More recent Power server chips (i.e. based on the 64 bit hash MMU)
support more than just the traditional 4k and 16M page sizes. This
can get quite complicated, because which page sizes are supported,
which combinations are supported within an MMU segment and how these
page sizes are encoded both in the SLB entry and the hash PTE can vary
depending on the CPU model (they are not specified by the
architecture). In addition the firmware or hypervisor may not permit
use of certain page sizes, for various reasons. Whether various page
sizes are supported on KVM, for example, depends on whether the PR or
HV variant of KVM is in use, and on the page size of the memory
backing the guest's RAM.
This patch adds information to the CPUState and cpu defs to describe
the supported page sizes and encodings. Since TCG does not yet
support any extended page sizes, we just set this to NULL in the
static CPU definitions, expanding this to the default 4k and 16M page
sizes when we initialize the cpu state. When using KVM, however, we
instead determine available page sizes using the new
KVM_PPC_GET_SMMU_INFO call. For old kernels without that call, we use
some defaults, with some guesswork which should do the right thing for
existing HV and PR implementations. The fallback might not be correct
for future versions, but that's ok, because they'll have
KVM_PPC_GET_SMMU_INFO.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
On target-ppc, our table of CPU types and features encodes the features as
found on the hardware, regardless of whether these features are actually
usable under TCG or KVM. We already have cases where the information from
the cpu table must be fixed up to account for limitations in the emulation
method we're using. e.g. TCG does not support the DFP and VSX instructions
and KVM needs different numbering of the CPUs in order to tell it the
correct thread to core mappings.
This patch cleans up these hacks to handle emulation limitations by
consolidating them into a pair of functions specifically for the purpose.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
[AF: Style and typo fixes, rename new functions and drop ppc_def_t arg]
Signed-off-by: Andreas Färber <afaerber@suse.de>
Scripted conversion:
sed -i "s/CPUState/CPUPPCState/g" target-ppc/*.[hc]
sed -i "s/#define CPUPPCState/#define CPUState/" target-ppc/cpu.h
Signed-off-by: Andreas Färber <afaerber@suse.de>
Acked-by: Anthony Liguori <aliguori@us.ibm.com>
The recent usage of MemoryRegion in kvm_ppc.h breaks builds with
CONFIG_USER_ONLY=y. This patch fixes it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
For convenience with kvm, x86 allows the user to specify -cpu host on the
qemu command line, which means make the guest cpu the same as the host
cpu. This patch implements the same option for ppc targets.
For now, this just read the host PVR (Processor Version Register) and
selects one of our existing CPU specs based on it. This means that the
option will not work if the host cpu is not supported by TCG, even if that
wouldn't matter for use under kvm.
In future, we can extend this in future to override parts of the cpu spec
based on information obtained from the host (via /proc/cpuinfo, the host
device tree, or explicit KVM calls). That will let us handle cases where
the real kvm-virtualized CPU doesn't behave exactly like the TCG-emulated
CPU. With appropriate annotation of the CPU specs we'll also then be able
to use host cpus under kvm even when there isn't a matching full TCG model.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Sufficiently recent PAPR specifications define properties "ibm,vmx"
and "ibm,dfp" on the CPU node which advertise whether the VMX vector
extensions (or the later VSX version) and/or the Decimal Floating
Point operations from IBM's recent POWER CPUs are available.
Currently we do not put these in the guest device tree and the guest
kernel will consequently assume they are not available. This is good,
because they are not supported under TCG. VMX is similar enough to
Altivec that it might be trivial to support, but VSX and DFP would
both require significant work to support in TCG.
However, when running under kvm on a host which supports these
instructions, there's no reason not to let the guest use them. This
patch, therefore, checks for the relevant support on the host CPU
and, if present, advertises them to the guest as well.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
The pseries machine of qemu implements the TCE mechanism used as a
virtual IOMMU for the PAPR defined virtual IO devices. Because the
PAPR spec only defines a small DMA address space, the guest VIO
drivers need to update TCE mappings very frequently - the virtual
network device is particularly bad. This means many slow exits to
qemu to emulate the H_PUT_TCE hypercall.
Sufficiently recent kernels allow this to be mitigated by implementing
H_PUT_TCE in the host kernel. To make use of this, however, qemu
needs to initialize the necessary TCE tables, and map them into itself
so that the VIO device implementations can retrieve the mappings when
they access guest memory (which is treated as a virtual DMA
operation).
This patch adds the necessary calls to use the KVM TCE acceleration.
If the kernel does not support acceleration, or there is some other
error creating the accelerated TCE table, then it will still fall back
to full userspace TCE implementation.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
At present, using the hypervisor aware Book3S-HV KVM will only work
with qemu on POWER7 CPUs. PPC970 CPUs also have hypervisor
capability, but they lack the VRMA feature which makes assigning guest
memory easier.
In order to allow KVM Book3S-HV on PPC970, we need to specially
allocate the first chunk of guest memory (the "Real Mode Area" or
RMA), so that it is physically contiguous.
Sufficiently recent host kernels allow such contiguous RMAs to be
allocated, with a kvm capability advertising whether the feature is
available and/or necessary on this hardware. This patch enables qemu
to use this support, thus allowing kvm acceleration of pseries qemu
machines on PPC970 hardware.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
---
agraf: fix to use memory api
Alex Graf has already made qemu support KVM for the pseries machine
when using the Book3S-PR KVM variant (which runs the guest in
usermode, emulating supervisor operations). This code allows gets us
very close to also working with KVM Book3S-HV (using the hypervisor
capabilities of recent POWER CPUs).
This patch moves us another step towards Book3S-HV support by
correctly handling SMT (multithreaded) POWER CPUs. There are two
parts to this:
* Querying KVM to check SMT capability, and if present, adjusting the
cpu numbers that qemu assigns to cause KVM to assign guest threads
to cores in the right way (this isn't automatic, because the POWER
HV support has a limitation that different threads on a single core
cannot be in different guests at the same time).
* Correctly informing the guest OS of the SMT thread to core mappings
via the device tree.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
When running PR style KVM, we need to tell the kernel that we want
to run in PAPR mode now. This means that we need to pass some more
register information down and enable papr mode. We also need to align
the HTAB to htab_size boundary.
Using this patch, -M pseries works with kvm even on non-hv kvm
implementations, as long as the preceding kernel patches are in.
Signed-off-by: Alexander Graf <agraf@suse.de>
---
v1 -> v2:
- match on CONFIG_PSERIES
v2 -> v3:
- remove HIOR pieces from PAPR patch (ABI breakage)
We have a bunch of helper functions that don't have any stubs for them in case
we don't have CONFIG_KVM enabled. That didn't bite us so far, because gcc can
optimize them out pretty well, but we should really provide them.
Signed-off-by: Alexander Graf <agraf@suse.de>
---
v1 -> v2:
- use uint64_t for clockfreq
We need to find out the host's clock-frequency when running on KVM, so
let's export a respective function.
Signed-off-by: Alexander Graf <agraf@suse.de>
---
v1 -> v2:
- enable 64bit values
When configured with --enable-debug, we compile without optimization.
This means that the function mpc8544_copy_soc_cell() in ppce500_mpc8544ds.c
is not optimized out, even though it is never called without kvm. That in
turn causes a link failure, because it calls the function
kvmppc_read_host_property() which is in kvm_ppc.o and therefore not
included in a --disable-kvm build.
This patch fixes the problem by providing a dummy stub for
kvmppc_read_host_property() in kvm_ppc.h when !CONFIG_KVM.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
This patch implements the infrastructure and hypercalls necessary for the
PAPR specified CRQ (Command Request Queue) mechanism. This general
request queueing system is used by many of the PAPR virtual IO devices,
including the virtual scsi adapter.
Signed-off-by: Ben Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David Gibson <dwg@au1.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
KVM on PowerPC used to have completely broken interrupt logic. Usually,
interrupts work by having a PIC that pulls a line up/down, so the CPU knows
that an interrupt is active. This line stays active until some action is
done to the PIC to release the line.
On KVM for PPC, we just checked if there was an interrupt pending and pulled
a line in the kernel module. We never released it though, hoping that kernel
space would just declare an interrupt as released when injected - which is
wrong.
To fix this, we need to completely redesign the interrupt injection logic.
Whenever an interrupt line gets triggered, we need to notify kernel space
that the line is up. Whenever it gets released, we do the same. This way
we can assure that the interrupt state is always known to kernel space.
This fixes random stalls in KVM guests on PowerPC that were waiting for
an interrupt while everyone else thought they received it already.
Signed-off-by: Alexander Graf <agraf@suse.de>
On KVM for PPC we need to tell the guest which instructions to use when
doing a hypercall. The clean way to do this is to go through an ioctl
from userspace and passing it on to the guest using the device tree.
So let's do the qemu part here: read out the hypercall and pass it on
to the guest's fw_cfg so openBIOS can read it out and expose it again.
Signed-off-by: Alexander Graf <agraf@suse.de>
Our guest systems need to know by how much the timebase increases every second,
so there usually is a "timebase-frequency" property in the cpu leaf of the
device tree.
This property is missing in OpenBIOS.
With qemu, Linux's fallback timebase speed and qemu's internal timebase speed
match up. With KVM, that is no longer true. The guest is running at the same
timebase speed as the host.
This leads to massive timing problems. On my test machine, a "sleep 2" takes
about 14 seconds with KVM enabled.
This patch exports the timebase frequency to OpenBIOS, so it can then put them
into the device tree.
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Implement hooks called by generic KVM code.
Also add code that will copy the host's CPU and timebase frequencies to the
guest, which is necessary on KVM because the guest can directly access the
timebase.
Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com>
Acked-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@6065 c046a42c-6fe2-441c-8c8c-71466251a162
