main responsible for fixing the Aarch64 race window we've seen.
After thinking a bit about the patches I had backported... I decided to "KISS" so I have kept current AIO logic intact - with existent, or not, BUGS - but did implement the stronger atomics, needed by Aarch64 and the extra barriers.
I've made some tests with and without i/o threads and it worked well.
Suggestion for next steps:
- Do a full battery of tests using your regression tests
(including the io-threaded test this time)
- Try to reproduce the Aarch64 issue (from LP: #1805256) without this patch
PROBLEM: I dont think we ever did reproduce this with Bionic's version.
- Decide (based on reproducer - or not) if you want to go on with this SRU.
Hello Christian,
This is a very clean backport of the commit:
commit 5710a3e09f
Author: Paolo Bonzini <email address hidden>
Date: Tue Apr 7 14:07:46 2020
async: use explicit memory barriers
...
main responsible for fixing the Aarch64 race window we've seen.
After thinking a bit about the patches I had backported... I decided to "KISS" so I have kept current AIO logic intact - with existent, or not, BUGS - but did implement the stronger atomics, needed by Aarch64 and the extra barriers.
I've made some tests with and without i/o threads and it worked well.
Suggestion for next steps:
- Do a full battery of tests using your regression tests
(including the io-threaded test this time)
- Try to reproduce the Aarch64 issue (from LP: #1805256) without this patch
PROBLEM: I dont think we ever did reproduce this with Bionic's version.
- Decide (based on reproducer - or not) if you want to go on with this SRU.
Let me know your thoughts, please.
I'm uploading to the PPA now...