target-i386: add SSE4a instruction support

This adds support for the AMD Phenom/Barcelona's SSE4a instructions.
Those include insertq and extrq, which are doing shift and mask on
XMM registers, in two versions (immediate shift/length values and
stored in another XMM register).
Additionally it implements movntss, movntsd, which are scalar
non-temporal stores (avoiding cache trashing). These are implemented
as normal stores, though.
SSE4a is guarded by the SSE4A CPUID bit (Fn8000_0001:ECX[6]).

Signed-off-by: Andre Przywara <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

target-i386: add lock mov cr0 = cr8

AMD CPUs featuring a shortcut to access CR8 even from 32-bit mode.
If you use the LOCK prefix with "mov CR0", it accesses CR8 instead.
This behavior is guarded by the CR8_LEGACY CPUID bit
(Fn8000_0001:ECX[1]).

Signed-off-by: Andre Przywara <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

tcg/i386: add support for ext{8,16}u_i32 TCG ops

Signed-off-by: Aurelien Jarno <[email protected]>

tcg/x86_64: add support for ext{8,16,32}u_i{32,64} TCG ops

Signed-off-by: Aurelien Jarno <[email protected]>

tcg: add ext{8,16,32}u_i{32,64} TCG ops

Currently zero extensions ops are implemented by a and op with a
constant. This is then catched in some backend, and replaced by
a zero extension instruction. While this works well on RISC
machines, this adds a useless register move on non-RISC machines.

Example on x86:
  ext16u_i32 r1, r2
is translated into
  mov    %eax,%ebx
  movzwl %bx, %ebx
while the optimized version should be:
  movzwl %ax, %ebx

This patch adds ext{8,16,32}u_i{32,64} TCG ops that can be
implemented in the backends to avoid emitting useless register
moves.

Signed-off-by: Aurelien Jarno <[email protected]>

linux-user: fix "#if 0"'d printf()

Make an "#if 0"'d printf() in load_elf_binary(), probably left to aid in
debugging, reflect what the actual code does. The current printf() will
only confuse those who "#if 1" it (it certainly confused me enough to
write this trivial patch).

Signed-off-by: Paul Bolle <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Check availability of uuid header / library

If available, the Universally Unique Identifier library
is used by the vdi block driver.

Other parts of QEMU (vl.c) could also use it.

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

hw/omap_dma: add matching {} in if 0

MULTI_REQ is never defined, so it doesn't matter much, but since
we have an if statement there, let's add {} to clarify what it
should do if it's uncommented, and indent the code properly.

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

oss/alsa: Do not invoke UB described in 7.15.1.1 (this time for ADC)

Signed-off-by: malc <[email protected]>

libuser is a generated directory

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

Use GNU ld to link roms on OpenSolaris

OpenSolaris ld seems to miss the equivalent of GNU ld's "-Ttext 0".

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

Fix warning about undefined madvise() on OpenSolaris

OpenSolaris headers can't export madvise() with a sane set of #defines.
For background, see MySQL bug #7156 (http://bugs.mysql.com/bug.php?id=7156)
for discussion about Solaris header problems.

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

linux-user: fix up oversealous nitpicking

Looks like linux-user code was correct, just unreadable: what it wanted
to do with "-=" was really assign a negative number, not decrement.  Fix
up accordingly.

Reported-by: Laurent Desnogues <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

tests: missing ; in if 0

Fix missing ; in commented out code

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

alsa: Change default buffer/period size

Increase buffer size but do not rely on ALSA picking up default period
size.

Signed-off-by: malc <[email protected]>

oss/alsa: Do not invoke UB described in 7.15.1.1

Additional argument (whether to try poll mode) is only passed with
VOICE_ENABLE command.

Thanks to Markus Armbruster for noticing the potential breakage.

Revert "Get rid of _t suffix"

In the very least, a change like this requires discussion on the list.

The naming convention is goofy and it causes a massive merge problem.  Something
like this _must_ be presented on the list first so people can provide input
and cope with it.

This reverts commit 99a0949b720a0936da2052cb9a46db04ffc6db29.

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

Get rid of _t suffix

Some not so obvious bits, slirp and Xen were left alone for the time
being.

Signed-off-by: malc <[email protected]>

Include microblaze binaries in tarbin.

Signed-off-by: Edgar E. Iglesias <[email protected]>

target-mips: make sure constants are in the second argument

Signed-off-by: Aurelien Jarno <[email protected]>

x86: use globals for CPU registers

Use globals for the 8 or 16 CPU registers on i386 and x86_64.

Signed-off-by: Aurelien Jarno <[email protected]>

mips: Fix spelling in comment

inofficial -> unofficial

Thanks to Blue Swirl.

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

escc: fix another coding style nit

Fix another place with =- to be "= -".
to avoid confusion with old-style "-="
(which we also have, and needs to be fixed).

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

slirp: fix unmatched bracket in if 0

Fix unmatched bracket in commented out code

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

net: fix coding style nit

Put space between = and - assigning a negative number
to avoid confusion with old-style "-="
(which we also have, and need to be fixed).

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

linux-user: fix coding style nit

Put space between = and & when taking a pointer,
to avoid confusion with old-style "&=".

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-mips: unmatched brackets in if 0

Fix unmatched braket in commented out code

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

linux-user: fix old style decrement usage

Modern compilers do not parse "=-" as decrement:
you must use "-=" for that.

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

twl92230: fix old style increment/decrement usage

Modern compilers do not parse "=-" as decrement:
you must use "-=" for that. Same for "=+"/"+=".

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

omap_dma: fix unbalanced { in commented out code

Fix unbalanced {} in commented out code.

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

escc: fix coding style nit

Put space between = and - assigning a negative number
to avoid confusion with old-style "-="
(which we also have, and needs to be fixed).

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

gdbstub: fix coding style nit

Put space between = and * when dereferencing a pointer,
to avoid confusion with old-style "*="

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

elfload: fix coding style nit

Put space between = and * when dereferencing a pointer,
to avoid confusion with old-style "*="

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

vvfat: fix coding style nit

Put space between = and & when taking a pointer,
to avoid confusion with old-style "&=".

Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

sdlaudio: make it suck less

Signed-off-by: malc <[email protected]>

Revert part of 6692b043198d58a12317009edb98654c6839f043

Committed by accident.

Signed-off-by: Aurelien Jarno <[email protected]>

Fix build with profiler enabled

Broken by 4a1418e07bdcfaa3177739e04707ecaec75d89e1

Signed-off-by: Aurelien Jarno <[email protected]>

TCG: fix DEF2 macro

Signed-off-by: Aurelien Jarno <[email protected]>

target-ppc: log instructions start in TCG code

Signed-off-by: Aurelien Jarno <[email protected]>

target-mips: log instructions start in TCG code

Signed-off-by: Aurelien Jarno <[email protected]>

Win32: avoid a warning

GetLastError() returns a DWORD.

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

BSD user: suppress a warning

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

BSD user: implement GUEST_BASE

Based on 379f6698d73f476de38682b3ff96ecb226728c43.

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

tcg/i386: generates dec/inc instead of sub/add when possible

We must take care that dec/inc do not compute CF, which is needed by
add2/sub2.

Signed-off-by: Aurelien Jarno <[email protected]>

tcg/i386: optimize and $0xff(ff), reg

Signed-off-by: Aurelien Jarno <[email protected]>

Compile some user files only once for all targets

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

tcg/x86_64: generated dec/inc instead of sub/add when possible

Signed-off-by: Aurelien Jarno <[email protected]>

Don't compile roms if not building system targets

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

Fix user emulator breakage

Fix breakage in the following conditions:
- use in-tree building
- build user targets after system targets

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

tcg/ppc: always use tcg_out_call

Signed-off-by: malc <[email protected]>

vl: Add failure check for SetEvent

Signed-off-by: malc <[email protected]>

vl: Do not use perror after failed Win32 API calls

Signed-off-by: malc <[email protected]>

posix-aio-compat: avoid signal race when spawning a thread

Signed-off-by: malc <[email protected]>

target-i386: kill a tmp register

Signed-off-by: Aurelien Jarno <[email protected]>

target-i386: use subfi instead of sub with a non-freed constant

Signed-off-by: Aurelien Jarno <[email protected]>

eepro100: Fix format strings in debug messages

size_t arguments need %zu instead of %d.

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

eepro100: Replace sprintf by snprintf

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

eepro100: Remove unused code

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

eepro100: Enhanced logging and comments

* Use TRACE macro to allow different logging flags.
* Add new debugging messages and clean existing ones.

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

fix -daemonize with kvm

Otherwise fork might not inherit state initialized by kvm_init().

Signed-off-by: Marcelo Tosatti <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Set SDL icon_title when using `-name XXX` for proper gnome-panel window list integration

qemu: improve sdl title information

Include the `-name XXX` commandline param in the gnome-panel window list
via icon_title.

https://bugs.edge.launchpad.net/ubuntu/+source/qemu-kvm/+bug/423076

Signed-off-by: Dominic Evans <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

qemu-option: rename bool -> boolean

We need this to allow the use of <stdbool.h>

Signed-off-by: Juan Quintela <[email protected]>
Acked-by: Gerd Hoffmann <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

ARM back-end: Use sxt[bh] instructions for ext{8, 6}s

This patch uses sxtb for ext8s_i32 and sxth for ext16s_i32 in ARM back-end.

Signed-off-by: Laurent Desnogues <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Fix and improve qint_from_int64_test

Use a long long integer constant to fix a compilation error (integer
constant is too large for 'long' type).

Use a better value for testing, as -1 makes the test pass even if
qi->value is of type uint32_t, float or double. This was suggested by
Reimar Döffinger <[email protected]>.

Also, make the test fail when qi->value is of type double or float by
casting qi->value to int64_t, to avoid value being promoted to the type
of qi->value.

Signed-off-by: Pierre Riteau <[email protected]>
Acked-by: Luiz Capitulino <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

monitor: Fix do_wav_capture() argument type

Currently do_wav_capture() path's argument type is 's' (string),
but it should be 'F' (filename), this way 'wavcapture' gets
command completion.

Signed-off-by: Luiz Capitulino <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

monitor: Fix do_commit() argument type

Currently do_commit() argument type is 's' (string), but it
should be 'B' (block), this way 'commit' gets command completion.

Signed-off-by: Luiz Capitulino <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Don't redefine NULL, please

Compiled (on linux) all the targets, and it compiled as expected.
What platform needs this redefinition?

Later, Juan.

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Fix coding style issue

Replace:

  if (-1 == foo())

with:

  if (foo() == -1)

While this coding style is not in direct contravention of our currently
ratified CODING_STYLE treaty, it could be argued that the Article 3 of
the European Convention on Human Rights (prohibiting torture and "inhuman
or degrading treatment") reads on the matter.

[This commit message was brought to you without humour, as is evidenced
by the absence of any emoticons]

Signed-off-by: Mark McLoughlin <[email protected]>
Cc: Avi Kivity <[email protected]>
Cc: Gerd Hoffmann <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

Remove PARAMS() macro

Only two disassemblers (alpha and sh4) were still using it. Just remove its
use there, and its aparations in dis-asm.h

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

__thread should be before real type

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

gcc wants 1st static and then const

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

Bring two last users of K&R definitions to ANSI c89

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

static and inline should came before the type of the functions

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

Use proper typedef syntax

Why this ever compiled is a mistery to me.

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

Revert "This files are compiled in libqemu.a now"

This reverts commit fe6549dfd76c278dbcd788b3c15c5e6e5ed32190.

tcg-runtime and host-utils are needed on 32 bit host and they are not part
of libqemu.a.

Thanks to Stefan Weil for reporting.

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

fix use after free

We are using the vs structure when it was just freed. Classic use after free,
fix it.

Signed-off-by: Glauber Costa <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

unbreak usb pass-through on linux.

Changes:
  * Re-add the 'dev->fd = fd;' line which the qdev patches dropped
    by mistake.
  * call qdev_init() so the newly created usb device is plugged into
    a usb port and thus actually visible to the guest.

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

clean: remove ide/*.o files on clean

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

ARM host: fix generated blocks linking

This patch fixes the linking of generated blocks on an ARM host.
No need to say this brings a very nice speedup :-)

Signed-off-by: Laurent Desnogues <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Fix spelling in comment

replace Convery -> Convert

Cc: Paul Brook <[email protected]>
Signed-off-by: Stefan Weil <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Suppress some variants of English in comments

Replace surpress, supress by suppress.

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

sparc64-8bit-asi

Sparc64 alternate space load/store helpers expect 8 bit ASI value,
while wrasi implementation sign-extends ASI operand causing
for example 0x80 to appear as 0xFFFFFF80. Resulting value falls
out of switch in helpers and causes obscure load/store faults.

- correct wrasi by masking lower 8 bits of xor result
- use lower 8 bits of ASI register in helpers

Signed-off-by: Igor V. Kovalenko <[email protected]>
Signed-off-by: Blue Swirl <[email protected]>

target-mips: remove MAX_OP_PER_INSTR workaround

Now that MAX_OP_PER_INSTR has been increased to a safer value, removed
the target-mips specific workaround.

Signed-off-by: Aurelien Jarno <[email protected]>

exec-all.h: increase OPC_BUF_SIZE

Increase OPC_BUF_SIZE to compensate the MAX_OP_PER_INSTR's increase.

Signed-off-by: Aurelien Jarno <[email protected]>

exec-all.h: increase MAX_OP_PER_INSTR to 96 from 64

The x86_64 ror instruction on a 32-bit host can generate up to 77 TCG
ops. Some more space should be left for opc that are added at the end
of the translation.

Thanks to Laurent Desnogues for the debugging help.

Signed-off-by: Aurelien Jarno <[email protected]>

This files are compiled in libqemu.a now

This fixes compilation of linux-user with today qemu, please apply.

Signed-off-by: Juan Quintela <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Fix device tree compile broken by ca20cf32ab3d945155141ef737f5d08ebb373e1d

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

Fix Sparse warning about invalid access past the end of 'mode'

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

Fix Sparse warning about obsolete struct initializer

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

Export tables properly to avoid a Sparse warning

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

Add 'static' to please Sparse

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

Fix Sparse warnings about using plain integer as NULL pointer

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

Fix user targets broken by 96e132e24ee5a693069e83b6a981693588b088c1

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

Compile host-utils only once

See also facd2857783d58387885ad7cb1e4a8386f241738 and
34005a0060c176b3025a9e7c5d064615a9f80325.

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

Compile TCG runtime library only once

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

Fix spelling in comment

registrs -> registers

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

Compile ne2000 only once

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

ioports: remove unused env parameter and compile only once

The CPU state parameter is not used, remove it and adjust callers. Now we
can compile ioport.c once for all targets.

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

Compile msix only once

Get page size in device init.

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

Compile loader only once

Callers must pass ELF machine, byte swapping and symbol LSB clearing
information to ELF loader. A.out loader needs page size information, pass
that too as a parameter.

Extract prototypes to a separate file. Move loader.[ch] and elf_ops.h under hw.

Adjust callers. Also use target_phys_addr_t instead of target_ulong for
addresses: loader addresses aren't virtual.

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