Am 9/30/2024 um 11:27 AM schrieb Alan Huang:
2024年9月30日 17:15，Alan Huang <mmpgouride@xxxxxxxxx> 写道：
2024年9月30日 16:57，Jonas Oberhauser <jonas.oberhauser@xxxxxxxxxxxxxxx> 写道：
Am 9/29/2024 um 12:26 AM schrieb Alan Huang:
2024年9月28日 23:55，Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx> wrote：
The motivation for introducing ptr_eq() is indeed because the
compiler barrier is not sufficient to prevent the compiler from
using one pointer instead of the other.
barrier_data(&b) prevents that.
I don't think one barrier_data can garantuee preventing this, because right after doing the comparison, the compiler still could do b=a.
In that case you would be guaranteed to use the value in b, but that value is not the value loaded into b originally but rather the value loaded into a, and hence your address dependency goes to the wrong load still.
After barrier_data(&b), *b will be loaded from memory, you mean even if *b is loaded from memory, the address dependency goes to the wrong load still?
Sorry, *b should b.
That's exactly what I meant to say. In my understanding, it can happen
like this:
a = READ_ONCE(*p);
...
b = READ_ONCE(*p);
if (a == b) {
b = a; // inserted by compiler
barrier_data(&b);
foo(*b); // compiler definitely use the current value in b
}
In the end, the address dependency is from the first load, and the CPU
can speculatively (with register renaming, forwarding etc) execute
a = READ_ONCE(*p);
b2 = a; // speculatively
tmp = load *b2 // speculatively
b1 = READ_ONCE(*p);
if (a == b1) { // confirmed
foo(tmp);
}
best wishes,
jonas
