Issue in WDT motorware API for 2806x

Jonathan Lock

Other Parts Discussed in Thread: MOTORWARE

Hey all. A potential problem with the WDOG_clearCounter function in "sw/drivers/wdog/src/32b/f28x/f2806x/wdog.h" is that it's definition is;

void WDOG_clearCounter(WDOG_Handle wdogHandle)
{
WDOG_Obj *wdog = (WDOG_Obj *)wdogHandle;

ENABLE_PROTECTED_REGISTER_WRITE_MODE;

// write first sequence
wdog->WDKEY = 0x55;

// write second sequence
wdog->WDKEY = 0xAA;

DISABLE_PROTECTED_REGISTER_WRITE_MODE;

return;
} // end of WDOG_clearCounter() function

When compiler optimizations are enabled the first write to WDKEY (0x55) can be removed, causing the WDT to trigger a reset. A potential way to solve this (that results in as little overhead as possible) is to change the definition to

void WDOG_clearCounter(WDOG_Handle wdogHandle)
{
volatile WDOG_Obj *wdog = (WDOG_Obj *)wdogHandle;

ENABLE_PROTECTED_REGISTER_WRITE_MODE;

// write first sequence
wdog->WDKEY = 0x55;

// write second sequence
wdog->WDKEY = 0xAA;

DISABLE_PROTECTED_REGISTER_WRITE_MODE;

return;
} // end of WDOG_clearCounter() function

Which ensures both writes are performed. Can someone verify/update this?

over 12 years ago

0 ChrisClearman over 12 years ago

TI__Guru** 118760 points

thanks, Jonathan. Let us look at this one.

BTW - if you post InstaSPIN/MotorWare topics in the InstaSPIN sub-forum you will be assured I am notified.

0 Trey German289 over 12 years ago

TI__Mastermind 22025 points

Jonathan,

Thanks for bringing this to our attention. I agree that this could be potentially be optimized out, but I'd like to ask if you've actually observed this?

I spent a few mins trying to recreate this with varying optimization levels and was unsuccessful. Declaring the struct as volatile should mitigate any potential issues nonetheless.

Regards,

0 Jonathan Lock over 12 years ago in reply to Trey German289

Intellectual 775 points

Chris; OK, I'll keep that in mind!

Trey; Yes, I have noticed that the compiler did remove the initial statement. I'm using the TI 6.1.3 compiler for the C2000 family (controlStick Piccolo 28069 in particular) and I noticed this issue with the compiler flags;

-v28 -ml -mt -O4 -g --define=FAST_ROM_V1p6 --diag_warning=225 --display_error_number --diag_wrap=off --fp_reassoc=on --fp_mode=relaxed

When compiling when the struct is not defined as volatile the WDOG_clearCounter function generates the following assembly code;

61       ENABLE_PROTECTED_REGISTER_WRITE_MODE;
00c77d:   7622        EALLOW
67       wdog->WDKEY = 0xAA;
00c77e:   56BFAADC    MOVB         *+XAR4[3], #0xaa, UNC
69       DISABLE_PROTECTED_REGISTER_WRITE_MODE;
00c780:   761A        EDIS

TBH I'm completely unversed in the C28x assembly instructions, but as there's only (what I assume to be) one move operation I don't see any way this could write both key values. When compiled with the struct declared as volatile I get;

61       ENABLE_PROTECTED_REGISTER_WRITE_MODE;
00c739:   7622        EALLOW
64       wdog->WDKEY = 0x55;
00c73a:   56BF55DC    MOVB         *+XAR4[3], #0x55, UNC
67       wdog->WDKEY = 0xAA;
00c73c:   56BFAADC    MOVB         *+XAR4[3], #0xaa, UNC
69       DISABLE_PROTECTED_REGISTER_WRITE_MODE;
00c73e:   761A        EDIS

Which to me seems like the expected assembly code. In general my application is fairly generic; I'm not using any RTOS and the compiler flags seem fairly standard to my newbie eyes ;)

0 Trey German289 over 12 years ago in reply to Jonathan Lock

TI__Mastermind 22025 points

Jonathan, Hmmm...that's odd. I was using 6.1.3 as well and I tried the -O4 optimization setting but didn't see that compiled out. I'll try it again.

Either way it doesn't hurt to put the volatile in there, so I'll go ahead and file a bug to get this added.

Thanks,

C2000™︎ microcontrollers

C2000 microcontrollers forum

Issue in WDT motorware API for 2806x