[linux-yocto] [PULL REQUEST] 3.10 - new preempt-rt branch - standard/preempt-rt/axxia/base]

[Bruce Ashfield]

On 1/9/2014, 11:11 AM, Hong H. Pham wrote:
> Hi Bruce.
>
>> On 13-12-02 12:18 PM, Paul Butler wrote:
>>> On Mon, Dec 02, 2013 at 12:12:20PM -0500, Bruce Ashfield wrote:
>>>> On 13-12-02 12:10 PM, Paul Butler wrote:
>>>>> The following changes since commit 
>>>>> ee9480cd91b2b46325a2da9aa6ae779d8e4163c0:
>>>>>
>>>>>     Merge branch 'standard/base' into standard/preempt-rt/base 
>>>>> (2013-11-25 15:11:10 -0500)
>>>>>
>>>>> are available in the git repository at:
>>>>
>>>> Before merging new -rt support, I'd like to see what sort of latency
>>>> and driver testing has been done on the board.
>>>>
>>>> There's no sense merging support if it has latency spikes, or the
>>>> drivers haven't been tested for locking safety and semantics.
>>>>
>>>
>>> Ok, same story for 3.4 -rt branch?
>>
>> Yep.
>
> I've posted a summary of the test results and test logs for the tests
> done on the 3.4 and 3.10 standard + preempt-rt axxia kernels.

Thanks Hong,

That's exactly what I was looking for. The docs/spreadsheets are easy to 
READ
and are more than enough test artifacts for what we are looking for.

>
> acp34xx test summary:
> https://docs.google.com/uc?id=0B-6CyQ6W462IeGM2VDVUR2FLLW8&export=download&revid=0B-6CyQ6W462IL0hRQ3BlT0FVelc0VHMyTzdtZGRWNWRkYkdVPQ 
>
>
> axm55xx test summary:
> https://docs.google.com/uc?id=0B-6CyQ6W462ITmY1RFpzTXVDY1U&export=download&revid=0B-6CyQ6W462IcndiYzh2cXhhQlJVVERMcWVmUnVzWnc4YXhnPQ 
>
>
> Test logs (acp34xx + axm55xx):
> https://docs.google.com/uc?id=0B-6CyQ6W462ISzVBQmRwMEhndVU&export=download&revid=0B-6CyQ6W462ISDVYUEVtbUVBTnNia2FZYkd5SEltUm9VOU1rPQ 
>
>
> I'll add some comments for the cyclictest graphs.  The latency 
> variance in
> the preempt-rt kernels is lower than the standard kernels (as one 
> would expect)
> but there are a few cases where the average latency is higher in the 
> preempt-rt
> kernels (axm55xx 3.10-rt).  I do not see any huge outliers (i.e. a 
> latency sample
> that's several hundred usecs) in the preempt-rt graphs, which to me 
> indicates
> that there are no latency spikes in the preempt-rt kernels.

Agreed. Average can be higher, but if the maximum is less than the standard
kernel's under load, we do have proof that -rt has made a difference where
it matters.

I'd actually like to see these stored as part of the BSP, right along 
side the
READMEs.

Paul: when you resend these pull requests, can you provide a link to them
in the BSP layer (github, if that is still the official location) ?

Cheers,

Bruce

>
> Hong
>
>>
>> Bruce
>>
>>>
> ....snip
>