The irq_vectors class of tracepoints adds two events for each interrupt: One at the entry and one at the return of each interrupt handler. They are very useful in the analysis of latency, as they point to the occurrence of an interrupt, and how long did it take to run.
However, there are two cases of interrupt which these tracepoints were not present: NMIs and regular device interrupt. So I suggested two new tracepoints, for these cases.
The output of these new tracepoints looks like this:
idle-0  d.h. 102.890935: external_interrupt_entry: vector=35 idle-0  d.h. 102.890960: external_interrupt_exit: vector=35 idle-0  d.Z. 179.594315: nmi_entry: vector=2 idle-0  d.Z. 179.594396: nmi_exit: vector=2
The discussion of the patch set can be found here: