New tmpfails on packages with 2+ tests in QEMU backends

A new category of tmpfails has recently appeared on ci.rocm.debian.net: Packages with 2+ tests (like rocrand) executed on the QEMU backend can tmpfail because of some communications failure.

I'm aware of this, because this happens consistently with Noble; it's the one showstopper for Ubuntu testing that's been puzzling me for weeks now.

The failure is odd, to say the least: The machine reboots after the first test, but comes back successfully, installs all test dependencies, and so on -- only to fail at the following step:

328s autopkgtest [15:46:05]: Setting up user "debci" to sudo without password...
726s qemu-system-x86_64: terminating on signal 15 from pid 330665 (/usr/bin/python3)
799s autopkgtest [14:07:49]: ERROR: testbed failure: sent `auxverb_debug_fail', got `timeout', expected `ok...'

I don't know why this would suddenly affect Debian, too. Least of all, why this issue would occur at this setup step, of all things.

I'm prioritizing this.