It actually happened on my customer kernel too now, after over 11h uptime. Magic sysrq is enabled; will try to gather more details.