qtest: fix out of tree build

Trace objects are also needed if tracing is enabled.

Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

qtest: add rtc-test test-case

Signed-off-by: Anthony Liguori <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

rtc: split out macros into a header file and use in test case

Signed-off-by: Anthony Liguori <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

qtest: add C version of test infrastructure

This also includes a qtest wrapper script to make it easier to launch qtest
tests directly.

Signed-off-by: Anthony Liguori <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

qtest: add clock management

This patch combines qtest and -icount together to turn the vm_clock
into a source that can be fully managed by the client.  To this end new
commands clock_step and clock_set are added.  Hooking them with libqtest
is left as an exercise to the reader.

Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

qtest: IRQ interception infrastructure

Since /i440fx/piix3 is being removed from the composition tree, the
IO-APIC is placed under /i440fx.  This is wrong and should be changed
as soon as the /i440fx/piix3 path is put back.

Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

qtest: add test framework

The idea behind qtest is pretty simple.  Instead of executing a CPU via TCG or
KVM, rely on an external process to send events to the device model that the CPU
would normally generate.

qtest presents itself as an accelerator.  In addition, a new option is added to
establish a qtest server (-qtest) that takes a character device.  This is what
allows the external process to send CPU events to the device model.

qtest uses a simple line based protocol to send the events.  Documentation of
that protocol is in qtest.c.

I considered reusing the monitor for this job.  Adding interrupts would be a bit
difficult.  In addition, logging would also be difficult.

qtest has extensive logging support.  All protocol commands are logged with
time stamps using a new command line option (-qtest-log).  Logging is important
since ultimately, this is a feature for debugging.

Signed-off-by: Anthony Liguori <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

test makefile overhaul

This introduces new test reporting infrastructure based on
gtester and gtester-report.

Also, all existing tests are moved to tests/, and tests/Makefile
is reorganized to factor out the commonalities in the rules.

Signed-off-by: Anthony Liguori <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'origin/master' into staging

* origin/master:
  qemu tcg: Remove one entry of INDEX_op_ld_i64 from ppc_op_defs

qemu tcg: Remove one entry of INDEX_op_ld_i64 from ppc_op_defs

There two entries of INDEX_op_ld_i64 in the ppc_op_defs.  That causes an
assertion failure in tcg_add_target_add_op_defs() when --enable-debug is
used on a ppc64 backend (that's ppc64 host, not target).

Signed-off-by: Li Zhang <[email protected]>
Signed-off-by: David Gibson <[email protected]>
Signed-off-by: malc <[email protected]>

Merge remote-tracking branch 'sstabellini/disk_io' into staging

* sstabellini/disk_io:
  xen_disk: when using AIO flush after the operation is completed
  xen_disk: open disk with BDRV_O_NOCACHE | BDRV_O_CACHE_WB | BDRV_O_NATIVE_AIO

Merge remote-tracking branch 'sstabellini/xen-fixes' into staging

* sstabellini/xen-fixes:
  xen_disk: detach the blkdev before bdrv_delete
  xen_console: ignore console disconnect events from console/0

Merge remote-tracking branch 'qmp/queue/qmp' into staging

* qmp/queue/qmp:
  qmp: document strict parsing
  qmp: parse commands in strict mode
  qmp: add and use q type specifier
  qapi: add strict mode to input visitor
  qapi: place outermost object on qiv stack
  qapi: untangle next_list
  qapi: allow freeing partially-allocated objects
  qapi: shortcut visits on errors
  qapi: fix memory leak on error
  qapi: fail hard on stack imbalance
  qapi: add a test case for type errors
  qapi: add struct-errors test case to test-qmp-output-visitor
  qapi: fix double free in qmp_output_visitor_cleanup()

Merge remote-tracking branch 'alon/libcacard' into staging

* alon/libcacard:
  libcacard/vcard_emul_nss: add warning for old coolkey
  libcacard/vcard_emul_nss: handle no readers at startup
  libcacard/vcard_emul_nss: don't stop thread when there are no slots

sparc: pass page aligned addresses to tlb_set_page

Mask incoming page address early so that resolved addresses
are page aligned. Remove further address masking.

Tested-by: Artyom Tarasenko <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

xen_disk: detach the blkdev before bdrv_delete

We need to detach the blkdev from the BlockDriverState before calling
bdrv_delete.

Signed-off-by: Stefano Stabellini <[email protected]>

xen_console: ignore console disconnect events from console/0

The first console has a different location compared to other PV devices
(console, rather than device/console/0) and doesn't obey the xenstore
state protocol. We already special case the first console in con_init
and con_initialise, we should also do it in con_disconnect.

Signed-off-by: Stefano Stabellini <[email protected]>

qmp: document strict parsing

Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qmp: parse commands in strict mode

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-off-by: Anthony Liguori <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qmp: add and use q type specifier

"O" is being used by the transaction and qom-set commands to mean "any
QObject", but it really means "do not validate the argument list".
Add a new specifier with the correct meaning.

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: add strict mode to input visitor

While QMP in general is designed so that it is possible to ignore
unknown arguments, in the case of the QMP server it is better to
reject them to detect bad clients.  In fact, we're already doing
this at the top level in the argument checker.  To extend this to
complex structures, add a mode to the input visitor where it checks
for unvisited keys and raises an error if it finds one.

Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: place outermost object on qiv stack

This is a slight change in the implementation of QMPInputVisitor
that helps when adding strict mode.

Const QObjects cannot be inc/decref-ed, and that's why QMPInputVisitor
relies heavily on weak references to inner objects.  I'm not removing
the weak references now, but since refcount+const is a lost battle in C
(C++ has "mutable") I think removing const is fine in this case.

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: untangle next_list

Right now, the semantics of next_list are complicated.  The caller must:

* call start_list

* call next_list for each element *including the first*

* on the first call to next_list, the second argument should point to
NULL and the result is the head of the list.  On subsequent calls,
the second argument should point to the last node (last result of
next_list) and next_list itself tacks the element at the tail of the
list.

This works for both input and output visitor, but having the visitor
write memory when it is only reading the list is ugly.  Plus, relying
on *list to detect the first call is tricky and undocumented.

We can initialize so->entry in next_list instead of start_list, leaving
it NULL in start_list.  This way next_list sees clearly whether it is
on the first call---as a bonus, it discriminates the cases based on
internal state of the visitor rather than external state.  We can
also pull the assignment of the list head from generated code up to
next_list.

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Michael Roth <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: allow freeing partially-allocated objects

Objects going through the dealloc visitor can be only partially allocated.
Detect the situation and avoid a segfault.  This also helps with the
input visitor, when there are errors.

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: shortcut visits on errors

We can exit very soon if we enter a visitor with a preexisting error.
This simplifies some cases because we will not have to deal with
obj being non-NULL while *obj is NULL.

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: fix memory leak on error

QmpInputVisitor would leak the malloced struct if the stack was
overflowed.  This can be easily fixed using error_propagate.

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: fail hard on stack imbalance

QmpOutputVisitor will segfault if an imbalanced end function is
called.  So we can abort in QmpInputVisitor too.

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: add a test case for type errors

There is no test case for parse errors, add one.

Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: add struct-errors test case to test-qmp-output-visitor

This test case verifies that invalid native enums are caught, and causes
qapi to tear down the QObject tree under construction, exercising the
previous patch.

Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Laszlo Ersek <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qapi: fix double free in qmp_output_visitor_cleanup()

Stack entries in QmpOutputVisitor are navigation links (weak references),
except the bottom (ie. least recently added) entry, which owns the root
QObject [1]. Make qmp_output_visitor_cleanup() drop the stack entries,
then release the QObject tree by the root.

Attempting to serialize an invalid enum inside a dictionary is an example
for triggering the double free.

[1] http://lists.nongnu.org/archive/html/qemu-devel/2012-03/msg03276.html

Signed-off-by: Laszlo Ersek <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

Merge remote-tracking branch 'mdroth/qga-pull-3-26-12' into staging

* mdroth/qga-pull-3-26-12:
  qemu-ga: fix bsd build, and re-org linux-specific implementations

Merge remote-tracking branch 'stefanha/trivial-patches' into staging

* stefanha/trivial-patches:
  trace-events: Fix broken build caused by wrong format specifier
  test: add test-qmp-commands to make check
  qapi: remove print statements from test-qmp-commands
  test: remove qemu-ga reference
  vl.c: fix '-cpu ?' segfault

Merge remote-tracking branch 'spice/spice.v51' into staging

* spice/spice.v51:
  ui/spice-display: use uintptr_t when casting qxl physical addresses
  ui/spice-display.c: Fix compilation warnings on 32 bit hosts

test: remove qemu-ga reference

This was added by mistake a while back.

Signed-off-by: Michael Roth <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

Man page: Add -global description

There's only TODO information in qemu man page for -global option. This is a basic description of this option with simple example.

Signed-off-by: Miroslav Rezanina <[email protected]>
v4:
 - break long line

v3:
 - add use case description
 - use prop instead of property

v2:
 - Use better value in example
Patch:
--
Signed-off-by: Anthony Liguori <[email protected]>

qapi: add c_fun to escape function names

Signed-off-by: Federico Simoncelli <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

libcacard/vcard_emul_nss: add warning for old coolkey

Older coolkey versions (before the future fix of RHBZ 802435) have
a fake card reader created if no reader is detected during module
initialization. Warn libcacard users if the faulty coolkey is detected
by checking for the fake reader name "E-Gate 0 0".

Signed-off-by: Alon Levy <[email protected]>

libcacard/vcard_emul_nss: handle no readers at startup

When starting with no readers, coolkey should show no slots (with
RHBZ 806038 fixed). Fix initialization to launch the event handling
thread for each module that isn't the internal module regardless of the
number of slots detected for it at initialization time, since slot
number may start as 0 and is dynamic.

RHBZ: 802435
Signed-off-by: Alon Levy <[email protected]>

libcacard/vcard_emul_nss: don't stop thread when there are no slots

Signed-off-by: Alon Levy <[email protected]>

trace-events: Fix broken build caused by wrong format specifier

mem is an uint64_t value, so %lx was wrong.

Signed-off-by: Stefan Weil <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

test: add test-qmp-commands to make check

All the deps are here but the test was never added to the list of tests
for make check

Signed-off-by: Michael Roth <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

qapi: remove print statements from test-qmp-commands

This is necessary for nicer make check integration.

Signed-off-by: Michael Roth <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

test: remove qemu-ga reference

This was added by mistake a while back.

Signed-off-by: Michael Roth <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

vl.c: fix '-cpu ?' segfault

Fix stupid copy&paste mistake at commit
ecf40beae7dcbb057d4f115207f9d8276832a774: I moved code around but kept
"optarg" on the cpu_list() call.

Reported-by: Jiri Denemark <[email protected]>
Signed-off-by: Eduardo Habkost <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

qemu-ga: fix bsd build, and re-org linux-specific implementations

tcg-sparc: Add debug_frame support.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

tcg-hppa: Add debug_frame support.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

tcg: Allow ELF_HOST_FLAGS and ELF_OSABI overrides in gdb-jit.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

tcg: Add debug_info to JIT ELF image.

This allows us to actually supply a function name in softmmu builds;
gdb doesn't pick up the minimal symbol table otherwise.  Also add a
bit of documentation and statically generate more of the ELF image.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Use noreturn marker in helper.h.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Make use of fp_status.flush_inputs_to_zero.

This softfp feature post-dates the last major update to the Alpha
fpu translation.  We can make use of this to eliminate at least
one helper function that was performing this operation by hand.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Move memory helpers to mem_helper.c.

This completes the transition away from AREG0.  This patch must
be last because it requires CONFIG_TCG_PASS_AREG0 set too.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Move palcode support helpers to sys_helper.c.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Move integer overflow helpers to int_helper.c.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Move fpcr helpers from op_helper.c to helper.c.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Move floating-point helpers to fpu_helper.c.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Move exception helpers to helper.c.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-alpha: Move integer helpers to int_helper.c.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

alpha-linux-user: Initialize fpu to round-to-normal.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

gdbserver: Don't send a GDB syscall until the system CPU is stopped

Fix an issue where the GDB server implementation was sending GDB syscall
requests while the system CPU was still running.  Syscall requests must
be sent while the CPU is stopped otherwise replies from the GDB client
might get dropped and the GDB server might be incorrectly transitioned
into a 'RUN_STATE_PAUSED' state.

Signed-off-by: Meador Inge <[email protected]>
[PMM: trivial rebase, reinstated comma after last item in RSState enum]
Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Andreas Färber <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

Support utf8 chars in pod docs

We've at least one UTF8 char in the qemu texi doc:

 $ grep Tibor qemu-doc.texi
 by Tibor "TS" Schütz.
 $ man ./qemu.1 | grep Tibor
        by Tibor "TS" SchA~Xtz.

This patch allows utf8 in man/pod docs.

Initially it was split into two parts and sent on 2012-02-02.
Resending it again (3rd time) now in merged form.  If any
other generalizations of $(POD2MAN) are needed it can be done
in a separate patch.  Current form of $(POD2MAN) is choosen
to be able to easily change it if some implementation does
not support utf8 or resulting output has issues with local
man(1) program/macros.

First, add @documentencoding in scripts/texi2pod.pl:

Currently our texi2pod ignores @documentencoding even if it is set
properly in *.texi files.  This results in a mojibake in documents
generated from qemu.pod (which is generated from qemu-doc.texi by
texi2pod), because the rest of the tools assumes ASCII encoding.

This patch recognizes first @documentencoding in input and places
it at the beginning of output as =encoding directive.

Second, run pod2man with --utf8 option to enable utf8 in manpages:

This option makes no difference for manpages which contains only
ascii chars.  But for manpages with actual UTF8 characters (qemu
docs contains these), this change allows to see real characters
instead of mojibakes or substitutes.

Signed-off-By: Michael Tokarev <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

tcg: Use the GDB JIT debugging interface.

This allows us to generate unwind info for the dynamicly generated
code in the code_gen_buffer.  Only i386 is converted at this point.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

fix screendump

Commit 45efb16124efef51de5157afc31984b5a47700f9 optimized a bit too
much.  We can skip the vga_invalidate_display() in case no console
switch happened because we don't need a full redraw then.  We can *not*
skip vga_hw_update() though, because the screen content will be stale
then in case nobody else calls vga_hw_update().

Trigger: vga textmode with vnc display and no client connected.

Reported-by: Avi Kivity <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>
Reviewed-by: Alon Levy <[email protected]>
Tested-by: Avi Kivity <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-sparc: Add compiler attribute to some functions which don't return

helper_raise_exception does not return, nor does do_unaligned_access.

Cc: Blue Swirl <[email protected]>
Signed-off-by: Stefan Weil <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-mips: Add compiler attribute to some functions which don't return

helper_raise_exception_err does not return, nor do helper_raise_exception
and do_unaligned_access.

Cc: Aurelien Jarno <[email protected]>
Signed-off-by: Stefan Weil <[email protected]>
Reviewed-by: Andreas Färber <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

Add support for target helper functions which don't return

Most functions which handle exceptions don't return.

With a compiler attribute (added by QEMU_NORETURN),
gcc can optimize the code.

Signed-off-by: Stefan Weil <[email protected]>
Reviewed-by: Andreas Färber <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

scripts: add gdb support script

Add a script that enhances gdb to be aware of QEMU data structures.

This patch adds a single gdb command, 'qemu mtree'.  The command is
similar to the monitor's 'info mtree', except that it prints MemoryRegion
addresses, and except for working from a core dump as well as a live instance.

Signed-off-by: Avi Kivity <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

xen_disk: when using AIO flush after the operation is completed

If ioreq->postsync call bdrv_flush when the AIO operation is actually
completed.

Signed-off-by: Stefano Stabellini <[email protected]>

xen_disk: open disk with BDRV_O_NOCACHE | BDRV_O_CACHE_WB | BDRV_O_NATIVE_AIO

Signed-off-by: Stefano Stabellini <[email protected]>

ui/spice-display: use uintptr_t when casting qxl physical addresses

The current intptr_t casts are a problem when the address's highest
bit is 1, and it is cast to a intptr_t and then to uint64_t, such
as at:
     surface.mem        = (intptr_t)ssd->buf;

This causes the sign bit to be extended which causes a wrong address to
be passed on to spice, which then complains when it gets the wrong
slot_id number, since the slot_id is taken from the higher bits.

The assertion happens early - during the first primary surface creation.

This fixes running "-vga qxl -spice" with 32 bit compiled
qemu-system-i386.

Signed-off-by: Alon Levy <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

ui/spice-display.c: Fix compilation warnings on 32 bit hosts

Fix compilation failures ("cast from pointer to integer of
different size [-Werror=pointer-to-int-cast]") by using
uintptr_t instead.

Signed-off-by: Peter Maydell <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

Merge remote-tracking branch 'sstabellini/saverestore-8' into staging

* sstabellini/saverestore-8:
  xen: do not allocate RAM during INMIGRATE runstate
  xen mapcache: check if memory region has moved.
  xen: record physmap changes to xenstore
  Set runstate to INMIGRATE earlier
  Introduce "xen-save-devices-state"
  cirrus_vga: do not reset videoram

Conflicts:
qapi-schema.json

Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'stefanha/trivial-patches' into staging

* stefanha/trivial-patches:
  qemu-ga: for w32, fix leaked handle ov.hEvent in ga_channel_write()
  ioapic: fix build with DEBUG_IOAPIC
  .gitignore: add qemu-bridge-helper and option rom build products
  cleanup obsolete typedef
  monitor: Remove unused bool field 'qapi' in mon_cmd_t struct
  ds1338: Add missing break statement
  vnc: Fix packed boolean struct members
  Remove type field in ModuleEntry as it's not used

Merge remote-tracking branch 'spice/spice.v50' into staging

* spice/spice.v50:
  spice: fix broken initialization
  hw/qxl.c: Fix compilation failures on 32 bit hosts
  qxl/qxl_render.c: add trace events
  qxl: switch qxl.c to trace-events
  qxl: init_pipe_signaling: exit on failure
  monitor: fix client_migrate_info error handling
  spice: set spice uuid and name

Merge remote-tracking branch 'bonzini/scsi-next' into staging

* bonzini/scsi-next:
  scsi: add get_dev_path
  virtio-scsi: call unregister_savevm properly
  scsi: copy serial number into VPD page 0x83
  scsi-cd: check ready condition before processing several commands
  get rid of CONFIG_VIRTIO_SCSI

Merge remote-tracking branch 'qemu-kvm/memory/urgent' into staging

* qemu-kvm/memory/urgent:
  exec: fix guest memory access for Xen

xen: do not allocate RAM during INMIGRATE runstate

Signed-off-by: Anthony PERARD <[email protected]>
Signed-off-by: Stefano Stabellini <[email protected]>

xen mapcache: check if memory region has moved.

This patch changes the xen_map_cache behavior. Before trying to map a guest
addr, mapcache will look into the list of range of address that have been moved
(physmap/set_memory). There is currently one memory space like this, the vram,
"moved" from were it's allocated to were the guest will look into.

This help to have a succefull migration.

Signed-off-by: Anthony PERARD <[email protected]>
Signed-off-by: Stefano Stabellini <[email protected]>

xen: record physmap changes to xenstore

Write to xenstore any physmap changes so that the hypervisor can be
aware of them.
Read physmap changes from xenstore on boot.

Signed-off-by: Stefano Stabellini <[email protected]>

Set runstate to INMIGRATE earlier

Set runstate to RUN_STATE_INMIGRATE as soon as we can on resume.

Signed-off-by: Stefano Stabellini <[email protected]>
Acked-by: Luiz Capitulino <[email protected]>

Introduce "xen-save-devices-state"

- add an "is_ram" flag to SaveStateEntry;

- register_savevm_live sets is_ram for live_savevm devices;

- introduce a "xen-save-devices-state" QAPI command that can be used to save
the state of all devices, but not the RAM or the block devices of the
VM.

Changes in v8:

- rename save-devices-state to xen-save-devices-state.

Changes in v7:

- rename save_devices to save-devices-state.

Changes in v6:

- remove the is_ram parameter from register_savevm_live and sets is_ram
if the device is a live_savevm device;

- introduce save_devices as a QAPI command, write a better description
for it;

- fix CODING_STYLE;

- introduce a new doc to explain the save format used by save_devices.

Signed-off-by: Stefano Stabellini <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Acked-by: Luiz Capitulino <[email protected]>

exec: fix guest memory access for Xen

In cpu_physical_memory_rw, a change has been introduced and qemu_get_ram_ptr is
no longuer called with the ram addr we want to access, but only with the
section address. This patch fixes this. (All other call to qemu_get_ram_ptr are
already called with the right address.)

This patch fixes Xen guest.

Signed-off-by: Anthony PERARD <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

Merge remote-tracking branch 'origin/master' into staging

* origin/master:
  tcg: fix sparc host for AREG0 free operation

scsi: add get_dev_path

Signed-off-by: Paolo Bonzini <[email protected]>

virtio-scsi: call unregister_savevm properly

This fixes a use-after-free when migrating after hot-unplug.

Signed-off-by: Paolo Bonzini <[email protected]>

scsi: copy serial number into VPD page 0x83

Currently QEMU passes the qdev device id to the guest in an ASCII-string
designator in page 0x83.  While this is fine, it does not match what
real hardware does; usually the ASCII-string designator there hosts
another copy of the serial number (there can be other designators,
for example with a world-wide name).  Do the same for QEMU SCSI
disks.

ATAPI does not support VPD pages, so it does not matter there.

Signed-off-by: Paolo Bonzini <[email protected]>

scsi-cd: check ready condition before processing several commands

This commit is more or less obvious.  What it caused is less obvious:
SCSI CD drives failed to eject under Linux, though for example the
"change" command worked okay.  This happens because of the autoclose
option in the Linux CD-ROM driver.

The actual chain of events is quite complex and somehow involves
udev helpers; the actual command that matters is READ TOC, though
honestly it's not really clear to me how because it should always be
invoked after autoclose, not before.

Signed-off-by: Paolo Bonzini <[email protected]>

get rid of CONFIG_VIRTIO_SCSI

Signed-off-by: Paolo Bonzini <[email protected]>

Merge remote-tracking branch 'qemu-kvm/memory/urgent' into staging

* qemu-kvm/memory/urgent: (42 commits)
  memory: check for watchpoints when getting code ram_addr
  exec: fix write tlb entry misused as iotlb
  Sparc: avoid AREG0 wrappers for memory access helpers
  Sparc: avoid AREG0 for memory access helpers
  TCG: add 5 arg helpers to def-helper.h
  softmmu templates: optionally pass CPUState to memory access functions
  i386: Remove REGPARM
  sparc64: implement PCI and ISA irqs
  sparc: reset CPU state on reset
  apb: use normal PCI device header for PBM device
  w64: Fix data type of next_tb and tcg_qemu_tb_exec
  softfloat: fix for C99
  vmstate: fix varrays with uint32_t indexes
  Fix large memory chunks allocation with tcg_malloc.
  hw/pxa2xx.c: Fix handling of pxa2xx_i2c variable offset within region
  hw/pxa2xx_lcd.c: drop target_phys_addr_t usage in device state
  hw/pxa2xx_dma.c: drop target_phys_addr_t usage in device state
  ARM: Remove unnecessary subpage workarounds
  malta: Fix display for LED array
  malta: Use symbolic hardware addresses
  ...

Merge remote-tracking branch 'qmp/queue/qmp' into staging

* qmp/queue/qmp:
  qapi: Convert migrate
  Purge migration of (almost) everything to do with monitors
  Error: Introduce error_copy()
  QError: Introduce new errors for the migration command

spice: fix broken initialization

Commit 1b71f7c14fab6f00c2680d4489fbee7baf796e4f moved MODULE_INIT_QOM to
way before MODULE_INIT_MACHINE, thereby breaking assumptions made in
spice-core.c which registered both a type initializer and a machine
intializer.

This fix removes the type registration, and replaces it with calling
qemu_spice_init in vl.c after command line parsing (second pass) is
done, and after timers are armed, required by spice server.

Signed-off-by: Alon Levy <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

hw/qxl.c: Fix compilation failures on 32 bit hosts

Fix compilation failures on 32 bit hosts (cast from pointer to
integer of different size; %ld expects 'long int' not uint64_t).

Reported-by: Steve Langasek <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

qxl/qxl_render.c: add trace events

Signed-off-by: Alon Levy <[email protected]>
Reviewed-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

qxl: switch qxl.c to trace-events

dprint is still used for qxl_init_common one time prints.

also switched parts of spice-display.c over, mainly all the callbacks to
spice server.

All qxl device trace events start with the qxl device id.

Signed-off-by: Alon Levy <[email protected]>
Reviewed-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

qxl: init_pipe_signaling: exit on failure

If pipe creation fails, exit, don't log and continue. Fix indentation at
the same time.

Signed-off-by: Alon Levy <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

monitor: fix client_migrate_info error handling

Report QERR_MISSING_PARAMETER when port is missing. Otherwise
QERR_UNDEFINED_ERROR will occur.

rhbz #795652

Signed-off-by: Yonit Halperin <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

spice: set spice uuid and name

This allows a Spice client to identify a VM

Signed-off-by: Gerd Hoffmann <[email protected]>

qemu-ga: for w32, fix leaked handle ov.hEvent in ga_channel_write()

In the function ga_channel_write(), the handle ov.hEvent is created
by the call to CreateEvent(). However, the handle is not closed
prior to the function return.

This patch closes the handle before the return of the function.

Kudos to Paolo Bonzini for spotting this bug.

Signed-off-by: Jeff Cody <[email protected]>
Acked-by: Michael Roth <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

ioapic: fix build with DEBUG_IOAPIC

ioapic.c:198: error: format ‘%08x’ expects type ‘unsigned int’, but argument 3 has type ‘uint64_t’

Signed-off-by: Jason Wang <[email protected]>
Reviewed-by: Andreas Färber <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

.gitignore: add qemu-bridge-helper and option rom build products

This adds a few previously missing generated files to .gitignore: the
qemu-bridge-helper binary, and more generated versions of the
linuxboot, multiboot and kvmvapic roms from pc-bios/optionrom.

Signed-off-by: David Gibson <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>