I have a bulldozer (AMD 9590) running -CURRENT (a few days ago) that exhibits some strange behavior lighting up a linux VM that we have transferred over from Digital Ocean.
If it boots with console as stdio (-f argument), it runs. If it boots without -f (console is a null-modem serial), then I get the following on the console (and the boot never finishes)
[ 54.548000] serial8250: too much work for irq4
[ 57.412000] serial8250: too much work for irq4
[ 61.040000] [sched_delayed] sched: RT throttling activated
[ 61.044000] serial8250: too much work for irq4
[ 63.892000] serial8250: too much work for irq4
[ 66.760000] serial8250: too much work for irq4
[ 69.656000] serial8250: too much work for irq4
Now I've done some research on this message. According to old posts (2009-2012-ish) Other virtual machines got this message from linux when they were either delivering interrupts on the serial when the serial was empty _or_ they were emptying the serial buffer too quickly and interrupting for more output.
On that latter point, apparently this is a legitimate failure mode of some physical chips, so the was much pushback on the linux side that the VMs need be fixed.
I don't see any published work arounds and I'd rather not leave screens running the console of machines ... so ...