This renames the kvm->arch.hpte_setup_done field to mmu_ready because
we will want to use it for radix guests too -- both for setting things
up before vcpu execution, and for excluding vcpus from executing while
MMU-related things get changed, such as in future switching the MMU
from radix to HPT mode or vice-versa.

This also moves the call to kvmppc_setup_partition_table() that was
done in kvmppc_hv_setup_htab_rma() for HPT guests, and the setting
of mmu_ready, into the caller in kvmppc_vcpu_run_hv().
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Calls to mmu_notifier_invalidate_page() were replaced by calls to
mmu_notifier_invalidate_range() and are now bracketed by calls to
mmu_notifier_invalidate_range_start()/end()

Remove now useless invalidate_page callback.

Changed since v1 (Linus Torvalds)
    - remove now useless kvm_arch_mmu_notifier_invalidate_page()
Signed-off-by: NJérôme Glisse <jglisse@redhat.com>
Tested-by: NMike Galbraith <efault@gmx.de>
Tested-by: NAdam Borowski <kilobyte@angband.pl>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: kvm@vger.kernel.org
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Enhance KVM to cause a guest exit with KVM_EXIT_NMI
exit reason upon a machine check exception (MCE) in
the guest address space if the KVM_CAP_PPC_FWNMI
capability is enabled (instead of delivering a 0x200
interrupt to guest). This enables QEMU to build error
log and deliver machine check exception to guest via
guest registered machine check handler.

This approach simplifies the delivery of machine
check exception to guest OS compared to the earlier
approach of KVM directly invoking 0x200 guest interrupt
vector.

This design/approach is based on the feedback for the
QEMU patches to handle machine check exception. Details
of earlier approach of handling machine check exception
in QEMU and related discussions can be found at:

https://lists.nongnu.org/archive/html/qemu-devel/2014-11/msg00813.html

Note:

This patch now directly invokes machine_check_print_event_info()
from kvmppc_handle_exit_hv() to print the event to host console
at the time of guest exit before the exception is passed on to the
guest. Hence, the host-side handling which was performed earlier
via machine_check_fwnmi is removed.

The reasons for this approach is (i) it is not possible
to distinguish whether the exception occurred in the
guest or the host from the pt_regs passed on the
machine_check_exception(). Hence machine_check_exception()
calls panic, instead of passing on the exception to
the guest, if the machine check exception is not
recoverable. (ii) the approach introduced in this
patch gives opportunity to the host kernel to perform
actions in virtual mode before passing on the exception
to the guest. This approach does not require complex
tweaks to machine_check_fwnmi and friends.
Signed-off-by: NAravinda Prasad <aravinda@linux.vnet.ibm.com>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NMahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This introduces a new KVM capability to control how KVM behaves
on machine check exception (MCE) in HV KVM guests.

If this capability has not been enabled, KVM redirects machine check
exceptions to guest's 0x200 vector, if the address in error belongs to
the guest. With this capability enabled, KVM will cause a guest exit
with the exit reason indicating an NMI.

The new capability is required to avoid problems if a new kernel/KVM
is used with an old QEMU, running a guest that doesn't issue
"ibm,nmi-register".  As old QEMU does not understand the NMI exit
type, it treats it as a fatal error.  However, the guest could have
handled the machine check error if the exception was delivered to
guest's 0x200 interrupt vector instead of NMI exit in case of old
QEMU.

[paulus@ozlabs.org - Reworded the commit message to be clearer,
 enable only on HV KVM.]
Signed-off-by: NAravinda Prasad <aravinda@linux.vnet.ibm.com>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NMahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

On POWER9, we no longer have the restriction that we had on POWER8
where all threads in a core have to be in the same partition, so
the CPU threads are now independent.  However, we still want to be
able to run guests with a virtual SMT topology, if only to allow
migration of guests from POWER8 systems to POWER9.

A guest that has a virtual SMT mode greater than 1 will expect to
be able to use the doorbell facility; it will expect the msgsndp
and msgclrp instructions to work appropriately and to be able to read
sensible values from the TIR (thread identification register) and
DPDES (directed privileged doorbell exception status) special-purpose
registers.  However, since each CPU thread is a separate sub-processor
in POWER9, these instructions and registers can only be used within
a single CPU thread.

In order for these instructions to appear to act correctly according
to the guest's virtual SMT mode, we have to trap and emulate them.
We cause them to trap by clearing the HFSCR_MSGP bit in the HFSCR
register.  The emulation is triggered by the hypervisor facility
unavailable interrupt that occurs when the guest uses them.

To cause a doorbell interrupt to occur within the guest, we set the
DPDES register to 1.  If the guest has interrupts enabled, the CPU
will generate a doorbell interrupt and clear the DPDES register in
hardware.  The DPDES hardware register for the guest is saved in the
vcpu->arch.vcore->dpdes field.  Since this gets written by the guest
exit code, other VCPUs wishing to cause a doorbell interrupt don't
write that field directly, but instead set a vcpu->arch.doorbell_request
flag.  This is consumed and set to 0 by the guest entry code, which
then sets DPDES to 1.

Emulating reads of the DPDES register is somewhat involved, because
it requires reading the doorbell pending interrupt status of all of the
VCPU threads in the virtual core, and if any of those VCPUs are
running, their doorbell status is only up-to-date in the hardware
DPDES registers of the CPUs where they are running.  In order to get
a reasonable approximation of the current doorbell status, we send
those CPUs an IPI, causing an exit from the guest which will update
the vcpu->arch.vcore->dpdes field.  We then use that value in
constructing the emulated DPDES register value.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This allows userspace to set the desired virtual SMT (simultaneous
multithreading) mode for a VM, that is, the number of VCPUs that
get assigned to each virtual core.  Previously, the virtual SMT mode
was fixed to the number of threads per subcore, and if userspace
wanted to have fewer vcpus per vcore, then it would achieve that by
using a sparse CPU numbering.  This had the disadvantage that the
vcpu numbers can get quite large, particularly for SMT1 guests on
a POWER8 with 8 threads per core.  With this patch, userspace can
set its desired virtual SMT mode and then use contiguous vcpu
numbering.

On POWER8, where the threading mode is "strict", the virtual SMT mode
must be less than or equal to the number of threads per subcore.  On
POWER9, which implements a "loose" threading mode, the virtual SMT
mode can be any power of 2 between 1 and 8, even though there is
effectively one thread per subcore, since the threads are independent
and can all be in different partitions.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This adds code to allow us to use a different value for the HFSCR
(Hypervisor Facilities Status and Control Register) when running the
guest from that which applies in the host.  The reason for doing this
is to allow us to trap the msgsndp instruction and related operations
in future so that they can be virtualized.  We also save the value of
HFSCR when a hypervisor facility unavailable interrupt occurs, because
the high byte of HFSCR indicates which facility the guest attempted to
access.

We save and restore the host value on guest entry/exit because some
bits of it affect host userspace execution.

We only do all this on POWER9, not on POWER8, because we are not
intending to virtualize any of the facilities controlled by HFSCR on
POWER8.  In particular, the HFSCR bit that controls execution of
msgsndp and related operations does not exist on POWER8.  The HFSCR
doesn't exist at all on POWER7.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This allows userspace (e.g. QEMU) to enable large decrementer mode for
the guest when running on a POWER9 host, by setting the LPCR_LD bit in
the guest LPCR value.  With this, the guest exit code saves 64 bits of
the guest DEC value on exit.  Other places that use the guest DEC
value check the LPCR_LD bit in the guest LPCR value, and if it is set,
omit the 32-bit sign extension that would otherwise be done.

This doesn't change the DEC emulation used by PR KVM because PR KVM
is not supported on POWER9 yet.

This is partly based on an earlier patch by Oliver O'Halloran.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Marc Zyngier suggested that we define the arch specific VCPU request
base, rather than requiring each arch to remember to start from 8.
That suggestion, along with Radim Krcmar's recent VCPU request flag
addition, snowballed into defining something of an arch VCPU request
defining API.

No functional change.

(Looks like x86 is running out of arch VCPU request bits.  Maybe
 someday we'll need to extend to 64.)
Signed-off-by: NAndrew Jones <drjones@redhat.com>
Acked-by: NChristoffer Dall <cdall@linaro.org>
Signed-off-by: NChristoffer Dall <cdall@linaro.org>

This patch makes KVM capable of using the XIVE interrupt controller
to provide the standard PAPR "XICS" style hypercalls. It is necessary
for proper operations when the host uses XIVE natively.

This has been lightly tested on an actual system, including PCI
pass-through with a TG3 device.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Cleanup pr_xxx(), unsplit pr_xxx() strings, etc., fix build
 failures by adding KVM_XIVE which depends on KVM_XICS and XIVE, and
 adding empty stubs for the kvm_xive_xxx() routines, fixup subject,
 integrate fixes from Paul for building PR=y HV=n]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT
and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO
without passing them to user space which saves time on switching
to user space and back.

This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM.
KVM tries to handle a TCE request in the real mode, if failed
it passes the request to the virtual mode to complete the operation.
If it a virtual mode handler fails, the request is passed to
the user space; this is not expected to happen though.

To avoid dealing with page use counters (which is tricky in real mode),
this only accelerates SPAPR TCE IOMMU v2 clients which are required
to pre-register the userspace memory. The very first TCE request will
be handled in the VFIO SPAPR TCE driver anyway as the userspace view
of the TCE table (iommu_table::it_userspace) is not allocated till
the very first mapping happens and we cannot call vmalloc in real mode.

If we fail to update a hardware IOMMU table unexpected reason, we just
clear it and move on as there is nothing really we can do about it -
for example, if we hot plug a VFIO device to a guest, existing TCE tables
will be mirrored automatically to the hardware and there is no interface
to report to the guest about possible failures.

This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to
the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd
and associates a physical IOMMU table with the SPAPR TCE table (which
is a guest view of the hardware IOMMU table). The iommu_table object
is cached and referenced so we do not have to look up for it in real mode.

This does not implement the UNSET counterpart as there is no use for it -
once the acceleration is enabled, the existing userspace won't
disable it unless a VFIO container is destroyed; this adds necessary
cleanup to the KVM_DEV_VFIO_GROUP_DEL handler.

This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user
space.

This adds real mode version of WARN_ON_ONCE() as the generic version
causes problems with rcu_sched. Since we testing what vmalloc_to_phys()
returns in the code, this also adds a check for already existing
vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect().

This finally makes use of vfio_external_user_iommu_id() which was
introduced quite some time ago and was considered for removal.

Tests show that this patch increases transmission speed from 220MB/s
to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card).
Signed-off-by: NAlexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: NAlex Williamson <alex.williamson@redhat.com>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

PR KVM page fault handler performs eaddr to pte translation for a guest,
however kvmppc_mmu_book3s_64_xlate() does not preserve WIMG bits
(storage control) in the kvmppc_pte struct. If PR KVM is running as
a second level guest under HV KVM, and PR KVM tries inserting HPT entry,
this fails in HV KVM if it already has this mapping.

This preserves WIMG bits between kvmppc_mmu_book3s_64_xlate() and
kvmppc_mmu_map_page().
Signed-off-by: NAlexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This patch provides the MMIO load/store emulation for instructions
of 'double & vector unsigned char & vector signed char & vector
unsigned short & vector signed short & vector unsigned int & vector
signed int & vector double '.

The instructions that this adds emulation for are:

- ldx, ldux, lwax,
- lfs, lfsx, lfsu, lfsux, lfd, lfdx, lfdu, lfdux,
- stfs, stfsx, stfsu, stfsux, stfd, stfdx, stfdu, stfdux, stfiwx,
- lxsdx, lxsspx, lxsiwax, lxsiwzx, lxvd2x, lxvw4x, lxvdsx,
- stxsdx, stxsspx, stxsiwx, stxvd2x, stxvw4x

[paulus@ozlabs.org - some cleanups, fixes and rework, make it
 compile for Book E, fix build when PR KVM is built in]
Signed-off-by: NBin Lu <lblulb@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Its value has never changed; we might as well make it part of the ABI instead
of using the return value of KVM_CHECK_EXTENSION(KVM_CAP_COALESCED_MMIO).

Because PPC does not always make MMIO available, the code has to be made
dependent on CONFIG_KVM_MMIO rather than KVM_COALESCED_MMIO_PAGE_OFFSET.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NRadim Krčmář <rkrcmar@redhat.com>

This adds a not yet working outline of the HPT resizing PAPR
extension.  Specifically it adds the necessary ioctl() functions,
their basic steps, the work function which will handle preparation for
the resize, and synchronization between these, the guest page fault
path and guest HPT update path.

The actual guts of the implementation isn't here yet, so for now the
calls will always fail.
Signed-off-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Currently the kvm_hpt_info structure stores the hashed page table's order,
and also the number of HPTEs it contains and a mask for its size.  The
last two can be easily derived from the order, so remove them and just
calculate them as necessary with a couple of helper inlines.
Signed-off-by: NDavid Gibson <david@gibson.dropbear.id.au>
Reviewed-by: NThomas Huth <thuth@redhat.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Currently, the powerpc kvm_arch structure contains a number of variables
tracking the state of the guest's hashed page table (HPT) in KVM HV.  This
patch gathers them all together into a single kvm_hpt_info substructure.
This makes life more convenient for the upcoming HPT resizing
implementation.
Signed-off-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

With radix, the guest can do TLB invalidations itself using the tlbie
(global) and tlbiel (local) TLB invalidation instructions.  Linux guests
use local TLB invalidations for translations that have only ever been
accessed on one vcpu.  However, that doesn't mean that the translations
have only been accessed on one physical cpu (pcpu) since vcpus can move
around from one pcpu to another.  Thus a tlbiel might leave behind stale
TLB entries on a pcpu where the vcpu previously ran, and if that task
then moves back to that previous pcpu, it could see those stale TLB
entries and thus access memory incorrectly.  The usual symptom of this
is random segfaults in userspace programs in the guest.

To cope with this, we detect when a vcpu is about to start executing on
a thread in a core that is a different core from the last time it
executed.  If that is the case, then we mark the core as needing a
TLB flush and then send an interrupt to any thread in the core that is
currently running a vcpu from the same guest.  This will get those vcpus
out of the guest, and the first one to re-enter the guest will do the
TLB flush.  The reason for interrupting the vcpus executing on the old
core is to cope with the following scenario:

	CPU 0			CPU 1			CPU 4
	(core 0)			(core 0)			(core 1)

	VCPU 0 runs task X      VCPU 1 runs
	core 0 TLB gets
	entries from task X
	VCPU 0 moves to CPU 4
							VCPU 0 runs task X
							Unmap pages of task X
							tlbiel

				(still VCPU 1)			task X moves to VCPU 1
				task X runs
				task X sees stale TLB
				entries

That is, as soon as the VCPU starts executing on the new core, it
could unmap and tlbiel some page table entries, and then the task
could migrate to one of the VCPUs running on the old core and
potentially see stale TLB entries.

Since the TLB is shared between all the threads in a core, we only
use the bit of kvm->arch.need_tlb_flush corresponding to the first
thread in the core.  To ensure that we don't have a window where we
can miss a flush, this moves the clearing of the bit from before the
actual flush to after it.  This way, two threads might both do the
flush, but we prevent the situation where one thread can enter the
guest before the flush is finished.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This adds code to  branch around the parts that radix guests don't
need - clearing and loading the SLB with the guest SLB contents,
saving the guest SLB contents on exit, and restoring the host SLB
contents.

Since the host is now using radix, we need to save and restore the
host value for the PID register.

On hypervisor data/instruction storage interrupts, we don't do the
guest HPT lookup on radix, but just save the guest physical address
for the fault (from the ASDR register) in the vcpu struct.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This adds a field in struct kvm_arch and an inline helper to
indicate whether a guest is a radix guest or not, plus a new file
to contain the radix MMU code, which currently contains just a
translate function which knows how to traverse the guest page
tables to translate an address.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This adds the implementation of the KVM_PPC_CONFIGURE_V3_MMU ioctl
for HPT guests on POWER9.  With this, we can return 1 for the
KVM_CAP_PPC_MMU_HASH_V3 capability.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

KVM_HALT_POLL_NS_DEFAULT is an arch specific constant which sets the
default value of the halt_poll_ns kvm module parameter which determines
the global maximum halt polling interval.

The current value for powerpc is 500000 (500us) which means that any
repetitive workload with a period of less than that can drive the cpu
usage to 100% where it may have been mostly idle without halt polling.
This presents the possibility of a large increase in power usage with
a comparatively small performance benefit.

Reduce the default to 10000 (10us) and a user can tune this themselves
to set their affinity for halt polling based on the trade off between power
and performance which they are willing to make.
Signed-off-by: NSuraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

POWER9 adds new capabilities to the tlbie (TLB invalidate entry)
and tlbiel (local tlbie) instructions.  Both instructions get a
set of new parameters (RIC, PRS and R) which appear as bits in the
instruction word.  The tlbiel instruction now has a second register
operand, which contains a PID and/or LPID value if needed, and
should otherwise contain 0.

This adapts KVM-HV's usage of tlbie and tlbiel to work on POWER9
as well as older processors.  Since we only handle HPT guests so
far, we need RIC=0 PRS=0 R=0, which ends up with the same instruction
word as on previous processors, so we don't need to conditionally
execute different instructions depending on the processor.

The local flush on first entry to a guest in book3s_hv_rmhandlers.S
is a loop which depends on the number of TLB sets.  Rather than
using feature sections to set the number of iterations based on
which CPU we're on, we now work out this number at VM creation time
and store it in the kvm_arch struct.  That will make it possible to
get the number from the device tree in future, which will help with
compatibility with future processors.

Since mmu_partition_table_set_entry() does a global flush of the
whole LPID, we don't need to do the TLB flush on first entry to the
guest on each processor.  Therefore we don't set all bits in the
tlb_need_flush bitmap on VM startup on POWER9.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This adds code to handle two new guest-accessible special-purpose
registers on POWER9: TIDR (thread ID register) and PSSCR (processor
stop status and control register).  They are context-switched
between host and guest, and the guest values can be read and set
via the one_reg interface.

The PSSCR contains some fields which are guest-accessible and some
which are only accessible in hypervisor mode.  We only allow the
guest-accessible fields to be read or set by userspace.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

When switching from/to a guest that has a transaction in progress,
we need to save/restore the checkpointed register state.  Although
XER is part of the CPU state that gets checkpointed, the code that
does this saving and restoring doesn't save/restore XER.

This fixes it by saving and restoring the XER.  To allow userspace
to read/write the checkpointed XER value, we also add a new ONE_REG
specifier.

The visible effect of this bug is that the guest may see its XER
value being corrupted when it uses transactions.

Fixes: e4e38121 ("KVM: PPC: Book3S HV: Add transactional memory support")
Fixes: 0a8eccef ("KVM: PPC: Book3S HV: Add missing code for transaction reclaim on guest exit")
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Reviewed-by: NThomas Huth <thuth@redhat.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This keeps a per vcpu cache for recently page faulted MMIO entries.
On a page fault, if the entry exists in the cache, we can avoid some
time-consuming paths, for example, looking up HPT, locking HPTE twice
and searching mmio gfn from memslots, then directly call
kvmppc_hv_emulate_mmio().

In current implenment, we limit the size of cache to four. We think
it's enough to cover the high-frequency MMIO HPTEs in most case.
For example, considering the case of using virtio device, for virtio
legacy devices, one HPTE could handle notifications from up to
1024 (64K page / 64 byte Port IO register) devices, so one cache entry
is enough; for virtio modern devices, we always need one HPTE to handle
notification for each device because modern device would use a 8M MMIO
register to notify host instead of Port IO register, typically the
system's configuration should not exceed four virtio devices per
vcpu, four cache entry is also enough in this case. Of course, if needed,
we could also modify the macro to a module parameter in the future.
Signed-off-by: NYongji Xie <xyjxie@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

POWER8 has one virtual timebase (VTB) register per subcore, not one
per CPU thread.  The HV KVM code currently treats VTB as a per-thread
register, which can lead to spurious soft lockup messages from guests
which use the VTB as the time source for the soft lockup detector.
(CPUs before POWER8 did not have the VTB register.)

For HV KVM, this fixes the problem by making only the primary thread
in each virtual core save and restore the VTB value.  With this,
the VTB state becomes part of the kvmppc_vcore structure.  This
also means that "piggybacking" of multiple virtual cores onto one
subcore is not possible on POWER8, because then the virtual cores
would share a single VTB register.

PR KVM emulates a VTB register, which is per-vcpu because PR KVM
has no notion of CPU threads or SMT.  For PR KVM we move the VTB
state into the kvmppc_vcpu_book3s struct.

Cc: stable@vger.kernel.org # v3.14+
Reported-by: NThomas Huth <thuth@redhat.com>
Tested-by: NThomas Huth <thuth@redhat.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Add VCPU stat counters to track affinity for passthrough
interrupts.

pthru_all: Counts all passthrough interrupts whose IRQ mappings are
           in the kvmppc_passthru_irq_map structure.
pthru_host: Counts all cached passthrough interrupts that were injected
	    from the host through kvm_set_irq (i.e. not handled in
	    real mode).
pthru_bad_aff: Counts how many cached passthrough interrupts have
               bad affinity (receiving CPU is not running VCPU that is
	       the target of the virtual interrupt in the guest).
Signed-off-by: NSuresh Warrier <warrier@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This patch introduces an IRQ mapping structure, the
kvmppc_passthru_irqmap structure that is to be used
to map the real hardware IRQ in the host with the virtual
hardware IRQ (gsi) that is injected into a guest by KVM for
passthrough adapters.

Currently, we assume a separate IRQ mapping structure for
each guest. Each kvmppc_passthru_irqmap has a mapping arrays,
containing all defined real<->virtual IRQs.

[paulus@ozlabs.org - removed irq_chip field from struct
 kvmppc_passthru_irqmap; changed parameter for
 kvmppc_get_passthru_irqmap from struct kvm_vcpu * to struct
 kvm *, removed small cached array.]
Signed-off-by: NSuresh Warrier <warrier@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

vcpu stats are used to collect information about a vcpu which can be viewed
in the debugfs. For example halt_attempted_poll and halt_successful_poll
are used to keep track of the number of times the vcpu attempts to and
successfully polls. These stats are currently not used on powerpc.

Implement incrementation of the halt_attempted_poll and
halt_successful_poll vcpu stats for powerpc. Since these stats are summed
over all the vcpus for all running guests it doesn't matter which vcpu
they are attributed to, thus we choose the current runner vcpu of the
vcore.

Also add new vcpu stats: halt_poll_success_ns, halt_poll_fail_ns and
halt_wait_ns to be used to accumulate the total time spend polling
successfully, polling unsuccessfully and waiting respectively, and
halt_successful_wait to accumulate the number of times the vcpu waits.
Given that halt_poll_success_ns, halt_poll_fail_ns and halt_wait_ns are
expressed in nanoseconds it is necessary to represent these as 64-bit
quantities, otherwise they would overflow after only about 4 seconds.

Given that the total time spend either polling or waiting will be known and
the number of times that each was done, it will be possible to determine
the average poll and wait times. This will give the ability to tune the kvm
module parameters based on the calculated average wait and poll times.
Signed-off-by: NSuraj Jitindar Singh <sjitindarsingh@gmail.com>
Reviewed-by: NDavid Matlack <dmatlack@google.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

vms and vcpus have statistics associated with them which can be viewed
within the debugfs. Currently it is assumed within the vcpu_stat_get() and
vm_stat_get() functions that all of these statistics are represented as
u32s, however the next patch adds some u64 vcpu statistics.

Change all vcpu statistics to u64 and modify vcpu_stat_get() accordingly.
Since vcpu statistics are per vcpu, they will only be updated by a single
vcpu at a time so this shouldn't present a problem on 32-bit machines
which can't atomically increment 64-bit numbers. However vm statistics
could potentially be updated by multiple vcpus from that vm at a time.
To avoid the overhead of atomics make all vm statistics ulong such that
they are 64-bit on 64-bit systems where they can be atomically incremented
and are 32-bit on 32-bit systems which may not be able to atomically
increment 64-bit numbers. Modify vm_stat_get() to expect ulongs.
Signed-off-by: NSuraj Jitindar Singh <sjitindarsingh@gmail.com>
Reviewed-by: NDavid Matlack <dmatlack@google.com>
Acked-by: NChristian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

This patch introduces new halt polling functionality into the kvm_hv kernel
module. When a vcore is idle it will poll for some period of time before
scheduling itself out.

When all of the runnable vcpus on a vcore have ceded (and thus the vcore is
idle) we schedule ourselves out to allow something else to run. In the
event that we need to wake up very quickly (for example an interrupt
arrives), we are required to wait until we get scheduled again.

Implement halt polling so that when a vcore is idle, and before scheduling
ourselves, we poll for vcpus in the runnable_threads list which have
pending exceptions or which leave the ceded state. If we poll successfully
then we can get back into the guest very quickly without ever scheduling
ourselves, otherwise we schedule ourselves out as before.

There exists generic halt_polling code in virt/kvm_main.c, however on
powerpc the polling conditions are different to the generic case. It would
be nice if we could just implement an arch specific kvm_check_block()
function, but there is still other arch specific things which need to be
done for kvm_hv (for example manipulating vcore states) which means that a
separate implementation is the best option.

Testing of this patch with a TCP round robin test between two guests with
virtio network interfaces has found a decrease in round trip time of ~15us
on average. A performance gain is only seen when going out of and
back into the guest often and quickly, otherwise there is no net benefit
from the polling. The polling interval is adjusted such that when we are
often scheduled out for long periods of time it is reduced, and when we
often poll successfully it is increased. The rate at which the polling
interval increases or decreases, and the maximum polling interval, can
be set through module parameters.

Based on the implementation in the generic kvm module by Wanpeng Li and
Paolo Bonzini, and on direction from Paul Mackerras.
Signed-off-by: NSuraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

The struct kvmppc_vcore is a structure used to store various information
about a virtual core for a kvm guest. The runnable_threads element of the
struct provides a list of all of the currently runnable vcpus on the core
(those in the KVMPPC_VCPU_RUNNABLE state). The previous implementation of
this list was a linked_list. The next patch requires that the list be able
to be iterated over without holding the vcore lock.

Reimplement the runnable_threads list in the kvmppc_vcore struct as an
array. Implement function to iterate over valid entries in the array and
update access sites accordingly.
Signed-off-by: NSuraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

The next commit will introduce a member to the kvmppc_vcore struct which
references MAX_SMT_THREADS which is defined in kvm_book3s_asm.h, however
this file isn't included in kvm_host.h directly. Thus compiling for
certain platforms such as pmac32_defconfig and ppc64e_defconfig with KVM
fails due to MAX_SMT_THREADS not being defined.

Move the struct kvmppc_vcore definition to kvm_book3s.h which explicitly
includes kvm_book3s_asm.h.
Signed-off-by: NSuraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

It doesn't make sense to create irqfds for a VM that doesn't have
in-kernel interrupt controller emulation.  There is an existing
interface for architecture code to tell the irqfd code whether or
not any interrupt controller has been initialized, called
kvm_arch_intc_initialized(), so let's implement that for powerpc.
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

Some wakeups should not be considered a sucessful poll. For example on
s390 I/O interrupts are usually floating, which means that _ALL_ CPUs
would be considered runnable - letting all vCPUs poll all the time for
transactional like workload, even if one vCPU would be enough.
This can result in huge CPU usage for large guests.
This patch lets architectures provide a way to qualify wakeups if they
should be considered a good/bad wakeups in regard to polls.

For s390 the implementation will fence of halt polling for anything but
known good, single vCPU events. The s390 implementation for floating
interrupts does a wakeup for one vCPU, but the interrupt will be delivered
by whatever CPU checks first for a pending interrupt. We prefer the
woken up CPU by marking the poll of this CPU as "good" poll.
This code will also mark several other wakeup reasons like IPI or
expired timers as "good". This will of course also mark some events as
not sucessful. As  KVM on z runs always as a 2nd level hypervisor,
we prefer to not poll, unless we are really sure, though.

This patch successfully limits the CPU usage for cases like uperf 1byte
transactional ping pong workload or wakeup heavy workload like OLTP
while still providing a proper speedup.

This also introduced a new vcpu stat "halt_poll_no_tuning" that marks
wakeups that are considered not good for polling.
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Radim Krčmář <rkrcmar@redhat.com> (for an earlier version)
Cc: David Matlack <dmatlack@google.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
[Rename config symbol. - Paolo]
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The KVM_MAX_VCPUS define provides the maximum number of vCPUs per guest, and
also the upper limit for vCPU ids. This is okay for all archs except PowerPC
which can have higher ids, depending on the cpu/core/thread topology. In the
worst case (single threaded guest, host with 8 threads per core), it limits
the maximum number of vCPUS to KVM_MAX_VCPUS / 8.

This patch separates the vCPU numbering from the total number of vCPUs, with
the introduction of KVM_MAX_VCPU_ID, as the maximal valid value for vCPU ids
plus one.

The corresponding KVM_CAP_MAX_VCPU_ID allows userspace to validate vCPU ids
before passing them to KVM_CREATE_VCPU.

This patch only implements KVM_MAX_VCPU_ID with a specific value for PowerPC.
Other archs continue to return KVM_MAX_VCPUS instead.
Suggested-by: NRadim Krcmar <rkrcmar@redhat.com>
Signed-off-by: NGreg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

This enables userspace view of TCE tables to start from non-zero offset
on a bus. This will be used for huge DMA windows.

This only changes the internal structure, the user interface needs to
change in order to use an offset.
Signed-off-by: NAlexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

At the moment the kvmppc_spapr_tce_table struct can only describe
4GB windows and handle fixed size (4K) pages. Dynamic DMA windows
support more so these limits need to be extended.

This replaces window_size (in bytes, 4GB max) with page_shift (32bit)
and size (64bit, in pages).

This should cause no behavioural change as this is changing
the internal structures only - the user interface still only
allows one to create a 32-bit table with 4KiB pages at this stage.
Signed-off-by: NAlexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: NDavid Gibson <david@gibson.dropbear.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

The problem:

On -rt, an emulated LAPIC timer instances has the following path:

1) hard interrupt
2) ksoftirqd is scheduled
3) ksoftirqd wakes up vcpu thread
4) vcpu thread is scheduled

This extra context switch introduces unnecessary latency in the
LAPIC path for a KVM guest.

The solution:

Allow waking up vcpu thread from hardirq context,
thus avoiding the need for ksoftirqd to be scheduled.

Normal waitqueues make use of spinlocks, which on -RT
are sleepable locks. Therefore, waking up a waitqueue
waiter involves locking a sleeping lock, which
is not allowed from hard interrupt context.

cyclictest command line:

This patch reduces the average latency in my tests from 14us to 11us.

Daniel writes:
Paolo asked for numbers from kvm-unit-tests/tscdeadline_latency
benchmark on mainline. The test was run 1000 times on
tip/sched/core 4.4.0-rc8-01134-g0905f04e:

  ./x86-run x86/tscdeadline_latency.flat -cpu host

with idle=poll.

The test seems not to deliver really stable numbers though most of
them are smaller. Paolo write:

"Anything above ~10000 cycles means that the host went to C1 or
lower---the number means more or less nothing in that case.

The mean shows an improvement indeed."

Before:

               min             max         mean           std
count  1000.000000     1000.000000  1000.000000   1000.000000
mean   5162.596000  2019270.084000  5824.491541  20681.645558
std      75.431231   622607.723969    89.575700   6492.272062
min    4466.000000    23928.000000  5537.926500    585.864966
25%    5163.000000  16132529.750000  5790.132275  16683.745433
50%    5175.000000  2281919.000000  5834.654000  23151.990026
75%    5190.000000  2382865.750000  5861.412950  24148.206168
max    5228.000000  4175158.000000  6254.827300  46481.048691

After
               min            max         mean           std
count  1000.000000     1000.00000  1000.000000   1000.000000
mean   5143.511000  2076886.10300  5813.312474  21207.357565
std      77.668322   610413.09583    86.541500   6331.915127
min    4427.000000    25103.00000  5529.756600    559.187707
25%    5148.000000  1691272.75000  5784.889825  17473.518244
50%    5160.000000  2308328.50000  5832.025000  23464.837068
75%    5172.000000  2393037.75000  5853.177675  24223.969976
max    5222.000000  3922458.00000  6186.720500  42520.379830

[Patch was originaly based on the swait implementation found in the -rt
 tree. Daniel ported it to mainline's version and gathered the
 benchmark numbers for tscdeadline_latency test.]
Signed-off-by: NDaniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: NPeter Zijlstra (Intel) <peterz@infradead.org>
Cc: linux-rt-users@vger.kernel.org
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1455871601-27484-4-git-send-email-wagi@monom.orgSigned-off-by: NThomas Gleixner <tglx@linutronix.de>