On Thu, Jun 11, 2020 at 11:58:12PM +0200, Peter Zijlstra wrote: > On Thu, Jun 11, 2020 at 11:55:38PM +0200, Peter Zijlstra wrote: > > I'll have to dig around a little more to see if I can't get rid of the > > relocation entirely. Also, I need to steal better arch_nop_insn() from > > the kernel :-)
Oh, I just realized that recordmcount does exactly this same thing, so I checked what that does to the relocation, and it turns out, it does the same thing I did. They change the relocation type to R_*_NONE too. So I suppose that's all right then. I suppose I ought to go look at the recordmcount to objtool patches to see if there's anything to share there.
