Makefile.target: add (clean-/build-)guest-tests targets

Now all the build infrastructure is in place we can build tests for
each guest that we support. That support mainly depends on having
cross compilers installed or docker setup. To keep all the logic for
that together we put the rules in tests/tcg/Makefile.include and
include it from the main Makefile.target.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg/Makefile: update to be called from Makefile.target

This make is now invoked from each individual target make with the
appropriate CC and EXTRA_CFLAGS set for each guest. It then includes
additional Makefile.targets from:

  - tests/tcg/multiarch (always)
  - tests/tcg/$(TARGET_BASE_ARCH) (if available)
  - tests/tcg/$(TARGET_NAME)

The order is important as the later Makefile's may want to suppress
TESTS from its base arch profile. Each included Makefile.target is
responsible for adding TESTS as well as defining any special build
instructions for individual tests.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for PowerPC

Now we have restored debian-image-powerpc-cross using Debian SID
compilers we can build for 32 bit powerpc. Although PPC32 supports a
range of pages sizes currently only 4k works so the others are
commented out for now.

We can also merge the ppc64 support under the base architecture
directory to avoid too much proliferation of directories.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

docker: move debian-powerpc-cross to sid based build

The original Jessie based cross builder hasn't worked for a while. The
state of the libraries is still perilous for cross-building QEMU but
we can use it for building TCG tests.

The debian-apt-fake.sh script can also be dropped as it is no longer
used.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for RISCV64

As before, using Debian SID compilers.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for mips64

As before, using Debian SID compilers.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for sparc64

As before, using Debian SID compilers.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for sh4

As before, using Debian SID compilers. While the compiler can be
coerced into generating big-endian code it seems the linker can't deal
with it so we only enable the building for little endian SH4.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for m68k

As before, using Debian SID compilers.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for HPPA

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg/alpha: add Alpha specific tests

These tests did use their own crt.o stub however that is a little
stone age so we drop crt.S and just statically link to the cross
compilers libraries.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for Alpha

We can't use our normal Debian based compilers as Alpha isn't an
officially supported architecture. However it is available as a port
and fortunately cross compilers for all these targets are included in
Debian Sid, the perpetual rolling/unstable/testing version of Debian.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for ppc64

Currently this just enables building the multiarch tests.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for s390x

This doesn't add any additional tests but enables building the
multiarch tests for s390x.

Signed-off-by: Alex Bennée <[email protected]>
Acked-by: Cornelia Huck <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>

tests/tcg/mips: include common mips hello-mips

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/docker/Makefile.include: fix mipsel-cross dependancy

This got broken in commit 4319db7 but generally only shows up when you
try and do massive parallel builds on fresh machines.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for MIPS

This doesn't add any additional tests but enables building the
multiarch tests for MIPS using docker cross compilers. We don't have a
cross compiler for mips64 big endian though.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: move MIPS specific tests into subdir

These only need to be built for MIPS guests.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg/arm: add fcvt test cases for AArch32/64

This runs through the usual float to float conversions and crucially
also runs with ARM Alternative Half Precision Format.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: enable building for AArch64

We only have compilers for the (default) little endian variants.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg/arm: fix up test-arm-iwmmxt test

We need to rename the source file to a .S so we can do a single-line
assemble and link invocation. We also specify the additional CFLAGS
for the compile as it's a non-standard ARM binary.

Signed-off-by: Alex Bennée <[email protected]>
[rth: force fpu configuration]
Signed-off-by: Richard Henderson <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

tests/tcg: enable building for ARM

This allows us to use the docker cross compiler image to build these
tests.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: move ARM specific tests into subdir

These only need to be built for ARM guests.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

tests/tcg/i386/test-i386: fix printf format

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg/i386/test-i386: use modern vector_size attributes

The compiler complains about the old __mode__ style attributes.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg/x86_64: add Makefile.target

The sources for x86_64 are shared in the i386 directory which will be
included thanks to TARGET_BASE_ARCH. However not all sources build so
we need to filter out the ones we can't build in the 64 bit world and
those that can't be built for 32 bit.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg/i386: add runner for test-i386-fprem

The runner needs to compare against a reference run. We also only run
this test when SPEED=slow as it takes a while.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg/i386: fix test-i386

We don't include anything from qemu itself for the build.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

tests/tcg/i386: Build fix for hello-i386

We have -Werror=missing-prototype, add a dummy prototype to avoid that
warning.

Signed-off-by: Fam Zheng <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Signed-off-by: Alex Bennée <[email protected]>

tests/tcg: enable building for i386

While you can construct a compile command that does work using the
x86_64 host compiler that most people use this is flakey. Different
distros handle this is different ways so we default to using a known
good i386 compiler via docker.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

tests/tcg: move i386 specific tests into subdir

These only need to be built for i386 guests. This includes a stub
tests/tcg/i386/Makfile.target which absorbs some of what was in
tests/tcg/Makefile.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

tests/tcg/multiarch: move most output to stdout

The default test run outputs to stdout so it can be re-directed.
Errors are still reported to stderr.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>

tests/tcg/multiarch: don't hard code paths/ports for linux-test

The fixed path and ports get in the way of running our tests and
builds in parallel. Instead of using TESTPATH we use mkdtemp() and
instead of a fixed port we allow the kernel to assign one and query it
afterwards.

Ideally test directory creation should be common functionally across
all TCG tests but this could complicate an already huge patch series
so we mark it as a TODO for next time.

Signed-off-by: Alex Bennée <[email protected]>

tests/tcg/multiarch: enable additional linux-test tests

Un-comment the remaining tests.

I removed the itimer value tests because I'm fairly sure a re-arming
timer will always have a different value in it when you grab it.

I've also fixed up the clone thread flags as QEMU will only allow a
clone to use flags which match glibc. However the test is still racey
so it remains disabled by default - it can be run by passing any
additional parameters on the command line.

Signed-off-by: Alex Bennée <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

tests/tcg/multiarch: Build fix for linux-test

To keep the compiler happy, and to fit in our buildsys flags:

- Make local functions "static"
- #ifdef out unused functions
- drop cutils/osdep dependencies

Signed-off-by: Fam Zheng <[email protected]>
[AJB: drop cutils/osdep dependencies]
Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

tests/tcg: move architecture independent tests into subdir

We will want to build these for all supported guest architectures so
lets move them all into one place. We also drop test_path at this
point because it needs qemu utils and glib bits which is hard to
support for cross compiling.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

docker: Makefile.include introduce DOCKER_SCRIPT

Define this in one place to make it easy to re-use.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>

docker: allow "cc" command to run in user context

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

docker: extend "cc" command to accept compiler

When calling our cross-compilation images we want to call something
other than the default cc.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>

docker: Add "cc" subcommand

Signed-off-by: Fam Zheng <[email protected]>
[AJB: add if args.paths check]
Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>

configure: set cross_cc_FOO for host compiler

We can build tests for the host system with the compiler that we have
selected.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

configure: allow user to specify --cross-cc-cflags-foo=

As an individual compiler may be able to support several targets with
the appropriate flags we need to expose this to the user as well.

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

configure: move i386_cc to cross_cc_i386

Also dont assume x86_64 compiler can build i386 binaries.

Signed-off-by: Alex Bennée <[email protected]>
Tested-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

configure: add support for --cross-cc-FOO

This allows us to specify cross compilers for our guests. This is
useful for building test images/programs. Currently we re-run the
compile test for each target. I couldn't think of a way to cache the
value for a given arch without getting messier configure code.

The cross compiler for the guest is visible to each target as
CROSS_CC_GUEST in config-target.mak. This is quoted to handle the case
of --cc="ccache gcc".

Signed-off-by: Alex Bennée <[email protected]>
Reviewed-by: Richard Henderson <[email protected]>

docker: docker.py wrap StringIO import for python3

Although the docker.py is nominally python2 we actually invoke it with
the configured python from the configure script.

Signed-off-by: Alex Bennée <[email protected]>

Merge remote-tracking branch 'remotes/cohuck/tags/s390x-20180619' into staging

- cleanup in virtio-ccw
- accommodate guests using vfio-ccw without specifying unlimited
  prefetch, but actually working fine
- add cpu model for the z14 Model ZR1
- add support for pxelinux.cfg-style network booting to the s390x
  firmware

# gpg: Signature made Tue 19 Jun 2018 10:33:06 BST
# gpg:                using RSA key DECF6B93C6F02FAF
# gpg: Good signature from "Cornelia Huck <[email protected]>"
# gpg:                 aka "Cornelia Huck <[email protected]>"
# gpg:                 aka "Cornelia Huck <[email protected]>"
# gpg:                 aka "Cornelia Huck <[email protected]>"
# gpg:                 aka "Cornelia Huck <[email protected]>"
# Primary key fingerprint: C3D0 D66D C362 4FF6 A8C0  18CE DECF 6B93 C6F0 2FAF

* remotes/cohuck/tags/s390x-20180619:
  pc-bios/s390-ccw: Update the s390-netboot.img binary
  pc-bios/s390-ccw: Optimize the s390-netboot.img for size
  pc-bios/s390-ccw/net: Try to load pxelinux.cfg file accoring to the UUID
  pc-bios/s390-ccw/net: Add support for pxelinux-style config files
  pc-bios/s390-ccw/net: Update code for the latest changes in SLOF
  roms: Update SLOF submodule to current status
  pc-bios/s390-ccw: define loadparm length
  s390x/cpumodels: add z14 Model ZR1
  s390x/ipl: Try to detect Linux vs non Linux for initial IPL PSW
  vfio-ccw: remove orb.c64 (64 bit data addresses) check
  vfio-ccw: add force unlimited prefetch property
  virtio-ccw: clean up notify

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

Merge remote-tracking branch 'remotes/mcayland/tags/qemu-openbios-20180618' into staging

qemu-openbios queue

# gpg: Signature made Mon 18 Jun 2018 19:28:08 BST
# gpg:                using RSA key 5BC2C56FAE0F321F
# gpg: Good signature from "Mark Cave-Ayland <[email protected]>"
# Primary key fingerprint: CC62 1AB9 8E82 200D 915C  C9C4 5BC2 C56F AE0F 321F

* remotes/mcayland/tags/qemu-openbios-20180618:
  Update OpenBIOS images to 8fe6f5f96f built from submodule.

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

Merge remote-tracking branch 'remotes/mcayland/tags/qemu-sparc-20180618' into staging

qemu-sparc queue

# gpg: Signature made Mon 18 Jun 2018 18:43:24 BST
# gpg:                using RSA key 5BC2C56FAE0F321F
# gpg: Good signature from "Mark Cave-Ayland <[email protected]>"
# Primary key fingerprint: CC62 1AB9 8E82 200D 915C  C9C4 5BC2 C56F AE0F 321F

* remotes/mcayland/tags/qemu-sparc-20180618:
  SPARC64: add icount support
  hw/sparc/sun4m: Fix problems with device introspection
  hw/sparc64/sun4u: Fix introspection by converting prom instance_init to realize

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

Merge remote-tracking branch 'remotes/rth/tags/pull-axp-20180618' into staging

Place parallel device properly, fixing vga

# gpg: Signature made Mon 18 Jun 2018 17:45:50 BST
# gpg:                using RSA key 64DF38E8AF7E215F
# gpg: Good signature from "Richard Henderson <[email protected]>"
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth/tags/pull-axp-20180618:
  hw/isa/smc37c669: Change the parallel I/O base to 378H

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

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

Block layer patches:

- Active mirror (blockdev-mirror copy-mode=write-blocking)
- bdrv_drain_*() fixes and test cases
- Fix crash with scsi-hd and drive_del

# gpg: Signature made Mon 18 Jun 2018 17:44:10 BST
# gpg:                using RSA key 7F09B272C88F2FD6
# gpg: Good signature from "Kevin Wolf <[email protected]>"
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74  56FE 7F09 B272 C88F 2FD6

* remotes/kevin/tags/for-upstream: (35 commits)
  iotests: Add test for active mirroring
  block/mirror: Add copy mode QAPI interface
  block/mirror: Add active mirroring
  job: Add job_progress_increase_remaining()
  block/mirror: Add MirrorBDSOpaque
  block/dirty-bitmap: Add bdrv_dirty_iter_next_area
  test-hbitmap: Add non-advancing iter_next tests
  hbitmap: Add @advance param to hbitmap_iter_next()
  block: Generalize should_update_child() rule
  block/mirror: Use source as a BdrvChild
  block/mirror: Wait for in-flight op conflicts
  block/mirror: Use CoQueue to wait on in-flight ops
  block/mirror: Convert to coroutines
  block/mirror: Pull out mirror_perform()
  block: fix QEMU crash with scsi-hd and drive_del
  test-bdrv-drain: Test graph changes in drain_all section
  block: Allow graph changes in bdrv_drain_all_begin/end sections
  block: ignore_bds_parents parameter for drain functions
  block: Move bdrv_drain_all_begin() out of coroutine context
  block: Allow AIO_WAIT_WHILE with NULL ctx
  ...

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

Merge remote-tracking branch 'remotes/armbru/tags/pull-monitor-2018-06-18' into staging

Monitor patches for 2018-06-18

# gpg: Signature made Mon 18 Jun 2018 14:50:29 BST
# gpg:                using RSA key 3870B400EB918653
# gpg: Good signature from "Markus Armbruster <[email protected]>"
# gpg:                 aka "Markus Armbruster <[email protected]>"
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867  4E5F 3870 B400 EB91 8653

* remotes/armbru/tags/pull-monitor-2018-06-18:
  monitor: add lock to protect mon_fdsets
  monitor: move init global earlier
  monitor: remove event_clock_type
  monitor: fix comment for monitor_lock
  monitor: more comments on lock-free elements
  monitor: protect mon->fds with mon_lock
  monitor: rename out_lock to mon_lock

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

Merge remote-tracking branch 'remotes/kraxel/tags/input-20180618-pull-request' into staging

input: ps2 post_load fix.

# gpg: Signature made Mon 18 Jun 2018 11:18:30 BST
# gpg:                using RSA key 4CB6D8EED3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>"
# gpg:                 aka "Gerd Hoffmann <[email protected]>"
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>"
# Primary key fingerprint: A032 8CFF B93A 17A7 9901  FE7D 4CB6 D8EE D3E8 7138

* remotes/kraxel/tags/input-20180618-pull-request:
  ps2: check PS2Queue wptr pointer in post_load routine

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

Merge remote-tracking branch 'remotes/kraxel/tags/vga-20180618-pull-request' into staging

vga: add ramfb, print virglrenderer version

# gpg: Signature made Mon 18 Jun 2018 10:57:38 BST
# gpg:                using RSA key 4CB6D8EED3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>"
# gpg:                 aka "Gerd Hoffmann <[email protected]>"
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>"
# Primary key fingerprint: A032 8CFF B93A 17A7 9901  FE7D 4CB6 D8EE D3E8 7138

* remotes/kraxel/tags/vga-20180618-pull-request:
  Add ramfb MAINTAINERS entry
  hw/display: add standalone ramfb device
  hw/display: add ramfb, a simple boot framebuffer living in guest ram
  configure: print virglrenderer version

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

Merge remote-tracking branch 'remotes/kraxel/tags/usb-20180618-pull-request' into staging

usb: revert ccid / qom patches.

# gpg: Signature made Mon 18 Jun 2018 10:21:11 BST
# gpg:                using RSA key 4CB6D8EED3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>"
# gpg:                 aka "Gerd Hoffmann <[email protected]>"
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>"
# Primary key fingerprint: A032 8CFF B93A 17A7 9901  FE7D 4CB6 D8EE D3E8 7138

* remotes/kraxel/tags/usb-20180618-pull-request:
  Revert "bus: do not unref the added child bus on realize"
  Revert "usb: release the created buses"
  Revert "usb-ccid: fix bus leak"

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

Merge remote-tracking branch 'remotes/dgibson/tags/ppc-for-3.0-20180618' into staging

ppc patch queue 2018-06-18

Next batch of ppc and spapr related patches for the 3.0 release.
    * Improved handling of Spectre/Meltdown mitigations for POWER8
    * Numerous Mac machine type cleanups and improvements
    * Cleanup to cpu realize/unrealize path for spapr
    * Create a place for machine-specific per-cpu information, and
      start moving some things to it
    * Assorted bugfixes

# gpg: Signature made Mon 18 Jun 2018 04:52:37 BST
# gpg:                using RSA key 6C38CACA20D9B392
# gpg: Good signature from "David Gibson <[email protected]>"
# gpg:                 aka "David Gibson (Red Hat) <[email protected]>"
# gpg:                 aka "David Gibson (ozlabs.org) <[email protected]>"
# gpg:                 aka "David Gibson (kernel.org) <[email protected]>"
# Primary key fingerprint: 75F4 6586 AE61 A66C C44E  87DC 6C38 CACA 20D9 B392

* remotes/dgibson/tags/ppc-for-3.0-20180618: (28 commits)
  spapr: fix xics_system_init() error path
  target/ppc, spapr: Move VPA information to machine_data
  ppc/pnv: introduce a pnv_chip_core_realize() routine
  spapr_cpu_core: introduce spapr_create_vcpu()
  spapr_cpu_core: add missing rollback on realization path
  spapr_cpu_core: fix potential leak in spapr_cpu_core_realize()
  spapr_cpu_core: convert last snprintf() to g_strdup_printf()
  pnv: Add cpu unrealize path
  pnv: Clean up cpu realize path
  pnv_core: Allocate cpu thread objects individually
  pnv: Fix some error handling cpu realize()
  spapr: Clean up cpu realize/unrealize paths
  sm501: Do not clear read only bits when writing registers
  mos6522: expose mos6522_update_irq() through MOS6522DeviceClass
  mos6522: remove additional interrupt flag filter from mos6522_update_irq()
  mos6522: only clear the shift register interrupt upon write
  xics_kvm: fix a build break
  mac_newworld: add PMU device
  adb: add property to disable direct reg 3 writes
  adb: fix read reg 3 byte ordering
  ...

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

Merge remote-tracking branch 'remotes/ehabkost/tags/python-next-pull-request' into staging

Python queue, 2018-06-15

* Add avocado_qemu: functional/acceptance test infrastructure

# gpg: Signature made Fri 15 Jun 2018 20:12:20 BST
# gpg:                using RSA key 2807936F984DC5A6
# gpg: Good signature from "Eduardo Habkost <[email protected]>"
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF  D1AA 2807 936F 984D C5A6

* remotes/ehabkost/tags/python-next-pull-request:
  configure: Enable out-of-tree acceptance tests
  Acceptance tests: add Linux kernel boot and console checking test
  scripts/qemu.py: introduce set_console() method
  Acceptance tests: add quick VNC tests
  scripts/qemu.py: allow adding to the list of extra arguments
  Add functional/acceptance tests infrastructure
  Remove COPYING.PYTHON

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

Update OpenBIOS images to 8fe6f5f96f built from submodule.

Signed-off-by: Mark Cave-Ayland <[email protected]>

Merge remote-tracking branch 'mreitz/tags/pull-block-2018-06-18' into queue-block

Block patches:
- Active mirror (blockdev-mirror copy-mode=write-blocking)

# gpg: Signature made Mon Jun 18 17:08:19 2018 CEST
# gpg:                using RSA key F407DB0061D5CF40
# gpg: Good signature from "Max Reitz <[email protected]>"
# Primary key fingerprint: 91BE B60A 30DB 3E88 57D1  1829 F407 DB00 61D5 CF40

* mreitz/tags/pull-block-2018-06-18:
  iotests: Add test for active mirroring
  block/mirror: Add copy mode QAPI interface
  block/mirror: Add active mirroring
  job: Add job_progress_increase_remaining()
  block/mirror: Add MirrorBDSOpaque
  block/dirty-bitmap: Add bdrv_dirty_iter_next_area
  test-hbitmap: Add non-advancing iter_next tests
  hbitmap: Add @advance param to hbitmap_iter_next()
  block: Generalize should_update_child() rule
  block/mirror: Use source as a BdrvChild
  block/mirror: Wait for in-flight op conflicts
  block/mirror: Use CoQueue to wait on in-flight ops
  block/mirror: Convert to coroutines
  block/mirror: Pull out mirror_perform()

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

Merge tag 'tags/s390x-2018-06-18' into staging

Add support for pxelinux.cfg-style network booting to the s390x firmware

# gpg: Signature made Mon 18 Jun 2018 03:59:06 PM CEST
# gpg:                using RSA key 2ED9D774FE702DB5
# gpg: Good signature from "Thomas Huth <[email protected]>" [full]
# gpg:                 aka "Thomas Huth <[email protected]>" [undefined]
# gpg:                 aka "Thomas Huth <[email protected]>" [undefined]
# gpg:                 aka "Thomas Huth <[email protected]>" [unknown]

* tag 'tags/s390x-2018-06-18':
  pc-bios/s390-ccw: Update the s390-netboot.img binary
  pc-bios/s390-ccw: Optimize the s390-netboot.img for size
  pc-bios/s390-ccw/net: Try to load pxelinux.cfg file accoring to the UUID
  pc-bios/s390-ccw/net: Add support for pxelinux-style config files
  pc-bios/s390-ccw/net: Update code for the latest changes in SLOF
  roms: Update SLOF submodule to current status
  pc-bios/s390-ccw: define loadparm length

Signed-off-by: Cornelia Huck <[email protected]>

iotests: Add test for active mirroring

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Reviewed-by: Alberto Garcia <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block/mirror: Add copy mode QAPI interface

This patch allows the user to specify whether to use active or only
background mode for mirror block jobs.  Currently, this setting will
remain constant for the duration of the entire block job.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Alberto Garcia <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block/mirror: Add active mirroring

This patch implements active synchronous mirroring.  In active mode, the
passive mechanism will still be in place and is used to copy all
initially dirty clusters off the source disk; but every write request
will write data both to the source and the target disk, so the source
cannot be dirtied faster than data is mirrored to the target.  Also,
once the block job has converged (BLOCK_JOB_READY sent), source and
target are guaranteed to stay in sync (unless an error occurs).

Active mode is completely optional and currently disabled at runtime.  A
later patch will add a way for users to enable it.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

job: Add job_progress_increase_remaining()

Signed-off-by: Max Reitz <[email protected]>
Message-id: 20180613181823[email protected]
Reviewed-by: Kevin Wolf <[email protected]>
Signed-off-by: Max Reitz <[email protected]>

block/mirror: Add MirrorBDSOpaque

This will allow us to access the block job data when the mirror block
driver becomes more complex.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block/dirty-bitmap: Add bdrv_dirty_iter_next_area

This new function allows to look for a consecutively dirty area in a
dirty bitmap.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Reviewed-by: John Snow <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

test-hbitmap: Add non-advancing iter_next tests

Add a function that wraps hbitmap_iter_next() and always calls it in
non-advancing mode first, and in advancing mode next.  The result should
always be the same.

By using this function everywhere we called hbitmap_iter_next() before,
we should get good test coverage for non-advancing hbitmap_iter_next().

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Reviewed-by: John Snow <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

hbitmap: Add @advance param to hbitmap_iter_next()

This new parameter allows the caller to just query the next dirty
position without moving the iterator.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Reviewed-by: John Snow <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block: Generalize should_update_child() rule

Currently, bdrv_replace_node() refuses to create loops from one BDS to
itself if the BDS to be replaced is the backing node of the BDS to
replace it: Say there is a node A and a node B.  Replacing B by A means
making all references to B point to A.  If B is a child of A (i.e. A has
a reference to B), that would mean we would have to make this reference
point to A itself -- so we'd create a loop.

bdrv_replace_node() (through should_update_child()) refuses to do so if
B is the backing node of A.  There is no reason why we should create
loops if B is not the backing node of A, though.  The BDS graph should
never contain loops, so we should always refuse to create them.

If B is a child of A and B is to be replaced by A, we should simply
leave B in place there because it is the most sensible choice.

A more specific argument would be: Putting filter drivers into the BDS
graph is basically the same as appending an overlay to a backing chain.
But the main child BDS of a filter driver is not "backing" but "file",
so restricting the no-loop rule to backing nodes would fail here.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Reviewed-by: Alberto Garcia <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block/mirror: Use source as a BdrvChild

With this, the mirror_top_bs is no longer just a technically required
node in the BDS graph but actually represents the block job operation.

Also, drop MirrorBlockJob.source, as we can reach it through
mirror_top_bs->backing.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Reviewed-by: Alberto Garcia <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block/mirror: Wait for in-flight op conflicts

This patch makes the mirror code differentiate between simply waiting
for any operation to complete (mirror_wait_for_free_in_flight_slot())
and specifically waiting for all operations touching a certain range of
the virtual disk to complete (mirror_wait_on_conflicts()).

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block/mirror: Use CoQueue to wait on in-flight ops

Attach a CoQueue to each in-flight operation so if we need to wait for
any we can use it to wait instead of just blindly yielding and hoping
for some operation to wake us.

A later patch will use this infrastructure to allow requests accessing
the same area of the virtual disk to specifically wait for each other.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block/mirror: Convert to coroutines

In order to talk to the source BDS (and maybe in the future to the
target BDS as well) directly, we need to convert our existing AIO
requests into coroutine I/O requests.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

block/mirror: Pull out mirror_perform()

When converting mirror's I/O to coroutines, we are going to need a point
where these coroutines are created.  mirror_perform() is going to be
that point.

Signed-off-by: Max Reitz <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Jeff Cody <[email protected]>
Reviewed-by: Alberto Garcia <[email protected]>
Message-id: 20180613181823[email protected]
Signed-off-by: Max Reitz <[email protected]>

monitor: add lock to protect mon_fdsets

Introduce a new global big lock for mon_fdsets.  Take it where needed.

The monitor_fdset_get_fd() handling is a bit tricky: now we need to call
qemu_mutex_unlock() which might pollute errno, so we need to make sure
the correct errno be passed up to the callers.  To make things simpler,
we let monitor_fdset_get_fd() return the -errno directly when error
happens, then in qemu_open() we move it back into errno.

Reviewed-by: Stefan Hajnoczi <[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Peter Xu <[email protected]>
Message-Id: <20180608035511[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

monitor: move init global earlier

Before this patch, monitor fd helpers might be called even earlier than
monitor_init_globals().  This can be problematic.

After previous work, now monitor_init_globals() does not depend on
accelerator initialization any more.  Call it earlier (before CLI
parsing; that's where the monitor APIs might be called) to make sure it
is called before any of the monitor APIs.

Suggested-by: Markus Armbruster <[email protected]>
Reviewed-by: Stefan Hajnoczi <[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Peter Xu <[email protected]>
Message-Id: <20180608035511[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

monitor: remove event_clock_type

Instead, use a dynamic function to detect which clock we'll use.  The
problem is that the old code will let monitor initialization depend on
configure_accelerator() (that's where qtest_enabled() start to take
effect).  After this change, we don't have such a dependency any more.
We just need to make sure configure_accelerator() is called when we
start to use it.  Now it's only used in monitor_qapi_event_queue() and
monitor_qapi_event_handler(), so we're good.

Suggested-by: Markus Armbruster <[email protected]>
Signed-off-by: Peter Xu <[email protected]>
Message-Id: <20180608035511[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
[monitor_get_event_clock() name and comment tweaked]
Reviewed-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

monitor: fix comment for monitor_lock

Fix typo in d622cb5879c.  Meanwhile move these variables close to each
other.  monitor_qapi_event_state can be declared static, add that.

Reported-by: Markus Armbruster <[email protected]>
Reviewed-by: Stefan Hajnoczi <[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Peter Xu <[email protected]>
Message-Id: <20180608035511[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

monitor: more comments on lock-free elements

Add some explicit comments for both Readline and cpu_set/cpu_get helpers
that they do not need the mon_lock protection.

Reviewed-by: Stefan Hajnoczi <[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Peter Xu <[email protected]>
Message-Id: <20180608035511[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

monitor: protect mon->fds with mon_lock

mon->fds were protected by BQL.  Now protect it by mon_lock so that it
can even be used in monitor iothread.

Reviewed-by: Stefan Hajnoczi <[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Peter Xu <[email protected]>
Message-Id: <20180608035511[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

monitor: rename out_lock to mon_lock

The out_lock is protecting a few Monitor fields.  In the future the
monitor code will start to run in multiple threads.  We are going to
turn it into a bigger lock to protect not only the out buffer but also
most of the rest.

Since at it, rearrange the Monitor struct a bit.

Reviewed-by: Stefan Hajnoczi <[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Peter Xu <[email protected]>
Message-Id: <20180608035511[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

pc-bios/s390-ccw: Update the s390-netboot.img binary

This binary now contains the support for pxelinux.cfg-style network
booting.

Signed-off-by: Thomas Huth <[email protected]>

pc-bios/s390-ccw: Optimize the s390-netboot.img for size

The -O2 optimization flag is passed via CFLAGS to the firmware Makefile,
but in netbook.mak, we've got some rules that only use QEMU_CFLAGS for
compiling the libc and libnet from SLOF, so these files get compiled
without optimization so far. Use CFLAGS here, too, to create faster
and smaller code.

We can additionally save some more bytes in the firmware images by compi-
ling the code with -fno-asynchronous-unwind-tables. This will omit some
ELF sections (used for stack unwinding for example) from the image that
we do not need in the firmware.

Acked-by: Christian Borntraeger <[email protected]>
Signed-off-by: Thomas Huth <[email protected]>

pc-bios/s390-ccw/net: Try to load pxelinux.cfg file accoring to the UUID

With the STSI instruction, we can get the UUID of the current VM instance,
so we can support loading pxelinux config files via UUID in the file name,
too.

Acked-by: Christian Borntraeger <[email protected]>
Tested-by: Viktor Mihajlovski <[email protected]>
Signed-off-by: Thomas Huth <[email protected]>

pc-bios/s390-ccw/net: Add support for pxelinux-style config files

Since it is quite cumbersome to manually create a combined kernel with
initrd image for network booting, we now support loading via pxelinux
configuration files, too. In these files, the kernel, initrd and command
line parameters can be specified seperately, and the firmware then takes
care of glueing everything together in memory after the files have been
downloaded. See this URL for details about the config file layout:
https://www.syslinux.org/wiki/index.php?title=PXELINUX

The user can either specify a config file directly as bootfile via DHCP
(but in this case, the file has to start either with "default" or a "#"
comment so we can distinguish it from binary kernels), or a folder (i.e.
the bootfile name must end with "/") where the firmware should look for
the typical pxelinux.cfg file names, e.g. based on MAC or IP address.
We also support the pxelinux.cfg DHCP options 209 and 210 from RFC 5071.

Acked-by: Christian Borntraeger <[email protected]>
Tested-by: Viktor Mihajlovski <[email protected]>
Signed-off-by: Thomas Huth <[email protected]>

pc-bios/s390-ccw/net: Update code for the latest changes in SLOF

The ip_version information now has to be stored in the filename_ip_t
structure, and there is now a common function called tftp_get_error_info()
which can be used to get the error string for a TFTP error code.
We can also get rid of some superfluous "(char *)" casts now.

Acked-by: Christian Borntraeger <[email protected]>
Tested-by: Viktor Mihajlovski <[email protected]>
Signed-off-by: Thomas Huth <[email protected]>

roms: Update SLOF submodule to current status

We need the latest version of SLOF's libnet for adding pxelinux.cfg
support in the s390-ccw bios, too.

Signed-off-by: Thomas Huth <[email protected]>

pc-bios/s390-ccw: define loadparm length

Loadparm is defined by the s390 architecture to be 8 bytes
in length. Let's define this size in the s390-ccw bios.

Suggested-by: Laszlo Ersek <[email protected]>
Signed-off-by: Collin Walling <[email protected]>
Reviewed-by: Laszlo Ersek <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Signed-off-by: Thomas Huth <[email protected]>

block: fix QEMU crash with scsi-hd and drive_del

Removing a drive with drive_del while it is being used to run an I/O
intensive workload can cause QEMU to crash.

An AIO flush can yield at some point:

blk_aio_flush_entry()
 blk_co_flush(blk)
  bdrv_co_flush(blk->root->bs)
   ...
    qemu_coroutine_yield()

and let the HMP command to run, free blk->root and give control
back to the AIO flush:

    hmp_drive_del()
     blk_remove_bs()
      bdrv_root_unref_child(blk->root)
       child_bs = blk->root->bs
       bdrv_detach_child(blk->root)
        bdrv_replace_child(blk->root, NULL)
         blk->root->bs = NULL
        g_free(blk->root) <============== blk->root becomes stale
       bdrv_unref(child_bs)
        bdrv_delete(child_bs)
         bdrv_close()
          bdrv_drained_begin()
           bdrv_do_drained_begin()
            bdrv_drain_recurse()
             aio_poll()
              ...
              qemu_coroutine_switch()

and the AIO flush completion ends up dereferencing blk->root:

  blk_aio_complete()
   scsi_aio_complete()
    blk_get_aio_context(blk)
     bs = blk_bs(blk)
 ie, bs = blk->root ? blk->root->bs : NULL
            ^^^^^
            stale

The problem is that we should avoid making block driver graph
changes while we have in-flight requests. Let's drain all I/O
for this BB before calling bdrv_root_unref_child().

Signed-off-by: Greg Kurz <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>

test-bdrv-drain: Test graph changes in drain_all section

This tests both adding and remove a node between bdrv_drain_all_begin()
and bdrv_drain_all_end(), and enabled the existing detach test for
drain_all.

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

block: Allow graph changes in bdrv_drain_all_begin/end sections

bdrv_drain_all_*() used bdrv_next() to iterate over all root nodes and
did a subtree drain for each of them. This works fine as long as the
graph is static, but sadly, reality looks different.

If the graph changes so that root nodes are added or removed, we would
have to compensate for this. bdrv_next() returns each root node only
once even if it's the root node for multiple BlockBackends or for a
monitor-owned block driver tree, which would only complicate things.

The much easier and more obviously correct way is to fundamentally
change the way the functions work: Iterate over all BlockDriverStates,
no matter who owns them, and drain them individually. Compensation is
only necessary when a new BDS is created inside a drain_all section.
Removal of a BDS doesn't require any action because it's gone afterwards
anyway.

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

block: ignore_bds_parents parameter for drain functions

In the future, bdrv_drained_all_begin/end() will drain all invidiual
nodes separately rather than whole subtrees. This means that we don't
want to propagate the drain to all parents any more: If the parent is a
BDS, it will already be drained separately. Recursing to all parents is
unnecessary work and would make it an O(n²) operation.

Prepare the drain function for the changed drain_all by adding an
ignore_bds_parents parameter to the internal implementation that
prevents the propagation of the drain to BDS parents. We still (have to)
propagate it to non-BDS parents like BlockBackends or Jobs because those
are not drained separately.

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

block: Move bdrv_drain_all_begin() out of coroutine context

Before we can introduce a single polling loop for all nodes in
bdrv_drain_all_begin(), we must make sure to run it outside of coroutine
context like we already do for bdrv_do_drained_begin().

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

block: Allow AIO_WAIT_WHILE with NULL ctx

bdrv_drain_all() wants to have a single polling loop for draining the
in-flight requests of all nodes. This means that the AIO_WAIT_WHILE()
condition relies on activity in multiple AioContexts, which is polled
from the mainloop context. We must therefore call AIO_WAIT_WHILE() from
the mainloop thread and use the AioWait notification mechanism.

Just randomly picking the AioContext of any non-mainloop thread would
work, but instead of bothering to find such a context in the caller, we
can just as well accept NULL for ctx.

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

test-bdrv-drain: Test that bdrv_drain_invoke() doesn't poll

This adds a test case that goes wrong if bdrv_drain_invoke() calls
aio_poll().

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

block: Defer .bdrv_drain_begin callback to polling phase

We cannot allow aio_poll() in bdrv_drain_invoke(begin=true) until we're
done with propagating the drain through the graph and are doing the
single final BDRV_POLL_WHILE().

Just schedule the coroutine with the callback and increase bs->in_flight
to make sure that the polling phase will wait for it.

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

test-bdrv-drain: Graph change through parent callback

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

block: Don't poll in parent drain callbacks

bdrv_do_drained_begin() is only safe if we have a single
BDRV_POLL_WHILE() after quiescing all affected nodes. We cannot allow
that parent callbacks introduce a nested polling loop that could cause
graph changes while we're traversing the graph.

Split off bdrv_do_drained_begin_quiesce(), which only quiesces a single
node without waiting for its requests to complete. These requests will
be waited for in the BDRV_POLL_WHILE() call down the call chain.

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

test-bdrv-drain: Test node deletion in subtree recursion

If bdrv_do_drained_begin() polls during its subtree recursion, the graph
can change and mess up the bs->children iteration. Test that this
doesn't happen.

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

block: Drain recursively with a single BDRV_POLL_WHILE()

Anything can happen inside BDRV_POLL_WHILE(), including graph
changes that may interfere with its callers (e.g. child list iteration
in recursive callers of bdrv_do_drained_begin).

Switch to a single BDRV_POLL_WHILE() call for the whole subtree at the
end of bdrv_do_drained_begin() to avoid such effects. The recursion
happens now inside the loop condition. As the graph can only change
between bdrv_drain_poll() calls, but not inside of it, doing the
recursion here is safe.

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

test-bdrv-drain: Add test for node deletion

This patch adds two bdrv-drain tests for what happens if some BDS goes
away during the drainage.

The basic idea is that you have a parent BDS with some child nodes.
Then, you drain one of the children.  Because of that, the party who
actually owns the parent decides to (A) delete it, or (B) detach all its
children from it -- both while the child is still being drained.

A real-world case where this can happen is the mirror block job, which
may exit if you drain one of its children.

Signed-off-by: Max Reitz <[email protected]>
Signed-off-by: Kevin Wolf <[email protected]>