Memory bandwidth counter width and overflow issue - steps to reproduce:

1. Run memory bandwidth stress workload (e.g., STREAM) on multiple CPUs to generate memory bandwidth.

# taskset -c 0-7 ./stream_c.exe
# pstree -lp | grep stream_c.exe
           |            |-sshd(4118)---bash(4184)---stream_c.exe(4388)-+-{stream_c.exe}(4389)
           |            |                                              |-{stream_c.exe}(4390)
           |            |                                              |-{stream_c.exe}(4391)
           |            |                                              |-{stream_c.exe}(4392)
           |            |                                              |-{stream_c.exe}(4393)
           |            |                                              |-{stream_c.exe}(4394)
           |            |                                              `-{stream_c.exe}(4395)


2. Mount Resctrl Filesystem, create control group "p1", add STREAM workload PIDs/TIDs to p1 group:

# mount -t resctrl resctrl /sys/fs/resctrl
# mkdir /sys/fs/resctrl/p1
# ./add_tasks.sh
-------------------------------------------------------------------------------------
#/usr/bin/bash
# add_tasks.sh

ID=`ps -efL | grep stream_c.exe | grep -v "grep" | awk '{print $4}'`
for id in $ID
do
        echo $id >/sys/fs/resctrl/p1/tasks
done
-------------------------------------------------------------------------------------


3. Read memory bandwidth counters with 1 second interval, observe the random unexpected values due to incorrect counter overflow handling:

# ./mbm_total_verbose.sh
-------------------------------------------------------------------------------------
#/usr/bin/bash
# mbm_total_verbose.sh

total_1=`cat /sys/fs/resctrl/p1/mon_data/mon_L3_00/mbm_total_bytes`
sleep 1
while true
do
        total_2=`cat /sys/fs/resctrl/p1/mon_data/mon_L3_00/mbm_total_bytes`
        echo "total b/w (bytes/s):" `expr $total_2 - $total_1` "($total_2 - $total_1)"
        total_1=$total_2
        sleep 1
done
-------------------------------------------------------------------------------------

total b/w (bytes/s): 21663042304 (222340959424 - 200677917120)
total b/w (bytes/s): 21951040640 (244292000064 - 222340959424)
total b/w (bytes/s): 21491510336 (265783510400 - 244292000064)
total b/w (bytes/s): 1125646660758656 (1125912444269056 - 265783510400)  < == unexpected values due to incorrect counter overflow handling
total b/w (bytes/s): 21957204096 (1125934401473152 - 1125912444269056)
total b/w (bytes/s): 21750773184 (1125956152246336 - 1125934401473152)
total b/w (bytes/s): 21397818496 (1125977550064832 - 1125956152246336)
total b/w (bytes/s): 21933403264 (1125999483468096 - 1125977550064832)
total b/w (bytes/s): 21947796032 (1126021431264128 - 1125999483468096)
total b/w (bytes/s): 21271977152 (1126042703241280 - 1126021431264128)
total b/w (bytes/s): 21854536000 (1126064557777280 - 1126042703241280)
total b/w (bytes/s): 21957381568 (1126086515158848 - 1126064557777280)
total b/w (bytes/s): 21358693184 (1126107873852032 - 1126086515158848)
total b/w (bytes/s): 21771448640 (1126129645300672 - 1126107873852032)
total b/w (bytes/s): 21960641920 (1126151605942592 - 1126129645300672)
total b/w (bytes/s): 21463460352 (1126173069402944 - 1126151605942592)
total b/w (bytes/s): 1125646683192192 (2251819752595136 - 1126173069402944)  < == unexpected values due to incorrect counter overflow handling
total b/w (bytes/s): 21958503872 (2251841711099008 - 2251819752595136)
total b/w (bytes/s): 21731181440 (2251863442280448 - 2251841711099008)
total b/w (bytes/s): 21410335936 (2251884852616384 - 2251863442280448)