Re: [RFC PATCH 6/8] preempt/dynamic: Provide preempt_schedule[_notrace]() static calls

[Peter Zijlstra]

On Tue, Jan 26, 2021 at 05:57:30PM -0600, Josh Poimboeuf wrote:
> On Fri, Jan 22, 2021 at 05:52:26PM +0100, Peter Zijlstra wrote:
> >  static int static_call_add_module(struct module *mod)
> >  {
> > -	return __static_call_init(mod, mod->static_call_sites,
> > -				  mod->static_call_sites + mod->num_static_call_sites);
> > +	struct static_call_site *start = mod->static_call_sites;
> > +	struct static_call_site *stop = start + mod->num_static_call_sites;
> > +	struct static_call_site *site;
> > +
> > +	for (site = start; site != stop; site++) {
> > +		unsigned long addr = (unsigned long)static_call_key(site);
> > +		struct static_call_ass *ass;
> > +
> > +		/*
> > +		 * Gotta fix up the keys that point to the trampoline.
> > +		 */
> > +		if (!kernel_text_address(addr))
> > +			continue;
> > +
> > +		ass = static_call_find_ass(addr);
> > +		if (!ass) {
> > +			pr_warn("Failed to fixup __raw_static_call() usage at: %ps\n",
> > +				static_call_addr(site));
> > +			return -EINVAL;
> > +		}
> > +		site->key = ((unsigned long)ass->key - (unsigned long)&site->key) |
> > +			    (site->key & STATIC_CALL_SITE_FLAGS);
> 
> Well, I hate it, but I'm not sure I have any better ideas.  It should be
> possible to use kallsyms, instead of the rb-tree/register nonsense.  Not
> sure about the performance impact though.  Might be a good reason to
> speed up kallsyms!

Oh right, let me see if I can make that work.

> Also I do have some naming suggestions ;-)

Nah, we need a little more fun back in the code :-)