CCS/AM3359: HWI misses

The RTOS team have been notified. They will respond here.

Hi Garrett,

We set HWI priority to 8 but miss some HWIs. If there is no other method, we have to consider other way.

thanks,
Uchikoshi

Hello,

To simply, I added below code into "NIMU_BasicExample" .

   Clock_Params_init(&clkParams);

   clkParams.startFlag = TRUE;

   clkParams.period = 1;  //  clock ticks = 100us in .cfg file, SWI priority is set to 15.

   Clock_create(clk0Fxn, 2, &clkParams, NULL);

Void clk0Fxn(UArg arg0)

{

   /* Toggle LED1 */

   GPIO_toggle(USER_LED1);

}

Hardware is BeagleBoneBlack. No Ethernet cable is connected on RJ45.

Below is the capture of toggle waveform.

Is there any way to prevent from missing swi isr with NDK?

Regards,

Uchikoshi

Uchikoshi-san,

Can you enable TI RTOS system analyzer trace by adding logging into your RTOS configuration:

var LoggingSetup = xdc.useModule('ti.uia.sysbios.LoggingSetup');
LoggingSetup.loadLogging = true;
LoggingSetup.loadLoggerSize = 1024;
LoggingSetup.mainLoggerSize = 32768;
LoggingSetup.sysbiosLoggerSize = 32768;
LoggingSetup.sysbiosSwiLogging = true;
LoggingSetup.sysbiosHwiLogging = true;
LoggingSetup.sysbiosSemaphoreLogging = true;
LoggingSetup.loadTaskLogging = true;
LoggingSetup.loadSwiLogging = true;
LoggingSetup.loadHwiLogging = true;
LoggingSetup.enableTaskProfiler = true;
LoggingSetup.sysbiosHwiLoggingRuntimeControl = true;
LoggingSetup.sysbiosSwiLoggingRuntimeControl = true;
LoggingSetup.eventUploadMode = LoggingSetup.UploadMode_JTAGSTOPMODE;

You can then look at the Execution graph in system analyzer to see if you can catch the HWI and SWI and correlate why this occurs in the system.

Also note, SWI have lower priority to HWI in the system so ensure no other HWI can pre-empt the SWI.

Regards,

Rahul

PS: Logging setup tools wiki:

http://processors.wiki.ti.com/index.php/UIA_2.0_LoggingSetup 

Hello Rahul,

I captured system execution graph. NDK task ndk_config_Global_stackThread() runs every 200us and it seems to prevent HWI from executing. 

execution.zip

Could you please check attached fog file and confirm if this is unavildable phenomenon in case of NDK.

If you need more information, please let me know.

Regards,

Uchikoshi  

Uchikoshi-san,

If you look at the system analyzer graph, there appears to be a semaphore post and pend that correlates with the time in which the system doesn`t respond to HWIs. HWI and SWI are meant to run to completion and shouldn`t pend on any semaphore. Semaphores are only recommended to be used in tasks. Can you check which thread in the system is posting the semaphore that may be preventing the system from not responding to HWI triggers.

Regardss,
Rahul

It would be useful to see the zoomed in picture where the global_stack task is running. So basically have this zoomed in (along with the X-axis units).

Are there other times when global_stack runs and it does not cause the issue.

Note: the bin file is not very useful without the .out *, .uia.xml, and *rta.xml files.

Todd

Hi Rahul, Todd

 

Thank you for your advice.

To simplify it, I just ran the NIMU_BasicExample in PDK as-is and turned on UIA for log. Can you see attached file?

 

nimu_bbbam335x_xa8fg.uia.zip

processor_sdk_rtos_am335x_3_00_00_04

pdk_am335x_1_0_3

NDK 2.24.3.35

SYS/BIOS 6.45.1.29

UIA 2.0.6.52

 

It looks like that ti_sysbios_knl_Clock_workFunc__E() posts semaphore. This is not every time and only before execution of ti_ndk_config_Global_stackThread().

 

I appreciate any of your advice.

 

Regards,

Uchikoshi