[PATCH v17 4/4] media: platform: mtk-mdp3: add Mediatek MDP3 driver

moudy.ho moudy.ho at mediatek.com
Tue Jun 7 03:01:41 PDT 2022


On Wed, 2022-06-01 at 10:25 +0800, CK Hu wrote:
> Hi, Moudy:
> 
> On Tue, 2022-05-31 at 14:13 +0800, Moudy Ho wrote:
> > This patch adds driver for Mediatek's Media Data Path ver.3 (MDP3).
> > It provides the following functions:
> >   color transform, format conversion, resize, crop, rotate, flip
> >   and additional image quality enhancement.
> > 
> > The MDP3 driver is mainly used for Google Chromebook products to
> > import the new architecture to set the HW settings as shown below:
> >   User -> V4L2 framework
> >     -> MDP3 driver -> SCP (setting calculations)
> >       -> MDP3 driver -> CMDQ (GCE driver) -> HW
> > 
> > Each modules' related operation control is sited in mtk-mdp3-comp.c
> > Each modules' register table is defined in file with "mdp_reg_"
> > prefix
> > GCE related API, operation control  sited in mtk-mdp3-cmdq.c
> > V4L2 m2m device functions are implemented in mtk-mdp3-m2m.c
> > Probe, power, suspend/resume, system level functions are defined in
> > mtk-mdp3-core.c
> > 
> 
> [snip]
> 
> > 
> > +int mdp_cmdq_send(struct mdp_dev *mdp, struct mdp_cmdq_param
> > *param)
> > +{
> > +	struct mdp_cmdq_cmd cmd;
> > +	struct mdp_path *path = NULL;
> > +	struct mdp_cmdq_cb_param *cb_param = NULL;
> > +	struct mdp_comp *comps = NULL;
> > +	struct device *dev = &mdp->pdev->dev;
> > +	int i, ret;
> > +
> > +	atomic_inc(&mdp->job_count);
> > +	if (atomic_read(&mdp->suspended)) {
> > +		atomic_dec(&mdp->job_count);
> > +		return -ECANCELED;
> > +	}
> > +
> > +	cb_param = kzalloc(sizeof(*cb_param), GFP_KERNEL);
> > +	if (!cb_param) {
> > +		ret = -ENOMEM;
> > +		goto err_cmdq_data;
> > +	}
> > +
> > +	cmd.pkt = cmdq_pkt_create(mdp->cmdq_clt, SZ_16K);
> > +	if (IS_ERR(cmd.pkt)) {
> > +		ret = PTR_ERR(cmd.pkt);
> > +		goto err_cmdq_data;
> > +	}
> > +
> > +	comps = kcalloc(param->config->num_components, sizeof(*comps),
> > +			GFP_KERNEL);
> > +	if (!comps) {
> > +		ret = -ENOMEM;
> > +		goto err_cmdq_data;
> > +	}
> > +
> > +	path = kzalloc(sizeof(*path), GFP_KERNEL);
> > +	if (!path) {
> > +		ret = -ENOMEM;
> > +		goto err_cmdq_data;
> > +	}
> > +
> > +	path->mdp_dev = mdp;
> > +	path->config = param->config;
> > +	path->param = param->param;
> > +	for (i = 0; i < param->param->num_outputs; i++) {
> > +		path->bounds[i].left = 0;
> > +		path->bounds[i].top = 0;
> > +		path->bounds[i].width =
> > +			param->param->outputs[i].buffer.format.width;
> > +		path->bounds[i].height =
> > +			param->param->outputs[i].buffer.format.height;
> > +		path->composes[i] = param->composes[i] ?
> > +			param->composes[i] : &path->bounds[i];
> > +	}
> > +
> > +	ret = mdp_path_ctx_init(mdp, path);
> > +	if (ret) {
> > +		dev_err(dev, "mdp_path_ctx_init error\n");
> > +		goto err_cmdq_data;
> > +	}
> > +
> > +	cmd.event = &mdp->event[0];
> > +	ret = mdp_path_config(mdp, &cmd, path);
> > +	if (ret) {
> > +		dev_err(dev, "mdp_path_config error\n");
> > +		goto err_cmdq_data;
> > +	}
> > +	cmdq_pkt_finalize(cmd.pkt);
> > +
> > +	for (i = 0; i < param->config->num_components; i++)
> > +		memcpy(&comps[i], path->comps[i].comp,
> > +		       sizeof(struct mdp_comp));
> > +
> > +	cb_param->mdp = mdp;
> > +	cb_param->user_cmdq_cb = param->cmdq_cb;
> > +	cb_param->user_cb_data = param->cb_data;
> > +	cb_param->pkt = cmd.pkt;
> > +	cb_param->comps = comps;
> > +	cb_param->num_comps = param->config->num_components;
> > +	cb_param->mdp_ctx = param->mdp_ctx;
> > +
> > +	cmd.pkt->async_cb.data = (void *)cb_param;
> 
> async_cb is part of proprietary callback mechanism and I'm replacing
> the proprietary one with standard one. My final patch is [1] (not
> upstreamed yet) and 'data' would be removed. The way to get data
> related to pkt:
> 
> strutc mdp_data {
>     struct cmdq_pkt pkt;
>     /* Other data such as mdp, user_cmdq_cb, ... */
> }
> 
> The callback function would callback with pkt and you could use pkt
> to
> find struct mdp_data{}.
> 
> [1] 
> 
https://git.kernel.org/pub/scm/linux/kernel/git/chunkuang.hu/linux.git/commit/?h=mediatek-cmdq6&id=a35f642aa89f02b7307725cdaa3bfb348b26d093
> 
> Regards,
> CK
> 
> 
Hi CK,

Thanks for the review.
Based on the patch you mentioned, I will make corresponding 
adjustments and publish in the next version.

Regards,
Moudy
> > +	mdp->cmdq_clt->client.rx_callback = mdp_handle_cmdq_callback;
> > +
> > +	mtk_mutex_prepare(mdp->mdp_mutex[MDP_PIPE_RDMA0]);
> > +	for (i = 0; i < param->config->num_components; i++)
> > +		mdp_comp_clock_on(&mdp->pdev->dev, path-
> > > comps[i].comp);
> > 
> > +
> > +	dma_sync_single_for_device(mdp->cmdq_clt->chan->mbox->dev,
> > +				   cmd.pkt->pa_base, cmd.pkt-
> > > cmd_buf_size,
> > 
> > +				   DMA_TO_DEVICE);
> > +	ret = mbox_send_message(mdp->cmdq_clt->chan, cmd.pkt);
> > +	if (ret < 0) {
> > +		dev_err(dev, "mbox send message fail %d!\n", ret);
> > +		goto err_clock_off;
> > +	}
> > +	mbox_client_txdone(mdp->cmdq_clt->chan, 0);
> > +
> > +	kfree(path);
> > +	return 0;
> > +
> > +err_clock_off:
> > +	mtk_mutex_unprepare(mdp->mdp_mutex[MDP_PIPE_RDMA0]);
> > +	mdp_comp_clocks_off(&mdp->pdev->dev, cb_param->comps,
> > +			    cb_param->num_comps);
> > +err_cmdq_data:
> > +	kfree(path);
> > +	atomic_dec(&mdp->job_count);
> > +	wake_up(&mdp->callback_wq);
> > +	if (cmd.pkt)
> > +		cmdq_pkt_destroy(cmd.pkt);
> > +	kfree(comps);
> > +	kfree(cb_param);
> > +	return ret;
> > +}
> 
> 




More information about the Linux-mediatek mailing list