IDE-DMA on VIA chipset issues

[beng]

Not Gentoo-specific. Seems like some advanced members, so thought I'd post
anyway. No distribution elitism here, eh?

VIA-VT82C686A EIDE-chipset based mobo (Chaintech 6AJA4)
WD UDMA66 HD on primary IDE -- 80-pin cable. No jumpers (master)
20x CDRom on secondary - master
Peanut Linux 9.2 /2.4.20 kernel w/prempt & lowlatency patches
256 MB SDRAM
2 duplicate Linux installations on umsdos and reiserfs partitions

Seems Linux having trouble w/the IDE controller using DMA. Messages
disappear if DMA disabled, but doing so is nearly unacceptable. Winders
95 & NT4 are not having any apparent HD problems w/DMA enabled. Next to
latest VIA 4in1 drivers for them (latest is supposedly a no-no). HD is in
excellent condition according to WD tools. No bad sectors.

Example of messages in dmesg:

hda: dma_intr: error=0x84 { DriveStatusError BadCRC }
hda: dma_intr: status=0x51 { DriveReady SeekComplete Error }
hda: dma_intr: error=0x84 { DriveStatusError BadCRC }
hda: dma_intr: status=0x51 { DriveReady SeekComplete Error }
hda: dma_intr: error=0x84 { DriveStatusError BadCRC }

and even a rare:

hda: ide0 reset: success

during extended HD I/O.

No file corruption in either FS, but during intensive HD I/O, there are
noticable lags & huge numbers of the above messages. Swapfile is NOT being
used during these episodes.

Google search brings up nothing useful. From the default settings, here is
what I have tried.

1. Used WD tools to set WD drive to UDMA33 only. Seemed to decrease message
occurence slightly. Kept it @ UDMA33.
2. Used hdparm to vary settings. Most severe change (still w/DMA enabled):

hdparm -c0u0 /dev/hda

(disables 32-bit I/O, interrupt unmasking) seemed to cut messages down somewhat during normal usage. However, same symptoms as default during intense I/O -- even the occassional ide0 reset message. Changing the CDRom's setting w/hdparm, mounting the CDRom, or removing it doesn't affect hda's messages.

3. In BIOS, disabled IDE prefetch. No change.

4. Some settings in /proc/ide/ide0/hda, like smart_thresholds & values may be
relavent, but catting those items brings up numbers I can't make sense of.

Toughy, huh?

[hardware_chump]

From a hardware perspective:

o Try changing out your IDE cable with another one.
o Try a shorter cable.
o Switch primary and secondary around (CDROM on primary and HD on secondary).
o Try rerouting your cable around noisy areas of the board and system (esp. power supplies). I've seen noisy switching power supplies cause UDMA mode failures.

-hc

[beng]

Missed some info in 1st post --

Tried 2 different, NEW 80-pin cables -- no difference.

Same drive, using either cable installed in my Sis5513 IDE-chipset box (runs at udma33) shows no such errors as above. No slowdowns during heavy I/O. hdparm -t /dev/hda shows the same results as in this VIA-chipset board -- 'bout 24mb/sec.

I have kernel-stuff for both boards' hardware enabled so I can run same kernel on both boxes.

[hardware_chump]

I forgot to add that Windows XP and presumably Via 4-in-1 IDE drivers will "downshift" when they run into errors. A true test is to do a clean install of Winows ME (maybe Windows 95 will work). Then, using default drivers, check the DMA box in the primary IDE. Reboot windows and time how long it takes. Then clear the DMA box. Reboot and time how long it takes. If it takes longer for windows to boot with DMA checked, you may be having DMA problems.

-hc

[Malakin]

Try updating the bios on the motherboard and try a stock 2.4.20 kernel without any patches.

[beng]

Got latest BIOS. Tried the 2.4.20 kernel w/o patches -- greeted w/an avalanche of syslog errors.

The error "messages" weren't necessarily important, but the lagging during heavy,
extended I/O was. So far w/what I've done, it doesn't lag anymore. What did I do.....

Removed the VIA82C686-chipset "support?" from the kernel! Kernel-log complains
about no driver, but no more lagging or error messages. Just running on generic ide drivers. I set hdparm parameters in rc.local. I do receive a new one-time message in dmesg:

blk: queue c028d9c4, I/O limit 4095Mb (mask 0xffffffff)

Don't know what that means. Time will tell...

[wylbur]

has anyone found out more about this? I've got a similar problem. Lots of weird and intermittant probs at the software level. Then I booted from a livecd and ran fsck.reiserfs which died while replaying the journal:

Checking internal tree../ 35 (of 97)/60 (of 156)hda: dma_intr: status=0x51 { DriveReady SeeksComplete Error }

<followed by a bunch more kernel messages of this kind>

The problem has occurred looks like a hardware problem.
Check your hard drive for badblocks.

bread: Cannot read the block (96594)

Aborted


/sbin/badblocks goes on and on reporting errors like the dma_intr above. I'm running gentoo-sources 2.4.20 with not much exciting in the make menuconfig. The machine will boot to this disk, but it won't emerge rsync (/usr/portage is on top of some of the weird sectors, I guess).

Dell's BIOS HD check utility reports that everything is ok with this drive (a Maxtor 60Gb IDE).

Ideas?! thanks - PB.

[gibbo]

There seems to be problems with some VIA chipset based motherboards and certain combinations of IDE devices.

I has the following setup:

IDE0: Master IBM DJNA-351520
Slave WD300BB-00AUA1

The BIOS would sometimes not recognise the IBM drive and wouldn't boot. A hard reset would fix this. At other times, I would get errors in syslog complaining about the WD slave drive.

I tried different kernel versions, hdparm settings but nothing fixed it completely. Finally I installed an additional CMD649 PCI IDE controller card and shifted both the HDD's to this. I've had no problems since then. I still use the VIA controller for a DVDROM & a CD-RW with no problems.

YMMV.

Nick.