I've committed your change. In the future please send a diff instead
of the whole file, it's much easier to deal with.
Thanks for the fix,
Ali
On Oct 19, 2007, at 11:24 AM, Geoffrey Blake wrote:
I found the problem within the alpha console code. First, a few
variables that should’ve been longs were ints, causing any memory
size of >2GB to be reported wrong. The second problem was the hard
coding the MDESC table size to the equivalent of 2GB only, so
anything over 2GB would fail booting as well.
I’ve attached the bug fixed console.c code to this message for
anyone who would be interested in running M5 sims with more than
2GB of simulated RAM.
Geoff
From: [EMAIL PROTECTED] [mailto:m5-users-
[EMAIL PROTECTED] On Behalf Of Ali Saidi
Sent: Thursday, October 18, 2007 12:17 PM
To: M5 users mailing list
Subject: Re: [m5-users] Question about swapdisk, and supported RAM
size
On Oct 18, 2007, at 11:59 AM, Geoffrey Blake wrote:
I’m still running m5-2.0b2 with my TM extensions but have run into
a problem. I am running a benchmark that takes up a very large
amount of virtual memory, and eventually needs to use the swap disk
in FS mode. My first question is, after looking at the code for a
little bit, it looks like the COW disk is brought into main memory
as it is written to, but does the disk image in memory just keep
growing if needed inside M5?
Yes, although the OS should swap out unused parts.
It seems this would be the case because after running my benchmark
for a few hours, the virtual memory footprint was 11GB and
continued growing. I’ve calculated my worst case memory footprint
for the benchmark in question at needing ~3GB of memory, so running
with 1024MB of RAM the swap disk will be heavily used.
Also, I’m trying to add more RAM than 1024MB to the simulation to
avoid using the swap disk image, but anything over that makes Linux
kernel panic. I’ve checked the boot log and I’m getting bizarre
memory ranges like the following:
Memory cluster 0 [0 - 392]
Memory cluster 1 [392 - -392]
Why would Linux be receiving a negative memory range? I’ll keep
looking, I guess it could be the console code passing bad values up
to the kernel, or something not using a 64bit integer.
I think the problem is likely the Alpha Console that sets up the
values before handing them off to the operating system. There are a
lot of ints in there that should probably be longs.
Ali
<console.c>
_______________________________________________
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
