Re: [Amforth] AVR8: The SREG:I-flag is an Unreliable Indicator of Interrupt Status

[<ho...@tj...>]

On 2026-02-08 15:38, Martin Nicholas via Amforth-devel wrote:
> Firstly, here's the docs:
> http://amforth.sourceforge.net/TG/AVR8.html#interrupts
> http://amforth.sourceforge.net/TG/recipes/Interrupt-Critical-Section.html
> 
> Source files:
> amforth-6.9/avr8/lib/hardware/int-q.frt
> etc.
> 
> Because of the way AmForth handles interrupts, the I-flag can be
> cleared (disabling the interrupts globally) at any time during the
> execution of a Forth primative. It can change to zero between any two
> M/C code instructions.
> 
> So "int?" has to check both the I-flag and the value of "isrflag"
> (T-flag) to come to a valid conclusion. "-int" cannot just perform a
> CLI instruction as the next NEXT may process a pending (pending before
> the CLI was executed) interrupt and re-enable them. Etc.
> 
> Reliable fetch of interrupt status.
> Checks for pending interrupts.
> 	I	T***	int?	Remarks
> 	=	=	====	=======
> 	0	0	 0	Interrupts disabled by the program.
> 	0	1	-1	Interrupt pending*
> 	1	0	-1	Normal operation.
> 	1	1	-1	Shouldn't ever be**
> 
>   *	This will be serviced by the next NEXT and SREG:I will be
> 	set once more.
>  **	The interrupt will still get serviced, provided it isn't
> 	overwritten.
> ***	The T-flag seems to be reserved as a holder for the ISR
> 	status, but currently it isn't used. Location "isrflag" is used
> 	instead and is a much slower.
> 
> Special care has to be taken to save and restore the I-flag. I have
> written macro "no_ints". It uses MCUCR:IVCE to
> disable interrupts without affecting the I-flag - see the docs.
> 
> Correctly preserve the interrupt status in @0:
> 
> .macro no_ints
> 	in	@0, MCUCR
> 	ori	@0, (1<<IVCE)
> 	out	MCUCR, @0
> 	in	@0, SREG
> 	cli
> .endmacro
> 
> Later Atmel cores automatically disable interrupts for a time if you
> fiddle with the SP register, so "no_ints" is not required here. Maybe
> there are other operations that do this too.

Hi Martin,

Thank you. Added to the list.

A preliminary check of the archives shows that the way interrupts were 
handled by the VM changed between releases 6.4 and 6.5

In 6.4 there is
brts DO_INTERRUPT

In 6.5 there is
cp isrflag, zerol
brne DO_INTERRUPT

The release notes for 6.5 [1] state that there was a bug in the 
interrupt handling on the avr8. Erich discovered and fixed the bug, and 
so may be able to help. Current avr8 documentation references the T flag 
but does not reference isrflag, so this is something that will need to 
be updated.

The file avr8//drivers/generic-isr.asm shows considerable related 
changes between the releases but I need more time to study it.

Best wishes,
Tristan

[1] https://sourceforge.net/p/amforth/mailman/message/35814877/