qemu-option: Disable two helpful messages that got broken recently
commit 8be7e7e4 and commit ec7b2ccb messed up the ordering of error
message and the helpful explanation that should follow it, like this:
$ qemu-system-x86_64 --nodefaults -S --vnc :0 --chardev null,id=,
Identifiers consist of letters, digits, '-', '.', '_', starting with a letter.
qemu-system-x86_64: -chardev null,id=,: Parameter 'id' expects an identifier
$ qemu-system-x86_64 --nodefaults -S --vnc :0 --machine kvm_shadow_mem=dunno
You may use k, M, G or T suffixes for kilobytes, megabytes, gigabytes and terabytes.
qemu-system-x86_64: -machine kvm_shadow_mem=dunno: Parameter 'kvm_shadow_mem' expects a size
error: Clean up error strings with embedded newlines
The arguments of error_report() should yield a short error string
without newlines.
A few places try to print additional help after the error message by
embedding newlines in the error string. That's nice, but let's do it
the right way.
Since I'm touching these lines anyway, drop a stray preposition and
some tabs. We don't use tabs for similar messages elsewhere.
Stefan Hajnoczi [Fri, 8 Feb 2013 07:49:10 +0000 (08:49 +0100)]
block/curl: disable extra protocols to prevent CVE-2013-0249
There is a buffer overflow in libcurl POP3/SMTP/IMAP. The workaround is
simple: disable extra protocols so that they cannot be exploited. Full
details here:
http://curl.haxx.se/docs/adv_20130206.html
QEMU only cares about HTTP, HTTPS, FTP, FTPS, and TFTP. I have tested
that this fix prevents the exploit on my host with
libcurl-7.27.0-5.fc18.
Jesse Larrew [Tue, 5 Feb 2013 23:47:17 +0000 (17:47 -0600)]
hw/virtio-net: disable multiqueue by default
The new multiqueue feature adds fields to the virtio device config, which
breaks Windows guests. Disable the feature by default until the Windows
drivers are fixed.
Jesse Larrew [Tue, 5 Feb 2013 23:47:16 +0000 (17:47 -0600)]
hw/virtio-net.c: set config size using host features
Currently, the config size for virtio devices is hard coded. When a new
feature is added that changes the config size, drivers that assume a static
config size will break. For purposes of backward compatibility, there needs
to be a way to inform drivers of the config size needed to accommodate the
set of features enabled.
aliguori: merged in
- hw/virtio-net: use existing macros to implement endof
- hw/virtio-net: fix config_size data type
Michael Roth [Thu, 7 Feb 2013 00:25:48 +0000 (18:25 -0600)]
net: fix infinite loop on exit
1ceef9f27359cbe92ef124bf74de6f792e71f6fb added handling for cleaning
up multiple queues in qemu_del_nic() for cases where multiqueue is in
use. To determine the number of queues it looks at nic->conf->queues,
then iterates through all the queues to cleanup the associated
NetClientStates. If no queues are found, no NetClientStates are deleted.
However, nic->conf->queues is only set when a peer is created via
-netdev or netdev_add, and is otherwise 0. This causes us to spin in
net_cleanup() if we attempt to shut down qemu before adding a host
device.
Since qemu_new_nic() unconditionally creates at least 1
queue/NetClientState at queue idx 0, make qemu_del_nic() always attempt
to clean it up.
Peter Maydell [Tue, 5 Feb 2013 20:44:23 +0000 (20:44 +0000)]
tests/test-string-input-visitor: Handle errors provoked by fuzz test
It's OK and expected for visitors to return errors when presented with
the fuzz test's random data. Since the fuzzer doesn't care about
errors, we pass in NULL rather than an Error**. This fixes a bug in
the fuzzer where it was passing the same Error** into each visitor,
with the effect that once one visitor returned an error, each later
visitor would notice that it had been passed in an Error** representing
an already set error, and do nothing.
For the case of visit_type_str() we also need to handle the case where
an error means that the visitor doesn't set our char*. We initialize
the pointer to NULL so we can safely g_free() it regardless of whether
the visitor allocated a string for us or not.
This fixes a problem where this test failed the MacOSX malloc()
consistency checks and might segfault on other platforms [due
to calling free() on an uninitialized pointer variable when
visit_type_str() failed.].
Peter Maydell [Thu, 31 Jan 2013 12:50:40 +0000 (12:50 +0000)]
linux-user: Restore cast to target type in get_user()
Commit 658f2dc97 accidentally dropped the cast to the target type of
the value loaded by get_user(). The most visible effect of this would
be that the sequence "uint64_t v; get_user_u32(v, addr)" would sign
extend the 32 bit loaded value into v rather than zero extending as
would be expected for a _u32 accessor. Put the cast back again to
restore the old behaviour.
Peter Maydell [Sat, 2 Feb 2013 15:13:02 +0000 (15:13 +0000)]
hw/pxa2xx: Fix transposed crn/crm values for pxa2xx cp14 perf regs
When the pxa2xx performance counter related cp14 registers were converted
from a switch-statement implementation to the new table driven cpregs
format in commit dc2a9045c, the crn and crm values for all these
registers were accidentally transposed. Fix this mistake, which was
causing OpenBSD for Zaurus to fail to boot.
Anthony Liguori [Wed, 6 Feb 2013 22:36:16 +0000 (16:36 -0600)]
Merge remote-tracking branch 'stefanha/tracing' into staging
# By Markus Armbruster
# Via Stefan Hajnoczi
* stefanha/tracing:
trace: Fix location of simpletrace.py in docs
trace: Clean up the "try to update atomic until it worked" loops
trace: Direct access of atomics is verboten, use the API
trace: Fix simple trace dropped event record for big endian
Anthony Liguori [Wed, 6 Feb 2013 22:36:11 +0000 (16:36 -0600)]
Merge remote-tracking branch 'stefanha/trivial-patches' into staging
# By Michael Tokarev (1) and Stefan Weil (1)
# Via Stefan Hajnoczi
* stefanha/trivial-patches:
vnc: recognize Hungarian doubleacutes
target-m68k: Fix comment
As a general rule, HMP commands must be built on top of the QMP API.
Luiz and others have worked long & hard to make HMP conform to this
rule.
Commit f1088908 added chardev-add, in violation of this rule. QMP
command chardev-add was added right before, with minimal features, and
the idea to complete it step by step, then switch over the HMP command
to use it.
Unfortunately, we're not there, yet, and we don't want to release with
chardev-add in a "HMP is more powerful than QMP" state.
Disable the HMP command for now, along with its chardev-remove buddy.
qemu-char: Saner naming of memchar stuff & doc fixes
New device, has never been released, so we can still improve things
without worrying about compatibility.
Naming is a mess. The code calls the device driver CirMemCharDriver,
the public API calls it "memory", "memchardev", or "memchar", and the
special commands are named like "memchar-FOO". "memory" is a
particularly unfortunate choice, because there's another character
device driver called MemoryDriver. Moreover, the device's distinctive
property is that it's a ring buffer, not that's in memory. Therefore:
* Rename CirMemCharDriver to RingBufCharDriver, and call the thing a
"ringbuf" in the API.
* Rename QMP and HMP commands from memchar-FOO to ringbuf-FOO.
* Rename device parameter from maxcapacity to size (simple words are
good for you).
* Clearly mark the parameter as optional in documentation.
* Fix error reporting so that chardev-add reports to current monitor,
not stderr.
* Replace cirmem in C identifiers by ringbuf.
* Rework documentation. Document the impact of our crappy UTF-8
handling on reading.
* QMP examples that even work.
I could split this up into multiple commits, but they'd change the
same documentation lines multiple times. Not worth it.
Inline trivial cirmem_chr_is_empty() into its only caller.
Rename qemu_chr_cirmem_count() to cirmem_count().
Fast ring buffer index wraparound. Without this, there's no point in
restricting size to a power two.
qemu_is_chr(chr, "memory") returns *zero* when chr is a memory
character device, which isn't what I'd expect. Replace it by the
saner and more obviously correct chr_is_cirmem(). Also avoids
encouraging testing for specific character devices elsewhere.
Since these values can possibly be sent from guest (for hw/9pfs), do a sanity check
on them. A 9p write request with 0 bytes caused qemu to abort without this patch
e1000: fix link down handling with auto negotiation
Fixes a couple of regression bugs introduced by b9d03e352cb6b31a66545763f6a1e20c9abf0c2c and related to
auto-negotiation:
- Auto-negotiation currently sets link up even if it was
forced down from the monitor.
- If Auto-negotiation was in progress during migration,
link will never come up.
As a fix, don't touch NC link_down field at all,
instead add code on receive path to check
guest link status.
The build is broken on ppc64-linux, possibly only with new binutils:
ld: hw/lm32/../milkymist-tmu2.o: undefined reference to symbol 'XFree'
ld: note: 'XFree' is defined in DSO /lib64/libX11.so.6 so try \
adding it to the linker command line
The misnamed HOST_LONG_BITS is really HOST_POINTER_BITS. Here we're
explicitly using an unsigned long, rather than uintptr_t, so it is
more correct to select the swap size via ULONG_MAX.
Anthony Liguori [Wed, 6 Feb 2013 11:12:06 +0000 (05:12 -0600)]
bios: recompile BIOS
SeaBIOS is really close to spilling over to 256k. Until we can better
handle migration across RAM block size changes, recompile SeaBIOS with
a compiler that causes the binary to still fit in 128k.
This was built with:
gcc version 4.7.2 20121109 (Red Hat 4.7.2-8) (GCC)
trace: Direct access of atomics is verboten, use the API
The GLib Reference Manual says:
It is very important that all accesses to a particular integer or
pointer be performed using only this API and that different sizes
of operation are not mixed or used on overlapping memory
regions. Never read or assign directly from or to a value --
always use this API.
trace: Fix simple trace dropped event record for big endian
We use atomic operations to keep track of dropped events.
Inconveniently, GLib supports only int and void * atomics, but the
counter dropped_events is uint64_t. Can't stop commit 62bab732: a
quick (gint *)&dropped_events bludgeons the compiler into submission.
That cast is okay only when int is exactly 64 bits wide, which it
commonly isn't.
If int is even wider, we clobber whatever follows dropped_events. Not
worth worrying about, as none of the machines that interest us have
such morbidly obese ints.
That leaves the common case: int narrower than 64 bits.
Harmless on little endian hosts: we just don't access the most
significant bits of dropped_events. They remain zero.
On big endian hosts, we use only the most significant bits of
dropped_events as counter. The least significant bits remain zero.
However, we write out the full value, which is the correct counter
shifted left a bunch of places.
Fix by changing the variables involved to int.
There's another, equally suspicious-looking (gint *)&trace_idx
argument to g_atomic_int_compare_and_exchange(), but that one casts
unsigned *, so it's okay. But it's also superfluous, because GLib's
atomic int operations work just fine for unsigned. Drop it.
Anthony Liguori [Mon, 4 Feb 2013 21:22:08 +0000 (15:22 -0600)]
s390x: silence warning from GCC on uninitialized values
As best I can tell, this is a false positive.
[aliguori@ccnode4 qemu-s390]$ make
CC s390x-softmmu/target-s390x/helper.o
/home/aliguori/git/qemu/target-s390x/helper.c: In function ‘do_interrupt’:
/home/aliguori/git/qemu/target-s390x/helper.c:673:17: error: ‘addr’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
/home/aliguori/git/qemu/target-s390x/helper.c:620:20: note: ‘addr’ was declared here
/home/aliguori/git/qemu/target-s390x/helper.c:673:17: error: ‘mask’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
/home/aliguori/git/qemu/target-s390x/helper.c:620:14: note: ‘mask’ was declared here
cc1: all warnings being treated as errors
make[1]: *** [target-s390x/helper.o] Error 1
make: *** [subdir-s390x-softmmu] Error 2
The load function passes the top-level PIIX4PMState pointer to
vmstate_load_state() to handle nested structs for APMState and
pci_status, which leads to corruption of the top-level PIIX4PMState,
since pointers to the nested structs are expected.
A segfault can be fairly reliably triggered by migrating from 1.2 and
issuing a reset, which will trigger a number of QOM operations which
rely on the now corrupted ObjectClass/Object members.
Fix this by passing in the expected pointers for vmstate_load_state().
Eduardo Habkost [Mon, 4 Feb 2013 18:27:52 +0000 (16:27 -0200)]
vl.c: validate -numa "cpus" parameter properly
- Accept empty strings without aborting
- Use parse_uint*() to parse numbers
- Abort if anything except '-' or end-of-string is found after the first
number.
- Check for endvalue < value
Also change the MAX_CPUMASK_BITS warning message from "A max of %d CPUs
are supported in a guest" to "qemu: NUMA: A max of %d VCPUs are
supported".
Eduardo Habkost [Mon, 4 Feb 2013 18:27:46 +0000 (16:27 -0200)]
vl.c: Fix off-by-one bug when handling "-numa node" argument
The numa_add() code was unconditionally adding 1 to the get_opt_name()
return value, making it point after the end of the string if no ','
separator is present.
Example of weird behavior caused by the bug:
$ qemu-img create -f qcow2 this-file-image-has,cpus=5,mem=1000,in-its-name.qcow2 5G
Formatting 'this-file-image-has,cpus=5,mem=1000,in-its-name.qcow2', fmt=qcow2 size=5368709120 encryption=off cluster_size=65536
$ ./x86_64-softmmu/qemu-system-x86_64 -S -monitor stdio -numa node 'this-file-image-has,cpus=5,mem=1000,in-its-name.qcow2'
QEMU 1.3.50 monitor - type 'help' for more information
(qemu) info numa
1 nodes
node 0 cpus: 0
node 0 size: 1000 MB
(qemu)
This changes the code to nove the pointer only if ',' is found.
Eduardo Habkost [Mon, 4 Feb 2013 18:27:45 +0000 (16:27 -0200)]
cutils: unsigned int parsing functions
There are lots of duplicate parsing code using strto*() in QEMU, and
most of that code is broken in one way or another. Even the visitors
code have duplicate integer parsing code[1]. This introduces functions
to help parsing unsigned int values: parse_uint() and parse_uint_full().
Parsing functions for signed ints and floats will be submitted later.
parse_uint_full() has all the checks made by opts_type_uint64() at
opts-visitor.c:
- Check for NULL (returns -EINVAL)
- Check for negative numbers (returns -EINVAL)
- Check for empty string (returns -EINVAL)
- Check for overflow or other errno values set by strtoll() (returns
-errno)
- Check for end of string (reject invalid characters after number)
(returns -EINVAL)
parse_uint() does everything above except checking for the end of the
string, so callers can continue parsing the remainder of string after
the number.
Unit tests included.
[1] string-input-visitor.c:parse_int() could use the same parsing code
used by opts-visitor.c:opts_type_int(), instead of duplicating that
logic.
Andreas Färber [Sun, 27 Jan 2013 06:26:05 +0000 (07:26 +0100)]
target-cris: Build fix for debug output
Around r3361 (81fdc5f8d2d681da8d255baf0713144f8656bac9) env->debug1 used
to contain the address of an MMU fault. This is now written into
env->pregs[PR_EDA] instead.
Paolo Bonzini [Fri, 1 Feb 2013 22:03:16 +0000 (23:03 +0100)]
bitops: unify bitops_ffsl with the one in host-utils.h, call it bitops_ctzl
We had two copies of a ffs function for longs with subtly different
semantics and, for the one in bitops.h, a confusing name: the result
was off-by-one compared to the library function ffsl.
Unify the functions into one, and solve the name problem by calling
the 0-based functions "bitops_ctzl" and "bitops_ctol" respectively.
This also fixes the build on platforms with ffsl, including Mac OS X
and Windows.
Anthony Liguori [Sat, 2 Feb 2013 00:02:50 +0000 (18:02 -0600)]
tap: unbreak -netdev tap,fd=X
The multiqueue patch series broke -netdev tap,fd=X which manifests
as libvirt not being able to start a guest. This was because it
passed NULL for the netdev name which results in an anonymous netdev
device regardless of what the user specified.
Paolo Bonzini [Fri, 25 Jan 2013 13:12:37 +0000 (14:12 +0100)]
qdev: drop extra references at creation time
qdev_free and qbus_free have to do unparent+unref, because nobody else
drops the initial reference (the one included by object_initialize)
before them.
For device_init_func and do_device_add, this is trivially correct,
since the DeviceState goes out of scope.
For qdev_create, qdev_try_create and qbus_init, it is a bit more tricky.
What we are doing here is just assuming that the caller knows what it's
doing, and won't call qdev_free/qbus_free while the device is still there.
This is a pretty reasonable assumption and (behind the scenes) is also
what GObject/GTK does. GTK actually has a "floating reference" that
goes away as soon as the caller does gtk_container_add or something
like that, but in the end qbus_init and qdev_try_create are already
adding the new object to its qdev parent! So in the end the two solutions
are the same.
Paolo Bonzini [Fri, 25 Jan 2013 13:12:34 +0000 (14:12 +0100)]
qdev: move unrealization of devices from finalize to unparent
Similarly, a bus holds a reference back to the device, and this will
prevent the device from going away as soon as this reference is counted
properly. To avoid this, move the unrealization of devices to the
unparent callback. This includes recursively unparenting all the buses
and (after the previous patch) the devices on those buses, which ensures
that the web of references completely disappears for all devices that
reside (in the qdev tree) below the one being unplugged.
After this patch, the qdev tree and the bus<->child relationship is
defined as "A is above B, iff unplugging A will automatically unplug B".
Paolo Bonzini [Fri, 25 Jan 2013 13:12:33 +0000 (14:12 +0100)]
qdev: move deletion of children from finalize to unparent
A device will never be finalized as long as it has a reference from
other devices that sit on its buses. To ensure that the references
go away, deassociate a bus from its children in the unparent callback
for the bus.
Paolo Bonzini [Fri, 25 Jan 2013 13:12:28 +0000 (14:12 +0100)]
qdev: change first argument of qbus_create_inplace to void *
Make it clear that no BUS() macro is needed in the callers (in fact it
wouldn't work because the object has not been initialized yet with the
right class).
liguang [Thu, 24 Jan 2013 05:03:26 +0000 (13:03 +0800)]
vl: correct error message when fail to init kvm
command:
qemu-system-x86_64 -hda disk.img -smp 32 --enable-kvm
error:
Number of SMP cpus requested (32) exceeds max cpus supported by KVM (16)
failed to initialize KVM: Invalid argument
No accelerator found!
well, it did find kvm, but failed to init,
so message "No accelerator found!" is confusing,
this commit remove the confusing error message.
Anthony Liguori [Tue, 29 Jan 2013 21:42:45 +0000 (15:42 -0600)]
sparc: disable qtest in make check
We've seen this repeatedly in buildbot but I can now reliably
reproduce it myself too. With a few hundred runs of 'make check',
qemu-system-sparc will hang consuming 100% CPU. I've attached GDB
to the hung process and unfortunately, I can't get anything useful
out of GDB (RIP is not a valid simple and there is nothing else on
the stack).
At any rate, since this only manifests in qemu-system-sparc and it
doesn't appear to be a qtest specific problem, I think we should
disable it until the problem is resolved.
Anthony Liguori [Fri, 1 Feb 2013 20:40:05 +0000 (14:40 -0600)]
Merge remote-tracking branch 'stefanha/block' into staging
# By Kevin Wolf (7) and others
# Via Stefan Hajnoczi
* stefanha/block:
block/raw-posix: Build fix for O_ASYNC
vmdk: Allow space in file name
parallels: Fix bdrv_open() error handling
dmg: Use g_free instead of free
dmg: Fix bdrv_open() error handling
vpc: Fix bdrv_open() error handling
cloop: Fix bdrv_open() error handling
bochs: Fix bdrv_open() error handling
sheepdog: pass vdi_id to sheep daemon for sd_close()
vmdk: Allow selecting SCSI adapter in image creation
block: Adds mirroring tests for resized images
block: Fix is_allocated_above with resized files
qemu-iotests: Add regression test for b7ab0fea
Jason Wang [Wed, 30 Jan 2013 11:12:40 +0000 (19:12 +0800)]
virtio-net: migration support for multiqueue
This patch add migration support for multiqueue virtio-net. Instead of bumping
the version, we conditionally send the info of multiqueue only when the device
support more than one queue to maintain the backward compatibility.
Jason Wang [Wed, 30 Jan 2013 11:12:39 +0000 (19:12 +0800)]
virtio-net: multiqueue support
This patch implements both userspace and vhost support for multiple queue
virtio-net (VIRTIO_NET_F_MQ). This is done by introducing an array of
VirtIONetQueue to VirtIONet.
Jason Wang [Wed, 30 Jan 2013 11:12:38 +0000 (19:12 +0800)]
virtio-net: separate virtqueue from VirtIONet
To support multiqueue virtio-net, the first step is to separate the virtqueue
related fields from VirtIONet to a new structure VirtIONetQueue. The following
patches will add an array of VirtIONetQueue to VirtIONet based on this patch.
Jason Wang [Wed, 30 Jan 2013 11:12:35 +0000 (19:12 +0800)]
vhost: multiqueue support
This patch lets vhost support multiqueue. The idea is simple, just launching
multiple threads of vhost and let each of vhost thread processing a subset of
the virtqueues of the device. After this change each emulated device can have
multiple vhost threads as its backend.
To do this, a virtqueue index were introduced to record to first virtqueue that
will be handled by this vhost_net device. Based on this and nvqs, vhost could
calculate its relative index to setup vhost_net device.
Since we may have many vhost/net devices for a virtio-net device. The setting of
guest notifiers were moved out of the starting/stopping of a specific vhost
thread. The vhost_net_{start|stop}() were renamed to
vhost_net_{start|stop}_one(), and a new vhost_net_{start|stop}() were introduced
to configure the guest notifiers and start/stop all vhost/vhost_net devices.
Jason Wang [Wed, 30 Jan 2013 11:12:34 +0000 (19:12 +0800)]
tap: multiqueue support
Recently, linux support multiqueue tap which could let userspace call TUNSETIFF
for a signle device many times to create multiple file descriptors as
independent queues. User could also enable/disabe a specific queue through
TUNSETQUEUE.
The patch adds the generic infrastructure to create multiqueue taps. To achieve
this a new parameter "queues" were introduced to specify how many queues were
expected to be created for tap by qemu itself. Alternatively, management could
also pass multiple pre-created tap file descriptors separated with ':' through a
new parameter fds like -netdev tap,id=hn0,fds="X:Y:..:Z". Multiple vhost file
descriptors could also be passed in this way.
Each TAPState were still associated to a tap fd, which mean multiple TAPStates
were created when user needs multiqueue taps. Since each TAPState contains one
NetClientState, with the multiqueue nic support, an N peers of NetClientState
were built up.
A new parameter, mq_required were introduce in tap_open() to create multiqueue
tap fds.
Jason Wang [Wed, 30 Jan 2013 11:12:33 +0000 (19:12 +0800)]
tap: introduce a helper to get the name of an interface
This patch introduces a helper tap_get_ifname() to get the device name of tap
device. This is needed when ifname is unspecified in the command line and qemu
were asked to create tap device by itself. In this situation, the name were
allocated by kernel, so if multiqueue is asked, we need to fetch its name after
creating the first queue.
Only linux has this support since it's the only platform that supports
multiqueue tap.
Jason Wang [Wed, 30 Jan 2013 11:12:32 +0000 (19:12 +0800)]
tap: support enabling or disabling a queue
This patch introduce a new bit - enabled in TAPState which tracks whether a
specific queue/fd is enabled. The tap/fd is enabled during initialization and
could be enabled/disabled by tap_enalbe() and tap_disable() which calls platform
specific helpers to do the real work. Polling of a tap fd can only done when
the tap was enabled.
Jason Wang [Wed, 30 Jan 2013 11:12:31 +0000 (19:12 +0800)]
tap: add Linux multiqueue support
This patch add basic multiqueue support for Linux. When multiqueue is needed, we
will first check whether kernel support multiqueue tap before creating more
queues. Two new functions tap_fd_enable() and tap_fd_disable() were introduced
to enable and disable a specific queue. Since the multiqueue is only supported
in Linux, return error on other platforms.
Jason Wang [Wed, 30 Jan 2013 11:12:29 +0000 (19:12 +0800)]
tap: import linux multiqueue constants
Import multiqueue constants from if_tun.h from 3.8-rc3. A new ifr flag
IFF_MULTI_QUEUE were introduced to create a multiqueue backend by calling
TUNSETIFF with the this flag and with the same interface name many times.
A new ioctl TUNSETQUEUE were introduced. When doing this ioctl with
IFF_DETACH_QUEUE, the queue were disabled in the linux kernel. When doing this
ioctl with IFF_ATTACH_QUEUE, the queue were enabled in the linux kernel.
Jason Wang [Wed, 30 Jan 2013 11:12:28 +0000 (19:12 +0800)]
net: multiqueue support
This patch adds basic multiqueue support for qemu. The idea is simple, an array
of NetClientStates were introduced in NICState, parse_netdev() were extended to
find and match all NetClientStates belongs to the backend and place their
pointers in NICConf. Then qemu_new_nic can setup a N:N mapping between NICStates
that belongs to a nic and NICStates belongs to the netdev. And a queue_index
were introduced in NetClientState to track its index. After this, each peers of
a NICState were abstracted as a queue.
After this change, all NetClientState that belongs to the same backend/nic has
the same id. When use want to change the link status, all NetClientStates that
belongs to the same backend/nic will be also changed. When user want to delete
a device or netdev, all NetClientStates that belongs to the same backend/nic
will be deleted also. Changing or deleting an specific queue is not allowed.
Jason Wang [Wed, 30 Jan 2013 11:12:27 +0000 (19:12 +0800)]
net: introduce NetClientState destructor
To allow allocating an array of NetClientState and free it once, this patch
introduces destructor of NetClientState. Which could do type specific free,
which could be used by multiqueue to free the array once.
projects / qemu.git / log