Hi Andreas, et al.,

Andreas, thanks for the note in reply.

I did try using the "--dual" option when starting full system, but this has
not worked.  The following code, in configs/common/FSConfig.py, line 533,
falls through to the fatal 'don't know how' line when using x86.
 Apparently Alpha and ARM are supported by 'tsunami' and 'realview', but
x86 has no dual simulation support.
    if hasattr(testSystem, 'realview'):
        self.etherlink.int0 = Parent.testsys.realview.ethernet.interface
        self.etherlink.int1 = Parent.drivesys.realview.ethernet.interface
    elif hasattr(testSystem, 'tsunami'):
        self.etherlink.int0 = Parent.testsys.tsunami.ethernet.interface
        self.etherlink.int1 = Parent.drivesys.tsunami.ethernet.interface
    else:
        fatal("Don't know how to connect these system together")

Further, regarding ruby vs. non-ruby: I would like the simulated machine to
be x86 muli-core cache-coherent and that would imply using the ruby memory
system (correct?)

Succinctly...
1. is it possible to run x86 full system using the "--dual" option
2. if yes, what do I need to change or fix?

Thank you,
Pete Stevenson


Date: Fri, 25 Apr 2014 08:22:58 +0100
> From: Andreas Hansson <[email protected]>
> To: gem5 users mailing list <[email protected]>
> Subject: Re: [gem5-users] x86 full system simulation with networked
>         (server) applications
> Message-ID: <cf7fcb0b.184c3%[email protected]>
> Content-Type: text/plain; charset="windows-1252"
>
> Hi Pete,
>
> X86_FS and X86_SE is now simply X86 (and full-system or syscall-emulation
> is decided based on how you run it). When you say network I assume you mean
> Ethernet, as in something connecting two systems? If so, there is no need
> to use Ruby. I don?t know about X86, but it most certainly works for Alpha
> and ARM.
>
> There are Ethernet cards as part of gem5, and you have at least one
> regression that shows how it is used. Note that there is currently no
> Ethernet switch model in gem5, so typically you end up with one ?driver
> system? that is generating the load, and one ?test system? that is the
> actual configuration you are evaluating. See e.g. config/fs.py for an
> example of this setup.
>
> Andreas
>
> From: Pete Stevenson <[email protected]<mailto:
> [email protected]>>
> Reply-To: gem5 users mailing list <[email protected]<mailto:
> [email protected]>>
> Date: Friday, 25 April 2014 05:55
> To: "[email protected]<mailto:[email protected]>" <[email protected]
> <mailto:[email protected]>>
> Subject: [gem5-users] x86 full system simulation with networked (server)
> applications
>
> Hi All -
>
> I am having trouble getting the x86 simulated full system to connect to
> the network.  I have been trying for a few days now to get this to work,
> and I have scoured the message lists, blog posts, FAQs, tutorials, etc. The
> most useful document I have found is this:
> https://www.mail-archive.com/[email protected]/msg00071.html
>
> I think it would be useful to the community if there were a guide to
> starting up an x86 multi-core full system simulation with gem5.
>
> Some more specific questions follow...
>
> Regarding simulated architecture: there is no longer an "X86_FS" config,
> so which config should I use?  (is it X86_MESI_CMP_directory?)
>
> Regarding the network, or lack thereof:  I compiled using the mesi/cmp
> config mentioned above, obtained files as referred to by the message above,
> namely the uncompressed linux kernel, x86_64-vmlinux-2.6.22.9.smp (very
> old!!), and the files linux-bigswap2.img, and x86root.img.  After starting
> the sim and connection in by telnet or m5term, I cannot ping anything and
> other tools (i.e. ifconfig, etc) don't work because there is nothing in
> /proc.  The meta question being, what else do I need to do... how do I get
> to full system sim + network + ability to use wget, apt-get or yum, etc.
>
> Regarding disk images: I have also tried making my own disk image, but
> after reading several blog posts and message board replies, this has not
> proved very fruitful (yet).  Perhaps this is the right path, but I'd like
> some indication that it is (i.e. before investing time in this further).
>
> Regarding feasibility:  is full system + server apps + network even
> possible?  If yes, then I am willing to invest my time here, but of course
> it would be good to know I am attempting something that is reasonably well
> supported.
>
> For context, I plan to extend the work of my Ph.D. thesis, namely
> investigations into deduplicated memory systems.
>
> Thank you,
> Pete Stevenson
>
>
>
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to