[PATCH][RFC] Add mt9p031 sensor support.

This RFC includes a power management implementation that causes
the sensor to show images with horizontal artifacts (usually
monochrome lines that appear on the image randomly).

Signed-off-by: Javier Martin <javier.martin@vista-silicon.com>

Hi Javier,

Thanks for the patch. Here's a review of the power handling code.

This RFC includes a power management implementation that causes
the sensor to show images with horizontal artifacts (usually
monochrome lines that appear on the image randomly).

Signed-off-by: Javier Martin <javier.martin@vista-silicon.com>

[snip]

diff --git a/drivers/media/video/mt9p031.c b/drivers/media/video/mt9p031.c
new file mode 100644
index 0000000..04d8812
--- /dev/null
+++ b/drivers/media/video/mt9p031.c

[snip]

@@ -0,0 +1,841 @@
+/*
+ * Driver for MT9P031 CMOS Image Sensor from Aptina
+ *
+ *
+ *
+ * Based on the MT9V032 driver and Bastian Hecht's code.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/delay.h>
+#include <linux/device.h>
+#include <linux/i2c.h>
+#include <linux/log2.h>
+#include <linux/pm.h>
+#include <linux/regulator/consumer.h>
+#include <linux/slab.h>
+#include <media/v4l2-subdev.h>
+#include <linux/videodev2.h>
+
+#include <media/mt9p031.h>
+#include <media/v4l2-chip-ident.h>

This header is not needed anymore.

+#include <media/v4l2-subdev.h>
+#include <media/v4l2-device.h>
+
+#define MT9P031_PIXCLK_FREQ 54000000
+
+/* mt9p031 selected register addresses */
+#define MT9P031_CHIP_VERSION 0x00
+#define MT9P031_CHIP_VERSION_VALUE 0x1801
+#define MT9P031_ROW_START 0x01
+#define MT9P031_ROW_START_DEF 54
+#define MT9P031_COLUMN_START 0x02
+#define MT9P031_COLUMN_START_DEF 16
+#define MT9P031_WINDOW_HEIGHT 0x03
+#define MT9P031_WINDOW_WIDTH 0x04
+#define MT9P031_H_BLANKING 0x05
+#define MT9P031_H_BLANKING_VALUE 0
+#define MT9P031_V_BLANKING 0x06
+#define MT9P031_V_BLANKING_VALUE 25
+#define MT9P031_OUTPUT_CONTROL 0x07
+#define MT9P031_OUTPUT_CONTROL_CEN 2
+#define MT9P031_OUTPUT_CONTROL_SYN 1
+#define MT9P031_SHUTTER_WIDTH_UPPER 0x08
+#define MT9P031_SHUTTER_WIDTH 0x09
+#define MT9P031_PIXEL_CLOCK_CONTROL 0x0a
+#define MT9P031_FRAME_RESTART 0x0b
+#define MT9P031_SHUTTER_DELAY 0x0c
+#define MT9P031_RST 0x0d
+#define MT9P031_RST_ENABLE 1
+#define MT9P031_RST_DISABLE 0
+#define MT9P031_READ_MODE_1 0x1e
+#define MT9P031_READ_MODE_2 0x20
+#define MT9P031_READ_MODE_2_ROW_MIR 0x8000
+#define MT9P031_READ_MODE_2_COL_MIR 0x4000
+#define MT9P031_ROW_ADDRESS_MODE 0x22
+#define MT9P031_COLUMN_ADDRESS_MODE 0x23
+#define MT9P031_GLOBAL_GAIN 0x35
+
+#define MT9P031_WINDOW_HEIGHT_MAX 1944
+#define MT9P031_WINDOW_WIDTH_MAX 2592
+#define MT9P031_WINDOW_HEIGHT_MIN 2
+#define MT9P031_WINDOW_WIDTH_MIN 18

Can you move those 4 constants right below MT9P031_WINDOW_HEIGHT and
MT9P031_WINDOW_WIDTH ? The max values are not correct, according to the
datasheet they should be 2005 and 2751. You can define *_DEF constants for the
default width and height.

+struct mt9p031 {
+ struct v4l2_subdev subdev;
+ struct media_pad pad;
+ struct v4l2_rect rect; /* Sensor window */
+ struct v4l2_mbus_framefmt format;
+ struct mt9p031_platform_data *pdata;
+ struct mutex power_lock; /* lock to protect power_count */
+ int power_count;
+ u16 xskip;
+ u16 yskip;
+ /* cache register values */
+ u16 output_control;
+ u16 h_blanking;
+ u16 v_blanking;
+ u16 column_address_mode;
+ u16 row_address_mode;
+ u16 column_start;
+ u16 row_start;
+ u16 window_width;
+ u16 window_height;
+ struct regulator *reg_1v8;
+ struct regulator *reg_2v8;
+};

[snip]

+static int restore_registers(struct i2c_client *client)
+{
+ int ret;
+ struct mt9p031 *mt9p031 = to_mt9p031(client);
+
+ /* Disable register update, reconfigure atomically */
+ ret = mt9p031_set_output_control(mt9p031, 0,
+ MT9P031_OUTPUT_CONTROL_SYN);
+ if (ret < 0)
+ return ret;
+
+ /* Blanking and start values - default... */
+ ret = reg_write(client, MT9P031_H_BLANKING, mt9p031->h_blanking);
+ if (ret < 0)
+ return ret;
+
+ ret = reg_write(client, MT9P031_V_BLANKING, mt9p031->v_blanking);
+ if (ret < 0)
+ return ret;
+
+ ret = reg_write(client, MT9P031_COLUMN_ADDRESS_MODE,
+ mt9p031->column_address_mode);
+ if (ret < 0)
+ return ret;
+
+ ret = reg_write(client, MT9P031_ROW_ADDRESS_MODE,
+ mt9p031->row_address_mode);
+ if (ret < 0)
+ return ret;
+
+ ret = reg_write(client, MT9P031_COLUMN_START,
+ mt9p031->column_start);
+ if (ret < 0)
+ return ret;
+
+ ret = reg_write(client, MT9P031_ROW_START,
+ mt9p031->row_start);
+ if (ret < 0)
+ return ret;
+
+ ret = reg_write(client, MT9P031_WINDOW_WIDTH,
+ mt9p031->window_width);
+ if (ret < 0)
+ return ret;
+
+ ret = reg_write(client, MT9P031_WINDOW_HEIGHT,
+ mt9p031->window_height);
+ if (ret < 0)
+ return ret;

All those registers will be written to in mt9p031_s_stream(), there's no need
to restore them when powering the chip up. You can remove this function
completely, as well as the register cache.

+ /* Re-enable register update, commit all changes */
+ ret = mt9p031_set_output_control(mt9p031,
+ MT9P031_OUTPUT_CONTROL_SYN, 0);
+ if (ret < 0)
+ return ret;
+ return 0;
+}

[snip]

+static int mt9p031_power_on(struct mt9p031 *mt9p031)
+{
+ struct i2c_client *client = v4l2_get_subdevdata(&mt9p031->subdev);
+ int ret;
+
+ /* Ensure RESET_BAR is low */
+ if (mt9p031->pdata->reset)
+ mt9p031->pdata->reset(&mt9p031->subdev, 1);
+ /* turn on digital supply first */
+ ret = regulator_enable(mt9p031->reg_1v8);
+ if (ret) {
+ dev_err(&client->dev,
+ "Failed to enable 1.8v regulator: %d\n", ret);
+ goto err_1v8;

You can return ret immediately.

+ }

According to the datasheet (see figure 31) you need a delay of minimum 1ms
here. regulator_enable() probably doesn't wait for the power to stabilize
before returning (correct me if I'm wrong), so a slightly longer delay would
be safer.

+ /* now turn on analog supply */
+ ret = regulator_enable(mt9p031->reg_2v8);
+ if (ret) {
+ dev_err(&client->dev,
+ "Failed to enable 2.8v regulator: %d\n", ret);
+ goto err_rst;
+ }

+ /* Now RESET_BAR must be high */
+ if (mt9p031->pdata->reset)

Similarly, you need to wait for the 2.8V power supply to stabilize before de-
asserting the reset signal. A short delay is probably required. You can put it
inside the 'if', as the delay is not needed if the reset signal can't be
controlled.

+ mt9p031->pdata->reset(&mt9p031->subdev, 0);
+
+ if (mt9p031->pdata->set_xclk)
+ mt9p031->pdata->set_xclk(&mt9p031->subdev, MT9P031_PIXCLK_FREQ);
+
+ /* soft reset */
+ ret = mt9p031_reset(client);
+ if (ret < 0) {
+ dev_err(&client->dev, "Failed to reset the camera\n");
+ goto err_rst;
+ }
+
+ ret = restore_registers(client);
+ if (ret < 0) {
+ dev_err(&client->dev, "Failed to restore registers\n");
+ goto err_rst;
+ }
+
+ return 0;
+err_rst:
+ regulator_disable(mt9p031->reg_1v8);

You should disable both regulators here, and only the 1.8V regulator if
enabling the 2.8V regulator fails.

+err_1v8:
+ return ret;
+
+}

[snip]

+static int mt9p031_set_power(struct v4l2_subdev *sd, int on)
+{
+ struct mt9p031 *mt9p031 = container_of(sd, struct mt9p031, subdev);
+ int ret = 0;
+
+ mutex_lock(&mt9p031->power_lock);
+
+ /*
+ * If the power count is modified from 0 to != 0 or from != 0 to 0,
+ * update the power state.
+ */
+ if (mt9p031->power_count == !on) {
+ if (on) {
+ ret = mt9p031_power_on(mt9p031);
+ if (ret) {
+ dev_err(mt9p031->subdev.v4l2_dev->dev,
+ "Failed to enable 2.8v regulator: %d\n", ret);

This message isn't correct anymore, as mt9p031_power_on() now handles two
regulators. As mt9p031_power_on() already prints an error message in the case
of failure, you can remove this message.

+ goto out;
+ }
+ } else {
+ mt9p031_power_off(mt9p031);
+ }
+ }
+
+ /* Update the power count. */
+ mt9p031->power_count += on ? 1 : -1;
+ WARN_ON(mt9p031->power_count < 0);
+
+out:
+ mutex_unlock(&mt9p031->power_lock);
+ return ret;
+}
+
+static int mt9p031_registered(struct v4l2_subdev *sd)
+{
+ struct mt9p031 *mt9p031 = container_of(sd, struct mt9p031, subdev);
+ struct i2c_client *client = v4l2_get_subdevdata(&mt9p031->subdev);
+ int ret;
+
+ ret = mt9p031_set_power(&mt9p031->subdev, 1);
+ if (ret) {
+ dev_err(&client->dev,
+ "Failed to power on device: %d\n", ret);
+ goto err_pwron;

You can return ret directly.

+ }
+
+ ret = mt9p031_video_probe(client);
+ if (ret)
+ goto err_evprobe;
+

Code from here

+ mt9p031->pad.flags = MEDIA_PAD_FL_SOURCE;
+ ret = media_entity_init(&mt9p031->subdev.entity, 1, &mt9p031->pad, 0);
+ if (ret)
+ goto err_evprobe;
+
+ mt9p031->subdev.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;

to here can be moved to the probe() function, it will simplify
mt9p031_registered(). The function will become so small that you can directly
inline mt9p031_video_probe() inside it.

Hi,
thank you for the review, I agree with you on all the suggested
changes except on this one:

Hi Javier,

OK, that sounds reasonable. However, that would include black pixels
that are located at the beginning of the array (0,0) to (16, 54),
which means that users would have to specify a crop value of (15,54)
to eliminate those initial black level pixels. Which seems quite
unnatural to me.
Another option could be setting (16,54) as default values and allowing
to introduce negative cropping values. Is this possible?
And finally, the most sensible idea IMHO could be not letting the user
to see pixels from (0,0) to (15,54) (setting 15,54 as minimum and
default )and, for black level compensation, ending pixels could be
used (2608,1998) to (2751, 2003).

What do you think?

> Hi Javier,
>
>> Hi,
>> thank you for the review, I agree with you on all the suggested
>> changes except on this one:
>>
>> >> This RFC includes a power management implementation that causes
>> >> the sensor to show images with horizontal artifacts (usually
>> >> monochrome lines that appear on the image randomly).
>> >>
>> >> Signed-off-by: Javier Martin <javier.martin@vista-silicon.com>
>> >
>> > [snip]
>> >
>> >> diff --git a/drivers/media/video/mt9p031.c
>> >> b/drivers/media/video/mt9p031.c new file mode 100644
>> >> index 0000000..04d8812
>> >> --- /dev/null
>> >> +++ b/drivers/media/video/mt9p031.c
>> >
>> > [snip]
>> >
>> >> +#define MT9P031_WINDOW_HEIGHT_MAX 1944
>> >> +#define MT9P031_WINDOW_WIDTH_MAX 2592
>> >> +#define MT9P031_WINDOW_HEIGHT_MIN 2
>> >> +#define MT9P031_WINDOW_WIDTH_MIN 18
>> >
>> > Can you move those 4 constants right below MT9P031_WINDOW_HEIGHT and
>> > MT9P031_WINDOW_WIDTH ? The max values are not correct, according to the
>> > datasheet they should be 2005 and 2751.
>>
>> In figure 4, it says active image size is 2592 x 1944
>> Why should I include active boundary and dark pixels?
>
> Users might want to get the dark pixels for black level compensation purpose.
> As the chip allows for that, it should be supported. The default should of
> course be the active area of 2592 x 1944 pixels.
>

OK, that sounds reasonable. However, that would include black pixels
that are located at the beginning of the array (0,0) to (16, 54),
which means that users would have to specify a crop value of (15,54)
to eliminate those initial black level pixels. Which seems quite
unnatural to me.
Another option could be setting (16,54) as default values and allowing
to introduce negative cropping values. Is this possible?
And finally, the most sensible idea IMHO could be not letting the user
to see pixels from (0,0) to (15,54) (setting 15,54 as minimum and
default )and, for black level compensation, ending pixels could be
used (2608,1998) to (2751, 2003).

No, you set your crop bounds to (0,0)-... but your default rectangle to
(15,54)-... and that's also what you set if noone issues an S_CROP.

Thanks
Guennadi