Le jeu. 9 juil. 2020 à 13:26, Romain Dolbeau <[email protected]> a écrit : > So - has anyone made any progress on this or are we still in need of a > bisect? If the latest, is there any known way to quickly cause a crash > to ensure if a tested kernel is good/bad?
I wanted to give a go at bisecting, so first I recovered a 4.17 package in the archive on my T5120 to have something that should work as backup - and in so doing, removed the Debian kernels I had... maybe that was a mistake, as I could have been running 5.6 at the time, I'm not sure. After failing to deliberately crash my home-cross-compiled vanilla 5.2 on the Sun Blade 2500 Red, I installed the current kernel in Sid: linux-image-5.7.0-1-sparc64-smp 5.7.6-1 And managed to do a full rebuild of GCC 10.1 (starting with recent binutils/gmp/mpfr/mpc/isl before a complete 3-stage bootstrap), and in parallel do a git checkout of Linux, some package installation and a configure/rebuild of ZFS. Took almost a day (with a shutdown/reboot the middle of stage 2), no crash in sight, though I tried via SSH only (the XVR-600 isn't supported in X). The machine has been rock-solid so far (running from a SAS drive on a flashed 1068)... For those with crashes - could you try the current kernel and see if it fixes the problem? And if it doesn't, what kind of workload do you have when the kernel crashes? I've seen the crashes myself but can't reproduce them anymore and I don't have the archive of the 5.6 I might have been running at the time... Cordially, -- Romain Dolbeau
