On Thu, 25 Aug 2016, Arnd Bergmann wrote: > As Russell King found out the hard way, a change I did to fix multiplatform > builds with this driver broke the old Assabet/Neponset platform: It turns > out that while the driver is runtime configurable in principle, the > runtime configuration does not cover the specific case of machines that > can not do any 16-bit I/O on the smc91x registers. > > The driver currently provides helpers to access 16-bit registers for > architectures that are known at compile-time to only have 8-bit I/O, > but my patch changed it to a runtime flag that never gets consulted > most register accesses. > > This introduces new SMC_out16()/SMC_in16 helpers (if anyone can suggest > a better name, I'm glad to modify this) that behaves like SMC_outw()/SMC_inw() > most of the time, but uses a pair of 8-bit accesses on platforms that > have no support for wider register accesses.
Why don't you fold this directly into SMC_outw() instead? Nicolas