Advansys on amd64 status

dobbs · Posted: Tue Nov 20, 2007 8:57 pm Post subject: Advansys on amd64 status

Background: I want to use my SCSI scanner. I have an "Advanced System Products, Inc ABP940-U / ABP960-U (rev 03)" PCI SCSI card and an AMD64 system.

I see from this ancient post that the advansys driver is "unstable" on amd64, but my kernel config (gentoo-sources-2.6.22-r9) doesn't have an option to "Select only drivers expected to compile cleanly". I do have development and/or incomplete code/drivers enabled.

I remember previously having the clean compile option, where did it go?

What is the advansys modules development status for amd64? Did it migrate to a different name that I missed? Am I SOL with this card?

saellaven · l33t Joined: 23 Jul 2006 Posts: 655

Working (with a couple assertion failures on loading) on AMD64/gentoo-sources-2.6.23 here.

scsi9 : AdvanSys SCSI 3.3K: PCI Ultra: IO 0xAC00-0xAC0F, IRQ 0x10

I make no guarantees that it'll work for you, but here's a patch that will let you enable it:

saellaven · l33t Joined: 23 Jul 2006 Posts: 655

I've traced the assertion down to a timestamp comparison (timestamp of the scsi_cmnd.this_residual (int) and jiffies (unsigned long)). I just submitted the following patch to linux-scsi and matthew wilcox (who has been working on the advansys driver).

dobbs · Posted: Fri Nov 23, 2007 9:36 am Post subject:

Much gratitude for the patches. Module loads fine and scanner probed. Not much else achieved (problems with xsane?).

Edit: Ok, scanner works beautifully now for no reason. Again, thank you for the patches.

saellaven · l33t Joined: 23 Jul 2006 Posts: 655

The advansys driver's maintainer disappeared and it fell behind after PCI and DMA changes to the rest of the kernel. It was actually marked BROKEN and hidden entirely from the menu in Linus' kernel for most of the year. Andrew Morton and I did some tweaks to it earlier this year to keep it working and Matthew Wilcox started really cleaning it up over the summer. Since it had been broken and there were only two people using the driver (at least as far as those of us who reported bugs on lkml), it was probably marked X86_32 because nobody knew if it would work on anything else given how far behind it had fallen.

I believe the current plan is to drop the X86_32 tag and switch it to VIRT_TO_BUS (quick look says everything but powerpc, sparc and superh platforms) instead, which is why I didn't submit the KConfig patch, just the fix for the timestamp problem. I did let Matthew and linux-scsi know that it was working on AMD64 for me, though. I'm guessing we'll see the official changes in 2.6.24 or 2.6.25 (if Linux deems the changes to be too drastic for this late in 2.6.24).

Of course, I'm not a kernel dev, I just read lkml and this particular bug lit a fire under my butt to fix it for my own system too. I'm glad I could help.
_________________
Ryzen 3700X, Asus Prime X570-Pro, 64 GB DDR4 3200, GeForce GTX 1660 Super
openrc-0.17, ~vanilla-sources, ~nvidia-drivers, ~gcc