* ----------------------- SANDBOX ACCESS VIOLATION SUMMARY ----------------------- * LOG FILE: "/var/tmp/portage/sys-libs/libomp-16.0.0/temp/sandbox.log" * VERSION 1.0 FORMAT: F - Function called FORMAT: S - Access Status FORMAT: P - Path as passed to function FORMAT: A - Absolute Path (not canonical) FORMAT: R - Canonical Path FORMAT: C - Command Line F: open_wr S: deny P: /dev/kfd A: /dev/kfd R: /dev/kfd C: /usr/lib/llvm/16/bin/amdgpu-arch * -------------------------------------------------------------------------------- Reproducible: Always
Created attachment 858441 [details] build.log
Created attachment 858443 [details] emerge --info
It seems adding `addpredict /dev/kfd` to `multilib_src_configure()` is enough to build successfully.
Do you have libcuda or libveo in /usr/lib*? I'm wondering why I don't get this error on my system.
(In reply to Michał Górny from comment #4) > Do you have libcuda or libveo in /usr/lib*? I'm wondering why I don't get > this error on my system. I'm sure libcuda is something Nvidia specific, probably libveo as well. I don't have either of them. This is about AMD cards, not Nvidia, the ebuild already contains `addpredict /dev/nvidiactl`.
Oh, I see that it's used by amdgpu-arch now. I guess it doesn't happen for me because I'm testing in nspawn container and there's no /dev/kfd there. Lemme try on a regular system.
The bug has been closed via the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=c928da1854b76d608773485d2416fdc4663b3635 commit c928da1854b76d608773485d2416fdc4663b3635 Author: Michał Górny <mgorny@gentoo.org> AuthorDate: 2023-03-21 12:57:51 +0000 Commit: Michał Górny <mgorny@gentoo.org> CommitDate: 2023-03-21 15:05:19 +0000 sys-libs/libomp: Fix trying to access AMD GPUs Prevent the CMake script from starting amdgpu-arch tool that accesses the GPU in order to determine AMDGPU-based tests can be run. We don't want the test suite to access the GPU, so just hard-disable that. Closes: https://bugs.gentoo.org/902449 Signed-off-by: Michał Górny <mgorny@gentoo.org> sys-libs/libomp/libomp-16.0.0.9999.ebuild | 2 ++ sys-libs/libomp/libomp-16.0.0.ebuild | 2 ++ sys-libs/libomp/libomp-17.0.0.9999.ebuild | 2 ++ sys-libs/libomp/libomp-17.0.0_pre20230314.ebuild | 2 ++ 4 files changed, 8 insertions(+)
(In reply to Michał Górny from comment #6) > Oh, I see that it's used by amdgpu-arch now. I guess it doesn't happen for > me because I'm testing in nspawn container and there's no /dev/kfd there. > Lemme try on a regular system. Dear Sam, I'm sorry to interrupt, but I also have an AMD arch and don't have cuda installed and run into this bug. I see, that the patch sets AMD arch to NOTFOUND. I'm contemplating on it, whether it is the best way providing optimal outcome/function to handle this and not instructing the ebuild to disregard access to the kfd dev, like it was proposed before? Thanks: Dw.