ssi: Convert uses of ssi_create_slave_no_init() with Coccinelle
Replace
dev = ssi_create_slave_no_init(bus, type_name);
...
qdev_init_nofail(dev);
by
dev = qdev_new(type_name);
...
qdev_realize_and_unref(dev, bus, &error_fatal);
Recent commit "qdev: New qdev_new(), qdev_realize(), etc." explains
why.
@@
type SSIBus;
identifier bus;
expression dev, qbus, expr;
expression list args;
@@
- bus = (SSIBus *)qbus;
+ bus = qbus; // TODO fix up decl
...
- dev = ssi_create_slave_no_init(bus, args);
+ dev = qdev_new(args);
... when != dev = expr
- qdev_init_nofail(dev);
+ qdev_realize_and_unref(dev, bus, &error_fatal);
@@
expression dev, bus, expr;
expression list args;
@@
- dev = ssi_create_slave_no_init(bus, args);
+ dev = qdev_new(args);
... when != dev = expr
- qdev_init_nofail(dev);
+ qdev_realize_and_unref(dev, BUS(bus), &error_fatal);
ssi: ssi_auto_connect_slaves() never does anything, drop
ssi_auto_connect_slaves(parent, cs_line, bus) iterates over @parent's
QOM children @dev of type TYPE_SSI_SLAVE. It puts these on @bus, and
sets cs_line[] to qdev_get_gpio_in_named(dev, SSI_GPIO_CS, 0).
Suspicious: there is no protection against overrunning cs_line[].
Turns out it's safe because ssi_auto_connect_slaves() never finds any
such children. Its called by realize methods of some (but not all)
devices providing an SSI bus, and gets passed the device.
SSI slave devices are always created with ssi_create_slave_no_init(),
optionally via ssi_create_slave(). This adds them to their SSI bus.
It doesn't set their QOM parent.
ssi_create_slave_no_init() is always immediately followed by
qdev_init_nofail(), with no QOM parent assigned, so
device_set_realized() puts the device into the /machine/unattached/
orphanage. None become QOM children of a device providing an SSI bus.
ssi_auto_connect_slaves() was added in commit b4ae3cfa57 "ssi: Add
slave autoconnect helper". I can't see which slaves it was supposed
to connect back then.
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.
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.
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).
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:
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).
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.
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.
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():
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.
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.
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.
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:
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.
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 ""
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:
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.
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:
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
object_property_set_bool(OBJECT(dev), true, "realized", ...) right
after qdev_init_nofail(dev) does nothing, because qdev_init_nofail()
already realizes. Drop.
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":
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":
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.
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.
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.
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:
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:
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
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:
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:
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.
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":
Thomas Huth [Thu, 11 Jun 2020 17:24:45 +0000 (19:24 +0200)]
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.
Pavel Dovgalyuk [Fri, 22 May 2020 06:45:54 +0000 (09:45 +0300)]
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.
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.
David Gibson [Thu, 4 Jun 2020 06:42:19 +0000 (16:42 +1000)]
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.
David Gibson [Thu, 4 Jun 2020 06:42:18 +0000 (16:42 +1000)]
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.
David Gibson [Thu, 4 Jun 2020 06:42:17 +0000 (16:42 +1000)]
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.
David Gibson [Thu, 4 Jun 2020 06:42:16 +0000 (16:42 +1000)]
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.
David Gibson [Thu, 4 Jun 2020 06:42:15 +0000 (16:42 +1000)]
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.
David Gibson [Thu, 4 Jun 2020 06:42:13 +0000 (16:42 +1000)]
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.
David Gibson [Thu, 4 Jun 2020 06:42:12 +0000 (16:42 +1000)]
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.
Anthony PERARD [Wed, 3 Jun 2020 16:04:42 +0000 (17:04 +0100)]
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.
Roman Bolshakov [Thu, 28 May 2020 19:37:52 +0000 (22:37 +0300)]
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.
Roman Bolshakov [Thu, 28 May 2020 19:37:46 +0000 (22:37 +0300)]
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".
Joseph Myers [Fri, 12 Jun 2020 13:45:23 +0000 (13:45 +0000)]
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).
Peter Maydell [Fri, 12 Jun 2020 14:53:49 +0000 (15:53 +0100)]
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 <[email protected]>
# gpg: Signature made Fri 12 Jun 2020 15:18:04 BST
# gpg: using RSA key 5D09FD0871C8F85B94CA8A0D281F0DB8D28D5469
# gpg: issuer "[email protected]"
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>" [full]
# gpg: aka "Michael S. Tsirkin <[email protected]>" [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
...
Peter Xu [Tue, 17 Mar 2020 19:59:08 +0000 (15:59 -0400)]
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.
Raphael Norwitz [Thu, 7 May 2020 21:37:58 +0000 (17:37 -0400)]
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 <[email protected]>
Message-Id: <CAFubqFtqNZw=Y-ar3N=3zTQi6LkKg_G-7W7OOHHbE7Y1fV7HAQ@mail.gmail.com> Reviewed-by: Eric Blake <[email protected]> Reviewed-by: Michael S. Tsirkin <[email protected]> Signed-off-by: Michael S. Tsirkin <[email protected]>
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.
Thomas Huth [Wed, 27 May 2020 15:31:52 +0000 (17:31 +0200)]
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.
Stefan Hajnoczi [Fri, 29 May 2020 16:13:38 +0000 (17:13 +0100)]
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.
Raphael Norwitz [Thu, 21 May 2020 05:00:59 +0000 (05:00 +0000)]
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.
projects / qemu.git / log