Prev: x86: Use EOI register in io-apic on intel platforms
Next: sysfs: Update s_iattr on link and unlink.
From: Kristleifur Daðason on 4 Dec 2009 14:10
On Fri, Dec 4, 2009 at 7:00 PM, Mikael Abrahamsson <swmike(a)swm.pp.se> wrote:
>
> On Thu, 3 Dec 2009, Kristleifur Dađason wrote:
>
>> On Thu, Dec 3, 2009 at 9:17 PM, Mikael Abrahamsson <swmike(a)swm.pp.se> wrote:
>>>
>>> On Wed, 4 Nov 2009, Andre Tomt wrote:
>>>
>>>> Identical issues with the AOC-SASLP-MV8 card with 8 Seagate Barracuda
>>>> 7200.11 1.5TB drives using the mvsas driver. Creating a md raid array seems
>>>> to trigger it nearly instantly, however other loads do not trigger it
>>>> as fast (if at all, only tested briefly with paralell dd).
>>>
>>> Would just like to chime in with a "me too" on this. I've tried both ubuntu
>>> 9.10 kernel (2.6.31) and their 10.04 pre-alpha (guess it's 2.6.32-rc8) and
>>> mvsas + AOC-SASLP-MV8 is pretty much unusable. If I start md creation it
>>> stalls immediately and gets nowhere, last time I tried it md claimed it had
>>> written 160 blocks then it stalled (all drive LEDs on the controller is
>>> constantly lit).
>>>
>>> If I boot without drives and then hot-plug the drives, it immediately
>>> oops:es.
>>>
>>> Any code/fixes I can try, this is a lab machine so I can easily do whatever
>>> needed with it...
>>>
>>
>> Hi,
>>
>> I had a bit of trouble finding some patches that were released by Andy Yan at
>> Marvell on November 9th, so here is a link:
>>
>> 1/7: http://kerneltrap.org/mailarchive/linux-scsi/2009/11/9/6558753
>
> They applied cleanly to the ubuntu 10.04 2.6.32 kernel source package but when compiling I received an error about "sas_change_queue_depth" having wrong amount of arguments:
>
> -               sas_change_queue_depth(sdev, MVS_QUEUE_SIZE);
> +               sas_change_queue_depth(sdev, MVS_QUEUE_SIZE, SCSI_QDEPTH_DEFAULT);
>
> So I removed that change and then things compiled, booted properly, but just froze when I plugged in the disks (including some garbled graphics on my X screen). That patch set doesn't seem to be ok.

(This is resent as my first email was HTML formatted and rejected by
the list. Edited and amended also.)

Cheers for trying them out! For the record, I think they were made
against 2.6.31, perhaps the original 2.6.31 rather than later 2.6.31.x
releases.

And also, I recommend using the mainline kernel for Ubuntu, especially
when trying these patches. It's available here:
http://kernel.ubuntu.com/~kernel-ppa/mainline/ - including source
packages.

-- Kristleifur
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Mikael Abrahamsson on 5 Dec 2009 06:00
On Fri, 4 Dec 2009, Kristleifur Da�ason wrote:

> Cheers for trying them out! For the record, I think they were made
> against 2.6.31, perhaps the original 2.6.31 rather than later 2.6.31.x
> releases.

It doesn't apply cleanly against vanilla 2.6.31, it does however apply
cleanly against vanilla 2.6.32, but with the following compile error:

drivers/scsi/mvsas/mv_sas.c: In function �mvs_slave_configure�:
drivers/scsi/mvsas/mv_sas.c:417: error: �SCSI_QDEPTH_DEFAULT� undeclared
(first use in this function)
drivers/scsi/mvsas/mv_sas.c:417: error: (Each undeclared identifier is
reported only once
drivers/scsi/mvsas/mv_sas.c:417: error: for each function it appears in.)
drivers/scsi/mvsas/mv_sas.c:417: error: too many arguments to function
�sas_change_queue_depth�
make[4]: *** [drivers/scsi/mvsas/mv_sas.o] Error 1

When removing this extra option, it compiles with vanilla 2.6.32 just
fine. I crashes the same way as the ubuntu kernel when hot-plugging a disk
as well (this time I did it without gdm/x running and it oopses, but
unfortunately the highest part of the oops scrolled off). It's 100%
reproducible anyway, so basically the controller is a no-go in 2.6.31.6
and 2.6.32 both vanilla and ubuntu versions, both without and including
the nov09 mvsas patch (at least the original version of the patch as
posted). Different failure scenarios though... With the stock kernel the
controller seems to get stuck in writing and nothing more happens but at
least the machine doesn't crash and burn like it does with the
mvsas-patch.

--
Mikael Abrahamsson email: swmike(a)swm.pp.se
From: Mikael Abrahamsson on 10 Dec 2009 14:30
On Sat, 5 Dec 2009, Mikael Abrahamsson wrote:

> When removing this extra option, it compiles with vanilla 2.6.32 just
> fine. I crashes the same way as the ubuntu kernel when hot-plugging a
> disk as well (this time I did it without gdm/x running and it oopses,
> but unfortunately the highest part of the oops scrolled off). It's 100%
> reproducible anyway, so basically the controller is a no-go in 2.6.31.6
> and 2.6.32 both vanilla and ubuntu versions, both without and including
> the nov09 mvsas patch (at least the original version of the patch as
> posted). Different failure scenarios though... With the stock kernel the
> controller seems to get stuck in writing and nothing more happens but at
> least the machine doesn't crash and burn like it does with the
> mvsas-patch.

Would it make sense for me to log this problem in bugzilla? There seems to
be a similar issue with equal hw:

<http://bugzilla.kernel.org/show_bug.cgi?id=14534>

did these fixes ever go upstream? Are they scheduled to go into the
stable rebuilds of 2.6.32 ? The patches mentioned in there, are these the
ones I already tried that didn't solve my problem?

--
Mikael Abrahamsson email: swmike(a)swm.pp.se
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Thomas Fjellstrom on 10 Dec 2009 16:10
On Thu December 10 2009, Mikael Abrahamsson wrote:
> On Sat, 5 Dec 2009, Mikael Abrahamsson wrote:
> > When removing this extra option, it compiles with vanilla 2.6.32 just
> > fine. I crashes the same way as the ubuntu kernel when hot-plugging a
> > disk as well (this time I did it without gdm/x running and it oopses,
> > but unfortunately the highest part of the oops scrolled off). It's 100%
> > reproducible anyway, so basically the controller is a no-go in 2.6.31.6
> > and 2.6.32 both vanilla and ubuntu versions, both without and including
> > the nov09 mvsas patch (at least the original version of the patch as
> > posted). Different failure scenarios though... With the stock kernel
> > the controller seems to get stuck in writing and nothing more happens
> > but at least the machine doesn't crash and burn like it does with the
> > mvsas-patch.
>
> Would it make sense for me to log this problem in bugzilla? There seems
> to be a similar issue with equal hw:
>
> <http://bugzilla.kernel.org/show_bug.cgi?id=14534>
>
> did these fixes ever go upstream? Are they scheduled to go into the
> stable rebuilds of 2.6.32 ? The patches mentioned in there, are these the
> ones I already tried that didn't solve my problem?
>

I'm really hoping Andy Yan or someone else on the project can get some time
to polish up the last patch set posted to linux-scsi. There were a few
comments regarding them, but no one has responded to them yet.

--
Thomas Fjellstrom
tfjellstrom(a)shaw.ca
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
First  |  Prev  | 
Pages: 1 2
Prev: x86: Use EOI register in io-apic on intel platforms
Next: sysfs: Update s_iattr on link and unlink.