[Info-vax] Volatile, was: Re: yet another sys$qiow question

[Stephen Hoffman]

On 2015-08-21 12:16:14 +0000, David Froble said:

> The problem arises when rather poor coding practice is used.  I really 
> would hope that none of the manuals or examples suggests reading the 
> IOSB prior to the operation being signaled as complete.
> 
> I understand what you're saying the volatile command does, but so far I 
> don't think I've seen an example that needs it, except for what I'd 
> consider improper coding.

I'm not entirely certain that the compilers will always do the right 
thing here — always, and in the absence of a volatile declaration — 
even when you're using what are currently considered proper coding 
practices — there's definitely some code-smell here, and it may well be 
the case that some compiler keeps the IOSB value in a register, meaning 
that the out-of-band change to the IOSB is not made visible to the 
application code.

The compiler has no insight into the asynchronous sharing of the IOSB 
absent that volatile type, and accordingly the compiler might decide to 
generate (entirely correct) code that keeps the IOSB value in a 
register, once the data been read into the register from the IOSB.

I'd hope that a out-of-compilation-scope function call — such as a 
system service call — that referencing the address of shared data would 
force a re-load of the data into the register, but I don't know that.

Registers are valuable resources in present architectures, so the 
generated code doesn't usually have a value sitting in a register for 
very long.  But register-caching is permissible in C and most other 
languages, absent the language-specific asynchronous / volatile syntax. 
  This is analogous to the while-spin loop that has been posted here, 
but it's a whole lot less visible.




-- 
Pure Personal Opinion | HoffmanLabs LLC