers then at least AMD64
> > can work without the PCI DMA API.
>
> 32bit all around as far I as I can tell

http://www.adaptec.com/worldwide/support/suppdetail.html?cat=%2fProduct%2fASR-2110S&prodkey=ASR-2110S
http://www.adaptec.com/worldwide/support/techspecs.html?prodkey=ASR-2110S&cat=%2fProduct%2fASR-2110S
Adaptec SCSI RAID 2110S claims to be a "64-bit/66MHz PCI-to-SCSI RAID
card". The physical card is longer than normal 32-bit PCI cards with
more pins that fit into a "64-bit PCI slot". Are Adaptec's claims of
64-bit hardware false?

32-bit Red Hat 9 with dpt_i2o is working great with Opteron and this
card, so I have a safe option to fallback on if we can't fix the 64-bit
operation.

Warren Togami
[email protected]

2003-06-06 08:11:02

by Andi Kleen

[permalink] [raw]
Subject: Re: 2.4.21-rc7 AMD64 dpt_i2o fails compile

> Adaptec SCSI RAID 2110S claims to be a "64-bit/66MHz PCI-to-SCSI RAID
> card". The physical card is longer than normal 32-bit PCI cards with
> more pins that fit into a "64-bit PCI slot". Are Adaptec's claims of
> 64-bit hardware false?

They are refering to 64bit PCI. That is not the same as support for 64bit
operating systems. An 64bit OS can also support non 64bit hardware, but it
needs proper support in the driver which the current linux driver doesn't
have.

You will have to wait until somebody fixes the driver.

-Andi

2003-06-06 15:15:37

by Alan

[permalink] [raw]
Subject: Re: 2.4.21-rc7 AMD64 dpt_i2o fails compile

On Gwe, 2003-06-06 at 06:33, Warren Togami wrote:
> http://www.adaptec.com/worldwide/support/suppdetail.html?cat=%2fProduct%2fASR-2110S&prodkey=ASR-2110S
> http://www.adaptec.com/worldwide/support/techspecs.html?prodkey=ASR-2110S&cat=%2fProduct%2fASR-2110S
> Adaptec SCSI RAID 2110S claims to be a "64-bit/66MHz PCI-to-SCSI RAID
> card". The physical card is longer than normal 32-bit PCI cards with
> more pins that fit into a "64-bit PCI slot". Are Adaptec's claims of
> 64-bit hardware false?

64bit bus isnt the same thing as 64bit capable or 64bit safe. The
current driver will need some work for the latter