On 03/28/2015 03:28 PM, Kevin Cernekee wrote: > Side note: > > AFAIK we still have a problem if somebody wants to build serial8250 + > (any other tty driver that occupies major 4 / minor 64) into the same > kernel, and use DT to pick the correct driver at runtime. Yep, exactly. > serial8250_init() starts registering ports before it knows whether the > system even has an 8250. This behavior is required to support hw configuration from userspace. > I talked to Rob about it earlier this week > and he suggested that you might have some thoughts on how to fix it. A big piece of that will land in 4.01 (once I get the actual 8250 split commit to Greg). The 8250 driver has been split into a legacy/universal 8250 driver and separate port ops module; one of the benefits of this is that the objectionable legacy behavior of the 8250 driver (especially limitations wrt ttyS sharing) can be left behind in that driver without breaking existing users. The idea is that an alternate 8250 driver(s) with none of the legacy baggage can be taught to share (4,64) ports with other drivers. Unfortunately, I haven't made further progress with because of other projects. Regards, Peter Hurley
