The way that Xen handles MSI PIRQs is kind of awful.
There is a special MSI message which targets a PIRQ. The vector in the
low bits of data must be zero. The low 8 bits of the PIRQ# are in the
destination ID field, the extended destination ID field is unused, and
instead the high bits of the PIRQ# are in the high 32 bits of the address.
Using the high bits of the address means that we can't intercept and
translate these messages in kvm_send_msi(), because they won't be caught
by the APIC — addresses like 0x1000fee46000 aren't in the APIC's range.
So we catch them in pci_msi_trigger() instead, and deliver the event
channel directly.
That isn't even the worst part. The worst part is that Xen snoops on
writes to devices' MSI vectors while they are *masked*. When a MSI
message is written which looks like it targets a PIRQ, it remembers
the device and vector for later.
When the guest makes a hypercall to bind that PIRQ# (snooped from a
marked MSI vector) to an event channel port, Xen *unmasks* that MSI
vector on the device. Xen guests using PIRQ delivery of MSI don't
ever actually unmask the MSI for themselves.
Now that this is working we can finally enable XENFEAT_hvm_pirqs and
let the guest use it all.
Tested with passthrough igb and emulated e1000e + AHCI.
CPU0 CPU1
0: 65 0 IO-APIC 2-edge timer
1: 0 14 xen-pirq 1-ioapic-edge i8042
4: 0 846 xen-pirq 4-ioapic-edge ttyS0
8: 1 0 xen-pirq 8-ioapic-edge rtc0
9: 0 0 xen-pirq 9-ioapic-level acpi
12: 257 0 xen-pirq 12-ioapic-edge i8042
24: 9600 0 xen-percpu -virq timer0
25: 2758 0 xen-percpu -ipi resched0
26: 0 0 xen-percpu -ipi callfunc0
27: 0 0 xen-percpu -virq debug0
28: 1526 0 xen-percpu -ipi callfuncsingle0
29: 0 0 xen-percpu -ipi spinlock0
30: 0 8608 xen-percpu -virq timer1
31: 0 874 xen-percpu -ipi resched1
32: 0 0 xen-percpu -ipi callfunc1
33: 0 0 xen-percpu -virq debug1
34: 0 1617 xen-percpu -ipi callfuncsingle1
35: 0 0 xen-percpu -ipi spinlock1
36: 8 0 xen-dyn -event xenbus
37: 0 6046 xen-pirq -msi ahci[0000:00:03.0]
38: 1 0 xen-pirq -msi-x ens4
39: 0 73 xen-pirq -msi-x ens4-rx-0
40: 14 0 xen-pirq -msi-x ens4-rx-1
41: 0 32 xen-pirq -msi-x ens4-tx-0
42: 47 0 xen-pirq -msi-x ens4-tx-1
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
This wires up the basic infrastructure but the actual interrupts aren't
there yet, so don't advertise it to the guest.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Just hook up the basic hypercalls to stubs in xen_evtchn.c for now.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Extract requests, return ENOSYS to all of them. This is enough to allow
older Linux guests to boot, as they need *something* back but it doesn't
matter much what.
A full implementation of a single-tentant internal XenStore copy-on-write
tree with transactions and watches is waiting in the wings to be sent in
a subsequent round of patches along with hooking up the actual PV disk
back end in qemu, but this is enough to get guests booting for now.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Just the basic shell, with the event channel hookup. It only dumps the
buffer for now; a real ring implmentation will come in a subsequent patch.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The provides the QEMU side of interdomain event channels, allowing events
to be sent to/from the guest.
The API mirrors libxenevtchn, and in time both this and the real Xen one
will be available through ops structures so that the PV backend drivers
can use the correct one as appropriate.
For now, this implementation can be used directly by our XenStore which
will be for emulated mode only.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Introduce support for one shot and periodic mode of Xen PV timers,
whereby timer interrupts come through a special virq event channel
with deadlines being set through:
1) set_timer_op hypercall (only oneshot)
2) vcpu_op hypercall for {set,stop}_{singleshot,periodic}_timer
hypercalls
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The guest is permitted to specify an arbitrary domain/bus/device/function
and INTX pin from which the callback IRQ shall appear to have come.
In QEMU we can only easily do this for devices that actually exist, and
even that requires us "knowing" that it's a PCMachine in order to find
the PCI root bus — although that's OK really because it's always true.
We also don't get to get notified of INTX routing changes, because we
can't do that as a passive observer; if we try to register a notifier
it will overwrite any existing notifier callback on the device.
But in practice, guests using PCI_INTX will only ever use pin A on the
Xen platform device, and won't swizzle the INTX routing after they set
it up. So this is just fine.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The GSI callback (and later PCI_INTX) is a level triggered interrupt. It
is asserted when an event channel is delivered to vCPU0, and is supposed
to be cleared when the vcpu_info->evtchn_upcall_pending field for vCPU0
is cleared again.
Thankfully, Xen does *not* assert the GSI if the guest sets its own
evtchn_upcall_pending field; we only need to assert the GSI when we
have delivered an event for ourselves. So that's the easy part, kind of.
There's a slight complexity in that we need to hold the BQL before we
can call qemu_set_irq(), and we definitely can't do that while holding
our own port_lock (because we'll need to take that from the qemu-side
functions that the PV backend drivers will call). So if we end up
wanting to set the IRQ in a context where we *don't* already hold the
BQL, defer to a BH.
However, we *do* need to poll for the evtchn_upcall_pending flag being
cleared. In an ideal world we would poll that when the EOI happens on
the PIC/IOAPIC. That's how it works in the kernel with the VFIO eventfd
pairs — one is used to trigger the interrupt, and the other works in the
other direction to 'resample' on EOI, and trigger the first eventfd
again if the line is still active.
However, QEMU doesn't seem to do that. Even VFIO level interrupts seem
to be supported by temporarily unmapping the device's BARs from the
guest when an interrupt happens, then trapping *all* MMIO to the device
and sending the 'resample' event on *every* MMIO access until the IRQ
is cleared! Maybe in future we'll plumb the 'resample' concept through
QEMU's irq framework but for now we'll do what Xen itself does: just
check the flag on every vmexit if the upcall GSI is known to be
asserted.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Specifically add listing, injection of event channels.
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Acked-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Add the array of virq ports to each vCPU so that we can deliver timers,
debug ports, etc. Global virqs are allocated against vCPU 0 initially,
but can be migrated to other vCPUs (when we implement that).
The kernel needs to know about VIRQ_TIMER in order to accelerate timers,
so tell it via KVM_XEN_VCPU_ATTR_TYPE_TIMER. Also save/restore the value
of the singleshot timer across migration, as the kernel will handle the
hypercalls automatically now.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
This finally comes with a mechanism for actually injecting events into
the guest vCPU, with all the atomic-test-and-set that's involved in
setting the bit in the shinfo, then the index in the vcpu_info, and
injecting either the lapic vector as MSI, or letting KVM inject the
bare vector.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
It calls an internal close_port() helper which will also be used from
EVTCHNOP_reset and will actually do the work to disconnect/unbind a port
once any of that is actually implemented in the first place.
That in turn calls a free_port() internal function which will be in
error paths after allocation.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
This adds the basic structure for maintaining the port table and reporting
the status of ports therein.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Include basic support for setting HVM_PARAM_CALLBACK_IRQ to the global
vector method HVM_PARAM_CALLBACK_TYPE_VECTOR, which is handled in-kernel
by raising the vector whenever the vCPU's vcpu_info->evtchn_upcall_pending
flag is set.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
Xen will "latch" the guest's 32-bit or 64-bit ("long mode") setting when
the guest writes the MSR to fill in the hypercall page, or when the guest
sets the event channel callback in HVM_PARAM_CALLBACK_IRQ.
KVM handles the former and sets the kernel's long_mode flag accordingly.
The latter will be handled in userspace. Keep them in sync by noticing
when a hypercall is made in a mode that doesn't match qemu's idea of
the guest mode, and resyncing from the kernel. Do that same sync right
before serialization too, in case the guest has set the hypercall page
but hasn't yet made a system call.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
For the shared info page and for grant tables, Xen shares its own pages
from the "Xen heap" to the guest. The guest requests that a given page
from a certain address space (XENMAPSPACE_shared_info, etc.) be mapped
to a given GPA using the XENMEM_add_to_physmap hypercall.
To support that in qemu when *emulating* Xen, create a memory region
(migratable) and allow it to be mapped as an overlay when requested.
Xen theoretically allows the same page to be mapped multiple times
into the guest, but that's hard to track and reinstate over migration,
so we automatically *unmap* any previous mapping when creating a new
one. This approach has been used in production with.... a non-trivial
number of guests expecting true Xen, without any problems yet being
noticed.
This adds just the shared info page for now. The grant tables will be
a larger region, and will need to be overlaid one page at a time. I
think that means I need to create separate aliases for each page of
the overall grant_frames region, so that they can be mapped individually.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
The ioapic sources reside in hw/intc already. Move the headers there
as well.
Signed-off-by: Bernhard Beschow <shentey@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20230213173033.98762-11-shentey@gmail.com>
[PMD: Keep ioapic_internal.h in hw/intc/, not under include/]
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
These IRQ counting functions will soon be required in binaries that
do not include the APIC code, too, so let's extract them into a
separate file that can be linked independently of the APIC code.
While we're at it, change the apic_* prefix into kvm_* since the
functions are used from the i8259 PIC (i.e. not the APIC), too.
Reviewed-by: Bernhard Beschow <shentey@gmail.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Message-Id: <20230110095351.611724-2-thuth@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Drop abs64() and use uabs64() from host-utils, which avoids
an undefined behavior when taking abs of the most negative value.
Signed-off-by: Luis Pires <luis.pires@eldorado.org.br>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20210910112624.72748-5-luis.pires@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Stop including cpu.h in files that don't need it.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210416171314.2074665-4-thuth@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Many files include hw/irq.h without needing it. Remove the superfluous
include statements.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20210327050236.2232347-1-thuth@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
The 'running' argument from VMChangeStateHandler does not require
other value than 0 / 1. Make it a plain boolean.
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <20210111152020.1422021-3-philmd@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Move the property types and property macros implemented in
qdev-properties-system.c to a new qdev-properties-system.h
header.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20201211220529.2290218-16-ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
The IOAPIC has an 'Extended Destination ID' field in its RTE, which maps
to bits 11-4 of the MSI address. Since those address bits fall within a
given 4KiB page they were historically non-trivial to use on real hardware.
The Intel IOMMU uses the lowest bit to indicate a remappable format MSI,
and then the remaining 7 bits are part of the index.
Where the remappable format bit isn't set, we can actually use the other
seven to allow external (IOAPIC and MSI) interrupts to reach up to 32768
CPUs instead of just the 255 permitted on bare metal.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <78097f9218300e63e751e077a0a5ca029b56ba46.camel@infradead.org>
[Fix UBSAN warning. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
KVM_CAP_IRQ_ROUTING is always available on x86, so replace checks
for kvm_has_gsi_routing() and KVM_CAP_IRQ_ROUTING with asserts.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200922201922.2153598-3-ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
QEMU's kvmclock device is only created when KVM PV feature bits for
kvmclock (KVM_FEATURE_CLOCKSOURCE/KVM_FEATURE_CLOCKSOURCE2) are
exposed to the guest. With 'kvm=off' cpu flag the device is not
created and we don't call KVM_GET_CLOCK/KVM_SET_CLOCK upon migration.
It was reported that without these call at least Hyper-V TSC page
clocksouce (which can be enabled independently) gets broken after
migration.
Switch to creating kvmclock QEMU device unconditionally, it seems
to always make sense to call KVM_GET_CLOCK/KVM_SET_CLOCK on migration.
Use KVM_CAP_ADJUST_CLOCK check instead of CPUID feature bits.
Reported-by: Antoine Damhet <antoine.damhet@blade-group.com>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200922151934.899555-1-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Restricting LostTickPolicy to machine.json pulls slightly less
QAPI-generated code into user-mode.
Acked-by: Markus Armbruster <armbru@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200913195348.1064154-2-philmd@redhat.com>
[Add rationale to commit message]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Some typedefs and macros are defined after the type check macros.
This makes it difficult to automatically replace their
definitions with OBJECT_DECLARE_TYPE.
Patch generated using:
$ ./scripts/codeconverter/converter.py -i \
--pattern=QOMStructTypedefSplit $(git grep -l '' -- '*.[ch]')
which will split "typdef struct { ... } TypedefName"
declarations.
Followed by:
$ ./scripts/codeconverter/converter.py -i --pattern=MoveSymbols \
$(git grep -l '' -- '*.[ch]')
which will:
- move the typedefs and #defines above the type check macros
- add missing #include "qom/object.h" lines if necessary
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20200831210740.126168-9-ehabkost@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20200831210740.126168-10-ehabkost@redhat.com>
Message-Id: <20200831210740.126168-11-ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Fix a typo in an error message for KVM_SET_IRQCHIP ioctl:
"KVM_GET_IRQCHIP" should be "KVM_SET_IRQCHIP".
Fixes: a39c1d47ac ("kvm: x86: Add user space part for in-kernel IOAPIC")
Signed-off-by: Kenta Ishiguro <kentaishiguro@slowstart.org>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Li Qiang <liq3ea@gmail.com>
Message-Id: <20200717123514.15406-1-kentaishiguro@slowstart.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Each architecture's sourceset is placed in an hw_arch dictionary, and picked up
from there when building the per-emulator static_library.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Devices may have component devices and buses.
Device realization may fail. Realization is recursive: a device's
realize() method realizes its components, and device_set_realized()
realizes its buses (which should in turn realize the devices on that
bus, except bus_set_realized() doesn't implement that, yet).
When realization of a component or bus fails, we need to roll back:
unrealize everything we realized so far. If any of these unrealizes
failed, the device would be left in an inconsistent state. Must not
happen.
device_set_realized() lets it happen: it ignores errors in the roll
back code starting at label child_realize_fail.
Since realization is recursive, unrealization must be recursive, too.
But how could a partly failed unrealize be rolled back? We'd have to
re-realize, which can fail. This design is fundamentally broken.
device_set_realized() does not roll back at all. Instead, it keeps
unrealizing, ignoring further errors.
It can screw up even for a device with no buses: if the lone
dc->unrealize() fails, it still unregisters vmstate, and calls
listeners' unrealize() callback.
bus_set_realized() does not roll back either. Instead, it stops
unrealizing.
Fortunately, no unrealize method can fail, as we'll see below.
To fix the design error, drop parameter @errp from all the unrealize
methods.
Any unrealize method that uses @errp now needs an update. This leads
us to unrealize() methods that can fail. Merely passing it to another
unrealize method cannot cause failure, though. Here are the ones that
do other things with @errp:
* virtio_serial_device_unrealize()
Fails when qbus_set_hotplug_handler() fails, but still does all the
other work. On failure, the device would stay realized with its
resources completely gone. Oops. Can't happen, because
qbus_set_hotplug_handler() can't actually fail here. Pass
&error_abort to qbus_set_hotplug_handler() instead.
* hw/ppc/spapr_drc.c's unrealize()
Fails when object_property_del() fails, but all the other work is
already done. On failure, the device would stay realized with its
vmstate registration gone. Oops. Can't happen, because
object_property_del() can't actually fail here. Pass &error_abort
to object_property_del() instead.
* spapr_phb_unrealize()
Fails and bails out when remove_drcs() fails, but other work is
already done. On failure, the device would stay realized with some
of its resources gone. Oops. remove_drcs() fails only when
chassis_from_bus()'s object_property_get_uint() fails, and it can't
here. Pass &error_abort to remove_drcs() instead.
Therefore, no unrealize method can fail before this patch.
device_set_realized()'s recursive unrealization via bus uses
object_property_set_bool(). Can't drop @errp there, so pass
&error_abort.
We similarly unrealize with object_property_set_bool() elsewhere,
always ignoring errors. Pass &error_abort instead.
Several unrealize methods no longer handle errors from other unrealize
methods: virtio_9p_device_unrealize(),
virtio_input_device_unrealize(), scsi_qdev_unrealize(), ...
Much of the deleted error handling looks wrong anyway.
One unrealize methods no longer ignore such errors:
usb_ehci_pci_exit().
Several realize methods no longer ignore errors when rolling back:
v9fs_device_realize_common(), pci_qdev_unrealize(),
spapr_phb_realize(), usb_qdev_realize(), vfio_ccw_realize(),
virtio_device_realize().
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: <20200505152926.18877-17-armbru@redhat.com>
Both gsi_handler() and kvm_pc_gsi_handler() have the same content,
except one comment. Move the comment, and de-duplicate the code.
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These are needed by microvm too, so move them outside of PC-specific files.
With this patch, microvm.c need not include pc.h anymore.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
