10:21 <milkylainen> Question. If I want to move the "hardware" directory containing the dts trees from the Tegra kernel tree, what includes do I have to change to fix the path?

10:22 <milkylainen> I found the scripts/Kbuild.include, but I'm not making much sense from it.

10:27 <milkylainen> An instruction how to build a single dtb would also be nice.

10:53 <cyndis> why do you want to move it? the build system should automatically build the DTBs from hardware/ while building the kernel

10:56 <cyndis> Kbuild.include or arch/arm64/boot/dts/Makefile are probably good places to hack around if you really want to change the path

10:59 <milkylainen> cyndis: Yup. But couldn't make it work.

11:00 <milkylainen> cyndis: Kernel should be pretty self contained. Make dtbs or a single dtb should not be a chore, nor require some build environment?

11:01 <cyndis> well, technically DTS's are not a part of the kernel, I'd say it's more a historical quirk that in upstream that is the case

11:02 <cyndis> downstream there are other users of DTS's as well in places where the linux kernel is not present so it makes some sense for them to be separate

11:03 <milkylainen> That's fine, but if I move the directory, for whatever build requirement reasons, how do I get it to build? I'm tinkering with the same Makefile and Kbuild.include you mentioned. So so far so good.. But to no avail so far.

11:05 <pgwipeout[m]> You could always just run dtc standalone without the build env. I've never tried with preprocessing requirements though.

11:06 <cyndis> you could try changing NV_BUILD_KERNEL_DTS_ROOT in Makefile.kernel

11:06 <milkylainen> cyndis: did, didn't work. I suspect some trickery with ../../ goes haywire if the path is not correct.

11:07 <cyndis> yeah, may be. other paths have probably never been tested

11:07 <milkylainen> pgwipeout[m]: I could. But a chore finding out where everything is located. It's easier just repointing the top level.. :)

11:08 <milkylainen> cyndis: mm. :\

11:14 <milkylainen> It seems it finds the files, but some translation for rules get broken. I get a lot of:

11:14 <milkylainen> arch/arm64/boot/dts/Makefile:132: target 'arch/arm64/boot/dts/_ddot_' given more than once in the same rule

11:21 <milkylainen> Hrrm. Can't make it work even at the expected directory offset?

11:21 <milkylainen> make V=1 ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- dtbs should be fine in a pristine unpacked tree? right? defconfig done obviously.

11:36 <pgwipeout[m]> Honestly I don't even think you need the cross compiler to make the dtbs

11:40 <milkylainen> Probably right.

11:41 <milkylainen> Either way, couldn't make a pristine tree build dtbs. I get the same error as when moving the hardware tree.

11:46 <milkylainen> pgwipeout[m]: Maybe need the cross cc? Tried with the host cc.. gcc: error: unrecognized command-line option ‘-mlittle-endian’

11:46 <milkylainen> Kernel tree seems to build an empty.o? Empty kernel module? Or something?

11:48 <pgwipeout[m]> Odd. Oh well.

11:49 <pgwipeout[m]> Yeah the path to build the dtbs from a pristine tree should just be running make defconfig (whichever one is yours) then make dtbs.

11:49 <pgwipeout[m]> Kernel only builds the dtbs for the configured arch and manufacturer(in case of arm64)

11:49 <pgwipeout[m]> s/Kernel only builds the dtbs for the configured arch and manufacturer(in case of arm64)/Kernel only builds the dtbs for the configured arch and manufacturer/

11:56 <milkylainen> https://forums.developer.nvidia.com/t/failed-to-make-l4t-kernel-dts/116399/5

11:56 <milkylainen> E =

11:56 <milkylainen> the-space = $E $E

11:57 <milkylainen> works now.

11:57 <milkylainen> Maybe should have mentioned that this was 32.4.4 and not the latest one.

11:57 <milkylainen> My bad.

11:58 <milkylainen> (32.4.4 is not my choice).

11:58 <milkylainen> pgwipeout[m]: -^

11:59 <pgwipeout[m]> Oh, this is the l4t kernel...

11:59 <pgwipeout[m]> Sorry, I spend my life in mainline. L4t is an absolute nightmare.

12:00 <milkylainen> pgwipeout[m]: heh. Well. Customer choice. I don't know how mainline supports Xavier AGX?

12:00 <pgwipeout[m]> If they have broken dtb builds...

12:00 <milkylainen> I'd love to use a vanilla. Like really... :)

12:02 <pgwipeout[m]> Surprised mainline doesn't have support for the AGX yet.

12:02 <pgwipeout[m]> But then again mainline pretty much neutered the Jetson Nano's GPU.

12:09 <milkylainen> I agree about the mainline part. But unfortunately a lot of customers do not view platforms as a long term comittment.

12:09 <milkylainen> "Oh. Working kernel? Whatever. Done, check, next"

12:21 <cyndis> mainline does have decent support for xavier, but not the gpu yet

12:26 <cyndis> the L4T UEFI setup (https://developer.download.nvidia.com/embedded/L4T/UEFI_Readme.html) even uses the mainline kernel. Of course it is currently experimental and has limited hardware support

12:27 <milkylainen> cyndis: and the gpu can't be added as a separate driver blob?

12:27 <milkylainen> cyndis: nice.

12:28 <cyndis> not at this time

12:28 <milkylainen> :\ . Nice with a vanilla UEFI setup anyway.

12:34 <cyndis> well, if you add a PCI-E gpu, that will probably work :) but the "desktop driver" doesn't support the iGPU

12:36 <milkylainen> :)

12:42 <milkylainen> Ok. So next peculiar behavior. Shouldn't I be able to build a single dtb by specifying its name?

12:43 <milkylainen> Right now, the "dtbs" target work fine. And running a specific <dtbname>.dtb after dtbs result in "nothing be done for", ie already made. But running it before results in a b0rk.

12:44 <milkylainen> I am guessing the copy rule doesn't get run unless the "dtbs" target is called.