Curiously, I can't find this message in recent kernels, it was patched out in 2018: https://github.com/torvalds/linux/commit/9d7c249a1ef9bf0d5696df14e6bc067004f

Curiously, I can't find this message in recent kernels, it was patched out in 2018: https://github.com/torvalds/linux/commit/9d7c249a1ef9bf0d5696df14e6bc067004f16979 https://github.com/torvalds/linux/commit/9d7c249a1ef9bf0d569... . Is it being re-added by distros or something? In terms of how it causes the issue, I think it's more that the serial driver will monopolise the (virtual) CPU for as long as it is receiving or sending text. If there's a large buffer being dumped all at once (a common default for non-interactive streams), it could stall for long enough to cause an underrun, which doesn't need to be very long, even if the average data rate isn't very high. Consider that the virtual 8250 serial port interface isn't exactly the most efficient interface to begin with, since it's at most copying 16 bytes from host to guest at a time