Hi James,

 

Sorry for the slow response.  I meant to respond to this earlier but I’ve been caught up in a variety of deadlines so I haven’t had time to look at it.  I apologize for not looking at your initial error message more closely; from reading it I was pretty sure it was another error that I ran into (which is what I described).

 

You are correct that it has multiple matches for the interrupt controller.  You can fix this by placing the switch CPUs at the Root level instead of the System level.  I’ve fixed the Wiki to reflect this, and I’m attaching a configuration file that has the same fix.  Right now the O3 CPU has a bug with sampling (it hangs after a bit), and doesn’t work with caches.  I’m working on fixing these, and hopefully we can release a new version soon that’ll have it fixed (or I’ll just release a patch).

 

Kevin

 


From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of James Anon
Sent: Wednesday, September 20, 2006 2:04 PM
To: M5 users mailing list
Subject: Re: [m5-users] Using the SwitchCpu function

 

Hi Kevin,

Thanks again for the response.  I think I am following your instructions to a tee, however I still seem to be running into an error (specifically, it seems as though there is an ambiguity in the python with respect to the parent of the CPU).  My configuration is for the full system mode, so it is a bit more complicated than the example in the FAQ, but it essentially works as follows:

run_cpu = AtomicSimpleCPU()
switch_cpu = TimingSimpleCPU(defer_registration=True)
mem_mode = 'timing'

run_cpu.clock = '2GHz'
switch_cpu.clock = '2GHz'

switch_cpu_list = [(run_cpu,switch_cpu)]

root = Root(clock = '1THz',
            system = makeLinuxAlphaSystem(mem_mode, bm[0]))
root.system.cpu = run_cpu
root.system.switch_cpu = switch_cpu

# I am a bit unsure about the next four lines, however the error is the
# same with or without them
run_cpu.connectMemPorts(root.system.membus)
run_cpu.mem = root.system.physmem
switch_cpu.connectMemPorts(root.system.membus)
switch_cpu.mem = root.system.physmem

m5.instantiate (root)

m5.simulate(20000) 
m5.switchCpus(switch_cpu_list)
exit_event = m5.simulate(maxtick)

This causes the following error:

Error in unproxying param 'cpu' of system.intrctrl
Traceback (most recent call last):
  File "<string>", line 1, in ?
  File "build/ALPHA_FS/python/m5/main.py", line 314, in main
    execfile(sys.argv[0], scope)
  File "configs/example/fsSwap.py", line 65, in ?
    m5.instantiate(root)
  File "build/ALPHA_FS/python/m5/__init__.py", line 87, in instantiate
    root.print_ini()
  File "build/ALPHA_FS/python/m5/config.py", line 517, in print_ini
    self._children[child].print_ini()
  File "build/ALPHA_FS/python/m5/config.py", line 517, in print_ini
    self._children[child].print_ini()
  File "build/ALPHA_FS/python/m5/config.py", line 505, in print_ini
    value = value.unproxy(self)
  File "build/ALPHA_FS/python/m5/config.py", line 665, in unproxy
    result, done = self.find(obj)
  File "build/ALPHA_FS/python/m5/config.py", line 748, in find
    return obj.find_any(self._pdesc.ptype)
  File "build/ALPHA_FS/python/m5/config.py", line 466, in find_any
    raise AttributeError, \
AttributeError: parent.any matched more than one: <bound method TimingSimpleCPU.path of <m5.objects.SimpleCPU.TimingSimpleCPU object at 0xb78ff78c>> <bound method AtomicSimpleCPU.path of <m5.objects.SimpleCPU.AtomicSimpleCPU object at 0xb78ff74c>>


This is the same error I have been getting (and actually why I replied before asking for a more specific example, as I thought I was doing something wrong in declaring the child to be instantiated).  As I said before, I know very little python, but it seems to be an ambiguity as to the parent of a CPU method.  Please let me know if you have any insight into what I can do to resolve this issue.

Thank you once again,

James



On 9/13/06, Kevin Lim <[EMAIL PROTECTED]> wrote:

Hi James,

 

Sorry for the slow response.  I've had some pretty big deadlines due recently, so I've been very busy with that.

 

I looked at a config I had working a while ago for sampling and I updated the Wiki with the information from it.  Basically you should be able to assign the CPU you will switch in as a child of the system that it will run in once it is switched in.  So if you had an instantiation of System called my_system, you should be able to do my_system.switch_cpu = my_switch_cpu.  In this case you're not actually setting a parameter within System, but merely indicating that it has a child object switch_cpu.

 

Hopefully this helps,

 

Kevin

 


From: [EMAIL PROTECTED] [mailto: [EMAIL PROTECTED]] On Behalf Of James Anon
Sent: Tuesday, September 05, 2006 4:07 PM
To: M5 users mailing list
Subject: Re: [m5-users] Using the SwitchCpu function

 

Hi Kevin,

Thanks for the response.  I was wondering if you could expand a bit upon exactly where to attach the child cpu list (possibly by just providing a sample configuration file that uses the swithCpu).  I've tried attaching it at both the root level, and at the system level; but the problem is m5 dies with the error that there is no such parameter at that level ( e.g. "AttributeError: Class Root has no parameter cpuList").  I've looked through the root and system objects and can't find any parameter associated with such a list.  I also tried declaring a vector of cpus instead of just one ( e.g.     root.system.cpu = [cpu,cpu2]) to see if I could get it to instantiate that way, but that usually ends in an error such as "AttributeError: parent.any matched more than one:".  Finally, I've also tried creating multiple systems ( root.testdsys and root.drivesys), but I don't think that is giving me what I desire.

I'm still pretty new to Python so I apologize if there is a simple way to look this up and I am just overlooking it, but as of this moment I'm at a lost for how to solve this issue. 

James

On 9/3/06, Kevin Lim <[EMAIL PROTECTED]> wrote:

I believe the problem here is that cpu and cpu2 need to be part of the system (somewhere) in order to be instantiated by m5's configuration parsing.  The error that it's giving you means that you didn't have those two CPUs as part of the configuration that actually gets instantiated.  Thus when it tries to switch between the CPUs, it finds that the CPUs have no corresponding object within the running instance of m5.  Try making your switch_cpu_list as a child of either the root object or a system object (I believe you should just be able to do "root.switch_cpus = switch_cpu_list" or something along those lines).  I'll try to clean up the Wiki to clarify that when I get a better example.

 

Kevin

 


From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of James Anon
Sent: Friday, September 01, 2006 2:35 PM
To: [email protected]
Subject: [m5-users] Using the SwitchCpu function

 

I am trying to use the m5.switchCpus function to quickly simulate pass the linux booting stage, and then switch to detailed mode after the full system is booted.  I followed the wiki page and have made minor modifications to the fs.py example script including:

cpu = AtomicSimpleCPU()
cpu2 = DetailedO3CPU()
mem_mode = 'timing'
.
.
.

switch_cpu_list = [(cpu,cpu2)]
m5.simulate(3663496423522)
m5.switchCpus(switch_cpu_list)
exit_event = m5.simulate(maxtick)


Most other options I've left the same from the fs.py script, but when I run the simulator, it works perfectly until the switch point and then dies with:

Switching CPUs
Traceback (most recent call last):
  File "<string>", line 1, in ?
  File "build/ALPHA_FS/python/m5/main.py", line 314, in main
    execfile(sys.argv[0], scope)
  File "configs/example/fsSwap.py", line 86, in ?
    m5.switchCpus(switch_cpu_list)
  File "build/ALPHA_FS/python/m5/__init__.py", line 206, in switchCpus
    new_cpu.takeOverFrom(old_cpus[index])
  File "build/ALPHA_FS/python/m5/config.py", line 570, in takeOverFrom
    self._ccObject.takeOverFrom(cpu_ptr)
AttributeError: 'NoneType' object has no attribute 'takeOverFrom'


Is there something I'm forgetting to instantiate?  Any help would be greatly appreciate.

Ps.  Thanks for setting up the WIKI page.  It has been a tremendous help, and in my opinion a great improvement over the documentation that was available for the previous version of m5.  I hope the community continues to add to it!


_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

 


_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

 

Attachment: fs.py
Description: Binary data

_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to