From: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
To: Rishikesh Donadkar <r-donadkar@ti.com>
Cc: y-abhilashchandra@ti.com, devarsht@ti.com, s-jain1@ti.com,
vigneshr@ti.com, mchehab@kernel.org, robh@kernel.org,
krzk+dt@kernel.org, p.zabel@pengutronix.de, conor+dt@kernel.org,
sakari.ailus@linux.intel.com, hverkuil-cisco@xs4all.nl,
jai.luthra@ideasonboard.com, changhuang.liang@starfivetech.com,
jack.zhu@starfivetech.com, sjoerd@collabora.com,
dan.carpenter@linaro.org, hverkuil+cisco@kernel.org,
linux-kernel@vger.kernel.org, linux-media@vger.kernel.org,
devicetree@vger.kernel.org, jai.luthra@linux.dev,
laurent.pinchart@ideasonboard.com, mripard@kernel.org
Subject: Re: [PATCH v9 15/19] media: ti: j721e-csi2rx: Change the drain architecture for multistream
Date: Thu, 15 Jan 2026 14:37:07 +0200 [thread overview]
Message-ID: <fbae9c01-6a2c-46ee-963a-8e5230b860b4@ideasonboard.com> (raw)
In-Reply-To: <20251230083220.2405247-16-r-donadkar@ti.com>
Hi,
On 30/12/2025 10:32, Rishikesh Donadkar wrote:
> On buffer starvation the DMA is marked IDLE, and the stale data in the
> internal FIFOs gets drained only on the next VIDIOC_QBUF call from the
> userspace. This approach works fine for a single stream case.
>
> But in multistream scenarios, buffer starvation for one stream i.e. one
> virtual channel, can block the shared HW FIFO of the CSI2RX IP. This can
> stall the pipeline for all other virtual channels, even if buffers are
> available for them.
One stream is filtered based on VC & DT, but the above only mentions VC.
And then later uses VC when referring to the stream. I think you can
drop the VC parts, and just talk about streams.
> This patch introduces a new architecture, that continuously drains data
> from the shared HW FIFO into a small (32KiB) buffer if no buffers are made
> available to the driver from the userspace. This ensures independence
> between different streams, where a slower downstream element for one
> camera does not block streaming for other cameras.
>
> Additionally, after a drain is done for a VC, the next frame will be a
> partial frame, as a portion of its data will have already been drained
> before a valid buffer is queued by user space to the driver.
This makes it sounds we drain a single 32KB piece. But won't we continue
draining that stream until the stream is stopped or the user provides a
buffer?
Also, does the DMA not offer us ways to drain a full frame? There's no
way to e.g. set the DMA TX increment to 0, so that it would just write
to a single location in memory? Or just set the target to void.
Tomi
> Use wait for completion barrier to make sure the shared hardware FIFO
> is cleared of the data at the end of stream after the source has stopped
> sending data.
>
> Reviewed-by: Jai Luthra <jai.luthra@ideasonboard.com>
> Reviewed-by: Yemike Abhilash Chandra <y-abhilashchandra@ti.com>
> Signed-off-by: Rishikesh Donadkar <r-donadkar@ti.com>
> ---
> .../platform/ti/j721e-csi2rx/j721e-csi2rx.c | 112 ++++++++----------
> 1 file changed, 50 insertions(+), 62 deletions(-)
>
> diff --git a/drivers/media/platform/ti/j721e-csi2rx/j721e-csi2rx.c b/drivers/media/platform/ti/j721e-csi2rx/j721e-csi2rx.c
> index fa6152464d4b6..e713293696eb1 100644
> --- a/drivers/media/platform/ti/j721e-csi2rx/j721e-csi2rx.c
> +++ b/drivers/media/platform/ti/j721e-csi2rx/j721e-csi2rx.c
> @@ -82,7 +82,6 @@ struct ti_csi2rx_buffer {
>
> enum ti_csi2rx_dma_state {
> TI_CSI2RX_DMA_STOPPED, /* Streaming not started yet. */
> - TI_CSI2RX_DMA_IDLE, /* Streaming but no pending DMA operation. */
> TI_CSI2RX_DMA_ACTIVE, /* Streaming and pending DMA operation. */
> };
>
> @@ -109,6 +108,7 @@ struct ti_csi2rx_ctx {
> struct v4l2_format v_fmt;
> struct ti_csi2rx_dma dma;
> struct media_pad pad;
> + struct completion drain_complete;
> u32 sequence;
> u32 idx;
> u32 vc;
> @@ -251,6 +251,10 @@ static const struct ti_csi2rx_fmt ti_csi2rx_formats[] = {
> static int ti_csi2rx_start_dma(struct ti_csi2rx_ctx *ctx,
> struct ti_csi2rx_buffer *buf);
>
> +/* Forward declarations needed by ti_csi2rx_drain_callback. */
> +static int ti_csi2rx_drain_dma(struct ti_csi2rx_ctx *ctx);
> +static int ti_csi2rx_dma_submit_pending(struct ti_csi2rx_ctx *ctx);
> +
> static const struct ti_csi2rx_fmt *find_format_by_fourcc(u32 pixelformat)
> {
> unsigned int i;
> @@ -617,9 +621,32 @@ static void ti_csi2rx_setup_shim(struct ti_csi2rx_ctx *ctx)
>
> static void ti_csi2rx_drain_callback(void *param)
> {
> - struct completion *drain_complete = param;
> + struct ti_csi2rx_ctx *ctx = param;
> + struct ti_csi2rx_dma *dma = &ctx->dma;
> + unsigned long flags;
> +
> + spin_lock_irqsave(&dma->lock, flags);
> +
> + if (dma->state == TI_CSI2RX_DMA_STOPPED) {
> + complete(&ctx->drain_complete);
> + spin_unlock_irqrestore(&dma->lock, flags);
> + return;
> + }
>
> - complete(drain_complete);
> + /*
> + * If dma->queue is empty, it indicates that no buffer has been
> + * provided by user space. In this case, initiate a transactions
> + * to drain the DMA. Since one drain of size DRAIN_BUFFER_SIZE
> + * will be done here, the subsequent frame will be a
> + * partial frame, with a size of frame_size - DRAIN_BUFFER_SIZE
> + */
> + if (list_empty(&dma->queue)) {
> + if (ti_csi2rx_drain_dma(ctx))
> + dev_warn(ctx->csi->dev, "DMA drain failed\n");
> + } else {
> + ti_csi2rx_dma_submit_pending(ctx);
> + }
> + spin_unlock_irqrestore(&dma->lock, flags);
> }
>
> /*
> @@ -637,12 +664,9 @@ static int ti_csi2rx_drain_dma(struct ti_csi2rx_ctx *ctx)
> {
> struct ti_csi2rx_dev *csi = ctx->csi;
> struct dma_async_tx_descriptor *desc;
> - struct completion drain_complete;
> dma_cookie_t cookie;
> int ret;
>
> - init_completion(&drain_complete);
> -
> desc = dmaengine_prep_slave_single(ctx->dma.chan, csi->drain.paddr,
> csi->drain.len, DMA_DEV_TO_MEM,
> DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
> @@ -652,7 +676,7 @@ static int ti_csi2rx_drain_dma(struct ti_csi2rx_ctx *ctx)
> }
>
> desc->callback = ti_csi2rx_drain_callback;
> - desc->callback_param = &drain_complete;
> + desc->callback_param = ctx;
>
> cookie = dmaengine_submit(desc);
> ret = dma_submit_error(cookie);
> @@ -661,13 +685,6 @@ static int ti_csi2rx_drain_dma(struct ti_csi2rx_ctx *ctx)
>
> dma_async_issue_pending(ctx->dma.chan);
>
> - if (!wait_for_completion_timeout(&drain_complete,
> - msecs_to_jiffies(DRAIN_TIMEOUT_MS))) {
> - dmaengine_terminate_sync(ctx->dma.chan);
> - dev_dbg(csi->dev, "DMA transfer timed out for drain buffer\n");
> - ret = -ETIMEDOUT;
> - goto out;
> - }
> out:
> return ret;
> }
> @@ -716,9 +733,11 @@ static void ti_csi2rx_dma_callback(void *param)
>
> ti_csi2rx_dma_submit_pending(ctx);
>
> - if (list_empty(&dma->submitted))
> - dma->state = TI_CSI2RX_DMA_IDLE;
> -
> + if (list_empty(&dma->submitted)) {
> + if (ti_csi2rx_drain_dma(ctx))
> + dev_warn(ctx->csi->dev,
> + "DMA drain failed on one of the transactions\n");
> + }
> spin_unlock_irqrestore(&dma->lock, flags);
> }
>
> @@ -754,6 +773,7 @@ static int ti_csi2rx_start_dma(struct ti_csi2rx_ctx *ctx,
> static void ti_csi2rx_stop_dma(struct ti_csi2rx_ctx *ctx)
> {
> struct ti_csi2rx_dma *dma = &ctx->dma;
> + struct ti_csi2rx_dev *csi = ctx->csi;
> enum ti_csi2rx_dma_state state;
> unsigned long flags;
> int ret;
> @@ -763,6 +783,8 @@ static void ti_csi2rx_stop_dma(struct ti_csi2rx_ctx *ctx)
> dma->state = TI_CSI2RX_DMA_STOPPED;
> spin_unlock_irqrestore(&dma->lock, flags);
>
> + init_completion(&ctx->drain_complete);
> +
> if (state != TI_CSI2RX_DMA_STOPPED) {
> /*
> * Normal DMA termination does not clean up pending data on
> @@ -771,11 +793,20 @@ static void ti_csi2rx_stop_dma(struct ti_csi2rx_ctx *ctx)
> * enforced before terminating DMA.
> */
> ret = ti_csi2rx_drain_dma(ctx);
> - if (ret && ret != -ETIMEDOUT)
> + if (ret)
> dev_warn(ctx->csi->dev,
> "Failed to drain DMA. Next frame might be bogus\n");
> }
>
> + /* We wait for the drain to complete so that the stream stops
> + * cleanly, making sure the shared hardware FIFO is cleared of
> + * data from the current stream. No more data will be coming from
> + * the source after this.
> + */
> + if (!wait_for_completion_timeout(&ctx->drain_complete,
> + msecs_to_jiffies(DRAIN_TIMEOUT_MS)))
> + dev_dbg(csi->dev, "DMA transfer timed out for drain buffer\n");
> +
> ret = dmaengine_terminate_sync(ctx->dma.chan);
> if (ret)
> dev_err(ctx->csi->dev, "Failed to stop DMA: %d\n", ret);
> @@ -838,57 +869,14 @@ static void ti_csi2rx_buffer_queue(struct vb2_buffer *vb)
> struct ti_csi2rx_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
> struct ti_csi2rx_buffer *buf;
> struct ti_csi2rx_dma *dma = &ctx->dma;
> - bool restart_dma = false;
> unsigned long flags = 0;
> - int ret;
>
> buf = container_of(vb, struct ti_csi2rx_buffer, vb.vb2_buf);
> buf->ctx = ctx;
>
> spin_lock_irqsave(&dma->lock, flags);
> - /*
> - * Usually the DMA callback takes care of queueing the pending buffers.
> - * But if DMA has stalled due to lack of buffers, restart it now.
> - */
> - if (dma->state == TI_CSI2RX_DMA_IDLE) {
> - /*
> - * Do not restart DMA with the lock held because
> - * ti_csi2rx_drain_dma() might block for completion.
> - * There won't be a race on queueing DMA anyway since the
> - * callback is not being fired.
> - */
> - restart_dma = true;
> - dma->state = TI_CSI2RX_DMA_ACTIVE;
> - } else {
> - list_add_tail(&buf->list, &dma->queue);
> - }
> + list_add_tail(&buf->list, &dma->queue);
> spin_unlock_irqrestore(&dma->lock, flags);
> -
> - if (restart_dma) {
> - /*
> - * Once frames start dropping, some data gets stuck in the DMA
> - * pipeline somewhere. So the first DMA transfer after frame
> - * drops gives a partial frame. This is obviously not useful to
> - * the application and will only confuse it. Issue a DMA
> - * transaction to drain that up.
> - */
> - ret = ti_csi2rx_drain_dma(ctx);
> - if (ret && ret != -ETIMEDOUT)
> - dev_warn(ctx->csi->dev,
> - "Failed to drain DMA. Next frame might be bogus\n");
> -
> - spin_lock_irqsave(&dma->lock, flags);
> - ret = ti_csi2rx_start_dma(ctx, buf);
> - if (ret) {
> - vb2_buffer_done(&buf->vb.vb2_buf, VB2_BUF_STATE_ERROR);
> - dma->state = TI_CSI2RX_DMA_IDLE;
> - spin_unlock_irqrestore(&dma->lock, flags);
> - dev_err(ctx->csi->dev, "Failed to start DMA: %d\n", ret);
> - } else {
> - list_add_tail(&buf->list, &dma->submitted);
> - spin_unlock_irqrestore(&dma->lock, flags);
> - }
> - }
> }
>
> static int ti_csi2rx_get_stream(struct ti_csi2rx_ctx *ctx)
next prev parent reply other threads:[~2026-01-15 12:37 UTC|newest]
Thread overview: 54+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-12-30 8:32 [PATCH v9 00/19] media: cadence,ti: CSI2RX Multistream Support Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 01/19] media: ti: j721e-csi2rx: Remove word size alignment on frame width Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 02/19] dt-bindings: media: ti,j721e-csi2rx-shim: Support 32 dma chans Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 03/19] media: ti: j721e-csi2rx: separate out device and context Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 04/19] media: ti: j721e-csi2rx: prepare SHIM code for multiple contexts Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 05/19] media: ti: j721e-csi2rx: allocate DMA channel based on context index Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 06/19] media: ti: j721e-csi2rx: add a subdev for the core device Rishikesh Donadkar
2026-01-14 15:21 ` Tomi Valkeinen
2026-01-15 6:36 ` Jai Luthra
2026-01-15 12:56 ` Tomi Valkeinen
2026-01-20 23:25 ` Laurent Pinchart
2026-01-21 7:38 ` Tomi Valkeinen
2026-01-21 10:52 ` Laurent Pinchart
2026-01-22 6:53 ` Jai Luthra
2026-01-22 9:53 ` Laurent Pinchart
2026-02-28 17:49 ` Jai Luthra
2026-01-21 9:10 ` Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 07/19] media: cadence: csi2rx: Move to .enable/disable_streams API Rishikesh Donadkar
2026-01-14 15:25 ` Tomi Valkeinen
2025-12-30 8:32 ` [PATCH v9 08/19] media: staging: starfive: Move to enabel-disable streams in starfive drivers Rishikesh Donadkar
2026-01-14 12:51 ` Jai Luthra
2026-01-14 13:05 ` Laurent Pinchart
2025-12-30 8:32 ` [PATCH v9 09/19] media: ti: j721e-csi2rx: get number of contexts from device tree Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 10/19] media: cadence: csi2rx: add get_frame_desc wrapper Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 11/19] media: ti: j721e-csi2rx: add support for processing virtual channels Rishikesh Donadkar
2026-01-14 15:31 ` Tomi Valkeinen
2026-01-16 10:28 ` Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 12/19] media: cadence: csi2rx: add multistream support Rishikesh Donadkar
2026-01-15 12:01 ` Tomi Valkeinen
2026-01-16 11:04 ` Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 13/19] media: ti: j721e-csi2rx: " Rishikesh Donadkar
2026-01-15 12:27 ` Tomi Valkeinen
2026-01-20 8:48 ` Rishikesh Donadkar
2026-01-15 12:28 ` Tomi Valkeinen
2026-01-20 8:52 ` Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 14/19] media: ti: j721e-csi2rx: Submit all available buffers Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 15/19] media: ti: j721e-csi2rx: Change the drain architecture for multistream Rishikesh Donadkar
2026-01-15 12:37 ` Tomi Valkeinen [this message]
2026-01-15 16:02 ` Jai Luthra
2026-01-16 10:11 ` Vignesh Raghavendra
2026-01-19 5:08 ` Rishikesh Donadkar
2026-01-19 5:04 ` Rishikesh Donadkar
2025-12-30 8:32 ` [PATCH v9 16/19] media: ti: j721e-csi2rx: Return the partial frame as error Rishikesh Donadkar
2026-01-06 11:15 ` Jai Luthra
2026-01-08 5:37 ` Rishikesh Donadkar
2026-01-15 12:39 ` Tomi Valkeinen
2025-12-30 8:32 ` [PATCH v9 17/19] media: cadence: csi2rx: Support runtime PM Rishikesh Donadkar
2026-01-14 17:04 ` Tomi Valkeinen
2025-12-30 8:32 ` [PATCH v9 18/19] media: ti: j721e-csi2rx: Support runtime suspend Rishikesh Donadkar
2026-01-15 12:46 ` Tomi Valkeinen
2026-01-19 6:04 ` Jai Luthra
2025-12-30 8:32 ` [PATCH v9 19/19] media: ti: j721e-csi2rx: Support system suspend using pm_notifier Rishikesh Donadkar
2026-01-15 12:50 ` Tomi Valkeinen
2026-01-19 5:25 ` Rishikesh Donadkar
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=fbae9c01-6a2c-46ee-963a-8e5230b860b4@ideasonboard.com \
--to=tomi.valkeinen@ideasonboard.com \
--cc=changhuang.liang@starfivetech.com \
--cc=conor+dt@kernel.org \
--cc=dan.carpenter@linaro.org \
--cc=devarsht@ti.com \
--cc=devicetree@vger.kernel.org \
--cc=hverkuil+cisco@kernel.org \
--cc=hverkuil-cisco@xs4all.nl \
--cc=jack.zhu@starfivetech.com \
--cc=jai.luthra@ideasonboard.com \
--cc=jai.luthra@linux.dev \
--cc=krzk+dt@kernel.org \
--cc=laurent.pinchart@ideasonboard.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-media@vger.kernel.org \
--cc=mchehab@kernel.org \
--cc=mripard@kernel.org \
--cc=p.zabel@pengutronix.de \
--cc=r-donadkar@ti.com \
--cc=robh@kernel.org \
--cc=s-jain1@ti.com \
--cc=sakari.ailus@linux.intel.com \
--cc=sjoerd@collabora.com \
--cc=vigneshr@ti.com \
--cc=y-abhilashchandra@ti.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®