ESP DMA and sbus (was: HyperSPARC patches (?))

[Martin Habets <errandir_news@mph.eclipse.co.uk>]

Hi,

I can reproduce this problem on my SS20 UP (TMS390Z50) using a tape
drive. I've been testing 2.6.13 and 2.6.16-rc2, with the tape on
ESP100A-FAST and ESP236-FAST. I test using tar and the 'btape ... test'
commands.
One thing I have noticed is that I only get the error while reading,
so far it has never happened while writing.

The exact error I'm getting is:
esp1: DMA error 4004430e
esp1: Resetting scsi bus
esp1: SCSI bus reset interrupt
st0: Error 80000 (sugg. bt 0x0, driver bt 0x0, host bt 0x8).
esp1: Warning, live target 0 not responding to selection.
st0: Error 60000 (sugg. bt 0x0, driver bt 0x0, host bt 0x6).
esp1: Warning, live target 0 not responding to selection.

This has only happened to me twice using just hard disk activity, under
extreme circumstances. Extreme meaning after it had run out of both
memory and swap space.
My 2 hard disks are attached to the ESP100A-FAST. When I attach my tape
there and trigger the problem, the hards disks also get the "not
responding to selection" errors. So the whole controller is affected.

Some questions:
1) Kernel bug 6344 mentiones the last working kernel is 2.4.32.
   Does anyone have this working with a later kernel?
2) Are there known changes in the dma/sbus area? Is there any merrit
   trying later kernels?
3) Is there any way to correct the situation? I have to reboot it to
   start from fresh. Removing/adding the device using /proc/scsi/scsi
   does not help.

I plan to try the eh locking patch posted next, and maybe something like
2.6.8 after that if it is still not working. Unless someone has a better
proposal.

Thanks,
-- 
Martin