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 <kwolf@redhat.com>

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 <hch@lst.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

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 <hch@lst.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: remove unused qcow2_create_refcount_update function

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: use always stderr for debugging

let all DEBUG_ALLOC2 printf goes to stderr

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

sheepdog: use coroutines

This makes the sheepdog block driver support bdrv_co_readv/writev
instead of bdrv_aio_readv/writev.

With this patch, Sheepdog network I/O becomes fully asynchronous.  The
block driver yields back when send/recv returns EAGAIN, and is resumed
when the sheepdog network connection is ready for the operation.

Signed-off-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: remove memory leak

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: Removed QCowAIOCB entirely

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: reindent and use while before the big jump

prepare to remove read/write callbacks

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: remove common from QCowAIOCB

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: remove cluster_offset from QCowAIOCB

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: remove l2meta from QCowAIOCB

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: removed cur_nr_sectors field in QCowAIOCB

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: Removed unused AIOCB fields

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow: remove old #undefined code

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow: Remove QCowAIOCB

Embed qcow_aio_read_cb into qcow_co_readv and qcow_aio_write_cb into qcow_co_writev

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow: move some blocks of code to avoid useless variable initialization

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow: QCowAIOCB field cleanup

remove unused field from this structure and put some of them in qcow_aio_read_cb and qcow_aio_write_cb

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow/qcow2: Allocate QCowAIOCB structure using stack

instead of calling qemi_aio_get use stack

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

posix-aio-compat: fix latency issues

In certain circumstances, posix-aio-compat can incur a lot of latency:
 - threads are created by vcpu threads, so if vcpu affinity is set,
   aio threads inherit vcpu affinity.  This can cause many aio threads
   to compete for one cpu.
 - we can create up to max_threads (64) aio threads in one go; since a
   pthread_create can take around 30μs, we have up to 2ms of cpu time
   under a global lock.

Fix by:
 - moving thread creation to the main thread, so we inherit the main
   thread's affinity instead of the vcpu thread's affinity.
 - if a thread is currently being created, and we need to create yet
   another thread, let thread being born create the new thread, reducing
   the amount of time we spend under the main thread.
 - drop the local lock while creating a thread (we may still hold the
   global mutex, though)

Note this doesn't eliminate latency completely; scheduler artifacts or
lack of host cpu resources can still cause it.  We may want pre-allocated
threads when this cannot be tolerated.

Thanks to Uli Obergfell of Red Hat for his excellent analysis and suggestions.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: include flush requests in info blockstats

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/curl: Handle failed reads gracefully.

Current behaviour if a read fails is for the acb to not get finished.
This causes an infinite loop in bdrv_read_em (block.c). The read failure
never gets reported to the  guest and if the error condition clears, the
process never recovers.

With this patch, when curl reports a failure we finish the acb as a
failure. This results in the guest receiving an I/O error (rather than
the read hanging indefinitely) and if the error condition subsequently
clears, retries work as expected.

The simplest test is to put an ISO on a web server you have control over
and open it with qemu-io. Then move the ISO out of the way and attempt
to read some data - you should see behaviour matching the above.

Signed-off-by: Nick Thomas <nick@bytemark.co.uk>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-img: print error codes when convert fails

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow: initialize coroutine mutex

commit 52b8eb60132b27ad53476490e9d7579003390cfa added a mutex,
but never initialized it.  This caused a segfault.

Reported-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: fix typo in documentation for qcow2_get_cluster_offset()

Documentation states the num is measured in clusters, but its
actually measured in sectors

Signed-off-by: Devin Nakamura <devin122@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-img: Use qemu_blockalign

Now that you can use cache=none for the output file in qemu-img, we should
properly align our buffers so that raw-posix doesn't have to use its (smaller)
bounce buffer.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>

qcow2: Fix DEBUG_* compilation

By introducing BlockDriverState compiling qcow2 with DEBUG_ALLOC and DEBUG_EXT
defined got broken.
Define a BdrvCheckResult structure locally which is now needed as the second
argument.

Also fix qcow2_read_extensions() needing BDRVQcowState.

Signed-off-by: Philipp Hahn <hahn@univention.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add cache=directsync parameter to -drive

This patch adds -drive cache=directsync for O_DIRECT | O_SYNC host file
I/O with no disk write cache presented to the guest.

This mode is useful when guests may not be sending flushes when
appropriate and therefore leave data at risk in case of power failure.
When cache=directsync is used, write operations are only completed to
the guest when data is safely on disk.

This new mode is like cache=writethrough but it bypasses the host page
cache.

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: parse cache mode flags in a single place

This patch introduces bdrv_parse_cache_flags() which sets open flags
given a cache mode.  Previously this was duplicated in blockdev.c and
qemu-img.c.

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

coroutine: Add CoRwlock support

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

xilinx: removed microbalze_pic_init from xilinx.h

This is a microblaze target specific function that belongs outside
of xilinx.h (which is a collection of target independent device model
instantiator functions)

Signed-off-by: Peter A. G. Crosthwaite <peter.crosthwaite@petalogix.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>

xilinx.h: Added missing includes

Added some missing #includes for this file. Previously this file
relied on its clients to pre-include its dependencies.

Signed-off-by: Peter A. G. Crosthwaite <peter.crosthwaite@petalogix.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>

sdl: Don't release input on mouse mode change in full-screen mode

While in full-screen mode, the input focus naturally belongs to the SDL
window. Avoid dropping it when switching from absolute to relative
mouse mode.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Replace qemu_system_cond with VCPU stop mechanism

We can express the VCPU thread wakeup with the stop mechanism, saving
both qemu_system_ready and the qemu_system_cond. For KVM threads, we can
just enter the main loop as long as the thread is stopped. The central
TCG thread is better held back before the loop as there can be side
effects of the services called even when all CPUs are stopped.

Creating VCPUs in stopped state will also be required for proper CPU
hotplugging support.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vga: Drop some unused fields

Memory region refactorings obsoleted them.

CC: Avi Kivity <avi@redhat.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vga: Use linear mapping + dirty logging in chain 4 memory access mode

Most VGA memory access modes require MMIO handling as they demand weird
logic to get a byte from or into the video RAM. However, there is one
exception: chain 4 mode with all memory planes enabled for writing. This
mode actually allows lineary mapping, which can then be combined with
dirty logging to accelerate KVM.

This patch accelerates specifically VBE accesses like they are used by
grub in graphical mode. Not only the standard VGA adapter benefits from
this, also vmware and spice in VGA mode.

CC: Gerd Hoffmann <kraxel@redhat.com>
CC: Avi Kivity <avi@redhat.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmware-vga: Eliminate vga_dirty_log_restart

After the conversion to the new Memory API, vga_dirty_log_restart became
seriously pointless. Remove it from vmware-vga and and then finally drop
the service.

CC: Andrzej Zaborowski <balrogg@gmail.com>
CC: Avi Kivity <avi@redhat.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmware-vga: Remove dead DIRECT_VRAM mode

The code was disabled since day 1 of vmware-vga, and now it does not
even build anymore. Time for a cleanup.

CC: Andrzej Zaborowski <balrogg@gmail.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmware-vga: Disable verbose mode

Elimiates 'vmsvga_value_write: guest runs Linux.' messages from the
console.

CC: Andrzej Zaborowski <balrogg@gmail.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vmware-vga: Register reset service

Fixes cold reset in vmware graphic modes. We need to split up the reset
function for this purpose, breaking out init-once bits.

Cc: Andrzej Zaborowski <balrogg@gmail.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

ioapic: Implement polarity

If the polarity bit is set in the redirection table, the input level
simply has to inverted as it is low active in this case.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

target-i386: Remove unused polarity arguments from APIC API

Polarity of external interrupts needs to be handled in the IOAPIC.
Passing it to the APIC is pointless. So remove all these arguments.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Do not kick vcpus in TCG mode

In TCG mode, iothread and vcpus run in lock-step. So it's pointless to
send a signal from qemu_cpu_kick to the vcpu thread - if we got here,
the receiver already left the vcpu loop.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Poll main loop after I/O events were received

Polling until select returns empty fdsets helps to reduce the switches
between iothread and vcpus. The benefit of this patch is best visible
when running an SMP guest on an SMP host in emulation mode.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Do not drop global mutex for polled main loop runs

If we call select without a timeout, it's more efficient to keep the
global mutex locked as we may otherwise just play ping pong with a
vcpu thread contending for it. This is particularly important for TCG
mode where we run in lock-step with the vcpu thread.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

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

microblaze-user: Deliver SIGFPE on div by zero

Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>

memory: Fix old_portio vs non-zero offset

The legacy functions that we're wrapping expect that offset
to be included in the register.  Indeed, they generally
expect the absolute address and then mask off the "high" bits.

The FDC is the first converted device with a non-zero offset.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Avi Kivity <avi@redhat.com>

memory: temporarily suppress the subregion collision warning

After 312b4234, the APIC and PCI devices are colliding with each other.  This
is harmless in practice because the APIC accesses are special cased and never
make there way onto the bus.

Avi is working on a proper fix, but until that's ready, avoid printing the
warning.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

440fx: fix PAM, PCI holes

The current implementation of PAM and the PCI holes is broken in several
ways:

  - PCI BARs are not restricted to the PCI hole (a BAR may hide memory)
  - PCI devices do not respect PAM (if a PCI device maps a region while
    PAM maps the region to RAM, the request will be honored)

This patch fixes things by introducing a pci address space, and using
memory region aliases to represent PAM regions, SMRAM, and PCI holes.

The memory hierarchy looks something like

system_memory
 |
 +--- low memory alias (0-0xe0000000)
 |      |
 |      +-- ram@0
 |
 +--- high memory alias (0x100000000-EOM)
 |      |
 |      +-- ram@0xe0000000
 |
 +--- pci hole alias (end of low memory-0x100000000)
 |      |
 |      +-- pci@end-of-low-memory
 |
 |
 +--- pam[n] (0xc0000-0xc3fff etc) (when set to pci, priority 1)
 |      |
 |      +-- pci@0xc4000 etc
 |
 +--- smram (0xa0000-0xbffff) (when set to pci/vga, priority 1)
        |
        +-- pci@0xa0000 etc

ram (simple ram region)

pci
 |
 +--- BARn
 |
 +--- VGA 0xa0000-0xbffff
 |
 +--- ROMs

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vga: drop get_system_memory() from vga devices and derivatives

Instead, use the bus accessors, or get the address space directly
from the board constructor.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

pci: add pci_address_space()

Returns the PCI address space.  Useful for bridges that can obscure
part of the PCI address space.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

isa: add isa_address_space()

A helper that returns the address space used by ISA devices.  Useful
for getting rid of isa_mem_base, multiple ISA buses, or ISA buses behind
bridges.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

sysbus: remove sysbus_init_mmio_cb()

This problem with this function is that it is not reversible - it is
impossible to know where things are registered and unregister them
exactly.  As there are no more users, we can remove it.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

ppce500_pci: convert to sysbus_init_mmio_cb2()

Not a huge step forward, but at least we now have a 1:1 relationship
between registration and unregistration.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

versatile_pci: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

arm11mpcore: use sysbus_init_mmio_cb2

This tells the sysbus code it need not use IO_MEM_UNASSIGNED.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

sh_pci: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

sysbus: add a variant of sysbus_init_mmio_cb with an unmap callback

sysbus_init_mmio_cb() uses the destructive IO_MEM_UNASSIGNED to remove a
region.  Provide an alternative that calls an unmap callback, so the removal
may be done non-destructively.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

ppc4xx_sdram: convert to memory API

Clumsy due to the lack of clipping support, needed for
changing exposed ram size.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

ppc405_uc: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

pcie_host: convert to memory API

Assuming that mmcfg size cannot change at runtime.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

onenand: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

omap_gpmc/nseries/tusb6010: convert to memory API

Somewhat clumsy since it needs a variable sized region.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

tusb6010: move declarations to new file tusb6010.h

Avoid #include hell.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

gt64xxx.c: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

armv7m: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

arm_timer: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

arm_sysctl: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

arm_gic: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

apic: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

apb_pci: convert to memory API

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

pcnet: fix wrong opaque (broken by bd8d6f7cadb6ace98c779135217a4ed7b5fccc23)

Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Improvements to libtool support.

Improvements to the libtool support in QEMU. Replace hard coded
libtool in the infrastructure with $(LIBTOOL) and allow
overriding the libtool binary used via the configure
script.

Reviewed-by: Andreas F=E4rber <andreas.faerber@web.de>
Signed-off-by: Brad Smith <brad@comstyle.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

w32: Fix qemu_ftruncate64

SetFilePointer returns INVALID_SET_FILE_POINTER when it fails.
In addition, GetLastError must be checked.

The first call of SetFilePointer did not use INVALID_SET_FILE_POINTER,
the second call used wrong error handling.

Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

build: list libraries after objects, for proper linkage

Without this change, when using -Wl,--as-needed with GNU linker, the
libraries would be discarded.

Signed-off-by: Diego Elio Pettenò <flameeyes@flameeyes.eu>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

fix code format

Fix code format to make checkpatch.pl happy.

Signed-off-by: Robert Wang <wdongxu@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

e1000: use MII status register for link up/down

Some guests will use the standard MII status register
to verify link state.  They will not notice link changes
unless this register is updated.

Verified with Linux 3.0 and Windows XP guests.

Without this patch, ethtool will report speed and duplex as
unknown when the link is down, but still report the link as
up.  This is because the Linux e1000 driver checks the
mac_reg[STATUS] register link state before it checks speed
and duplex, but uses the phy_reg[PHY_STATUS] register for
the actual link state check.  Fix by updating both registers
on link state changes.

Linux guest before:

 (qemu) set_link e1000.0 off

 kvm-sid:~# ethtool eth0
 Settings for eth0:
        Supported ports: [ TP ]
        Supported link modes:   10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
        Supports auto-negotiation: Yes
        Advertised link modes:  10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
        Advertised pause frame use: No
        Advertised auto-negotiation: Yes
        Speed: Unknown!
        Duplex: Unknown! (255)
        Port: Twisted Pair
        PHYAD: 0
        Transceiver: internal
        Auto-negotiation: on
        MDI-X: Unknown
        Supports Wake-on: umbg
        Wake-on: d
        Current message level: 0x00000007 (7)
                               drv probe link
        Link detected: yes

 (qemu) set_link e1000.0 on

Linux guest after:

 (qemu) set_link e1000.0 off
 [   63.384221] e1000: eth0 NIC Link is Down

 kvm-sid:~# ethtool eth0
 Settings for eth0:
        Supported ports: [ TP ]
        Supported link modes:   10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
        Supports auto-negotiation: Yes
        Advertised link modes:  10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
        Advertised pause frame use: No
        Advertised auto-negotiation: Yes
        Speed: Unknown!
        Duplex: Unknown! (255)
        Port: Twisted Pair
        PHYAD: 0
        Transceiver: internal
        Auto-negotiation: on
        MDI-X: Unknown
        Supports Wake-on: umbg
        Wake-on: d
        Current message level: 0x00000007 (7)
                               drv probe link
        Link detected: no

 (qemu) set_link e1000.0 on
 [   84.304582] e1000: eth0 NIC Link is Up 1000 Mbps Full Duplex, Flow Control: RX

Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: document the functions that will be the public interface

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: remove qemu_chr_send_event()

It's dead code.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_get_msgfd() -> qemu_chr_fe_get_msgfd()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_close() -> qemu_chr_delete()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: qemu_chr_open_opts() -> qemu_chr_new_from_opts()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_open() -> qemu_chr_new()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_set_echo() -> qemu_chr_fe_set_echo()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: qemu_chr_ioctl() -> qemu_chr_fe_ioctl()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_guest_close() -> qemu_chr_fe_close()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_guest_open() -> qemu_chr_fe_open()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_can_read() -> qemu_chr_be_can_read()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_read() -> qemu_chr_be_write()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_printf() -> qemu_chr_fe_printf()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

char: rename qemu_chr_write() -> qemu_chr_fe_write()

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

tcg/ppc64: fix 16/32 mixup

Signed-off-by: malc <av1474@comtv.ru>

Merge branch 'master' of git://git.qemu.org/qemu

tcg/ppc64: implement not_i32/64 and ext32u_i64

Signed-off-by: malc <av1474@comtv.ru>

tcg/ppc32: implement deposit_i32

Signed-off-by: malc <av1474@comtv.ru>

Merge remote-tracking branch 'pmaydell/armhw-for-upstream' into staging

guest agent: remove uneeded dependencies

This patch tries to cull any uneeded library dependencies from the guest
agent to improve portability across various distros. We do so by being
as explicit as possible about in-tree dependencies rather than relying
on existing *-obj-y targets, and by manually setting LIBS for the
qemu-ga target to avoid pulling in LIBS_TOOLS libraries discovered by
configure.

Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

guest agent: remove g_strcmp0 usage

g_strcmp0 isn't in all version of glib 2.0, so don't use it to avoid
build breakage on older distros.

Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

memory: crack wide ioport accesses into smaller ones when needed

The memory API supports cracking wide accesses into narrower ones
when needed; but this was no implemented for the pio address space,
causing lsi53c895a's IO BAR to malfunction.

Fix by correctly cracking wide accesses when needed.

Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>