On the first warm day of this spring, I got 2GB ram compatible with my very old DS20.

Previously there was 512MByte ram in it (4 modules).
The 2G ram is 8 modules.
I removed the 512MB ram ( just to have some spare ram to test any alpha around ), and installed the 8x256MB ram.

I booted the debian, and top reported the 2GB ram. No problmes in dmesg.
But after a half day, I can't connect to the alpha over ssh.
In the log there is a periodical error:

pci_map_single failed: could not allocate dma page tables

After a reboot it's ok for a while...

I used memtest, it tested 1GB memory, that was ok.

Unfortunately I can't figure out srm's memory tester
command's usage...

Now I test it with 1G ram, but it would be nice if I could use the 2G.

But that is two banks.

Maybe the DS20 is using some kind of non contiguous memory model?
My kernel don't have that option... is it necessary?

I checked my earlier logs, but I couldn't find the error above
there...

Maybe my alpha system is faulty now? I shouldn't open the cabinet after 14 years? Please help

If it did have 2G and now shows 1G then I'd think that one of the 4x banks has failed.

http://h18002.www1.hp.com/alphaserve...ry_install.pdf

I removed 1GB, and now it seems to be fine, but I wonder that unexpected errors caused by not the memory modules itself... they have ECC, and srm do some basic testing.. so I think that maybe the problem is that the linux kernel is not handling two banks correctly?
And if it's bank fail, then why that strange error in the logs?

"I booted the debian, and top reported the 2GB ram. No problmes in dmesg." So we know it did work. There is not maybe linux here.

"I used memtest, it tested 1GB memory, that was ok." No, it was not ok at all.

Can't say what bad memory does. dma would make me think drive controller also not just limited to ram.