i2o - problems with controllers with additional memory

[Dave]

After trying the workaround detailed by Warren Togami in:

http://www.redhat.com/archives/fedora-list/2004-May/msg04389.html

to install Fedora Core 2 on an Adaptec 3200S (not ZCR) without any luck, I 
recently tried the latest Rawhide, unfortunately also without success, only to 
discover the problem was elsewhere. It didn't appear to be anaconda failing to 
detect the array, but rather the kernel failing to map the controller:

PCI: Found IRQ 10 for device 0000:00:09.1
i2o: I2O controller found on bus 0 at 73.
i2o: PCI I2O controller at C8000000 size=134217728
i2o: Unable to map controller.
i2o: DMA / IO allocation for I2O controller  failed
I2O controller: probe of 0000:00:09.1 failed with error -12

i2o is failing in ioremap() - VmallocTotal is 106M, so this isn't surprising. 
  Yes, I should have spotted this originally, but it took me a Rawhide 
download and two days to notice.

lspci gave:

00:09.0 PCI bridge: Adaptec (formerly DPT) PCI Bridge (rev 02) (prog-if 00 
[Normal decode])
	Flags: bus master, medium devsel, latency 64
	Bus: primary=00, secondary=02, subordinate=02, sec-latency=64
	I/O behind bridge: 0000a000-0000afff
	Capabilities: [68] Power Management version 2

00:09.1 I2O: Adaptec (formerly DPT) SmartRAID V Controller (rev 02) (prog-if 01)
	Subsystem: Adaptec (formerly DPT) 3000S Ultra3 Dual Channel
	Flags: medium devsel, IRQ 10
	BIST result: 00
	Memory at c8000000 (32-bit, prefetchable) [size=128M]
	Capabilities: [80] Power Management version 2

This appears to be the same problem reported by Markus Lidel on the LKML in June:

http://www.ussg.iu.edu/hypermail/linux/kernel/0406.0/0070.html

The controller had 128Mb SDRAM memory module on board, I removed this module 
and replaced the standard 32Mb module the board would normally ship with, and 
the kernel was able to correctly load the module.

The kernel used was 2.6.8-1.541 - I'm aware there is a slightly more recent 
2.6.8-1.549 kernel outside of the development download, but wasn't going to 
expend the effort of rebuilding a boot CD from scratch.

Ironically even Redhat 7.3 manages to detect and handle the controller with 
128Mb in without issue (2.4 kernels' dpt_i2o). You'll be glad to know that 
anaconda in rawhide does detect the controller without the i2o_proc hoopla, 
though. =)

I realise FC3 Test 2 was delayed, are there any i2o or ioremap changes pending 
that will allow me to use the controller with the larger memory module present?

I guess Warren is best placed to answer this - I do now have the controller 
working with 32Mb, but I'm willing to cooperate with any testing the i2o on 
linux team might want, if only out of curiosity. The controller works on the 
2.4 kernel and *cough* other operating systems without issue, so it would be 
nice to see the 2.6 kernel also handling this correctly. There's also a second 
128Mb module here if anyone feels like seeing what 256Mb will do, probably 
with FC2 and an updated kernel rather than Rawhide now, though. =)


David Zambonini