Re: [PATCH] Only process_die notifier in ia64_do_page_fault if KPROBES is configured.

[Andi Kleen]

On Saturday 27 August 2005 01:05, Christoph Lameter wrote:
> On Fri, 26 Aug 2005, Rusty Lynch wrote:
> > Just to be sure everyone understands the overhead involved, kprobes only
> > registers a single notifier.  If kprobes is disabled (CONFIG_KPROBES is
> > off) then the overhead on a page fault is the overhead to execute an
> > empty notifier chain.
>
> Its the overhead of using registers to pass parameters, performing a
> function call that does nothing etc. A waste of computing resources. All
> of that unconditionally in a performance critical execution path that
> is executed a gazillion times for an optional feature that I frankly
> find not useful at all and that is disabled by default.

In the old days notifier_call_chain used to be inline. Then someone looking
at code size out of lined it. Perhaps it should be inlined again or notifier.h
could supply a special faster inline version for time critical code.

Then it would be simple if (global_var != NULL) { ... } in the fast path.
In addition the call chain could be declared __read_mostly.

I suspect with these changes Christoph's concerns would go away, right?

-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/