On 12/9/22 02:52, Oliver Upton wrote:
The combination of the pool-based ucall implementation + page_fault_test
resulted in some 'fun' bugs. As has always been the case, KVM selftests
is a house of cards.
Small series to fix up the issues on kvm/queue. Patches 1-2 can probably
be squashed into Paolo's merge resolution, if desired.
Tested on Ampere Altra and a Skylake box, since there was a decent
amount of munging in architecture-generic code.
v1 -> v2:
- Collect R-b from Sean (thanks!)
- Use a common routine for split and contiguous VA spaces, with
commentary on why arm64 is different since we all get to look at it
now. (Sean)
- Don't identity map the ucall MMIO hole
- Fix an off-by-one issue in the accounting of virtual memory,
discovered in fighting with #2
- Fix an infinite loop in ucall_alloc(), discovered fighting with the
ucall_init() v. kvm_vm_elf_load() ordering issue
Queued 3+5, thanks.
Paolo
