I printed out the instruction trace by enabling TRACE_RIP. I've found that
when xen is running inside of marss, all instructions in Dom0 Linux kernel
space is printed as "kernel: 0", and all xen instructions are printed as
"kernel: 1". But in a non-virtualized environment, kernel instructions
should be printed as "kernel: 1". This may due to the modified privilege
level (cs register) in xen environments. I am wondering if this will affect
the instructions decoding procedure in ptlsim.

I tried to look up the instruction where assertion is triggered (the last
instruction rip in the ptl_rip_trace file). I found that the rip is
pointing to an incorrect address, which is in between of two valid
instruction addresses. I think this may be the reason of the assertion
failure.

Does anyone have any suggestions how should I proceed to further diagnosis
this problem? Any suggestions would be appreciated. Thanks.


Xin







On Wed, Apr 17, 2013 at 4:02 PM, Xin Xu <[email protected]> wrote:

> I tried to use -snapshot on a clean image file. But I still see assertion
> failures. I found that it seems the assertion failures usually occur when
> xen DomU is running some tasks. If the DomU is idle, the simulation seems
> be to fine. So I guess that it is not the simulation corrupting the image
> file.
>
> I am relatively new on marss. I am wondering how should I find out where
> is the problem. Is that possible that there are some instructions that are
> not correctly handled in the simulation mode?
>
> Thank you
>
>
> On Tue, Apr 16, 2013 at 11:50 AM, Xin Xu <[email protected]> wrote:
>
>> I see. Sorry for misunderstanding. I will try that and update results
>> here. Thanks for you suggestion.
>>
>>
>> On Tue, Apr 16, 2013 at 11:46 AM, Paul Rosenfeld <[email protected]>wrote:
>>
>>> What I meant was that if the first time you run the simulation it works
>>> fine but on subsequent runs it errors out, then this might be an indication
>>> that the simulation is corrupting your image. However, if the simulation
>>> errors out on a fresh disk image (no simulations previously run on it),
>>> then it might be an indication that something else is wrong.
>>>
>>> But either way, I think the -snapshot argument is a good precaution.
>>>
>>>
>>> On Tue, Apr 16, 2013 at 11:41 AM, Xin Xu <[email protected]> wrote:
>>>
>>>> Hi Paul,
>>>>
>>>> The error occurs when after simulation mode is started, it seems that
>>>> the Assertion `ctx.page_fault_addr != 0' failure occur very soon after
>>>> the simulation starts. The output shows that
>>>>
>>>> Completed             0 cycles,             0 commits:         0 Hz,
>>>>       0 insns/sec
>>>>
>>>> But for the Assertion `physreg->data' failure, there are some
>>>> instructions are committed:
>>>>
>>>> Completed        351000 cycles,        225581 commits:    420941 Hz,
>>>>  266862 insns/sec
>>>>
>>>> I will try snapshot parameter when run simulations.
>>>>
>>>> By the way, apparently the rootsize=60GB is too large for
>>>> ubuntu-vm-builder, it always produce an image with 5GB space. I now changed
>>>> it to 20GB. The produced image is in correct size now. I will do further
>>>> test to see if this can change the situation.
>>>>
>>>> Thanks
>>>>
>>>>
>>>>
>>>> On Tue, Apr 16, 2013 at 11:27 AM, Paul Rosenfeld 
>>>> <[email protected]>wrote:
>>>>
>>>>> Sorry, forgot to reply all.
>>>>>
>>>>> Does the error happen on the very first time you try to run a disk
>>>>> image or is it intermittent?
>>>>>
>>>>> I'd recommend you add the -snapshot option when you are planning on
>>>>> running simulation. This will throw away any changes to the disk that
>>>>> happen during your session when the simulation ends and hopefully make the
>>>>> corruption go away.
>>>>>
>>>>>
>>>>> On Tue, Apr 16, 2013 at 11:24 AM, Xin Xu <[email protected]>wrote:
>>>>>
>>>>>> Hi Paul,
>>>>>>
>>>>>> Thanks for your reply. Here is the command,
>>>>>>
>>>>>> qemu/qemu-system-x86_64 -m 4096 -hda tmpDSQFvf.qcow2 -net
>>>>>> nic,model=ne2k_pci -net user
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Tue, Apr 16, 2013 at 11:21 AM, Paul Rosenfeld <
>>>>>> [email protected]> wrote:
>>>>>>
>>>>>>> What is the command are you using to start marss (specifically, are
>>>>>>> you using the -snapshot option)?
>>>>>>>
>>>>>>>
>>>>>>> On Mon, Apr 15, 2013 at 2:58 PM, Xin Xu <[email protected]>wrote:
>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> I am trying to set up a customized image with xen hypervisor using
>>>>>>>> the ubuntu-vm-builder tool. I follow the instructions in this page
>>>>>>>> http://marss86.org/~marss86/index.php/Custom_Disk_Images
>>>>>>>> and create a precise 12.04 ubuntu image with default disk size 4GB.
>>>>>>>> This image can be correctly loaded and executed within the marssx86 in
>>>>>>>> emulation mode. However, once I start the simulation mode using these
>>>>>>>> commands
>>>>>>>>
>>>>>>>> simconfig -machine xeon_single_core
>>>>>>>> simconfig -run -stopinsns 10m -stats output.log
>>>>>>>>
>>>>>>>> after certain numbers of instructions, the simulation breaks with
>>>>>>>> assertion failures. I observed two types of assertions:
>>>>>>>>
>>>>>>>> Assertion `ctx.page_fault_addr != 0' failed
>>>>>>>> Assertion `physreg->data' failed
>>>>>>>>
>>>>>>>>
>>>>>>>> I searched in the mailing list, it seems that the problem is caused
>>>>>>>> by the corrupted images. I create several images with 12.04 and 11.10. 
>>>>>>>> They
>>>>>>>> all have the same problem. I am not sure where is the problem.
>>>>>>>>
>>>>>>>> One thing I would like to mention is even if I specify the rootsize
>>>>>>>> in the config file of ubuntu-vm-builder to other sizes such as 
>>>>>>>> 60gb/40gb,
>>>>>>>> but the created image is always 4gb. I am not sure if this is related 
>>>>>>>> or
>>>>>>>> not.
>>>>>>>>
>>>>>>>> I also tried the raw format image that created by virtualbox, the
>>>>>>>> problem is same. The emulation part is OK even if an application is 
>>>>>>>> running
>>>>>>>> in domu, but the simulation always breaks.
>>>>>>>>
>>>>>>>> Any suggestions would be appreciated. Thanks
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> http://www.marss86.org
>>>>>>>> Marss86-Devel mailing list
>>>>>>>> [email protected]
>>>>>>>> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>
_______________________________________________
http://www.marss86.org
Marss86-Devel mailing list
[email protected]
https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel

Reply via email to