Re: rcu_read_lock lost its compiler barrier
From: Paul E. McKenney
Date: Thu Jun 06 2019 - 08:31:41 EST
On Thu, Jun 06, 2019 at 05:28:55PM +0800, Herbert Xu wrote:
> On Thu, Jun 06, 2019 at 02:06:19AM -0700, Paul E. McKenney wrote:
> > Or is your point instead that given the initial value of "a" being
> > zero and the value stored to "a" being one, there is no way that
> > any possible load and store tearing (your slicing and dicing) could
> > possibly mess up the test of the value loaded from "a"?
> Exactly. If you can dream up of a scenario where the compiler can
> get this wrong I'm all ears.
I believe that this is safe in practice, as long as you exercise
constant vigilance. (OK, OK, I might be overdramatizing...)
I cannot immediately think of a way that the compiler could get this
wrong even in theory, but similar code sequences can be messed up.
The reason for this is that in theory, the compiler could use the
stored-to location as temporary storage, like this:
a = whatever; // Compiler uses "a" as a temporary
whatever = a;
a = 1; // Intended store
The compiler is allowed to do this (again, in theory and according to a
strict interpretation of the standard) because you haven't told it that
anything else is paying attention to variable "a". As a result, the
compiler is within its rights to use "a" as temporary storage immediately
prior to any plain C-language store to "a".
In practice, I don't know of any compilers that actually do this, nor
have I heard anyone suggesting that they might soon actually do this.
And even if they could, your example would still work because your example
doesn't care about anything other than zero and non-zero, so wouldn't
get confused by the compiler storing a temporary value of 42 or whatever.
> > > But I do concede that in the general RCU case you must have the
> > > READ_ONCE/WRITE_ONCE calls for rcu_dereference/rcu_assign_pointer.
> > OK, good that we are in agreement on this part, at least! ;-)
> Well only because we're allowing crazy compilers that can turn
> a simple word-aligned word assignment (a = b) into two stores.
In my experience, the insanity of compilers increases with time,