[LITMUS^RT] RFC: record interrupt interference with Feather-Trace

[Björn Brandenburg]

On Jan 25, 2012, at 3:07 AM, Mac Mollison wrote:

> 
> Why not instead just disable interrupts completely for all the code
> paths that are being traced? Is there a reason related to the design of
> Linux that makes this a bad idea?
> 
> That way your overhead measurements are totally accurate and you truly
> don't have to mention any kind of caveat with the numbers when writing
> papers.
> 

Because a) we don't want to disturb how the system works normally (jn the absence of probes) and b) it's not feasible to disable interrupts in all code paths (e.g., ARM enables IRQs in the context switch path, system calls occur with interrupts enabled, semaphore locking protocols work with interrupts enabled, etc.).

- Björn