Re: [PATCH 2/2] drm/mediatek: Apply CMDQ control flow

From: CK Hu
Date: Fri Aug 30 2019 - 03:58:52 EST


Hi, Bibby:

On Fri, 2019-08-30 at 15:41 +0800, Bibby Hsieh wrote:
> Unlike other SoCs, MT8183 does not have "shadow"
> registers for performaing an atomic video mode
> set or page flip at vblank/vsync.
>
> The CMDQ (Commend Queue) in MT8183 is used to help
> update all relevant display controller registers
> with critical time limation.
>
> Signed-off-by: YT Shen <yt.shen@xxxxxxxxxxxx>
> Signed-off-by: CK Hu <ck.hu@xxxxxxxxxxxx>
> Signed-off-by: Philipp Zabel <p.zabel@xxxxxxxxxxxxxx>
> Signed-off-by: Bibby Hsieh <bibby.hsieh@xxxxxxxxxxxx>
> Signed-off-by: Yongqiang Niu <yongqiang.niu@xxxxxxxxxxxx>
> ---
> drivers/gpu/drm/mediatek/mtk_drm_crtc.c | 190 +++++++++++++++++---
> drivers/gpu/drm/mediatek/mtk_drm_crtc.h | 2 +
> drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.c | 34 ++++
> drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.h | 2 +
> drivers/gpu/drm/mediatek/mtk_drm_plane.c | 4 +
> 5 files changed, 206 insertions(+), 26 deletions(-)
>
> diff --git a/drivers/gpu/drm/mediatek/mtk_drm_crtc.c b/drivers/gpu/drm/mediatek/mtk_drm_crtc.c
> index 092e502ed27b..329ca5a14c39 100644
> --- a/drivers/gpu/drm/mediatek/mtk_drm_crtc.c
> +++ b/drivers/gpu/drm/mediatek/mtk_drm_crtc.c
> @@ -10,7 +10,9 @@
> #include <drm/drm_plane_helper.h>
> #include <drm/drm_probe_helper.h>
> #include <linux/clk.h>
> +#include <linux/of_address.h>
> #include <linux/pm_runtime.h>
> +#include <linux/soc/mediatek/mtk-cmdq.h>
>
> #include "mtk_drm_drv.h"
> #include "mtk_drm_crtc.h"
> @@ -41,6 +43,10 @@ struct mtk_drm_crtc {
> unsigned int layer_nr;
> bool pending_planes;
> bool cursor_update;
> +
> + struct cmdq_client *cmdq_client;
> + u32 cmdq_event;
> +
> void __iomem *config_regs;
> const struct mtk_mmsys_reg_data *mmsys_reg_data;
> struct mtk_disp_mutex *mutex;
> @@ -57,6 +63,12 @@ struct mtk_crtc_state {
> unsigned int pending_width;
> unsigned int pending_height;
> unsigned int pending_vrefresh;
> + struct cmdq_pkt *cmdq_handle;
> +};
> +
> +struct mtk_cmdq_cb_data {
> + struct drm_crtc_state *state;
> + struct cmdq_pkt *cmdq_handle;
> };
>
> static inline struct mtk_drm_crtc *to_mtk_crtc(struct drm_crtc *c)
> @@ -208,6 +220,46 @@ static void mtk_crtc_ddp_clk_disable(struct mtk_drm_crtc *mtk_crtc)
> clk_disable_unprepare(mtk_crtc->ddp_comp[i]->clk);
> }
>
> +static void ddp_cmdq_cursor_cb(struct cmdq_cb_data data)
> +{
> +
> +#if IS_ENABLED(CONFIG_MTK_CMDQ)

I would like the whole ddp_cmdq_cursor_cb() and ddp_cmdq_cb() inside
this part.

> + struct mtk_cmdq_cb_data *cb_data = data.data;
> +
> + DRM_DEBUG_DRIVER("%s\n", __func__);
> +
> + cmdq_pkt_destroy(cb_data->cmdq_handle);
> + kfree(cb_data);
> +#endif
> +
> +}
> +
> +static void ddp_cmdq_cb(struct cmdq_cb_data data)
> +{
> +
> +#if IS_ENABLED(CONFIG_MTK_CMDQ)
> + struct mtk_cmdq_cb_data *cb_data = data.data;
> + struct drm_crtc_state *crtc_state = cb_data->state;
> + struct drm_atomic_state *atomic_state = crtc_state->state;
> + struct drm_crtc *crtc = crtc_state->crtc;
> + struct mtk_drm_crtc *mtk_crtc = to_mtk_crtc(crtc);
> +
> + DRM_DEBUG_DRIVER("%s\n", __func__);
> +
> + if (mtk_crtc->pending_needs_vblank) {
> + /* cmdq_vblank_event must be read after cmdq_needs_event */
> + smp_rmb();
> +
> + mtk_drm_crtc_finish_page_flip(mtk_crtc);
> + mtk_crtc->pending_needs_vblank = false;
> + }
> + mtk_atomic_state_put_queue(atomic_state);
> + cmdq_pkt_destroy(cb_data->cmdq_handle);
> + kfree(cb_data);
> +#endif
> +
> +}
> +
> static int mtk_crtc_ddp_hw_init(struct mtk_drm_crtc *mtk_crtc)
> {
> struct drm_crtc *crtc = &mtk_crtc->base;
> @@ -283,7 +335,8 @@ static int mtk_crtc_ddp_hw_init(struct mtk_drm_crtc *mtk_crtc)
> if (prev == DDP_COMPONENT_OVL0)
> mtk_ddp_comp_bgclr_in_on(comp);
>
> - mtk_ddp_comp_config(comp, width, height, vrefresh, bpc);
> + mtk_ddp_comp_config(comp, width, height,
> + vrefresh, bpc, NULL);
> mtk_ddp_comp_start(comp);
> }
>
> @@ -303,7 +356,7 @@ static int mtk_crtc_ddp_hw_init(struct mtk_drm_crtc *mtk_crtc)
> } else
> local_layer = i;
> mtk_ddp_comp_layer_config(comp, local_layer,
> - plane_state);
> + plane_state, NULL);
> }
>
> return 0;
> @@ -361,7 +414,7 @@ static void mtk_crtc_ddp_config(struct drm_crtc *crtc)
> if (state->pending_config) {
> mtk_ddp_comp_config(comp, state->pending_width,
> state->pending_height,
> - state->pending_vrefresh, 0);
> + state->pending_vrefresh, 0, NULL);
>
> state->pending_config = false;
> }
> @@ -381,7 +434,7 @@ static void mtk_crtc_ddp_config(struct drm_crtc *crtc)
> local_layer = i;
>
> mtk_ddp_comp_layer_config(comp, local_layer,
> - plane_state);
> + plane_state, NULL);
> plane_state->pending.config = false;
> }
> }
> @@ -405,26 +458,69 @@ void mtk_drm_crtc_cursor_update(struct drm_crtc *crtc, struct drm_plane *plane,
> return;
>
> mutex_lock(&priv->hw_lock);
> - plane_helper_funcs->atomic_update(plane, plane_state);
> - for (i = 0; i < mtk_crtc->layer_nr; i++) {
> - struct drm_plane *plane = &mtk_crtc->planes[i];
> - struct mtk_plane_state *plane_state;
> + if (IS_ENABLED(CONFIG_MTK_CMDQ) && mtk_crtc->cmdq_client) {
> + struct mtk_crtc_state *mtk_crtc_state =
> + to_mtk_crtc_state(crtc->state);
> + struct mtk_cmdq_cb_data *cb_data;
> +
> + mtk_crtc_state->cmdq_handle =
> + cmdq_pkt_create(mtk_crtc->cmdq_client,
> + PAGE_SIZE);
> + cmdq_pkt_clear_event(mtk_crtc_state->cmdq_handle,
> + mtk_crtc->cmdq_event);
> + cmdq_pkt_wfe(mtk_crtc_state->cmdq_handle, mtk_crtc->cmdq_event);
> + plane_helper_funcs->atomic_update(plane, plane_state);
> + cb_data = kmalloc(sizeof(*cb_data), GFP_KERNEL);
> + cb_data->cmdq_handle = mtk_crtc_state->cmdq_handle;
> + cmdq_pkt_flush_async(mtk_crtc_state->cmdq_handle,
> + ddp_cmdq_cursor_cb, cb_data);
> + } else {
> + plane_helper_funcs->atomic_update(plane, plane_state);
>
> - plane_state = to_mtk_plane_state(plane->state);
> - if (plane_state->pending.dirty) {
> - plane_state->pending.config = true;
> - plane_state->pending.dirty = false;
> + for (i = 0; i < mtk_crtc->layer_nr; i++) {
> + struct drm_plane *plane = &mtk_crtc->planes[i];
> + struct mtk_plane_state *plane_state;
> +
> + plane_state = to_mtk_plane_state(plane->state);
> + if (plane_state->pending.dirty) {
> + plane_state->pending.config = true;
> + plane_state->pending.dirty = false;
> + }
> + }
> + mtk_crtc->pending_planes = true;
> + mtk_crtc->cursor_update = true;
> + if (priv->data->shadow_register) {
> + mtk_disp_mutex_acquire(mtk_crtc->mutex);
> + mtk_crtc_ddp_config(crtc);
> + mtk_disp_mutex_release(mtk_crtc->mutex);
> }
> }
> - mtk_crtc->pending_planes = true;
> - mtk_crtc->cursor_update = true;
> + mutex_unlock(&priv->hw_lock);
> +}
>
> - if (priv->data->shadow_register) {
> - mtk_disp_mutex_acquire(mtk_crtc->mutex);
> - mtk_crtc_ddp_config(crtc);
> - mtk_disp_mutex_release(mtk_crtc->mutex);

My idea for the control flow is:

1. In the end of mtk_drm_crtc_atomic_flush(), do something like shadow
register control,

if (priv->data->cmdq) {
/* abort cmd in queue */
/* Call mtk_crtc_ddp_config() to generate command */
/* send cmd to queue */
}
2. In mtk_crtc_ddp_config(), if cmdq is used, state->pending_config and
mtk_crtc->pending_planes would not be set to false. These two variable
would be set to false in cmdq callback.

Regards,
CK

> +void mtk_drm_crtc_plane_update(struct drm_crtc *crtc, struct drm_plane *plane,
> + struct mtk_plane_state *state)
> +{
> + struct mtk_drm_crtc *mtk_crtc = to_mtk_crtc(crtc);
> + struct mtk_ddp_comp *comp = mtk_crtc->ddp_comp[0];
> + struct drm_crtc_state *crtc_state = crtc->state;
> + struct mtk_crtc_state *mtk_crtc_state = to_mtk_crtc_state(crtc_state);
> + struct cmdq_pkt *cmdq_handle = mtk_crtc_state->cmdq_handle;
> + unsigned int comp_layer_nr = mtk_ddp_comp_layer_nr(comp);
> + unsigned int local_layer;
> + unsigned int plane_index = plane - mtk_crtc->planes;
> +
> + DRM_DEBUG_DRIVER("%s\n", __func__);
> + if (mtk_crtc->cmdq_client) {
> + if (plane_index >= comp_layer_nr) {
> + comp = mtk_crtc->ddp_comp[1];
> + local_layer = plane_index - comp_layer_nr;
> + } else {
> + local_layer = plane_index;
> + }
> + mtk_ddp_comp_layer_config(comp, local_layer, state,
> + cmdq_handle);
> }
> - mutex_unlock(&priv->hw_lock);
> }
>
> static void mtk_drm_crtc_atomic_enable(struct drm_crtc *crtc,
> @@ -487,6 +583,15 @@ static void mtk_drm_crtc_atomic_begin(struct drm_crtc *crtc,
> WARN_ON(drm_crtc_vblank_get(crtc) != 0);
> mtk_crtc->event = state->base.event;
> state->base.event = NULL;
> + /* Make sure the above parameter is set before update */
> + smp_wmb();
> + mtk_crtc->pending_needs_vblank = true;
> + }
> + if (IS_ENABLED(CONFIG_MTK_CMDQ) && mtk_crtc->cmdq_client) {
> + state->cmdq_handle = cmdq_pkt_create(mtk_crtc->cmdq_client,
> + PAGE_SIZE);
> + cmdq_pkt_clear_event(state->cmdq_handle, mtk_crtc->cmdq_event);
> + cmdq_pkt_wfe(state->cmdq_handle, mtk_crtc->cmdq_event);
> }
> }
>
> @@ -494,13 +599,29 @@ static void mtk_drm_crtc_atomic_flush(struct drm_crtc *crtc,
> struct drm_crtc_state *old_crtc_state)
> {
> struct drm_atomic_state *old_atomic_state = old_crtc_state->state;
> + struct drm_crtc_state *crtc_state = crtc->state;
> + struct mtk_crtc_state *state = to_mtk_crtc_state(crtc_state);
> + struct cmdq_pkt *cmdq_handle = state->cmdq_handle;
> struct mtk_drm_crtc *mtk_crtc = to_mtk_crtc(crtc);
> struct mtk_drm_private *priv = crtc->dev->dev_private;
> + struct mtk_cmdq_cb_data *cb_data;
> unsigned int pending_planes = 0;
> int i;
>
> - if (mtk_crtc->event)
> - mtk_crtc->pending_needs_vblank = true;
> + DRM_DEBUG_DRIVER("[CRTC:%u] [STATE:%p(%p)->%p(%p)]\n", crtc->base.id,
> + old_crtc_state, old_crtc_state->state,
> + crtc_state, crtc_state->state);
> +
> + if (IS_ENABLED(CONFIG_MTK_CMDQ) && mtk_crtc->cmdq_client) {
> + drm_atomic_state_get(old_atomic_state);
> + cb_data = kmalloc(sizeof(*cb_data), GFP_KERNEL);
> + cb_data->state = old_crtc_state;
> + cb_data->cmdq_handle = cmdq_handle;
> + cmdq_pkt_flush_async(cmdq_handle, ddp_cmdq_cb, cb_data);
> +
> + return;
> + }
> +
> for (i = 0; i < mtk_crtc->layer_nr; i++) {
> struct drm_plane *plane = &mtk_crtc->planes[i];
> struct mtk_plane_state *plane_state;
> @@ -521,7 +642,8 @@ static void mtk_drm_crtc_atomic_flush(struct drm_crtc *crtc,
>
> if (crtc->state->color_mgmt_changed)
> for (i = 0; i < mtk_crtc->ddp_comp_nr; i++)
> - mtk_ddp_gamma_set(mtk_crtc->ddp_comp[i], crtc->state);
> + mtk_ddp_gamma_set(mtk_crtc->ddp_comp[i],
> + crtc->state, NULL);
>
> if (priv->data->shadow_register) {
> mtk_disp_mutex_acquire(mtk_crtc->mutex);
> @@ -578,10 +700,13 @@ void mtk_crtc_ddp_irq(struct drm_crtc *crtc, struct mtk_ddp_comp *comp)
> struct mtk_drm_crtc *mtk_crtc = to_mtk_crtc(crtc);
> struct mtk_drm_private *priv = crtc->dev->dev_private;
>
> - if (!priv->data->shadow_register)
> - mtk_crtc_ddp_config(crtc);
> -
> - mtk_drm_finish_page_flip(mtk_crtc);
> + if (mtk_crtc->cmdq_client) {
> + drm_crtc_handle_vblank(crtc);
> + } else {
> + if (!priv->data->shadow_register)
> + mtk_crtc_ddp_config(crtc);
> + mtk_drm_finish_page_flip(mtk_crtc);
> + }
> }
>
> int mtk_drm_crtc_create(struct drm_device *drm_dev,
> @@ -678,5 +803,18 @@ int mtk_drm_crtc_create(struct drm_device *drm_dev,
> drm_crtc_enable_color_mgmt(&mtk_crtc->base, 0, false, MTK_LUT_SIZE);
> priv->num_pipes++;
>
> + if (IS_ENABLED(CONFIG_MTK_CMDQ)) {
> + mtk_crtc->cmdq_client = cmdq_mbox_create(dev,
> + drm_crtc_index(&mtk_crtc->base), 2000);
> + of_property_read_u32_index(dev->of_node, "mediatek,gce-events",
> + drm_crtc_index(&mtk_crtc->base),
> + &mtk_crtc->cmdq_event);
> + if (IS_ERR(mtk_crtc->cmdq_client)) {
> + dev_dbg(dev, "mtk_crtc %d failed to create mailbox client, writing register by CPU now\n",
> + drm_crtc_index(&mtk_crtc->base));
> + mtk_crtc->cmdq_client = NULL;
> + }
> + }
> +
> return 0;
> }
> diff --git a/drivers/gpu/drm/mediatek/mtk_drm_crtc.h b/drivers/gpu/drm/mediatek/mtk_drm_crtc.h
> index 46e903be68ec..6b2423c88416 100644
> --- a/drivers/gpu/drm/mediatek/mtk_drm_crtc.h
> +++ b/drivers/gpu/drm/mediatek/mtk_drm_crtc.h
> @@ -19,6 +19,8 @@ void mtk_crtc_ddp_irq(struct drm_crtc *crtc, struct mtk_ddp_comp *comp);
> int mtk_drm_crtc_create(struct drm_device *drm_dev,
> const enum mtk_ddp_comp_id *path,
> unsigned int path_len);
> +void mtk_drm_crtc_plane_update(struct drm_crtc *crtc, struct drm_plane *plane,
> + struct mtk_plane_state *state);
> void mtk_drm_crtc_cursor_update(struct drm_crtc *crtc, struct drm_plane *plane,
> struct drm_plane_state *plane_state);
>
> diff --git a/drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.c b/drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.c
> index 76416c1cbb28..056ebc6a8199 100644
> --- a/drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.c
> +++ b/drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.c
> @@ -395,6 +395,40 @@ int mtk_ddp_comp_init(struct device *dev, struct device_node *node,
> if (IS_ERR(comp->clk))
> return PTR_ERR(comp->clk);
>
> + if (IS_ENABLED(CONFIG_MTK_CMDQ)) {
> + struct platform_device *comp_pdev;
> + struct resource res;
> + struct cmdq_client_reg *cmdq_reg;
> + int ret = 0;
> +
> + if (of_address_to_resource(node, 0, &res) != 0) {
> + dev_err(dev, "Missing reg in %s node\n",
> + node->full_name);
> + return -EINVAL;
> + }
> + comp->regs_pa = res.start;
> +
> + comp_pdev = of_find_device_by_node(node);
> + if (!comp_pdev) {
> + dev_warn(dev, "Waiting for component device %s\n",
> + node->full_name);
> + return -EPROBE_DEFER;
> + }
> +
> + cmdq_reg = kzalloc(sizeof(*cmdq_reg), GFP_KERNEL);
> + if (!cmdq_reg)
> + return -EINVAL;
> +
> + ret = cmdq_dev_get_client_reg(&comp_pdev->dev, cmdq_reg, 0);
> + if (ret != 0)
> + dev_dbg(&comp_pdev->dev,
> + "get mediatek,gce-client-reg fail!\n");
> + else
> + comp->subsys = cmdq_reg->subsys;
> +
> + kfree(cmdq_reg);
> + }
> +
> return 0;
> }
>
> diff --git a/drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.h b/drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.h
> index 6bbc35f92815..0faec2dad5a3 100644
> --- a/drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.h
> +++ b/drivers/gpu/drm/mediatek/mtk_drm_ddp_comp.h
> @@ -99,6 +99,8 @@ struct mtk_ddp_comp {
> int irq;
> enum mtk_ddp_comp_id id;
> const struct mtk_ddp_comp_funcs *funcs;
> + resource_size_t regs_pa;
> + u8 subsys;
> };
>
> static inline void mtk_ddp_comp_config(struct mtk_ddp_comp *comp,
> diff --git a/drivers/gpu/drm/mediatek/mtk_drm_plane.c b/drivers/gpu/drm/mediatek/mtk_drm_plane.c
> index 59dbdaf07425..ef46c0f43039 100644
> --- a/drivers/gpu/drm/mediatek/mtk_drm_plane.c
> +++ b/drivers/gpu/drm/mediatek/mtk_drm_plane.c
> @@ -202,6 +202,8 @@ static void mtk_plane_atomic_update(struct drm_plane *plane,
> state->pending.height = drm_rect_height(&plane->state->dst);
> wmb(); /* Make sure the above parameters are set before update */
> state->pending.dirty = true;
> +
> + mtk_drm_crtc_plane_update(crtc, plane, state);
> }
>
> static void mtk_plane_atomic_disable(struct drm_plane *plane,
> @@ -212,6 +214,8 @@ static void mtk_plane_atomic_disable(struct drm_plane *plane,
> state->pending.enable = false;
> wmb(); /* Make sure the above parameter is set before update */
> state->pending.dirty = true;
> + /* Fetch CRTC from old plane state when disabling. */
> + mtk_drm_crtc_plane_update(old_state->crtc, plane, state);
> }
>
> static const struct drm_plane_helper_funcs mtk_plane_helper_funcs = {