ICH7 sata-ahci + software raid warning

ICH7 sata-ahci + software raid warning

[Christian Pernegger]

I'm (still) trying to setup a md array on the ICH7 SATA controller of
an Intel SE7230NH1-E with 4 WD5000YS disks.

On this controller (in ahci mode) I have not yet managed to get a disk
mark as failed.

- a bad cable just led to hangs and timeouts
- pulling the power on one of the SATA drives (while the system is running)
  had the exact same result.

Apparently the error handling up to 2.6.17 is so buggy that it never
returns an error that might cause md to fail the disk. (Bad blocks
maybe, can't test that.)

Regards,

C.

Re: ICH7 sata-ahci + software raid warning

[Justin Piszcz]

Why not just --set-faulty or --fail with mdadm?

On Mon, 10 Jul 2006, Christian Pernegger wrote:

> I'm (still) trying to setup a md array on the ICH7 SATA controller of
> an Intel SE7230NH1-E with 4 WD5000YS disks.
>
> On this controller (in ahci mode) I have not yet managed to get a disk
> mark as failed.
>
> - a bad cable just led to hangs and timeouts
> - pulling the power on one of the SATA drives (while the system is running)
> had the exact same result.
>
> Apparently the error handling up to 2.6.17 is so buggy that it never
> returns an error that might cause md to fail the disk. (Bad blocks
> maybe, can't test that.)
>
> Regards,
>
> C.
> -
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Re: ICH7 sata-ahci + software raid warning

[Francois Barre]

2006/7/11, Justin Piszcz <jpiszcz@lucidpixels.com>:
> Why not just --set-faulty or --fail with mdadm?
>

If you answer to "Bad blocks maybe, can't test that.", I'll say AFAIK,
--set-faulty will only stay at the md layer, and won't go through the
underlying driver (sata here, ata elsewhere, ...), so this will not
help diagnose the driver behaviour on block read problem.

If you answer to the fact that md does not mark the drive faulty, I
guess Christian was refering to an automatic to mark the disk faulty,
not a manual way, in order, for exemple, to trigger a resync with a
spare drive just after a drive failure has been detected (and not only
after the admin found it).

As an ugly workaround, a daemon script could look at the current drive
status regularly (in /sys, in /proc or trying to read /dev/sd?), and
mark the drive faulty on error...

Re: ICH7 sata-ahci + software raid warning

[Justin Piszcz]

On Tue, 11 Jul 2006, Francois Barre wrote:

> 2006/7/11, Justin Piszcz <jpiszcz@lucidpixels.com>:
>> Why not just --set-faulty or --fail with mdadm?
>> 
>
> If you answer to "Bad blocks maybe, can't test that.", I'll say AFAIK,
> --set-faulty will only stay at the md layer, and won't go through the
> underlying driver (sata here, ata elsewhere, ...), so this will not
> help diagnose the driver behaviour on block read problem.
>
> If you answer to the fact that md does not mark the drive faulty, I
> guess Christian was refering to an automatic to mark the disk faulty,
> not a manual way, in order, for exemple, to trigger a resync with a
> spare drive just after a drive failure has been detected (and not only
> after the admin found it).
>
> As an ugly workaround, a daemon script could look at the current drive
> status regularly (in /sys, in /proc or trying to read /dev/sd?), and
> mark the drive faulty on error...
>

For ICH5 and all other controllers (non-AHCI) of course, I've always seen 
md mark it faulty on a bad disk/sector/etc.

I thought you were just trying to mark it faulty manually for the purpose 
of rebuilding it.

Justin.

Re: ICH7 sata-ahci + software raid warning

[Christian Pernegger]

> For ICH5 and all other controllers (non-AHCI) of course, I've always seen
> md mark it faulty on a bad disk/sector/etc.

ICH7 (ahci) does not. At least not a whole disk dieing, don't know
about bad sectors.

> I thought you were just trying to mark it faulty manually for the purpose
> of rebuilding it.

Ah, no I wasn't :)

C.

Re: ICH7 sata-ahci + software raid warning

[Christian Pernegger]

> Why not just --set-faulty or --fail with mdadm?

This is not about testing, but about real errors.

1) I depend on the md layer to tell me if a disk is faulty ... failing
the disk manually would only help if I already knew the disk was
faulty and just wanted to remove it.

2) If the affected array has root on it it's game over: no logins of
any sort, no issuing commands, no manual failing, since any process
that accesses a fs will hang.

C.