qemu.git
6 years agoRevert "tests: add qmp/qom-set-without-value test"
Peter Maydell [Mon, 24 Sep 2018 13:04:40 +0000 (14:04 +0100)]
Revert "tests: add qmp/qom-set-without-value test"

Markus spotted some issues with this new test case which
unfortunately I didn't notice had been flagged until after
I'd applied the pull request. Revert the relevant commit.

This reverts commit 2b70ea92766f5a1a735a44e28c92cdfba3c4054f.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/mcayland/tags/qemu-openbios.for-upstream-201808...
Peter Maydell [Mon, 24 Sep 2018 12:28:38 +0000 (13:28 +0100)]
Merge remote-tracking branch 'remotes/mcayland/tags/qemu-openbios.for-upstream-20180831' into staging

qemu-openbios.for-upstream queue

# gpg: Signature made Fri 31 Aug 2018 12:26:55 BST
# gpg:                using RSA key 5BC2C56FAE0F321F
# gpg: Good signature from "Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>"
# Primary key fingerprint: CC62 1AB9 8E82 200D 915C  C9C4 5BC2 C56F AE0F 321F

* remotes/mcayland/tags/qemu-openbios.for-upstream-20180831:
  Update OpenBIOS images to a1280807 built from submodule.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/huth-gitlab/tags/pull-request-2018-08-31' into...
Peter Maydell [Mon, 24 Sep 2018 11:28:36 +0000 (12:28 +0100)]
Merge remote-tracking branch 'remotes/huth-gitlab/tags/pull-request-2018-08-31' into staging

Removal of deprecated options and improvements for the qtests

# gpg: Signature made Fri 31 Aug 2018 09:10:23 BST
# gpg:                using RSA key 2ED9D774FE702DB5
# gpg: Good signature from "Thomas Huth <th.huth@gmx.de>"
# gpg:                 aka "Thomas Huth <thuth@redhat.com>"
# gpg:                 aka "Thomas Huth <huth@tuxfamily.org>"
# gpg:                 aka "Thomas Huth <th.huth@posteo.de>"
# Primary key fingerprint: 27B8 8847 EEE0 2501 18F3  EAB9 2ED9 D774 FE70 2DB5

* remotes/huth-gitlab/tags/pull-request-2018-08-31:
  tests: add a qmp success-response test
  tests: add qmp/qom-set-without-value test
  tests: add qmp/object-add-without-props test
  tests: add qmp_assert_error_class()
  tests/libqos: Utilize newer glib spawn check
  net: Remove the deprecated -tftp, -bootp, -redir and -smb options
  Remove the deprecated options -startdate, -localtime and -rtc-td-hack
  Remove the deprecated -nodefconfig option
  Remove the deprecated -balloon option

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/cohuck/tags/s390x-20180829' into staging
Peter Maydell [Mon, 24 Sep 2018 10:43:00 +0000 (11:43 +0100)]
Merge remote-tracking branch 'remotes/cohuck/tags/s390x-20180829' into staging

- various fixes and improvements in the tcg code
- split off the individual virtio-ccw devices into separate files

# gpg: Signature made Wed 29 Aug 2018 10:38:03 BST
# gpg:                using RSA key DECF6B93C6F02FAF
# gpg: Good signature from "Cornelia Huck <conny@cornelia-huck.de>"
# gpg:                 aka "Cornelia Huck <huckc@linux.vnet.ibm.com>"
# gpg:                 aka "Cornelia Huck <cornelia.huck@de.ibm.com>"
# gpg:                 aka "Cornelia Huck <cohuck@kernel.org>"
# gpg:                 aka "Cornelia Huck <cohuck@redhat.com>"
# Primary key fingerprint: C3D0 D66D C362 4FF6 A8C0  18CE DECF 6B93 C6F0 2FAF

* remotes/cohuck/tags/s390x-20180829:
  target/s390x: use regular spaces in translate.c
  hw/s390x: Move virtio-ccw-blk code to a separate file
  hw/s390x: Move virtio-ccw-net code to a separate file
  hw/s390x: Move virtio-ccw-input code to a separate file
  hw/s390x: Move virtio-ccw-gpu code to a separate file
  hw/s390x: Move vhost-vsock-ccw code to a separate file
  hw/s390x: Move virtio-ccw-crypto code to a separate file
  hw/s390x: Move virtio-ccw-9p code to a separate file
  hw/s390x: Move virtio-ccw-rng code to a separate file
  hw/s390x: Move virtio-ccw-scsi code to a separate file
  hw/s390x: Move virtio-ccw-balloon code to a separate file
  hw/s390x: Move virtio-ccw-serial code to a separate file
  hw/s390x/virtio-ccw: Consolidate calls to virtio_ccw_unrealize()
  target/s390x: fix PACK reading 1 byte less and writing 1 byte more
  target/s390x: add EX support for TRT and TRTR
  target/s390x: fix IPM polluting irrelevant bits
  target/s390x: fix CSST decoding and runtime alignment check
  target/s390x: add BAL and BALR instructions
  tests/tcg: add a simple s390x test

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2018-08-28' into staging
Peter Maydell [Mon, 24 Sep 2018 09:46:33 +0000 (10:46 +0100)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2018-08-28' into staging

QAPI patches for 2018-08-28

# gpg: Signature made Tue 28 Aug 2018 17:23:32 BST
# gpg:                using RSA key 3870B400EB918653
# gpg: Good signature from "Markus Armbruster <armbru@redhat.com>"
# gpg:                 aka "Markus Armbruster <armbru@pond.sub.org>"
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867  4E5F 3870 B400 EB91 8653

* remotes/armbru/tags/pull-qapi-2018-08-28:
  qapi: Add comments to aid debugging generated introspection
  qapi: Minor introspect.py cleanups
  qapi: Update docs for generator changes since commit 9ee86b85267
  qapi: Emit a blank line before dummy declaration
  qapi: Drop qapi_event_send_FOO()'s Error ** argument
  qapi: Fix build_params() for empty parameter list

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoUpdate OpenBIOS images to a1280807 built from submodule.
Mark Cave-Ayland [Fri, 31 Aug 2018 11:19:07 +0000 (12:19 +0100)]
Update OpenBIOS images to a1280807 built from submodule.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
6 years agotests: add a qmp success-response test
Marc-André Lureau [Thu, 30 Aug 2018 15:58:10 +0000 (17:58 +0200)]
tests: add a qmp success-response test

Verify the usage of this schema feature and the API behaviour.  This
should be the only case where qmp_dispatch() returns NULL.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agotests: add qmp/qom-set-without-value test
Marc-André Lureau [Thu, 30 Aug 2018 15:58:09 +0000 (17:58 +0200)]
tests: add qmp/qom-set-without-value test

test_qom_set_without_value() is about a bug in infrastructure used by
the QMP core, fixed in commit c489780203.  We covered the bug in
infrastructure unit tests (commit bce3035a44).  I wrote that test
earlier, to cover QMP level as well, the test could go into qmp-test.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agotests: add qmp/object-add-without-props test
Marc-André Lureau [Thu, 30 Aug 2018 15:58:08 +0000 (17:58 +0200)]
tests: add qmp/object-add-without-props test

test_object_add_without_props() tests a bug in qmp_object_add() we
fixed in commit e64c75a975.  Sadly, we don't have systematic
object-add tests.  This lone test can go into qmp-cmd-test for want of
a better home.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agotests: add qmp_assert_error_class()
Marc-André Lureau [Thu, 30 Aug 2018 15:58:07 +0000 (17:58 +0200)]
tests: add qmp_assert_error_class()

This helper will simplify a bunch of code checking for QMP errors and
can be shared by various tests.  Note that test-qga does check for
error description as well, so don't replace the code there for now.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agotests/libqos: Utilize newer glib spawn check
Eric Blake [Tue, 21 Aug 2018 19:05:16 +0000 (14:05 -0500)]
tests/libqos: Utilize newer glib spawn check

During development, I got a 'make check' failure that claimed:

qemu-img returned status code 32512
**
ERROR:tests/libqos/libqos.c:202:mkimg: assertion failed: (!rc)

But 32512 is too big for a normal exit status value, which means we
failed to use WEXITSTATUS() to shift the bits to the desired value
for printing.  However, instead of worrying about how to portably
parse g_spawn()'s rc in the proper platform-dependent manner, it's
better to just rely on the fact that we now require glib 2.40 (since
commit e7b3af815) and can therefore use glib's portable checker
instead, where the message under my same condition improves to:

Child process exited with code 127
**
ERROR:tests/libqos/libqos.c:192:mkimg: assertion failed: (ret && !err)

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agonet: Remove the deprecated -tftp, -bootp, -redir and -smb options
Thomas Huth [Wed, 22 Aug 2018 13:43:30 +0000 (15:43 +0200)]
net: Remove the deprecated -tftp, -bootp, -redir and -smb options

These options likely do not work as expected as soon as the user
tries to use more than one network interface at once. The parameters
have been marked as deprecated since QEMU v2.6, so users had plenty
of time to move their scripts to the new syntax. Time to remove the
old parameters now.

Reviewed-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Acked-by: Peter Krempa <pkrempa@redhat.com>
Acked-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agoRemove the deprecated options -startdate, -localtime and -rtc-td-hack
Thomas Huth [Tue, 21 Aug 2018 13:18:06 +0000 (15:18 +0200)]
Remove the deprecated options -startdate, -localtime and -rtc-td-hack

Deprecated since two releases, nobody complained, thus it's time to
remove them now.

Acked-by: Peter Krempa <pkrempa@redhat.com>
Acked-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agoRemove the deprecated -nodefconfig option
Thomas Huth [Tue, 21 Aug 2018 10:59:56 +0000 (12:59 +0200)]
Remove the deprecated -nodefconfig option

It's the same as -no-user-config and marked as deprecated since three
releases already. Time to remove it now.

Acked-by: Peter Krempa <pkrempa@redhat.com>
Acked-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agoRemove the deprecated -balloon option
Thomas Huth [Tue, 21 Aug 2018 09:58:47 +0000 (11:58 +0200)]
Remove the deprecated -balloon option

The "-balloon" option has been replaced by "-device virtio-balloon".
It's been marked as deprecated since two releases, and nobody
complained, so let's remove it now.

Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Peter Krempa <pkrempa@redhat.com>
Acked-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
6 years agoqapi: Add comments to aid debugging generated introspection
Eric Blake [Mon, 27 Aug 2018 21:39:43 +0000 (16:39 -0500)]
qapi: Add comments to aid debugging generated introspection

We consciously chose in commit 1a9a507b to hide QAPI type names
from the introspection output on the wire, but added a command
line option -u to unmask the type name when doing a debug build.
The unmask option still remains useful to some other forms of
automated analysis, so it will not be removed; however, when it
is not in use, the generated .c file can be hard to read.  At
the time when we first introduced masking, the generated file
consisted only of a monolithic C string, so there was no clean
way to inject any comments.

Later, in commit 7d0f982b, we switched the generation to output
a QLit object, in part to make it easier for future addition of
conditional compilation.  In fact, commit d626b6c1 took advantage
of this by passing a tuple instead of a bare object for encoding
the output of conditionals.  By extending that tuple, we can now
interject strategic comments.

For now, type name debug aid comments are only output once per
meta-type, rather than at all uses of the number used to encode
the type within the introspection data.  But this is still a lot
more convenient than having to regenerate the file with the
unmask operation temporarily turned on - merely search the
generated file for '"NNN" =' to learn the corresponding source
name and associated definition of type NNN.

The generated qapi-introspect.c changes only with the addition
of comments, such as:

| @@ -14755,6 +15240,7 @@
|          { "name", QLIT_QSTR("[485]"), },
|          {}
|      })),
| +    /* "485" = QCryptoBlockInfoLUKSSlot */
|      QLIT_QDICT(((QLitDictEntry[]) {
|          { "members", QLIT_QLIST(((QLitObject[]) {
|              QLIT_QDICT(((QLitDictEntry[]) {

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180827213943.33524-3-eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
[Rebased, update to qapi-code-gen.txt corrected]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
6 years agoqapi: Minor introspect.py cleanups
Eric Blake [Mon, 27 Aug 2018 21:39:42 +0000 (16:39 -0500)]
qapi: Minor introspect.py cleanups

Commit 7d0f982b changed generated introspection output to no longer
produce long lines in the generated .c file, but failed to adjust
comments to match.  Add some clarity that the shorter length that
matters most is the overall QMP response on the wire.

Commit 25b1ef31 triggers a pep8 formatting nit.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180827213943.33524-2-eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
6 years agoqapi: Update docs for generator changes since commit 9ee86b85267
Markus Armbruster [Tue, 28 Aug 2018 12:07:36 +0000 (14:07 +0200)]
qapi: Update docs for generator changes since commit 9ee86b85267

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180828120736.32323-3-armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
6 years agoqapi: Emit a blank line before dummy declaration
Markus Armbruster [Tue, 28 Aug 2018 12:07:35 +0000 (14:07 +0200)]
qapi: Emit a blank line before dummy declaration

We emit a dummy variable in each .c file "to shut up OSX toolchain
warnings about empty .o files" (commit 252dc3105fc).  Separate it from
the code preceding it (if any) with a blank line.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180828120736.32323-2-armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
6 years agoqapi: Drop qapi_event_send_FOO()'s Error ** argument
Peter Xu [Wed, 15 Aug 2018 13:37:37 +0000 (21:37 +0800)]
qapi: Drop qapi_event_send_FOO()'s Error ** argument

The generated qapi_event_send_FOO() take an Error ** argument.  They
can't actually fail, because all they do with the argument is passing it
to functions that can't fail: the QObject output visitor, and the
@qmp_emit callback, which is either monitor_qapi_event_queue() or
event_test_emit().

Drop the argument, and pass &error_abort to the QObject output visitor
and @qmp_emit instead.

Suggested-by: Eric Blake <eblake@redhat.com>
Suggested-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180815133747.25032-4-peterx@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
[Commit message rewritten, update to qapi-code-gen.txt corrected]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
6 years agoqapi: Fix build_params() for empty parameter list
Markus Armbruster [Wed, 15 Aug 2018 13:37:36 +0000 (21:37 +0800)]
qapi: Fix build_params() for empty parameter list

build_params() returns '' instead of 'void' when there are no
parameters.  Can't happen now, but the next commit will change that.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
[peterx: compose the patch from email replies]
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180815133747.25032-3-peterx@redhat.com>

6 years agotarget/s390x: use regular spaces in translate.c
Pavel Zbitskiy [Wed, 22 Aug 2018 14:40:37 +0000 (10:40 -0400)]
target/s390x: use regular spaces in translate.c

In a few places translate.c contains non-breaking spaces (0xc2 0xa0)
instead of regular ones (0x20):

  7c 7c c2 a0 63 63
  7c 7c 20    63 63
  |  |        c  c

This confuses some text editors.

Signed-off-by: Pavel Zbitskiy <pavel.zbitskiy@gmail.com>
Message-Id: <20180822144039.5796-2-pavel.zbitskiy@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-blk code to a separate file
Thomas Huth [Wed, 25 Jul 2018 18:08:30 +0000 (20:08 +0200)]
hw/s390x: Move virtio-ccw-blk code to a separate file

The code should only be enabled if CONFIG_VIRTIO_BLK has been set.
This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532542110-9017-1-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-net code to a separate file
Thomas Huth [Wed, 25 Jul 2018 18:07:36 +0000 (20:07 +0200)]
hw/s390x: Move virtio-ccw-net code to a separate file

The code should only be enabled if CONFIG_VIRTIO_NET has been set.
This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532542056-8927-1-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-input code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:24 +0000 (14:20 +0200)]
hw/s390x: Move virtio-ccw-input code to a separate file

The code should only be enabled if CONFIG_VIRTIO_INPUT has been set.
This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-11-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-gpu code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:23 +0000 (14:20 +0200)]
hw/s390x: Move virtio-ccw-gpu code to a separate file

The code should only be enabled if CONFIG_VIRTIO_GPU has been set. This
can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-10-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move vhost-vsock-ccw code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:22 +0000 (14:20 +0200)]
hw/s390x: Move vhost-vsock-ccw code to a separate file

The code should only be enabled if CONFIG_VHOST_VSOCK has been set.
This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-9-git-send-email-thuth@redhat.com>
[CH: updated MAINTAINERS]
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-crypto code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:21 +0000 (14:20 +0200)]
hw/s390x: Move virtio-ccw-crypto code to a separate file

The code should only be enabled if CONFIG_VIRTIO_CRYPTO has been set.
This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-8-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-9p code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:20 +0000 (14:20 +0200)]
hw/s390x: Move virtio-ccw-9p code to a separate file

The code should only be enabled if CONFIG_VIRTIO_9P and CONFIG_VIRTFS
have been set. This can be done best if the code resides in a separate
file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-7-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-rng code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:19 +0000 (14:20 +0200)]
hw/s390x: Move virtio-ccw-rng code to a separate file

The code should only be enabled if CONFIG_VIRTIO_RNG has been set.
This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-6-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-scsi code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:18 +0000 (14:20 +0200)]
hw/s390x: Move virtio-ccw-scsi code to a separate file

The code should only be enabled if CONFIG_VIRTIO_SCSI has been set.
This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-5-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-balloon code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:17 +0000 (14:20 +0200)]
hw/s390x: Move virtio-ccw-balloon code to a separate file

The code should only be enabled if CONFIG_VIRTIO_BALLOON has been
set. This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-4-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x: Move virtio-ccw-serial code to a separate file
Thomas Huth [Wed, 25 Jul 2018 12:20:16 +0000 (14:20 +0200)]
hw/s390x: Move virtio-ccw-serial code to a separate file

The code should only be enabled if CONFIG_VIRTIO_SERIAL has been set.
This can be done best if the code resides in a separate file.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-3-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agohw/s390x/virtio-ccw: Consolidate calls to virtio_ccw_unrealize()
Thomas Huth [Wed, 25 Jul 2018 12:20:15 +0000 (14:20 +0200)]
hw/s390x/virtio-ccw: Consolidate calls to virtio_ccw_unrealize()

Currently, every virtio-ccw device explicitely sets its unrealize
function to virtio_ccw_unrealize() in its class_init function.
We can simplify this by using a common unrealize function, just like
it is already done for the realize functions.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1532521224-27235-2-git-send-email-thuth@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agotarget/s390x: fix PACK reading 1 byte less and writing 1 byte more
Pavel Zbitskiy [Tue, 21 Aug 2018 02:51:03 +0000 (22:51 -0400)]
target/s390x: fix PACK reading 1 byte less and writing 1 byte more

PACK fails on the test from the Principles of Operation: F1F2F3F4
becomes 0000234C instead of 0001234C due to an off-by-one error.
Furthermore, it overwrites one extra byte to the left of F1.

If len_dest is 0, then we only want to flip the 1st byte and never loop
over the rest. Therefore, the loop condition should be > and not >=.

If len_src is 1, then we should flip the 1st byte and pack the 2nd.
Since len_src is already decremented before the loop, the first
condition should be >=, and not >.

Likewise for len_src == 2 and the second condition.

Signed-off-by: Pavel Zbitskiy <pavel.zbitskiy@gmail.com>
Message-Id: <20180821025104.19604-7-pavel.zbitskiy@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agotarget/s390x: add EX support for TRT and TRTR
Pavel Zbitskiy [Tue, 21 Aug 2018 02:51:02 +0000 (22:51 -0400)]
target/s390x: add EX support for TRT and TRTR

Improves "b213c9f5: target/s390x: Implement TRTR" by introducing the
intermediate functions, which are compatible with dx_helper type.

Signed-off-by: Pavel Zbitskiy <pavel.zbitskiy@gmail.com>
Message-Id: <20180821025104.19604-6-pavel.zbitskiy@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agotarget/s390x: fix IPM polluting irrelevant bits
Pavel Zbitskiy [Tue, 21 Aug 2018 02:51:01 +0000 (22:51 -0400)]
target/s390x: fix IPM polluting irrelevant bits

Suppose psw.mask=0x0000000080000000, cc=2, r1=0 and we do "ipm 1".
This command must touch only bits 32-39, so the expected output
is r1=0x20000000. However, currently qemu yields r1=0x20008000,
because irrelevant parts of PSW leak into r1 during program mask
transfer.

Signed-off-by: Pavel Zbitskiy <pavel.zbitskiy@gmail.com>
Message-Id: <20180821025104.19604-5-pavel.zbitskiy@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agotarget/s390x: fix CSST decoding and runtime alignment check
Pavel Zbitskiy [Tue, 21 Aug 2018 02:51:00 +0000 (22:51 -0400)]
target/s390x: fix CSST decoding and runtime alignment check

CSST is defined as:

    C(0xc802, CSST,    SSF,   CASS, la1, a2, 0, 0, csst, 0)

It means that the first parameter is handled by in1_la1().
in1_la1() fills addr1 field, and not in1.

Furthermore, when extract32() is used for the alignment check, the
third parameter should specify the number of trailing bits that must
be 0. For FC these numbers are:

    FC=0 (word, 4 bytes):        2
    FC=1 (double word, 8 bytes): 3
    FC=2 (quad word, 16 bytes):  4

For SC these numbers correspond to the size:

    SC=0: 0
    SC=1: 1
    SC=2: 2
    SC=3: 3
    SC=4: 4

Signed-off-by: Pavel Zbitskiy <pavel.zbitskiy@gmail.com>
Message-Id: <20180821025104.19604-4-pavel.zbitskiy@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agotarget/s390x: add BAL and BALR instructions
Pavel Zbitskiy [Tue, 21 Aug 2018 02:50:59 +0000 (22:50 -0400)]
target/s390x: add BAL and BALR instructions

These instructions are provided for compatibility purposes and are
used only by old software, in the new code BAS and BASR are preferred.
The difference between the old and new instruction exists only in the
24-bit mode.

In addition, fix BAS polluting high 32 bits of the first operand in
24- and 31-bit addressing modes.

Signed-off-by: Pavel Zbitskiy <pavel.zbitskiy@gmail.com>
Message-Id: <20180821025104.19604-3-pavel.zbitskiy@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agotests/tcg: add a simple s390x test
Pavel Zbitskiy [Tue, 21 Aug 2018 02:50:58 +0000 (22:50 -0400)]
tests/tcg: add a simple s390x test

Copied from alpha.

Signed-off-by: Pavel Zbitskiy <pavel.zbitskiy@gmail.com>
Message-Id: <20180821025104.19604-2-pavel.zbitskiy@gmail.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
6 years agoMerge remote-tracking branch 'remotes/armbru/tags/pull-error-2018-08-27-v2' into...
Peter Maydell [Mon, 27 Aug 2018 15:44:20 +0000 (16:44 +0100)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-error-2018-08-27-v2' into staging

Error reporting patches for 2018-08-27

# gpg: Signature made Mon 27 Aug 2018 14:18:15 BST
# gpg:                using RSA key 3870B400EB918653
# gpg: Good signature from "Markus Armbruster <armbru@redhat.com>"
# gpg:                 aka "Markus Armbruster <armbru@pond.sub.org>"
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867  4E5F 3870 B400 EB91 8653

* remotes/armbru/tags/pull-error-2018-08-27-v2:
  intel-iommu: replace more vtd_err_* traces
  intel-iommu: start to use error_report_once
  qemu-error: introduce {error|warn}_report_once

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/juanquintela/tags/check/20180827' into staging
Peter Maydell [Mon, 27 Aug 2018 13:55:57 +0000 (14:55 +0100)]
Merge remote-tracking branch 'remotes/juanquintela/tags/check/20180827' into staging

check/next for 20180827

# gpg: Signature made Mon 27 Aug 2018 10:58:51 BST
# gpg:                using RSA key F487EF185872D723
# gpg: Good signature from "Juan Quintela <quintela@redhat.com>"
# gpg:                 aka "Juan Quintela <quintela@trasno.org>"
# Primary key fingerprint: 1899 FF8E DEBF 58CC EE03  4B82 F487 EF18 5872 D723

* remotes/juanquintela/tags/check/20180827:
  check: Move wdt_ib700 test to common
  check: Move endianess test to common
  check: Move VMXNET3 test to common
  check: Only test boot-serial when sga is compiled in
  check: Only test ivshm when it is compiled in
  x86_64-softmmu: Configuration is identical to i386-softmmu
  check: Only test usb-xhci-nec when it is compiled in
  check: Only test isa-testdev when it is compiled in
  configure: We don't want to clean configuration files

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agointel-iommu: replace more vtd_err_* traces
Peter Xu [Wed, 15 Aug 2018 09:53:28 +0000 (17:53 +0800)]
intel-iommu: replace more vtd_err_* traces

Replace all the trace_vtd_err_*() hooks with the new error_report_once()
since they are similar to trace_vtd_err() - dumping the first error
would be mostly enough, then we have them on by default too.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180815095328.32414-4-peterx@redhat.com>
[Use "%x" instead of "%" PRIx16 to print uint16_t, whitespace tidied up]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
6 years agointel-iommu: start to use error_report_once
Peter Xu [Wed, 15 Aug 2018 09:53:27 +0000 (17:53 +0800)]
intel-iommu: start to use error_report_once

Replace existing trace_vtd_err() with error_report_once() then stderr
will capture something if any of the error happens, meanwhile we don't
suffer from any DDOS.  Then remove the trace point.  Since at it,
provide more information where proper (now we can pass parameters into
the report function).

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180815095328.32414-3-peterx@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
[Two format strings fixed, whitespace tidied up]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
6 years agoMerge remote-tracking branch 'remotes/kraxel/tags/ui-20180827-v4-pull-request' into...
Peter Maydell [Mon, 27 Aug 2018 11:30:51 +0000 (12:30 +0100)]
Merge remote-tracking branch 'remotes/kraxel/tags/ui-20180827-v4-pull-request' into staging

ui: misc fixes which piled up during 3.0 release freeze

# gpg: Signature made Mon 27 Aug 2018 09:53:07 BST
# gpg:                using RSA key 4CB6D8EED3E87138
# 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>"
# Primary key fingerprint: A032 8CFF B93A 17A7 9901  FE7D 4CB6 D8EE D3E8 7138

* remotes/kraxel/tags/ui-20180827-v4-pull-request:
  util: promote qemu_egl_rendernode_open() to libqemuutil
  dmabuf: add y0_top, pass it to spice
  ui/vnc: Remove useless parenthesis around DIV_ROUND_UP macro
  ui/sdl2: Fix broken -full-screen CLI option
  spice-display: fix qemu_spice_cursor_refresh_bh locking
  spice-display: access ptr_x/ptr_y under Mutex
  vnc: remove support for deprecated tls, x509, x509verify options
  doc: switch to modern syntax for VNC TLS setup
  sdl2: redraw correctly when scanout_mode enabled.
  ui: use enum to string helpers
  vnc: fix memleak of the "vnc-worker-output" name
  ui/sdl2: Remove the obsolete SDL_INIT_NOPARACHUTE flag

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agocheck: Move wdt_ib700 test to common
Juan Quintela [Tue, 14 Aug 2018 12:11:34 +0000 (14:11 +0200)]
check: Move wdt_ib700 test to common

It is protected by CONFIG_WDT_IB700.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
6 years agocheck: Move endianess test to common
Juan Quintela [Tue, 14 Aug 2018 12:09:32 +0000 (14:09 +0200)]
check: Move endianess test to common

It is already protected by CONFIG_ISA_TESTDEV in all architectures.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
6 years agocheck: Move VMXNET3 test to common
Juan Quintela [Tue, 14 Aug 2018 12:07:22 +0000 (14:07 +0200)]
check: Move VMXNET3 test to common

We protect it with CONFIG_VMXNET3_PCI now, so no need to also put it
on i386.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
6 years agocheck: Only test boot-serial when sga is compiled in
Juan Quintela [Mon, 13 Aug 2018 11:01:42 +0000 (13:01 +0200)]
check: Only test boot-serial when sga is compiled in

This is only for x86* architecture.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
6 years agocheck: Only test ivshm when it is compiled in
Juan Quintela [Tue, 10 Jul 2018 21:39:01 +0000 (23:39 +0200)]
check: Only test ivshm when it is compiled in

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
6 years agox86_64-softmmu: Configuration is identical to i386-softmmu
Juan Quintela [Tue, 31 Jul 2018 15:38:14 +0000 (17:38 +0200)]
x86_64-softmmu: Configuration is identical to i386-softmmu

If we ever changed that, just make the things that are different
explicit.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
6 years agocheck: Only test usb-xhci-nec when it is compiled in
Juan Quintela [Tue, 10 Jul 2018 21:07:09 +0000 (23:07 +0200)]
check: Only test usb-xhci-nec when it is compiled in

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
6 years agocheck: Only test isa-testdev when it is compiled in
Juan Quintela [Tue, 10 Jul 2018 14:52:36 +0000 (16:52 +0200)]
check: Only test isa-testdev when it is compiled in

Once there, untangle endianness-test and boot-serial-test.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
---

boot-serial-test don't depend on isa-testdev.  Thanks Thomas.

6 years agoconfigure: We don't want to clean configuration files
Juan Quintela [Mon, 16 Jul 2018 08:41:30 +0000 (10:41 +0200)]
configure: We don't want to clean configuration files

If you don't want to compile everything, you configure
config-devices.mak.  And then make clean remove it, and make will
create a default one without your configuration.  Fix it by not
removing it on clean target.  Remove it instead on distclean.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
---

Remove it instead on distclean.

6 years agoutil: promote qemu_egl_rendernode_open() to libqemuutil
Marc-André Lureau [Fri, 13 Jul 2018 13:09:06 +0000 (15:09 +0200)]
util: promote qemu_egl_rendernode_open() to libqemuutil

vhost-user-gpu will share the same code to open a DRM node.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20180713130916.4153-20-marcandre.lureau@redhat.com>

[ kraxel: buildfix: util/drm.o must be CONFIG_OPENGL not CONFIG_LINUX ]

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
6 years agoqemu-error: introduce {error|warn}_report_once
Peter Xu [Wed, 15 Aug 2018 09:53:26 +0000 (17:53 +0800)]
qemu-error: introduce {error|warn}_report_once

There are many error_report()s that can be used in frequently called
functions, especially on IO paths.  That can be unideal in that
malicious guest can try to trigger the error tons of time which might
use up the log space on the host (e.g., libvirt can capture the stderr
of QEMU and put it persistently onto disk).  In VT-d emulation code, we
have trace_vtd_error() tracer.  AFAIU all those places can be replaced
by something like error_report() but trace points are mostly used to
avoid the DDOS attack that mentioned above.  However using trace points
mean that errors are not dumped if trace not enabled.

It's not a big deal in most modern server managements since we have
things like logrotate to maintain the logs and make sure the quota is
expected.  However it'll still be nice that we just provide another way
to restrict message generations.  In most cases, this kind of
error_report()s will only provide valid information on the first message
sent, and all the rest of similar messages will be mostly talking about
the same thing.  This patch introduces *_report_once() helpers to allow
a message to be dumped only once during one QEMU process's life cycle.
It will make sure: (1) it's on by deffault, so we can even get something
without turning the trace on and reproducing, and (2) it won't be
affected by DDOS attack.

To implement it, I stole the printk_once() macro from Linux.

CC: Eric Blake <eblake@redhat.com>
CC: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180815095328.32414-2-peterx@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
[Whitespace adjusted, comments improved]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
6 years agoMerge remote-tracking branch 'remotes/otubo/tags/pull-seccomp-20180823' into staging
Peter Maydell [Sat, 25 Aug 2018 12:08:57 +0000 (13:08 +0100)]
Merge remote-tracking branch 'remotes/otubo/tags/pull-seccomp-20180823' into staging

pull-seccomp-20180823

# gpg: Signature made Thu 23 Aug 2018 15:46:13 BST
# gpg:                using RSA key DF32E7C0F0FFF9A2
# gpg: Good signature from "Eduardo Otubo (Senior Software Engineer) <otubo@redhat.com>"
# Primary key fingerprint: D67E 1B50 9374 86B4 0723  DBAB DF32 E7C0 F0FF F9A2

* remotes/otubo/tags/pull-seccomp-20180823:
  seccomp: set the seccomp filter to all threads
  configure: require libseccomp 2.2.0
  seccomp: prefer SCMP_ACT_KILL_PROCESS if available
  seccomp: use SIGSYS signal instead of killing the thread

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/awilliam/tags/vfio-fixes-20180823.1' into staging
Peter Maydell [Sat, 25 Aug 2018 09:59:06 +0000 (10:59 +0100)]
Merge remote-tracking branch 'remotes/awilliam/tags/vfio-fixes-20180823.1' into staging

VFIO fixes 2018-08-23

 - Fix coverity reported issue with use of realpath (Alex Williamson)

 - Cleanup file descriptor in error path (Alex Williamson)

 - Fix postcopy use of new balloon inhibitor (Alex Williamson)

# gpg: Signature made Thu 23 Aug 2018 17:46:41 BST
# gpg:                using RSA key 239B9B6E3BB08B22
# gpg: Good signature from "Alex Williamson <alex.williamson@redhat.com>"
# gpg:                 aka "Alex Williamson <alex@shazbot.org>"
# gpg:                 aka "Alex Williamson <alwillia@redhat.com>"
# gpg:                 aka "Alex Williamson <alex.l.williamson@gmail.com>"
# Primary key fingerprint: 42F6 C04E 540B D1A9 9E7B  8A90 239B 9B6E 3BB0 8B22

* remotes/awilliam/tags/vfio-fixes-20180823.1:
  postcopy: Synchronize usage of the balloon inhibitor
  vfio/pci: Fix failure to close file descriptor on error
  vfio/pci: Handle subsystem realpath() returning NULL

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/armbru/tags/pull-qobject-2018-08-24' into staging
Peter Maydell [Sat, 25 Aug 2018 09:11:54 +0000 (10:11 +0100)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-qobject-2018-08-24' into staging

QObject patches for 2018-08-24

# gpg: Signature made Fri 24 Aug 2018 20:28:53 BST
# gpg:                using RSA key 3870B400EB918653
# gpg: Good signature from "Markus Armbruster <armbru@redhat.com>"
# gpg:                 aka "Markus Armbruster <armbru@pond.sub.org>"
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867  4E5F 3870 B400 EB91 8653

* remotes/armbru/tags/pull-qobject-2018-08-24: (58 commits)
  json: Update references to RFC 7159 to RFC 8259
  json: Support %% in JSON strings when interpolating
  json: Improve safety of qobject_from_jsonf_nofail() & friends
  json: Keep interpolation state in JSONParserContext
  tests/drive_del-test: Fix harmless JSON interpolation bug
  json: Clean up headers
  qobject: Drop superfluous includes of qemu-common.h
  json: Make JSONToken opaque outside json-parser.c
  json: Unbox tokens queue in JSONMessageParser
  json: Streamline json_message_process_token()
  json: Enforce token count and size limits more tightly
  qjson: Have qobject_from_json() & friends reject empty and blank
  json: Assert json_parser_parse() consumes all tokens on success
  json: Fix streamer not to ignore trailing unterminated structures
  json: Fix latent parser aborts at end of input
  qjson: Fix qobject_from_json() & friends for multiple values
  json: Improve names of lexer states related to numbers
  json: Replace %I64d, %I64u by %PRId64, %PRIu64
  json: Leave rejecting invalid interpolation to parser
  json: Pass lexical errors and limit violations to callback
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/amarkovic/tags/mips-queue-aug-2018' into staging
Peter Maydell [Fri, 24 Aug 2018 22:10:15 +0000 (23:10 +0100)]
Merge remote-tracking branch 'remotes/amarkovic/tags/mips-queue-aug-2018' into staging

MIPS queue August 2018 v6

# gpg: Signature made Fri 24 Aug 2018 16:52:27 BST
# gpg:                using RSA key D4972A8967F75A65
# gpg: Good signature from "Aleksandar Markovic <amarkovic@wavecomp.com>"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 8526 FBF1 5DA3 811F 4A01  DD75 D497 2A89 67F7 5A65

* remotes/amarkovic/tags/mips-queue-aug-2018: (45 commits)
  target/mips: Add definition of nanoMIPS I7200 CPU
  mips_malta: Fix semihosting argument passing for nanoMIPS bare metal
  mips_malta: Add setting up GT64120 BARs to the nanoMIPS bootloader
  mips_malta: Add basic nanoMIPS boot code for Malta board
  elf: Don't check FCR31_NAN2008 bit for nanoMIPS
  elf: On elf loading, treat both EM_MIPS and EM_NANOMIPS as legal for MIPS
  elf: Relax MIPS' elf_check_arch() to accept EM_NANOMIPS too
  elf: Add EM_NANOMIPS value as a valid one for e_machine field
  target/mips: Fix ERET/ERETNC behavior related to ADEL exception
  target/mips: Add updating BadInstr and BadInstrX for nanoMIPS
  target/mips: Add availability control via bit NMS
  target/mips: Add emulation of DSP ASE for nanoMIPS - part 6
  target/mips: Add emulation of DSP ASE for nanoMIPS - part 5
  target/mips: Add emulation of DSP ASE for nanoMIPS - part 4
  target/mips: Add emulation of DSP ASE for nanoMIPS - part 3
  target/mips: Add emulation of DSP ASE for nanoMIPS - part 2
  target/mips: Add emulation of DSP ASE for nanoMIPS - part 1
  target/mips: Implement MT ASE support for nanoMIPS
  target/mips: Fix pre-nanoMIPS MT ASE instructions availability control
  target/mips: Add emulation of nanoMIPS 32-bit branch instructions
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agojson: Update references to RFC 7159 to RFC 8259
Markus Armbruster [Thu, 23 Aug 2018 16:40:25 +0000 (18:40 +0200)]
json: Update references to RFC 7159 to RFC 8259

RFC 8259 (December 2017) obsoletes RFC 7159 (March 2014).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180823164025.12553-59-armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
6 years agojson: Support %% in JSON strings when interpolating
Markus Armbruster [Thu, 23 Aug 2018 16:40:24 +0000 (18:40 +0200)]
json: Support %% in JSON strings when interpolating

The previous commit makes JSON strings containing '%' awkward to
express in templates: you'd have to mask the '%' with an Unicode
escape \u0025.  No template currently contains such JSON strings.
Support the printf conversion specification %% in JSON strings as a
convenience anyway, because it's trivially easy to do.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-58-armbru@redhat.com>

6 years agojson: Improve safety of qobject_from_jsonf_nofail() & friends
Markus Armbruster [Thu, 23 Aug 2018 16:40:23 +0000 (18:40 +0200)]
json: Improve safety of qobject_from_jsonf_nofail() & friends

The JSON parser optionally supports interpolation.  This is used to
build QObjects by parsing string templates.  The templates are C
literals, so parse errors (such as invalid interpolation
specifications) are actually programming errors.  Consequently, the
functions providing parsing with interpolation
(qobject_from_jsonf_nofail(), qobject_from_vjsonf_nofail(),
qdict_from_jsonf_nofail(), qdict_from_vjsonf_nofail()) pass
&error_abort to the parser.

However, there's another, more dangerous kind of programming error:
since we use va_arg() to get the value to interpolate, behavior is
undefined when the variable argument isn't consistent with the
interpolation specification.

The same problem exists with printf()-like functions, and the solution
is to have the compiler check consistency.  This is what
GCC_FMT_ATTR() is about.

To enable this type checking for interpolation as well, we carefully
chose our interpolation specifications to match printf conversion
specifications, and decorate functions parsing templates with
GCC_FMT_ATTR().

Note that this only protects against undefined behavior due to type
errors.  It can't protect against use of invalid interpolation
specifications that happen to be valid printf conversion
specifications.

However, there's still a gaping hole in the type checking: GCC
recognizes '%' as start of printf conversion specification anywhere in
the template, but the parser recognizes it only outside JSON strings.
For instance, if someone were to pass a "{ '%s': %d }" template, GCC
would require a char * and an int argument, but the parser would
va_arg() only an int argument, resulting in undefined behavior.

Avoid undefined behavior by catching the programming error at run
time: have the parser recognize and reject '%' in JSON strings.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-57-armbru@redhat.com>

6 years agojson: Keep interpolation state in JSONParserContext
Markus Armbruster [Thu, 23 Aug 2018 16:40:22 +0000 (18:40 +0200)]
json: Keep interpolation state in JSONParserContext

The recursive descent parser passes along a pointer to
JSONParserContext.  It additionally passes a pointer to interpolation
state (a va_alist *) as needed to reach its consumer
parse_interpolation().

Stuffing the latter pointer into JSONParserContext saves us the
trouble of passing it along, so do that.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-56-armbru@redhat.com>

6 years agotests/drive_del-test: Fix harmless JSON interpolation bug
Markus Armbruster [Thu, 23 Aug 2018 16:40:21 +0000 (18:40 +0200)]
tests/drive_del-test: Fix harmless JSON interpolation bug

test_after_failed_device_add() does this:

    response = qmp("{'execute': 'device_add',"
                   " 'arguments': {"
                   "   'driver': 'virtio-blk-%s',"
                   "   'drive': 'drive0'"
                   "}}", qvirtio_get_dev_type());

Wrong.  An interpolation specification must be a JSON token, it
doesn't work within JSON string tokens.  The code above doesn't use
the value of qvirtio_get_dev_type(), and sends arguments

    {"driver": "virtio-blk-%s", "drive": "drive0"}}

The command fails because there is no driver named "virtio-blk-%".
Harmless, since the test wants the command to fail.  Screwed up in
commit 2f84a92ec63.

Fix the obvious way.  The command now fails because the drive is
empty, like it did before commit 2f84a92ec63.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-55-armbru@redhat.com>

6 years agojson: Clean up headers
Markus Armbruster [Thu, 23 Aug 2018 16:40:20 +0000 (18:40 +0200)]
json: Clean up headers

The JSON parser has three public headers, json-lexer.h, json-parser.h,
json-streamer.h.  They all contain stuff that is of no interest
outside qobject/json-*.c.

Collect the public interface in include/qapi/qmp/json-parser.h, and
everything else in qobject/json-parser-int.h.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-54-armbru@redhat.com>

6 years agoqobject: Drop superfluous includes of qemu-common.h
Markus Armbruster [Thu, 23 Aug 2018 16:40:19 +0000 (18:40 +0200)]
qobject: Drop superfluous includes of qemu-common.h

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-53-armbru@redhat.com>

6 years agojson: Make JSONToken opaque outside json-parser.c
Markus Armbruster [Thu, 23 Aug 2018 16:40:18 +0000 (18:40 +0200)]
json: Make JSONToken opaque outside json-parser.c

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-52-armbru@redhat.com>

6 years agojson: Unbox tokens queue in JSONMessageParser
Markus Armbruster [Thu, 23 Aug 2018 16:40:17 +0000 (18:40 +0200)]
json: Unbox tokens queue in JSONMessageParser

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-51-armbru@redhat.com>

6 years agojson: Streamline json_message_process_token()
Markus Armbruster [Thu, 23 Aug 2018 16:40:16 +0000 (18:40 +0200)]
json: Streamline json_message_process_token()

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-50-armbru@redhat.com>

6 years agojson: Enforce token count and size limits more tightly
Markus Armbruster [Thu, 23 Aug 2018 16:40:15 +0000 (18:40 +0200)]
json: Enforce token count and size limits more tightly

Token count and size limits exist to guard against excessive heap
usage.  We check them only after we created the token on the heap.
That's assigning a cowboy to the barn to lasso the horse after it has
bolted.  Close the barn door instead: check before we create the
token.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-49-armbru@redhat.com>

6 years agoqjson: Have qobject_from_json() & friends reject empty and blank
Markus Armbruster [Thu, 23 Aug 2018 16:40:14 +0000 (18:40 +0200)]
qjson: Have qobject_from_json() & friends reject empty and blank

The last case where qobject_from_json() & friends return null without
setting an error is empty or blank input.  Callers:

* block.c's parse_json_protocol() reports "Could not parse the JSON
  options".  It's marked as a work-around, because it also covered
  actual bugs, but they got fixed in the previous few commits.

* qobject_input_visitor_new_str() reports "JSON parse error".  Also
  marked as work-around.  The recent fixes have made this unreachable,
  because it currently gets called only for input starting with '{'.

* check-qjson.c's empty_input() and blank_input() demonstrate the
  behavior.

* The other callers are not affected since they only pass input with
  exactly one JSON value or, in the case of negative tests, one error.

Fail with "Expecting a JSON value" instead of returning null, and
simplify callers.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-48-armbru@redhat.com>

6 years agojson: Assert json_parser_parse() consumes all tokens on success
Markus Armbruster [Thu, 23 Aug 2018 16:40:13 +0000 (18:40 +0200)]
json: Assert json_parser_parse() consumes all tokens on success

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-47-armbru@redhat.com>

6 years agojson: Fix streamer not to ignore trailing unterminated structures
Markus Armbruster [Thu, 23 Aug 2018 16:40:12 +0000 (18:40 +0200)]
json: Fix streamer not to ignore trailing unterminated structures

json_message_process_token() accumulates tokens until it got the
sequence of tokens that comprise a single JSON value (it counts curly
braces and square brackets to decide).  It feeds those token sequences
to json_parser_parse().  If a non-empty sequence of tokens remains at
the end of the parse, it's silently ignored.  check-qjson.c cases
unterminated_array(), unterminated_array_comma(), unterminated_dict(),
unterminated_dict_comma() demonstrate this bug.

Fix as follows.  Introduce a JSON_END_OF_INPUT token.  When the
streamer receives it, it feeds the accumulated tokens to
json_parser_parse().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-46-armbru@redhat.com>

6 years agojson: Fix latent parser aborts at end of input
Markus Armbruster [Thu, 23 Aug 2018 16:40:11 +0000 (18:40 +0200)]
json: Fix latent parser aborts at end of input

json-parser.c carefully reports end of input like this:

    token = parser_context_pop_token(ctxt);
    if (token == NULL) {
        parse_error(ctxt, NULL, "premature EOI");
        goto out;
    }

Except parser_context_pop_token() can't return null, it fails its
assertion instead.  Same for parser_context_peek_token().  Broken in
commit 65c0f1e9558, and faithfully preserved in commit 95385fe9ace.
Only a latent bug, because the streamer throws away any input that
could trigger it.

Drop the assertions, so we can fix the streamer in the next commit.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-45-armbru@redhat.com>

6 years agoqjson: Fix qobject_from_json() & friends for multiple values
Markus Armbruster [Thu, 23 Aug 2018 16:40:10 +0000 (18:40 +0200)]
qjson: Fix qobject_from_json() & friends for multiple values

qobject_from_json() & friends use the consume_json() callback to
receive either a value or an error from the parser.

When they are fed a string that contains more than either one JSON
value or one JSON syntax error, consume_json() gets called multiple
times.

When the last call receives a value, qobject_from_json() returns that
value.  Any other values are leaked.

When any call receives an error, qobject_from_json() sets the first
error received.  Any other errors are thrown away.

When values follow errors, qobject_from_json() returns both a value
and sets an error.  That's bad.  Impact:

* block.c's parse_json_protocol() ignores and leaks the value.  It's
  used to to parse pseudo-filenames starting with "json:".  The
  pseudo-filenames can come from the user or from image meta-data such
  as a QCOW2 image's backing file name.

* vl.c's parse_display_qapi() ignores and leaks the error.  It's used
  to parse the argument of command line option -display.

* vl.c's main() case QEMU_OPTION_blockdev ignores the error and leaves
  it in @err.  main() will then pass a pointer to a non-null Error *
  to net_init_clients(), which is forbidden.  It can lead to assertion
  failure or other misbehavior.

* check-qjson.c's multiple_values() demonstrates the badness.

* The other callers are not affected since they only pass strings with
  exactly one JSON value or, in the case of negative tests, one
  error.

The impact on the _nofail() functions is relatively harmless.  They
abort when any call receives an error.  Else they return the last
value, and leak the others, if any.

Fix consume_json() as follows.  On the first call, save value and
error as before.  On subsequent calls, if any, don't save them.  If
the first call saved a value, the next call, if any, replaces the
value by an "Expecting at most one JSON value" error.  Take care not
to leak values or errors that aren't saved.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-44-armbru@redhat.com>

6 years agojson: Improve names of lexer states related to numbers
Markus Armbruster [Thu, 23 Aug 2018 16:40:09 +0000 (18:40 +0200)]
json: Improve names of lexer states related to numbers

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-43-armbru@redhat.com>

6 years agojson: Replace %I64d, %I64u by %PRId64, %PRIu64
Markus Armbruster [Thu, 23 Aug 2018 16:40:08 +0000 (18:40 +0200)]
json: Replace %I64d, %I64u by %PRId64, %PRIu64

Support for %I64d got added in commit 2c0d4b36e7f "json: fix PRId64 on
Win32".  We had to hard-code I64d because we used the lexer's finite
state machine to check interpolations.  No more, so clean this up.

Additional conversion specifications would be easy enough to implement
when needed.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-42-armbru@redhat.com>

6 years agojson: Leave rejecting invalid interpolation to parser
Markus Armbruster [Thu, 23 Aug 2018 16:40:07 +0000 (18:40 +0200)]
json: Leave rejecting invalid interpolation to parser

Both lexer and parser reject invalid interpolation specifications.
The parser's check is useless.

The lexer ends the token right after the first bad character.  This
tends to lead to suboptimal error reporting.  For instance, input

    [ %04d ]

produces the tokens

    JSON_LSQUARE  [
    JSON_ERROR    %0
    JSON_INTEGER  4
    JSON_KEYWORD  d
    JSON_RSQUARE  ]

The parser then yields an error, an object and two more errors:

    error: Invalid JSON syntax
    object: 4
    error: JSON parse error, invalid keyword
    error: JSON parse error, expecting value

Dumb down the lexer to accept [A-Za-z0-9]*.  The parser's check is now
used.  Emit a proper error there.

The lexer now produces

    JSON_LSQUARE  [
    JSON_INTERP   %04d
    JSON_RSQUARE  ]

and the parser reports just

    JSON parse error, invalid interpolation '%04d'

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-41-armbru@redhat.com>

6 years agojson: Pass lexical errors and limit violations to callback
Markus Armbruster [Thu, 23 Aug 2018 16:40:06 +0000 (18:40 +0200)]
json: Pass lexical errors and limit violations to callback

The callback to consume JSON values takes QObject *json, Error *err.
If both are null, the callback is supposed to make up an error by
itself.  This sucks.

qjson.c's consume_json() neglects to do so, which makes
qobject_from_json() null instead of failing.  I consider that a bug.

The culprit is json_message_process_token(): it passes two null
pointers when it runs into a lexical error or a limit violation.  Fix
it to pass a proper Error object then.  Update the callbacks:

* monitor.c's handle_qmp_command(): the code to make up an error is
  now dead, drop it.

* qga/main.c's process_event(): lumps the "both null" case together
  with the "not a JSON object" case.  The former is now gone.  The
  error message "Invalid JSON syntax" is misleading for the latter.
  Improve it to "Input must be a JSON object".

* qobject/qjson.c's consume_json(): no update; check-qjson
  demonstrates qobject_from_json() now sets an error on lexical
  errors, but still doesn't on some other errors.

* tests/libqtest.c's qmp_response(): the Error object is now reliable,
  so use it to improve the error message.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-40-armbru@redhat.com>

6 years agojson: Treat unwanted interpolation as lexical error
Markus Armbruster [Thu, 23 Aug 2018 16:40:05 +0000 (18:40 +0200)]
json: Treat unwanted interpolation as lexical error

The JSON parser optionally supports interpolation.  The lexer
recognizes interpolation tokens unconditionally.  The parser rejects
them when interpolation is disabled, in parse_interpolation().
However, it neglects to set an error then, which can make
json_parser_parse() fail without setting an error.

Move the check for unwanted interpolation from the parser's
parse_interpolation() into the lexer's finite state machine.  When
interpolation is disabled, '%' is now handled like any other
unexpected character.

The next commit will improve how such lexical errors are handled.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-39-armbru@redhat.com>

6 years agojson: Rename token JSON_ESCAPE & friends to JSON_INTERP
Markus Armbruster [Thu, 23 Aug 2018 16:40:04 +0000 (18:40 +0200)]
json: Rename token JSON_ESCAPE & friends to JSON_INTERP

The JSON parser optionally supports interpolation.  The code calls it
"escape".  Awkward, because it uses the same term for escape sequences
within strings.  The latter usage is consistent with RFC 8259 "The
JavaScript Object Notation (JSON) Data Interchange Format" and ISO C.
Call the former "interpolation" instead.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-38-armbru@redhat.com>

6 years agojson: Don't create JSON_ERROR tokens that won't be used
Markus Armbruster [Thu, 23 Aug 2018 16:40:03 +0000 (18:40 +0200)]
json: Don't create JSON_ERROR tokens that won't be used

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-37-armbru@redhat.com>

6 years agojson: Don't pass null @tokens to json_parser_parse()
Markus Armbruster [Thu, 23 Aug 2018 16:40:02 +0000 (18:40 +0200)]
json: Don't pass null @tokens to json_parser_parse()

json_parser_parse() normally returns the QObject on success.  Except
it returns null when its @tokens argument is null.

Its only caller json_message_process_token() passes null @tokens when
emitting a lexical error.  The call is a rather opaque way to say json
= NULL then.

Simplify matters by lifting the assignment to json out of the emit
path: initialize json to null, set it to the value of
json_parser_parse() when there's no lexical error.  Drop the special
case from json_parser_parse().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-36-armbru@redhat.com>

6 years agojson: Redesign the callback to consume JSON values
Markus Armbruster [Thu, 23 Aug 2018 16:40:01 +0000 (18:40 +0200)]
json: Redesign the callback to consume JSON values

The classical way to structure parser and lexer is to have the client
call the parser to get an abstract syntax tree, the parser call the
lexer to get the next token, and the lexer call some function to get
input characters.

Another way to structure them would be to have the client feed
characters to the lexer, the lexer feed tokens to the parser, and the
parser feed abstract syntax trees to some callback provided by the
client.  This way is more easily integrated into an event loop that
dispatches input characters as they arrive.

Our JSON parser is kind of between the two.  The lexer feeds tokens to
a "streamer" instead of a real parser.  The streamer accumulates
tokens until it got the sequence of tokens that comprise a single JSON
value (it counts curly braces and square brackets to decide).  It
feeds those token sequences to a callback provided by the client.  The
callback passes each token sequence to the parser, and gets back an
abstract syntax tree.

I figure it was done that way to make a straightforward recursive
descent parser possible.  "Get next token" becomes "pop the first
token off the token sequence".  Drawback: we need to store a complete
token sequence.  Each token eats 13 + input characters + malloc
overhead bytes.

Observations:

1. This is not the only way to use recursive descent.  If we replaced
   "get next token" by a coroutine yield, we could do without a
   streamer.

2. The lexer reports errors by passing a JSON_ERROR token to the
   streamer.  This communicates the offending input characters and
   their location, but no more.

3. The streamer reports errors by passing a null token sequence to the
   callback.  The (already poor) lexical error information is thrown
   away.

4. Having the callback receive a token sequence duplicates the code to
   convert token sequence to abstract syntax tree in every callback.

5. Known bug: the streamer silently drops incomplete token sequences.

This commit rectifies 4. by lifting the call of the parser from the
callbacks into the streamer.  Later commits will address 3. and 5.

The lifting removes a bug from qjson.c's parse_json(): it passed a
pointer to a non-null Error * in certain cases, as demonstrated by
check-qjson.c.

json_parser_parse() is now unused.  It's a stupid wrapper around
json_parser_parse_err().  Drop it, and rename json_parser_parse_err()
to json_parser_parse().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-35-armbru@redhat.com>

6 years agojson: Have lexer call streamer directly
Markus Armbruster [Thu, 23 Aug 2018 16:40:00 +0000 (18:40 +0200)]
json: Have lexer call streamer directly

json_lexer_init() takes the function to process a token as an
argument.  It's always json_message_process_token().  Makes the code
harder to understand for no actual gain.  Drop the indirection.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-34-armbru@redhat.com>

6 years agojson-parser: simplify and avoid JSONParserContext allocation
Marc-André Lureau [Thu, 23 Aug 2018 16:39:59 +0000 (18:39 +0200)]
json-parser: simplify and avoid JSONParserContext allocation

parser_context_new/free() are only used from json_parser_parse(). We
can fold the code there and avoid an allocation altogether.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20180719184111.5129-9-marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180823164025.12553-33-armbru@redhat.com>

6 years agojson: remove useless return value from lexer/parser
Marc-André Lureau [Thu, 23 Aug 2018 16:39:58 +0000 (18:39 +0200)]
json: remove useless return value from lexer/parser

The lexer always returns 0 when char feeding. Furthermore, none of the
caller care about the return value.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20180326150916.9602-10-marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180823164025.12553-32-armbru@redhat.com>

6 years agocheck-qjson: Fix and enable utf8_string()'s disabled part
Markus Armbruster [Thu, 23 Aug 2018 16:39:57 +0000 (18:39 +0200)]
check-qjson: Fix and enable utf8_string()'s disabled part

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-31-armbru@redhat.com>

6 years agojson: Fix \uXXXX for surrogate pairs
Markus Armbruster [Thu, 23 Aug 2018 16:39:56 +0000 (18:39 +0200)]
json: Fix \uXXXX for surrogate pairs

The JSON parser treats each half of a surrogate pair as unpaired
surrogate.  Fix it to recognize surrogate pairs.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-30-armbru@redhat.com>

6 years agojson: Reject invalid \uXXXX, fix \u0000
Markus Armbruster [Thu, 23 Aug 2018 16:39:55 +0000 (18:39 +0200)]
json: Reject invalid \uXXXX, fix \u0000

The JSON parser translates invalid \uXXXX to garbage instead of
rejecting it, and swallows \u0000.

Fix by using mod_utf8_encode() instead of flawed wchar_to_utf8().

Valid surrogate pairs are now differently broken: they're rejected
instead of translated to garbage.  The next commit will fix them.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-29-armbru@redhat.com>

6 years agojson: Simplify parse_string()
Markus Armbruster [Thu, 23 Aug 2018 16:39:54 +0000 (18:39 +0200)]
json: Simplify parse_string()

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-28-armbru@redhat.com>

6 years agojson: Leave rejecting invalid escape sequences to parser
Markus Armbruster [Thu, 23 Aug 2018 16:39:53 +0000 (18:39 +0200)]
json: Leave rejecting invalid escape sequences to parser

Both lexer and parser reject invalid escape sequences in strings.  The
parser's check is useless.

The lexer ends the token right after the first non-well-formed byte.
This tends to lead to suboptimal error reporting.  For instance, input

    {"abc\@ijk": 1}

produces the tokens

    JSON_LCURLY   {
    JSON_ERROR    "abc\@
    JSON_KEYWORD  ijk
    JSON_ERROR   ": 1}\n

The parser then reports three errors

    Invalid JSON syntax
    JSON parse error, invalid keyword 'ijk'
    Invalid JSON syntax

before it recovers at the newline.

Drop the lexer's escape sequence checking, and make it accept the same
characters after backslash it accepts elsewhere in strings.  It now
produces

    JSON_LCURLY   {
    JSON_STRING   "abc\@ijk"
    JSON_COLON    :
    JSON_INTEGER  1
    JSON_RCURLY

and the parser reports just

    JSON parse error, invalid escape sequence in string

While there, fix parse_string()'s inaccurate function comment.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-27-armbru@redhat.com>

6 years agojson: Accept overlong \xC0\x80 as U+0000 ("modified UTF-8")
Markus Armbruster [Thu, 23 Aug 2018 16:39:52 +0000 (18:39 +0200)]
json: Accept overlong \xC0\x80 as U+0000 ("modified UTF-8")

Since the JSON grammer doesn't accept U+0000 anywhere, this merely
exchanges one kind of parse error for another.  It's purely for
consistency with qobject_to_json(), which accepts \xC0\x80 (see commit
e2ec3f97680).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-26-armbru@redhat.com>

6 years agojson: Leave rejecting invalid UTF-8 to parser
Markus Armbruster [Thu, 23 Aug 2018 16:39:51 +0000 (18:39 +0200)]
json: Leave rejecting invalid UTF-8 to parser

Both the lexer and the parser (attempt to) validate UTF-8 in JSON
strings.

The lexer rejects bytes that can't occur in valid UTF-8: \xC0..\xC1,
\xF5..\xFF.  This rejects some, but not all invalid UTF-8.  It also
rejects ASCII control characters \x00..\x1F, in accordance with RFC
8259 (see recent commit "json: Reject unescaped control characters").

When the lexer rejects, it ends the token right after the first bad
byte.  Good when the bad byte is a newline.  Not so good when it's
something like an overlong sequence in the middle of a string.  For
instance, input

    {"abc\xC0\xAFijk": 1}\n

produces the tokens

    JSON_LCURLY   {
    JSON_ERROR    "abc\xC0
    JSON_ERROR    \xAF
    JSON_KEYWORD  ijk
    JSON_ERROR   ": 1}\n

The parser then reports four errors

    Invalid JSON syntax
    Invalid JSON syntax
    JSON parse error, invalid keyword 'ijk'
    Invalid JSON syntax

before it recovers at the newline.

The commit before previous made the parser reject invalid UTF-8
sequences.  Since then, anything the lexer rejects, the parser would
reject as well.  Thus, the lexer's rejecting is unnecessary for
correctness, and harmful for error reporting.

However, we want to keep rejecting ASCII control characters in the
lexer, because that produces the behavior we want for unclosed
strings.

We also need to keep rejecting \xFF in the lexer, because we
documented that as a way to reset the JSON parser
(docs/interop/qmp-spec.txt section 2.6 QGA Synchronization), which
means we can't change how we recover from this error now.  I wish we
hadn't done that.

I think we should treat \xFE the same as \xFF.

Change the lexer to accept \xC0..\xC1 and \xF5..\xFD.  It now rejects
only \x00..\x1F and \xFE..\xFF.  Error reporting for invalid UTF-8 in
strings is much improved, except for \xFE and \xFF.  For the example
above, the lexer now produces

    JSON_LCURLY   {
    JSON_STRING   "abc\xC0\xAFijk"
    JSON_COLON    :
    JSON_INTEGER  1
    JSON_RCURLY

and the parser reports just

    JSON parse error, invalid UTF-8 sequence in string

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-25-armbru@redhat.com>

6 years agojson: Report first rather than last parse error
Markus Armbruster [Thu, 23 Aug 2018 16:39:50 +0000 (18:39 +0200)]
json: Report first rather than last parse error

Quiz time!  When a parser reports multiple errors, but the user gets
to see just one, which one is (on average) the least useful one?

Yes, you're right, it's the last one!  You're clearly familiar with
compilers.

Which one does QEMU report?

Right again, the last one!  You're clearly familiar with QEMU.

Reproducer: feeding

    {"abc\xC2ijk": 1}\n

to QMP produces

    {"error": {"class": "GenericError", "desc": "JSON parse error, key is not a string in object"}}

Report the first error instead.  The reproducer now produces

    {"error": {"class": "GenericError", "desc": "JSON parse error, invalid UTF-8 sequence in string"}}

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-24-armbru@redhat.com>

6 years agojson: Reject invalid UTF-8 sequences
Markus Armbruster [Thu, 23 Aug 2018 16:39:49 +0000 (18:39 +0200)]
json: Reject invalid UTF-8 sequences

We reject bytes that can't occur in valid UTF-8 (\xC0..\xC1,
\xF5..\xFF in the lexer.  That's insufficient; there's plenty of
invalid UTF-8 not containing these bytes, as demonstrated by
check-qjson:

* Malformed sequences

  - Unexpected continuation bytes

  - Missing continuation bytes after start bytes other than
    \xC0..\xC1, \xF5..\xFD.

* Overlong sequences with start bytes other than \xC0..\xC1,
  \xF5..\xFD.

* Invalid code points

Fixing this in the lexer would be bothersome.  Fixing it in the parser
is straightforward, so do that.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-23-armbru@redhat.com>

6 years agocheck-qjson: Document we expect invalid UTF-8 to be rejected
Markus Armbruster [Thu, 23 Aug 2018 16:39:48 +0000 (18:39 +0200)]
check-qjson: Document we expect invalid UTF-8 to be rejected

The JSON parser rejects some invalid sequences, but accepts others
without correcting the problem.

We should either reject all invalid sequences, or minimize overlong
sequences and replace all other invalid sequences by a suitable
replacement character.  A common choice for replacement is U+FFFD.

I'm going to implement the former.  Update the comments in
utf8_string() to expect this.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-22-armbru@redhat.com>

6 years agojson: Tighten and simplify qstring_from_escaped_str()'s loop
Markus Armbruster [Thu, 23 Aug 2018 16:39:47 +0000 (18:39 +0200)]
json: Tighten and simplify qstring_from_escaped_str()'s loop

Simplify loop control, and assert that the string ends with the
appropriate quote (the lexer ensures it does).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-21-armbru@redhat.com>

6 years agojson: Revamp lexer documentation
Markus Armbruster [Thu, 23 Aug 2018 16:39:46 +0000 (18:39 +0200)]
json: Revamp lexer documentation

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-20-armbru@redhat.com>