Skip to contentSkip to conversation

Problems getting dma working with the sc1200.c ide driver

DISCUSSION Axel Waggershauser 1 participant 1 message Feb 10, 2004

lore
Share & tools
lore
#0 lore raw

Hi,

I have the <subject> problem and tried to contact Mark Lord, which seems to be the author of the driver. There is no response yet, so I decided to post it here...

I have a peace of hardware called "X-board" in front of me (if your are
interested: http://www.jumptec.de/product/data/xboard/xb-861.html). It is based on the Geode SC1200 cpu. This module is plugged into a development board which I connected a standard IDE drive to.

I tried to get dma access working with a 2.4 and the 2.6.2 kernel without success. The 2.6.2 kernel compiled with CONFIG_IDEDMA_PCI_AUTO=y hangs at boot with the last line on the screen containing "sc1200_set_xfer_mode(UDMA2)". Both kernel versions compiled without CONFIG_IDEDMA_PCI_AUTO=y boot fine, but trying to enable dma with hdparm after boot fails with "SC1200: set xfer mode failure". I took a look at the 2.6.2 driver code and found that if IDEDMA_PCI_AUTO is not configured every attempt to enable dma later on fails because sc1200_autoselect_dma_mode() returns 0 since hwif->autodma is false. So without the IDEDMA_PCI_AUTO option (which locks my machine) there seems to be no way to enable dma, or is there?

I did not further try to track down the lockup, yet. I could start debugging this on the level of "inserting printks". I have no knowledge of the IDE part of the kernel, though.

I have seen lines like:

	if (!((d->device == PCI_DEVICE_ID_CYRIX_5530_IDE && d->vendor == PCI_VENDOR_ID_CYRIX)
	     ||(d->device == PCI_DEVICE_ID_NS_SCx200_IDE && d->vendor == PCI_VENDOR_ID_NS)))
	{
	    hwif->autodma = 0;
	}

which I do not understand in their context but made me suspect that dma is not really expected to work on this setup.

I would appreciate any help or hints to a solution.

Thanks,
Axel.