Re: [RFC PATCH] LKMM: Add ctrl_dep() macro for control dependency

On Wed, Oct 13, 2021 at 05:01:04PM -0700, Paul E. McKenney wrote:

On Sun, Oct 10, 2021 at 04:02:02PM +0200, Florian Weimer wrote:

quoted

* Linus Torvalds:

quoted

On Fri, Oct 1, 2021 at 9:26 AM Florian Weimer [off-list ref] wrote:

quoted

Will any conditional branch do, or is it necessary that it depends in
some way on the data read?

The condition needs to be dependent on the read.

(Easy way to see it: if the read isn't related to the conditional or
write data/address, the read could just be delayed to after the
condition and the store had been done).

That entirely depends on how the hardware is specified to work.  And
the hardware could recognize certain patterns as always producing the
same condition codes, e.g., AND with zero.  Do such tests still count?
It depends on what the specification says.

What I really dislike about this: Operators like & and < now have side
effects, and is no longer possible to reason about arithmetic
expressions in isolation.

Is there a reasonable syntax that might help with these issues?

Yes, I know, we for sure have conflicting constraints on "reasonable"
on copy on this email.  What else is new?  ;-)

I could imagine a tag of some sort on the load and store, linking the
operations that needed to be ordered.  You would also want that same
tag on any conditional operators along the way?  Or would the presence
of the tags on the load and store suffice?

Here's a easy cop-out.  Imagine a version of READ_ONCE that is 
equivalent to:

	a normal READ_ONCE on TSO architectures,

	a load-acquire on more weakly ordered architectures.

Call it READ_ONCE_FOR_COND, for the sake of argument.  Then as long as 
people are careful to use READ_ONCE_FOR_COND when loading the values 
that a conditional expression depends on, and WRITE_ONCE for the 
important stores in the branches of the "if" statement, all 
architectures will have the desired ordering.  (In fact, if there are 
multiple loads involved in the condition then only the last one has to 
be READ_ONCE_FOR_COND; the others can just be READ_ONCE.)

Of course, this is not optimal on non-TSO archictecture.  That's why I 
called it a cop-out.  But at least it is simple and easy.

Alan Stern