On Tue, 14 Sep 1999, Tom Livingston wrote:
> note that actual PCI setup is:
> AGP Slot: Trident 975 (IRQ disabled)
> PCI 1 PDC20246: IDE controller on PCI bus 00 dev 78 hdm/n/o/p
> PCI 2 PDC20246: IDE controller on PCI bus 00 dev 68 hdi/j/k/l
> PCI 3 PDC20246: IDE controller on PCI bus 00 dev 58 hde/f/g/h
> PCI 4 Empty
> PCI 5 3com 3c905b 10/100 ethernet (IRQ shared with PCI 5)
>
> When I moved the PDC in slot 3 to slot 4, hde/f/g/h were unavailable, driver
> wasn't able to read them. This is apparently because of the 3com card
> sharing the IRQ. Note that any card I put in PCI 4 won't work at this
> point, a second ethernet card in PCI 4 causes a lockup within a few minutes.
This I expected and you have verified with a repeat.
> So I pulled the ethernet card, so I was pci1: pdc, pci2: pdc, pci3: empty,
> pci4: pdc, pci5: empty. I ran the same tests, and I still saw corruption
> when reading from the drive on the HPT-366 whether or not in was in the raid
> array.
This is not good...
At this point I am lost for answers. The only options are to vary the
FIFO size limits for with IRQ's are generated and serviced or prepare fast
leading IRQ's that are pre-issued.
> You did notice that I am using maxtor 9xxx series UDMA hard drives, right?
> I was worried that the issues with the "broken ASIC" were popping up.
It is a pure hardware timing issue. HighPoint gave me the ATA-33 drive
discriptions that fail.
> You may have noticed my report of the same system using IDE & SMP being
> extremely unstable, it will crash within minutes of heavy load... Do you
> think these problems are related?
Nope, just using single channel onboard.
Under heavy kernel build loads with extra overhead, it is clean.
Andre Hedrick
The Linux IDE guy
