Re: 2.4.19-pre3 udma ide hang with heavy disk activity

[Geoffrey Hoff]

I just took a look at these two patches and they are already
incorporated in linux-2.4.19-pre5 which gives me the same results as
2.4.19-pre3.  Do you have any other suggestions?

Thanks to everyone who works on the IDE subsystem.  I'm sure someone
will point me to a solution in no time flat.

* Andre Hedrick <[e-mail]> [020331 16:38]:
> 
> Please apply these which have been submitted to Alan Cox but originate
> from Vojtech.
> 
> Cheers,
> 
> Andre Hedrick
> LAD Storage Consulting Group
> 
> On Sun, 31 Mar 2002, Geoffrey Hoff wrote:
> 
> > On kernels 2.4.19-pre3 and later I get a disk hang any time I generate
> > heavy disk activity.  My test is a quick script that untars 2.4.0 and
> > then applys patch up to 2.4.19 and repeats.  With kernels 19-pre3
> > through 19-pre5, I can usually get through about the third patch and then
> > my hard drive activity led goes solid and I can no longer access any
> > disk.  This is happening specifically with a VIA controller (vt82c596b
> > rev 22) with a Maxtor 91536U6.  Attached are part of my dmesg,
> > proc/ide/via, hdinfo of /dev/hda and part of my .config.  If I use
> > hdparm to set the drive to mdma2 in 19-pre3 I never see any problems.  It
> > only happens in any udma mode.  I compiled 19-pre3 with the via
> > driver disabled, but the problem continues.  I also tried 19-pre5 as I
> > looked and saw that it contains more ide updates, but I get the same
> > results.
> > 
> > In 19-pre2 and previous kernels, If the drive is running in UDMA66, I
> > ocassionally get checksum errors and retries so I usually on boot put
> > the drive in UDMA33 mode.  I have never seen any error of any kind in
> > this mode.  I have tried modes udma0, 2, and 4 on 2.4.19-pre3 and they
> > all eventually hang.  When this hang occurs, I get no kernel messages at
> > all.  I set dmesg to level 8 so I should see any message on the console.
> > If it weren't for ext3 and reiserfs, I would be very tired of fscks by
> > now.  If there is anything that I can try to help narrow this problem
> > down, please let me know.