AIO context in userspace is represented as a simple ring buffer, which
can be consumed directly without entering the kernel, which obviously
can bring some performance gain. QEMU does not use timeout value for
waiting for events completions, so we can consume all events from
userspace.
Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Message-id: 1468931263-32667-2-git-send-email-roman.penyaev@profitbricks.com
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: qemu-devel@nongnu.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
It is generally not expected that io_submit() fails other than with
-EAGAIN, but corner cases like SELinux refusing I/O when permissions are
revoked are still possible. In this case, we shouldn't abort, but just
return an I/O error for the request.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1470741619-23231-1-git-send-email-kwolf@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Invoking io_setup(MAX_EVENTS) we ask kernel to create ring buffer for us
with specified number of events. But kernel ring buffer allocation logic
is a bit tricky (ring buffer is page size aligned + some percpu allocation
are required) so eventually more than requested events number is allocated.
From a userspace side we have to follow the convention and should not try
to io_submit() more or logic, which consumes completed events, should be
changed accordingly. The pitfall is in the following sequence:
MAX_EVENTS = 128
io_setup(MAX_EVENTS)
io_submit(MAX_EVENTS)
io_submit(MAX_EVENTS)
/* now 256 events are in-flight */
io_getevents(MAX_EVENTS) = 128
/* we can handle only 128 events at once, to be sure
* that nothing is pended the io_getevents(MAX_EVENTS)
* call must be invoked once more or hang will happen. */
To prevent the hang or reiteration of io_getevents() call this patch
restricts the number of in-flights, which is now limited to MAX_EVENTS.
Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1468415004-31755-1-git-send-email-roman.penyaev@profitbricks.com
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: qemu-devel@nongnu.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This has better performance because it executes fewer system calls
and does not use a bottom half per disk.
Originally proposed by Ming Lei.
[Changed #include "raw-aio.h" to "block/raw-aio.h" in win32-aio.c to fix
build error as reported by Peter Maydell <peter.maydell@linaro.org>.
--Stefan]
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1467650000-51385-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
squash! linux-aio: share one LinuxAioState within an AioContext
In practice the entry argument is always known at creation time, and
it is confusing that sometimes qemu_coroutine_enter is used with a
non-NULL argument to re-enter a coroutine (this happens in
block/sheepdog.c and tests/test-coroutine.c). So pass the opaque value
at creation time, for consistency with e.g. aio_bh_new.
Mostly done with the following semantic patch:
@ entry1 @
expression entry, arg, co;
@@
- co = qemu_coroutine_create(entry);
+ co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry2 @
expression entry, arg;
identifier co;
@@
- Coroutine *co = qemu_coroutine_create(entry);
+ Coroutine *co = qemu_coroutine_create(entry, arg);
...
- qemu_coroutine_enter(co, arg);
+ qemu_coroutine_enter(co);
@ entry3 @
expression entry, arg;
@@
- qemu_coroutine_enter(qemu_coroutine_create(entry), arg);
+ qemu_coroutine_enter(qemu_coroutine_create(entry, arg));
@ reentry @
expression co;
@@
- qemu_coroutine_enter(co, NULL);
+ qemu_coroutine_enter(co);
except for the aforementioned few places where the semantic patch
stumbled (as expected) and for test_co_queue, which would otherwise
produce an uninitialized variable warning.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Partial write most likely means that there is not space rather than
"something wrong happens". Thus it would be more natural to return
ENOSPC rather than EINVAL.
The problem actually happens with NBD server, which has reported EINVAL
rather then ENOSPC on the first error using its protocol, which makes
report to the user wrong.
Signed-off-by: Denis V. Lunev <den@openvz.org>
CC: Pavel Borzenkov <pborzenkov@virtuozzo.com>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
linux-aio uses a BH in order to make sure that the remaining completions
are processed even in nested event loops of completion callbacks in
order to avoid deadlocks.
There is no need, however, to have the BH overhead for the first call
into qemu_laio_completion_bh() or after all pending completions have
already been processed. Therefore, this patch calls directly into
qemu_laio_completion_bh() in qemu_laio_completion_cb() and cancels
the BH after qemu_laio_completion_bh() has processed all pending
completions.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
The raw-posix block driver actually supports byte-aligned requests now
on non-O_DIRECT images, like it already (and previously incorrectly)
claimed in bs->request_alignment.
For some block drivers this means that a RMW cycle can be avoided when
they write sub-sector metadata e.g. for cluster allocation.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
In order to use the modern byte-based .bdrv_co_preadv/pwritev()
interface, this patch switches raw-posix to coroutine-based interfaces
as a first step. In terms of semantics and performance, it doesn't make
a difference with the existing code whether we go from a coroutine to a
callback-based interface already in block/io.c or only in linux-aio.c
As there have been concerns in the past that this change may be a step
in the wrong direction with respect to a possible AIO fast path, the
old callback-based interface for linux-aio is left around and can be
reactivated when a fast path (e.g. directly from virtio-blk dataplane,
bypassing the whole block layer) is implemented.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Replace void* with an opaque LinuxAioState type.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Extract the handling of io_plug "depth" from linux-aio.c and let the
main bdrv_drain loop do nothing but wait on I/O.
Like the two newly introduced functions, bdrv_io_plug and bdrv_io_unplug
now operate on all children. The visit order is now symmetrical between
plug and unplug, making it possible for formats to implement plug/unplug.
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.
This commit was created with scripts/clean-includes.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
All callers pass in false, and the real external ones will switch to
true in coming patches.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
There is no need to do another O(n) pass on the list; the iocb to
split the list at is already available through the array we passed to
io_submit.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1418305950-30924-6-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
These are unused.
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1418305950-30924-5-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
It does not identify an index in an array anymore.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1418305950-30924-4-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Avoid that unplug submits requests when io_submit reported that it
couldn't accept more; at the same time, try more io_submit calls if it
could handle the whole set of requests that were passed, so that the
"blocked" flag is reset as soon as possible.
After the previous patch, laio_submit already tried to avoid submitting
requests to a blocked queue, by comparing s->io_q.idx with "==" instead
of the more natural ">=". Switch to the simpler expression now that we
have the "blocked" flag.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1418305950-30924-3-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Keep a queue of requests that were not submitted; pass them to
the kernel when a completion is reported, unless the queue is
plugged.
The array of iocbs is rebuilt every time from scratch. This
avoids keeping the iocbs array and list synchronized.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1418305950-30924-2-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
I'll use it with block backends shortly, and the name is going to fit
badly there. It's a block layer thing anyway, not just a block driver
thing.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
I'll use BlockDriverAIOCB with block backends shortly, and the name is
going to fit badly there. It's a block layer thing anyway, not just a
block driver thing.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Just call io_cancel (2), if it fails, it means the request is not
canceled, so the event loop will eventually call
qemu_laio_process_completion.
In qemu_laio_process_completion, change to call the cb unconditionally.
It is required by bdrv_aio_cancel_async.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
If two Linux AIO request completions are fetched in the same
io_getevents() call, QEMU will deadlock if request A's callback waits
for request B to complete using an aio_poll() loop. This was reported
to happen with the mirror blockjob.
This patch moves completion processing into a BH and makes it resumable.
Nested event loops can resume completion processing so that request B
will complete and the deadlock will not occur.
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Ming Lei <ming.lei@canonical.com>
Cc: Marcin Gibuła <m.gibula@beyond.pl>
Reported-by: Marcin Gibuła <m.gibula@beyond.pl>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-by: Marcin Gibuła <m.gibula@beyond.pl>
when hotplug virtio-scsi disks using laio, the aio_nr will
increase in laio_init() by io_setup(), we can see the number by
# cat /proc/sys/fs/aio-nr
128
if the aio_nr attach the maxnum, which found from
# cat /proc/sys/fs/aio-max-nr
65536
the hotplug process will fail because of aio context leak.
Fix it by io_destroy in laio_cleanup().
Reported-by: daifulai <daifulai@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch implements .bdrv_io_plug, .bdrv_io_unplug and
.bdrv_flush_io_queue callbacks for linux-aio Block Drivers,
so that submitting I/O as a batch can be supported on linux-aio.
[Unprocessed requests are completed with -EIO instead of a bogus ret
value.
--Stefan]
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Hot unplugging -drive aio=native,file=test.img,format=raw images leaves
the Linux AIO event notifier and struct qemu_laio_state allocated.
Luckily nothing will use the event notifier after the BlockDriverState
has been closed so the handler function is never called.
It's still worth fixing this resource leak.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Drop the assumption that we're using the main AioContext for Linux AIO.
Convert the Linux AIO event notifier to use aio_set_event_notifier().
The .bdrv_detach/attach_aio_context() interfaces also need to be
implemented to move the event notifier handler from the old to the new
AioContext.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The .io_flush() handler no longer exists and has no users. Drop the
io_flush argument to aio_set_fd_handler() and related functions.
The AioFlushEventNotifierHandler and AioFlushHandler typedefs are no
longer used and are dropped too.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
.io_flush() is no longer called so drop qemu_laio_completion_cb(). It
turns out that count is now unused so drop that too.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Now that AIOPool no longer keeps a freelist, it isn't really a "pool"
anymore. Rename it to AIOCBInfo and make it const since it no longer
needs to be modified.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>