CodeWarrior startup routine and S12Z ECC RAM initialization

[kef2]

Hi,

After migration from Freescale to NXP it became unclear were to report bugs or concerns. Is Community really right place for this?

ECCSTAT RDY bit explanation from S12ZVC RM:

ECC Ready— Shows the status of the ECC module.
0 Internal SRAM initialization is ongoing, access to the SRAM is disabled
1 Internal SRAM initialization is done, access to the SRAM is enabled

CW startup routine seems not waiting for RDY==1. But setting reset vector to iaprstvect() routine and power cycling board I see 0x0070 loop count @ 0x3000. :

void _Startup(void);

#pragma NO_RETURN

void iaprstvect(void)

{

__asm{

         CLR D2

rdlp:   INC D2

         BRCLR.B ECCSTAT, #0, rdlp

         ST D2, 0x3000  // save loop count

         JMP _Startup

}

Looks like it takes quite a lot at power on to initialize ECC RAM. Long enough for default startup routine to set stack pointer and jump (BSR) to DoZeroOut() routine before ECC RAM initialization completes! Are there chances for big failure like reading wrong address from stack when returning back from DoZeroOut() to _Startup()?

[danielmartynek]

Hi Edward,

It doesn't matter whether you post it here or create a ticket.

But in this case, it doesn't seem to be a bug.

I tried the code you posted right in the _Startup function like this on S12ZVCA19 devkit

__EXTERN_C void _Startup(void) {
__asm{
       CLR D2
 rdlp: INC D2
       BRCLR.B ECCSTAT, #0, rdlp
       ST D2, 0x3000 // save loop count
}
 
__asm{
       LD S, #__SEG_END_SSTACK-1 /* initialize SP */
#ifdef __ALIGN_STACK_OPT__
 TFR S, D6 /* align SP to 4*/
 AND D6, #-4
 TFR D6, S
#endif 
}
 DoZeroOut();
 DoCopyDown();
#ifdef __cplusplus
 __static_init();
#endif
 main();
} ‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍

The counter showed 0x70.

But if the code is modified to write the SRAM while RDY == 0, the execution stalls at the ST instruction and the SRAM is written once the SRAM has become ready.

__EXTERN_C void _Startup(void) {
 __asm{
       CLR D2
 rdlp: INC D2
       ST D2, 0x3000 // save loop count 
       BRCLR.B ECCSTAT, #0, rdlp 
}
 
 __asm {
       LD S, #__SEG_END_SSTACK-1 /* initialize SP */
#ifdef __ALIGN_STACK_OPT__
 TFR S, D6 /* align SP to 4*/
 AND D6, #-4
 TFR D6, S
#endif 
}
 DoZeroOut();
 DoCopyDown();
#ifdef __cplusplus
 __static_init();
#endif
 main();
}‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍

Regards,

Daniel

[kef2]

Hi Daniel,

Thank you! I tried enabling TIM0 and using TCNT to measure RDY delay time. Waiting for RDY=1 with leading write to RAM vs no leading write to RAM produces about the same reading in TCNT and different loop counter. Yes, it looks like a stall. Clearly S12ZVC RM is wrong and should be fixed. This is what we have:

7.3.4 Memory Initialization

To avoid spurious ECC error reporting, memory operations that allow a read before a first write (like the

read-modify-write operation of the unaligned access) require that the memory contains valid ECC values

before the first read-modify-write access is performed. The ECC module provides logic to initialize the

complete memory content with zero during the power up phase. During the initialization process the access

to the SRAM is disabled and the RDY status bit is cleared. If the initialization process is done, SRAM

access is possible and the RDY status bit is set.

Access disabled(ignored) vs stalled are not the same things. Bold part sounds as if initialization may never complete.

Regards,

Edward

[danielmartynek]

Hi Edward,

I agree, it will be reported.

Thank you,

BR, Daniel