Re: max_scsi_luns and 2.4.19-pre10.

Re: max_scsi_luns and 2.4.19-pre10.

[Kurt Garloff]

[-- Attachment #1: Type: text/plain, Size: 1541 bytes --]

Hi Austin,

enough guesses have been there not answering your questions ...

On Tue, Jun 25, 2002 at 07:46:25PM -0500, Austin Gonyou wrote:
> This originally was asking for help regarding QLA2200's, but I've since
> discovered it's a kernel param problem that I'm not sure how to solve.
> 
> Using a default RH kernel (from SGI XFS installer) and passing
> max_scsi_luns=128 in grub, and for scsi_mod, it seems to work. 

In 2.4.19pre1 a patch was merged into mainline which introduced a flag
BLIST_LARGELUN and set it for EMC Symmetrix devices. Some distributors
(incl. RH and SuSE) did ship kernels with this patch included.
http://van-dijk.net/linuxkernel/200206/0347.html
(An older patch for 2.4.16 exists as well.)

The flag does allow a device to use more than 8 LUNs despite it reporting
as SCSI Version 2 devices (which can not support more than 8 LUNs normally
...) 
The flag also needs to be set for some more devices, look for DGC, DELL, CMD
and CNSi/CNSI devices that already have the BLIST_SPARSELUN flag.

But as you did not post the output of /proc/scsi/scsi nor the syslog
meesages from your SCSI subsystem nobody knows what devices you're using or
what actually happens. Just speculations ...

PS: The better list for such questions is linux-scsi@vger.kernel.org

Regards,
-- 
Kurt Garloff  <garloff@suse.de>                          Eindhoven, NL
GPG key: See mail header, key servers         Linux kernel development
SuSE Linux AG, Nuernberg, DE                            SCSI, Security

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: max_scsi_luns and 2.4.19-pre10.

[Austin Gonyou]

On Wed, 2002-06-26 at 07:33, Kurt Garloff wrote:
> Hi Austin,
> 
> enough guesses have been there not answering your questions ...

Sure I hear that. But I posted an earlier question about QLA2200 and a
PV 660F and not seeing > 8 luns with 2.4.19-pre10. 


> On Tue, Jun 25, 2002 at 07:46:25PM -0500, Austin Gonyou wrote:
> > This originally was asking for help regarding QLA2200's, but I've since
> > discovered it's a kernel param problem that I'm not sure how to solve.
> > 
> > Using a default RH kernel (from SGI XFS installer) and passing
> > max_scsi_luns=128 in grub, and for scsi_mod, it seems to work. 
> 
> In 2.4.19pre1 a patch was merged into mainline which introduced a flag
> BLIST_LARGELUN and set it for EMC Symmetrix devices. Some distributors
> (incl. RH and SuSE) did ship kernels with this patch included.
> http://van-dijk.net/linuxkernel/200206/0347.html
> (An older patch for 2.4.16 exists as well.)


I'll take a look at that, and see if I can merge it into -aa4. 

> The flag does allow a device to use more than 8 LUNs despite it reporting
> as SCSI Version 2 devices (which can not support more than 8 LUNs normally
> ...) 
> The flag also needs to be set for some more devices, look for DGC, DELL, CMD
> and CNSi/CNSI devices that already have the BLIST_SPARSELUN flag.

This would be a DELL device, so I'll see about changing it from
SPARESLUN to LARGELUN?

> But as you did not post the output of /proc/scsi/scsi nor the syslog
> meesages from your SCSI subsystem nobody knows what devices you're using or
> what actually happens. Just speculations ...

There's nothing to post from /proc/scsi/scsi or the syslog other than
there's no more than 8 devices on my FC chain. I guess the real point
here is that if you're using FC, you're probably going to use more than
8 luns, even if not immediately. Especially for large Databases. 

> PS: The better list for such questions is linux-scsi@vger.kernel.org

That makes sense I'll post to that list immediately and see what I can
get. Sorry for the confusion. 


> Regards,
> -- 
> Kurt Garloff  <garloff@suse.de>                          Eindhoven, NL
> GPG key: See mail header, key servers         Linux kernel development
> SuSE Linux AG, Nuernberg, DE                            SCSI, Security
-- 
Austin Gonyou <austin@digitalroadkill.net>

Re: max_scsi_luns and 2.4.19-pre10.

[Austin Gonyou]

On Wed, 2002-06-26 at 07:33, Kurt Garloff wrote:
> Hi Austin,
> 
> enough guesses have been there not answering your questions ...
> 
> On Tue, Jun 25, 2002 at 07:46:25PM -0500, Austin Gonyou wrote:
> > This originally was asking for help regarding QLA2200's, but I've since
> > discovered it's a kernel param problem that I'm not sure how to solve.
> > 
> > Using a default RH kernel (from SGI XFS installer) and passing
> > max_scsi_luns=128 in grub, and for scsi_mod, it seems to work. 
> 
> In 2.4.19pre1 a patch was merged into mainline which introduced a flag
> BLIST_LARGELUN and set it for EMC Symmetrix devices. Some distributors
> (incl. RH and SuSE) did ship kernels with this patch included.
> http://van-dijk.net/linuxkernel/200206/0347.html
> (An older patch for 2.4.16 exists as well.)

Ahh...I see now. I looked for DELL in the scsi_scan.c and saw the
symmetrix there right above it. I should be able to add LARGELUN, since
it's #defined above and go for it. I'll see what this does, it's
starting to make sense now. TIA.



> The flag does allow a device to use more than 8 LUNs despite it reporting
> as SCSI Version 2 devices (which can not support more than 8 LUNs normally
> ...) 
> The flag also needs to be set for some more devices, look for DGC, DELL, CMD
> and CNSi/CNSI devices that already have the BLIST_SPARSELUN flag.
> 
> But as you did not post the output of /proc/scsi/scsi nor the syslog
> meesages from your SCSI subsystem nobody knows what devices you're using or
> what actually happens. Just speculations ...
> 
> PS: The better list for such questions is linux-scsi@vger.kernel.org
> 
> Regards,
> -- 
> Kurt Garloff  <garloff@suse.de>                          Eindhoven, NL
> GPG key: See mail header, key servers         Linux kernel development
> SuSE Linux AG, Nuernberg, DE                            SCSI, Security
-- 
Austin Gonyou <austin@digitalroadkill.net>

Re: max_scsi_luns and 2.4.19-pre10.

[Kurt Garloff]

[-- Attachment #1: Type: text/plain, Size: 2022 bytes --]

Hi Austin,

On Wed, Jun 26, 2002 at 09:18:45AM -0500, Austin Gonyou wrote:
> On Wed, 2002-06-26 at 07:33, Kurt Garloff wrote:
> > enough guesses have been there not answering your questions ...
> 
> Sure I hear that. But I posted an earlier question about QLA2200 and a

You don't think that somebody who reads your message and tries to post a
helpful comment scans the list for earlier messages of yours, do you?

> PV 660F and not seeing > 8 luns with 2.4.19-pre10.
  ^^^^^^^

This device needs BLIST_LARGELUN.

> I'll take a look at that, and see if I can merge it into -aa4. 

The patch should be in there, just not the additional devices that need
BLIST_LARGELUN.

> > The flag does allow a device to use more than 8 LUNs despite it reporting
> > as SCSI Version 2 devices (which can not support more than 8 LUNs normally
> > ...) 
> > The flag also needs to be set for some more devices, look for DGC, DELL, CMD
> > and CNSi/CNSI devices that already have the BLIST_SPARSELUN flag.
> 
> This would be a DELL device, so I'll see about changing it from
> SPARESLUN to LARGELUN?

No. Add " | BLIST_LARGELUN" .

> > But as you did not post the output of /proc/scsi/scsi nor the syslog
> > meesages from your SCSI subsystem nobody knows what devices you're using or
> > what actually happens. Just speculations ...
> 
> There's nothing to post from /proc/scsi/scsi or the syslog other than
> there's no more than 8 devices on my FC chain. I guess the real point
> here is that if you're using FC, you're probably going to use more than
> 8 luns, even if not immediately. Especially for large Databases. 

People could have seen what SCSI device you're using.
So I could have told you instead of guessing and risking to add to the noise
myself.

Regards,
-- 
Kurt Garloff  <garloff@suse.de>                          Eindhoven, NL
GPG key: See mail header, key servers         Linux kernel development
SuSE Linux AG, Nuernberg, DE                            SCSI, Security

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: max_scsi_luns and 2.4.19-pre10.

[Austin Gonyou]

On Wed, 2002-06-26 at 11:07, Kurt Garloff wrote:
> Hi Austin,
> 
> On Wed, Jun 26, 2002 at 09:18:45AM -0500, Austin Gonyou wrote:
> > On Wed, 2002-06-26 at 07:33, Kurt Garloff wrote:
> > > enough guesses have been there not answering your questions ...
> > 
> > Sure I hear that. But I posted an earlier question about QLA2200 and a
> 
> You don't think that somebody who reads your message and tries to post a
> helpful comment scans the list for earlier messages of yours, do you?
> 
Well, I do, especially if someone says it's an urgent thing to try and
get the whole story. If I don't have time, I'll ask someone to give more
info, which you did and I respect that and thank you.

> > PV 660F and not seeing > 8 luns with 2.4.19-pre10.
>   ^^^^^^^
> 
> This device needs BLIST_LARGELUN.

I got that from your earlier post and addressed it directly. Thanks for
the pointer. I didn't know the 660 was defined explicitly in the
scsi_scan.c I've since recompiled several kernel versions to assert this
as being true with the BLIST_LARGELUN option and it's definitely fixed
my problems there.

> > I'll take a look at that, and see if I can merge it into -aa4. 
> 
> The patch should be in there, just not the additional devices that need
> BLIST_LARGELUN.
> 
> > > The flag does allow a device to use more than 8 LUNs despite it reporting
> > > as SCSI Version 2 devices (which can not support more than 8 LUNs normally
> > > ...) 
> > > The flag also needs to be set for some more devices, look for DGC, DELL, CMD
> > > and CNSi/CNSI devices that already have the BLIST_SPARSELUN flag.
> > 
> > This would be a DELL device, so I'll see about changing it from
> > SPARESLUN to LARGELUN?
> 
> No. Add " | BLIST_LARGELUN" .


Yep, Did that. Again Thanks!

> > > But as you did not post the output of /proc/scsi/scsi nor the syslog
> > > meesages from your SCSI subsystem nobody knows what devices you're using or
> > > what actually happens. Just speculations ...
> > 
> > There's nothing to post from /proc/scsi/scsi or the syslog other than
> > there's no more than 8 devices on my FC chain. I guess the real point
> > here is that if you're using FC, you're probably going to use more than
> > 8 luns, even if not immediately. Especially for large Databases. 
> 
> People could have seen what SCSI device you're using.
> So I could have told you instead of guessing and risking to add to the noise
> myself.

I hear you loud and clear. I wouldn't have made so much noise myself had
it not been a *top* priority item that I couldn't find info on anywhere
else, plus the boss putting pressure too. Soon, this system will be a
6650 with PV 660F, 8GB RAM, using Linux with Oracle8.1.7.2, XFS, 0(1)
scheduler and nearly a TB of total DB. 

I have some results from my benchmarking which are quite astounding too.


> Regards,
> -- 
> Kurt Garloff  <garloff@suse.de>                          Eindhoven, NL
> GPG key: See mail header, key servers         Linux kernel development
> SuSE Linux AG, Nuernberg, DE                            SCSI, Security
-- 
Austin Gonyou <austin@digitalroadkill.net>