On Wed, Mar 2, 2016, at 11:51, Stuart Bennett wrote: > On 27/02/16 22:36, Henrique de Moraes Holschuh wrote: > > In the thermald bug report, you will see lots of posts asking for the values > > of several MSRs, and also to run "turbostat" and attach its output. > > > > Can you repeat those tests, and report the value of those MSRs in both the > > "working" and "system is stuck in low power" states? > > I've done the tests under three scenarios:
... > turbostat output attached -- the only major differences I can spot are > in MSR_CORE_PERF_LIMIT_REASONS and PKG_% (indicating major package > throttling). ... > Clearly the only substantial change here is in MSR 0x198 following > degradation. This is exactly what happens when thermald messes with these Xeons, as long as your MSR 0x1b0 is also reading "7". So, could you check the output of "rdmsr 0x1b0" ? After you do the MSR 0x1b0 tests, please rename away msr.ko from your system (e.g. to msr.ko.disabled), so that it becomes utterly impossible for userspace to change the MSRs (well, except through the perf path, but one thing at a time). Once msr.ko is disabled, please reboot to the new microcode. Does the system still regress in the new microcode, even without the msr.ko kernel driver ever being loaded? -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique de Moraes Holschuh <h...@debian.org>
