[logging] How do I convert dmesg timestamp to custom date format?

I am trying to understand the dmesg timestamp and find it hard to convert that to change it to java date/custom date format.

Sample dmesg log:

[14614.647880] airo(eth1): link lost (missed beacons)

So how do I convert 14614.647880 to a standard date?

With the help of dr answer, I wrote a workaround that makes the conversion to put in your .bashrc. It won't break anything if you don't have any timestamp or already correct timestamps.

dmesg_with_human_timestamps () {
    $(type -P dmesg) "$@" | perl -w -e 'use strict;
        my ($uptime) = do { local @ARGV="/proc/uptime";<>}; ($uptime) = ($uptime =~ /^(\d+)\./);
        foreach my $line (<>) {
            printf( ($line=~/^\[\s*(\d+)\.\d+\](.+)/) ? ( "[%s]%s\n", scalar localtime(time - $uptime + $1), $2 ) : $line )
        }'
}
alias dmesg=dmesg_with_human_timestamps

Also, a good reading on the dmesg timestamp conversion logic & how to enable timestamps when there are none: https://supportcenter.checkpoint.com/supportcenter/portal?eventSubmit_doGoviewsolutiondetails=&solutionid=sk92677

If you don't have the -T option for dmesg as for example on Andoid, you can use the busybox version. The following solves also some other issues:

1. The [0.0000] format is preceded by something that looks like misplaced color information, prefixes like <6>.
2. Make integers from floats.

It is inspired by this blog post.

#!/bin/sh                                                                                                               
# Translate dmesg timestamps to human readable format                                                                   

# uptime in seconds                                                                                                     
uptime=$(cut -d " " -f 1 /proc/uptime)                                                                                  

# remove fraction                                                                                                       
uptime=$(echo $uptime | cut -d "." -f1)                                                                                 

# run only if timestamps are enabled                                                                                    
if [ "Y" = "$(cat /sys/module/printk/parameters/time)" ]; then                                                          
  dmesg | sed "s/[^\[]*\[/\[/" | sed "s/^\[[ ]*\?\([0-9.]*\)\] \(.*\)/\\1 \\2/" | while read timestamp message; do      
    timestamp=$(echo $timestamp | cut -d "." -f1)                                                                       
    ts1=$(( $(busybox date +%s) - $uptime + $timestamp ))                                                               
    ts2=$(busybox date -d "@${ts1}")                                                                                    
    printf "[%s] %s\n" "$ts2" "$message"                                                                                
  done                                                                                                                  
else                                                                                                                    
  echo "Timestamps are disabled (/sys/module/printk/parameters/time)"                                                   
fi                                                                                                                      

Note, however, that this implementation is quite slow.

In recent versions of dmesg, you can just call dmesg -T.

So KevZero requested a less kludgy solution, so I came up with the following:

sed -r 's#^\[([0-9]+\.[0-9]+)\](.*)#echo -n "[";echo -n $(date --date="@$(echo "$(grep btime /proc/stat|cut -d " " -f 2)+\1" | bc)" +"%c");echo -n "]";echo -n "\2"#e'

Here's an example:

$ dmesg|tail | sed -r 's#^\[([0-9]+\.[0-9]+)\](.*)#echo -n "[";echo -n $(date --date="@$(echo "$(grep btime /proc/stat|cut -d " " -f 2)+\1" | bc)" +"%c");echo -n "]";echo -n "\2"#e'
[2015-12-09T04:29:20 COT] cfg80211:   (57240000 KHz - 63720000 KHz @ 2160000 KHz), (N/A, 0 mBm), (N/A)
[2015-12-09T04:29:23 COT] wlp3s0: authenticate with dc:9f:db:92:d3:07
[2015-12-09T04:29:23 COT] wlp3s0: send auth to dc:9f:db:92:d3:07 (try 1/3)
[2015-12-09T04:29:23 COT] wlp3s0: authenticated
[2015-12-09T04:29:23 COT] wlp3s0: associate with dc:9f:db:92:d3:07 (try 1/3)
[2015-12-09T04:29:23 COT] wlp3s0: RX AssocResp from dc:9f:db:92:d3:07 (capab=0x431 status=0 aid=6)
[2015-12-09T04:29:23 COT] wlp3s0: associated
[2015-12-09T04:29:56 COT] thinkpad_acpi: EC reports that Thermal Table has changed
[2015-12-09T04:29:59 COT] i915 0000:00:02.0: BAR 6: [??? 0x00000000 flags 0x2] has bogus alignment
[2015-12-09T05:00:52 COT] thinkpad_acpi: EC reports that Thermal Table has changed

If you want it to perform a bit better, put the timestamp from proc into a variable instead :)

you will need to reference the "btime" in /proc/stat, which is the Unix epoch time when the system was latest booted. Then you could base on that system boot time and then add on the elapsed seconds given in dmesg to calculate timestamp for each events.

A caveat which the other answers don't seem to mention is that the time which is shown by dmesg doesn't take into account any sleep/suspend time. So there are cases where the usual answer of using dmesg -T doesn't work, and shows a completely wrong time.

A workaround for such situations is to write something to the kernel log at a known time and then use that entry as a reference to calculate the other times. Obviously, it will only work for times after the last suspend.

So to display the correct time for recent entries on machines which may have been suspended since their last boot, use something like this from my other answer here:

# write current time to kernel ring buffer so it appears in dmesg output
echo "timecheck: $(date +%s) = $(date +%F_%T)" | sudo tee /dev/kmsg

# use our "timecheck" entry to get the difference
# between the dmesg timestamp and real time
offset=$(dmesg | grep timecheck | tail -1 \
| perl -nle '($t1,$t2)=/^.(\d+)\S+ timecheck: (\d+)/; print $t2-$t1')

# pipe dmesg output through a Perl snippet to
# convert it's timestamp to correct readable times
dmesg | tail \
| perl -pe 'BEGIN{$offset=shift} s/^\[(\d+)\S+/localtime($1+$offset)/e' $offset

With older Linux distros yet another option is to use wrapping script, e.g. in Perl or Python.

See solutions here:

http://linuxaria.com/article/how-to-make-dmesg-timestamp-human-readable?lang=en http://jmorano.moretrix.com/2012/03/dmesg-human-readable-timestamps/

For systems without "dmesg -T" such as RHEL/CentOS 6, I liked the "dmesg_with_human_timestamps" function provided by lucas-cimon earlier. It has a bit of trouble with some of our boxes with large uptime though. Turns out that kernel timestamps in dmesg are derived from an uptime value kept by individual CPUs. Over time this gets out of sync with the real time clock. As a result, the most accurate conversion for recent dmesg entries will be based on the CPU clock rather than /proc/uptime. For example, on a particular CentOS 6.6 box here:

# grep "\.clock" /proc/sched_debug  | head -1
  .clock                         : 32103895072.444568
# uptime
 15:54:05 up 371 days, 19:09,  4 users,  load average: 3.41, 3.62, 3.57
# cat /proc/uptime
32123362.57 638648955.00

Accounting for the CPU uptime being in milliseconds, there's an offset of nearly 5 1/2 hours here. So I revised the script and converted it to native bash in the process:

dmesg_with_human_timestamps () {
    FORMAT="%a %b %d %H:%M:%S %Y"

    now=$(date +%s)
    cputime_line=$(grep -m1 "\.clock" /proc/sched_debug)

    if [[ $cputime_line =~ [^0-9]*([0-9]*).* ]]; then
        cputime=$((BASH_REMATCH[1] / 1000))
    fi

    dmesg | while IFS= read -r line; do
        if [[ $line =~ ^\[\ *([0-9]+)\.[0-9]+\]\ (.*) ]]; then
            stamp=$((now-cputime+BASH_REMATCH[1]))
            echo "[$(date +"${FORMAT}" --date=@${stamp})] ${BASH_REMATCH[2]}"
        else
            echo "$line"
        fi
    done
}

alias dmesgt=dmesg_with_human_timestamps