Working from OPENBSD_7_8
Attaching acpi uart UAR1 from an AMD Ryzen 5 CPU
BIOS ASROCK B550 PHantom Gaming 4 / American Megatrends rev 0x50011
Highly cut down - full dmesg, etc on request
I -think- this occurs for all amd64

in dev/ic/com.c line 1603 com_fifo_probe line 34 (+ or -)
this: (release version 1.180)
         for (len = 0; len < 256; len++) {
                 timo = 2000;
                 while (!ISSET(com_read_reg(sc, com_lsr),
                     LSR_RXRDY) && --timo)
                         delay(1);
                 if (!timo || com_read_reg(sc, com_data) != (len + 1))
                         break;
         }
results bad len = 0

this: (test)
         uint8_t c = 0x5A;
         for (len = 0; len < 256; len++) {
                 timo = 2000;
                 while (!ISSET(com_read_reg(sc, com_lsr),
                     LSR_RXRDY) && --timo)
                         delay(1);
                 if (!timo || (c = com_read_reg(sc, com_data)) != (len + 1))
                         break;
         }
         printf("fifo probe stopped at %d with %02x\n", len, c);
results good len = 16

It looks like cc is optimizing out the com_read_reg(sc, com_data)
unless the value is used later.

The printf is ugly. I suspect it wouldn't be approved.
Making com_read_reg volatile didn't fix it.
bus.h definition of bus_space_read doesn't look like it uses volatile

Two possible fixes which would make 30-40 lines of diff.
Both limit variable scopes which often makes cc happier.
Extracting that part of com_fifo_probe into another routine
Putting braces around the scope of timo, len, and other temps

Is there a better way to deal with this?

thanks
Geoff Steckel