From mboxrd@z Thu Jan 1 00:00:00 1970 From: tracemein@gmail.com (Naren) Date: Mon, 15 Feb 2010 12:42:36 +0100 Subject: [ltt-dev] UST - Problems in recording trace on a current process In-Reply-To: <20100215010447.riemtxaxus4cgwgw@www.imp.polymtl.ca> References: <19fa8ef31002081349v36a13174m1a031048048fd45c@mail.gmail.com> <4B72D009.2060702@polymtl.ca> <19fa8ef31002100853n189221f3u3015c85afc65db5c@mail.gmail.com> <201002101515.49624.pierre-marc.fournier@polymtl.ca> <19fa8ef31002101408l3b5eb66bnc062c2785910d63e@mail.gmail.com> <20100215010447.riemtxaxus4cgwgw@www.imp.polymtl.ca> Message-ID: <19fa8ef31002150342v17a3755t99d255a9cc4ea011@mail.gmail.com> Message-ID: <20100215114236.am5Qli_QKb_S9miMCXb2KfXLl2b_yoLgHMgMRJLznik@z> Hi Pierre, >>You could load in lttv both a userspace trace and a kernel trace that was >>recorded at the same time. You would then get appropriate events for drawing >>the CFV as well as userspace events. For that you would need to make sure the >>TSC clock is used in ust (see kernelcompat.h). I verified the clock source in kernelcompat.h. It uses gettimeofday from sys/time.h. As you suggested, its imperative that a tsc clock source is used when userspace and kernel tracing are combined. I went through the timesource patch of lttng,x86-trace-clock.patch, a patch dealing with switching clocksource of x86 architecture, to tsc. Can you suggest as to how we can change the clocksource to tsc in ust? Regards, Naren -------------- next part -------------- An HTML attachment was scrubbed... URL: