Add new macro QEMU_PACKED for packed C structures

A packed struct needs different gcc attributes for compilations
with MinGW compilers because glib-2.0 adds compiler flag
-mms-bitfields which modifies the packing algorithm.

Attribute gcc_struct reverses the negative effects of -mms-bitfields.
QEMU_PACKED sets this attribute and must be used for any packed
struct which is affected by -mms-bitfields.

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

rename qemu_malloc and related to glib names for coherence

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

Fix install(1) usage to be compatible with OpenBSD's install(1).

Fix install(1) usage to be compatible with OpenBSD's install(1).

When creating a directory via the -d flag the -p flag cannot be
used at the same time. Also in the context of installing QEMU it
doesn't make sense to use the -p flag anyway so use the [default]
-c flag instead.

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

Rename qemu -> qemu-system-i386

This has been discussed before in the past.  The special casing really makes no
sense anymore.  This seems like a good change to make for 1.0.

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

main: force enabling of I/O thread

Enabling the I/O thread by default seems like an important part of declaring
1.0.  Besides allowing true SMP support with KVM, the I/O thread means that the
TCG VCPU doesn't have to multiplex itself with the I/O dispatch routines which
currently requires a (racey) signal based alarm system.

I know there have been concerns about performance.  I think so far the ones that
have come up (virtio-net) are most likely due to secondary reasons like
decreased batching.

I think we ought to force enabling I/O thread early in 1.0 development and
commit to resolving any lingering issues.

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

trace: fix out-of-tree builds

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

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

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

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

build: sort objects to remove duplicates for link

Avoid duplicate object files during the link.  There are legitimate
cases where a link command-line would include duplicate object files
because two independent subsystems both depend on common infrastructure.

Use GNU make's $(sort) function to remove duplicate object files from
the link command-line.

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

main: switch qemu_set_fd_handler to g_io_add_watch

This patch changes qemu_set_fd_handler to be implemented in terms of
g_io_add_watch().  The semantics are a bit different so some glue is required.

qemu_set_fd_handler2 is much harder to convert because of its use of polling.

The glib main loop has the major of advantage of having a proven thread safe
architecture.  By using the glib main loop instead of our own, it will allow us
to eventually introduce multiple I/O threads.

I'm pretty sure that this will work on Win32, but I would appreciate some help
testing.  I think the semantics of g_io_channel_unix_new() are really just tied
to the notion of a "unix fd" and not necessarily unix itself.

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

Add glib support to main loop

This allows GSources to be used to register callback events in QEMU.  This is
useful as it allows us to take greater advantage of glib and also because it
allows us to write code that is more easily testable outside of QEMU since we
can make use of glib's main loop in unit tests.

All new code should use glib's callback mechanisms for registering fd events
which are very well documented at:

http://developer.gnome.org/glib/stable/glib-The-Main-Event-Loop.html

And:

http://developer.gnome.org/gio/stable/

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

tcg/ppc/tcg-target.c: Avoid 'set but not used' gcc warnings

Move the declaration and initialisation of some variables in
tcg_out_qemu_ld and tcg_out_qemu_st inside CONFIG_SOFTMMU, to
avoid the "variable set but not used" warning of gcc 4.6.

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

simpletrace: fix process() argument count

The simpletrace.process() function invokes analyzer methods with the
wrong number of arguments if a timestamp should be included.  This patch
fixes the issue so that trace analysis scripts can make use of
timestamps.

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

trace: enable all events

Given that all events with programmatically-controlled state are disabled by
default, we can delete the "disable" property from all events.

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

trace: [stderr] add support for dynamically enabling/disabling events

Uses the generic interface provided in "trace/control.h" in order to provide
a programmatic interface as well as command line and monitor controls.

Signed-off-by: Fabien Chouteau <[email protected]>
Signed-off-by: Lluís Vilanova <[email protected]>

trace: [simple] disable all trace points by default

Note that this refers to the backend-specific state (whether the output must be
generated), not the event "disabled" property (which always uses the "nop"
backend).

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

trace: always use the "nop" backend on events with the "disable" keyword

Any event with the keyword/property "disable" generates an empty trace event
using the "nop" backend, regardless of the current backend.

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

trace: add "-trace events" argument to control initial state

The "-trace events" argument can be used to provide a file with a list of trace
event names that will be enabled prior to starting execution, thus providing
early tracing.

This saves the user from manually toggling event states through the monitor
interface or whichever backend-specific interface.

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

trace: always compile support for controlling and querying trace event states

The current interface is generic for this small set of operations, and thus
other backends can easily modify the "trace/control.c" file to add their own
implementation.

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

trace: separate trace event control and query routines from the simple backend

Generalize the 'st_print_trace_events' and 'st_change_trace_event_state' into
backend-specific 'trace_print_events' and 'trace_event_set_state' (respectively)
in the "trace/control.h" file.

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

trace: generalize the "property" concept in the trace-events file

This adds/modifies the following functions:

* get_name: Get _only_ the event name
* has_property: Return whether an event has a property (keyword before the event
  name)

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

trace: avoid conditional code compilation during option parsing

A default implementation for backend-specific routines is provided in
"trace/default.c", which backends can override by setting "trace_default=no" in
"configure".

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

trace: move backend-specific code into the trace/ directory

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

trace: [make] replace 'ifeq' with values in CONFIG_TRACE_*

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

trace: [configure] rename CONFIG_*_TRACE into CONFIG_TRACE_*

Provides a more hierarchical view of the variable domain.

Also adds the CONFIG_TRACE_* variables for all backends.

[Stefan added missing 'test' in stap if statement]

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

build: [simple] Include qemu-timer-common.o in trace-obj-y

Helper programs like qemu-ga use tracing primitives, but qemu-timer-common.o
(also used by simpletrace.o) is not necessarily included in the linkage line.

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

build: Fix linkage of QEMU_PROG

Using '$^' to establish the files to link with will remove any repeated entries
in the list of dependencies.

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

tusb6010: Convert to qdev

Convert the tusb6010 to qdev.

Signed-off-by: Juha Riihimäki <[email protected]>
[Riku Voipio: Fixes and restructuring patchset]
Signed-off-by: Riku Voipio <[email protected]>
[Peter Maydell: More fixes and cleanups for upstream submission]
Signed-off-by: Peter Maydell <[email protected]>
Signed-off-by: Edgar E. Iglesias <[email protected]>

memory: Fix memory_region_get_ram_ptr for ROM devices

Mask out the sub-page bits that are used by ROM device for storing the
io-index and the IO_MEM_ROMD flag.

Signed-off-by: Jan Kiszka <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

Merge branch 'omap-for-upstream' of git://git.linaro.org/people/pmaydell/qemu-arm into pm

Merge remote-tracking branch 'kwolf/for-anthony' into staging

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

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

Merge remote-tracking branch 'mst/for_anthony' into staging

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

qemu-img: Require larger zero areas for sparse handling

By default, require 4k of consecutive zero bytes for qemu-img to make the
output file sparse by not issuing a write request for the zeroed parts. Add an
-S option to allow users to tune this setting.

This helps to avoid situations where a lot of zero sectors and data sectors are
mixed and qemu-img tended to issue many tiny 512 byte writes.

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

Fix spelling in comments and debug messages (recieve -> receive)

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

hw/pci-stub: fix comment typo

[Stefan fixed "doesn't" -> "don't"]

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

memory: fix rom_device I/O mode

When adding a rom_device in I/O mode, we incorrectly masked off the low
bits, resulting in a pure RAM map.  Fix my masking off the high bits and
IO_MEM_ROMD, yielding a pure I/O map.

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

omap_gpmc: Implement prefetch engine

This commit implements the prefetch engine feature of the GPMC
which can be used for NAND devices. This includes both interrupt
driven and DMA-filling modes.

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

omap: Wire up the DMA request line to the GPMC

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

omap_gpmc: Pull prefetch engine data into sub-struct

Refactor the gpmc state structure so items relating to
the prefetch engine are in their own sub-struct and have
more useful names.

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

omap_gpmc: Accept a zero mask field on omap3630

OMAP3630 adds an extra bit of address masking, so a mask of
0xb1111 is valid. Unfortunately the GPMC_REVISION is the same as
on the OMAP3430 which only has three bits of address masking, so
we have to derive this feature directly from the OMAP revision
rather than from the GPMC revision.

Signed-off-by: Juha Riihimäki <[email protected]>
[Riku Voipio: Fixes and restructuring patchset]
Signed-off-by: Riku Voipio <[email protected]>
[Peter Maydell: More fixes and cleanups for upstream submission]
Signed-off-by: Peter Maydell <[email protected]>

hw/omap.h: Add OMAP 3630 to omap_mpu_model enumeration

Add the OMAP 3630 to the omap_mpu_model enumeration, and add the
corresponding cpu_is_omap3630() function.

(OMAP3 isn't supported yet but this is useful in upgrading common
components to be "OMAP3 ready". We already have this for OMAP3430.)

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

omap_gpmc: Support NAND devices

Support accesses to NAND devices, both by mapping them into
the GPMC address space, and via the NAND_COMMAND, NAND_ADDRESS
and NAND_DATA GPMC registers.

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

omap_gpmc: Reindent misindented switch statements

Whitespace-only change fixing indentation.

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

omap_gpmc: Calculate revision from OMAP model

Signed-off-by: Juha Riihimäki <[email protected]>
[Riku Voipio: Fixes and restructuring patchset]
Signed-off-by: Riku Voipio <[email protected]>
[Peter Maydell: More fixes and cleanups for upstream submission]
Signed-off-by: Peter Maydell <[email protected]>

omap_gpmc: Take omap_mpu_state* in omap_gpmc_init

Take a pointer to the omap mpu state struct in omap_gpmc_init.
Some details of GPMC behaviour depend on the OMAP version we
are a part of.

Signed-off-by: Juha Riihimäki <[email protected]>
[Riku Voipio: Fixes and restructuring patchset]
Signed-off-by: Riku Voipio <[email protected]>
[Peter Maydell: More fixes and cleanups for upstream submission]
Signed-off-by: Peter Maydell <[email protected]>

omap_gpmc: Fix handling of FIFOTHRESHOLDSTATUS bit

The OMAP3 TRM is inconsistent about whether the GPMC FIFOTHRESHOLDSTATUS
bit should be set when FIFOPOINTER > FIFOTHRESHOLD or when it is >=
FIFOTHRESHOLD. Apparently the underlying functional spec from which
the TRM was created states that the behaviour is ">=", and this also
makes more conceptual sense.

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

omap_gpmc: Wire up the GPMC IRQ correctly

The omap_gpmc wasn't actually wiring up its IRQ, so
anything that provoked an interrupt would be using
uninitialised data for its IRQ number.

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

omap_gpmc: GPMC_IRQSTATUS is write-one-to-clear

Fix a bug in the handling of writes to GPMC_IRQSTATUS:
it behaves as "write one to clear, writing zero is ignored".

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

omap_gpmc: Refactor omap_gpmc_cs_map and omap_gpmc_cs_unmap

Refactor the omap_gpmc_cs_map/unmap functions:
 * take the omap_gpmc_s* and a chipselect id rather than the
   omap_gpmc_cs_file_s*, so they have access to the general gpmc
   member fields
 * extract the base and mask from the config registers in the functions
   rather than at every callsite
 * check for CSVALID in the functions rather than at every callsite

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

omap_gpmc: Clean up omap_gpmc_attach MemoryRegion conversion

Now that all callers of omap_gpmc_attach pass in a MemoryRegion*,
we can remove the base_update and unmap function pointer arguments,
and the opaque pointer that was passed into these callbacks.

We can also remove the base and size fields from omap_gpmc_cs_file_s
as these are no longer necessary (you don't need the base/size
to unmap a MemoryRegion the way you did to undo a mapping made
with cpu_register_physical_memory()).

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

hw/onenand: Minor spacing fixes

Minor whitespace-only cleanup (separated out from the qdevifying
patch for clarity).

Signed-off-by: Juha Riihimäki <[email protected]>
[Riku Voipio: Fixes and restructuring patchset]
Signed-off-by: Riku Voipio <[email protected]>
[Peter Maydell: More fixes and cleanups for upstream submission]
Signed-off-by: Peter Maydell <[email protected]>

hw/onenand: Qdevify

Qdevify the ONENAND device.

Signed-off-by: Juha Riihimäki <[email protected]>
[Riku Voipio: Fixes and restructuring patchset]
Signed-off-by: Riku Voipio <[email protected]>
[Peter Maydell: More fixes and cleanups for upstream submission]
Signed-off-by: Peter Maydell <[email protected]>

hw/onenand: Remove unnecessary argument from onenand_command()

Refactor onenand_command() -- since it is essentially a method of
the device object, it doesn't make sense to pass in something as
an argument which is one of the object's own member fields.

Signed-off-by: Juha Riihimäki <[email protected]>
[Riku Voipio: Fixes and restructuring patchset]
Signed-off-by: Riku Voipio <[email protected]>
[Peter Maydell: More fixes and cleanups for upstream submission]
Signed-off-by: Peter Maydell <[email protected]>

hw/sysbus: Add sysbus_mmio_get_region()

Add a sysbus_mmio_get_region() which allows users of sysbus
devices to turn a (SysBusDevice*, mmioidx) tuple into a
MemoryRegion*. This enables some useful simplifications of
devices which pass through another device's mmio region
(either directly or by implementing some kind of memory
controller device).

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

sh_pci: Fix sh_pci memory alias confusion

The a7 area was set up as an alias of itself, rather than the p4 area.  This
sent the memory core into infinite recursion.

Fix by aliasing the a7 area to the p4 area.

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

Fix disabling interrupts in sun4u

clear interrupt request if the interrupt priority < CPU pil
clear hardware interrupt request if interrupts are disabled

Signed-off-by: Artyom Tarasenko <[email protected]>
[[email protected]: added a comment about magic 2]
Signed-off-by: Blue Swirl <[email protected]>

xilinx: Convert most xilinx devices to MemoryRegion

This converts ethlite, intc, timer and uartlite to use
MemoryRegions.

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

TCG: improve optimizer debugging

Use enum TCGOpcode instead of plain old int so that the name of
current op can be seen in GDB. Add a default case to switch
so that GCC does not complain about unhandled enum cases.

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

dyngen-exec.h: cleanup

Remove unused or otherwise available stuff.

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

g364fb: convert to qdev

Extract G364 ROM contents from device emulation to machine emulation,
so device emulation can be reused in other machines (Commodore Amiga)

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

g364fb: use trace framework

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

checkpatch: fix braces {} handling

checkpatch.pl doesn't report warning for if/else statements with missing
'else' braces:

if (something) {
    foo;
} else
    bar;

The patch has been tested using the last 100 commits.

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

Fix linker scripts

Remove PROVIDE_HIDDEN and ONLY_IF_{RO,RW} from linker scripts to make
them work with older binutils versions.  Fixes *-bsd-user build on
OpenBSD 4.9 which ships binutils 2.15.

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

Fix build on OpenBSD with BSD userland emu and smartcard NSS enabled

The first issue is the hard coded POSIX Real Time extensions library in the
libcacard/Makefile. From looking at the code it doesn't seem this is necessary
anyway. Robert Relyea seems to think it most likely isn't necessary.

The second issue was the missing exclusion of the BSD userland binary
builds from the addition of this Makefile target for the smartcard NSS
code which breaks the builds if smartcard NSS support is enabled.

pastebin clip of the build failure..

http://pastebin.com/raw.php?i=BLCKd3s6

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

block: latency accounting

Account the total latency for read/write/flush requests.  This allows
management tools to average it based on a snapshot of the nr ops
counters and allow checking for SLAs or provide statistics.

Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

vhost-net: cleanup host notifiers at last step

When the vhost notifier is disabled, the userspace handler runs
immediately: virtio_pci_set_host_notifier_internal might
call virtio_queue_notify_vq.
Since the VQ state and the tap backend state aren't
recovered yet, this causes
"Guest moved used index from XXX to YYY" assertions.

The solution is to split out host notifier handling
from vhost VQ setup and disable notifiers as our last step
when we stop vhost-net. For symmetry enable them first thing
on start.

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

linux-user: Correct a few missuses of host addresses

Fix a few cases where we were passing host pointers to the
guest.

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

memory: add opaque parameter to memory_region_init_rom_device()

The MemoryRegionOps callbacks expect it.

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

microblaze: Add an MSR_PVR constant and use it.

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

microblaze: Make the MSR PVR bit non writable

Instead of hardcoding it to 1.

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

Revert "Merge remote-tracking branch 'qemu-kvm/memory/batch' into staging"

This reverts commit 8ef9ea85a2cc1007eaefa53e6871f1f83bcef22d, reversing
changes made to 444dc48298c480e42e15a8fe676be737d8a6b2a1.

From Avi:

  Please revert the entire pull (git revert 8ef9ea85a2cc1) while I work this
  out - it isn't trivial.

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

vga: Silence bogus gcc warning about uninitialized variables

Some gcc versions do not properly detect that all possible cases are
covered and base and size are always initialized. Please gcc by defining
a pseudo default case.

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

block: explicit I/O accounting

Decouple the I/O accounting from bdrv_aio_readv/writev/flush and
make the hardware models call directly into the accounting helpers.

This means:
 - we do not count internal requests from image formats in addition
   to guest originating I/O
 - we do not double count I/O ops if the device model handles it
   chunk wise
 - we only account I/O once it actuall is done
 - can extent I/O accounting to synchronous or coroutine I/O easily
 - implement I/O latency tracking easily (see the next patch)

I've conveted the existing device model callers to the new model,
device models that are using synchronous I/O and weren't accounted
before haven't been updated yet.  Also scsi hasn't been converted
to the end-to-end accounting as I want to defer that after the pending
scsi layer overhaul.

Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

monitor: fix build breakage for !CONFIG_VNC

Commit c62f6d1 (monitor: fix build breakage with --disable-vnc)
conditionalised some VNC setup code but left an unused variable.  Move
the variable into the conditional code to fix the build breakage.

Cc: Luiz Capitulino <[email protected]>
Cc: Markus Armbruster <[email protected]>
Signed-off-by: Jamie Iles <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

qcow2: remove unused qcow2_create_refcount_update function

Signed-off-by: Frediano Ziglio <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

qcow2: use always stderr for debugging

let all DEBUG_ALLOC2 printf goes to stderr

Signed-off-by: Frediano Ziglio <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

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

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

Merge remote-tracking branch 'aneesh/for-upstream-2' into staging

memory: fix memory_region_init_rom_device() not initializing ->ops

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

disasm: update comment

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

Fix up some style nits of last uq/master merge

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

milkymist: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

milkymist-softusb: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

milkymist-minimac2: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

mcf5208: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

mainstone: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

lm32_boards: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

g364fb: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

dummy_m68k: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

pflash_cfi01/pflash_cfi02: convert to memory API

cfi02 is annoying in that is ignores some address bits; we probably
want explicit support in the memory API for that.

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

Makefile.hw: allow hw/ files to include glib headers

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

piix_pci: wrap memory update in a transaction

The code will remap all PAMs, even if just one is updated, resulting
in reduced performance.  Wrap in a transaction to detect that those
other PAMs have not changed.

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

cirrus: wrap memory update in a transaction

This prevents spurious unmapping and remapping of the vga windows,
which reduces performance.

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

leon3: convert to memory API

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>

integratorcp: convert to memory API (RAM/flash only)

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>