Re: [PATCH v2 6/6] drm: rcar-du: Add group hooks for atomic-commit
From: Laurent Pinchart
Date: Sun May 12 2019 - 12:42:56 EST
Hi Kieran,
Thank you for the patch.
On Fri, Mar 15, 2019 at 05:01:10PM +0000, Kieran Bingham wrote:
> The group can now be handled independently from the CRTC tracking its
> own state.
>
> Introduce an rcar_du_group_atomic_check() call which will iterate the
> CRTCs to determine the per-state use-count of the group.
>
> This use count then allows us to determine if the group should be
> configured or disabled in our commit tail through the introduction of
> two new calls rcar_du_group_atomic_{pre,post}_commit().
>
> The existing rcar_du_group_{get,put}() functions are now redundant and
> removed along with their interactions in the CRTC get/put calls.
>
> The groups share clocking with the CRTCs within the group, and so are
> accessible only when one of its CRTCs has been powered through
> rcar_du_crtc_atomic_exit_standby().
>
> Signed-off-by: Kieran Bingham <kieran.bingham+renesas@xxxxxxxxxxxxxxxx>
>
> ---
> v2:
> - All register sequences now maintained.
> - Clock management is no longer handled by the group
> (_crtc_{exit,enter}_standby handles this for us)
>
>
>
> drivers/gpu/drm/rcar-du/rcar_du_crtc.c | 8 --
> drivers/gpu/drm/rcar-du/rcar_du_group.c | 101 ++++++++++++++++--------
> drivers/gpu/drm/rcar-du/rcar_du_group.h | 14 +++-
> drivers/gpu/drm/rcar-du/rcar_du_kms.c | 6 ++
> 4 files changed, 85 insertions(+), 44 deletions(-)
>
> diff --git a/drivers/gpu/drm/rcar-du/rcar_du_crtc.c b/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
> index 2606de788688..fce8bd1d9d25 100644
> --- a/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
> +++ b/drivers/gpu/drm/rcar-du/rcar_du_crtc.c
> @@ -511,14 +511,8 @@ static int rcar_du_crtc_enable(struct rcar_du_crtc *rcrtc)
> if (ret < 0)
> goto error_clock;
>
> - ret = rcar_du_group_get(rcrtc->group);
> - if (ret < 0)
> - goto error_group;
> -
> return 0;
>
> -error_group:
> - clk_disable_unprepare(rcrtc->extclock);
> error_clock:
> clk_disable_unprepare(rcrtc->clock);
> return ret;
You can simplify this function further by inlining the
clk_disable_unprepare() in the single error path and removing the
error_clock label.
> @@ -526,8 +520,6 @@ static int rcar_du_crtc_enable(struct rcar_du_crtc *rcrtc)
>
> static void rcar_du_crtc_disable(struct rcar_du_crtc *rcrtc)
> {
> - rcar_du_group_put(rcrtc->group);
> -
> clk_disable_unprepare(rcrtc->extclock);
> clk_disable_unprepare(rcrtc->clock);
> }
> diff --git a/drivers/gpu/drm/rcar-du/rcar_du_group.c b/drivers/gpu/drm/rcar-du/rcar_du_group.c
> index 9c82d666f170..1f9504bca0f3 100644
> --- a/drivers/gpu/drm/rcar-du/rcar_du_group.c
> +++ b/drivers/gpu/drm/rcar-du/rcar_du_group.c
> @@ -24,6 +24,7 @@
> */
>
> #include <linux/clk.h>
> +#include <linux/err.h>
> #include <linux/io.h>
>
> #include <drm/drm_atomic.h>
> @@ -172,38 +173,6 @@ static void rcar_du_group_setup(struct rcar_du_group *rgrp)
> mutex_unlock(&rgrp->lock);
> }
>
> -/*
> - * rcar_du_group_get - Acquire a reference to the DU channels group
> - *
> - * Acquiring the first reference setups core registers. A reference must be held
> - * before accessing any hardware registers.
> - *
> - * This function must be called with the DRM mode_config lock held.
> - *
> - * Return 0 in case of success or a negative error code otherwise.
> - */
> -int rcar_du_group_get(struct rcar_du_group *rgrp)
> -{
> - if (rgrp->use_count)
> - goto done;
> -
> - rcar_du_group_setup(rgrp);
> -
> -done:
> - rgrp->use_count++;
> - return 0;
> -}
> -
> -/*
> - * rcar_du_group_put - Release a reference to the DU
> - *
> - * This function must be called with the DRM mode_config lock held.
> - */
> -void rcar_du_group_put(struct rcar_du_group *rgrp)
> -{
> - --rgrp->use_count;
> -}
> -
> static void __rcar_du_group_start_stop(struct rcar_du_group *rgrp, bool start)
> {
> struct rcar_du_device *rcdu = rgrp->dev;
> @@ -384,6 +353,74 @@ static const struct drm_private_state_funcs rcar_du_group_state_funcs = {
> .atomic_destroy_state = rcar_du_group_atomic_destroy_state,
> };
>
> +#define for_each_oldnew_group_in_state(__state, __obj, __old_state, __new_state, __i) \
> + for_each_oldnew_private_obj_in_state((__state), (__obj), (__old_state), (__new_state), (__i)) \
> + for_each_if((__obj)->funcs == &rcar_du_group_state_funcs)
> +
> +static struct rcar_du_group_state *
> +rcar_du_get_group_state(struct drm_atomic_state *state,
> + struct rcar_du_group *rgrp)
> +{
> + struct drm_private_state *pstate;
> +
> + pstate = drm_atomic_get_private_obj_state(state, &rgrp->private);
> + if (IS_ERR(pstate))
> + return ERR_CAST(pstate);
> +
> + return to_rcar_group_state(pstate);
> +}
> +
> +int rcar_du_group_atomic_check(struct drm_device *dev,
> + struct drm_atomic_state *state)
> +{
> + struct drm_crtc *crtc;
> + struct drm_crtc_state *crtc_state;
> + unsigned int i;
> +
> + for_each_new_crtc_in_state(state, crtc, crtc_state, i) {
> + struct rcar_du_crtc *rcrtc = to_rcar_crtc(crtc);
> + struct rcar_du_group_state *rstate;
> +
> + if (crtc_state->active_changed || crtc_state->mode_changed) {
> + rstate = rcar_du_get_group_state(state, rcrtc->group);
> + if (IS_ERR(rstate))
> + return PTR_ERR(rstate);
> +
> + if (crtc_state->active)
> + rstate->use_count++;
This doesn't seem right to me. A commit with just a page flip will have
neither active_changed not mode_changed set. The group use count will
thus be 0. If the next commit enables the second CRTC in the group, then
you will call rcar_du_group_setup() again in
rcar_du_group_atomic_pre_commit().
I think you should increment use_count when crtc_state->active
regardless of active_changed and mode_changed.
> + }
> + }
> +
> + return 0;
> +}
> +
> +int rcar_du_group_atomic_pre_commit(struct drm_device *dev,
> + struct drm_atomic_state *state)
> +{
> + struct drm_private_state *old_pstate, *new_pstate;
> + struct drm_private_obj *obj;
> + unsigned int i;
> +
> + for_each_oldnew_group_in_state(state, obj, old_pstate, new_pstate, i) {
> + struct rcar_du_group *rgrp = to_rcar_group(obj);
> + struct rcar_du_group_state *old_state, *new_state;
> +
> + old_state = to_rcar_group_state(old_pstate);
> + new_state = to_rcar_group_state(new_pstate);
> +
> + if (!old_state->use_count && new_state->use_count)
> + rcar_du_group_setup(rgrp);
> + }
> +
> + return 0;
> +}
> +
> +int rcar_du_group_atomic_post_commit(struct drm_device *dev,
> + struct drm_atomic_state *state)
> +{
> + return 0;
> +}
As for the similar CRTC patch, I think you can drop
rcar_du_group_atomic_post_commit(), turn
rcar_du_group_atomic_pre_commit() into a void function, and rename it to
rcar_du_group_atomic_setup().
The rest looks good to me, so with this fixed,
Reviewed-by: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
> +
> /*
> * rcar_du_group_init - Initialise and reset a group object
> *
> diff --git a/drivers/gpu/drm/rcar-du/rcar_du_group.h b/drivers/gpu/drm/rcar-du/rcar_du_group.h
> index 4b812e167987..288c09a6d7d0 100644
> --- a/drivers/gpu/drm/rcar-du/rcar_du_group.h
> +++ b/drivers/gpu/drm/rcar-du/rcar_du_group.h
> @@ -26,7 +26,6 @@ struct rcar_du_device;
> * @index: group index
> * @channels_mask: bitmask of populated DU channels in this group
> * @num_crtcs: number of CRTCs in this group (1 or 2)
> - * @use_count: number of users of the group (rcar_du_group_(get|put))
> * @used_crtcs: number of CRTCs currently in use
> * @lock: protects the dptsr_planes field and the DPTSR register
> * @dptsr_planes: bitmask of planes driven by dot-clock and timing generator 1
> @@ -43,7 +42,6 @@ struct rcar_du_group {
>
> unsigned int channels_mask;
> unsigned int num_crtcs;
> - unsigned int use_count;
> unsigned int used_crtcs;
>
> struct mutex lock;
> @@ -59,9 +57,12 @@ struct rcar_du_group {
> /**
> * struct rcar_du_group_state - Driver-specific group state
> * @state: base DRM private state
> + * @use_count: number of users of the group
> */
> struct rcar_du_group_state {
> struct drm_private_state state;
> +
> + unsigned int use_count;
> };
>
> #define to_rcar_group_state(s) \
> @@ -70,14 +71,19 @@ struct rcar_du_group_state {
> u32 rcar_du_group_read(struct rcar_du_group *rgrp, u32 reg);
> void rcar_du_group_write(struct rcar_du_group *rgrp, u32 reg, u32 data);
>
> -int rcar_du_group_get(struct rcar_du_group *rgrp);
> -void rcar_du_group_put(struct rcar_du_group *rgrp);
> void rcar_du_group_start_stop(struct rcar_du_group *rgrp, bool start);
> void rcar_du_group_restart(struct rcar_du_group *rgrp);
> int rcar_du_group_set_routing(struct rcar_du_group *rgrp);
>
> int rcar_du_set_dpad0_vsp1_routing(struct rcar_du_device *rcdu);
>
> +int rcar_du_group_atomic_check(struct drm_device *dev,
> + struct drm_atomic_state *state);
> +int rcar_du_group_atomic_pre_commit(struct drm_device *dev,
> + struct drm_atomic_state *state);
> +int rcar_du_group_atomic_post_commit(struct drm_device *dev,
> + struct drm_atomic_state *state);
> +
> int rcar_du_group_init(struct rcar_du_device *rcdu, struct rcar_du_group *rgrp,
> unsigned int index);
> void rcar_du_group_cleanup(struct rcar_du_group *rgrp);
> diff --git a/drivers/gpu/drm/rcar-du/rcar_du_kms.c b/drivers/gpu/drm/rcar-du/rcar_du_kms.c
> index eb01bea1ab83..77f0ff38298b 100644
> --- a/drivers/gpu/drm/rcar-du/rcar_du_kms.c
> +++ b/drivers/gpu/drm/rcar-du/rcar_du_kms.c
> @@ -271,6 +271,10 @@ static int rcar_du_atomic_check(struct drm_device *dev,
> if (ret)
> return ret;
>
> + ret = rcar_du_group_atomic_check(dev, state);
> + if (ret)
> + return ret;
> +
> if (rcar_du_has(rcdu, RCAR_DU_FEATURE_VSP1_SOURCE))
> return 0;
>
> @@ -305,6 +309,7 @@ static void rcar_du_atomic_commit_tail(struct drm_atomic_state *old_state)
>
> /* Apply the atomic update. */
> rcar_du_crtc_atomic_exit_standby(dev, old_state);
> + rcar_du_group_atomic_pre_commit(dev, old_state);
> rcar_du_crtc_atomic_pre_commit(dev, old_state);
>
> drm_atomic_helper_commit_modeset_disables(dev, old_state);
> @@ -313,6 +318,7 @@ static void rcar_du_atomic_commit_tail(struct drm_atomic_state *old_state)
> drm_atomic_helper_commit_modeset_enables(dev, old_state);
>
> rcar_du_crtc_atomic_post_commit(dev, old_state);
> + rcar_du_group_atomic_post_commit(dev, old_state);
> rcar_du_crtc_atomic_enter_standby(dev, old_state);
>
> drm_atomic_helper_commit_hw_done(old_state);
--
Regards,
Laurent Pinchart