On 03/11/2020 10:19, Jacopo Mondi wrote: > Hi Tomi, > thanks for testing > > On Tue, Nov 03, 2020 at 09:19:17AM +0200, Tomi Valkeinen wrote: >> Hi Jacopo, >> >> On 29/10/2020 00:57, Jacopo Mondi wrote: >>> Hi Hugues Tomi and Sam >>> >>> this small series collects Tomi's patch on adjusting htot which has been >>> floating around for some time with a rework of the clock tree based on >>> Hugues' and Sam's work on setting pclk_period. It also address the need to >>> suppport LINK_FREQUENCY control as pointed out by Hugues. >>> >>> I'm sort of happy with the result as I've removed quite some chrun and the clock >>> tree calculation is more linear. All modes work except full-resolution which a >>> bit annoys me, as I can't select it through s_fmt (to be honest I have not >>> investigated that in detail, that's why an RFC). >>> >>> Framerate is better than before, but still off for some combinations: >>> 640x480@30 gives me ~40 FPS, 1920x1080@15 gives me ~7. >>> The other combinations I've tested looks good. >>> >>> Can I have your opinion on these changes and if they help you with your >>> platforms? >>> >>> I've only been able to test YUYV, support for formats with != bpp will need >>> some work most probably, but that was like this before (although iirc Hugues >>> has captured JPEG, right ?) >>> >>> There's a bit more cleanup on top to be done (I've left TODOs around) and >>> probably the HBLANK calculation should be checked to see if it works with the >>> new htot values. >> >> Unfortunately the second patch seems to break capture on AM6 EVM + OV5640. The effect is pretty odd. >> The picture is very dark, with odd vertical lines, but it's still capturing something as I can see a >> correctly shaped shadow of my hand if I wave my hand over the sensor. > > This saddens me quite a lot :( The current clock tree programming > procedure is horrid and it has been bugging me for 2 years now :( > > Is capture broken in all modes, or have you tested a single one ? I tested 640x480, 720x480, 720x576. I have only this sensor to test the CSI RX on AM6 EVM, so I would not be surprised if there are issues in the CSI RX driver (too). But this is super frustrating to debug, as the sensor is a badly documented black box, and I don't have means to probe the CSI lines... Tomi -- Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
