Merge remote-tracking branch 'afaerber/qom-cpu.v5' into staging

* afaerber/qom-cpu.v5: (43 commits)
  qom: Introduce CPU class
  Rename CPUState -> CPUArchState
  xtensa hw/: Don't use CPUState
  sparc hw/: Don't use CPUState
  sh4 hw/: Don't use CPUState
  s390x hw/: Don't use CPUState
  ppc hw/: Don't use CPUState
  mips hw/: Don't use CPUState
  microblaze hw/: Don't use CPUState
  m68k hw/: Don't use CPUState
  lm32 hw/: Don't use CPUState
  i386 hw/: Don't use CPUState
  cris hw/: Don't use CPUState
  arm hw/: Don't use CPUState
  alpha hw/: Don't use CPUState
  xtensa-semi: Don't use CPUState
  m68k-semi: Don't use CPUState
  arm-semi: Don't use CPUState
  target-xtensa: Don't overuse CPUState
  target-unicore32: Don't overuse CPUState
  ...

vnc: Limit r/w access to size of allocated memory

This fixes memory reads and writes which exceeded the upper limit
of allocated memory vd->guest.ds->data and vd->server->data.

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

vnc: don't mess up with iohandlers in the vnc thread

The threaded VNC servers messed up with QEMU fd handlers without
any kind of locking, and that can cause some nasty race conditions.

Using qemu_mutex_lock_iothread() won't work because vnc_dpy_cpy(),
which will wait for the current job queue to finish, can be called with
the iothread lock held.

Instead, we now store the data in a temporary buffer, and use a bottom
half to notify the main thread that new data is available.

vnc_[un]lock_ouput() is still needed to access VncState members like
abort, csock or jobs_buffer.

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

qom: Introduce CPU class

Reintroduce CPUState as QOM object: It's abstract and derived directly
from TYPE_OBJECT for compatibility with the user emulators.
The identifier CPUState avoids conflicts between CPU() and the struct.

Introduce $(qom-twice-y) to build it separately for system and for user
emulators.

Prepare a virtual reset method, (re)introduce cpu_reset() as wrapper.

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

Rename CPUState -> CPUArchState

Scripted conversion:
  for file in *.[hc] hw/*.[hc] hw/kvm/*.[hc] linux-user/*.[hc] linux-user/m68k/*.[hc] bsd-user/*.[hc] darwin-user/*.[hc] tcg/*/*.[hc] target-*/cpu.h; do
    sed -i "s/CPUState/CPUArchState/g" $file
  done

All occurrences of CPUArchState are expected to be replaced by QOM CPUState,
once all targets are QOM'ified and common fields have been extracted.

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

xtensa hw/: Don't use CPUState

Scripted conversion:
  for file in hw/xtensa_*.[hc]; do
    sed -i "s/CPUState/CPUXtensaState/g" $file
  done

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

sparc hw/: Don't use CPUState

Scripted conversion:
  for file in hw/sun4m.c hw/sun4u.c hw/grlib.h hw/leon3.c; do
    sed -i "s/CPUState/CPUSPARCState/g" $file
  done

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

sh4 hw/: Don't use CPUState

Scripted conversion:
  for file in hw/sh.h hw/shix.c hw/r2d.c; do
    sed -i "s/CPUState/CPUSH4State/g" $file
  done

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

s390x hw/: Don't use CPUState

Scripted conversion:
  for file in hw/s390-*.[hc]; do
    sed -i "s/CPUState/CPUS390XState/g" $file
  done

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

ppc hw/: Don't use CPUState

Scripted conversion:
  for file in hw/ppc*.[hc] hw/mpc8544_guts.c hw/spapr*.[hc] hw/virtex_ml507.c hw/xics.c; do
    sed -i "s/CPUState/CPUPPCState/g" $file
  done

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

mips hw/: Don't use CPUState

Scripted conversion:
  for file in hw/mips_*.[hc]; do
    sed -i "s/CPUState/CPUMIPSState/g" $file
  done

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

microblaze hw/: Don't use CPUState

Scripted conversion:
  for file in hw/microblaze_*.[hc] hw/petalogix_ml605_mmu.c hw/petalogix_s3adsp1800_mmu.c; do
    sed -i "s/CPUState/CPUMBState/g" $file
  done

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

m68k hw/: Don't use CPUState

Scripted conversion:
  for file in hw/an5206.c hw/dummy_m68k.c hw/mcf.h hw/mcf5206.c hw/mcf5208.c hw/mcf_intc.c; do
    sed -i "s/CPUState/CPUM68KState/g" $file
  done

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

lm32 hw/: Don't use CPUState

Scripted conversion:
  for file in hw/lm32_boards.c hw/milkymist.c; do
    sed -i "s/CPUState/CPULM32State/g" $file
  done

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

i386 hw/: Don't use CPUState

Scripted conversion:
  for file in hw/apic.h hw/kvm/apic.c hw/kvmvapic.c hw/pc.c hw/vmport.c hw/xen_machine_pv.c; do
    sed -i "s/CPUState/CPUX86State/g" $file
  done

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

cris hw/: Don't use CPUState

Scripted conversion:
  for file in hw/cris-boot.[hc] hw/cris_pic_cpu.c hw/axis_dev88.c hw/etraxfs.h hw/etraxfs_ser.c; do
    sed -i "s/CPUState/CPUCRISState/g" $file
  done

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

arm hw/: Don't use CPUState

Scripted conversion:
  for file in hw/arm-misc.h hw/arm_boot.c hw/arm_pic.c hw/armv7m.c hw/exynos4210.h hw/highbank.c hw/integratorcp.c hw/musicpal.c hw/omap.h hw/pxa.h hw/pxa2xx_gpio.c hw/pxa2xx_pic.c hw/realview.c hw/strongarm.h hw/versatilepb.c hw/vexpress.c  hw/xilinx_zynq.c ; do
    sed -i "s/CPUState/CPUARMState/g" $file
  done

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

alpha hw/: Don't use CPUState

Scripted conversion:
  for file in hw/alpha_*.[hc]; do
    sed -i "s/CPUState/CPUAlphaState/g" $file
  done

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

xtensa-semi: Don't use CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUXtensaState/g" xtensa-semi.c

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

m68k-semi: Don't use CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUM68KState/g" m68k-semi.c

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

arm-semi: Don't use CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUARMState/g" arm-semi.c

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

target-xtensa: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUXtensaState/g" target-xtensa/*.[hc]
  sed -i "s/#define CPUXtensaState/#define CPUState/" target-xtensa/cpu.h

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

target-unicore32: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUUniCore32State/g" target-unicore32/*.[hc]
  sed -i "s/#define CPUUniCore32State/#define CPUState/" target-unicore32/cpu.h

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

target-sparc: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUSPARCState/g" target-sparc/*.[hc]
  sed -i "s/#define CPUSPARCState/#define CPUState/" target-sparc/cpu.h

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

target-sh4: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUSH4State/g" target-sh4/*.[hc]
  sed -i "s/#define CPUSH4State/#define CPUState/" target-sh4/cpu.h

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

target-s390x: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUS390XState/g" target-s390x/*.[hc]
  sed -i "s/#define CPUS390XState/#define CPUState/" target-s390x/cpu.h

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

target-ppc: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUPPCState/g" target-ppc/*.[hc]
  sed -i "s/#define CPUPPCState/#define CPUState/" target-ppc/cpu.h

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

target-mips: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUMIPSState/g" target-mips/*.[hc]
  sed -i "s/#define CPUMIPSState/#define CPUState/" target-mips/cpu.h

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

target-microblaze: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUMBState/g" target-microblaze/*.[hc]
  sed -i "s/#define CPUMBState/#define CPUState/" target-microblaze/cpu.h

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

target-m68k: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUM68KState/g" target-m68k/*.[hc]
  sed -i "s/#define CPUM68KState/#define CPUState/" target-m68k/cpu.h

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

target-lm32: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPULM32State/g" target-lm32/*.[hc]
  sed -i "s/#define CPULM32State/#define CPUState/" target-lm32/cpu.h

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

target-i386: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUX86State/g" target-i386/*.[hc]
  sed -i "s/#define CPUX86State/#define CPUState/" target-i386/cpu.h

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

target-cris: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUCRISState/g" target-cris/*.[hc]
  sed -i "s/#define CPUCRISState/#define CPUState/" target-cris/cpu.h

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

target-arm: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUARMState/g" target-arm/*.[hc]
  sed -i "s/#define CPUARMState/#define CPUState/" target-arm/cpu.h

Signed-off-by: Andreas Färber <[email protected]>
Acked-by: Anthony Liguori <[email protected]>
Acked-by: Peter Maydell <[email protected]>

target-alpha: Don't overuse CPUState

Scripted conversion:
  sed -i "s/CPUState/CPUAlphaState/g" target-alpha/*.[hc]
  sed -i "s/#define CPUAlphaState/#define CPUState/" target-alpha/cpu.h

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

bsd-user: Don't overuse CPUState

Use CPU*State where applicable.

Signed-off-by: Andreas Färber <[email protected]>

darwin-user: Don't overuse CPUState

Use CPU*State where applicable.

Signed-off-by: Andreas Färber <[email protected]>

linux-user: Don't overuse CPUState

In target-specific code use CPU*State.

While at it, fix indentation on those lines.

Signed-off-by: Andreas Färber <[email protected]>

hw/mc146818: Drop unneeded #includes

pc.h and apic.h are not needed; apic.h would drag in x86 CPUState and
is now included directly for TARGET_I386.

isa.h is already #included from mc146818rtc.h.

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

target-unicore32: Rename to CPUUniCore32State

This aids in refactoring CPUState by adopting the common naming scheme.

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

target-sparc: Typedef struct CPUSPARCState early

Will be needed for qemu_irq_ack callback.

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

target-lm32/microblaze: Typedef struct CPU{MB,LM32}State

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

monitor: Avoid CPUState in read/write functions

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

monitor: Don't access registers through CPUState

Use CPUX86State etc. instead (hand-converted).

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

Rename cpu_reset() to cpu_state_reset()

Frees the identifier cpu_reset for QOM CPUs (manual rename).

Don't hide the parameter type behind explicit casts, use static
functions with strongly typed argument to indirect.

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

PPC: 405: Use proper CPU reset

On ppc405ep there is a register that allows for software to reset the
core, but not the whole system. Implement this reset using a reset
interrupt.

This gets rid of a bunch of #if 0'ed code.

Reported-by: Andreas Färber <[email protected]>
Signed-off-by: Alexander Graf <[email protected]>
Signed-off-by: Andreas Färber <[email protected]>

qom: Introduce object_class_get_list()

This function allows to obtain a singly-linked list of classes, which
can be sorted by the caller.

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

qom/object.c: rename type_class_init() to type_initialize()

Function name type_class_init() gave us a wrong impression of separation
of type's "class" and "object" entities initialization. Name type_initialize()
is more appropriate for type_class_init() function (considering what operations
it performs).

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

qom: if @instance_size==0, assign size of object to parent object size

QOM documentation states that for objects of type with @instance_size == 0 size
will be assigned to match parent object's size. But currently this feauture is
not implemented and qemu asserts during creation of object with zero instance_size.

Set appropriate value for type instance_size during type_class_init() call.
object_initialize_with_type() must call type_class_init() before asserting
type->instance_size, and object_new_with_type() must call type_class_init() before
object allocation.

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

kvmvapic: align start address as well as size

The kvmvapic code remaps a section of ROM as RAM to allow the guest to
maintain state there.  It is careful to align the section size to a page
boundary, to avoid creating subpages, but neglects to do the same for
the start address.  These leads to an assert later on when the memory
core tries to create a page which is half RAM and half ROM.

Fix by aligning the start address to a page boundary.

This can be triggered by running qemu-system-x86_64 -enable-kvm -vga none.

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

Merge remote-tracking branch 'kiszka/queues/slirp' into staging

* kiszka/queues/slirp:
  slirp: Fix compiler warning for w64
  slirp: Cleanup resources on instance removal
  slirp: Remove unneeded if_queued
  slirp: Fix queue walking in if_start
  slirp: Prevent recursion of if_start
  slirp: Keep next_m always valid

Merge remote-tracking branch 'kraxel/usb.44' into staging

* kraxel/usb.44:
  Endian fix an assertion in usb-msd
  uhci: alloc can't fail, drop check.
  uhci: new uhci_handle_td return code for tds still in flight
  uhci: renumber uhci_handle_td return codes
  uhci: use enum for uhci_handle_td return codes
  uhci: tracing support
  uhci: cancel on schedule stop.
  uhci: fix uhci_async_cancel_all
  uhci: pass addr to uhci_async_alloc
  usb: improve packet state sanity checks
  usb-ohci: DMA writeback bug fixes
  usb-ehci: drop unused isoch_pause variable
  usb: zap hw/ush-{ohic,uhci}.h + init wrappers
  usb: the big rename

qom: Add QOM support to user emulators

Link the Object base class and the module infrastructure for class
registration. Introduce $(universal-obj-y) for objects that are more
common than $(common-obj-y), so that those only get built once.

Call QOM module init for type registration.

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

qom: Register QOM infrastructure early

The constructors for QOM TYPE_INTERFACE were executed rather late in
vl.c's main(). Call them very early so that QOM can safely be used for
machines and CPUs.

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

kvmclock: Always register type

Currently, the "kvmclock" type is only registered when kvm_enabled().

This breaks when moving type registration to before command line
parsing (so that QOM types can be used for CPU and machine).

Since the QOM classes are lazy-initialized anyway and kvmclock_create()
has another kvm_enabled() check, simply drop the KVM check in
kvmclock_register_types().

kvm-i8259, kvm-apic and kvm-ioapic do not suffer from such a check.

Reviewed-by: please.
Signed-off-by: Andreas Färber <[email protected]>
Reviewed-by: Jan Kiszka <[email protected]>
Cc: Marcelo Tosatti <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

slirp: Fix compiler warning for w64

Casting a pointer to an integer value must use uintptr_t or intptr_t
(not long) for portable code. MinGW-w64 requires this because
sizeof(long) != sizeof(void *) for w64 hosts, so casting to long
raises a compiler warning.

I use uintptr_t instead of intptr_t because changing the sign does not
matter here and casting pointers to unsigned values seems more
reasonable (the unsigned value is a non negative offset.

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

slirp: Cleanup resources on instance removal

Close & free sockets when shutting down a slirp instance, also release
all buffers.

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

slirp: Remove unneeded if_queued

There is now a trivial check on entry of if_start for pending packets,
so we can drop the additional tracking via if_queued.

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

slirp: Fix queue walking in if_start

Another attempt to get this right: We need to carefully walk both the
fastq and the batchq in if_start while trying to send packets to
possibly not yet resolved hosts on the virtual network.

So far we just requeued a delayed packet where it was and then started
walking the queues from the top again - that couldn't work. Now we pre-
calculate the next packet in the queue so that the current one can
safely be removed if it was sent successfully. We also need to take into
account that the next packet can be from the same session if the current
one was sent and there are no other sessions.

CC: Fabien Chouteau <[email protected]>
CC: Zhi Yong Wu <[email protected]>
CC: Stefan Weil <[email protected]>
Tested-by: Stefan Weil <[email protected]>
Signed-off-by: Jan Kiszka <[email protected]>

slirp: Prevent recursion of if_start

if_start can be called recursively via if_encap. Avoid this as our
scheme of dequeuing packets is not compatible with this.

CC: Fabien Chouteau <[email protected]>
CC: Zhi Yong Wu <[email protected]>
CC: Stefan Weil <[email protected]>
Signed-off-by: Jan Kiszka <[email protected]>

slirp: Keep next_m always valid

Make sure that next_m always points to a packet if batchq is non-empty.
This will simplify walking the queues in if_start.

CC: Fabien Chouteau <[email protected]>
CC: Zhi Yong Wu <[email protected]>
CC: Stefan Weil <[email protected]>
Signed-off-by: Jan Kiszka <[email protected]>

Endian fix an assertion in usb-msd

This fixes a broken endian assumption in an assertion in usb-msd.

Cc: Gerd Hoffman <[email protected]>
Signed-off-by: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: David Gibson <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

uhci: alloc can't fail, drop check.

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

uhci: new uhci_handle_td return code for tds still in flight

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

uhci: renumber uhci_handle_td return codes

Step #2 (separate for better bisectability): renumber so the silly '-1'
goes away.  Pick a range which doesn't overlap the old values.

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

uhci: use enum for uhci_handle_td return codes

Step #1 (separate for better bisectability): replace numbers with names.

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

uhci: tracing support

Zap DPRINTF, add tracepoints instead.

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

uhci: cancel on schedule stop.

Cancel any in-flight transaction when the guest stops the uhci schedule.

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

uhci: fix uhci_async_cancel_all

It should also free all queues.

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

uhci: pass addr to uhci_async_alloc

Also do async->td initialization in uhci_async_alloc now.
Prepares for adding tracepoints.

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

usb: improve packet state sanity checks

Add a new function to check whenever the packet state is as expected,
log more informations in case it isn't.

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

usb-ohci: DMA writeback bug fixes

This patch fixes two bugs in the OHCI device where the device writes
back data to system memory that should be exclusively under the
control of the guest side driver.

In OHCI specification Section 5.2.7, it mentioned "In all cases, Host
Controller Driver is responsible for the insertion and removal of all
Endpoint Descriptors in the various Host Controller Endpoint
Descriptor lists".  In the ohci_frame_boundary(), ohci_put_hcca()
writes the entire hcca back including the interrupt ED lists which
should be under driver control. This violates the specification and
can race with a host driver updating that list at the same time.

In the OHCI Spec Section 4.6, Transfer Descriptor Queue Processing, it
mentioned "Since the TD pointed to by TailP is not accessed by the HC,
the Host Controller Driver can initialize that TD and link at least
one other to it without creating a coherency or synchronization
problem".  While the function ohci_put_ed() writes the entire endpoint
descriptor back including the TailP which should under driver
control. This violate the specification and can race with a host
driver updating the TD list at the same time.

In each case the solution is to make sure we don't write data which is
under driver control.

Cc: Gerd Hoffman <[email protected]>
Signed-off-by: Wei Yang <[email protected]>
Signed-off-by: David Gibson <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

usb-ehci: drop unused isoch_pause variable

Signed-off-by: Hans de Goede <[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

usb: zap hw/ush-{ohic,uhci}.h + init wrappers

Remove the uhci and ohci init wrappers, which all wrapped a
pci_create_simple() one-liner.  Switch callsites to call
pci_create_simple directly.  Remove the header files where
the wrappers where declared.

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

usb: the big rename

Reorganize usb source files.  Create a new hw/usb/ directory and move
all usb source code to that place.  Also make filenames a bit more
descriptive.  Host adapters are prefixed with "hch-" now, usb device
emulations are prefixed with "dev-".  Fixup paths Makefile and include
paths to make it compile.  No code changes.

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

Merge remote-tracking branch 'mdroth/qga-pull-3-12-2012' into staging

* mdroth/qga-pull-3-12-2012:
  qemu-ga: add guest-sync-delimited
  qemu-ga: add guest-network-get-interfaces command
  qemu-ga: add win32 guest-suspend-ram command
  qemu-ga: add win32 guest-suspend-disk command.
  qemu-ga: add guest-suspend-hybrid
  qemu-ga: add guest-suspend-ram
  qemu-ga: add guest-suspend-disk

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

* kwolf/for-anthony:
  test-coroutine: add performance test for nesting
  coroutine: adding configure option for sigaltstack coroutine backend
  coroutine: adding configure choose mechanism for coroutine backend
  coroutine: adding sigaltstack method (.c source)
  qcow2: Reduce number of I/O requests
  qcow2: Add qcow2_alloc_clusters_at()
  qcow2: Factor out count_cow_clusters
  qmp: convert blockdev-snapshot-sync to a wrapper around transactions
  add mode field to blockdev-snapshot-sync transaction item
  rename blockdev-group-snapshot-sync
  qapi: complete implementation of unions
  use QSIMPLEQ_FOREACH_SAFE when freeing list elements
  Add 'make check-block'
  make check: Add qemu-iotests subset
  qemu-iotests: Mark some tests as quick
  qcow2: Add error messages in qcow2_truncate
  block: handle -EBUSY in bdrv_commit_all()
  qcow2: Add some tracing
  qed: do not evict in-use L2 table cache entries
  Group snapshot: Fix format name for backing file

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

* qemu-kvm/memory/core:
  memory: get rid of cpu_register_io_memory()
  memory: dispatch directly via MemoryRegion
  exec: fix code tlb entry misused as iotlb in get_page_addr_code()
  memory: store section indices in iotlb instead of io indices
  memory: make phys_page_find() return an unadjusted section

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

* stefanha/tracing:
  vga: add trace event for ppm_save
  console: add some trace events
  maintainers: Add docs/tracing.txt to Tracing
  docs: correct ./configure line in tracing.txt
  trace: make trace_thread_create() use its function arg
  tracetool: Omit useless QEMU_*_ENABLED() check
  trace: Provide a per-event status define for conditional compilation

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

* qmp/queue/qmp:
  qapi-schema.json: fix comment for type ObjectPropretyInfo
  qapi-schema: fix typos and explain 'spice' auth
  qjson.h: include compiler.h for GCC_FMT_ATTR

qemu-ga: add guest-sync-delimited

guest-sync leaves it as an exercise to the user as to how to reliably
obtain the response to guest-sync if the client had previously read in a
partial response (due qemu-ga previously being restarted mid-"sentence"
due to reboot, forced restart, etc).

qemu-ga handles this situation on its end by having a client precede
their guest-sync request with a 0xFF byte (invalid UTF-8), which
qemu-ga/QEMU JSON parsers will treat as a flush event. Thus we can
reliably flush the qemu-ga parser state in preparation for receiving
the guest-sync request.

guest-sync-delimited provides the same functionality for a client: when
a guest-sync-delimited is issued, qemu-ga will precede it's response
with a 0xFF byte that the client can use as an indicator to flush its
buffer/parser state in preparation for reliably receiving the
guest-sync-delimited response.

It is also useful as an optimization for clients, since, after issuing a
guest-sync-delimited, clients can safely discard all stale data read
from the channel until the 0xFF is found.

More information available on the wiki:

http://wiki.qemu.org/Features/QAPI/GuestAgent#QEMU_Guest_Agent_Protocol

Signed-off-by: Michael Roth <[email protected]>

qemu-ga: add guest-network-get-interfaces command

This command returns an array of:

 [ifname, hwaddr, [ipaddr, ipaddr_family, prefix] ]

for each interface in the system.
Currently, only IPv4 and IPv6 are supported.

Signed-off-by: Michal Privoznik <[email protected]>

qemu-ga: add win32 guest-suspend-ram command

S3 sleep implementation for windows.

qemu-ga: add win32 guest-suspend-disk command.

Implement guest-suspend-disk RPC for Windows. Functionally this should be
equivalent to the posix implementation.

Signed-off-by: Gal Hammer <[email protected]>

qemu-ga: add guest-suspend-hybrid

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

qemu-ga: add guest-suspend-ram

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

qemu-ga: add guest-suspend-disk

As the command name implies, this command suspends the guest to disk.

The suspend operation is implemented by two functions: bios_supports_mode()
and guest_suspend(). Both functions are generic enough to be used by
other suspend modes (introduced by next commits).

Both functions will try to use the scripts provided by the pm-utils
package if it's available. If it's not available, a manual method,
which consists of directly writing to '/sys/power/state', will be used.

To reap terminated children, a new signal handler is installed in the
parent to catch SIGCHLD signals and a non-blocking call to waitpid()
is done to collect their exit statuses. The statuses, however, are
discarded.

The approach used to query the guest for suspend support deserves some
explanation. It's implemented by bios_supports_mode() and shown below:

  qemu-ga
     |
 create pipe
     |
   fork()
     -----------------
     |               |
     |               |
     |             fork()
     |               --------------------------
     |               |                        |
     |               |                        |
     |               |               exec('pm-is-supported')
     |               |
     |              wait()
     |       write exit status to pipe
     |              exit
     |
  read pipe

This might look complex, but the resulting code is quite simple.
The purpose of that approach is to allow qemu-ga to reap its children
(semi-)automatically from its SIGCHLD handler.

Implementing this the obvious way, that's, doing the exec() call from
the first child process, would force us to introduce a more complex way
to reap qemu-ga's children. Like registering PIDs to be reaped and
having a way to wait for them when returning their exit status to
qemu-ga is necessary. The approach explained above avoids that complexity.

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

cpu-all.h: Don't accidentally sign extend in g2h()

Cast the argument of the g2h() macro to a target_ulong so that
it isn't accidentally sign-extended if it is a signed 32 bit
type and long is a 64 bit type. In particular, this fixes a
bug where it would return the wrong value for 32 bit guests
on 64 bit hosts when passed in one of the arg* values from
do_syscall() [which are all abi_long and thus signed types].
This could result in spurious failure of mlock(), among others.

Reviewed-by: Andreas F=E4rber <[email protected]>
Signed-off-by: Peter Maydell <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

implement vnc_dpy_setdata

The comment is wrong, we have to do something in the setdata callback.
Changing the framebuffer backing storage (happens when the guest pans
the display) renders the whole screen content invalid.

Trigger #1: cirrus vga + 32bit linux guest + vesafb with ypan enabled.
Trigger #2: std vga + http://patchwork.ozlabs.org/patch/145479/

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

initialize CPU model list after handling -readconfig options

To properly load cpudefs using -readconfig, we have to call
cpudef_init() after finishing the command-line option handling.

Consequently, the handling of "-cpu ?" has to be done after the
command-line option handling loop, too.

Without this patch, "-readconfig configfile -cpu ?" fails to list the
CPU definitions read from 'configfile'.

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

add Opteron_G4 CPU model (v2)

This patch addes a Bulldozer-based Opteron_G4 CPU model.

This version has the ffxsr bit actually disabled, to match what was
documented below. Thanks to Andre Przywara for spotting the bug.

I am trying to be conservative with the new model, so I am enabling only
features known to be useful to guests, and not enabling anything that
was not tested or found to be useful to a guest.

List of missing flags in comparison to real hardware:

- vme: host-specific feature.
- osxsave: it is not set here because it is set by the guest OS, not by KVM
- monitor: this is filtered out by the KVM module, so no point in
  enabling it.
- mmxext: untested, so not enabled.
- Perf*, Topology*, lwp, ibs: not emulated by KVM.
- wdt, skinit, osvw, altmovcr8, extapicspace, cmplegacy: untested,
  so not enabled.

List of new flags, in comparison to the Opteron_G3 model:

- xsave: xsave feature, already implemented by Qemu
- avx, aes, sse4.x, ssse3, pclmulqdq: all new state the new instructions
  could use is handled by the xsave state loading/saving code on Qemu.
- pdpe1gb: 1GB pages, supported by the KVM kernel module.
- ffxsr: untested, so not enabled
- fma4, xop: all new state the new instructions could use is handled by
  the xsave loading/saving code on Qemu.
- 3dnowprefetch: safe to pass through, though the flag is not used by
  Linux guests, at least.

Below is the comparison between the current Opteron_G3 model
and the new model being added.

- The "full" line contains the flags found on actual hardware.
- The "missing" line shows the flags that are present on actual
  hardware, but not on the added Opteron_G4 model.
- The "new" line shows the flags that were not on the Opteron_G3 model
  but are on Opteron_G4.

feature_edx:
  Opteron_G3: sse2 sse fxsr mmx clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de     fpu
  full:       sse2 sse fxsr mmx clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de vme fpu
  Opteron_G4: sse2 sse fxsr mmx clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de     fpu
  missing:                                                                                              vme

feature_ecx:
  Opteron_G3:                       popcnt               cx16       monitor           sse3
  full:       avx osxsave xsave aes popcnt sse4.2 sse4.1 cx16 ssse3 monitor pclmulqdq sse3
  Opteron_G4: avx         xsave aes popcnt sse4.2 sse4.1 cx16 ssse3         pclmulqdq sse3
  missing:        osxsave                                           monitor
  new:        avx         xsave aes        sse4.2 sse4.1      ssse3         pclmulqdq

extfeature_edx:
  Opteron_G3: lm rdtscp               fxsr mmx        nx pse36 pat cmov mca pge mtrr syscall apic cx8 mce pae msr tsc pse de     fpu
  full:       lm rdtscp pdpe1gb ffxsr fxsr mmx mmxext nx pse36 pat cmov mca pge mtrr syscall apic cx8 mce pae msr tsc pse de vme fpu
  Opteron_G4: lm rdtscp pdpe1gb       fxsr mmx        nx pse36 pat cmov mca pge mtrr syscall apic cx8 mce pae msr tsc pse de     fpu
  missing:                                     mmxext                                                                        vme
  new:                  pdpe1gb

extfeature_ecx:
  Opteron_G3:                                                                misalignsse sse4a abm                        svm           lahf_lm
  full:       Perf* Topology* fma4 lwp wdt skinit xop ibs osvw 3dnowprefetch misalignsse sse4a abm altmovcr8 extapicspace svm cmplegacy lahf_lm
  Opteron_G4:                 fma4                xop          3dnowprefetch misalignsse sse4a abm                        svm           lahf_lm
  new:                        fma4                xop          3dnowprefetch
  missing:    Perf* Topology*      lwp wdt skinit     ibs osvw                                     altmovcr8 extapicspace     cmplegacy

Changes v1 -> v2:
 - Actually disable ffxsr bit

Cc: Andre Przywara <[email protected]>
Signed-off-by: Eduardo Habkost <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

add SandyBridge CPU model

This patches add the definition of a SandyBridge CPU model.

Summary of differences:

Flags present on actual hardware, but not on the added model definition:

- pbe, tm, ht, ss, acpi, vme, xTPR, tm2, eist, smx: host-specific
  features, not exposed to guest.
- ds, ds-cpl, dtes64, pdcm: emulation not supported by KVM (although it
  may be added in the future if implementing PMU virtualization)
- pcid, vmx, monitor: not emulated by Qemu/KVM right now.
- osxsave: set by the guest OS, not by Qemu.

Flags added, that were not present on Westmere model:

- xsave: already supported by Qemu
- avx, pclmulqdq: all new state the new instructions could use is
  handled by xsave state loading/saving code.
- tsc-deadline, x2apic, rdtscp: already supported by Qemu/KVM.

Below there's a comparison of the features on the current Westmere CPU
model, and the SandyBridge CPU model.

- The "full" line contains the flags found on actual hardware.
- The "missing" line shows the flags that are present on actual
  hardware, but not on the added SandyBridge model.
- The "new" line shows the flags that were not on the Westmere model,
  but are on SandyBridge.

feature_edx:
  Westmere:                 sse2 sse fxsr mmx         clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de     fpu
  full:        pbe tm ht ss sse2 sse fxsr mmx ds acpi clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pge msr tsc pse de vme fpu
  SandyBridge:              sse2 sse fxsr mmx         clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de     fpu
  missing:     pbe tm ht ss                   ds acpi                                                                         vme

feature_ecx:
  Westmere:                      aes              popcnt        sse4.2 sse4.1                cx16 ssse3                                                  sse3
  full:        avx osxsave xsave aes tsc-deadline popcnt x2apic sse4.2 sse4.1 pcid pdcm xTPR cx16 ssse3 tm2 eist smx vmx ds-cpl monitor dtes64 pclmulqdq sse3
  SandyBridge: avx         xsave aes tsc-deadline popcnt x2apic sse4.2 sse4.1                cx16 ssse3                                        pclmulqdq sse3
  missing:         osxsave                                                    pcid pdcm xTPR            tm2 eist smx vmx ds-cpl monitor dtes64
  new:         avx         xsave     tsc-deadline        x2apic                                                                                pclmulqdq

extfeature_edx:
  Westmere:    i64        nx syscall
  full:        i64 rdtscp nx syscall
  SandyBridge: i64 rdtscp nx syscall
  new:             rdtscp

extfeature_ecx:
  Westmere:    lahf_lm
  full:        lahf_lm
  SandyBridge: lahf_lm

Cc: "Dugger, Donald D" <[email protected]>
Cc: "Zhang, Xiantao" <[email protected]>
Acked-by: Xiantao Zhang <[email protected]>
Signed-off-by: Eduardo Habkost <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

add "tsc-deadline" flag name to feature_ecx table

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

qom: fix device hot-unplug

Property removal modifies the list, so it is not safe to continue
iteration.  We know anyway that each object can have only one
parent (see object_property_add_child), so exit after finding
the requested object.

Reported-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

qdev: accept empty string properties

These were stored as NULL due to wrong cut-and-paste from set_pointer.

Reported-by: Gerhard Wiesinger <[email protected]>
Tested-by: Gerhard Wiesinger <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Anthony Liguori <[email protected]>

test-coroutine: add performance test for nesting

The performance test will also check for nesting. It will do
a certain quantity of cycles, and each of one will do a depth
nesting process.

This is useful for benchmarking the creation of coroutines,
given that nesting is creation-intensive (and the other perf
test does not benchmark that).

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

coroutine: adding configure option for sigaltstack coroutine backend

It's possible to use sigaltstack backend with --with-coroutine=sigaltstack

v2: changed from enable/disable configure flags

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

coroutine: adding configure choose mechanism for coroutine backend

Configure tries, as a default, ucontext functions for the
coroutines. But now the user can force another backend by
--with-coroutine=BACKEND option

v2: Using --with-coroutine=BACKEND instead of enable
disable individual configure options

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

coroutine: adding sigaltstack method (.c source)

This file is based in both coroutine-ucontext.c and
pth_mctx.c (from the GNU Portable Threads library).

The mechanism used to change stacks is the sigaltstack
function (variant 2 of the pth library).

v2: Some corrections. Moving global variables into
thread storage (CoroutineThreadState).

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

qcow2: Reduce number of I/O requests

If the first part of a write request is allocated, but the second isn't
and it can be allocated so that the resulting area is contiguous, handle
it at once. This is a common case for sequential writes.

After this patch, alloc_cluster_offset() only checks if the clusters are
already allocated or how many new clusters can be allocated contigouosly.
The actual cluster allocation is split off into a new function
do_alloc_cluster_offset().

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