Re: MPS prstack

[Helmut Eller <eller.helmut@gmail.com>]

On Mon, May 27 2024, Po Lu wrote:

> Helmut Eller <eller.helmut@gmail.com> writes:
>
>> Yes.  I think 
>>
>>   prstack.stack[prstack.sp] = e;
>>   __atomic_store_n (&prstack.sp, prstack.sp + 1, __ATOMIC_SEQ_CST);
>>
>> is quite sufficient.
>
> How so?  This doesn't guarantee that MPS cannot pause the thread after E
> is saved into prstack but before prstack.sp is incremented,

Right.  But we are sure that prstack.sp is not incremented before E is
saved.  And scan_prstack doesn't care what is beyond prstack.sp.

> or that E
> will remain on the control stack or in the register file until it is.

Yes.  But that's another issue.  That could be solved by making
print_stack_push a non-inline function that takes a reference to E as
argument.