Here is a new variation of the SIGP handling discussed a few weeks ago [1]. Notable changes: - Patches 1 and 6 from v1 were picked for 5.16 (Thank you!) [2] - Patches 2 through 5 were removed, and replaced with this iteration that relies on a KVM capability and IOCTL I opted to use David's suggestion [3] for the kernel to automatically set a vcpu "busy" and userspace to reset it when complete. I made it dependent on the existing USER_SIGP stuff, which maybe isn't great for potential non-SIGP scenarios in the future, but this at least shows how it could work. According to the Principles of Operation, only a subset of SIGP orders would generate a "busy" condition, and a different subset would even notice it. But I did the entirety of the SIGP orders, even the invalid ones that would otherwise return some status bits and CC1 instead of the CC2 (BUSY) condition. Perhaps that's too much, perhaps not. As I'm writing this, I'm realizing that I probably need to look at the cpu reset paths clearer, to ensure the "busy" indicator is actually reset to zero. Since this is an RFC, I've left the CAP/IOCTL definitions as a standalone patch, so I see it easier when working with the QEMU code. Ultimately this would be squashed together, and might have some refit after the merge window anyway. I'll send the QEMU series shortly, which takes advantage of this. Thoughts? [1] https://lore.kernel.org/r/20211008203112.1979843-1-farman@xxxxxxxxxxxxx/ [2] https://lore.kernel.org/r/20211031121104.14764-1-borntraeger@xxxxxxxxxx/ [3] https://lore.kernel.org/r/3e3b38d1-b338-0211-04ab-91f913c1f557@xxxxxxxxxx/ Eric Farman (2): Capability/IOCTL/Documentation KVM: s390: Extend the USER_SIGP capability Documentation/virt/kvm/api.rst | 27 +++++++++++++++++++++ arch/s390/include/asm/kvm_host.h | 2 ++ arch/s390/kvm/kvm-s390.c | 18 ++++++++++++++ arch/s390/kvm/kvm-s390.h | 10 ++++++++ arch/s390/kvm/sigp.c | 40 ++++++++++++++++++++++++++++++++ include/uapi/linux/kvm.h | 4 ++++ 6 files changed, 101 insertions(+) -- 2.25.1
