The WHPX build is broken since commit 12e9493df92 which removed the
"hw/boards.h" where MachineState is declared:
$ ./configure \
--enable-hax --enable-whpx
$ make x86_64-softmmu/all
[...]
CC x86_64-softmmu/target/i386/whpx-all.o
target/i386/whpx-all.c: In function 'whpx_accel_init':
target/i386/whpx-all.c:1378:25: error: dereferencing pointer to
incomplete type 'MachineState' {aka 'struct MachineState'}
whpx->mem_quota = ms->ram_size;
^~
make[1]: *** [rules.mak:69: target/i386/whpx-all.o] Error 1
CC x86_64-softmmu/trace/generated-helpers.o
make[1]: Target 'all' not remade because of errors.
make: *** [Makefile:471: x86_64-softmmu/all] Error 2
Restore this header, partially reverting commit 12e9493df92.
Alex Bennée [Thu, 19 Sep 2019 13:07:36 +0000 (14:07 +0100)]
configure: preserve PKG_CONFIG for subdir builds
The slirp sub-module complains about not being able to find the glib
library on cross-compiles because it is using the default pkg-config
tool (which isn't installed in our cross-build docker images).
Preserve PKG_CONFIG in our host config and pass it down to slirp.
Alex Bennée [Wed, 18 Sep 2019 13:31:42 +0000 (14:31 +0100)]
tests/tcg: add simple record/replay smoke test for aarch64
This adds two new tests that re-use the memory test to check basic
record replay functionality is still working. We have to define our
own runners rather than using the default pattern as we want to change
the test name but re-use the memory binary.
We declare the test binaries as PHONY as they don't really exist.
[AJB: A better test would output some sort of timer value or other
otherwise variable value so we could compare the record and replay
outputs and ensure they match]
Alex Bennée [Mon, 16 Sep 2019 13:50:56 +0000 (14:50 +0100)]
tests/tcg: add generic version of float_convs
This is broadly similar to the existing fcvt test for ARM but using
the generic float testing framework. We should be able to pare down
the ARM fcvt test case to purely half-precision with or without the
Alt HP provision.
Alex Bennée [Fri, 13 Sep 2019 11:12:17 +0000 (12:12 +0100)]
tests/tcg: add float_madds test to multiarch
This is a generic floating point multiply and accumulate test for
single precision floating point values. I've split of the common float
functions into a helper library so additional tests can use the same
common code.
As I don't have references for all architectures I've allowed some
flexibility for tests to pass without reference files. They can be
added as we get collect them.
Alex Bennée [Wed, 11 Sep 2019 16:49:55 +0000 (17:49 +0100)]
tests/tcg: clean-up some comments after the de-tangling
These were missed in the recent de-tangling so have been updated to be
more actuate. I've also built up ARM_TESTS in a manner similar to
AARCH64_TESTS for better consistency.
Alex Bennée [Thu, 19 Sep 2019 16:51:27 +0000 (17:51 +0100)]
tests/docker: reduce scary warnings by cleaning up clean up
There was in the clean-up code caused by attempting to inspect images
which finished before we got there. Clean up the clean up code by:
- only track the one instance at a time
- use --filter for docker ps instead of doing it by hand
- just call docker rm -f to be done with it
- use uuid.uuid4() for a random uid
Remove a redundant masking of ignore. Once that's gone it is
obvious that the system-mode inner test is redundant with the
outer test. Move the fpcr_exc_enable masking up and tidy.
target/alpha: Mask IOV exception with INV for user-only
The kernel masks the integer overflow exception with the
software invalid exception mask. Include IOV in the set
of exception bits masked by fpcr_exc_enable.
The CONFIG_USER_ONLY adjustment blindly mashed the swcr
exception enable bits into the fpcr exception disable bits.
However, fpcr_exc_enable has already converted the exception
disable bits into the exception status bits in order to make
it easier to mask status bits at runtime.
Instead, merge the swcr enable bits with the fpcr before we
convert to status bits.
We were setting the wrong bit. The fp_status.flush_to_zero
setting is overwritten by either the constant 1 or the value
of fpcr_flush_to_zero depending on bits within an fp insn.
Peter Maydell [Thu, 26 Sep 2019 13:23:58 +0000 (14:23 +0100)]
Merge remote-tracking branch 'remotes/dgilbert/tags/pull-migration-20190925a' into staging
Migration pull 2019-09-25
me: test fixes from (should stop hangs in postcopy tests).
me: An RDMA cleanup hang fix
Wei: Tidy ups around postcopy
Marc-Andre: mem leak fix
# gpg: Signature made Wed 25 Sep 2019 15:59:41 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-20190925a:
migration/postcopy: Recognise the recovery states as 'in_postcopy'
tests/migration/postcopy: trim migration bandwidth
tests/migration: Fail on unexpected migration states
migration/rdma.c: Swap synchronize_rcu for call_rcu
migration/rdma: Don't moan about disconnects at the end
migration: remove sent parameter in get_queued_page_not_dirty
migration/postcopy: unsentmap is not necessary for postcopy
migration/postcopy: not necessary to do discard when canonicalizing bitmap
migration: fix vmdesc leak on vmstate_save() error
Peter Maydell [Thu, 26 Sep 2019 09:13:39 +0000 (10:13 +0100)]
Merge remote-tracking branch 'remotes/ericb/tags/pull-nbd-2019-09-24-v2' into staging
nbd patches for 2019-09-24
- Improved error message for plaintext client of encrypted server
- Fix various assertions when -object iothread is in use
- Silence a Coverity error for use-after-free on error path
* remotes/ericb/tags/pull-nbd-2019-09-24-v2:
util/qemu-sockets: fix keep_alive handling in inet_connect_saddr
tests: Use iothreads during iotest 223
nbd: Grab aio context lock in more places
nbd/server: attach client channel to the export's AioContext
nbd/client: Add hint when TLS is missing
migration/postcopy: Recognise the recovery states as 'in_postcopy'
Various parts of the migration code do different things when they're
in postcopy mode; prior to this patch this has been 'postcopy-active'.
This patch extends 'in_postcopy' to include 'postcopy-paused' and
'postcopy-recover'.
In particular, when you set the max-postcopy-bandwidth parameter, this
only affects the current migration fd if we're 'in_postcopy';
this leads to a race in the postcopy recovery test where it increases
the speed from 4k/sec to unlimited, but that increase can get ignored
if the change is made between the point at which the reconnection
happens and it transitions back to active.
tests/migration/postcopy: trim migration bandwidth
On slow hosts with tcg we were sometimes finding that the migration
would complete during precopy and never get into the postcopy test.
Trim back the bandwidth a bit to make that much less likely.
tests/migration: Fail on unexpected migration states
We've got various places where we wait for a migration to enter
a given state; but if we enter an unexpected state we tend to fail
in odd ways; add a mechanism for explicitly testing for any state
which we shouldn't be in.
migration/rdma.c: Swap synchronize_rcu for call_rcu
This fixes a deadlock that can occur on the migration source after
a failed RDMA migration; as the source tries to cleanup it
clears a pair of pointers and uses synchronize_rcu to wait; this
is happening on the main thread. With the CPUs running
a CPU thread can be an rcu reader and attempt to grab the main lock
(kvm_handle_io->address_space_write->flatview_write->flatview_write_continue->
prepare_mmio_access->qemu_mutex_lock_iothread_impl)
Replace the synchronize_rcu with a call_rcu to postpone the freeing.
Wei Yang [Mon, 19 Aug 2019 06:18:41 +0000 (14:18 +0800)]
migration/postcopy: not necessary to do discard when canonicalizing bitmap
All pages, either partially sent or partially dirty, will be discarded in
postcopy_send_discard_bm_ram(), since we update the unsentmap to be
unsentmap = unsentmap | dirty in ram_postcopy_send_discard_bitmap().
This is not necessary to do discard when canonicalizing bitmap. And by
doing so, we separate the page discard into two individual steps:
* remotes/armbru/tags/pull-qapi-2019-09-24: (37 commits)
qapi: Assert .visit() and .check_clash() run only after .check()
qapi: Fix excessive QAPISchemaEntity.check() recursion
qapi: Fix to .check() empty structs just once
qapi: Delete useless check_exprs() code for simple union kind
qapi: Clean up around check_known_keys()
qapi: Simplify check_keys()
qapi: Normalize 'if' in check_exprs(), like other sugar
qapi: Fix missing 'if' checks in struct, union, alternate 'data'
qapi: Reject blank 'if' conditions in addition to empty ones
qapi: Fix broken discriminator error messages
qapi: Remove null from schema language
qapi: Improve reporting of lexical errors
qapi: Use quotes more consistently in frontend error messages
tests/qapi-schema: Demonstrate suboptimal lexical errors
tests/qapi-schema: Demonstrate insufficient 'if' checking
tests/qapi-schema: Demonstrate broken discriminator errors
tests/qapi-schema: Demonstrate misleading optional tag error
tests/qapi-schema: Delete two redundant tests
tests/qapi-schema: Cover unknown pragma
qapi: Tweak code to match docs/devel/qapi-code-gen.txt
...
Peter Maydell [Tue, 24 Sep 2019 14:36:31 +0000 (15:36 +0100)]
Merge remote-tracking branch 'remotes/aperard/tags/pull-xen-20190924' into staging
Xen queue
* Update of maintainer email address
* Fixes for xen-bus and xen-block
# gpg: Signature made Tue 24 Sep 2019 12:27:56 BST
# gpg: using RSA key F80C006308E22CFD8A92E7980CF5572FD7FB55AF
# gpg: issuer "[email protected]"
# gpg: Good signature from "Anthony PERARD <[email protected]>" [marginal]
# gpg: aka "Anthony PERARD <[email protected]>" [marginal]
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 5379 2F71 024C 600F 778A 7161 D8D5 7199 DF83 42C8
# Subkey fingerprint: F80C 0063 08E2 2CFD 8A92 E798 0CF5 572F D7FB 55AF
* remotes/aperard/tags/pull-xen-20190924:
xen-bus: only set the xen device frontend state if it is missing
xen-block: treat XenbusStateUnknown the same as XenbusStateClosed
MAINTAINERS: update my email address
xen: perform XenDevice clean-up in XenBus watch handler
xen: introduce separate XenWatchList for XenDevice objects
xen / notify: introduce a new XenWatchList abstraction
xen-bus: check whether the frontend is active during device reset...
* remotes/kraxel/tags/audio-20190924-pull-request:
audio: split ctl_* functions into enable_* and volume_*
audio: common rate control code for timer based outputs
audio: unify input and output mixeng buffer management
audio: remove remains of the old backend api
wavaudio: port to the new audio backend api
spiceaudio: port to the new audio backend api
sdlaudio: port to the new audio backend api
paaudio: port to the new audio backend api
ossaudio: port to the new audio backend api
noaudio: port to the new audio backend api
dsoundaudio: port to the new audio backend api
coreaudio: port to the new audio backend api
alsaaudio: port to the new audio backend api
audio: api for mixeng code free backends
audio: fix ALSA period-length typo in documentation
audio: fix buffer-length typo in documentation
Eric Blake [Tue, 17 Sep 2019 02:39:17 +0000 (21:39 -0500)]
nbd: Grab aio context lock in more places
When iothreads are in use, the failure to grab the aio context results
in an assertion failure when trying to unlock things during blk_unref,
when trying to unlock a mutex that was not locked. In short, all
calls to nbd_export_put need to done while within the correct aio
context. But since nbd_export_put can recursively reach itself via
nbd_export_close, and recursively grabbing the context would deadlock,
we can't do the context grab directly in those functions, but must do
so in their callers.
Hoist the use of the correct aio_context from nbd_export_new() to its
caller qmp_nbd_server_add(). Then tweak qmp_nbd_server_remove(),
nbd_eject_notifier(), and nbd_esport_close_all() to grab the right
context, so that all callers during qemu now own the context before
nbd_export_put() can call blk_unref().
Remaining uses in qemu-nbd don't matter (since that use case does not
support iothreads).
nbd/server: attach client channel to the export's AioContext
On creation, the export's AioContext is set to the same one as the
BlockBackend, while the AioContext in the client QIOChannel is left
untouched.
As a result, when using data-plane, nbd_client_receive_next_request()
schedules coroutines in the IOThread AioContext, while the client's
QIOChannel is serviced from the main_loop, potentially triggering the
assertion at qio_channel_restart_[read|write].
To fix this, as soon we have the export corresponding to the client,
we call qio_channel_attach_aio_context() to attach the QIOChannel
context to the export's AioContext. This matches with the logic at
blk_aio_attached().
Eric Blake [Sat, 7 Sep 2019 17:20:55 +0000 (12:20 -0500)]
nbd/client: Add hint when TLS is missing
I received an off-list report of failure to connect to an NBD server
expecting an x509 certificate, when the client was attempting something
similar to this command line:
$ ./x86_64-softmmu/qemu-system-x86_64 -name 'blah' -machine q35 -nodefaults \
-object tls-creds-x509,id=tls0,endpoint=client,dir=$path_to_certs \
-device virtio-scsi-pci,id=virtio_scsi_pci0,bus=pcie.0,addr=0x6 \
-drive id=drive_image1,if=none,snapshot=off,aio=threads,cache=none,format=raw,file=nbd:localhost:9000,werror=stop,rerror=stop,tls-creds=tls0 \
-device scsi-hd,id=image1,drive=drive_image1,bootindex=0
qemu-system-x86_64: -drive id=drive_image1,if=none,snapshot=off,aio=threads,cache=none,format=raw,file=nbd:localhost:9000,werror=stop,rerror=stop,tls-creds=tls0: TLS negotiation required before option 7 (go)
server reported: Option 0x7 not permitted before TLS
The problem? As specified, -drive is trying to pass tls-creds to the
raw format driver instead of the nbd protocol driver, but before we
get to the point where we can detect that raw doesn't know what to do
with tls-creds, the nbd driver has already failed because the server
complained. The fix to the broken command line? Pass
'...,file.tls-creds=tls0' to ensure the tls-creds option is handed to
nbd, not raw. But since the error message was rather cryptic, I'm
trying to improve the error message.
With this patch, the error message adds a line:
qemu-system-x86_64: -drive id=drive_image1,if=none,snapshot=off,aio=threads,cache=none,format=raw,file=nbd:localhost:9000,werror=stop,rerror=stop,tls-creds=tls0: TLS negotiation required before option 7 (go)
Did you forget a valid tls-creds?
server reported: Option 0x7 not permitted before TLS
And with luck, someone grepping for that error message will find this
commit message and figure out their command line mistake. Sadly, the
only mention of file.tls-creds in our docs relates to an --image-opts
use of PSK encryption with qemu-img as the client, rather than x509
certificate encryption with qemu-kvm as the client.
Entity checking goes back to commit ac88219a6c "qapi: New QAPISchema
intermediate representation", v2.5.0. It's designed to work as
follows: QAPISchema.check() calls .check() for all the schema's
entities. An entity's .check() recurses into another entity's
.check() only if the C struct generated for the former contains the C
struct generated for the latter (pointers don't count). This is used
to detect "object contains itself".
There are two instances of this:
* An object's C struct contains its base's C struct
* An object's C struct contains its variants' C structs
QAPISchemaObjectTypeVariants().check calls v.type.check(). Since
commit b807a1e1e3 "qapi: Check for QAPI collisions involving variant
members", v2.6.0.
Thus, only object types can participate in recursion.
QAPISchemaObjectType.check() is made for that: it checks @self when
called the first time, recursing into base and variants, it reports an
"contains itself" error when this recursion reaches an object being
checked, and does nothing it reaches an object that has been checked
already.
The other .check() may safely assume they get called exactly once.
Sadly, this design has since eroded:
* QAPISchemaCommand.check() and QAPISchemaEvent.check() call
.args_type.check(). Since commit c818408e44 "qapi: Implement boxed
types for commands/events", v2.7.0. Harmless, since args_type can
only be an object type.
* QAPISchemaEntity.check() calls ._ifcond.check() when inheriting the
condition from another type. Since commit 4fca21c1b0 qapi: leave
the ifcond attribute undefined until check(), v3.0.0. This makes
simple union wrapper types recurse into the wrapped type (nothing
else uses this condition inheritance). The .check() of types used
as simple union branch type get called multiple times.
* QAPISchemaObjectType.check() calls its super type's .check()
*before* the conditional handling multiple calls. Also since commit 4fca21c1b0. QAPISchemaObjectType.check()'s guard against multiple
checking doesn't protect QAPISchemaEntity.check().
* QAPISchemaArrayType.check() calls .element_type.check(). Also since
commit 4fca21c1b0. The .check() of types used as array element
types get called multiple times.
Commit 56a4689582 "qapi: Fix array first used in a different module"
(v4.0.0) added more code relying on this .element_type.check().
The absence of explosions suggests the .check() involved happen to be
effectively idempotent.
Fix the unwanted recursion anyway:
* QAPISchemaCommand.check() and QAPISchemaEvent.check() calling
.args_type.check() is unnecessary. Delete the calls.
* Fix QAPISchemaObjectType.check() to call its super type's .check()
after the conditional handling multiple calls.
* A QAPISchemaEntity's .ifcond becomes valid at .check(). This is due
to arrays and simple unions.
Most types get ifcond and info passed to their constructor.
Array types don't: they get it from their element type, which
becomes known only in .element_type.check().
The implicit wrapper object types for simple union branches don't:
they get it from the wrapped type, which might be an array.
Ditch the idea to set .ifcond in .check(). Instead, turn it into a
property and compute it on demand. Safe because it's only used
after the schema has been checked.
Most types simply return the ifcond passed to their constructor.
Array types forward to their .element_type instead, and the wrapper
types forward to the wrapped type.
* A QAPISchemaEntity's .module becomes valid at .check(). This is
because computing it needs info and schema.fname, and because array
types get it from their element type instead.
Make it a property just like .ifcond.
Additionally, have QAPISchemaEntity.check() assert it gets called at
most once, so the design invariant will stick this time. Neglecting
that was entirely my fault.
QAPISchemaObjectType.check() does nothing for types that have been
checked already. Except the "has been checked" predicate is broken
for empty types: self.members is [] then, which isn't true. The bug
is harmless, but fix it anyway: use self.member is not None instead.
qapi: Delete useless check_exprs() code for simple union kind
Commit bceae7697f "qapi script: support enum type as discriminator in
union" made check_exprs() add the implicit enum types of simple unions
to global @enum_types. I'm not sure it was needed even then. It's
certainly not needed now. Delete it.
discriminator_find_enum_define() and add_name() parameter @implicit
are now dead. Bury them.
qapi: Normalize 'if' in check_exprs(), like other sugar
We normalize shorthand to longhand forms in check_expr(): enumeration
values with normalize_enum(), feature values with
normalize_features(), struct members, union branches and alternate
branches with normalize_members(). If conditions are an exception: we
normalize them in QAPISchemaEntity.check() and
QAPISchemaMember.__init(), with listify_cond(). The idea goes back to
commit 2cbc94376e "qapi: pass 'if' condition into QAPISchemaEntity
objects", v3.0.0.
Normalize in check_expr() instead, with new helper normalize_if().
qapi: Fix missing 'if' checks in struct, union, alternate 'data'
Commit 87adbbffd4..3e270dcacc "qapi: Add 'if' to (implicit
struct|union|alternate) members" (v4.0.0) neglected test coverage, and
promptly failed to check the conditions. Review fail.
Recent commit "tests/qapi-schema: Demonstrate insufficient 'if'
checking" added test coverage, demonstrating the bug. Fix it by add
the missing check_if().
check_union() checks the discriminator exists in base and makes sense.
Two error messages mention the base. These are broken for anonymous
bases, as demonstrated by tests flat-union-invalid-discriminator and
flat-union-invalid-if-discriminator.err. The third one doesn't
bother.
First broken when commit ac4338f8eb "qapi: Allow anonymous base for
flat union" (v2.6.0) neglected to adjust the "not a member of base"
error message. Commit ccadd6bcba "qapi: Add 'if' to implicit struct
members" (v4.0.0) then cloned the flawed error message.
We represent the parse tree as OrderedDict. We fetch optional dict
members with .get(). So far, so good.
We represent null literals as None. .get() returns None both for
"absent" and for "present, value is the null literal". Uh-oh.
Test features-if-invalid exposes this bug: "'if': null" is
misinterpreted as absent "if".
We added null to the schema language to "allow [...] an explicit
default value" (commit e53188ada5 "qapi: Allow true, false and null in
schema json", v2.4.0). Hasn't happened; null is still unused except
as generic invalid value in tests/.
To fix, we'd have to replace .get() by something more careful, or
represent null differently. Feasible, but we got more and bigger fish
to fry right now. Remove the null literal from the schema language.
Replace null in tests by another invalid value.
Test features-if-invalid now behaves as it should.
The error message for forgotten quotes around a name shows just the
name's first character, which isn't as nice as it could be. Same for
attempting to use a number.
tests/qapi-schema: Demonstrate misleading optional tag error
Test flat-union-optional-discriminator declares its union tag as
'*switch': 'Enum', and points to it with 'discriminator': '*switch'.
This gets rejected as "discriminator of flat union 'MyUnion' uses
invalid name '*switch'". Correct; member 'discriminator' doesn't
accept a '*' prefix.
However, this merely tests name validity checking, which we already
cover elsewhere. More interesting is testing the valid name 'switch'.
This reports "discriminator 'switch' is not a member of base struct
'Base'", which is misleading.
Copy the existing 'discriminator': '*switch' test to
flat-union-discriminator-bad-name, and rewrite its comment. Change
flat-union-optional-discriminator to test 'discriminator': 'switch',
and mark it FIXME.
Tests duplicate-key and double-data test the same thing. The former
predates the latter, and it has a better name. Delete the latter, and
tweak the former's comment.
Tests include-format-err and include-extra-junk test the same thing.
The former predates the latter, but the latter has a better name and a
comment. Delete the former.
qapi: Tweak code to match docs/devel/qapi-code-gen.txt
The previous commit made qapi-code-gen.txt define "(top-level)
expression" as either "directive" or "definition". The code still
uses "expression" when it really means "definition". Tidy up.
The previous commit made qapi-code-gen.txt use "object" rather than
"dictionary". The code still uses "dictionary". Tidy up.
docs/devel/qapi-code-gen: Improve QAPI schema language doc
We document the language by giving patterns of valid JSON objects.
The patterns contain placeholders we don't define anywhere; their
names have to speak for themselves. I guess they do, but I'd prefer a
bit more rigor. Provide a grammar instead, and rework the text
accordingly.
Documentation for QAPI schema conditionals (commit 967c885108, 6cc32b0e14, 87adbbffd4..3e270dcacc) and feature flags (commit 6a8c0b5102) was bolted on. The sections documenting types, commands
and events don't mention them. Section "Features" and "Configuring
the schema" then provide additional syntax for types, commands and
events. I hate that. Fix the sections documenting types, commands
and events to provide accurate syntax, and point to "Features" and
"Configuring the schema" for details.
We talk about "(top-level) expressions other than include and pragma".
Adopt more convenient terminology: a (top-level) expression is either
a directive (include or pragma) or a definition (anything else).
Avoid the terms "dictionary" and "key". Stick to JSON terminology
"object" and "member name" instead.
We have some compatibility advice buried in sections "Enumeration
types" and "Struct types". Compatibility is actually about commands
and events. It devolves to the types used there. All kinds of types,
not just enumerations and structs.
Replace the existing advice by a new section "Compatibility
considerations".
docs/devel/qapi-code-gen: Reorder sections for readability
Section "QMP/Guest agent schema" starts with a brief introduction,
then subsection "Comments", then subsection "Schema overview" (more
elaborate introduction), and only then talks about schema entities
like types, commands, and so forth.
Subsection "Comments" is long and tiring: almost 500 words, mostly
about doc comments. Move the doc comment part to its own subsection
"Documentation comments" at the very end of "QMP/Guest agent schema".
Subsection "Schema overview" explains naming rules at considerable
length: 250 words. Move this part to its own subsection "Naming rules
and reserved names" right after the subsections on schema entities.
Subsection "Enumeration types" is wedged between "Struct types" and
"Union types". Move it before "Struct types".
Absent flat union branches default to the empty struct (since commit 800877bb16 "qapi: allow empty branches in flat unions"). But an
attempt to omit all of them is rejected with "Union 'FOO' has no
branches". Harmless oddity, but it's easy to avoid, so do that.
A union or alternate without branches makes no sense and doesn't work:
it can't be instantiated. A union or alternate with just one branch
works, but is degenerate. We accept the former, but reject the
latter. Weird. docs/devel/qapi-code-gen.txt doesn't mention the
difference. It claims an alternate definition is "is similar to a
simple union type".
Permit degenerate alternates to make them consistent with unions.
qapi: Drop support for escape sequences other than \\
Since the previous commit restricted strings to printable ASCII,
\uXXXX's only use is obfuscation. Drop it.
This leaves \\, \/, \', and \". Since QAPI schema strings are all
names, and names are restricted to ASCII letters, digits, hyphen, and
underscore, none of them is useful.
The latter three have no test coverage. Drop them.
Keep \\ to avoid (more) gratuitous incompatibility with JSON.
All Unicode characters may be placed within the quotation marks,
except for the characters that MUST be escaped: quotation mark,
reverse solidus, and the control characters (U+0000 through
U+001F).
The QAPI schema parser accepts both less and more than JSON: it
accepts only ASCII with \u (less), and accepts control characters
other than LF (new line) unescaped. How it treats unescaped non-ASCII
input differs between Python 2 and Python 3.
Make it accept strictly less: require printable ASCII. Drop support
for \b, \f, \n, \r, \t.
docs/devel/qapi-code-gen: Minor specification fixes
The specification claims "Each expression that isn't an include
directive may be preceded by a documentation block", but the code also
rejects them for pragma directives. The code is correct. Fix the
specification.
The specification reserves member names starting with 'has_', but the
code also reserves name 'u'. Fix the specification.
The specification claims "The string 'max' is not allowed as an enum
value". Untrue. Fix the specification. While there, delete the
naming advice, because it's redundant with the naming rules in section
"Schema overview"
The specification claims "No branch of the union can be named 'max',
as this would collide with the implicit enum". Untrue. Fix the
specification.
The specification claims "It is not allowed to name an event 'MAX',
since the generator also produces a C enumeration of all event names
with a generated _MAX value at the end." Untrue. Fix the
specification.
The specification claims "All branches of the union must be complex
types", but the code permits only struct types. The code is correct.
Fix the specification.
The specification claims a command's return type "must be the string
name of a complex or built-in type, a one-element array containing the
name of a complex or built-in type" unless the command is in pragma
'returns-whitelist'. The code does not permit built-in types. Fix
the specification.
Commands and events can define their argument type inline (default) or
by referring to another type ('boxed': true, since commit c818408e44
"qapi: Implement boxed types for commands/events", v2.7.0). The
unboxed inline definition is an (anonymous) struct type. The boxed
type may be a struct, union, or alternate type.
The latter is problematic: docs/interop/qemu-spec.txt requires the
value of the 'data' key to be a json-object, but any non-degenerate
alternate type has at least one branch that isn't.
Fortunately, we haven't made use of alternates in this context outside
tests/. Drop support for them.
QAPISchemaAlternateType.is_empty() is now unused. Drop it, too.
qapi: Drop check_type()'s redundant parameter @allow_optional
check_type() uses @allow_optional only when @value is a dictionary and
@allow_dict is True. All callers that pass allow_dict=True also pass
allow_optional=True.
Therefore, @allow_optional is always True when check_type() uses it.
Drop the redundant parameter.
scripts/git.orderfile: Match QAPI schema more precisely
Pattern *.json also matches the tests/qapi-schema/*.json. Separates
them from the tests/qapi-schema/*.{err,exit,out} in diffs. I hate
that. Change the pattern to match just the "real" QAPI schemata.
Mark Syms [Wed, 18 Sep 2019 11:57:44 +0000 (12:57 +0100)]
xen-bus: only set the xen device frontend state if it is missing
Some toolstack implementations will set the frontend xenstore
keys to Initialising which will then trigger the in guest PV
drivers to begin initialising and some implementations will
then set their state to Closing. If this has occurred then
device realize must not overwrite the frontend keys as then
the handshake will stall.
Signed-off-by: Mark Syms <[email protected]>
Also avoid creating the frontend area if it already exists.
Paul Durrant [Wed, 18 Sep 2019 11:57:02 +0000 (12:57 +0100)]
xen-block: treat XenbusStateUnknown the same as XenbusStateClosed
When a frontend gracefully disconnects from an offline backend, it will
set its own state to XenbusStateClosed. The code in xen-block.c correctly
deals with this and sets the backend into XenbusStateClosed. Unfortunately
it is possible for toolstack to actually delete the frontend area
before the state key has been read, leading to an apparent frontend state
of XenbusStateUnknown. This prevents the backend state from transitioning
to XenbusStateClosed and hence leaves it limbo.
This patch simply treats a frontend state of XenbusStateUnknown the same
as XenbusStateClosed, which will unblock the backend in these circumstances.
Paul Durrant [Fri, 13 Sep 2019 08:21:58 +0000 (09:21 +0100)]
xen: perform XenDevice clean-up in XenBus watch handler
Cleaning up offline XenDevice objects directly in
xen_device_backend_changed() is dangerous as xen_device_unrealize() will
modify the watch list that is being walked. Even the QLIST_FOREACH_SAFE()
used in notifier_list_notify() is insufficient as *two* notifiers (for
the frontend and backend watches) are removed, thus potentially rendering
the 'next' pointer unsafe.
The solution is to use the XenBus backend_watch handler to do the clean-up
instead, as it is invoked whilst walking a separate watch list.
This patch therefore adds a new 'inactive_devices' list to XenBus, to which
offline devices are added by xen_device_backend_changed(). The XenBus
backend_watch registration is also changed to not only invoke
xen_bus_enumerate() but also a new xen_bus_cleanup() function, which will
walk 'inactive_devices' and perform the necessary actions.
For safety an extra 'online' check is also added to xen_bus_type_enumerate()
to make sure that no attempt is made to create a new XenDevice object for a
backend that is offline.
NOTE: This patch also includes some cosmetic changes:
- substitute the local variable name 'backend_state'
in xen_bus_type_enumerate() with 'state', since there
is no ambiguity with any other state in that context.
- change xen_device_state_is_active() to
xen_device_frontend_is_active() (and pass a XenDevice directly)
since the state tests contained therein only apply to a frontend.
- use 'state' rather then 'xendev->backend_state' in
xen_device_backend_changed() to shorten the code.
Paul Durrant [Fri, 13 Sep 2019 08:21:57 +0000 (09:21 +0100)]
xen: introduce separate XenWatchList for XenDevice objects
This patch uses the XenWatchList abstraction to add a separate watch list
for each device. This is more scalable than walking a single notifier
list for all watches and is also necessary to implement a bug-fix in a
subsequent patch.
Paul Durrant [Fri, 13 Sep 2019 08:21:56 +0000 (09:21 +0100)]
xen / notify: introduce a new XenWatchList abstraction
Xenstore watch call-backs are already abstracted away from XenBus using
the XenWatch data structure but the associated NotifierList manipulation
and file handle registration is still open coded in various xen_bus_...()
functions.
This patch creates a new XenWatchList data structure to allow these
interactions to be abstracted away from XenBus as well. This is in
preparation for a subsequent patch which will introduce separate watch lists
for XenBus and XenDevice objects.
NOTE: This patch also introduces a new notifier_list_empty() helper function
for the purposes of adding an assertion that a XenWatchList is not
freed whilst its associated NotifierList is still occupied.
/*
* If the toolstack (or unplug request callback) has set the backend
- * state to Closing, but there is no active frontend (i.e. the
- * state is not Connected) then set the backend state to Closed.
+ * state to Closing, but there is no active frontend then set the
+ * backend state to Closed.
*/
if (xendev->backend_state == XenbusStateClosing &&
- xendev->frontend_state != XenbusStateConnected) {
+ !xen_device_state_is_active(state)) {
xen_device_backend_set_state(xendev, XenbusStateClosed);
}
mistakenly replaced the check of 'xendev->frontend_state' with a check
(now in a helper function) of 'state', which actually equates to
'xendev->backend_state'.
make check-unit: use after free in test-opts-visitor
In the struct OptsVisitor, the 'repeated_opts' member points to a list
in the 'unprocessed_opts' hash table after the list has been destroyed.
A subsequent call to visit_type_int() references the deleted list.
It results in use-after-free issue reproduced by running the test case
under the Valgrind: valgrind tests/test-opts-visitor.
A new mode ListMode::LM_TRAVERSED is declared to mark the list
traversal completed.
qapi: Make visit_next_list()'s comment less confusing
visit_next_list() returns non-null on success, null on failure. The
comment's phrasing "until NULL return or error occurs" is needlessly
confusing. Scratch the "or error occurs" part.
Peter Maydell [Mon, 23 Sep 2019 22:04:57 +0000 (23:04 +0100)]
Merge remote-tracking branch 'remotes/cleber/tags/python-next-pull-request' into staging
Improvements include:
* Relaxed error code pattern on machine_m68k_nextcube.py
* Better naming and separation of tests in x86_cpu_model_versions.py
* New checks on guest side on linux_ssh_mips_malta.py
* Use of avocado.utils.ssh, and thus "/usr/bin/ssh" for SSH
interaction
# gpg: Signature made Mon 23 Sep 2019 14:38:01 BST
# gpg: using RSA key 7ABB96EB8B46B94D5E0FE9BB657E8D33A5F209F3
# gpg: Good signature from "Cleber Rosa <[email protected]>" [marginal]
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 7ABB 96EB 8B46 B94D 5E0F E9BB 657E 8D33 A5F2 09F3
* remotes/cleber/tags/python-next-pull-request:
Acceptance tests: use avocado.utils.ssh for SSH interaction
tests/acceptance: Add new test cases in linux_ssh_mips_malta.py
tests/acceptance: Refactor and improve reporting in linux_ssh_mips_malta.py
Acceptance test x86_cpu_model_versions: split into smaller tests
Acceptance test x86_cpu_model_versions: fix mismatches between test and messages
Acceptance test x86_cpu_model_versions: shutdown VMs
Acceptance test machine_m68k_nextcube.py: relax the error code pattern
* remotes/borntraeger/tags/s390x-20190923:
s390x/cpumodel: Add the z15 name to the description of gen15a
s390x/kvm: Officially require at least kernel 3.15
pc-bios/s390-ccw: Rebuild the s390-netboot.img firmware image
pc-bios/s390-ccw/net: fix a possible memory leak in get_uuid()
pc-bios/s390-ccw: Do not pre-initialize empty array
Kővágó, Zoltán [Thu, 19 Sep 2019 21:24:22 +0000 (23:24 +0200)]
audio: split ctl_* functions into enable_* and volume_*
This way we no longer need vararg functions, improving compile time
error detection. Also now it's possible to check actually what commands
are supported, without needing to manually update ctl_caps.
Kővágó, Zoltán [Thu, 19 Sep 2019 21:24:21 +0000 (23:24 +0200)]
audio: common rate control code for timer based outputs
This commit removes the ad-hoc rate-limiting code from noaudio and
wavaudio, and replaces them with a (slightly modified) code from
spiceaudio. This way multiple write calls (for example when the
circular buffer wraps around) do not cause problems.
Kővágó, Zoltán [Thu, 19 Sep 2019 21:24:20 +0000 (23:24 +0200)]
audio: unify input and output mixeng buffer management
Usage notes: hw->samples became hw->{mix,conv}_buf->size, except before
initialization (audio_pcm_hw_alloc_resources_*), hw->samples gives the
initial size of the STSampleBuffer. The next commit tries to fix this
inconsistency.
projects / qemu.git / log