AT91SAM7S64 IAR port

jra01 wrote on Tuesday, September 20, 2005:

I discovered a problem with crashing system. The cause showed up to be that IAR expects stack to be initiated to 0, but freeRTOS initates to 01010101, 02020202 …

Therefore I have made a change to port.c, in "pxPortInitialiseStack":

    *pxTopOfStack = ( portSTACK_TYPE ) 0xaaaaaaaa;    /* R14 */
    pxTopOfStack–;   
    *pxTopOfStack = ( portSTACK_TYPE ) pxOriginalTOS; /* Stack used when task starts goes in R13. */
    pxTopOfStack–;
    for (i = 0; i < 12; ++i) {
        *pxTopOfStack = ( portSTACK_TYPE ) 0x0;    /* R12…R1 */
        pxTopOfStack–;   
    }

After that it works.

Regards

Jokke

rtel wrote on Tuesday, September 20, 2005:

This is definitely not the case.  When a task starts it does not make any assumptions as to the stack or register states.

I suspect you are calling main() in the wrong processor mode.  main() must be called in supervisor mode.  Are you using the startup code that came with the demo project?

jra01 wrote on Tuesday, September 20, 2005:

This has nothing to do with main. In a normal task I discovered (looking at the assembler listing) that the compiler places a read in a vector, using an index that is in some cases still not initialized. This works fine as long as the index is 0 until it is used, but I got problems with reads out of defined memory, since, in this case, R5 was 0x05050505, as is initialised in port.c.
Maybe this is a compiler error, but my guess was that they made this assumption.
The code is a state machine, and it might not be obvious enough for the compiler that it starts in a state where the index is set, before moving to the state where it is used. On the other hand I do not get any warning of "used before set"(maybe IAR does not give such warnings?).
I am using the startup from the demo project, and I made a lot of versions of the software before this came up.
I guess I will have to send you the code and the assembler listing before we reach consensus on this…
Regards
Jokke

rtel wrote on Tuesday, September 20, 2005:

If you could send me the smallest project possible that exhibits this behaviour I would be very grateful!  Send it to the email address from the FreeRTOS WEB site Contacts page (r dot barry at fr…org).  Thanks.

rtel wrote on Tuesday, September 20, 2005:

From the received code it looks as if there could be a compiler error - I will post the results but might take a while.

jra01 wrote on Monday, September 26, 2005:

I upgraded the IAR tools to lates version (4.30), and there the compiler bug seems to be solved!
No problem with freeRTOS!
SOLVED