On Thu, 2020-03-26 at 16:04 +0800, Zhang Rui wrote: > On Wed, 2020-03-25 at 23:59 -0700, Kenneth R. Crudup wrote: > > On Thu, 26 Mar 2020, Zhang Rui wrote: > > > > > According to the dmesg below, I don't think it is the psys domain > > > that > > > is missing. > > > > Yeah, take another look (and I've re-attached the dmesg excerpts). > > I > > will > > always get the domains "pp0-core", "package" and "pp1-gpu" and > > "fixed > > counters" reports 3, but every now and then I get an additional > > domain > > "psys" (and a "fixed counters" report of 4). > > yeah, you're right. > what does "rdmsr 0x64D" return when there are 3 fixed counters? > if it returns 0, which I think it is the case, does reading the msr > from other cpus return non-zero value, by using rdmsr -p parameter? > > > > > > One possibility is that you're switching between different > > > kernels, > > > and > > > one kernel is built with CONFIG_PROC_THERMAL_MMIO_RAPL set, and > > > one > > > is > > > not. can you make a double check? > > > > Nah, that's not the case. I only pull forward from Linus' tree and > > have > > always included that config option. > > > > What is relevant is I *think* (and can verify later) that I get the > > "psys" domain on a cold boot from power off, but not if I reboot. > > yeah. it is good to compare the "rdmsr 0x64d" output in both cases. BTW, it is better to get the output of "rdmsr 0x65C" as well. thanks, rui > > tthanks, > rui > > > > > if that is not the problem, please attach the output of "grep . > > > /sys/class/powercap/intel-rapl*/name" in both cases. > > > > OK, I'll do that in the morning and get back to you. > > > > -Kenny > >
