Peter Maydell [Mon, 19 Jul 2021 10:52:57 +0000 (11:52 +0100)]
qemu-options.hx: Fix formatting of -machine memory-backend option
The documentation of the -machine memory-backend has some minor
formatting errors:
* Misindentation of the initial line meant that the whole option
section is incorrectly indented in the HTML output compared to
the other -machine options
* The examples weren't indented, which meant that they were formatted
as plain run-on text including outputting the "::" as text.
* The a) b) list has no rst-format markup so it is rendered as
a single run-on paragraph
* remotes/rth-gitlab/tags/pull-tcg-20210726:
tests/unit: Remove unused variable from test_io
linux-user/syscall: Remove unused variable from execve
hw/pci-hist/pnv_phb4: Fix typo in pnv_phb4_ioda_write
hw/ppc/spapr_events: Remove unused variable from check_exception
hw/audio/adlib: Remove unused variable in adlib_callback
net/checksum: Remove unused variable in net_checksum_add_iov
util/selfmap: Discard mapping on error
accel/tcg: Remove unused variable in cpu_exec
nbd/server: Mark variable unused in nbd_negotiate_meta_queries
bitops.h: revert db1ffc32dd ("qemu/bitops.h: add bitrev8 implementation")
accel/tcg: Remove unnecessary check on icount_extra in cpu_loop_exec_tb()
accel/tcg: Don't use CF_COUNT_MASK as the max value of icount_decr.u16.low
* remotes/kraxel/tags/fixes-20210726-pull-request:
ui/gtk: add a keyboard fifo to the VTE consoles
ui: update keycodemapdb submodule commit
ui/cocoa: Fix the type of main's argv
ui/egl-headless: Remove a check for CONFIG_OPENGL
ui/spice: Use HAVE_SPICE_GL for OpenGL checks
ui/gtk: Fix relative mouse with multiple monitors
From clang-13:
util/selfmap.c:26:21: error: variable 'errors' set but not used \
[-Werror,-Wunused-but-set-variable]
Quite right of course, but there's no reason not to check errors.
First, incrementing errors is incorrect, because qemu_strtoul
returns an errno not a count -- just or them together so that
we have a non-zero value at the end.
Second, if we have an error, do not add the struct to the list,
but free it instead.
Commit db1ffc32dd ("qemu/bitops.h: add bitrev8 implementation") introduced
a bitrev8() function to reverse the bit ordering required for storing the
MAC address in the q800 PROM.
This function is not required since QEMU implements its own revbit8()
function which does exactly the same thing. Remove the extraneous
bitrev8() function and switch its only caller in hw/m68k/q800.c to
use revbit8() instead.
Peter Maydell [Sun, 25 Jul 2021 17:44:05 +0000 (18:44 +0100)]
accel/tcg: Remove unnecessary check on icount_extra in cpu_loop_exec_tb()
In cpu_loop_exec_tb(), we decide whether to look for a TB with
exactly insns_left instructions in it using the condition
(!cpu->icount_extra && insns_left > 0 && insns_left < tb->icount)
The check for icount_extra == 0 is unnecessary, because we just set
insns_left = MIN(0xffff, cpu->icount_budget);
icount_extra = icount_budget - insns_left;
and so icount_extra can only be non-zero if icount_budget > 0xffff
and insns_left == 0xffff. But in that case insns_left >= tb->icount
because 0xffff is much larger than TCG_MAX_INSNS, so the condition
will be false anyway.
Remove the unnecessary check, and instead assert:
* that we are only going to execute a partial TB here if the
icount budget has run out (ie icount_extra == 0)
* that the number of insns we're going to execute does fit into
the CF_COUNT_MASK
Peter Maydell [Sun, 25 Jul 2021 17:44:04 +0000 (18:44 +0100)]
accel/tcg: Don't use CF_COUNT_MASK as the max value of icount_decr.u16.low
In cpu_loop_exec_tb() we were bounding the number of insns we might
try to execute in a TB using CF_COUNT_MASK. This is incorrect,
because we can validly put up to 0xffff into icount_decr.u16.low. In
particular, since commit 78ff82bb1b67c0d7 reduced CF_COUNT_MASK to
511 this meant that we would incorrectly only try to execute 511
instructions in a 512-instruction TB, which could result in QEMU
hanging when in icount mode.
Use the actual maximum value, which is 0xffff. (This brings this code
in to line with the similar logic in icount_prepare_for_run() in
tcg-accel-ops-icount.c.)
Peter Maydell [Mon, 26 Jul 2021 12:36:51 +0000 (13:36 +0100)]
Merge remote-tracking branch 'remotes/quic/tags/pull-hex-20210725' into staging
The Hexagon target was silently failing the SIGSEGV test because
the signal handler was not called.
Patch 1/2 fixes the Hexagon target
Patch 2/2 drops include qemu.h from target/hexagon/op_helper.c
**** Changes in v2 ****
Drop changes to linux-test.c due to intermittent failures on riscv
# gpg: Signature made Sun 25 Jul 2021 22:39:38 BST
# gpg: using RSA key 7B0244FB12DE4422
# gpg: Good signature from "Taylor Simpson (Rock on) <[email protected]>" [undefined]
# 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: 3635 C788 CE62 B91F D4C5 9AB4 7B02 44FB 12DE 4422
* remotes/quic/tags/pull-hex-20210725:
target/hexagon: Drop include of qemu.h
Hexagon (target/hexagon) remove put_user_*/get_user_*
Peter Maydell [Mon, 26 Jul 2021 10:00:14 +0000 (11:00 +0100)]
Merge remote-tracking branch 'remotes/stsquad/tags/pull-for-6.1-rc1-230721-1' into staging
Doc, metadata, plugin and testing updates for 6.1-rc1:
- git ignore some file editor detritus
- add overview on device emulation terminology
- remove needless if leg in configure custom devices logic
- numerous gitdm/mailmap updates
- fix plugin_exit race for linux-user
- fix a few bugs in cache modelling plugin
- fix plugin calculation of physical address
- handle pure assembler/linker tcg tests outside of docker
- add tricore build to gitlab
- remove superfluous MacOSX task
- generalise the OpenBSI gitlab rules
# gpg: Signature made Fri 23 Jul 2021 17:28:26 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-for-6.1-rc1-230721-1: (28 commits)
gitlab-ci: Extract OpenSBI job rules to reusable section
gitlab-ci: Remove the second superfluous macos task
gitlab: enable a very minimal build with the tricore container
tests/tcg/configure.sh: add handling for assembler only builds
plugins: Fix physical address calculation for IO regions
plugins/cache: Fixed "function decl. is not a prototype" warnings
plugins/cache: limited the scope of a mutex lock
plugins/cache: Fixed a bug with destroying FIFO metadata
tcg/plugins: implement a qemu_plugin_user_exit helper
contrib/gitdm: add more individual contributor entries.
contrib/gitdm: add a new interns group-map for GSoC/Outreachy work
contrib/gitdm: add an explicit academic entry for BU
contrib/gitdm: add group-map for Netflix
contrib/gitdm: add domain-map for NVIDIA
contrib/gitdm: add domain-map for Crudebyte
contrib/gitdm: un-ironically add a mapping for LWN
contrib/gitdm: add domain-map/group-map for Wind River
contrib/gitdm: add domain-map for Eldorado
contrib/gitdm: add domain-map/group-map mappings for Samsung
gitdm.config: sort the corporate GroupMap entries
...
block/nvme: Fix VFIO_MAP_DMA failed: No space left on device
When the NVMe block driver was introduced (see commit bdd6a90a9e5,
January 2018), Linux VFIO_IOMMU_MAP_DMA ioctl was only returning
-ENOMEM in case of error. The driver was correctly handling the
error path to recycle its volatile IOVA mappings.
To fix CVE-2019-3882, Linux commit 492855939bdb ("vfio/type1: Limit
DMA mappings per container", April 2019) added the -ENOSPC error to
signal the user exhausted the DMA mappings available for a container.
The block driver started to mis-behave:
qemu-system-x86_64: VFIO_MAP_DMA failed: No space left on device
(qemu)
(qemu) info status
VM status: paused (io-error)
(qemu) c
VFIO_MAP_DMA failed: No space left on device
(qemu) c
VFIO_MAP_DMA failed: No space left on device
(The VM is not resumable from here, hence stuck.)
Fix by handling the new -ENOSPC error (when DMA mappings are
exhausted) without any distinction to the current -ENOMEM error,
so we don't change the behavior on old kernels where the CVE-2019-3882
fix is not present.
An easy way to reproduce this bug is to restrict the DMA mapping
limit (65535 by default) when loading the VFIO IOMMU module:
Since commit 8eb13bbbac ("ui/gtk: vte: fix sending multiple
characeters") it's very easy to lock up QEMU with the GTK ui.
If you configure a guest with a serial device and the guest
doesn't listen on this device, QEMU will lock up after
entering two characters in the serial console. That's because
current code uses a busy loop for the chardev write retries
and the busy loop doesn't terminate in this case.
To fix this problem add a fifo to the VTE consoles and use the
chr_accept_input() callback function to write the remaining
characters in the queue to the chardev.
The fifo has a size of 4096 bytes, so one can copy and paste
a fairly large URL or file path.
ui/egl-headless is only built when CONFIG_OPENGL is defined because it
depends on CONFIG_OPENGL without condition. Remove a redundant
conditonal in ui/egl-headless.c
Some code in ui/spice used CONFIG_OPENGL for OpenGL conditionals, but
SPICE also depends on CONFIG_GBM and SPICE server whose version is
0.13.1 or later for OpenGL. Always use HAVE_SPICE_GL, which defines the
precise condition.
Dennis Wölfing [Tue, 20 Jul 2021 14:39:41 +0000 (16:39 +0200)]
ui/gtk: Fix relative mouse with multiple monitors
To handle relative mouse input the event handler needs to move the mouse
away from the screen edges. Failing to do so results in the mouse
getting stuck at invisible walls. However the current implementation for
this is broken on hosts with multiple monitors.
With multiple monitors the mouse can be located outside of the current
monitor which is not handled by the current code. Also the monitor
itself might be located at coordinates different from (0, 0).
* remotes/bonzini-gitlab/tags/for-upstream:
qom: use correct field name when getting/setting alias properties
qapi: introduce forwarding visitor
gitlab: only let pages be published from default branch
MAINTAINERS: Add memory_mapping.h and memory_mapping.c to "Memory API"
MAINTAINERS: Add Peter Xu and myself as co-maintainer of "Memory API"
MAINTAINERS: Replace Eduardo as "Host Memory Backends" maintainer
i386: do not call cpudef-only models functions for max, host, base
target/i386: Added consistency checks for CR3
meson: fix dependencies for modinfo #2
gitlab-ci: Extract OpenSBI job rules to reusable section
All jobs depending on 'docker-opensbi' job must use at most all
the rules that triggers it. The simplest way to ensure that
is to always use the same rules. Extract all the rules to a
reusable section, and include this section (with the 'extends'
keyword) in both 'docker-opensbi' and 'build-opensbi' jobs.
The problem was introduced in commit c6fc0fc1a71 ("gitlab-ci.yml:
Add jobs to build OpenSBI firmware binaries"), but was revealed in
commit 91e9c47e50a ("docker: OpenSBI build job depends on OpenSBI
container").
This fix is similar to the one used with the EDK2 firmware job in
commit ac0595cf6b3 ("gitlab-ci: Extract EDK2 job rules to reusable
section").
Thomas Huth [Tue, 20 Jul 2021 23:27:02 +0000 (00:27 +0100)]
gitlab-ci: Remove the second superfluous macos task
While there might have been bigger differnces between the -base and
the -xcode images in the beginning, they almost vanished in the
current builds, e.g. when comparing the output of the "configure"
step after cleaning up the differences due to temporary path names,
I only get:
$ diff -u /tmp/base.txt /tmp/xcode.txt
--- /tmp/base.txt 2021-07-16 09:16:24.211427940 +0200
+++ /tmp/xcode.txt 2021-07-16 09:16:43.029684274 +0200
@@ -19,14 +19,14 @@
Build type: native build
Project name: qemu
Project version: 6.0.50
-C compiler for the host machine: cc (clang 12.0.0 "Apple clang version 12.0.0 (clang-1200.0.32.29)")
+C compiler for the host machine: cc (clang 12.0.0 "Apple clang version 12.0.0 (clang-1200.0.32.28)")
C linker for the host machine: cc ld64 609.8
Host machine cpu family: x86_64
Host machine cpu: x86_64
Program sh found: YES (/bin/sh)
Program python3 found: YES (/usr/local/opt/[email protected]/bin/python3.9)
Program bzip2 found: YES (/usr/bin/bzip2)
-C++ compiler for the host machine: c++ (clang 12.0.0 "Apple clang version 12.0.0 (clang-1200.0.32.29)")
+C++ compiler for the host machine: c++ (clang 12.0.0 "Apple clang version 12.0.0 (clang-1200.0.32.28)")
C++ linker for the host machine: c++ ld64 609.8
Objective-C compiler for the host machine: clang (clang 12.0.0)
Objective-C linker for the host machine: clang ld64 609.8
Since we're not using Xcode itself at all, it seems like it does not
make much sense anymore to waste compute cycles with two images here.
Thus let's delete the -xcode job now.
Alex Bennée [Tue, 20 Jul 2021 23:27:01 +0000 (00:27 +0100)]
gitlab: enable a very minimal build with the tricore container
Rather than base of the shared Debian 10 container which would require
us to bring in even more dependencies just bring in what is needed for
building tricore-softmmu in GitLab. We don't even remove the container
from the DOCKER_PARTIAL_IMAGES lest we cause more confusion.
Alex Bennée [Tue, 20 Jul 2021 23:27:00 +0000 (00:27 +0100)]
tests/tcg/configure.sh: add handling for assembler only builds
Up until this point we only handled local compilers or assumed we had
everything in the container. This falls down when we are building QEMU
inside the container.
This special handling only affects tricore for now but I put it in a
case just in case we add any other "special" targets. Setting
CROSS_CC_GUEST is a bit of a hack just to ensure the test runs as we
gate on a detected compiler even though the Makefile won't actually
use it. It also means we display something sane in the configure
output.
plugins/cache: Fixed a bug with destroying FIFO metadata
This manifests itself when associativity degree is greater than the
number of sets and FIFO is used, otherwise it's also a memory leak
whenever FIFO was used.
Alex Bennée [Tue, 20 Jul 2021 23:26:54 +0000 (00:26 +0100)]
tcg/plugins: implement a qemu_plugin_user_exit helper
In user-mode emulation there is a small race between preexit_cleanup
and exit_group() which means we may end up calling instrumented
instructions before the kernel reaps child threads. To solve this we
implement a new helper which ensures the callbacks are flushed along
with any translations before we let the host do it's a thing.
While we are at it make the documentation of
qemu_plugin_register_atexit_cb clearer as to what the user can expect.
Alex Bennée [Tue, 20 Jul 2021 23:26:40 +0000 (00:26 +0100)]
.mailmap: fix up some broken commit authors
Fixes: 49a6f3bffb ("target/arm: Correct the encoding of MDCCSR_EL0 and DBGDSCRint") Fixes: 5a07192a04 ("target/i386: Fix handling of k_gs_base register in 32-bit mode in gdbstub") Signed-off-by: Alex Bennée <[email protected]> Cc: Nick Hudson <[email protected]> Cc: Marek Dolata <[email protected]>
Message-Id: <20210720232703[email protected]>
Alex Bennée [Tue, 20 Jul 2021 23:26:37 +0000 (00:26 +0100)]
docs: add a section on the generalities of vhost-user
While we do mention some of this stuff in the various daemons and
manuals the subtleties of the socket and memory sharing are sometimes
missed. This document attempts to give some background on vhost-user
daemons in general terms.
Alex Bennée [Tue, 20 Jul 2021 23:26:36 +0000 (00:26 +0100)]
docs: collect the disparate device emulation docs into one section
While we are at it add a brief preamble that explains some of the
common concepts in QEMU's device emulation which will hopefully lead
to less confusing about our dizzying command line options.
Paolo Bonzini [Sun, 18 Jul 2021 06:50:44 +0000 (08:50 +0200)]
qom: use correct field name when getting/setting alias properties
Alias targets have a different name than the alias property itself
(e.g. a machine's pflash0 might be an alias of a property named 'drive').
When the target's getter or setter invokes the visitor, it will use
a different name than what the caller expects, and the visitor will
not be able to find it (or will consume erroneously).
The solution is for alias getters and setters to wrap the incoming
visitor, and forward the sole field that the target is expecting while
renaming it appropriately.
This bug has been there forever, but it was exposed after -M parsing
switched from QemuOptions and StringInputVisitor to keyval and
QObjectInputVisitor. Before, the visitor ignored the name. Now, it
checks "drive" against what was passed on the command line and finds
that no such property exists.
Fixes: https://gitlab.com/qemu-project/qemu/-/issues/484 Reported-by: Alex Williamson <[email protected]> Signed-off-by: Paolo Bonzini <[email protected]>
Paolo Bonzini [Sun, 18 Jul 2021 06:49:22 +0000 (08:49 +0200)]
qapi: introduce forwarding visitor
This new adaptor visitor takes a single field of the adaptee, and exposes it
with a different name.
This will be used for QOM alias properties. Alias targets can of course
have a different name than the alias property itself (e.g. a machine's
pflash0 might be an alias of a property named 'drive'). When the target's
getter or setter invokes the visitor, it will use a different name than
what the caller expects, and the visitor will not be able to find it
(or will consume erroneously).
The solution is for alias getters and setters to wrap the incoming
visitor, and forward the sole field that the target is expecting while
renaming it appropriately.
"Remember that GitLab Pages are by default branch/tag agnostic
and their deployment relies solely on what you specify in
.gitlab-ci.yml. You can limit the pages job with the only
parameter, whenever a new commit is pushed to a branch used
specifically for your pages."
The current "pages" job is not limited, so it is happily publishing
docs content from any branch/tag in qemu.git that gets pushed to.
This means we're potentially publishing from the "staging" branch
or worse from outdated "stable-NNN" branches
This change restricts it to only publish from the default branch
in the main repository. For contributor forks, however, we allow
it to publish from any branch, since users will have arbitrarily
named topic branches in flight at any time.
i386: do not call cpudef-only models functions for max, host, base
Some cpu properties have to be set only for cpu models in builtin_x86_defs,
registered with x86_register_cpu_model_type, and not for
cpu models "base", "max", and the subclass "host".
These properties are the ones set by function x86_cpu_apply_props,
(also including kvm_default_props, tcg_default_props),
and the "vendor" property for the KVM and HVF accelerators.
After recent refactoring of cpu, which also affected these properties,
they were instead set unconditionally for all x86 cpus.
This has been detected as a bug with Nested on AMD with cpu "host",
as svm was not turned on by default, due to the wrongful setting of
kvm_default_props via x86_cpu_apply_props, which set svm to "off".
Rectify the bug introduced in commit "i386: split cpu accelerators"
and document the functions that are builtin_x86_defs-only.
All MBZ in CR3 must be zero (APM2 15.5)
Added checks in both helper_vmrun and helper_write_crN.
When EFER.LMA is zero the upper 32 bits needs to be zeroed.
modinfo runs the preprocessor and therefore needs all generated input files
to be there. The "depends" clause does not work in Meson 0.55.3, so for
now use "input".
Part #2: Update the rule for target-specific modules too.
* remotes/bonzini-gitlab/tags/for-upstream:
configure: Let --without-default-features disable vhost-kernel and vhost-vdpa
configure: Fix the default setting of the "xen" feature
configure: Allow vnc to get disabled with --without-default-features
configure: Fix --without-default-features propagation to meson
meson: fix dependencies for modinfo
configure: Drop obsolete check for the alloc_size attribute
target/i386: Added consistency checks for EFER
target/i386: Added consistency checks for CR4
target/i386: Added V_INTR_PRIO check to virtual interrupts
qemu-config: restore "machine" in qmp_query_command_line_options()
usb: fix usb-host dependency check
chardev-spice: add missing module_obj directive
vl: Parse legacy default_machine_opts
qemu-config: fix memory leak on ferror()
qemu-config: never call the callback after an error, fix leak
Before commit 49afbca3b00e8e517d54964229a794b51768deaf ("virtio-gpu: drop
use_virgl_renderer"), use_virgl_renderer was preventing calling GL
functions from non-GL context threads. The innocuously looking
g->parent_obj.use_virgl_renderer = false;
was set the first time virtio_gpu_gl_reset() was called, during
pc_machine_reset() in the main thread. Further virtio_gpu_gl_reset()
calls in IO threads, without associated GL context, were thus skipping
GL calls and avoided warnings or crashes (see also
https://gitlab.freedesktop.org/virgl/virglrenderer/-/issues/226).
Since commit 551dbd0846d2 ("migration: check pre_save return in
vmstate_save_state") the pre_save hook can fail. So lets finally
use that to drop the guest-triggerable assert in qxl_pre_save().
* remotes/rth-gitlab/tags/pull-tcg-20210721: (27 commits)
accel/tcg: Record singlestep_enabled in tb->cflags
accel/tcg: Hoist tb_cflags to a local in translator_loop
accel/tcg: Remove TranslatorOps.breakpoint_check
accel/tcg: Move breakpoint recognition outside translation
accel/tcg: Merge tb_find into its only caller
target/avr: Implement gdb_adjust_breakpoint
hw/core: Introduce CPUClass.gdb_adjust_breakpoint
target/i386: Implement debug_check_breakpoint
target/arm: Implement debug_check_breakpoint
hw/core: Introduce TCGCPUOps.debug_check_breakpoint
accel/tcg: Use CF_NO_GOTO_{TB, PTR} in cpu_exec_step_atomic
accel/tcg: Handle -singlestep in curr_cflags
accel/tcg: Drop CF_NO_GOTO_PTR from -d nochain
accel/tcg: Add CF_NO_GOTO_TB and CF_NO_GOTO_PTR
target/alpha: Drop goto_tb path in gen_call_pal
accel/tcg: Move curr_cflags into cpu-exec.c
accel/tcg: Reduce CF_COUNT_MASK to match TCG_MAX_INSNS
accel/tcg: Push trace info building into atomic_common.c.inc
trace: Fold mem-internal.h into mem.h
accel/tcg: Expand ATOMIC_MMU_LOOKUP_*
...
Thomas Huth [Tue, 13 Jul 2021 09:31:55 +0000 (11:31 +0200)]
configure: Let --without-default-features disable vhost-kernel and vhost-vdpa
The vhost_kernel and vhost_vdpa variables should be pre-initialized with
the $default_feature setting so that these features get disabled when
the user runs the configure scripts with --without-default-features.
Thomas Huth [Tue, 13 Jul 2021 09:31:54 +0000 (11:31 +0200)]
configure: Fix the default setting of the "xen" feature
The "xen" variable should either contain "enabled", "disabled" or
nothing (for auto detection). But when the user currently runs the
configure script with --without-default-features, it gets set to
"no" instead. This does not work as expected, the feature will still
be enabled if the Xen headers are present. Thus set the variable
to "disabled" instead if default_feature switch has been set.
Paolo Bonzini [Wed, 21 Jul 2021 16:51:57 +0000 (18:51 +0200)]
meson: fix dependencies for modinfo
modinfo runs the preprocessor and therefore needs all generated input files
to be there. The "depends" clause does not work in Meson 0.55.3, so for
now use "input".
EFER.SVME has to be set, and EFER reserved bits must
be zero.
In addition the combinations
* EFER.LMA or EFER.LME is non-zero and the processor does not support LM
* non-zero EFER.LME and CR0.PG and zero CR4.PAE
* non-zero EFER.LME and CR0.PG and zero CR0.PE
* non-zero EFER.LME, CR0.PG, CR4.PAE, CS.L and CS.D
are all invalid.
(AMD64 Architecture Programmer's Manual, V2, 15.5)
target/i386: Added V_INTR_PRIO check to virtual interrupts
The APM2 states that The processor takes a virtual INTR interrupt
if V_IRQ and V_INTR_PRIO indicate that there is a virtual interrupt pending
whose priority is greater than the value in V_TPR.
Stefan Hajnoczi [Wed, 21 Jul 2021 15:10:55 +0000 (16:10 +0100)]
qemu-config: restore "machine" in qmp_query_command_line_options()
Commit d8fb7d0969d5c32b3d1b9e20b63ec6c0abe80be4 ("vl: switch -M parsing
to keyval") stopped adding the "machine" QemuOptsList. This causes
"machine" options to not show up in QMP query-command-line-options
output. For example, libvirt cannot detect that kernel_irqchip support
is available.
Adjust the "machine" opts enumeration in
qmp_query_command_line_options() so that options are properly reported.
To avoid leaking memory on the error path, reorder the
code as:
- check the parameters first
- check resource already existing
- finally allocate memory
* remotes/ericb/tags/pull-bitmaps-2021-07-21:
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps'
qemu-img: Fail fast on convert --bitmaps with inconsistent bitmap
iotests: Improve and rename test 291 to qemu-img-bitmap
Peter Maydell [Sat, 17 Jul 2021 10:30:17 +0000 (11:30 +0100)]
target/hexagon: Drop include of qemu.h
The qemu.h file is a CONFIG_USER_ONLY header; it doesn't appear on
the include path for softmmu builds. Currently we include it
unconditionally in target/hexagon/op_helper.c. We used to need it
for the put_user_*() and get_user_*() functions, but now that we have
removed the uses of those from op_helper.c, the only reason it's
still there is that we're implicitly relying on it pulling in some
other headers.
Explicitly include the headers we need for other functions, and drop
the include of qemu.h.
Eric Blake [Wed, 21 Jul 2021 15:53:48 +0000 (10:53 -0500)]
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps'
The point of 'qemu-img convert --bitmaps' is to be a convenience for
actions that are already possible through a string of smaller
'qemu-img bitmap' sub-commands. One situation not accounted for
already is that if a source image contains an inconsistent bitmap (for
example, because a qemu process died abruptly before flushing bitmap
state), the user MUST delete those inconsistent bitmaps before
anything else useful can be done with the image.
We don't want to delete inconsistent bitmaps by default: although a
corrupt bitmap is only a loss of optimization rather than a corruption
of user-visible data, it is still nice to require the user to opt in
to the fact that they are aware of the loss of the bitmap. Still,
requiring the user to check 'qemu-img info' to see whether bitmaps are
consistent, then use 'qemu-img bitmap --remove' to remove offenders,
all before using 'qemu-img convert', is a lot more work than just
adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which
opts in to skipping the broken bitmaps.
After testing the new option, also demonstrate the way to manually fix
things (either deleting bad bitmaps, or re-creating them as empty) so
that it is possible to convert without the option.
Eric Blake [Fri, 9 Jul 2021 15:39:50 +0000 (10:39 -0500)]
qemu-img: Fail fast on convert --bitmaps with inconsistent bitmap
Waiting until the end of the convert operation (a potentially
time-consuming task) to finally detect that we can't copy a bitmap is
bad, comparing to failing fast up front. Furthermore, this prevents
us from leaving a file behind with a bitmap that is not marked as
inconsistent even though it does not have sane contents.
This fixes the problems exposed in the previous patch to the iotest:
it adds a fast failure up front, and even if we don't fail early, it
ensures that any bitmap we add but do not properly populate is removed
again rather than left behind incomplete.
Eric Blake [Fri, 9 Jul 2021 15:39:49 +0000 (10:39 -0500)]
iotests: Improve and rename test 291 to qemu-img-bitmap
Enhance the test to demonstrate existing less-than-stellar behavior of
qemu-img with a qcow2 image containing an inconsistent bitmap: we
don't diagnose the problem until after copying the entire image (a
potentially long time), and when we do diagnose the failure, we still
end up leaving an empty bitmap in the destination. This mess will be
cleaned up in the next patch.
While at it, rename the test now that we support useful iotest names,
and fix a missing newline in the error message thus exposed.
projects / qemu.git / log