Hello Karel, On Tuesday, August 12, 2008 at 14:22:47 +0200, Karel Zak wrote: > The patch is below. It's based on gettimeofday() rather than on > SIGALRM (alarm(2)). The gettimeofday has some overhead, but it > shouldn't be a big problem Both patches fix Tomasz hang, so they are good. Indeed, the gettimeofday() and related manipulations seem to have a small overhead, around 5% of the total time in loop. The big part is ioctl(RTC_RD_TIME) which is a heavy operation. On my slow test machine, the RTC_RD_TIME alone takes 57 µs, the rest of the loop takes 3 µs, and the time wasted after the loop adds 35 µs. So, on one side 5% is not much. OTOS using alarm() for timeout, a sharper loop, and an immediate timestamping of the detected tick, the equivalent busywait_second_change() function in adjtimex 1.26 takes 57 µs total. No overhead, neither in loop nor after. BTW busywait_for_rtc_clock_tick() is called only when RTC_UIE_ON fails. What about also calling it when RTC_UIE_ON succeeds, but waiting for the interrupt timeouts? This is unfortunately a quite common case these days, with all those misconfigured HPETs. Affected people would then get normal tick synchronization. Alain. -- To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
