Re: [PATCH v1] arm64: mm: Don't sleep in split_kernel_leaf_mapping() when in atomic context

[David Hildenbrand (Red Hat)]

On 03.11.25 18:28, Ryan Roberts wrote:
> On 03/11/2025 15:38, David Hildenbrand (Red Hat) wrote:
> > >    }
> > >
> > > +static inline bool force_pte_mapping(void)
> > > +{
> > > +    bool bbml2 = system_capabilities_finalized() ?
> > > +        system_supports_bbml2_noabort() : cpu_supports_bbml2_noabort();
> >
> > You are only moving this function. Still, there is some room for improvement I
> > want to point out :)
> >
> > bbml2 could be a const (or a helper function like bbml2_supported).
> >
> > > +
> > > +    return (!bbml2 && (rodata_full || arm64_kfence_can_set_direct_map() ||
> > > +               is_realm_world())) ||
> > > +        debug_pagealloc_enabled();
> >
> >
> > I suspect this could be made a bit easier to read.
> >
> >      if (debug_pagealloc_enabled())
> >          return true;
> >      if (bbml2)
> >          return false;
> >      return rodata_full || arm64_kfence_can_set_direct_map() || is_realm_world();
>
> Yeah, I guess that's a bit nicer. I'd prefer to tidy it up in as separate commit
> though. (feel free ;-) )

Separate commit is fine (hoping you can do it once this lands :P ).

> > > +}
> > > +
> > >    static DEFINE_MUTEX(pgtable_split_lock);
> > >
> > >    int split_kernel_leaf_mapping(unsigned long start, unsigned long end)
> > > @@ -723,6 +733,16 @@ int split_kernel_leaf_mapping(unsigned long start,
> > > unsigned long end)
> > >        if (!system_supports_bbml2_noabort())
> > >            return 0;
> > >
> > > +    /*
> > > +     * If the region is within a pte-mapped area, there is no need to try to
> > > +     * split. Additionally, CONFIG_DEBUG_PAGEALLOC and CONFIG_KFENCE may
> > > +     * change permissions from softirq context so for those cases (which are
> > > +     * always pte-mapped), we must not go any further because taking the
> > > +     * mutex below may sleep.
> > > +     */
> > > +    if (force_pte_mapping() || is_kfence_address((void *)start))
> > > +        return 0;
> > > +
> >
> > We're effectively performing two system_supports_bbml2_noabort() checks,
> > similarly in
> > arch_kfence_init_pool().
> >
> > I wonder if there is a clean way to avoid that.
>
> I thought about this too. But system_supports_bbml2_noabort() is actually a
> magic alternatives patching thing; 

Makes sense, so likely just another nop in the final code.

> the code is updated so it's zero overhead. I
> decided this was the simplest and clearest way to do it. But I'm open to other
> ideas...

Given that we have two such call sequences, I was wondering if we could 
have a helper that better expresses+documents the desired semantics.

static bool pte_leaf_split_possible()
{
	/*
	 * !BBML2_NOABORT systems should never run into scenarios where
         * we would have to split. So exit early and let calling code
	 * detect it + raise a warning.
	 */
	if (!system_supports_bbml2_noabort())
		return false;
	return force_pte_mapping();
}

Something like that maybe.

--
Cheers

David