Re: sata_nv issues with MCP51 SATA controller

From: Jon Ivar Rykkelid
Date: Thu Sep 13 2007 - 11:05:57 EST


Jeff Garzik wrote:
Jon Ivar Rykkelid wrote:

Hi, I'm resending (didn't see my first attempt appear on the maillist):



I'm having serious disk-issues when using the on-board nvidia controller
for my HDDs (My motherboard is a Gigabyte GA-N650SLI-DS4 with nvidia
chipset, cpu is intel Core2Quad)

excerpt from "lspci":
00:0d.0 IDE interface: nVidia Corporation MCP51 IDE (rev a1)
00:0e.0 IDE interface: nVidia Corporation MCP51 Serial ATA Controller
(rev a1)
00:0f.0 IDE interface: nVidia Corporation MCP51 Serial ATA Controller
(rev a1)

I have a normal IDE/P-ATA-disk attached to the "IDE"-controller and that
works fine (/dev/hda)

However, any number of disks (I have tried 2 and 4) connected to the
SATA-controller(s), will eventually fail. - See attached log (excerpt /
anything relevant from /var/log/messages)

At first, disks were REALLY unstable, but then I disabled S.M.A.R.T.
(both in BIOS and Linux), and I updated from the CentOS5 (equivalent of
RHEL5) kernel (2.6.18) to the latest (at that time) official kernel from
kernel.org:

> uname -a
Linux mirakel 2.6.22.5-custom_jir #2 SMP Thu Aug 30 22:06:21 CEST 2007
i686 i686 i386 GNU/Linux

Now it will normally take a day or two before SATA crashes, so things
are better, but still rather useless.

First error when sata_nv get into problems is always:
"exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x2 frozen"
(as shown in the attached log-file.) - when this happens to one device,
it'll almost instantly happen to the other disk attached to that
controller as well. A couple of minutes (or so) later, the disk(s)
connected to the other controller will start acting up as well (in the
same manner). - I/O freezes, and nothing helps except a reboot...

As I run a rather large (software / md) RAID-5 disk array on this server
(I'm doing a bit of video editing), every crash means a time-consuming
rebuild of the disk-array...

I have given up on the sata_nv / nvidia-controllers for the time being.
I now resort to some old PCI-connected sata-controllers which work fine
(but slow, as they are outdated and "overloaded").

So, if anyone has a good solution / suggestion / improved driver (over
the one supplied with the official 2.6.22.5-kernel) I am eager to give
it a go and see if the situation can be resolved.

does adma=0 module option do anything?

Jeff
Thanks for the suggestion, but sata_nv is not built modular in my current kernel, so "no can do" at the moment
(However, if some expert REALLY thinks this will fix things, I will CERTAINLY recompile and give it a go)

As I said before, it all works for some time (a day or two) before it crashes with the current kernel & no "S.M.A.R.T.". With my current setup I have always had the time to fully rebuild my disk-array before a new crash. - In the case of 4 disks attached to the nvidia controllers (disregarding the disks on other controllers), this means that the sata_nv-driver / controllers alone have read at least 750GB and written 250GB of data before the crash (with no resets working) - soft reboot fixes everything. - I'm pretty confident that this is a driver issue.

As Tejun Heo <htejun@xxxxxxxxx> writes "the whole controller seems to have went down at once and it's not even IRQ routing problem - resets are failing."

The error-messages / crash-symptoms were the same with SMART enabled and the original CentOS5-kernel, except that with that setup, the crashes were much more frequent.

Any help?

BR
Jon Ivar

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/