LSIFC909 problem

LSIFC909 problem

[egi]

Since kenel > 2.6.8 my system doesn't regonize any more my FC-drives.
I installed latest kernel 2.6.23-git7 incl patch but no chance.
I get the following message during the boot:

Oct 21 08:57:06 localhost kernel: Fusion MPT base driver 3.04.05
Oct 21 08:57:06 localhost kernel: Copyright (c) 1999-2007 LSI Logic
Corporation
Oct 21 08:57:06 localhost kernel: Fusion MPT FC Host driver 3.04.05
Oct 21 08:57:06 localhost kernel: mptbase: Initiating ioc0 bringup
Oct 21 08:57:06 localhost kernel: ioc0: LSIFC909 B1:
Capabilities={Initiator,LAN}
Oct 21 08:57:06 localhost kernel: mptbase: ioc0: IOCStatus(0x0022):
Config Page Invalid Page: type=00h, page=02h, action=00h, form=00000000h
Oct 21 08:57:06 localhost kernel: mptbase: ioc0: IOCStatus(0x0022):
Config Page Invalid Page: type=09h, page=00h, action=00h, form=00000000h
Oct 21 08:57:06 localhost kernel: scsi0 : ioc0: LSIFC909 B1,
FwRev=01000000h, Ports=1, MaxQ=1023, IRQ=21
modules:

mptfc                  17508  0
mptscsih               32672  1 mptfc
mptbase                70112  2 mptfc,mptscsih
scsi_transport_fc      40548  1 mptfc


If I boot my system with kernel 2.6.8-rc2 the fc-drives are available.
But with Kernel > 2.6.8-rc2 nothing works any more.

Can someone help me?

RE: LSIFC909 problem

[Moore, Eric]

On Sunday, October 21, 2007 1:34 AM, egi wrote: 
> 
> Since kenel > 2.6.8 my system doesn't regonize any more my FC-drives.
> I installed latest kernel 2.6.23-git7 incl patch but no chance.
> I get the following message during the boot:
> 
> Oct 21 08:57:06 localhost kernel: Fusion MPT base driver 3.04.05
> Oct 21 08:57:06 localhost kernel: Copyright (c) 1999-2007 LSI Logic
> Corporation
> Oct 21 08:57:06 localhost kernel: Fusion MPT FC Host driver 3.04.05
> Oct 21 08:57:06 localhost kernel: mptbase: Initiating ioc0 bringup
> Oct 21 08:57:06 localhost kernel: ioc0: LSIFC909 B1:
> Capabilities={Initiator,LAN}
> Oct 21 08:57:06 localhost kernel: mptbase: ioc0: IOCStatus(0x0022):
> Config Page Invalid Page: type=00h, page=02h, action=00h, 
> form=00000000h
> Oct 21 08:57:06 localhost kernel: mptbase: ioc0: IOCStatus(0x0022):
> Config Page Invalid Page: type=09h, page=00h, action=00h, 
> form=00000000h
> Oct 21 08:57:06 localhost kernel: scsi0 : ioc0: LSIFC909 B1,
> FwRev=01000000h, Ports=1, MaxQ=1023, IRQ=21
> modules:


Sorry for delayed repsonse, but the FC909 is not supported anymore.   I
need to suppy a patch to remove this support.  Here is feedback from
Micheal Reed.  About a two years ago the mptfc was rewrote to support
the fibre channel transport layer.


Micheal Reed wrote:

Looking at the 2.6.5 sles9 driver, I see that the WWNN and WWPN values
are still read from fc device page 0, given a "channel" and "id" as
input.  The "channel" and "id", which I equate to "bus" and "target",
are discovered the old fasioned way, by probing every possible target on
the bus to see which respond.  This is done by a call to
scsi_scan_host().

The new transport code definitely does it differently, by reading fc
device page 0, passing in the port_id of the previous port to retrieve
data for the next port, and registering the discovered targets with the
fc transport.

In the new code, the "sorting" of the targets won't work, which isn't a
showstopper.

The CurrentTargetID and CurrentBus values are taken from fc device page
0 and placed in the VirtTarget structure, so this is the cause for
concern.  This VirtTarget structure is stored in the scsi_target's
hostdata field and is used to map the fc transport's target id to the
firmware's target id.

The 909 doesn't offer the needed functionality to support usage of the
fc transport.

If the firmware accepted commands for a target via its port id instead
of bus/target there might be a way around this, but, that would be a lot
of design effort for not a lot of return, even if it were possible.

Re: LSIFC909 problem

[Matthew Wilcox]

On Mon, Oct 22, 2007 at 10:11:24AM -0600, Moore, Eric wrote:
> Sorry for delayed repsonse, but the FC909 is not supported anymore.   I
> need to suppy a patch to remove this support.  Here is feedback from
> Micheal Reed.  About a two years ago the mptfc was rewrote to support
> the fibre channel transport layer.

Sounds like we need a new driver written to support the FC909 then.
Unless we could pretend the FC909 is a parallel scsi card or something
... that wasn't quite clear from Michael's mail.

-- 
Intel are signing my paycheques ... these opinions are still mine
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."

RE: LSIFC909 problem

[Moore, Eric]

On Monday, October 22, 2007 10:18 AM,  Matthew Wilcox wrote:
> Sounds like we need a new driver written to support the FC909 then.
> Unless we could pretend the FC909 is a parallel scsi card or something
> ... that wasn't quite clear from Michael's mail.
> 

ok, are you suggesting for FC909 we call scsi_scan_host, else chips we
"hook into transport layer"?

The way I understood from Stephen Shirron, is the FC909 is using mpi
version 1.0, and all the other FC cards are on mpi version 1.2 or 1.5.
The mpi defines the interface between driver and firmware.   The older
mpi version 1.0 is missing feature needed for the FC transport to work.


Eric

Re: LSIFC909 problem

[Matthew Wilcox]

On Mon, Oct 22, 2007 at 10:33:41AM -0600, Moore, Eric wrote:
> On Monday, October 22, 2007 10:18 AM,  Matthew Wilcox wrote:
> > Sounds like we need a new driver written to support the FC909 then.
> > Unless we could pretend the FC909 is a parallel scsi card or something
> > ... that wasn't quite clear from Michael's mail.
> > 
> 
> ok, are you suggesting for FC909 we call scsi_scan_host, else chips we
> "hook into transport layer"?

Yes.  That reminds me, I need to convert fusion to the parallel scanning
regime.  I haven't forgotten our conversation at OLS2006.

> The way I understood from Stephen Shirron, is the FC909 is using mpi
> version 1.0, and all the other FC cards are on mpi version 1.2 or 1.5.
> The mpi defines the interface between driver and firmware.   The older
> mpi version 1.0 is missing feature needed for the FC transport to work.

That sounds plausible.

-- 
Intel are signing my paycheques ... these opinions are still mine
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."

Re: LSIFC909 problem

[egi]

Matthew Wilcox wrote:
> On Mon, Oct 22, 2007 at 10:33:41AM -0600, Moore, Eric wrote:
>> On Monday, October 22, 2007 10:18 AM,  Matthew Wilcox wrote:
>>> Sounds like we need a new driver written to support the FC909 then.
>>> Unless we could pretend the FC909 is a parallel scsi card or something
>>> ... that wasn't quite clear from Michael's mail.
>>>
>> ok, are you suggesting for FC909 we call scsi_scan_host, else chips we
>> "hook into transport layer"?
> 
> Yes.  That reminds me, I need to convert fusion to the parallel scanning
> regime.  I haven't forgotten our conversation at OLS2006.
> 
>> The way I understood from Stephen Shirron, is the FC909 is using mpi
>> version 1.0, and all the other FC cards are on mpi version 1.2 or 1.5.
>> The mpi defines the interface between driver and firmware.   The older
>> mpi version 1.0 is missing feature needed for the FC transport to work.
> 
> That sounds plausible.
> 
Thank you a lot for your infos. You writes that this card isn't any more 
supported.
What's suspectfor me is why the fc909 is still listed on the Kconfig as 
supported controller. I request this lsi909 card 8 mnoth ago. This was a 
new card and the seller told me that this card was often sold and he had 
also a lot of this cards allways in his stock. (maybe i' m the first 
user of this card which shuold run with linux ;-) ..
You wrote any changes which must be done to support this card with the 
current kernel. Is this a lot of work for you?