dear M5 team,
thank you for replying.
Though I changed source code as below, simulation halt soon.
And I also tried to add benchmark on ALPHA_FS,
but FATAL message is printed.
in simple/cpu.hh
------------------
1.add static int NUM_CPUs to the member of class SimpleCPU
this indicates the number of CPUs in the system.
2.add int cpu_id
this indicates the id of cpu
in simple/cpu.cc
-------------------
xc = new ExecContext(this, /* thread_num */ 0, p->process, /* asid */ 0);
this code change as below
xc = new ExecContext(this, /* thread_num */ 0, p->process, cpu_id);
and
memReq->asid = cpu_id;
I think these changes occur troubles somewhere,
but I couldn't find.
So I also tried to run single app on ALPHA_FS as below
to run multi_apps ultimately.
1. I create disk image and copied benchmark binary to the disk image using
Real-machine.
dd if=/dev/zero of=./my_disk.img ibs=1024 count=10240
fdisk (option)
mkfs -t ext2 my_disk.img
mount -o loop -t ext2 my_disk.img /mount/point
cp benchmark /mount/point
umount /mount/point
2. discribed .py file to use the my_disk.img as IDE slave disk.
IDE master is the disk image from m5_system_1.1.
3. run ALPHA_FS/m5.debug and connected to it's console using m5term
4. mount dev/hdb /mnt on running ALPHA_FS
5. ls /mnt shows existing benchmark binary in disk image
6. run the benchmark , but FATAL message is printed
modprobe: FATAL: Could not load /lib/modules/2.6.11.8/modules.dep: No such
file or directory
benchmark is from SPEC2000, such as bzip2.
How can I add benchmarks in FS mode?
Thanks,
------
Tomonobu Mihara<[EMAIL PROTECTED]> ,Japan
Steve Reinhardt <[EMAIL PROTECTED]> wrote:
>
> At first glance your .py file below looks fine to me. Although there
> are no physical addresses in syscall emulation mode, there is a
> translation step where the address space ID (ASID) is pasted into the
> upper bits of the address to make addresses unique across different
> processes (see the dummyTranslation() method in cpu/exec_context.hh).
> The problem you're having is that these ASIDs are currently only set for
> different threads on the same CPU (i.e. using SMT on FullCPU). For
> example, in the SimpleCPU constructor (cpu/simple/cpu.cc) the ASID
> argument of the ExecContext constructor it calls is hard-wired to 0. If
> you change that code to give each workload a different ASID you should
> be fine.
>
> In syscall emulation mode, we've only used the MP model to run parallel
> apps, so this hasn't been a problem for us. (We've only used the SMT
> model to run multiprogram workloads as well, so as we found recently the
> code doesn't support running parallel apps on SMT either.) We're
> currently embarking on a pretty thorough revamping of the memory system
> that will provide a more realistic physical address space for syscall
> emulation mode (among other things), so hopefully by the time M5 2.0
> comes out this won't be a problem. That won't happen soon though (we're
> shooting for this summer).
>
> Steve
>
> Mihara Tomonobu wrote:
> > dear M5 team,
> >
> > I would like to simulate Multiprocessing on CMP,
> > but I wonder how to describe .py file.
> >
> > I think that address space of workloads in NO_FullSim mode is not
> > translated to physical address, because workloads belonging other
> > process share data (as my m5stats.txt).
> >
> > Can I separate virtual address, or transelate virtual address into physical
> > address easily?
> >
> > Thanks.
> > ----
> > Tomonobu Mihara, Japan
> >
> >
> >
> > -------------------------------------------------------
> > This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
> > for problems? Stop! Download the new AJAX search engine that makes
> > searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
> > http://ads.osdn.com/?ad_id=7637&alloc_id=16865&op=click
> > _______________________________________________
> > m5sim-users mailing list
> > [email protected]
> > https://lists.sourceforge.net/lists/listinfo/m5sim-users
>
>
>
> -------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
> for problems? Stop! Download the new AJAX search engine that makes
> searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
> http://ads.osdn.com/?ad_id=7637&alloc_id=16865&op=click
> _______________________________________________
> m5sim-users mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/m5sim-users
-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems? Stop! Download the new AJAX search engine that makes
searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
_______________________________________________
m5sim-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/m5sim-users
