Re: [PATCH 3/4] pmdomain: qcom: rpmhpd: Let genpd enforce the boot performance floor
From: Ulf Hansson
Date: Thu Nov 12 2026 - 09:20:35 EST
On Tue, Oct 13, 2026 at 2:49 PM Abel Vesa <abel.vesa@xxxxxxxxxxxxxxxx> wrote:
>
> RPMh power domains are registered as powered off without knowing their
> initial hardware state. Until the provider's sync_state() callback runs,
> rpmhpd_aggregate_corner() replaces requests with the highest supported
> corner, leaving genpd unaware of this boot performance constraint.
>
> Set GENPD_FLAG_POWER_UNKNOWN and provide the highest mapped performance
> level as boot_performance_state. Reject an empty mapping before indexing
> its last entry. Let genpd enforce the floor and retain consumer-enabled
> domains while boot protection is active, instead of clamping corners
> inside the provider.
>
> Retain state_synced for the hardware-specific sleep-vote policy. Include
> the supplied corner in sleep votes for active-only domains until the
> provider's sync_state() callback releases that protection. This also
> applies when translating an enabled peer's vote. Keep the existing sync
> callback to reconcile votes for domains left software-OFF by genpd.
Hmm, perhaps we should start by just enabling GENPD_FLAG_POWER_UNKNOWN
- and making that as a preparation patch, placed first in the series?
Would that not be a more cautious approach, taking it step by step?
Kind regards
Uffe
>
> Assisted-by: LLM
> Signed-off-by: Abel Vesa <abel.vesa@xxxxxxxxxxxxxxxx>
> ---
> drivers/pmdomain/qcom/rpmhpd.c | 17 ++++++++---------
> 1 file changed, 8 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/pmdomain/qcom/rpmhpd.c b/drivers/pmdomain/qcom/rpmhpd.c
> index fa8fed5bd3e5..1488f31bcdda 100644
> --- a/drivers/pmdomain/qcom/rpmhpd.c
> +++ b/drivers/pmdomain/qcom/rpmhpd.c
> @@ -40,7 +40,7 @@
> * @res_name: Resource name used for cmd-db lookup
> * @addr: Resource address as looped up using resource name from
> * cmd-db
> - * @state_synced: Indicator that sync_state has been invoked for the rpmhpd resource
> + * @state_synced: Whether boot protection of sleep votes has been released
> */
> struct rpmhpd {
> struct device *dev;
> @@ -982,7 +982,7 @@ static void to_active_sleep(struct rpmhpd *pd, unsigned int corner,
> {
> *active = corner;
>
> - if (pd->active_only)
> + if (pd->active_only && pd->state_synced)
> *sleep = 0;
> else
> *sleep = *active;
> @@ -1006,13 +1006,7 @@ static int rpmhpd_aggregate_corner(struct rpmhpd *pd, unsigned int corner)
> unsigned int peer_active_corner = 0, peer_sleep_corner = 0;
> unsigned int peer_enabled_corner;
>
> - if (pd->state_synced) {
> - to_active_sleep(pd, corner, &this_active_corner, &this_sleep_corner);
> - } else {
> - /* Clamp to highest corner if sync_state hasn't happened */
> - this_active_corner = pd->level_count - 1;
> - this_sleep_corner = pd->level_count - 1;
> - }
> + to_active_sleep(pd, corner, &this_active_corner, &this_sleep_corner);
>
> if (peer && peer->enabled) {
> peer_enabled_corner = max(peer->corner, peer->enable_corner);
> @@ -1210,10 +1204,15 @@ static int rpmhpd_probe(struct platform_device *pdev)
> ret = rpmhpd_update_level_mapping(rpmhpds[i]);
> if (ret)
> return ret;
> + if (!rpmhpds[i]->level_count)
> + return -EINVAL;
>
> rpmhpds[i]->pd.power_off = rpmhpd_power_off;
> rpmhpds[i]->pd.power_on = rpmhpd_power_on;
> rpmhpds[i]->pd.set_performance_state = rpmhpd_set_performance_state;
> + rpmhpds[i]->pd.flags |= GENPD_FLAG_POWER_UNKNOWN;
> + rpmhpds[i]->pd.boot_performance_state =
> + rpmhpds[i]->level[rpmhpds[i]->level_count - 1];
> pm_genpd_init(&rpmhpds[i]->pd, NULL, true);
>
> data->domains[i] = &rpmhpds[i]->pd;
>
> --
> 2.54.0
>