Re: [PATCH v3 21/30] maple_tree: Add cp_is_new_root() helper
From: Liam R. Howlett
Date: Mon Feb 02 2026 - 10:08:28 EST
* SeongJae Park <sj@xxxxxxxxxx> [260131 19:10]:
> Hello,
>
> On Fri, 30 Jan 2026 15:59:26 -0500 "Liam R. Howlett" <Liam.Howlett@xxxxxxxxxx> wrote:
>
> > Add a helper to do what is needed when the maple copy node contains a
> > new root node. This is useful for future commits and is
> > self-documenting code.
> >
> > Signed-off-by: Liam R. Howlett <Liam.Howlett@xxxxxxxxxx>
> > ---
> > lib/maple_tree.c | 70 ++++++++++++++++++++++++++----------------------
> > 1 file changed, 38 insertions(+), 32 deletions(-)
> >
> > diff --git a/lib/maple_tree.c b/lib/maple_tree.c
> > index 5280fa6d2d6ec..42038e42a4c7e 100644
> > --- a/lib/maple_tree.c
> > +++ b/lib/maple_tree.c
> > @@ -3337,6 +3337,43 @@ static void mas_spanning_rebalance(struct ma_state *mas,
> > mas_spanning_rebalance_loop(mas, mast, count);
> > }
> >
> > +static inline bool cp_is_new_root(struct maple_copy *cp, struct ma_state *mas)
> > +{
> > + if (cp->min || cp->max != ULONG_MAX)
> > + return false;
> > +
> > + if (cp->d_count != 1) {
> > + enum maple_type mt = maple_arange_64;
> > +
> > + if (!mt_is_alloc(mas->tree))
> > + mt = maple_range_64;
> > +
> > + cp->data = cp->d_count;
> > + cp->s_count = 0;
> > + dst_setup(cp, mas, mt);
> > + init_cp_src(cp);
> > + node_copy(mas, cp->src[0].node, 0, cp->data, cp->max, maple_copy,
> > + cp->dst[0].node, 0, mt);
> > + node_finalise(cp->dst[0].node, mt, cp->end + 1);
> > + /*
> > + * Warning, see cp_leaf_init() comment and rcu_assign_pointer()
> > + * documentation. Since this is a new root, there are no
> > + * read-side operations that can view it until it is insert into
> > + * the tree after an rcu_assign_pointer() call.
> > + */
> > + RCU_INIT_POINTER(cp->slot[0], mt_mk_node(cp->dst[0].node, mt));
>
> I just found the above makes my build test using an old version compiler fails.
> Fortunately, seems it is same to the one we discussed before [1], and same
> mitigation like below attached patch works, at least for my test setup.
Thanks SJ.
This is still with gcc 8.1.0?
I thought debian stable would be old enough.
Thanks,
Liam
>
> [1] https://lore.kernel.org/dwhxxuil4zkesmyj6xviyyyfedrcd65h6qd4bplmcrsg36purj@f523i7t6nxag
>
>
> Thanks,
> SJ
>
> [...]
> === >8 ===
> From ecc4e468d72c431d53043c8a61fddb6ddf2ecf7c Mon Sep 17 00:00:00 2001
> From: SeongJae Park <sj@xxxxxxxxxx>
> Date: Sat, 31 Jan 2026 16:02:56 -0800
> Subject: [PATCH] lib/mape_tree: temporal build fix
>
> Without the fix, build with old compilers fails like below:
>
> CC lib/maple_tree.o
> In file included from .../arch/arm64/include/asm/rwonce.h:67,
> from .../include/linux/compiler.h:380,
> from .../include/linux/array_size.h:5,
> from .../include/linux/kernel.h:16,
> from .../include/linux/maple_tree.h:11,
> from .../lib/maple_tree.c:56:
> .../lib/maple_tree.c: In function 'cp_is_new_root':
> .../include/linux/rcupdate.h:555:36: error: dereferencing pointer to incomplete type 'struct maple_enode'
> 555 | #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
> | ^~~~
> .../include/asm-generic/rwonce.h:55:33: note: in definition of macro '__WRITE_ONCE'
> 55 | *(volatile typeof(x) *)&(x) = (val); \
> | ^~~
> .../include/linux/rcupdate.h:1046:3: note: in expansion of macro 'WRITE_ONCE'
> 1046 | WRITE_ONCE(p, RCU_INITIALIZER(v)); \
> | ^~~~~~~~~~
> .../include/linux/rcupdate.h:1046:17: note: in expansion of macro 'RCU_INITIALIZER'
> 1046 | WRITE_ONCE(p, RCU_INITIALIZER(v)); \
> | ^~~~~~~~~~~~~~~
> .../lib/maple_tree.c:3364:3: note: in expansion of macro 'RCU_INIT_POINTER'
> 3364 | RCU_INIT_POINTER(cp->slot[0], mt_mk_node(cp->dst[0].node, mt));
> | ^~~~~~~~~~~~~~~~
>
> Signed-off-by: SeongJae Park <sj@xxxxxxxxxx>
> ---
> lib/maple_tree.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/lib/maple_tree.c b/lib/maple_tree.c
> index c522419e99f4e..eb2855269332a 100644
> --- a/lib/maple_tree.c
> +++ b/lib/maple_tree.c
> @@ -3361,7 +3361,8 @@ static inline bool cp_is_new_root(struct maple_copy *cp, struct ma_state *mas)
> * read-side operations that can view it until it is insert into
> * the tree after an rcu_assign_pointer() call.
> */
> - RCU_INIT_POINTER(cp->slot[0], mt_mk_node(cp->dst[0].node, mt));
> + RCU_INIT_POINTER(cp->slot[0],
> + (void *)mt_mk_node(cp->dst[0].node, mt));
> cp->height++;
> }
> WARN_ON_ONCE(cp->dst[0].node != mte_to_node(
> --
> 2.47.3
>