Xenix device-drivers/DMA problem

[Wm. E. Davidsen Jr]

In article <649 at ocsmd.ocs.com> motti at ocsmd.ocs.com (Motti Bazar) writes:

| Second and more problematic, I can leave with multiple calls (it will
| be slower). Then What I'm doing is buffering things in a buffer that I
| allocated in the driver. The problem is that when I go and call the DMA
| setup routine (dma_param(....)) and then go to sleep (sleep(....)) I see
| the operation (drive busy light goes on) but then I get a nice panic
| message telling me about a parity error and the machine is dead with
| the hard-disk busy light on forever. Somebody have any idea ?????

  I have seen this on a number of machines when I was evaluating 386's.
I'm sorry to say that it usually comes when the following occur:
	hard disk i/o
	DMA to floppy/tape/whatever
	CPU in 32 bit mode

  The last is important. IBM diagnostics and MS-DOS just don't cause the
problem. My *guess* is that the memory is on the ragged edge and loses a
refresh causing a parity error. My *experience* tells me that changing
the memory to faster memory (or just other memory) frequently helps,
slowing the CPU sometimes helps.

  I don't know what your machine is, but a lot of clones seem to be
running on the edge of memory spec for 32 bit. Dell redesigned their
motherboard between the 310 and 325 series, and replaced a MB for us,
due to just this type of problem. They admitted and fixed the problem, a
lot of vendors call it software because their diagnostics run just fine.
-- 
	bill davidsen		(wedu at crd.GE.COM)
  {uunet | philabs}!steinmetz!crdos1!davidsen
"Stupidity, like virtue, is its own reward" -me