linux-user: init_guest_space: Try to make ARM space+commpage continuous

At a fixed distance after the usable memory that init_guest_space maps, for
32-bit ARM targets we also need to map a commpage.  The normal
init_guest_space logic doesn't keep this in mind when searching for an
address range.

If !host_start, then try to find a big continuous segment where we can put
both the usable memory and the commpage; we then munmap that segment and
set current_start to that address; and let the normal code mmap the usable
memory and the commpage separately.  That is: if we don't have hint of
where to start looking for memory, come up with one that is better than
NULL.  Depending on host_size and guest_start, there may or may not be a
gap between the usable memory and the commpage, so this is slightly more
restrictive than it needs to be; but it's only a hint, so that's OK.

We only do that for !host start, because if host_start, then either:
 - we got an address passed in with -B, in which case we don't want to
   interfere with what the user said;
 - or host_start is based off of the ELF image's loaddr.  The check "if
   (host_start && real_start != current_start)" suggests that we really
   want lowest available address that is >= loaddr.  I don't know why that
   is, but I'm trusting that Paul Brook knew what he was doing when he
   wrote the original version of that check in
   c581deda322080e8beb88b2e468d4af54454e4b3 way back in 2010.

Signed-off-by: Luke Shumaker <lukeshu@parabola.nu>
Message-Id: <20171228180814.9749-11-lukeshu@lukeshu.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>

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

Machine and x86 queue, 2018-03-19

* cpu_model/cpu_type cleanups
* x86: Fix on Intel Processor Trace CPUID checks

# gpg: Signature made Mon 19 Mar 2018 20:07:14 GMT
# gpg:                using RSA key 2807936F984DC5A6
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>"
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF  D1AA 2807 936F 984D C5A6

* remotes/ehabkost/tags/machine-next-pull-request:
  i386: Disable Intel PT if packets IP payloads have LIP values
  cpu: drop unnecessary NULL check and cpu_common_class_by_name()
  cpu: get rid of unused cpu_init() defines
  Use cpu_create(type) instead of cpu_init(cpu_model)
  cpu: add CPU_RESOLVING_TYPE macro
  tests: add machine 'none' with -cpu test
  nios2: 10m50_devboard: replace cpu_model with cpu_type

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

RISC-V: Fix riscv_isa_string memory size bug

This version uses a constant size memory buffer sized for
the maximum possible ISA string length. It also uses g_new
instead of g_new0, uses more efficient logic to append
extensions and adds manual zero termination of the string.

Cc: Palmer Dabbelt <palmer@sifive.com>
Cc: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Michael Clark <mjc@sifive.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
[PMM: Use qemu_tolower() rather than tolower()]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge remote-tracking branch 'remotes/ericb/tags/pull-qapi-2018-03-12-v4' into staging

qapi patches for 2018-03-12, 2.12 softfreeze

- Marc-André Lureau: 0/4 qapi: generate a literal qobject for introspection
- Max Reitz: 0/7 block: Handle null backing link
- Daniel P. Berrange: chardev: tcp: postpone TLS work until machine done
- Peter Xu: 00/23 QMP: out-of-band (OOB) execution support
- Vladimir Sementsov-Ogievskiy: 0/2 block latency histogram
- Eric Blake: qapi: Pass '-u' when doing non-silent diff

# gpg: Signature made Mon 19 Mar 2018 19:59:04 GMT
# gpg:                using RSA key A7A16B4A2527436A
# gpg: Good signature from "Eric Blake <eblake@redhat.com>"
# gpg:                 aka "Eric Blake (Free Software Programmer) <ebb9@byu.net>"
# gpg:                 aka "[jpeg image of size 6874]"
# Primary key fingerprint: 71C2 CC22 B1C4 6029 27D2  F3AA A7A1 6B4A 2527 436A

* remotes/ericb/tags/pull-qapi-2018-03-12-v4: (38 commits)
  qapi: Pass '-u' when doing non-silent diff
  qapi: add block latency histogram interface
  block/accounting: introduce latency histogram
  tests: qmp-test: add oob test
  tests: qmp-test: verify command batching
  qmp: add command "x-oob-test"
  monitor: enable IO thread for (qmp & !mux) typed
  qmp: isolate responses into io thread
  qmp: support out-of-band (oob) execution
  qapi: introduce new cmd option "allow-oob"
  monitor: send event when command queue full
  qmp: add new event "command-dropped"
  monitor: separate QMP parser and dispatcher
  monitor: let suspend/resume work even with QMPs
  monitor: let suspend_cnt be thread safe
  monitor: introduce monitor_qmp_respond()
  qmp: introduce QMPCapability
  monitor: allow using IO thread for parsing
  monitor: let mon_list be tail queue
  monitor: unify global init
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

i386: Disable Intel PT if packets IP payloads have LIP values

Intel processor trace should be disabled when
CPUID.(EAX=14H,ECX=0H).ECX.[bit31] is set.
Generated packets which contain IP payloads will have LIP
values when this bit is set, or IP payloads will have RIP
values.
Currently, The information of CPUID 14H is constant to make
live migration safty and this bit is always 0 in guest even
if host support LIP values.
Guest sees the bit is 0 will expect IP payloads with RIP
values, but the host CPU will generate IP payloads with
LIP values if this bit is set in HW.
To make sure the value of IP payloads correctly, Intel PT
should be disabled when bit[31] is set.

Signed-off-by: Luwei Kang <luwei.kang@intel.com>
Message-Id: <1520969191-18162-1-git-send-email-luwei.kang@intel.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

qapi: Pass '-u' when doing non-silent diff

Ed-script diffs are awful compared to context diffs.  Fix another
'diff -q' while in the area (if the files are different, being
noisy makes it easier to diagnose why).

While at it, diff .err before .out, because if a test fails, .err
is more likely to contain the most important information for
fixing the failure.

Fixes: 46ec4fce
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20180315125116.804342-1-eblake@redhat.com>

qapi: add block latency histogram interface

Set (and clear) histograms through new command
block-latency-histogram-set and show new statistics in
query-blockstats results.

For now, the command is marked experimental with prefix 'x-',
to gain experience with the interface without being stuck
with design decisions.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180309165212.97144-3-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[eblake: fix typos, mention x- prefix in commit message]
Signed-off-by: Eric Blake <eblake@redhat.com>

block/accounting: introduce latency histogram

Introduce latency histogram statics for block devices.
For each accounted operation type, the latency region [0, +inf) is
divided into subregions by several points. Then, calculate
hits for each subregion.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180309165212.97144-2-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

tests: qmp-test: add oob test

Test the new OOB capability.  Here we used the new "x-oob-test" command.
First, we send a lock=true and oob=false command to hang the main
thread.  Then send another lock=false and oob=true command (which will
be run inside parser this time) to free that hanged command.

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-24-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: grammar tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>

tests: qmp-test: verify command batching

OOB introduced DROP event for flow control.  This should not affect old
QMP clients.  Add a command batching check to make sure of it.

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-23-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

qmp: add command "x-oob-test"

This command is only used to test OOB functionality.  It should not be
used for any other purposes.

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-22-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: grammar tweak]
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: enable IO thread for (qmp & !mux) typed

Start to use dedicate IO thread for QMP monitors that are not using
MUXed chardev.

Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-21-peterx@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

qmp: isolate responses into io thread

For those monitors who have enabled IO thread, we'll offload the
responding procedure into IO thread.  The main reason is that chardev is
not thread safe, and we need to do all the read/write IOs in the same
thread.  For use_io_thr=true monitors, that thread is the IO thread.

We do this isolation in similar pattern as what we have done to the
request queue: we first create one response queue for each monitor, then
instead of replying directly in the main thread, we queue the responses
and kick the IO thread to do the rest of the job for us.

A funny thing after doing this is that, when the QMP clients send "quit"
to QEMU, it's possible that we close the IOThread even earlier than
replying to that "quit".  So another thing we need to do before cleaning
up the monitors is that we need to flush the response queue (we don't
need to do that for command queue; after all we are quitting) to make
sure replies for handled commands are always flushed back to clients.

Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-20-peterx@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

qmp: support out-of-band (oob) execution

Having "allow-oob":true for a command does not mean that this command
will always be run in out-of-band mode.  The out-of-band quick path will
only be executed if we specify the extra "run-oob" flag when sending the
QMP request:

    { "execute":   "command-that-allows-oob",
      "arguments": { ... },
      "control":   { "run-oob": true } }

The "control" key is introduced to store this extra flag.  "control"
field is used to store arguments that are shared by all the commands,
rather than command specific arguments.  Let "run-oob" be the first.

Note that in the patch I exported qmp_dispatch_check_obj() to be used to
check the request earlier, and at the same time allowed "id" field to be
there since actually we always allow that.

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-19-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: rebase to qobject_to(), spelling fix]
Signed-off-by: Eric Blake <eblake@redhat.com>

qapi: introduce new cmd option "allow-oob"

Here "oob" stands for "Out-Of-Band".  When "allow-oob" is set, it means
the command allows out-of-band execution.

The "oob" idea is proposed by Markus Armbruster in following thread:

  https://lists.gnu.org/archive/html/qemu-devel/2017-09/msg02057.html

This new "allow-oob" boolean will be exposed by "query-qmp-schema" as
well for command entries, so that QMP clients can know which commands
can be used in out-of-band calls. For example the command "migrate"
originally looks like:

  {"name": "migrate", "ret-type": "17", "meta-type": "command",
   "arg-type": "86"}

And it'll be changed into:

  {"name": "migrate", "ret-type": "17", "allow-oob": false,
   "meta-type": "command", "arg-type": "86"}

This patch only provides the QMP interface level changes.  It does not
contain the real out-of-band execution implementation yet.

Suggested-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-18-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: rebase on introspection done by qlit]
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: send event when command queue full

Set maximum QMP command queue length to 8.  If the queue is full,
instead of queuing the command, we directly return a "command-dropped"
event, telling the client that a specific command is dropped.

Note that this flow control mechanism is only valid if OOB is enabled.
If it's not, the effective queue length will always be 1, which strictly
follows original behavior of QMP command handling (which never drops
messages).

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-17-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: commit message grammar, abort on failure to send event]
Signed-off-by: Eric Blake <eblake@redhat.com>

qmp: add new event "command-dropped"

This event will be emitted if one QMP command is dropped.  Also,
declare an enum for the reasons.

Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-16-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: rebase to master]
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: separate QMP parser and dispatcher

Originally QMP goes through these steps:

  JSON Parser --> QMP Dispatcher --> Respond
      /|\    (2)                (3)     |
   (1) |                               \|/ (4)
       +---------  main thread  --------+

This patch does this:

  JSON Parser     QMP Dispatcher --> Respond
      /|\ |           /|\       (4)     |
       |  | (2)        | (3)            |  (5)
   (1) |  +----->      |               \|/
       +---------  main thread  <-------+

So the parsing job and the dispatching job is isolated now.  It gives us
a chance in follow up patches to totally move the parser outside.

The isolation is done using one QEMUBH. Only one dispatcher QEMUBH is
used for all the monitors.

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-15-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: grammar tweaks, rebase to qobject_to()]
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: let suspend/resume work even with QMPs

This patches allows QMP monitors to be suspended/resumed.

One thing to mention is that for QMPs that are using IOThreads, we need
an explicit kick for the IOThread in case it is sleeping.

Meanwhile, we need to take special care on non-interactive HMPs.
Currently only gdbserver is using that.  For these monitors, we still
don't allow suspend/resume operations.

Since at it, add traces for the operations.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-14-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: let suspend_cnt be thread safe

Monitor code now can be run in more than one thread.  Let it be thread
safe when accessing suspend_cnt counter.

Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-13-peterx@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: introduce monitor_qmp_respond()

A tiny refactoring, preparing to split the QMP dispatcher away.

Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-12-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: rebase to qobject_to() usage]
Signed-off-by: Eric Blake <eblake@redhat.com>

qmp: introduce QMPCapability

There were no QMP capabilities defined.  Define the first capability,
"oob", to allow out-of-band messages.

After this patch, we will allow QMP clients to enable QMP capabilities
when sending the first "qmp_capabilities" command.  Originally we are
starting QMP session with no arguments like:

  { "execute": "qmp_capabilities" }

Now we can enable some QMP capabilities using (take OOB as example,
which is the only capability that we support):

  { "execute": "qmp_capabilities",
    "arguments": { "enable": [ "oob" ] } }

When the "arguments" key is not provided, no capability is enabled.

For capability "oob", the monitor needs to be run on a dedicated IO
thread, otherwise the command will fail.  For example, trying to enable
OOB on a MUXed typed QMP monitor will fail.

One thing to mention is that QMP capabilities are per-monitor, and also
when the connection is closed due to some reason, the capabilities will
be reset.

Also, touch up qmp-test.c to test the new bits.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-11-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: touch up commit message]
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: allow using IO thread for parsing

For each Monitor, add one field "use_io_thr" to show whether it will be
using the dedicated monitor IO thread to handle input/output.  When set,
monitor IO parsing work will be offloaded to the dedicated monitor IO
thread, rather than the original main loop thread.

This only works for QMP.  HMP will always be run on the main loop
thread.

Currently we're still keeping use_io_thr off always.  Will turn it on
later at some point.

One thing to mention is that we cannot set use_io_thr for every QMP
monitor.  The problem is that MUXed typed chardevs may not work well
with it now. When MUX is used, frontend of chardev can be the monitor
plus something else.  The only thing we know would be safe to be run
outside main thread so far is the monitor frontend. All the rest of the
frontends should still be run in main thread only.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-10-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: squash in Peter's followup patch to avoid test failures]
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: let mon_list be tail queue

It was QLIST.  I want to use this list to do monitor priority job later,
which need tail insertion ability.  So switching to a tail queue.

Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-9-peterx@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: unify global init

There are many places where the monitor initializes its globals:

- monitor_init_qmp_commands() at the very beginning
- single function to init monitor_lock
- in the first entry of monitor_init() using "is_first_init"

Unify them a bit.

monitor_lock is not used before monitor_init() (as confirmed by code
analysis and gdb watchpoints); so we are safe delaying what was a
constructor-time initialization of the mutex into the later first call
to monitor_init().

Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-8-peterx@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: move the cur_mon hack deeper for QMP

In monitor_qmp_read(), we have the hack to temporarily replace the
cur_mon pointer.  Now we move this hack deeper inside the QMP dispatcher
routine since the Monitor pointer can be actually obtained using
container_of() upon the parser object, just like most of the other JSON
parser users do.

This does not make much sense as a single patch.  However, this will be
a big step for the next patch, when the QMP dispatcher routine will be
split from the QMP parser.

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-7-peterx@redhat.com>
[eblake: rebase context of qobject_to() macro]
Signed-off-by: Eric Blake <eblake@redhat.com>

monitor: move skip_flush into monitor_data_init

It's part of the data init.  Collect it.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-6-peterx@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

qobject: let object_property_get_str() use new API

We can simplify object_property_get_str() using the new
qobject_get_try_str().

Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-5-peterx@redhat.com>
[eblake: rebase context of qobject_to() macro]
Signed-off-by: Eric Blake <eblake@redhat.com>

qobject: introduce qobject_get_try_str()

A quick way to fetch string from qobject when it's a QString.

Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-4-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: rebase to qobject_to() macro]
Signed-off-by: Eric Blake <eblake@redhat.com>

qobject: introduce qstring_get_try_str()

The only difference from qstring_get_str() is that it allows the qstring
to be NULL.  If so, NULL is returned.

CC: Eric Blake <eblake@redhat.com>
CC: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-3-peterx@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

docs: update QMP documents for OOB commands

Update both the developer and spec for the new QMP OOB (Out-Of-Band)
command.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180309090006.10018-2-peterx@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: grammar tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>

chardev: tcp: postpone TLS work until machine done

TLS handshake may create background GSource tasks, while we won't know
the correct GMainContext until the whole chardev (including frontend)
inited.  Let's postpone the initial TLS handshake until machine done.

For dynamically created tcp chardev, we don't postpone that by checking
the init_machine_done variable.

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
[peterx: add missing include line, do unit test]
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180308140714.28906-1-peterx@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

block: Deprecate "backing": ""

We have a clear replacement, so let's deprecate it.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-Id: <20180224154033.29559-8-mreitz@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

block: Handle null backing link

Instead of converting all "backing": null instances into "backing": "",
handle a null value directly in bdrv_open_inherit().

This enables explicitly null backing links for json:{} filenames.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-Id: <20180224154033.29559-7-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: rebase to qobject_to() parameter order and qapi headers split]
Signed-off-by: Eric Blake <eblake@redhat.com>

qapi: Make more of qobject_to()

This patch reworks some places which use either qobject_type() checks
plus qobject_to(), where the latter alone is sufficient, or NULL checks
plus qobject_type() checks where we can simply do a qobject_to() != NULL
check.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-Id: <20180224154033.29559-6-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: rebase to qobject_to() parameter ordering]
Signed-off-by: Eric Blake <eblake@redhat.com>

qapi: Remove qobject_to_X() functions

They are no longer needed now.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-Id: <20180224154033.29559-5-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

qapi: Replace qobject_to_X(o) by qobject_to(X, o)

This patch was generated using the following Coccinelle script:

@@
expression Obj;
@@
(
- qobject_to_qnum(Obj)
+ qobject_to(QNum, Obj)
|
- qobject_to_qstring(Obj)
+ qobject_to(QString, Obj)
|
- qobject_to_qdict(Obj)
+ qobject_to(QDict, Obj)
|
- qobject_to_qlist(Obj)
+ qobject_to(QList, Obj)
|
- qobject_to_qbool(Obj)
+ qobject_to(QBool, Obj)
)

and a bit of manual fix-up for overly long lines and three places in
tests/check-qjson.c that Coccinelle did not find.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-Id: <20180224154033.29559-4-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: swap order from qobject_to(o, X), rebase to master, also a fix
to latent false-positive compiler complaint about hw/i386/acpi-build.c]
Signed-off-by: Eric Blake <eblake@redhat.com>

qapi: Add qobject_to()

This is a dynamic casting macro that, given a QObject type, returns an
object as that type or NULL if the object is of a different type (or
NULL itself).

The macro uses lower-case letters because:
1. There does not seem to be a hard rule on whether qemu macros have to
   be upper-cased,
2. The current situation in qapi/qmp is inconsistent (compare e.g.
   QINCREF() vs. qdict_put()),
3. qobject_to() will evaluate its @obj parameter only once, thus it is
   generally not important to the caller whether it is a macro or not,
4. I prefer it aesthetically.

The macro parameter order is chosen with typename first for
consistency with other QAPI macros like QAPI_CLONE(), as well as
for legibility (read it as "qobject to" type "applied to" obj).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20180224154033.29559-3-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
[eblake: swap parameter order to list type first, avoid clang ubsan
warning on QOBJECT(NULL) and container_of(NULL,type,base)]
Signed-off-by: Eric Blake <eblake@redhat.com>

Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20180319' into staging

target-arm queue:
 * fsl-imx6: Fix incorrect Ethernet interrupt defines
 * dump: Update correct kdump phys_base field for AArch64
 * char: i.MX: Add support for "TX complete" interrupt
 * bcm2836/raspi: Fix various bugs resulting in panics trying
   to boot a Debian Linux kernel on raspi3

# gpg: Signature made Mon 19 Mar 2018 18:30:33 GMT
# gpg:                using RSA key 3C2525ED14360CDE
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>"
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>"
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>"
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* remotes/pmaydell/tags/pull-target-arm-20180319:
  hw/arm/raspi: Provide spin-loop code for AArch64 CPUs
  hw/arm/bcm2836: Hardcode correct CPU type
  hw/arm/bcm2836: Use correct affinity values for BCM2837
  hw/arm/bcm2836: Create proper bcm2837 device
  hw/arm/bcm2836: Rename bcm2836 type/struct to bcm283x
  hw/arm/bcm2386: Fix parent type of bcm2386
  hw/arm/boot: If booting a kernel in EL2, set SCR_EL3.HCE
  hw/arm/boot: assert that secure_boot and secure_board_setup are false for AArch64
  hw/arm/raspi: Don't do board-setup or secure-boot for raspi3
  char: i.MX: Add support for "TX complete" interrupt
  char: i.MX: Simplify imx_update()
  dump: Update correct kdump phys_base field for AArch64
  fsl-imx6: Swap Ethernet interrupt defines

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

hw/arm/raspi: Provide spin-loop code for AArch64 CPUs

The raspi3 has AArch64 CPUs, which means that our smpboot
code for keeping the secondary CPUs in a pen needs to have
a version for A64 as well as A32. Without this, the
secondary CPUs go into an infinite loop of taking undefined
instruction exceptions.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180313153458.26822-10-peter.maydell@linaro.org

hw/arm/bcm2836: Hardcode correct CPU type

Now we have separate types for BCM2386 and BCM2387, we might as well
just hard-code the CPU type they use rather than having it passed
through as an object property. This then lets us put the initialization
of the CPU object in init rather than realize.

Note that this change means that it's no longer possible on
the command line to use -cpu to ask for a different kind of
CPU than the SoC supports. This was never a supported thing to
do anyway; we were just not sanity-checking the command line.

This does require us to only build the bcm2837 object on
TARGET_AARCH64 configs, since otherwise it won't instantiate
due to the missing cortex-a53 device and "make check" will fail.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180313153458.26822-9-peter.maydell@linaro.org

hw/arm/bcm2836: Use correct affinity values for BCM2837

The BCM2837 sets the Aff1 field of the MPIDR affinity values for the
CPUs to 0, whereas the BCM2836 uses 0xf. Set this correctly, as it
is required for Linux to boot.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180313153458.26822-8-peter.maydell@linaro.org

hw/arm/bcm2836: Create proper bcm2837 device

The bcm2837 is pretty similar to the bcm2836, but it does have
some differences. Notably, the MPIDR affinity aff1 values it
sets for the CPUs are 0x0, rather than the 0xf that the bcm2836
uses, and if this is wrong Linux will not boot.

Rather than trying to have one device with properties that
configure it differently for the two cases, create two
separate QOM devices for the two SoCs. We use the same approach
as hw/arm/aspeed_soc.c and share code and have a data table
that might differ per-SoC. For the moment the two types don't
actually have different behaviour.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180313153458.26822-7-peter.maydell@linaro.org

hw/arm/bcm2836: Rename bcm2836 type/struct to bcm283x

Our BCM2836 type is really a generic one that can be any of
the bcm283x family. Rename it accordingly. We change only
the names which are visible via the header file to the
rest of the QEMU code, leaving private function names
in bcm2836.c as they are.

This is a preliminary to making bcm283x be an abstract
parent class to specific types for the bcm2836 and bcm2837.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180313153458.26822-6-peter.maydell@linaro.org

hw/arm/bcm2386: Fix parent type of bcm2386

The TypeInfo and state struct for bcm2386 disagree about what the
parent class is -- the TypeInfo says it's TYPE_SYS_BUS_DEVICE,
but the BCM2386State struct only defines the parent_obj field
as DeviceState. This would have caused problems if anything
actually tried to treat the object as a TYPE_SYS_BUS_DEVICE.
Fix the TypeInfo to use TYPE_DEVICE as the parent, since we don't
need any of the additional functionality TYPE_SYS_BUS_DEVICE
provides.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180313153458.26822-5-peter.maydell@linaro.org

hw/arm/boot: If booting a kernel in EL2, set SCR_EL3.HCE

If we're directly booting a Linux kernel and the CPU supports both
EL3 and EL2, we start the kernel in EL2, as it expects. We must also
set the SCR_EL3.HCE bit in this situation, so that the HVC
instruction is enabled rather than UNDEFing. Otherwise at least some
kernels will panic when trying to initialize KVM in the guest.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20180313153458.26822-4-peter.maydell@linaro.org

hw/arm/boot: assert that secure_boot and secure_board_setup are false for AArch64

Add some assertions that if we're about to boot an AArch64 kernel,
the board code has not mistakenly set either secure_boot or
secure_board_setup. It doesn't make sense to set secure_boot,
because all AArch64 kernels must be booted in non-secure mode.

It might in theory make sense to set secure_board_setup, but
we don't currently support that, because only the AArch32
bootloader[] code calls this hook; bootloader_aarch64[] does not.
Since we don't have a current need for this functionality, just
assert that we don't try to use it. If it's needed we'll add
it later.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180313153458.26822-3-peter.maydell@linaro.org

hw/arm/raspi: Don't do board-setup or secure-boot for raspi3

For the rpi1 and 2 we want to boot the Linux kernel via some
custom setup code that makes sure that the SMC instruction
acts as a no-op, because it's used for cache maintenance.
The rpi3 boots AArch64 kernels, which don't need SMC for
cache maintenance and always expect to be booted non-secure.
Don't fill in the aarch32-specific parts of the binfo struct.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180313153458.26822-2-peter.maydell@linaro.org

char: i.MX: Add support for "TX complete" interrupt

Add support for "TX complete"/TXDC interrupt generate by real HW since
it is needed to support guests other than Linux.

Based on the patch by Bill Paul as found here:
https://bugs.launchpad.net/qemu/+bug/1753314

Cc: qemu-devel@nongnu.org
Cc: qemu-arm@nongnu.org
Cc: Bill Paul <wpaul@windriver.com>
Cc: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Bill Paul <wpaul@windriver.com>
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Message-id: 20180315191141.6789-2-andrew.smirnov@gmail.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

char: i.MX: Simplify imx_update()

Code of imx_update() is slightly confusing since the "flags" variable
doesn't really corespond to anything in real hardware and server as a
kitchensink accumulating events normally reported via USR1 and USR2
registers.

Change the code to explicitly evaluate state of interrupts reported
via USR1 and USR2 against corresponding masking bits and use the to
detemine if IRQ line should be asserted or not.

NOTE: Check for UTS1_TXEMPTY being set has been dropped for two
reasons:

    1. Emulation code implements a single character FIFO, so this flag
       will always be set since characters are trasmitted as a part of
       the code emulating "push" into the FIFO

    2. imx_update() is really just a function doing ORing and maksing
       of reported events, so checking for UTS1_TXEMPTY should happen,
       if it's ever really needed should probably happen outside of
       it.

Cc: qemu-devel@nongnu.org
Cc: qemu-arm@nongnu.org
Cc: Bill Paul <wpaul@windriver.com>
Cc: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Message-id: 20180315191141.6789-1-andrew.smirnov@gmail.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

dump: Update correct kdump phys_base field for AArch64

For guest kernel that supports KASLR, the load address can change every
time when guest VM runs. To find the physical base address correctly,
current QEMU dump searches VMCOREINFO for the string "NUMBER(phys_base)=".
However this string pattern is only available on x86_64. AArch64 uses a
different field, called "NUMBER(PHYS_OFFSET)=". This patch makes sure
QEMU dump uses the correct string on AArch64.

Signed-off-by: Wei Huang <wei@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-id: 1520615003-20869-1-git-send-email-wei@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

fsl-imx6: Swap Ethernet interrupt defines

The sabrelite machine model used by qemu-system-arm is based on the
Freescale/NXP i.MX6Q processor. This SoC has an on-board ethernet
controller which is supported in QEMU using the imx_fec.c module
(actually called imx.enet for this model.)

The include/hw/arm/fsm-imx6.h file defines the interrupt vectors for the
imx.enet device like this:

 #define FSL_IMX6_ENET_MAC_1588_IRQ 118
 #define FSL_IMX6_ENET_MAC_IRQ 119

According to https://www.nxp.com/docs/en/reference-manual/IMX6DQRM.pdf,
page 225, in Table 3-1. ARM Cortex A9 domain interrupt summary,
interrupts are as follows.

150 ENET MAC 0 IRQ
151 ENET MAC 0 1588 Timer interrupt

where

150 - 32 == 118
151 - 32 == 119

In other words, the vector definitions in the fsl-imx6.h file are reversed.

Fixing the interrupts alone causes problems with older Linux kernels:
The Ethernet interface will fail to probe with Linux v4.9 and earlier.
Linux v4.1 and earlier will crash due to a bug in Ethernet driver probe
error handling. This is a Linux kernel problem, not a qemu problem:
the Linux kernel only worked by accident since it requested both interrupts.

For backward compatibility, generate the Ethernet interrupt on both interrupt
lines. This was shown to work from all Linux kernel releases starting with
v3.16.

Link: https://bugs.launchpad.net/qemu/+bug/1753309
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Message-id: 1520723090-22130-1-git-send-email-linux@roeck-us.net
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

cpu: drop unnecessary NULL check and cpu_common_class_by_name()

both do nothing as for the first all callers
   parse_cpu_model() and qmp_query_cpu_model_()
should provide non NULL value, so just abort if it's not so.

While at it drop cpu_common_class_by_name() which is not need
any more as every target has CPUClass::class_by_name callback
by now, though abort in case a new arch will forget to define one.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <1518013857-4372-1-git-send-email-imammedo@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

cpu: get rid of unused cpu_init() defines

cpu_init(cpu_model) were replaced by cpu_create(cpu_type) so
no users are left, remove it.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au> (ppc)
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <1518000027-274608-6-git-send-email-imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

Use cpu_create(type) instead of cpu_init(cpu_model)

With all targets defining CPU_RESOLVING_TYPE, refactor
cpu_parse_cpu_model(type, cpu_model) to parse_cpu_model(cpu_model)
so that callers won't have to know internal resolving cpu
type. Place it in exec.c so it could be called from both
target independed vl.c and *-user/main.c.

That allows us to stop abusing cpu type from
  MachineClass::default_cpu_type
as resolver class in vl.c which were confusing part of
cpu_parse_cpu_model().

Also with new parse_cpu_model(), the last users of cpu_init()
in null-machine.c and bsd/linux-user targets could be switched
to cpu_create() API and cpu_init() API will be removed by
follow up patch.

With no longer users left remove MachineState::cpu_model field,
new code should use MachineState::cpu_type instead and
leave cpu_model parsing to generic code in vl.c.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <1518000027-274608-5-git-send-email-imammedo@redhat.com>
[ehabkost: Fix bsd-user build error]
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

cpu: add CPU_RESOLVING_TYPE macro

it will be used for providing to cpu name resolving class for
parsing cpu model for system and user emulation code.

Along with change add target to null-machine tests, so
that when switch to CPU_RESOLVING_TYPE happens,
it would ensure that null-machine usecase still works.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Laurent Vivier <laurent@vivier.eu> (m68k)
Acked-by: David Gibson <david@gibson.dropbear.id.au> (ppc)
Acked-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de> (tricore)
Message-Id: <1518000027-274608-4-git-send-email-imammedo@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
[ehabkost: Added macro to riscv too]
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

tests: add machine 'none' with -cpu test

Check that "$QEMU -M none -cpu FOO" starts QEMU without error

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <1518000027-274608-3-git-send-email-imammedo@redhat.com>
[ehabkost: include qmp/qdict.h instead of qmp/types.h]
[ehabkost: add riscv targets to machine-none-test]
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

nios2: 10m50_devboard: replace cpu_model with cpu_type

use cpu_create() instead of being removed cpu_generic_init()

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <1518000027-274608-2-git-send-email-imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

compiler: Add QEMU_BUILD_BUG_MSG() macro

_Static_assert() allows us to specify messages, and that may come in
handy.  Even without _Static_assert(), encouraging developers to put a
helpful message next to the QEMU_BUILD_BUG_* may make debugging easier
whenever it breaks.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20180224154033.29559-2-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

qapi: generate a literal qobject for introspection

Replace the generated json string with a literal qobject. The later is
easier to deal with, at run time as well as compile time: adding #if
conditionals will be easier than in a json string.

The output of query-qmp-schema is not changed.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180305172951.2150-5-marcandre.lureau@redhat.com>
[eblake: fix python 3 failure]
Signed-off-by: Eric Blake <eblake@redhat.com>

qlit: add qobject_from_qlit()

Instantiate a QObject* from a literal QLitObject.

LitObject only supports int64_t for now.  uint64_t and double aren't
implemented.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180305172951.2150-4-marcandre.lureau@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

qlit: use QType instead of int

Suggested-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180305172951.2150-3-marcandre.lureau@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

qapi2texi: minor python code simplification

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180305172951.2150-2-marcandre.lureau@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

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

Block layer patches

# gpg: Signature made Mon 19 Mar 2018 11:01:45 GMT
# gpg:                using RSA key 7F09B272C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>"
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74  56FE 7F09 B272 C88F 2FD6

* remotes/kevin/tags/for-upstream: (46 commits)
  iotests: Avoid realpath, for CentOS 6
  block: fix iotest 146 output expectations
  iscsi: fix iSER compilation
  block: Fix leak of ignore_children in error path
  vvfat: Fix inherit_options flags
  block/mirror: change the semantic of 'force' of block-job-cancel
  vpc: Require aligned size in .bdrv_co_create
  vpc: Support .bdrv_co_create
  vhdx: Support .bdrv_co_create
  vdi: Make comments consistent with other drivers
  qed: Support .bdrv_co_create
  qcow: Support .bdrv_co_create
  qemu-iotests: Enable write tests for parallels
  parallels: Support .bdrv_co_create
  iotests: Add regression test for commit base locking
  block: Fix flags in reopen queue
  vdi: Implement .bdrv_co_create
  vdi: Move file creation to vdi_co_create_opts
  vdi: Pull option parsing from vdi_co_create
  qemu-iotests: Test luks QMP image creation
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

iotests: Avoid realpath, for CentOS 6

CentOS 6 lacks a realpath binary on the base install, which makes
all iotests runs fail since the 2.11 release:

001         - output mismatch (see 001.out.bad)
./check: line 815: realpath: command not found
diff: missing operand after `/home/dummy/qemu/tests/qemu-iotests/001.out'
diff: Try `diff --help' for more information.

Many of the uses of 'realpath' in the check script were being
used on the output of 'type -p' - but that is already an
absolute file name.  While a canonical name can often be
shorter (realpath gets rid of /../), it can also be longer (due
to symlink expansion); and we really don't care if the name is
canonical, merely that it was an executable file with an
absolute path.  These were broken in commit cceaf1db.

The remaining use of realpath was to convert a possibly relative
filename into an absolute one before calling diff to make it
easier to copy-and-paste the filename for moving the .bad file
into place as the new reference file even when running iotests
out-of-tree (see commit 93e53fb6), but $PWD can achieve the same
purpose.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: fix iotest 146 output expectations

Commit bff5554843 added "force_size" into the common.filter for
_filter_img_create(), but test 146 still expects it in the output.

Signed-off-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iscsi: fix iSER compilation

This fails in Fedora 28.

Reported-by: Andreas Schwab <schwab@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Fix leak of ignore_children in error path

Reported-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vvfat: Fix inherit_options flags

Overriding flags violates the precedence rules of
bdrv_reopen_queue_child. Just like the read-only option, no-flush should
be put into the options. The same is done in bdrv_temp_snapshot_options.

Reported-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/mirror: change the semantic of 'force' of block-job-cancel

When doing drive mirror to a low speed shared storage, if there was heavy
BLK IO write workload in VM after the 'ready' event, drive mirror block job
can't be canceled immediately, it would keep running until the heavy BLK IO
workload stopped in the VM.

Libvirt depends on the current block-job-cancel semantics, which is that
when used without a flag after the 'ready' event, the command blocks
until data is in sync.  However, these semantics are awkward in other
situations, for example, people may use drive mirror for realtime
backups while still wanting to use block live migration.  Libvirt cannot
start a block live migration while another drive mirror is in progress,
but the user would rather abandon the backup attempt as broken and
proceed with the live migration than be stuck waiting for the current
drive mirror backup to finish.

The drive-mirror command already includes a 'force' flag, which libvirt
does not use, although it documented the flag as only being useful to
quit a job which is paused.  However, since quitting a paused job has
the same effect as abandoning a backup in a non-paused job (namely, the
destination file is not in sync, and the command completes immediately),
we can just improve the documentation to make the force flag obviously
useful.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Jeff Cody <jcody@redhat.com>
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Max Reitz <mreitz@redhat.com>
Cc: Eric Blake <eblake@redhat.com>
Cc: John Snow <jsnow@redhat.com>
Reported-by: Huaitong Han <huanhuaitong@didichuxing.com>
Signed-off-by: Huaitong Han <huanhuaitong@didichuxing.com>
Signed-off-by: Liang Li <liliangleo@didichuxing.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vpc: Require aligned size in .bdrv_co_create

Perform the rounding to match a CHS geometry only in the legacy code
path in .bdrv_co_create_opts. QMP now requires that the user already
passes a CHS aligned image size, unless force-size=true is given.

CHS alignment is required to make the image compatible with Virtual PC,
but not for use with newer Microsoft hypervisors.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

vpc: Support .bdrv_co_create

This adds the .bdrv_co_create driver callback to vpc, which
enables image creation over QMP.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

vhdx: Support .bdrv_co_create

This adds the .bdrv_co_create driver callback to vhdx, which
enables image creation over QMP.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

vdi: Make comments consistent with other drivers

This makes the .bdrv_co_create(_opts) implementation of vdi look more
like the other recently converted block drivers.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

qed: Support .bdrv_co_create

This adds the .bdrv_co_create driver callback to qed, which
enables image creation over QMP.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

qcow: Support .bdrv_co_create

This adds the .bdrv_co_create driver callback to qcow, which
enables image creation over QMP.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>

qemu-iotests: Enable write tests for parallels

Originally we added parallels as a read-only format to qemu-iotests
where we did just some tests with a binary image. Since then, write and
image creation support has been added to the driver, so we can now
enable it in _supported_fmt generic.

The driver doesn't support migration yet, though, so we need to add it
to the list of exceptions in 181.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>

parallels: Support .bdrv_co_create

This adds the .bdrv_co_create driver callback to parallels, which
enables image creation over QMP.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>

iotests: Add regression test for commit base locking

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Fix flags in reopen queue

Reopen flags are not synchronized according to the
bdrv_reopen_queue_child precedence until bdrv_reopen_prepare. It is a
bit too late: we already check the consistency in bdrv_check_perm before
that.

This fixes the bug that when bdrv_reopen a RO node as RW, the flags for
backing child are wrong. Before, we could recurse with flags.rw=1; now,
role->inherit_options + update_flags_from_options will make sure to
clear the bit when necessary.  Note that this will not clear an
explicitly set bit, as in the case of parallel block jobs (e.g.
test_stream_parallel in 030), because the explicit options include
'read-only=false' (for an intermediate node used by a different job).

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vdi: Implement .bdrv_co_create

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vdi: Move file creation to vdi_co_create_opts

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vdi: Pull option parsing from vdi_co_create

In preparation of QAPI-fying VDI image creation, we have to create a
BlockdevCreateOptionsVdi type which is received by a (future)
vdi_co_create().

vdi_co_create_opts() now converts the QemuOpts object into such a
BlockdevCreateOptionsVdi object.  The protocol-layer file is still
created in vdi_co_do_create() (and BlockdevCreateOptionsVdi.file is set
to an empty string), but that will be addressed by a follow-up patch.

Note that cluster-size is not part of the QAPI schema because it is not
supported by default.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-iotests: Test luks QMP image creation

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

luks: Catch integer overflow for huge sizes

When you request an image size close to UINT64_MAX, the addition of the
crypto header may cause an integer overflow. Catch it instead of
silently truncating the image size.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

luks: Turn invalid assertion into check

The .bdrv_getlength implementation of the crypto block driver asserted
that the payload offset isn't after EOF. This is an invalid assertion to
make as the image file could be corrupted. Instead, check it and return
-EIO if the file is too small for the payload offset.

Zero length images are fine, so trigger -EIO only on offset > len, not
on offset >= len as the assertion did before.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

luks: Support .bdrv_co_create

This adds the .bdrv_co_create driver callback to luks, which enables
image creation over QMP.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

luks: Create block_crypto_co_create_generic()

Everything that refers to the protocol layer or QemuOpts is moved out of
block_crypto_create_generic(), so that the remaining function is
suitable to be called by a .bdrv_co_create implementation.

LUKS is the only driver that actually implements the old interface, and
we don't intend to use it in any new drivers, so put the moved out code
directly into a LUKS function rather than creating a generic
intermediate one.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>

luks: Separate image file creation from formatting

The crypto driver used to create the image file in a callback from the
crypto subsystem. If we want to implement .bdrv_co_create, this needs to
go away because that callback will get a reference to an already
existing block node.

Move the image file creation to block_crypto_create_generic().

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>

tests/test-blockjob: test cancellations

Whatever the state a blockjob is in, it should be able to be canceled
by the block layer.

Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: test manual job dismissal

Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: Expose manual property

Expose the "manual" property via QAPI for the backup-related jobs.
As of this commit, this allows the management API to request the
"concluded" and "dismiss" semantics for backup jobs.

Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: add block-job-finalize

Instead of automatically transitioning from PENDING to CONCLUDED, gate
the .prepare() and .commit() phases behind an explicit acknowledgement
provided by the QMP monitor if auto_finalize = false has been requested.

This allows us to perform graph changes in prepare and/or commit so that
graph changes do not occur autonomously without knowledge of the
controlling management layer.

Transactions that have reached the "PENDING" state together can all be
moved to invoke their finalization methods by issuing block_job_finalize
to any one job in the transaction.

Jobs in a transaction with mixed job->auto_finalize settings will all
remain stuck in the "PENDING" state, as if the entire transaction was
specified with auto_finalize = false. Jobs that specified
auto_finalize = true, however, will still not emit the PENDING event.

Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: add PENDING status and event

For jobs utilizing the new manual workflow, we intend to prohibit
them from modifying the block graph until the management layer provides
an explicit ACK via block-job-finalize to move the process forward.

To distinguish this runstate from "ready" or "waiting," we add a new
"pending" event and status.

For now, the transition from PENDING to CONCLUDED/ABORTING is automatic,
but a future commit will add the explicit block-job-finalize step.

Transitions:
Waiting -> Pending:   Normal transition.
Pending -> Concluded: Normal transition.
Pending -> Aborting:  Late transactional failures and cancellations.

Removed Transitions:
Waiting -> Concluded: Jobs must go to PENDING first.

Verbs:
Cancel: Can be applied to a pending job.

             +---------+
             |UNDEFINED|
             +--+------+
                |
             +--v----+
   +---------+CREATED+-----------------+
   |         +--+----+                 |
   |            |                      |
   |         +--+----+     +------+    |
   +---------+RUNNING<----->PAUSED|    |
   |         +--+-+--+     +------+    |
   |            | |                    |
   |            | +------------------+ |
   |            |                    | |
   |         +--v--+       +-------+ | |
   +---------+READY<------->STANDBY| | |
   |         +--+--+       +-------+ | |
   |            |                    | |
   |         +--v----+               | |
   +---------+WAITING<---------------+ |
   |         +--+----+                 |
   |            |                      |
   |         +--v----+                 |
   +---------+PENDING|                 |
   |         +--+----+                 |
   |            |                      |
+--v-----+   +--v------+               |
|ABORTING+--->CONCLUDED|               |
+--------+   +--+------+               |
                |                      |
             +--v-+                    |
             |NULL<--------------------+
             +----+

Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: add waiting status

For jobs that are stuck waiting on others in a transaction, it would
be nice to know that they are no longer "running" in that sense, but
instead are waiting on other jobs in the transaction.

Jobs that are "waiting" in this sense cannot be meaningfully altered
any longer as they have left their running loop. The only meaningful
user verb for jobs in this state is "cancel," which will cancel the
whole transaction, too.

Transitions:
Running -> Waiting:   Normal transition.
Ready   -> Waiting:   Normal transition.
Waiting -> Aborting:  Transactional cancellation.
Waiting -> Concluded: Normal transition.

Removed Transitions:
Running -> Concluded: Jobs must go to WAITING first.
Ready   -> Concluded: Jobs must go to WAITING first.

Verbs:
Cancel: Can be applied to WAITING jobs.

             +---------+
             |UNDEFINED|
             +--+------+
                |
             +--v----+
   +---------+CREATED+-----------------+
   |         +--+----+                 |
   |            |                      |
   |         +--v----+     +------+    |
   +---------+RUNNING<----->PAUSED|    |
   |         +--+-+--+     +------+    |
   |            | |                    |
   |            | +------------------+ |
   |            |                    | |
   |         +--v--+       +-------+ | |
   +---------+READY<------->STANDBY| | |
   |         +--+--+       +-------+ | |
   |            |                    | |
   |         +--v----+               | |
   +---------+WAITING<---------------+ |
   |         +--+----+                 |
   |            |                      |
+--v-----+   +--v------+               |
|ABORTING+--->CONCLUDED|               |
+--------+   +--+------+               |
                |                      |
             +--v-+                    |
             |NULL<--------------------+
             +----+

Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: add prepare callback

Some jobs upon finalization may need to perform some work that can
still fail. If these jobs are part of a transaction, it's important
that these callbacks fail the entire transaction.

We allow for a new callback in addition to commit/abort/clean that
allows us the opportunity to have fairly late-breaking failures
in the transactional process.

The expected flow is:

- All jobs in a transaction converge to the PENDING state,
  added in a forthcoming commit.
- Upon being finalized, either automatically or explicitly
  by the user, jobs prepare to complete.
- If any job fails preparation, all jobs call .abort.
- Otherwise, they succeed and call .commit.

Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: add block_job_txn_apply function

Simply apply a function transaction-wide.
A few more uses of this in forthcoming patches.

Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: add commit, abort, clean helpers

The completed_single function is getting a little mucked up with
checking to see which callbacks exist, so let's factor them out.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: ensure abort is called for cancelled jobs

Presently, even if a job is canceled post-completion as a result of
a failing peer in a transaction, it will still call .commit because
nothing has updated or changed its return code.

The reason why this does not cause problems currently is because
backup's implementation of .commit checks for cancellation itself.

I'd like to simplify this contract:

(1) Abort is called if the job/transaction fails
(2) Commit is called if the job/transaction succeeds

To this end: A job's return code, if 0, will be forcibly set as
-ECANCELED if that job has already concluded. Remove the now
redundant check in the backup job implementation.

We need to check for cancellation in both block_job_completed
AND block_job_completed_single, because jobs may be cancelled between
those two calls; for instance in transactions. This also necessitates
an ABORTING -> ABORTING transition to be allowed.

The check in block_job_completed could be removed, but there's no
point in starting to attempt to succeed a transaction that we know
in advance will fail.

This does NOT affect mirror jobs that are "canceled" during their
synchronous phase. The mirror job itself forcibly sets the canceled
property to false prior to ceding control, so such cases will invoke
the "commit" callback.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockjobs: add block_job_dismiss

For jobs that have reached their CONCLUDED state, prior to having their
last reference put down (meaning jobs that have completed successfully,
unsuccessfully, or have been canceled), allow the user to dismiss the
job's lingering status report via block-job-dismiss.

This gives management APIs the chance to conclusively determine if a job
failed or succeeded, even if the event broadcast was missed.

Note: block_job_do_dismiss and block_job_decommission happen to do
exactly the same thing, but they're called from different semantic
contexts, so both aliases are kept to improve readability.

Note 2: Don't worry about the 0x04 flag definition for AUTO_DISMISS, she
has a friend coming in a future patch to fill the hole where 0x02 is.

Verbs:
Dismiss: operates on CONCLUDED jobs only.
Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>