10:51 <hexdump0815> just to answer my question regarding erasing the legacy uboot on a s905w tv box: "store init 3" (as described in the khadas vim doc) did not work - afterwards the uboot was still there, but "mmc erase boot" worked and erased the uboot

10:53 <hexdump0815> afterwards the box was trying to boot from the sdcard and boots fine with the legacy sd uboot extracted from the stock firmware from that tv box - but more interesting: it also boots relatively fine using a khadas vim uboot (taken from the balbes tv box images)

10:55 <hexdump0815> i only say relatively fine as it boots about 1 out of about 10 tries - in those cases it fails to get the scpi clocks and runs fine afterwards - in the other cases it seems to get the scpi clocks and gets a kernel panic shortly after - i assume this is because it gets the too high s905x clocks from the khadas bl31 (or was it bl301)

10:56 <hexdump0815> so most probably one needs to buid a uboot using a proper s905w bl31 (or bl301) - so the question is: is such a blob somewhere available for the s905w?

10:57 <hexdump0815> repk: is it maybe possible to extract that blob from an existing uboot - not sure because of that complex building process of the bootloader from the compiled uboot and all the different blxy blobs

11:05 <xdarklight> hexdump0815: RE S805 MXQ: I send the patch to improve support yesterday. you are CC'ed on it. can you please reply with your Tested-by (that's the only missing bit before Kevin will queue it for v5.3)?

11:13 <hexdump0815> xdarklight: thanks a lot - will test it most probably tomorrow

11:17 <xdarklight> awesome, thank you! :)

17:11 <xdarklight> narmstrong: I have tested a PWM calculation similar to what ukleinek suggested. the only difference I could find was Khadas VIM's PWM LED - good news: the new algorithm is more precise (with the old calculation the output was -59500ns off the requested, the new is off by +2500ns)

17:12 <xdarklight> all other applications (32.768kHz wifi clock, VCCK on the 32-bit SoCs) are using a pre_div of 0 (means 1) and the calculated values are 100% identical there

17:13 <xdarklight> footnote: I measured this with my Saleae Logic clone at 24MHz sampling rate

17:13 <narmstrong> xdarklight: thanks for testing :-)

17:46 <xdarklight> narmstrong: you're welcome :)

17:47 <xdarklight> btw, do you have any more ideas (for example where to start debugging) why I'm not getting any CVBS output with modetest (or in other words: why I only get output when writing garbage to /dev/fb0)?

17:56 <narmstrong> Weird because fb0 is a drm client, so using modetest should actually do the same, maybe on the first setup sequence works and the reinit doesn’t

17:56 <narmstrong> You could try disabling fb console totally and only use modetest -s

18:15 <xdarklight> rebuilding with: # CONFIG_FRAMEBUFFER_CONSOLE is not set

18:17 <xdarklight> still nothing :(

18:18 <xdarklight> (cat /dev/urandom > /dev/fb0 still "works")