| Message ID | 20261102100533.30106-1-info@humanlearning.ch |
|---|---|
| State | New |
| Headers | show |
| Series |
|
| Related | show |
Hi, On 2-Nov-26 11:05, François Roux wrote: > Without an entry for ov02c10, the software ISP used by the simple > pipeline handler has no gain model or black level for this sensor and > logs: > > IPASoft: Failed to create camera sensor helper for ov02c10 > > AGC then still runs, but on raw analogue-gain register values instead > of a linear gain, which makes it converge poorly. > > The kernel driver (drivers/media/i2c/ov02c10.c) advertises analogue-gain > register values in the range 0x10..0xf8 with a default of 0x10, and > writes them to the 16-bit register at 0x3508 as (val << 4). That > register is graded in 1/256 steps, so both the control range and the > register encoding agree on 1/16 steps: the default of 0x10 is unity gain > and the maximum of 0xf8 is 15.5x. Without a datasheet, this was checked > by measurement on a second machine (Lenovo Yoga Slim 7x, X1E80100): at a > fixed exposure, the signal above the pedestal follows code / 16 to > within 1.5% from 0x18 to 0xf8 (1.49 at 0x18, 4.03 at 0x40, 8.11 at 0x80, > 15.32 at 0xf8). > > The sensor outputs 10-bit Bayer data (MEDIA_BUS_FMT_SGRBG10_1X10). The > driver programs a black level target of 0x40 in the BLC registers > 0x4002/0x4003, which scales to 4096 at the 16-bit width expected by > blackLevel(), as for ov01a10, ov08d10, ov2740 and ov5675, which use > the same 0x40 target at 10 bits. A > dark-frame measurement matches it: the mean pedestal is 64.2 LSB, and > changing the target to 0x80 while streaming moves it to 128.2 LSB on all > four colour channels from the next frame. > > Tested on the front camera of a Microsoft Surface Pro 12 (Qualcomm > X1P42100) driven by the qcom CAMSS pipeline. Before the change the IPA > reported "Exposure 4-2320, gain 16-248 (1)"; after it reports > "Exposure 4-2320, gain 1-15.5 (0.145)". > > Acked-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com> > Tested-by: Birk Skyum <birk.skyum@pm.me> > Tested-by: Christian Cortner <ccortner@gmail.com> > Signed-off-by: François Roux <info@humanlearning.ch> Thanks, patch looks good to me: Reviewed-by: Hans de Goede <johannes.goede@oss.qualcomm.com> Regards, Hans > --- > src/ipa/libipa/camera_sensor_helper.cpp | 12 ++++++++++++ > 1 file changed, 12 insertions(+) > > > base-commit: df8e348d64b4932a0676256ed24f0da3ee9438bc > > diff --git a/src/ipa/libipa/camera_sensor_helper.cpp b/src/ipa/libipa/camera_sensor_helper.cpp > index 359040ec..72f974ee 100644 > --- a/src/ipa/libipa/camera_sensor_helper.cpp > +++ b/src/ipa/libipa/camera_sensor_helper.cpp > @@ -713,6 +713,18 @@ public: > }; > REGISTER_CAMERA_SENSOR_HELPER("ov01a10", CameraSensorHelperOv01a10) > > +class CameraSensorHelperOv02c10 : public CameraSensorHelper > +{ > +public: > + CameraSensorHelperOv02c10() > + { > + /* From Linux kernel driver: 0x40 at 10bits. */ > + blackLevel_ = 4096; > + gain_ = AnalogueGainLinear{ 1, 0, 0, 16 }; > + } > +}; > +REGISTER_CAMERA_SENSOR_HELPER("ov02c10", CameraSensorHelperOv02c10) > + > class CameraSensorHelperOv08d10 : public CameraSensorHelper > { > public:
Quoting François Roux (2026-11-02 10:05:33) > Without an entry for ov02c10, the software ISP used by the simple > pipeline handler has no gain model or black level for this sensor and > logs: > > IPASoft: Failed to create camera sensor helper for ov02c10 > > AGC then still runs, but on raw analogue-gain register values instead > of a linear gain, which makes it converge poorly. > > The kernel driver (drivers/media/i2c/ov02c10.c) advertises analogue-gain > register values in the range 0x10..0xf8 with a default of 0x10, and > writes them to the 16-bit register at 0x3508 as (val << 4). That > register is graded in 1/256 steps, so both the control range and the > register encoding agree on 1/16 steps: the default of 0x10 is unity gain Why are we constraining away the extra steps? Is the driver wrong ? Should it be offering the full granularity ? > and the maximum of 0xf8 is 15.5x. Without a datasheet, this was checked > by measurement on a second machine (Lenovo Yoga Slim 7x, X1E80100): at a > fixed exposure, the signal above the pedestal follows code / 16 to > within 1.5% from 0x18 to 0xf8 (1.49 at 0x18, 4.03 at 0x40, 8.11 at 0x80, > 15.32 at 0xf8). > > The sensor outputs 10-bit Bayer data (MEDIA_BUS_FMT_SGRBG10_1X10). The > driver programs a black level target of 0x40 in the BLC registers > 0x4002/0x4003, which scales to 4096 at the 16-bit width expected by > blackLevel(), as for ov01a10, ov08d10, ov2740 and ov5675, which use > the same 0x40 target at 10 bits. A I don't think we need to list every other sensor to say "The driver configures a black level pedestal of 0x40 at 10 bit". > dark-frame measurement matches it: the mean pedestal is 64.2 LSB, and > changing the target to 0x80 while streaming moves it to 128.2 LSB on all > four colour channels from the next frame. > > Tested on the front camera of a Microsoft Surface Pro 12 (Qualcomm > X1P42100) driven by the qcom CAMSS pipeline. Before the change the IPA > reported "Exposure 4-2320, gain 16-248 (1)"; after it reports > "Exposure 4-2320, gain 1-15.5 (0.145)". > > Acked-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com> > Tested-by: Birk Skyum <birk.skyum@pm.me> > Tested-by: Christian Cortner <ccortner@gmail.com> > Signed-off-by: François Roux <info@humanlearning.ch> > --- > src/ipa/libipa/camera_sensor_helper.cpp | 12 ++++++++++++ > 1 file changed, 12 insertions(+) > > diff --git a/src/ipa/libipa/camera_sensor_helper.cpp b/src/ipa/libipa/camera_sensor_helper.cpp > index 359040ec..72f974ee 100644 > --- a/src/ipa/libipa/camera_sensor_helper.cpp > +++ b/src/ipa/libipa/camera_sensor_helper.cpp > @@ -713,6 +713,18 @@ public: > }; > REGISTER_CAMERA_SENSOR_HELPER("ov01a10", CameraSensorHelperOv01a10) > > +class CameraSensorHelperOv02c10 : public CameraSensorHelper > +{ > +public: > + CameraSensorHelperOv02c10() > + { > + /* From Linux kernel driver: 0x40 at 10bits. */ > + blackLevel_ = 4096; > + gain_ = AnalogueGainLinear{ 1, 0, 0, 16 }; > + } > +}; > +REGISTER_CAMERA_SENSOR_HELPER("ov02c10", CameraSensorHelperOv02c10) > + > class CameraSensorHelperOv08d10 : public CameraSensorHelper > { > public: > > base-commit: df8e348d64b4932a0676256ed24f0da3ee9438bc > -- > 2.56.0 >
diff --git a/src/ipa/libipa/camera_sensor_helper.cpp b/src/ipa/libipa/camera_sensor_helper.cpp index 359040ec..72f974ee 100644 --- a/src/ipa/libipa/camera_sensor_helper.cpp +++ b/src/ipa/libipa/camera_sensor_helper.cpp @@ -713,6 +713,18 @@ public: }; REGISTER_CAMERA_SENSOR_HELPER("ov01a10", CameraSensorHelperOv01a10) +class CameraSensorHelperOv02c10 : public CameraSensorHelper +{ +public: + CameraSensorHelperOv02c10() + { + /* From Linux kernel driver: 0x40 at 10bits. */ + blackLevel_ = 4096; + gain_ = AnalogueGainLinear{ 1, 0, 0, 16 }; + } +}; +REGISTER_CAMERA_SENSOR_HELPER("ov02c10", CameraSensorHelperOv02c10) + class CameraSensorHelperOv08d10 : public CameraSensorHelper { public: