linux-user: remove unused image_info members

Reviewed-by: Peter Maydell <[email protected]>
Signed-off-by: Stefan Brüns <[email protected]>
Signed-off-by: Riku Voipio <[email protected]>

linux-user: Treat --foo options the same as -foo

The system mode binaries provide a similar alias
and it makes common options like --version and --help
work as expected.

Signed-off-by: Meador Inge <[email protected]>
Signed-off-by: Riku Voipio <[email protected]>

linux-user: use EXIT_SUCCESS and EXIT_FAILURE

As suggested by Laurent, use EXIT_SUCCESS and EXIT_FAILURE from
stdlib.h instead of numeric values.

Cc: Laurent Vivier <[email protected]>
Signed-off-by: Riku Voipio <[email protected]>

linux-user: Add proper error messages for bad options

This patch adds better support for diagnosing option
parser errors.  The previous implementation just printed
the usage text and exited when a bad option or argument
was found.  This made it very difficult to determine why
the usage was being displayed and it was doubly confusing
for cases like '--help' (it wasn't clear that --help was
actually an error).

Signed-off-by: Meador Inge <[email protected]>
Signed-off-by: Riku Voipio <[email protected]>

linux-user: Add -help

This option is already available on the system mode
binaries.  It would be better if long options were
supported (i.e. --help), but this is okay for now.

Signed-off-by: Meador Inge <[email protected]>
Signed-off-by: Riku Voipio <[email protected]>

linux-user: Exit 0 when -h is used

Signed-off-by: Meador Inge <[email protected]>
Signed-off-by: Riku Voipio <[email protected]>

Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging

* First batch of MAINTAINERS updates
* IOAPIC fixes (to pass kvm-unit-tests with -machine kernel_irqchip=off)
* NBD API upgrades from Daniel
* strtosz fixes from Marc-André
* improved support for readonly=on on scsi-generic devices
* new "info ioapic" and "info lapic" monitor commands
* Peter Crosthwaite's ELF_MACHINE cleanups
* docs patches from Thomas and Daniel

# gpg: Signature made Fri 25 Sep 2015 11:20:52 BST using RSA key ID 78C7AE83
# gpg: Good signature from "Paolo Bonzini <[email protected]>"
# gpg:                 aka "Paolo Bonzini <[email protected]>"

* remotes/bonzini/tags/for-upstream: (52 commits)
  doc: Refresh URLs in the qemu-tech documentation
  docs: describe the QEMU build system structure / design
  typedef: add typedef for QemuOpts
  i386: interrupt poll processing
  i386: partial revert of interrupt poll fix
  ppc: Rename ELF_MACHINE to be PPC specific
  i386: Rename ELF_MACHINE to be x86 specific
  alpha: Remove ELF_MACHINE from cpu.h
  mips: Remove ELF_MACHINE from cpu.h
  sparc: Remove ELF_MACHINE from cpu.h
  s390: Remove ELF_MACHINE from cpu.h
  sh4: Remove ELF_MACHINE from cpu.h
  xtensa: Remove ELF_MACHINE from cpu.h
  tricore: Remove ELF_MACHINE from cpu.h
  or32: Remove ELF_MACHINE from cpu.h
  lm32: Remove ELF_MACHINE from cpu.h
  unicore: Remove ELF_MACHINE from cpu.h
  moxie: Remove ELF_MACHINE from cpu.h
  cris: Remove ELF_MACHINE from cpu.h
  m68k: Remove ELF_MACHINE from cpu.h
  ...

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

Merge remote-tracking branch 'remotes/awilliam/tags/vfio-update-20150925.0' into staging

VFIO updates 2015-09-25

 - Remove use of g_malloc0_n for glib2.22 compat

# gpg: Signature made Fri 25 Sep 2015 17:58:04 BST using RSA key ID 3BB08B22
# gpg: Good signature from "Alex Williamson <[email protected]>"
# gpg:                 aka "Alex Williamson <[email protected]>"
# gpg:                 aka "Alex Williamson <[email protected]>"
# gpg:                 aka "Alex Williamson <[email protected]>"

* remotes/awilliam/tags/vfio-update-20150925.0:
  vfio/pci: Remove use of g_malloc0_n() from quirks

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

Merge remote-tracking branch 'remotes/cody/tags/block-pull-request' into staging

# gpg: Signature made Fri 25 Sep 2015 16:47:31 BST using RSA key ID C0DE3057
# gpg: Good signature from "Jeffrey Cody <[email protected]>"
# gpg:                 aka "Jeffrey Cody <[email protected]>"
# gpg:                 aka "Jeffrey Cody <[email protected]>"

* remotes/cody/tags/block-pull-request:
  sheepdog: refine discard support
  sheepdog: use per AIOCB dirty indexes for non overlapping requests
  Backup: don't do copy-on-read in before_write_notifier
  block: Introduce a new API bdrv_co_no_copy_on_readv()
  sheepdog: add reopen support
  block/nfs: cache allocated filesize for read-only files
  block/nfs: fix calculation of allocated file size

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

Merge remote-tracking branch 'remotes/vivier-misc/tags/pull-muldiv64-20150925' into staging

Remove muldiv64() by using period instead of frequency

# gpg: Signature made Fri 25 Sep 2015 14:54:37 BST using RSA key ID 3F2FBE3C
# gpg: Good signature from "Laurent Vivier <[email protected]>"
# gpg:                 aka "Laurent Vivier <[email protected]>"
# gpg:                 aka "Laurent Vivier (Red Hat) <[email protected]>"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: CD2F 75DD C8E3 A4DC 2E4F  5173 F30C 38BD 3F2F BE3C

* remotes/vivier-misc/tags/pull-muldiv64-20150925:
  net: remove muldiv64()
  bt: remove muldiv64()
  hpet: remove muldiv64()
  arm: clarify the use of muldiv64()
  openrisc: remove muldiv64()
  mips: remove muldiv64()
  pcnet: remove muldiv64()
  rtl8139: remove muldiv64()
  i6300esb: remove muldiv64()

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

Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

virtio,pc features, fixes

New features:
    vhost-user multiqueue support
    virtio-ccw virtio 1 support

Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Fri 25 Sep 2015 07:40:35 BST using RSA key ID D28D5469
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>"
# gpg:                 aka "Michael S. Tsirkin <[email protected]>"

* remotes/mst/tags/for_upstream:
  MAINTAINERS: add more devices to the PCI section
  MAINTAINERS: add more devices to the PC section
  vhost-user: add a new message to disable/enable a specific virt queue.
  vhost-user: add multiple queue support
  vhost: introduce vhost_backend_get_vq_index method
  vhost-user: add VHOST_USER_GET_QUEUE_NUM message
  vhost: rename VHOST_RESET_OWNER to VHOST_RESET_DEVICE
  vhost-user: add protocol feature negotiation
  vhost-user: use VHOST_USER_XXX macro for switch statement
  virtio-ccw: enable virtio-1
  virtio-ccw: feature bits > 31 handling
  virtio-ccw: support ring size changes
  virtio: ring sizes vs. reset
  pc: Introduce pc-*-2.5 machine classes
  q35: Move options common to all classes to pc_i440fx_machine_options()
  q35: Move options common to all classes to pc_q35_machine_options()
  virtio-net: unbreak self announcement and guest offloads after migration
  virtio: right size for virtio_queue_get_avail_size

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

sheepdog: refine discard support

This patch refines discard support of the sheepdog driver. The
existing discard mechanism was implemented on SD_OP_DISCARD_OBJ, which
was introduced before fine grained reference counting on newer
sheepdog. It doesn't care about relations of snapshots and clones and
discards objects unconditionally.

With this patch, the driver just updates an inode object for updating
reference. Removing the object is done in sheep process side.

Cc: Teruaki Ishizaki <[email protected]>
Cc: Vasiliy Tolstov <[email protected]>
Cc: Jeff Cody <[email protected]>
Signed-off-by: Hitoshi Mitake <[email protected]>
Tested-by: Vasiliy Tolstov <[email protected]>
Message-id: 1441076590[email protected]
Signed-off-by: Jeff Cody <[email protected]>

sheepdog: use per AIOCB dirty indexes for non overlapping requests

In the commit 96b14ff85acf, requests for overlapping areas are
serialized. However, it cannot handle a case of non overlapping
requests. In such a case, min_dirty_data_idx and max_dirty_data_idx
can be overwritten by the requests and invalid inode update can
happen e.g. a case like create(1, 2) and create(3, 4) are issued in
parallel.

This patch lets SheepdogAIOCB have dirty data indexes instead of
BDRVSheepdogState for avoiding the above situation.

This patch also does trivial renaming for better description:
overwrapping -> overlapping

Cc: Teruaki Ishizaki <[email protected]>
Cc: Vasiliy Tolstov <[email protected]>
Cc: Jeff Cody <[email protected]>
Signed-off-by: Hitoshi Mitake <[email protected]>
Tested-by: Vasiliy Tolstov <[email protected]>
Message-id: 1441076590[email protected]
Signed-off-by: Jeff Cody <[email protected]>

net: remove muldiv64()

muldiv64() is used to convert nanoseconds to microseconds.

    x = muldiv64(qemu_clock_get_ns(..), 1000000, get_ticks_per_sec());

As  get_ticks_per_sec() is 10^9, it can be replaced by:

    x = qemu_clock_get_us(..);

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

bt: remove muldiv64()

Originally, timers were ticks based, and it made sense to
add ticks to current time to know when to trigger an alarm.

But since commit:

7447545 change all other clock references to use nanosecond resolution accessors

All timers use nanoseconds and we need to convert ticks to nanoseconds.

As get_ticks_per_sec() is 10^9,

    a = muldiv64(b, get_ticks_per_sec(), 100);
    y = muldiv64(x, get_ticks_per_sec(), 1000000);

can be converted to

    a = b * 10000000;
    y = x * 1000;

Signed-off-by: Laurent Vivier <[email protected]>
Reviewed-by: Paolo Bonzini <[email protected]>

hpet: remove muldiv64()

hpet defines a clock period in femtoseconds but
then converts it to nanoseconds to use the internal
timers.

We can define the period in nanoseconds and use it
directly, this allows to remove muldiv64().

We only need to convert the period to femtoseconds
to put it in internal hpet capability register.

Signed-off-by: Laurent Vivier <[email protected]>
Reviewed-by: Paolo Bonzini <[email protected]>

arm: clarify the use of muldiv64()

muldiv64() is used to convert microseconds into CPU ticks.

But it is not clear and not commented. This patch uses macro
to clearly identify what is used: time, CPU frequency and ticks.
For an elapsed time and a given frequency, we compute how many ticks
 we have.

Signed-off-by: Laurent Vivier <[email protected]>
Reviewed-by: Peter Crosthwaite <[email protected]>
Acked-by: Peter Maydell <[email protected]>

openrisc: remove muldiv64()

Originally, timers were ticks based, and it made sense to
add ticks to current time to know when to trigger an alarm.

But since commit:

7447545 change all other clock references to use nanosecond resolution accessors

All timers use nanoseconds and we need to convert ticks to nanoseconds, by
doing something like:

    y = muldiv64(x, get_ticks_per_sec(), TIMER_FREQ)

where x is the number of device ticks and y the number of system ticks.

y is used as nanoseconds in timer functions,
it works because 1 tick is 1 nanosecond.
(get_ticks_per_sec() is 10^9)

But as openrisc timer frequency is 20 MHz, we can also do:

    y = x * 50; /* 20 MHz period is 50 ns */

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

mips: remove muldiv64()

Originally, timers were ticks based, and it made sense to
add ticks to current time to know when to trigger an alarm.

But since commit:

7447545 change all other clock references to use nanosecond resolution accessors

All timers use nanoseconds and we need to convert ticks to nanoseconds, by
doing something like:

    y = muldiv64(x, get_ticks_per_sec(), TIMER_FREQ)

where x is the number of device ticks and y the number of system ticks.

y is used as nanoseconds in timer functions,
it works because 1 tick is 1 nanosecond.
(get_ticks_per_sec() is 10^9)

But as MIPS timer frequency is 100 MHz, we can also do:

    y = x * 10; /* 100 MHz period is 10 ns */

Signed-off-by: Laurent Vivier <[email protected]>
Reviewed-by: Leon Alrae <[email protected]>

pcnet: remove muldiv64()

Originally, timers were ticks based, and it made sense to
add ticks to current time to know when to trigger an alarm.

But since commit:

7447545 change all other clock references to use nanosecond resolution accessors

All timers use nanoseconds and we need to convert ticks to nanoseconds, by
doing something like:

    y = muldiv64(x, get_ticks_per_sec(), PCI_FREQUENCY)

where x is the number of device ticks and y the number of system ticks.

y is used as nanoseconds in timer functions,
it works because 1 tick is 1 nanosecond.
(get_ticks_per_sec() is 10^9)

But as PCI frequency is 33 MHz, we can also do:

    y = x * 30; /* 33 MHz PCI period is 30 ns */

Which is much more simple.

This implies a 33.333333 MHz PCI frequency,
but this is correct.

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

rtl8139: remove muldiv64()

Originally, timers were ticks based, and it made sense to
add ticks to current time to know when to trigger an alarm.

But since commit:

7447545 change all other clock references to use nanosecond resolution accessors

All timers use nanoseconds and we need to convert ticks to nanoseconds, by
doing something like:

    y = muldiv64(x, get_ticks_per_sec(), PCI_FREQUENCY)

where x is the number of device ticks and y the number of system ticks.

y is used as nanoseconds in timer functions,
it works because 1 tick is 1 nanosecond.
(get_ticks_per_sec() is 10^9)

But as PCI frequency is 33 MHz, we can also do:

    y = x * 30; /* 33 MHz PCI period is 30 ns */

Which is much more simple.

This implies a 33.333333 MHz PCI frequency,
but this is correct.

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

i6300esb: remove muldiv64()

Originally, timers were ticks based, and it made sense to
add ticks to current time to know when to trigger an alarm.

But since commit:

7447545 change all other clock references to use nanosecond resolution accessors

All timers use nanoseconds and we need to convert ticks to nanoseconds, by
doing something like:

    y = muldiv64(x, get_ticks_per_sec(), PCI_FREQUENCY)

where x is the number of device ticks and y the number of system ticks.

y is used as nanoseconds in timer functions,
it works because 1 tick is 1 nanosecond.
(get_ticks_per_sec() is 10^9)

But as PCI frequency is 33 MHz, we can also do:

    y = x * 30; /* 33 MHz PCI period is 30 ns */

Which is much more simple.

This implies a 33.333333 MHz PCI frequency,
but this is correct.

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

Backup: don't do copy-on-read in before_write_notifier

We will copy data in before_write_notifier to do backup.
It is a nested I/O request, so we cannot do copy-on-read.

The steps to reproduce it:
1. -drive copy-on-read=on,...  // qemu option
2. drive_backup -f disk0 /path_to_backup.img // monitor command

Signed-off-by: Wen Congyang <[email protected]>
Tested-by: Jeff Cody <[email protected]>
Message-id: 1441682913[email protected]
Signed-off-by: Jeff Cody <[email protected]>

block: Introduce a new API bdrv_co_no_copy_on_readv()

In some cases, we need to disable copy-on-read, and just
read the data.

Signed-off-by: Wen Congyang <[email protected]>
Message-id: 1441682913[email protected]
Signed-off-by: Jeff Cody <[email protected]>

sheepdog: add reopen support

With reopen supported, block-commit (and offline commit) is now supported for
image files whose base image uses the Sheepdog protocol driver.

Cc: [email protected]
Cc: Jeff Cody <[email protected]>
Cc: Kevin Wolf <[email protected]>
Cc: Stefan Hajnoczi <[email protected]>
Signed-off-by: Liu Yuan <[email protected]>
Message-id: 1440730438[email protected]
Signed-off-by: Jeff Cody <[email protected]>

block/nfs: cache allocated filesize for read-only files

If the file is readonly its not expected to grow so
save the blocking call to nfs_fstat_async and use
the value saved at connection time. Also important
the monitor (and thus the main loop) will not hang
if block device info is queried and the NFS share
is unresponsive.

Signed-off-by: Peter Lieven <[email protected]>
Reviewed-by: Jeff Cody <[email protected]>
Reviewed-by: Max Reitz <[email protected]>
Message-id: 1440671441[email protected]
Signed-off-by: Jeff Cody <[email protected]>

block/nfs: fix calculation of allocated file size

st.st_blocks is always counted in 512 byte units. Do not
use st.st_blksize as multiplicator which may be larger.

Cc: [email protected]
Signed-off-by: Peter Lieven <[email protected]>
Reviewed-by: Max Reitz <[email protected]>
Reviewed-by: Jeff Cody <[email protected]>
Message-id: 1440067607[email protected]
Signed-off-by: Jeff Cody <[email protected]>

doc: Refresh URLs in the qemu-tech documentation

The TwoOStwo and Willows page seem to have disappeared completely,
and also some of the other links were not pointing to the right
locations anymore.

Signed-off-by: Thomas Huth <[email protected]>
Message-Id: <1443173916[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

docs: describe the QEMU build system structure / design

Developers who are new to QEMU, or have a background familiarity
with GNU autotools, can have trouble getting their head around the
home-grown QEMU build system. This document attempts to explain
the structure / design of the configure script and the various
Makefile pieces that live across the source tree.

Signed-off-by: Daniel P. Berrange <[email protected]>
Message-Id: <1443102098[email protected]>
Acked-by: Laszlo Ersek <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

typedef: add typedef for QemuOpts

This patch moves typedefs for QemuOpts and related types
to qemu/typedefs.h file.

Reviewed-by: Paolo Bonzini <[email protected]>
Signed-off-by: Pavel Dovgalyuk <[email protected]>
Message-Id: <20150917162501[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

i386: interrupt poll processing

This patch updates x86_cpu_exec_interrupt function.
It can process two interrupt request at a time (poll and another one).
This makes its execution non-deterministic. Determinism is requred
for recorded icount execution.

Signed-off-by: Pavel Dovgalyuk <[email protected]>
Message-Id: <20150917162410[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

i386: partial revert of interrupt poll fix

Processing CPU_INTERRUPT_POLL requests in cpu_has_work functions
break the determinism of cpu_exec. This patch is required to make
interrupts processing deterministic.

Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Pavel Dovgalyuk <[email protected]>
Message-Id: <20150917162331[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

ppc: Rename ELF_MACHINE to be PPC specific

Rename ELF_MACHINE to be PPC specific. This is used as-is by the
various PPC bootloaders and is locally defined to ELF_MACHINE in linux
user in PPC specific ifdeffery.

This removes another architecture specific definition from the global
namespace (as desired by multi-arch).

Cc: Alexander Graf <[email protected]>
Cc: [email protected]
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

i386: Rename ELF_MACHINE to be x86 specific

Rename ELF_MACHINE to be I386 specific. This is used as-is by the
multiboot loader.

Linux-user previously used this definition but will not anymore,
falling back to the default bahaviour of using ELF_ARCH as ELF_MACHINE.

This removes another architecture specific definition from the global
namespace.

Cc: Paolo Bonzini <[email protected]>
Cc: Richard Henderson <[email protected]>
Cc: Eduardo Habkost <[email protected]>
Reviewed-by: Eduardo Habkost <[email protected]>
Acked-by: Eduardo Habkost <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

alpha: Remove ELF_MACHINE from cpu.h

ELF_MACHINE is unused by target alpha.

Cc: Richard Henderson <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

mips: Remove ELF_MACHINE from cpu.h

The only generic code relying on this is linux-user, but linux users'
default behaviour of defaulting ELF_MACHINE to ELF_ARCH will handle
this.

The bootloaders can just pass EM_MIPS directly, as that is
architecture specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Aurelien Jarno <[email protected]>
Cc: Leon Alrae <[email protected]>
Reviewed-by: Aurelien Jarno <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

sparc: Remove ELF_MACHINE from cpu.h

The bootloaders can just pass EM_SPARC or EM_SPARCV9 directly, as
they are architecture specific code (to one or the other).

This removes another architecture specific definition from the global
namespace.

Cc: Mark Cave-Ayland <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

s390: Remove ELF_MACHINE from cpu.h

The bootloader can just pass EM_S390 directly, as that
is architecture specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Richard Henderson <[email protected]>
Cc: Alexander Graf <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

sh4: Remove ELF_MACHINE from cpu.h

The only generic code relying on this is linux-user, but linux users'
default behaviour of defaulting ELF_MACHINE to ELF_ARCH will handle
this.

This removes another architecture specific definition from the global
namespace.

Cc: Aurelien Jarno <[email protected]>
Acked-by: Aurelien Jarno <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

xtensa: Remove ELF_MACHINE from cpu.h

The bootloaders can just pass EM_XTENSA directly, as that
is architecture specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Max Filippov <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

tricore: Remove ELF_MACHINE from cpu.h

The bootloader can just pass EM_TRICORE directly, as that
is architecture specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Bastian Koppelmann <[email protected]>
Acked-By: Bastian Koppelmann <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

or32: Remove ELF_MACHINE from cpu.h

The only generic code relying on this is linux-user, but linux users'
default behaviour of defaulting ELF_MACHINE to ELF_ARCH will handle
this.

The bootloader can just pass EM_OPENRISC directly, as that is
architecture specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Jia Liu <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

lm32: Remove ELF_MACHINE from cpu.h

The bootloaders can just pass EM_LATTICEMICO32 directly, as that is
architecture specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Michael Walle <[email protected]>
Acked-By: Michael Walle <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

unicore: Remove ELF_MACHINE from cpu.h

The only generic code relying on this is linux-user, but linux users'
default behaviour of defaulting ELF_MACHINE to ELF_ARCH will handle
this.

This removes another architecture specific definition from the global
namespace.

Cc: Guan Xuetao <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

moxie: Remove ELF_MACHINE from cpu.h

The bootloader can just pass EM_MOXIE directly, as that is architecture
specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Anthony Green <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

cris: Remove ELF_MACHINE from cpu.h

The only generic code relying on this is linux-user, but linux users'
default behaviour of defaulting ELF_MACHINE to ELF_ARCH will handle
this.

The bootloader can just pass EM_CRIS directly, as that is architecture
specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Edgar E. Iglesias <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

m68k: Remove ELF_MACHINE from cpu.h

The only generic code relying on this is linux-user, but linux users'
default behaviour of defaulting ELF_MACHINE to ELF_ARCH will handle
this.

The machine model bootloaders can just pass EM_68K directly, as that
is architecture specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Laurent Vivier <[email protected]>
Cc: Greg Ungerer <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Reviewed-by: Greg Ungerer <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

mb: Remove ELF_MACHINE from cpu.h

The only generic code relying on this is linux-user, but linux-users'
default behaviour or setting ELF_MACHINE to ELF_ARCH will handle this.

The microblaze bootloader can just pass EM_MICROBLAZE directly, as that
is architecture specific code.

This removes another architecture specific definition from the global
namespace.

Cc: Edgar E. Iglesias <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

arm: Remove ELF_MACHINE from cpu.h

The only generic code relying on this is linux-user. Linux user
already has a lot of #ifdef TARGET_ customisation so instead, define
ELF_ARCH as either EM_ARM or EM_AARCH64 appropriately.

The armv7m bootloader can just pass EM_ARM directly, as that
is architecture specific code. Note that arm_boot already has its own
logic selecting an arm specific elf machine so this makes V7M more
consistent with arm_boot.

This removes another architecture specific definition from the global
namespace.

Cc: Peter Maydell <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

elf: Update EM_MOXIE definition

EM_MOXIE now has a proper assigned elf code. Use it. Register the old
interim value as EM_MOXIE_OLD and accept either in elf loading.

Cc: Anthony Green <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

elf_ops: Fix coding style for EM alias case statement

Fix the coding style for these cases as per CODING_STYLE. Reverse the
Yoda conditions and add missing if braces.

Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

linux-user: elfload: Provide default for elf_check_arch

For many arch's this macro is defined as the predicatable behaviour
of checking the argument for eqaulity against ELF_ARCH. Provide a
default define as such, so only archs with special handling (usually
allowing multiple EM values) need to provide a def.

Arches that do any of:

1: provide this def exactly the same way as the new default
        (alpha, x86_64)
2: check against ELF_MACHINE while defining ELF_ARCH == ELF_MACHINE
        (arm, aarch64)
3: check against EM_FOO directly while defining ELF_ARCH == EM_FOO
        (unicore32, sparc32, ppc32, mips, openrisc, sh4, cris, m86k)

have their elf_check_arch removed as the default will provide the
correct behaviour.

Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

linux_user: elfload: Default ELF_MACHINE to ELF_ARCH

In most (but not all) cases, ELF_MACHINE and ELF_ARCH are safely the
same. Default ELF_MACHINE to ELF_ARCH. This makes defining ELF_MACHINE
optional for target-*/cpu.h when they are known to match.

Reviewed-by: Richard Henderson <[email protected]>
Acked-By: Riku Voipio <[email protected]>
Signed-off-by: Peter Crosthwaite <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

hmp: implemented io apic dump state for TCG

Added support emulator for the hmp command "info ioapic"

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

hmp: added io apic dump state

Added the hmp command to query io apic state, may be usefull after guest
crashes to understand IRQ routing in guest.

Implementation is only for kvm here. The dump will look like
(qemu) info ioapic
ioapic id=0x00 sel=0x26 (redir[11])
pin 0  0x0000000000010000 dest=0 vec=0   active-hi edge  masked fixed  physical
pin 1  0x0000000000000031 dest=0 vec=49  active-hi edge         fixed  physical
...
pin 23 0x0000000000010000 dest=0 vec=0   active-hi edge  masked fixed  physical
IRR        (none)
Remote IRR (none)

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

ioapic_internal.h: added more constants

Added the masks for easy  access to fields of the redirection table entry

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

hmp: added local apic dump state

Added the hmp command to query local apic registers state, may be
usefull after guest crashes to understand IRQ routing in guest.

(qemu) info lapic
dumping local APIC state for CPU 0

LVT0    0x00010700 active-hi edge  masked                      ExtINT (vec 0)
LVT1    0x00000400 active-hi edge                              NMI
LVTPC   0x00010000 active-hi edge  masked                      Fixed  (vec 0)
LVTERR  0x000000fe active-hi edge                              Fixed  (vec 254)
LVTTHMR 0x00010000 active-hi edge  masked                      Fixed  (vec 0)
LVTT    0x000000ef active-hi edge                 one-shot     Fixed  (vec 239)
Timer   DCR=0x3 (divide by 16) initial_count = 61360
SPIV    0x000001ff APIC enabled, focus=off, spurious vec 255
ICR     0x000000fd physical edge de-assert no-shorthand
ICR2    0x00000001 cpu 1 (X2APIC ID)
ESR     0x00000000
ISR     (none)
IRR     239

APR 0x00 TPR 0x00 DFR 0x0f LDR 0x00 PPR 0x00

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

monitor: make monitor_fprintf and mon_get_cpu externally visible

monitor_fprintf and mon_get_cpu will be used in the target-specific monitor,
so it is advisable to make it external.

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

apic_internal.h: fix formatting and drop unused consts

Fix formatting of local apic definitions and drop unused constant
APIC_INPUT_POLARITY, APIC_SEND_PENDING. Magic numbers in shifts are
replaced with constants defined just above.

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

apic_internal.h: added more constants

These constants are needed for optimal access to
bit fields local apic registers without magic numbers.

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

apic_internal.h: rename ESR_ILLEGAL_ADDRESS to APIC_ESR_ILLEGAL_ADDRESS

Added prefix APIC_ for determining the constant of a particular subsystem,
improve the overall readability and match other constant names.

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

apic_internal.h: make some apic_get_* functions externally visible

Move apic_get_bit(), apic_set_bit() to apic_internal.h, make the apic_get_ppr
symbol external. It's necessary to work with isr, tmr, irr and ppr outside
hw/intc/apic.c

Signed-off-by: Pavel Butsykin <[email protected]>
Signed-off-by: Denis V. Lunev <[email protected]>
CC: Paolo Bonzini <[email protected]>
CC: Andreas Färber <[email protected]>
Message-Id: <1442927901[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

ioapic: fix contents of arbitration register

The arbitration register should read to the same value as the
IOAPIC id register.  Fixes kvm-unit-tests ioapic.flat.

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

ioapic: coalesce level interrupts

If a level-triggered interrupt goes down and back up before the
corresponding EOI, it should be coalesced.  This fixes one testcase
in kvm-unit-tests' ioapic.flat.

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

MAINTAINERS: add maintainer for network device front-ends

Only "Odd Fixes" status, but let's add a point of contact.

Cc: Jason Wang <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

MAINTAINERS: add maintainer for character device front-ends

Only "Odd Fixes" status, but let's add a point of contact.

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

MAINTAINERS: add IPack section

Reviewed-by: Alberto Garcia <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

MAINTAINERS: Add more s390 files

Cc: Alexander Graf <[email protected]>
Reviewed-by: Christian Borntraeger <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

MAINTAINERS: Add disassemblers to the various backends

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

MAINTAINERS: there is no PPC64 TCG backend anymore

PPC32 and PPC64 were unified.

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

get_maintainer.pl: \C is deprecated

"Match a single C-language char (octet) even if that is part of a larger
UTF-8 character.  Thus it breaks up characters into their UTF-8 bytes,
so you may end up with malformed pieces of UTF-8."

Just use a period instead.

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

vhost-scsi: include linux/vhost.h

Replace ad-hoc declarations with the linux header.

Signed-off-by: Marc-André Lureau <[email protected]>
Message-Id: <1442585920[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

Makefile: fix build when VPATH is outside GIT tree

Steve Ellcey / Leon Alrae reported that QEMU fails to build when
the VPATH directory is outside of the GIT tree, and the system
emulators & tools build is disabled. eg

   cd ..
   mkdir build
   cd build
   ../qemu/configure --disable-system --disable-tools
   make
   (...)
   make[1]: *** No rule to make target `../qom/object.o', needed by `qemu-aarch64'. Stop.
   make: *** [subdir-aarch64-linux-user] Error 2

The problem is due to the fact that some sub directory deps
were listed against SOFTMMU_SUBDIR_RULES instead of SUBDIR_RULES,
so were only processed for system emulators, not user emalutors.

Signed-off-by: Daniel P. Berrange <[email protected]>
Message-Id: <1442570495[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

scsi-generic: let guests recognize readonly=on on passthrough devices

Passed-through SCSI devices can be opened with the readonly=on option.
When this happens, Linux filters away write commands so that the guest
cannot overwrite the contents of the device.

However, the guest does not know that the device is read-only, and
accepts writes.  The writes only fail later when the page cache is
flushed.

This patch modifies scsi-generic to modify the MODE SENSE data and
set the read-only bit in the device-specific parameters, so that
the guest OS treats the disk as write protected.

Reviewed-by: Fam Zheng <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

checkpatch: do not recommend qemu_strtok over strtok

If anything it should recommend strtok_r!

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

tests: add some qemu_strtosz() tests

While reading the function I decided to write some tests.

Signed-off-by: Marc-André Lureau <[email protected]>
Message-Id: <1442419377[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

utils: rename strtosz to use qemu prefix

Not only it makes sense, but it gets rid of checkpatch warning:
WARNING: consider using qemu_strtosz in preference to strtosz

Also remove get rid of tabs to please checkpatch.

Signed-off-by: Marc-André Lureau <[email protected]>
Message-Id: <1442419377[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

qemu-nbd: convert to use the QAPI SocketAddress object

The qemu-nbd program currently uses a QemuOpts objects
when setting up sockets. Switch it over to use the
QAPI SocketAddress objects instead.

Signed-off-by: Daniel P. Berrange <[email protected]>
Message-Id: <1442411543[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

nbd: convert to use the QAPI SocketAddress object

The nbd block driver currently uses a QemuOpts object
when setting up sockets. Switch it over to use the
QAPI SocketAddress object instead.

Signed-off-by: Daniel P. Berrange <[email protected]>
Message-Id: <1442411543[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

MAINTAINERS: add more devices to the PCI section

Cc: Michael S. Tsirkin <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

MAINTAINERS: add more devices to the PC section

For chipset devices, I can co-maintain it with Michael.

Cc: Michael S. Tsirkin <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

Merge remote-tracking branch 'remotes/weil/tags/pull-wxx-20150924' into staging

wxx patch queue

# gpg: Signature made Thu 24 Sep 2015 20:24:50 BST using RSA key ID 677450AD
# gpg: Good signature from "Stefan Weil <[email protected]>"
# gpg:                 aka "Stefan Weil <[email protected]>"
# gpg:                 aka "Stefan Weil <[email protected]>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 4923 6FEA 75C9 5D69 8EC2  B78A E08C 21D5 6774 50AD

* remotes/weil/tags/pull-wxx-20150924:
  oslib-win32: only provide localtime_r/gmtime_r if missing
  gtk: avoid redefining _WIN32_WINNT macro
  qemu-thread: add a fast path to the Win32 QemuEvent
  slirp: Fix non blocking connect for w32
  nsis: Add QEMU version information to Windows registry

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

oslib-win32: only provide localtime_r/gmtime_r if missing

The oslib-win32 file currently provides a localtime_r and
gmtime_r replacement unconditionally. Some versions of
Mingw-w64 would provide crude macros for localtime_r/gmtime_r
which QEMU takes care to disable. Latest versions of Mingw-w64
now provide actual functions for localtime_r/gmtime_r, but
with a twist that you have to include unistd.h or pthread.h
before including time.h.  By luck some files in QEMU have
such an include order, resulting in compile errors:

  CC    util/osdep.o
In file included from include/qemu-common.h:48:0,
                 from util/osdep.c:48:
include/sysemu/os-win32.h:77:12: error: redundant redeclaration of 'gmtime_r' [-Werror=redundant-decls]
 struct tm *gmtime_r(const time_t *timep, struct tm *result);
            ^
In file included from include/qemu-common.h:35:0,
                 from util/osdep.c:48:
/usr/i686-w64-mingw32/sys-root/mingw/include/time.h:272:107: note: previous definition of 'gmtime_r' was here
In file included from include/qemu-common.h:48:0,
                 from util/osdep.c:48:
include/sysemu/os-win32.h:79:12: error: redundant redeclaration of 'localtime_r' [-Werror=redundant-decls]
 struct tm *localtime_r(const time_t *timep, struct tm *result);
            ^
In file included from include/qemu-common.h:35:0,
                 from util/osdep.c:48:
/usr/i686-w64-mingw32/sys-root/mingw/include/time.h:269:107: note: previous definition of 'localtime_r' was here

This change adds a configure test to see if localtime_r
exits, and only enables the QEMU impl if missing. We also
re-arrange qemu-common.h try attempt to guarantee that all
source files get unistd.h before time.h and thus see the
localtime_r/gmtime_r defs.

[sw: Use "official" spellings for Mingw-w64, MinGW in comments.]
[sw: Terminate sentences with a dot in comments.]

Signed-off-by: Daniel P. Berrange <[email protected]>
Reviewed-by: Denis V. Lunev <[email protected]>
Signed-off-by: Stefan Weil <[email protected]>

gtk: avoid redefining _WIN32_WINNT macro

When building for Mingw64 target on Fedora 22 a warning
is issued about _WIN32_WINNT being redefined.

In file included from ui/gtk.c:40:0:
include/ui/gtk.h:5:0: warning: "_WIN32_WINNT" redefined
 # define _WIN32_WINNT 0x0601 /* needed to get definition of MAPVK_VK_TO_VSC */
  ^
In file included from /usr/i686-w64-mingw32/sys-root/mingw/include/crtdefs.h:10:0,
                 from /usr/i686-w64-mingw32/sys-root/mingw/include/stdio.h:9,
                 from /home/berrange/src/virt/qemu/include/qemu/fprintf-fn.h:12,
                 from /home/berrange/src/virt/qemu/include/qemu-common.h:18,
                 from ui/gtk.c:37:
/usr/i686-w64-mingw32/sys-root/mingw/include/_mingw.h:225:0: note: this is the location of the previous definition
 #define _WIN32_WINNT 0x502
 ^

Rather than try to get MAPVK_VK_TO_VSC defined indirectly
by defining _WIN32_WINNT, instead just define it explicitly
if missing.

Signed-off-by: Daniel P. Berrange <[email protected]>
Signed-off-by: Stefan Weil <[email protected]>
Acked-by: Paolo Bonzini <[email protected]>

qemu-thread: add a fast path to the Win32 QemuEvent

QemuEvents are used heavily by call_rcu.  We do not want them to be slow,
but the current implementation does a kernel call on every invocation
of qemu_event_* and won't cut it.

So, wrap a Win32 manual-reset event with a fast userspace path.  The
states and transitions are the same as for the futex and mutex/condvar
implementations, but the slow path is different of course.  The idea
is to reset the Win32 event lazily, as part of a test-reset-test-wait
sequence.  Such a sequence is, indeed, how QemuEvents are used by
RCU and other subsystems!

The patch includes a formal model of the algorithm.

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

slirp: Fix non blocking connect for w32

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

nsis: Add QEMU version information to Windows registry

The uninstall keys include an option key "DisplayVersion" which we set
now. By default the version value is read from file VERSION, but it is
also possible to pass VERSION=#.#.# to make.

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

Merge remote-tracking branch 'remotes/elmarco/tags/rm-libcacard' into staging

Remove libcacard

# gpg: Signature made Wed 23 Sep 2015 22:37:11 BST using RSA key ID 75969CE5
# gpg: Good signature from "Marc-André Lureau <[email protected]>"
# gpg:                 aka "Marc-André Lureau <[email protected]>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 87A9 BD93 3F87 C606 D276  F62D DAE8 E109 7596 9CE5

* remotes/elmarco/tags/rm-libcacard:
  libcacard: use the standalone project

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

vhost-user: add a new message to disable/enable a specific virt queue.

Add a new message, VHOST_USER_SET_VRING_ENABLE, to enable or disable
a specific virt queue, which is similar to attach/detach queue for
tap device.

virtio driver on guest doesn't have to use max virt queue pair, it
could enable any number of virt queue ranging from 1 to max virt
queue pair.

Signed-off-by: Changchun Ouyang <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Jason Wang <[email protected]>
Tested-by: Marcel Apfelbaum <[email protected]>

vhost-user: add multiple queue support

This patch is initially based a patch from Nikolay Nikolaev.

This patch adds vhost-user multiple queue support, by creating a nc
and vhost_net pair for each queue.

Qemu exits if find that the backend can't support the number of requested
queues (by providing queues=# option). The max number is queried by a
new message, VHOST_USER_GET_QUEUE_NUM, and is sent only when protocol
feature VHOST_USER_PROTOCOL_F_MQ is present first.

The max queue check is done at vhost-user initiation stage. We initiate
one queue first, which, in the meantime, also gets the max_queues the
backend supports.

In older version, it was reported that some messages are sent more times
than necessary. Here we came an agreement with Michael that we could
categorize vhost user messages to 2 types: non-vring specific messages,
which should be sent only once, and vring specific messages, which should
be sent per queue.

Here I introduced a helper function vhost_user_one_time_request(), which
lists following messages as non-vring specific messages:

        VHOST_USER_SET_OWNER
        VHOST_USER_RESET_DEVICE
        VHOST_USER_SET_MEM_TABLE
        VHOST_USER_GET_QUEUE_NUM

For above messages, we simply ignore them when they are not sent the first
time.

Signed-off-by: Nikolay Nikolaev <[email protected]>
Signed-off-by: Changchun Ouyang <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Jason Wang <[email protected]>
Tested-by: Marcel Apfelbaum <[email protected]>

vhost: introduce vhost_backend_get_vq_index method

Minusing the idx with the base(dev->vq_index) for vhost-kernel, and
then adding it back for vhost-user doesn't seem right. Here introduces
a new method vhost_backend_get_vq_index() for getting the right vq
index for following vhost messages calls.

Suggested-by: Jason Wang <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Jason Wang <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Tested-by: Marcel Apfelbaum <[email protected]>

vhost-user: add VHOST_USER_GET_QUEUE_NUM message

This is for querying how many queues the backend supports if it has mq
support(when VHOST_USER_PROTOCOL_F_MQ flag is set from the quried
protocol features).

vhost_net_get_max_queues() is the interface to export that value, and
to tell if the backend supports # of queues user requested, which is
done in the following patch.

Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Tested-by: Marcel Apfelbaum <[email protected]>

vhost: rename VHOST_RESET_OWNER to VHOST_RESET_DEVICE

Quote from Michael:

    We really should rename VHOST_RESET_OWNER to VHOST_RESET_DEVICE.

Suggested-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Marcel Apfelbaum <[email protected]>
Tested-by: Marcel Apfelbaum <[email protected]>

vhost-user: add protocol feature negotiation

Support a separate bitmask for vhost-user protocol features,
and messages to get/set protocol features.

Invoke them at init.

No features are defined yet.

[ leverage vhost_user_call for request handling -- Yuanhan Liu ]

Signed-off-by: Michael S. Tsirkin <address@hidden>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Marcel Apfelbaum <[email protected]>
Tested-by: Marcel Apfelbaum <[email protected]>

vhost-user: use VHOST_USER_XXX macro for switch statement

So that we could let vhost_user_call to handle extented requests,
such as VHOST_USER_GET/SET_PROTOCOL_FEATURES, instead of invoking
vhost_user_read/write and constructing the msg again by ourself.

Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Reviewed-by: Marcel Apfelbaum <[email protected]>
Tested-by: Marcel Apfelbaum <[email protected]>

virtio-ccw: enable virtio-1

Let's enable revision 1 for virtio-ccw devices. We can always offer
VERSION_1 as drivers in legacy mode won't be able to see it anyway.

We have to introduce a way to set a lower maximum revision for a device
to accommodate the following cases:
- compat machines (to enforce legacy only)
- virtio-blk with scsi support (version 1 + scsi is fenced by common
  code, with a user-configured max revision of 0 we can allow scsi
  via not offering VERSION_1)

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

virtio-ccw: feature bits > 31 handling

We currently switch off the VERSION_1 feature bit if the guest has
not negotiated at least revision 1. As no feature bits beyond 31 are
valid however unless VERSION_1 has been negotiated, make sure that
legacy guests never see a feature bit beyond 31.

Reviewed-by: David Hildenbrand <[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

virtio-ccw: support ring size changes

Wire up changing the ring size for virtio-1 devices.

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

virtio: ring sizes vs. reset

We allow guests to change the size of the virtqueue rings by supplying
a number of buffers that is different from the number of buffers the
device was initialized with. Current code has some problems, however,
since reset does not reset the ringsizes to the default values (as this
is not saved anywhere).

Let's extend the core code to keep track of the default ringsizes and
migrate them once the guest changed them for any of the virtqueues
for a device.

Reviewed-by: Jason Wang <[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

pc: Introduce pc-*-2.5 machine classes

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