Hi all, I was wondering if cpu frequency scaling in dom0 (using the cpufreq=dom0-kernel boot parameter) may cause problems with HVM domUs? This is on Xen 3.2.1. PV domUs seem to work just fine. They adjust to the frequency change on the go. The HVM I run using the unmodified_driver drivers (also from the 3.2.1 release) seem to have very slow timers if dom0 lowers the frequency. Birger _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
>From: Birger Brunswiek >Sent: 2008年7月24日 4:58 > >Hi all, >I was wondering if cpu frequency scaling in dom0 (using the >cpufreq=dom0-kernel >boot parameter) may cause problems with HVM domUs? This is on >Xen 3.2.1. > >PV domUs seem to work just fine. They adjust to the frequency >change on the go. >The HVM I run using the unmodified_driver drivers (also from >the 3.2.1 release) >seem to have very slow timers if dom0 lowers the frequency. >Current HVM can''t cope with freq/tsc change, unless all rdtsc are trapped which then may bring other side-effects for heavy overhead. It''s better for you to try cpufreq on a platform with constant TSC, e.g. any Intel processors supporting VT. Thanks, Kevin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Tian, Kevin wrote:>> From: Birger Brunswiek >> Sent: 2008年7月24日 4:58 >> >> Hi all, >> I was wondering if cpu frequency scaling in dom0 (using the >> cpufreq=dom0-kernel >> boot parameter) may cause problems with HVM domUs? This is on >> Xen 3.2.1. >> >> PV domUs seem to work just fine. They adjust to the frequency >> change on the go. >> The HVM I run using the unmodified_driver drivers (also from >> the 3.2.1 release) >> seem to have very slow timers if dom0 lowers the frequency. >> > > Current HVM can''t cope with freq/tsc change, unless all rdtsc > are trapped which then may bring other side-effects for heavy > overhead. It''s better for you to try cpufreq on a platform with > constant TSC, e.g. any Intel processors supporting VT.I see ... it''s just when the last time I looked for Intel CPUs it way quite difficult to get those with VT in the consumer segment. That''s quite different with AMD. Does anyone know if the modules (?) which are responsible to get the freq scaling right in PVs can be used/ported to an HVM? Similar to the unmodified drivers to use a paravirtualised disk and netif. Birger _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
>From: Birger Brunswiek [mailto:birger.b@gmx.net] >Sent: 2008年7月25日 3:43 > >Does anyone know if the modules (?) which are responsible to >get the freq >scaling right in PVs can be used/ported to an HVM? Similar to >the unmodified >drivers to use a paravirtualised disk and netif. >That pv time stuff is hardly put as a module, which is in the core of kernel, and thus hardly fit into HVM. One alternative is to use other clocksource for HVM, instead of tsc. Thanks, Kevin _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Possibly Parallel Threads
- FW: cpufreq info propagation
- cpu frequency scaling in domUs
- dict crashes with multiple map definitions
- Is: drivers/cpufreq/cpufreq-xen.c Was:Re: [PATCH 2 of 2] linux-xencommons: Load processor-passthru
- [PATCH V2 1/2] cpufreq, xenpm: fix cpufreq and xenpm mismatch