Re: [PATCH v7 4/5] media: platform: mediatek: isp: add mediatek ISP3.0 camsv

["CK Hu (胡俊光)" <ck.hu@mediatek.com>]

Hi, Julien:

On Thu, 2024-11-21 at 09:53 +0100, Julien Stephan wrote:
> External email : Please do not click links or open attachments until you have verified the sender or the content.
> 
> 
> From: Phi-bang Nguyen <pnguyen@baylibre.com>
> 
> This driver provides a path to bypass the SoC ISP so that image data
> coming from the SENINF can go directly into memory without any image
> processing. This allows the use of an external ISP.
> 
> Signed-off-by: Phi-bang Nguyen <pnguyen@baylibre.com>
> Signed-off-by: Florian Sylvestre <fsylvestre@baylibre.com>
> [Paul Elder fix irq locking]
> Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
> Co-developed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
> Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
> Co-developed-by: Julien Stephan <jstephan@baylibre.com>
> Signed-off-by: Julien Stephan <jstephan@baylibre.com>
> ---

[snip]

> +static irqreturn_t isp_irq_camsv30(int irq, void *data)
> +{
> +       struct mtk_cam_dev *cam_dev = (struct mtk_cam_dev *)data;
> +       struct mtk_cam_dev_buffer *buf;
> +       unsigned int irq_status;
> +
> +       spin_lock(&cam_dev->buf_list_lock);

spin_lock region should be as small as possible.
Move to where you start to access buf_list.

> +
> +       irq_status = mtk_camsv30_read(cam_dev, CAMSV_INT_STATUS);
> +
> +       if (irq_status & INT_ST_MASK_CAMSV_ERR)
> +               dev_err(cam_dev->dev, "irq error 0x%lx\n",
> +                       irq_status & INT_ST_MASK_CAMSV_ERR);
> +
> +       /* De-queue frame */
> +       if (irq_status & CAMSV_IRQ_PASS1_DON) {
> +               cam_dev->sequence++;
> +
> +               buf = list_first_entry_or_null(&cam_dev->buf_list,
> +                                              struct mtk_cam_dev_buffer,
> +                                              list);
> +               if (buf) {
> +                       buf->v4l2_buf.sequence = cam_dev->sequence;
> +                       buf->v4l2_buf.vb2_buf.timestamp =
> +                               ktime_get_ns();
> +                       vb2_buffer_done(&buf->v4l2_buf.vb2_buf,
> +                                       VB2_BUF_STATE_DONE);

These jobs could be done after buffer is deleted from list, and move these jobs out of this spin_lock region.
vb2_buffer_done() looks like does many things. So it's worth to move vb2_buffer_done() out of this spin_lock region.

spin_lock();

processed_buf = list_first_entry_or_null();
if (processed_buf)
	list_del(&processed_buf->list);

spin_unlock();

if (processed_buf) {
	...
	vb2_buffer_done();
}

Regards,
CK

> +                       list_del(&buf->list);
> +               }
> +
> +               buf = list_first_entry_or_null(&cam_dev->buf_list,
> +                                              struct mtk_cam_dev_buffer,
> +                                              list);
> +               if (buf)
> +                       mtk_camsv30_update_buffers_add(cam_dev, buf);
> +       }
> +
> +       spin_unlock(&cam_dev->buf_list_lock);
> +
> +       return IRQ_HANDLED;
> +}
> +

>