On 05/24/12 10:58, Damjan Marion wrote: > On May 24, 2012, at 4:30 PM, Warner Losh wrote: >> On May 24, 2012, at 3:40 AM, Damjan Marion wrote: >>> On May 24, 2012, at 10:25 AM, Aleksandr Rybalko wrote: >>> >>> I was looking into this few months ago but I didn't found a value in doing >>> this in embedded world where we already have custom kernel for each SoC/board. >>> >>> Maybe we will have GENERIC arm kernel one day, but there is long road.... >> I'm working on that, at least for all Atmel kernels. We'll have at least three kernels though: armv4 little endian, armv4 big endian and armv6 little endian. Even for atmel, some of the id registers are such we may need multiple kernels. > I guess they all use same interrupt controller. > Currently there can be only support for one intc built in kernel so that needs serious rework. > > Damjan This is something we've solved quite nicely on PowerPC. The current code allows interrupt controllers to be regular device drivers, including support for systems with multiple interrupt controllers and domains, so I'd suggest pulling that code over if Warner hasn't done it yet. FDT (or real Open Firmware) really helps here. There's a lot of code in the PPC tree for running the same kernels on systems with very different properties (MMUs, firmware, paravirtualized and bare metal, multiple hypervisors, PICs, etc.) that may be helpful for ARM/MIPS GENERIC. The exact same PPC64 kernel can boot a PS3 under a paravirtualized hypervisor with custom firmware, a Powermac G5, or an IBM machine, with or without hypervisor, for instance. -NathanReceived on Thu May 24 2012 - 14:21:33 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:27 UTC