Comment 142 for bug 625239

On 09/14/2010 05:19 PM, Seth wrote:
> This could be brittle, because all this could do
> is reduce the time window for the race condition to happen
I can confirm my suspicion to be right because after - granted: many-
some reboots I saw the log message again in successive reboots:

    /var/log/syslog:Sep 14 20:38:39 localhost gdm-binary[984]: WARNING:
    GdmDisplay: display lasted 0.695752 seconds
    /var/log/syslog:Sep 14 20:42:04 localhost gdm-binary[1060]: WARNING:
    GdmDisplay: display lasted 0.198205 seconds

This was running _with_ my fixed gdm package, but from the analysis we
know that in this case the X desktop would have ended up shared with a
getty. If you are interested, you can easily verify for yourself how the
"nvidia not ready" condition will generate this message by using the
attached reproduction script:

    #!/bin/bash
    stop gdm
    sleep 2
    pgrep -l X && pkill X
    rm -fv /var/run/gdm/firstserver.stamp
    while lsmod | grep -q nvidia
    do
        echo unloading removing nvidia module
        rmmod nvidia
        sleep 1
    done
    lsmod | grep nv | nl
    {
        modprobe nvidia-current &
        start gdm &
        wait
    }
    sleep 5; chvt 1
    grep "display lasted" /var/log/syslog | tail
    date +"Time is now: %D %T"

Such a script, unfortunately, is not suitable for testing upstart
mutations (because the other subsystems interacting are not being
restarted; only a boot will give the right order and timing for these...)