QEMU needs to perform memslots operations like merging and splitting,
and each operation requires more than a single ioctl.
Therefore if a vcpu is concurrently reading the same memslots,
it could end up reading something that was temporarly deleted.
For example, merging two memslots into one would imply:
DELETE(m1)
DELETE(m2)
CREATE(m1+m2)

And a vcpu could attempt to read m2 right after it is deleted, but
before the new one is created.

This approach is 100% QEMU-based. No KVM API modification is involved,
but implies that QEMU must make sure no new ioctl is running and all
vcpus are stopped.

The logic and code are basically taken from David Hildenbrand
proposal given a while ago while reviewing a previous attempt where
I suggested to solve the above problem directly in KVM by extending
its API.

This is the original code:
https://github.com/davidhildenbrand/qemu/commit/86b1bf546a8d00908e33f7362b0b61e2be8dbb7a

I just split the patch in three smaller patches, and used a
QemuLockCnt instead of counter + mutex.

RHBZ: https://bugzilla.redhat.com/show_bug.cgi?id=1979276

Emanuele
---
v2:
* use QemuEvent instead of spinning in ioctl_inhibit_begin
* move the blocker API in a separate file and header, so that other accel can
  use it if they want.

David Hildenbrand (1):
  kvm: Atomic memslot updates

Emanuele Giuseppe Esposito (2):
  accel: introduce accelerator blocker API
  KVM: keep track of running ioctls

 accel/accel-blocker.c          | 139 +++++++++++++++++++++++++++++++++
 accel/kvm/kvm-all.c            | 107 ++++++++++++++++++++++---
 accel/meson.build              |   2 +-
 hw/core/cpu-common.c           |   2 +
 include/hw/core/cpu.h          |   3 +
 include/sysemu/accel-blocker.h |  45 +++++++++++
 include/sysemu/kvm_int.h       |   8 ++
 7 files changed, 294 insertions(+), 12 deletions(-)
 create mode 100644 accel/accel-blocker.c
 create mode 100644 include/sysemu/accel-blocker.h

-- 
2.31.1