2 <util/atomic.h> Atomically and Non-Atomically Executed Code Blocks

Note:

The macros in this header file require the ISO/IEC 9899:1999 ("ISO C99") feature of for loop variables that are declared inside the for loop itself. For that reason, this header file can only be used if the standard level of the compiler (option --std=) is set to either c99 or gnu99.

The macros in this header file deal with code blocks that are guaranteed to be excuted Atomically or Non-Atmomically. The term "Atomic" in this context refers to the unability of the respective code to be interrupted.

These macros operate via automatic manipulation of the Global Interrupt Status (I) bit of the SREG register. Exit paths from both block types are all managed automatically without the need for special considerations, i. e. the interrupt status will be restored to the same value it has been when entering the respective block.

A typical example that requires atomic access is a 16 (or more) bit variable that is shared between the main execution path and an ISR. While declaring such a variable as volatile ensures that the compiler will not optimize accesses to it away, it does not guarantee atomic access to it. Assuming the following example:

There is a chance where the main context will exit its wait loop when the variable ctr just reached the value 0xFF. This happens because the compiler cannot natively access a 16-bit variable atomically in an 8-bit CPU. So the variable is for example at 0x100, the compiler then tests the low byte for 0, which succeeds. It then proceeds to test the high byte, but that moment the ISR triggers, and the main context is interrupted. The ISR will decrement the variable from 0x100 to 0xFF, and the main context proceeds. It now tests the high byte of the variable which is (now) also 0, so it concludes the variable has reached 0, and terminates the loop.

Using the macros from this header file, the above code can be rewritten like:

This will install the appropriate interrupt protection before accessing variable ctr, so it is guaranteed to be consistently tested. If the global interrupt state were uncertain before entering the ATOMIC_BLOCK, it should be executed with the parameter ATOMIC_RESTORESTATE rather than ATOMIC_FORCEON.

See Problems with reordering code for things to be taken into account with respect to compiler optimizations.