On Wed, 23 Aug 2017 21:24:13 +0200
Maxime Ripard [off-list ref] wrote:

On Wed, Aug 23, 2017 at 10:41:18AM +0800, Yong wrote:

quoted

quoted

quoted

quoted

quoted

+static irqreturn_t sun6i_csi_isr(int irq, void *dev_id)
+{
+	struct sun6i_csi_dev *sdev = (struct sun6i_csi_dev *)dev_id;
+	struct regmap *regmap = sdev->regmap;
+	u32 status;
+
+	regmap_read(regmap, CSI_CH_INT_STA_REG, &status);
+
+	if ((status & CSI_CH_INT_STA_FIFO0_OF_PD) ||
+	    (status & CSI_CH_INT_STA_FIFO1_OF_PD) ||
+	    (status & CSI_CH_INT_STA_FIFO2_OF_PD) ||
+	    (status & CSI_CH_INT_STA_HB_OF_PD)) {
+		regmap_write(regmap, CSI_CH_INT_STA_REG, status);
+		regmap_update_bits(regmap, CSI_EN_REG, CSI_EN_CSI_EN, 0);
+		regmap_update_bits(regmap, CSI_EN_REG, CSI_EN_CSI_EN,
+				   CSI_EN_CSI_EN);

You need to enable / disable it at every frame? How do you deal with
double buffering? (or did you choose to ignore it for now?)

These *_OF_PD status bits indicate an overflow error condition.

Shouldn't we return an error code then? The names of these flags could
be better too.

Then, where and how to deal with the error coce.

If you want to deal with FIFO overflow, I'm not sure you have anything
to do. It means, you've been to slow to queue buffers, so I guess
stopping the pipeline until more buffers are queued would make
sense. And we should probably increase the sequence number while doing
so to notify the userspace that some frames were lost.

If there is no queued buffers, the CSI must has been already stoped by
sun6i_video_frame_done. So, the FIFO overflow may only occur on some 
unpredictable conditions or something I don't know.

For sequence number, I can't actually get the number of the lost frames.

Maybe I misunderstood you. Did you mean use IRQ_RETVAL(error) instead
of IRQ_HANDLED?

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

Thanks,
Yong