The batch notification BH needs to know which virtqueues to notify when
multiqueue is enabled. Use a bitmap to track the virtqueues with
pending notifications.
At this point there is only one virtqueue so hard-code virtqueue index
0. A later patch will switch to real virtqueue indices.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1466511196-12612-3-git-send-email-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
All users have been converted to the new ioevent callbacks.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Have vhost and dataplane use the new api for transports that
have been converted.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Block layer is prepared to unspecialize dataplane, an evidence is this
almost complete list of unblocked operations. It has all types except
two (actually three if DATAPLANE itself counts but blockdev.c makes sure
attaching twice is not possible): MIRROR_TARGET and BACKUP_TARGET.
blockdev-mirror refuses to start if target is attached, so the first is
not a problem.
By removing BACKUP_TARGET, blockdev-backup will become permissive to
write to a virtio-blk dataplane disk, but that is not worse than
non-dataplane given the latter is already possible. In either case,
blockdev.c always checks the target and source are on the same
AioContext, or bring them together if possible.
Signed-off-by: Fam Zheng <famz@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Message-id: 1463969978-24970-4-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Eliminating the reentrancy is actually a nice thing that we can do
with the API that Michael proposed, so let's make it first class.
This also hides the complex assign/set_handler conventions from
callers of virtio_queue_aio_set_host_notifier_handler, which in
fact was always called with assign=true.
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
In addition to handling IO in vcpu thread and in io thread, dataplane
introduces yet another mode: handling it by AioContext.
This reuses the same handler as previous modes, which triggers races as
these were not designed to be reentrant. Use a separate handler just
for aio, and disable regular handlers when dataplane is active.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We must not call virtio_blk_data_plane_notify if dataplane is
disabled: we would hit a segmentation fault in notify_guest_bh as
s->guest_notifier has not been setup and is NULL.
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Commit 57cb38b included qapi/error.h into qemu/osdep.h to get the
Error typedef. Since then, we've moved to include qemu/osdep.h
everywhere. Its file comment explains: "To avoid getting into
possible circular include dependencies, this file should not include
any other QEMU headers, with the exceptions of config-host.h,
compiler.h, os-posix.h and os-win32.h, all of which are doing a
similar job to this file and are under similar constraints."
qapi/error.h doesn't do a similar job, and it doesn't adhere to
similar constraints: it includes qapi-types.h. That's in excess of
100KiB of crap most .c files don't actually need.
Add the typedef to qemu/typedefs.h, and include that instead of
qapi/error.h. Include qapi/error.h in .c files that need it and don't
get it now. Include qapi-types.h in qom/object.h for uint16List.
Update scripts/clean-includes accordingly. Update it further to match
reality: replace config.h by config-target.h, add sysemu/os-posix.h,
sysemu/os-win32.h. Update the list of includes in the qemu/osdep.h
comment quoted above similarly.
This reduces the number of objects depending on qapi/error.h from "all
of them" to less than a third. Unfortunately, the number depending on
qapi-types.h shrinks only a little. More work is needed for that one.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
[Fix compilation without the spice devel packages. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
In disabled mode, virtio-blk dataplane seems to be enabled, but flow
actually goes through the normal virtio path. This patch simplifies a bit
the handling of disabled mode. In disabled mode, virtio_blk_handle_output
might be called even if s->dataplane is not NULL.
This is a bit tricky, because the current check for s->dataplane will
always trigger, causing a continuous stream of calls to
virtio_blk_data_plane_start. Unfortunately, these calls will not
do anything. To fix this, set the "started" flag even in disabled
mode, and skip virtio_blk_data_plane_start if the started flag is true.
The resulting changes also prepare the code for the next patch, were
virtio-blk dataplane will reuse the same virtio_blk_handle_output function
as "regular" virtio-blk.
Because struct VirtIOBlockDataPlane is opaque in virtio-blk.c, we have
to move s->dataplane->started inside struct VirtIOBlock.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Make the API more similar to the regular virtqueue API. This will
help when modifying the code to not use vring.c anymore.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
The return code of virtqueue_pop/vring_pop is unused except to check for
errors or 0. We can thus easily move allocation inside the functions
and just return a pointer to the VirtQueueElement.
The advantage is that we will be able to allocate only the space that
is needed for the actual size of the s/g list instead of the full
VIRTQUEUE_MAX_SIZE items. Currently VirtQueueElement takes about 48K
of memory, and this kind of allocation puts a lot of stress on malloc.
By cutting the size by two or three orders of magnitude, malloc can
use much more efficient algorithms.
The patch is pretty large, but changes to each device are testable
more or less independently. Splitting it would mostly add churn.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Put the code for setting up and removing op blockers into an own
function, respectively. Then, we can invoke those functions whenever a
BDS is removed from an virtio-blk BB or inserted into it.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@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>
Done with this Coccinelle semantic patch
@@
expression FMT, E1, E2;
expression list ARGS;
@@
- error_setg(E1, FMT, ARGS, error_get_pretty(E2));
+ error_propagate(E1, E2);/*###*/
+ error_prepend(E1, FMT/*@@@*/, ARGS);
followed by manual cleanup, first because I can't figure out how to
make Coccinelle transform strings, and second to get rid of now
superfluous error_propagate().
We now use or propagate the original error whole instead of just its
message obtained with error_get_pretty(). This avoids suppressing its
hint (see commit 50b7b00), but I can't see how the errors touched in
this commit could come with hints. It also improves the message
printed with &error_abort when we screw up (see commit 1e9b65b).
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
It's necessary to distinguish source and target before we can add
blockdev-mirror, because we would want a concrete type of operation to
check on target bs before starting.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 1450932306-13717-2-git-send-email-famz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
The official way of enabling dataplane is through the "iothread"
property that references an iothread object created by "-object
iothread". Since the old "x-data-plane=on" way now even crashes, it's
probably easier to just drop it:
$ qemu-system-x86_64 -drive file=null-co://,id=d0,if=none \
-device virtio-blk-pci,drive=d0,x-data-plane=on
ERROR:/home/fam/work/qemu/qom/object.c:1515:
object_get_canonical_path_component: assertion failed: (obj->parent != NULL)
Aborted
Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 1449485967-19240-1-git-send-email-famz@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
They will be excluded by type in the nested event loops in block layer,
so that unwanted events won't be processed there.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@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>
After qemu_iovec_destroy, the QEMUIOVector's size is zeroed and
the zero size ultimately is used to compute virtqueue_push's len
argument. Therefore, reads from virtio-blk devices did not
migrate their results correctly. (Writes were okay).
Save the size in virtio_blk_handle_request, and use it when the request
is completed.
Based on a patch by Wen Congyang.
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Message-id: 1427997044-392-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The vring.c code currently assumes that guest and host endianness match,
which is not true for a number of cases:
- emulating targets with a different endianness than the host
- bi-endian targets, where the correct endianness depends on the virtio
device
- upcoming support for the virtio-1 standard mandates little-endian
accesses even for big-endian targets and hosts
Make sure to use accessors that depend on the virtio device.
Note that dataplane now needs to be built per-target.
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Fam Zheng <famz@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Tested-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1422289602-17874-2-git-send-email-cornelia.huck@de.ibm.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
this patch finally introduces multiread support to virtio-blk. While
multiwrite support was there for a long time, read support was missing.
The complete merge logic is moved into virtio-blk.c which has
been the only user of request merging ever since. This is required
to be able to merge chunks of requests and immediately invoke callbacks
for those requests. Secondly, this is required to switch to
direct invocation of coroutines which is planned at a later stage.
The following benchmarks show the performance of running fio with
4 worker threads on a local ram disk. The numbers show the average
of 10 test runs after 1 run as warmup phase.
| 4k | 64k | 4k
MB/s | rd seq | rd rand | rd seq | rd rand | wr seq | wr rand
--------------+--------+---------+--------+---------+--------+--------
master | 1221 | 1187 | 4178 | 4114 | 1745 | 1213
multiread | 1829 | 1189 | 4639 | 4110 | 1894 | 1216
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Like BLOCK_OP_TYPE_BACKUP_SOURCE and BLOCK_OP_TYPE_BACKUP_TARGET,
block-commit involves two asymmetric devices.
This change is not user-visible (yet), because commit only works with
device names.
But once we enable backing reference in blockdev-add, or specifying
node-name in block-commit command, we don't want the user to start two
commit jobs on the same backing chain, which will corrupt things because
of the final bdrv_swap.
Before we have per category blockers, splitting this type is still
better.
[Resolved virtio-blk dataplane conflict by replacing
BLOCK_OP_TYPE_COMMIT with both BLOCK_OP_TYPE_COMMIT_{SOURCE, TARGET}.
They are safe since the block job runs in the same AioContext as the
dataplane IOThread.
--Stefan]
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The transaction QMP command performs operations atomically on a group of
drives. This command needs to acquire AioContext in order to work
safely when virtio-blk dataplane IOThreads are accessing drives.
The transactional nature of the command means that actions are split
into prepare, commit, abort, and clean functions. Acquire the
AioContext in prepare and don't release it until one of the other
functions is called. This prevents the IOThread from running the
AioContext before the transaction has completed.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 1416566940-4430-4-git-send-email-stefanha@redhat.com
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add dataplane support to the change-backing-file QMP commands. By
acquiring the AioContext we avoid race conditions with the dataplane
thread which may also be accessing the BlockDriverState.
Note that this command operates on both bs and a node in its chain
(image_bs). The bdrv_chain_contains(bs, image_bs) check guarantees that
bs and image_bs are in the same AioContext.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
By acquiring the AioContext we avoid race conditions with the dataplane
thread which may also be accessing the BlockDriverState.
Fix up eject, change, and block_passwd in a single patch because
qmp_eject() and qmp_change_blockdev() both call eject_device(). Also
fix block_passwd while we're tackling a command that takes a block
encryption password.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add dataplane support to the blockdev-snapshot-delete-internal-sync QMP
command. By acquiring the AioContext we avoid race conditions with the
dataplane thread which may also be accessing the BlockDriverState.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that blockjobs use AioContext they are safe for use with dataplane.
Unblock them!
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 1413889440-32577-12-git-send-email-stefanha@redhat.com
Device models should access their block backends only through the
block-backend.h API. Convert them, and drop direct includes of
inappropriate headers.
Just four uses of BlockDriverState are left:
* The Xen paravirtual block device backend (xen_disk.c) opens images
itself when set up via xenbus, bypassing blockdev.c. I figure it
should go through qmp_blockdev_add() instead.
* Device model "usb-storage" prompts for keys. No other device model
does, and this one probably shouldn't do it, either.
* ide_issue_trim_cb() uses bdrv_aio_discard() instead of
blk_aio_discard() because it fishes its backend out of a BlockAIOCB,
which has only the BlockDriverState.
* PC87312State has an unused BlockDriverState[] member.
The next two commits take care of the latter two.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This is consistent with how VirtIOFOOConf variables are named
elsewhere, and makes blk available for BlockBackend variables.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit 3718d8ab65 ("block: Replace in_use
with operation blocker") broke the error path because it consumed
local_err instead of propagating it.
The caller has no way to know that the function failed. This caused
virtio-blk to start "successfully" even though there was a fatal
dataplane error.
Steps to reproduce:
$ qemu-system-x86_64 -enable-kvm -object iothread,id=iothread0 \
-drive if=none,id=drive0,file=a.img \
(qemu) drive_mirror drive0 /tmp/foo.img
(qemu) device_add virtio-blk-pci,iothread=iothread0,drive=drive0
Expected result:
Since the mirror block job is using drive0 it is not possible to start
virtio-blk data-plane.
device_add fails and the PCI adapter is not added.
Actual result:
device_add completes and the PCI adapter is added.
Cc: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that drive_del acquires the AioContext we can safely allow deleting
the drive. As with non-dataplane mode, all I/Os submitted by the guest
after drive_del will return EIO.
This patch makes hot unplug work with virtio-blk dataplane. Previously
drive_del reported an error because the device was busy.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Now that block_resize acquires the AioContext we can safely allow
resizing the disk.
Reported-by: Andrey Korolyov <andrey@xdel.ru>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If we fail to set up guest or host notifiers, there's no use trying again
every time the guest kicks, so disable dataplane in that case.
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The dataplane code is currently doing a hard exit if it fails to set
up either guest or host notifiers. In practice, this may mean that a
guest suddenly dies after a dataplane device failed to come up (e.g.,
when a file descriptor limit is hit for tne nth device).
Let's just try to unwind the setup instead and return.
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Setting up guest or host notifiers may fail, but the user will have
no idea why: Let's print the error returned by the callback.
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Now requests are submitted as a batch, so it is natural
to notify guest as a batch too.
This may suppress interrupt notification to VM a lot:
- in my test, decreased by ~13K/sec
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The callback has to be saved and reset in virtio_blk_data_plane_start(),
otherwise dataplane's requests will be completed in qemu aio context.
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The memory allocation between hw/block/virtio-blk.c,
hw/block/dataplane/virtio-blk.c, and hw/virtio/dataplane/vring.c is
messy. Structs are allocated in different files than they are freed in.
This is risky and makes memory leaks easier.
Embed VirtQueueElement in VirtIOBlockReq to reduce the amount of memory
allocation we need to juggle. This also makes vring.c and virtio.c
slightly more similar.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
VirtIOBlockReq is freed later by virtio_blk_free_request() in
hw/block/virtio-blk.c. Remove this extraneous g_slice_free().
This patch fixes the following segfault:
0x00005555556373af in virtio_blk_rw_complete (opaque=0x5555565ff5e0, ret=0) at hw/block/virtio-blk.c:99
99 bdrv_acct_done(req->dev->bs, &req->acct);
(gdb) print req
$1 = (VirtIOBlockReq *) 0x5555565ff5e0
(gdb) print req->dev
$2 = (VirtIOBlock *) 0x0
(gdb) bt
#0 0x00005555556373af in virtio_blk_rw_complete (opaque=0x5555565ff5e0, ret=0) at hw/block/virtio-blk.c:99
#1 0x0000555555840ebe in bdrv_co_em_bh (opaque=0x5555566152d0) at block.c:4675
#2 0x000055555583de77 in aio_bh_poll (ctx=ctx@entry=0x5555563a8150) at async.c:81
#3 0x000055555584b7a7 in aio_poll (ctx=0x5555563a8150, blocking=blocking@entry=true) at aio-posix.c:188
#4 0x00005555556e520e in iothread_run (opaque=0x5555563a7fd8) at iothread.c:41
#5 0x00007ffff42ba124 in start_thread () from /usr/lib/libpthread.so.0
#6 0x00007ffff16d14bd in clone () from /usr/lib/libc.so.6
Reported-by: Max Reitz <mreitz@redhat.com>
Cc: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Before commit 580b6b2aa2(dataplane: use the QEMU block
layer for I/O), dataplane for virtio-blk submits block
I/O as a batch.
This commit 580b6b2aa2 replaces the custom linux AIO
implementation(including submit I/O as a batch) with QEMU
block layer, but this commit causes ~40% throughput regression
on virtio-blk performance, and removing submitting I/O
as a batch is one of the causes.
This patch applies the newly introduced bdrv_io_plug() and
bdrv_io_unplug() interfaces to support submitting I/O
at batch for Qemu block layer, and in my test, the change
can improve throughput by ~30% with 'aio=native'.
Following my fio test script:
[global]
direct=1
size=4G
bsrange=4k-4k
timeout=40
numjobs=4
ioengine=libaio
iodepth=64
filename=/dev/vdc
group_reporting=1
[f]
rw=randread
Result on one of my small machine(host: x86_64, 2cores, 4thread, guest: 4cores):
- qemu master: 65K IOPS
- qemu master with these patches: 92K IOPS
- 2.0.0 release(dataplane using custom linux aio): 104K IOPS
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The x-data-plane=on|off option is no longer useful because the
iothread=<iothread> option conveys the same information plus which
IOThread to use.
Do not delete x-data-plane=on|off yet as a convenience to people using
this legacy experimental option. We will drop it in QEMU 2.2.
Instead, turn on data-plane when either x-data-plane=on or
iothread=<iothread> are used. The following command-line uses
data-plane:
qemu -device virtio-blk-pci,iothread=foo,drive=drive0
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
If the virtio transport does not support notifiers (like s390-virtio),
we can't use dataplane. Bail out early and let the user know what is
wrong.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The old name is misleading in its new usage, so rename it.
Signed-off-by: Fam Zheng <famz@redhat.com>
Tested-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This drops request handling code from dataplane, and uses code from
hw/block/virtio-blk.c.
It starts to use multiwrite as non-dataplane does.
Dataplane sets VirtIOBlock.complete_request to vring version, and calls
into non-dataplane's process handling. In complete_request_early,
qiov.size is added to vring push length, because it's also called in rw
completion now.
Signed-off-by: Fam Zheng <famz@redhat.com>
Tested-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
VirtIOBlockReq is allocated in process_request, and freed in command
functions.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In current virtio spec, inhdr is a single byte, and is unlikely to
change for both functionality and compatibility considerations.
Non-dataplane uses .in, and we are on the way to converge them. So
let's unify it to get cleaner code.
Remove .inhdr and use .in.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Field "inhdr" is added temporarily for a more mechanical change, and
will be dropped in the next commit.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Since it's set but not used.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
