vl: Use size_t for sizes in get_boot_devices_list()

Code mixes uint32_t, int and size_t.  Very unlikely to go wrong in
practice, but clean it up anyway.

Signed-off-by: Markus Armbruster <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

fw_cfg: Use void *, size_t instead of uint8_t *, uint32_t for blobs

Many callers pass size_t, which gets silently truncated to uint32_t.
Harmless, because all practical sizes are well below 4GiB.  Clean it
up anyway.  Size overflow now fails assertions.

Bonus: saves a whole bunch of silly casts.

Signed-off-by: Markus Armbruster <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

pc: Clean up bochs_bios_init()'s (non-)use of sizeof

Signed-off-by: Markus Armbruster <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

sun4: Fix unchecked strdup() by switching to fw_cfg_add_string()

Signed-off-by: Markus Armbruster <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

pc: Fix unchecked strdup() by switching to fw_cfg_add_string()

Signed-off-by: Markus Armbruster <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

fw_cfg: New fw_cfg_add_string()

Signed-off-by: Markus Armbruster <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

fw_cfg: Dumb down fw_cfg_add_*() not to return success / failure

No caller is checking the value, so all errors get ignored, usually
silently.  assert() instead.

Signed-off-by: Markus Armbruster <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

fw_cfg: Replace debug prints by tracepoints

Signed-off-by: Markus Armbruster <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

optimize: optimize using nonzero bits

This adds two optimizations using the non-zero bit mask.  In some cases
involving shifts or ANDs the value can become zero, and can thus be
optimized to a move of zero.  Second, useless zero-extension or an
AND with constant can be detected that would only zero bits that are
already zero.

The main advantage of this optimization is that it turns zero-extensions
into moves, thus enabling much better copy propagation (around 1% code
reduction).  Here is for example a "test $0xff0000,%ecx + je" before
optimization:

 mov_i64 tmp0,rcx
 movi_i64 tmp1,$0xff0000
 discard cc_src
 and_i64 cc_dst,tmp0,tmp1
 movi_i32 cc_op,$0x1c
 ext32u_i64 tmp0,cc_dst
 movi_i64 tmp12,$0x0
 brcond_i64 tmp0,tmp12,eq,$0x0

and after (without patch on the left, with on the right):

 movi_i64 tmp1,$0xff0000                 movi_i64 tmp1,$0xff0000
 discard cc_src                          discard cc_src
 and_i64 cc_dst,rcx,tmp1                 and_i64 cc_dst,rcx,tmp1
 movi_i32 cc_op,$0x1c                    movi_i32 cc_op,$0x1c
 ext32u_i64 tmp0,cc_dst
 movi_i64 tmp12,$0x0                     movi_i64 tmp12,$0x0
 brcond_i64 tmp0,tmp12,eq,$0x0           brcond_i64 cc_dst,tmp12,eq,$0x0

Other similar cases: "test %eax, %eax + jne" where eax is already 32-bit
(after optimization, without patch on the left, with on the right):

 discard cc_src                          discard cc_src
 mov_i64 cc_dst,rax                      mov_i64 cc_dst,rax
 movi_i32 cc_op,$0x1c                    movi_i32 cc_op,$0x1c
 ext32u_i64 tmp0,cc_dst
 movi_i64 tmp12,$0x0                     movi_i64 tmp12,$0x0
 brcond_i64 tmp0,tmp12,ne,$0x0           brcond_i64 rax,tmp12,ne,$0x0

"test $0x1, %dl + je":

 movi_i64 tmp1,$0x1                      movi_i64 tmp1,$0x1
 discard cc_src                          discard cc_src
 and_i64 cc_dst,rdx,tmp1                 and_i64 cc_dst,rdx,tmp1
 movi_i32 cc_op,$0x1a                    movi_i32 cc_op,$0x1a
 ext8u_i64 tmp0,cc_dst
 movi_i64 tmp12,$0x0                     movi_i64 tmp12,$0x0
 brcond_i64 tmp0,tmp12,eq,$0x0           brcond_i64 cc_dst,tmp12,eq,$0x0

In some cases TCG even outsmarts GCC. :)  Here the input code has
"and $0x2,%eax + movslq %eax,%rbx + test %rbx, %rbx" and the optimizer,
thanks to copy propagation, does the following:

 movi_i64 tmp12,$0x2                     movi_i64 tmp12,$0x2
 and_i64 rax,rax,tmp12                   and_i64 rax,rax,tmp12
 mov_i64 cc_dst,rax                      mov_i64 cc_dst,rax
 ext32s_i64 tmp0,rax                  -> nop
 mov_i64 rbx,tmp0                     -> mov_i64 rbx,cc_dst
 and_i64 cc_dst,rbx,rbx               -> nop

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

optimize: track nonzero bits of registers

Add a "mask" field to the tcg_temp_info struct.  A bit that is zero
in "mask" will always be zero in the corresponding temporary.
Zero bits in the mask can be produced from moves of immediates,
zero-extensions, ANDs with constants, shifts; they can then be
be propagated by logical operations, shifts, sign-extensions,
negations, deposit operations, and conditional moves.  Other
operations will just reset the mask to all-ones, i.e. unknown.

[rth: s/target_ulong/tcg_target_ulong/]

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

optimize: only write to state when clearing optimizer data

The next patch will add to the TCG optimizer a field that should be
non-zero in the default case.  Thus, replace the memset of the
temps array with a loop.  Only the state field has to be up-to-date,
because others are not used except if the state is TCG_TEMP_COPY
or TCG_TEMP_CONST.

[rth: Extracted the loop to a function.]

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

Merge branch 'ppc-for-upstream' of git://repo.or.cz/qemu/agraf

* 'ppc-for-upstream' of git://repo.or.cz/qemu/agraf:
  PPC: KVM: Add support for EPR with KVM
  openpic: export e500 epr enable into a ppc.c function
  Update Linux kernel headers
  PPC: e500: Change in-memory order of load blobs
  PPC: Provide zero SVR for -cpu e500mc and e5500
  PPC: E500: Calculate loading blob offsets properly
  openpic: set mixed mode as supported
  openpic: unify gcr mode mask updates
  openpic: move gcr write into a function

Merge branch 's390-for-upstream' of git://repo.or.cz/qemu/agraf

* 's390-for-upstream' of git://repo.or.cz/qemu/agraf:
  s390: Add a hypercall registration interface.
  target-s390x: Unregister reset callback on finalization
  s390x: fix indentation
  s390: Add CPU reset handler
  s390x: Remove inline function ebcdic_put and related data from cpu.h
  S390: Enable -cpu help and QMP query-cpu-definitions
  s390: Move IPL code into a separate device
  s390: new contributions GPLv2 or later

Merge branch 'axp-next' of git://github.com/rth7680/qemu

* 'axp-next' of git://github.com/rth7680/qemu:
  alpha-linux-user: Correct select
  alpha-linux-user: Translate fcntl l_type

s390: Add a hypercall registration interface.

Allow virtio machines to register for different diag500 function
codes and convert s390-virtio to use it.

Signed-off-by: Cornelia Huck <[email protected]>
Signed-off-by: Alexander Graf <[email protected]>

target-s390x: Unregister reset callback on finalization

Since commit "s390: Add CPU reset handler" the CPU's instance_init
registers a reset callback. Unregister that on instance_finalize.

Signed-off-by: Andreas Färber <[email protected]>
Signed-off-by: Alexander Graf <[email protected]>

s390x: fix indentation

In one of the last commits we accidently got 3-space indentation into
the tree. Fix it up so it's 4 spaces wide.

Reported-by: Andreas Faerber <[email protected]>
Signed-off-by: Alexander Graf <[email protected]>

s390: Add CPU reset handler

Add a CPU reset handler to have all CPUs in a PoP compliant
state.

Signed-off-by: Jens Freimann <[email protected]>
[agraf: move hw/hw.h into existing ifdef]
Signed-off-by: Alexander Graf <[email protected]>

s390x: Remove inline function ebcdic_put and related data from cpu.h

The function is only used in misc_helper.c, so move it to that file.

This reduces the size of debug executables (compiled without optimization)
because they get unused code and data for each compilation which includes
cpu.h.

Executables with optimization don't change their size.

ebcdic2ascii is currently unused and could be removed (not done here).

The array ascii2ebcdic must be accessed with an unsigned index, therefore
(int)ascii[i] was replaced by (uint8_t)ascii[i]. The old code would have
failed for a signed char less than 0. The current code only converts
"QEMU" and spaces to EBCDIC, so there is no problem today.

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

S390: Enable -cpu help and QMP query-cpu-definitions

This enables qemu -cpu help to return a list of supported CPU models
on s390 and also to query for cpu definitions in the monitor.
Initially only cpu model = host is returned. This needs to be reworked
into a full-fledged CPU model handling later on.
This change is needed to allow libvirt exploiters (like OpenStack)
to specify a CPU model.

Signed-off-by: Viktor Mihajlovski <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
Reviewed-by: Christian Borntraeger <[email protected]>
[agraf: fix s390x-linux-user, adjust header locations]
Signed-off-by: Alexander Graf <[email protected]>

s390: Move IPL code into a separate device

Lets move the code to setup IPL for external kernel
or via the zipl rom into a separate file. This allows to

- define a reboot handler, setting up the PSW appropriately
- enhance the boot code to IPL disks that contain a bootmap that
  was created with zipl under LPAR or z/VM (future patch)
- reuse that code for several machines (e.g. virtio-ccw and virtio-s390)
- allow different machines to provide different defaults

Signed-off-by: Christian Borntraeger <[email protected]>
Signed-off-by: Jens Freimann <[email protected]>
[agraf: symbolify initial psw, adjust header file location, fix for QOM]
Signed-off-by: Alexander Graf <[email protected]>

s390: new contributions GPLv2 or later

IBMs s390 contributions were meant to to be gplv2 or later (since
we were contributing to qemu). Several of the s390 specific files
link to gpl code anyway, so lets clarify the licence statement for
new contributions for those files that we have touched multiple
times or will likely touch again.

This patch does not touch files that mostly deal with tcg.

Signed-off-by: Christian Borntraeger <[email protected]>
Signed-off-by: Alexander Graf <[email protected]>

PPC: KVM: Add support for EPR with KVM

This patch links KVM EPR support to the existing TCG support we have now.

Signed-off-by: Alexander Graf <[email protected]>

openpic: export e500 epr enable into a ppc.c function

Enabling and disabling the EPR capability (mpic_proxy) is a system
wide operation. As such, it belongs into the ppc.c file, since that's
where PPC specific machine wide logic happens.

Signed-off-by: Alexander Graf <[email protected]>

Update Linux kernel headers

Based on kvm.git a843fac (next) plus dfdebc24 (master).

Signed-off-by: Alexander Graf <[email protected]>

PPC: e500: Change in-memory order of load blobs

Today, we load

  <kernel> <initrd> <dtb>

into memory in that order. However, Linux has a bug where it can only
handle the dtb if it's within the first 64MB of where <kernel> starts.

So instead, let's change the order to

  <kernel> <dtb> <initrd>

making Linux happy.

Signed-off-by: Alexander Graf <[email protected]>

PPC: Provide zero SVR for -cpu e500mc and e5500

Even though our -cpu types for e500mc and e5500 are no real CPUs that
actually have version registers, a guest might still want to access
said version register and that has to succeed for a guest to be happy.

So let's expose a zero SVR value on E500_SVR SPR reads.

Signed-off-by: Alexander Graf <[email protected]>

PPC: E500: Calculate loading blob offsets properly

We have 3 blobs we need to load when booting the system:

  - kernel
  - initrd
  - dtb

We place them in physical memory in that order. At least we should.
This patch fixes the location calculation up to take any module into
account, fixing the dtb offset along the way.

Reported-by: Stuart Yoder <[email protected]>
Signed-off-by: Alexander Graf <[email protected]>

openpic: set mixed mode as supported

The Raven MPIC implementation supports the "Mixed" mode to work with
an i8259. While we don't implement mixed mode, we should mark it as
a supported mode in the mode bitmap.

Signed-off-by: Alexander Graf <[email protected]>

openpic: unify gcr mode mask updates

The mode mask already masks out bits we don't care about, so the
actual handling code can stay intact regardless.

Signed-off-by: Alexander Graf <[email protected]>

openpic: move gcr write into a function

The GCR register contains too much functionality to be covered inside
of the register switch statement. Move it out into a separate function.

Signed-off-by: Alexander Graf <[email protected]>

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

# By Wenchao Xia
# Via Luiz Capitulino
* luiz/queue/qmp:
  HMP: add sub command table to info
  HMP: move define of mon_cmds
  HMP: add infrastructure for sub command
  HMP: delete info handler
  HMP: add QDict to info callback handler

virtio-pci: fix irqfd cleanup argument order

Order of arguments of kvm_virtio_pci_irqfd_release
got mixed up in all calls.
As a result users see assertions during cleanup.

Reported-by: Laszlo Ersek <[email protected]>
Reviewed-by: Laszlo Ersek <[email protected]>
Tested-by: Laszlo Ersek <[email protected]>
Reviewed-by: Stefan Hajnoczi <[email protected]>
Tested-by: Wanlong Gao <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

qom: Extend documentation on QOM method concepts

Add a documentation section "Methods" and discuss among others how to
handle overriding virtual methods.

Clarify DeviceClass::realize documentation and refer to the above.

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

Makefile: drop recursive libcacard clean

Commit eb8eb53e5846a957cf333f2e1ec8cb6e0c04 ("libcacard: rewrite
Makefile in non-recursive style") refactored libcacard/Makefile so it
can be included by the top-level Makefile.

The top-level clean target still loops over subdirectories, including
libcacard/, to invoke recursive clean.  Remove libcacard from the
recursive clean since its files are already included at the top level.

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

HMP: add sub command table to info

  Now info command takes a table of sub info commands,
and changed do_info() to do_info_help() to do help funtion
only.
 Note that now "info <unknown-topic>" returns error instead
of list of info topics.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

HMP: move define of mon_cmds

  Because mon_cmds may use info_cmds, so adjust the declare sequence
of them.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

HMP: add infrastructure for sub command

  This patch make parsing of hmp command aware of that it may
have sub command. Also discard simple encapsulation function
monitor_find_command(). For case "@command ", space after
@command is filtered out.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

HMP: delete info handler

  Now cmd and info handler have same format, so delete info handler.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

HMP: add QDict to info callback handler

  This patch change all info call back function to take
additional QDict * parameter, which allow those command
take parameter. Now it is set to NULL at default case.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

bswap: improve gluing

OpenBSD system compiler (gcc 4.2.1) has problems with concatenation
of macro arguments in macro functions:
  CC    aes.o
In file included from /src/qemu/include/qemu-common.h:126,
                 from /src/qemu/aes.c:30:
/src/qemu/include/qemu/bswap.h: In function 'leul_to_cpu':
/src/qemu/include/qemu/bswap.h:461: warning: implicit declaration of function 'bswapHOST_LONG_BITS'
/src/qemu/include/qemu/bswap.h:461: warning: nested extern declaration of 'bswapHOST_LONG_BITS'

Function leul_to_cpu() is only used in kvm-all.c, so the warnings
are not fatal on OpenBSD without -Werror.

Fix by applying glue(). Also add do {} while(0) wrapping and fix
semicolon use while at it.

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

Merge remote-tracking branch 'kraxel/chardev.1' into staging

# By Gerd Hoffmann
# Via Gerd Hoffmann
* kraxel/chardev.1:
  chardev: add pty chardev support to chardev-add (qmp)
  chardev: add socket chardev support to chardev-add (qmp)
  chardev: add parallel chardev support to chardev-add (qmp)
  chardev: add serial chardev support to chardev-add (qmp)
  chardev: add file chardev support to chardev-add (qmp)
  chardev: add hmp hotplug commands
  chardev: add qmp hotplug commands, with null chardev support
  chardev: reduce chardev ifdef mess a bit
  chardev: fix QemuOpts lifecycle
  chardev: add error reporting for qemu_chr_new_from_opts

build: fix Win32 clean build

The version.o file did not appear explicitly as a dependency, and
this caused clean builds to fail.  Force its build by making the
Makefile depend on version.o.

(We cannot add it to libqemuutil.a, because it doesn't export any
symbol and thus would not be pulled by the linker).

Cc: Blue Swirl <[email protected]>
Cc: Stefan Weil <[email protected]>
Tested-by: Stefan Weil <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

tmp105: Add temperature QOM property

This obsoletes tmp105_set() and allows for better error handling.

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

tmp105: QOM'ify

Introduce TYPE_ constant and cast macro.
Move the state struct to the new header to allow for future embedding.

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

tests: Add tmp105 qtest test case

Exercise all four commands of the TMP105, testing for an issue in the
I2C TX path.

The test case uses the N800's OMAP I2C and is the first for ARM.

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

tmp105: Fix I2C protocol bug

An early length postincrement in the TMP105's I2C TX path led to
transfers of more than one byte to place the second byte in the third
byte's place within the buffer and the third byte to get discarded.

Fix this by explictly incrementing the length after the checks but
before the callback is called, which again checks the length.

Adjust the Coding Style while at it.

Signed-off-by: Alex Horn <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>
Reviewed-by: Anthony Liguori <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

tmp105: Split out I2C message constants from header

Allows value sharing with qtest.

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

libqtest: Prepare I2C libqos

This adds a simple I2C API and a driver implementation for omap_i2c.

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

audio: Replace non-portable asprintf in debug code by g_strdup_printf

sw->name already uses the correct g_free to free the allocated memory.

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

usb-storage: Drop useless null test in usb_msd_handle_data()

scsi_req_new() never returns null, and scsi_req_enqueue() dereferences
the pointer, so checking for null is useless.

Spotted by Coverity.

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

ui: Drop useless null tests in parse_keyboard_layout()

Spotted by Coverity.

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

pseries: Replace non-portable asprintf by g_strdup_printf

g_strdup_printf already handles OOM errors, so some error handling in
QEMU code can be removed.

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

Add new DEFAULT_MACHINE_OPTIONS to q35 and ppc405

Without this default q35/ppc405 based machines would no longer boot
after commit e4ada29e909787f629626660b1561f6a680187d3

Signed-off-by: Knut Omang <[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

alpha-linux-user: Correct select

Alpha, like s390x, passes all select arguments in registers.

Signed-off-by: Laurent Vivier <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

alpha-linux-user: Translate fcntl l_type

The values of F_RDLCK, F_WRLCK, F_UNLCK, F_EXLCK, F_SHLCK
differ between alpha and other linux architectures.

This patch allows to run "dpkg" (database lock).

Signed-off-by: Laurent Vivier <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

chardev: add pty chardev support to chardev-add (qmp)

The ptsname is returned directly, so there is no need to
use query-chardev to figure the pty device path.

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

chardev: add socket chardev support to chardev-add (qmp)

qemu_chr_open_socket is split into two functions.  All initialization
after creating the socket file handler is split away into the new
qemu_chr_open_socket_fd function.

chr->filename doesn't get filled from QemuOpts any more.  Qemu gathers
the information using getsockname and getnameinfo instead.  This way it
will also work correctly for file handles passed via file descriptor
passing.

Finally qmp_chardev_open_socket() is the actual qmp hotplug
implementation which basically just calls socket_listen or
socket_connect and the new qemu_chr_open_socket_fd function.

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

chardev: add parallel chardev support to chardev-add (qmp)

Also alias the old parport name to parallel for -chardev.

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

chardev: add serial chardev support to chardev-add (qmp)

Similar to file, except that no separate in/out files are supported
because it's pointless for direct device access.  Also the special
tty ioctl hooks (pass through linespeed settings etc) are activated
on Unix.

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

chardev: add file chardev support to chardev-add (qmp)

Add support for file chardevs.  Output file is mandatory,
input file is optional.

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

chardev: add hmp hotplug commands

Add chardev-add and chardev-remove commands to the human monitor.
chardev-add accepts the same syntax as -chardev, chardev-remove
expects a chardev id.

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

chardev: add qmp hotplug commands, with null chardev support

Add chardev-add and chardev-remove qmp commands.  Hotplugging
a null chardev is supported for now, more will be added later.

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

chardev: reduce chardev ifdef mess a bit

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

chardev: fix QemuOpts lifecycle

qemu_chr_new_from_opts handles QemuOpts release now, so callers don't
have to worry.  It will either be saved in CharDriverState, then
released in qemu_chr_delete, or in the error case released instantly.

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

chardev: add error reporting for qemu_chr_new_from_opts

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

qdev: Prepare "realized" property

Introduce the QOM realizefn suggested by Anthony.
Detailed documentation is supplied in the qdev header.

For now this implements a default DeviceClass::realize callback that
just wraps DeviceClass::init, which it deprecates.
Once all devices have been converted to DeviceClass::realize,
DeviceClass::init is to be removed.

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

qdev: Fold state enum into bool realized

Whether the device was initialized or not is QOM-level information and
currently unused. Drop it from device. This leaves the boolean state of
whether or not DeviceClass::init was called or not, a.k.a. "realized".

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

pseries: set no default boot order

This patch removes the default boot order for pseries machine. This allows
the machine to handle a NULL boot order in case no -boot option is provided.
Thus it helps SLOF firmware to verify if boot order is specified in command
line or not. If no boot order is provided SLOF tries to boot from the
device set in the nvram.

Reviewed-by: Anthony Liguori <[email protected]>
Acked-by: Alexander Graf <[email protected]>
Signed-off-by: Avik Sil <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

Make default boot order machine specific

This patch makes default boot order machine specific instead of
set globally. The default boot order can be set per machine in
QEMUMachine boot_order. This also allows a machine to receive a
NULL boot order when -boot isn't used and take an appropriate action
accordingly. This helps machine boots from the devices as set in
guest's non-volatile memory location in case no boot order is
provided by the user.

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

acl: Free memory allocated with g_malloc() with g_free()

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

acl: Fix acl_remove not to mess up the ACL

It leaks memory and fails to adjust qemu_acl member nentries.  Future
acl_add become confused: can misreport the position, and can silently
fail to add.

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

sdl: Fix heap smash in sdl_zoom_rgb{16,32} for int > 32 bits

Careless use of malloc(): allocate Uint32[N], assign to int *, use
int[N].

Fix by converting to g_new().

Functions can't fail anymore, so make them return void.  Caller
ignored the value anyway.

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

kvm: add stub for kvm_irqchip_update_msi_route

ppc64 build needs this stub to build with virtio enabled.

Signed-off-by: Michael S. Tsirkin <[email protected]>
Tested-by: Andreas Färber <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'afaerber/memory-ioport' into staging

* afaerber/memory-ioport:
  acpi_piix4: Do not use old_portio-style callbacks
  xen_platform: Do not use old_portio-style callbacks
  hw/dma.c: Fix conversion of ioport_register* to MemoryRegion

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

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

* stefanha/trivial-patches:
  configure: try pkg-config for curses
  qom: Make object_resolve_path_component() path argument const
  Add libcacard/trace/generated-tracers.c to .gitignore

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

Merge remote-tracking branch 'stefanha/block' into staging

* stefanha/block:
  block: Fix how mirror_run() frees its buffer
  win32-aio: Fix how win32_aio_process_completion() frees buffer
  scsi-disk: qemu_vfree(NULL) is fine, simplify
  w32: Make qemu_vfree() accept NULL like the POSIX implementation
  sheepdog: clean up sd_aio_setup()
  sheepdog: multiplex the rw FD to flush cache
  block: clear dirty bitmap when discarding
  ide: issue discard asynchronously but serialize the pieces
  ide: fix TRIM with empty range entry
  block: make discard asynchronous
  raw: support discard on block devices
  raw-posix: remember whether discard failed
  raw-posix: support discard on more filesystems
  block: fix initialization in bdrv_io_limits_enable()
  qcow2: Fix segfault on zero-length write

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

Merge remote-tracking branch 'afaerber/qom-cpu' into staging

* afaerber/qom-cpu:
  target-i386: Use switch in check_hw_breakpoints()
  target-i386: Avoid goto in hw_breakpoint_insert()
  target-i386: Introduce hw_{local,global}_breakpoint_enabled()
  target-i386: Define DR7 bit field constants
  target-i386: Move kvm_check_features_against_host() check to realize time
  target-i386: cpu_x86_register() consolidate freeing resources
  target-i386: Move setting defaults out of cpu_x86_parse_featurestr()
  target-i386: check/enforce: Check all feature words
  target-i386/cpu.c: Add feature name array for ext4_features
  target-i386: kvm_check_features_against_host(): Use feature_word_info
  target-i386/cpu: Introduce FeatureWord typedefs
  target-i386: Disable kvm_mmu by default
  kvm: Add fake KVM constants to avoid #ifdefs on KVM-specific code
  exec: Return CPUState from qemu_get_cpu()
  xen: Simplify halting of first CPU
  kvm: Pass CPUState to kvm_init_vcpu()
  cpu: Move cpu_index field to CPUState
  cpu: Move numa_node field to CPUState
  target-mips: Clean up mips_cpu_map_tc() documentation
  cpu: Move nr_{cores,threads} fields to CPUState

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

Merge remote-tracking branch 'afaerber-or/prep-up' into staging

* afaerber-or/prep-up:
  pc87312: Avoid define conflict on mingw32
  pc87312: Replace register_ioport_*() with MemoryRegion

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

Merge remote-tracking branch 'mdroth/qga-pull-1-14-2013-2' into staging

* mdroth/qga-pull-1-14-2013-2:
  qga: add missing commas in json docs

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

Merge remote-tracking branch 'mdroth/qga-pull-1-14-2013' into staging

* mdroth/qga-pull-1-14-2013:
  qemu-ga: Handle errors uniformely in ga_channel_open()
  qemu-ga: Plug fd leak on ga_channel_open() error paths
  qemu-ga: Plug fd leak on ga_channel_listen_accept() error path
  qemu-ga: Plug file descriptor leak on ga_open_pidfile() error path
  qemu-ga: Drop pointless lseek() from ga_open_pidfile()
  qemu-ga: Document intentional fall through in channel_event_cb()
  qemu-ga: add ga_open_logfile()
  qemu-ga: ga_open_pidfile(): use qemu_open()

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

Merge remote-tracking branch 'sstabellini/xen-2013-01-14' into staging

* sstabellini/xen-2013-01-14:
  xen_disk: implement BLKIF_OP_FLUSH_DISKCACHE, remove BLKIF_OP_WRITE_BARRIER
  xen_disk: add persistent grant support to xen_disk backend
  xen_disk: fix memory leak

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

acpi_piix4: Do not use old_portio-style callbacks

Signed-off-by: Hervé Poussineau <[email protected]>
[AF: Used HWADDR_PRIx for hwaddr PIIX4_DPRINTF()]
Signed-off-by: Andreas Färber <[email protected]>

xen_platform: Do not use old_portio-style callbacks

Signed-off-by: Hervé Poussineau <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

hw/dma.c: Fix conversion of ioport_register* to MemoryRegion

The commit 582299336879504353e60c7937fbc70fea93f3da introduced a 1-shift for
some offset in DMA emulation.

Before the previous commit, which converted ioport_register_* to
MemoryRegion, the DMA controller registered 8 ioports with the following
formula:
base + ((8 + i) << d->shift) where 0 <= i < 8
When an IO occured within a Memory Region, DMA callback receives an
offset relative to the start address. Here the start address is:
base + (8 << d->shift).
The offset should be: (i << d->shift). After the shift is reverted, the
offsets are 0..7 not 1..8.

Fixes LP#1089996.

Reported-by: Andreas Gustafsson <[email protected]>
Signed-off-by: Julien Grall <[email protected]>
Tested-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

configure: try pkg-config for curses

Static linkikng against ncurses may require explicit -ltinfo.
In case -lcurses and -lncurses both didn't work give pkg-config a
chance.

Fixes #1094786 for me.

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

block: Fix how mirror_run() frees its buffer

It allocates with qemu_blockalign(), therefore it must free with
qemu_vfree(), not g_free().

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

win32-aio: Fix how win32_aio_process_completion() frees buffer

win32_aio_submit() allocates it with qemu_blockalign(), therefore it
must be freed with qemu_vfree(), not g_free().

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Kevin Wolf <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

scsi-disk: qemu_vfree(NULL) is fine, simplify

Signed-off-by: Markus Armbruster <[email protected]>
Acked-by: Paolo Bonzini <[email protected]>
Reviewed-by: Kevin Wolf <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

w32: Make qemu_vfree() accept NULL like the POSIX implementation

On POSIX, qemu_vfree() accepts NULL, because it's merely wrapper
around free().  As far as I can tell, the Windows implementation
doesn't.  Breeds bugs that bite only under Windows.

Make the Windows implementation behave like the POSIX implementation.

Signed-off-by: Markus Armbruster <[email protected]>
Reviewed-by: Kevin Wolf <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

sheepdog: clean up sd_aio_setup()

The last two parameters of sd_aio_setup() are never used, so remove them.

Cc: MORITA Kazutaka <[email protected]>
Cc: Kevin Wolf <[email protected]>
Cc: Stefan Hajnoczi <[email protected]>
Signed-off-by: Liu Yuan <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

sheepdog: multiplex the rw FD to flush cache

This will reduce sockfds connected to the sheep server to one, which simply the
future hacks.

Cc: MORITA Kazutaka <[email protected]>
Cc: Kevin Wolf <[email protected]>
Cc: Stefan Hajnoczi <[email protected]>
Signed-off-by: Liu Yuan <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

qom: Make object_resolve_path_component() path argument const

A usage with a hardcoded partial path such as

  object_resolve_path_component(obj, "foo")

is totally valid but currently leads to a compilation error. Fix this.

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

Add libcacard/trace/generated-tracers.c to .gitignore

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

block: clear dirty bitmap when discarding

Note that resetting bits in the dirty bitmap is done _before_ actually
processing the request.  Writes, instead, set bits after the request
is completed.

This way, when there are concurrent write and discard requests, the
outcome will always be that the blocks are marked dirty.  This scenario
should never happen, but it is safer to do it this way.

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

ide: issue discard asynchronously but serialize the pieces

Now that discard can take a long time, make it asynchronous.
Each LBA range entry is processed separately because discard
can be an expensive operation.

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

ide: fix TRIM with empty range entry

ATA-ACS-3 says "If the two byte range length is zero, then the LBA
Range Entry shall be discarded as padding."  iovecs are used as if
they are linearized, so it is incorrect to discard the rest of
this iovec.

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

block: make discard asynchronous

This is easy with the thread pool, because we can use s->is_xfs and
s->has_discard from the worker function.

QEMU has a widespread assumption that each I/O operation writes less
than 2^32 bytes.  This patch doesn't fix it throughout of course,
but it starts correcting struct RawPosixAIOData so that there is
no regression with respect to the synchronous discard implementation.

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

raw: support discard on block devices

Block devices use a ioctl instead of fallocate, so add a separate
implementation.

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

raw-posix: remember whether discard failed

Avoid sending system calls repeatedly if they shall fail.  This
does not apply to XFS: if the filesystem-specific ioctl fails,
something weird is happening.

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