[lug] Interface CRC error on USB connected SATA drive

Jed S. Baer blug at jbaer.cotse.net
Sat Sep 10 19:18:08 MDT 2016


On Sat, 10 Sep 2016 13:26:34 -0600
Lee Woodworth wrote:

> There are two cache levels I am thinking about, the OS buffer cache and
> the drive cache. Somebody that knows more than me could tell you for
> sure, but I think modified buffer-cache pages get written to the target
> backing store but do not immediately get erased/released. Unmodified
> pages get overwritten as needed. E.g. vmstat shows 431MB in the buffer
> cache of file server here after 6 hours of minimal activity. Drive
> cache policy is hardware dependent which is why I use a power-off when
> possible to ensure its empty.

Yeah, the OS cache can be flushed using fsync. I think we just have
to trust the HD manufacturer for the drive cache. But I do get the point
of making sure that I'm reading from what got written to disk, and not
from any cache.

> Since you are using a themaltake dock, it is possible you are using the
> uas driver even if it is USB 2. I would see if there is a uas directory
> in somewhere /sys and whether it has a device symlink/directory.

Currently, using uas. I'll make a to-do item to see about changing that.
I assume I have to do some module blacklisting, and/or some other
incantations.

> Looks like your errors are not simple ones to diagnose. Maybe A/B
> testing using the same USB port+cable for the dock vs. another known
> good external enclosure would tell you whether the dock is the issue.

Well, absent further evidence, I think I'm going to go with what I found
initially, which is that it's probably some signaling or data on the
wire, and an intermittent failure maybe from stuffing 167GB through the
interface as fast as it could be made to go. And that the driver retried
and it worked.

I believe I have everything fixed now so that errors will get noticed
if/when they happen too. I guess I sorta understand why a distro would
not include various "server" type things, such as smart, in a
desktop-target distribution. But then if things do go wrong, well, I
guess they just do.

Thank you for the suggestions.


More information about the LUG mailing list