On Wed, 25 Apr 2012, Seongil O wrote:
Hi! all,
Is it possible to generate a checkpoint on ALPHA RUBY FS mode?
And if it is possible, could you tell me weather there are coherences which
do not support the checkpointing feature?
I got following error message during checkpointing on the
ALPHA_MOESI_hammer full system mode.
I built gem5 in latest version in reop.gem5.org
scons version: 2.1.0
swig version: 1.3.40
I used linux-2.6.27 kernel with tsb_ospal.
The dist image was the image including precompiled Parsec benchmark.
$ ./build/ALPHA/gem5.opt configs/example/ruby_fs.py --cpu-type=timing -n 2
--caches --l2cache --topology=Crossbar
On m5 terminal, I typed "checkpoint" command
# m5 checkpoint
And then, on the gem5 running console
panic: FIFO ordering violated: [MessageBuffer: consumer-yes [ ]] [Queue
to Throttle 0 1] name: [Queue to Throttle 0 1] current time: 3679822942500
delta: 500 arrival_time: 3679822943000 last arrival_time: 3680958830000
@ cycle 3679822942500
[enqueue:build/ALPHA/mem/ruby/buffers/MessageBuffer.cc, line 198]
Memory Usage: 1600392 KBytes
Program aborted at cycle 3679822942500
Aborted
It seems that "m5 checkpoint" command did not completely fence the memory
access to RUBY during checkpoint generation.
Is there any way to solve this problem?
My suggestion would be to create checkpoint using an rcS script. That way
if you face the problem stated above, then the problem could be easily
reproduced. Or else you might want to run for some fixed number of ticks,
closed to 3679822942500, and then checkpoint the state at the end of the
simulation. If you face problems with this approach, again those can be
reproduced.
If you want to solve it on your own, take a look at the code in
mem/ruby/buffers/MessageBuffer.cc and try to figure out why the arrival
time of a later message got set to a time less than a message that had
arrived earlier.
--
Nilay
_______________________________________________
gem5-users mailing list
gem5-users@gem5.org
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
