OSNews

Sounds awesome... you can certainly do a lot of interesting things if you control everything from the microprocessor to the OS, including the compiler toolchain. You'd think with all the geniuses at IBM, they could afford a resonable source control system. Perhaps they could even just buy one... perforce, for instance. Or Rational, which apparently has its own.

Are you saying that linux isn't pageable? I thought that support was added to some degree with 2.6. The page fault in kernel space is a bane of Windows Driver development as well (ever seen the IRQL_NOT_LESS_OR_EQUAL blue screen? It's the most common one and 99% of the time it's from accessing paged out code).

In a sense, Windows has similarities to AIX in terms of its inability to be open-sourced. For instance, Pretzold goes into some detail in Beautiful Code on the way the BitBlt graphics operation is performed in Windows: the function uses (or at least used to use) its parameters to autogenerate code on the stack which it would then execute per-pixel to achieve the image transfer and transformation. Apparently this kind of runtime codegen in graphics operations is really common, but I would be utterly unsuprised if it were part of the RPC mechanism as well. The point of this discussion is that there are some things that Open Source cannot do for reasons of distributable architecture and organization. It doesn't help that Linus is conservative in certain ways and is ideosyncratic in his likes and dislikes (a kgdb would go a long way in increasing the aggresiveness of things people can do in the linux kernel). It would be nice if the kernel and gcc communities would interact more positively to allow for some crazy IBM-esque optimizations.

Thanks for your post, though. You've given some new insight into how IBM does its kernel engineering. Are you still with that team?