We already detect if a device is being hot plugged before CAS to trigger
a CAS reboot and during migration to migrate the state of the associated
DRC. But hot unplugging a device is also an asynchronous operation that
requires the guest to take action. This means that if the guest is migrated
after the hot unplug event was sent but before it could release the device
with RTAS, the destination QEMU doesn't know about the pending unplug
operation and doesn't actually remove the device when the guest finally
releases it.
Similarly, if the unplug request is fired before CAS, the guest isn't
notified of the change, just like with hotplug. It ends up booting with
the device still present in the DT and configures it, just like it was
never removed. Even weirder, since the event is still queued, it will
be eventually processed when some other unrelated event is posted to
the guest.
Enhance spapr_drc_transient() to also return true if an unplug request is
pending. This fixes the issue at CAS with a CAS reboot request and
causes the DRC state to be migrated. Some extra care is still needed to
inform the destination that an unplug request is pending : migrate the
unplug_requested field of the DRC in an optional subsection. This might
break backwards migration, but this is still better than ending with
an inconsistent guest.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <158169248798.3465937.1108351365840514270.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We currently don't support hotplug of devices between boot and CAS. If
this happens a CAS reboot is triggered. We detect this during CAS using
the spapr_drc_needed() function which is essentially a VMStateDescription
.needed callback. Even if the condition for CAS reboot happens to be the
same as for DRC migration, it looks wrong to piggyback a migration helper
for this.
Introduce a helper with slightly more explicit name and use it in both CAS
and DRC migration code. Since a subsequent patch will enhance this helper
to cover the case of hot unplug, let's go for spapr_drc_transient(). While
here convert spapr_hotplugged_dev_before_cas() to the "transient" wording as
well.
This doesn't change any behaviour.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <158169248180.3465937.9531405453362718771.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
As reported by Coverity defect CID 1419397, the 'j' variable goes up to
63 and shouldn't be used to left shift a 32-bit integer.
The result of the operation goes to a 64-bit integer : use a 64-bit
constant.
Reported-by: Coverity CID 1419397 Bad bit shift operation
Fixes: 9ae1329ee2 "ppc/pnv: Add models for POWER8 PHB3 PCIe Host bridge"
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <158153364010.3229002.8004283672455615950.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Some default features of the pseries machine are only available with
KVM. Warnings are printed when the pseries machine is used with another
accelerator:
qemu-system-ppc64: warning: TCG doesn't support requested feature,
cap-ccf-assist=on
qemu-system-ppc64: warning: Firmware Assisted Non-Maskable
Interrupts(FWNMI) not supported in TCG
qemu-system-ppc64: warning: TCG doesn't support requested feature,
cap-ccf-assist=on
qemu-system-ppc64: warning: Firmware Assisted Non-Maskable
Interrupts(FWNMI) not supported in TCG
qemu-system-ppc64: warning: TCG doesn't support requested feature,
cap-ccf-assist=on
qemu-system-ppc64: warning: Firmware Assisted Non-Maskable
Interrupts(FWNMI) not supported in TCG
This is annoying for CI since it usually runs without KVM. We already
disable features that emit similar warnings thanks to properties of
the pseries machine, but this is open-coded in various
places. Consolidate the set of properties in a single place. Extend it
to silence the above warnings. And use it in the various tests that
start pseries machines.
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <158059697130.1820292.7823434132030453110.stgit@bahia.lan>
Reviewed-by: Thomas Huth <thuth@redhat.com>
[dwg: Correct minor grammatical error]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
According to the description of "ibm,client-architecture-support" that
can found in LoPAPR "B.6.2.3 Root Node Methods":
If multiple partition processors or threads are active at the time of
the ibm,client-architecture-support method call, or an error is detected
in the format of the ibm,architecture.vec structure, the err? boolean
shall be TRUE; else FALSE.
We certainly don't want to temper with the platform or with the PCR of
the other vCPUs if they happen to be active. Ensure we have only one
active vCPU and fail CAS otherwise. This is just for conformance and
robustness, it doesn't fix any known bugs.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157969867170.571404.12117797348882189656.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Most of the option vector helpers have assertions to check their
arguments aren't null. The guest can provide an arbitrary address
for the CAS structure that would result in such null arguments.
Fail CAS with H_PARAMETER and print a warning instead of aborting
QEMU.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <157925255250.397143.10855183619366882459.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
init_in_iov_from_pdu might not be able to allocate the full buffer size
requested, which comes from the client and could be larger than the
transport has available at the time of the request. Specifically, this
can happen with read operations, with the client requesting a read up to
the max allowed, which might be more than the transport has available at
the time.
Today the implementation of init_in_iov_from_pdu throws an error, both
Xen and Virtio.
Instead, change the V9fsTransport interface so that the size becomes a
pointer and can be limited by the implementation of
init_in_iov_from_pdu.
Change both the Xen and Virtio implementations to set the size to the
size of the buffer they managed to allocate, instead of throwing an
error. However, if the allocated buffer size is less than P9_IOHDRSZ
(the size of the header) still throw an error as the case is unhandable.
Signed-off-by: Stefano Stabellini <stefano.stabellini@xilinx.com>
CC: groug@kaod.org
CC: anthony.perard@citrix.com
CC: roman@zededa.com
CC: qemu_oss@crudebyte.com
[groug: fix 32-bit build]
Signed-off-by: Greg Kurz <groug@kaod.org>
The proper way to do that would be to use device_class_set_parent_realize(),
but defining a Pnv8PsiClass and a Pnv9PsiClass types with a parent_realize
pointer adds a fair amount of code. Calling pnv_psi_realize() explicitely
is fine for now.
This should probably be achieved with a device realize hook in the
PSI base class and device_class_set_parent_realize() in the children
classes.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <157841476667.66386.13659183399113837990.stgit@bahia.tlslab.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
And call it from a QEMU reset handler. This allows each PNV child class to
override the reset hook if needed, eg. POWER8 doesn't but POWER9 does.
The proper way to do that would be to use device_class_set_parent_reset(),
but defining a Pnv8PsiClass and a Pnv9PsiClass types with a parent_reset
pointer adds a fair amount of code. Calling pnv_psi_reset() explicitely is
fine for now.
A subsequent patch will consolidate the call to qemu_register_reset() in
a single place.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <157841476035.66386.17838417527621752518.stgit@bahia.tlslab.ibm.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
And use it instead of reaching out to the machine. This allows to get rid
of a call to qdev_get_machine() and to reduce the scope of another one so
that it is only used within the argument list of error_append_hint(). This
is an acceptable tradeoff compared to all it would require to know about
the maximum number of CPUs here without calling qdev_get_machine().
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20200106145645.4539-10-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Now that the spapr and pnv machines do set the "xive-fabric" link, the
use of the XIVE fabric pointer becomes mandatory. This is checked with
an assert() in a new realize hook. Since the XIVE router is realized at
machine init for the all the machine's life time, no risk to abort an
already running guest (ie. not a hotplug path).
This gets rid of a qdev_get_machine() call.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20200106145645.4539-6-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
In order to get rid of qdev_get_machine(), first add a pointer to the
XIVE fabric under the XIVE router and make it configurable through a
QOM link property.
Configure it in the spapr and pnv machine. In the case of pnv, the XIVE
routers are under the chip, so this is done with a QOM alias property of
the POWER9 pnv chip.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20200106145645.4539-5-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The XIVE router base class currently inherits an empty realize hook
from the sysbus device base class, but it will soon implement one
of its own to perform some sanity checks. Do the preliminary plumbing
to have it called.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <20200106145645.4539-4-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The number of CPU chips of the powernv machine is configurable through a
"num-chips" property. This doesn't fit well with the CPU topology, eg.
some configurations can come up with more CPUs than the maximum of CPUs
set in the toplogy. This causes assertion to be hit with mttcg:
-machine powernv,num-chips=2 -smp cores=2 -accel tcg,thread=multi
ERROR:
tcg/tcg.c:789:tcg_register_thread: assertion failed: (n < ms->smp.max_cpus)
Aborted (core dumped)
Mttcg mandates the CPU topology to be dimensioned to the actual number
of CPUs, depending on the number of chips the user asked for. That is,
'-machine num-chips=N' should always have a '-smp' companion with a
topology that meats the resulting number of CPUs, typically
'-smp sockets=N'.
It thus seems that "num-chips" doesn't bring anything but forcing the user
to specify the requested number of chips on the command line twice. Simplify
the command line by computing the number of chips based on the CPU topology
exclusively. The powernv machine isn't a production thing ; it is mostly
used by developpers to prepare the bringup of real HW. Because of this and
for simplicity, this deliberately ignores the official deprecation process
and dumps "num-chips" right away : '-smp sockets=N' is now the only way to
control the number of CPU chips.
This is done at machine init because smp_parse() is called after instance
init.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157830658266.533764.2214183961444213947.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The XIVE router base class currently inherits an empty realize hook
from the sysbus device base class, but it will soon implement one
of its own to perform some sanity checks. Do the preliminary plumbing
to have it called.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20191219181155.32530-6-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
QOM interfaces allow a limited form of multiple inheritance, at the
condition of being stateless. That is, they cannot be instantiated
and a pointer to an interface shouldn't be dereferenceable in any way.
This is achieved by making the QOM instance type an incomplete type,
which is, as mentioned by Markus Armbruster, the closest you can get
to abstract class in C.
Incomplete types are widely used to hide implementation details, but
people usually expect to find at least one place where the type is
fully defined. The fact that it doesn't happen with QOM interfaces is
quite disturbing, especially since it isn't documented anywhere as
recently discussed in this thread:
https://lists.gnu.org/archive/html/qemu-devel/2019-12/msg01579.html
Amend the documentation in the object.h header file to provide more
details about why and how to implement QOM interfaces using incomplete
types.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The XSCOM bus is implemented with a QOM interface, which is mostly
generic from a CPU type standpoint, except for the computation of
addresses on the Pervasive Connect Bus (PCB) network. This is handled
by the pnv_xscom_pcba() function with a switch statement based on
the chip_type class level attribute of the CPU chip.
This can be achieved using QOM. Also the address argument is masked with
PNV_XSCOM_SIZE - 1, which is for POWER8 only. Addresses may have different
sizes with other CPU types. Have each CPU chip type handle the appropriate
computation with a QOM xscom_pcba() method.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157623843543.360005.13996472463887521794.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The pnv_chip_core_realize() function configures the XSCOM MMIO subregion
for each core of a single chip. The base address of the subregion depends
on the CPU type. Its computation is currently open-code using the
pnv_chip_is_powerXX() helpers. This can be achieved with QOM. Introduce
a method for this in the base chip class and implement it in child classes.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157623841311.360005.4705705734873339545.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The pnv_pic_print_info() callback checks the type of the chip in order
to forward to the request appropriate interrupt controller. This can
be achieved with QOM. Introduce a method for this in the base chip class
and implement it in child classes.
This also prepares ground for the upcoming interrupt controller of POWER10
chips.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157623840755.360005.5002022339473369934.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We add an extra node to advertise power management on some machines,
namely powernv9 and powernv10. This is achieved by using the
pnv_is_power9() and pnv_is_power10() helpers.
This can be achieved with QOM. Add a method to the base class for
powernv machines and have it implemented by machine types that
support power management instead.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157623839642.360005.9243510140436689941.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The pnv_dt_create() function generates different contents for the
"compatible" property of the root node in the DT, depending on the
CPU type. This is open coded with multiple ifs using pnv_is_powerXX()
helpers.
It seems cleaner to achieve with QOM. Introduce a base class for the
powernv machine and a compat attribute that each child class can use
to provide the value for the "compatible" property.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157623839085.360005.4046508784077843216.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
[dwg: Folded in small fix Greg spotted after posting]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The Processor Service Interface (PSI) model has a chip_type class level
attribute, which is used to generate the content of the "compatible" DT
property according to the CPU type.
Since the PSI model already has specialized classes for each supported
CPU type, it seems cleaner to achieve this with QOM. Provide the content
of the "compatible" property with a new class level attribute.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157623837974.360005.14706607446188964477.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
PnvXScomInterface is an interface instance. It should never be
dereferenced. Drop the dummy type definition for extra safety,
which is the common practice with QOM interfaces.
While here also convert the bogus OBJECT_CHECK() to INTERFACE_CHECK().
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157608025541.186670.1577861507610404326.stgit@bahia.lan>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The power7_set_irq() and power9_set_irq() functions set this but it is
never used actually. Modern Book3s compatible CPUs are only supported
by the pnv and spapr machines. They have an interrupt controller, XICS
for POWER7/8 and XIVE for POWER9, whose models don't require to track
IRQ input states at the CPU level.
Drop these lines to avoid confusion.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157548862861.3650476.16622818876928044450.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When a CPU is reset, QEMU makes sure no interrupt is pending by clearing
CPUPPCstate::pending_interrupts in ppc_cpu_reset(). In the case of a
complete machine emulation, eg. a sPAPR machine, an external interrupt
request could still be pending in KVM though, eg. an IPI. It will be
eventually presented to the guest, which is supposed to acknowledge it at
the interrupt controller. If the interrupt controller is emulated in QEMU,
either XICS or XIVE, ppc_set_irq() won't deassert the external interrupt
pin in KVM since it isn't pending anymore for QEMU. When the vCPU re-enters
the guest, the interrupt request is still pending and the vCPU will try
again to acknowledge it. This causes an infinite loop and eventually hangs
the guest.
The code has been broken since the beginning. The issue wasn't hit before
because accel=kvm,kernel-irqchip=off is an awkward setup that never got
used until recently with the LC92x IBM systems (aka, Boston).
Add a ppc_irq_reset() function to do the necessary cleanup, ie. deassert
the IRQ pins of the CPU in QEMU and most importantly the external interrupt
pin for this vCPU in KVM.
Reported-by: Satheesh Rajendran <sathnaga@linux.vnet.ibm.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157548861740.3650476.16879693165328764758.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The XIVE KVM devices now has an attribute to configure the number of
interrupt servers. This allows to greatly optimize the usage of the VP
space in the XIVE HW, and thus to start a lot more VMs.
Only set this attribute if available in order to support older POWER9
KVM.
The XIVE KVM device now reports the exhaustion of VPs upon the
connection of the first VCPU. Check that in order to have a chance
to provide a hint to the user.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157478679392.67101.7843580591407950866.stgit@bahia.tlslab.ibm.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The XICS-on-XIVE KVM devices now has an attribute to configure the number
of interrupt servers. This allows to greatly optimize the usage of the VP
space in the XIVE HW, and thus to start a lot more VMs.
Only set this attribute if available in order to support older POWER9 KVM
and pre-POWER9 XICS KVM devices.
The XICS-on-XIVE KVM device now reports the exhaustion of VPs upon the
connection of the first VCPU. Check that in order to have a chance to
provide a hint to the user.
`
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157478678846.67101.9660531022460517710.stgit@bahia.tlslab.ibm.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The XIVE and XICS-on-XIVE KVM devices on POWER9 hosts can greatly reduce
their consumption of some scarce HW resources, namely Virtual Presenter
identifiers, if they know the maximum number of vCPUs that may run in the
VM.
Prepare ground for this by passing the value down to xics_kvm_connect()
and kvmppc_xive_connect(). This is purely mechanical, no functional
change.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157478678301.67101.2717368060417156338.stgit@bahia.tlslab.ibm.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Allocating a big void * array to store multiple objects isn't a
recommended practice for various reasons:
- no compile time type checking
- potential dangling pointers if a reference on an individual is
taken and the array is freed later on
- duplicate boiler plate everywhere the array is browsed through
Allocate an array of pointers and populate it instead.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20191125065820.927-4-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When using the XIVE KVM device, the trigger page is directly accessible
in QEMU. Unlike with XICS, no need to ask KVM to fire the interrupt. A
simple store on the trigger page does the job.
Just call xive_esb_trigger().
This may improve performance of emulated devices that go through
qemu_set_irq(), eg. virtio devices created with ioeventfd=off or
configured by the guest to use LSI interrupts, which aren't really
recommended setups.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157408992731.494439.3405812941731584740.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Failing to set any of the ICS property should really never happen:
- object_property_add_child() always succeed unless the child object
already has a parent, which isn't the case here obviously since the
ICS has just been created with object_new()
- the ICS has an "nr-irqs" property than can be set as long as the ICS
isn't realized
In both cases, an error indicates there is a bug in QEMU. Propagating the
error, ie. exiting QEMU since spapr_irq_init() is called with &error_fatal
doesn't make much sense. Abort instead. This is consistent with what is
done with XIVE : both qdev_create() and qdev_prop_set_uint32() abort QEMU
on error.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157403285265.409804.8683093665795248192.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The ICP object has both a pointer and an ICP_PROP_CPU property pointing
to the cpu. Confusing bugs could arise if these ever go out of sync.
Change the property definition so that it explicitly sets the pointer.
The property isn't optional : not being able to set the link is a bug
and QEMU should rather abort than exit in this case.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157403284709.409804.16142099083325945141.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The ICP object has both a pointer and an ICP_PROP_XICS property pointing
to the XICS fabric. Confusing bugs could arise if these ever go out of
sync.
Change the property definition so that it explicitly sets the pointer.
The property isn't optional : not being able to set the link is a bug
and QEMU should rather abort than exit in this case.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157403284152.409804.17114564311521923733.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The ICS object has both a pointer and an ICS_PROP_XICS property pointing
to the XICS fabric. Confusing bugs could arise if these ever go out of
sync.
Change the property definition so that it explicitely sets the pointer.
The property isn't optional : not being able to set the link is a bug
and QEMU should rather abort than exit in this case.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157403283596.409804.17347207690271971987.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The core object has both a pointer and a "chip" property pointing to the
chip object. Confusing bugs could arise if these ever go out of sync.
Change the property definition so that it explicitely sets the pointer.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157383336007.165747.1524120147081367440.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The homer object has both a pointer and a "chip" property pointing to the
chip object. Confusing bugs could arise if these ever go out of sync.
Change the property definition so that it explicitely sets the pointer.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157383335451.165747.32301068645427993.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The END source object has both a pointer and a "xive" property pointing to
the router object. Confusing bugs could arise if these ever go out of sync.
Change the property definition so that it explicitely sets the pointer.
The property isn't optional : not being able to set the link is a bug
and QEMU should rather abort than exit in this case.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157383333784.165747.5298512574054268786.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The source object has both a pointer and a "xive" property pointing to the
notifier object. Confusing bugs could arise if these ever go out of sync.
Change the property definition so that it explicitely sets the pointer.
The property isn't optional : not being able to set the link is a bug
and QEMU should rather abort than exit in this case.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157383333227.165747.12901571295951957951.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Commit 7e46261368 converted virtfs-proxy-helper to using libcap-ng. There
aren't any users of libcap anymore. No need to install libcap-dev.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Alex Bennée <alex.bennee@linaro.org>
When a VCPU gets connected to the XIVE interrupt controller, we add a
const link targetting the CPU object to the TCTX object. Similar links
are added to the ICP object when using the XICS interrupt controller.
As explained in <qom/object.h>:
* The caller must ensure that @target stays alive as long as
* this property exists. In the case @target is a child of @obj,
* this will be the case. Otherwise, the caller is responsible for
* taking a reference.
We're in the latter case for both XICS and XIVE. Add the missing
calls to object_ref() and object_unref().
This doesn't fix any known issue because the life cycle of the TCTX or
ICP happens to be shorter than the one of the CPU or XICS fabric, but
better safe than sorry.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <157192724770.3146912.15400869269097231255.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
SpaprInterruptControllerClass and PnvChipClass have an intc_create() method
that calls the appropriate routine, ie. icp_create() or xive_tctx_create(),
to establish the link between the VCPU and the presenter component of the
interrupt controller during realize.
There aren't any symmetrical call to be called when the VCPU gets unrealized
though. It is assumed that object_unparent() is the only thing to do.
This is questionable because the parenting logic around the CPU and
presenter objects is really an implementation detail of the interrupt
controller. It shouldn't be open-coded in the machine code.
Fix this by adding an intc_destroy() method that undoes what was done in
intc_create(). Also NULLify the presenter pointers to avoid having
stale pointers around. This will allow to reliably check if a vCPU has
a valid presenter.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157192724208.3146912.7254684777515287626.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
If memory allocation fails when using -mem-path, QEMU is supposed to print
out a message to indicate that fallback to anonymous RAM is deprecated. This
is done with error_printf() which does output buffering. As a consequence,
the message is only printed at the next flush, eg. when quiting QEMU, and
it also lacks a trailing newline:
qemu-system-ppc64: unable to map backing store for guest RAM: Cannot allocate memory
qemu-system-ppc64: warning: falling back to regular RAM allocation
QEMU 4.1.50 monitor - type 'help' for more information
(qemu) q
This is deprecated. Make sure that -mem-path specified path has sufficient resources to allocate -m specified RAM amountgreg@boss02:~/Work/qemu/qemu-spapr$
Add the missing \n to fix both issues.
Fixes: cb79224b7e "deprecate -mem-path fallback to anonymous RAM"
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <157304440026.351774.14607704217028190097.stgit@bahia.lan>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
The error message in object_class_property_add() was copied from
object_property_add() in commit 16bf7f522a. Clarify that it is
about a class, not an object.
While here, have the format string in both functions to fit in a
single line for better grep-ability, despite the checkpatch warning.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Message-Id: <157287383591.234942.311840593519058490.stgit@bahia.tlslab.ibm.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Since vCPUs aren't plugged into a bus, we manually register a reset
handler for each vCPU. We also call this handler at realize time
to ensure hot plugged vCPUs are reset before being exposed to the
guest. This results in vCPUs being reset twice at machine reset.
It doesn't break anything but it is slightly suboptimal and above
all confusing.
The hotplug path in device_set_realized() already knows how to reset
a hotplugged device if the device reset handler is present. Implement
one for sPAPR CPU cores that resets all vCPUs under a core.
While here rename spapr_cpu_reset() to spapr_reset_vcpu() for
consistency with spapr_realize_vcpu() and spapr_unrealize_vcpu().
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
[clg: add documentation on the reset helper usage ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20191022163812.330-3-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Support for setting VSMT is available in KVM since linux-4.13. Most distros
that support KVM on POWER already have it. It thus seem reasonable enough
to have the default machine to set VSMT to smp_threads.
This brings contiguous VCPU ids and thus brings their upper bound down to
the machine's max_cpus. This is especially useful for XIVE KVM devices,
which may thus allocate only one VP descriptor per VCPU.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157010411885.246126.12610015369068227139.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Some device types of the XICS model are exposed to the QEMU command
line:
$ ppc64-softmmu/qemu-system-ppc64 -device help | grep ic[sp]
name "icp"
name "ics"
name "ics-spapr"
name "pnv-icp", desc "PowerNV ICP"
These are internal devices that shouldn't be instantiable by the
user. By the way, they can't be because their respective realize
functions expect link properties that can't be set from the command
line:
qemu-system-ppc64: -device icp: required link 'xics' not found:
Property '.xics' not found
qemu-system-ppc64: -device ics: required link 'xics' not found:
Property '.xics' not found
qemu-system-ppc64: -device ics-spapr: required link 'xics' not found:
Property '.xics' not found
qemu-system-ppc64: -device pnv-icp: required link 'xics' not found:
Property '.xics' not found
Hide them by setting dc->user_creatable to false in the base class
"icp" and "ics" init functions.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157017826724.337875.14822177178282524024.stgit@bahia.lan>
Message-Id: <157045578962.865784.8551555523533955113.stgit@bahia.lan>
[dwg: Folded reason comment into base patch]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Some device types of the XIVE model are exposed to the QEMU command
line:
$ ppc64-softmmu/qemu-system-ppc64 -device help | grep xive
name "xive-end-source", desc "XIVE END Source"
name "xive-source", desc "XIVE Interrupt Source"
name "xive-tctx", desc "XIVE Interrupt Thread Context"
These are internal devices that shouldn't be instantiable by the
user. By the way, they can't be because their respective realize
functions expect link properties that can't be set from the command
line:
qemu-system-ppc64: -device xive-source: required link 'xive' not found:
Property '.xive' not found
qemu-system-ppc64: -device xive-end-source: required link 'xive' not found:
Property '.xive' not found
qemu-system-ppc64: -device xive-tctx: required link 'cpu' not found:
Property '.cpu' not found
Hide them by setting dc->user_creatable to false in their respective
class init functions.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157017473006.331610.2983143972519884544.stgit@bahia.lan>
Message-Id: <157045578401.865784.6058183726552779559.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
[dwg: Folded comment update into base patch]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Latest submissions for 9pfs made me realize that I no longer have time
and motivation to actively support it. I'll stay around for odd fixes
though.
Signed-off-by: Greg Kurz <groug@kaod.org>
It is more convenient to use the return value of the function to notify
errors, rather than to be tied up setting up the &local_err boilerplate.
Signed-off-by: Greg Kurz <groug@kaod.org>
Coverity is reporting in CID 1405304 that tpm_execute() may pass a NULL
tpm_proxy->host_path pointer to open(). This is based on the fact that
h_tpm_comm() does a NULL check on tpm_proxy->host_path and then passes
tpm_proxy to tpm_execute().
The check in h_tpm_comm() is abusive actually since a spapr-proxy-tpm
requires a non NULL host_path property, as checked during realize.
Fixes: 0fb6bd0732
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156805260916.1779401.11054185183758185247.stgit@bahia.lan>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
A recent change in spapr_machine_reset() showed that resetting the compat
mode in spapr_machine_reset() for the boot vCPU and in spapr_cpu_reset()
for all other vCPUs was fragile. The fix was thus to reset the compat mode
for all vCPUs in spapr_machine_reset(), but we still have to propagate
it to hot-plugged CPUs. This is still performed from spapr_cpu_reset(),
hence resulting in ppc_set_compat() being called twice for every vCPU at
machine reset. Apart from wasting cycles, which isn't really an issue
during machine reset, this seems to indicate that spapr_cpu_reset() isn't
the best place to set the compat mode.
A natural candidate for CPU-hotplug specific code is spapr_core_plug().
Also, it sits in the same file as spapr_machine_reset() : this makes
it easier for someone who wants to know when the compat PVR is set.
Call ppc_set_compat() from there. This doesn't need to be done for
initial vCPUs since the compat PVR is 0 and spapr_machine_reset() sets
the appropriate value later. No need to do this on manually added vCPUS
on the destination QEMU during migration since the compat PVR is
part of the migrated vCPU state. Both conditions can be checked with
spapr_drc_hotplugged().
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156701285312.499757.7807417667750711711.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
PHBs already take care of clearing the MSIs from the bitmap during reset
or unplug. No need to do this globally from the machine code. Rather add
an assert to ensure that PHBs have acted as expected.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156415228966.1064338.190189424190233355.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
[dwg: Fix crash in qtest case where spapr->irq_map can be NULL at the
new assert()]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When the machine is reset, the MSI bitmap is cleared but the allocated
MSIs are not freed. Some operating systems, such as AIX, can detect the
previous configuration and assert.
Empty the MSI cache, this performs the needed cleanup.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156415228410.1064338.4486161194061636096.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When freeing MSIs, we need to:
- remove them from the machine's MSI bitmap
- remove them from the IC backend
- remove them from the PHB's MSI cache
This is currently open coded in two places in rtas_ibm_change_msi(),
and we're about to need this in spapr_phb_reset() as well. Instead of
duplicating this code again, make it a destroy function for the PHB's
MSI cache. Removing an MSI device from the cache will call the destroy
function internally.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156415227855.1064338.5657793835271464648.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The ics_set_kvm_state_one() function is called either to restore the
state of an interrupt source during migration or to set the interrupt
source to a default state during reset.
Since always, ie. 2013, the code only sets the MASKED bit if the 'current
priority' and the 'saved priority' are different. This is likely true
when restoring an interrupt that had been previously masked with the
ibm,int-off RTAS call. However this is always false in the case of
reset since both 'current priority' and 'saved priority' are equal to
0xff, and the MASKED bit is never set.
The legacy KVM XICS device gets away with that because it ends updating
its internal structure the same way, whether the MASKED bit is set or
the priority is 0xff.
The XICS-on-XIVE device for POWER9 is different. It sticks to the KVM
documentation [1] and _really_ relies on the MASKED bit to correctly
set. If not, it will configure the interrupt source in the XIVE HW, even
though the guest hasn't configured the interrupt yet. This disturbs the
complex logic implemented in XICS-on-XIVE and may result in the loss of
subsequent queued events.
Always set the MASKED bit if interrupt is masked as expected by the KVM
XICS-on-XIVE device. This has no impact on the legacy KVM XICS.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/virtual/kvm/devices/xics.txt
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156217454083.559957.7359208229523652842.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Make kvmppc_xive_disconnect() able to undo the changes of a partial
execution of kvmppc_xive_connect() and use it to perform rollback.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <156198735673.293938.7313195993600841641.stgit@bahia>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Hot-unplugging a PHB with a VFIO device connected to it crashes QEMU:
-device spapr-pci-host-bridge,index=1,id=phb1 \
-device vfio-pci,host=0034:01:00.3,id=vfio0
(qemu) device_del phb1
[ 357.207183] iommu: Removing device 0001:00:00.0 from group 1
[ 360.375523] rpadlpar_io: slot PHB 1 removed
qemu-system-ppc64: memory.c:2742:
do_address_space_destroy: Assertion `QTAILQ_EMPTY(&as->listeners)' failed.
'as' is the IOMMU address space, which indeed has a listener registered
to by vfio_connect_container() when the VFIO device is realized. This
listener is supposed to be unregistered by vfio_disconnect_container()
when the VFIO device is finalized. Unfortunately, the VFIO device hasn't
reached finalize yet at the time the PHB unrealize function is called,
and address_space_destroy() gets called with the VFIO listener still
being registered.
All regions have just been unmapped from the address space. Listeners
aren't needed anymore at this point. Remove them before destroying the
address space.
The VFIO code will try to remove them _again_ at device finalize,
but it is okay since memory_listener_unregister() is idempotent.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156110925375.92514.11649846071216864570.stgit@bahia.lan>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[dwg: Correct spelling error pointed out by aik]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Make xics_kvm_disconnect() able to undo the changes of a partial execution
of xics_kvm_connect() and use it to perform rollback.
Note that kvmppc_define_rtas_kernel_token(0) never fails, no matter the
RTAS call has been defined or not.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156077922319.433243.609897156640506891.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Passing both errp and &local_err to functions is a recipe for messing
things up.
Since we must use &local_err for icp_kvm_realize(), use &local_err
everywhere where rollback must happen and have a single call to
error_propagate() them all. While here, add errno to the error
message.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156077921212.433243.11716701611944816815.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
kvm_enabled() expands to (0) when CONFIG_KVM is not defined. The first
CONFIG_KVM guard is thus useless and it is likely that the compiler
will optimize the code out in the case of the second guard. And even
if it doesn't, we have a stub for kvmppc_get_hypercall().
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156051054077.224162.9332715375637801197.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Greg Kurz