Note that those livelock thresholds are global, so they may mess up (slow down greatly) disk I/O and network I/O too. I won't have time until July to put together a workaround, but perhaps Sephe can extend his kern.livelock_limit_hi feature into a kern.livelock_limit_lo feature that is then flagged for the ACPI interrupt as a more permanent workaround.
-Matt On Mon, Jun 12, 2017 at 1:39 AM, Francois Tigeot <[email protected]> wrote: > Hi, > > On Wed, May 17, 2017 at 08:58:26PM +0200, Luca Franchini wrote: > > > > thank you for answering and for your work too. I'll will stay tuned for > > next updates. > > Thanks Luca! > > With the help of dillon@, I noticed the first core of my Skylake laptop > was > being pegged by acpi interrupts: > > [systat -pv 1 output] > timer ipi extint user% sys% intr% idle% smpcol > cpu0 238 11 25316 0.0 51.1 48.1 0.8 0 > > [vmstat -i output] > interrupt total rate > acpi0 89216043 23348 > > Setting a much lower treshold for misbehaving interrupts made the laptop > usable. These are the sysctls I used: > > kern.livelock_lowater=1000 > kern.livelock_limit=2000 > > It's far from perfect though and power usage is still excessive; I'll try > to learn more about the acpi subsystem in order to find out what exactly > is going wrong here. > > -- > Francois Tigeot >
