Hello, Since a month or so I've noticed that my only text box that supports P-states would seem to eventually get all its CPUs stuck in the lower frequency P-state, and no amount of load would result in a P-state change.
The issue has always been intermittent, and not easy to reproduce. Looking at the on-demand governor I've found the two issues fixed on this series. With those fixed I haven't been able to reproduce the issue, but as said I don't have a deterministic way to trigger it. Anyway, the fixes look legit, so regardless of whether there's something else causing my issue, we should take those. I'm a bit surprised no-one has noticed wonky performance with the on-demand governor before; the changes that introduce those issue have been there since the governor was introduced. Neither I know why I've only noticed the low frequency stickiness issue ~last month, if it was indeed caused by the issues here. Thanks, Roger. Roger Pau Monne (2): xen/cpufreq: fix adjusting of sampling window on early exit xen/cpufreq: fix usages of align_timer() in the on-demand governor xen/drivers/cpufreq/cpufreq_ondemand.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) -- 2.51.0
