VIRTIO_BLK_F_SCSI is supposed to mean whether the host can *parse*
SCSI requests, not *execute* them. You could run QEMU with scsi=on
and a file-backed disk, and QEMU would fail all SCSI requests even
though it advertises VIRTIO_BLK_F_SCSI.
Because we need to do this to fix a migration compatibility problem
related to how QEMU is invoked by management, we must do this
unconditionally even on older machine types. This more or less assumes
that no one ever invoked QEMU with scsi=off.
Here is how testing goes:
- old QEMU, scsi=on -> new QEMU, scsi=on
- new QEMU, scsi=on -> old QEMU, scsi=on
- old QEMU, scsi=off -> new QEMU, scsi=on
- new QEMU, scsi=off -> old QEMU, scsi=on
ok (new QEMU has VIRTIO_BLK_F_SCSI, adding host features is fine)
- old QEMU, scsi=off -> new QEMU, scsi=off
ok (new QEMU has VIRTIO_BLK_F_SCSI, adding host features is fine)
- old QEMU, scsi=on -> new QEMU, scsi=off
ok, bug fixed
- new QEMU, scsi=on -> old QEMU, scsi=off
doesn't work (same as: old QEMU, scsi=on -> old QEMU, scsi=off)
- new QEMU, scsi=off -> old QEMU, scsi=off
broken by the patch
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
We will have to add another field to the virtio-blk configuration in
the next patch. Avoid a proliferation of arguments to virtio_blk_init.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Move it from virtio_blk_exit_pci to virtio_blk_exit.
This is included here because the next patch removes proxy->block.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Linux really looks only at scsi->errors for SG_IO requests; it does
not look at the virtio request status at all. Because of this, when
a SG_IO request is failed early with virtio_blk_req_complete(req,
VIRTIO_BLK_S_UNSUPP), without writing hdr.status, it will look like
a success to the guest.
This is their bug, but we can make it safe for older guests now by
forcing scsi->errors to have a non-zero value whenever a request
has to be failed.
But if we fix the bug in the guest driver, we will have another problem
because QEMU returns VIRTIO_BLK_S_IOERR if the status is non-zero, and
Linux translates that to -EIO. Rather, the guest should succeed the
request and pass the non-zero status via the userspace-provided SG_IO
structure. So, remove the case where virtio_blk_handle_scsi can
return VIRTIO_BLK_S_IOERR.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Since most property types do not have a parse property now, this was
broken. Fix it by looking at the setter instead.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Acked-by: Andreas F=E4rber <afaerber@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This was erroneously dropped in d6c730086c
(pc: reduce duplication in compat machine types).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Otherwise, non-string properties without a legacy counterpart are missed.
Also fix error propagation in object_property_print() itself.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This is a partial revert of commits a369da5 (vga: improve VGA logic,
committed 2012-01-22) and c5bd4f3 (vga: fix -nodefaults -device VGA,
2012-01-24) which broke command-line option parsing in different ways.
Since commit a369da5 it has become impossible to specify a VGA device
entirely with QemuOpts-enabled options, i.e. without needing an explicit
"-vga none".
In addition, until commit c5bd4f3 -nodefaults would not disable the device
you specified with the legacy "-vga" option, independent of the order.
Since commit c5bd4f3 QEMU -nodefaults will override a previous -vga
option.
I did not reintroduce machine->no_vga. Boards can simply ignore the
vga_interface_type variable, and most will indeed do so.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Unallocated sectors should really never be accessed by the guest,
so there's no need to copy them during the streaming process.
If they are read by the guest during streaming, guest-initiated
copy-on-read will copy them (we're in the base == NULL case, which
enables copy on read). If they are read after we disconnect the
image from the base, they will read as zeroes anyway.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This fixes inability to make progress in streaming if the quota is set
to less than the amount of data that an I/O operation has to write.
In this case, limit->dispatched + n will always be above the quota and,
due to the "goto retry" to recheck cancellation and allocation, streaming
will livelock.
This can be reproduced with "block_job_set_speed ide0-hd0 1b". Of course,
with this patch the requested limit will not be obeyed. That could be
done with another patch that caps is_allocated's n argument by the slice
quota.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The change of the argument name from value to speed was not propagated there.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When an image is modified to point to the new backing file, the backing
file format is set to NULL, which means auto-probe. This is wrong, in
fact it is a small security problem.
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The test on sectors not allocated can fail if the L1/L2 tables are
not on disk yet. Allow tests to shutdown the VM early.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Because sector_num is not updated, the loop would either go on
forever or return garbage.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
qemu-io's cvtstr function sometimes will incorrectly omit the
decimal part of the number, and sometimes will incorrectly include
it. This patch fixes both. The former is more serious, and can
be seen in the patches to 027.out and 033.out.
The changes to all other files were scripted with sed, so there were
no "surprises" beyond 027.out and 033.out.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
qemu-img info should use the same logic as qemu when printing the
backing file path, or debugging becomes quite tricky. We can also
simplify the output in case the backing file has an absolute path
or a protocol.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_close should leave fields in the same state as bdrv_new. It is
not up to bdrv_open_common to fix the mess.
Also, backing_format was not being re-initialized.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
path_has_protocol will erroneously return "true" if the colon is part
of a filename. These names are common with stable device names produced
by udev. We cannot fully protect against this in case the filename
does not have a path component (e.g. if the current directory is
/dev/disk/by-path), but in the common case there will be a slash before
and path_has_protocol can easily detect that and return false.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
On Windows, all the logic is already in is_windows_drive and
is_windows_drive_prefix. On POSIX, there is no need to look
out for colons.
The win32 code changes the behaviour in some cases, we could have
something like "d:foo.img". The old code would treat it as relative
path, the new one as absolute. Now the path is absolute, because to
go from c:/program files/blah to d:foo.img you cannot say c:/program
files/blah/d:foo.img. You have to say d:foo.img. But you could also
say it's relative because (I think, at least it was like that in DOS
15 years ago) d:foo.img is relative to the current path of drive D.
Considering how path_is_absolute is used by path_combine, I think it's
better to treat it as absolute.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The limitation on not having I/O after cancellation cannot really be
kept. Even streaming has a very small race window where you could
cancel a job and have it report completion. If this window is hit,
bdrv_change_backing_file() will yield and possibly cause accesses to
dangling pointers etc.
So, let's just assume that we cannot know exactly what will happen
after the coroutine has set busy to false. We can set a very lax
condition:
- if we cancel the job, the coroutine won't set it to false again
(and hence will not call co_sleep_ns again).
- block_job_cancel_sync will wait for the coroutine to exit, which
pretty much ensures no race.
Instead, we track the coroutine that executes the job and put very
strict conditions on what to do while it is quiescent (busy = false).
First of all, the coroutine must never set busy = false while the job
has been cancelled. Second, the coroutine can be reentered arbitrarily
while it is quiescent, so you cannot really do anything but co_sleep_ns at
that time. This condition is obeyed by the block_job_sleep_ns function.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This function abstracts the pretty complex semantics of the "busy"
member of BlockJob.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We are reusing bs->file across close/open, which may not cause any
known bugs but is a recipe for trouble. Prefer bdrv_delete, and
enjoy the new invariant in the implementation of bdrv_delete.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This is another bug caused by not doing a full cleanup of the BDS
across close/open. This was found with mirroring by Shaolong Hu,
but it can probably be reproduced also with eject or change.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_append must also copy open_flags to the top, because the snapshot
has BDRV_O_NO_BACKING set. This causes interesting results if you
later use drive-reopen (not upstream) to reopen the image, and lose
the backing file in the process.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
QED's opaque data includes a pointer back to the BlockDriverState.
This breaks when bdrv_append shuffles data between bs_new and bs_top.
To avoid this, add a "rebind" function that tells the driver about
the new relationship between the BlockDriverState and its opaque.
The patch also adds rebind to VVFAT for completeness, even though
it is not used with live snapshots.
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
A trailing space is left when qemu-img has no arguments, for example if
-nocache is not used. This becomes an empty argument after split()
and causes qemu-io to fail.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Also reuse elsewhere the new constant for sizeof(unsigned long) * 8.
The dirty bitmap is allocated in bits but declared as unsigned long.
Thus, its memory block is accessed beyond its end unless the image
is a multiple of 64 chunks (i.e. a multiple of 64 MB).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_img_create will temporarily open the backing file to probe its size.
However, this could be done with a read-write open if the wrong flags are
passed to bdrv_img_create. Since there is really no documentation on
what flags can be passed, assume that bdrv_img_create receives the flags
with which the new image will be opened; sanitize them when opening
the backing file.
Reported-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
These are needed to print "info block" output correctly. QCOW2 does this
because it needs it to write the header, but QED does not, and common code
is the right place to do it.
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Optional inquiry information is declared obsolete in the latest versions
of the standard; invalid CDBs or unsupported VPD pages are supported
can be diagnosed with trace_scsi_inquiry.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The requirements on the REQUEST SENSE buffer size are not in my copy of SPC
(SPC-4 r27) and not observed by LIO. Rip them out.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The requirements on the INQUIRY buffer size are not in my copy of SPC
(SPC-4 r27) and not observed by LIO. Rip them out.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The transfer length for these commands is different from the transfer
length of the corresponding disk commands, so parse it specially.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Interpreting cdb[4] == 0 as a request to transfer 256 blocks is only
needed for READ_6 and WRITE_6. No other command in that range needs
that special-casing, and the resulting overrun breaks scsi-testsuite's
attempt to use command 2 as a known-invalid command.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Linux expects REQ_FUA to be advertised only if WRITE+FUA is faster than
WRITE+SYNCHRONIZE CACHE, so we should not set the DPOFUA bit. However,
it is useful to have it for testing purposes, so add a qdev property to
set it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It is pointless to add a uint32_t field for every new feature.
Since we will need a new feature soon, convert accesses to "removable"
to look at bit 0 only.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Recently introduced FUA support also gave us a use-after-free
of the BlockAcctCookie within a SCSIDiskReq, due to unbalanced
reference counting.
The patch fixes this by making scsi_do_read look like a combination
of scsi_*_complete + scsi_*_data. It does both a ref (like
scsi_read_data) and an unref (like scsi_flush_complete).
Reported-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The do...while loop can never loop, because select will just not return
0 when invoked with infinite timeout.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The definition of when qemu_aio_flush should loop is much simpler
than it looks. It just has to call qemu_aio_wait until it makes
no progress and all flush callbacks return false. qemu_aio_wait
is the logical place to tell the caller about this.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Right now, nbd_wr_sync will hang if no data at all is available on the
socket and the other side is not going to provide any. Relax this by
making it loop only for writes or partial reads. This fixes a race
where one thread is executing qemu_aio_wait() and another is executing
main_loop_wait(). Then, the select() call in main_loop_wait() can return
stale data and call the "readable" callback with no data in the socket.
Reported-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In the next patch we need to look at the return code of nbd_wr_sync.
To avoid percolating the socket_error() ugliness all around, let's
handle errors by returning negative errno values.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
GCC (pedantically, but correctly) considers that a negative ssize_t may
become positive when casted to int. This may cause uninitialized variable
warnings when a function returns such a negative ssize_t and is inlined.
Propagate ssize_t return types to avoid this.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
posix_aio_read already calls qemu_aio_process_queue, and dually
qemu_aio_process_queue is always followed by a select loop that calls
posix_aio_read.
No races are possible, so there is no need for a separate process_queue
callback.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
QED uses vm_clock timers so that images are not touched during and after
migration. This however does not apply to qemu-io and qemu-img.
Treat vm_clock as a synonym for rt_clock there, and enable it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This will let timers run during aio_read and aio_write commands,
though not during synchronous commands.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In the next patch we want to reenter the coroutine from
block_job_cancel_sync and cancel the timer.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
To force unit access, add a flush operation after the actual write.
WRITE AND VERIFY commands always flush according to SBC, so do it
even though we do not perform the reread.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
First scsi_flush_complete, like scsi_dma_complete, is always called with
an active AIOCB.
Second, always test for "ret < 0" to check for errors.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Adding multiqueue is as simple as creating more than one virtqueues,
and saving the queue number for each request.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Serializing virtio-scsi requests needs a simple way to get from a
VirtQueue to the number of the queue. The virtio_queue_get_id
provides this.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In order to restore requests correctly from a multitude of virtqueues,
we need to store the id of the request queue that each request came
from.
Do this even for single-queue, by storing a hard-coded zero, to
simplify future implementation of multiqueue.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
scsibus_get_dev_path is leaking id if it is not NULL. Fix it.
Reported-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The special target should not be needed anymore, and caused (perhaps
due to a Make bug) a failure with "make -j2". In any case, the
main makefile is a better place for such special targets rather
than an included makefile.
Reported-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Tested-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
These are pretty sane requirements to move forward with glib usage.
2.12 is the version found in RHEL/CentOS 5, and 2.20 is the
first version to support g_poll. Without g_poll, we cannot
integrate well with the glib main loop.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
On w32, glib implements g_poll using WaitForMultipleObjects
or MsgWaitForMultipleObjects. This means that we can simplify
our code by switching to g_poll, and at the same time prepare for
adding back glib sources.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Right now, the main loop is not interrupted when data arrives on a
socket. To fix this, register each socket to interrupt the main loop
with WSAEventSelect. This does not replace select, it only communicates
a change in socket state that requires a select call.
Since the interrupt fires only once per recv call, or only once
after a send call returns EWOULDBLOCK we can activate it on all events
unconditionally. If QEMU is momentarily uninterested on some condition,
the main loop will not busy wait. Instead, it may get one extra wakeup,
but then it will ignore the condition until progress occurs and/or
qemu_set_fd_handler is called to set a callback. At this point the
condition will be tested via select and the callback will be invoked
even if it is still disabled on the event.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Using select with glib pollfds is wrong under w32. Restrict
the code to the POSIX case.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
The timeval-based timeout is not needed until we actually invoke select,
so compute it only then. Also group the two calls that modify the
timeout, glib_select_fill and os_host_main_loop_wait.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
I am not sure that these are really proper GtkDoc, but they follow
the existing documentation in block_int.h.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
There is no need to do this in every implementation of set_speed
(even though there is only one right now).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Streaming can issue I/O while qcow2_close is running. This causes the
L2 caches to become very confused or, alternatively, could cause a
segfault when the streaming coroutine is reentered after closing its
block device. The fix is to cancel streaming jobs when closing their
underlying device.
The cancellation must be synchronous, on the other hand qemu_aio_wait
will not restart a coroutine that is sleeping in co_sleep. So add
a flag saying whether streaming has in-flight I/O. If the busy flag
is false, the coroutine is quiescent and, when cancelled, will not
issue any new I/O.
This protects streaming against closing, but not against deleting.
We have a reference count protecting us against concurrent deletion,
but I still added an assertion to ensure nothing bad happens.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We do not want jobs to keep a device busy for a possibly very long
time, and management could become confused because they thought a
device was not even there anymore. So, cancel long-running jobs
as soon as their device is going to disappear.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Finally reindent all code and change goto statements to a loop.
Acked-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reads and writes to the underlying file can also occur with the simple
non-vectored I/O interfaces.
Acked-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
vdi.c really works as if it implemented bdrv_read and bdrv_write. However,
because only vector I/O is supported by the asynchronous callbacks, it
went through extra pain to bounce-buffer the I/O. This can be handled
by the block layer now that the format is coroutine-based.
Acked-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Most of the AIOCB really holds local variables that need to persist
across callback invocation. It can go away now.
Acked-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now inline the former AIO callbacks into vdi_co_readv and vdi_co_writev.
While many cleanups are possible, the code now really looks synchronous.
Acked-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The next step is to take code that only triggers after the first operation,
and move it at the end of vdi_aio_read_cb and vdi_aio_write_cb.
Acked-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Even a basic conversion changing the bdrv_aio_readv/bdrv_aio_writev calls
to bdrv_co_readv/bdrv_co_writev, and callbacks to goto statements can
eliminate a lot of code. This is because error handling is simplified
and indirections through bottom halves can go away.
After this patch, I/O to the underlying file already happens via
coroutines, but the code still looks a lot like if asynchronous I/O was
being used.
Acked-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
A strong limitation of QOM right now is that unconverted ports
(e.g. all...) do not give a canonical path to devices that are
part of the board. This in turn makes it impossible to replace
PROP_PTR with a QOM link for example.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
We want the composition tree to to be in order by the time we call
qdev_init, so that a single set of the toplevel realize property can
propagate all the way down the composition tree.
This is not the case so far. Unfortunately, this is incompatible
with calling qdev_init in the constructor wrappers for devices,
so for now we need to unattach some devices that are created through
those wrappers. This will be fixed by removing qdev_init and instead
setting the toplevel realize property after machine init.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Paolo Bonzini