mvme6100 support for CVS head

Kate Feng feng1 at bnl.gov
Fri Jul 17 18:31:06 UTC 2009


Joel Sherrill wrote:
> Kate Feng wrote:
>> Daron Chabot wrote:
>>  
>>> On Fri, Jul 17, 2009 at 7:06 AM, Kate Feng<feng1 at bnl.gov> wrote:
>>>      
>>>> Joel Sherrill wrote:
>>>>          
>>>>> Feng, Kate wrote:
>>>>>              
>>>>>> I wrote and upgraded the mvme5500 BSP to disco BSP to support the
>>>>>> mvme6100 board.  At that time, the performance of the network on
>>>>>> the mvme6100 board was not important, but the decision for the
>>>>>> choice of H/W for my project was urgent.  Thus, the code for
>>>>>> the network is not really ready for a formal release yet.  It 
>>>>>> seems that
>>>>>> the
>>>>>> network code for beatnik BSP needs to be  optimized or rewritten  
>>>>>> as well
>>>>>> based on the version I obtained from Till Straumann a while ago.  
>>>>>> I do
>>>>>> not know if he had any chance to improve the netowrk
>>>>>> performance ever since then.
>>>>>>                   
>>>>> So .. it should work to ...
>>>>>
>>>>> a) build for mvme5500 and run on mvme6100?
>>>>>               
>>>> Not the mvme5500 BSP  in the current RTEMS release.
>>>>           
>>> As in my original posting, I am not referring to the current (4.9.2)
>>> release, but to the mvme5500 BSP as it stands
>>> in the cvs HEAD version.
>>>
>>> This enum from mvme5500/include/bsp.h seems to imply run-time
>>> detection of board type:
>>> /* Board type */
>>> typedef enum {
>>>     undefined = 0,
>>>     MVME5500,
>>>     MVME6100
>>> } BSP_BoardTypes;
>>>
>>> So, as Joel asked, with the version on cvs HEAD, is it possible to
>>> build for the 5500 and _run_ on the 6100 ?
>>>       
>> I will have to submit the network driver for the mvme6100.   I just wish
>> one to be aware that the current driver is not written in the typical 
>> RTEMS
>> real-time fashion, and its performance could be boosted -up to at least
>> an estimated factor of two. Thus, one should not use that as a template
>> for other network driver.
>>
>> Speaking of  the CVS head, back in April, I  had to modify some 
>> shared files
>> to solve the issue with the RTEMS work space and C program heap
>> before I submitted the updated code for mvme5500 BSP.  I did not send
>> the patch for the shared files back then because I only had it tested 
>> for the case
>> when rtems_unified_work_area is false.  Joel, is the code updated ?
>> The code should be updated before mvme5500,  mvme6100, and I think
>> any other PPC can work.  Did I miss something ?
>>
>>   
> I have no idea what the problem you are referring to is. :)
>
> Give me a description or half-baked old patch and I will look
> at it.
> -
The attached is the half-baked patch for the CVS head, which I only
tested it with mvme5500 BSP briefly.  Regarding the network driver
for the mvme6100, it was  ported quickly from the beatnik BSP, which
was written by Till Straumann.   Thus, the performance would be
similar  to that of the beatnik BSP.  If both Joel and Till does not mind
that I release a work-around version in the CVS head, I will find time
to package it, if that helps.  However, I only used the driver to boot
up the BSP so that I could test my project. Perhaps,
has one rewritten it to improve its performance since the mvme6100
boards are heavily used for others'  projects ?

Please advise.

Sincerely,
Kate
-------------- next part --------------
A non-text attachment was scrubbed...
Name: CVS-RTEMS4.10-libbsp.diff
Type: text/x-patch
Size: 4463 bytes
Desc: not available
URL: <http://lists.rtems.org/pipermail/users/attachments/20090717/5a56f6d5/attachment-0001.bin>


More information about the users mailing list