Re: new Areca driver in 2.6.16-rc6-mm2 appears to be broken

From: erich
Date: Thu Apr 20 2006 - 05:34:04 EST


Dear Dear Jens Axboe,

I will try debench on the fs with larger xfer size when I enter my Lab. next time.
In the previous testing I do iometer and bonnie++ bench with ext3 riserfs file system.
I have test the transfer size from 2K to 5M with random/sequence read/write each term 2 hours.
When those three testing platforms all have completion.
There were not any one message appear as this.
I have done same testing with ext2 512 sectors and all done well.
I will find the difference of each LBA value , request length between 512 sectors and 4096 sectors.

Best Regards
Erich Chen
----- Original Message ----- From: "Jens Axboe" <axboe@xxxxxxx>
To: "erich" <erich@xxxxxxxxxxxx>
Cc: <dax@xxxxxxxxxxxx>; <billion.wu@xxxxxxxxxxxx>; "Al Viro" <viro@xxxxxxxxxxxxxxxx>; "Andrew Morton" <akpm@xxxxxxxx>; "Randy.Dunlap" <rdunlap@xxxxxxxxxxxx>; "Matti Aarnio" <matti.aarnio@xxxxxxxxxxx>; <linux-kernel@xxxxxxxxxxxxxxx>; "James Bottomley" <James.Bottomley@xxxxxxxxxxxx>; "Chris Caputo" <ccaputo@xxxxxxx>
Sent: Thursday, April 20, 2006 4:23 PM
Subject: Re: new Areca driver in 2.6.16-rc6-mm2 appears to be broken



(don't top post!)

On Thu, Apr 20 2006, erich wrote:
Dear Dear Jens Axboe,

Thanks for your notification and advice.
Areca's firmware has max sg entries of 38 limit.
In my debug driver I had add this condition check.
But no one request more than 38 sg.
Both transfer length all have a lot of requests equal with 38 sg.
But why it ocur only at 4096 sectors?
If the /sys/block/sda/queue/max_sectors_kb equal 256 all operation running
well.
But if I modify it more than 256, the bug appeared.
I will do more research about why there were a lot of requests equal with
38 sg in all file system.

It was just a suggestion, the bug might very well be just the size of
the transfer itself and nothing SG related. All I can say for sure is
that I'd be very surprised if this fs corruption isn't due to the
hardware mangling the data for large transfers.

And only it ocur at the volume that format with mkfs.ext2.

Most likely a coincidence, try running eg dbench or other stress tests
on the fs with larger xfer size and I'm sure it'll corrupt eventually as
well.


--
Jens Axboe


-
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/