Comment # 7 on bug 1092035 from Fei Li

(In reply to Fei Li from comment #5)
> (In reply to magist3r sergei from comment #4)
> > Hi. I used geekbench benchmark.
> > Here is my general result: https://browser.geekbench.com/v4/cpu/8134716
> > And here after cpu performance drop:
> > https://browser.geekbench.com/v4/cpu/8134665
> > 
> > CPU performance drop is present right after starting VM with *-IBRS profile
> > and restores only with CPU power reset (reboot or suspend/resume). Even
> > stopping all VM's not solves the problem.
> So your problem is stopping the VM with *-IBRS can not restore the host CPU
> performance, and instead the performance can only be restored by rebooting
> the host, right? This sounds not like the same with this bug's title. :)
> Will try to reproduce it on my environment then and see what's wrong with
> this.
> 
I use the geekbench-tool you shared to do the tests, and did not reproduce your
phenomenon. In my environment[1] (my machine is Skylake, currently I do not
have a Ivybridge one), CPU downgrades not so obvious[2].

[1] My environment:
host # rpm -qa | grep ucode-intel
ucode-intel-20180312-22.1.x86_64
host # uname -a
Linux linux 4.4.126-48-default #1 SMP Sat Apr 7 05:22:50 UTC 2018 (f24992c)
x86_64 x86_64 x86_64 GNU/Linux
host # grep -H . /sys/devices/system/cpu/vulnerabilities/*
/sys/devices/system/cpu/vulnerabilities/meltdown:Mitigation: PTI
/sys/devices/system/cpu/vulnerabilities/spectre_v1:Mitigation: __user pointer
sanitization
/sys/devices/system/cpu/vulnerabilities/spectre_v2:Mitigation: IBRS+IBPB
host # dmesg | grep spec
 //no message means fully applied
And qemu (2.9.1-41.1.x86_64) libvirt (3.3.0-15.1) are up to date, too.

guest # cat /proc/cpuinfo
Model name:   Intel Xeon E3-12xx v2 (Ivy Bridge, IBRS)
...
flags:        fpu vme ... spec_ctl ...
...
guest # grep -H . /sys/devices/system/cpu/vulnerabilities/*
/sys/devices/system/cpu/vulnerabilities/meltdown:Mitigation: PTI
/sys/devices/system/cpu/vulnerabilities/spectre_v1:Mitigation: __user pointer
sanitization
/sys/devices/system/cpu/vulnerabilities/spectre_v2:Mitigation: Full generic
retpoline + IBPB


[2] My test results
1. In the host without any vm is running, run `./geekbench4` under the
decompressed Geekbench-4.2.2-Linux, result is 4544/15058, see
https://browser.geekbench.com/v4/cpu/8284798
2. `virsh start opensuse42.3`, run `./geekbench4`, result is 4541/14357, see
https://browser.geekbench.com/v4/cpu/8285056
3. `virsh destroy opensuse42.3`, run `./geekbench4`, result is 4501/14143, see
https://browser.geekbench.com/v4/cpu/8285106

BTW, I use same steps on a Haswell v2 machine to do the tests too. The result
is also not obvious before starting a vm and destroy a vm. FYI, pls see
8284586/8284623/8284656.

Umm, have no idea what is the difference between our environments.. But at
least, the config you have attached are same.