PR KVM supports an ePAPR compliant hypercall interface in parallel to the
normal sPAPR one. Expose the ePAPR /hypervisor node and properties to the
guest so it can use it.
This enables magic page sharing on PR KVM with -M pseries.
However we had a few nasty bugs in the magic page implementation on vcpus
newer than 970 (p7, p8) that KVM now has workarounds for. It indicates that
it does have these workarounds through the PPC_FIXUP_HCALL capability.
To not expose broken guest kernels to issues on host kernels that don't
have the fixups in place, we don't expose working hypercall instructions
when the fixups are not available so that the guest can never active the
magic page.
Signed-off-by: Alexander Graf <agraf@suse.de>
This adds @bus_offset into sPAPRTCETable to tell where TCE table starts
from. It is set to 0 for emulated devices. Dynamic DMA windows will use
other offset.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
At the moment only 4K pages are supported by sPAPRTCETable. Since sPAPR
spec allows other page sizes and we are going to implement them, we need
page size to be configrable.
This adds @page_shift into sPAPRTCETable and replaces SPAPR_TCE_PAGE_SHIFT
with it where it is possible.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
This removes window_size as it is basically a copy of nb_table
shifted by SPAPR_TCE_PAGE_SHIFT. As new dynamic DMA windows are
going to support windows as big as the entire RAM and this number
will be bigger that 32 capacity, we will have to do something
about @window_size anyway and removal seems to be the right way to go.
This removes dma_window_start/dma_window_size from sPAPRPHBState as
they are no longer used.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
qdev_init_nofail() was replaced by object_property_set_bool("realized")
all over the QEMU so do we.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
At the moment sPAPRPHBState contains a @tcet pointer to the only
TCE table. However sPAPR spec allows having more than one DMA window.
Since the TCE object is already a child of SPAPR PHB object, there is
no need to keep an additional pointer to it in sPAPRPHBState so remove it.
This changes the way sPAPRPHBState::reset performs reset of sPAPRTCETable
objects.
This changes the default DMA window properties calculation.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently the default DMA window is represented by a single MemoryRegion.
However there can be more than just one window so we need
a "root" memory region to be separated from the actual DMA window(s).
This introduces a "root" IOMMU memory region and adds a subregion for
the default DMA 32bit window. Following patches will add other
subregion(s).
This initializes a default DMA window subregion size to the guest RAM
size as this window can be switched into "bypass" mode which implements
direct DMA mapping.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
The spapr-pci PHB initializes IOMMU for emulated devices only.
The upcoming VFIO support will do it different. However both emulated
and VFIO PHB types share most of the initialization code.
For the type specific things a new finish_realize() callback is
introduced.
This introduces sPAPRPHBClass derived from PCIHostBridgeClass and
adds the callback pointer.
This implements finish_realize() for emulated devices.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[agraf: Fix compilation]
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently only single TCE entry per request is supported (H_PUT_TCE).
However PAPR+ specification allows multiple entry requests such as
H_PUT_TCE_INDIRECT and H_STUFF_TCE. Having less transitions to the host
kernel via ioctls, support of these calls can accelerate IOMMU operations.
This implements H_STUFF_TCE and H_PUT_TCE_INDIRECT.
This advertises "multi-tce" capability to the guest if the host kernel
supports it (KVM_CAP_SPAPR_MULTITCE) or guest is running in TCG mode.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
At the moment the "ibm,hypertas-functions" list is fixed. However some
calls should be listed there if they are supported by QEMU or the host
kernel.
This enables hyperrtas_prop to grow on stack by adding
a SPAPR_HYPERRTAS_ADD macro. "qemu,hypertas-functions" is converted as well.
The first user of this is going to be a "multi-tce" property.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
SPAPR IOMMU is a bus-less device and therefore its only ID in
migration stream is an instance id which is not reliable ID
as it depends on the command line parameters order. Since
libvirt may change the order, we need something better than that.
This removes VMSD descriptor from the class definitiion and
registers it with @liobn as an intance ID to let the destination
side find the right device to receive migration data.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
Modern Linux kernels support last POWERPC CPUs so when a kernel boots,
in most cases it can find a matching cpu_spec in the kernel's cpu_specs
list. However if the kernel is quite old, it may be missing a definition
of the actual CPU. To provide an ability for old kernels to work on modern
hardware, a Processor Compatibility Mode has been introduced
by the PowerISA specification.
>From the hardware prospective, it is supported by the Processor
Compatibility Register (PCR) which is defined in PowerISA. The register
enables one of the compatibility modes (2.05/2.06/2.07).
Since PCR is a hypervisor privileged register and cannot be
directly accessed from the guest, the mode selection is done via
ibm,client-architecture-support (CAS) RTAS call using which the guest
specifies what "raw" and "architected" CPU versions it supports.
QEMU works out the best match, changes a "cpu-version" property of
every CPU and notifies the guest about the change by setting these
properties in the buffer passed as a response on a custom H_CAS hypercall.
This implements ibm,client-architecture-support parameters parsing
(now only for PVRs) and cooks the device tree diff with new values for
"cpu-version", "ibm,ppc-interrupt-server#s" and
"ibm,ppc-interrupt-server#s" properties.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
This puts a limit to the number of threads per core based on the current
compatibility mode. Although PowerISA specs do not specify the maximum
threads per core number, the linux guest still expects that
PowerISA2.05-compatible CPU supports only 2 threads per core as this
is what POWER6 (2.05 compliant CPU) implements, the same is for
POWER7 (2.06, 4 threads) and POWER8 (2.07, 8 threads).
This calls spapr_fixup_cpu_smt_dt() with the maximum allowed number of
threads which affects ibm,ppc-interrupt-server#s and
ibm,ppc-interrupt-gserver#s properties.
The number of CPU nodesremains unchanged.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
In PPC code we usually use the "cs" name for a CPUState* variables
and "cpu" for PowerPCCPU. So let's change spapr_fixup_cpu_dt() to
use same rules as spapr_create_fdt_skel() does.
This adds missing nodes creation if they do not already exist in
the current device tree, this is going to be used from
the client-architecture-support handler.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
The PAPR+ specification defines a ibm,client-architecture-support (CAS)
RTAS call which purpose is to provide a negotiation mechanism for
the guest and the hypervisor to work out the best compatibility parameters.
During the negotiation process, the guest provides an array of various
options and capabilities which it supports, the hypervisor adjusts
the device tree and (optionally) reboots the guest.
At the moment the Linux guest calls CAS method at early boot so SLOF
gets called. SLOF allocates a memory buffer for the device tree changes
and calls a custom KVMPPC_H_CAS hypercall. QEMU parses the options,
composes a diff for the device tree, copies it to the buffer provided
by SLOF and returns to SLOF. SLOF updates the device tree and returns
control to the guest kernel. Only then the Linux guest parses the device
tree so it is possible to avoid unnecessary reboot in most cases.
The device tree diff is a header with an update format version
(defined as 1 in this patch) followed by a device tree with the properties
which require update.
If QEMU detects that it has to reboot the guest, it silently does so
as the guest expects reboot to happen because this is usual pHyp firmware
behavior.
This defines custom KVMPPC_H_CAS hypercall. The current SLOF already
has support for it.
This implements stub which returns very basic tree (root node,
no properties) to the guest.
As the return buffer does not contain any change, no change in behavior is
expected.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
This adds basic support for the "compat" CPU option. By specifying
the compat property, the user can manually switch guest CPU mode from
"raw" to "architected".
This defines feature disable bits which are not used yet as, for example,
PowerISA 2.07 says if 2.06 mode is selected, the TM bit does not matter -
transactional memory (TM) will be disabled because 2.06 does not define
it at all. The same is true for VSX and 2.05 mode. So just setting a mode
must be ok.
This does not change the existing behavior as the actual compatibility
mode support is coming in next patches.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[agraf: fix compilation on 32bit hosts]
Signed-off-by: Alexander Graf <agraf@suse.de>
The upcoming support of the "ibm,client-architecture-support"
reconfiguration call will be able to change dynamically the number
of threads per core (SMT mode). From the device tree prospective
this does not change the number of CPU nodes (as it is one node per
a CPU core) but affects content and size of the ibm,ppc-interrupt-server#s
and ibm,ppc-interrupt-gserver#s properties.
This moves ibm,ppc-interrupt-server#s and ibm,ppc-interrupt-gserver#s
out of the device tree skeleton.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
This adds a "ibm,chip-id" property for CPU nodes which should be the same
for all cores in the same CPU socket. The recent guest kernels use this
information to associate threads with sockets.
Refer to the kernel commit 256f2d4b463d3030ebc8d2b54f427543814a2bdc
for more details.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
This allows guests to have a different timebase origin from the host.
This is needed for migration, where a guest can migrate from one host
to another and the two hosts might have a different timebase origin.
However, the timebase seen by the guest must not go backwards, and
should go forwards only by a small amount corresponding to the time
taken for the migration.
This is only supported for recent POWER hardware which has the TBU40
(timebase upper 40 bits) register. That includes POWER6, 7, 8 but not
970.
This adds kvm_access_one_reg() to access a special register which is not
in env->spr. This requires kvm_set_one_reg/kvm_get_one_reg patch.
The feature must be present in the host kernel.
This bumps vmstate_spapr::version_id and enables new vmstate_ppc_timebase
only for it. Since the vmstate_spapr::minimum_version_id remains
unchanged, migration from older QEMU is supported but without
vmstate_ppc_timebase.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
Almost all platforms QEMU emulates have some sort of firmware they can load
to expose a guest environment that closely resembles the way it would look
like on real hardware.
This patch introduces such a firmware on our e500 platforms. U-boot is the
default firmware for most of these systems and as such our preferred choice.
For backwards compatibility reasons (and speed and simplicity) we skip u-boot
when you use -kernel and don't pass in -bios. For all other combinations like
-kernel and -bios or no -kernel you get u-boot as firmware.
This allows you to modify the boot environment, execute a networked boot through
the e1000 emulation and execute u-boot payloads.
Signed-off-by: Alexander Graf <agraf@suse.de>
We want to move to a model where firmware loads our kernel. To achieve
this we need to be able to tell firmware where the kernel lies.
Let's copy the mechanism we already use for -M pseries and expose the
kernel load address and size through the device tree.
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch adds pci pin to irq_num routing callback.
This callback is called from pci_device_route_intx_to_irq to
find which pci device maps to which irq.
This fix is required for pci-device passthrough using vfio.
Also without this patch we gets below prints
"
PCI: Bug - unimplemented PCI INTx routing (e500-pcihost)
qemu-system-ppc64: PCI: Bug - unimplemented PCI INTx routing (e500-pcihost) "
and Legacy interrupt does not work with pci device passthrough.
Signed-off-by: Bharat Bhushan <Bharat.Bhushan@freescale.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
[agraf: remove double semicolon]
Signed-off-by: Alexander Graf <agraf@suse.de>
- Use PCI_NUM_PINS rather than hardcoding
- use "pin" wherever possible
Signed-off-by: Bharat Bhushan <Bharat.Bhushan@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
At the moment XICS does not support interrupts reuse so sPAPR PHB
implements this. sPAPRPHBState holds array of 32 spapr_pci_msi to
describe PCI config address, first MSI and number of MSIs. Once
allocated for a device, QEMU tries reusing this config until the number
of MSIs changes.
Existing SPAPR guests call ibm,change-msi in a loop until the handler
returns the requested number of vectors.
Recently introduced check for the maximum number of MSI/MSIX vectors
supported by a device only works for a device which is new for PHB's
MSI cache. If it is already there, the check is not performed which
leads to new IRQ block allocation. This happens during PCI hotplug
even when the user hot plug the same device which he just hot unplugged.
This moves the check earlier.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
While there, also moved the hard coded value for CLOCKFREQ to a #define.
Signed-off-by: BALATON Zoltan <balaton@eik.bme.hu>
Signed-off-by: Alexander Graf <agraf@suse.de>
Current guest kernels try allocating as many vectors as the quota is.
For example, in the case of virtio-net (which has just 3 vectors)
the guest requests 4 vectors (that is the quota in the test) and
the existing ibm,change-msi handler returns 4. But before it returns,
it calls msix_set_message() in a loop and corrupts memory behind
the end of msix_table.
This limits the number of vectors returned by ibm,change-msi to
the maximum supported by the actual device.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Cc: qemu-stable@nongnu.org
[agraf: squash in bugfix from aik]
Signed-off-by: Alexander Graf <agraf@suse.de>
In the past, IO space could not be mapped into the memory address space
so we introduced a workaround for that. Nowadays it does not look
necessary so we can remove the workaround and make sPAPR PCI
configuration simplier.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Acked-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
After previous Peter patch, they are redundant. This way we don't
assign them except when needed. Once there, there were lots of case
where the ".fields" indentation was wrong:
.fields = (VMStateField []) {
and
.fields = (VMStateField []) {
Change all the combinations to:
.fields = (VMStateField[]){
The biggest problem (appart from aesthetics) was that checkpatch complained
when we copy&pasted the code from one place to another.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Acked-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Total removal of QEMUMachineInitArgs struct. QEMUMachineInitArgs's fields
are copied into MachineState. Removed duplicated fields from MachineState.
All the other changes are only mechanical refactoring, no semantic changes.
Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com> (s390)
Reviewed-by: Michael S. Tsirkin <mst@redhat.com> (PC)
[AF: Renamed ms -> machine, use MACHINE_GET_CLASS()]
Signed-off-by: Andreas Färber <afaerber@suse.de>
This fixes warnings from the static code analysis (smatch).
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
PortioList is an abstraction used for construction of MemoryRegionPortioList
from MemoryRegionPortio. It can be used later to unmap created memory regions.
It also requires proper cleanup because some of the memory inside is allocated
dynamically.
By moving PortioList ot device state we make it possible to cleanup later and
avoid leaking memory.
This change spans several target platforms. The following testcases cover all
changed lines:
qemu-system-ppc -M prep
qemu-system-i386 -vga qxl
qemu-system-i386 -M isapc -soundhw adlib -device ib700,id=watchdog0,bus=isa.0
Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
No need to go through qemu_machine field. Use
MachineClass fields directly.
Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
In order to eliminate the QEMUMachine indirection,
add its fields directly to MachineClass.
Do not yet remove qemu_machine field because it is
still in use by sPAPR.
Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>
[AF: Copied fields for sPAPR, too]
Signed-off-by: Andreas Färber <afaerber@suse.de>
The spinning struct is in guest endianness, so we need to initialize
its variables in guest endianness too.
This fixes booting e500 guests with SMP on x86 for me.
Signed-off-by: Alexander Graf <agraf@suse.de>
There are 3 different variants of the decrementor for BookE and BookS.
The BookE variant sets TSR[DIS] to 1 when the DEC value becomes 1 or 0. TSR[DIS]
is then the indicator whether the decrementor interrupt line is asserted or not.
The old BookS variant treats DEC as an edge interrupt that gets triggered when
the DEC value's top bit turns 1 from 0.
The new BookS variant maintains the assertion bit inside DEC itself. Whenever
the DEC value becomes negative (top bit set) the DEC interrupt line is asserted.
So far we implemented mostly the old BookS variant. Let's do them all properly.
This fixes booting pseries ppc64 guest images in TCG mode for me.
Signed-off-by: Alexander Graf <agraf@suse.de>
We now reset SPRs to their reset values on CPU reset. So if we want
to have an SPR persistently changed, we need to change its default
reset value rather than the value itself manually.
Do this for SPR_BOOKE_PIR, fixing e500v2 SMP boot.
Reported-by: Frederic Konrad <fred.konrad@greensocs.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Tested-by: KONRAD Frederic <fred.konrad@greensocs.com>
Add U suffix to various places where we were doing "1 << 31",
which is undefined behaviour, and also to other constant
definitions in the same groups, for consistency.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
This extends the pseries machine type with the interface to fix firmware
pathnames for devices which have @bootindex property.
This fixes SCSI disks' device node names (which are wildcard nodes in
the device-tree), for spapr-vscsi, virtio-scsi and usb-storage.
This fixes PHB name from "pci" to "pci@XXXX" where XXXX is a BUID as
there is no bus on top of sPAPRPHBState where PHB firmware name could
be fixed using the BusClass::get_fw_dev_path() mechanism.
This stores the boot list in the /chosen/qemu,boot-list property of
the device tree. "\n" are replaced by spaces to support OF1275.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This changes VIO bridge fw name from spapr-vio-bridge to vdevice and
vscsi/veth node names from QEMU object names to VIO specific device tree
names.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This changes resource code definitions to ones used in the host kernel.
This fixes H_SET_MODE_RESOURCE_LE (switch between big endian and
little endian) to sync registers from KVM before changing LPCR value.
This adds a set_spr() helper to update an SPR in a CPU's context to avoid
possible races and makes use of it to change LPCR.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
We wanted to loop till index is 8. On 8 we return with H_PTEG_FULL. If we
are successful in loading hpte with any other index, we continue with that
index value.
Reported-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
PCI memory region is 0x3f000000 bytes starting at 0xc0000000.
However, keep compatibility with Open Hack'Ware expectations
by adding a hack for Open Hack'Ware display.
Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Remove now duplicated code from prep board.
Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Before spapr_vga_init will returned false if the vga is specified by
the command '-device VGA' because vga_interface_type was evaluated to
VGA_NONE. With the change in previous patch of this series,
spapr_vga_init should return true if it's told that the vga will be
initialized in flow of the generic devices initialization.
To keep '-nodefaults' have the semantics of bare minimum, it adds a
check of 'has_defaults' in usb_enabled() to avoid that a USB controller
is added by '-nodefautls, -device VGA' implicitly.
This patch also makes two cleanups:
1. skip initialization for VGA_NONE
2. remove the useless 'break'
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mark Wu <wudxw@linux.vnet.ibm.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Raven datasheet explains where firmware lives in system memory, so do
it there instead of in board code. Other boards using the same PCI
host will not have to copy the firmware loading code.
Signed-off-by: Hervé Poussineau <hpoussin@reactos.org>
[AF: Drop BIOS size workaround in favor of replacing our firmware blob]
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Alexander Graf