Hi all,
Some of you may remember the recent discussions to straighten out
some serialization issues with the SIGNAL PROCESSOR instruction,
and its interaction with userspace. This resulted in a few kernel
patches [1] that reliably solve the issues I was seeing.
The attached kvm-unit-tests series adapts the existing smp tests
such that it can reproduce the problems I saw when those patches
are reverted (typically by simultaneously doing a kernel compile),
and also demonstrate that those patches indeed fix the issue.
There's some cleanup in here too, based on my understanding of
the smp tests as I was walking through here. Thoughts?
[1] 812de04661c4 KVM: s390: Clarify SIGP orders versus STOP/RESTART
67cf68b6a5cc KVM: s390: Add a routine for setting userspace CPU state
8eeba194a32e KVM: s390: Simplify SIGP Set Arch handling
Eric Farman (6):
lib: s390x: smp: Retry SIGP SENSE on CC2
s390x: smp: Test SIGP RESTART against stopped CPU
s390x: smp: Fix checks for SIGP STOP STORE STATUS
s390x: smp: Create and use a non-waiting CPU stop
s390x: smp: Create and use a non-waiting CPU restart
lib: s390x: smp: Convert remaining smp_sigp to _retry
lib/s390x/smp.c | 53 ++++++++++++++++++++++++++++++++++++++++++++++---
lib/s390x/smp.h | 2 ++
s390x/smp.c | 39 +++++++++++++++++++-----------------
3 files changed, 73 insertions(+), 21 deletions(-)
--
2.32.0
