This test failure is caused by the kernel not setting dscr_inherit when the user dscr MSR is written to. Which is caused by FSCR not raising a facility unavailable interrupt. Which is ironic as one of the other patches that is backported is 993e3d96fd08c3ebf7566e43be9b8cd622063e6d ("powerpc/64s: Don't let DT CPU features set FSCR_DSCR"), which should prevent this situation.
The reason dscr_inherit_test does not fail is because it writes to the kernel MSR first. Changing dscr_inherit_exec_test to do the same, it passes. Testing 5.8 shows that it's fixed somehow, though I haven't found the exact commit that would fix this. Reverting 912c0a7f2b5daa3cbb2bc10f303981e493de73bd ("powerpc/64s: Save FSCR to init_task.thread.fscr after feature init") seems reasonable here. But failing to inherit dscr would cause cache performance regressions that are hard to justify respining the kernel for. Cascardo. ** Changed in: linux (Ubuntu Eoan) Status: Confirmed => Won't Fix -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1888332 Title: [Potential Regression] dscr_inherit_exec_test from powerpc in ubuntu_kernel_selftests failed on B/E/F To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu-kernel-tests/+bug/1888332/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs