Hi Jai
+ Laurent + Sakari for a question below
On Tue, Jul 16, 2024 at 03:02:48PM GMT, Jai Luthra wrote:
> On Jul 12, 2024 at 16:18:36 +0200, Jacopo Mondi wrote:
> > On Thu, Jun 27, 2024 at 06:40:01PM GMT, Jai Luthra wrote:
> > > With single stream capture, it was simpler to use the video device as
> > > the media entity representing the main TI CSI2RX device. Now with multi
> > > stream capture coming into the picture, the model has shifted to each
> > > video device having a link to the main device's subdev. The routing
> > > would then be set on this subdev.
> > >
> > > Add this subdev, link each context to this subdev's entity and link the
> > > subdev's entity to the source. Also add an array of media pads. It will
> > > have one sink pad and source pads equal to the number of contexts.
> > >
> > > Co-developed-by: Pratyush Yadav <p.yadav@xxxxxx>
> > > Signed-off-by: Pratyush Yadav <p.yadav@xxxxxx>
> > > Signed-off-by: Jai Luthra <j-luthra@xxxxxx>
> > > ---
> > > .../media/platform/ti/j721e-csi2rx/j721e-csi2rx.c | 237 ++++++++++++++++++---
> > > 1 file changed, 212 insertions(+), 25 deletions(-)
> > >
> > > diff --git a/drivers/media/platform/ti/j721e-csi2rx/j721e-csi2rx.c b/drivers/media/platform/ti/j721e-csi2rx/j721e-csi2rx.c
> > > index 361b0ea8e0d9..13d7426ab4ba 100644
> > > --- a/drivers/media/platform/ti/j721e-csi2rx/j721e-csi2rx.c
> > > +++ b/drivers/media/platform/ti/j721e-csi2rx/j721e-csi2rx.c
> > > @@ -51,6 +51,11 @@
> > > #define MAX_WIDTH_BYTES SZ_16K
> > > #define MAX_HEIGHT_LINES SZ_16K
> > >
> > > +#define TI_CSI2RX_PAD_SINK 0
> > > +#define TI_CSI2RX_PAD_FIRST_SOURCE 1
> > > +#define TI_CSI2RX_NUM_SOURCE_PADS 1
> > > +#define TI_CSI2RX_NUM_PADS (1 + TI_CSI2RX_NUM_SOURCE_PADS)
> > > +
> > > #define DRAIN_TIMEOUT_MS 50
> > > #define DRAIN_BUFFER_SIZE SZ_32K
> > >
> > > @@ -97,6 +102,7 @@ struct ti_csi2rx_ctx {
> > > struct mutex mutex; /* To serialize ioctls. */
> > > struct v4l2_format v_fmt;
> > > struct ti_csi2rx_dma dma;
> > > + struct media_pad pad;
> > > u32 sequence;
> > > u32 idx;
> > > };
> > > @@ -104,12 +110,15 @@ struct ti_csi2rx_ctx {
> > > struct ti_csi2rx_dev {
> > > struct device *dev;
> > > void __iomem *shim;
> > > + struct mutex mutex; /* To serialize ioctls. */
> > > + unsigned int enable_count;
> > > struct v4l2_device v4l2_dev;
> > > struct media_device mdev;
> > > struct media_pipeline pipe;
> > > - struct media_pad pad;
> > > + struct media_pad pads[TI_CSI2RX_NUM_PADS];
> > > struct v4l2_async_notifier notifier;
> > > struct v4l2_subdev *source;
> > > + struct v4l2_subdev subdev;
> > > struct ti_csi2rx_ctx ctx[TI_CSI2RX_NUM_CTX];
> > > /* Buffer to drain stale data from PSI-L endpoint */
> > > struct {
> > > @@ -431,6 +440,15 @@ static int csi_async_notifier_complete(struct v4l2_async_notifier *notifier)
> > > struct ti_csi2rx_dev *csi = dev_get_drvdata(notifier->v4l2_dev->dev);
> > > int ret, i;
> > >
> > > + /* Create link from source to subdev */
> > > + ret = v4l2_create_fwnode_links_to_pad(csi->source,
> >
> > Isn't this a single link from the image source to the RX's sink pad ?
> > Wouldn't media_create_pad_link() do here ?
>
> The source here is always cdns-csi2rx bridge (and not the sensor) which
> has multiple pads.
>
> If we use media_create_pad_link() here, we would need to figure out the
> correct source pad using fwnode APIs. v4l2_create_fwnode_links_to_pad()
> helper already does that for us.
>
> But because of your comment I realized that we are missing
> .get_fwnode_pad() callbacks in both cdns-csi2rx and this driver.
>
> It is working by sheer luck as media_entity_get_fwnode_pad() returns the
> first pad as fallback. I will fix this in v3.
>
Ack
> >
> >
> > > + &csi->pads[TI_CSI2RX_PAD_SINK],
> > > + MEDIA_LNK_FL_IMMUTABLE |
> > > + MEDIA_LNK_FL_ENABLED);
> > > + if (ret)
> > > + return ret;
> > > +
> > > + /* Create and link video nodes for all DMA contexts */
> > > for (i = 0; i < TI_CSI2RX_NUM_CTX; i++) {
> > > struct ti_csi2rx_ctx *ctx = &csi->ctx[i];
> > > struct video_device *vdev = &ctx->vdev;
> > > @@ -438,13 +456,17 @@ static int csi_async_notifier_complete(struct v4l2_async_notifier *notifier)
> > > ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
> > > if (ret)
> > > goto unregister_dev;
> > > - }
> > >
> > > - ret = v4l2_create_fwnode_links_to_pad(csi->source, &csi->pad,
> > > - MEDIA_LNK_FL_IMMUTABLE |
> > > - MEDIA_LNK_FL_ENABLED);
> > > - if (ret)
> > > - goto unregister_dev;
> > > + ret = media_create_pad_link(&csi->subdev.entity,
> > > + TI_CSI2RX_PAD_FIRST_SOURCE + ctx->idx,
> > > + &vdev->entity, 0,
> > > + MEDIA_LNK_FL_IMMUTABLE |
> > > + MEDIA_LNK_FL_ENABLED);
> > > + if (ret) {
> > > + video_unregister_device(vdev);
> > > + goto unregister_dev;
> >
> > Should you call media_entity_remove_links() on the csi2 entity on the
> > error path ?
> >
>
> Will fix.
>
> > > + }
> > > + }
> > >
> > > ret = v4l2_device_register_subdev_nodes(&csi->v4l2_dev);
> > > if (ret)
> > > @@ -859,7 +881,7 @@ static int ti_csi2rx_start_streaming(struct vb2_queue *vq, unsigned int count)
> > > dma->state = TI_CSI2RX_DMA_ACTIVE;
> > > spin_unlock_irqrestore(&dma->lock, flags);
> > >
> > > - ret = v4l2_subdev_call(csi->source, video, s_stream, 1);
> > > + ret = v4l2_subdev_call(&csi->subdev, video, s_stream, 1);
> > > if (ret)
> > > goto err_dma;
> > >
> > > @@ -887,7 +909,7 @@ static void ti_csi2rx_stop_streaming(struct vb2_queue *vq)
> > > writel(0, csi->shim + SHIM_CNTL);
> > > writel(0, csi->shim + SHIM_DMACNTX(ctx->idx));
> > >
> > > - ret = v4l2_subdev_call(csi->source, video, s_stream, 0);
> > > + ret = v4l2_subdev_call(&csi->subdev, video, s_stream, 0);
> > > if (ret)
> > > dev_err(csi->dev, "Failed to stop subdev stream\n");
> > >
> > > @@ -905,6 +927,119 @@ static const struct vb2_ops csi_vb2_qops = {
> > > .wait_finish = vb2_ops_wait_finish,
> > > };
> > >
> > > +static inline struct ti_csi2rx_dev *to_csi2rx_dev(struct v4l2_subdev *sd)
> > > +{
> > > + return container_of(sd, struct ti_csi2rx_dev, subdev);
> > > +}
> > > +
> > > +static int ti_csi2rx_sd_set_fmt(struct v4l2_subdev *sd,
> > > + struct v4l2_subdev_state *state,
> > > + struct v4l2_subdev_format *format)
> > > +{
> > > + struct v4l2_mbus_framefmt *fmt;
> > > + int ret = 0;
> > > +
> > > + /* No transcoding, don't allow setting source fmt */
> > > + if (format->pad >= TI_CSI2RX_PAD_FIRST_SOURCE)
> >
> >
> > > TI_CSI2RX_PAD_SINK
> >
> > might be clearer
>
> Will fix.
>
> >
> > > + return v4l2_subdev_get_fmt(sd, state, format);
> > > +
> > > + if (!find_format_by_code(format->format.code))
> > > + format->format.code = ti_csi2rx_formats[0].code;
> > > +
> > > + format->format.field = V4L2_FIELD_NONE;
> >
> > What about other colorspace related fields ?
>
> The HW does not care about colorspace/encoding etc. and acts as
> passthrough, so we accept whatever userspace sets for those.
>
Or should it rather propagate to userspace what is set by the
downstream entity ?
> >
> > > +
> > > + fmt = v4l2_subdev_state_get_format(state, format->pad, format->stream);
> > > + if (!fmt) {
> >
> > Can this happen ?
>
> Yes __v4l2_subdev_state_get_format() may return NULL if userspace tried
> to set format with an invalid `stream` (i.e. no routes exist for that
> stream)
>
The core makes sure the stream is valid afaict
See check_state() in v4l2-subdev.c
> >
> > > + ret = -EINVAL;
> > > + goto out;
> > > + }
> > > + *fmt = format->format;
> > > +
> > > + fmt = v4l2_subdev_state_get_format(state, TI_CSI2RX_PAD_FIRST_SOURCE,
> > > + format->stream);
> > > + if (!fmt) {
> > > + ret = -EINVAL;
> >
> > ditto
> >
> > > + goto out;
> >
> > kind of missing the point of a jump here where you can simply return
> > -EINVAL
>
> Will fix.
>
> >
> > > + }
> > > + *fmt = format->format;
> > > +
> > > +out:
> > > + return ret;
> > > +}
> > > +
> > > +static int ti_csi2rx_sd_init_state(struct v4l2_subdev *sd,
> > > + struct v4l2_subdev_state *state)
> > > +{
> > > + struct v4l2_subdev_format format = {
> > > + .pad = TI_CSI2RX_PAD_SINK,
> > > + .format = {
> > > + .width = 640,
> > > + .height = 480,
> > > + .code = MEDIA_BUS_FMT_UYVY8_1X16,
> > > + .field = V4L2_FIELD_NONE,
> > > + .colorspace = V4L2_COLORSPACE_SRGB,
> > > + .ycbcr_enc = V4L2_YCBCR_ENC_601,
> > > + .quantization = V4L2_QUANTIZATION_LIM_RANGE,
> > > + .xfer_func = V4L2_XFER_FUNC_SRGB,
> > > + },
> > > + };
> > > +
> > > + return ti_csi2rx_sd_set_fmt(sd, state, &format);
> > > +}
> > > +
> > > +static int ti_csi2rx_sd_s_stream(struct v4l2_subdev *sd, int enable)
> >
> > Even if I never used that API myself, but I have a feeling
> > enable_streams() (which is meant to replace s_stream for multiplexed
> > subdev) would help you avoiding the manual enable_count booking here
> > below ?
>
> In PATCH 12/13 we switch to using enable_streams() APIs here.
>
> We still keep this enable_count bookkeeping, but I agree it can be
> dropped with minor changes. Will fix in v3.
>
Thanks, if it gets troublesome to remove it and you later switch to
the new API then don't bother
> >
> > > +{
> > > + struct ti_csi2rx_dev *csi = to_csi2rx_dev(sd);
> > > + int ret = 0;
> > > +
> > > + mutex_lock(&csi->mutex);
> > > +
> > > + if (enable) {
> > > + if (csi->enable_count > 0) {
> > > + csi->enable_count++;
> > > + goto out;
> > > + }
> > > +
> > > + ret = v4l2_subdev_call(csi->source, video, s_stream, 1);
> > > + if (ret)
> > > + goto out;
> > > +
> > > + csi->enable_count++;
> > > + } else {
> > > + if (csi->enable_count == 0) {
> > > + ret = -EINVAL;
> > > + goto out;
> > > + }
> > > +
> > > + if (--csi->enable_count > 0)
> > > + goto out;
> > > +
> > > + ret = v4l2_subdev_call(csi->source, video, s_stream, 0);
> > > + }
> > > +
> > > +out:
> > > + mutex_unlock(&csi->mutex);
> > > + return ret;
> > > +}
> > > +
> > > +static const struct v4l2_subdev_pad_ops ti_csi2rx_subdev_pad_ops = {
> > > + .get_fmt = v4l2_subdev_get_fmt,
> > > + .set_fmt = ti_csi2rx_sd_set_fmt,
> > > +};
> > > +
> > > +static const struct v4l2_subdev_video_ops ti_csi2rx_subdev_video_ops = {
> > > + .s_stream = ti_csi2rx_sd_s_stream,
> > > +};
> > > +
> > > +static const struct v4l2_subdev_ops ti_csi2rx_subdev_ops = {
> > > + .video = &ti_csi2rx_subdev_video_ops,
> > > + .pad = &ti_csi2rx_subdev_pad_ops,
> > > +};
> > > +
> > > +static const struct v4l2_subdev_internal_ops ti_csi2rx_internal_ops = {
> > > + .init_state = ti_csi2rx_sd_init_state,
> > > +};
> > > +
> > > static void ti_csi2rx_cleanup_dma(struct ti_csi2rx_ctx *ctx)
> > > {
> > > dma_release_channel(ctx->dma.chan);
> > > @@ -912,6 +1047,7 @@ static void ti_csi2rx_cleanup_dma(struct ti_csi2rx_ctx *ctx)
> > >
> > > static void ti_csi2rx_cleanup_v4l2(struct ti_csi2rx_dev *csi)
> > > {
> > > + v4l2_subdev_cleanup(&csi->subdev);
> > > media_device_unregister(&csi->mdev);
> > > v4l2_device_unregister(&csi->v4l2_dev);
> > > media_device_cleanup(&csi->mdev);
> > > @@ -973,14 +1109,22 @@ static int ti_csi2rx_link_validate(struct media_link *link)
> > > struct v4l2_subdev_format source_fmt = {
> > > .which = V4L2_SUBDEV_FORMAT_ACTIVE,
> > > .pad = link->source->index,
> > > + .stream = 0,
> >
> > I don't think it hurts, but this seems not to be strictly related to
> > this patch ?
> >
> > > };
> > > + struct v4l2_subdev_state *state;
> > > const struct ti_csi2rx_fmt *ti_fmt;
> > > int ret;
> > >
> > > - ret = v4l2_subdev_call_state_active(csi->source, pad,
> > > - get_fmt, &source_fmt);
> > > - if (ret)
> > > - return ret;
> > > + state = v4l2_subdev_lock_and_get_active_state(&csi->subdev);
> > > + ret = v4l2_subdev_call(&csi->subdev, pad, get_fmt, state, &source_fmt);
> > > + v4l2_subdev_unlock_state(state);
> > > +
> > > + if (ret) {
> > > + dev_dbg(csi->dev,
> > > + "Skipping validation as no format present on \"%s\":%u:0\n",
> > > + link->source->entity->name, link->source->index);
> >
> > If get_fmt returns an error, should you simply continue or fail ? In
> > which cases get_fmt on a subdev fails ? Only if the pad or the stream
> > are not valid right ?
>
> The subdev introduced in this patch has multiple source pads, all
> actively linked to separate video nodes, one for each DMA channel.
>
> We continue (return 0) here as we don't want __media_pipeline_start to
> fail because of pads that are "unused" i.e. have no routes and no
> formats set.
>
> If the user actually routes a stream to any pad the pipeline validation
> will check that a format is also set (and thus the link will get
> validated here).
>
> I wonder if we can keep the links mutable, and expect userspace to
> disable links to unused video nodes. Will have to think more about that
> approach.
>
We recently had a discussion about this (link enablement for 'active'
dma paths) with Sakari and Laurent on the Mali-C55 ISP. Let's see if
they have feedbacks.
> >
> > > + return 0;
> > > + }
> > >
> > > if (source_fmt.format.width != csi_fmt->width) {
> > > dev_dbg(csi->dev, "Width does not match (source %u, sink %u)\n",
> > > @@ -1010,8 +1154,9 @@ static int ti_csi2rx_link_validate(struct media_link *link)
> > >
> > > if (ti_fmt->fourcc != csi_fmt->pixelformat) {
> > > dev_dbg(csi->dev,
> > > - "Cannot transform source fmt 0x%x to sink fmt 0x%x\n",
> > > - ti_fmt->fourcc, csi_fmt->pixelformat);
> > > + "Cannot transform \"%s\":%u format %p4cc to %p4cc\n",
> > > + link->source->entity->name, link->source->index,
> > > + &ti_fmt->fourcc, &csi_fmt->pixelformat);
> > > return -EPIPE;
> > > }
> > >
> > > @@ -1022,6 +1167,10 @@ static const struct media_entity_operations ti_csi2rx_video_entity_ops = {
> > > .link_validate = ti_csi2rx_link_validate,
> > > };
> > >
> > > +static const struct media_entity_operations ti_csi2rx_subdev_entity_ops = {
> > > + .link_validate = v4l2_subdev_link_validate,
> > > +};
> > > +
> > > static int ti_csi2rx_init_dma(struct ti_csi2rx_ctx *ctx)
> > > {
> > > struct dma_slave_config cfg = {
> > > @@ -1053,7 +1202,8 @@ static int ti_csi2rx_init_dma(struct ti_csi2rx_ctx *ctx)
> > > static int ti_csi2rx_v4l2_init(struct ti_csi2rx_dev *csi)
> > > {
> > > struct media_device *mdev = &csi->mdev;
> > > - int ret;
> > > + struct v4l2_subdev *sd = &csi->subdev;
> > > + int ret, i;
> > >
> > > mdev->dev = csi->dev;
> > > mdev->hw_revision = 1;
> > > @@ -1065,16 +1215,50 @@ static int ti_csi2rx_v4l2_init(struct ti_csi2rx_dev *csi)
> > >
> > > ret = v4l2_device_register(csi->dev, &csi->v4l2_dev);
> > > if (ret)
> > > - return ret;
> > > + goto cleanup_media;
> > >
> > > ret = media_device_register(mdev);
> > > - if (ret) {
> > > - v4l2_device_unregister(&csi->v4l2_dev);
> > > - media_device_cleanup(mdev);
> > > - return ret;
> > > - }
> > > + if (ret)
> > > + goto unregister_v4l2;
> > > +
> > > + v4l2_subdev_init(sd, &ti_csi2rx_subdev_ops);
> > > + sd->internal_ops = &ti_csi2rx_internal_ops;
> > > + sd->entity.function = MEDIA_ENT_F_VID_IF_BRIDGE;
> > > + sd->flags = V4L2_SUBDEV_FL_HAS_DEVNODE;
> > > + strscpy(sd->name, dev_name(csi->dev), sizeof(sd->name));
> > > + sd->dev = csi->dev;
> > > + sd->entity.ops = &ti_csi2rx_subdev_entity_ops;
> > > +
> > > + csi->pads[TI_CSI2RX_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> > > +
> > > + for (i = TI_CSI2RX_PAD_FIRST_SOURCE; i < TI_CSI2RX_NUM_PADS; i++)
> >
> > for (unsigned int i = 0;
>
> Will fix.
>
> >
> > > + csi->pads[i].flags = MEDIA_PAD_FL_SOURCE;
> > > +
> > > + ret = media_entity_pads_init(&sd->entity, ARRAY_SIZE(csi->pads),
> > > + csi->pads);
> > > + if (ret)
> > > + goto unregister_media;
> > > +
> > > + ret = v4l2_subdev_init_finalize(sd);
> > > + if (ret)
> > > + goto unregister_media;
> > > +
> > > + ret = v4l2_device_register_subdev(&csi->v4l2_dev, sd);
> > > + if (ret)
> > > + goto cleanup_subdev;
> > >
> > > return 0;
> > > +
> > > +cleanup_subdev:
> > > + v4l2_subdev_cleanup(sd);
> > > +unregister_media:
> > > + media_device_unregister(mdev);
> > > +unregister_v4l2:
> > > + v4l2_device_unregister(&csi->v4l2_dev);
> > > +cleanup_media:
> > > + media_device_cleanup(mdev);
> > > +
> > > + return ret;
> > > }
> > >
> > > static int ti_csi2rx_init_ctx(struct ti_csi2rx_ctx *ctx)
> > > @@ -1101,9 +1285,9 @@ static int ti_csi2rx_init_ctx(struct ti_csi2rx_ctx *ctx)
> > >
> > > ti_csi2rx_fill_fmt(fmt, &ctx->v_fmt);
> > >
> > > - csi->pad.flags = MEDIA_PAD_FL_SINK;
> >
> > Ah was this re-initializing the same pad multiple times ?
>
> Not sure I understand the comment fully. We have only 1 video context
> here, more contexts will be introduced in subsequent patches.
Yeah, I meant that before thsi patch there was a single csi->pad which
I thought was re-initialized multiple times. But before this patch you
had no contexts, so no re-initialization
>
> >
> > > + ctx->pad.flags = MEDIA_PAD_FL_SINK;
> > > vdev->entity.ops = &ti_csi2rx_video_entity_ops;
> > > - ret = media_entity_pads_init(&ctx->vdev.entity, 1, &csi->pad);
> > > + ret = media_entity_pads_init(&ctx->vdev.entity, 1, &ctx->pad);
> > > if (ret)
> > > return ret;
> > >
> > > @@ -1159,6 +1343,8 @@ static int ti_csi2rx_probe(struct platform_device *pdev)
> > > if (!csi->drain.vaddr)
> > > return -ENOMEM;
> > >
> > > + mutex_init(&csi->mutex);
> > > +
> > > ret = ti_csi2rx_v4l2_init(csi);
> > > if (ret)
> > > goto err_v4l2;
> > > @@ -1191,6 +1377,7 @@ static int ti_csi2rx_probe(struct platform_device *pdev)
> > > ti_csi2rx_cleanup_ctx(&csi->ctx[i]);
> > > ti_csi2rx_cleanup_v4l2(csi);
> > > err_v4l2:
> > > + mutex_destroy(&csi->mutex);
> > > dma_free_coherent(csi->dev, csi->drain.len, csi->drain.vaddr,
> > > csi->drain.paddr);
> > > return ret;
> > > @@ -1213,7 +1400,7 @@ static void ti_csi2rx_remove(struct platform_device *pdev)
> > >
> > > ti_csi2rx_cleanup_notifier(csi);
> > > ti_csi2rx_cleanup_v4l2(csi);
> > > -
> > > + mutex_destroy(&csi->mutex);
> > > dma_free_coherent(csi->dev, csi->drain.len, csi->drain.vaddr,
> > > csi->drain.paddr);
> > > }
> > >
> > > --
> > > 2.43.0
> > >
> > >
>
> --
> Thanks,
> Jai
>
> GPG Fingerprint: 4DE0 D818 E5D5 75E8 D45A AFC5 43DE 91F9 249A 7145
Attachment:
signature.asc
Description: PGP signature
