17:09 <camaronut> Hi all.  I've been googling and debugging an issue with Dunfell builds.  When using a machine with over 16 threads, I occasionally see failures building the openssl rpm.  The failure occurs in nativesdk-rpm's code.  rpm's implementation of cpio gets unhappy when trying to write the trailer on the first openssl package.  I added some debugging

17:09 <camaronut> to rpm (build/pack.c and lib/cpio.c) and see that the failure is when the code compares cpio->fileend != cpio->offset.  fileend is zero, and offset is -1.

17:11 <camaronut> curious if anyone else has seen similar issues.  My gut is telling me this is some sort of stomp, but I have no proof.  I see that the OpenMP parallelization changed upstream compared to the version in Dunfell.

17:11 <RP> rburton: didn't you see this?

17:13 <camaronut> Devil's advocate regarding OpenMP: this happens on the first package that's written, which is outside the "#pragma omp parallel" block.

17:45 <RP> camaronut: https://git.yoctoproject.org/poky/commit/?id=3c5e126401c8bcd4d7759c28cc644f18f7eb65b7 may have been the fix in master

17:59 <camaronut> RP: interesting.  Not sure if limiting threads would solve the issue or force the issue underground.  :)   We've seen this failure in a 72 thread machine with 64GB RAM, and another machine with (I think) 16 threads and 32GB RAM.

18:02 <camaronut> RP: another observation: dmesg does not show any errors on either machine.  No OOM killer is invoked.

19:04 <camaronut> Another bit: I'm able to reproduce the openssl rpm problem by running bitbake -c cleanall openssl && bitbake openssl in a loop.  The last time I reproed it I hit it in 2 iterations.

21:07 <Konsgn> hmm, what is best practice for modifying u-boot dts files? I did my devtool modify, devtool menuconfig, but what should i be doing to add knowledge of a tpm2.x chip to uboot?