[PATCH v2 2/2] arm64: ftrace: add support for far branches to dynamic ftrace

[will.deacon@arm.com (Will Deacon)]

On Wed, Jun 07, 2017 at 03:50:10PM +0000, Ard Biesheuvel wrote:
> On 7 June 2017 at 15:46, Steven Rostedt <rostedt@goodmis.org> wrote:
> > On Mon, 5 Jun 2017 18:15:35 +0100
> > Will Deacon <will.deacon@arm.com> wrote:
> >
> >
> >> > +           tramp = (unsigned long *)mod->arch.ftrace_trampoline->sh_addr;
> >> > +           if (tramp[0] != addr) {
> >> > +                   if (tramp[0] != 0) {
> >> > +                           pr_err("ftrace: far branches to multiple entry points unsupported inside a single module\n");
> >> > +                           return AARCH64_BREAK_FAULT;
> >> > +                   }
> >> > +
> >> > +                   /* point the trampoline to our ftrace entry point */
> >> > +                   module_disable_ro(mod);
> >> > +                   tramp[0] = addr;
> >> > +                   module_enable_ro(mod, true);
> >>
> >> I'm not sure what the barrier semantics are for module_enable_ro, but I'd be
> >> inclined to stick in a smp_wmb() here to order the write of the trampoline
> >> data before the writing of the branch instruction.
> >
> > I would assume that module_disable/enable_ro() has proper barriers for
> > modifying the page tables with respect to code around it, otherwise it
> > would probably be an issues elsewhere in the kernel. Specifically in
> > the module code itself.
> >
> > I don't see how a smp_wmb() would be needed here, especially since this
> > is serialized code, and not something done by multiple CPUs.
> >
> 
> But other cores could be invoking the function we are patching here,
> no? So when such a core observes (and executes) the updated
> instruction before it observes the updated target field of the
> trampoline, it will branch to address 0x0.

I think Steve's saying that the TLB invalidation buried in the set_memory_*
functions called by module_enable_ro will have sufficient barriers to order
the write of the trampoline with respect to writing the branch instruction,
but this isn't quite right because module_enable_ro is affected by things
like CONFIG_STRICT_MODULE_RWX and rodata_enabled.

Will