On Mon, Nov 10, 2025 at 04:48:35PM +0100, Hanna Czenczek wrote:
> Hi,
>
> See the v1 cover letter for a general overview:
>
> https://lists.nongnu.org/archive/html/qemu-block/2025-10/msg00501.html
>
> Changes in v2:
> - Kept `.ret = EINPGORESS`-style initializations where they already had
> been (curl, nvme)
> - Dropped trivial BH waking code (i.e. which can be directly replaced by
> aio_co_wake()) in iscsi, nfs, nvme
> - curl: Yield in curl_do_preadv() (former curl_setup_preadv()) and
> curl_find_buf() instead of returning whether curl_co_preadv() has to
> yield or not
> - nvme: Added a patch that annotates some functions (primarily BHs and
> CBs) with which AioContext they (must) run in
> - qcow2 cache-cleaning timer: Run the timer as a coroutine instead of in
> a timer CB; use a CoQueue to await it exiting instead of polling
> (well, we still need to poll in case we don’t run in a coroutine, but
> that’s standard procedure, I believe)
> - The need to initialize the CoQueue showed that there is a code path
> in qcow2 that doesn’t initialize its CoMutex. Added a patch to do
> that.
> - Also added a patch to have the timer use realtime instead of virtual
> time.
>
>
> git backport-diff against v1:
>
> Key:
> [----] : patches are identical
> [####] : number of functional differences between upstream/downstream patch
> [down] : patch is downstream-only
> The flags [FC] indicate (F)unctional and (C)ontextual differences, respectively
>
> 001/19:[----] [--] 'block: Note on aio_co_wake use if not yet yielding'
> 002/19:[----] [--] 'rbd: Run co BH CB in the coroutine’s AioContext'
> 003/19:[0019] [FC] 'iscsi: Run co BH CB in the coroutine’s AioContext'
> 004/19:[0022] [FC] 'nfs: Run co BH CB in the coroutine’s AioContext'
> 005/19:[0060] [FC] 'curl: Fix coroutine waking'
> 006/19:[----] [--] 'gluster: Do not move coroutine into BDS context'
> 007/19:[----] [--] 'nvme: Kick and check completions in BDS context'
> 008/19:[0038] [FC] 'nvme: Fix coroutine waking'
> 009/19:[down] 'nvme: Note in which AioContext some functions run'
> 010/19:[----] [--] 'block/io: Take reqs_lock for tracked_requests'
> 011/19:[down] 'qcow2: Re-initialize lock in invalidate_cache'
> 012/19:[0145] [FC] 'qcow2: Fix cache_clean_timer'
> 013/19:[down] 'qcow2: Schedule cache-clean-timer in realtime'
> 014/19:[----] [--] 'ssh: Run restart_coroutine in current AioContext'
> 015/19:[----] [--] 'blkreplay: Run BH in coroutine’s AioContext'
> 016/19:[----] [--] 'block: Note in which AioContext AIO CBs are called'
> 017/19:[----] [--] 'iscsi: Create AIO BH in original AioContext'
> 018/19:[----] [--] 'null-aio: Run CB in original AioContext'
> 019/19:[----] [--] 'win32-aio: Run CB in original context'
>
>
> Hanna Czenczek (19):
> block: Note on aio_co_wake use if not yet yielding
> rbd: Run co BH CB in the coroutine’s AioContext
> iscsi: Run co BH CB in the coroutine’s AioContext
> nfs: Run co BH CB in the coroutine’s AioContext
> curl: Fix coroutine waking
> gluster: Do not move coroutine into BDS context
> nvme: Kick and check completions in BDS context
> nvme: Fix coroutine waking
> nvme: Note in which AioContext some functions run
> block/io: Take reqs_lock for tracked_requests
> qcow2: Re-initialize lock in invalidate_cache
> qcow2: Fix cache_clean_timer
> qcow2: Schedule cache-clean-timer in realtime
> ssh: Run restart_coroutine in current AioContext
> blkreplay: Run BH in coroutine’s AioContext
> block: Note in which AioContext AIO CBs are called
> iscsi: Create AIO BH in original AioContext
> null-aio: Run CB in original AioContext
> win32-aio: Run CB in original context
>
> block/qcow2.h | 5 +-
> include/block/aio.h | 15 ++++
> include/block/block_int-common.h | 7 +-
> block/blkreplay.c | 3 +-
> block/curl.c | 45 +++++++---
> block/gluster.c | 17 ++--
> block/io.c | 3 +
> block/iscsi.c | 63 ++++++--------
> block/nfs.c | 41 ++++-----
> block/null.c | 7 +-
> block/nvme.c | 113 ++++++++++++++++--------
> block/qcow2.c | 145 ++++++++++++++++++++++++-------
> block/rbd.c | 12 +--
> block/ssh.c | 22 ++---
> block/win32-aio.c | 31 +++++--
> 15 files changed, 347 insertions(+), 182 deletions(-)
>
> --
> 2.51.1
>
I reviewed the nvme changes and the core block layer changes:
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>