- 08 Feb, 2023 40 commits
-
-
Hans de Goede authored
The ov2680 only supports a single format, there is no need to use a define for this. Reviewed-by:
Andy Shevchenko <andy@kernel.org> Signed-off-by:
Hans de Goede <hdegoede@redhat.com> Signed-off-by:
Mauro Carvalho Chehab <mchehab@kernel.org>
-
Hans de Goede authored
Remove a bunch of unused data-types and defines from ov2680.h. Reviewed-by:
-
Hans de Goede authored
3 fixes for the framesize list: 1. Drop modes < 640x480, these are made by significant cropping, leading to such a small remainig field-of-view that they are not really usable 2. 1616x1082 is presumably intended to be 1600x1080 + 16 pixels padding in both dimensions, but the height is wrong. Change this to 1616x1096. 3. The 800x600 mode is missing the 16 pixels padding and 720x592 is missing 16 pixels padding in its width and the 720x576 base mode is a mode with non square pixels, while the sensor has square pixels. Replace both with 768x576 + 16 pixels padding -> 784x592 Reviewed-by:
-
Hans de Goede authored
Drop struct ov2680_resolution and the ov2680_res_preview[] array, this is now only used in ov2680_enum_frame_size() and only the width + height are used there. Replace this with a new struct v4l2_frmsize_discrete ov2680_frame_sizes[] array. No functional changes. Reviewed-by:
-
Hans de Goede authored
Since we now calculate timings baded on the desired width and height, any width and height are valid as long as they don't exceed the sensor's dimensions. Drop the v4l2_find_nearest_size() call and instead clamp the requested width and height. Reviewed-by:
-
Hans de Goede authored
Fix and simplify ov2680_enum_frame_interval(), the index is not an index into ov2680_res_preview[], so using N_PREVIEW is wrong. Instead it is an index indexing the different framerates for the resolution specified in fie->width, fie->height. Since the ov2680 code only supports a single fixed 30 fps, index must always be 0 and we don't need to check the other fie input values. Reviewed-by:
-
Hans de Goede authored
The res member of struct ov2680_device isn't read anywhere anymore, drop it. Reviewed-by:
-
Hans de Goede authored
The fps, lines-per-frame and skip-frames values are the same for all resolutions, use defines for these. The ov2680_res_preview[] incorrectly sets fps to 60 for some low-res modes, this is incorrect with the current fixed (resolution independent) lines-per-frame value. Note this not drop the now no longer used fps, lines-per-frame and skip-frames struct ov2680_resolution members. The entire struct is going away in the next patches so that would just cause unnecessary changes. Reviewed-by:
-
Hans de Goede authored
Instead of using a long fixed register settings list for each resolution, calculate the register settings based on the requested width + height. This will allow future enhancements like adding hblank and vblank controls and adding selection support. Signed-off-by:
-
Hans de Goede authored
By default the ov2680 automatically sets the window to match the outputsize and automatically adjusts it to keep the bayer pattern stable when enabling hflip/vflip. This does not work for the 1616x1216 mode because there is no room to adjust the window there. To make flipping work in the 1616 wide modes the register lists for those modes set bit 0 of 0x5708 (manual_win_en) to 1 and ov2680_set_bayer_order() updates the bayer-order on the pad to match. But ov2680_set_bayer_order() is always called, so when enabling flipping on modes with a width of less then 1616 now results in the wrong bayer order being reported on the pad since the sensor is auto-adjusting the window in this case. Specify the correct (== output-size) window-size in all resolutions register-list and always set the manual_win_en bit, so that the bayer order is changed on hflip/vflip enable on all resolutions. Signed-off-by:
-
Hans de Goede authored
Add a test pattern control. This is a 1:1 copy of the test pattern control in the main drivers/media/i2c/ov2680.c driver. Reviewed-by:
-
Hans de Goede authored
Add exposure and gain controls. This allows controlling the exposure and gain through standard v4l2 IOCTLs. Note the register defines for the exposure and gain registers are renamed to match the datasheet. Reviewed-by:
-
Hans de Goede authored
Exposure and gain control should use standard v4l2 controls, not a custom ioctl. The next patch in this series will re-add support as standard controls, this is split into 2 patches for easier reviewing. Reviewed-by:
-
Hans de Goede authored
Rework the flip ctrls to be more like those of mainline (non staging) drivers. This is modelled after the main ov2680 and ov5693 drivers. This also introduces __ov2680_get_pad_format() to make the ov2680 code more compliant with the mainline v4l2-subdev APIs. Note the OV2680_FLIP_REG and OV2680_MIRROR_REG defines are renamed to OV2680_REG_FORMAT1 and OV2680_REG_FORMAT2 to match the datasheet. Signed-off-by:
-
Hans de Goede authored
Use the new ov_16bit_addr_reg_helpers.h instead of duplicating the ovxxxx sensor I2C register access helpers found in many different sensor drivers. Signed-off-by:
-
Hans de Goede authored
The following drivers under drivers/media/i2c: ov08x40.c, ov13858.c, ov13b10.c, ov2680.c, ov2685.c, ov2740.c, ov4689.c, ov5670.c, ov5675.c, ov5695.c, ov8856.c, ov9282.c and ov9734.c, as well as various "atomisp" sensor drivers in drivers/staging, *all* use register access helpers with the following function prototypes: int ovxxxx_read_reg(struct ovxxxx_dev *sensor, u16 reg, unsigned int len, u32 *val); int ovxxxx_write_reg(struct ovxxxx_dev *sensor, u16 reg, unsigned int len, u32 val); To read/write registers on Omnivision OVxxxx image sensors wich expect a 16 bit register address in big-endian format and which have 1-3 byte wide registers, in big-endian format (for the higher width registers). Add a new ov_16bit_addr_reg_helpers.h header file with static inline versions of these register access helpers, so that this code duplication can be removed. Signed-off-by: -
Hans de Goede authored
Don't use MEDIA_ENT_F_V4L2_SUBDEV_UNKNOWN to initialize the function field of various media-entity links. This fixes the following warnings showing up in dmesg: atomisp-isp2 0000:00:03.0: Entity type for entity ATOM ISP CSI2-port0 was not initialized! atomisp-isp2 0000:00:03.0: Entity type for entity ATOM ISP CSI2-port1 was not initialized! atomisp-isp2 0000:00:03.0: Entity type for entity ATOM ISP CSI2-port2 was not initialized! atomisp-isp2 0000:00:03.0: Entity type for entity tpg_subdev was not initialized! atomisp-isp2 0000:00:03.0: Entity type for entity ATOMISP_SUBDEV_0 was not initialized! atomisp-isp2 0000:00:03.0: Entity type for entity ATOMISP_SUBDEV_1 was not initialized! Signed-off-by:
-
Hans de Goede authored
Looking at csi2_link_setup(), this function can never work, it does a switch-case like this: switch (local->index | is_media_entity_v4l2_subdev(remote->entity)) with cases like this: case ATOMISP_SUBDEV_PAD_SOURCE | MEDIA_ENT_F_V4L2_SUBDEV_UNKNOWN where ATOMISP_SUBDEV_PAD_SOURCE matches an index (0-1) and MEDIA_ENT_F_V4L2_SUBDEV_UNKNOWN is 0x00020000, but is_media_entity_v4l2_subdev(remote->entity) does not return MEDIA_ENT_F_* values, it return a bool, so 0 or 1 which means that non of the cases can ever match the input value. Looking at the rest of the function all it ever does (if it would actually hit one of the cases) is set the atomisp_mipi_csi2_device struct's output member. And checking the rest of the atomisp code that member is never read. Also userspace does not actually setup media-controller links when using the atomisp /dev/video$ nodes since all the links are fixed. So csi2_link_setup() never runs. Remove the unnecessary and broken csi2_link_setup() function and also remove the unused atomisp_mipi_csi2_device struct's output member. Signed-off-by:
-
Hans de Goede authored
Looking at isp_subdev_link_setup(), this function can never work, it does a switch-case like this: switch (local->index | is_media_entity_v4l2_subdev(remote->entity)) with cases like this: case ATOMISP_SUBDEV_PAD_SINK | MEDIA_ENT_F_V4L2_SUBDEV_UNKNOWN where ATOMISP_SUBDEV_PAD_SINK matches an index (0-4) and MEDIA_ENT_F_V4L2_SUBDEV_UNKNOWN is 0x00020000, but is_media_entity_v4l2_subdev(remote->entity) does not return MEDIA_ENT_F_* values, it return a bool, so 0 or 1 which means that non of the cases can ever match the input value. Looking at the rest of the function all it ever does (if it would actually hit one of the cases) is set the atomisp_sub_device struct's input member. And checking the rest of the atomisp code that member is never read. Also userspace does not actually setup media-controller links when using the atomisp /dev/video$ nodes since all the links are fixed. So isp_subdev_link_setup() never runs. Remove the unnecessary and broken isp_subdev_link_setup() function and also remove the unused atomisp_sub_device struct's input member. Reviewed-by:
-
Hans de Goede authored
The device core will call ACPI to turn the device (i2c_client) for a sensor on / put it in D0 before calling its probe() method. This takes a reference on all of the ACPI power-resources belonging to the device. Since the atomisp_gmin_platform code disables ACPI power-resource management and does its own pm, this reference never gets released. This is a problem for ACPI power-resources which are shared with other devices since those now never get turned off again (nor back on again). Explicitly put the device in D3 before disabling the ACPI power-resource management. Note that atomisp_register_i2c_module() runs near the end of the sensor driver's probe() function, after the driver is done with probing the hw. So the power-resouces (the same resources as directly controlled by the atomisp platform code) getting turned off (a second time, as they are already off) is not a problem. Reviewed-by:
-
Hans de Goede authored
atomisp_set_fmt() first does: v4l2_fill_mbus_format(&vformat.format, ...); vformat.format.height += padding_h; vformat.format.width += padding_w; ret = v4l2_subdev_call(isp->inputs[asd->input_curr].camera, pad, set_fmt, NULL, &vformat); if (ret) return ret; f->fmt.pix.width = vformat.format.width - padding_w; f->fmt.pix.height = vformat.format.height - padding_h; this happens with the original padding w/h = 16 values and then later on it calls: ret = atomisp_set_fmt_to_snr(vdev, &s_fmt, f->fmt.pix.pixelformat, padding_w, padding_h, dvs_env_w, dvs_env_h); Which repeats the above structure. If at that point padding w/h are changed to 12 then it will now request a different output-size of the sensor driver. The sensor drivers so far have actually been ignoring this since they use v4l2_find_nearest_size() on a fixed resolution list and the nearest resolution will be the one from the earlier calls where padding w/h was 16. But there really is no reason for sensor drivers to use a fixed resolution list. They make lower resolutions using cropping so they can make any resolution as long as width/height are even numbers. Dropping the fixed-resolution list limit from sensors on BYT results in trying to start streaming failing because the resolution set to the sensor now no longer matches with the resolution used during the initial part of the configuration done by atomisp_set_fmt(). Drop the BYT specific overriding of the padding_w/h to 12, so that the padding in the first and second s_fmt calls made to the sensor matches, to fix stream start failing when the fixed resolution list is dropped. Reviewed-by: -
Hans de Goede authored
ffmt is a local variable pointing to a substruct of another local variable which really just makes the code harder to read / follow, so drop it. Signed-off-by:
-
Hans de Goede authored
The DSDT of all Windows BYT / CHT devices which I have seen has proper ACPI powermagement for the clk and regulators used by the sensors. So there is no need for the whole custom atomisp_gmin custom code to disable the ACPI pm and directly poke at the PMIC for this. Add new atomisp_register_sensor_no_gmin() + atomisp_unregister_subdev() helpers which allow registering a sensor with the atomisp code without using any of the atomisp_gmin power-management code. Reviewed-by:
-
Hans de Goede authored
atomisp_gmin_find_subdev() can be used to lookup a subdev given its i2c-adapter + i2c-client-address. But the only caller of it reads this from the intel_v4l2_subdev_table struct and that same struct already contains a pointer to the v4l2_subdev. So this function is not necessary, drop it and modify its only caller to directly take the subdev from the intel_v4l2_subdev_table struct. Also drop struct intel_v4l2_subdev_i2c_board_info since that now no longer is used. Reviewed-by:
-
Hans de Goede authored
The s_power callback for v4l2-subdevs has been deprecated, allow sensor drivers without a s_power callback to work by ignoring the -ENOIOCTLCMD return value. Reviewed-by:
-
Hans de Goede authored
The atomisp driver creates 8 /dev/video# device nodes. 4 nodes (preview / video / viewfinder / capture) for each of 2 possible streams aka atomisp-sub-device-s (asd-s). Both streams start with asd->input_curr set to 0 (to the first sensor), opening + releasing a file-handle on one of the nodes of an asd, while streaming from the other asd causes the sensor to get turned off, leading to the stream failing. The atomisp-code already tracks which asd "owns" a specific sensor, use this to only turn the sensor off if it is owned by the asd. Reviewed-by:
-
Hans de Goede authored
The memory for all of struct atomisp_video_pipe is kzalloc()-ed in atomisp_subdev_init() so there is no need to memset parts of struct atomisp_video_pipe to 0. Reviewed-by:
-
Hans de Goede authored
atomisp_init_pipe() does 3 things: 1. Init a bunch of list-heads / locks 2. Init the vb_queue for the videodev (aka pipe) 3. zero the per-frame parameters related variables of the pipe 1. and 2. really should not be done at file-open time, but once at probe. Currently the code is getting away with doing this on every videodev-open because only 1 open is allowed at a time. 1. is already done at probe time by atomisp_init_subdev_pipe(), move 2. to atomisp_init_subdev_pipe() so that it is also done once at probe. As for 3. The per-frame parameters can only be set from a qbuf ioctl, which can only happen after a reqbufs ioctl and atomisp_buf_cleanup already zeros the per-frame parameters when the buffers are released, so 3. is not necessary at all. Reviewed-by:
-
Hans de Goede authored
Make atomisp behave like any other drivers and have it load the firmware at probe time (as it was already doing by default). The deferred firmware loading support needlessly complicates the v4l2_file_operations.open callback (atomisp_open()), getting in the way of allowing multiple opens like a normal v4l2 device would. Reviewed-by:
-
Hans de Goede authored
Remove the no longer used bin_factor_x, bin_factor_y and bin_mode members from the resolution info inside various atomisp camera sensor drivers. Reviewed-by:
-
Hans de Goede authored
The bin-factor-x and bin-factor-y ctrls are only used internally to get a single value to pass to atomisp_css_input_set_binning_factor(), which is supposed to tune the lens-shading correction for the binning factor. But all sensor drivers return either 0 or 1 for this, with 0 meaning unset and 1 meaning no-binning. Even though some modes do actually do binning ... Also note that the removed atomisp_get_sensor_bin_factor() would fall back to 0 if either the x and y factor differ or if the ctrls are not implemented (not all sensor drivers implement them). Simply always pass 0 to atomisp_css_input_set_binning_factor(). This is part of a patch-series which tries to remove atomisp specific / custom code from the sensor drivers, with as end goal to make the atomisp drivers regular camera sensor drivers. Reviewed-by:
-
Hans de Goede authored
This ioctl returns a number of fixed sensor parameters + a number of mode-specific parameters. With libcamera these fixed parameters are instead stored in a table with sensor-name to parameters mappings (camera_sensor_properties.cpp); and the variable parameters can be derived from the set fmt. So this custom ioctl is not necessary; and it currently has no users. Remove the ioctl and all the sensor drivers xxxx_get_intg_factor() helpers which return this info. This is part of a patch-series which tries to remove atomisp specific / custom code from the sensor drivers, with as end goal to make the atomisp drivers regular camera sensor drivers. Reviewed-by:
-
Hans de Goede authored
This ioctl simply returns a couple of fixed sensor parameters. With libcamera these fixed parameters are instead stored in a table with sensor-name to parameters mappings (camera_sensor_properties.cpp), so this custom ioctl is not necessary; and it currently has no users. Remove the ioctl and also remove the custom v4l2-ctrls underpinning the ioctl. This is part of a patch-series which tries to remove atomisp specific / custom code from the sensor drivers, with as end goal to make the atomisp drivers regular camera sensor drivers. Reviewed-by:
-
Hans de Goede authored
On BYT on poweron/runtime-resume the code is doing: 1. Do nothing 2. msleep(10) 3. Start actual poweron sequence Since the runtime resume can happen at any moment, waiting 10ms after it does not really make any sense. According to both the comment and to: https://github.com/intel/ProductionKernelQuilts/blob/master/uefi/cht-m1stable/patches/cam-0341-atomisp-WA-sleep-10ms-when-power-up-ISP-on-byt.patch Which is the patch which originally added this this was added as a workaround for a single test failing on a single model tablet/laptop. So lets just drop this. Reviewed-by:
-
Hans de Goede authored
punit_ddr_dvfs_enable() is only used on CHT devices and there dmesg gets filled with: "DDR DVFS, door bell is not cleared within 3ms" messages, so clearly the doorbell checking is not working. This check was added by: https://github.com/intel/ProductionKernelQuilts/blob/master/uefi/cht-m1stable/patches/cam-0340-atomisp-add-door-bell-for-ddr-dvfs-on-cht.patch Which commit message says: "PUNIT interface added to check Req_ACK of freq status". This suggests that the doorbell mechanism may only be available with certain PUNIT fw versions and it seems that many CHT devices do not have this fw version; that or the doorbell mechanism is not working for other reasons. Revert cam-0340-atomisp-add-door-bell-for-ddr-dvfs-on-cht.patch, replacing the doorbell check with a msleep(20) this fixes dmesg getting filled with error messages. Signed-off-by:
-
Hans de Goede authored
These are clearly debug messages, printing these all the time is not useful. Silence these by simply removing them altogether. Reviewed-by:
-
Hans de Goede authored
The atomisp does not use standard PCI power-management through the PCI config space. Instead this driver directly tells the P-Unit to disable the ISP over the IOSF. The standard PCI subsystem pm_ops will try to access the config space before (resume) / after (suspend) this driver has turned the ISP on / off, resulting in the following errors: Unable to change power state from D0 to D3hot, device inaccessible Unable to change power state from D3cold to D0, device inaccessible Getting logged into dmesg a whole bunch of time during boot as well as every time the camera is used. To avoid these errors use a custom pm_domain instead of standard driver pm-callbacks so that all the PCI subsys suspend / resume handling is skipped and call pci_save_state() / pci_restore_state() ourselves. Signed-off-by:
-
Hans de Goede authored
Remove the atomisp_sw_contex struct, it has only 1 member: running_freq, instead store running_freq directly. While at it also change running_freq from an int to an unsigned int, all values stored in it are unsigned and it is compared to the also unsigned new_freq variable. Reviewed-by:
-
Hans de Goede authored
The Crystal Cove PMIC used on some BYT/CHT devices has different revisions when paired with Bay Trail (BYT) vs Cherry Trail (CHT) SoCs. The current hardcoded values are only valid for CHT devices, change the code so that it uses the correct register values on both BYT and CHT. Reviewed-by:
-
Hans de Goede authored
Check buffer index is in range inside atomisp_qbuf_wrapper() before using it do index pipe->frame_request_config_id[]. Signed-off-by:
-
