Splitting interrupt handlers


"Real" operating system had to face the IRQ mess
problem "S": the CPU can serve a single interrupt at a time
problem "P": interrupts disrupt predictable (and RT) reponse

The first approach is splitting the interrupt handling
A "top half" handler does thevery minimum stuff
It communicates with hardware, and acknowledges the interrupt
A "bottom half" handler deals with OS data structures
This happens with interrupts enabled


     IRQ (irq disabled in CPU)         _____
     bottom half (irq enabled)        /     \____________
     process context (enabled)  _____/                   \__________________


This mitigates problem S above, but not problem P
And the bottom half still cannot sleep or schedule
It runs in a privileged context, where the process is stalled