sdk/emulator/qemu.git
9 years agoMerge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
Peter Maydell [Mon, 11 May 2015 11:01:09 +0000 (12:01 +0100)]
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging

- build bugfix from Fam and new configure check from Emilio
- two improvements to "info mtere" from Gerd
- KVM support for memory transaction attributes
- one more small step towards unlocked MMIO dispatch
- one piece of the qemu-nbd errno fixes
- trivial-ish patches from Denis and Thomas

# gpg: Signature made Fri May  8 13:47:29 2015 BST using RSA key ID 78C7AE83
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>"
# gpg:                 aka "Paolo Bonzini <pbonzini@redhat.com>"
# 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: 46F5 9FBD 57D6 12E7 BFD4  E2F7 7E15 100C CD36 69B1
#      Subkey fingerprint: F133 3857 4B66 2389 866C  7682 BFFB D25F 78C7 AE83

* remotes/bonzini/tags/for-upstream:
  qemu-nbd: only send a limited number of errno codes on the wire
  rules.mak: Force CFLAGS for all objects in DSO
  configure: require __thread support
  exec: move rcu_read_lock/unlock to address_space_translate callers
  kvm: add support for memory transaction attributes
  mtree: also print disabled regions
  mtree: tag & indent a bit better
  apic_common: improve readability of apic_reset_common
  kvm: Silence warning from valgrind

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9 years agoMerge remote-tracking branch 'remotes/kraxel/tags/pull-gtk-20150508-1' into staging
Peter Maydell [Mon, 11 May 2015 09:43:08 +0000 (10:43 +0100)]
Merge remote-tracking branch 'remotes/kraxel/tags/pull-gtk-20150508-1' into staging

gtk: add ui_info support, cleanups + fixes.

# gpg: Signature made Fri May  8 12:47:04 2015 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <kraxel@redhat.com>"
# gpg:                 aka "Gerd Hoffmann <gerd@kraxel.org>"
# gpg:                 aka "Gerd Hoffmann (private) <kraxel@gmail.com>"

* remotes/kraxel/tags/pull-gtk-20150508-1:
  gtk: update mouse position in mouse_set()
  gtk: create gtk.h
  gtk: add ui_info support
  console: add dpy_ui_info_supported
  console: delayed ui_info guest notification

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9 years agoMerge remote-tracking branch 'remotes/kraxel/tags/pull-usb-20150508-1' into staging
Peter Maydell [Mon, 11 May 2015 08:42:20 +0000 (09:42 +0100)]
Merge remote-tracking branch 'remotes/kraxel/tags/pull-usb-20150508-1' into staging

usb: qomify, bugfixes for xhci & uhci.

# gpg: Signature made Fri May  8 12:39:28 2015 BST using RSA key ID D3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <kraxel@redhat.com>"
# gpg:                 aka "Gerd Hoffmann <gerd@kraxel.org>"
# gpg:                 aka "Gerd Hoffmann (private) <kraxel@gmail.com>"

* remotes/kraxel/tags/pull-usb-20150508-1:
  uhci: controller is halted after reset
  usb: usb-serial QOMify
  usb: usb-redir QOMify
  usb: usb-wacom-tablet QOMify
  usb: usb-uas QOMify
  usb: usb-storage QOMify
  usb: usb-ccid QOMify
  usb: usb-net QOMify
  usb-mtp: fix segmentation fault
  usb: usb-mtp QOMify
  usb: usb-hub QOMify
  usb: usb-hid QOMify
  usb: usb-bt QOMify
  usb: usb-audio QOMify
  uhci: QOMify
  xhci: fix events for setup trb.
  Revert "xhci: generate a Transfer Event for each Transfer TRB with the IOC bit set"
  xhci: set timer to retry xfers
  usb: fix usb-net segfault

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9 years agoMerge remote-tracking branch 'remotes/cohuck/tags/s390x-20150508' into staging
Peter Maydell [Sun, 10 May 2015 20:40:54 +0000 (21:40 +0100)]
Merge remote-tracking branch 'remotes/cohuck/tags/s390x-20150508' into staging

Assorted s390x patches:
- updates for virtio-ccw and s390-virtio, making them more similar
  to virtio-pci
- improvements regarding per-vcpu interrupts and migration

# gpg: Signature made Fri May  8 09:45:09 2015 BST using RSA key ID C6F02FAF
# gpg: Good signature from "Cornelia Huck <huckc@linux.vnet.ibm.com>"
# gpg:                 aka "Cornelia Huck <cornelia.huck@de.ibm.com>"

* remotes/cohuck/tags/s390x-20150508:
  s390x/kvm: migrate vcpu interrupt state
  s390x: move fpu regs into a subsection of the vmstate
  s390x/kvm: use ioctl KVM_S390_IRQ for vcpu interrupts
  virtio-ccw: implement ->device_plugged
  virtio-ccw: change realization sequence
  s390-virtio: clear {used,avail}_event_idx on reset as well
  s390-virtio: use common features
  s390-virtio: Accommodate guests using virtqueues too early

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9 years agoqemu-nbd: only send a limited number of errno codes on the wire
Paolo Bonzini [Thu, 7 May 2015 15:25:10 +0000 (17:25 +0200)]
qemu-nbd: only send a limited number of errno codes on the wire

Right now, NBD includes potentially platform-specific error values in
the wire protocol.

Luckily, most common error values are more or less universal: in
particular, of all errno values <= 34 (up to ERANGE), they are all the
same on supported platforms except for 11 (which is EAGAIN on Windows and
Linux, but EDEADLK on Darwin and the *BSDs).  So, in order to guarantee
some portability, only keep a handful of possible error codes and squash
everything else to EINVAL.

This patch defines a limited set of errno values that are valid for the
NBD protocol, and specifies recommendations for what error to return
in specific corner cases.  The set of errno values is roughly based on
the errors listed in the read(2) and write(2) man pages, with some
exceptions:

- ENOMEM is added for servers that implement copy-on-write or other
  formats that require dynamic allocation.

- EDQUOT is not part of the universal set of errors; it can be changed
  to ENOSPC on the wire format.

- EFBIG is part of the universal set of errors, but it is also changed
  to ENOSPC because it is pretty similar to ENOSPC or EDQUOT.

Incoming values will in general match system errno values, but not
on the Hurd which has different errno values (they have a "subsystem
code" equal to 0x10 in bits 24-31).  The Hurd is probably not something
to which QEMU has been ported, but still do the right thing and
reverse-map the NBD errno values to the system errno values.

The corresponding patch to the NBD protocol description can be found at
http://article.gmane.org/gmane.linux.drivers.nbd.general/3154.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agorules.mak: Force CFLAGS for all objects in DSO
Fam Zheng [Thu, 7 May 2015 06:55:15 +0000 (14:55 +0800)]
rules.mak: Force CFLAGS for all objects in DSO

Because of the trick of process-archive-undefs, all .mo objects, even
with --enable-modules, are dependencies of executables.

This breaks CFLAGS propogation because the compiling of module object
will happen too early before building for DSO.

With GCC 5, the linking would fail because .o doesn't have -fPIC. Also,
BUILD_DSO will be missed. (module-common.o will have it, so the stamp
symbol was still liked in .so).

Fix the problem by forcing the CFLAGS on individual .o-cflags during
unnest-vars.

Reported-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Fam Zheng <famz@redhat.com>
Cc: qemu-stable@nongnu.org # 2.3
Message-Id: <1430981715-31465-1-git-send-email-famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agouhci: controller is halted after reset
Gerd Hoffmann [Thu, 7 May 2015 07:24:00 +0000 (09:24 +0200)]
uhci: controller is halted after reset

... and the status register should say so.

Fixes "usbus0: controller did not stop" error printed by freebsd.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-serial QOMify
Gonglei [Wed, 6 May 2015 12:55:36 +0000 (20:55 +0800)]
usb: usb-serial QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-redir QOMify
Gonglei [Wed, 6 May 2015 12:55:35 +0000 (20:55 +0800)]
usb: usb-redir QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-wacom-tablet QOMify
Gonglei [Wed, 6 May 2015 12:55:34 +0000 (20:55 +0800)]
usb: usb-wacom-tablet QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-uas QOMify
Gonglei [Wed, 6 May 2015 12:55:33 +0000 (20:55 +0800)]
usb: usb-uas QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-storage QOMify
Gonglei [Wed, 6 May 2015 12:55:32 +0000 (20:55 +0800)]
usb: usb-storage QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-ccid QOMify
Gonglei [Wed, 6 May 2015 12:55:31 +0000 (20:55 +0800)]
usb: usb-ccid QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-net QOMify
Gonglei [Wed, 6 May 2015 12:55:30 +0000 (20:55 +0800)]
usb: usb-net QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb-mtp: fix segmentation fault
Gonglei [Wed, 6 May 2015 12:55:29 +0000 (20:55 +0800)]
usb-mtp: fix segmentation fault

When x-root property not be configured, will cause segfault
because of null pointer accessing. Add a check for s->root
property avoid segfault.

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-mtp QOMify
Gonglei [Wed, 6 May 2015 12:55:28 +0000 (20:55 +0800)]
usb: usb-mtp QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-hub QOMify
Gonglei [Wed, 6 May 2015 12:55:27 +0000 (20:55 +0800)]
usb: usb-hub QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-hid QOMify
Gonglei [Wed, 6 May 2015 12:55:26 +0000 (20:55 +0800)]
usb: usb-hid QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-bt QOMify
Gonglei [Wed, 6 May 2015 12:55:25 +0000 (20:55 +0800)]
usb: usb-bt QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: usb-audio QOMify
Gonglei [Wed, 6 May 2015 12:55:24 +0000 (20:55 +0800)]
usb: usb-audio QOMify

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agouhci: QOMify
Gonglei [Wed, 6 May 2015 12:55:23 +0000 (20:55 +0800)]
uhci: QOMify

Cc: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agoxhci: fix events for setup trb.
Gerd Hoffmann [Tue, 28 Apr 2015 07:31:44 +0000 (09:31 +0200)]
xhci: fix events for setup trb.

When we find a IOC bit set on a setup trb and therefore queue an event,
that should not stop events being generated for following data trbs.
So clear the 'reported' flag.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agoRevert "xhci: generate a Transfer Event for each Transfer TRB with the IOC bit set"
Gerd Hoffmann [Tue, 28 Apr 2015 07:19:35 +0000 (09:19 +0200)]
Revert "xhci: generate a Transfer Event for each Transfer TRB with the IOC bit set"

This makes xhci generate multiple short packet events in case of
multi-trb transfers.  Which is wrong.  We need to fix this in a
different way.

This reverts commit aa6857891df614c620e6e9fc4bc4af6e0e49cafd.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agoxhci: set timer to retry xfers
Gerd Hoffmann [Tue, 28 Apr 2015 07:19:14 +0000 (09:19 +0200)]
xhci: set timer to retry xfers

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agousb: fix usb-net segfault
Michal Kazior [Wed, 29 Apr 2015 11:34:59 +0000 (11:34 +0000)]
usb: fix usb-net segfault

The dev->config pointer isn't set until guest
system initializes usb devices (via
usb_desc_set_config). However qemu networking can
go through some motions prior to that, e.g.:

 #0  is_rndis (s=0x555557261970) at hw/usb/dev-network.c:653
 #1  0x000055555585f723 in usbnet_can_receive (nc=0x55555641e820) at hw/usb/dev-network.c:1315
 #2  0x000055555587635e in qemu_can_send_packet (sender=0x5555572660a0) at net/net.c:470
 #3  0x0000555555878e34 in net_hub_port_can_receive (nc=0x5555562d7800) at net/hub.c:101
 #4  0x000055555587635e in qemu_can_send_packet (sender=0x5555562d7980) at net/net.c:470
 #5  0x000055555587dbca in tap_can_send (opaque=0x5555562d7980) at net/tap.c:172

The command to reproduce most reliably was:

 qemu-system-i386 -usb -device usb-net,vlan=0 -net tap,vlan=0

This wasn't strictly a problem with tap. Other
networking endpoints (vde, user) could trigger
this problem as well.

Fixes: https://bugs.launchpad.net/qemu/+bug/1050823
Cc: qemu-stable@nongnu.org
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agoconfigure: require __thread support
Emilio G. Cota [Wed, 29 Apr 2015 11:09:02 +0000 (13:09 +0200)]
configure: require __thread support

The codebase doesn't build without __thread support.
Formalise this requirement by adding a check for it in the
configure script.

Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agos390x/kvm: migrate vcpu interrupt state
Jens Freimann [Mon, 2 Mar 2015 16:44:24 +0000 (17:44 +0100)]
s390x/kvm: migrate vcpu interrupt state

This patch adds support to migrate vcpu interrupts.
We use ioctl KVM_S390_GET_IRQ_STATE and _SET_IRQ_STATE
to get/set the complete interrupt state for a vcpu.

Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
9 years agos390x: move fpu regs into a subsection of the vmstate
David Hildenbrand [Mon, 30 Mar 2015 11:22:47 +0000 (13:22 +0200)]
s390x: move fpu regs into a subsection of the vmstate

Let's move the floating point registers into a seperate subsection and
bump up the version id. This cleans up the current vmstate and will
allow for a future extension with vector registers in a compatible way.

This patch is based on a patch from Eric Farman.

Reviewed-by: Eric Farman <farman@linux.vnet.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
9 years agos390x/kvm: use ioctl KVM_S390_IRQ for vcpu interrupts
Jens Freimann [Thu, 18 Dec 2014 16:38:05 +0000 (17:38 +0100)]
s390x/kvm: use ioctl KVM_S390_IRQ for vcpu interrupts

KVM_S390_INT uses only two parameter fields. This is not
enough to pass all required information for certain interrupts.

A new ioctl KVM_S390_IRQ is available which allows us to
inject all local interrupts as defined in the Principles of
Operation. It takes a struct kvm_s390_irq as a parameter
which can store interrupt payload data for all interrupts.

Let's use the new ioctl for injecting vcpu interrupts.

Tested-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
9 years agovirtio-ccw: implement ->device_plugged
Cornelia Huck [Tue, 21 Apr 2015 14:36:56 +0000 (16:36 +0200)]
virtio-ccw: implement ->device_plugged

Let's move operations that are only valid after the backend has been
realized to a ->device_plugged callback, just as virtio-pci does.
Also reorder setting up the host feature bits to the sequence used
by virtio-pci.

While we're at it, also add a ->device_unplugged callback to stop
ioeventfd, just to be on the safe side.

Reviewed-by: Shannon Zhao <shannon.zhao@linaro.org>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Message-Id: <1429627016-30656-3-git-send-email-cornelia.huck@de.ibm.com>

9 years agovirtio-ccw: change realization sequence
Cornelia Huck [Tue, 21 Apr 2015 14:36:55 +0000 (16:36 +0200)]
virtio-ccw: change realization sequence

virtio-ccw has an odd sequence of realizing devices: first the
device-specific relization (net, block, ...), then the generic
realization. It feels less odd to have the generic realization
callback trigger the device-specific realization instead (and this
also matches what virtio-pci does).

One thing to note: We need to defer initializing the cu model in the
sense id data until after the device-specific realization has been
performed, as we need to refer to the virtio device's device_id.

Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Message-Id: <1429627016-30656-2-git-send-email-cornelia.huck@de.ibm.com>

9 years agos390-virtio: clear {used,avail}_event_idx on reset as well
Christian Borntraeger [Mon, 4 May 2015 13:27:25 +0000 (15:27 +0200)]
s390-virtio: clear {used,avail}_event_idx on reset as well

The old s390-virtio transport clears the vring used/avail indices in
the shared area on reset. When we enabled event_idx for virtio-blk, we
noticed that this is not enough: We also need to clear the published
used/avail event indices, or reboot will fail.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
9 years agos390-virtio: use common features
Cornelia Huck [Wed, 29 Apr 2015 13:30:58 +0000 (15:30 +0200)]
s390-virtio: use common features

We used to avoid enabling event_idx for virtio-blk devices via
s390-virtio, but we now have a workaround in place for guests trying
to use the device before setting DRIVER_OK. Therefore, let's add
DEFINE_VIRTIO_COMMON_FEATURES to the base device so all devices get
those common features - and make s390-virtio use the same mechanism
as the other transports do.

Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Shannon Zhao <shannon.zhao@linaro.org>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
9 years agos390-virtio: Accommodate guests using virtqueues too early
Christian Borntraeger [Thu, 30 Apr 2015 15:53:13 +0000 (17:53 +0200)]
s390-virtio: Accommodate guests using virtqueues too early

Feature updates are not a synchronuous operation for the legacy
s390-virtio transport. This transport syncs the guest feature bits
(those from finalize) on the set_status hypercall. Before that qemu
thinks that features are zero, which means QEMU will misbehave, e.g.
it will not write the event index, even if the guest asks for it.

Let's detect the case where a kick happens before the driver is ready
and force sync the features.
With this workaround, it is now safe to switch to the common feature
bit handling code as used by all other transports.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
9 years agohw/ptimer: Do not artificially limit timers when using icount
Edgar E. Iglesias [Wed, 10 Sep 2014 08:33:58 +0000 (18:33 +1000)]
hw/ptimer: Do not artificially limit timers when using icount

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
9 years agoMerge remote-tracking branch 'remotes/juanquintela/tags/migration/20150507-1' into...
Peter Maydell [Thu, 7 May 2015 17:22:03 +0000 (18:22 +0100)]
Merge remote-tracking branch 'remotes/juanquintela/tags/migration/20150507-1' into staging

migration/next for 20150507

# gpg: Signature made Thu May  7 17:42:19 2015 BST using RSA key ID 5872D723
# gpg: Good signature from "Juan Quintela <quintela@redhat.com>"
# gpg:                 aka "Juan Quintela <quintela@trasno.org>"

* remotes/juanquintela/tags/migration/20150507-1:
  migration: Fix migration state update issue
  migration: avoid divide by zero in xbzrle cache miss rate
  migration: Add hmp interface to set and query parameters
  migration: Add qmp commands to set and query parameters
  migration: Use an array instead of 3 parameters
  migration: Add interface to control compression
  migration: Add the core code for decompression
  migration: Make compression co-work with xbzrle
  migration: Add the core code of multi-thread compression
  migration: Split save_zero_page from ram_save_page
  arch_init: Add and free data struct for decompression
  arch_init: Alloc and free data struct for compression
  qemu-file: Add compression functions to QEMUFile
  migration: Add the framework of multi-thread decompression
  migration: Add the framework of multi-thread compression
  docs: Add a doc about multiple thread compression

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9 years agomigration: Fix migration state update issue
Liang Li [Thu, 7 May 2015 18:31:22 +0000 (02:31 +0800)]
migration: Fix migration state update issue

If live migration is very fast and can be completed in 1 second,
the dirty_sync_count of MigrationState will not be updated.
Then you will see "dirty sync count: 0" in qemu monitor even if
the actual dirty sync count is not 0.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr.David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: avoid divide by zero in xbzrle cache miss rate
Michael Chapman [Wed, 15 Apr 2015 03:59:14 +0000 (13:59 +1000)]
migration: avoid divide by zero in xbzrle cache miss rate

This bug manifested itself as a VM that could not be resumed by libvirt
following a migration:

  # virsh resume example
  error: Failed to resume domain example
  error: internal error: cannot parse json {"return":
    {"xbzrle-cache":
      {..., "cache-miss-rate": -nan, ...},
      ...
    }
  }: lexical error: malformed number, a digit is required after the minus sign.

This patch also ensures xbzrle_cache_miss_prev and iterations_prev are
cleared at the start of the migration.

Signed-off-by: Michael Chapman <mike@very.puzzling.org>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Add hmp interface to set and query parameters
Liang Li [Mon, 23 Mar 2015 08:32:29 +0000 (16:32 +0800)]
migration: Add hmp interface to set and query parameters

Add the hmp interface to tune and query the parameters used in
live migration.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Add qmp commands to set and query parameters
Liang Li [Mon, 23 Mar 2015 08:32:28 +0000 (16:32 +0800)]
migration: Add qmp commands to set and query parameters

Add the qmp commands to tune and query the parameters used in live
migration.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Use an array instead of 3 parameters
Liang Li [Mon, 23 Mar 2015 08:32:27 +0000 (16:32 +0800)]
migration: Use an array instead of 3 parameters

Put the three parameters related to multiple thread (de)compression
into an int array, and use an enum type to index the parameter.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Add interface to control compression
Liang Li [Mon, 23 Mar 2015 08:32:26 +0000 (16:32 +0800)]
migration: Add interface to control compression

The multiple compression threads can be turned on/off through
qmp and hmp interface before doing live migration.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Dr.David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Add the core code for decompression
Liang Li [Mon, 23 Mar 2015 08:32:25 +0000 (16:32 +0800)]
migration: Add the core code for decompression

Implement the core logic of multiple thread decompression,
the decompression can work now.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Make compression co-work with xbzrle
Liang Li [Mon, 23 Mar 2015 08:32:24 +0000 (16:32 +0800)]
migration: Make compression co-work with xbzrle

Now, multiple thread compression can co-work with xbzrle. when
xbzrle is on, multiple thread compression will only work at the
first round of RAM data sync.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Dr.David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Add the core code of multi-thread compression
Liang Li [Mon, 23 Mar 2015 08:32:23 +0000 (16:32 +0800)]
migration: Add the core code of multi-thread compression

Implement the core logic of the multiple thread compression. At this
point, multiple thread compression can't co-work with xbzrle yet.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Split save_zero_page from ram_save_page
Liang Li [Mon, 23 Mar 2015 08:32:22 +0000 (16:32 +0800)]
migration: Split save_zero_page from ram_save_page

Split the function save_zero_page from ram_save_page so that we can
reuse it later.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agoarch_init: Add and free data struct for decompression
Liang Li [Mon, 23 Mar 2015 08:32:21 +0000 (16:32 +0800)]
arch_init: Add and free data struct for decompression

Define the data structure and variables used to do multiple thread
decompression, and add the code to initialize and free them.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Dr.David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agoarch_init: Alloc and free data struct for compression
Liang Li [Mon, 23 Mar 2015 08:32:20 +0000 (16:32 +0800)]
arch_init: Alloc and free data struct for compression

Define the data structure and variables used to do multiple thread
compression, and add the code to initialize and free them.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Dr.David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agoqemu-file: Add compression functions to QEMUFile
Liang Li [Mon, 23 Mar 2015 08:32:19 +0000 (16:32 +0800)]
qemu-file: Add compression functions to QEMUFile

qemu_put_compression_data() compress the data and put it to QEMUFile.
qemu_put_qemu_file() put the data in the buffer of source QEMUFile to
destination QEMUFile.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Add the framework of multi-thread decompression
Liang Li [Mon, 23 Mar 2015 08:32:18 +0000 (16:32 +0800)]
migration: Add the framework of multi-thread decompression

Add the code to create and destroy the multiple threads those will be
used to do data decompression. Left some functions empty just to keep
clearness, and the code will be added later.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Dr.David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agomigration: Add the framework of multi-thread compression
Liang Li [Mon, 23 Mar 2015 08:32:17 +0000 (16:32 +0800)]
migration: Add the framework of multi-thread compression

Add the code to create and destroy the multiple threads those will
be used to do data compression. Left some functions empty to keep
clearness, and the code will be added later.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Dr.David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agodocs: Add a doc about multiple thread compression
Liang Li [Mon, 23 Mar 2015 08:32:16 +0000 (16:32 +0800)]
docs: Add a doc about multiple thread compression

Give some details about the multiple thread (de)compression and
how to use it in live migration.

Signed-off-by: Liang Li <liang.z.li@intel.com>
Signed-off-by: Yang Zhang <yang.z.zhang@intel.com>
Reviewed-by: Dr.David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
9 years agoMerge remote-tracking branch 'remotes/rth/tags/tcg-next-20150505' into staging
Peter Maydell [Wed, 6 May 2015 10:16:35 +0000 (11:16 +0100)]
Merge remote-tracking branch 'remotes/rth/tags/tcg-next-20150505' into staging

size reduction merge

# gpg: Signature made Wed May  6 00:21:43 2015 BST using RSA key ID 4DD0279B
# gpg: Good signature from "Richard Henderson <rth7680@gmail.com>"
# gpg:                 aka "Richard Henderson <rth@redhat.com>"
# gpg:                 aka "Richard Henderson <rth@twiddle.net>"

* remotes/rth/tags/tcg-next-20150505:
  tcg: optimise memory layout of TCGTemp

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9 years agogtk: update mouse position in mouse_set()
Gerd Hoffmann [Tue, 1 Jul 2014 17:12:45 +0000 (19:12 +0200)]
gtk: update mouse position in mouse_set()

Without that the next mouse motion event uses the old position
as base for relative move calculation, giving wrong results and
making your mouse pointer jump around.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agogtk: create gtk.h
Gerd Hoffmann [Wed, 4 Mar 2015 14:37:27 +0000 (15:37 +0100)]
gtk: create gtk.h

Move various gtk bits (includes, data structures) to a header file.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agogtk: add ui_info support
Gerd Hoffmann [Fri, 13 Mar 2015 11:47:00 +0000 (12:47 +0100)]
gtk: add ui_info support

Pass new display size to the guest after window resizes.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agoconsole: add dpy_ui_info_supported
Gerd Hoffmann [Fri, 13 Mar 2015 11:21:14 +0000 (12:21 +0100)]
console: add dpy_ui_info_supported

Allow ui code to check whenever the emulated
display supports display change notifications.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agoconsole: delayed ui_info guest notification
Gerd Hoffmann [Thu, 12 Mar 2015 11:51:13 +0000 (12:51 +0100)]
console: delayed ui_info guest notification

So we don't flood the guest with display change notifications
while the user resizes the window.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
9 years agoMerge remote-tracking branch 'remotes/armbru/tags/pull-qmp-2015-05-05' into staging
Peter Maydell [Tue, 5 May 2015 17:22:12 +0000 (18:22 +0100)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-qmp-2015-05-05' into staging

drop qapi nested structs

# gpg: Signature made Tue May  5 17:40:40 2015 BST using RSA key ID EB918653
# gpg: Good signature from "Markus Armbruster <armbru@redhat.com>"
# gpg:                 aka "Markus Armbruster <armbru@pond.sub.org>"

* remotes/armbru/tags/pull-qmp-2015-05-05: (40 commits)
  qapi: Check for member name conflicts with a base class
  qapi: Support (subset of) \u escapes in strings
  qapi: Tweak doc references to QMP when QGA is also meant
  qapi: Drop dead visitor code related to nested structs
  qapi: Drop support for inline nested types
  qapi: Drop inline nested structs in query-pci
  qapi: Drop inline nested struct in query-version
  qapi: Drop tests for inline nested structs
  qapi: Merge UserDefTwo and UserDefNested in tests
  qapi: Forbid 'type' in schema
  qapi: Use 'struct' instead of 'type' in schema
  qapi: Document 'struct' metatype
  qapi: Prefer 'struct' over 'type' in generator
  qapi: More rigorous checking for type safety bypass
  qapi: Whitelist commands that don't return dictionary
  qapi: Require valid names
  qapi: More rigourous checking of types
  qapi: Add some type check tests
  qapi: Unify type bypass and add tests
  qapi: Allow true, false and null in schema json
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9 years agoqapi: Check for member name conflicts with a base class
Eric Blake [Mon, 4 May 2015 15:05:37 +0000 (09:05 -0600)]
qapi: Check for member name conflicts with a base class

Our type inheritance for both 'struct' and for flat 'union' merges
key/value pairs from the base class with those from the type in
question.  Although the C code currently boxes things so that there
is a distinction between which member is referred to, the QMP wire
format does not allow passing a key more than once in a single
object.  Besides, if we ever change the generated C code to not be
quite so boxy, we'd want to avoid duplicate member names there,
too.

Fix a testsuite entry added in an earlier patch, as well as adding
a couple more tests to ensure we have appropriate coverage.  Ensure
that collisions are detected, regardless of whether there is a
difference in opinion on whether the member name is optional.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Support (subset of) \u escapes in strings
Eric Blake [Mon, 4 May 2015 15:05:36 +0000 (09:05 -0600)]
qapi: Support (subset of) \u escapes in strings

The handling of \ inside QAPI strings was less than ideal, and
really only worked JSON's \/, \\, \", and our extension of \'
(an obvious extension, when you realize we use '' instead of ""
for strings).  For other things, like '\n', it resulted in a
literal 'n' instead of a newline.

Of course, at the moment, we really have no use for escaped
characters, as QAPI has to map to C identifiers, and we currently
support ASCII only for that.  But down the road, we may add
support for default values for string parameters to a command
or struct; if that happens, it would be nice to correctly support
all JSON escape sequences, such as \n or \uXXXX.  This gets us
closer, by supporting Unicode escapes in the ASCII range.

Since JSON does not require \OCTAL or \xXX escapes, and our QMP
implementation does not understand them either, I intentionally
reject it here, but it would be an easy addition if we desired it.
Likewise, intentionally refusing the NUL byte means we don't have
to worry about C strings being shorter than the qapi input.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Tweak doc references to QMP when QGA is also meant
Eric Blake [Mon, 4 May 2015 15:05:35 +0000 (09:05 -0600)]
qapi: Tweak doc references to QMP when QGA is also meant

We have more than one qapi schema in use by more than one protocol.
Add a new term 'Client JSON Protocol' for use throughout the
document, to avoid confusion on whether something refers only to
QMP and not QGA.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Drop dead visitor code related to nested structs
Eric Blake [Mon, 4 May 2015 15:05:34 +0000 (09:05 -0600)]
qapi: Drop dead visitor code related to nested structs

Now that we no longer have nested structs to visit, the use of
prefix strings is no longer required.  Remove the code that is
no longer reachable.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Drop support for inline nested types
Eric Blake [Mon, 4 May 2015 15:05:33 +0000 (09:05 -0600)]
qapi: Drop support for inline nested types

A future patch will be using a 'name':{dictionary} entry in the
QAPI schema to specify a default value for an optional argument
(see previous commit messages for more details why); but existing
use of inline nested structs conflicts with that goal. Now that
all commands have been changed to avoid inline nested structs,
nuke support for them, and turn it into a hard error. Update the
testsuite to reflect tighter parsing rules.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Drop inline nested structs in query-pci
Eric Blake [Mon, 4 May 2015 15:05:32 +0000 (09:05 -0600)]
qapi: Drop inline nested structs in query-pci

A future patch will be using a 'name':{dictionary} entry in the
QAPI schema to specify a default value for an optional argument
(see previous commit message for more details why); but existing
use of inline nested structs conflicts with that goal. This patch
fixes one of only two commands relying on nested types, by
breaking the nesting into an explicit type; it means that the
type is now boxed instead of unboxed in C code, but the QMP wire
format is unaffected by this change.

Prefer the safer g_new0() while making the conversion, and reduce
some long lines.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Drop inline nested struct in query-version
Eric Blake [Mon, 4 May 2015 15:05:31 +0000 (09:05 -0600)]
qapi: Drop inline nested struct in query-version

A future patch will be using a 'name':{dictionary} entry in the
QAPI schema to specify a default value for an optional argument
(see previous commit message for more details why); but existing
use of inline nested structs conflicts with that goal. This patch
fixes one of only two commands relying on nested types, by
breaking the nesting into an explicit type; it means that the
type is now boxed instead of unboxed in C code, but the QMP wire
format is unaffected by this change.

Prefer the safer g_new0() while making the conversion.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Drop tests for inline nested structs
Eric Blake [Mon, 4 May 2015 15:05:30 +0000 (09:05 -0600)]
qapi: Drop tests for inline nested structs

A future patch will be using a 'name':{dictionary} entry in the
QAPI schema to specify a default value for an optional argument;
but existing use of inline nested structs conflicts with that goal.

More precisely, a definition in the QAPI schema associates a name
with a set of properties:

Example 1: { 'struct': 'Foo', 'data': { MEMBERS... } }
associates the global name 'Foo' with properties (meta-type struct)
and MEMBERS...

Example 2: 'mumble': TYPE
within MEMBERS... above associates 'mumble' with properties (type
TYPE) and (optional false) within type Foo

The syntax of example 1 is extensible; if we need another property,
we add another name/value pair to the dictionary (such as
'base':TYPE).  The syntax of example 2 is not extensible, because
the right hand side can only be a type.

We have used name encoding to add a property: "'*mumble': 'int'"
associates 'mumble' with (type int) and (optional true).  Nice,
but doesn't scale.  So the solution is to change our existing uses
to be syntactic sugar to an extensible form:

   NAME: TYPE   --> NAME:  { 'type': TYPE, 'optional': false }
   *ONAME: TYPE --> ONAME: { 'type': TYPE, 'optional': true }

This patch fixes the testsuite to avoid inline nested types, by
breaking the nesting into explicit types; it means that the type
is now boxed instead of unboxed in C code, but makes no difference
on the wire (and if desired, a later patch could change the
generator to not do so much boxing in C).  When touching code to
add new allocations, also convert existing allocations to
consistently prefer typesafe g_new0 over g_malloc0 when a type
name is involved.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Merge UserDefTwo and UserDefNested in tests
Eric Blake [Mon, 4 May 2015 15:05:29 +0000 (09:05 -0600)]
qapi: Merge UserDefTwo and UserDefNested in tests

In the testsuite, UserDefTwo and UserDefNested were identical
structs other than the member names.  Reduce code duplication by
having just one type, and choose names that also favor reuse.
This will also make it easier for a later patch to get rid of
inline nested types in QAPI.  When touching code related to
allocations, convert g_malloc0(sizeof(Type)) to the more typesafe
g_new0(Type, 1).

Ensure that 'make check-qapi-schema check-unit' still passes.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Forbid 'type' in schema
Eric Blake [Mon, 4 May 2015 15:05:28 +0000 (09:05 -0600)]
qapi: Forbid 'type' in schema

Referring to "type" as both a meta-type (built-in, enum, union,
alternate, or struct) and a specific type (the name that the
schema uses for declaring structs) is confusing.  Finish up the
conversion to using "struct" in qapi schema by removing the hack
in the generator that allowed 'type'.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Use 'struct' instead of 'type' in schema
Eric Blake [Mon, 4 May 2015 15:05:27 +0000 (09:05 -0600)]
qapi: Use 'struct' instead of 'type' in schema

Referring to "type" as both a meta-type (built-in, enum, union,
alternate, or struct) and a specific type (the name that the
schema uses for declaring structs) is confusing.  Do the bulk of
the conversion to "struct" in qapi schema, with a fairly
mechanical:

for f in `find -name '*.json'; do sed -i "s/'type'/'struct'/"; done

followed by manually filtering out the places where we have a
'type' embedded in 'data'.  Then tweak a couple of tests whose
output changes slightly due to longer lines.

I also verified that the generated files for QMP and QGA (such
as qmp-commands.h) are the same before and after, as assurance
that I didn't leave in any accidental member name changes.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Document 'struct' metatype
Eric Blake [Mon, 4 May 2015 15:05:26 +0000 (09:05 -0600)]
qapi: Document 'struct' metatype

Referring to "type" as both a meta-type (built-in, enum, union,
alternate, or struct) and a specific type (the name that the
schema uses for declaring structs) is confusing.  Now that the
generator accepts 'struct' as a synonym for 'type', update all
documentation to use saner wording.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Prefer 'struct' over 'type' in generator
Eric Blake [Mon, 4 May 2015 15:05:25 +0000 (09:05 -0600)]
qapi: Prefer 'struct' over 'type' in generator

Referring to "type" as both a meta-type (built-in, enum, union,
alternate, or struct) and a specific type (the name that the
schema uses for declaring structs) is confusing.  The confusion
is only made worse by the fact that the generator mostly already
refers to struct even when dealing with expr['type'].  This
commit changes the generator to consistently refer to it as
struct everywhere, plus a single back-compat tweak that allows
accepting the existing .json files as-is, so that the meat of
this change is separate from the mindless churn of that change.

Fix the testsuite fallout for error messages that change, and
in some cases, become more legible.  Improve comments to better
match our intentions where a struct (rather than any complex
type) is required.  Note that in some cases, an error message
now refers to 'struct' while the schema still refers to 'type';
that will be cleaned up in the later commit to the schema.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: More rigorous checking for type safety bypass
Eric Blake [Mon, 4 May 2015 15:05:24 +0000 (09:05 -0600)]
qapi: More rigorous checking for type safety bypass

Now that we have a way to validate every type, we can also be
stricter about enforcing that callers that want to bypass
type safety in generated code.  Prior to this patch, it didn't
matter what value was associated with the key 'gen', but it
looked odd that 'gen':'yes' could result in bypassing the
generated code.  These changes also enforce the changes made
earlier in the series for documentation and consolidation of
using '**' as the wildcard type, as well as 'gen':false as the
canonical spelling for requesting type bypass.

Note that 'gen':false is a one-way switch away from the default;
we do not support 'gen':true (similar for 'success-response').
In practice, this doesn't matter.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Whitelist commands that don't return dictionary
Eric Blake [Mon, 4 May 2015 15:05:23 +0000 (09:05 -0600)]
qapi: Whitelist commands that don't return dictionary

...or an array of dictionaries.  Although we have to cater to
existing commands, returning a non-dictionary means the command
is not extensible (no new name/value pairs can be added if more
information must be returned in parallel).  By making the
whitelist explicit, any new command that falls foul of this
practice will have to be self-documenting, which will encourage
developers to either justify the action or rework the design to
use a dictionary after all.

It's a little bit sloppy that we share a single whitelist among
three clients (it's too permissive for each).  If this is a
problem, a future patch could tighten things by having the
generator take the whitelist as an argument (as in
scripts/qapi-commands.py --legacy-returns=...), or by having
the generator output C code that requires explicit use of the
whitelist (as in:
 #ifndef FROBNICATE_LEGACY_RETURN_OK
 # error Command 'frobnicate' should return a dictionary
 #endif
then having the callers define appropriate macros).  But until
we need such fine-grained separation (if ever), this patch does
the job just fine.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Require valid names
Eric Blake [Mon, 4 May 2015 15:05:22 +0000 (09:05 -0600)]
qapi: Require valid names

Previous commits demonstrated that the generator overlooked various
bad naming situations:
- types, commands, and events need a valid name
- enum members must be valid names, when combined with prefix
- union and alternate branches cannot be marked optional

Valid upstream names match [a-zA-Z][a-zA-Z0-9_-]*; valid downstream
names match __[a-zA-Z][a-zA-Z0-9._-]*.  Enumerations match the
weaker [a-zA-Z0-9._-]+ (in part thanks to QKeyCode picking an enum
that starts with a digit, which we can't change now due to
backwards compatibility).  Rather than call out three separate
regex, this patch just uses a broader combination that allows both
upstream and downstream names, as well as a small hack that
realizes that any enum name is merely a suffix to an already valid
name prefix (that is, any enum name is valid if prepending _ fits
the normal rules).

We could reject new enumeration names beginning with a digit by
whitelisting existing exceptions.  We could also be stricter
about the distinction between upstream names (no leading
underscore, no use of dot) and downstream (mandatory leading
double underscore), but it is probably not worth the bother.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: More rigourous checking of types
Eric Blake [Mon, 4 May 2015 15:05:21 +0000 (09:05 -0600)]
qapi: More rigourous checking of types

Now that we know every expression is valid with regards to
its keys, we can add further tests that those keys refer to
valid types.  With this patch, all uses of a type (the 'data':
of command, type, union, alternate, and event; the 'returns':
of command; the 'base': of type and union) must resolve to an
appropriate subset of metatypes  declared by the current qapi
parse; this includes recursing into each member of a data
dictionary.  Dealing with '**' and nested anonymous structs
will be done in later patches.

Update the testsuite to match improved output.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Add some type check tests
Eric Blake [Mon, 4 May 2015 15:05:20 +0000 (09:05 -0600)]
qapi: Add some type check tests

Demonstrate that the qapi generator silently parses confusing
types, which may cause other errors later on. Later patches
will update the expected results as the generator is made stricter.

Most of the new tests focus on blatant errors.  But
returns-whitelist is a case where we have historically allowed
returning something other than a JSON object from particular
commands; we have to keep that behavior to avoid breaking clients,
but it would be nicer to avoid adding such commands in the future,
because any return that is not an (array of) object cannot be
easily extended if future qemu wants to return additional
information.  The QMP protocol already documents that clients
should ignore unknown dictionary keys, but does not require
clients to have to handle more than one type of JSON object.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Unify type bypass and add tests
Eric Blake [Mon, 4 May 2015 15:05:19 +0000 (09:05 -0600)]
qapi: Unify type bypass and add tests

For a few QMP commands, we are forced to pass an arbitrary type
without tracking it properly in QAPI.  Among the existing clients,
this unnamed type was spelled 'dict', 'visitor', and '**'; this
patch standardizes on '**', matching the documentation changes
earlier in the series.

Meanwhile, for the 'gen' key, we have been ignoring the value,
although the schema consistently used "'no'" ('success-response'
was hard-coded to checking for 'no').  But now that we can support
a literal "false" in the schema, we might as well use that rather
than ignoring the value or special-casing a random string.  Note
that these are one-way switches (use of 'gen':true is not the same
as omitting 'gen'). Also, the use of '**' requires 'gen':false,
but the use of 'gen':false does not mandate the use of '**'.

There is no difference to the generated code.  Add some tests on
what we'd like to guarantee, although it will take later patches
to clean up test results and actually enforce the use of a bool
parameter.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Allow true, false and null in schema json
Fam Zheng [Mon, 4 May 2015 15:05:18 +0000 (09:05 -0600)]
qapi: Allow true, false and null in schema json

In the near term, we will use it for a sensible-looking
'gen':false inside command declarations, instead of the
current ugly 'gen':'no'.

In the long term, it will allow conversion from shorthand
with defaults mentioned only in side-band documentation:
 'data':{'*flag':'bool', '*string':'str'}
into an explicit default value documentation, as in:
 'data':{'flag':{'type':'bool', 'optional':true, 'default':true},
         'string':{'type':'str', 'optional':true, 'default':null}}

We still don't parse integer values (also necessary before
we can allow explicit defaults), but that can come in a later
series.

Update the testsuite to match an improved error message.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Better error messages for duplicated expressions
Eric Blake [Mon, 4 May 2015 15:05:17 +0000 (09:05 -0600)]
qapi: Better error messages for duplicated expressions

The previous commit demonstrated that the generator overlooked
duplicate expressions:
- a complex type or command reusing a built-in type name
- redeclaration of a type name, whether by the same or different
metatype
- redeclaration of a command or event
- collision of a type with implicit 'Kind' enum for a union
- collision with an implicit MAX enum constant

Since the c_type() function in the generator treats all names
as being in the same namespace, this patch adds a global array
to track all known names and their source, to prevent collisions
before it can cause further problems.  While valid .json files
won't trigger any of these cases, we might as well be nicer to
developers that make a typo while trying to add new QAPI code.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Add tests of redefined expressions
Eric Blake [Mon, 4 May 2015 15:05:16 +0000 (09:05 -0600)]
qapi: Add tests of redefined expressions

Demonstrate that the qapi generator doesn't deal very well with
redefined expressions.  At the parse level, they are silently
accepted; and while the testsuite just stops at parsing, I've
further tested that many of them cause generator crashes or
invalid C code if they were appended to qapi-schema-test.json.
A later patch will tighten things up and adjust the testsuite
to match.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Better error messages for bad expressions
Eric Blake [Mon, 4 May 2015 15:05:15 +0000 (09:05 -0600)]
qapi: Better error messages for bad expressions

The previous commit demonstrated that the generator overlooked some
fairly basic broken expressions:
- missing metataype
- metatype key has a non-string value
- unknown key in relation to the metatype
- conflicting metatype (this patch treats the second metatype as an
unknown key of the first key visited, which is not necessarily the
first key the user typed)

Add check_keys to cover these situations, and update testcases to
match.  A couple other tests (enum-missing-data, indented-expr) had
to change since the validation added here occurs so early.
Conversely, changes to ident-with-escape results show that we still
have problems where our handling of escape sequences differs from
true JSON, which will matter down the road if we allow arbitrary
default string values for optional parameters (but for now is not
too bad, as we currently can avoid unicode escaping as we don't
need to represent anything beyond C identifier material).

While valid .json files won't trigger any of these cases, we might
as well be nicer to developers that make a typo while trying to add
new QAPI code.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Add some expr tests
Eric Blake [Mon, 4 May 2015 15:05:14 +0000 (09:05 -0600)]
qapi: Add some expr tests

Demonstrate that the qapi generator doesn't deal well with
expressions that aren't up to par. Later patches will improve
the expected results as the generator is made stricter.  Only
a few of the the added tests actually behave sanely at
rejecting obvious problems or demonstrating success.

Note that in some cases, we reject bad QAPI merely because our
pseudo-JSON parser does not yet know how to parse numbers.  This
series does not address that, but when a later series adds support
for numeric defaults of integer fields, the testsuite will ensure
that we don't lose the error (and hopefully that the error
message quality is improved).

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Use 'alternate' to replace anonymous union
Eric Blake [Mon, 4 May 2015 15:05:13 +0000 (09:05 -0600)]
qapi: Use 'alternate' to replace anonymous union

Previous patches have led up to the point where I create the
new meta-type "'alternate':'Foo'".  See the previous patches
for documentation; I intentionally split as much work into
earlier patches to minimize the size of this patch, but a lot
of it is churn due to testsuite fallout after updating to the
new type.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Document new 'alternate' meta-type
Eric Blake [Mon, 4 May 2015 15:05:12 +0000 (09:05 -0600)]
qapi: Document new 'alternate' meta-type

The next patch will quit special-casing "'union':'Foo',
'discriminator':{}" and instead use "'alternate':'Foo'".

Separating docs from implementation makes it easier to focus
on wording without holding up code.  In particular, making
alternate a separate type makes for a nice type hierarchy:

          /-------- meta-type ------\
         /              |            \
    simple types    alternate     complex types
    |         |                   |           |
 built-in   enum             type(struct)   union
 |       \    /                            /    \
numeric  string                         simple  flat

A later patch will then clean up 'type' vs. 'struct'
confusion.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Rename anonymous union type in test
Eric Blake [Mon, 4 May 2015 15:05:11 +0000 (09:05 -0600)]
qapi: Rename anonymous union type in test

Reduce churn in the future patch that replaces anonymous unions
with a new metatype 'alternate' by changing 'AnonUnion' to
'Alternate'.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Segregate anonymous unions into alternates in generator
Eric Blake [Mon, 4 May 2015 15:05:10 +0000 (09:05 -0600)]
qapi: Segregate anonymous unions into alternates in generator

Special-casing 'discriminator == {}' for handling anonymous unions
is getting awkward; since this particular type is not always a
dictionary on the wire, it is easier to treat it as a completely
different class of type, "alternate", so that if a type is listed
in the union_types array, we know it is not an anonymous union.

This patch just further segregates union handling, to make sure that
anonymous unions are not stored in union_types, and splitting up
check_union() into separate functions.  A future patch will change
the qapi grammar, and having the segregation already in place will
make it easier to deal with the distinct meta-type.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Prepare for catching more semantic parse errors
Eric Blake [Mon, 4 May 2015 15:05:09 +0000 (09:05 -0600)]
qapi: Prepare for catching more semantic parse errors

This patch widens the scope of a try block (with the attending
reindentation required by Python) in preparation for a future
patch adding more instances of QAPIExprError inside the block.
It's easier to separate indentation from semantic changes, so
this patch has no real behavior change.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Tighten checking of unions
Eric Blake [Mon, 4 May 2015 15:05:08 +0000 (09:05 -0600)]
qapi: Tighten checking of unions

Previous commits demonstrated that the generator had several
flaws with less-than-perfect unions:
- a simple union that listed the same branch twice (or two variant
names that map to the same C enumerator, including the implicit
MAX sentinel) ended up generating invalid C code
- an anonymous union that listed two branches with the same qtype
ended up generating invalid C code
- the generator crashed on anonymous union attempts to use an
array type
- the generator was silently ignoring a base type for anonymous
unions
- the generator allowed unknown types or nested anonymous unions
as a branch in an anonymous union

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Forbid base without discriminator in unions
Eric Blake [Mon, 4 May 2015 15:05:07 +0000 (09:05 -0600)]
qapi: Forbid base without discriminator in unions

None of the existing QMP or QGA interfaces uses a union with a
base type but no discriminator; it is easier to avoid this in the
generator to save room for other future extensions more likely to
be useful.  An earlier commit added a union-base-no-discriminator
test to ensure that we eventually give a decent error message;
likewise, removing UserDefUnion outright is okay, because we moved
all the tests we wish to keep into the tests of the simple union
UserDefNativeListUnion in the previous commit.  Now is the time to
actually forbid simple union with base, and remove the last
vestiges from the testsuite.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Clean up test coverage of simple unions
Eric Blake [Mon, 4 May 2015 15:05:06 +0000 (09:05 -0600)]
qapi: Clean up test coverage of simple unions

The tests of UserDefNativeListUnion serve to validate code
generation of simple unions without a base type, except that it
did not have full coverage in the strict test.  The next commits
will remove tests and support for simple unions with a base type,
so there is no real loss at repurposing that test here as
opposed to churn of adding a new test then deleting the old one.

Fix some indentation and long lines while at it.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Add some union tests
Eric Blake [Mon, 4 May 2015 15:05:05 +0000 (09:05 -0600)]
qapi: Add some union tests

Demonstrate that the qapi generator doesn't deal well with unions
that aren't up to par. Later patches will update the expected
reseults as the generator is made stricter.  A few tests work
as planned, but most show poor or missing error messages.

Of particular note, qapi-code-gen.txt documents 'base' only for
flat unions, but the tests here demonstrate that we currently allow
a 'base' to a simple union, although it is exercised only in the
testsuite.  Later patches will remove this undocumented feature, to
give us more flexibility in adding other future extensions to union
types.  For example, one possible extension is the idea of a
type-safe simple enum, where added fields tie the discriminator to
a user-defined enum type rather than creating an implicit enum from
the names in 'data'.  But adding such safety on top of a simple
enum with a base type could look ambiguous with a flat enum;
besides, the documentation also mentions how any simple union can
be represented by an equivalent flat union.  So it will be simpler
to just outlaw support for something we aren't using.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Better error messages for bad enums
Eric Blake [Mon, 4 May 2015 15:05:04 +0000 (09:05 -0600)]
qapi: Better error messages for bad enums

The previous commit demonstrated that the generator had several
flaws with less-than-perfect enums:
- an enum that listed the same string twice (or two variant
strings that map to the same C enumerator) ended up generating
an invalid C enum
- because the generator adds a _MAX terminator to each enum,
the use of an enum member 'max' can also cause this clash
- if an enum omits 'data', the generator left a python stack
trace rather than a graceful message
- an enum that used a non-array 'data' was silently accepted by
the parser
- an enum that used non-string members in the 'data' member
was silently accepted by the parser

Add check_enum to cover these situations, and update testcases
to match.  While valid .json files won't trigger any of these
cases, we might as well be nicer to developers that make a typo
while trying to add new QAPI code.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Add some enum tests
Eric Blake [Mon, 4 May 2015 15:05:03 +0000 (09:05 -0600)]
qapi: Add some enum tests

Demonstrate that the qapi generator doesn't deal well with enums
that aren't up to par. Later patches will update the expected
results as the generator is made stricter.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Require ASCII in schema
Eric Blake [Mon, 4 May 2015 15:05:02 +0000 (09:05 -0600)]
qapi: Require ASCII in schema

Python 2 and Python 3 have a wild history of whether strings
default to ascii or unicode, where Python 3 requires checking
isinstance(foo, basestr) to cover all strings, but where that
code is not portable to Python 2.  It's simpler to just state
that we don't care about Unicode strings, and to just always
use the simpler isinstance(foo, str) everywhere.

I'm no python expert, so I'm basing it on this conversation:
https://lists.gnu.org/archive/html/qemu-devel/2014-09/msg05278.html

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Fix generation of 'size' builtin type
Eric Blake [Mon, 4 May 2015 15:05:01 +0000 (09:05 -0600)]
qapi: Fix generation of 'size' builtin type

We were missing the 'size' builtin type (which means that QAPI using
[ 'size' ] would fail to compile).

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Simplify builtin type handling
Eric Blake [Mon, 4 May 2015 15:05:00 +0000 (09:05 -0600)]
qapi: Simplify builtin type handling

There was some redundancy between builtin_types[] and
builtin_type_qtypes{}.  Merge them into one.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Document type-safety considerations
Eric Blake [Mon, 4 May 2015 15:04:59 +0000 (09:04 -0600)]
qapi: Document type-safety considerations

Go into more details about the various types of valid expressions
in a qapi schema, including tweaks to document fixes being done
later in the current patch series.  Also fix some stale and missing
documentation in the QMP specification.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agoqapi: Add copyright declaration on docs
Eric Blake [Mon, 4 May 2015 15:04:58 +0000 (09:04 -0600)]
qapi: Add copyright declaration on docs

While our top-level COPYING with its GPLv2+ license applies to
any documentation file that omits explicit instructions, these
days it's better to be a good example of calling out our
intentions.  Correct use of GPL requires the use of a copyright
statement, so I'm adding notice to two QAPI documents, by
attributing these files to the initial authors and major
contributors.  I used:

$ git blame --line-porcelain $file \
  | sed -n 's/^author //p' | sort | uniq -c | sort -rn

to determine authorship of these two files.  qmp-spec.txt blames
entirely to Red Hat (easy, since my contribution falls in that
category); while qapi-code-gen.txt has multiple contributors
representing multiple entities.  But since it was originally
supplied by Michael Roth, the notice I added there copies the
notice he has used in other files.  As there is no intended
change in license from the implicit one previously present from
the top level, I have not bothered to CC other contributors;
if we want to weaken things to something looser (such as LGPL)
so that there is no question that someone re-implementing the
spec is not forced to use GPL, that would be a different commit.

CC: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
9 years agotcg: optimise memory layout of TCGTemp
Emilio G. Cota [Fri, 3 Apr 2015 00:07:53 +0000 (20:07 -0400)]
tcg: optimise memory layout of TCGTemp

This brings down the size of the struct from 56 to 32 bytes on 64-bit,
and to 20 bytes on 32-bit. This leads to memory savings:

Before:
$ find . -name 'tcg.o' | xargs size
   text    data     bss     dec     hex filename
  41131   29800      88   71019   1156b ./aarch64-softmmu/tcg/tcg.o
  37969   29416      96   67481   10799 ./x86_64-linux-user/tcg/tcg.o
  39354   28816      96   68266   10aaa ./arm-linux-user/tcg/tcg.o
  40802   29096      88   69986   11162 ./arm-softmmu/tcg/tcg.o
  39417   29672      88   69177   10e39 ./x86_64-softmmu/tcg/tcg.o

After:
$ find . -name 'tcg.o' | xargs size
   text    data     bss     dec     hex filename
  40883   29800      88   70771   11473 ./aarch64-softmmu/tcg/tcg.o
  37473   29416      96   66985   105a9 ./x86_64-linux-user/tcg/tcg.o
  38858   28816      96   67770   108ba ./arm-linux-user/tcg/tcg.o
  40554   29096      88   69738   1106a ./arm-softmmu/tcg/tcg.o
  39169   29672      88   68929   10d41 ./x86_64-softmmu/tcg/tcg.o

Note that using an entire byte for some enums that need less than
that wastes a few bits (noticeable in 32 bits, where we use
20 bytes instead of 16) but avoids extraction code, which overall
is a win--I've tested several variations of the patch, and the appended
is the best performer for OpenSSL's bntest by a very small margin:

Before:
$ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null
[...]
 Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs):

      10538.479833 task-clock (msec)  # 0.999 CPUs utilized  ( +-  0.38% )
               772 context-switches   # 0.073 K/sec          ( +-  2.03% )
                 0 cpu-migrations     # 0.000 K/sec          ( +-100.00% )
             2,207 page-faults        # 0.209 K/sec          ( +-  0.08% )
      10.552871687 seconds time elapsed                      ( +-  0.39% )

After:
$ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null
 Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs):

      10459.968847 task-clock (msec)  # 0.999 CPUs utilized  ( +-  0.30% )
               739 context-switches   # 0.071 K/sec          ( +-  1.71% )
                 0 cpu-migrations     # 0.000 K/sec          ( +- 68.14% )
             2,204 page-faults        # 0.211 K/sec          ( +-  0.10% )
      10.473900411 seconds time elapsed                      ( +-  0.30% )

Suggested-by: Stefan Weil <sw@weilnetz.de>
Suggested-by: Richard Henderson <rth@twiddle.net>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <rth@twiddle.net>