On Fri, 14 Dec 2007 01:24:13 +0000
"Steven M. Bellovin" <smb@cs.columbia.edu> wrote:

> For me, it's been random, with about a 50% probability of either a
> shutdown or no problem.
> 
I ran

	envstat -d acpibat0 -s charge,'charge state' -i 30 >/tmp/battlog

overnight and looked at the results.  The machine was plugged
in the entire time, and I started with a full battery.  Something is
clearly bogus here:

b135$ grep charge: /tmp/battlog | sort | uniq -c
 150         charge:     66.970 Wh (99.51%)
 158         charge:     67.000 Wh (99.55%)
 149         charge:     67.030 Wh (99.60%)
  13         charge:     67.040 Wh (99.61%)
 156         charge:     67.060 Wh (99.64%)
 164         charge:     67.080 Wh (99.67%)
 163         charge:     67.120 Wh (99.73%)
 118         charge:     67.220 Wh (99.88%)
   1         charge:    105.440 Wh (156.67%)
   1         charge:    113.120 Wh (168.08%)
   6         charge:    120.800 Wh (179.49%)
   1         charge:    128.480 Wh (190.91%)
   1         charge:    133.600 Wh (198.51%)
   1         charge:    143.840 Wh (213.73%)
   1         charge:    169.440 Wh (251.77%)
b136$ grep 'charge state:' /tmp/battlog | sort | uniq -c
 980   charge state:     NORMAL
 103   charge state:   CRITICAL

The battery state is wrong almost 10% of the time.  The charge level
does better, but still has a non-trivial error rate.

My tests on the AC adapter showed no failures overnight, but I need to
do some boot-time tests.

		--Steve Bellovin, http://www.cs.columbia.edu/~smb