> The fix is to use a volatile pointer for the accesses in spinlocks. This > prevents gcc from optimizing the accesses. > > I have now run 5.6.1[78] kernels for about 4 days without a stall or any > other obvious kernel issue. It does not cleanly apply to 5.7.2, but the parts that do apply have made my C8000 run longer in the gcc testsuite than before. I will run it for a few days more to be sure, but it looks good so far. My testcase is: run several CMake nightlies every day by cron, run emerge jobs for Gentoo stabilization in a chroot. For the last few attempts it always tried to run the gcc 8.4.0 testsuite at the end, which killed the machine eventually. Eike
Attachment:
signature.asc
Description: This is a digitally signed message part.
