Now that we stay in coroutine context for the whole request when doing
reads or writes, we can add coroutine_fn annotations to many functions
that can do I/O or yield directly.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
This fixes the last place where we degraded from AIO to actual blocking
synchronous I/O requests. Putting it into a coroutine means that instead
of blocking, the coroutine simply yields while doing I/O.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Now that we process a request in the same coroutine from beginning to
end and don't drop out of it any more, we can look like a proper
coroutine-based driver and simply call qed_aio_next_io() and get a
return value from it instead of spawning an additional coroutine that
reenters the parent when it's done.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Now that we're running in coroutine context, the ad-hoc serialisation
code (which drops a request that has to wait out of coroutine context)
can be replaced by a CoQueue.
This means that when we resume a serialised request, it is running in
coroutine context again and its I/O isn't blocking any more.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Most of the qed code is now synchronous and matches the coroutine model.
One notable exception is the serialisation between requests which can
still schedule a callback. Before we can replace this with coroutine
locks, let's convert the driver's external interfaces to the coroutine
versions.
We need to be careful to handle both requests that call the completion
callback directly from the calling coroutine (i.e. fully synchronous
code) and requests that involve some callback, so that we need to yield
and wait for the completion callback coming from outside the coroutine.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Manos Pitsidianakis <el13635@mail.ntua.gr>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Instead of calling itself recursively as the last thing, just convert
qed_aio_next_io() into a loop.
This patch is best reviewed with 'git show -w' because most of it is
just whitespace changes.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Don't recurse into qed_aio_next_io() and qed_aio_complete() here, but
just return an error code and let the caller handle it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Don't recurse into qed_aio_next_io() and qed_aio_complete() here, but
just return an error code and let the caller handle it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Don't recurse into qed_aio_next_io() and qed_aio_complete() here, but
just return an error code and let the caller handle it.
While refactoring qed_aio_write_alloc() to accomodate the change,
qed_aio_write_zero_cluster() ended up with a single line, so I chose to
inline that line and remove the function completely.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Don't recurse into qed_aio_next_io() and qed_aio_complete() here, but
just return an error code and let the caller handle it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Don't recurse into qed_aio_next_io() and qed_aio_complete() here, but
just return an error code and let the caller handle it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Don't recurse into qed_aio_next_io() and qed_aio_complete() here, but
just return an error code and let the caller handle it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
qed_commit_l2_update() is unconditionally called at the end of
qed_aio_write_l1_update(). Inline it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Note that this code is generally not running in coroutine context, so
this is an actual blocking synchronous operation. We'll fix this in a
moment.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Note that this code is generally not running in coroutine context, so
this is an actual blocking synchronous operation. We'll fix this in a
moment.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
The GenericCB infrastructure isn't used any more. Remove it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Note that this code is generally not running in coroutine context, so
this is an actual blocking synchronous operation. We'll fix this in a
moment.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Note that this code is generally not running in coroutine context, so
this is an actual blocking synchronous operation. We'll fix this in a
moment.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
With this change, qed_aio_write_prefill() and qed_aio_write_postfill()
collapse into a single function. This is reflected by a rename of the
combined function to qed_aio_write_cow().
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Note that this code is generally not running in coroutine context, so
this is an actual blocking synchronous operation. We'll fix this in a
moment.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Note that this code is generally not running in coroutine context, so
this is an actual blocking synchronous operation. We'll fix this in a
moment.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Instead of passing the return value to a callback, return it to the
caller so that the callback can be inlined there.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Note that this code is generally not running in coroutine context, so
this is an actual blocking synchronous operation. We'll fix this in a
moment.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
The qed driver serialises allocating write requests. When the active
allocation is finished, the AIO callback is called, but after this, the
next allocating request is immediately processed instead of leaving the
coroutine. Resuming another allocation request in the same request
coroutine means that the request now runs in the wrong coroutine.
The following is one of the possible effects of this: The completed
request will generally reenter its request coroutine in a bottom half,
expecting that it completes the request in bdrv_driver_pwritev().
However, if the second request actually yielded before leaving the
coroutine, the reused request coroutine is in an entirely different
place and is reentered prematurely. Not a good idea.
Let's make sure that we exit the coroutine after completing the first
request by resuming the next allocating request only with a bottom
half.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
We already have functions for doing these calculations, so let's use
them instead of doing everything by hand. This makes the code a bit
more readable.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If the guest tries to write data that results on the allocation of a
new cluster, instead of writing the guest data first and then the data
from the COW regions, write everything together using one single I/O
operation.
This can improve the write performance by 25% or more, depending on
several factors such as the media type, the cluster size and the I/O
request size.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of passing a single buffer pointer to do_perform_cow_write(),
pass a QEMUIOVector. This will allow us to merge the write requests
for the COW regions and the actual data into a single one.
Although do_perform_cow_read() does not strictly need to change its
API, we're doing it here as well for consistency.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reading both COW regions requires two separate requests, but it's
perfectly possible to merge them and perform only one. This generally
improves performance, particularly on rotating disk drives. The
downside is that the data in the middle region is read but discarded.
This patch takes a conservative approach and only merges reads when
the size of the middle region is <= 16KB.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch splits do_perform_cow() into three separate functions to
read, encrypt and write the COW regions.
perform_cow() can now read both regions first, then encrypt them and
finally write them to disk. The memory allocation is also done in
this function now, using one single buffer large enough to hold both
regions.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of calling perform_cow() twice with a different COW region
each time, call it just once and make perform_cow() handle both
regions.
This patch simply moves code around. The next one will do the actual
reordering of the COW operations.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Qcow2COWRegion has two attributes:
- The offset of the COW region from the start of the first cluster
touched by the I/O request. Since it's always going to be positive
and the maximum request size is at most INT_MAX, we can use a
regular unsigned int to store this offset.
- The size of the COW region in bytes. This is guaranteed to be >= 0,
so we should use an unsigned type instead.
In x86_64 this reduces the size of Qcow2COWRegion from 16 to 8 bytes.
It will also help keep some assertions simpler now that we know that
there are no negative numbers.
The prototype of do_perform_cow() is also updated to reflect these
changes.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We are using the return value of qcow2_encrypt_sectors() to detect
problems but we are throwing away the returned Error since we have no
way to report it to the user. Therefore we can simply get rid of the
local Error variable and pass NULL instead.
Alternatively we could try to figure out a way to pass the original
error instead of simply returning -EIO, but that would be more
invasive, so let's keep the current approach.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
There used to be throttle_timers_{detach,attach}_aio_context() calls
in bdrv_set_aio_context(), but since 7ca7f0f6db
they are now in blk_set_aio_context().
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Calling aio_poll() directly may have been fine previously, but this is
the future, man! The difference between an aio_poll() loop and
BDRV_POLL_WHILE() is that BDRV_POLL_WHILE() releases the AioContext
around aio_poll().
This allows the IOThread to run fd handlers or BHs to complete the
request. Failure to release the AioContext causes deadlocks.
Using BDRV_POLL_WHILE() partially fixes a 'savevm' hang with -object
iothread.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Call bdrv_inc/dec_in_flight() for vmstate reads/writes. This seems
unnecessary at first glance because vmstate reads/writes are done
synchronously while the guest is stopped. But we need the bdrv_wakeup()
in bdrv_dec_in_flight() so the main loop sees request completion.
Besides, it's cleaner to count vmstate reads/writes like ordinary
read/write requests.
The bdrv_wakeup() partially fixes a 'savevm' hang with -object iothread.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
commit_complete() can't assume that after its block_job_completed() the
job is actually immediately freed; someone else may still be holding
references. In this case, the op blockers on the intermediate nodes make
the graph reconfiguration in the completion code fail.
Call block_job_remove_all_bdrv() manually so that we know for sure that
any blockers on intermediate nodes are given up.
Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
We would like to use a same QObject type to represent numbers, whether
they are int, uint, or floats. Getters will allow some compatibility
between the various types if the number fits other representations.
Add a few more tests while at it.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20170607163635.17635-7-marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
[parse_stats_intervals() simplified a bit, comment in
test_visitor_in_int_overflow() tidied up, suppress bogus warnings]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
I'm not trying too hard yet. Later, with multiqueue support,
this may cause mutex contention or cacheline bouncing.
Cc: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20170605123908.18777-20-pbonzini@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
block_acct_destroy is called unconditionally in blk_delete, but there is
no BlockAcctStats function that is called unconditionally in blk_new.
Split block_acct_init in two, so that it will be possible to create a
QemuMutex in block_acct_init and destroy it in block_acct_cleanup.
Cc: Alberto Garcia <berto@igalia.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20170605123908.18777-19-pbonzini@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
This is the common code to account operations that produced actual I/O.
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20170605123908.18777-18-pbonzini@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
It protects only the list of dirty bitmaps; in the next patch we will
also protect their content.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20170605123908.18777-15-pbonzini@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Kevin Wolf
Alberto Garcia
Stefan Hajnoczi
Marc-André Lureau
Paolo Bonzini