By setting a Kconfig option, the architecture can control when
guest notifications will be presented by the apf backend.
There is the default batch mechanism, working as before, where the vcpu
thread should pull in this information.
Opposite to this, there is now the direct mechanism, that will push the
information to the guest.
This way s390 can use an already existing architecture interface.
Still the vcpu thread should call check_completion to cleanup leftovers.
Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
(cherry picked from commit e0ead41a6d)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
If kvm_io_bus_register_dev() fails then it returns success but it should
return an error code.
I also did a little cleanup like removing an impossible NULL test.
Cc: stable@vger.kernel.org
Fixes: 2b3c246a68 ('KVM: Make coalesced mmio use a device per zone')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit aac5c4226e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Commit 7940876e13 ("kvm: make local
functions static") broke KVM PPC builds due to removing (rather than
moving) the stub version of kvm_vcpu_eligible_for_directed_yield().
This patch reintroduces it.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Alexander Graf <agraf@suse.de>
[Move the #ifdef inside the function. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 4a55dd7273)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The THP code in KVM/ARM is a bit restrictive in not allowing a THP
to be used if the VMA is not 2MB aligned. Actually, it is not so much
the VMA that matters, but the associated memslot:
A process can perfectly mmap a region with no particular alignment
restriction, and then pass a 2MB aligned address to KVM. In this
case, KVM will only use this 2MB aligned region, and will ignore
the range between vma->vm_start and memslot->userspace_addr.
It can also choose to place this memslot at whatever alignment it
wants in the IPA space. In the end, what matters is the relative
alignment of the user space and IPA mappings with respect to a
2M page. They absolutely must be the same if you want to use THP.
Cc: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 136d737fd2)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Running 'make namespacecheck' found lots of functions that
should be declared static, since only used in one file.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
(cherry picked from commit 7940876e13)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The SMC-based PSCI emulation for Guest is going to be very different
from the in-kernel HVC-based PSCI emulation hence for now just inject
undefined exception when Guest executes SMC instruction.
Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Signed-off-by: marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit e5cf9dcdbf)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
This patch allows us to have X-Gene guest VCPU when using KVM arm64
on APM X-Gene host.
We add KVM_ARM_TARGET_XGENE_POTENZA for X-Gene Potenza compatible
guest VCPU and we return KVM_ARM_TARGET_XGENE_POTENZA in kvm_target_cpu()
when running on X-Gene host with Potenza core.
[maz: sanitized the commit log]
Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit e28100bd8e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Current max VCPUs per-Guest is set to 4 which is preventing
us from creating a Guest (or VM) with 8 VCPUs on Host (e.g.
X-Gene Storm SOC) with 8 Host CPUs.
The correct value of max VCPUs per-Guest should be same as
the max CPUs supported by GICv2 which is 8 but, increasing
value of max VCPUs per-Guest can make things slower hence
we add Kconfig option to let KVM users select appropriate
max VCPUs per-Guest.
Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit da7814700a)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Implement support for the CPU interface register access driven by MMIO
address offsets from the CPU interface base address. Useful for user
space to support save/restore of the VGIC state.
This commit adds support only for the same logic as the current VGIC
support, and no more. For example, the active priority registers are
handled as RAZ/WI, just like setting priorities on the emulated
distributor.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit fa20f5aea5)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Handle MMIO accesses to the two registers which should support both the
case where the VMs want to read/write either of these registers and the
case where user space reads/writes these registers to do save/restore of
the VGIC state.
Note that the added complexity compared to simple set/clear enable
registers stems from the bookkeping of source cpu ids. It may be
possible to change the underlying data structure to simplify the
complexity, but since this is not in the critical path at all, this will
do.
Also note that reading this register from a live guest will not be
accurate compared to on hardware, because some state may be living on
the CPU LRs and the only way to give a consistent read would be to force
stop all the VCPUs and request them to unqueu the LR state onto the
distributor. Until we have an actual user of live reading this
register, we can live with the difference.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 90a5355ee7)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
To properly access the VGIC state from user space it is very unpractical
to have to loop through all the LRs in all register access functions.
Instead, support moving all pending state from LRs to the distributor,
but leave active state LRs alone.
Note that to accurately present the active and pending state to VCPUs
reading these distributor registers from a live VM, we would have to
stop all other VPUs than the calling VCPU and ask each CPU to unqueue
their LR state onto the distributor and add fields to track active state
on the distributor side as well. We don't have any users of such
functionality yet and there are other inaccuracies of the GIC emulation,
so don't provide accurate synchronized access to this state just yet.
However, when the time comes, having this function should help.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit cbd333a4bf)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Add infrastructure to handle distributor and cpu interface register
accesses through the KVM_{GET/SET}_DEVICE_ATTR interface by adding the
KVM_DEV_ARM_VGIC_GRP_DIST_REGS and KVM_DEV_ARM_VGIC_GRP_CPU_REGS groups
and defining the semantics of the attr field to be the MMIO offset as
specified in the GICv2 specs.
Missing register accesses or other changes in individual register access
functions to support save/restore of the VGIC state is added in
subsequent patches.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit c07a0191ef)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The arch-generic KVM code expects the cpu field of a vcpu to be -1 if
the vcpu is no longer assigned to a cpu. This is used for the optimized
make_all_cpus_request path and will be used by the vgic code to check
that no vcpus are running.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit e9b152cb95)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Rename the vgic_ranges array to vgic_dist_ranges to be more specific and
to prepare for handling CPU interface register access as well (for
save/restore of VGIC state).
Pass offset from distributor or interface MMIO base to
find_matching_range function instead of the physical address of the
access in the VM memory map. This allows other callers unaware of the
VM specifics, but with generic VGIC knowledge to reuse the function.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 1006e8cb22)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Define CPU interface offsets for the GICC_ABPR, GICC_APR, and GICC_IIDR
registers. Define distributor registers for the GICD_SPENDSGIR and the
GICD_CPENDSGIR. KVM/ARM needs to know about these definitions to fully
support save/restore of the VGIC.
Also define some masks and shifts for the various GICH_VMCR fields.
Cc: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 0307e1770f)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Support setting the distributor and cpu interface base addresses in the
VM physical address space through the KVM_{SET,GET}_DEVICE_ATTR API
in addition to the ARM specific API.
This has the added benefit of being able to share more code in user
space and do things in a uniform manner.
Also deprecate the older API at the same time, but backwards
compatibility will be maintained.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit ce01e4e887)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Support creating the ARM VGIC device through the KVM_CREATE_DEVICE
ioctl, which can then later be leveraged to use the
KVM_{GET/SET}_DEVICE_ATTR, which is useful both for setting addresses in
a more generic API than the ARM-specific one and is useful for
save/restore of VGIC state.
Adds KVM_CAP_DEVICE_CTRL to ARM capabilities.
Note that we change the check for creating a VGIC from bailing out if
any VCPUs were created, to bailing out if any VCPUs were ever run. This
is an important distinction that shouldn't break anything, but allows
creating the VGIC after the VCPUs have been created.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 7330672bef)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Rework the VGIC initialization slightly to allow initialization of the
vgic cpu-specific state even if the irqchip (the VGIC) hasn't been
created by user space yet. This is safe, because the vgic data
structures are already allocated when the CPU is allocated if VGIC
support is compiled into the kernel. Further, the init process does not
depend on any other information and the sacrifice is a slight
performance degradation for creating VMs in the no-VGIC case.
The reason is that the new device control API doesn't mandate creating
the VGIC before creating the VCPU and it is unreasonable to require user
space to create the VGIC before creating the VCPUs.
At the same time move the irqchip_in_kernel check out of
kvm_vcpu_first_run_init and into the init function to make the per-vcpu
and global init functions symmetric and add comments on the exported
functions making it a bit easier to understand the init flow by only
looking at vgic.c.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit e1ba0207a1)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
For migration to work we need to save (and later restore) the state of
each core's virtual generic timer.
Since this is per VCPU, we can use the [gs]et_one_reg ioctl and export
the three needed registers (control, counter, compare value).
Though they live in cp15 space, we don't use the existing list, since
they need special accessor functions and the arch timer is optional.
Acked-by: Marc Zynger <marc.zyngier@arm.com>
Signed-off-by: Andre Przywara <andre.przywara@linaro.org>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 39735a3a39)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The current KVM implementation of PSCI returns INVALID_PARAMETERS if the
waitqueue for the corresponding CPU is not active. This does not seem
correct, since KVM should not care what the specific thread is doing,
for example, user space may not have called KVM_RUN on this VCPU yet or
the thread may be busy looping to user space because it received a
signal; this is really up to the user space implementation. Instead we
should check specifically that the CPU is marked as being turned off,
regardless of the VCPU thread state, and if it is, we shall
simply clear the pause flag on the CPU and wake up the thread if it
happens to be blocked for us.
Further, the implementation seems to be racy when executing multiple
VCPU threads. There really isn't a reasonable user space programming
scheme to ensure all secondary CPUs have reached kvm_vcpu_first_run_init
before turning on the boot CPU.
Therefore, set the pause flag on the vcpu at VCPU init time (which can
reasonably be expected to be completed for all CPUs by user space before
running any VCPUs) and clear both this flag and the feature (in case the
feature can somehow get set again in the future) and ping the waitqueue
on turning on a VCPU using PSCI.
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 478a8237f6)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Upon CPU shutdown and consequent warm-reboot, the hypervisor CPU state
must be re-initialized. This patch implements a CPU PM notifier that
upon warm-boot calls a KVM hook to reinitialize properly the hypervisor
state so that the CPU can be safely resumed.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
(cherry picked from commit 1fcf7ce0c6)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Since the commit 15ad7146 ("KVM: Use the scheduler preemption notifiers
to make kvm preemptible"), the remaining stuff in this function is a
simple cond_resched() call with an extra need_resched() check which was
there to avoid dropping VCPUs unnecessarily. Now it is meaningless.
Signed-off-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit c08ac06ab3)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
In multiple functions the vcpu_id is used as an offset into a bitfield. Ag
malicious user could specify a vcpu_id greater than 255 in order to set or
clear bits in kernel memory. This could be used to elevate priveges in the
kernel. This patch verifies that the vcpu_id provided is less than 255.
The api documentation already specifies that the vcpu_id must be less than
max_vcpus, but this is currently not checked.
Reported-by: Andrew Honig <ahonig@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Andrew Honig <ahonig@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 338c7dbadd)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
KVM initialisation fails on architectures implementing virt_to_idmap()
because virt_to_phys() on such architectures won't fetch you the correct
idmap page.
So update the KVM ARM code to use the virt_to_idmap() to fix the issue.
Since the KVM code is shared between arm and arm64, we create
kvm_virt_to_phys() and handle the redirection in respective headers.
Cc: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 4fda342cc7)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Using the address of 'empty_zero_page' as source address in order to
clear a page is wrong. On some architectures empty_zero_page is only the
pointer to the struct page of the empty_zero_page. Therefore the clear
page operation would copy the contents of a couple of struct pages instead
of clearing a page. For kvm only arm/arm64 are affected by this bug.
To fix this use the ZERO_PAGE macro instead which will return the struct
page address of the empty_zero_page on all architectures.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
(cherry picked from commit 8a3caa6d74)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Using virt_to_phys on percpu mappings is horribly wrong as it may be
backed by vmalloc. Introduce kvm_kaddr_to_phys which translates both
types of valid kernel addresses to the corresponding physical address.
At the same time resolves a typing issue where we were storing the
physical address as a 32 bit unsigned long (on arm), truncating the
physical address for addresses above the 4GB limit. This caused
breakage on Keystone.
Cc: <stable@vger.kernel.org> [3.10+]
Reported-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Tested-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 40c2729bab)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
When booting a vcpu using PSCI, make sure we start it with the
endianness of the caller. Otherwise, secondaries can be pretty
unhappy to execute a BE kernel in LE mode...
This conforms to PSCI spec Rev B, 5.13.3.
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit ce94fe93d5)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Do the necessary byteswap when host and guest have different
views of the universe. Actually, the only case we need to take
care of is when the guest is BE. All the other cases are naturally
handled.
Also be careful about endianness when the data is being memcopy-ed
from/to the run buffer.
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 6d89d2d9b5)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
It was used in conjunction with KVM_SET_MEMORY_REGION ioctl which was
removed by b74a07beed in 2010, QEMU stopped using it in 2008, so
it is time to remove the code finally.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
(cherry picked from commit 80f5b5e700)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
So far we've succeeded at making KVM and VFIO mostly unaware of each
other, but areas are cropping up where a connection beyond eventfds
and irqfds needs to be made. This patch introduces a KVM-VFIO device
that is meant to be a gateway for such interaction. The user creates
the device and can add and remove VFIO groups to it via file
descriptors. When a group is added, KVM verifies the group is valid
and gets a reference to it via the VFIO external user interface.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit ec53500fae)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Add a kvm ioctl which states which system functionality kvm emulates.
The format used is that of CPUID and we return the corresponding CPUID
bits set for which we do emulate functionality.
Make sure ->padding is being passed on clean from userspace so that we
can use it for something in the future, after the ioctl gets cast in
stone.
s/kvm_dev_ioctl_get_supported_cpuid/kvm_dev_ioctl_get_cpuid/ while at
it.
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 9c15bb1d0a)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
I don't know if this was due to cut and paste, or somebody was really
using a D20 to pick the error code for kvm_init_debugfs as suggested by
Linus (EFAULT is 14, so the possibility cannot be entirely ruled out).
In any case, this patch fixes it.
Reported-by: Tim Gardner <tim.gardner@canonical.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 0c8eb04a62)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
On an (even slightly) oversubscribed system, spinlocks are quickly
becoming a bottleneck, as some vcpus are spinning, waiting for a
lock to be released, while the vcpu holding the lock may not be
running at all.
The solution is to trap blocking WFEs and tell KVM that we're
now spinning. This ensures that other vpus will get a scheduling
boost, allowing the lock to be released more quickly. Also, using
CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT slightly improves the performance
when the VM is severely overcommited.
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit d241aac798)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
In kvm_iommu_map_pages(), we need to know the page size via call
kvm_host_page_size(). And it will check whether the target slot
is valid before return the right page size.
Currently, we will map the iommu pages when creating a new slot.
But we call kvm_iommu_map_pages() during preparing the new slot.
At that time, the new slot is not visible by domain(still in preparing).
So we cannot get the right page size from kvm_host_page_size() and
this will break the IOMMU super page logic.
The solution is to map the iommu pages after we insert the new slot
into domain.
Signed-off-by: Yang Zhang <yang.z.zhang@Intel.com>
Tested-by: Patrick Lu <patrick.lu@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit e0230e1327)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The KVM PSCI code blindly assumes that vcpu_id and MPIDR are
the same thing. This is true when vcpus are organized as a flat
topology, but is wrong when trying to emulate any other topology
(such as A15 clusters).
Change the KVM PSCI CPU_ON code to look at the MPIDR instead
of the vcpu_id to pick a target CPU.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 79c648806f)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The L2CTLR register contains the number of CPUs in this cluster.
Make sure the register content is actually relevant to the vcpu
that is being configured by computing the number of cores that are
part of its cluster.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 9cbb6d969c)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
In order to be able to support more than 4 A7 or A15 CPUs,
we need to fix the MPIDR computing to reflect the fact that
both A15 and A7 can only exist in clusters of at most 4 CPUs.
Fix the MPIDR computing to allow virtual clusters to be exposed
to the guest.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 2d1d841bd4)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Support transparent huge pages in KVM/ARM and KVM/ARM64. The
transparent_hugepage_adjust is not very pretty, but this is also how
it's solved on x86 and seems to be simply an artifact on how THPs
behave. This should eventually be shared across architectures if
possible, but that can always be changed down the road.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 9b5fdb9781)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Support huge pages in KVM/ARM and KVM/ARM64. The pud_huge checking on
the unmap path may feel a bit silly as the pud_huge check is always
defined to false, but the compiler should be smart about this.
Note: This deals only with VMAs marked as huge which are allocated by
users through hugetlbfs only. Transparent huge pages can only be
detected by looking at the underlying pages (or the page tables
themselves) and this patch so far simply maps these on a page-by-page
level in the Stage-2 page tables.
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit ad361f093c)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Update comments to reflect what is really going on and add the TWE bit
to the comments in kvm_arm.h.
Also renames the function to kvm_handle_wfx like is done on arm64 for
consistency and uber-correctness.
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 86ed81aa2e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
On an (even slightly) oversubscribed system, spinlocks are quickly
becoming a bottleneck, as some vcpus are spinning, waiting for a
lock to be released, while the vcpu holding the lock may not be
running at all.
This creates contention, and the observed slowdown is 40x for
hackbench. No, this isn't a typo.
The solution is to trap blocking WFEs and tell KVM that we're
now spinning. This ensures that other vpus will get a scheduling
boost, allowing the lock to be released more quickly. Also, using
CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT slightly improves the performance
when the VM is severely overcommited.
Quick test to estimate the performance: hackbench 1 process 1000
2xA15 host (baseline): 1.843s
2xA15 guest w/o patch: 2.083s
4xA15 guest w/o patch: 80.212s
8xA15 guest w/o patch: Could not be bothered to find out
2xA15 guest w/ patch: 2.102s
4xA15 guest w/ patch: 3.205s
8xA15 guest w/ patch: 6.887s
So we go from a 40x degradation to 1.5x in the 2x overcommit case,
which is vaguely more acceptable.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 58d5ec8f8e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
