gistfile1.txt

I used ftrace to try and find where the problem is, and have tracked it to spi-bcm2835.

Pi4 spends 19.907 ms sending the pixel data and Pi zero spends 29.990 ms.

trace-cmd is available on the 'pi' image.

Pi4
---

# trace-cmd record -o /tmp/trace.dat -p function_graph -l mipi_dbi_fb_dirty -l spi_transfer_one_message
# trace-cmd report -i /tmp/trace.dat

     kworker/0:2-348   [000]   577.767036: funcgraph_entry:                     |  mipi_dbi_fb_dirty() {
     kworker/0:2-348   [000]   577.770096: funcgraph_entry:      + 27.518 us    |    spi_transfer_one_message();
     kworker/0:2-348   [000]   577.770145: funcgraph_entry:      + 28.889 us    |    spi_transfer_one_message();
     kworker/0:2-348   [000]   577.770198: funcgraph_entry:      + 26.500 us    |    spi_transfer_one_message();
     kworker/0:2-348   [000]   577.770247: funcgraph_entry:      + 28.944 us    |    spi_transfer_one_message();
     kworker/0:2-348   [000]   577.770299: funcgraph_entry:      + 26.055 us    |    spi_transfer_one_message();
     kworker/0:2-348   [000]   577.770378: funcgraph_entry:      # 19907.074 us |    spi_transfer_one_message();
     kworker/0:2-348   [000]   577.790293: funcgraph_exit:       # 23258.000 us |  }

The SPI hw can only handle buffers which length fits in a 16-bit register DLEN.
The driver works around this by splitting up the transfer.

# trace-cmd record -o /tmp/trace.dat -p function_graph -l mipi_dbi_fb_dirty -l spi_transfer_one_message -l bcm2835_spi_transfer_one

     kworker/0:2-348   [000]   674.462743: funcgraph_entry:                     |  mipi_dbi_fb_dirty() {
...
     kworker/0:2-348   [000]   674.466270: funcgraph_entry:                     |    spi_transfer_one_message() {
     kworker/0:2-348   [000]   674.466283: funcgraph_entry:      + 21.352 us    |      bcm2835_spi_transfer_one();
     kworker/0:2-348   [000]   674.474748: funcgraph_entry:      + 22.500 us    |      bcm2835_spi_transfer_one();
     kworker/0:2-348   [000]   674.483216: funcgraph_entry:      + 21.963 us    |      bcm2835_spi_transfer_one();
     kworker/0:2-348   [000]   674.486192: funcgraph_exit:       # 19922.037 us |    }
     kworker/0:2-348   [000]   674.486220: funcgraph_exit:       # 23477.908 us |  }


Pi Zero
-------

# trace-cmd record -o /tmp/trace.dat -p function_graph -l mipi_dbi_fb_dirty -l spi_transfer_one_message
# trace-cmd report -i /tmp/trace.dat

     kworker/0:3-59    [000]   261.795885: funcgraph_entry:                     |  mipi_dbi_fb_dirty() {
     kworker/0:3-59    [000]   261.804197: funcgraph_entry:      ! 112.000 us   |    spi_transfer_one_message();
     kworker/0:3-59    [000]   261.804387: funcgraph_entry:      + 54.000 us    |    spi_transfer_one_message();
     kworker/0:3-59    [000]   261.804508: funcgraph_entry:      + 47.000 us    |    spi_transfer_one_message();
     kworker/0:3-59    [000]   261.804615: funcgraph_entry:      + 49.000 us    |    spi_transfer_one_message();
     kworker/0:3-59    [000]   261.804723: funcgraph_entry:      + 42.000 us    |    spi_transfer_one_message();
     kworker/0:3-59    [000]   261.804918: funcgraph_entry:      # 29990.000 us |    spi_transfer_one_message();
     kworker/0:3-59    [000]   261.834974: funcgraph_exit:       # 39099.000 us |  }