OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

[Andreas Nagel]

[-- Attachment #1: Type: text/plain, Size: 2250 bytes --]

Hello,

in order to familiarize myself with Media Controller and V4L2 I am 
creating a small example program for capturing some frames through the 
OMAP3 ISP.
The hardware used is a TAO-3530 on a Tsunami daughterboard from 
Technexion. My video source is a standard DVD player connected to the 
daughterboards S-VIDEO port. That port itself is wired to a TVP5146 
decoder chip from TI.
A precompiled Android image with a demo app proofs, that the hardware is 
working fine.

My example program is mostly based on the following wiki page and the 
capture example in the V4L2 documentation.
http://processors.wiki.ti.com/index.php/Writing_V4L2_Media_Controller_Applications_on_Dm36x_Video_Capture

My code sets up the ISP pipeline, configures the format on all the 
subdevices pads and the actual video device. Works fine so far.
Then I passed user pointers (aquired with malloc) to the device driver 
for the capture buffers. Before issuing VIDIOC_STREAMON, I enqueue my 
buffers with VIDIOC_QBUF, which fails with errno = EIO. I don't know, 
why this is happening or where to got from here.

When using memory-mapped buffers instead, mapping the addresses to 
userspace works fine as well as VIDIOC_QBUF calls. But then 
VIDIOC_STREAMON fails with EINVAL. According to V4L documentation, 
EINVAL means
a) buffertype (V4L2_BUF_TYPE_VIDEO_CAPTURE in this case) not supported
b) no buffers have been allocated (memory mapping)
c) or enqueued yet

Because I tested V4L2_CAP_VIDEO_CAPTURE capability, I guess option a) 
does not apply. Buffers have been enqueud, so c) doesn't apply either.
What about b) ? As I chose memory-mapped buffers here, the device 
drivers manages the buffers. How can I make sure, that buffers were 
actually allocated?

And am I missing something else?

I attached my example code. If you need more information, I will provide it.

Note: I have to use the Technexion 2.6.37 kernel, which is based on the 
TI kernel. It's the only kernel that comes with the ISP driver and Media 
Controller API onboard and I guess, TI or TN included this stuff 
somehow. Normally, this shouldn't be available until 2.6.39. Sadly, I 
cannot use another kernel, because Technexion doesn't push board support 
anywhere.


Best regards,
Andreas




[-- Attachment #2: capture.c --]
[-- Type: text/x-csrc, Size: 9515 bytes --]

/*
 * capture.c
 *
 *  Created on: 29.10.2012
 *      Author: andreas
 */

#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <linux/media.h>
#include <linux/v4l2-subdev.h>
#include <string.h>
#include <errno.h>
#include <sys/ioctl.h>
#include <sys/mman.h>


#include "capture.h"

#define CLEAR(x) memset(&x, 0, sizeof(x))

int main(void) {

	/*
	 * Open media device.
	 */
	int media_fd = open(DEVNODE_ISP, O_RDWR);
	if (media_fd < 0) {
		puts("Can't open media device.");
		return -1;
	}

	/*
	 * Get the required entities.
	 */
	entities_t entities;
	CLEAR(entities);

	if (get_entities(media_fd, &entities) != 0) {
		puts("Required entity IDs could not be determined.");
		return -1;
	}

	/*
	 * Connect tvp --> ccdc
	 */
	struct media_link_desc link;
	CLEAR(link);

	link.flags |= MEDIA_LINK_FLAG_ENABLED;
	link.source.entity = entities.tvp514x;
	link.source.index = PAD_TVP514X;
	link.source.flags = MEDIA_PAD_FLAG_OUTPUT;
	link.sink.entity = entities.ccdc;
	link.sink.index = PAD_CCDC_SINK;
	link.sink.flags = MEDIA_PAD_FLAG_INPUT;

	if (ioctl(media_fd, MEDIA_IOC_SETUP_LINK, &link) == 0) {
		puts("tvp --> ccdc: enabled");
	}
	else {
		printf("Error setting up link [tvp --> ccdc]. Error %d, %s.\n", errno, strerror(errno));
		return -1;
	}

	/*
	 * Connect ccdc --> ccdc_output
	 */
	CLEAR(link);

	link.flags |= MEDIA_LINK_FLAG_ENABLED;
	link.source.entity = entities.ccdc;
	link.source.index = PAD_CCDC_SOURCE;
	link.source.flags = MEDIA_PAD_FLAG_OUTPUT;
	link.sink.entity = entities.ccdc_out;
	link.sink.index = PAD_CCDC_OUTPUT;
	link.sink.flags = MEDIA_PAD_FLAG_INPUT;

	if (ioctl(media_fd, MEDIA_IOC_SETUP_LINK, &link) == 0) {
		puts("ccdc --> ccdc-out: enabled");
	}
	else {
		printf(	"Error setting up link [ccdc --> ccdc-out]. Error %d, %s.\n",
				errno,
				strerror(errno));
		return -1;
	}

	/*
	 * Open capture device
	 */
	int video_fd = open(DEVNODE_CCDC_OUT, O_RDWR | O_NONBLOCK, 0);
	if (video_fd < 0) {
		puts("Can't open capture device.");
		return -1;
	}

	/*
	 * Check some capabilities.
	 */
	struct v4l2_capability cap;
	printf("Checking device capabilites...");
	if (ioctl(video_fd, VIDIOC_QUERYCAP, &cap) != 0) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
	}
	else {
		puts("ok.");
		printf(	"Device %s is a video capture device... %s\n",
				DEVNODE_CCDC_OUT,
				cap.capabilities & V4L2_CAP_VIDEO_CAPTURE ? "yes" : "no");

		printf(	"Device %s supports streaming... %s\n",
				DEVNODE_CCDC_OUT,
				cap.capabilities & V4L2_CAP_STREAMING ? "yes" : "no");
	}

	/*
	 * Setting camera as input
	 */
	struct v4l2_input input;
	CLEAR(input);
	input.type = V4L2_INPUT_TYPE_CAMERA;
	input.index = 70;		// required for the tvp514x decoder chip

	printf("Setting camera input... ");
	if (-1 == ioctl(video_fd, VIDIOC_S_INPUT, &input.index)) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
		return -1;
	}
	else {
		puts("ok.");
	}

	/*
	 * Set format on TVP output pad.
	 */
	int tvp_fd = open(DEVNODE_TVP514X, O_RDWR);
	if (tvp_fd < 0) {
		puts("Can't open tvp device.");
		return -1;
	}

	struct v4l2_subdev_format fmt;
	CLEAR(fmt);

	fmt.pad = PAD_TVP514X;
	fmt.which = V4L2_SUBDEV_FORMAT_ACTIVE;
	fmt.format.code = V4L2_MBUS_FMT_UYVY8_2X8;
	fmt.format.width = WIDTH;
	fmt.format.height = HEIGHT;
	fmt.format.field = V4L2_FIELD_INTERLACED;

	printf("Setting format on TVP subdev... ");
	if (ioctl(tvp_fd, VIDIOC_SUBDEV_S_FMT, &fmt) != 0) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
		return -1;
	}
	else {
		puts("ok.");
	}

	/*
	 * Set format on CCDC input pad.
	 */
	int ccdc_fd = open(DEVNODE_CCDC, O_RDWR);
	if (ccdc_fd < 0) {
		puts("Can't open CCDC subdev.");
		return -1;
	}

	CLEAR(fmt);
	fmt.pad = PAD_CCDC_SINK;
	fmt.which = V4L2_SUBDEV_FORMAT_ACTIVE;
	fmt.format.code = V4L2_MBUS_FMT_UYVY8_2X8;
	fmt.format.width = WIDTH;
	fmt.format.height = HEIGHT;
	fmt.format.field = V4L2_FIELD_INTERLACED;

	printf("Setting format on CCDC sink pad... ");
	if (ioctl(ccdc_fd, VIDIOC_SUBDEV_S_FMT, &fmt) != 0) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
		return -1;
	}
	else {
		puts("ok.");
	}

	/*
	 * Set format on CCDC output pad.
	 */
	CLEAR(fmt);
	fmt.pad = PAD_CCDC_SOURCE;
	fmt.which = V4L2_SUBDEV_FORMAT_ACTIVE;
	fmt.format.code = V4L2_MBUS_FMT_UYVY8_2X8;
	fmt.format.width = WIDTH;
	fmt.format.height = HEIGHT;
	fmt.format.colorspace = V4L2_COLORSPACE_SMPTE170M; // <-- ITU BT.601
	fmt.format.field = V4L2_FIELD_INTERLACED;

	printf("Setting format on CCDC source pad... ");
	if (ioctl(ccdc_fd, VIDIOC_SUBDEV_S_FMT, &fmt) != 0) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
		return -1;
	}
	else {
		puts("ok.");
	}

	/*
	 * Set format on video node
	 */
	struct v4l2_format fmtx;
	CLEAR(fmtx);
	fmtx.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
	fmtx.fmt.pix.width = WIDTH;
	fmtx.fmt.pix.height = HEIGHT;
	fmtx.fmt.pix.pixelformat = V4L2_PIX_FMT_YUYV;
	fmtx.fmt.pix.field = V4L2_FIELD_INTERLACED;

	printf("Setting format on video node... ");
	if (ioctl(video_fd, VIDIOC_S_FMT, &fmtx) != 0) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
		return -1;
	}
	else {
		puts("ok.");
	}

	/*
	 * Get pitch (bytes per line).
	 */
	printf("Getting pitch... ");
	if (ioctl(video_fd, VIDIOC_G_FMT, &fmtx) != 0) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
		return -1;
	}
	else {
		printf("%d Bytes.\n", fmtx.fmt.pix.bytesperline);
	}

	/*
	 * Request buffers.
	 */
	enum v4l2_memory io_method = V4L2_MEMORY_MMAP;

	struct v4l2_requestbuffers req;
	CLEAR(req);
	req.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
	req.count = NUM_BUFS;
	req.memory = io_method;

	printf("Request buffers (%s)... ", io_method == V4L2_MEMORY_MMAP ? "memory-mapped" : "userptr");
	if (ioctl(video_fd, VIDIOC_REQBUFS, &req) != 0) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
		return -1;
	}
	else {
		printf("ok. Got %d buffers.\n", req.count);
	}

	/*
	 * Query buffers (because of memory mapping).
	 */
	void *capture_buffers[NUM_BUFS];

	int i;
	for (i = 0; i < NUM_BUFS; i++) {
		struct v4l2_buffer buf;
		CLEAR(buf);

		buf.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
		buf.memory = io_method;
		buf.index = i;

		printf("Querying buffer %d... ", i);
		if (ioctl(video_fd, VIDIOC_QUERYBUF, &buf) != 0) {
			printf("failed. Error %d (%s).\n", errno, strerror(errno));
			break;
		}
		else {
			puts("ok.");
		}

		capture_buffers[i] = mmap(	NULL,
									buf.length,
									PROT_READ | PROT_WRITE,
									MAP_SHARED,
									video_fd,
									buf.m.offset);

		printf("Mapping of buffer %d... ", i);
		if (MAP_FAILED == capture_buffers[i]) {
			puts("failed.");
			return -1;
		}
		else {
			puts("ok.");
		}
	}

	/*
	 * Queue buffers
	 */
	for (i = 0; i < NUM_BUFS; i++) {

		struct v4l2_buffer buf;
		CLEAR(buf);

		buf.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
		buf.memory = io_method;
		buf.index = i;

		printf("Enqueuing buffer %d... ", i);
		if (ioctl(video_fd, VIDIOC_QBUF, &buf) != 0) {
			printf("failed. Error %d (%s).\n", errno, strerror(errno));
			return -1;
		}
		else {
			puts("ok.");
		}
	}

	/*
	 * Start streaming
	 */
	enum v4l2_buf_type type;
	CLEAR(type);

	type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
	printf("Start streaming... ");
	if (ioctl(video_fd, VIDIOC_STREAMON, &type) != 0) {
		printf("failed. Error %d (%s).\n", errno, strerror(errno));
	}
	else {
		puts("ok.");
	}

	return 0;
}

//int allocate_cmem_buffers(buf_info_t *bufs) {
//	void *pool;
//
//	CMEM_AllocParams cParams;
//	CMEM_init();
//
//	cParams.type = CMEM_POOL;
//	cParams.flags = CMEM_NONCACHED;
//	cParams.alignment = 32;
//	pool = CMEM_allocPool(0, &cParams);
//	if (pool == NULL ) {
//		puts("Failed to allocate CMEM pool.");
//		return -1;
//	}
//
//	int i;
//	for (i = 0; i < NUM_BUFS; i++) {
////		bufs[i]->user_addr = CMEM_alloc(BUFSIZ, &cParams);
////		if (!bufs[i]->user_addr) {
////			puts("Error allocating cmem buffer.");
////			return -1;
////		}
////		bufs[i]->phy_addr = CMEM_getPhys(bufs[i].user_addr);
////		if (bufs[i]->phy_addr == 0){
////			puts("Error getting physical address.");
////			return -1;
////		}
//		printf("Buffer %d allocated.\n", i);
//
//	}
//
//	return 0;
//}

int get_entities(int media_fd, entities_t *e) {
	/*
	 * This function will determine the entity ids of the needed subdevices.
	 * If all devices are found, it returns 0. In any other case non-zero.
	 */
	int count = 20;

	/*
	 * Set success count to number of integers in struct.
	 * If all entities are found, this count will be 0, indiciating success.
	 */
	int success = sizeof(entities_t) / sizeof(int);

	struct media_entity_desc entity[count];
	int ret;
	int index;
	for (index = 0; index < count; index++) {
		memset(&entity[index], 0, sizeof(struct media_entity_desc));
		entity[index].id = index | MEDIA_ENTITY_ID_FLAG_NEXT;

		ret = ioctl(media_fd, MEDIA_IOC_ENUM_ENTITIES, &entity[index]);
		if (ret < 0) {
			if (errno == EINVAL)
				break;
		}
		else {
			printf("[%2d] %s\t\t", entity[index].id, entity[index].name);

			if (!strcmp(entity[index].name, ENTITY_CCDC_OUT_NAME)) {
				e->ccdc_out = entity[index].id;
				success--;
				puts("(selected)");
			}
			else if (!strcmp(entity[index].name, ENTITY_TVP514X_NAME)) {
				e->tvp514x = entity[index].id;
				success--;
				puts("(selected)");
			}
			else if (!strcmp(entity[index].name, ENTITY_CCDC_NAME)) {
				e->ccdc = entity[index].id;
				success--;
				puts("(selected)");
			}
			else
				puts("");
		}

		/* Break loop, if all devices are found. */
		if (success == 0)
			break;

	}

	return success;
}

[-- Attachment #3: capture.h --]
[-- Type: text/x-chdr, Size: 969 bytes --]

/*
 * capture.h
 *
 *  Created on: 30.10.2012
 *      Author: andreas
 */

#ifndef CAPTURE_H_
#define CAPTURE_H_

#define ENTITY_CCDC_NAME "OMAP3 ISP CCDC"
#define ENTITY_CCDC_OUT_NAME "OMAP3 ISP CCDC output"
#define ENTITY_TVP514X_NAME "tvp514x 2-005d"

#define DEVNODE_CCDC "/dev/v4l-subdev2"
#define DEVNODE_CCDC_OUT "/dev/video2"
#define DEVNODE_TVP514X "/dev/v4l-subdev8"
#define DEVNODE_ISP "/dev/media0"

#define PAD_TVP514X 0
#define PAD_CCDC_SINK 0
#define PAD_CCDC_SOURCE 1
#define PAD_CCDC_OUTPUT 0

#define WIDTH 720
#define HEIGHT 576

#define NUM_BUFS 4
#define BUFSIZE WIDTH*HEIGHT*2



/*
 * Holds the entity ids of the specified devices.
 * Only integers here!!
 */
typedef struct entities {
	int tvp514x;
	int ccdc;
	int ccdc_out;
} entities_t;

typedef struct buf_info {
	void *user_addr;
	unsigned long phy_addr;
} buf_info_t;



//int allocate_cmem_buffers(buf_info_t *bufs);
int get_entities(int media_fd, entities_t *e);

#endif /* CAPTURE_H_ */

Re: OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

[Sakari Ailus]

Hi Andreas,

On Mon, Nov 05, 2012 at 04:47:43PM +0100, Andreas Nagel wrote:
> Hello,
> 
> in order to familiarize myself with Media Controller and V4L2 I am
> creating a small example program for capturing some frames through
> the OMAP3 ISP.
> The hardware used is a TAO-3530 on a Tsunami daughterboard from
> Technexion. My video source is a standard DVD player connected to
> the daughterboards S-VIDEO port. That port itself is wired to a
> TVP5146 decoder chip from TI.
> A precompiled Android image with a demo app proofs, that the
> hardware is working fine.
> 
> My example program is mostly based on the following wiki page and
> the capture example in the V4L2 documentation.
> http://processors.wiki.ti.com/index.php/Writing_V4L2_Media_Controller_Applications_on_Dm36x_Video_Capture
> 
> My code sets up the ISP pipeline, configures the format on all the
> subdevices pads and the actual video device. Works fine so far.
> Then I passed user pointers (aquired with malloc) to the device
> driver for the capture buffers. Before issuing VIDIOC_STREAMON, I
> enqueue my buffers with VIDIOC_QBUF, which fails with errno = EIO. I
> don't know, why this is happening or where to got from here.

One possibility could be that mapping the buffer to ISP MMU fails for a
reason or another. Do you set the length field in the buffer?

> When using memory-mapped buffers instead, mapping the addresses to
> userspace works fine as well as VIDIOC_QBUF calls. But then
> VIDIOC_STREAMON fails with EINVAL. According to V4L documentation,
> EINVAL means
> a) buffertype (V4L2_BUF_TYPE_VIDEO_CAPTURE in this case) not supported
> b) no buffers have been allocated (memory mapping)
> c) or enqueued yet
> 
> Because I tested V4L2_CAP_VIDEO_CAPTURE capability, I guess option
> a) does not apply. Buffers have been enqueud, so c) doesn't apply
> either.
> What about b) ? As I chose memory-mapped buffers here, the device
> drivers manages the buffers. How can I make sure, that buffers were
> actually allocated?
> 
> And am I missing something else?

The formats on the pads at different ends of the links in the pipeline must
match. In most cases, they have to be exactly the same.

Have you used the media-ctl test program here?

<URL:http://git.ideasonboard.org/media-ctl.git>

media-ctl -p gives you (and us) lots of information that helps figuring out
what could go wrong here.

> I attached my example code. If you need more information, I will provide it.
> 
> Note: I have to use the Technexion 2.6.37 kernel, which is based on
> the TI kernel. It's the only kernel that comes with the ISP driver
> and Media Controller API onboard and I guess, TI or TN included this
> stuff somehow. Normally, this shouldn't be available until 2.6.39.
> Sadly, I cannot use another kernel, because Technexion doesn't push
> board support anywhere.

This might work but I think it'd be best if you could just use the mainline
kernel. Others are mostly unsupported.

Cc Laurent.

Kind regards,

-- 
Sakari Ailus
e-mail: sakari.ailus@iki.fi	XMPP: sailus@retiisi.org.uk

Re: OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

[Andreas Nagel]

[-- Attachment #1: Type: text/plain, Size: 1164 bytes --]

Hi Sakari,

thanks for helping.

> My code sets up the ISP pipeline, configures the format on all the
> subdevices pads and the actual video device. Works fine so far.
> Then I passed user pointers (aquired with malloc) to the device
> driver for the capture buffers. Before issuing VIDIOC_STREAMON, I
> enqueue my buffers with VIDIOC_QBUF, which fails with errno = EIO. I
> don't know, why this is happening or where to got from here.
>> One possibility could be that mapping the buffer to ISP MMU fails for 
>> a reason or another. Do you set the length field in the buffer? 

Yes, the length was set when using userptr.

>>
>> And am I missing something else?
> The formats on the pads at different ends of the links in the pipeline must
> match. In most cases, they have to be exactly the same.
>
> Have you used the media-ctl test program here?
>
> <URL:http://git.ideasonboard.org/media-ctl.git>
>
> media-ctl -p gives you (and us) lots of information that helps figuring out
> what could go wrong here.

All pads do indeed have the same format.
I ran media-ctl, as you suggested. You can see the topology in the 
attached textfile.



Kind regards,
Andreas

[-- Attachment #2: topology.txt --]
[-- Type: text/plain, Size: 3819 bytes --]

Opening media device /dev/media0
Enumerating entities
Found 16 entities
Enumerating pads and links
Media controller API version 0.0.0

Media device information
------------------------
driver          omap3isp
model           TI OMAP3 ISP
serial          
bus info        
hw revision     0x0
driver version  0.0.0

Device topology
- entity 1: OMAP3 ISP CCP2 (2 pads, 2 links)
            type V4L2 subdev subtype Unknown
            device node name /dev/v4l-subdev0
	pad0: Sink
		<- "OMAP3 ISP CCP2 input":0 []
	pad1: Source
		-> "OMAP3 ISP CCDC":0 []

- entity 2: OMAP3 ISP CCP2 input (1 pad, 1 link)
            type Node subtype V4L
            device node name /dev/video0
	pad0: Source
		-> "OMAP3 ISP CCP2":0 []

- entity 3: OMAP3 ISP CSI2a (2 pads, 2 links)
            type V4L2 subdev subtype Unknown
            device node name /dev/v4l-subdev1
	pad0: Sink
	pad1: Source
		-> "OMAP3 ISP CSI2a output":0 []
		-> "OMAP3 ISP CCDC":0 []

- entity 4: OMAP3 ISP CSI2a output (1 pad, 1 link)
            type Node subtype V4L
            device node name /dev/video1
	pad0: Sink
		<- "OMAP3 ISP CSI2a":1 []

- entity 5: OMAP3 ISP CCDC (3 pads, 9 links)
            type V4L2 subdev subtype Unknown
            device node name /dev/v4l-subdev2
	pad0: Sink
		<- "OMAP3 ISP CCP2":1 []
		<- "OMAP3 ISP CSI2a":1 []
		<- "tvp514x 2-005d":0 []
	pad1: Source
		-> "OMAP3 ISP CCDC output":0 []
		-> "OMAP3 ISP resizer":0 []
	pad2: Source
		-> "OMAP3 ISP preview":0 []
		-> "OMAP3 ISP AEWB":0 [ENABLED,IMMUTABLE]
		-> "OMAP3 ISP AF":0 [ENABLED,IMMUTABLE]
		-> "OMAP3 ISP histogram":0 [ENABLED,IMMUTABLE]

- entity 6: OMAP3 ISP CCDC output (1 pad, 1 link)
            type Node subtype V4L
            device node name /dev/video2
	pad0: Sink
		<- "OMAP3 ISP CCDC":1 []

- entity 7: OMAP3 ISP preview (2 pads, 4 links)
            type V4L2 subdev subtype Unknown
            device node name /dev/v4l-subdev3
	pad0: Sink
		<- "OMAP3 ISP CCDC":2 []
		<- "OMAP3 ISP preview input":0 []
	pad1: Source
		-> "OMAP3 ISP preview output":0 []
		-> "OMAP3 ISP resizer":0 []

- entity 8: OMAP3 ISP preview input (1 pad, 1 link)
            type Node subtype V4L
            device node name /dev/video3
	pad0: Source
		-> "OMAP3 ISP preview":0 []

- entity 9: OMAP3 ISP preview output (1 pad, 1 link)
            type Node subtype V4L
            device node name /dev/video4
	pad0: Sink
		<- "OMAP3 ISP preview":1 []

- entity 10: OMAP3 ISP resizer (2 pads, 4 links)
             type V4L2 subdev subtype Unknown
             device node name /dev/v4l-subdev4
	pad0: Sink
		<- "OMAP3 ISP CCDC":1 []
		<- "OMAP3 ISP preview":1 []
		<- "OMAP3 ISP resizer input":0 []
	pad1: Source
		-> "OMAP3 ISP resizer output":0 []

- entity 11: OMAP3 ISP resizer input (1 pad, 1 link)
             type Node subtype V4L
             device node name /dev/video5
	pad0: Source
		-> "OMAP3 ISP resizer":0 []

- entity 12: OMAP3 ISP resizer output (1 pad, 1 link)
             type Node subtype V4L
             device node name /dev/video6
	pad0: Sink
		<- "OMAP3 ISP resizer":1 []

- entity 13: OMAP3 ISP AEWB (1 pad, 1 link)
             type V4L2 subdev subtype Unknown
             device node name /dev/v4l-subdev5
	pad0: Sink
		<- "OMAP3 ISP CCDC":2 [ENABLED,IMMUTABLE]

- entity 14: OMAP3 ISP AF (1 pad, 1 link)
             type V4L2 subdev subtype Unknown
             device node name /dev/v4l-subdev6
	pad0: Sink
		<- "OMAP3 ISP CCDC":2 [ENABLED,IMMUTABLE]

- entity 15: OMAP3 ISP histogram (1 pad, 1 link)
             type V4L2 subdev subtype Unknown
             device node name /dev/v4l-subdev7
	pad0: Sink
		<- "OMAP3 ISP CCDC":2 [ENABLED,IMMUTABLE]

- entity 17: tvp514x 2-005d (1 pad, 1 link)
             type V4L2 subdev subtype Unknown
             device node name /dev/v4l-subdev8
	pad0: Source
		-> "OMAP3 ISP CCDC":0 []

Re: OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

[Laurent Pinchart]

Hi,

On Wednesday 07 November 2012 12:22:27 Andreas Nagel wrote:
> 
> > My code sets up the ISP pipeline, configures the format on all the
> > subdevices pads and the actual video device. Works fine so far.
> > Then I passed user pointers (aquired with malloc) to the device
> > driver for the capture buffers. Before issuing VIDIOC_STREAMON, I
> > enqueue my buffers with VIDIOC_QBUF, which fails with errno = EIO. I
> > don't know, why this is happening or where to got from here.
> > 
> >> One possibility could be that mapping the buffer to ISP MMU fails for
> >> a reason or another. Do you set the length field in the buffer?
> 
> Yes, the length was set when using userptr.
> 
> >> And am I missing something else?
> > 
> > The formats on the pads at different ends of the links in the pipeline
> > must match. In most cases, they have to be exactly the same.
> > 
> > Have you used the media-ctl test program here?
> > 
> > <URL:http://git.ideasonboard.org/media-ctl.git>
> > 
> > media-ctl -p gives you (and us) lots of information that helps figuring
> > out what could go wrong here.
> 
> All pads do indeed have the same format.
> I ran media-ctl, as you suggested. You can see the topology in the
> attached textfile.

media-ctl doesn't show pad formats, that's a bit weird. Are you using a recent 
version ?

Does the TVP5146 generate interlaced frames ?

-- 
Regards,

Laurent Pinchart

Re: OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

[Sakari Ailus]

On Thu, Nov 08, 2012 at 10:26:11AM +0100, Laurent Pinchart wrote:
> media-ctl doesn't show pad formats, that's a bit weird. Are you using a recent 
> version ?

This could as well be an issue with the kernel API --- I think that kernel
has a version which isn't in mainline. So the IOCTL used to access the media
bus formats are quite possibly different.

Regards,

-- 
Sakari Ailus
e-mail: sakari.ailus@iki.fi	XMPP: sailus@retiisi.org.uk

Re: OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

[Andreas Nagel]

Sakari Ailus schrieb am 08.11.2012 10:29:
> On Thu, Nov 08, 2012 at 10:26:11AM +0100, Laurent Pinchart wrote:
>> media-ctl doesn't show pad formats, that's a bit weird. Are you using a recent
>> version ?
> This could as well be an issue with the kernel API --- I think that kernel
> has a version which isn't in mainline. So the IOCTL used to access the media
> bus formats are quite possibly different.
>
> Regards,
>

Hi Sakari,
hi Laurent,


first, I could resolve my issues.

When I allocated buffers with the CMEM library from TI (provides aligned 
and contiguous memory buffers), I was able to use user pointers. And 
VIDIOC_STREAMON just failed because of a wrong but similar written 
pixelformat. Since yesterday, I am now able to capture frames and save 
them as YUV data in a file.

My Technexion kernel is based on the TI linux kernel and they 
(Technexion) probably backported some version of Media Controller into 
that kernel. In order to build Laurents media-ctl application, I had to 
rename all MEDIA_* constants in the source files (e.g. MEDIA_PAD_FL_SINK 
into MEDIA_PAD_FLAG_INPUT).
It's probably true, that this implementation is quite different from the 
one in mainline or other kernels. That might also be the reason, why 
media-ctl -p does not print pad formats.
But as long as Technexion keep board support for themselves, I cannot 
use another kernel. I already tried that with some mainline and 
linux-omap versions(3.2, 3.4, 3.6). They don't boot, or if they actually 
do, I don't see anything because network and tty is not available.

Lastly, the TVP5146 indeed generates interlaced frames.

All the best,
Andreas

Re: OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

[Sakari Ailus]

Hi Andreas,

On Sat, Nov 10, 2012 at 02:49:12PM +0100, Andreas Nagel wrote:
> Sakari Ailus schrieb am 08.11.2012 10:29:
> >On Thu, Nov 08, 2012 at 10:26:11AM +0100, Laurent Pinchart wrote:
> >>media-ctl doesn't show pad formats, that's a bit weird. Are you using a recent
> >>version ?
> >This could as well be an issue with the kernel API --- I think that kernel
> >has a version which isn't in mainline. So the IOCTL used to access the media
> >bus formats are quite possibly different.
> >
> >Regards,
> >
> 
> Hi Sakari,
> hi Laurent,
> 
> 
> first, I could resolve my issues.
> 
> When I allocated buffers with the CMEM library from TI (provides
> aligned and contiguous memory buffers), I was able to use user
> pointers. And VIDIOC_STREAMON just failed because of a wrong but
> similar written pixelformat. Since yesterday, I am now able to
> capture frames and save them as YUV data in a file.

Good to hear you got this resolved. Whether the memory is contiguous
shouldn't matter. The ISP has got an MMU. malloc() typically allocated only
page aligned buffer (AFAIR) but to be really sure, one can also use
posix_memalign for the purpose.

Kind regards,

-- 
Sakari Ailus
e-mail: sakari.ailus@iki.fi	XMPP: sailus@retiisi.org.uk

Re: OMAP3 ISP: VIDIOC_STREAMON and VIDIOC_QBUF calls fail

[Laurent Pinchart]

Hi Andreas,

On Saturday 10 November 2012 14:49:12 Andreas Nagel wrote:
> Sakari Ailus schrieb am 08.11.2012 10:29:
> > On Thu, Nov 08, 2012 at 10:26:11AM +0100, Laurent Pinchart wrote:
> >> media-ctl doesn't show pad formats, that's a bit weird. Are you using a
> >> recent version ?
> > 
> > This could as well be an issue with the kernel API --- I think that kernel
> > has a version which isn't in mainline. So the IOCTL used to access the
> > media bus formats are quite possibly different.
> > 
> > Regards,
> 
> Hi Sakari,
> hi Laurent,
> 
> first, I could resolve my issues.
> 
> When I allocated buffers with the CMEM library from TI (provides aligned
> and contiguous memory buffers), I was able to use user pointers. And
> VIDIOC_STREAMON just failed because of a wrong but similar written
> pixelformat. Since yesterday, I am now able to capture frames and save
> them as YUV data in a file.
> 
> My Technexion kernel is based on the TI linux kernel and they
> (Technexion) probably backported some version of Media Controller into
> that kernel. In order to build Laurents media-ctl application, I had to
> rename all MEDIA_* constants in the source files (e.g. MEDIA_PAD_FL_SINK
> into MEDIA_PAD_FLAG_INPUT).
> It's probably true, that this implementation is quite different from the
> one in mainline or other kernels. That might also be the reason, why
> media-ctl -p does not print pad formats.
> But as long as Technexion keep board support for themselves, I cannot
> use another kernel. I already tried that with some mainline and
> linux-omap versions(3.2, 3.4, 3.6). They don't boot, or if they actually
> do, I don't see anything because network and tty is not available.

You would need to port at least board code from your vendor kernel to mainline 
in order to boot a mainline kernel on your board. That's what I would advise 
you to do if time and resources permit, as the TI kernel should be avoided 
like plague. However, you mentioned CMEM above, which hints that you might be 
using TI's userspace media stack. If that's the case, good luck, I feel your 
pain, but I can't help.

> Lastly, the TVP5146 indeed generates interlaced frames.

We don't have interlaced support in the mainline OMAP3 ISP driver. I don't 
know how the driver got patched in your kernel, but I can't offert support for 
that code.

-- 
Regards,

Laurent Pinchart