RFC: panfrost: allow to override "GPU bitness"
Currently when compiling Panfrost data width in structs equals pointer size of target arch, e.g. 32-bit when compiling for arm and 64-bit for aarch64. T6xx and T7xx use 32-bit structs, while T8xx uses 64-bit. If compiling for arm userspace and T8xx we cannot use
uintptr_t in structs because they are 32-bit wide, we have to use 64-bit wide pointers.
I consider this MR a temporary solution to allow running panfrost on T8xx from arm userspace (in my case) and T7xx from aarch64 userspace (Allwinner H6).