I'm really busy at the moment, but when I get some time I'll take a
look too. One thing we have done to debug checkpoints in the past is
to take multiple checkpoints say ever 10000000 ticks and then restore
from those checkpoints also taking checkpoints ever 1000000 ticks.
For every 1000000 tick multiple you can have the original (which
should be correct) and the restored checkpoint (which hopefully
matches the original). If the checkpoints don't match then something
went wrong.
I think the first step in debugging Richard's problem is looking at
what is calling PageTable::allocate(). Whatever it is probably
shouldn't be because that page is already allocated.
Ali
On Oct 24, 2007, at 1:27 PM, Vilas Sridharan wrote:
Richard,
No, as a matter of fact it is not. If you can, try applu or apsi
-- those worked for me.
I had two different failure types: benchmarks where I couldn't
even get a valid checkpoint, and benchmarks that failed after
restoring the checkpoint and beginning simulation. Ammp is in the
latter category. I'm using Simpoints' early simulation points to
do this, so I'm sometimes fast-forwarding quite a ways before
taking a checkpoint. For the ones that failed after restoring, I
don't think they always failed right away - some executed for many
cycles after restoring before they failed (this was a while ago,
though, so I don't remember for sure). I think they almost all
have the same error message, though -- some sort of page table issue.
If I get some time, I'll try and re-create some of the problems and
see if they still occur, and if I can debug them. It sure would be
nice to get these running.
-Vilas
On 10/24/07, Richard Strong <[EMAIL PROTECTED]> wrote:
Is ammp among those you got to work? I only ask because this is the
one I am trying to get work first.
-R
On 10/24/07, Vilas Sridharan < [EMAIL PROTECTED]> wrote:
Richard,
I've made checkpoints using m5 2.0b2 for (some of) the spec2k
benchmarks and gotten them to work. I've had trouble with some of
the other ones (gcc, for instance) -- I'm not sure what's causing
it and haven't yet had time to debug it (I think I usually also get
a page table error, though a different one). But I've definitely
gotten the checkpoints to work in SE mode, at least in 2.0b2. The
code may have changed for 2.0b3, but one of the developers will
have to answer that.
-Vilas
On 10/23/07, Richard Strong <[EMAIL PROTECTED]> wrote:
Hi,
I am trying to get checkpoints working with ALPHA_SE and have run
into some problems. When I try to resume a checkpoint I get a fatal
message in src/mem/page_table.cc:110 fatal("PageTable::allocate:
address 0x%x already mapped", vaddr);. It appears that an address
has already been mapped in the page table Why would this happen?
Has anyone at least tried making checkpoint for a spec2000
benchmark and resuming it for ALPHA_SE? Perhaps I am running code
at the start of the binary after I restore the checkpoint?
-Richard
I make a checkpoint using:
./build/ALPHA_SE/m5.debug configs/example/se.py --
max_checkpoints=2 --take_checkpoints="10000,100" --
checkpoint_dir=./checkpoints
I then try to resume a checkpoint with a AtomicSimpleCPU by the
command:
$m5-2.0b3*11:57:44=>./build/ALPHA_SE/m5.debug configs/example/
se.py -r 1 --checkpoint_dir=./checkpoints
M5 Simulator System
Copyright (c) 2001-2006
The Regents of The University of Michigan
All Rights Reserved
M5 compiled Oct 23 2007 11:57:34
M5 started Tue Oct 23 12:15:07 2007
M5 executing on rickshin-2.local
command line: ./build/ALPHA_SE/m5.debug configs/example/se.py -r 1
--checkpoint_dir=./checkpoints
Global frequency set at 1000000000000 ticks per second
Restoring from checkpoint
fatal: PageTable::allocate: address 0x11ff92000 already mapped
@ cycle 10000
[allocate:build/ALPHA_SE/mem/page_table.cc, line 110]
Memory Usage: 0 KBytes
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
