On Tue, 22 Nov 2022 09:38:35 GMT, Maurizio Cimadamore <mcimadam...@openjdk.org>
wrote:
>> src/jdk.sctp/unix/classes/sun/nio/ch/sctp/SctpMultiChannelImpl.java line 590:
>>
>>> 588: int pos)
>>> 589: throws IOException {
>>> 590: try (var guard = NIO_ACCESS.acquireScope(bb)) {
>>
>> Why was the old code not using reachability fences? Bug or feature?
>
> I see that there's a subsequent buffer call if `n > 0`, so that's probably
> why the fence was skipped? (I also assume that the code calling this method
> will access the buffer before/after, so reachability is never truly an issue
> - but for session-backed buffers this needs fixing).
>
> Also, stepping back, I note how, if `receive0` was a native call using
> Linker, perhaps we wouldn't need all this manual address computation - we'd
> just get a memory segment slice from the buffer and pass that to the handle
> (which will perform the correct liveness check). E.g. maybe a better long
> term solution would be to panama-ize this code?
Yes, once the memory/linker APIs are permanent then the SCTP implementation
would be a good candidate to redo.
-------------
PR: https://git.openjdk.org/jdk/pull/11260
