a qustion in portENTER_SWITCHING_ISR in at91s

aresqi wrote on Monday, June 26, 2006:

I have a question in portENTER_SWITCHING_ISR in at91sam7s with gcc port. Please help me.

in the macro


________asm volatile ( "SUB        R11, LR, #4" );                           

why do we need the R11 – frame pointer? I think compile will use fp automatively, is it right?
Or we use it just want to save some IRQ stack space??

nobody wrote on Monday, June 26, 2006:

This is what I think:

The frame pointer switch is only required because the demo uses "naked" interrupt service routines. 

If the function was not naked then the compiler would move the stack pointer to make room for the local variables.  Variable are than accessed using an offset from the frame pointer.

As the function is naked the stack never gets adjusted, and the stack frame never setup.  This means the offsets from the frame pointer will clobber data.

The code you highlight allows the task stack to be used for local storage instead, which can be used safely.  Offsets are then into unused parts of the task stack.  The stack still uses IRQ stack, so nothing gets corrupted.

Take a look at the STR9 port files.  This saves the context before the VIC is read.  I think this could be a better solution but has the overhead of the context being saved for every IRQ even if it is not going to cause a context switch.