isa: Convert uses of isa_create() with Coccinelle

Replace

    dev = isa_create(bus, type_name);
    ...
    qdev_init_nofail(dev);

by

    dev = isa_new(type_name);
    ...
    isa_realize_and_unref(dev, bus, &error_fatal);

Recent commit "qdev: New qdev_new(), qdev_realize(), etc." explains
why.

Coccinelle script:

    @@
    expression dev, bus, expr;
    expression list args;
    expression d;
    @@
    -    dev = isa_create(bus, args);
    +    dev = isa_new(args);
    (
         d = &dev->qdev;
    |
         d = DEVICE(dev);
    )
         ... when != dev = expr
    -    qdev_init_nofail(d);
    +    isa_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression dev, bus, expr;
    expression list args;
    @@
    -    dev = isa_create(bus, args);
    +    dev = isa_new(args);
         ... when != dev = expr
    -    qdev_init_nofail(DEVICE(dev));
    +    isa_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression dev, bus, expr;
    expression list args;
    @@
    -    dev = DEVICE(isa_create(bus, args));
    +    ISADevice *isa_dev; // TODO move
    +    isa_dev = isa_new(args);
    +    dev = DEVICE(isa_dev);
         ... when != dev = expr
    -    qdev_init_nofail(dev);
    +    isa_realize_and_unref(isa_dev, bus, &error_fatal);

Missing #include "qapi/error.h" added manually, whitespace changes
minimized manually.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-20-armbru@redhat.com>

isa: New isa_new(), isa_realize_and_unref() etc.

I'm converting from qdev_create()/qdev_init_nofail() to
qdev_new()/qdev_realize_and_unref(); recent commit "qdev: New
qdev_new(), qdev_realize(), etc." explains why.

ISA devices use qdev_create() through isa_create() and
isa_try_create().

Provide isa_new(), isa_try_new(), and isa_realize_and_unref() for
converting ISA devices.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-19-armbru@redhat.com>

pci: pci_create(), pci_create_multifunction() are now unused, drop

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-18-armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>

pci: Convert uses of pci_create() etc. manually

Same transformation as in the previous commit.  Manual, because
convincing Coccinelle to transform these cases is not worthwhile.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-17-armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>

pci: Convert uses of pci_create() etc. with Coccinelle

Replace

    dev = pci_create(bus, type_name);
    ...
    qdev_init_nofail(dev);

by

    dev = pci_new(type_name);
    ...
    pci_realize_and_unref(dev, bus, &error_fatal);

and similarly for pci_create_multifunction().

Recent commit "qdev: New qdev_new(), qdev_realize(), etc." explains
why.

Coccinelle script:

    @@
    expression dev, bus, expr;
    expression list args;
    @@
    -    dev = pci_create(bus, args);
    +    dev = pci_new(args);
         ... when != dev = expr
    -    qdev_init_nofail(&dev->qdev);
    +    pci_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression dev, bus, expr;
    expression list args;
    expression d;
    @@
    -    dev = pci_create(bus, args);
    +    dev = pci_new(args);
    (
         d = &dev->qdev;
    |
         d = DEVICE(dev);
    )
         ... when != dev = expr
    -    qdev_init_nofail(d);
    +    pci_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression dev, bus, expr;
    expression list args;
    @@
    -    dev = pci_create(bus, args);
    +    dev = pci_new(args);
         ... when != dev = expr
    -    qdev_init_nofail(DEVICE(dev));
    +    pci_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression dev, bus, expr;
    expression list args;
    @@
    -    dev = DEVICE(pci_create(bus, args));
    +    PCIDevice *pci_dev; // TODO move
    +    pci_dev = pci_new(args);
    +    dev = DEVICE(pci_dev);
         ... when != dev = expr
    -    qdev_init_nofail(dev);
    +    pci_realize_and_unref(pci_dev, bus, &error_fatal);

    @@
    expression dev, bus, expr;
    expression list args;
    @@
    -    dev = pci_create_multifunction(bus, args);
    +    dev = pci_new_multifunction(args);
         ... when != dev = expr
    -    qdev_init_nofail(&dev->qdev);
    +    pci_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression bus, expr;
    expression list args;
    identifier dev;
    @@
    -    PCIDevice *dev = pci_create_multifunction(bus, args);
    +    PCIDevice *dev = pci_new_multifunction(args);
         ... when != dev = expr
    -    qdev_init_nofail(&dev->qdev);
    +    pci_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression dev, bus, expr;
    expression list args;
    @@
    -    dev = pci_create_multifunction(bus, args);
    +    dev = pci_new_multifunction(args);
         ... when != dev = expr
    -    qdev_init_nofail(DEVICE(dev));
    +    pci_realize_and_unref(dev, bus, &error_fatal);

Missing #include "qapi/error.h" added manually, whitespace changes
minimized manually, @pci_dev declarations moved manually.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-16-armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>

hw/ppc: Eliminate two superfluous QOM casts

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-15-armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>

pci: New pci_new(), pci_realize_and_unref() etc.

I'm converting from qdev_create()/qdev_init_nofail() to
qdev_new()/qdev_realize_and_unref(); recent commit "qdev: New
qdev_new(), qdev_realize(), etc." explains why.

PCI devices use qdev_create() through pci_create() and
pci_create_multifunction().

Provide pci_new(), pci_new_multifunction(), and
pci_realize_and_unref() for converting PCI devices.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-14-armbru@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>

qdev: Convert uses of qdev_set_parent_bus() manually

Same transformation as in the previous commit.  Manual, because
convincing Coccinelle to transform these cases is somewhere between
not worthwhile and infeasible (at least for me).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-13-armbru@redhat.com>

qdev: Convert uses of qdev_set_parent_bus() with Coccinelle

In addition to the qdev_create() patterns converted so far, we have a
qdev_set_parent_bus() pattern.  Mostly when we embed a device in a
parent device rather than allocating it on the heap.

This pattern also puts devices in the dangerous "no QOM parent, but
plugged into bus" state I explained in recent commit "qdev: New
qdev_new(), qdev_realize(), etc."

Apply same solution: convert to qdev_realize().  Coccinelle script:

    @@
    expression dev, bus, errp;
    symbol true;
    @@
    -    qdev_set_parent_bus(DEVICE(dev), bus);
         ...
    -    object_property_set_bool(OBJECT(dev), true, "realized", errp);
    +    qdev_realize(DEVICE(dev), bus, errp);

    @ depends on !(file in "qdev-monitor.c") && !(file in "hw/core/qdev.c")@
    expression dev, bus, errp;
    symbol true;
    @@
    -    qdev_set_parent_bus(dev, bus);
         ...
    -    object_property_set_bool(OBJECT(dev), true, "realized", errp);
    +    qdev_realize(dev, bus, errp);

    @@
    expression dev, bus;
    symbol true;
    @@
    -    qdev_set_parent_bus(DEVICE(dev), bus);
         ...
    -    qdev_init_nofail(DEVICE(dev));
    +    qdev_realize(DEVICE(dev), bus, &error_fatal);

Unconverted uses of qdev_set_parent_bus() remain.  They'll be
converted later in this series.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-12-armbru@redhat.com>
[Also convert new hw/virtio/vhost-user-vsock-pci.c]

qdev: Convert uses of qdev_create() manually

Same transformation as in the previous commit.  Manual, because
convincing Coccinelle to transform these cases is somewhere between
not worthwhile and infeasible (at least for me).

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-11-armbru@redhat.com>

qdev: Convert uses of qdev_create() with Coccinelle

This is the transformation explained in the commit before previous.
Takes care of just one pattern that needs conversion.  More to come in
this series.

Coccinelle script:

    @ depends on !(file in "hw/arm/highbank.c")@
    expression bus, type_name, dev, expr;
    @@
    -    dev = qdev_create(bus, type_name);
    +    dev = qdev_new(type_name);
         ... when != dev = expr
    -    qdev_init_nofail(dev);
    +    qdev_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression bus, type_name, dev, expr;
    identifier DOWN;
    @@
    -    dev = DOWN(qdev_create(bus, type_name));
    +    dev = DOWN(qdev_new(type_name));
         ... when != dev = expr
    -    qdev_init_nofail(DEVICE(dev));
    +    qdev_realize_and_unref(DEVICE(dev), bus, &error_fatal);

    @@
    expression bus, type_name, expr;
    identifier dev;
    @@
    -    DeviceState *dev = qdev_create(bus, type_name);
    +    DeviceState *dev = qdev_new(type_name);
         ... when != dev = expr
    -    qdev_init_nofail(dev);
    +    qdev_realize_and_unref(dev, bus, &error_fatal);

    @@
    expression bus, type_name, dev, expr, errp;
    symbol true;
    @@
    -    dev = qdev_create(bus, type_name);
    +    dev = qdev_new(type_name);
         ... when != dev = expr
    -    object_property_set_bool(OBJECT(dev), true, "realized", errp);
    +    qdev_realize_and_unref(dev, bus, errp);

    @@
    expression bus, type_name, expr, errp;
    identifier dev;
    symbol true;
    @@
    -    DeviceState *dev = qdev_create(bus, type_name);
    +    DeviceState *dev = qdev_new(type_name);
         ... when != dev = expr
    -    object_property_set_bool(OBJECT(dev), true, "realized", errp);
    +    qdev_realize_and_unref(dev, bus, errp);

The first rule exempts hw/arm/highbank.c, because it matches along two
control flow paths there, with different @type_name.  Covered by the
next commit's manual conversions.

Missing #include "qapi/error.h" added manually.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-10-armbru@redhat.com>
[Conflicts in hw/misc/empty_slot.c and hw/sparc/leon3.c resolved]

qdev: Convert to qdev_unrealize() manually

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-9-armbru@redhat.com>

qdev: Convert to qdev_unrealize() with Coccinelle

For readability, and consistency with qbus_realize().

Coccinelle script:

    @ depends on !(file in "hw/core/qdev.c")@
    typedef DeviceState;
    DeviceState *dev;
    symbol false, error_abort;
    @@
    -    object_property_set_bool(OBJECT(dev), false, "realized", &error_abort);
    +    qdev_unrealize(dev);

    @ depends on !(file in "hw/core/qdev.c") && !(file in "hw/core/bus.c")@
    expression dev;
    symbol false, error_abort;
    @@
    -    object_property_set_bool(OBJECT(dev), false, "realized", &error_abort);
    +    qdev_unrealize(DEVICE(dev));

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-8-armbru@redhat.com>

qdev: Convert to qbus_realize(), qbus_unrealize()

I'm going to convert device realization to qdev_realize() with the
help of Coccinelle.  Convert bus realization to qbus_realize() first,
to get it out of Coccinelle's way.  Readability improves.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-7-armbru@redhat.com>

qdev: Put qdev_new() to use with Coccinelle

Let's start simple and put qdev_new() to use.  Coccinelle script:

    @ depends on !(file in "hw/core/qdev.c")@
    expression type_name;
    @@
    -    DEVICE(object_new(type_name))
    +    qdev_new(type_name)

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-6-armbru@redhat.com>

qdev: New qdev_new(), qdev_realize(), etc.

We commonly plug devices into their bus right when we create them,
like this:

    dev = qdev_create(bus, type_name);

Note that @dev is a weak reference.  The reference from @bus to @dev
is the only strong one.

We realize at some later time, either with

    object_property_set_bool(OBJECT(dev), true, "realized", errp);

or its convenience wrapper

    qdev_init_nofail(dev);

If @dev still has no QOM parent then, realizing makes the
/machine/unattached/ orphanage its QOM parent.

Note that the device returned by qdev_create() is plugged into a bus,
but doesn't have a QOM parent, yet.  Until it acquires one,
unrealizing the bus will hang in bus_unparent():

    while ((kid = QTAILQ_FIRST(&bus->children)) != NULL) {
        DeviceState *dev = kid->child;
        object_unparent(OBJECT(dev));
    }

object_unparent() does nothing when its argument has no QOM parent,
and the loop spins forever.

Device state "no QOM parent, but plugged into bus" is dangerous.

Paolo suggested to delay plugging into the bus until realize.  We need
to plug into the parent bus before we call the device's realize
method, in case it uses the parent bus.  So the dangerous state still
exists, but only within realization, where we can manage it safely.

This commit creates infrastructure to do this:

    dev = qdev_new(type_name);
    ...
    qdev_realize_and_unref(dev, bus, errp)

Note that @dev becomes a strong reference here.
qdev_realize_and_unref() drops it.  There is also plain
qdev_realize(), which doesn't drop it.

The remainder of this series will convert all users to this new
interface.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Cc: Alistair Francis <alistair@alistair23.me>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Acked-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-5-armbru@redhat.com>

Revert "hw/versatile: realize the PCI root bus as part of the versatile init"

This reverts commit b1af7959a66610669e1a019b9a84f6ed3a7936c6.

Realizing a device automatically realizes its buses, in
device_set_realized().  Realizing them in realize methods is
redundant, unless the methods themselves require them to be realized
early.  pci_vpb_realize() doesn't.  Drop the redundant bus
realization.

Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-4-armbru@redhat.com>

Revert "hw/prep: realize the PCI root bus as part of the prep init"

This reverts commit 685f9a3428f625f580af0123aa95f4838d86cac3.

Realizing a device automatically realizes its buses, in
device_set_realized().  Realizing them in realize methods is
redundant, unless the methods themselves require them to be realized
early.  raven_pcihost_realizefn() doesn't.  Drop the redundant bus
realization.

Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-3-armbru@redhat.com>

qdev: Rename qbus_realize() to qbus_init()

qbus_realize() does not actually realize.  Rename it to qbus_init().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200610053247.1583243-2-armbru@redhat.com>

qdev: Assert onboard devices all get realized properly

This would have caught some of the bugs I just fixed.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200609122339.937862-25-armbru@redhat.com>

sd: Hide the qdev-but-not-quite thing created by sd_init()

Commit 260bc9d8aa "hw/sd/sd.c: QOMify" QOMified only the device
itself, not its users.  It kept sd_init() around for non-QOMified
users.

More than four years later, three such users remain: omap1 (machines
cheetah, sx1, sx1-v1) and omap2 (machines n800, n810) are not
QOMified, and pl181 (machines integratorcp, realview-eb,
realview-eb-mpcore, realview-pb-a8 realview-pbx-a9, versatileab,
versatilepb, vexpress-a15, vexpress-a9) is not QOMified properly.

The issue I presently have with this: an "sd-card" device should plug
into an "sd-bus" (its DeviceClass member bus_type says so), but
sd_init() leaves it unplugged.  This is normally a bug (I just fixed
some instances), and I'd like to assert proper pluggedness to prevent
regressions.  However, the qdev-but-not-quite thing returned by
sd_init() would fail the assertion.  Meh.

Make sd_init() hide it from QOM/qdev.  Visible in "info qom-tree",
here's the change for cheetah:

     /machine (cheetah-machine)
       [...]
       /unattached (container)
         [...]
         /device[5] (serial-mm)
           /serial (serial)
           /serial[0] (qemu:memory-region)
    -    /device[6] (sd-card)
    -    /device[7] (omap-gpio)
    +    /device[6] (omap-gpio)
         [rest of device[*] renumbered...]

Cc: "Philippe Mathieu-Daudé" <philmd@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200609122339.937862-24-armbru@redhat.com>

qdev: Assert devices are plugged into a bus that can take them

This would have caught some of the bugs I just fixed.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-Id: <20200609122339.937862-23-armbru@redhat.com>

sparc/leon3: Fix to put grlib,* devices on sysbus

leon3_generic_hw_init() creates a "grlib,ahbpnp" and a "grlib,apbpnp"
sysbus device in a way that leaves them unplugged.

Create them the common way that puts them into the main system bus.
Affects machine leon3_generic.  Visible in "info qtree":

     bus: main-system-bus
       type System
    +  dev: grlib,ahbpnp, id ""
    +    mmio 00000000fffff000/0000000000001000
    +  dev: grlib,apbpnp, id ""
    +    mmio 00000000800ff000/0000000000001000
       dev: grlib,irqmp, id ""

Cc: Fabien Chouteau <chouteau@adacore.com>
Cc: KONRAD Frederic <frederic.konrad@adacore.com>
Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Cc: Artyom Tarasenko <atar4qemu@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: KONRAD Frederic <frederic.konrad@adacore.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200609122339.937862-22-armbru@redhat.com>
Acked-by: Artyom Tarasenko <atar4qemu@gmail.com>

riscv: Fix type of SiFive[EU]SocState, member parent_obj

Device "riscv.sifive.e.soc" is a direct subtype of TYPE_DEVICE, but
its instance struct SiFiveESoCState's member @parent_obj is
SysBusDevice instead of DeviceState.  Correct that.

Same for "riscv.sifive.u.soc"'s instance struct SiFiveUSoCState.

Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Alistair Francis <Alistair.Francis@wdc.com>
Cc: Sagar Karandikar <sagark@eecs.berkeley.edu>
Cc: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Cc: qemu-riscv@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20200609122339.937862-21-armbru@redhat.com>

riscv: Fix to put "riscv.hart_array" devices on sysbus

riscv_sifive_e_soc_init(), riscv_sifive_u_soc_init(),
spike_board_init(), spike_v1_10_0_board_init(),
spike_v1_09_1_board_init(), and riscv_virt_board_init() create
"riscv-hart_array" sysbus devices in a way that leaves them unplugged.

Create them the common way that puts them into the main system bus.
Affects machines sifive_e, sifive_u, spike, spike_v1.10, spike_v1.9.1,
and virt.  Visible in "info qtree", here's the change for sifive_e:

     bus: main-system-bus
       type System
    +  dev: riscv.hart_array, id ""
    +    num-harts = 1 (0x1)
    +    hartid-base = 0 (0x0)
    +    cpu-type = "sifive-e31-riscv-cpu"
       dev: sifive_soc.gpio, id ""

Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Alistair Francis <Alistair.Francis@wdc.com>
Cc: Sagar Karandikar <sagark@eecs.berkeley.edu>
Cc: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Cc: qemu-riscv@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20200609122339.937862-20-armbru@redhat.com>

display/sm501 display/ati: Fix to realize "i2c-ddc"

sm501_init() and ati_vga_realize() create an "i2c-ddc" device, but
neglect to realize it.  Affects machines sam460ex, shix, r2d, and
fulong2e.

In theory, a device becomes real only on realize.  In practice, the
transition from unreal to real is a fuzzy one.  The work to make a
device real can be spread between realize methods (fine),
instance_init methods (wrong), and board code wiring up the device
(fine as long as it effectively happens on realize).  Depending on
what exactly is done where, a device can work even when we neglect
to realize it.

This one appears to work.  Nevertheless, it's a clear misuse of the
interface.  Even when it works today (more or less by chance), it can
break tomorrow.

Fix by realizing it right away.  Visible in "info qom-tree"; here's
the change for sam460ex:

     /machine (sam460ex-machine)
       [...]
       /unattached (container)
         [...]
    -    /device[14] (sii3112)
    +    /device[14] (i2c-ddc)
    +    /device[15] (sii3112)
         [rest of device[*] renumbered...]

Fixes: 4a1f253adb45ac6019971193d5077c4d5d55886a
Fixes: c82c7336de58876862e6b4dccbda29e9240fd388
Cc: BALATON Zoltan <balaton@eik.bme.hu>
Cc: qemu-ppc@nongnu.org
Cc: Magnus Damm <magnus.damm@gmail.com>
Cc: Philippe Mathieu-Daudé <f4bug@amsat.org>
Cc: Aleksandar Markovic <aleksandar.qemu.devel@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Tested-by: BALATON Zoltan <balaton@eik.bme.hu>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200609122339.937862-19-armbru@redhat.com>

pnv/psi: Correct the pnv-psi* devices not to be sysbus devices

pnv_chip_power8_instance_init() creates a "pnv-psi-POWER8" sysbus
device in a way that leaves it unplugged.
pnv_chip_power9_instance_init() and pnv_chip_power10_instance_init()
do the same for "pnv-psi-POWER9" and "pnv-psi-POWER10", respectively.

These devices aren't actually sysbus devices.  Correct that.

Cc: "Cédric Le Goater" <clg@kaod.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20200609122339.937862-18-armbru@redhat.com>

ppc/pnv: Put "*-pnv-chip" and "pnv-xive" on the main system bus

pnv_init() creates "power10_v1.0-pnv-chip", "power8_v2.0-pnv-chip",
"power8e_v2.1-pnv-chip", "power8nvl_v1.0-pnv-chip", or
"power9_v2.0-pnv-chip" sysbus devices in a way that leaves them
unplugged.

pnv_chip_power9_instance_init() creates a "pnv-xive" sysbus device in
a way that leaves it unplugged.

Create them the common way that puts them into the main system bus.
Affects machines powernv8, powernv9, and powernv10.  Visible in "info
qtree".  Here's the change for powernv9:

     bus: main-system-bus
       type System
    +  dev: power9_v2.0-pnv-chip, id ""
    +    chip-id = 0 (0x0)
    +    ram-start = 0 (0x0)
    +    ram-size = 1879048192 (0x70000000)
    +    nr-cores = 1 (0x1)
    +    cores-mask = 72057594037927935 (0xffffffffffffff)
    +    nr-threads = 1 (0x1)
    +    num-phbs = 6 (0x6)
    +    mmio 000603fc00000000/0000000400000000
    [...]
    +  dev: pnv-xive, id ""
    +    ic-bar = 1692157036462080 (0x6030203100000)
    +    vc-bar = 1689949371891712 (0x6010000000000)
    +    pc-bar = 1690499127705600 (0x6018000000000)
    +    tm-bar = 1692157036986368 (0x6030203180000)

Cc: "Cédric Le Goater" <clg@kaod.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20200609122339.937862-17-armbru@redhat.com>

macio: Fix macio-bus to be a subtype of System bus

The devices we plug into the macio-bus are all sysbus devices
(DeviceClass member bus_type is TYPE_SYSTEM_BUS), but macio-bus does
not derive from TYPE_SYSTEM_BUS.  Fix that.

"info qtree" now shows the devices' mmio ranges, as it should

Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200609122339.937862-16-armbru@redhat.com>

macio: Put "macio-nvram" device on the macio bus

macio_oldworld_init() creates a "macio-nvram", sysbus device, but
neglects to but it on a bus.

Put it on the macio bus.  Affects machine g3beige.  Visible in "info
qtree":

             bus: macio.0
               type macio-bus
               [...]
    +          dev: macio-nvram, id ""
    +            size = 8192 (0x2000)
    +            it_shift = 4 (0x4)

This also makes it a QOM child of macio-oldworld.  Visible in "info
qom-tree":

     /machine (g3beige-machine)
       [...]
       /unattached (container)
         [...]
         /device[6] (macio-oldworld)
           [...]
    -    /device[7] (macio-nvram)
    -      /macio-nvram[0] (qemu:memory-region)
    +      /nvram (macio-nvram)
    +        /macio-nvram[0] (qemu:memory-region)
         [rest of device[*] renumbered...]

Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200609122339.937862-15-armbru@redhat.com>

ppc4xx: Drop redundant device realization

object_property_set_bool(OBJECT(dev), true, "realized", ...) right
after qdev_init_nofail(dev) does nothing, because qdev_init_nofail()
already realizes.  Drop.

Cc: BALATON Zoltan <balaton@eik.bme.hu>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: BALATON Zoltan <balaton@eik.bme.hu>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20200609122339.937862-14-armbru@redhat.com>

MAINTAINERS: Make section PowerNV cover pci-host/pnv* as well

Cc: Cédric Le Goater <clg@kaod.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <20200609122339.937862-13-armbru@redhat.com>

pnv/phb4: Delete unused "pnv-phb4-pec-stack" devices

The number of stacks is controlled by property "num-stacks".
pnv_pec_instance_init() creates the maximum supported number, because
the property has not been set then.  pnv_pec_realize() realizes only
the wanted number.  Works, although it can leave unrealized devices
hanging around in the QOM composition tree.  Affects machine powernv9.

Delete the unused devices by making pnv_pec_realize() unparent them.
Visible in "info qom-tree":

     /machine (powernv9-machine)
       /chip[0] (power9_v2.0-pnv-chip)
         [...]
         /pec[0] (pnv-phb4-pec)
           /stack[0] (pnv-phb4-pec-stack)
             [...]
    -      /stack[1] (pnv-phb4-pec-stack)
    -        /phb (pnv-phb4)
    -          /pcie-mmcfg-mmio[0] (qemu:memory-region)
    -          /root (pnv-phb4-root-port)
    -          /source (xive-source)
    -      /stack[2] (pnv-phb4-pec-stack)
    -        /phb (pnv-phb4)
    -          /pcie-mmcfg-mmio[0] (qemu:memory-region)
    -          /root (pnv-phb4-root-port)
    -          /source (xive-source)
           /xscom-pec-0.0-nest[0] (qemu:memory-region)
           /xscom-pec-0.0-pci[0] (qemu:memory-region)
         /pec[1] (pnv-phb4-pec)
           /stack[0] (pnv-phb4-pec-stack)
             [...]
           /stack[1] (pnv-phb4-pec-stack)
             [...]
    -      /stack[2] (pnv-phb4-pec-stack)
    -        /phb (pnv-phb4)
    -          /pcie-mmcfg-mmio[0] (qemu:memory-region)
    -          /root (pnv-phb4-root-port)
    -          /source (xive-source)
           /xscom-pec-0.1-nest[0] (qemu:memory-region)
           /xscom-pec-0.1-pci[0] (qemu:memory-region)
         /pec[2] (pnv-phb4-pec)
           /stack[0] (pnv-phb4-pec-stack)
             [...]
           /stack[1] (pnv-phb4-pec-stack)
             [...]
           /stack[2] (pnv-phb4-pec-stack)
             [...]

Cc: Cédric Le Goater <clg@kaod.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20200609122339.937862-12-armbru@redhat.com>

macio: Delete unused "macio-gpio" devices

These devices go with the "via-pmu" device, which is controlled by
property "has-pmu".  macio_newworld_init() creates it unconditionally,
because the property has not been set then.  macio_newworld_realize()
realizes it only when the property is true.  Works, although it can
leave an unrealized device hanging around in the QOM composition tree.
Affects machine mac99 with via=cuda (default).

Delete the unused device by making macio_newworld_realize() unparent
it.  Visible in "info qom-tree":

     /machine (mac99-machine)
       [...]
       /unattached (container)
         /device[9] (macio-newworld)
           [...]
           /escc-legacy-port[8] (qemu:memory-region)
           /escc-legacy-port[9] (qemu:memory-region)
           /escc-legacy[0] (qemu:memory-region)
    -      /gpio (macio-gpio)
    -        /gpio[0] (qemu:memory-region)
           /ide[0] (macio-ide)
             /ide.0 (IDE)
             /pmac-ide[0] (qemu:memory-region)

Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-Id: <20200609122339.937862-11-armbru@redhat.com>

macio: Fix to realize "mos6522-cuda" and "mos6522-pmu" devices

cuda_init() creates a "mos6522-cuda" device, but it's never realized.
Affects machines mac99 with via=cuda (default) and g3beige.

pmu_init() creates a "mos6522-pmu" device, but it's never realized.
Affects machine mac99 with via=pmu and via=pmu-adb,

In theory, a device becomes real only on realize.  In practice, the
transition from unreal to real is a fuzzy one.  The work to make a
device real can be spread between realize methods (fine),
instance_init methods (wrong), and board code wiring up the device
(fine as long as it effectively happens on realize).  Depending on
what exactly is done where, a device can work even when we neglect
to realize it.

These two appear to work.  Nevertheless, it's a clear misuse of the
interface.  Even when it works today (more or less by chance), it can
break tomorrow.

Fix by realizing them in cuda_realize() and pmu_realize(),
respectively.

Fixes: 6dca62a0000f95e0b7020aa00d0ca9b2c421f341
Cc: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200609122339.937862-10-armbru@redhat.com>

mac_via: Fix to realize "mos6522-q800-via*" devices

mac_via_realize() creates a "mos6522-q800-via1" and a
"mos6522-q800-via2" device, but neglects to realize them.  Affects
machine q800.

In theory, a device becomes real only on realize.  In practice, the
transition from unreal to real is a fuzzy one.  The work to make a
device real can be spread between realize methods (fine),
instance_init methods (wrong), and board code wiring up the device
(fine as long as it effectively happens on realize).  Depending on
what exactly is done where, a device can work even when we neglect
to realize it.

These two appear to work.  Nevertheless, it's a clear misuse of the
interface.  Even when it works today (more or less by chance), it can
break tomorrow.

Fix by realizing them right away.

Fixes: 6dca62a0000f95e0b7020aa00d0ca9b2c421f341
Cc: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-Id: <20200609122339.937862-9-armbru@redhat.com>
Tested-by: Laurent Vivier <laurent@vivier.eu>
Acked-by: Laurent Vivier <laurent@vivier.eu>

auxbus: Fix aux-to-i2c-bridge to be a subtype of aux-slave

We plug aux-to-i2c-bridge into the aux-bus, even though its
DeviceClass member bus_type is null, not TYPE_AUX_BUS.  Fix that by
deriving it from TYPE_AUX_SLAVE instead of TYPE_DEVICE.

Cc: KONRAD Frederic <fred.konrad@greensocs.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200609122339.937862-8-armbru@redhat.com>

armv7m: Delete unused "ARM,bitband-memory" devices

These devices are optional, and enabled by property "enable-bitband".
armv7m_instance_init() creates them unconditionally, because the
property has not been set then.  armv7m_realize() realizes them only
when the property is true.  Works, although it leaves unrealized
devices hanging around in the QOM composition tree.  Affects machines
microbit, mps2-an505, mps2-an521, musca-a, and musca-b1.

Delete the unused devices by making armv7m_realize() unparent them.
Visible in "info qom-tree"; here's the change for microbit:

     /machine (microbit-machine)
       /microbit.twi (microbit.i2c)
         /microbit.twi[0] (qemu:memory-region)
       /nrf51 (nrf51-soc)
         /armv6m (armv7m)
           /armv7m-container[0] (qemu:memory-region)
    -      /bitband[0] (ARM,bitband-memory)
    -        /bitband[0] (qemu:memory-region)
    -      /bitband[1] (ARM,bitband-memory)
    -        /bitband[0] (qemu:memory-region)
           /cpu (cortex-m0-arm-cpu)

Cc: Peter Maydell <peter.maydell@linaro.org>
Cc: qemu-arm@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20200609122339.937862-7-armbru@redhat.com>

arm/aspeed: Rework NIC attachment

The number of MACs supported by an Aspeed SoC is defined by "macs_num"
under the SoC model, that is two for the AST2400 and AST2500 and four
for the AST2600. The model initializes the maximum number of supported
MACs but the number of realized devices is capped by the number of
network device back-ends defined on the command line. This can leave
unrealized devices hanging around in the QOM composition tree.

To get virtual hardware that matches the physical hardware, you have
to pass exactly as many -nic options as there are MACs, and some of
them must be -nic none:

* Machines ast2500-evb, palmetto-bmc, romulus-bmc, sonorapass-bmc,
  swift-bmc, and witherspoon-bmc: two -nic, and the second one must be
  -nic none.

* Machine ast2600-evb: four -nic, the first one must be -nic none.

* Machine tacoma-bmc: four nic, the first two and the last one must be
  -nic none.

Modify the machine initialization to define which MACs are attached to
a network device back-end using a bit-field property "macs-mask" and
let the SoC realize all network devices.

The default setting of "macs-mask" is "use MAC0" only, which works for
all our AST2400 and AST2500 machines. The AST2600 machines have
different configurations. The AST2600 EVB machine activates MAC1, MAC2
and MAC3 and the Tacoma BMC machine activates MAC2.

Incompatible CLI change: -nic options now apply to *active* MACs:
MAC1, MAC2, MAC3 for ast2600-evb, MAC2 for tacoma-bmc, and MAC0 for
all the others.

The machines now always get all MACs as they should. Visible in "info
qom-tree", here's the change for tacoma-bmc:

     /machine (tacoma-bmc-machine)
       /peripheral (container)
       /peripheral-anon (container)
       /soc (ast2600-a1)
         [...]
         /ftgmac100[0] (ftgmac100)
           /ftgmac100[0] (qemu:memory-region)
         /ftgmac100[1] (ftgmac100)
    +      /ftgmac100[0] (qemu:memory-region)
         /ftgmac100[2] (ftgmac100)
    +      /ftgmac100[0] (qemu:memory-region)
         /ftgmac100[3] (ftgmac100)
    +      /ftgmac100[0] (qemu:memory-region)
         [...]
         /mii[0] (aspeed-mmi)
           /aspeed-mmi[0] (qemu:memory-region)
         /mii[1] (aspeed-mmi)
    +      /aspeed-mmi[0] (qemu:memory-region)
         /mii[2] (aspeed-mmi)
    +      /aspeed-mmi[0] (qemu:memory-region)
         /mii[3] (aspeed-mmi)
    +      /aspeed-mmi[0] (qemu:memory-region)

Also visible in "info qtree"; here's the change for tacoma-bmc:

       dev: ftgmac100, id ""
         gpio-out "sysbus-irq" 1
         aspeed = true
    -    mac = "52:54:00:12:34:56"
    -    netdev = "hub0port0"
    +    mac = "52:54:00:12:34:57"
    +    netdev = ""
         mmio 000000001e660000/0000000000002000
       dev: ftgmac100, id ""
    -    aspeed = false
    -    mac = "00:00:00:00:00:00"
    +    gpio-out "sysbus-irq" 1
    +    aspeed = true
    +    mac = "52:54:00:12:34:58"
         netdev = ""
    +    mmio 000000001e680000/0000000000002000
       dev: ftgmac100, id ""
    -    aspeed = false
    -    mac = "00:00:00:00:00:00"
    -    netdev = ""
    +    gpio-out "sysbus-irq" 1
    +    aspeed = true
    +    mac = "52:54:00:12:34:56"
    +    netdev = "hub0port0"
    +    mmio 000000001e670000/0000000000002000
       dev: ftgmac100, id ""
    -    aspeed = false
    -    mac = "00:00:00:00:00:00"
    +    gpio-out "sysbus-irq" 1
    +    aspeed = true
    +    mac = "52:54:00:12:34:59"
         netdev = ""
    +    mmio 000000001e690000/0000000000002000
       [...]
       dev: aspeed-mmi, id ""
         mmio 000000001e650000/0000000000000008
       dev: aspeed-mmi, id ""
    +    mmio 000000001e650008/0000000000000008
       dev: aspeed-mmi, id ""
    +    mmio 000000001e650010/0000000000000008
       dev: aspeed-mmi, id ""
    +    mmio 000000001e650018/0000000000000008

Inactive MACs will have no peer and QEMU may warn the user with :

    qemu-system-arm: warning: nic ftgmac100.0 has no peer
    qemu-system-arm: warning: nic ftgmac100.1 has no peer
    qemu-system-arm: warning: nic ftgmac100.3 has no peer

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
[Commit message expanded]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200609122339.937862-6-armbru@redhat.com>

arm/aspeed: Compute the number of CPUs from the SoC definition

Commit ece09beec457 ("aspeed: introduce a configurable number of CPU
per machine") was a convient change during bringup but the Aspeed SoCs
have a fixed number of CPUs : one for the AST2400 and AST2500, and two
for the AST2600.

When the number of CPUs configured with -smp is less than the SoC's
fixed number, the "unconfigured" CPUs are left unrealized. This can
happen for machines ast2600-evb and tacoma-bmc, where the SoC's fixed
number is 2. To get virtual hardware that matches the physical
hardware, you have to pass -smp cpus=2 (or its sugared form -smp 2).

We normally reject -smp cpus=N when N exceeds the machine's limit.
Except we ignore cpus=2 (and only cpus=2) with a warning for machines
ast2500-evb, palmetto-bmc, romulus-bmc, sonorapass-bmc, swift-bmc, and
witherspoon-bmc.

Remove the "num-cpu" property from the SoC state and use the fixed
number of CPUs defined in the SoC class instead. Compute the default,
min, max number of CPUs of the machine directly from the SoC class
definition.

Machines ast2600-evb and tacoma-bmc now always get their second CPU as
they should. Visible in "info qom-tree"; here's the change for
ast2600-evb:

     /machine (ast2600-evb-machine)
       /peripheral (container)
       /peripheral-anon (container)
       /soc (ast2600-a1)
         /a7mpcore (a15mpcore_priv)
           /a15mp-priv-container[0] (qemu:memory-region)
           /gic (arm_gic)
             /gic_cpu[0] (qemu:memory-region)
             /gic_cpu[1] (qemu:memory-region)
    +        /gic_cpu[2] (qemu:memory-region)
             /gic_dist[0] (qemu:memory-region)
             /gic_vcpu[0] (qemu:memory-region)
             /gic_viface[0] (qemu:memory-region)
             /gic_viface[1] (qemu:memory-region)
    +        /gic_viface[2] (qemu:memory-region)
             /unnamed-gpio-in[0] (irq)
             [...]
    +        /unnamed-gpio-in[160] (irq)
             [same for 161 to 190...]
    +        /unnamed-gpio-in[191] (irq)

Also visible in "info qtree"; here's the change for ast2600-evb:

     bus: main-system-bus
       type System
       dev: a15mpcore_priv, id ""
         gpio-in "" 128
    -    gpio-out "sysbus-irq" 5
    -    num-cpu = 1 (0x1)
    +    gpio-out "sysbus-irq" 10
    +    num-cpu = 2 (0x2)
         num-irq = 160 (0xa0)
         mmio 0000000040460000/0000000000008000
       dev: arm_gic, id ""
    -    gpio-in "" 160
    -    num-cpu = 1 (0x1)
    +    gpio-in "" 192
    +    num-cpu = 2 (0x2)
         num-irq = 160 (0xa0)
         revision = 2 (0x2)
         has-security-extensions = true
         has-virtualization-extensions = true
         num-priority-bits = 8 (0x8)
         mmio ffffffffffffffff/0000000000001000
         mmio ffffffffffffffff/0000000000002000
         mmio ffffffffffffffff/0000000000001000
         mmio ffffffffffffffff/0000000000002000
         mmio ffffffffffffffff/0000000000000100
    +    mmio ffffffffffffffff/0000000000000100
    +    mmio ffffffffffffffff/0000000000000200
         mmio ffffffffffffffff/0000000000000200

The other machines now reject -smp cpus=2 just like -smp cpus=3 and up.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
[Commit message expanded]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200609122339.937862-5-armbru@redhat.com>

sd/pxa2xx_mmci: Fix to realize "pxa2xx-mmci" device

pxa2xx_mmci_init() creates a "pxa2xx-mmci" device, but neglects to
realize it.  Affects machines akita, borzoi, connex, mainstone, spitz,
terrier, tosa, verdex, and z2.

In theory, a device becomes real only on realize.  In practice, the
transition from unreal to real is a fuzzy one.  The work to make a
device real can be spread between realize methods (fine),
instance_init methods (wrong), and board code wiring up the device
(fine as long as it effectively happens on realize).  Depending on
what exactly is done where, a device can work even when we neglect
to realize it.

This one appears to work.  Nevertheless, it's a clear misuse of the
interface.  Even when it works today (more or less by chance), it can
break tomorrow.

Fix by realizing it right away.  Visible in "info qom-tree"; here's
the change for akita:

     /machine (akita-machine)
       [...]
       /unattached (container)
         [...]
    +    /device[5] (pxa2xx-mmci)
    +      /pxa2xx-mmci[0] (qemu:memory-region)
    +      /sd-bus (pxa2xx-mmci-bus)
         [rest of device[*] renumbered...]

Fixes: 7a9468c92517e19037bfe2272f64f5dadaf9db15
Cc: Andrzej Zaborowski <balrogg@gmail.com>
Cc: Peter Maydell <peter.maydell@linaro.org>
Cc: qemu-arm@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200609122339.937862-4-armbru@redhat.com>

display/xlnx_dp: Fix to realize "i2c-ddc" and "aux-to-i2c-bridge"

xlnx_dp_init() creates these two devices, but they're never realized.
Affects machine xlnx-zcu102.

In theory, a device becomes real only on realize.  In practice, the
transition from unreal to real is a fuzzy one.  The work to make a
device real can be spread between realize methods (fine),
instance_init methods (wrong), and board code wiring up the device
(fine as long as it effectively happens on realize).  Depending on
what exactly is done where, a device can work even when we neglect to
realize it.

These two appear to work.  Nevertheless, it's a clear misuse of the
interface.  Even when it works today (more or less by chance), it can
break tomorrow.

Fix by realizing them in xlnx_dp_realize().

Fixes: 58ac482a66de09a7590f705e53fc6a3fb8a055e8
Cc: KONRAD Frederic <fred.konrad@greensocs.com>
Cc: Alistair Francis <alistair@alistair23.me>
Cc: "Edgar E. Iglesias" <edgar.iglesias@gmail.com>
Cc: Peter Maydell <peter.maydell@linaro.org>
Cc: qemu-arm@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20200609122339.937862-3-armbru@redhat.com>

arm/stm32f405: Fix realization of "stm32f2xx-adc" devices

stm32f405_soc_initfn() creates six such devices, but
stm32f405_soc_realize() realizes only one.  Affects machine
netduinoplus2.

In theory, a device becomes real only on realize.  In practice, the
transition from unreal to real is a fuzzy one.  The work to make a
device real can be spread between realize methods (fine),
instance_init methods (wrong), and board code wiring up the device
(fine as long as it effectively happens on realize).  Depending on
what exactly is done where, a device can work even when we neglect
to realize it.

The five unrealized devices appear to stay unreal: neither MMIO nor
IRQ get wired up.

Fix stm32f405_soc_realize() to realize and wire up all six.  Visible
in "info qtree":

     bus: main-system-bus
       type System
       dev: stm32f405-soc, id ""
         cpu-type = "cortex-m4-arm-cpu"
       dev: stm32f2xx-adc, id ""
         gpio-out "sysbus-irq" 1
    -    mmio ffffffffffffffff/00000000000000ff
    +    mmio 0000000040012000/00000000000000ff
       dev: stm32f2xx-adc, id ""
         gpio-out "sysbus-irq" 1
    -    mmio ffffffffffffffff/00000000000000ff
    +    mmio 0000000040012100/00000000000000ff
       dev: stm32f2xx-adc, id ""
         gpio-out "sysbus-irq" 1
    -    mmio ffffffffffffffff/00000000000000ff
    +    mmio 0000000040012200/00000000000000ff
       dev: stm32f2xx-adc, id ""
         gpio-out "sysbus-irq" 1
    -    mmio ffffffffffffffff/00000000000000ff
    +    mmio 0000000040012300/00000000000000ff
       dev: stm32f2xx-adc, id ""
         gpio-out "sysbus-irq" 1
    -    mmio 0000000040012000/00000000000000ff
    +    mmio 0000000040012400/00000000000000ff
       dev: stm32f2xx-adc, id ""
         gpio-out "sysbus-irq" 1
    -    mmio ffffffffffffffff/00000000000000ff
    +    mmio 0000000040012500/00000000000000ff
       dev: armv7m, id ""

Fixes: 529fc5fd3e18ace8f739afd02dc0953354f39442
Cc: Alistair Francis <alistair@alistair23.me>
Cc: Peter Maydell <peter.maydell@linaro.org>
Cc: qemu-arm@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200609122339.937862-2-armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>

qom: Make "info qom-tree" show children sorted

"info qom-tree" prints children in unstable order.  This is a pain
when diffing output for different versions to find change.  Print it
sorted.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200527084754.7531-3-armbru@redhat.com>
Tested-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>

qom: Constify object_get_canonical_path{,_component}()'s parameter

Suggested-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200527084754.7531-2-armbru@redhat.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>

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

* Miscellaneous fixes and feature enablement (many)
* SEV refactoring (David)
* Hyper-V initial support (Jon)
* i386 TCG fixes (x87 and SSE, Joseph)
* vmport cleanup and improvements (Philippe, Liran)
* Use-after-free with vCPU hot-unplug (Nengyuan)
* run-coverity-scan improvements (myself)
* Record/replay fixes (Pavel)
* -machine kernel_irqchip=split improvements for INTx (Peter)
* Code cleanups (Philippe)
* Crash and security fixes (PJP)
* HVF cleanups (Roman)

# gpg: Signature made Fri 12 Jun 2020 16:57:04 BST
# gpg:                using RSA key F13338574B662389866C7682BFFBD25F78C7AE83
# gpg:                issuer "pbonzini@redhat.com"
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>" [full]
# gpg:                 aka "Paolo Bonzini <pbonzini@redhat.com>" [full]
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4  E2F7 7E15 100C CD36 69B1
#      Subkey fingerprint: F133 3857 4B66 2389 866C  7682 BFFB D25F 78C7 AE83

* remotes/bonzini/tags/for-upstream: (116 commits)
  target/i386: Remove obsolete TODO file
  stubs: move Xen stubs to accel/
  replay: fix replay shutdown for console mode
  exec/cpu-common: Move MUSB specific typedefs to 'hw/usb/hcd-musb.h'
  hw/usb: Move device-specific declarations to new 'hcd-musb.h' header
  exec/memory: Remove unused MemoryRegionMmio type
  checkpatch: reversed logic with acpi test checks
  target/i386: sev: Unify SEVState and SevGuestState
  target/i386: sev: Remove redundant handle field
  target/i386: sev: Remove redundant policy field
  target/i386: sev: Remove redundant cbitpos and reduced_phys_bits fields
  target/i386: sev: Partial cleanup to sev_state global
  target/i386: sev: Embed SEVState in SevGuestState
  target/i386: sev: Rename QSevGuestInfo
  target/i386: sev: Move local structure definitions into .c file
  target/i386: sev: Remove unused QSevGuestInfoClass
  xen: fix build without pci passthrough
  i386: hvf: Drop HVFX86EmulatorState
  i386: hvf: Move mmio_buf into CPUX86State
  i386: hvf: Move lazy_flags into CPUX86State
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
# Conflicts:
# hw/i386/acpi-build.c

target/i386: Remove obsolete TODO file

The last real change to this file is from 2012, so it is very likely
that this file is completely out-of-date and ignored today. Let's
simply remove it to avoid confusion if someone finds it by accident.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20200611172445.5177-1-thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

stubs: move Xen stubs to accel/

Keep them close to the other accelerator-dependent stubs, so as to remove
stubs that are not needed by tools.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

replay: fix replay shutdown for console mode

When QEMU is used without any graphical window,
QEMU execution is terminated with the signal (e.g., Ctrl-C).
Signal processing in QEMU does not include
qemu_system_shutdown_request call. That is why shutdown
event is not recorded by record/replay in this case.
This patch adds shutdown event to the end of the record log.
Now every replay will shutdown the machine at the end.

Signed-off-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
Message-Id: <159012995470.27967.18129611453659045726.stgit@pasha-ThinkPad-X280>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

exec/cpu-common: Move MUSB specific typedefs to 'hw/usb/hcd-musb.h'

The CPUReadMemoryFunc/CPUWriteMemoryFunc typedefs are legacy
remnant from before the conversion to MemoryRegions.
Since they are now only used in tusb6010.c and hcd-musb.c,
move them to "hw/usb/musb.h" and rename them appropriately.

Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200601141536.15192-4-f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

hw/usb: Move device-specific declarations to new 'hcd-musb.h' header

Move the declarations for the MUSB-HDRC USB2.0 OTG compliant core
into a separate header.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200601141536.15192-3-f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

exec/memory: Remove unused MemoryRegionMmio type

Since commit 62a0db942dec ('memory: Remove old_mmio accessors')
this structure is unused. Remove it.

Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200601141536.15192-2-f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

checkpatch: reversed logic with acpi test checks

Logic reversed: allowed list should just be ignored. Instead we
only take that into account :(

Fixes: e11b06a880ca ("checkpatch: ignore allowed diff list")
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20200602053614.54745-1-mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Unify SEVState and SevGuestState

SEVState is contained with SevGuestState.  We've now fixed redundancies
and name conflicts, so there's no real point to the nested structure.  Just
move all the fields of SEVState into SevGuestState.

This eliminates the SEVState structure, which as a bonus removes the
confusion with the SevState enum.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200604064219.436242-10-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Remove redundant handle field

The user can explicitly specify a handle via the "handle" property wired
to SevGuestState::handle.  That gets passed to the KVM_SEV_LAUNCH_START
ioctl() which may update it, the final value being copied back to both
SevGuestState::handle and SEVState::handle.

AFAICT, nothing will be looking SEVState::handle before it and
SevGuestState::handle have been updated from the ioctl().  So, remove the
field and just use SevGuestState::handle directly.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200604064219.436242-9-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Remove redundant policy field

SEVState::policy is set from the final value of the policy field in the
parameter structure for the KVM_SEV_LAUNCH_START ioctl().  But, AFAICT
that ioctl() won't ever change it from the original supplied value which
comes from SevGuestState::policy.

So, remove this field and just use SevGuestState::policy directly.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200604064219.436242-8-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Remove redundant cbitpos and reduced_phys_bits fields

The SEVState structure has cbitpos and reduced_phys_bits fields which are
simply copied from the SevGuestState structure and never changed.  Now that
SEVState is embedded in SevGuestState we can just access the original copy
directly.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200604064219.436242-7-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Partial cleanup to sev_state global

The SEV code uses a pretty ugly global to access its internal state.  Now
that SEVState is embedded in SevGuestState, we can avoid accessing it via
the global in some cases.  In the remaining cases use a new global
referencing the containing SevGuestState which will simplify some future
transformations.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200604064219.436242-6-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Embed SEVState in SevGuestState

Currently SevGuestState contains only configuration information.  For
runtime state another non-QOM struct SEVState is allocated separately.

Simplify things by instead embedding the SEVState structure in
SevGuestState.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200604064219.436242-5-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Rename QSevGuestInfo

At the moment this is a purely passive object which is just a container for
information used elsewhere, hence the name.  I'm going to change that
though, so as a preliminary rename it to SevGuestState.

That name risks confusion with both SEVState and SevState, but I'll be
working on that in following patches.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200604064219.436242-4-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Move local structure definitions into .c file

Neither QSevGuestInfo nor SEVState (not to be confused with SevState) is
used anywhere outside target/i386/sev.c, so they might as well live in
there rather than in a (somewhat) exposed header.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200604064219.436242-3-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: sev: Remove unused QSevGuestInfoClass

This structure is nothing but an empty wrapper around the parent class,
which by QOM conventions means we don't need it at all.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200604064219.436242-2-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

xen: fix build without pci passthrough

Xen PCI passthrough support may not be available and thus the global
variable "has_igd_gfx_passthru" might be compiled out. Common code
should not access it in that case.

Unfortunately, we can't use CONFIG_XEN_PCI_PASSTHROUGH directly in
xen-common.c so this patch instead move access to the
has_igd_gfx_passthru variable via function and those functions are
also implemented as stubs. The stubs will be used when QEMU is built
without passthrough support.

Now, when one will want to enable igd-passthru via the -machine
property, they will get an error message if QEMU is built without
passthrough support.

Fixes: 46472d82322d0 ('xen: convert "-machine igd-passthru" to an accelerator property')
Reported-by: Roger Pau Monné <roger.pau@citrix.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Message-Id: <20200603160442.3151170-1-anthony.perard@citrix.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Drop HVFX86EmulatorState

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-14-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Move mmio_buf into CPUX86State

There's no similar field in CPUX86State, but it's needed for MMIO traps.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-13-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Move lazy_flags into CPUX86State

The lazy flags are still needed for instruction decoder.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-12-r.bolshakov@yadro.com>
[Move struct to target/i386/cpu.h - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Drop regs in HVFX86EmulatorState

HVFX86EmulatorState carries it's own copy of x86 registers. It can be
dropped in favor of regs in generic CPUX86State.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-11-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Drop copy of RFLAGS defines

Use the ones provided in target/i386/cpu.h instead.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-10-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Drop rflags from HVFX86EmulatorState

HVFX86EmulatorState carries it's own copy of x86 flags. It can be
dropped in favor of eflags in generic CPUX86State.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-9-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Drop fetch_rip from HVFX86EmulatorState

The field is used to print address of instructions that have no parser
in decode_invalid(). RIP from VMCS is saved into fetch_rip before
decoding starts but it's also saved into env->eip in load_regs().
Therefore env->eip can be used instead of fetch_rip.

While at it, correct address printed in decode_invalid(). It prints an
address before the unknown instruction.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-8-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Use IP from CPUX86State

Drop and replace rip field from HVFX86EmulatorState in favor of eip from
common CPUX86State.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-7-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

i386: hvf: Use ins_len to advance IP

There's no need to read VMCS twice, instruction length is already
available in ins_len.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-6-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

i386: hvf: Drop unused variable

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-5-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

i386: hvf: Clean stray includes in sysemu

They have no use.

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Reviewed-by: Claudio Fontana <cfontana@suse.de>
Message-Id: <20200528193758.51454-4-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

i386: hvf: Drop useless declarations in sysemu

They're either declared elsewhere or have no use.

While at it, rename _hvf_cpu_synchronize_post_init() to
do_hvf_cpu_synchronize_post_init().

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-3-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

i386: hvf: Move HVFState definition into hvf

"sysemu/hvf.h" is intended for inclusion in generic code. However it
also contains several hvf definitions and declarations, including
HVFState that are used only inside "hvf.c". "hvf-i386.h" would be more
appropriate place to define HVFState as it's only included by "hvf.c"
and "x86_task.c".

Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200528193758.51454-2-r.bolshakov@yadro.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/ppc: Restrict PPCVirtualHypervisorClass to system-mode

The code related to PPC Virtual Hypervisor is pointless in user-mode.

Acked-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200526172427.17460-5-f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

sysemu/hvf: Only declare hvf_allowed when HVF is available

When HVF is not available, the hvf_allowed variable does not exist.

Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200526172427.17460-4-f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

sysemu/tcg: Only declare tcg_allowed when TCG is available

When TCG is not available, the tcg_allowed variable does not exist.

Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200526172427.17460-3-f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

sysemu/accel: Restrict machine methods to system-mode

Restrict init_machine(), setup_post() and has_memory()
to system-mode.

Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200526172427.17460-2-f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: correct fix for pcmpxstrx substring search

This corrects a bug introduced in my previous fix for SSE4.2 pcmpestri
/ pcmpestrm / pcmpistri / pcmpistrm substring search, commit
ae35eea7e4a9f21dd147406dfbcd0c4c6aaf2a60.

That commit fixed a bug that showed up in four GCC tests with one libc
implementation.  The tests in question generate random inputs to the
intrinsics and compare results to a C implementation, but they only
test 1024 possible random inputs, and when the tests use the cases of
those instructions that work with word rather than byte inputs, it's
easy to have problematic cases that show up much less frequently than
that.  Thus, testing with a different libc implementation, and so a
different random number generator, showed up a problem with the
previous patch.

When investigating the previous test failures, I found the description
of these instructions in the Intel manuals (starting from computing a
16x16 or 8x8 set of comparison results) confusing and hard to match up
with the more optimized implementation in QEMU, and referred to AMD
manuals which described the instructions in a different way.  Those
AMD descriptions are very explicit that the whole of the string being
searched for must be found in the other operand, not running off the
end of that operand; they say "If the prototype and the SUT are equal
in length, the two strings must be identical for the comparison to be
TRUE.".  However, that statement is incorrect.

In my previous commit message, I noted:

  The operation in this case is a search for a string (argument d to
  the helper) in another string (argument s to the helper); if a copy
  of d at a particular position would run off the end of s, the
  resulting output bit should be 0 whether or not the strings match in
  the region where they overlap, but the QEMU implementation was
  wrongly comparing only up to the point where s ends and counting it
  as a match if an initial segment of d matched a terminal segment of
  s.  Here, "run off the end of s" means that some byte of d would
  overlap some byte outside of s; thus, if d has zero length, it is
  considered to match everywhere, including after the end of s.

The description "some byte of d would overlap some byte outside of s"
is accurate only when understood to refer to overlapping some byte
*within the 16-byte operand* but at or after the zero terminator; it
is valid to run over the end of s if the end of s is the end of the
16-byte operand.  So the fix in the previous patch for the case of d
being empty was correct, but the other part of that patch was not
correct (as it never allowed partial matches even at the end of the
16-byte operand).  Nor was the code before the previous patch correct
for the case of d nonempty, as it would always have allowed partial
matches at the end of s.

Fix with a partial revert of my previous change, combined with
inserting a check for the special case of s having maximum length to
determine where it is necessary to check for matches.

In the added test, test 1 is for the case of empty strings, which
failed before my 2017 patch, test 2 is for the bug introduced by my
2017 patch and test 3 deals with the case where a match of an initial
segment at the end of the string is not valid when the string ends
before the end of the 16-byte operand (that is, the case that would be
broken by a simple revert of the non-empty-string part of my 2017
patch).

Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2006121344290.9881@digraph.polyomino.org.uk>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

virtio,acpi,pci: features, fixes, cleanups, tests

Max slots negotiation for vhost-user.
Free page reporting for balloon.
Partial TPM2 ACPI support for ARM.
Support for NVDIMMs having their own proximity domains.
New vhost-user-vsock device.

Fixes, cleanups in ACPI, PCI, virtio.
New tests for TPM ACPI.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
# gpg: Signature made Fri 12 Jun 2020 15:18:04 BST
# gpg:                using RSA key 5D09FD0871C8F85B94CA8A0D281F0DB8D28D5469
# gpg:                issuer "mst@redhat.com"
# gpg: Good signature from "Michael S. Tsirkin <mst@kernel.org>" [full]
# gpg:                 aka "Michael S. Tsirkin <mst@redhat.com>" [full]
# Primary key fingerprint: 0270 606B 6F3C DF3D 0B17  0970 C350 3912 AFBE 8E67
#      Subkey fingerprint: 5D09 FD08 71C8 F85B 94CA  8A0D 281F 0DB8 D28D 5469

* remotes/mst/tags/for_upstream: (58 commits)
  virtio-pci: fix queue_enable write
  pci: Display PCI IRQ pin in "info pci"
  Fix parameter type in vhost migration log path
  acpi: ged: rename event memory region
  acpi: fadt: add hw-reduced sleep register support
  acpi: madt: skip pci override on pci-less systems.
  acpi: create acpi-common.c and move madt code
  acpi: make build_madt() more generic.
  virtio: add vhost-user-vsock-pci device
  virtio: add vhost-user-vsock base device
  vhost-vsock: add vhost-vsock-common abstraction
  hw/pci: Fix crash when running QEMU with "-nic model=rocker"
  libvhost-user: advertise vring features
  Lift max ram slots limit in libvhost-user
  Support individual region unmap in libvhost-user
  Support adding individual regions in libvhost-user
  Support ram slot configuration in libvhost-user
  Refactor out libvhost-user fault generation logic
  Lift max memory slots limit imposed by vhost-user
  Transmit vhost-user memory regions individually
  ...

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

virtio-pci: fix queue_enable write

Spec said: The driver uses this to selectively prevent the device from
executing requests from this virtqueue. 1 - enabled; 0 - disabled.

Though write 0 to queue_enable is forbidden by the spec, we should not
assume that the value is 1.

Fix this by ignore the write value other than 1.

Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20200610054351.15811-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

pci: Display PCI IRQ pin in "info pci"

Sometimes it would be good to be able to read the pin number along
with the IRQ number allocated.  Since we'll dump the IRQ number, no
reason to not dump the pin information.  For example, the vfio-pci
device will overwrite the pin with the hardware pin number.  It would
be nice to know the pin number of one assigned device from QMP/HMP.

CC: Dr. David Alan Gilbert <dgilbert@redhat.com>
CC: Alex Williamson <alex.williamson@redhat.com>
CC: Michael S. Tsirkin <mst@redhat.com>
CC: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
CC: Julia Suvorova <jusual@redhat.com>
CC: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200317195908.283800-1-peterx@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Acked-by: Markus Armbruster <armbru@redhat.com>

Fix parameter type in vhost migration log path

The ‘enable’ parameter to the vhost_migration_log() function is given as
an int, but "true"/"false" values are passed in wherever it is invoked.
Inside the function itself it is only ever compared with bool values.
Therefore the parameter value itself should be changed to bool.

Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <CAFubqFtqNZw=Y-ar3N=3zTQi6LkKg_G-7W7OOHHbE7Y1fV7HAQ@mail.gmail.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

acpi: ged: rename event memory region

Rename memory region and callbacks and ops to carry "evt" in the name
because a second region will be added shortly.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Message-Id: <20200520132003.9492-10-kraxel@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Igor Mammedow <imammedo@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

acpi: fadt: add hw-reduced sleep register support

Add fields to struct AcpiFadtData and update build_fadt() to properly
generate sleep register entries.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20200520132003.9492-9-kraxel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

acpi: madt: skip pci override on pci-less systems.

Needed for microvm.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200520132003.9492-8-kraxel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

acpi: create acpi-common.c and move madt code

We'll need madt support for microvm.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20200520132003.9492-7-kraxel@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

acpi: make build_madt() more generic.

Remove PCMachineState dependency from build_madt().
Pass AcpiDeviceIf as separate argument instead of
depending on PCMachineState->acpi_dev.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20200520132003.9492-6-kraxel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio: add vhost-user-vsock-pci device

Add the PCI version of vhost-user-vsock

Launch QEMU like this:

  qemu -chardev socket,path=/tmp/vm.vsock,id=chr0 \
       -device vhost-user-vsock-pci,chardev=chr0

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20200522122512.87413-4-sgarzare@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio: add vhost-user-vsock base device

This patch introduces a vhost-user device for vsock, using the
vhost-vsock-common parent class.

The vhost-user-vsock device can be used to implement the virtio-vsock
device emulation in user-space.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20200522122512.87413-3-sgarzare@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vhost-vsock: add vhost-vsock-common abstraction

This patch prepares the introduction of vhost-user-vsock, moving
the common code usable for both vhost-vsock and vhost-user-vsock
devices, in the new vhost-vsock-common parent class.

While moving the code, fixed checkpatch warnings about block comments.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20200522122512.87413-2-sgarzare@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

hw/pci: Fix crash when running QEMU with "-nic model=rocker"

QEMU currently aborts when being started with "-nic model=rocker" or with
"-net nic,model=rocker". This happens because the "rocker" device is not
a normal NIC but a switch, which has different properties. Thus we should
only consider real NIC devices for "-nic" and "-net". These devices can
be identified by the "netdev" property, so check for this property before
adding the device to the list.

Reported-by: Michael Tokarev <mjt@tls.msk.ru>
Fixes: 52310c3fa7dc854d ("net: allow using any PCI NICs in -net or -nic")
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20200527153152.9211-1-thuth@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

libvhost-user: advertise vring features

libvhost-user implements several vring features without advertising
them. There is no way for the vhost-user master to detect support for
these features.

Things more or less work today because QEMU assumes the vhost-user
backend always implements certain feature bits like
VIRTIO_RING_F_EVENT_IDX. This is not documented anywhere.

This patch explicitly advertises features implemented in libvhost-user
so that the vhost-user master does not need to make undocumented
assumptions.

Feature bits that libvhost-user now advertises can be removed from
vhost-user-blk.c. Devices should not be responsible for advertising
vring feature bits, that is libvhost-user's job.

Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20200529161338.456017-1-stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Lift max ram slots limit in libvhost-user

Historically, VMs with vhost-user devices could hot-add memory a maximum
of 8 times. Now that the VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS
protocol feature has been added, VMs with vhost-user backends which
support this new feature can support a configurable number of ram slots
up to the maximum supported by the target platform.

This change adds VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS support for
backends built with libvhost-user, and increases the number of supported
ram slots from 8 to 32.

Memory hot-add, hot-remove and postcopy migration were tested with
the vhost-user-bridge sample.

Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1588533678-23450-11-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

Support individual region unmap in libvhost-user

When the VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS protocol feature is
enabled, on memory hot-unplug qemu will transmit memory regions to
remove individually using the new message VHOST_USER_REM_MEM_REG
message. With this change, vhost-user backends build with libvhost-user
can now unmap individual memory regions when receiving the
VHOST_USER_REM_MEM_REG message.

Qemu only sends VHOST_USER_REM_MEM_REG messages when the
VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS feature is negotiated, and
support for that feature has not yet been added in libvhost-user, this
new functionality is not yet used.

Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1588533678-23450-10-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

Support adding individual regions in libvhost-user

When the VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS is enabled, qemu will
transmit memory regions to a backend individually using the new message
VHOST_USER_ADD_MEM_REG. With this change vhost-user backends built with
libvhost-user can now map in new memory regions when VHOST_USER_ADD_MEM_REG
messages are received.

Qemu only sends VHOST_USER_ADD_MEM_REG messages when the
VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS feature is negotiated, and
since it is not yet supported in libvhost-user, this new functionality
is not yet used.

Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1588533678-23450-9-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

Support ram slot configuration in libvhost-user

The VHOST_USER_GET_MAX_MEM_SLOTS message allows a vhost-user backend to
specify a maximum number of ram slots it is willing to support. This
change adds support for libvhost-user to process this message. For now
the backend will reply with 8 as the maximum number of regions
supported.

libvhost-user does not yet support the vhost-user protocol feature
VHOST_USER_PROTOCOL_F_CONFIGIRE_MEM_SLOTS, so qemu should never
send the VHOST_USER_GET_MAX_MEM_SLOTS message. Therefore this new
functionality is not currently used.

Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1588533678-23450-8-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Refactor out libvhost-user fault generation logic

In libvhost-user, the incoming postcopy migration path for setting the
backend's memory tables has become convolued. In particular, moving the
logic which starts generating faults, having received the final ACK from
qemu can be moved to a separate function. This simplifies the code
substantially.

This logic will also be needed by the postcopy path once the
VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS feature is supported.

Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1588533678-23450-7-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>