Optimizations

[EtchedPixels]

Mix of system stuff and guidance needed here

Generically:

• Hash buffer cache on bigger boxes
• Avoid cases we double look up buffers
• Keep a true run queue (and maybe a punishment queue) to speed up the scheduler IRQ path a shade, although it's not actually that hot a path in normal usage
• Hooks to optimize path walking as an option

Platform Specific

• Document that adding an optimizing block copies between user<->kernel is a big win so worth doing platform specific usermem for in many cases
• Do a demo set for the TRS80 model 1 (remembering we'll also need buffer<->user direct copies when the buffers go banked)

[EtchedPixels]

TRS80 model 1 parts done Some other usermem optimized cases also done for a few platforms

[EtchedPixels]

Update:

• big buffer cache turns out to be a lose anyway (our I/O is relatively fast and we are not heavily multi-user, nor do we have any asynchronous DMA I/O like old minicomputers have)
• Path and some other walk optimisations are done
• Copier optimisations done in a nice example form in sbcv2

Scheduler queue TBD