sata_mv important note

sata_mv important note

[Jeff Garzik]

For sata_mv users, you should be aware of three things:

1) The Marvell driver is experimental, and not yet considered ready for 
production use.  As Kconfig notes: HIGHLY EXPERIMENTAL.

2) There are PCI Message Signalled Interrupt (MSI) problems that are not 
yet diagnosed.  Workaround is to disable MSI.

3) There are still some errata that are not yet implemented.  Thus, for 
some systems, you may see either lockups or data corruption.  That's the 
price of running a HIGHLY EXPERIMENTAL driver.

Rest assured that all problem reports are read, even if you don't 
receive a reply.  After enough feedback is received, the exact problem 
becomes more clear.

	Jeff

Re: sata_mv important note

[Sebastian Kuzminsky]

Jeff Garzik <jgarzik@pobox.com> wrote:
> For sata_mv users, you should be aware of three things:
> 
> 1) The Marvell driver is experimental, and not yet considered ready for 
> production use.  As Kconfig notes: HIGHLY EXPERIMENTAL.

Right, understood.

Can anyone recommend a 4xSATA (or more) controller on a PCI-X card that
_is_ ready for production use?  eSATA ports are prefered but not required.
Support for hotplug, power management (spindown) and SMART is prefered
but not required.

Anyone?


> Rest assured that all problem reports are read, even if you don't 
> receive a reply.  After enough feedback is received, the exact problem 
> becomes more clear.

Thanks Jeff!


-- 
Sebastian Kuzminsky

Re: sata_mv important note

[Arkadiusz Miskiewicz]

On Tuesday 17 January 2006 17:24, Sebastian Kuzminsky wrote:
> Jeff Garzik <jgarzik@pobox.com> wrote:
> > For sata_mv users, you should be aware of three things:
> >
> > 1) The Marvell driver is experimental, and not yet considered ready for
> > production use.  As Kconfig notes: HIGHLY EXPERIMENTAL.
>
> Right, understood.

I'm using:

03:03.0 SCSI storage controller: Marvell Technology Group Ltd. MV88SX6041 
4-port SATA II PCI-X Controller (rev 07)

but with http://www.keffective.com/mvsata/FC3/mvSata-3.4.2a-patched.tbz driver
and it works nicely (+ 2.8GHz Xeon HT, smp kernel). I was quite suprised to 
see that there are no problems with it in typical usage (while I'm sure that 
this driver is far away from kernel standards).

-- 
Arkadiusz Miśkiewicz                    PLD/Linux Team
http://www.t17.ds.pwr.wroc.pl/~misiek/  http://ftp.pld-linux.org/

Re: sata_mv important note

[Mark Lord]

Sebastian Kuzminsky wrote:
>
> Can anyone recommend a 4xSATA (or more) controller on a PCI-X card that
> _is_ ready for production use?  eSATA ports are prefered but not required.
> Support for hotplug, power management (spindown) and SMART is prefered
> but not required.

The Pacific Digital QStor cards come in PCIX variants,
and support SATA2.  Hotplug et al is supported in the hardware,
but not in the Libata driver.  ATAPI is not supported at all.

Cheers

Re: sata_mv important note

[Sebastian Kuzminsky]

Arkadiusz Miskiewicz <arekm@pld-linux.org> wrote:
> On Tuesday 17 January 2006 17:24, Sebastian Kuzminsky wrote:
> > Jeff Garzik <jgarzik@pobox.com> wrote:
> > > For sata_mv users, you should be aware of three things:
> > >
> > > 1) The Marvell driver is experimental, and not yet considered ready for
> > > production use.  As Kconfig notes: HIGHLY EXPERIMENTAL.
> >
> > Right, understood.
> 
> I'm using:
> 
> 03:03.0 SCSI storage controller: Marvell Technology Group Ltd. MV88SX6041=20
> 4-port SATA II PCI-X Controller (rev 07)
> 
> but with http://www.keffective.com/mvsata/FC3/mvSata-3.4.2a-patched.tbz driver
> and it works nicely (+ 2.8GHz Xeon HT, smp kernel). I was quite suprised to
> see that there are no problems with it in typical usage (while I'm sure that
> this driver is far away from kernel standards).

I'm using:

0000:02:01.0 IDE interface: Marvell Technology Group Ltd. MV88SX6081 8-port SATA II PCI-X Controller (rev 09)

I'm running the stock 2.6.15 kernel & the in-kernel driver.  I have four
disks on this controller.  The controller and disks seem perfectly stable,
I've been running four parallel "badblocks -n" processes (one on each
disk) for almost 5 days now.  Using the disks as PVs in LVM works fine,
and building a RAID-6 out of them also works fine.

But when I build a RAID-6 out of them, and use the array as a PV
for LVM, the system locks up within seconds (no errors, no sysrq,
no CapsLock-blinky, no network-pingy).  This behavior is perfectly
repeatable.

The problem goes away and everything works if I turn on all the debugging
options in the kernel config (but I dont get any debug output from
the kernel).

Arkadiusz, if possible, please see if you can replicate my hang.


-- 
Sebastian Kuzminsky

Re: sata_mv important note

[Mark Lord]

Sebastian Kuzminsky wrote:
>
> I'm using:
> 
> 0000:02:01.0 IDE interface: Marvell Technology Group Ltd. MV88SX6081 8-port SATA II PCI-X Controller (rev 09)
> 
> I'm running the stock 2.6.15 kernel & the in-kernel driver.  I have four
> disks on this controller.  The controller and disks seem perfectly stable,
> I've been running four parallel "badblocks -n" processes (one on each
> disk) for almost 5 days now.  Using the disks as PVs in LVM works fine,
> and building a RAID-6 out of them also works fine.
> 
> But when I build a RAID-6 out of them, and use the array as a PV
> for LVM, the system locks up within seconds (no errors, no sysrq,
> no CapsLock-blinky, no network-pingy).  This behavior is perfectly
> repeatable.
> 
> The problem goes away and everything works if I turn on all the debugging
> options in the kernel config (but I dont get any debug output from
> the kernel).
> 
> Arkadiusz, if possible, please see if you can replicate my hang.

I will try and reproduce and fix this issue when my Marvell RAID system
arrives here in a week or two.

Cheers

Re: sata_mv important note

[Bartlomiej Zolnierkiewicz]

On 1/17/06, Sebastian Kuzminsky <seb@highlab.com> wrote:
> Arkadiusz Miskiewicz <arekm@pld-linux.org> wrote:
> > On Tuesday 17 January 2006 17:24, Sebastian Kuzminsky wrote:
> > > Jeff Garzik <jgarzik@pobox.com> wrote:
> > > > For sata_mv users, you should be aware of three things:
> > > >
> > > > 1) The Marvell driver is experimental, and not yet considered ready for
> > > > production use.  As Kconfig notes: HIGHLY EXPERIMENTAL.
> > >
> > > Right, understood.
> >
> > I'm using:
> >
> > 03:03.0 SCSI storage controller: Marvell Technology Group Ltd. MV88SX6041=20
> > 4-port SATA II PCI-X Controller (rev 07)
> >
> > but with http://www.keffective.com/mvsata/FC3/mvSata-3.4.2a-patched.tbz driver
> > and it works nicely (+ 2.8GHz Xeon HT, smp kernel). I was quite suprised to
> > see that there are no problems with it in typical usage (while I'm sure that
> > this driver is far away from kernel standards).
>
> I'm using:
>
> 0000:02:01.0 IDE interface: Marvell Technology Group Ltd. MV88SX6081 8-port SATA II PCI-X Controller (rev 09)
>
> I'm running the stock 2.6.15 kernel & the in-kernel driver.  I have four
> disks on this controller.  The controller and disks seem perfectly stable,
> I've been running four parallel "badblocks -n" processes (one on each
> disk) for almost 5 days now.  Using the disks as PVs in LVM works fine,
> and building a RAID-6 out of them also works fine.
>
> But when I build a RAID-6 out of them, and use the array as a PV
> for LVM, the system locks up within seconds (no errors, no sysrq,
> no CapsLock-blinky, no network-pingy).  This behavior is perfectly
> repeatable.

Have you tried using "nmi_watchdog=1" kernel parameter?

> The problem goes away and everything works if I turn on all the debugging
> options in the kernel config (but I dont get any debug output from
> the kernel).

Could you find the specific option which makes the hang go away?

Bartlomiej

Re: sata_mv important note

[Sebastian Kuzminsky]

Bartlomiej Zolnierkiewicz <bzolnier@gmail.com> wrote:
> On 1/17/06, Sebastian Kuzminsky <seb@highlab.com> wrote:
> > 0000:02:01.0 IDE interface: Marvell Technology Group Ltd. MV88SX6081 8-port SATA II PCI-X Controller (rev 09)
> >
> > I'm running the stock 2.6.15 kernel & the in-kernel driver.  I have four
> > disks on this controller.  The controller and disks seem perfectly stable,
> > I've been running four parallel "badblocks -n" processes (one on each
> > disk) for almost 5 days now.  Using the disks as PVs in LVM works fine,
> > and building a RAID-6 out of them also works fine.
> >
> > But when I build a RAID-6 out of them, and use the array as a PV
> > for LVM, the system locks up within seconds (no errors, no sysrq,
> > no CapsLock-blinky, no network-pingy).  This behavior is perfectly
> > repeatable.
> 
> Have you tried using "nmi_watchdog=1" kernel parameter?

I just tried this and it hung again, with nothing in the logs or on
the console.

Pretty wierd.

I just had another hard lockup with sata_mv -> Raid-6, but without LVM.
This is new for me, first time I've seen it lock up without LVM.  I was
resyncing the raid array and running 'badblocks -svn' on it (/dev/md1)
at the same time, and it locked.

I'm going to shelve the Marvell 6081 controller for a while, and go buy
something else...


-- 
Sebastian Kuzminsky