A system feature called Pause-Loop Exiting can cause instability and performance degradation on a virtualized system that will have a web server running in a heavily utilized guest. Turn it off by setting the ple_gap module parameter to 0 for kvm_intel.
Example: Add file kvmopts.conf to /etc/modprobe.d/ with contents:
options kvm_intel ple_gap=0
Restart the kvm_intel module if it is already running.
# modprobe -r kvm_intel
# modprobe kvm_intel
or
insmod kvm-intel ple_gap=0
From Intel spec:
If the “PAUSE exiting” VM-execution control is 0 and the “PAUSE-loop exiting” VM-execution control is 1, the following treatment applies.
The processor determines the amount of time between this execution of PAUSE and the previous
execution of PAUSE at CPL 0. If this amount of time exceeds the value of the VM-execution control field
PLE_Gap, the processor considers this execution to be the first execution of PAUSE in a loop. (It also
does so for the first execution of PAUSE at CPL 0 after VM entry.)
Otherwise, the processor determines the amount of time since the most recent execution of PAUSE that
was considered to be the first in a loop. If this amount of time exceeds the value of the VM-execution
control field PLE_Window, a VM exit occurs.
For purposes of these computations, time is measured based on a counter that runs at the same rate as
the timestamp counter (TSC).
【人话总结】
永远把两次间隔超过ple_gap的pause指令其中的后一条作为第一次loop内pause，从这个第一次loop内pause开始计算次数直到超过PLE_window就触发exit
也就是间隔在ple_gap之内的连续pause指令构成了pause loop，大于PLE_window的pause loop将触发退出