Re: [PATCH] hw/timer: fix int underflow

[Paolo Bonzini]

On 1/14/25 11:14, Peter Maydell wrote:
> > So, to my opinion, explicit cast to "long long" is necessary
> > here to get the expected behavior.
>
> I wasn't saying the existing code was necessarily correct,
> or that your proposed change was necessarily wrong.
> I was saying your patch didn't come with any analysis of
> what the actual hardware behaviour is, which is
> how you would determine whether the fix you propose
> is the correct one, or if it should be some other
> change instead. (Some of my response was trying to
> provide some of that analysis.)
I would say that the patch is more or less obvious in the sense that the 
intent of the code is to model a counter (which counts up by 1 and 
cannot miss a step) with a timer within a process that could be 
preempted.  You probably will not find an answer in the datasheet, other 
than by analyzing the signal diagrams, because this scenario simply 
cannot exist with a hardware counter.

However, the patch has another problem, which is that it can cause a 
stack overflow at imx_gpt_compute_next_timeout(s, event).

Paolo