Skip to contentSkip to conversation

OMAP3 Bridge Problems

DISCUSSION Lane Brooks 2 participants 13 messages Aug 3, 2010 – Sep 3, 2010 Media (V4L/DVB)

lore
Share & tools
lore
#0 lore raw

Laurent and team,

I am using the OMAP3 ISP code from the devel branch on gitorious that I back ported to a 2.6.31 kernel. Raw bayer streaming to the CCDC output works fine. I am using parallel input with the bridge disabled in that mode.

I am having a problem when I switch the sensor to output YUV422 data. The YUV422 stream is a 8x2. If I only switch the sensor to YUV422 mode, then I can get the YUV422 data at the CCDC output, but the CCDC pads an extra zero byte in there and I only get half the image. So that works as expected. I was then hoping all I would have to do is enable the bridge to get the YUV422_8x2 data packed into the YUV422_16x1 automatically, but instead I get select timeouts.

My question:

- Are there other things I need to when I enable the parallel bridge? For example, do I need to change a clock rate somewhere? From the TRM, it seems like it should just work without any changes, but maybe I am missing something.

Thanks,
Lane

#1 lore raw

Hi Lane,

Quoted 1 line · On Tuesday 03 August 2010 17:26:48 Lane Brooks wrote:

Laurent and team,

Could you please CC me when sending a mail that requires my attention ? I follow the linux-media mailing list, but sometimes mails can slip by.

Quoted 13 lines

I am using the OMAP3 ISP code from the devel branch on gitorious that I back ported to a 2.6.31 kernel. Raw bayer streaming to the CCDC output works fine. I am using parallel input with the bridge disabled in that mode.

I am having a problem when I switch the sensor to output YUV422 data. The YUV422 stream is a 8x2. If I only switch the sensor to YUV422 mode, then I can get the YUV422 data at the CCDC output, but the CCDC pads an extra zero byte in there and I only get half the image. So that works as expected. I was then hoping all I would have to do is enable the bridge to get the YUV422_8x2 data packed into the YUV422_16x1 automatically, but instead I get select timeouts.

Quoted 6 lines

My question:

- Are there other things I need to when I enable the parallel bridge? For example, do I need to change a clock rate somewhere? From the TRM, it seems like it should just work without any changes, but maybe I am missing something.

Good question. ISP bridge and YUV modes support are not implemented in the driver, but you're probably already aware of that.

I unfortunately have no straightforward answer. Try tracing the ISP interrupts and monitoring the CCDC SBL busy bits to see if the CCDC writes images to memory correctly.

Signature · 3 lines
Regards,

Laurent Pinchart
#2 lore raw
Quoted 3 lines · Laurent Pinchart

Hi Lane,

On Tuesday 03 August 2010 17:26:48 Lane Brooks wrote:

[snip]

Quoted 7 lines
Quoted 7 lines

My question:

- Are there other things I need to when I enable the parallel bridge? For example, do I need to change a clock rate somewhere? From the TRM, it seems like it should just work without any changes, but maybe I am missing something.

Quoted 6 lines

Good question. ISP bridge and YUV modes support are not implemented in the driver, but you're probably already aware of that.

I unfortunately have no straightforward answer. Try tracing the ISP interrupts and monitoring the CCDC SBL busy bits to see if the CCDC writes images to memory correctly.

I found at least some of the problem. In my platform data I was enabling the bridge using the #defines in ispreg.h as in

static struct isp_platform_data bmi_isp_platform_data = {
.parallel = {
.data_lane_shift = 3,
.clk_pol = 0,
.bridge = ISPCTRL_PAR_BRIDGE_LENDIAN,
},
.subdevs = bmi_camera_subdevs,
};

The bridge related #defines in ispreg.h, however, have a shift of 2 applied to them. The problem is that the shift is applied again in isp.c when the options are actually applied. In other words, the bridge parameters are being shifted up twice, which is causing corruption to the control register and causing my hanging problems when I try to enable the bridge. It seems there are several other such cases in the ispreg.h where double shifting might occur if the user tries to use them in the platform data.

My question:
Is this an oversight or is it this way on purpose? Am I not supposed to be using these defines in my platform definitions? It seems that *some* of the parameters in ispreg.h should not be shifted up (like the bridge options).

Thanks,
Lane

#3 lore raw
Quoted 12 lines · Lane Brooks
Quoted 3 lines · Laurent Pinchart

Hi Lane,

On Tuesday 03 August 2010 17:26:48 Lane Brooks wrote:

[snip]

Quoted 7 lines
Quoted 7 lines

My question:

- Are there other things I need to when I enable the parallel bridge? For example, do I need to change a clock rate somewhere? From the TRM, it seems like it should just work without any changes, but maybe I am missing something.

Quoted 8 lines
Quoted 8 lines

Good question. ISP bridge and YUV modes support are not implemented in the
driver, but you're probably already aware of that.

I unfortunately have no straightforward answer. Try tracing the ISP interrupts
and monitoring the CCDC SBL busy bits to see if the CCDC writes images to memory correctly.

[snip]

Laurent,

I was able to get YUV CCDC capture mode working with rather minimal effort. Attached is a patch with the initial effort. Can you comment?

Thanks,
Lane

drivers/media/video/isp/ispccdc.c +60 -4
index 90bcc6c..cc91fa1 100644
@@ -1563,6 +1563,15 @@ __ccdc_get_format(struct isp_ccdc_device *ccdc, struct v4l2_subdev_fh *fh,
  * @pad: Pad number
  * @fmt: Format
  */
+
+static enum v4l2_mbus_pixelcode sink_fmts[] = {
+	V4L2_MBUS_FMT_SGRBG10_1X10,
+	V4L2_MBUS_FMT_YUYV16_1X16,
+	V4L2_MBUS_FMT_UYVY16_1X16,
+	V4L2_MBUS_FMT_YVYU16_1X16,
+	V4L2_MBUS_FMT_VYUY16_1X16,
+};
+
 static void
 ccdc_try_format(struct isp_ccdc_device *ccdc, struct v4l2_subdev_fh *fh,
 		unsigned int pad, struct v4l2_mbus_framefmt *fmt,
@@ -1571,14 +1580,22 @@ ccdc_try_format(struct isp_ccdc_device *ccdc, struct v4l2_subdev_fh *fh,
 	struct v4l2_mbus_framefmt *format;
 	unsigned int width = fmt->width;
 	unsigned int height = fmt->height;
-
+	int i;
 	switch (pad) {
 	case CCDC_PAD_SINK:
 		/* Check if the requested pixel format is supported.
 		 * TODO: If the CCDC output formatter pad is connected directly
 		 * to the resizer, only YUV formats can be used.
 		 */
-		fmt->code = V4L2_MBUS_FMT_SGRBG10_1X10;
+		for(i=0; i<ARRAY_SIZE(sink_fmts); ++i) {
+			if(sink_fmts[i] == fmt->code) 
+				break;
+		}
+		/* if the requested type is not in the list of
+		 * supported types, then change it to the first format
+		 * code in the list of supported. */
+		if(i==ARRAY_SIZE(sink_fmts)) 
+			fmt->code = sink_fmts[0];
 
 		/* Clamp the input size. */
 		fmt->width = clamp_t(u32, width, 32, 4096);
@@ -1630,10 +1647,10 @@ static int ccdc_enum_mbus_code(struct v4l2_subdev *sd,
 			       struct v4l2_subdev_fh *fh,
 			       struct v4l2_subdev_pad_mbus_code_enum *code)
 {
-	if (code->index != 0)
+	if (code->index >= ARRAY_SIZE(sink_fmts))
 		return -EINVAL;
 
-	code->code = V4L2_MBUS_FMT_SGRBG10_1X10;
+	code->code = sink_fmts[code->index];
 
 	return 0;
 }
@@ -1719,6 +1736,45 @@ static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh,
 
 	/* Propagate the format from sink to source */
 	if (pad == CCDC_PAD_SINK) {
+		u32 syn_mode, ispctrl_val;
+		struct isp_device *isp = to_isp_device(ccdc);
+		if (!isp_get(isp))
+			return -EBUSY;
+
+		syn_mode    = isp_reg_readl(isp, OMAP3_ISP_IOMEM_CCDC, 
+					    ISPCCDC_SYN_MODE);
+		ispctrl_val = isp_reg_readl(isp, OMAP3_ISP_IOMEM_MAIN, 
+					    ISP_CTRL);
+		syn_mode    &= ISPCCDC_SYN_MODE_INPMOD_MASK;
+		ispctrl_val &= ~(ISPCTRL_PAR_BRIDGE_MASK 
+				 << ISPCTRL_PAR_BRIDGE_SHIFT);
+		switch(format->code) {
+		case V4L2_MBUS_FMT_YUYV16_1X16:
+		case V4L2_MBUS_FMT_UYVY16_1X16:
+		case V4L2_MBUS_FMT_YVYU16_1X16:
+		case V4L2_MBUS_FMT_VYUY16_1X16:
+			syn_mode |= ISPCCDC_SYN_MODE_INPMOD_YCBCR16;
+
+			/* TODO: In YCBCR16 mode, the bridge has to be
+			 * enabled, so we enable it here and force it
+			 * big endian. Whether to do big or little endian
+			 * should somehow come from the platform data.*/
+			ispctrl_val |= ISPCTRL_PAR_BRIDGE_BENDIAN 
+				<< ISPCTRL_PAR_BRIDGE_SHIFT;
+			break;
+		default:
+			syn_mode |= ISPCCDC_SYN_MODE_INPMOD_RAW;
+			ispctrl_val |= isp->pdata->parallel.bridge
+				<< ISPCTRL_PAR_BRIDGE_SHIFT;
+			break;
+		}
+		isp_reg_writel(isp, syn_mode, OMAP3_ISP_IOMEM_CCDC, 
+			       ISPCCDC_SYN_MODE);
+		isp_reg_writel(isp, ispctrl_val, OMAP3_ISP_IOMEM_MAIN, 
+			       ISP_CTRL);
+		isp_put(isp);
+
+
 		format = __ccdc_get_format(ccdc, fh, CCDC_PAD_SOURCE_OF, which);
 		memcpy(format, fmt, sizeof(*format));
 		ccdc_try_format(ccdc, fh, CCDC_PAD_SOURCE_OF, format, which);
drivers/media/video/isp/ispreg.h +4 -4
index 7efcfaa..4c191af 100644
@@ -732,10 +732,10 @@
 #define ISPCTRL_PAR_SER_CLK_SEL_MASK		0xFFFFFFFC
 
 #define ISPCTRL_PAR_BRIDGE_SHIFT		2
-#define ISPCTRL_PAR_BRIDGE_DISABLE		(0x0 << 2)
-#define ISPCTRL_PAR_BRIDGE_LENDIAN		(0x2 << 2)
-#define ISPCTRL_PAR_BRIDGE_BENDIAN		(0x3 << 2)
-#define ISPCTRL_PAR_BRIDGE_MASK			(0x3 << 2)
+#define ISPCTRL_PAR_BRIDGE_DISABLE		0x0
+#define ISPCTRL_PAR_BRIDGE_LENDIAN		0x2
+#define ISPCTRL_PAR_BRIDGE_BENDIAN		0x3
+#define ISPCTRL_PAR_BRIDGE_MASK			0x3
 
 #define ISPCTRL_PAR_CLK_POL_SHIFT		4
 #define ISPCTRL_PAR_CLK_POL_INV			(1 << 4)
#4 lore raw

Hi Lane,

Thanks for the patch.

On Thursday 05 August 2010 20:53:50 Lane Brooks wrote:

[snip]

Quoted 2 lines

I was able to get YUV CCDC capture mode working with rather minimal effort. Attached is a patch with the initial effort. Can you comment?

When sending patches for review, please send them inline instead of attaching them to the mail. It makes the review easier.

Review

Quoted 1 line
(quoted diff) +0 -0

b/drivers/media/video/isp/ispccdc.c

drivers/media/video/isp/ispccdc.c __ccdc_get_format(struct isp_ccdc_device *ccdc,

Quoted 4 lines

index 90bcc6c..cc91fa1 100644

drivers/media/video/isp/ispccdc.c +0 -0
@@ -1563,6 +1563,15 @@ __ccdc_get_format(struct isp_ccdc_device *ccdc, 

struct v4l2_subdev_fh *fh,

Quoted 5 lines
drivers/media/video/isp/ispccdc.c +2 -0
  * @pad: Pad number
  * @fmt: Format
  */
+
+static enum v4l2_mbus_pixelcode sink_fmts[] = {

drivers/media/video/isp/ispccdc.c __ccdc_get_format(struct isp_ccdc_device *ccdc, struct v4l2_subdev_fh *fh, · line 1577

Quoted 10 lines
drivers/media/video/isp/ispccdc.c +7 -0
+	V4L2_MBUS_FMT_SGRBG10_1X10,
+	V4L2_MBUS_FMT_YUYV16_1X16,
+	V4L2_MBUS_FMT_UYVY16_1X16,
+	V4L2_MBUS_FMT_YVYU16_1X16,
+	V4L2_MBUS_FMT_VYUY16_1X16,
+};
+
 static void
 ccdc_try_format(struct isp_ccdc_device *ccdc, struct v4l2_subdev_fh *fh,
 		unsigned int pad, struct v4l2_mbus_framefmt *fmt,

There's a very similar patch that is currently pending in my queue. It adds support for other Bayer patterns. Your overall approach is good, but the two patches will conflict.

Once the other one will get committed, your patch will become much simpler. I thus won't comment on the parts that will disappear then.

drivers/media/video/isp/ispccdc.c static int ccdc_set_format(struct v4l2_subdev *sd,

Quoted 1 line
drivers/media/video/isp/ispccdc.c +0 -0
@@ -1719,6 +1736,45 @@ static int ccdc_set_format(struct v4l2_subdev *sd, 

struct v4l2_subdev_fh *fh,

Quoted 6 lines
drivers/media/video/isp/ispccdc.c +3 -0
 	/* Propagate the format from sink to source */
 	if (pad == CCDC_PAD_SINK) {
+		u32 syn_mode, ispctrl_val;
+		struct isp_device *isp = to_isp_device(ccdc);
+		if (!isp_get(isp))

drivers/media/video/isp/ispccdc.c static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh, · line 1751

Quoted 10 lines
drivers/media/video/isp/ispccdc.c +10 -0
+			return -EBUSY;
+
+		syn_mode    = isp_reg_readl(isp, OMAP3_ISP_IOMEM_CCDC, 
+					    ISPCCDC_SYN_MODE);
+		ispctrl_val = isp_reg_readl(isp, OMAP3_ISP_IOMEM_MAIN, 
+					    ISP_CTRL);
+		syn_mode    &= ISPCCDC_SYN_MODE_INPMOD_MASK;
+		ispctrl_val &= ~(ISPCTRL_PAR_BRIDGE_MASK 
+				 << ISPCTRL_PAR_BRIDGE_SHIFT);
+		switch(format->code) {

Documentation/CodingStyle requires a space after the switch keyword. Please run scripts/checkpatch.pl before submitting patches.

Quoted 11 lines
drivers/media/video/isp/ispccdc.c +11 -0
+		case V4L2_MBUS_FMT_YUYV16_1X16:
+		case V4L2_MBUS_FMT_UYVY16_1X16:
+		case V4L2_MBUS_FMT_YVYU16_1X16:
+		case V4L2_MBUS_FMT_VYUY16_1X16:
+			syn_mode |= ISPCCDC_SYN_MODE_INPMOD_YCBCR16;
+
+			/* TODO: In YCBCR16 mode, the bridge has to be
+			 * enabled, so we enable it here and force it
+			 * big endian. Whether to do big or little endian
+			 * should somehow come from the platform data.*/
+			ispctrl_val |= ISPCTRL_PAR_BRIDGE_BENDIAN 

drivers/media/video/isp/ispccdc.c static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh, · line 1772

Quoted 10 lines
drivers/media/video/isp/ispccdc.c +10 -0
+				<< ISPCTRL_PAR_BRIDGE_SHIFT;
+			break;
+		default:
+			syn_mode |= ISPCCDC_SYN_MODE_INPMOD_RAW;
+			ispctrl_val |= isp->pdata->parallel.bridge
+				<< ISPCTRL_PAR_BRIDGE_SHIFT;
+			break;
+		}
+		isp_reg_writel(isp, syn_mode, OMAP3_ISP_IOMEM_CCDC, 
+			       ISPCCDC_SYN_MODE);

Writing to the ISPCCDC_SYN_MODE register should be moved to ccdc_configure(). Just move the switch statement there right after the

	format = &ccdc->formats[CCDC_PAD_SINK];

line (without the ispctrl_val settings), it should be enough.

drivers/media/video/isp/ispccdc.c static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh, · line 1774

Quoted 2 lines
drivers/media/video/isp/ispccdc.c +2 -0
+		isp_reg_writel(isp, ispctrl_val, OMAP3_ISP_IOMEM_MAIN, 
+			       ISP_CTRL);

The ISP_CTRL register should be written in isp_select_bridge_input() only. As you correctly mention, whether the data is in little endian or big endian format should come from platform data, so I think it's fine to force board files to set the isp->pdata->parallel.bridge field to the correct value.

Review

Quoted 7 lines
drivers/media/video/isp/ispccdc.c +3 -0
+		isp_put(isp);
+
+
 		format = __ccdc_get_format(ccdc, fh, CCDC_PAD_SOURCE_OF, which);
 		memcpy(format, fmt, sizeof(*format));
 		ccdc_try_format(ccdc, fh, CCDC_PAD_SOURCE_OF, format, which);
(quoted diff) +0 -0

b/drivers/media/video/isp/ispreg.h

Quoted 5 lines

index 7efcfaa..4c191af 100644

drivers/media/video/isp/ispreg.h +0 -0
@@ -732,10 +732,10 @@
 #define ISPCTRL_PAR_SER_CLK_SEL_MASK		0xFFFFFFFC

drivers/media/video/isp/ispreg.h line 738

Quoted 10 lines
drivers/media/video/isp/ispreg.h +4 -4
 
 #define ISPCTRL_PAR_BRIDGE_SHIFT		2
-#define ISPCTRL_PAR_BRIDGE_DISABLE		(0x0 << 2)
-#define ISPCTRL_PAR_BRIDGE_LENDIAN		(0x2 << 2)
-#define ISPCTRL_PAR_BRIDGE_BENDIAN		(0x3 << 2)
-#define ISPCTRL_PAR_BRIDGE_MASK			(0x3 << 2)
+#define ISPCTRL_PAR_BRIDGE_DISABLE		0x0
+#define ISPCTRL_PAR_BRIDGE_LENDIAN		0x2
+#define ISPCTRL_PAR_BRIDGE_BENDIAN		0x3
+#define ISPCTRL_PAR_BRIDGE_MASK			0x3

You should remove the shift in isp_select_bridge_input() instead. Could you please submit a patch that does just that ? Don't forget to sign it and include a meaningful commit message.

Signature · 3 lines
Regards,

Laurent Pinchart
#5 lore raw
Quoted 16 lines · Laurent Pinchart

Hi Lane,

Thanks for the patch.

On Thursday 05 August 2010 20:53:50 Lane Brooks wrote:

[snip]

Quoted 2 lines

I was able to get YUV CCDC capture mode working with rather minimal effort. Attached is a patch with the initial effort. Can you comment?

Writing to the ISPCCDC_SYN_MODE register should be moved to ccdc_configure(). Just move the switch statement there right after the

	format =&ccdc->formats[CCDC_PAD_SINK];
Quoted 8 lines

line (without the ispctrl_val settings), it should be enough.

drivers/media/video/isp/ispccdc.c static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh, · line 1774

Quoted 2 lines
drivers/media/video/isp/ispccdc.c +2 -0
+		isp_reg_writel(isp, ispctrl_val, OMAP3_ISP_IOMEM_MAIN,
+			       ISP_CTRL);

The ISP_CTRL register should be written in isp_select_bridge_input() only. As you correctly mention, whether the data is in little endian or big endian format should come from platform data, so I think it's fine to force board files to set the isp->pdata->parallel.bridge field to the correct value.

Putting the bridge settings in the platform data is tricky because they need to change depending on the selected format. For example, for my board, when in SGRBG mode, the bridge needs disabled. When in YUV16 mode, however, I need need to select BIG/LITTLE endian depending on whether it is YUYV or UYVY or ... I am not quite sure how to capture that in the platform data without enumerating every supported format code in the platform data. The current patch knows (based on the OMAP TRM) that YUV16 mode requires the bridge to be enabled. So in the platform data I specify the bridge state for SGBRG mode and force the bridge to BIG endian in YUV16 mode. This leaves the sensor to switch the phasing based on YUYV, YVYU, etc. mode. I am not sure who should be in charge of doing byte swapping in general, but if the input and output modes are the same, then big endian should be used to avoid a byte swap. In other words, the mode is completely determinable based on the formats, so perhaps I should implement it that way. If the input and output port require a byte swap, then go little endian, otherwise go big endian.

The reason why I put both writes to the ISPCTRL and SYN_MODE registers where I did. Moving them to other places will require querying the selected format code. Is that what you want as well?

Lane

#6 lore raw

Hi Lane,

Quoted 28 lines · On Monday 09 August 2010 00:56:27 Lane Brooks wrote:
Quoted 18 lines · Laurent Pinchart

Hi Lane,

Thanks for the patch.

On Thursday 05 August 2010 20:53:50 Lane Brooks wrote:

[snip]

Quoted 2 lines

I was able to get YUV CCDC capture mode working with rather minimal effort. Attached is a patch with the initial effort. Can you comment?

Writing to the ISPCCDC_SYN_MODE register should be moved to ccdc_configure(). Just move the switch statement there right after the

	format =&ccdc->formats[CCDC_PAD_SINK];

line (without the ispctrl_val settings), it should be enough.

Quoted 8 lines

drivers/media/video/isp/ispccdc.c static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh, · line 1774

Quoted 2 lines
drivers/media/video/isp/ispccdc.c +2 -0
+		isp_reg_writel(isp, ispctrl_val, OMAP3_ISP_IOMEM_MAIN,
+			       ISP_CTRL);

The ISP_CTRL register should be written in isp_select_bridge_input() only. As you correctly mention, whether the data is in little endian or big endian format should come from platform data, so I think it's fine to force board files to set the isp->pdata->parallel.bridge field to the correct value.

Quoted 5 lines

Putting the bridge settings in the platform data is tricky because they need to change depending on the selected format. For example, for my board, when in SGRBG mode, the bridge needs disabled. When in YUV16 mode, however, I need need to select BIG/LITTLE endian depending on whether it is YUYV or UYVY or ...

Ah right... So your sensor can output both Bayer and YUV data ? What sensor is that BTW ?

Quoted 11 lines

I am not quite sure how to capture that in the platform data without enumerating every supported format code in the platform data. The current patch knows (based on the OMAP TRM) that YUV16 mode requires the bridge to be enabled. So in the platform data I specify the bridge state for SGBRG mode and force the bridge to BIG endian in YUV16 mode. This leaves the sensor to switch the phasing based on YUYV, YVYU, etc. mode. I am not sure who should be in charge of doing byte swapping in general, but if the input and output modes are the same, then big endian should be used to avoid a byte swap. In other words, the mode is completely determinable based on the formats, so perhaps I should implement it that way. If the input and output port require a byte swap, then go little endian, otherwise go big endian.

OK I understand. The best solution (for now) would then be to modify isp_configure_bridge(). I wrote a few patches that modify how platform data is handled, but I can't commit them at the moment (they depend on other patches that I still need to clean up).

Quoted 3 lines

The reason why I put both writes to the ISPCTRL and SYN_MODE registers where I did. Moving them to other places will require querying the selected format code. Is that what you want as well?

For SYN_MODE, definitely. For ISPCTRL, you can hack isp_configure_bridge() to retrieve the current CCDC input format, and we'll write a proper fix right after I commit my platform data restructuring patches.

Signature · 3 lines
Regards,

Laurent Pinchart
#7 lore raw
Quoted 28 lines · Laurent Pinchart

Hi Lane,

Quoted 25 lines · On Monday 09 August 2010 00:56:27 Lane Brooks wrote:
Quoted 24 lines · Laurent Pinchart

Hi Lane,

Thanks for the patch.

On Thursday 05 August 2010 20:53:50 Lane Brooks wrote:

[snip]

Quoted 2 lines

I was able to get YUV CCDC capture mode working with rather minimal effort. Attached is a patch with the initial effort. Can you comment?

Writing to the ISPCCDC_SYN_MODE register should be moved to ccdc_configure(). Just move the switch statement there right after the

	format =&ccdc->formats[CCDC_PAD_SINK];

line (without the ispctrl_val settings), it should be enough.

drivers/media/video/isp/ispccdc.c static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh, · line 1774

Quoted 2 lines
drivers/media/video/isp/ispccdc.c +2 -0
+		isp_reg_writel(isp, ispctrl_val, OMAP3_ISP_IOMEM_MAIN,
+			       ISP_CTRL);

The ISP_CTRL register should be written in isp_select_bridge_input() only. As you correctly mention, whether the data is in little endian or big endian format should come from platform data, so I think it's fine to force board files to set the isp->pdata->parallel.bridge field to the correct value.

Quoted 8 lines
Quoted 5 lines

Putting the bridge settings in the platform data is tricky because they need to change depending on the selected format. For example, for my board, when in SGRBG mode, the bridge needs disabled. When in YUV16 mode, however, I need need to select BIG/LITTLE endian depending on whether it is YUYV or UYVY or ...

Ah right... So your sensor can output both Bayer and YUV data ? What sensor is that BTW ?

Aptina MT9T111. It can even output JPEG.

Quoted 16 lines
Quoted 11 lines

I am not quite sure how to capture that in the platform data without enumerating every supported format code in the platform data. The current patch knows (based on the OMAP TRM) that YUV16 mode requires the bridge to be enabled. So in the platform data I specify the bridge state for SGBRG mode and force the bridge to BIG endian in YUV16 mode. This leaves the sensor to switch the phasing based on YUYV, YVYU, etc. mode. I am not sure who should be in charge of doing byte swapping in general, but if the input and output modes are the same, then big endian should be used to avoid a byte swap. In other words, the mode is completely determinable based on the formats, so perhaps I should implement it that way. If the input and output port require a byte swap, then go little endian, otherwise go big endian.

OK I understand. The best solution (for now) would then be to modify isp_configure_bridge(). I wrote a few patches that modify how platform data is handled, but I can't commit them at the moment (they depend on other patches that I still need to clean up).

Quoted 6 lines
Quoted 3 lines

The reason why I put both writes to the ISPCTRL and SYN_MODE registers where I did. Moving them to other places will require querying the selected format code. Is that what you want as well?

For SYN_MODE, definitely. For ISPCTRL, you can hack isp_configure_bridge() to retrieve the current CCDC input format, and we'll write a proper fix right after I commit my platform data restructuring patches.

I'll wait for your patches then. Let me know.

#8 lore raw

Hi Lane,

Quoted 59 lines · On Monday 09 August 2010 15:38:56 Lane Brooks wrote:
Quoted 25 lines · Laurent Pinchart
Quoted 24 lines · On Monday 09 August 2010 00:56:27 Lane Brooks wrote:
Quoted 22 lines · Laurent Pinchart

On Thursday 05 August 2010 20:53:50 Lane Brooks wrote:

[snip]

Quoted 2 lines

I was able to get YUV CCDC capture mode working with rather minimal effort. Attached is a patch with the initial effort. Can you comment?

Writing to the ISPCCDC_SYN_MODE register should be moved to ccdc_configure(). Just move the switch statement there right after the

	format =&ccdc->formats[CCDC_PAD_SINK];

line (without the ispctrl_val settings), it should be enough.

drivers/media/video/isp/ispccdc.c static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh, · line 1774

Quoted 2 lines
drivers/media/video/isp/ispccdc.c +2 -0
+		isp_reg_writel(isp, ispctrl_val, OMAP3_ISP_IOMEM_MAIN,
+			       ISP_CTRL);

The ISP_CTRL register should be written in isp_select_bridge_input() only. As you correctly mention, whether the data is in little endian or big endian format should come from platform data, so I think it's fine to force board files to set the isp->pdata->parallel.bridge field to the correct value.

Quoted 8 lines
Quoted 5 lines

Putting the bridge settings in the platform data is tricky because they need to change depending on the selected format. For example, for my board, when in SGRBG mode, the bridge needs disabled. When in YUV16 mode, however, I need need to select BIG/LITTLE endian depending on whether it is YUYV or UYVY or ...

Ah right... So your sensor can output both Bayer and YUV data ? What sensor is that BTW ?

Aptina MT9T111. It can even output JPEG.

Quoted 22 lines
Quoted 4 lines

I am not quite sure how to capture that in the platform data without enumerating every supported format code in the platform data. The current patch knows (based on the OMAP TRM) that YUV16 mode requires the bridge to be enabled. So in the platform data I specify the bridge

Quoted 8 lines

state for SGBRG mode and force the bridge to BIG endian in YUV16 mode. This leaves the sensor to switch the phasing based on YUYV, YVYU, etc. mode. I am not sure who should be in charge of doing byte swapping in general, but if the input and output modes are the same, then big endian should be used to avoid a byte swap. In other words, the mode is completely determinable based on the formats, so perhaps I should implement it that way. If the input and output port require a byte swap, then go little endian, otherwise go big endian.

OK I understand. The best solution (for now) would then be to modify isp_configure_bridge(). I wrote a few patches that modify how platform data is handled, but I can't commit them at the moment (they depend on other patches that I still need to clean up).

Quoted 3 lines

The reason why I put both writes to the ISPCTRL and SYN_MODE registers where I did. Moving them to other places will require querying the selected format code. Is that what you want as well?

Quoted 6 lines
Quoted 4 lines

For SYN_MODE, definitely. For ISPCTRL, you can hack isp_configure_bridge() to retrieve the current CCDC input format, and we'll write a proper fix right after I commit my platform data restructuring patches.

I'll wait for your patches then. Let me know.

The omap3camera repository is now deprecated. Please use

http://meego.gitorious.org/maemo-multimedia/omap3isp-rx51

where you will find the latest patches based on 2.6.35. Have fun :-)

I will be on holidays next week, so I won't be able to provide assistance before week 37.

Signature · 3 lines
Regards,

Laurent Pinchart
#9 lore raw
In reply to Lane Brooks
Quoted 12 lines · Lane Brooks
Quoted 3 lines · Laurent Pinchart

Hi Lane,

On Tuesday 03 August 2010 17:26:48 Lane Brooks wrote:

[snip]

Quoted 7 lines
Quoted 7 lines

My question:

- Are there other things I need to when I enable the parallel bridge? For example, do I need to change a clock rate somewhere? From the TRM, it seems like it should just work without any changes, but maybe I am missing something.

Quoted 7 lines
Quoted 6 lines

Good question. ISP bridge and YUV modes support are not implemented in the driver, but you're probably already aware of that.

I unfortunately have no straightforward answer. Try tracing the ISP interrupts and monitoring the CCDC SBL busy bits to see if the CCDC writes images to memory correctly.

[snip]

Laurent,

I was able to get CCDC capture of YUV data working with rather minimal effort. Attached is a patch with the initial work. Can you comment?

Thanks,
Lane

drivers/media/video/isp/ispccdc.c +59 -4
index 90bcc6c..9226406 100644
@@ -1563,6 +1563,15 @@ __ccdc_get_format(struct isp_ccdc_device *ccdc, struct v4l2_subdev_fh *fh,
  * @pad: Pad number
  * @fmt: Format
  */
+
+static enum v4l2_mbus_pixelcode sink_fmts[] = {
+	V4L2_MBUS_FMT_SGRBG10_1X10,
+	V4L2_MBUS_FMT_YUYV16_1X16,
+	V4L2_MBUS_FMT_UYVY16_1X16,
+	V4L2_MBUS_FMT_YVYU16_1X16,
+	V4L2_MBUS_FMT_VYUY16_1X16,
+};
+
 static void
 ccdc_try_format(struct isp_ccdc_device *ccdc, struct v4l2_subdev_fh *fh,
 		unsigned int pad, struct v4l2_mbus_framefmt *fmt,
@@ -1571,14 +1580,22 @@ ccdc_try_format(struct isp_ccdc_device *ccdc, struct v4l2_subdev_fh *fh,
 	struct v4l2_mbus_framefmt *format;
 	unsigned int width = fmt->width;
 	unsigned int height = fmt->height;
-
+	int i;
 	switch (pad) {
 	case CCDC_PAD_SINK:
 		/* Check if the requested pixel format is supported.
 		 * TODO: If the CCDC output formatter pad is connected directly
 		 * to the resizer, only YUV formats can be used.
 		 */
-		fmt->code = V4L2_MBUS_FMT_SGRBG10_1X10;
+		for(i=0; i<ARRAY_SIZE(sink_fmts); ++i) {
+			if(sink_fmts[i] == fmt->code) 
+				break;
+		}
+		/* if the requested type is not in the list of
+		 * supported types, then change it to the first format
+		 * code in the list of supported. */
+		if(i==ARRAY_SIZE(sink_fmts)) 
+			fmt->code = sink_fmts[0];
 
 		/* Clamp the input size. */
 		fmt->width = clamp_t(u32, width, 32, 4096);
@@ -1630,10 +1647,10 @@ static int ccdc_enum_mbus_code(struct v4l2_subdev *sd,
 			       struct v4l2_subdev_fh *fh,
 			       struct v4l2_subdev_pad_mbus_code_enum *code)
 {
-	if (code->index != 0)
+	if (code->index >= ARRAY_SIZE(sink_fmts))
 		return -EINVAL;
 
-	code->code = V4L2_MBUS_FMT_SGRBG10_1X10;
+	code->code = sink_fmts[code->index];
 
 	return 0;
 }
@@ -1719,6 +1736,44 @@ static int ccdc_set_format(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh,
 
 	/* Propagate the format from sink to source */
 	if (pad == CCDC_PAD_SINK) {
+		u32 syn_mode, ispctrl_val;
+		struct isp_device *isp = to_isp_device(ccdc);
+		if (!isp_get(isp))
+			return -EBUSY;
+
+		syn_mode    = isp_reg_readl(isp, OMAP3_ISP_IOMEM_CCDC, 
+					    ISPCCDC_SYN_MODE);
+		ispctrl_val = isp_reg_readl(isp, OMAP3_ISP_IOMEM_MAIN, 
+					    ISP_CTRL);
+		syn_mode    &= ISPCCDC_SYN_MODE_INPMOD_MASK;
+		ispctrl_val &= ~ISPCTRL_PAR_BRIDGE_MASK;
+		switch(format->code) {
+		case V4L2_MBUS_FMT_YUYV16_1X16:
+		case V4L2_MBUS_FMT_UYVY16_1X16:
+		case V4L2_MBUS_FMT_YVYU16_1X16:
+		case V4L2_MBUS_FMT_VYUY16_1X16:
+			syn_mode |= ISPCCDC_SYN_MODE_INPMOD_YCBCR16;
+
+			/* TODO: In YCBCR16 mode, the bridge has to be
+			 * enabled, so we enable it here and force it
+			 * big endian. Whether to do big or little endian
+			 * should somehow come from the platform data.*/
+			ispctrl_val |= ISPCTRL_PAR_BRIDGE_BENDIAN 
+				<< ISPCTRL_PAR_BRIDGE_SHIFT;
+			break;
+		default:
+			syn_mode |= ISPCCDC_SYN_MODE_INPMOD_RAW;
+			ispctrl_val |= isp->pdata->parallel.bridge
+				<< ISPCTRL_PAR_BRIDGE_SHIFT;
+			break;
+		}
+		isp_reg_writel(isp, syn_mode, OMAP3_ISP_IOMEM_CCDC, 
+			       ISPCCDC_SYN_MODE);
+		isp_reg_writel(isp, ispctrl_val, OMAP3_ISP_IOMEM_MAIN, 
+			       ISP_CTRL);
+		isp_put(isp);
+
+
 		format = __ccdc_get_format(ccdc, fh, CCDC_PAD_SOURCE_OF, which);
 		memcpy(format, fmt, sizeof(*format));
 		ccdc_try_format(ccdc, fh, CCDC_PAD_SOURCE_OF, format, which);
drivers/media/video/isp/ispreg.h +3 -3
index 7efcfaa..9bbeb9b 100644
@@ -732,9 +732,9 @@
 #define ISPCTRL_PAR_SER_CLK_SEL_MASK		0xFFFFFFFC
 
 #define ISPCTRL_PAR_BRIDGE_SHIFT		2
-#define ISPCTRL_PAR_BRIDGE_DISABLE		(0x0 << 2)
-#define ISPCTRL_PAR_BRIDGE_LENDIAN		(0x2 << 2)
-#define ISPCTRL_PAR_BRIDGE_BENDIAN		(0x3 << 2)
+#define ISPCTRL_PAR_BRIDGE_DISABLE		0x0
+#define ISPCTRL_PAR_BRIDGE_LENDIAN		0x2
+#define ISPCTRL_PAR_BRIDGE_BENDIAN		0x3
 #define ISPCTRL_PAR_BRIDGE_MASK			(0x3 << 2)
 
 #define ISPCTRL_PAR_CLK_POL_SHIFT		4
#10 lore raw
In reply to Lane Brooks

Hi Lane,

Quoted 40 lines · On Thursday 05 August 2010 18:06:51 Lane Brooks wrote:
Quoted 1 line · Laurent Pinchart

On Tuesday 03 August 2010 17:26:48 Lane Brooks wrote:

[snip]

Quoted 7 lines
Quoted 6 lines

My question:

- Are there other things I need to when I enable the parallel bridge? For example, do I need to change a clock rate somewhere? From the TRM, it seems like it should just work without any changes, but maybe I am missing something.

Quoted 6 lines

Good question. ISP bridge and YUV modes support are not implemented in the driver, but you're probably already aware of that.

I unfortunately have no straightforward answer. Try tracing the ISP interrupts and monitoring the CCDC SBL busy bits to see if the CCDC writes images to memory correctly.

I found at least some of the problem. In my platform data I was enabling the bridge using the #defines in ispreg.h as in

static struct isp_platform_data bmi_isp_platform_data = {
.parallel = {
.data_lane_shift = 3,
.clk_pol = 0,
.bridge = ISPCTRL_PAR_BRIDGE_LENDIAN,
},
.subdevs = bmi_camera_subdevs,
};

The bridge related #defines in ispreg.h, however, have a shift of 2 applied to them. The problem is that the shift is applied again in isp.c when the options are actually applied. In other words, the bridge parameters are being shifted up twice, which is causing corruption to the control register and causing my hanging problems when I try to enable the bridge. It seems there are several other such cases in the ispreg.h where double shifting might occur if the user tries to use them in the platform data.

Quoted 5 lines

My question:
Is this an oversight or is it this way on purpose? Am I not supposed to be using these defines in my platform definitions? It seems that *some* of the parameters in ispreg.h should not be shifted up (like the bridge options).

It's a bug, thanks for pointing it out. The value shouldn't be shifted again in isp_select_bridge_input(). Do you want to submit a patch ?

Signature · 3 lines
Regards,

Laurent Pinchart
#11 lore raw
Quoted 41 lines · Laurent Pinchart

Hi Lane,

Quoted 38 lines · On Thursday 05 August 2010 18:06:51 Lane Brooks wrote:
Quoted 1 line · Laurent Pinchart

On Tuesday 03 August 2010 17:26:48 Lane Brooks wrote:

[snip]

Quoted 6 lines
Quoted 6 lines

My question:

- Are there other things I need to when I enable the parallel bridge? For example, do I need to change a clock rate somewhere? From the TRM, it seems like it should just work without any changes, but maybe I am missing something.

Quoted 6 lines

Good question. ISP bridge and YUV modes support are not implemented in the driver, but you're probably already aware of that.

I unfortunately have no straightforward answer. Try tracing the ISP interrupts and monitoring the CCDC SBL busy bits to see if the CCDC writes images to memory correctly.

I found at least some of the problem. In my platform data I was enabling the bridge using the #defines in ispreg.h as in

static struct isp_platform_data bmi_isp_platform_data = {
.parallel = {
.data_lane_shift = 3,
.clk_pol = 0,
.bridge = ISPCTRL_PAR_BRIDGE_LENDIAN,
},
.subdevs = bmi_camera_subdevs,
};

The bridge related #defines in ispreg.h, however, have a shift of 2 applied to them. The problem is that the shift is applied again in isp.c when the options are actually applied. In other words, the bridge parameters are being shifted up twice, which is causing corruption to the control register and causing my hanging problems when I try to enable the bridge. It seems there are several other such cases in the ispreg.h where double shifting might occur if the user tries to use them in the platform data.

Quoted 8 lines
Quoted 5 lines

My question:
Is this an oversight or is it this way on purpose? Am I not supposed to be using these defines in my platform definitions? It seems that *some* of the parameters in ispreg.h should not be shifted up (like the bridge options).

It's a bug, thanks for pointing it out. The value shouldn't be shifted again in isp_select_bridge_input(). Do you want to submit a patch ?

The isp_parallel_platform_data struct specifies the bridge definition as 2 bits, so if the shift is removed from isp_select_bridge instead of from the ispreg.h file, then the platform_data definition needs modified as well. Is that what you want?

#12 lore raw

Hi Lane,

Quoted 52 lines · On Monday 09 August 2010 00:34:24 Lane Brooks wrote:
Quoted 51 lines · Laurent Pinchart
Quoted 49 lines · On Thursday 05 August 2010 18:06:51 Lane Brooks wrote:
Quoted 1 line · Laurent Pinchart

On Tuesday 03 August 2010 17:26:48 Lane Brooks wrote:

[snip]

Quoted 7 lines
Quoted 6 lines

My question:

- Are there other things I need to when I enable the parallel bridge? For example, do I need to change a clock rate somewhere? From the TRM, it seems like it should just work without any changes, but maybe I am missing something.

Quoted 6 lines

Good question. ISP bridge and YUV modes support are not implemented in the driver, but you're probably already aware of that.

I unfortunately have no straightforward answer. Try tracing the ISP interrupts and monitoring the CCDC SBL busy bits to see if the CCDC writes images to memory correctly.

I found at least some of the problem. In my platform data I was enabling the bridge using the #defines in ispreg.h as in

static struct isp_platform_data bmi_isp_platform_data = {

.parallel = {

.data_lane_shift = 3,
.clk_pol = 0,
.bridge = ISPCTRL_PAR_BRIDGE_LENDIAN,

},
.subdevs = bmi_camera_subdevs,

};

The bridge related #defines in ispreg.h, however, have a shift of 2 applied to them. The problem is that the shift is applied again in isp.c when the options are actually applied. In other words, the bridge parameters are being shifted up twice, which is causing corruption to the control register and causing my hanging problems when I try to enable the bridge. It seems there are several other such cases in the ispreg.h where double shifting might occur if the user tries to use them in the platform data.

My question:
Is this an oversight or is it this way on purpose? Am I not supposed to be using these defines in my platform definitions? It seems that *some* of the parameters in ispreg.h should not be shifted up (like the bridge options).

Quoted 7 lines
Quoted 2 lines

It's a bug, thanks for pointing it out. The value shouldn't be shifted again in isp_select_bridge_input(). Do you want to submit a patch ?

The isp_parallel_platform_data struct specifies the bridge definition as 2 bits, so if the shift is removed from isp_select_bridge instead of from the ispreg.h file, then the platform_data definition needs modified as well. Is that what you want?

Good point. I think it's important to keep the registers definitions consistent (they all have - or should have - the shift applied), so please extend the bridge field to 4 bits.

Signature · 3 lines
Regards,

Laurent Pinchart