After resuming from S3, kernel schedules load on only one NUMA node. I am using 3.18.3, but remember the problem existed roughly since 3.18. Reproducible: Always Steps to Reproduce: 1. If you don't have NUMA machine, just compile kernel with CONFIG_NUMA_EMU and boot it with numa=fake=2, for example 2. Suspend to RAM and resume 3. Run some process you know will be multithreaded (e.g. cd /usr/src/linux && make -j) Actual Results: Using top/htop/whatever, you can see that CPUs from one NUMA node (e.g. 0 on my system, not sure about others). Expected Results: All cores/threads are under load.
Could you participate in the upstream bug? It looks like the symptoms described there are similar to yours.
(In reply to Mike Pagano from comment #1) > Could you participate in the upstream bug? It looks like the symptoms > described there are similar to yours. Oh, definitely. I wasn't sure it applies to vanilla kernel as well.
We'll watch the upstream bug and backport any patches we can.