(Non-) Ctrl Dependency in litmus-tests.txt?

(Non-) Ctrl Dependency in litmus-tests.txt?

[Paul Heidekrüger]

Hi all,

I was going through litmus-tests.txt and came across the following:

> LIMITATIONS
> ===========
> 
> Limitations of the Linux-kernel memory model (LKMM) include:
> 
> 1.Compiler optimizations are not accurately modeled.  Of course,
> 	the use of READ_ONCE() and WRITE_ONCE() limits the compiler's
> 	ability to optimize, but under some circumstances it is possible
> 	for the compiler to undermine the memory model.  For more
> 	information, see Documentation/explanation.txt (in particular,
> 	the "THE PROGRAM ORDER RELATION: po AND po-loc" and "A WARNING"
> 	sections).
> 
> 	Note that this limitation in turn limits LKMM's ability to
> 	accurately model address, control, and data dependencies.
> 	For example, if the compiler can deduce the value of some variable
> 	carrying a dependency, then the compiler can break that dependency
> 	by substituting a constant of that value.
> 
> 	Conversely, LKMM sometimes doesn't recognize that a particular
> 	optimization is not allowed, and as a result, thinks that a
> 	dependency is not present (because the optimization would break it).
> 	The memory model misses some pretty obvious control dependencies
> 	because of this limitation.  A simple example is:
> 
> 		r1 = READ_ONCE(x);
> 		if (r1 == 0)
> 			smp_mb();
> 		WRITE_ONCE(y, 1);
> 
> 	There is a control dependency from the READ_ONCE to the WRITE_ONCE,
> 	even when r1 is nonzero, but LKMM doesn't realize this and thinks
> 	that the write may execute before the read if r1 != 0.  (Yes, that
> 	doesn't make sense if you think about it, but the memory model's
> 	intelligence is limited.)

I'm unclear as to why the documentation sees a control dependency from
the READ_ONCE() to the WRITE_ONCE() here.

Quoting from explanation.txt:
> Finally, a read event and another memory access event are linked by a
> control dependency if the value obtained by the read affects whether
> the second event is executed at all.

Architectures might consider this control-dependent, yes, but since the
value of the if condition does not affect whether the WRITE_ONCE() is
executed at all, I'm not sure why this should be considered
control-dependent in LKMM? 

I might have another question about explanation.txt's definition of
control dependencies as per above, but will address it more thoroughly
in another email :-)

Many thanks,
Paul 

Re: (Non-) Ctrl Dependency in litmus-tests.txt?

[Alan Stern]

On Fri, Jun 03, 2022 at 04:12:37PM +0200, Paul Heidekrüger wrote:
> Hi all,
> 
> I was going through litmus-tests.txt and came across the following:
> 
> > LIMITATIONS
> > ===========
> > 
> > Limitations of the Linux-kernel memory model (LKMM) include:
> > 
> > 1.Compiler optimizations are not accurately modeled.  Of course,
> > 	the use of READ_ONCE() and WRITE_ONCE() limits the compiler's
> > 	ability to optimize, but under some circumstances it is possible
> > 	for the compiler to undermine the memory model.  For more
> > 	information, see Documentation/explanation.txt (in particular,
> > 	the "THE PROGRAM ORDER RELATION: po AND po-loc" and "A WARNING"
> > 	sections).
> > 
> > 	Note that this limitation in turn limits LKMM's ability to
> > 	accurately model address, control, and data dependencies.
> > 	For example, if the compiler can deduce the value of some variable
> > 	carrying a dependency, then the compiler can break that dependency
> > 	by substituting a constant of that value.
> > 
> > 	Conversely, LKMM sometimes doesn't recognize that a particular
> > 	optimization is not allowed, and as a result, thinks that a
> > 	dependency is not present (because the optimization would break it).
> > 	The memory model misses some pretty obvious control dependencies
> > 	because of this limitation.  A simple example is:
> > 
> > 		r1 = READ_ONCE(x);
> > 		if (r1 == 0)
> > 			smp_mb();
> > 		WRITE_ONCE(y, 1);
> > 
> > 	There is a control dependency from the READ_ONCE to the WRITE_ONCE,
> > 	even when r1 is nonzero, but LKMM doesn't realize this and thinks
> > 	that the write may execute before the read if r1 != 0.  (Yes, that
> > 	doesn't make sense if you think about it, but the memory model's
> > 	intelligence is limited.)
> 
> I'm unclear as to why the documentation sees a control dependency from
> the READ_ONCE() to the WRITE_ONCE() here.
> 
> Quoting from explanation.txt:
> > Finally, a read event and another memory access event are linked by a
> > control dependency if the value obtained by the read affects whether
> > the second event is executed at all.
> 
> Architectures might consider this control-dependent, yes, but since the
> value of the if condition does not affect whether the WRITE_ONCE() is
> executed at all, I'm not sure why this should be considered
> control-dependent in LKMM? 
> 
> I might have another question about explanation.txt's definition of
> control dependencies as per above, but will address it more thoroughly
> in another email :-)

You're right; strictly speaking this isn't a control dependency.  In 
fact it's not a dependency at all, just an ordering restriction that's 
connected with a conditional test.

If you would like to submit a patch updating the text, please feel free 
to do so.

Alan