Re: [avr-gcc-list] Re: [avr-libc-dev] [bug #29774] prologue/epilogue stack pointer manipulation not interrupt safe in XMega

[Matthew Vernon]

Ruud Vlaming <address@hidden> said:

> On Tuesday 08 June 2010, Matthew wrote:
> 
> > So here's what I have found so far. When executing SEI, one more 
> > instruction can execute before the first interrupt is serviced. When 
> > enabling interrupts by restoring the status register, interrupts can 
> > occur immediately. When writing to SPL, interrupts are automatically 
> > disabled until the next memory write (up to 4 cycles).
> Interesting info!
> 
> Can you provide the sources for that or did you test it yourself?
> And do these statements hold for xmega en mega alike?
> 
> > Therefore, the prologue and epilogue do not need to disable interrupts 
> > at all. A write to SPL followed by a write to SPH should be interrupt 
> > safe. I have not tested this yet.
> This seems hard to test, but is also very interesting.
> 
> Ruud
> 

The info about writing to SPL disabling interrupts is in the XMega A Manual
(doc8077.pdf) at the bottom of section 3.8. Note that the interrupt
architecture is NOT the same in the mega series.

The test case is quite simple with a JTAGICE. Set up an interrupt that is
always pending. The code will execute one opcode at a time between interrupts.
A breakpoint can be set at the line in the disassembly that you want to test
(such as restoring status register). Once at that point set a breakpoint
inside the interrupt and at the line after the opcode being tested. Getting
the interrupt first indicates that interrupts are not disabled.

--Matthew


--