[Bulk] [CentOS] hda dma seek errors

Bryan J. Smith b.j.smith at ieee.org
Wed Aug 3 20:17:33 UTC 2005


BRUCE STANLEY <bruce.stanley at prodigy.net> wrote:
> Hi Bryan!
> I should also mentions that the drives are installed in
> removabl drive bays so that I can swap then out with other
> drives without messing with the cables.  

Ewwwww (even more).  ATA is nasty for hot-swap.
SATA's SCA-like, staggered power/transient is much better.

> I have seen very little problems of the nature discribed
> with FC2, Mandrake 8.1, or  Suse 8.2.

Most of the problems I've seen were early 2.4 and mid 2.4
(circa 2.4.18-20).

Early 2.4 issues were due to Intel ICH and ViA 82xx ATA logic
having issues with IBM (which were also Western Digital at
the time) drives.  Long story short, IBM screwed up on the
ATA spec.  Intel shared specs and got the problem fixed.  ViA
horded specs and the problem continued.

Mid 2.4, Hendrick ripped out his ATA code because of some
vendors continually violating the GPL.  I can't remember what
the resolution of that was (I think he put them back in
within a few revisions).

Bus arbitration between the "dumb" AT Attachment (ATA)
channel and the Integrated Drive Electonics (IDE) on the
drive is always a PITA for the host/OS.  Sometimes some ATA
channels just don't have registers/support, or more often,
the IDE is not to spec, and the ATA channel doesn't work as
designed (and the host/OS is left resetting the channel to
try to reconnect the two).

It was enough that I made it a general rule to deploy 3Ware
for production systems, even desktops.  Why?  Because 3Ware
is both the host/OS (ASIC/firmware) and the ATA channel, so
it can more easily "tame" troublesome IDE.  I.e., the 3Ware
ASIC/firmware is driving the IDE directly, and works in
tandem with the 3Ware ATA channel logic.
 
> I have not seen any problems when using Windows 2000 pro.

Of course, because ATA channel vendors provide their drivers,
with all that nice code negotiated under NDA, etc...
At the same time, have you ever tried the "included" NT ATA
drivers?  Not always pretty.  ;->

> Is this problem occuring more with newer versions of the
> Red Hat base kernals?

See my comments above on kernel 2.4.

> It could still be a hardware problem of some sort
> with the system.

The ATA spec is _rarely_ followed "to-the-letter."

IBM has been guilty many, many times of this (as well as
with buggy BIOS Int13h services too), which affected
Western Digital as well until the switch to Hitachi (and
IBM's sell-off to the same).

Intel's early ICH ATA for the i8xx chipsets, from the former
PIIX ATA on the earlier i4xx chipsets also caused a lot of
issues.  But Intel worked them out rather quickly.

ViA has always horded it's specs, and Linux ATA issues
plagued many of its early 82xx ATA logic, up until about 8233
or 8235.

AMD has only designed 1 ATA logic, the original one in the
751, and it is still used in the 760 as well as the AMD8111.

SiS used to be the same with its 551x ATA logic, although I
have not tried its latest chipsets (since the Socket-462).

> When I get home tonight I'll open up the box and re-seat
> the cables just to be sure that they are ok.

The exact make/model of your southbridge ATA would help --
use "lspci -v".  If it is ViA, it should be VT823x.  Get the
exact model for me and I can help you more.

The exact output of "hdparm -i /dev/hda" and "hdparm -i
/dev/hdb" would help as well.



-- 
Bryan J. Smith                | Sent from Yahoo Mail
mailto:b.j.smith at ieee.org     |  (please excuse any
http://thebs413.blogspot.com/ |   missing headers)



More information about the CentOS mailing list