RFR(M): 8154736: enhancement of cmpxchg and copy_to_survivor for ppc64
aph at redhat.com
Tue Nov 1 09:44:15 UTC 2016
On 31/10/16 21:30, David Holmes wrote:
> On 31/10/2016 7:32 PM, Andrew Haley wrote:
>> On 30/10/16 21:26, David Holmes wrote:
>>> On 31/10/2016 4:36 AM, Andrew Haley wrote:
>>>> And, while we're on the subject, is memory_order_conservative actually
>>>> defined anywhere?
>>> No. It was chosen to represent the current status quo that the Atomic::
>>> ops should all be (by default) full bi-directional fences.
>> Does that mean that a CAS is actually stronger than a load acquire
>> followed by a store release? And that a CAS is a release fence even
>> when it fails and no store happens?
> Yes. Yes.
> // All of the atomic operations that imply a read-modify-write
> // action guarantee a two-way memory barrier across that
> // operation. Historically these semantics reflect the strength
> // of atomic operations that are provided on SPARC/X86. We assume
> // that strength is necessary unless we can prove that a weaker
> // form is sufficiently safe.
Mmmm, but that doesn't say anything about a CAS that fails. But fair
enough, I accept your interpretation.
> But there is some contention as to whether the actual implementations
> obey this completely.
Linux/AArch64 uses GCC's __sync_val_compare_and_swap, which is specified
"full barrier". That is, no memory operand is moved across the
operation, either forward or backward. Further, instructions are
issued as necessary to prevent the processor from speculating loads
across the operation and from queuing stores after the operation.
... which reads the same as the language you quoted above, but looking
at the assembly code I'm sure that it's really no stronger than a seq
cst load followed by a seq cst store.
I guess maybe I could give up fighting this and implement all AArch64
CAS sequences as
CAS(seq_cst); full fence
or, even more extremely,
full fence; CAS(relaxed); full fence
but it all seems unreasonably heavyweight.
>> And that a conservative load is a *store* barrier?
> Not sure what you mean. Atomic::load is not a r-m-w action so not
> expected to be a two-way memory barrier.
More information about the hotspot-gc-dev