[SRU Bionic/Focal] LP: #1888332

[Thadeu Lima de Souza Cascardo]

This was tested as KVM guests on a P9. The 3 failing tests pass after the fix.
No regressions compared to the kernel in -updates are noticed when running the
ptrace, tm and dscr kernel selftests.

[Impact]
Code that touches DSCR user MSR does not set dscr_inherit, which breaks DSCR
restore doing context switches and inheritance when forking. DSCR is used to
control cache hinting. This is caused by lack of kernel interrupt when the DSCR
user MSR is written, which is controlled by FSCR which would otherwise cause a
facility unavailable interrupt.

[Test case]
apt-get source linux
cd linux-5.4.0/tools/testing/tools/selftests/powerpc/
make -j 32
make -C dscr run_tests
make -C ptrace run_tests
make -C tm run_tests

Look up for "not ok" versus "ok", specially for dscr_inherit_exec_test,
ptrace-tar and tm-resched-tar.

[Potential regression]
Manipulating DSCR might break on different machines (with DTs containing
/cpus/ibm,powerpc-cpu-features, for example). Code that does so might crash
because the facility unavailable interrupt handling might be not working
correctly.