[Qemu-devel] MTTCG sync-up call summary 28/09/2015

[Alex Bennée]

Here follows a slightly abbreviated form of my notes for the sync-up
call. Please reply if anything is missing on unclear ;-)

- Alvise has posted v5 addressing Paolo's comments
  - bitmap addresses TLB size issues
  - consensus we can go with Alvise's and build on them with ideas from Emilo's 
series
  - Alvise will have a look through during re-basing to see if any potential 
conflicts
  - Hope to take best of both worlds approach with slow and fast path later
- Fred patch status
  - Waiting for latest re-base from Paolo
  - Signal Free TCG and linux-user fixes now merged upstream so re-base can be 
done there
- Paolo
  - Tree breakage AJB reported seems to be in merge-base but exacerbated by 
Paolo's MTTCG branch
    - upstream master seems much more stable, we shall see when Fred re-bases
    - Alex will hammer it when ready to check bug has disappeared
  - Outstanding work from Paolo's review:
    - missing breakpoint support
    - invalidating flag
    - locking on tb_get_next
  - Reverse playback may conflict a little when it arrives in tree
    - but still a week or so of getting into Paolo's tree
- Memory Barriers
  - Paper referenced by Emilo's patch set is good read 
[[https://www.cs.princeton.edu/research/techreps/TR-981-15][here]]
  - Alex will take Emilo's work for ARM on ARM testing
  - Would be nice to demo ARM on ARM/AArch64 in November
  - Alex plans to tests with all ARM barrier types by end of week
- Any work good for new people to ease in?
  - Look at breakpoints, is GDB stub working?
  - Benchmarking: perhaps get some independent numbers on both atomic 
implementations?

-- 
Alex Bennée