Alan Cox wrote:
When the package system works, it works great. When it breaks, it's a
steep climb to put it right.
It should never break.
If you are seeing breakage it is a good idea to run memtest86 and similar
things on the system as the usual cause I've seen for this is faulty ram
Alan: according to memtest86+ you are on the money, it is finding errors
between 969M and 1027M. This lies across the boundary between
2xidentical 512MB sticks, and a third {older} 512MB stick.
I am about to test singular chips, but also noticed the badram= patch
for the kernel. As far as I can see from google, this hasn't been
accepted into mainline, and would require manual compilation {and again
on kernel updates}. Perhaps I am wrong with this ?
Also, I noticed in F7 the kdump capability could reserve memory regions
? crashkernel=128M@16M, could an {mis-}application of it be used to stop
the kernel from accessing the bad ram ?
DaveT.
