Hi!

On Tue, 3 Jun 2025 03:18:23 +0000, Michael Kelley wrote:

The problem is timing dependent.  In some cases, the "pr/ttyAMA0"
thread is evidently able to get the message out before panic() calls
pr_emerg(). In my case running as a guest on Hyper-V, the pl011 device
in the guest VM is emulated by Hyper-V. Each pl011 register access
traps to the hypervisor, which slows things down and probably makes
the problem more likely. But from what I can see, the underlying
race condition exists regardless.

I believe the issue is not amba-pl011 specific, so maybe John and Petr
better be Cced as well.

Sincerely,
Ryo Takakura