[Devel] RE: i2o hardware hangs (ASR-2010S)
Salyzyn, Mark
mark_salyzyn at adaptec.com
Mon Aug 7 09:06:35 PDT 2006
The dpt_i2o driver has the advantage of all the requests tracked by the
scsi subsystem, returning them as scsi queue full to be retried.
Adpt_fail_posted_scbs is small miracle of simplicity. The i2o driver
will have to maintain it's own queue of commands to add this
functionality (!)
Vasily, it will necessarily be up to you as to whether you switch to
dpt_i2o to get the hardening you require today, or work out a deal with
Markus to add timeout/reset functionality to the i2o driver. If you wish
to attack this issue on your own and provide a patch to Markus, I am
here to provide technical advice about the DPT/Adaptec I2O cards, but
must admit a basic ignorance to Markus' driver sources and architecture.
My recommendations for the i2o driver reset procedure is to use a
rolling timeout, every new command completion resets the global timer.
This will allow starved or long commands to process. Once the timer hits
3 minutes for RAID (Block or SCSI) targets that have multiple
inheritances, 30 seconds for SCSI DASD targets, or some insmod tunable,
it resets the adapter. I recommend that when we hit ten seconds, or some
insmod tunable, that we call a card specific health check routine. I do
not recommend health check polling because we have noticed a reduction
in Adapter performance in some systems and generic i2o cards would
require a command to check, so that is why I tie it to the ten seconds
past last completion. For the DPT/Adaptec series of adapters, it checks
the BlinkLED status (code fragment in dpt_i2o driver at
adpt_read_blink_led), and if set, immediately record the fact and resets
the adapter. For cards other than the DPT/Adaptec series, I recommend a
short timeout Get Status request to see if the Firmware is in a run
state and is responsive to this simple command. The reset code will need
to retry all commands itself, I do not believe the block system has an
error status that can be used for it to retry the commands. If the Reset
Iop in the reset adapter code is unresponsive, then the known targets
need to be placed offline.
Sincerely -- Mark Salyzyn
> -----Original Message-----
> From: Markus Lidel [mailto:Markus.Lidel at shadowconnect.com]
> Sent: Monday, August 07, 2006 10:33 AM
> To: Salyzyn, Mark
> Cc: Vasily Averin; devel at openvz.org
> Subject: Re: i2o hardware hangs (ASR-2010S)
>
>
> Hello,
>
> Salyzyn, Mark wrote:
> > 64 bit (address and datapath) works in the driver I have provided,
> > although we have heard of some SM motherboards that work
> with these ZCR
> > cards that have broken bridges. The interference issue required both
> > drivers to register the address range, the sources I have provided
> > perform the registration, you may have to check with Markus
> to see if
> > the version of the i2o driver utilizes the same.
> > It was decided by the community to deprecate the dpt_i2o
> driver in the
> > 2.6 kernel, it still remains but any bugfixes are rejected
> unless they
> > are minor. Adaptec is committed to supporting the dpt_i2o
> driver for OEM
> > customers. Markus has taken efforts to incorporate the
> dpt_i2o features,
> > 64 bit etc, in the i2o driver. I do hope he has
> incorporated a timeout
> > and recovery mechanism, it is not dpt_i2o specific. I look
> forward to
> > his comments.
>
> At the moment there is no recovery mechanism in case of a
> timeout in the
> I2O driver. I think it could be a little bit problematic to reset the
> controller in case a timeout occur, because all open
> operations are lost
> in this case. But i agree that at least an error message
> should be logged
> to inform the user something is going wrong.
>
> >> -----Original Message-----
> >> From: Vasily Averin [mailto:vvs at sw.ru]
> >> Sent: Monday, August 07, 2006 4:05 AM
> >> To: Salyzyn, Mark
> >> Cc: Markus Lidel; devel at openvz.org
> >> Subject: Re: i2o hardware hangs (ASR-2010S)
> >>
> >>
> >> Hello Mark,
> >>
> >> thank you for your assistance.
> >>
> >> Salyzyn, Mark wrote:
> >>> Markus, when the commands time out, do you perform a reset
> >> iop sequence?
> >>> I thought you added the BlinkLED code detection that is in
> >> the dpt_i2o
> >>> driver, if not, we should make sure it is there so that we
> >> get a report
> >>> in the console and an accompanying reset. Vasily, you
> >> console log did
> >>> not report anything at the time of failure, I would have
> >> expected some
> >>> timeout reports.
> >> Unfortunately console logs does not have any errors or
> >> timeout reports.
> >> If you wish, I can send you console logs directly.
> >>
> >> However as far as I understand i2o layer does not have any
> >> sort of timeout/error
> >> handlers (I hope Markus correct me if I'm err), and it would
> >> be great if this
> >> feature will be appear in the future.
> >>
> >>> If it will help, Vasily, contact me for the latest
> dpt_i2o driver as
> >>> that is the driver I am most familiar with; it may be of
> interest to
> >>> determine if the problem duplicates with the dpt_i2o
> driver. Keep in
> >>> mind that the i2o driver is a block driver, dpt_i2o is a
> >> scsi driver.
> >>
> >> Unfortunately we do not know how we can reproduce this issue.
> >> Currently it
> >> occurs on the production nodes only and customers are very
> >> against of any
> >> experiments on these nodes.
> >>
> >> Therefore it is not to easy to switch from i2o layer to your
> >> dpt_i2o driver.
> >>
> >> Currently we have not dpt_i2o driver in our kernels. The most
> >> important reasons are:
> >> - this driver did have some problems on 64-bit kernels (but
> >> it is resolved
> >> already, I'm I right?).
> >> - it is not included into 2.6-based Red Hat distributiuons.
> >> - it did not worked when I've tried to compile it into kernel.
> >> - when I've tried to build it as module, I've discovered that
> >> it conflicts with
> >> i2o drivers: initscripts on the some distributions (FC4?)
> >> have tried to load
> >> both of these modules (one from initrd, second -- when
> >> detects according PCIID)
> >> and it hangs the node. I've not found any working combination
> >> and therefore
> >> we've decided to not include dpt_i2o driver into our 2.6 kernels.
> >>
> >> However, Mark, I'm ready to check your new driver on our
> >> internal testnodes, and
> >> if last issue (modules conflicts) is fixed I'll try to
> >> include your driver into
> >> our kernels.
> >>
> >> Thank you,
> >> Vasily Averin
> >>
> >>> Sincerely -- Mark Salyzyn
> >>>
> >>>> -----Original Message-----
> >>>> From: linux-scsi-owner at vger.kernel.org
> >>>> [mailto:linux-scsi-owner at vger.kernel.org] On Behalf Of
> Vasily Averin
> >>>> Sent: Friday, August 04, 2006 7:50 AM
> >>>> To: linux-scsi at vger.kernel.org; Markus Lidel
> >>>> Cc: devel at openvz.org
> >>>> Subject: i2o hardware hangs (ASR-2010S)
> >>>>
> >>>>
> >>>> Hello Markus,
> >>>>
> >>>> We experience problems with I2O hardware on 2.6 kernels,
> >>>> probably this can help
> >>>> you or maybe you even know the answer. Can you please,
> take a look?
> >>>>
> >>>> After migration to 2.6 kernels our customers began to claim
> >>>> that i2o-based
> >>>> nodes hang. We have investigated these claims and discovered
> >>>> that i2o disks on
> >>>> theses nodes stopped the processing of any IO requests.
> >>>> Please, note, it is not
> >>>> a single issue, it happens from time to time.
> >>>>
> >>>> Our kernel-space watchdog module has produced the following
> >>>> output to serial console
> >>>>
> >>>> Jul 31 07:38:37
> >>>> (80,0) i2o/hda r(77135616 1632632476 15538880) w(69903626
> >>>> 1034743472 407332291)
> >>>> Jul 31 07:39:38
> >>>> (80,0) i2o/hda r(77148190 1633252850 15543968) w(69906364
> >>>> 1034764548 407338084)
> >>>> (80,0) i2o/hda r(77157038 1633672916 15546672) w(69912375
> >>>> 1034808048 407351490)
> >>>> (80,0) i2o/hda r(77169933 1634285356 15550897) w(69916317
> >>>> 1034845588 407364374)
> >>>> (80,0) i2o/hda r(77178290 1634941276 15555039) w(69919031
> >>>> 1034865212 407369386)
> >>>> (80,0) i2o/hda r(77192170 1635427776 15559925) w(69922676
> >>>> 1034892406 407377617)
> >>>> (80,0) i2o/hda r(77216478 1635774384 15570783) w(69927294
> >>>> 1034921708 407385382)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928376 407387163)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928378 407387163)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928384 407387164)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928384 407387164)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928384 407387164)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928386 407387164)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928390 407387164)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928390 407387164)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928390 407387164)
> >>>> (80,0) i2o/hda r(77221642 1635925752 15572389) w(69927966
> >>>> 1034928390 407387164)
> >>>>
> >>>> where r(reads, read_sectors, read_merges) w(writes,
> >>>> write_sectors, write_merges)
> >>>>
> >>>> Magic keys works, according to showProcess processors are in
> >>>> idle, ShowTraces
> >>>> shows a few thousand processes in D-state, but we can not
> >>>> find any deadlocks, it
> >>>> looks like the processes waits until I/O finished.
> >>>> Unfortunately i2o layer has
> >>>> no any error handlers and there is no any chance that the
> >>>> node will return
> >>> >from this coma.
> >>>> Described incident has occurred after ~2 weeks uptime. It was
> >>>> Supermicro X5DP8
> >>>> motherboard /8Gb memory /Adaptec ASR-2010S I2O Zero
> Channel. Kernel
> >>>> 2.6.8-022stab078.9-enterprise, sources/configs are accessible
> >>>> on openvz.org.
> >>>>
> >>>> In the bootlogs I've found mtrr message. As far as I know you
> >>>> have fixed this
> >>>> issue, however I'm not sure that it can leads to described hang.
> >>>>
> >>>> I2O Core - (C) Copyright 1999 Red Hat Software
> >>>> i2o: max_drivers=4
> >>>> i2o: Checking for PCI I2O controllers...
> >>>> ACPI: PCI interrupt 0000:06:01.0[A] -> GSI 72 (level,
> low) -> IRQ 72
> >>>> i2o: I2O controller found on bus 6 at 8.
> >>>> i2o: PCI I2O controller
> >>>> BAR0 at 0xF8400000 size=1048576
> >>>> BAR1 at 0xFB000000 size=16777216
> >>>> mtrr: type mismatch for fb000000,1000000 old: uncachable new:
> >>>> write-combining
> >>>> i2o: could not enable write combining MTRR
> >>>> iop0: Installed at IRQ 72
> >>>> iop0: Activating I2O controller...
> >>>> iop0: This may take a few minutes if there are many devices
> >>>> iop0: HRT has 1 entries of 16 bytes each.
> >>>> Adapter 00000012: TID 0000:[HPC*]:PCI 1: Bus 1 Device 22
> Function 0
> >>>> iop0: Controller added
> >>>> I2O Block Storage OSM v0.9
> >>>> (c) Copyright 1999-2001 Red Hat Software.
> >>>> block-osm: registered device at major 80
> >>>> block-osm: New device detected (TID: 211)
> >>>> Using anticipatory io scheduler
> >>>> i2o/hda: i2o/hda1 i2o/hda2 < i2o/hda5 i2o/hda6 >
> >>>>
> >>>> # cat /proc/mtrr
> >>>> reg00: base=0xf8000000 (3968MB), size= 128MB: uncachable, count=1
> >>>> reg01: base=0x00000000 ( 0MB), size=8192MB: write-back, count=1
> >>>> reg02: base=0x200000000 (8192MB), size= 128MB:
> write-back, count=1
> >>>> reg03: base=0xf7f80000 (3967MB), size= 512KB: uncachable, count=1
> >>>>
> >>>> I would repeat, it is not a single fault, we have received
> >>>> similar claims once
> >>>> and again. For some time we believed that it was due some
> >>>> hardware faults,
> >>>> however some doubts are cast upon it. The same nodes worked
> >>>> well long time ago
> >>>> without any troubles under 2.4-based kernels with dpt_i2o
> >>>> driver and we have not
> >>>> observed any of i2o hardware troubles so frequently.
> >>>>
> >>>> Is it possible that our kernel (based on 2.6.8.1 mainstream)
> >>>> have some bugs in
> >>>> i2o drivers? However we're using driver sources taken from
> >>>> RHEL4U2 kernel, and I
> >>>> cannot find any similar claims from RHEL4 customers.
> >>>>
> >>>> Is it possible than we have some other related kernels bugs?
> >>>> In this case why we
> >>>> have such kind of issues only on i2o-based nodes?
> >>>>
> >>>> Could you please give me some hints which allow me to
> >>>> continue investigation of
> >>>> this issue. If you have any suggestions I'll check them
> next time.
> >>>>
> >>>> Thank you,
> >>>> Vasily Averin
> >>>>
> >>>> SWsoft Virtuozzo/OpenVZ Linux kernel team
>
>
> Best regards,
>
>
> Markus Lidel
> ------------------------------------------
> Markus Lidel (Senior IT Consultant)
>
> Shadow Connect GmbH
> Carl-Reisch-Weg 12
> D-86381 Krumbach
> Germany
>
> Phone: +49 82 82/99 51-0
> Fax: +49 82 82/99 51-11
>
> E-Mail: Markus.Lidel at shadowconnect.com
> URL: http://www.shadowconnect.com
>
More information about the Devel
mailing list