On Sun, Sep 27, 2026 at 06:03:59PM +0800, Kunwu Chan wrote:
> Add support for scaling atomic SRCU grace periods using
> srcu_read_lock_atomic()/srcu_read_unlock_atomic() and
> synchronize_srcu_atomic().
>
> Signed-off-by: Kunwu Chan <[email protected]>
Nice, thank you!!!
Some suggestions below.
Thanx, Paul
> ---
> kernel/rcu/rcuscale.c | 47 ++++++++++++++++++++++++++++++++++++++++++-
> 1 file changed, 46 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/rcu/rcuscale.c b/kernel/rcu/rcuscale.c
> index 5aca31c3c01d..86f1a912ebee 100644
> --- a/kernel/rcu/rcuscale.c
> +++ b/kernel/rcu/rcuscale.c
> @@ -264,6 +264,51 @@ static struct rcu_scale_ops srcu_ops = {
> .name = "srcu"
> };
>
> +static struct srcu_struct srcu_atomic_ctlp;
The trailing "p" suggests that this is a pointer, for example, see the
difference between srcu_ctl_scale and srcu_ctlp. You could make this
consistent with "srcud" and save a few lines of code as follows:
static struct srcu_struct srcua;
> +static int srcu_scale_atomic_read_lock(void)
> +{
> + return srcu_read_lock_atomic(&srcu_atomic_ctlp);
> +}
> +
> +static void srcu_scale_atomic_read_unlock(int idx)
> +{
> + srcu_read_unlock_atomic(&srcu_atomic_ctlp, idx);
> +}
> +
> +static void srcu_scale_atomic_synchronize(void)
> +{
> + synchronize_srcu_atomic(&srcu_atomic_ctlp);
> +}
> +
> +static void srcu_atomic_scale_init(void)
> +{
srcu_ctlp = &srcua;
> + init_srcu_struct_atomic(srcu_ctlp);
> +}
> +
> +static void srcu_atomic_scale_cleanup(void)
> +{
> + cleanup_srcu_struct(&srcu_atomic_ctlp);
> +}
You can then drop srcu_atomic_scale_cleanup() and just use the existing
srcu_sync_scale_cleanup().
> +static unsigned long srcu_atomic_scale_completed(void)
> +{
> + return srcu_batches_completed(&srcu_atomic_ctlp);
> +}
You can also drop srcu_atomic_scale_completed() and just use the existing
srcu_scale_completed(). Of course, both this and the above require
adjusting the srcu_atomic_ops initializer.
> +static struct rcu_scale_ops srcu_atomic_ops = {
> + .ptype = SRCU_FLAVOR,
> + .init = srcu_atomic_scale_init,
> + .cleanup = srcu_atomic_scale_cleanup,
> + .readlock = srcu_scale_atomic_read_lock,
> + .readunlock = srcu_scale_atomic_read_unlock,
> + .get_gp_seq = srcu_atomic_scale_completed,
> + .gp_diff = rcu_seq_diff,
> + .exp_completed = srcu_atomic_scale_completed,
> + .sync = srcu_scale_atomic_synchronize,
You can also then do this to get more debug data:
.stats = srcu_scale_stats,
> + .name = "srcu_atomic"
Then this could be "srcua". Or maybe more typing on the rcuscale
scale_type module parameter is OK. If so, maybe we should change
"srcud" to "srcu_dynamic", but this would require quite a few other
changes as well. (At least if we want the rcutorture names to be
consistent.)
Thoughts?
> +};
> +
> static struct srcu_struct srcud;
>
> static void srcu_sync_scale_init(void)
> @@ -1176,7 +1221,7 @@ rcu_scale_init(void)
> long i;
> long j;
> static struct rcu_scale_ops *scale_ops[] = {
> - &rcu_ops, &srcu_ops, &srcud_ops,
> + &rcu_ops, &srcu_ops, &srcu_atomic_ops, &srcud_ops,
> TASKS_OPS TASKS_RUDE_OPS TASKS_TRACING_OPS
> HAZPTR_SCALE_OPS
> };
> --
> 2.43.0
>
