ceph: remove atomic_t #14502
ceph: remove atomic_t #14502
Conversation
chardan
force-pushed
the
jfw-wip-halflife_atomic_t
branch
from
46e83e0
to
ce13d34
Compare
src/include/Spinlock.h
Outdated
| { | ||
| return pthread_spin_destroy(&l->lock); | ||
| lock.clear(std::memory_order_release); |
There was a problem hiding this comment.
again, do we have a micro benchmark result to compare the performance with and without the change?
There was a problem hiding this comment.
Not yet. I'll see what I can do.
There was a problem hiding this comment.
It's quick-n-dirty, but here's a start. Please feel free to kick the tires:
https://github.com/chardan/ceph-spinlock-minibench
|
Hmm, like other cleanups, this will be easier if we break it up into a few different commits and PRs targetting different subsystems. They'll have to go through different teuthology suites and possibly get rebased to avoid conflicts with other PRs. Can you break this into osd/, mds/, librbd/, rgw/, and then everything else (common)? The last patch that removes the old atomic.h will come separately and last. |
chardan
force-pushed
the
jfw-wip-halflife_atomic_t
branch
from
ce13d34
to
765330d
Compare
|
@liewegas If there are too many separate commits, I can also squish some of them together. |
…f spin locks. Removes simple_spin.cc, simple_spin.h; moves spinlock_t -> std::atomic_flag. Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
This commit is a "catch all". Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
…omic Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
|
@chardan commits are fine. Can you break it into a few separate PRs though so that one conflict in say rbd doesn't block everything else from test/merge? Thanks! |
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
Signed-off-by: Jesse Williamson <jwilliamson@suse.de>
chardan
force-pushed
the
jfw-wip-halflife_atomic_t
branch
from
765330d
to
8edbb61
Compare
|
Further adventures continue at: |
This is intended to remove atomic_t from the codebase, replacing uses of
it with the C++ standard library. It should add no features and cause no
regressions. The "include/atomic.h" header file and "common/simple_spin.h"
are removed.
Please be sure to inspect default values that I've assigned to atomics-- I
occasionally guessed-- and check for any undefined behavior I may have
introduced. I'm also not entirely sure if the removal of InstanceReplayer.h
is something I may have done through misuse of git, or something expected.
Guidance appreciated.
I've taken a pretty mechanical approach here-- for example, replacing
read() with load(). I wouldn't mind using std::atomic<>'s type coercion for a
more natural syntax (x = y as opposed to x = y.load()).
Another refinement might be to apply a template to help with situations
where we compare-and-swap, but intended to discard the swapped-to value (
avoiding having to add exposed variables for compare_exchange_*).
Please let me know if you'd like to see either or both of these included
as additional commits.
Thank you!