This implements H_XIRR_X hypercall in addition to H_XIRR as
it is mandatory for PAPR+ and there is no way for the guest to
detect whether it is supported or not so just add it.
As the Partition Adjunct Option is not supported at the moment,
the CPPR parameter of the hypercall is ignored.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
Recent (host) kernels support emulating the PAPR defined "XICS" interrupt
controller system within KVM. This patch allows qemu to initialize and
configure the in-kernel XICS, and keep its state in sync with qemu's XICS
state as necessary.
This should give considerable performance improvements. e.g. on a simple
IPI ping-pong test between hardware threads, using qemu XICS gives us
around 5,000 irqs/second, whereas the in-kernel XICS gives us around
70,000 irqs/s on the same hardware configuration.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
[Mike Qiu <qiudayu@linux.vnet.ibm.com>: fixed mistype which caused ics_set_kvm_state() to fail]
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
This adds a cpu_setup callback to the XICS device class (as XICS-KVM
will do it different), xics_cpu_setup() will call it if it is set.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
The upcoming XICS-KVM support will use bits of emulated XICS code.
So this introduces new level of hierarchy - "xics-common" class. Both
emulated XICS and XICS-KVM will inherit from it and override class
callbacks when required.
The new "xics-common" class implements:
1. replaces static "nr_irqs" and "nr_servers" properties with
the dynamic ones and adds callbacks to be executed when properties
are set.
2. xics_cpu_setup() callback renamed to xics_common_cpu_setup() as
it is a common part for both XICS'es
3. xics_reset() renamed to xics_common_reset() for the same reason.
The emulated XICS changes:
1. the part of xics_realize() which creates ICPs is moved to
the "nr_servers" property callback as realize() is too late to
create/initialize devices and instance_init() is too early to create
devices as the number of child devices comes via the "nr_servers"
property.
2. added ics_initfn() which does a little part of what xics_realize() did.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
The upcoming support of in-kernel XICS will redefine migration callbacks
for both ICS and ICP so classes and callback pointers are added.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
On the real hardware, RTAS is called in real mode and therefore
top 4 bits of the address passed in the call are ignored.
So does the patch.
This converts h_rtas() to use existing rtas_ld() handlers.
This fixed rtas_ld()/rtas_st() to ignore top 4 bits.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
H_SET_MODE is used for controlling various partition settings. One
of these settings is the endianness a guest takes its exceptions in.
Signed-off-by: Anton Blanchard <anton@samba.org>
[agraf: fix whitespace]
Signed-off-by: Alexander Graf <agraf@suse.de>
On the sPAPR platform a guest allocates MSI/MSIX vectors via RTAS
hypercalls which return global IRQ numbers to a guest so it only
operates with those and never touches MSIMessage.
Therefore MSIMessage handling is completely hidden in QEMU.
Previously every sPAPR PCI host bridge implemented its own MSI window
to catch msi_notify()/msix_notify() calls from QEMU devices (virtio-pci
or vfio) and route them to the guest via qemu_pulse_irq().
MSIMessage used to be encoded as:
.addr - address within the PHB MSI window;
.data - the device index on PHB plus vector number.
The MSI MR write function translated this MSIMessage to a global IRQ
number and called qemu_pulse_irq().
However the total number of IRQs is not really big (at the moment it is
1024 IRQs starting from 4096) and even 16bit data field of MSIMessage
seems to be enough to store an IRQ number there.
This simplifies MSI handling in sPAPR PHB. Specifically, this does:
1. remove a MSI window from a PHB;
2. add a single memory region for all MSIs to sPAPREnvironment
and spapr_pci_msi_init() to initialize it;
3. encode MSIMessage as:
* .addr - a fixed address of SPAPR_PCI_MSI_WINDOW==0x40000000000ULL;
* .data as an IRQ number.
4. change IRQ allocator to align first IRQ number in a block for MSI.
MSI uses lower bits to specify the vector number so the first IRQ has to
be aligned. MSIX does not need any special allocator though.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Basically, in HW the layout of the interrupt network is:
- One ICP per processor thread (the "presenter"). This contains the
registers to fetch a pending interrupt (ack), EOI, and control the
processor priority.
- One ICS per logical source of interrupts (ie, one per PCI host
bridge, and a few others here or there). This contains the per-interrupt
source configuration (target processor(s), priority, mask) and the
per-interrupt internal state.
Under PAPR, there is a single "virtual" ICS ... somewhat (it's a bit
oddball what pHyp does here, arguably there are two but we can ignore
that distinction). There is no register level access. A pair of firmware
(RTAS) calls is used to configure each virtual interrupt.
So our model here is somewhat the same. We have one ICS in the emulated
XICS which arguably *is* the emulated XICS, there's no point making it a
separate "device", that would just be gross, and each VCPU has an
associated ICP.
Yet we call the "XICS" struct icp_state and then the ICPs
'struct icp_server_state'. It's particularly confusing when all of the
functions have xics_prefixes yet take *icp arguments.
Rename:
struct icp_state -> XICSState
struct icp_server_state -> ICPState
struct ics_state -> ICSState
struct ics_irq_state -> ICSIRQState
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-id: 1374175984-8930-12-git-send-email-aliguori@us.ibm.com
[aik: added ics_resend() on post_load]
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
At present, the savevm / migration support for the pseries machine will not
work when KVM is enabled. That's because KVM manages the guest's hash page
table in the host kernel, so qemu has no visibility of it. This patch
fixes this by using new kernel interfaces to extract and reinsert the
guest's hash table during the migration process.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Message-id: 1374175984-8930-11-git-send-email-aliguori@us.ibm.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This adds the necessary pieces to implement savevm / migration for the
pseries machine. The most complex part here is migrating the hash
table - for the paravirtualized pseries machine the guest's hash page
table is not stored within guest memory, but externally and the guest
accesses it via hypercalls.
This patch uses a hypervisor reserved bit of the HPTE as a dirty bit
(tracking changes to the HPTE itself, not the page it references).
This is used to implement a live migration style incremental save and
restore of the hash table contents.
Normally a hash table is 16MB but it can get bigger depending on how
much RAM the guest has. Due to its nature, updates to it are random so
the live migration style is used for it.
In addition it adds VMStateDescription information to save and restore
the (few) remaining pieces of state information needed by the pseries
machine.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1374175984-8930-9-git-send-email-aliguori@us.ibm.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Model TCE tables as a device that's hooked up as a child object to
the owner. Besides the code cleanup, we get a few nice benefits:
1) free actually works now (it was dead code before)
2) the TCE information is visible in the device tree
3) we can expose table information as properties such that if we
change the window_size, we can use globals to keep migration
working.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-id: 1374175984-8930-6-git-send-email-aliguori@us.ibm.com
[dwg: pseries: savevm support for PAPR TCE tables]
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[alexey: ppc kvm: fix to compile]
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This patch adds helpers to allow PAPR VIO devices to save state common
to all VIO devices during savevm.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1374175984-8930-3-git-send-email-aliguori@us.ibm.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The DBDMA engine really just reads bytes from a producing device (IDE
in our case) and shoves these bytes into memory. It doesn't care whether
any alignment takes place or not.
Our code today however assumes that block accesses always happen on
sector (512 byte) boundaries. This is a fair assumption for most cases.
However, Mac OS X really likes to do unaligned, incomplete accesses
that it finishes with the next DMA request.
So we need to read / write the unaligned bits independent of the actual
asynchronous request, because that one can only handle 512-byte-aligned
data. We also need to cache these unaligned sectors until the next DMA
request, at which point the data might be successfully flushed from the
pipe.
Signed-off-by: Alexander Graf <agraf@suse.de>
Soon we will introduce intermediate processing pauses which will
allow the bottom half to restart a DMA request that couldn't be
fulfilled yet.
For that to work, move the processing variable into the io struct
which is what DMA providers work with.
While touching it, also change it into a bool
Signed-off-by: Alexander Graf <agraf@suse.de>
The DBDMA controller has a bottom half to asynchronously process DMA
request queues.
This bh was stored as a gross static variable. Move it into the device
struct instead.
While at it, move all users of it to the new generic kick function.
Signed-off-by: Alexander Graf <agraf@suse.de>
The DBDMA engine really is running all the time, waiting for input. However
we don't want to waste cycles constantly polling.
So introduce a kick function that data providers can call to notify the
DBDMA controller of new input.
Signed-off-by: Alexander Graf <agraf@suse.de>
We usually keep struct and constant definitions in header files. Move
them there to stay consistent and to make access to fields easier.
Signed-off-by: Alexander Graf <agraf@suse.de>
Since current_cpu is CPUState it no longer depends on CPUPPCState.
Move ppce500_set_mpic_proxy() to a new hw/ppc/ppc_e500.h because
hw/ppc/ppc.h is too heavily using CPUPPCState and PowerPCCPU.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Support in fwcfg has been around for exposure of the clock-frequency
CPU property. OpenBIOS reads it, we just never exposed it.
Since Mac OS X is very picky about its clock frequency values, let's
just take a known good value and always expose that.
Reported-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Signed-off-by: Alexander Graf <agraf@suse.de>
RTAS is a hypervisor provided binary blob that a guest loads and
calls into to execute certain functions. It's similar to the
vsyscall page in Linux or the short lived VMCI paravirt interface
from VMware.
The QEMU implementation of the RTAS blob is simply a passthrough
that proxies all RTAS calls to the hypervisor via an hypercall.
While we pass a CPU argument for hypercall handling in QEMU, we
don't pass it for RTAS calls. Since some RTAs calls require
making hypercalls (normally RTAS is implemented as guest code) we
have nasty hacks to allow that.
Add a CPU argument to RTAS call handling so we can more easily
invoke hypercalls just as guest code would.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Introduce type constant and cast macro.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Enables support for the in-kernel MPIC that thas been merged into the
KVM next branch. This includes irqfd/KVM_IRQ_LINE support from Alex
Graf (along with some other improvements).
Note from Alex regarding kvm_irqchip_create():
On x86, one would call kvm_irqchip_create() to initialize an
in-kernel interrupt controller. That function then goes ahead and
initializes global capability variables as well as the default irq
routing table.
On ppc, we can't call kvm_irqchip_create() because we can have
different types of interrupt controllers. So we want to do all the
things that function would do for us in the in-kernel device init
handler.
Signed-off-by: Scott Wood <scottwood@freescale.com>
[agraf: squash in kvm_irqchip_commit_routes patch, fix non-kvm build,
fix ppcemb]
Signed-off-by: Alexander Graf <agraf@suse.de>
The DMAContext is a simple pointer to an AddressSpace that is now always
already available. Make everyone hold the address space directly,
and clean up the DMA API to use the AddressSpace directly.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Fetch the root region from the sPAPRTCETable, and use it to build
an AddressSpace and DMAContext.
Now, everywhere we have a DMAContext we also have access to the
corresponding AddressSpace (either because we create it just before
the DMAContext, or because dma_context_memory's AddressSpace is
trivially address_space_memory).
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Now we can stop using a "translating" DMAContext, but we do not yet modify
the sPAPRTCETable users to get an AddressSpace; they keep using the table
via a DMAContext.
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The TCE table is currently returned as a DMAContext, and non-type-safe
APIs are called later passing back the DMAContext. Since we want to move
away from DMAContext, use an opaque type instead, and add an accessor
to retrieve the DMAContext from it.
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Many of these should be cleaned up with proper qdev-/QOM-ification.
Right now there are many catch-all headers in include/hw/ARCH depending
on cpu.h, and this makes it necessary to compile these files per-target.
However, fixing this does not belong in these patches.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>