kvm: Make KVMState be the TYPE_KVM_ACCEL instance struct
Now that we create an accel object before calling machine_init, we can
simply use the accel object to save all KVMState data, instead of
allocationg KVMState manually.
accel: Create accel object when initializing machine
Create an actual TYPE_ACCEL object when initializing a machine. This
will allow accelerator classes to implement some initialization on
instance_init, and to save state on the TYPE_ACCEL object.
accel: Pass MachineState object to accel init functions
Most of the machine options and machine state information is in the
MachineState object, not on the MachineClass. This will allow init
functions to use the MachineState object directly instead of
qemu_get_machine_opts() or the current_machine global.
Today, all accelerator init functions affect some global state:
* tcg_init() calls tcg_exec_init() and affects globals such as tcg_tcx,
page size globals, and possibly others;
* kvm_init() changes the kvm_state global, cpu_interrupt_handler, and possibly
others;
* xen_init() changes the xen_xc global, and registers a change state handler.
With the new accelerator QOM classes, initialization may now be split in two
steps:
* instance_init() will do basic initialization that doesn't affect any global
state and don't need MachineState or MachineClass data. This will allow
probing code to safely create multiple accelerator objects on the fly just
for reporting host/accelerator capabilities, for example.
* accel_init_machine()/init_machine() will save the accelerator object in
MachineState, and do initialization steps which still affect global state,
machine state, or that need data from MachineClass or MachineState.
To clarify the difference between those two steps, rename init() to
init_machine().
As qtest_availble() returns 1 only when CONFIG_POSIX is set, keep
setting AccelClass.available to keep current behavior (this is different
from what we did for KVM and Xen).
This also allows us to make qtest_init_accel() static.
Note that this has an user-visible side-effect: instead of reporting
"Xen is not supported for this target", QEMU binaries not supporting Xen
will report "xen accelerator does not exist".
As xen_available() always return 1 when CONFIG_XEN is enabled, we don't
need to set AccelClass.available anymore. xen_enabled() is not being
removed yet, but only because vl.c is still using it.
Note that this has an user-visible side-effect: instead of reporting
"KVM is not supported for this target", QEMU binaries not supporting KVM
will report "kvm accelerator does not exist".
As kvm_availble() always return 1 when CONFIG_KVM is enabled, we don't
need to set AccelClass.available anymore. kvm_enabled() is not being
completely removed yet only because qmp_query_kvm() still uses it.
accel: Report unknown accelerator as "not found" instead of "does not exist"
As the accelerator classes won't be registered anymore if they are not
enabled at compile time, saying "does not exist" may be misleading, as
the accelerator may be simply disabled. Change the wording to just say
"not found".
When we move accel classes outside accel.c, the available() function
won't be necessary anymore, because the classes will be registered only
if the accelerator code is really enabled at build time.
Corey Minyard [Thu, 2 Oct 2014 16:17:37 +0000 (11:17 -0500)]
qemu-char: Add reconnecting to client sockets
Adds a "reconnect" option to socket backends that gives a reconnect
timeout. This only applies to client sockets. If the other end
of a socket closes the connection, qemu will attempt to reconnect
after the given number of seconds.
Corey Minyard [Thu, 2 Oct 2014 16:17:36 +0000 (11:17 -0500)]
qemu-char: set socket filename to disconnected when not connected
This way we can tell if the socket is connected or not. It also splits
the string conversions out into separate functions to make this more
convenient.
Corey Minyard [Thu, 2 Oct 2014 16:17:34 +0000 (11:17 -0500)]
qemu-char: Rework qemu_chr_open_socket() for reconnect
Move all socket configuration to qmp_chardev_open_socket().
qemu_chr_open_socket_fd() just opens the socket. This is getting ready
for the reconnect code, which will call open_sock_fd() on a reconnect
attempt.
Peter Maydell [Tue, 30 Sep 2014 15:45:34 +0000 (16:45 +0100)]
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
This update brings dataplane to virtio-scsi (NOT
yet 100% thread-safe, though, which makes it really, really
experimental. It also brings asynchronous cancellation to
the SCSI subsystem and implements it in virtio-scsi. This
is a pretty important feature. Almost all the work here
was done by Fam Zheng.
I also included the virtio refcount fixes from Gonglei,
because they had a small conflict with virtio-scsi dataplane.
This pull request is using the new subkey 4E6B09D7.
# gpg: Signature made Tue 30 Sep 2014 12:31:02 BST using RSA key ID 4E6B09D7
# gpg: Good signature from "Paolo Bonzini <[email protected]>"
# gpg: aka "Paolo Bonzini <[email protected]>"
* remotes/bonzini/tags/for-upstream: (39 commits)
block/iscsi: handle failure on malloc of the allocationmap
util: introduce bitmap_try_new
virtio-scsi: Handle TMF request cancellation asynchronously
scsi: Introduce scsi_req_cancel_async
scsi: Introduce scsi_req_cancel_complete
scsi: Drop SCSIReqOps.cancel_io
scsi: Unify request unref in scsi_req_cancel
scsi-generic: Handle canceled request in scsi_command_complete
scsi: Drop scsi_req_abort
virtio-scsi: Process ".iothread" property
virtio-scsi: Call bdrv_io_plug/bdrv_io_unplug in cmd request handling
virtio-scsi: Batched prepare for cmd reqs
virtio-scsi: Two stages processing of cmd request
virtio-scsi: Add migration state notifier for dataplane code
virtio-scsi: Hook up with dataplane
virtio-scsi-dataplane: Code to run virtio-scsi on iothread
virtio-scsi: Add VirtIOSCSIVring in VirtIOSCSIReq
virtio-scsi: Add 'iothread' property to virtio-scsi
virtio: add a wrapper for virtio-backend initialization
virtio-9p: fix virtio-9p child refcount in transports
...
Peter Maydell [Tue, 30 Sep 2014 12:09:39 +0000 (13:09 +0100)]
Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging
pci, pc, virtio, misc bugfixes
A bunch of bugfixes.
Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Mon 29 Sep 2014 17:59:57 BST using RSA key ID D28D5469
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>"
# gpg: aka "Michael S. Tsirkin <[email protected]>"
* remotes/mst/tags/for_upstream:
vl: Adjust the place of calling mlockall to speedup VM's startup
pc-dimm: Don't check dimm->node when there is non-NUMA config
pci-hotplug-old: avoid losing error message
Revert "virtio-pci: fix migration for pci bus master"
loader: g_realloc(p, 0) frees and returns NULL, simplify
Devices will call this function to start an asynchronous cancellation. The
bus->info->cancel will be called after the request is canceled.
Devices will probably need to track a separate TMF request that triggers this
cancellation, and wait until the cancellation is done before completing it. So
we store a notifier list in SCSIRequest and in scsi_req_cancel_complete we
notify them.
Before, scsi_req_cancel will take ownership of the canceled request and unref
it. We did this because we didn't know whether AIO CB will be called or not
during the cancelling, so we set the io_canceled flag before calling it, and
skip unref in the potentially called callbacks, which is not very nice.
Now, bdrv_aio_cancel has a stricter contract that the completion callbacks are
always called, so we can remove the checks of req->io_canceled and just unref
it in callbacks.
It will also make implementing asynchronous cancellation easier.
Peter Maydell [Tue, 30 Sep 2014 10:52:41 +0000 (11:52 +0100)]
Merge remote-tracking branch 'remotes/rth/tags/tcg-next-201400729' into staging
tcg updates
# gpg: Signature made Mon 29 Sep 2014 19:58:04 BST using RSA key ID 4DD0279B
# gpg: Good signature from "Richard Henderson <[email protected]>"
# gpg: aka "Richard Henderson <[email protected]>"
# gpg: aka "Richard Henderson <[email protected]>"
* remotes/rth/tags/tcg-next-201400729:
tcg: Always enable TCGv type checking
qemu/compiler: Define QEMU_ARTIFICIAL
tcg-aarch64: Use 32-bit loads for qemu_ld_i32
tcg-sparc: Use UMULXHI instruction
tcg-sparc: Rename ADDX/SUBX insns
tcg-sparc: Use ADDXC in setcond_i64
tcg-sparc: Fix setcond_i32 uninitialized value
tcg-sparc: Use ADDXC in addsub2_i64
tcg-sparc: Support addsub2_i64
Peter Maydell [Tue, 30 Sep 2014 10:02:06 +0000 (11:02 +0100)]
Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20140929' into staging
target-arm:
* more EL2/EL3 preparation work
* don't handle c15_cpar changes via tb_flush()
* fix some unused function warnings in ARM devices
* build the GDB XML for 32 bit CPUs into qemu-*-aarch64
* implement guest breakpoint support
# gpg: Signature made Mon 29 Sep 2014 19:25:37 BST using RSA key ID 14360CDE
# gpg: Good signature from "Peter Maydell <[email protected]>"
* remotes/pmaydell/tags/pull-target-arm-20140929:
target-arm: Add support for VIRQ and VFIQ
target-arm: Add IRQ and FIQ routing to EL2 and 3
target-arm: A64: Emulate the SMC insn
target-arm: Add a Hypervisor Trap exception type
target-arm: A64: Emulate the HVC insn
target-arm: A64: Correct updates to FAR and ESR on exceptions
target-arm: Don't take interrupts targeting lower ELs
target-arm: Break out exception masking to a separate func
target-arm: A64: Refactor aarch64_cpu_do_interrupt
target-arm: Add SCR_EL3
target-arm: Add HCR_EL2
target-arm: Don't handle c15_cpar changes via tb_flush()
hw/input/tsc210x.c: Delete unused array tsc2101_rates
hw/display/pxa2xx_lcd.c: Remove unused function pxa2xx_dma_rdst_set
hw/intc/imx_avic.c: Remove unused function imx_avic_set_prio()
hw/display/blizzard.c: Delete unused function blizzard_rgb2yuv
configure: Build GDB XML for 32 bit ARM CPUs into qemu aarch64 binaries
target-arm: Implement handling of breakpoint firing
target-arm: Implement setting guest breakpoints
This enables the virtio-scsi-dataplane code by setting the iothread
in virtio-scsi device, and makes any function that is called by
back from dataplane to cooperate with the caller: they need to be
vring/iothread aware when handling the requests and using scsi devices
on the bus.
virtio-scsi-dataplane: Code to run virtio-scsi on iothread
This implements the core part of dataplane feature of virtio-scsi.
A few fields are added in VirtIOSCSICommon to maintain the dataplane
status. These fields are managed by a new source file:
virtio-scsi-dataplane.c.
Most code in this file will run on an iothread, unless otherwise
commented as in a global mutex context, such as those functions to
start, stop and setting the iothread property.
Upon start, we set up guest/host event notifiers, in a same way as
virtio-blk does. The handlers then pop request from vring and call into
virtio-scsi.c functions to process it. So we need to make sure make all
those called functions work with iothread, too.
Fam Zheng [Wed, 6 Aug 2014 05:35:06 +0000 (13:35 +0800)]
virtio-scsi: Add 'iothread' property to virtio-scsi
Similar to this property in virtio-blk for dataplane, add it as a QOM
link in virtio-scsi and an alias in virtio-scsi-pci and virtio-scsi-ccw,
in order to assign an iothread to the device.
virtio-9p: fix virtio-9p child refcount in transports
object_initialize() leaves the object with a refcount of 1.
object_property_add_child() adds its own reference which is
dropped again when the property is deleted.
The upshot of this is that we always have a refcount >= 1. Upon
unplug the virtio-9p child is not finalized!
Drop our reference after the child property has been added to the
parent.
virtio-9p: use aliases instead of duplicate qdev properties
virtio-9p-pci all duplicate the qdev properties of their
V9fsState child. This approach does not work well with
string or pointer properties since we must be careful
about leaking or double-freeing them.
Use the QOM alias property to forward property accesses to the
V9fsState child. This way no duplication is necessary.
virtio-balloon: fix virtio-balloon child refcount in transports
object_initialize() leaves the object with a refcount of 1.
object_property_add_child() adds its own reference which is dropped
again when the property is deleted.
The upshot of this is that we always have a refcount >= 1. Upon hot
unplug the virtio-balloon child is not finalized!
Drop our reference after the child property has been added to the
parent.
virtio-rng: fix virtio-rng child refcount in transports
object_initialize() leaves the object with a refcount of 1.
object_property_add_child() adds its own reference which is dropped
again when the property is deleted.
The upshot of this is that we always have a refcount >= 1. Upon hot
unplug the virtio-rng child is not finalized!
Drop our reference after the child property has been added to the
parent.
virtio-rng: use aliases instead of duplicate qdev properties
virtio-rng-{pci, s390, ccw} all duplicate the
qdev properties of their VirtIORNG child.
This approach does not work well with string or pointer
properties since we must be careful about leaking or
double-freeing them.
Use the QOM alias property to forward property accesses to the
VirtIORNG child. This way no duplication is necessary.
virtio-serial: fix virtio-serial child refcount in transports
object_initialize() leaves the object with a refcount of 1.
object_property_add_child() adds its own reference which is dropped
again when the property is deleted.
The upshot of this is that we always have a refcount >= 1. Upon hot
unplug the virtio-serial child is not finalized!
Drop our reference after the child property has been added to the
parent.
virtio-serial: use aliases instead of duplicate qdev properties
virtio-serial-{pci, s390, ccw} all duplicate the
qdev properties of their VirtIOSerial child.
This approach does not work well with string or pointer
properties since we must be careful about leaking or
double-freeing them.
Use the QOM alias property to forward property accesses to the
VirtIOSerial child. This way no duplication is necessary.
virtio/vhost-scsi: fix virtio-scsi/vhost-scsi child refcount in transports
object_initialize() leaves the object with a refcount of 1.
object_property_add_child() adds its own reference which is dropped
again when the property is deleted.
The upshot of this is that we always have a refcount >= 1. Upon hot
unplug the virtio-scsi/vhost-scsi child is not finalized!
Drop our reference after the child property has been added to the
parent.
virtio/vhost-scsi: use aliases instead of duplicate qdev properties
{virtio, vhost}-scsi-{pci, s390, ccw} all duplicate the
qdev properties of their VirtIOSCSI/VHostSCSI child.
This approach does not work well with string or pointer
properties since we must be careful about leaking or
double-freeing them.
Use the QOM alias property to forward property accesses to the
VirtIOSCSI/VHostSCSI child. This way no duplication is necessary.
virtio-net: fix virtio-net child refcount in transports
object_initialize() leaves the object with a refcount of 1.
object_property_add_child() adds its own reference which is dropped
again when the property is deleted.
The upshot of this is that we always have a refcount >= 1. Upon hot
unplug the virtio-net child is not finalized!
Drop our reference after the child property has been added to the
parent.
virtio-net: use aliases instead of duplicate qdev properties
virtio-net-pci, virtio-net-s390, and virtio-net-ccw all duplicate the
qdev properties of their VirtIONet child. This approach does not work
well with string or pointer properties since we must be careful about
leaking or double-freeing them.
Use the QOM alias property to forward property accesses to the
VirtIONet child. This way no duplication is necessary.
The combination of always_inline + artificial allows tiny inline
functions to be written that do not interfere with debugging.
In particular, gdb will not step into an artificial function.
The always_inline attribute was introduced in gcc 4.2,
and the artificial attribute was introduced in gcc 4.3.
The "old" qemu_ld opcode did not specify the size of the result,
and so we had to assume full register width. With the new opcodes,
we can narrow the result.
Peter Maydell [Mon, 29 Sep 2014 17:48:48 +0000 (18:48 +0100)]
target-arm: Don't handle c15_cpar changes via tb_flush()
At the moment we try to handle c15_cpar with the strategy of:
* emit generated code which makes assumptions about its value
* when the register value changes call tb_flush() to throw
away the now-invalid generated code
This works because XScale CPUs are always uniprocessor, but
it's confusing because it suggests that the same approach can
be taken for other registers. It also means we do a tb_flush()
on CPU reset, which makes multithreaded linux-user binaries
even more likely to fail than would otherwise be the case.
Replace it with a combination of TB flags for the access
checks done on cp0/cp1 for the XScale and iwMMXt instructions,
plus a runtime check for cp2..cp13 coprocessor accesses.
Peter Maydell [Mon, 29 Sep 2014 17:48:47 +0000 (18:48 +0100)]
configure: Build GDB XML for 32 bit ARM CPUs into qemu aarch64 binaries
The qemu-aarch64 and qemu-system-aarch64 binaries include support
for all the 32 bit ARM CPUs as well as the 64 bit ones. This means
we need to build in the GDB XML files for the 32 bit CPUs too.
Otherwise gdb will complain:
warning: while parsing target description (at line 1): Could not load XML document "arm-core.xml"
when you try to connect to our gdbserver to debug a 32 bit CPU
running in a qemu-aarch64 or qemu-system-aarch64 binary.
Peter Maydell [Mon, 29 Sep 2014 17:48:46 +0000 (18:48 +0100)]
target-arm: Implement handling of breakpoint firing
Implement handling of breakpoint event firing to correctly
inject the debug exception into the guest.
Since the breakpoint and watchpoint control register format is
very similar we adjust wp_matches() to also handle breakpoints
as well rather than using a separate function.
Peter Maydell [Mon, 29 Sep 2014 17:48:46 +0000 (18:48 +0100)]
target-arm: Implement setting guest breakpoints
This patch adds support for setting guest breakpoints
based on values the guest writes to the DBGBVR and DBGBCR
registers. (It doesn't include the code to handle when
these breakpoints fire, so has no guest-visible effect.)
vl: Adjust the place of calling mlockall to speedup VM's startup
If we configure mlock=on and memory policy=bind at the same time,
It will consume lots of time for system to treat with memory,
especially when call mbind behind mlockall.
Adjust the place of calling mlockall, calling mbind before mlockall
can remarkably reduce the time of VM's startup.
pc-dimm: Don't check dimm->node when there is non-NUMA config
It should not break memory hotplug feature if there is non-NUMA option.
This patch would also allow to use pc-dimm as replacement for initial memory
for non-NUMA configs.
Note: After this patch, the memory hotplug can work normally for Linux guest OS
when there is non-NUMA option and NUMA option. But not support Windows guest OS
to hotplug memory with no-NUMA config, actully, it's Windows limitation.
loader: g_realloc(p, 0) frees and returns NULL, simplify
Once upon a time, it was decided that qemu_realloc(ptr, 0) should
abort. Switching to glib retired that bright idea. A bit of code
that was added to cope with it (commit 3e372cf) is still around. Bury
it.
Stefan Hajnoczi [Mon, 29 Sep 2014 15:40:12 +0000 (16:40 +0100)]
libqos: use microseconds instead of iterations for virtio timeout
Some hosts are slow or overloaded so test execution takes a long time.
Test cases use timeouts to protect against an infinite loop stalling the
test forever (especially important in automated test setups).
Commit 6cd14054b67774cc58a51fca6660cfa1d3c08059 ("libqos virtio:
Increase ISR timeout") increased the clock_step() value in an attempt to
lengthen the virtio interrupt wait timeout, but timeout failures are
still occuring on the Travis automated testing platform.
This is because clock_step() only affects the guest's virtual time.
Virtio requests can be bottlenecked on host disk I/O latency - which
cannot be improved by stepping the clock, so the fix was ineffective.
This patch changes the qvirtio_wait_queue_isr() and
qvirtio_wait_config_isr() timeout mechanism from loop iterations to
microseconds. This way the test case can specify an absolute 30 second
timeout. Number of loop iterations is not a reliable timeout mechanism
since the speed depends on many factors including host performance.
Tests should no longer timeout on overloaded Travis instances.
Stefan Hajnoczi [Mon, 29 Sep 2014 15:40:11 +0000 (16:40 +0100)]
libqos: improve event_index test with timeout
The virtio event_index feature lets the device driver tell the device
how many requests to process before raising the next interrupt.
virtio-blk-test.c tries to verify that the device does not raise an
interrupt unnecessarily.
Unfortunately the test has a race condition. It spins checking for an
interrupt up to 100 times and then assumes the request has finished. On
a slow host the I/O request could still be in flight and the test would
fail.
This patch waits for the request to complete, or until a 30-second
timeout is reached. If an interrupt is raised while waiting the test
fails since the device was not supposed to raise interrupts.
Peter Maydell [Mon, 29 Sep 2014 11:26:14 +0000 (12:26 +0100)]
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Block patches
# gpg: Signature made Fri 26 Sep 2014 19:57:52 BST using RSA key ID C88F2FD6
# gpg: Good signature from "Kevin Wolf <[email protected]>"
* remotes/kevin/tags/for-upstream:
qemu-iotests: Fail test if explicit test case number is unknown
block: Validate node-name
vpc: fix beX_to_cpu() and cpu_to_beX() confusion
docs: add blkdebug block driver documentation
block: Catch simultaneous usage of options and their aliases
block: Specify -drive legacy option aliases in array
block: Improve message for device name clashing with node name
qemu-nbd: Destroy the BlockDriverState properly
block: Keep DriveInfo alive until BlockDriverState dies
blockdev: Disentangle BlockDriverState and DriveInfo creation
blkdebug: show an error for invalid event names
Simply switch function pointers when entering/leaving vga mode.
Allows to remove wrapper functions which do nothing but dispatch
calls depending on the current qxl mode.
Add a function to allow display emulations to switch the hwops
function pointers. This is useful for devices which have two
completely different operation modes. Typical case is the vga
compatibility mode vs. native mode in qxl and the upcoming
virtio-vga device.
projects / qemu.git / log