Enable two D457s #144
Enable two D457s #144
Conversation
|
The title should be "enable two D457s on one max9296" which could be more precise to what the series are doing. Potentially these changes could support multiple max9296s with each has 2 D457 connected, we've kept that in mind but because the HW setup we can't really test it. It's kept in mind that one D457 should work the same as before. When using two D457 on both link A/B of one max9296, because of the max9296 limitation of 4 streams, it's not possible to stream all streams of 2 D457s but the current setup is fine to make the most useful 2depth+2color and 2ir+2color work. There's a known issue to be fixed later: DFU functionality only work for Link B D457 now. |
The issue has been fixed. It was mainly a fw (firmware update counter) and hardware issue, not driver related. |
|
@xzhangxa , I'm not sure I understand - each camera can generate up to two streams simultaneously (and use two lanes) ?
|
Hi @ev-mp The quick answer is that these changes will not change the single D457 functionality, when 2 D457s are connected on both link A/B input of max9296, each D457 can still be used as before: all 4 streams streaming together, and metadata is still available. And it's possible to use 1-4 streams from both D457s. Though because of max9295/9296 4 pipe limitation, 8 streams of 2 D457s have to share the 4 pipe. As in the table, when one stream uses a pipe the other D457's corresponding stream will not be able to run (user will get an error). The stream pairs are chosen so that the most common dual-D457 usage 2depth+2color, 2ir+2color are possible. If necessary @jnzw and I could give a short introduction about all the changes, just summarize more than the PR description. |
|
Rebased onto latest Jetpack-5.0.2 branch. |
|
Rebased again onto Jetpack-5.0.2 branch. |
|
This should be rebased as we merged deb PR's |
Signed-off-by: Xin Zhang <xin.x.zhang@intel.com>
Signed-off-by: Xin Zhang <xin.x.zhang@intel.com>
* Set up GMSL links and init max9295 and max9296 settings from d4xx * Add proxy ser device and gmsl-link in DT Signed-off-by: Junze Wu <junze.wu@intel.com>
Calling the GMSL APIs only on the first sensor based on the device tree link to max9295, so for a typical 2 D457 on link A/B to the same max9296, these initial setting will be executed twice like the reference drivers. This should be considered as a temporary solution. Merging 4 sensors into one D457 device or changing GMSL API should be better. Signed-off-by: Xin Zhang <xin.x.zhang@intel.com> Signed-off-by: Junze Wu <junze.wu@intel.com>
Signed-off-by: Junze Wu <junze.wu@intel.com>
* Disable Depth, IR, IMU sensors of 1st D457 in DT * Enable second serializer (ser_b) in DT * Change dser's max-src from 1 to 2 * Add second D457 RGB sensor with metadata enabled Signed-off-by: Junze Wu <junze.wu@intel.com>
This patch is part of the original PR: IntelRealSense/realsense_mipi_platform_driver#107 Only the left cleanup is moved here, the IR metadata change in that PR is not included. Signed-off-by: Xin Zhang <xin.x.zhang@intel.com>
* Modify pipe Z to pull Depth & EMB * Modify pipe U to pull RGB & EMB * Modify DT to add two depth sensors Signed-off-by: Junze Wu <junze.wu@intel.com>
Signed-off-by: Kaili Xu <kaili.xu@intel.com>
1. Remove hardcode register settings for each pipeline, and change to use dynamic load pipe config base on the info read from device tree. 2. Change API max9295/9296_update_pipe() to max9295/9296_set_pipe(), and remove setting pipe related registers in max9295/9296_init_setting(), do it in max9295/9296_set_pipe() base on the info read from device tree. 3. Remove update pipe for IR pipe in tegra channel setting, and do it in d4xx start streaming stage. In addition, not limited to the IR pipe type, all types of pipe setting are performed here. 4. Remove param: "ir_type","ir_type_value", and probe/init check from max9295/9296, since all the pipe setting flow and needed info are maintained in d4xx. 5. Split 0x0312 from map_pipe_opt in max9296, and set this register for each pipe, in order to support dynamic switching of pipelines developed in the next step. 6. Add max9295/9296_reset_control() in ds5_remove(). Signed-off-by: Kaili Xu <kaili.xu@intel.com> Signed-off-by: Xin Zhang <xin.x.zhang@intel.com>
Signed-off-by: Junze Wu <junze.wu@intel.com>
* Add max9296_power_on/off to d4xx * Fix max9295_reset_control to reset all only if ref count is 0 Signed-off-by: Junze Wu <junze.wu@intel.com>
* Increase num-channels to 8 * Add sensor nodes with gmsl-link sections * Add vi and nvcsi nodes * Remove source node from DT * vc-id is set as follows as a workaround for conflict | 0 | 1 | 2 | 3 | | depth0 | color0 | ir0 | imu0 | | ir1 | imu1 | depth1 | color1 | Signed-off-by: Junze Wu <junze.wu@intel.com> Signed-off-by: Xin Zhang <xin.x.zhang@intel.com>
* Acquire available pipe id when streaming starts * Release video pipe when streaming is done * Allow streams with same vc-ids to share a pipe Signed-off-by: Junze Wu <junze.wu@intel.com>
Default to y for setup with maxim serdes. Signed-off-by: Xin Zhang <xin.x.zhang@intel.com>
Signed-off-by: Kaili Xu <kaili.xu@intel.com>
* Update SerDes pipe setup * Add API for max9296 one-shot reset * Do one-shot reset only when start/stop Y12I stream Signed-off-by: Junze Wu <junze.wu@intel.com>
Signed-off-by: Junze Wu <junze.wu@intel.com>
- Remove some serdes address settings since the original 9295/9296 code doesn't set them, and removing them doesn't impact D457 usage. - Still need to set pipes at init time to make it work, but reuse existing code to reduce duplicated lines. - Use existing address macros instead of hardcode values. Signed-off-by: Xin Zhang <xin.x.zhang@intel.com>
* Create new DT without serializer B and the 4 nodes of 2nd camera * Update shell scripts to apply correct DT * Updated README Signed-off-by: Junze Wu <junze.wu@intel.com>
Signed-off-by: Junze Wu <junze.wu@intel.com>
Signed-off-by: Junze Wu <junze.wu@intel.com>
* Fix Y12I resolution bug (1600x1300 -> 1280x800)(Depth xor IR,RGB xor IMU)