Current QMP's argument checker is more complex than it should be
and has (at least) one serious bug: it ignores unknown arguments.
To solve both problems we introduce a new argument checker. It's
added on top of the existing one, so that there are no regressions
during the transition.
This commit introduces the first part of the new checker, which
is run by qmp_check_client_args() and does the following:
1. Check if all mandatory arguments were provided
2. Set flags for argument validation
In order to do that, we transform the args_type string (from
qemu-montor.hx) into a qdict and iterate over it.
Next commit adds the new checker's second part: type checking and
invalid argument detection.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Historically, user monitor arguments beginning with '-' (eg. '-f')
were passed as integers down to handlers.
I've maintained this behavior in the new monitor because we didn't
have a boolean type at the very beginning of QMP. Today we have it
and this behavior is causing trouble to QMP's argument checker.
This commit fixes the problem by doing the following changes:
1. User Monitor
Before: the optional arg was represented as a QInt, we'd pass 1
down to handlers if the user specified the argument or
0 otherwise
This commit: the optional arg is represented as a QBool, we pass
true down to handlers if the user specified the
argument, otherwise _nothing_ is passed
2. QMP
Before: the client was required to pass the arg as QBool, but we'd
convert it to QInt internally. If the argument wasn't passed,
we'd pass 0 down
This commit: still require a QBool, but doesn't do any conversion and
doesn't pass any default value
3. Convert existing handlers (do_eject()/do_migrate()) to the new way
Before: Both handlers would expect a QInt value, either 0 or 1
This commit: Change the handlers to accept a QBool, they handle the
following cases:
A) true is passed: the option is enabled
B) false is passed: the option is disabled
C) nothing is passed: option not specified, use
default behavior
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
It's composed of functions qdict_first() and qdict_next(), plus
functions to access QDictEntry values.
This API was suggested by Markus Armbruster <armbru@redhat.com> and
it offers full control over the iteration process.
The usage is simple, the following example prints all keys in 'qdict'
(it's hopefully better than any English description):
QDict *qdict;
const QDictEntry *ent;
[...]
for (ent = qdict_first(qdict); ent; ent = qdict_next(qdict, ent)) {
printf("%s ", qdict_entry_key(ent));
}
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Next commit will introduce a new QDict iteration API which
returns QDictEntry entries, but we don't want users to directly
access its members since QDictEntry should be private to QDict.
In the near future this kind of data type will be turned into a
forward reference.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Let's call a 'hash' only what is returned by our hash function,
anything else is a 'bucket'.
This helps avoiding confusion with regard to how we traverse
our table.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
The 'by the guest' part is misleading, it could be disabled by
the host too.
We will likely need more surgery if we care for the distinction,
just dropping the problematic part is good enough for now.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
The current asynchronous command API doesn't return a QMP response
when the async command fails.
This is easy to reproduce with the balloon command (the sole async
command we have so far): run qemu w/o the '-balloon virtio' option
and try to issue the balloon command via QMP: no response will be
sent to the client.
This commit fixes the problem by making qmp_async_cmd_handler()
return the handler's error code and then calling
monitor_protocol_emitter() if the handler has returned an error.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
This fixes the following scenario using QMP.
First, put a bogus argument "foo" to "type", which results in an error.
{"execute": "netdev_add", "arguments": { "type": "foo", "id": "netdev1" } }
Then, call it again with correct argument "user".
{"execute": "netdev_add", "arguments": { "type": "user", "id": "netdev1" } }
This results in "DuplicatedId" error.
Because the first command was invalid, it should be able to reuse the
same "id", and the second command should work.
Reported-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Yoshiaki Tamura <tamura.yoshiaki@lab.ntt.co.jp>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Remove the arbitrary limitation of 1024 characters per return string and
read complete lines instead. Required for device_show.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
As sending "qmp_capabilities" on session start became mandatory, both
python examples were broken.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
As we want to add more flags to monitor commands, convert the only so
far existing one accordingly.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
We now have to move forward to the next argument type via next_arg_type.
This patch fixes completion for 'eject' and maybe also other commands.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Given too many arguments or an invalid command, we were leaking the
duplicated argument strings.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
scanf calls must not use PRI constants, they have probably the wrong size and
corrupt memory. We could replace them by SCN ones, but strtol is simpler than
scanf here anyway. While at it, also fix the parsers to reject garbage after
the number ("4096xyz" was accepted before).
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Richard Henderson <rth@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Commit 50e32ea8f3 changed the behaviour
for the return type of net_client_init() when a nic type with no init
method was specified. 'none' is one such nic type. Instead of returning
0, which gets interpreted as an index into the nd_table[] array, we
switched to returning -1, which signifies an error as well.
That broke VM start with '-net none'. Testing was only done with the
monitor command 'pci_add', which doesn't fail.
The correct fix would still be to return 0+ values from
net_client_init() only when the return value can be used as an index to
refer to an entry in nd_table[]. With the current code, callers can
erroneously poke into nd_table[0] when -net nic is used, which can lead
to badness.
However, this commit just returns to the previous behaviour before the
offending commit.
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
We currently only clear SVM_EVTINJ_VALID after successful interrupt
delivery. This apparently does not match real hardware which clears the
whole event_inj field on every vmexit, including unsuccessful interrupt
delivery.
Reported-by: Erik van der Kouwe <vdkouwe@cs.vu.nl>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The commit 8e65b7c049 introduced
expire_time of UHCIState. But expire_time is not in vmstate, the
second uhci_frame_timer will not be fired immediately after loadvm.
Signed-off-by: TeLeMan <geleman@gmail.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
For all i, ports_map[i] is used in and only in the i-th iteration.
Replace the dynamic array by a scalar variable.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Acked-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
This patch avoids handling write watchpoints on read-only memory access.
It also breaks the searching loop for watchpoint once the setup for
handling watchpoint later is done.
Signed-off-by: Jun Koi <junkoi2004@gmail.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
lsi_bad_phase has a bug in the choice of pmjad1/pmjad2. This does
not matter with Linux guests because it uses just one routine for
both, but it breaks Windows 64-bit guests. This is the text
from the spec:
"[The PMJCTL] bit controls which decision mechanism is used
when jumping on phase mismatch. When this bit is cleared the
LSI53C895A will use Phase Mismatch Jump Address 1 (PMJAD1) when
the WSR bit is cleared and Phase Mismatch Jump Address 2 (PMJAD2)
when the WSR bit is set. When this bit is set the LSI53C895A will
use jump address one (PMJAD1) on data out (data out, command,
message out) transfers and jump address two (PMJAD2) on data in
(data in, status, message in) transfers."
Which means:
CCNTL0.PMJCTL
0 SCNTL2.WSR = 0 PMJAD1
0 SCNTL2.WSR = 1 PMJAD2
1 out PMJAD1
1 in PMJAD2
In qemu, what you get instead is:
CCNTL0.PMJCTL
0 out PMJAD1
0 in PMJAD2 <<<<<
1 out PMJAD1
1 in PMJAD1 <<<<<
Considering that qemu always has SCNTL2.WSR cleared, the two marked cases
(corresponding to phase mismatch on input) are always jumping to the
wrong PMJAD register. The patch implements the correct semantics.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
