On Tue, Apr 30, 2019 at 09:44:45AM -0400, Steven Rostedt wrote:
> On Tue, 30 Apr 2019 12:46:48 +0200
> Peter Zijlstra <pet...@infradead.org> wrote:
>
> > On Mon, Apr 29, 2019 at 02:52:50PM -0400, Steven Rostedt wrote:
> > > On Mon, 29 Apr 2019 11:06:58 -0700
> > > Linus Torvalds <torva...@linux-foundation.org> wrote:
> > >
> > > > +void replace_call(void *addr, const void *opcode, size_t len, void
> > > > *target)
> > > > +{
> > > > + bp_int3_call_target = target;
> > > > + bp_int3_call_return = addr + len;
> > > > + bp_int3_handler_irqoff = emulate_call_irqoff;
> > > > + text_poke_bp(addr, opcode, len, emulate_call_irqon);
> > > > +}
> > >
> > > Note, the function tracer does not use text poke. It does it in batch
> > > mode. It can update over 40,000 calls in one go:
> >
> > Note that Daniel is adding batch stuff to text_poke(), because
> > jump_label/static_key stuffs also end up wanting to change more than one
> > site at a time and IPI spraying the machine for every single instance is
> > hurting.
> >
> > So ideally ftrace would start to use the 'normal' code when that
> > happens.
>
> Sure, but that's a completely different issue. We would need to solve
> the 'emulate' call for batch mode first.
I don't see a problem there; when INT3 happens; you bsearch() the batch
array to find the one you hit, you frob that into the percpu variables
and do the thing. Or am I loosing the plot again?
