Alex Bennée [Tue, 11 Oct 2016 16:16:24 +0000 (17:16 +0100)]
tests/docker/Makefile.include: add a generic docker-run target
This re-factors the docker makefile to include a docker-run target which
can be controlled entirely from environment variables specified on the
make command line. This allows us to run against any given docker image
we may have in our repository, for example:
make docker-run TEST="test-quick" IMAGE="debian:arm64" \
EXECUTABLE=./aarch64-linux-user/qemu-aarch64
The existing docker-foo@bar targets still work but the inline
verification has been dropped because we already don't hit that due to
other pattern rules in rules.mak.
Alex Bennée [Tue, 11 Oct 2016 16:16:21 +0000 (17:16 +0100)]
tests/docker: add travis dockerfile
This target grabs the latest Travis containers from their repository at
quay.io and then installs QEMU's build dependencies. With this it is
possible to run on broadly the same setup as they have on travis-ci.org.
* remotes/kraxel/tags/pull-ui-20161013-1:
input-linux: initialize key state
ui: rename vnc_init_state to vnc_start_protocol
ui: move some initialization out of vnc_init_state
ui: remove bogus call to reset_keys() in vnc_init_state
ui: remove bogus call to graphic_hw_update() in vnc_listen_io
ui: refactor method for setting up VncDisplay auth types
ui: rename misleading 'VncDisplay' variables
ui: remove 'ws_tls' field from VncState
ui: remove 'enabled' and 'ws_enabled' fields from VncState
ui: remove misleading comment from vnc_init_state
Revert "char: use a fixed idx for child muxed chr"
That commit mis-used mux char: the frontend are multiplexed, not the
backend. Fix the regression preventing "c-a c" to switch the focus. The
following patches will fix the crash (when leaving or removing frontend)
by tracking frontends with handler tags.
* remotes/dgibson/tags/ppc-for-2.7-20161013:
ppc: Check the availability of transactional memory
hw/ppc/spapr: Fix the selection of the processor features
hw/ppc/spapr: Move code related to "ibm,pa-features" to a separate function
linux-headers: update
Gerd Hoffmann [Wed, 12 Oct 2016 13:03:04 +0000 (15:03 +0200)]
input-linux: initialize key state
Query input device keys, initialize state accordingly, so the correct
state is reflected in case any key is pressed at initialization time.
There is a high chance for this to actually happen for the 'enter' key
in case you start qemu with a terminal command (directly or virsh).
When finding any pressed keys the input grab is delayed until all keys
are lifted, to avoid confusing guest and host with appearently stuck
keys.
ui: move some initialization out of vnc_init_state
Most of the fields in VncState are initialized in the
vnc_connect() method, but some are done in vnc_init_state()
instead.
The purpose of having vnc_init_state() is to delay starting
of the VNC wire protocol until after the websockets handshake
has completed. As such the vnc_init_state() method only needs
to be used for initialization that is dependant on the wire
protocol running.
This also lets us get rid of the initialized boolean flag
from the VncState struct.
ui: remove bogus call to graphic_hw_update() in vnc_listen_io
Just before accepting a new client connection the vnc_listen_io
method calls graphic_hw_update(). This is bogus because there
is a call to this method already in vnc_state_init() and the
client doesn't need up2date graphics console before reaching
that.
ui: refactor method for setting up VncDisplay auth types
There is a lot of repeated code in the auth type setup method,
particularly around checking TLS credential types. Refactor
it to reduce duplication and instead of having one method
do both plain and websockets at once, call it separately
for each.
The 'ws_tls' field in VncState is only ever representing
the result of 'tlscreds != NULL' and is thus pointless.
Replace use of 'ws_tls' with a direct check against
'tlscreds'
Thomas Huth [Wed, 28 Sep 2016 11:16:30 +0000 (13:16 +0200)]
ppc: Check the availability of transactional memory
KVM-PR currently does not support transactional memory, and the
implementation in TCG is just a fake. We should not announce TM
support in the ibm,pa-features property when running on such a
system, so disable it by default and only enable it if the KVM
implementation supports it (i.e. recent versions of KVM-HV).
These changes are based on some earlier work from Anton Blanchard
(thanks!).
Thomas Huth [Wed, 28 Sep 2016 11:16:29 +0000 (13:16 +0200)]
hw/ppc/spapr: Fix the selection of the processor features
The current code uses pa_features_206 for POWERPC_MMU_2_06, and
for everything else, it uses pa_features_207. This is bad in some
cases because there is also a "degraded" MMU version of ISA 2.06,
called POWERPC_MMU_2_06a, which should of course use the flags for
2.06 instead. And there is also the possibility that the user runs
the pseries machine with a POWER5+ or even 970 processor. In that
case we certainly do not want to set the flags for 2.07, and rather
simply skip the setting of the pa-features property instead.
Thomas Huth [Wed, 28 Sep 2016 11:16:28 +0000 (13:16 +0200)]
hw/ppc/spapr: Move code related to "ibm,pa-features" to a separate function
The function spapr_populate_cpu_dt() has become quite big
already, and since we likely have to extend the pa-features
property for every new processor generation, it is nicer
if we put the related code into a separate function.
* remotes/kraxel/tags/pull-usb-20161012-1:
usb-redir: allocate buffers before waking up the host adapter
usb: Fix incorrect default DMA offset.
usb: fix serial generator
xhci: make xhci_epid_to_usbep accept XHCIEPContext
xhci: drop XHCITransfer->{slotid,epid}
xhci: add & use xhci_kick_epctx()
xhci: drop XHCITransfer->xhci
xhci: use linked list for transfers
xhci: drop unused comp_xfer field
xhci: decouple EV_QUEUE from TD_QUEUE
xhci: limit the number of link trbs we are willing to process
Hans de Goede [Mon, 10 Oct 2016 10:45:13 +0000 (12:45 +0200)]
usb-redir: allocate buffers before waking up the host adapter
Needed to make sure usb redirection is prepared to actually handle the
callback from the usb host adapter. Without this interrupt endpoints
don't work on xhci.
Note: On ehci the usb_wakeup() call only schedules a BH for the actual
work, which hides this bug because the allocation happens before ehci
calls back even without this patch.
The default DMA offset is set to 3. When the property is not set by
the consumer, the default causes DMA access to be shifted by 3
bytes. In PXA, this results in incorrect DMA access, leading to error
notification in the USB controller driver. A better default would be
0, so that there is no offset, when the consumer does not specify one.
Gerd Hoffmann [Wed, 5 Oct 2016 09:33:18 +0000 (11:33 +0200)]
usb: fix serial generator
snprintf return value is *not* the number of chars written into the
buffer, but the number of chars needed. So in case the buffer is too
small you can go alloc a bigger one and try again. But that also means
you can't simply use the return value for the next snprintf call
without checking beforehand that things did actually fit.
Problem is that usb_desc_create_serial didn't perform that check, so a
loooong path string (can happen with deep pci-bridge nesting) results in
the third snprintf call smashing the stack.
Fix this by throwing out all the snpintf calls and use g_strdup_printf
instead.
xhci_kick_epctx is a xhci_kick_ep variant which takes an XHCIEPContext
as input instead of slotid and epid. So in case we have a XHCIEPContext
at hand at the callsite we can just pass it directly.
xhci has a fixed number of 24 (TD_QUEUE) XHCITransfer structs per
endpoint, which turns out to be a problem for usb3 devices with 32 (or
more) bulk streams. xhci re-checks the trb rings on every finished
transfer to make sure it'll pick up any pending work. But that scheme
breaks in case the first transfer of a ring can't be started because we
ran out of XHCITransfer structs already.
So remove static XHCITransfer array from XHCIEPContext. Use a linked
list instead, and allocate/free XHCITransfer as needed. Add helper
functions to allocate & initialize and to cleanup & release
XHCITransfer structs. That also simplifies trb management, we never
have to realloc XHCITransfer->trbs because we don't reuse XHCITransfer
structs any more.
New dynamic limit for in-flight xhci transfers per endpoint is
number-of-streams + 16.
EV_QUEUE must not change because an array of that size is part of live
migration data. Hard-code current value there, so we can touch TD_QUEUE
without breaking live migration.
Peter Maydell [Wed, 12 Oct 2016 10:05:21 +0000 (11:05 +0100)]
Merge remote-tracking branch 'remotes/stefanha/tags/tracing-pull-request' into staging
# gpg: Signature made Wed 12 Oct 2016 09:43:03 BST
# gpg: using RSA key 0x9CA4ABB381AB73C8
# gpg: Good signature from "Stefan Hajnoczi <[email protected]>"
# gpg: aka "Stefan Hajnoczi <[email protected]>"
# Primary key fingerprint: 8695 A8BF D3F9 7CDA AC35 775A 9CA4 ABB3 81AB 73C8
* remotes/stefanha/tags/tracing-pull-request:
trace: Add missing execution mode of guest events
trace: introduce a formal group name for trace events
trace: pass trace-events to tracetool as a positional param
trace: push reading of events up a level to tracetool main
trace: rename _read_events to read_events
trace: get rid of generated-events.h/generated-events.c
trace: dynamically allocate event IDs at runtime
trace: dynamically allocate trace_dstate in CPUState
trace: provide mechanism for registering trace events
trace: don't abort qemu if ftrace can't be initialized
trace: emit name <-> ID mapping in simpletrace header
trace: remove the TraceEventID and TraceEventVCPUID enums
trace: give each trace event a named TraceEvent struct
trace: break circular dependency in event-internal.h
trace: remove duplicate control.h includes in generated-tracers.h
trace: remove global 'uint16 dstate[]' array
trace: remove some now unused functions
trace: convert code to use event iterators
trace: add trace event iterator APIs
trace: move colo trace events to net/ sub-directory
trace: introduce a formal group name for trace events
The declarations in the generated-tracers.h file are
assuming there's only ever going to be one instance
of this header, as they are not namespaced. When we
have one header per event group, if a single source
file needs to include multiple sets of trace events,
the symbols will all clash.
This change thus introduces a '--group NAME' arg to the
'tracetool' program. This will cause all the symbols in
the generated header files to be given a unique namespace.
If no group is given, the group name 'common' is used,
which is suitable for the current usage where there is
only one global trace-events file used for code generation.
trace: pass trace-events to tracetool as a positional param
Instead of reading the contents of 'trace-events' from stdin,
accept the filename as a positional parameter. This also
allows for reading from multiple files, though this facility
is not used at this time.
trace: push reading of events up a level to tracetool main
Move the reading of events out of the 'tracetool.generate'
method and into tracetool.main, so that the latter is not
tied to generating from a single source of events.
trace: get rid of generated-events.h/generated-events.c
Currently the generated-events.[ch] files contain the
event dstates, constants and TraceEvent structs, while the
generated-tracers.[ch] files contain the actual trace
probe logic. With the removal of usage of the event enums
from the API there is no longer any compelling reason for
the separation between these files. The generated-events.h
content is only ever needed from the generated-tracers.[ch]
files.
The enums/constants/structs from generated-events.[ch] are
thus moved into the generated-tracers.[ch], so that there
is one less file to be generated.
Instead of having the code generator assign event IDs and
event VCPU IDs, assign them when the events are registered
at runtime. This will allow code to be generated from
individual trace-events without having to figure out
globally unique numbering at build time.
trace: dynamically allocate trace_dstate in CPUState
The CPUState struct has a bitmap tracking which VCPU
events are currently active. This is indexed based on
the event ID values, and sized according the maximum
TraceEventVCPUID enum value.
When we start dynamically assigning IDs at runtime,
we can't statically declare a bitmap without making
an assumption about the max event count. This problem
can be solved by dynamically allocating the per-CPU
dstate bitmap.
trace: provide mechanism for registering trace events
Remove the notion of there being a single global array
of trace events, by introducing a method for registering
groups of events.
The module_call_init() needs to be invoked at the start
of any program that wants to make use of the trace
support. Currently this covers system emulators qemu-nbd,
qemu-img and qemu-io.
[Squashed the following fix from Daniel P. Berrange
<[email protected]>:
The bsd-user/linux-user programs make use of the CPU emulation
code and this now requires that the trace events subsystem
is enabled, otherwise it'll crash trying to allocate an empty
trace events bitmap for the CPU object.
trace: don't abort qemu if ftrace can't be initialized
If the ftrace backend is compiled into QEMU, any attempt
to start QEMU while non-root will fail due to the
inability to open /sys/kernel/debug/tracing/tracing_on.
Add a fallback into the code so that it connects up the
trace_marker_fd variable to /dev/null when getting
EACCES on the 'trace_on' file. This allows QEMU to
run, with ftrace turned into a no-op.
[Fixed s/setting/getting/ and s/EACCESS/EACCES/ errors pointed out by
Eric Blake <[email protected]>.
--Stefan]
trace: emit name <-> ID mapping in simpletrace header
Currently simpletrace assumes that events are given IDs
starting from 0, based on the order in which they appear
in the trace-events file, with no gaps. When the
trace-events file is split up, this assumption becomes
problematic.
To deal with this, extend the simpletrace format so that
it outputs a table of event name <-> ID mappings. That
will allow QEMU to assign arbitrary IDs to events without
breaking simpletrace parsing.
The v3 simple trace format was
FILE HEADER
EVENT TRACE RECORD 0
EVENT TRACE RECORD 1
...
EVENT TRACE RECORD N
The v4 simple trace format is now
FILE HEADER
EVENT MAPPING RECORD 0
EVENT MAPPING RECORD 1
...
EVENT MAPPING RECORD M
EVENT TRACE RECORD RECORD 0
EVENT TRACE RECORD RECORD 1
...
EVENT TRACE RECORD N
Although this shows all the mapping records being emitted
upfront, this is not required by the format. While the main
simpletrace backend will emit all mappings at startup,
the systemtap simpletrace.stp script will emit the mappings
at first use. eg
FILE HEADER
...
EVENT MAPPING RECORD 0
EVENT TRACE RECORD RECORD 0
EVENT TRACE RECORD RECORD 1
EVENT MAPPING RECORD 1
EVENT TRACE RECORD RECORD 2
...
EVENT TRACE RECORD N
This is more space efficient given that most trace records
only include a subset of events.
In modifying the systemtap simpletrace code, a 'begin' probe
was added to emit the trace event header, so you no longer
need to add '--no-header' when running simpletrace.py for
systemtap generated trace files.
trace: remove the TraceEventID and TraceEventVCPUID enums
The TraceEventID and TraceEventVCPUID enums constants are
no longer actually used for anything critical.
The TRACE_EVENT_COUNT limit is used to determine the size
of the TraceEvents array, and can be removed if we just
NULL terminate the array instead.
The TRACE_VCPU_EVENT_COUNT limit is used as a magic value
for marking non-vCPU events, and also for declaring the
size of the trace dstate mask in the CPUState struct.
The former usage can be replaced by a dedicated constant
TRACE_EVENT_VCPU_NONE, defined as (uint32_t)-1. For the
latter usage, we can simply define a constant for the
number of VCPUs, avoiding the need for the full enum.
The only other usages of the enum values can be replaced
by accesing the id/vcpu_id fields via the named TraceEvent
structs.
trace: give each trace event a named TraceEvent struct
Currently we only expose a TraceEvent array, which must
be indexed via the TraceEventID enum constants. This
changes the generator to expose a named TraceEvent
instance for each event, with an _EVENT suffix.
trace: break circular dependency in event-internal.h
Currently event-internal.h includes generated-events.h,
while generated-events.h includes event-internal.h
causing a circular dependency.
event-internal.h requires that the content of
generated-events.h comes first, so that it can see
the typedefs for TraceEventID and TraceEventVCPUID.
Switching the TraceEvent struct to use uint32_t
for the two ID fields removes the dependency on
the typedef, allowing events-internal.h to be a
self-contained header. This will then let the patch
following this move event-internal.h to the top of
generated-events.h, so we can expose TraceEvent
struct variables in generated-events.h
Instead of having a global dstate array, declare a single
'uint16 TRACE_${EVENT_NAME}_DSTATE' variable for each
trace event. Record a pointer to this variable in the
TraceEvent struct too.
By turning trace_event_get_state_dynamic_by_id into a
macro, this still hits the fast path, and cache affinity
is ensured by declaring all the uint16 vars adjacent to
each other.
Currently methods which want to iterate over trace events,
do so using the trace_event_count() and trace_event_id()
methods. This leaks the concept of a single ID enum to
the callers. There is an alternative trace_event_pattern()
method which can be used in an iteration context, but its
design is stateless, so is not easy to expand it in the
future.
This defines a formal iterator API will provide a future-
proof way of iterating over events.
The iterator is also able to apply a pattern match filter
to events, further removing the need for the pattern
We add TCP,UDP,ICMP packet comparison to replace
IP packet comparison. This can increase the
accuracy of the package comparison.
Less checkpoint more efficiency.
Eric Blake [Tue, 11 Oct 2016 15:46:23 +0000 (10:46 -0500)]
build: Work around SIZE_MAX bug in OSX headers
C99 requires SIZE_MAX to be declared with the same type as the
integral promotion of size_t, but OSX mistakenly defines it as
an 'unsigned long long' expression even though size_t is only
'unsigned long'. Rather than futzing around with whether size_t
is 32- or 64-bits wide (which would be needed if we cared about
using SIZE_T in a #if expression), just hard-code it with a cast.
This is not a strict C99-compliant definition, because it doesn't
work in the preprocessor, but if we later need that, the build
will break on Mac to inform us to improve our replacement at that
time.
See also https://patchwork.ozlabs.org/patch/542327/ for an
instance where the wrong type trips us up if we don't fix it
for good in osdep.h.
Some versions of glibc make a similar mistake with SSIZE_MAX; the
goal is that the approach of this patch could be copied to work
around that problem if it ever becomes important to us.
Peter Maydell [Mon, 10 Oct 2016 14:19:20 +0000 (15:19 +0100)]
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Block layer patches
# gpg: Signature made Mon 10 Oct 2016 12:33:14 BST
# gpg: using RSA key 0x7F09B272C88F2FD6
# gpg: Good signature from "Kevin Wolf <[email protected]>"
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74 56FE 7F09 B272 C88F 2FD6
* remotes/kevin/tags/for-upstream:
dmg: Move libbz2 code to dmg-bz2.so
module: Don't load the same module if requested multiple times
scripts: Allow block module to not define BlockDriver
block: Add qdev ID to DEVICE_TRAY_MOVED
block-backend: Remember if attached device is non-qdev
block: Add node name to BLOCK_IO_ERROR event
block: Add bdrv_runtime_opts to query-command-line-options
block: use aio_bh_schedule_oneshot
async: add aio_bh_schedule_oneshot
block: use bdrv_add_before_write_notifier
* remotes/mjt/tags/trivial-patches-fetch: (26 commits)
net/filter-mirror: Fix mirror initial check typo
virtio: rename the bar index field name in VirtIOPCIProxy
linux-user: include <poll.h> instead of <sys/poll.h>
char: fix missing return in error path for chardev TLS init
CODING_STYLE: Fix a typo ("have" vs. "has")
bitmap: refine and move BITMAP_{FIRST/LAST}_WORD_MASK
build-sys: fix find-in-path
m68k: change default system clock for m5208evb
exec: remove unused compacted argument
usb: ehci: fix memory leak in ehci_process_itd
qapi: make the json schema files more regular.
maint: Add module_block.h to .gitignore
MAINTAINERS: Some updates related to the SH4 machines
MAINTAINERS: Add some more MIPS related files
MAINTAINERS: Add usermode related config files
MAINTAINERS: Add some more pattern to recognize all win32 related files
MAINTAINERS: Add some more rocker related files
MAINTAINERS: Add header files to CRIS section
MAINTAINERS: Add some more files to the virtio section
MAINTAINERS: Add some SPARC machine related files
...
Peter Maydell [Mon, 10 Oct 2016 10:45:09 +0000 (11:45 +0100)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2016-10-07' into staging
QAPI patches for 2016-10-07
# gpg: Signature made Fri 07 Oct 2016 18:55:40 BST
# gpg: using RSA key 0x3870B400EB918653
# gpg: Good signature from "Markus Armbruster <[email protected]>"
# gpg: aka "Markus Armbruster <[email protected]>"
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867 4E5F 3870 B400 EB91 8653
* remotes/armbru/tags/pull-qapi-2016-10-07:
docs: Belatedly update for move of QMP/* to docs/
docs: Belatedly update for move of qmp-commands.txt
qmp: Disable query-cpu-* commands when they're unavailable
MAINTAINERS: Pass the QObject staff from Luiz to Markus
MAINTAINERS: Pass the HMP staff from Luiz to David
qapi: return a 'missing parameter' error
qapi: assert list entry has a value
qapi: add assert about root value
tests/test-qmp-input-strict: Cover missing struct members
qapi: Fix crash when 'any' or 'null' parameter is missing
qmp: fix object-add assert() without props
* remotes/bonzini/tags/for-upstream: (39 commits)
qemu-doc: merge qemu-tech and qemu-doc
qemu-tech: rewrite some parts
qemu-tech: reorganize content
qemu-tech: move TCG test documentation to tests/tcg/README
qemu-tech: move user mode emulation features from qemu-tech
qemu-tech: document lazy condition code evaluation in cpu.h
qemu-tech: move text from qemu-tech to tcg/README
qemu-doc: drop installation and compilation notes
qemu-doc: replace introduction with the one from the internals manual
qemu-tech: drop index
test-qht: perform lookups under rcu_read_lock
qht: fix unlock-after-free segfault upon resizing
qht: simplify qht_reset_size
qemu-nbd: Shrink image size by specified offset
qemu_kill_report: Report PID name too
util: Introduce qemu_get_pid_name
char: update read handler in all cases
char: use a fixed idx for child muxed chr
i8259: give ISA device when registering ISA ioports
.travis.yml: add gcc sanitizer build
...
Halil Pasic [Thu, 6 Oct 2016 12:55:44 +0000 (14:55 +0200)]
virtio-gpu: convert VMSTATE_VIRTIO_DEVICE
Use the new VMSTATE_VIRTIO_DEVICE macro. The device virtio-gpu is
special because it actually does not adhere to the virtio migration
schema, because device state is last.
In most cases the functions passed to VMSTATE_VIRTIO_DEVICE
only call the virtio_load and virtio_save wrappers. Some include some
pre- and post- massaging too. The massaging is better expressed
as such in the VMStateDescription.
Let us prepare for changing the semantic of the VMSTATE_VIRTIO_DEVICE
macro so that it is more similar to the other VMSTATE_*_DEVICE macros
in a sense that it is a field definition.
The preprocessor conditionals are going to be removed as soon as
every usage is converted to the new semantic.
The vhost-user & colo code is poking at the QemuOpts instance
in the CharDriverState struct, not realizing that it is valid
for this to be NULL. e.g. the following crash shows a codepath
where it will be NULL:
Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x000055baf6ab4adc in qemu_opt_foreach (opts=0x0, func=0x55baf696b650 <net_vhost_chardev_opts>, opaque=0x7ffc51368c00, errp=0x7ffc51368e48) at util/qemu-option.c:617
617 QTAILQ_FOREACH(opt, &opts->head, next) {
[Current thread is 1 (Thread 0x7f1d4970bb40 (LWP 6603))]
(gdb) bt
#0 0x000055baf6ab4adc in qemu_opt_foreach (opts=0x0, func=0x55baf696b650 <net_vhost_chardev_opts>, opaque=0x7ffc51368c00, errp=0x7ffc51368e48) at util/qemu-option.c:617
#1 0x000055baf696b7da in net_vhost_parse_chardev (opts=0x55baf8ff9260, errp=0x7ffc51368e48) at net/vhost-user.c:314
#2 0x000055baf696b985 in net_init_vhost_user (netdev=0x55baf8ff9250, name=0x55baf879d270 "hostnet2", peer=0x0, errp=0x7ffc51368e48) at net/vhost-user.c:360
#3 0x000055baf6960216 in net_client_init1 (object=0x55baf8ff9250, is_netdev=true, errp=0x7ffc51368e48) at net/net.c:1051
#4 0x000055baf6960518 in net_client_init (opts=0x55baf776e7e0, is_netdev=true, errp=0x7ffc51368f00) at net/net.c:1108
#5 0x000055baf696083f in netdev_add (opts=0x55baf776e7e0, errp=0x7ffc51368f00) at net/net.c:1186
#6 0x000055baf69608c7 in qmp_netdev_add (qdict=0x55baf7afaf60, ret=0x7ffc51368f50, errp=0x7ffc51368f48) at net/net.c:1205
#7 0x000055baf6622135 in handle_qmp_command (parser=0x55baf77fb590, tokens=0x7f1d24011960) at /path/to/qemu.git/monitor.c:3978
#8 0x000055baf6a9d099 in json_message_process_token (lexer=0x55baf77fb598, input=0x55baf75acd20, type=JSON_RCURLY, x=113, y=19) at qobject/json-streamer.c:105
#9 0x000055baf6abf7aa in json_lexer_feed_char (lexer=0x55baf77fb598, ch=125 '}', flush=false) at qobject/json-lexer.c:319
#10 0x000055baf6abf8f2 in json_lexer_feed (lexer=0x55baf77fb598, buffer=0x7ffc51369170 "}R\204\367\272U", size=1) at qobject/json-lexer.c:369
#11 0x000055baf6a9d13c in json_message_parser_feed (parser=0x55baf77fb590, buffer=0x7ffc51369170 "}R\204\367\272U", size=1) at qobject/json-streamer.c:124
#12 0x000055baf66221f7 in monitor_qmp_read (opaque=0x55baf77fb530, buf=0x7ffc51369170 "}R\204\367\272U", size=1) at /path/to/qemu.git/monitor.c:3994
#13 0x000055baf6757014 in qemu_chr_be_write_impl (s=0x55baf7610a40, buf=0x7ffc51369170 "}R\204\367\272U", len=1) at qemu-char.c:387
#14 0x000055baf6757076 in qemu_chr_be_write (s=0x55baf7610a40, buf=0x7ffc51369170 "}R\204\367\272U", len=1) at qemu-char.c:399
#15 0x000055baf675b3b0 in tcp_chr_read (chan=0x55baf90244b0, cond=G_IO_IN, opaque=0x55baf7610a40) at qemu-char.c:2927
#16 0x000055baf6a5d655 in qio_channel_fd_source_dispatch (source=0x55baf7610df0, callback=0x55baf675b25a <tcp_chr_read>, user_data=0x55baf7610a40) at io/channel-watch.c:84
#17 0x00007f1d3e80cbbd in g_main_context_dispatch () from /usr/lib64/libglib-2.0.so.0
#18 0x000055baf69d3720 in glib_pollfds_poll () at main-loop.c:213
#19 0x000055baf69d37fd in os_host_main_loop_wait (timeout=126000000) at main-loop.c:258
#20 0x000055baf69d38ad in main_loop_wait (nonblocking=0) at main-loop.c:506
#21 0x000055baf676587b in main_loop () at vl.c:1908
#22 0x000055baf676d3bf in main (argc=101, argv=0x7ffc5136a6c8, envp=0x7ffc5136a9f8) at vl.c:4604
(gdb) p opts
$1 = (QemuOpts *) 0x0
The crash occurred when attaching vhost-user net via QMP:
Code using chardevs should not be poking at the internals of the
CharDriverState struct. What vhost-user wants is a chardev that is
operating as reconnectable network service, along with the ability
to do FD passing over the connection. The colo code simply wants
a network service. Add a feature concept to the char drivers so
that chardev users can query the actual features they wish to have
supported. The QemuOpts member is removed to prevent future mistakes
in this area.
Greg Kurz [Fri, 30 Sep 2016 15:13:40 +0000 (17:13 +0200)]
virtio-scsi: convert virtio_scsi_bad_req() to use virtio_error()
The virtio_scsi_bad_req() function is called when a guest sends a
request with missing or ill-sized headers. This generally happens
when the virtio_scsi_parse_req() function returns an error.
With this patch, virtio_scsi_bad_req() will mark the device as broken,
detach the request from the virtqueue and free it, instead of forcing
QEMU to exit.
In nearly all locations where virtio_scsi_bad_req() is called, the only
thing to do next is to return to the caller.
The virtio_scsi_handle_cmd_req_prepare() function is an exception though.
It is called in a loop by virtio_scsi_handle_cmd_vq() and passed requests
freshly popped from a cmd virtqueue; virtio_scsi_handle_cmd_req_prepare()
does some sanity checks on the request and returns a boolean flag to
indicate whether the request should be queued or not. In the latter case,
virtio_scsi_handle_cmd_req_prepare() has detected a non-fatal error and
sent a response back to the guest.
We have now a new condition to take into account: the device is broken
and should stop all processing.
The return value of virtio_scsi_handle_cmd_req_prepare() is hence changed
to an int. A return value of zero means that the request should be queued.
Other non-fatal error cases where the request shoudn't be queued return
a negative errno (values are vaguely inspired by the error condition, but
the only goal here is to discriminate the case we're interested in).
And finally, if virtio_scsi_bad_req() was called, -EINVAL is returned. In
this case, virtio_scsi_handle_cmd_vq() detaches and frees already queued
requests, instead of submitting them.
Greg Kurz [Fri, 30 Sep 2016 15:13:32 +0000 (17:13 +0200)]
virtio-net: handle virtio_net_flush_tx() errors
All these errors are caused by a buggy guest: let's switch the device to
the broken state instead of terminating QEMU. Also we detach the element
from the virtqueue and free it.
If this happens, virtio_net_flush_tx() also returns -EINVAL, so that all
callers can stop processing the virtqueue immediatly.
Greg Kurz [Fri, 30 Sep 2016 15:13:24 +0000 (17:13 +0200)]
virtio-net: handle virtio_net_receive() errors
All these errors are caused by a buggy guest: let's switch the device to
the broken state instead of terminating QEMU. Also we detach the element
from the virtqueue and free it.
Greg Kurz [Fri, 30 Sep 2016 15:13:16 +0000 (17:13 +0200)]
virtio-net: handle virtio_net_handle_ctrl() error
This error is caused by a buggy guest: let's switch the device to the
broken state instead of terminating QEMU. Also we detach the element
from the virtqueue and free it.
All these errors are caused by a buggy guest: QEMU should not exit.
With this patch, if virtio_blk_handle_request() detects a buggy request, it
marks the device as broken and returns an error to the caller so it takes
appropriate action.
In the case of virtio_blk_handle_vq(), we detach the request from the
virtqueue, free its allocated memory and stop popping new requests.
We don't need to bother about multireq since virtio_blk_handle_request()
errors out early and mrb.num_reqs == 0.
In the case of virtio_blk_dma_restart_bh(), we need to detach and free all
queued requests as well.
Greg Kurz [Fri, 30 Sep 2016 15:12:58 +0000 (17:12 +0200)]
virtio-9p: handle handle_9p_output() error
A broken guest may send a request without providing buffers for the reply
or for the request itself, and virtqueue_pop() will return an element with
either in_num == 0 or out_num == 0.
All 9P requests are expected to start with the following 7-byte header:
uint32_t size_le;
uint8_t id;
uint16_t tag_le;
If iov_to_buf() fails to return these 7 bytes, then something is wrong in
the guest.
In both cases, it is wrong to crash QEMU, since the root cause lies in the
guest.
This patch hence does the following:
- keep the check of in_num since pdu_complete() assumes it has enough
space to store the reply and we will send something broken to the guest
- let iov_to_buf() handle out_num == 0, since it will return 0 just like
if the guest had provided an zero-sized buffer.
- call virtio_error() to inform the guest that the device is now broken,
instead of aborting
- detach the request from the virtqueue and free it
Ports enter a "throttled" state when writing to the chardev would block.
The current output VirtQueueElement is kept around until the chardev
becomes writable again.
There are several places in the virtio-serial lifecycle where the
VirtQueueElement should be thrown away. For example, if the virtio
device is reset then virtqueue elements are no longer valid.
This patch adds the discard_throttle_data() function to unmap the
scatter-gather list and decrement vq->inuse. This ensures that the
VirtQueueElement is freed properly.
Stefan Hajnoczi [Mon, 19 Sep 2016 13:28:03 +0000 (14:28 +0100)]
virtio: add virtio_detach_element()
During device reset or similar situations a VirtQueueElement needs to be
freed without pushing it onto the used ring or rewinding the virtqueue.
Extract a new function to do this.
Later patches add virtio_detach_element() calls to existing device so
that scatter-gather lists are unmapped and vq->inuse goes back to zero
during device reset. Currently some devices don't bother and simply
call g_free(elem) which is not a clean way to throw away a
VirtQueueElement.
Igor Mammedov [Wed, 5 Oct 2016 15:51:24 +0000 (17:51 +0200)]
acpi: provide _PXM method for CPU devices if QEMU is started numa enabled
Workaround for long standing issue where Linux kernel
assigns hotplugged CPU to 1st numa node as it discards
proximity for possible CPUs from SRAT after it's parsed.
_PXM method allows linux query proximity directly from
hotplugged CPU object, which allows Linux to assing CPU
to the correct numa node.
Add support for enabling the virtio 1.0 "emergency write"
(VIRTIO_CONSOLE_F_EMERG_WRITE) feature. The previous patch introduced
the plumbing required for this; now we expose the virtio feature to
the guest. The feature is disabled for compatibility machines to avoid
exposing a new feature to existing guests.
As required by the virtio 1.0 spec, the emergency write functionality
is available to the guest even if the guest doesn't negotatiate the
feature, as well as before feature negotation.
virtio-serial: add plumbing for virtio console emergency write support
Add the infrastructure required for the virtio 1.0 "emergency write"
(VIRTIO_CONSOLE_F_EMERG_WRITE) feature. Because we don't touch the
size of the configuration area, guests will not be able to actually
make use of this without further patches.
char: fix missing return in error path for chardev TLS init
If the qio_channel_tls_new_(server|client) methods fail,
we disconnect the client. Unfortunately a missing return
means we then go on to try and run the TLS handshake on
a NULL I/O channel. This gives predictably segfaulty
results.
The main way to trigger this is to request a bogus TLS
priority string for the TLS credentials. e.g.
projects / qemu.git / log