Perhaps this is an issue with the order of initialization after the checkpoint restore?
If when is 1875000000000 and curTick is 1915819058000 the value this is used to reschedule the when event should be 3790819058000 (sum). Maybe curTick isn't getting added back or curTick hasn't been advanced to the restored tick yet? Ali On 22.01.2012 21:49, Anders Handler wrote: > Hi Ali, > The problem is reproducible. I think a x86 FS boot will make the error. > In my current setup the checkpoint is made at time tick 1915819058000. The curTick() yields 1915819058000, but _when is only 1875000000000, causing the bad value. > Just saving the _when value might work, but I think it will just cause an event scheduled in the past error. But I will need to test. > / Anders > > On Sat, Jan 21, 2012 at 4:43 PM, Ali Saidi <sa...@umich.edu [5]> wrote: > >> Hi Anders, >> >> No clue. It looks like Nilay last edited that code 3 months ago. So maybe he can shed some light on it. Is the problem reproducible in the checkpoints? >> >> tickEvent is scheduled when the device is created and isn't ever de-scheduled, so it should always have a positive value. >> >> _when is positive in the checkpoint, correct? >> >> because it's rescheduled for curTIck() + _when which would then always be positive. We certainly shouldn't be saving a negative tick value in the checkpoint file and I'm a little confused as to why we bother subtracting curTick in the serialize() method and adding it back in the unserialize(). Just leaving it alone should be good enough. >> >> Thanks, >> Ali >> >> On Jan 19, 2012, at 4:51 AM, Anders Handler wrote: >> >> > Hi, >> > >> > When I create a checkpoint I get a negative value stored in rtcClockTickOffset, which will make Gem5 crash when restoring the checkpoint, complaining that an event cannot be scheduled in the past. >> > >> > The failure is that the "RTCEvent event" holds an old tick value (_when) which I think it should not. The code is in: >> > src/dev/mc146818.hh >> > src/dev/mc146818.cc >> > >> > The workaround I am using is manually removing the "-" from the rtcClockTickOffset variable in the checkpoint file (m5.cpt). >> > >> > I run x86. >> > >> > Anyone has a clue how to fix? >> > >> > Best regards >> > Anders > _______________________________________________ >> > gem5-users mailing list >> > gem5-users@gem5.org [1] >> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users [2] >> >> _______________________________________________ >> gem5-users mailing list >> gem5-users@gem5.org [3] >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users [4] Links: ------ [1] mailto:gem5-users@gem5.org [2] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users [3] mailto:gem5-users@gem5.org [4] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users [5] mailto:sa...@umich.edu
_______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
