[DTrace-devel] [PATCH 2/7] Implement the sched provider

[Kris Van Hees]

On Sun, May 14, 2023 at 07:20:19PM -0400, Eugene Loh via DTrace-devel wrote:
> A few more comments.
> 
> I see probe "remain-cpu" mentioned in probe_args[], but it's not otherwise
> mentioned in the file.  When will it fire?

Same as various other probes in there...  I list the sched probes that are to
be supported, with the arguments they are to provide.  Those probes that do not
have support yet in the implementation are in #if 0 ... #endif blocks.

> Also, this patch has "on-cpu" firing with schedule_tail(). Meanwhile, if I
> look at kernel code (admittedly a little old) for DTv1, it has
> DTRACE_SCHED(on__cpu) in finish_task_switch(), which is called from
> schedule_tail() but also from context_switch(). So, does the new meaning of
> "on-cpu" differ from the DTv1 meaning?  In general, how closely do we expect
> these probes to match meanings from DTv1?

As discuss in the past, we plan on supporting all standard SDT DTrace probes.
But the support is being added incrementally as we determine the best ways to
make the probes available.  Right now, on-cpu is not fully supported because
the code path used for context switches is still being investigated to find the
right spot to place probes at.