Today, I recompiled that same kernel, but with different drivers built-in
(nothing related to SCSI, but that did change the kernel size and boot
timing), I'm getting a similar error as above (referenced scb not valid)
followed by a panic (using the 6.2.1 driver).
Do you know if anything changed in the PCI code that might affect
the initialization of PCI devices? It is almost as if some of our
PCI register writes are not being honored (e.g. prefetching has
suddenly been enabled for this card even though the prefetch bit
if off in all of its BARs). 6.2.1 has been in the kernel for some
time and was not changed in going from .10 -> .12.
--
Justin
** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/
Hi,
I can confirm this for my machine. It seems recompiling and changing
unrelated things does seem to impact the performance of the driver.
Since changing the memory footprint seems to impact the driver, I was
wondering about a few things:
- Is hardware "DMA" being used to swap in and out scb blocks of memory?
If so, are there scb memory "alignment" issues at work here when doing DMA?
- what about issues with locking when doing the DMA to prevent other reads
and or writes colliding? Ben's trouble with SMP machines might be
indicative of something here?
- data cache flushing issues?
Is there any specific code we should focus on in the driver when trying to
figure this thing out?
Thanks,
Kevin
On October 15, 2001 11:33, Justin T. Gibbs wrote:
quoted
Today, I recompiled that same kernel, but with different drivers
built-in (nothing related to SCSI, but that did change the kernel size
and boot timing), I'm getting a similar error as above (referenced
scb not valid) followed by a panic (using the 6.2.1 driver).
Do you know if anything changed in the PCI code that might affect
the initialization of PCI devices? It is almost as if some of our
PCI register writes are not being honored (e.g. prefetching has
suddenly been enabled for this card even though the prefetch bit
if off in all of its BARs). 6.2.1 has been in the kernel for some
time and was not changed in going from .10 -> .12.
** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/