Dear gem5 users,

I hope you can help me with this.

I'm trying to run full-system simulations of PARSEC and SPLASH2 over a
typical configuration with ALPHA cores, 32kB L1I and LID, as well as 512kb
of L2 per core. I use a regular mesh as interconnect.

I succeeded with MESI and Hammer, but *I'm getting stuck with Token
Coherence.*

Most of 4-core and 8-core simulations did well, many 16-core and almost all
(*not all*) 32-core simulations fail due to SOFT LOCKUP (see below).
64-core simulations do not even complete the boot-up.

--- My command line is something like this:

*build/ALPHA/gem5.opt --stats-file=blackscholes-4core.txt
configs/example/fs.py --ruby -n 4 --l1i_size=32kB --l1d_size=32kB
--l2_size=2MB --num-l2caches=4 --topology=Mesh --num-dirs=4
--garnet-network=fixed --maxinsts=10000000000000 --prog-interval=100Hz
--cpu-type=timing --script=util/parsec/blackscholes/blackscholes_4.rcS*

--- I'm getting this message on the terminal (does not make the program
abort, but I think it does not let the program to progress either) in, for
instance, 16-core cholesky, lu, vips, ferret or 32-core blackscholes, vips,
streamcluster radix.  (error messages are nearly identical from application
to application)
*BUG: soft lockup - CPU#2 stuck for 61s! [RADIX:1006]*
*Modules linked in:*
*pc = [<fffffc00003679c4>]  ra = [<fffffc00003679b0>]  ps = 0000    Not
tainted*
*pc is at smp_call_function_mask+0x174/0x290*
*ra is at smp_call_function_mask+0x160/0x290*
*v0 = 0000000000000040  t0 = 0000000000000003  t1 = 0000000000000000*
*t2 = 0000000000000000  t3 = 0000000000000000  t4 = 0000000000000040*
*t5 = 0000000000000453  t6 = fffffc001e616080  t7 = fffffc001e704000*
*a0 = 0000000000000040  a1 = 0000000000000040  a2 = 0000000000000000*
*a3 = 0000000000000001  a4 = fffffc001e743f50  a5 = fffffc001f5ba830*
*t8 = 0000000000000000  t9 = 000000532dce80cf  t10= a600000000000000*
*t11= 0000000000000000  pv = fffffc00004ead80  at = fffffc001e616080*
*gp = fffffc000085bf40  sp = fffffc001e707d18*

--- In 64-core, I had deadlock warnings during bootup (which seem to be
true). I increased the L2 memory to 1MB per core and it did progress beyond
the deadlock point, but then got stuck in another place when booting up.


It must be a memory problem for sure, but I don't know exactly what is
causing this.
I will keep trying, but still...

Does anyone have an intuition or has had the same issue?

Thanks a lot,
Sergi
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to