https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=292615
--- Comment #10 from Olivier Certner <[email protected]> --- (In reply to Martin Cracauer from comment #8) That's exactly what we are trying to discover here. (In reply to Olivier Certner from comment #9) Following-up on feedback. * For Martin: The output strangely shows that the content of the CPPC_CAPABILITY_1 register is 0, which seems wrong, as if it has not been initialized (contrary to what the AMD's APM says). We are using some of the values in there (so, all 0s in this case) to then program CPPC_REQUEST (but programming it with these 0s shouldn't be a problem, again according to the APM). Here's what we first could do (other ideas welcome). Going to produce a patch that, when CPPC_CAPABILITY_1 is not initialized, does not use its values to set CPPC_REQUEST, which I'll ask you to test. Probably going to add more debug knobs in a second step. Some problem I'm facing is that, if CPPC_CAPABILITY_1 is not initialized despite CPPC having being enabled via CPPC_ENABLE, I was considering declaring that CPPC support is broken, but that is too late at this point, since we are not supposed to be able to fallback to regular P-states once CPPC has been enabled (according to the APM, it cannot be disabled). But I need to enable it and then read CPPC_CAPABILITY_1 to see whether it is initialized correctly (it is supposed to be 0 before CPPC is enabled). * For Cy: I'm hearing you, but have a hard time finding anything you reported really problematic. That CPU temperature can reach 100°C and then cause frequency throttling during buildworld isn't at all surprising to me. This is, e.g., what I've been observing on a few Intel laptops of mine. Maybe your Framework does not handle it gracefully, but then that looks like a platform problem more than anything else. You shouldn't be doing what you're doing with powerd and regular P-states in the first place, irrespective of CPPC support. So, at the moment, I don't think the default value of the new knob should be changed. * More generally: I'm also starting to wonder if we may not be doing it wrong with that AMD-specific implementation that short-circuits ACPI. Even if that is what we also have been doing with Intel processors. I'll try to look around in other OSes to see what they are doing. Any ideas welcome. -- You are receiving this mail because: You are the assignee for the bug.
