[lug] decoding ram locations from kernel oops

[D. Stimits]

D. Stimits wrote:
> I'm curious about something. I have a rare OOPS since adding ram. 
> memtest86 has run for days without any error, I can compile kernels over 
> and over with make -j8, no problem. All the errors occur under X, so I'm 
> thinking perhaps the problem which on the surface appears to be ram 
> might actually be the video card ram, and not the system ram (I have a 
> hard time reproducing this error). Is there any way to verify the actual 
> physical ram location based on an OOPS message? Especially, when AGP and 
> hardware accel is involved, is there any way to know if the address 
> involved lives inside of the video card? Here's the line that appears 
> over and over whenever one of these rare OOPS occur, which causes me to 
> believe memory is involved:
> kernel: EIP is at rw_verify_area+0x5a/0xa0

One additional question...even if I can't tell what the exact source of 
ram hardware which results in the 0x5a/0xa0, can I be sure that this 
address means the same location each time? Maybe virtual mappings have 
an effect...or would OOPS addresses for memory be consistent?

What I'm thinking of is that I have ram arranged in 4 dimms, 2 are 128 
MB each, 2 are 256 MB each. I could swap the 128's with each other, if 
it changes and if I know addresses in an OOPS are related to physical 
hardware not to virtual settings, then seeing the address change would 
tell me it is one of those two dimms. I could do the same with swapping 
the 256's (of course it might take a long time, these OOPS are pretty rare).

D. Stimits, stimits AT comcast DOT net