On Wed, 29 May 2019 15:49:46 +0200 Peter Zijlstra <[email protected]> wrote:
> > That's basically what I was suggesting as the solution to this ;-) > > You were wanting changes to preempt_disable() and task_struct, neither > of which is required. The above only needs some per-cpu storage in the > tracer implementation. Only changes were to the trace preempt_disable() code. Which I still think needs to be done regardless to differentiate between when we are tracing preempt disable and when we are not. And no modification would need to be done to task_struct as we already have a place to add tracing flags. -- Steve
