https://github.com/raspberrypi/linux/pull/5941
Once CI has done the business (should be about 8:15pm GMT) then you should be able to do "sudo rpi-update pulls/5941" and install the kernel from that PR.
I have tried it, and hit the issue around the FIFO in the TC358743.
With 1080p60 it works fine. 1080p50 wraps around. Manually sending "i2ctransfer -y -f 4 w4@0x0f 0x00 0x06 0x00 0x10" to reset the FIFO trigger level to 16 makes it work. That's a little confusing as 1080p50 is using the same pixel clock as 1080p60, but with an extended horizontal front porch, so effectively giving more time to transfer each line, and the CSI2 bus is running faster than the HDMI data.
Please note that this hasn't changed with Pi5 - it's always been a little hit and miss with this driver over FIFO trigger levels. The original author only cared about 720p60 and 1080p60, so various settings are a compromise.
I do have the magic formula to compute the FIFO in a datasheet from Toshiba, but it's covered under NDAs. I may see whether we can simplify it sufficiently to make it generic, but no promises. There is a project where we want to use this chip in more earnest, so it may get done under that.
Once CI has done the business (should be about 8:15pm GMT) then you should be able to do "sudo rpi-update pulls/5941" and install the kernel from that PR.
I have tried it, and hit the issue around the FIFO in the TC358743.
With 1080p60 it works fine. 1080p50 wraps around. Manually sending "i2ctransfer -y -f 4 w4@0x0f 0x00 0x06 0x00 0x10" to reset the FIFO trigger level to 16 makes it work. That's a little confusing as 1080p50 is using the same pixel clock as 1080p60, but with an extended horizontal front porch, so effectively giving more time to transfer each line, and the CSI2 bus is running faster than the HDMI data.
Please note that this hasn't changed with Pi5 - it's always been a little hit and miss with this driver over FIFO trigger levels. The original author only cared about 720p60 and 1080p60, so various settings are a compromise.
I do have the magic formula to compute the FIFO in a datasheet from Toshiba, but it's covered under NDAs. I may see whether we can simplify it sufficiently to make it generic, but no promises. There is a project where we want to use this chip in more earnest, so it may get done under that.
Statistics: Posted by 6by9 — Fri Feb 09, 2024 7:18 pm