Eduardo Habkost [Mon, 5 Oct 2020 20:52:26 +0000 (16:52 -0400)]
docs: Move QTest documentation to its own document
The qtest and libqtest doc comments will be parsed to generate
API documentation, so move QTest documentation to its own
document where the API and format documentation and will be
included.
Paolo Bonzini [Tue, 6 Oct 2020 07:05:29 +0000 (09:05 +0200)]
exec: split out non-softmmu-specific parts
Over the years, most parts of exec.c that were not specific to softmmu
have been moved to accel/tcg; what's left is mostly the low-level part
of the memory API, which includes RAMBlock and AddressSpaceDispatch.
However exec.c also hosts 4-500 lines of code for the target specific
parts of the CPU QOM object, plus a few functions for user-mode
emulation that do not have a better place (they are not TCG-specific so
accel/tcg/user-exec.c is not a good place either).
Move these parts to a new file, so that exec.c can be moved to
softmmu/physmem.c.
While the FW_CFG_DATA_GENERATOR_INTERFACE is only consumed
by a device only available using system-mode (fw_cfg), it is
implemented by a crypto component (tls-cipher-suites) which
is always available when crypto is used.
Commit 69699f3055 introduced the following error in the
qemu-storage-daemon binary:
Since QOM dependencies are resolved at runtime, this issue
could not be triggered at linktime, and we don't have test
running the qemu-storage-daemon binary.
Peter Maydell [Mon, 12 Oct 2020 12:12:04 +0000 (13:12 +0100)]
Merge remote-tracking branch 'remotes/dgilbert/tags/pull-migration-20201012a' into staging
v3 Migration+ virtiofsd pull 2020-10-12
V3
Remove the postcopy recovery changes
Migration:
Dirtyrate measurement API cleanup
Virtiofsd:
Missing qemu_init_exec_dir call
Support for setting the group on socket creation
Stop a gcc warning
Avoid tempdir in sandboxing
# gpg: Signature made Mon 12 Oct 2020 12:43:30 BST
# gpg: using RSA key 45F5C71B4A0CB7FB977A9FA90516331EBC5BFDE7
# gpg: Good signature from "Dr. David Alan Gilbert (RH2) <[email protected]>" [full]
# Primary key fingerprint: 45F5 C71B 4A0C B7FB 977A 9FA9 0516 331E BC5B FDE7
* remotes/dgilbert/tags/pull-migration-20201012a:
migration/dirtyrate: present dirty rate only when querying the rate has completed
migration/dirtyrate: record start_time and calc_time while at the measuring state
virtiofsd: avoid /proc/self/fd tempdir
virtiofsd: Call qemu_init_exec_dir
tools/virtiofsd: add support for --socket-group
virtiofsd: Silence gcc warning
migration/dirtyrate: present dirty rate only when querying the rate has completed
Make dirty_rate field optional, present dirty rate only when querying
the rate has completed.
The qmp results is shown as follow:
@unstarted:
{"return":{"status":"unstarted","start-time":0,"calc-time":0},"id":"libvirt-12"}
@measuring:
{"return":{"status":"measuring","start-time":102931,"calc-time":1},"id":"libvirt-85"}
@measured:
{"return":{"status":"measured","dirty-rate":4,"start-time":150146,"calc-time":1},"id":"libvirt-15"}
migration/dirtyrate: record start_time and calc_time while at the measuring state
Querying could include both the start-time and the calc-time while at the measuring
state, allowing a caller to determine when they should expect to come back looking
for a result.
Stefan Hajnoczi [Tue, 6 Oct 2020 09:58:26 +0000 (10:58 +0100)]
virtiofsd: avoid /proc/self/fd tempdir
In order to prevent /proc/self/fd escapes a temporary directory is
created where /proc/self/fd is bind-mounted. This doesn't work on
read-only file systems.
Avoid the temporary directory by bind-mounting /proc/self/fd over /proc.
This does not affect other processes since we remounted / with MS_REC |
MS_SLAVE. /proc must exist and virtiofsd does not use it so it's safe to
do this.
Path traversal can be tested with the following function:
static void test_proc_fd_escape(struct lo_data *lo)
{
int fd;
int level = 0;
ino_t last_ino = 0;
fd = lo->proc_self_fd;
for (;;) {
struct stat st;
if (fstat(fd, &st) != 0) {
perror("fstat");
return;
}
if (last_ino && st.st_ino == last_ino) {
fprintf(stderr, "inode number unchanged, stopping\n");
return;
}
last_ino = st.st_ino;
Alex Bennée [Fri, 25 Sep 2020 12:51:29 +0000 (13:51 +0100)]
tools/virtiofsd: add support for --socket-group
If you like running QEMU as a normal user (very common for TCG runs)
but you have to run virtiofsd as a root user you run into connection
problems. Adding support for an optional --socket-group allows the
users to keep using the command line.
Gcc worries fd might be used unset, in reality it's always set if
fi is set, and only used if fi is set so it's safe. Initialise it to -1
just to keep gcc happy for now.
* remotes/armbru/tags/pull-qapi-2020-10-10: (34 commits)
qapi/visit.py: add type hint annotations
qapi/visit.py: remove unused parameters from gen_visit_object
qapi/visit.py: assert tag_member contains a QAPISchemaEnumType
qapi/types.py: remove one-letter variables
qapi/types.py: add type hint annotations
qapi/gen.py: delint with pylint
qapi/gen.py: update write() to be more idiomatic
qapi/gen.py: Remove unused parameter
qapi/gen.py: add type hint annotations
qapi/gen: Make _is_user_module() return bool
qapi/source.py: delint with pylint
qapi/source.py: add type hint annotations
qapi/commands.py: add type hint annotations
qapi/commands.py: Don't re-bind to variable of different type
qapi/events.py: Move comments into docstrings
qapi/events.py: add type hint annotations
qapi: establish mypy type-checking baseline
qapi/common.py: move build_params into gen.py
qapi/common.py: Convert comments into docstrings, and elaborate
qapi/common.py: add type hint annotations
...
Peter Maydell [Sun, 11 Oct 2020 13:34:07 +0000 (14:34 +0100)]
Merge remote-tracking branch 'remotes/stsquad/tags/pull-various-091020-1' into staging
Testing, gitdm and plugin fixes:
- fix acceptance regressions in MIPS and IDE
- speed up cirrus msys2/mingw builds
- add genisoimage to more docker images
- slew of gitdb updates
- fix some windows compile issues for plugins
- add V=1 to cirrus output
- disable rxsim in gitlab CI
# gpg: Signature made Fri 09 Oct 2020 17:30:29 BST
# gpg: using RSA key 6685AE99E75167BCAFC8DF35FBD0DB095A9E2A44
# gpg: Good signature from "Alex Bennée (Master Work Key) <[email protected]>" [full]
# Primary key fingerprint: 6685 AE99 E751 67BC AFC8 DF35 FBD0 DB09 5A9E 2A44
* remotes/stsquad/tags/pull-various-091020-1: (22 commits)
tests/acceptance: disable machine_rx_gdbsim on GitLab
cirrus: use V=1 when running tests on FreeBSD and macOS
plugin: Fixes compiling errors on msys2/mingw
plugins: Fixes a issue when dlsym failed, the handle not closed
.mailmap: Fix more contributor entries
contrib/gitdm: Add Yandex to the domain map
contrib/gitdm: Add Yadro to the domain map
contrib/gitdm: Add SUSE to the domain map
contrib/gitdm: Add Nir Soffer to Red Hat domain
contrib/gitdm: Add Qualcomm to the domain map
contrib/gitdm: Add Nuvia to the domain map
contrib/gitdm: Add Google to the domain map
contrib/gitdm: Add ByteDance to the domain map
contrib/gitdm: Add Baidu to the domain map
contrib/gitdm: Add more individual contributors
contrib/gitdm: Add more academic domains
tests/docker: Add genisoimage to the docker file
cirrus: msys2/mingw speed is up, add excluded target back
cirrus: Fixing and speedup the msys2/mingw CI
hw/ide: restore replay support of IDE
...
John Snow [Fri, 9 Oct 2020 16:15:52 +0000 (12:15 -0400)]
qapi/gen.py: update write() to be more idiomatic
Make the file handling here just a tiny bit more idiomatic.
(I realize this is heavily subjective.)
Use exist_ok=True for os.makedirs and remove the exception,
use fdopen() to wrap the file descriptor in a File-like object,
and use a context manager for managing the file pointer.
John Snow [Fri, 9 Oct 2020 16:15:46 +0000 (12:15 -0400)]
qapi/source.py: add type hint annotations
Annotations do not change runtime behavior.
This commit *only* adds annotations.
A note on typing of __init__: mypy requires init functions with no
parameters to document a return type of None to be considered fully
typed. In the case when there are input parameters, None may be omitted.
Since __init__ may never return any value, it is preferred to omit the
return annotation whenever possible.
John Snow [Fri, 9 Oct 2020 16:15:41 +0000 (12:15 -0400)]
qapi/events.py: add type hint annotations
Annotations do not change runtime behavior.
This commit *only* adds annotations.
Note: __init__ does not need its return type annotated, as it is special.
https://mypy.readthedocs.io/en/stable/class_basics.html#annotating-init-methods
John Snow [Fri, 9 Oct 2020 16:15:39 +0000 (12:15 -0400)]
qapi/common.py: move build_params into gen.py
Including it in common.py creates a circular import dependency; schema
relies on common, but common.build_params requires a type annotation
from schema. To type this properly, it needs to be moved outside the
cycle.
John Snow [Fri, 9 Oct 2020 16:15:38 +0000 (12:15 -0400)]
qapi/common.py: Convert comments into docstrings, and elaborate
As docstrings, they'll show up in documentation and IDE help.
The docstring style being targeted is the Sphinx documentation
style. Sphinx uses an extension of ReST with "domains". We use the
(implicit) Python domain, which supports a number of custom "info
fields". Those info fields are documented here:
https://www.sphinx-doc.org/en/master/usage/restructuredtext/domains.html#info-field-lists
Primarily, we use `:param X: descr`, `:return[s]: descr`, and `:raise[s]
Z: when`. Everything else is the Sphinx dialect of ReST.
(No, nothing checks or enforces this style that I am aware of. Sphinx
either chokes or succeeds, but does not enforce a standard of what is
otherwise inside the docstring. Pycharm does highlight when your param
fields are not aligned with the actual fields present. It does not
highlight missing return or exception statements. There is no existing
style guide I am aware of that covers a standard for a minimally
acceptable docstring. I am debating writing one.)
John Snow [Fri, 9 Oct 2020 16:15:34 +0000 (12:15 -0400)]
qapi/common.py: delint with pylint
At this point, that just means using a consistent strategy for constant names.
constants get UPPER_CASE and names not used externally get a leading underscore.
As a preference, while renaming constants to be UPPERCASE, move them to
the head of the file. Generally, it's nice to be able to audit the code
that runs on import in one central place.
John Snow [Fri, 9 Oct 2020 16:15:33 +0000 (12:15 -0400)]
qapi/common.py: Add indent manager
Code style tools really dislike the use of global keywords, because it
generally involves re-binding the name at runtime which can have strange
effects depending on when and how that global name is referenced in
other modules.
John Snow [Fri, 9 Oct 2020 16:15:31 +0000 (12:15 -0400)]
qapi: add pylintrc
Using `pylint --generate-rcfile > pylintrc`, generate a skeleton
pylintrc file. Sections that are not presently relevant (by the end of
this series) are removed leaving just the empty section as a search
engine / documentation hint to future authors.
I am targeting pylint 2.6.0. In the future (and hopefully before 5.2 is
released), I aim to have gitlab CI running the specific targeted
versions of pylint, mypy, flake8, etc in a job.
2.5.x will work if you additionally pass --disable=bad-whitespace.
This warning was removed from 2.6.x, for lack of consistent support.
Right now, quite a few modules are ignored as they are known to fail as
of this commit. modules will be removed from the known-bad list
throughout this and following series as they are repaired.
Note: Normally, pylintrc would go in the folder above the module, but as
that folder is shared by many things, it is going inside the module
folder (for now). Due to a bug in pylint 2.5+, pylint does not
correctly recognize when it is being run from "inside" a package, and
must be run *outside* of the package.
John Snow [Fri, 9 Oct 2020 16:15:30 +0000 (12:15 -0400)]
qapi: delint using flake8
Petty style guide fixes and line length enforcement. Not a big win, not
a big loss, but flake8 passes 100% on the qapi module, which gives us an
easy baseline to enforce hereafter.
A note on the flake8 exception: flake8 will warn on *any* bare except,
but pylint's is context-aware and will suppress the warning if you
re-raise the exception.
John Snow [Fri, 9 Oct 2020 16:15:29 +0000 (12:15 -0400)]
qapi: enforce import order/styling with isort
While we're mucking around with imports, we might as well formalize the
style we use. Let's use isort to do it for us.
lines_after_imports=2: Use two lines after imports, to match PEP8's
desire to have "two lines before and after" class definitions, which are
likely to start immediately after imports.
force_sort_within_sections: Intermingles "from x" and "import x" style
statements, such that sorting is always performed strictly on the module
name itself.
force_grid_wrap=4: Four or more imports from a single module will force
the one-per-line style that's more git-friendly. This will generally
happen for 'typing' imports.
multi_line_output=3: Uses the one-per-line indented style for long
imports.
include_trailing_comma: Adds a comma to the last import in a group,
which makes git conflicts nicer to deal with, generally.
line_length: 72 is chosen to match PEP8's "docstrings and comments" line
length limit. If you have a single line import that exceeds 72
characters, your names are too long!
John Snow [Fri, 9 Oct 2020 16:15:27 +0000 (12:15 -0400)]
qapi: Prefer explicit relative imports
All of the QAPI include statements are changed to be package-aware, as
explicit relative imports.
A quirk of Python packages is that the name of the package exists only
*outside* of the package. This means that to a module inside of the qapi
folder, there is inherently no such thing as the "qapi" package. The
reason these imports work is because the "qapi" package exists in the
context of the caller -- the execution shim, where sys.path includes a
directory that has a 'qapi' folder in it.
When we write "from qapi import sibling", we are NOT referencing the folder
'qapi', but rather "any package named qapi in sys.path". If you should
so happen to have a 'qapi' package in your path, it will use *that*
package.
When we write "from .sibling import foo", we always reference explicitly
our sibling module; guaranteeing consistency in *where* we are importing
these modules from.
This can be useful when working with virtual environments and packages
in development mode. In development mode, a package is installed as a
series of symlinks that forwards to your same source files. The problem
arises because code quality checkers will follow "import qapi.x" to the
"installed" version instead of the sibling file and -- even though they
are the same file -- they have different module paths, and this causes
cyclic import problems, false positive type mismatch errors, and more.
It can also be useful when dealing with hierarchical packages, e.g. if
we allow qemu.core.qmp, qemu.qapi.parser, etc.
John Snow [Fri, 9 Oct 2020 16:15:26 +0000 (12:15 -0400)]
qapi: move generator entrypoint into package
As part of delinting and adding type hints to the QAPI generator, it's
helpful for the entrypoint to be part of the package, only leaving a
very tiny entrypoint shim outside of the package.
John Snow [Fri, 9 Oct 2020 16:15:24 +0000 (12:15 -0400)]
qapi: modify docstrings to be sphinx-compatible
A precise style guide and a package-wide overhaul is forthcoming pending
further discussion and consensus. For now, merely avoid obvious errors
that cause Sphinx documentation build problems, using a style loosely
based on PEP 257 and Sphinx Autodoc. It is chosen for interoperability
with our existing Sphinx framework, and because it has loose recognition
in the Pycharm IDE.
Eric Blake [Wed, 30 Sep 2020 12:11:03 +0000 (07:11 -0500)]
nbd: Simplify meta-context parsing
We had a premature optimization of trying to read as little from the
wire as possible while handling NBD_OPT_SET_META_CONTEXT in phases.
But in reality, we HAVE to read the entire string from the client
before we can get to the next command, and it is easier to just read
it all at once than it is to read it in pieces. And once we do that,
several functions end up no longer performing I/O, so they can drop
length and errp parameters, and just return a bool instead of
modifying through a pointer.
Our iotests still pass; I also checked that libnbd's testsuite (which
covers more corner cases of odd meta context requests) still passes.
There are cases where the sequence of trace messages produced differs
(for example, when no bitmap is exported, a query for "qemu:" now
produces two trace lines instead of one), but trace points are for
debug and have no effect on what the client sees.
Eric Blake [Wed, 30 Sep 2020 12:11:02 +0000 (07:11 -0500)]
nbd/server: Reject embedded NUL in NBD strings
The NBD spec is clear that any string sent from the client must not
contain embedded NUL characters. If the client passes "a\0", we
should reject that option request rather than act on "a".
Testing this is not possible with a compliant client, but I was able
to use gdb to coerce libnbd into temporarily behaving as such a
client.
Eric Blake [Wed, 30 Sep 2020 12:11:01 +0000 (07:11 -0500)]
qemu-nbd: Honor SIGINT and SIGHUP
Honoring just SIGTERM on Linux is too weak; we also want to handle
other common signals, and do so even on BSD. Why? Because at least
'qemu-nbd -B bitmap' needs a chance to clean up the in-use bit on
bitmaps when the server is shut down via a signal.
block/nbd: nbd_co_reconnect_loop(): don't connect if drained
In a recent commit 12c75e20a269ac we've improved
nbd_co_reconnect_loop() to not make drain wait for additional sleep.
Similarly, we shouldn't try to connect, if previous sleep was
interrupted by drain begin, otherwise drain_begin will have to wait for
the whole connection attempt.
reconnect-delay has a design flaw: we handle it in the same loop where
we do connection attempt. So, reconnect-delay may be exceeded by
unpredictable time of connection attempt.
Let's instead use separate timer.
How to reproduce the bug:
1. Create an image on node1:
qemu-img create -f qcow2 xx 100M
4. Type 'read 0 512' in qemu-io interface to check that connection
works
Be careful: you should make steps 5-7 in a short time, less than 15
seconds.
5. Kill nbd server on node1
6. Run 'read 0 512' in qemu-io interface again, to be sure that nbd
client goes to reconnect loop.
7. On node1 run the following command
sudo iptables -A INPUT -p tcp --dport 10809 -j DROP
This will make the connect() call of qemu-io at node2 take a long time.
And you'll see that read command in qemu-io will hang for a long time,
more than 15 seconds specified by reconnect-delay parameter. It's the
bug.
8. Don't forget to drop iptables rule on node1:
sudo iptables -D INPUT -p tcp --dport 10809 -j DROP
Important note: Step [5] is necessary to reproduce _this_ bug. If we
miss step [5], the read command (step 6) will hang for a long time and
this commit doesn't help, because there will be not long connect() to
unreachable host, but long sendmsg() to unreachable host, which should
be fixed by enabling and adjusting keep-alive on the socket, which is a
thing for further patch set.
block/nbd: correctly use qio_channel_detach_aio_context when needed
Don't use nbd_client_detach_aio_context() driver handler where we want
to finalize the connection. We should directly use
qio_channel_detach_aio_context() in such cases. Driver handler may (and
will) contain another things, unrelated to the qio channel.
block/nbd: fix drain dead-lock because of nbd reconnect-delay
We pause reconnect process during drained section. So, if we have some
requests, waiting for reconnect we should cancel them, otherwise they
deadlock the drained section.
How to reproduce:
1. Create an image:
qemu-img create -f qcow2 xx 100M
2. Start NBD server:
qemu-nbd xx
3. Start vm with second nbd disk on node2, like this:
4. Access the vm through vnc (or some other way?), and check that NBD
drive works:
dd if=/dev/sdb of=/dev/null bs=1M count=10
- the command should succeed.
5. Now, kill the nbd server, and run dd in the guest again:
dd if=/dev/sdb of=/dev/null bs=1M count=10
Now Qemu is trying to reconnect, and dd-generated requests are waiting
for the connection (they will wait up to 60 seconds (see
reconnect-delay option above) and than fail). But suddenly, vm may
totally hang in the deadlock. You may need to increase reconnect-delay
period to catch the dead-lock.
VM doesn't respond because drain dead-lock happens in cpu thread with
global mutex taken. That's not good thing by itself and is not fixed
by this commit (true way is using iothreads). Still this commit fixes
drain dead-lock itself.
Note: probably, we can instead continue to reconnect during drained
section. To achieve this, we may move negotiation to the connect thread
to make it independent of bs aio context. But expanding drained section
doesn't seem good anyway. So, let's now fix the bug the simplest way.
Compiling C object libblock.fa.p/nbd_server.c.o
../nbd/server.c: In function ‘nbd_co_client_start’:
../nbd/server.c:625:14: error: ‘namelen’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
625 | rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
626 | errp);
| ~~~~~
../nbd/server.c:564:14: note: ‘namelen’ was declared here
564 | uint32_t namelen;
| ^~~~~~~
cc1: all warnings being treated as errors
As I cannot see how this can happen, let uns silence the warning.
Alex Bennée [Wed, 7 Oct 2020 16:00:38 +0000 (17:00 +0100)]
tests/acceptance: disable machine_rx_gdbsim on GitLab
While I can get the ssh test to fail on my test setup this seems a lot
more stable except when on GitLab. Hopefully we can re-enable both
once the serial timing patches have been added.
Paolo Bonzini [Wed, 7 Oct 2020 16:00:37 +0000 (17:00 +0100)]
cirrus: use V=1 when running tests on FreeBSD and macOS
Using "V=1" makes it easier to identify hanging tests, especially
since they are run with -j1. It is already used on Windows builds,
do the same for FreeBSD and macOS.
Yonggang Luo [Wed, 7 Oct 2020 16:00:20 +0000 (17:00 +0100)]
cirrus: Fixing and speedup the msys2/mingw CI
Use cache of cirrus caching msys2
The install of msys2 are refer to https://github.com/msys2/setup-msys2
The first time install msys2 would be time consuming, so increase timeout_in to 90m
according to https://cirrus-ci.org/faq/#instance-timed-out
Apply patch of https://lists.gnu.org/archive/html/qemu-devel/2020-10/msg00072.html
Alex Bennée [Wed, 7 Oct 2020 16:00:19 +0000 (17:00 +0100)]
hw/ide: restore replay support of IDE
A recent change to weak reset handling broke replay due to the use of
aio_bh_schedule_oneshot instead of the replay aware
replay_bh_schedule_oneshot_event.
In commit 102ca9667d we set "start-powered-off" on all vCPUs
included in the CPS (Coherent Processing System) but forgot to
start the vCPUS on when they are powered on in the CPC (Cluster
Power Controller).
This fixes the following tests:
$ avocado run tests/acceptance/machine_mips_malta.py
(1/3) test_mips_malta_i6400_framebuffer_logo_1core: PASS (3.67 s)
(2/3) test_mips_malta_i6400_framebuffer_logo_7cores: INTERRUPTED: Test interrupted by SIGTERM (30.22 s)
(3/3) test_mips_malta_i6400_framebuffer_logo_8cores: INTERRUPTED: Test interrupted by SIGTERM (30.25 s)
RESULTS : PASS 1 | ERROR 0 | FAIL 0 | SKIP 0 | WARN 0 | INTERRUPT 2 | CANCEL 0
Paolo Bonzini [Wed, 7 Oct 2020 16:00:17 +0000 (17:00 +0100)]
configure: fix performance regression due to PIC objects
Because most files in QEMU are grouped into static libraries, Meson conservatively
compiles them with -fPIC. This is overkill and produces slowdowns up to 20% on
some TCG tests.
As a stopgap measure, use the b_staticpic option to limit the slowdown to
--enable-pie. https://github.com/mesonbuild/meson/pull/7760 will allow
us to use b_staticpic=false and let Meson do the right thing.
* remotes/dgibson/tags/ppc-for-5.2-20201009:
specs/ppc-spapr-numa: update with new NUMA support
spapr_numa: consider user input when defining associativity
spapr_numa: change reference-points and maxdomain settings
spapr_numa: forbid asymmetrical NUMA setups
spapr: add spapr_machine_using_legacy_numa() helper
ppc/pnv: Increase max firmware size
spapr: Add a return value to spapr_check_pagesize()
spapr: Add a return value to spapr_nvdimm_validate()
spapr: Simplify error handling in spapr_cpu_core_realize()
spapr: Add a return value to spapr_set_vcpu_id()
spapr: Simplify error handling in prop_get_fdt()
spapr: Add a return value to spapr_drc_attach()
spapr: Simplify error handling in spapr_vio_busdev_realize()
spapr: Simplify error handling in do_client_architecture_support()
spapr: Get rid of cas_check_pvr() error reporting
spapr: Simplify error handling in callers of ppc_set_compat()
ppc: Fix return value in cpu_post_load() error path
ppc: Add a return value to ppc_set_compat() and ppc_set_compat_all()
spapr: Fix error leak in spapr_realize_vcpu()
spapr: Handle HPT allocation failure in nested guest
* remotes/armbru/tags/pull-monitor-2020-10-09:
block: Convert 'block_resize' to coroutine
block: Add bdrv_lock()/unlock()
block: Add bdrv_co_enter()/leave()
util/async: Add aio_co_reschedule_self()
hmp: Add support for coroutine command handlers
qmp: Move dispatcher to a coroutine
qapi: Add a 'coroutine' flag for commands
monitor: Make current monitor a per-coroutine property
qmp: Call monitor_set_cur() only in qmp_dispatch()
qmp: Assert that no other monitor is active
hmp: Update current monitor only in handle_hmp_command()
monitor: Use getter/setter functions for cur_mon
monitor: Add Monitor parameter to monitor_get_cpu_index()
monitor: Add Monitor parameter to monitor_set_cpu()
Kevin Wolf [Mon, 5 Oct 2020 15:58:55 +0000 (17:58 +0200)]
block: Convert 'block_resize' to coroutine
block_resize performs some I/O that could potentially take quite some
time, so use it as an example for the new 'coroutine': true annotation
in the QAPI schema.
bdrv_truncate() requires that we're already in the right AioContext for
the BlockDriverState if called in coroutine context. So instead of just
taking the AioContext lock, move the QMP handler coroutine to the
context.
Call blk_unref() only after switching back because blk_unref() may only
be called in the main thread.
Kevin Wolf [Mon, 5 Oct 2020 15:58:54 +0000 (17:58 +0200)]
block: Add bdrv_lock()/unlock()
Inside of coroutine context, we can't directly use aio_context_acquire()
for the AioContext of a block node because we already own the lock of
the current AioContext and we need to avoid double locking to prevent
deadlocks.
This provides helper functions to lock the AioContext of a node only if
it's not the same as the current AioContext.
Kevin Wolf [Mon, 5 Oct 2020 15:58:51 +0000 (17:58 +0200)]
hmp: Add support for coroutine command handlers
Often, QMP command handlers are not only called to handle QMP commands,
but also from a corresponding HMP command handler. In order to give them
a consistent environment, optionally run HMP command handlers in a
coroutine, too.
The implementation is a lot simpler than in QMP because for HMP, we
still block the VM while the coroutine is running.
projects / qemu.git / log