Big Fat Bug
I have found an interesting bug. With a 256 MB DIMM installed in
addition to the stock 64 MB module, 'free' incorrectly reports only
256 MB memory installed. 'cat /proc/cpuinfo' correctly indicates that
320 MB are installed... but apparently the info from /proc/meminfo is
incorrect.
This is Not Good, as the other possibly related symptom is that with
the 256 MB module installed Linux becomes unusably unstable. The 100%
kill test is to attempt to compile the kernel... but often the first
command I run locks the system cold (e.g. I booted a few minutes ago,
and my first 'ps aux' locked up.) I have currently removed the
module, which permits me to write this message from Linux.
Many of you will recall my problems getting Linux installed -- which I
initially blamed on a bad DIMM.
I have now gone through three such DIMMs from at least two different
manufacturers. The most recent module is Simple Tech part no
STA-MAIN/256. This is a PC100 rated module. (Which b.t.w. fits in
the lower slot.)
All modules have exhibited exactly the same symptoms: the Lombard
becomes unusably unstable, subject to random hard lockups. Removing
the module is the uniform method to fix the problem. The problem also
persists when using only the 256 MB module. Nor does choice of slots
for the 64/256 combination change matters.
I had thought that MacOS was subject to the same problem, just less
frequently but now I am not so sure. I have gotten a couple (two)
random lockups, but I can't correlate these to any particular
reproducable phenomena. Moreover, I can push the system very hard in
MacOS and not get any crashes.
MacOS correctly identifies the amount of RAM, and RAMometer has run
its RAM test just fine for a total of several hours. (Good way to
make sure the internal fan is working!) Other stress tests involving running
realtime audio dsp, netscape, copying 500 MB of data over the LAN,
then copying that locally, plus various other apps running in the
cracks all at the same time for an extended period do .... nothing.
Runs great. (wow.)
This problem has also occured with a variety of kernels, compiled
pre-gcc 2.95 and/or obtained in binary form from external sources.
1) Has anyone else noticed a problem similar to this, especially
w.r.t. incorrectly reported memory sizes? It is possible that this
could be a logic board problem with my Powerbook.
2) Where do bugreports for Vger kernels go to? Or else for Paul
Mackerras' kernels (I don't have Mackerras' contact info
presently..)
-- John
Reply to: