On Fri, Apr 19, 2013 at 11:05:47AM +0100, Peter Maydell wrote: > On 19 April 2013 10:39, Will Deacon <will.deacon@xxxxxxx> wrote: > > The point of mach-virt is that it is completely parameterised. So, if you're > > not emulating an 8250, then don't tell the kernel that you have one! > > Similarly, if you *do* emulate it, then either create a device-tree node for > > it or pass the appropriate earlyprintk= string on the command line. > > > > As far as kvmtool is concerned, we'd probably have a new command-line option > > for arm64, allowing you to specify the early console device. > > Please make the kernel pick the device out of the device tree > blob. The whole point of device tree is that it's how to tell > the kernel where things live -- making kvmtool/QEMU and/or the > user also have to mess with the kernel command line is awkward > and annoying. For a normal console device, it indeed needs to get it from the DT. For early console, you want it earlier than DT parsing so we pass it on the kernel command line via the earlyprintk= parameter. arm64 earlyprintk support for 8250 is already queued for 3.10 (and in -next). -- Catalin _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
