Re: Maxtor drive doesn't wake from sleep

Re: Maxtor drive doesn't wake from sleep

[Stefan Richter]

Mi wrote at linux1394-devel:
> Hello,
> 
> I have 3 external firewire Maxtor One Touch III drives, which appear to 
> go into a "sleep" state after between one and two hours, and cannot be 
> accessed after that.
> 
> After a lot of searches and tests, I have found a few clues, but still 
> wonder why they don't wake up automatically as needed. Is there anything 
> I can do to my system so that it "Just Works"? Is this a known problem?

There are very few vendors which implement auto-spin-down in FireWire 
disk enclosures. And of these few, only few get it right.

> To have the drive re-appear, I found I can
> 
>   rmmod sbp2
>   modprobe sbp2
> 
> and I later found scsi-spin which is also able to wake up the drive:
> 
>   scsi-spin -u /dev/sdd
> 
> But since these drives are on a server and suposed to be mounted with 
> automount, these manual steps to wake them are nt much help.
> 
> Note that there is absolutely nothing in the logs when the drive goes to 
> sleep. Only errors when I try to access them:
> 
>    kernel: Device sdd not ready.
>    kernel: end_request: I/O error, dev sdd, sector 0

I don't think this could be solved in the FireWire drivers. It seems 
sd_mod is the place to look for a solution. The sbp2 driver is not aware 
that there is something wrong with the disk, else there would be error 
messages.

That's why I added linux-scsi to the recipients. I'd be glad if somebody 
of the SCSI folk could comment.

> My system is Debian stable (Sarge / 3.1) with kernel 2.6:
> 
> # uname -a
> Linux gc 2.6.8-2-686 #1 Tue Aug 16 13:22:48 UTC 2005 i686 GNU/Linux

This is quite an old kernel. I don't know though if newer kernels 
contain changes to sd_mod which are relevant to the problem.

> In case it's relevant, the Firewire card is
> 
>  0000:02:02.0 FireWire (IEEE 1394): Texas Instruments TSB82AA2 
> IEEE-1394b Link Layer Controller (rev 01) (prog-if 10 [OHCI])
>        Subsystem: Timedia Technology Co Ltd: Unknown device 3110
>        Flags: bus master, medium devsel, latency 32, IRQ 209
>        Memory at fe9ff800 (32-bit, non-prefetchable) [size=2K]
>        Memory at fe9f8000 (32-bit, non-prefetchable) [size=16K]
>        Capabilities: [44] Power Management version 2
> 
> and an excerpt from lsmod (I kept the loaded modules which looked 
> relevant + the ones I didn't know):
> 
> sbp2                   24392  0
> ohci1394               35492  0
> ieee1394              111512  2 sbp2,ohci1394
> shpchp                101900  0
> pciehp                 99020  0
> pci_hotplug            34640  2 shpchp,pciehp
> ide_scsi               17412  0
> capability              4520  0
> commoncap               7232  1 capability
> mbcache                 9348  2 ext2,ext3
> sd_mod                 21728  11
> scsi_mod              125228  5 sbp2,aic79xx,ide_scsi,sd_mod,aic7xxx
> 
> 
> I don't know how to find out the module versions, but they are the 
> modules which come with that stock Debian kernel.
> 
> 
> Thanks for any help
> 
> M

-- 
Stefan Richter
-=====-=-==- --== ===-=
http://arcgraph.de/sr/

Re: Maxtor drive doesn't wake from sleep

[Douglas Gilbert]

Stefan Richter wrote:
> Mi wrote at linux1394-devel:
> 
>> Hello,
>>
>> I have 3 external firewire Maxtor One Touch III drives, which appear
>> to go into a "sleep" state after between one and two hours, and cannot
>> be accessed after that.
>>
>> After a lot of searches and tests, I have found a few clues, but still
>> wonder why they don't wake up automatically as needed. Is there
>> anything I can do to my system so that it "Just Works"? Is this a
>> known problem?
> 
> 
> There are very few vendors which implement auto-spin-down in FireWire
> disk enclosures. And of these few, only few get it right.
> 
>> To have the drive re-appear, I found I can
>>
>>   rmmod sbp2
>>   modprobe sbp2
>>
>> and I later found scsi-spin which is also able to wake up the drive:
>>
>>   scsi-spin -u /dev/sdd
>>
>> But since these drives are on a server and suposed to be mounted with
>> automount, these manual steps to wake them are nt much help.
>>
>> Note that there is absolutely nothing in the logs when the drive goes
>> to sleep. Only errors when I try to access them:
>>
>>    kernel: Device sdd not ready.
>>    kernel: end_request: I/O error, dev sdd, sector 0
> 
> 
> I don't think this could be solved in the FireWire drivers. It seems
> sd_mod is the place to look for a solution. The sbp2 driver is not aware
> that there is something wrong with the disk, else there would be error
> messages.
> 
> That's why I added linux-scsi to the recipients. I'd be glad if somebody
> of the SCSI folk could comment.

Stefan,
I believe the sd driver should issue a START STOP UNIT
(start) command when confronted with a "... initializing
command required" error [sk,asc,ascq: 2,4,2]. It also
needs to tell the block layer to go away and amuse itself
while it polls the device until it is ready.

When I suggested this some time ago, the response was that
whatever app was responsible for stopping the device bore
the responsibility for spinning it up again (i.e. "not our
problem").

The power condition state machine of ATA and SCSI differ
see: http://www.torque.net/sg/power.html
For devices that use SAT as a guide (and Maxtor seems to be)
the reported problem is going to become more signficant:
   a) external device decides to spin itself down after
      a period of inactivity
   b) user later accesses mounted file system and sd driver
      receives a "2,4,2" error

Doug Gilbert

Re: Maxtor drive doesn't wake from sleep

[Stefan Richter]

Douglas Gilbert wrote:
>>Mi wrote at linux1394-devel:
>>>I have 3 external firewire Maxtor One Touch III drives, which appear
>>>to go into a "sleep" state after between one and two hours, and cannot
>>>be accessed after that.
...
>>>I later found scsi-spin which is also able to wake up the drive:
>>>  scsi-spin -u /dev/sdd
...
>>>Note that there is absolutely nothing in the logs when the drive goes
>>>to sleep. Only errors when I try to access them:
>>>   kernel: Device sdd not ready.
>>>   kernel: end_request: I/O error, dev sdd, sector 0
...
> I believe the sd driver should issue a START STOP UNIT
> (start) command when confronted with a "... initializing
> command required" error [sk,asc,ascq: 2,4,2]. It also
> needs to tell the block layer to go away and amuse itself
> while it polls the device until it is ready.
> 
> When I suggested this some time ago, the response was that
> whatever app was responsible for stopping the device bore
> the responsibility for spinning it up again (i.e. "not our
> problem").
> 
> The power condition state machine of ATA and SCSI differ
> see: http://www.torque.net/sg/power.html
> For devices that use SAT as a guide (and Maxtor seems to be)
> the reported problem is going to become more signficant:
>    a) external device decides to spin itself down after
>       a period of inactivity
>    b) user later accesses mounted file system and sd driver
>       receives a "2,4,2" error

Thanks a lot for the pointer. This may indeed become more relevant now, 
including SBP-2 enclosures because vendors have mixed IEEE 1394 and 
(e)SATA product families now.

Mi, I would be glad to help and implement what Douglas described. 
However there is still so much on my sbp2 todo-list that I'm not yet 
considering to go hack other drivers.
-- 
Stefan Richter
-=====-=-==- --== ===-=
http://arcgraph.de/sr/