Re: [Qemu-devel] [PATCH v5 00/10] aio_context_acquire/release pushdown, part 1

[Stefan Hajnoczi <stefanha@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 2991 bytes --]

On Thu, Jan 12, 2017 at 07:07:50PM +0100, Paolo Bonzini wrote:
> This is the first step of pushing down the AioContext lock.  Bottom halves
> are already protected by their own lock, use it also for walking_bh
> and for the handlers list (including walking_handlers).  The (lock,
> walking_foo) pair is wrapped into the QemuLockCnt primitive.
> 
> The only difference from v3 is a smattering of tiny nice improvements
> to QemuLockCnt.
> 
> Paolo
> 
> v4->v5:
>         remove stray tabs [patchew]
> 
> v3->v4:
>         Avoid useless atomic_mb_read in non-futex lockcnt [Stefan]
>         Use atomic_read in qemu_lockcnt_count [Stefan]
>         Tweak comment for qemu_lockcnt_cmpxchg_or_wait [Fam]
>         Use if/else in qemu_lockcnt_dec_and_lock [Fam]
>         Comment QEMU_LOCKCNT_STATE_* definitions [Fam]
> 
> v2->v3:
>         fix missing unlock
> 
> v1->v2:
>         rebased on top of polling patches
>         moved documentation to include/
>         new patch "aio-posix: split aio_dispatch_handlers out of aio_dispatch"
>         replaced atomic_read with atomic_mb_read in qemu_lockcnt_inc
>         replaced futex* function and macro names with qemu_futex*
>         removed atomic access to node->pfd.revents (because there is no
>           concurrent aio_poll anymore after the BDRV_POLL_WHILE series)
>         left "aio: push aio_context_acquire/release down to dispatching"
>           to a later series
> 
> Paolo Bonzini (10):
>   aio: rename bh_lock to list_lock
>   qemu-thread: introduce QemuLockCnt
>   aio: make ctx->list_lock a QemuLockCnt, subsuming ctx->walking_bh
>   qemu-thread: optimize QemuLockCnt with futexes on Linux
>   aio-posix: split aio_dispatch_handlers out of aio_dispatch
>   aio: tweak walking in dispatch phase
>   aio-posix: remove walking_handlers, protecting AioHandler list with
>     list_lock
>   aio-win32: remove walking_handlers, protecting AioHandler list with
>     list_lock
>   aio: document locking
>   async: optimize aio_bh_poll
> 
>  aio-posix.c                 | 116 +++++++------
>  aio-win32.c                 |  83 +++++----
>  async.c                     |  45 ++---
>  docs/lockcnt.txt            | 277 +++++++++++++++++++++++++++++++
>  docs/multiple-iothreads.txt |  13 +-
>  include/block/aio.h         |  38 ++---
>  include/qemu/futex.h        |  36 ++++
>  include/qemu/thread.h       | 112 +++++++++++++
>  util/Makefile.objs          |   1 +
>  util/lockcnt.c              | 397 ++++++++++++++++++++++++++++++++++++++++++++
>  util/qemu-thread-posix.c    |  35 +---
>  util/qemu-thread-win32.c    |   2 +-
>  util/trace-events           |  10 ++
>  13 files changed, 1006 insertions(+), 159 deletions(-)
>  create mode 100644 docs/lockcnt.txt
>  create mode 100644 include/qemu/futex.h
>  create mode 100644 util/lockcnt.c
> 
> -- 
> 2.9.3
> 

Thanks, applied to my block tree:
https://github.com/stefanha/qemu/commits/block

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]