Liane Praza wrote: > Mike's two design questions are key to attacking this problem. I'd stay > away from constraining driver/kernel configuration to be read only by a > userland application.
Mixed feelings. At a purist level, you are of course right. At a somewhat different level, if we can manage to drive the number of tunables that are critical during early boot down to zero and hold it there, a userland model might be simpler. (Note that I said "critical"... doesn't mean that an early-boot driver can't have tunables, only that it has to be able to run acceptably without them until the config application can run.) Other than generally supporting putting tunables into SMF, that's probably about all I have to say. I jumped in because in a previous life device instance identification was one of my key areas of interest; I have no involvement with devices and the kernel these days.
