The process start_time value (as revealed by "ps -eo lstart" or "ps
-eo start" or "ps -eo start_time") is horribly wrong under coLinux.
This may be related to a discussion of this issue in the Linux Kernel
Mailing List for several months this summer:
However, there is an important difference. In the LKML discussion,
they were mentioning how the start_time values could be off by a few
_seconds_ per day of uptime. In the coLinux case, this becomes a few
*hours* per day of uptime.
I noticed this problem because I was seeing completely absurd values
for the elapsed time of processes. For example, a process that just
started was reported as having more than 7 hours of elapsed time.
I suspect that whatever fix for this problem gets adopted for the
mainline Linux kernel may not be good enough for coLinux.
The obvious solution is just to store in the start_time field the time
from the same source that gettimeofday consults. That way, the
coLinux-specific fixes to gettimeofday will also apply to process