Re: objtool warning for next-20221118

[Josh Poimboeuf <jpoimboe@xxxxxxxxxx>]

On Thu, Nov 24, 2022 at 08:47:47AM +0100, Juergen Gross wrote:
> > > +++ b/arch/x86/xen/smp_pv.c
> > > @@ -385,17 +385,9 @@ static void xen_pv_play_dead(void) /* used only
> > > with HOTPLUG_CPU */
> > >   {
> > >       play_dead_common();
> > >       HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(smp_processor_id()), 
> > > NULL);
> > > -    cpu_bringup();
> > > -    /*
> > > -     * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down)
> > > -     * clears certain data that the cpu_idle loop (which called us
> > > -     * and that we return from) expects. The only way to get that
> > > -     * data back is to call:
> > > -     */
> > > -    tick_nohz_idle_enter();
> > > -    tick_nohz_idle_stop_tick_protected();
> > > -    cpuhp_online_idle(CPUHP_AP_ONLINE_IDLE);
> > > +    /* FIXME: converge cpu_bringup_and_idle() and start_secondary() */
> > > +    cpu_bringup_and_idle();
> > 
> > I think this will leak stack memory. Multiple cpu offline/online cycles of
> > the same cpu will finally exhaust the idle stack.

Doh!  Of course...

I was actually thinking ahead, to where eventually xen_pv_play_dead()
can call start_cpu0(), which can be changed to automatically reset the
stack pointer like this:

SYM_CODE_START(start_cpu0)
        ANNOTATE_NOENDBR
        UNWIND_HINT_EMPTY
        movq    PER_CPU_VAR(pcpu_hot + X86_top_of_stack), %rax
        leaq    -PTREGS_SIZE(%rax), %rsp
        jmp     .Ljump_to_C_code
SYM_CODE_END(start_cpu0)

but that would only be possible be after more cleanups which converge
cpu_bringup_and_idle() with start_secondary().

> The attached patch seems to work fine.

The patch looks good to me.

It doesn't solve Paul's original issue where arch_cpu_idle_dead() needs
to be __noreturn.  But that should probably be a separate patch anyway.

> The __noreturn annotation seems to trigger an objtool warning, though, in
> spite of the added BUG() at the end of xen_pv_play_dead():
>
> arch/x86/xen/smp_pv.o: warning: objtool: xen_pv_play_dead() falls through to
> next function xen_pv_cpu_die()

You'll need to tell objtool that xen_cpu_bringup_again() is noreturn by
adding "xen_cpu_bringup_again" to global_noreturns[] in
tools/objtool/check.c.

(Yes it's a pain, I'll be working an improved solution to the noreturn
thing...)

-- 
Josh