From: Peter Zijlstra
> Sent: 11 July 2018 12:10
..
> Adding SYNC to WRITE_ONCE()/atomic* will hurt performance lots and will
> ultimately not guarantee anything more; and as Will said, keep you
> chasing dragons where people forgot to use WRITE_ONCE() where they maybe
> should've.
Also WRITE_ONCE() is there to solve an entirely different problem.
If you have a function that does:
<lots of code without any function calls>
foo->bar = 1;
the compiler is allowed to write other (unrelated) values
to foo->bar in the generated code for <lots of code>.
A long time ago I used a compiler that managed to convert:
if (foo->bar == 2)
foo->bar = 3;
into:
if (foo->bar == 2) {
foo->bar = 0;
foo->bar = 3;
}
When an interrupt read the value 0 a lot of linked list got screwed up.
WRITE_ONCE() is there to ensure that doesn't happen.
(In my case 'foo' was a 2-bit wide bitfield, and I suspect you
can't use WRITE_ONCE() on bitfields.)
David
