08:48 <Kwiboo> detlevc: great, I had some issues last time around with hevc on rkvdec1, the iommu flush did not fully solve the issue and some reset was still required, my fluster result where irregular when trying to run multiple parallel decoding streams

08:50 <Kwiboo> hoping to cycle back to video decoding next few days, re-spinning my old rkvdec1 h264 442/hi10 and hevc series and ffmpeg v4l2-request, just one or two u-boot series I want to finish up before that ;-)

09:50 <naoki> I almost forgot what I did last month...

09:50 <naoki> I think I need to send some patches.

13:05 <qschulz> Kwiboo: dsimic: shoot, forgot to Cc you on the v3 of the DTSO migration for Rockchip boards in the kernel

13:07 <qschulz> https://lore.kernel.org/linux-rockchip/20250128-pre-ict-jaguar-v3-0-7be2f09d390a@cherry.de/T/#u (wait a bit for all the mails to make it to the archive :) )

13:32 <CounterPillow> Does anyone have Part2 of the RV1106 TRM, if that even exists? I can only find Part1. I'm specifically looking for the TRNG section of that SoC

13:40 <mriesch> qschulz: there is rk3588-edgeble-neu6a-wifi.dtb and rk3588-edgeble-neu6a-wifi.dtbs <- that's correct right? or is there a stray s

13:41 <mriesch> probably correct, as it is applied consistently. also, i didn't check the makefile magic behind it

13:47 <qschulz> mriesch: you misread (but I did misread myself the first time :) )

13:47 <qschulz> there's rk3588-edgeble-neu6a-wifi.dtb and rk3588-edgeble-neu6a-wifi-dtbs

13:48 <mriesch> so there is. right.

13:48 <mriesch> :-)

13:48 <qschulz> If you did read the comment I added in the v3 in the Makefile, did it help or make things worse?

13:48 <qschulz> comprehension-wise

13:48 <qschulz> I also find it confusing but it seems to work and what we have in ti/ too

13:49 <mriesch> it's all good, maybe i should switch to a larger font size

13:50 <qschulz> mriesch: I had another thought also.. maybe we should prefix those "dummy" DTBs with overlay-test maybe>?

13:51 <qschulz> mmind00: ^ maybe also

13:51 <qschulz> because it does actually generate those .dtb files

13:51 <mriesch> qschulz: no strong opinion there

13:52 <mmind00> qschulz: I would just go with what the other arches do ;-)

13:57 <qschulz> mmind00: how boring :)

13:58 <qschulz> they do the same as in my patch wrt naming. Some mix dtb rules with dtbo application dtbs, some have it separate like in my patch

13:58 <qschulz> s/rules/targets/

14:00 <mmind00> qschulz: at least the Makefile positions I like very much ... especially with that nice introductory comment :-)

14:04 <qschulz> mmind00: thanks, i'll wait a few days before sending a v4 for the nitpicks from mriesch

14:05 <mmind00> qschulz: sounds good ... -rc1 is still till the weekend away :-)

14:05 <mmind00> and there is of course Fosdem

14:25 <detlevc> Kwiboo: Nice, I have a branch there if you want to toy with it a bit https://gitlab.collabora.com/detlev/linux/-/commits/add-rkvdec2-driver-iommu

23:19 <macromorgan> I figured out Optee, and think I might have a few bug fixes in my future... long story short I figured out how to get it to automatically add the memory holes so you don't have to...

23:20 <macromorgan> long story short for the rk3588 we need to 1) increase the CFG_DTB_MAX_SIZE in Optee to match A-TF (which is 0x20000). I'm not sure if this is a good thing or a bad thing though. We also need to tell Optee CFG_AUTO_MAX_PA_BITS=y CFG_CORE_LARGE_PHYS_ADDR=y.

23:21 <macromorgan> 2) just build A-TF as usual but with SPD=opteed

23:22 <macromorgan> 3) in U-Boot, you need to set CONFIG_OPTEE_LIB=y, CONFIG_OPTEE_TZDRAM_SIZE=0x02000000, and unset CONFIG_SPL_ATF_NO_PLATFORM_PARAM. You also need to change kernel_comp_addr_r so that it doesn't stomp all over Optee (I changed it to 0xa400000) which is just after the reserved memory for Optee.

23:23 <macromorgan> 4) remove the Optee device from the Linux device tree under firmware. U-Boot will re-add this and when it does also add the reserved memory nodes. If the optee node is present, however, U-Boot will just skip over it.

23:41 <Kwiboo> macromorgan: for optee itself you should probably change to use CFG_TZDRAM and CFG_SHMEM similar to rk3399 and px30 instead of the vendor addresses that are interfering with u-boot load addresses

23:43 <macromorgan> not sure I understand, those values are already set aren't they?

23:44 <Kwiboo> https://github.com/OP-TEE/optee_os/blob/master/core/arch/arm/plat-rockchip/conf.mk#L65-L68 <<-- better if these matches what rk3399 and px30 use a few lines above

23:45 <Kwiboo> also suggest you change CFG_DTB_MAX_SIZE to something more than 128 KiB, if DT symbols are enabled for the DT the rk3588 is more than 128 KiB

23:46 <macromorgan> we'd need to change it to match everywhere though right? I know A-TF has the max size at 128KB for now, Optee at 64KB, and not sure if I need to check U-Boot or anywhere else

23:46 <macromorgan> I mean, I'm down with such a change, it's just I don't know if it would work or not for everyone

23:49 <macromorgan> I'll do it to 640k... that ought to be enough for anybody :-p

23:55 <Kwiboo> hehe :-), if you are sending a patch to change it, 128k is probably not big enough once rk3588*.dtsi is fully featured with more dt nodes, and a patch for tf-a is also needed