More questions about "asm volatile"

[David Brown]

On 11/04/2011 13:16, Ingo Rohloff wrote:
> Hello,
>
> I just read a lot of stuff about what "asm volatile" means, or to be
> more precise what it DOES NOT mean. Unfortunately
> "http://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html" explains a lot
> about what "asm volatile" does NOT mean, but not about what it
> means.
>
> So here are my questions:
>
> 1st Question: The document mentioned above says: "Similarly, you
> can't expect a sequence of volatile asm instructions to remain
> perfectly consecutive. If you want consecutive output, use a single
> asm."
>
> Does that mean that the order of "volatile asm" instructions might be
> swapped, or does this just mean that the different "volatile asm"
> instructions might be interspersed with other instructions ?
>
> If there is a guarantee that "volatile asm" instructions CANNOT be
> swapped, then I think the "Extended-Asm.html" document should state
> this.
>
> 2nd Question: The document mentioned above says "Note that even a
> volatile asm instruction can be moved relative to other code,
> including across jump instructions. " and there are a number of old
> discussions about that behavior.
>
> Unfortunately neither the document above nor the discussions I have
> read make any statement if this just means that "volatile" has no
> meaning or if there are circumstances in which there still is SOME
> guarantee.
>
> What I am wondering in particular: If I have this code:
>
>
> extern volatile int *gPtr; void foo() { int v; v=*gPtr; asm volatile
> (<Something>  ); *gPtr=5; }
>
> Is it possible that the "asm volatile" is swapped with the accesses
> to the "volatile" pointer, or does the "asm volatile" interact with
> "volatile" pointer accesses, so that it is guaranteed that a "asm
> volatile" will not be swapped with a "volatile" pointer access ? Of
> course I am interested in the general case: Is there a guarantee that
> "asm volatile" will stay in order with accesses to volatile pointers
> ?
>
> Again if this is guaranteed, the "Extended-Asm.html" document
> mentioned above should say so.
>
> with best regards Ingo
>

The point of "volatile" is that volatile accesses, and volatile asm, 
cannot be re-ordered with respect to other /volatile/ accesses. 
Non-volatile accesses can be moved around them, but code such as you 
wrote above will be safe.