move helpers.h to helper.h

This provides a consistent naming scheme across all targets.

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

libcacard: fix opposite usage of isspace

Signed-off-by: Alon Levy <[email protected]>
Tested-by: Hans de Goede <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

target-mips: clear softfpu exception state for comparison instructions

MIPS FPU instructions should start with a clean softfpu status. This
is done for the most instructions, but not for comparison ones.

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

target-mips: fix c.ps.* instructions

Contrary to cabs.ps.* instructions, c.ps.* should not compare the absolute
value of the operand, but directly the operands.

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

target-mips: don't hardcode softfloat exception bits

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

target-mips: simplify FP comparisons

As the softfloat comparison functions already test for NaN, there is no
need to always call the float*_unordered*() functions.

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

target-ppc: fix SPE comparison functions

efstst*() functions are fast SPE funtions which do not take into account
special values (infinites, NaN, etc.), while efscmp*() functions are
IEEE754 compliant.

Given that float32_*() functions are IEEE754 compliant, the efscmp*()
functions are correctly implemented, while efstst*() are not. This
patch reverse the implementation of this two groups of functions and
fix the comments. It also use float32_eq() instead of float32_eq_quiet()
as qNaNs should not be ignored.

Cc: Alexander Graf <[email protected]>
Cc: Nathan Froyd <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

softfloat: improve description of comparison functions

Make clear for all comparison functions which ones trigger an exception
for all NaNs, and which one only for sNaNs.

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

softfloat: move float*_eq and float*_eq_quiet

I am not a big fan of code moving, but having the signaling version in
the middle of quiet versions and vice versa doesn't make the code easy
to read.

This patch is a simple code move, basically swapping locations of
float*_eq and float*_eq_quiet.

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

softfloat: rename float*_eq_signaling() into float*_eq()

float*_eq_signaling functions have a different semantics than other
comparison functions. Fix that by renaming float*_quiet_signaling() into
float*_eq().

Note that it is purely mechanical, and the behaviour should be unchanged.

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

softfloat: rename float*_eq() into float*_eq_quiet()

float*_eq functions have a different semantics than other comparison
functions. Fix that by first renaming float*_quiet() into float*_eq_quiet().

Note that it is purely mechanical, and the behaviour should be unchanged.
That said it clearly highlight problems due to this different semantics,
they are fixed later in this patch series.

Cc: Alexander Graf <[email protected]>
Acked-by: Edgar E. Iglesias <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

target-i386: fix CMPUNORDPS/D and CMPORDPS/D instructions

SSE instructions CMPUNORDPS/D and CMPORDPS/D do not trigger an invalid
exception if operands are qNANs.

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

target-mips: use new float*_unordered*() functions

Use the new float*_unordered*() functions from softfloat instead of
redefining a private version.

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

target-alpha: use new float64_unordered_quiet() function

Use float64_unordered_quiet() in helper_cmptun() instead of doing the
the comparison manually.

According to the "Alpha Compiler Writer's Guide", we should use the
_quiet version here, as CMPTUN and CMPTEQ should generate InvalidOp
for SNaNs but not for QNaNs.

Thanks to Peter Maydell <[email protected]> and Richard
Henderson <[email protected]> for digging into the manuals.

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

softfloat-native: add float*_unordered_quiet() functions

Add float*_unordered_quiet() functions to march the softfloat versions.
As FPU status is not tracked with softfloat-native, they don't differ
from the signaling version.

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

softfloat: add float*_unordered_{,quiet}() functions

Add float*_unordered() functions to softfloat, matching the softfloat-native
ones. Also add float*_unordered_quiet() functions to match the others
comparison functions.

This allow target-i386/ops_sse.h to be compiled with softfloat.

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

target-i386: add floatx_{add,mul,sub} and use them

Add floatx_{add,mul,sub} defines, and use them instead of using direct
C operations.

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

target-i386: use float unions from cpu-all.h

Use float unions from cpu-all.h instead of redefining new (wrong for arm)
ones in target-i386. This also allows building cpu-exec.o with softfloat.

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

target-i386: use CPU_LDoubleU instead of a private union

Use CPU_LDoubleU in cpu_dump_state() instead of redefining a union for
doing the conversion.

Based on a patch from Laurent Vivier <[email protected]>.

Cc: Laurent Vivier <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

cpu-all.h: define CPU_LDoubleU

Add a CPU_LDoubleU type, matching the floatx80 definition and the long
double type on x86 hosts.

Based on a patch from Laurent Vivier <[email protected]>.

Cc: Laurent Vivier <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

softfloat: use GCC builtins to count the leading zeros

Softfloat has its own implementation to count the leading zeros. However
a lot of architectures have either a dedicated instruction or an
optimized to do that. When using GCC >= 3.4, this patch uses GCC builtins
instead of the handcoded implementation.

Note that I amware that QEMU_GNUC_PREREQ is defined in osdep.h and that
clz32() and clz64() are defined in host-utils.h, but I think it is better
to keep the softfloat implementation self contained.

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

usb: fix spelling errors in usb-linux.c

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

usb: initialise data element in Linux USB_DISCONNECT ioctl

This isn't used, but leaving it empty causes valgrind noise.

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

usb: trivial spelling fixes

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

Makefile: Clean up after "make pdf"

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

Fix some typos in comments and documentation

helpfull -> helpful
usefull -> useful
cotrol -> control

and a grammar fix.

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

migration: move some declarations to migration.h

Move a few migration related declarations to migration.h.

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

Move CPU related functions to cpus.h

Move declarations of CPU related functions to cpus.h. Adjust the only user.

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

Move clock related functions to qemu-timer.h

Move declarations for clock related functions from sysemu.h to qemu-timer.h.

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

Move generic or OS function declarations to qemu-common.h

Move generic or OS related function declarations and macro
TFR to qemu-common.h.

Move win32 include directives to qemu-os-win32.h. While moving,
also add #include <winsock2.h> to fix a recent mingw32
build breakage.

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

Use qemu-common.h or qemu-timer.h in place of sysemu.h

In some cases qemu-common.h or qemu-timer.h can be used in place
of sysemu.h.

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

Remove unused sysemu.h include directives

Remove unused sysemu.h include directives to speed up build
with the following patches.

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

json-lexer: fix conflict with mingw32 ERROR definition

The name ERROR is too generic, it conflicts with mingw32 ERROR definition.

Replace ERROR with IN_ERROR.

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

sparc: Fix assertion caused by empty memory slot with 0 byte

If the memory size given on the command line is equal to the
maximum size of memory defined by the hardware, there is no
"empty slot" after physical memory.

The following command

qemu-system-sparc -m 256

raised an assertion:
exec.c:2614: cpu_register_physical_memory_offset: Assertion `size' failed

This can be fixed either at the caller side (don't call empty_slot_init)
or in empty_slot_init (do nothing) when size == 0. The second solution
was choosen here because it is more robust.

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

qemu_next_deadline should not consider host-time timers

It is purely for icount-based virtual timers.  And now that we got the
code right, rename the function to clarify the intended scope.

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

Revert wrong fixes for -icount in the iothread case

This reverts commits 225d02cd and c9f7383c.  While some parts of
the latter could be saved, I preferred a smooth, complete revert.

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

enable vm_clock to "warp" in the iothread+icount case

The previous patch however is not enough, because if the virtual CPU
goes to sleep waiting for a future timer interrupt to wake it up, qemu
deadlocks.  The timer interrupt never comes because time is driven by
icount, but the vCPU doesn't run any insns.

You could say that VCPUs should never go to sleep in icount
mode if there is a pending vm_clock timer; rather time should
just warp to the next vm_clock event with no sleep ever taking place.
Even better, you can sleep for some time related to the
time left until the next event, to avoid that the warps are too visible
externally; for example, you could be sending network packets continously
instead of every 100ms.

This is what this patch implements.  qemu_clock_warp is called: 1)
whenever a vm_clock timer is adjusted, to ensure the warp_timer is
synchronized; 2) at strategic points in the CPU thread, to make sure
the insn counter is synchronized before the CPU starts running.
In any case, the warp_timer is disabled while the CPU is running,
because the insn counter will then be making progress on its own.

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

really fix -icount in the iothread case

The correct fix for -icount is to consider the biggest difference
between iothread and non-iothread modes.  In the traditional model,
CPUs run _before_ the iothread calls select (or WaitForMultipleObjects
for Win32).  In the iothread model, CPUs run while the iothread
isn't holding the mutex, i.e. _during_ those same calls.

So, the iothread should always block as long as possible to let
the CPUs run smoothly---the timeout might as well be infinite---and
either the OS or the CPU thread itself will let the iothread know
when something happens.  At this point, the iothread wakes up and
interrupts the CPU.

This is exactly the approach that this patch takes: when cpu_exec_all
returns in -icount mode, and it is because a vm_clock deadline has
been met, it wakes up the iothread to process the timers.  This is
really the "bulk" of fixing icount.

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

tests/test-mmap.c: Check mmap() return value before using it

Correct the position of a "stop if MAP_FAILED" check in the mmap()
tests, so that if mmap() does fail we print a failure message
rather than segfaulting inside memcpy().

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

target-arm: Don't overflow when calculating value for signed VABAL

In the VABAL instruction we take the absolute difference of two
values of size x and store it in a result of size 2x. This means
we have to be careful to calculate the absolute difference using
a wide enough type that we don't accidentally overflow.

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

Revert SeaBIOS change due to overzealous commit -a

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

lm32: fix build breakage due to uninitialized variable 'r'

gcc 4.5.2 correctly complains that r is potentially uninitialized in this
function.

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

vpc.c: Use get_option_parameter() does the search

Use get_option_parameter() to instead of duplicating the loop, and
use BDRV_SECTOR_SIZE to instead of 512

Signed-off-by: Mitnick Lyu <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

atapi: GESN: implement 'media' subcommand

Implement the 'media' sub-command of the GET_EVENT_STATUS_NOTIFICATION
command.  This helps us report tray open, tray closed, no media, media
present states to the guest.

Newer Linux kernels (2.6.38+) rely on this command to revalidate discs
after media change.

This patch also sends out tray open/closed status to the guest driver
when requested e.g. via the CDROM_DRIVE_STATUS ioctl (thanks Markus).
Without such notification, the guest and qemu's tray open/close status
was frequently out of sync, causing installers like Anaconda detecting
no disc instead of tray open, confusing them terribly.

Signed-off-by: Amit Shah <[email protected]>
Acked-by: Jes Sorensen <[email protected]>
Acked-by: Paolo Bonzini <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

atapi: GESN: Standardise event response handling for future additions

Handle GET_EVENT_STATUS_NOTIFICATION's No Event Available response in a
generic way so that future additions to the code to handle other
response types is easier.

Signed-off-by: Amit Shah <[email protected]>
Acked-by: Jes Sorensen <[email protected]>
Acked-by: Paolo Bonzini <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

atapi: GESN: Use structs for commonly-used field types

Instead of using magic numbers, use structs that are more descriptive of
the fields being used.

Signed-off-by: Amit Shah <[email protected]>
Acked-by: Jes Sorensen <[email protected]>
Acked-by: Paolo Bonzini <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

atapi: Move GET_EVENT_STATUS_NOTIFICATION command handling to its own function

This makes the code more readable.

Also, there's a block like:

if () {
  ...
} else {
  ...
}

Split that into

if () {
  ...
  return;
}
...

Signed-off-by: Amit Shah <[email protected]>
Acked-by: Jes Sorensen <[email protected]>
Acked-by: Paolo Bonzini <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

atapi: Allow GET_EVENT_STATUS_NOTIFICATION after media change

After a media change, the only commands allowed from the guest were
REQUEST_SENSE and INQUIRY.  The guest may also issue
GET_EVENT_STATUS_NOTIFICATION commands to get media
changed notification.

Signed-off-by: Amit Shah <[email protected]>
Acked-by: Jes Sorensen <[email protected]>
Acked-by: Paolo Bonzini <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

atapi: Report correct errors on guest eject request

Table 629 of the MMC-5 spec mentions two different error conditions when
a CDROM eject is requested: a) while a disc is inserted and b) while a
disc is not inserted.

Ensure we return the appropriate error for the present condition of the
drive and disc status.

Signed-off-by: Amit Shah <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

atapi: Drives can be locked without media present

Drivers are free to lock drives without any media present.  Such a
condition should not result in an error condition.

See Table 341 in MMC-5 spec for details.

Signed-off-by: Amit Shah <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

qed: Add support for zero clusters

Zero clusters are similar to unallocated clusters except instead of reading
their value from a backing file when one is available, the cluster is always
read as zero.

This implements read support only.  At this stage, QED will never write a
zero cluster.

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

docs: Describe zero data clusters in QED specification

Zero data clusters are a space-efficient way of storing zeroed regions
of the image.

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

target-arm: Detect tininess before rounding for FP operations

The ARM architecture mandates that we detect tininess before rounding,
so set the softfloat fp_status up appropriately.

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

softfloat: Add setter function for tininess detection mode

Add a setter function for the underflow tininess detection mode,
in line with the similar functions for other parts of the float status
structure.

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

target-arm: Handle UNDEF cases for VDUP (scalar)

Handle the UNDEF cases for VDUP(scalar):
 imm4 == x000
 Q == 1 && Vd<0> == 1

Signed-off-by: Juha Riihimäki <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

target-arm: Treat UNPREDICTABLE VTBL, VTBX case as UNDEF

Catch the UNPREDICTABLE case for Neon VTBL,VTBX, and UNDEF it
rather than allowing the helper function to index off the end
of the register file.

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

target-arm: Handle UNDEF cases for Neon 2 register misc forms

Add missing UNDEF checks for Neon "two register miscellaneous" forms:
 * all instructions except VMOVN,VQMOVN must UNDEF
   if Q==1 && (Vd<0> == 1 || Vm<0> == 1)
 * VMOVN,VQMOVN,VCVT.F16.F32 UNDEF if Q == 1 || Vm<0> == 1
 * VSHLL,VCVT.F32.F16 UNDEF if Q == 1 || Vd<0> == 1
(The only other UNDEF case is VZIP,VUZP if Q == 0 && size == 10,
which we already handle.)

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

target-arm: Simplify checking of size field in Neon 2reg-misc forms

Many of the Neon "2 register misc" instruction forms require invalid
size fields to cause the instruction to UNDEF. Pull this information
out into an array; this simplifies the code and also means we can do
the check early and avoid the problem of leaking TCG temporaries in
the illegal_op case.

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

target-arm: Handle UNDEF cases for VEXT

VEXT must UNDEF if Q == 1 && (Vd<0> == 1 || Vr<0> == 1 || Vm<0> == 1)

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

target-arm: Handle UNDEF cases for Neon 2 regs + scalar forms

Add missing checks for cases which must UNDEF in the Neon "2 registers and
a scalar" data processing instruction space.

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

target-arm: Handle UNDEF cases for Neon 3-regs-different-widths

Add missing UNDEF checks for instructions in the Neon "3 registers of
different widths" data processing space.

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

target-arm: Handle UNDEF cases for Neon invalid modified-immediates

For Neon "one register and a modified immediate value" forms, the
combination op=1 cmode=1111 is unallocated and should UNDEF.
All instructions of this form also UNDEF if Q == 1 and Vd<0> == 1.
We also add a comment on the only UNPREDICTABLE in this space.

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

target-arm: Collapse VSRI case into VSHL, VSLI

Collapse some switch cases for VSRI into those for VSHL, VSLI,
since the bodies are the same. (This is not completely obvious
for the size < 3 case, but since for VSRI we know U=1 the
GEN_NEON_INTEGER_OP() expansion is equivalent to the open-coded
VSHL/VSLI case.)

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

target-arm: Handle UNDEF cases for Neon "2 regs and shift" insns

Correctly handle all the UNDEF cases for Neon instructions of the
"2 registers and shift" form, and make sure that we check for these
cases early enough not to leak TCG temporaries.

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

target-arm: Simplify three-register pairwise code

Since we know that the case of (pairwise && q) has been caught
earlier, we can simplify the register setup code for each pass
in the three-register-same-size Neon loop.

Signed-off-by: Juha Riihimäki <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

target-arm: Handle UNDEF cases for Neon 3-regs-same insns

Correct the handling of UNDEF cases for the NEON "3 registers same
size" forms, by adding missing checks and rationalising some others
so they are done early enough to avoid leaking TCG temporaries.

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

target-arm: Use lookup table for size check on Neon 3-reg-same insns

Simplify the checks for invalid size values for the Neon "three registers
of the same size" instruction forms (and add them where they were missing)
by using a lookup table.

This includes adding symbolic constants for the op values in this space,
since we now use them in multiple places.

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

fix acpi regression

This bug is introduced by commit 23910d3f.

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

Replace cpu_physical_memory_rw were possible

Using cpu_physical_memory_read, cpu_physical_memory_write and ldub_phys
improves readability and allows removing some type casts.

lduw_phys and ldl_phys were not used because both require aligned
addresses. Therefore it is not possible to simply replace existing
calls by one of these functions.

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

monitor: Remove some type casts which are no longer needed

All other type casts in calls of cpu_physical_memory_read are
used by hardware emulations and will be fixed by separate patches.

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

exec: Remove a type cast which is no longer needed

All other type casts in calls of cpu_physical_memory_write are
used by hardware emulations and will be fixed by separate patches.

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

cpu-common: Modify cpu_physical_memory_read and cpu_physical_memory_write

A lot of calls don't operate on bytes but on words or on structured data.
So instead of a pointer to uint8_t, a void pointer is the better choice.

This allows removing many type casts.

(Some very early implementations of memcpy used char pointers
which were replaced by void pointers for the same reason).

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

configure: disable opengl per default

There is a bug in nvidia's binary GPU driver, which causes a segmentation
fault if linked to libGL.

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

target-ppc: remove #ifdef FLOAT128

Now that PPC defaults to softfloat which always provides float128
support, there is no need to keep two version of the code, depending if
float128 support is available or not. Suggested by Peter Maydell.

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

target-sh4: get rid of CPU_{Float,Double}U

SH4 is always using softfloat, so it's possible to have helpers directly
taking float32 or float64 value. This allow to get rid of conversions
through CPU_{Float,Double}U.

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

unicore32: necessary modifications for other files to support unicore32

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

unicore32: add necessry headers in linux-user/unicore32 for unicore32 support

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

unicore32: add target-unicore32 directory for unicore32-linux-user support

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

microblaze: Add partial decoding of stream insns

Based on a patch from: Alejandro Cabrera <[email protected]>

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

microblaze: Add stream-insn related constants

Based on a patch from: Alejandro Cabrera <[email protected]>

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

microblaze: Correct ec mask in debug print

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

microblaze: Add constant for exception-code mask

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

microblaze: Correct MMU_ZONES mask

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

ppc: remove a write-only variable

Remove a write-only variable, spotted by GCC 4.6.0:
/src/qemu/hw/ppc.c: In function 'power7_set_irq':
/src/qemu/hw/ppc.c:255:9: error: variable 'cur_level' set but not used [-Werror=unused-but-set-variable]

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

s390x: fix virtio feature bitmap

The feature bitmap in the s390 virtio machine is little endian. To
address for that, we need to bswap the values after reading them out.

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

s390x: set alignment for long to 8

The alignment for longs on s390x is 8. That's the only place where it differs
from the default alignments found in configure already. The example alignment
program from Laurent printed the following on a real s390x:

  alignof(short) 2
  alignof(int) 4
  alignof(long) 8
  alignof(long long) 8

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

musicpal: Fix registration of MMIO-less sysbus devices

The proper way to signal that a sysbus devices need no MMIO region is to
pass -1 to sysbus_create_simple.

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

acpi, acpi_piix: factor out GPE logic

factor out ACPI GPE logic. Later it will be used by ICH9 ACPI.

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

arm: basic support for ARMv4/ARMv4T emulation

Currently target-arm/ assumes at least ARMv5 core. Add support for
handling also ARMv4/ARMv4T. This changes the following instructions:

BX(v4T and later)

BKPT, BLX, CDP2, CLZ, LDC2, LDRD, MCRR, MCRR2, MRRC, MCRR, MRC2, MRRC,
MRRC2, PLD QADD, QDADD, QDSUB, QSUB, STRD, SMLAxy, SMLALxy, SMLAWxy,
SMULxy, SMULWxy, STC2 (v5 and later)

All instructions that are "v5TE and later" are also bound to just v5, as
that's how it was before.

This patch doesn _not_ include disabling of cp15 access and base-updated
data abort model (that will be required to emulate chips based on a
ARM7TDMI), because:
* no ARM7TDMI chips are currently emulated (or planned)
* those features aren't strictly necessary for my purposes (SA-1 core
  emulation).

All v5 models are handled as they are v5T. Internally we still have a
check if the model is a v5(T) or v5TE, but as all emulated cores are
v5TE, those two cases are simply aliased (for now).

Patch is heavily based on patch by Filip Navara <[email protected]>
which in turn is based on work by Ulrich Hecht <[email protected]> and Vincent
Sanders <[email protected]>.

Signed-off-by: Dmitry Eremin-Solenikov <[email protected]>
Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

Fix conversions from pointer to tcg_target_long

tcg_gen_exit_tb takes a parameter of type tcg_target_long,
so the type casts of pointer to long should be replaced by
type casts of pointer to tcg_target_long (suggested by Blue Swirl).

These changes are needed for build environments where
sizeof(long) != sizeof(void *), especially for w64.

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

vnc: tight: Fix crash after 2GB of output

fix 2Gb integer overflow in in VNC tight and zlib encodings

As found by Roland Dreier <[email protected]> (excellent
catch!), when amount of VNC compressed data produced by zlib
and sent to client exceeds 2Gb, integer overflow occurs because
currently, we calculate amount of data produced at each step by
comparing saved total_out with new total_out, and total_out is
something which grows without bounds.  Compare it with previous
avail_out instead of total_out, and leave total_out alone.

The same code is used in vnc-enc-tight.c and vnc-enc-zlib.c,
so fix both cases.

There, there's no actual need to save previous_out value, since
capacity-offset (which is how that value is calculated) stays
the same so it can be recalculated again after call to deflate(),
but whole thing becomes less readable this way.

Reported-by: Roland Dreier <[email protected]>
Signed-off-by: Michael Tokarev <[email protected]>
Signed-off-by: Corentin Chary <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

smbus_eeprom: consolidate smbus eeprom creation oc pc_piix, mips_mapta, mips_fulong

consolidate smbus initialization for pc_piix, mips_malta and mips_fulong.

Cc: Aurelien Jarno <[email protected]>
Cc: Huacai Chen <[email protected]>
Signed-off-by: Isaku Yamahata <[email protected]>
Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

lan9118: Ignore write to MAC_VLAN1 register

On Mon, 4 Apr 2011 20:15:30 +0200, Aurelien Jarno <[email protected]> wrote:
> Is it really safe ignoring write to this register? If yes, it's probably
> a good idea to explain why in a comment. In any case, if supporting this
> register is easy to do, it would be the best option.

I think it is safe.  Please see an updated comment below.

And though implementing this register might be possible, I suppose it
is not worth to supporting FrameTooLong detection, for now at least.

Thank you for comments.

>8---------------------------------------------------------------------
From: Atsushi Nemoto <[email protected]>
Date: Tue, 5 Apr 2011 23:12:07 +0900
Subject: [PATCH] lan9118: Ignore write to MAC_VLAN1 register

Since linux 2.6.38, smsc911x driver writes to VLAN1 registger.
Since this register only affects FrameTooLong detection, ignoring
write to this register should be safe.

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

acpi, acpi_piix, vt82c686: factor out PM1_CNT logic

factor out ACPI PM1_CNT logic. This will be used by ich9 acpi.

Cc: Blue Swirl <[email protected]>
Cc: Huacai Chen <[email protected]>
Cc: Aurelien Jarno <[email protected]>
Signed-off-by: Isaku Yamahata <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

acpi, acpi_piix, vt82c686: factor out PM1a EVT logic

factor out ACPI PM1a EVT logic.
Later this will be used by ich9 acpi.

Cc: Blue Swirl <[email protected]>
Cc: Huacai Chen <[email protected]>
Cc: Aurelien Jarno <[email protected]>
Signed-off-by: Isaku Yamahata <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

acpi, acpi_piix, vt82c686: factor out PM_TMR logic

factor out PM_TMR logic. Later This will be used by ich9 acpi.
Also fixes the same bug in vt82c686.c that was fixed by the following
commits.

> commit 055479feab63607b8042bb8ebb2e0523f17cbc4e
> Author: aliguori <aliguori@c046a42c-6fe2-441c-8c8c-71466251a162>
> Date:   Wed Jan 21 16:31:20 2009 +0000
>
>     Always return latest pmsts instead of the old one (Xiantao Zhang)
>
>     It may lead to the issue when booting windows guests with acpi=1
>     if return the old pmsts.
>
>     Signed-off-by: Xiantao Zhang <[email protected]>
>     Signed-off-by: Anthony Liguori <[email protected]>

Cc: Blue Swirl <[email protected]>
Cc: Huacai Chen <[email protected]>
Cc: Aurelien Jarno <[email protected]>
Signed-off-by: Isaku Yamahata <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

hw/pflash_cfi02: Fix lazy reset of ROMD mode

When checking pfl->rom_mode for when to lazily reenter ROMD mode,
the value was check was the opposite of what it should have been.
This prevent the part from returning to ROMD mode after a write
was made to the CFI rom region.

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

configure: avoid basename usage message

basename prints a missing-argument error when sdlconfig is empty
and we're cross-compiling.

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

mpc85xx_pci_map_irq: change "unknow" to "unknown"

Signed-off-by: Scott Wood <[email protected]>
Acked-by: Alexander Graf <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

event: trivial coding style fixes

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