On Mon, 15 Mar 2021 08:18:30 +0200 "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@xxxxxxxxx> wrote: > There are multiple options that affect the event's timestamps while > recording trace or displaying the trace file. Some of these options can > be used together and the order and priorities for all of them should be > strictly defined: > trace-cmd record --date , --ts-offset - only one of these can be used > trace-cmd record host-guest trace - guest timestamps are auto > synchronized with host time stamps. If no trace clock is set by the user > and "kvm" synch plugin is available, then "x86-tsc" trace clock is used > and tsc timestamps are converted to nanoseconds in both host and guest > trace events. > trace-cmd report --ts-offset, --ts2secs, --nodate > > Event timestamps corrections are applied in this order, when the trace > file is opened for reading: > 1. If it is a guest trace file and there is information for > synchronization with the host events and is this synchronization is > enabled: synchronize guest events with host events. > 2. If the user has specified --ts2secs, apply it. > 3. If the user has not specified --ts2secs and there is information in > the trace file metadata for tsc to nanosecond conversion, apply it. > 4. If the user has specified --ts-offset or --date, apply it. The > offsets specified by "report" command have higher priority that the > offsets specified by "record" command. > I'll have to play with this a bit, but for now this looks fine. -- Steve
|