I can switch between being able/unable to reproduce this bug by just doing git checkout 64/65, without touching any dependencies. git-bisect, with test being running build 100 times, points me to the commit 0c3c88fc930f3755e6f4fb0879783e2585863e85 as the first one after which builds do not fail anymore. Manual testing confirms as much: I am able to reproduce the bug before, but not after, said commit. I am not sure what to make of it: the commit does not seem like it might cause significant change. I could also add that the bug does not seem to be connected with virual machines or number of CPUs. I reproduced it both on 1-CPU KVM/QEMU and real system (on two different computers with different numbers of CPUs). Hope that information would provide someone with some insight.