Date: Fri, 31 Jan 2020 17:40:42 +0000 (UTC) From: Conrad Meyer <cem@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r357336 - in head: share/man/man4 sys/x86/cpufreq Message-ID: <202001311740.00VHegIT085315@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: cem Date: Fri Jan 31 17:40:41 2020 New Revision: 357336 URL: https://svnweb.freebsd.org/changeset/base/357336 Log: hwpstate(4): Ignore CurPstateLimit by default Add a sysctl knob to allow users to re-enable it, and document the knob and default in cpufreq.4. (While here, add a few unrelated updates to cpufreq.4.) It seems that the register value in some hardware simply reflects the configured P-state. This results in an inadvertent and unintended outcome where the P-state can only walk down, and then the driver becomes "stuck" in the slowest possible P-state. The Linux driver never consults this register, so that's some evidence that ignoring the contents are relatively harmless. PR: 234733 Reported by: sigsys AT gmail.com, Erich Dollanksy <freebsd.ed.lists AT sumeritec.com> Modified: head/share/man/man4/cpufreq.4 head/sys/x86/cpufreq/hwpstate_amd.c Modified: head/share/man/man4/cpufreq.4 ============================================================================== --- head/share/man/man4/cpufreq.4 Fri Jan 31 15:56:08 2020 (r357335) +++ head/share/man/man4/cpufreq.4 Fri Jan 31 17:40:41 2020 (r357336) @@ -24,7 +24,7 @@ .\" .\" $FreeBSD$ .\" -.Dd January 22, 2020 +.Dd January 31, 2020 .Dt CPUFREQ 4 .Os .Sh NAME @@ -73,9 +73,13 @@ has passed (e.g., the system has cooled sufficiently). If a sysctl cannot be set due to an override condition, it will return .Er EPERM . .Pp -The frequency cannot be changed if TSC is in use as the timecounter. +The frequency cannot be changed if TSC is in use as the timecounter and the +hardware does not support invariant TSC. This is because the timecounter system needs to use a source that has a constant rate. +(On invariant TSC hardware, the TSC runs at the P0 rate regardless of the +configured P-state.) +Modern hardware mostly has invariant TSC. The timecounter source can be changed with the .Pa kern.timecounter.hardware sysctl. @@ -105,6 +109,15 @@ some systems. .It Va debug.cpufreq.verbose Print verbose messages. This setting is also accessible via a tunable with the same name. +.It Va debug.hwpstate_pstate_limit +If enabled, the AMD hwpstate driver limits administrative control of P-states +(including by +.Xr powerd 8 ) +to the value in the 0xc0010061 MSR, known as "PStateCurLim[CurPstateLimit]." +It is disabled (0) by default. +On some hardware, the limit register seems to simply follow the configured +P-state, which results in the inability to ever raise the P-state back to P0 +from a reduced frequency state. .El .Sh SUPPORTED DRIVERS The following device drivers offer absolute frequency control via the @@ -112,11 +125,15 @@ The following device drivers offer absolute frequency interface. Usually, only one of these can be active at a time. .Pp -.Bl -tag -compact -width ".Pa acpi_perf" +.Bl -tag -compact -width ".Pa hwpstate_intel" .It Pa acpi_perf ACPI CPU performance states .It Pa est Intel Enhanced SpeedStep +.It Pa hwpstate +AMD Cool'n'Quiet2 used in K10 through Family 17h +.It Pa hwpstate_intel +Intel SpeedShift driver .It Pa ichss Intel SpeedStep for ICH .It Pa powernow Modified: head/sys/x86/cpufreq/hwpstate_amd.c ============================================================================== --- head/sys/x86/cpufreq/hwpstate_amd.c Fri Jan 31 15:56:08 2020 (r357335) +++ head/sys/x86/cpufreq/hwpstate_amd.c Fri Jan 31 17:40:41 2020 (r357336) @@ -131,6 +131,12 @@ static int hwpstate_verify; SYSCTL_INT(_debug, OID_AUTO, hwpstate_verify, CTLFLAG_RWTUN, &hwpstate_verify, 0, "Verify P-state after setting"); +static bool hwpstate_pstate_limit; +SYSCTL_BOOL(_debug, OID_AUTO, hwpstate_pstate_limit, CTLFLAG_RWTUN, + &hwpstate_pstate_limit, 0, + "If enabled (1), limit administrative control of P-states to the value in " + "CurPstateLimit"); + static device_method_t hwpstate_methods[] = { /* Device interface */ DEVMETHOD(device_identify, hwpstate_identify), @@ -161,7 +167,8 @@ static driver_t hwpstate_driver = { DRIVER_MODULE(hwpstate, cpu, hwpstate_driver, hwpstate_devclass, 0, 0); /* - * Go to Px-state on all cpus considering the limit. + * Go to Px-state on all cpus, considering the limit register (if so + * configured). */ static int hwpstate_goto_pstate(device_t dev, int id) @@ -170,14 +177,15 @@ hwpstate_goto_pstate(device_t dev, int id) uint64_t msr; int cpu, i, j, limit; - /* get the current pstate limit */ - msr = rdmsr(MSR_AMD_10H_11H_LIMIT); - limit = AMD_10H_11H_GET_PSTATE_LIMIT(msr); - if (limit > id) { - HWPSTATE_DEBUG(dev, - "Restricting requested P%d to P%d due to HW limit\n", id, - limit); - id = limit; + if (hwpstate_pstate_limit) { + /* get the current pstate limit */ + msr = rdmsr(MSR_AMD_10H_11H_LIMIT); + limit = AMD_10H_11H_GET_PSTATE_LIMIT(msr); + if (limit > id) { + HWPSTATE_DEBUG(dev, "Restricting requested P%d to P%d " + "due to HW limit\n", id, limit); + id = limit; + } } cpu = curcpu;
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202001311740.00VHegIT085315>