2003-01-22 09:27:23

by Przemysław Maciuszko

[permalink] [raw]
Subject: Problem with Qlogic 2200 and 2.4.20

Hello.
I have a strange problem with 2.4.20 (also 2.4.19) and Qlogic FC 2200.

The machine runs test news-server, so disk load is high.
After few minutes of running I get the following errors on console:

qlogifc0 : no handle slots, this should not happen
hostdata->queued is 19, in_ptr: 63
qlogifc0 : no handle slots, this should not happen
hostdata->queued is 19, in_ptr: 6a
qlogifc0 : no handle slots, this should not happen
hostdata->queued is 19, in_ptr: 70

and so on.

After this machine locks up completetly and hard reboot must be done.
When there is no load on disks machine runs fine for many hours.
I'm using stripped logical volume on disk connected through this Qlogic FC
(from IBM's Shark) and using ext3.
LVM version 1.0.6

--
Przemys?aw Maciuszko
Agora S.A.


2003-01-22 15:22:44

by Thomas Tonino

[permalink] [raw]
Subject: Re: Problem with Qlogic 2200 and 2.4.20

Przemys?aw Maciuszko wrote:
> Hello.
> I have a strange problem with 2.4.20 (also 2.4.19) and Qlogic FC 2200.
>
> The machine runs test news-server, so disk load is high.
> After few minutes of running I get the following errors on console:
>
> qlogifc0 : no handle slots, this should not happen
> hostdata->queued is 19, in_ptr: 63
> qlogifc0 : no handle slots, this should not happen
> hostdata->queued is 19, in_ptr: 6a
> qlogifc0 : no handle slots, this should not happen
> hostdata->queued is 19, in_ptr: 70
>
> and so on.

This is a long standing problem. Andrew Patterson gave a patch on the list:

http://groups.google.com/groups?selm=linux.scsi.1019759258.2413.1.camel%40lvadp.fc.hp.com

Without this patch, 2.4.18 would lock up in a few minutes syncing a software
raid. With the patch, it has been running multiple megabytes I/O a second for
months. And yes it can sync software RAIDs too.

The driver seems to be working for some people without the patch, but the
problem appears so quickly on some systems that it is strange the fix has not
been incorporated yet. In my case, I tested with ISP2200 cards and a JBOD with
Cheetah 10k4 and 10k5 drives.


Thomas

2003-01-23 07:46:20

by GrandMasterLee

[permalink] [raw]
Subject: Re: Problem with Qlogic 2200 and 2.4.20

On Wed, 2003-01-22 at 09:31, Thomas Tonino wrote:
> Przemys?aw Maciuszko wrote:
> > Hello.
> > I have a strange problem with 2.4.20 (also 2.4.19) and Qlogic FC 2200.
> >
> > The machine runs test news-server, so disk load is high.
> > After few minutes of running I get the following errors on console:
> >
> > qlogifc0 : no handle slots, this should not happen
> > hostdata->queued is 19, in_ptr: 63
> > qlogifc0 : no handle slots, this should not happen
> > hostdata->queued is 19, in_ptr: 6a
> > qlogifc0 : no handle slots, this should not happen
> > hostdata->queued is 19, in_ptr: 70
> >
> > and so on.
>
> This is a long standing problem. Andrew Patterson gave a patch on the list:
>
> http://groups.google.com/groups?selm=linux.scsi.1019759258.2413.1.camel%40lvadp.fc.hp.com

Just to chime in, are you using the qlogicfc driver that comes with the
kernel? If so, Try using qlogic's 6.01 driver set instead and see if
your problem goes away. I've had other problems, mostly stack related,
but I've since found my fixes and will be rolling another kernel into
production soon, but keeping that driver version. I've got qla2300's and
2200's using 6.01.

I believe the kernel driver is still based on 5.38 or less driver
version, which has problems like you're experiencing.

--
GrandMasterLee

2003-01-23 18:45:15

by Paul Jakma

[permalink] [raw]
Subject: Re: Problem with Qlogic 2200 and 2.4.20

On 23 Jan 2003, GrandMasterLee wrote:

> Just to chime in, are you using the qlogicfc driver that comes with
> the kernel? If so, Try using qlogic's 6.01 driver set instead and
> see if your problem goes away. I've had other problems, mostly stack
> related, but I've since found my fixes

hmm.. i'd be very interested in them. I have found the qlogic v6
driver to dreadfully unstable under heavy load (eg multiple
bonnie++'s) on SMP.

> GrandMasterLee

regards,
--
Paul Jakma Sys Admin Alphyra
[email protected]
Warning: /never/ send email to [email protected] or [email protected]

2003-01-24 01:37:57

by GrandMasterLee

[permalink] [raw]
Subject: Re: Problem with Qlogic 2200 and 2.4.20

On Thu, 2003-01-23 at 12:52, Paul Jakma wrote:
> On 23 Jan 2003, GrandMasterLee wrote:
>
> > Just to chime in, are you using the qlogicfc driver that comes with
> > the kernel? If so, Try using qlogic's 6.01 driver set instead and
> > see if your problem goes away. I've had other problems, mostly stack
> > related, but I've since found my fixes
>
> hmm.. i'd be very interested in them. I have found the qlogic v6
> driver to dreadfully unstable under heavy load (eg multiple
> bonnie++'s) on SMP.

I think you'll have to qualify that with hardware type etc. As for the
6.x driver versions, so far, no instability, except when in beta, was
noticed. My DBs are 1TB and greater and we load very much data into them
every day. Few hundred MB/day. Some get's reconciled away somewhere and
some is trash which is later disposed of, but they never stop loading.
It's a pretty High load I'd say, could be higher though.


> > GrandMasterLee
>
> regards,

2003-01-24 10:45:36

by Thomas Tonino

[permalink] [raw]
Subject: Re: Problem with Qlogic 2200 and 2.4.20

Paul Jakma wrote:

>>Just to chime in, are you using the qlogicfc driver that comes with
>>the kernel? If so, Try using qlogic's 6.01 driver set instead and
>>see if your problem goes away. I've had other problems, mostly stack
>>related, but I've since found my fixes
>
> hmm.. i'd be very interested in them. I have found the qlogic v6
> driver to dreadfully unstable under heavy load (eg multiple
> bonnie++'s) on SMP.

I was planning to go use the qlogic 6.x driver, but only after a test.

For me, the in-kernel driver with the patch from Andrew Patterson is very stable
on SMP. We'll have to see how the 6.x driver works out.

The in-kernel driver without patch is hopeless on single processor too. Uptime
of 15 minutes or so when doing a resync, a bit more, but not much more than an
hour, when doing less intensive IO.

In both cases with the "no handle slots, this should not happen" message. Both
on Broadcom PIII boards - one Dell, one IBM.


Thomas

2003-01-25 02:56:23

by Paul Jakma

[permalink] [raw]
Subject: Re: Problem with Qlogic 2200 and 2.4.20

On 23 Jan 2003, GrandMasterLee wrote:

> I think you'll have to qualify that with hardware type etc.

Tyan S2462NG, dual 1.7GHz Athlons (MP), QLA2310.

> As for the 6.x driver versions, so far, no instability,

Its stable for me with the RH 5.31-RH driver. With v6.x it can be
easily locked up with heavy IO, eg multiple concurrent bonnie++'s.
(might take a few hours sometimes though). Spins in the interrupt
handler.

> of, but they never stop loading. It's a pretty High load I'd say,
> could be higher though.

hmm.

regards,
--
Paul Jakma Sys Admin Alphyra
[email protected]
Warning: /never/ send email to [email protected] or [email protected]