Re: [avr-gcc-list] generic queue library for AVR GCC?

[E. Weddington]

Rusty Wright wrote:

> Eric, wouldn't it be safer to disable interrupts before you copy the
> SREG?  I.e., isn't it possible for an interrupt to occur between the
> copy of SREG and the cli?  
Yes, it's possible. But no, you don't want to disable interrupts before 
saving SREG.

The whole point of saving SREG and restoring SREG is to restore the 
state of global interrupts. Remember, when you clear global interrupts 
(CLI) it changes a flag (bit) in the SREG register. This is why you want 
to save the SREG register *before* doing the cli(). When you restore the 
SREG register, you restore the state of the global interrupts after the 
"critical section" is done, *regardless of whether the global interrupts 
were enabled or disabled*.

For example, you do a critical section like:

cli();
// critical section
sei();

However, this enables global interrupts after the critical section. The 
problem is: were global interrupts enabled, or disabled before the cli() 
and the critical section? Saving and restoring the status register 
handles both cases:

sreg = SREG;
cli();
// critical section
SREG = sreg;

It doesn't matter whether there is an interrupt between the sreg = SREG; 
and the cli();. The point is that the state, no matter what it is, has 
been preseved and subsequently restored.


> Also, regarding the push/pop, most c compilers store local variables
> on the stack so with your method you'll end up pushing and popping
> anyhow.
You're absolutely right. I misspoke. My main point was really that one 
does not have to do this in inline assembly, that it can be done just as 
well in pure C.

Eric

PS, Don't forget to respond to the *list* and not just personally.