KVM: ARM/arm64: fix non-const declaration of function returning const

Sparse kicks up about a type mismatch for kvm_target_cpu:

arch/arm64/kvm/guest.c:271:25: error: symbol 'kvm_target_cpu' redeclared with different type (originally declared at ./arch/arm64/include/asm/kvm_host.h:45) - different modifiers

so fix this by adding the missing const attribute to the function
declaration.

Cc: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 6951e48bff0b55d2a8e825a953fc1f8e3a34bf1c)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm/arm64: KVM: Support KVM_CAP_READONLY_MEM

When userspace loads code and data in a read-only memory regions, KVM
needs to be able to handle this on arm and arm64.  Specifically this is
used when running code directly from a read-only flash device; the
common scenario is a UEFI blob loaded with the -bios option in QEMU.

Note that the MMIO exit on writes to a read-only memory is ABI and can
be used to emulate block-erase style flash devices.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 98047888bb9fd57734028c44ec17413ddd623958)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Introduce gfn_to_hva_memslot_prot

To support read-only memory regions on arm and arm64, we have a need to
resolve a gfn to an hva given a pointer to a memslot to avoid looping
through the memslots twice and to reuse the hva error checking of
gfn_to_hva_prot(), add a new gfn_to_hva_memslot_prot() function and
refactor gfn_to_hva_prot() to use this function.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 64d831269ccbca1fc6d739a0f3c8aa24afb43a5e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: add kvm_arch_sched_in

Introduce preempt notifiers for architecture specific code.
Advantage over creating a new notifier in every arch is slightly simpler
code and guaranteed call order with respect to kvm_sched_in.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit e790d9ef6405633b007339d746b709aed43a928d)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: avoid unnecessary synchronize_rcu

We dont have to wait for a grace period if there is no oldpid that
we are going to free. putpid also checks for NULL, so this patch
only fences synchronize_rcu.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 7103f60de8bed21a0ad5d15d2ad5b7a333dda201)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Move more code under CONFIG_HAVE_KVM_IRQFD

Commits e4d57e1ee1ab (KVM: Move irq notifier implementation into
eventfd.c, 2014-06-30) included the irq notifier code unconditionally
in eventfd.c, while it was under CONFIG_HAVE_KVM_IRQCHIP before.

Similarly, commit 297e21053a52 (KVM: Give IRQFD its own separate enabling
Kconfig option, 2014-06-30) moved code from CONFIG_HAVE_IRQ_ROUTING
to CONFIG_HAVE_KVM_IRQFD but forgot to move the pieces that used to be
under CONFIG_HAVE_KVM_IRQCHIP.

Together, this broke compilation without CONFIG_KVM_XICS.  Fix by adding
or changing the #ifdefs so that they point at CONFIG_HAVE_KVM_IRQFD.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit c77dcacb397519b6ade8f08201a4a90a7f4f751e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Give IRQFD its own separate enabling Kconfig option

Currently, the IRQFD code is conditional on CONFIG_HAVE_KVM_IRQ_ROUTING.
So that we can have the IRQFD code compiled in without having the
IRQ routing code, this creates a new CONFIG_HAVE_KVM_IRQFD, makes
the IRQFD code conditional on it instead of CONFIG_HAVE_KVM_IRQ_ROUTING,
and makes all the platforms that currently select HAVE_KVM_IRQ_ROUTING
also select HAVE_KVM_IRQFD.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Tested-by: Eric Auger <eric.auger@linaro.org>
Tested-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 297e21053a52f060944e9f0de4c64fad9bcd72fc)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Move irq notifier implementation into eventfd.c

This moves the functions kvm_irq_has_notifier(), kvm_notify_acked_irq(),
kvm_register_irq_ack_notifier() and kvm_unregister_irq_ack_notifier()
from irqchip.c to eventfd.c.  The reason for doing this is that those
functions are used in connection with IRQFDs, which are implemented in
eventfd.c.  In future we will want to use IRQFDs on platforms that
don't implement the GSI routing implemented in irqchip.c, so we won't
be compiling in irqchip.c, but we still need the irq notifiers.  The
implementation is unchanged.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Tested-by: Eric Auger <eric.auger@linaro.org>
Tested-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit e4d57e1ee1ab59f0cef0272800ac6c52e0ec814a)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Move all accesses to kvm::irq_routing into irqchip.c

Now that struct _irqfd does not keep a reference to storage pointed
to by the irq_routing field of struct kvm, we can move the statement
that updates it out from under the irqfds.lock and put it in
kvm_set_irq_routing() instead.  That means we then have to take a
srcu_read_lock on kvm->irq_srcu around the irqfd_update call in
kvm_irqfd_assign(), since holding the kvm->irqfds.lock no longer
ensures that that the routing can't change.

Combined with changing kvm_irq_map_gsi() and kvm_irq_map_chip_pin()
to take a struct kvm * argument instead of the pointer to the routing
table, this allows us to to move all references to kvm->irq_routing
into irqchip.c.  That in turn allows us to move the definition of the
kvm_irq_routing_table struct into irqchip.c as well.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Tested-by: Eric Auger <eric.auger@linaro.org>
Tested-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 9957c86d659a4d5a2bed25ccbd3bfc9c3f25e658)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: irqchip: Provide and use accessors for irq routing table

This provides accessor functions for the KVM interrupt mappings, in
order to reduce the amount of code that accesses the fields of the
kvm_irq_routing_table struct, and restrict that code to one file,
virt/kvm/irqchip.c.  The new functions are kvm_irq_map_gsi(), which
maps from a global interrupt number to a set of IRQ routing entries,
and kvm_irq_map_chip_pin, which maps from IRQ chip and pin numbers to
a global interrupt number.

This also moves the update of kvm_irq_routing_table::chip[][]
into irqchip.c, out of the various kvm_set_routing_entry
implementations.  That means that none of the kvm_set_routing_entry
implementations need the kvm_irq_routing_table argument anymore,
so this removes it.

This does not change any locking or data lifetime rules.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Tested-by: Eric Auger <eric.auger@linaro.org>
Tested-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 8ba918d488caded2c4368b0b922eb905fe3bb101)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Don't keep reference to irq routing table in irqfd struct

This makes the irqfd code keep a copy of the irq routing table entry
for each irqfd, rather than a reference to the copy in the actual
irq routing table maintained in kvm/virt/irqchip.c.  This will enable
us to change the routing table structure in future, or even not have a
routing table at all on some platforms.

The synchronization that was previously achieved using srcu_dereference
on the read side is now achieved using a seqcount_t structure.  That
ensures that we don't get a halfway-updated copy of the structure if
we read it while another thread is updating it.

We still use srcu_read_lock/unlock around the read side so that when
changing the routing table we can be sure that after calling
synchronize_srcu, nothing will be using the old routing.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Tested-by: Eric Auger <eric.auger@linaro.org>
Tested-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 56f89f3629ffd1a21d38c3d0bea23deac0e284ce)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: fix 64bit CP15 VM access for 32bit guests

Commit f0a3eaff71b8 (ARM64: KVM: fix big endian issue in
access_vm_reg for 32bit guest) changed the way we handle CP15
VM accesses, so that all 64bit accesses are done via vcpu_sys_reg.

This looks like a good idea as it solves indianness issues in an
elegant way, except for one small detail: the register index is
doesn't refer to the same array! We end up corrupting some random
data structure instead.

Fix this by reverting to the original code, except for the introduction
of a vcpu_cp15_64_high macro that deals with the endianness thing.

Tested on Juno with 32bit SMP guests.

Cc: Victor Kamensky <victor.kamensky@linaro.org>
Reviewed-by: 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 dedf97e8ff2c7513b1370e36b56e08b6bd0f0290)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: arm64: GICv3: mandate page-aligned GICV region

Just like GICv2 was fixed in 63afbe7a0ac1
(kvm: arm64: vgic: fix hyp panic with 64k pages on juno platform),
mandate the GICV region to be both aligned on a page boundary and
its size to be a multiple of page size.

This prevents a guest from being able to poke at regions where we
have no idea what is sitting there.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit fb3ec67942e92e5713e05b7691b277d0a0c0575d)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: GICv3: move system register access to msr_s/mrs_s

Commit 72c583951526 (arm64: gicv3: Allow GICv3 compilation with
older binutils) changed the way we express the GICv3 system registers,
but couldn't change the occurences used by KVM as the code wasn't
merged yet.

Just fix the accessors.

Cc: Will Deacon <will.deacon@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit f4c321eb268e932786c112e0b902ee942d91a336)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

kvm: arm64: vgic: fix hyp panic with 64k pages on juno platform

If the physical address of GICV isn't page-aligned, then we end up
creating a stage-2 mapping of the page containing it, which causes us to
map neighbouring memory locations directly into the guest.

As an example, consider a platform with GICV at physical 0x2c02f000
running a 64k-page host kernel. If qemu maps this into the guest at
0x80010000, then guest physical addresses 0x80010000 - 0x8001efff will
map host physical region 0x2c020000 - 0x2c02efff. Accesses to these
physical regions may cause UNPREDICTABLE behaviour, for example, on the
Juno platform this will cause an SError exception to EL3, which brings
down the entire physical CPU resulting in RCU stalls / HYP panics / host
crashing / wasted weeks of debugging.

SBSA recommends that systems alias the 4k GICV across the bounding 64k
region, in which case GICV physical could be described as 0x2c020000 in
the above scenario.

This patch fixes the problem by failing the vgic probe if the physical
base address or the size of GICV aren't page-aligned. Note that this
generated a warning in dmesg about freeing enabled IRQs, so I had to
move the IRQ enabling later in the probe.

Cc: Christoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Joel Schopp <joel.schopp@amd.com>
Cc: Don Dutile <ddutile@redhat.com>
Acked-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Joel Schopp <joel.schopp@amd.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 63afbe7a0ac184ef8485dac4914e87b211b5bfaa)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Allow KVM_CHECK_EXTENSION on the vm fd

The KVM_CHECK_EXTENSION is only available on the kvm fd today. Unfortunately
on PPC some of the capabilities change depending on the way a VM was created.

So instead we need a way to expose capabilities as VM ioctl, so that we can
see which VM type we're using (HV or PR). To enable this, add the
KVM_CHECK_EXTENSION ioctl to our vm ioctl portfolio.

Signed-off-by: Alexander Graf <agraf@suse.de>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 92b591a4c46b103ebd3fc0d03a084e1efd331253)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Rename and add argument to check_extension

In preparation to make the check_extension function available to VM scope
we add a struct kvm * argument to the function header and rename the function
accordingly. It will still be called from the /dev/kvm fd, but with a NULL
argument for struct kvm *.

Signed-off-by: Alexander Graf <agraf@suse.de>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 784aa3d7fb6f729c06d5836c9d9569f58e4d05ae)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

kvm: Resolve missing-field-initializers warnings

Resolve missing-field-initializers warnings seen in W=2 kernel
builds by having macros generate more elaborated initializers.
That is enough to silence the warnings.

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 25f97ff451a4aab534afc1290af97d23ea0b4fb3)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: make it easier to check the CPU part number correctly

Ensure that platform maintainers check the CPU part number in the right
manner: the CPU part number is meaningless without also checking the
CPU implement(e|o)r (choose your preferred spelling!)  Provide an
interface which returns both the implementer and part number together,
and update the definitions to include the implementer.

Mark the old function as being deprecated... indeed, using the old
function with the definitions will now always evaluate as false, so
people must update their un-merged code to the new function.  While
this could be avoided by adding new definitions, we'd also have to
create new names for them which would be awkward.

Acked-by: Nicolas Pitre <nico@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
(cherry picked from commit af040ffc9ba1e079ee4c0748aff64fa3d4716fa5)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: enable trapping of all debug registers

Enable trapping of the debug registers, preventing the guests to
mess with the host state (and allowing guests to use the debug
infrastructure as well).

Reviewed-by: Anup Patel <anup.patel@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit d329de09333aeee127aaf22eb7cee9c2dc4cf475)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: implement lazy world switch for debug registers

Implement switching of the debug registers. While the number
of registers is massive, CPUs usually don't implement them all
(A57 has 6 breakpoints and 4 watchpoints, which gives us a total
of 22 registers "only").

Also, we only save/restore them when MDSCR_EL1 has debug enabled,
or when we've flagged the debug registers as dirty. It means that
most of the time, we only save/restore MDSCR_EL1.

Reviewed-by: Anup Patel <anup.patel@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit b0e626b380872b663918230fafdac128c34fea56)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: add trap handlers for AArch32 debug registers

Add handlers for all the AArch32 debug registers that are accessible
from EL0 or EL1. The code follow the same strategy as the AArch64
counterpart with regards to tracking the dirty state of the debug
registers.

Reviewed-by: Anup Patel <anup.patel@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit bdfb4b389c8d8f07e2d5b8e1291e01c789ba4aad)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: check ordering of all system register tables

We now have multiple tables for the various system registers
we trap. Make sure we check the order of all of them, as it is
critical that we get the order right (been there, done that...).

Reviewed-by: Anup Patel <anup.patel@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit e6a9551760623d1703487e8a16bb9c3ea8a7e7a8)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: use separate tables for AArch32 32 and 64bit traps

An interesting "feature" of the CP14 encoding is that there is
an overlap between 32 and 64bit registers, meaning they cannot
live in the same table as we did for CP15.

Create separate tables for 64bit CP14 and CP15 registers, and
let the top level handler use the right one.

Reviewed-by: Anup Patel <anup.patel@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit a9866ba0cddfc497335fa02a175c4578b96722ff)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: common infrastructure for handling AArch32 CP14/CP15

As we're about to trap a bunch of CP14 registers, let's rework
the CP15 handling so it can be generalized and work with multiple
tables.

Reviewed-by: Anup Patel <anup.patel@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 72564016aae45f42e488f926bc803f9a2e1c771c)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: add trap handlers for AArch64 debug registers

Add handlers for all the AArch64 debug registers that are accessible
from EL0 or EL1. The trapping code keeps track of the state of the
debug registers, allowing for the switch code to implement a lazy
switching strategy.

Reviewed-by: Anup Patel <anup.patel@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 0c557ed4983b7abe152212b5b1726c2a789b2c61)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: move DBG_MDSCR_* to asm/debug-monitors.h

In order to be able to use the DBG_MDSCR_* macros from the KVM code,
move the relevant definitions to the obvious include file.

Also move the debug_el enum to a portion of the file that is guarded
by #ifndef __ASSEMBLY__ in order to use that file from assembly code.

Acked-by: Will Deacon <will.deacon@arm.com>
Reviewed-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 51ba248164d0eeb8b4f94d405430c18a56c6ac9a)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: rename pm_fake handler to trap_raz_wi

pm_fake doesn't quite describe what the handler does (ignoring writes
and returning 0 for reads).

As we're about to use it (a lot) in a different context, rename it
with a (admitedly cryptic) name that make sense for all users.

Reviewed-by: Anup Patel <anup.patel@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 7609c1251f9d8bbcd6a05ba22153e50cf4f88cff)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM64: KVM: fix big endian issue in access_vm_reg for 32bit guest

Fix issue with 32bit guests running on top of BE KVM host.
Indexes of high and low words of 64bit cp15 register are
swapped in case of big endian code, since 64bit cp15 state is
restored or saved with double word write or read instruction.

Define helper macro to access low words of 64bit cp15 register.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit f0a3eaff71b8bd5d5acfda1f0cf3eedf49755622)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM64: KVM: set and get of sys registers in BE case

Since size of all sys registers is always 8 bytes. Current
code is actually endian agnostic. Just clean it up a bit.
Removed comment about little endian. Change type of pointer
from 'void *' to 'u64 *' to enforce stronger type checking.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 26c99af1018c35020cfad1d20f02acb224807655)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM64: KVM: fix vgic_bitmap_get_reg function for BE 64bit case

Fix vgic_bitmap_get_reg function to return 'right' word address of
'unsigned long' bitmap value in case of BE 64bit image.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 9662fb4854e1319b4affda47f279c3f210316def)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM64: KVM: store kvm_vcpu_fault_info est_el2 as word

esr_el2 field of struct kvm_vcpu_fault_info has u32 type.
It should be stored as word. Current code works in LE case
because existing puts least significant word of x1 into
esr_el2, and it puts most significant work of x1 into next
field, which accidentally is OK because it is updated again
by next instruction. But existing code breaks in BE case.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit ba083d20d8cfa9e999043cd89c4ebc964ccf8927)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM64: KVM: MMIO support BE host running LE code

In case of guest CPU running in LE mode and host runs in
BE mode we need byteswap data, so read/write is emulated correctly.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit b30070862edbdb252f9d0d3a1e61b8dc4c68e3d2)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: enable KVM in Kconfig on big-endian systems

Previous patches addresses ARMV7 big-endian virtualiztion,
kvm related issues, so enable ARM_VIRT_EXT for big-endian
now.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit f5aa462147a209dab40b02f0f70234784b913bf9)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: one_reg coproc set and get BE fixes

Fix code that handles KVM_SET_ONE_REG, KVM_GET_ONE_REG ioctls to work in BE
image. Before this fix get/set_one_reg functions worked correctly only in
LE case - reg_from_user was taking 'void *' kernel address that actually could
be target/source memory of either 4 bytes size or 8 bytes size, and code copied
from/to user memory that could hold either 4 bytes register, 8 byte register
or pair of 4 bytes registers.

In order to work in endian agnostic way reg_from_user to reg_to_user functions
should copy register value only to kernel variable with size that matches
register size. In few place where size mismatch existed fix issue on macro
caller side.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 73891f72c414afff6da6f01e7af2ff5a44a8b823)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: MMIO support BE host running LE code

In case of status register E bit is not set (LE mode) and host runs in
BE mode we need byteswap data, so read/write is emulated correctly.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 27f194fd360a96cc64bebb2d69dd5abd67984b8a)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: vgic mmio should hold data as LE bytes array in BE case

According to recent clarifications of mmio.data array meaning -
the mmio.data array should hold bytes as they would appear in
memory. Vgic is little endian device. And in case of BE image
kernel side that emulates vgic, holds data in BE form. So we
need to byteswap cpu<->le32 vgic registers when we read/write them
from mmio.data[].

Change has no effect in LE case because cpu already runs in le32.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 1c9f04717ca8326e8df759d5dda9cd1b3d968b5b)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: __kvm_vcpu_run function return result fix in BE case

The __kvm_vcpu_run function returns a 64-bit result in two registers,
which has to be adjusted for BE case.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 6d7311b520864531c81f0e0237e96146d8057d77)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: handle 64bit values passed to mrcc or from mcrr instructions in BE case

In some cases the mcrr and mrrc instructions in combination with the ldrd
and strd instructions need to deal with 64bit value in memory. The ldrd
and strd instructions already handle endianness within word (register)
boundaries but to get effect of the whole 64bit value represented correctly,
rr_lo_hi macro is introduced and is used to swap registers positions when
the mcrr and mrrc instructions are used. That has the effect of swapping
two words.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 19b0e60a63f758a28329aa40f4270a6c98c2dcb7)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: fix vgic V7 assembler code to work in BE image

The vgic h/w registers are little endian; when BE asm code
reads/writes from/to them, it needs to do byteswap after/before.
Byteswap code uses ARM_BE8 wrapper to add swap only if
CONFIG_CPU_BIG_ENDIAN is configured.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 64054c25cf7e060cd6780744fefe7ed3990e4f21)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: vgic: enable GICv2 emulation on top on GICv3 hardware

Add the last missing bits that enable GICv2 emulation on top of
GICv3 hardware.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 67b2abfedb7b861bead93400fa315c5c30879d51)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: vgic: add GICv3 world switch

Introduce the GICv3 world switch code used to save/restore the
GICv3 context.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 754d37726010d872f1f714a8ce8920acdfa4978c)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: add the GICv3 backend

Introduce the support code for emulating a GICv2 on top of GICv3
hardware.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit b2fb1c0d378399e1427a91bb991c094f2ca09a2f)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: move HCR_EL2.{IMO,FMO} manipulation into the vgic switch code

GICv3 requires the IMO and FMO bits to be tightly coupled with some
of the interrupt controller's register switch.

In order to have similar code paths, move the manipulation of these
bits to the GICv2 switch code.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit ac3c3747e2db2f326ffc601651de544cdd33a8e9)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: split GICv2 world switch from hyp code

Move the GICv2 world switch code into its own file, and add the
necessary indirection to the arm64 switch code.

Also introduce a new type field to the vgic_params structure.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 1a9b13056dde7e3092304d6041ccc60a913042ea)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: remove __kvm_hyp_code_{start,end} from hyp.S

We already have __hyp_text_{start,end} to express the boundaries
of the HYP text section, and __kvm_hyp_code_{start,end} are getting
in the way of a more modular world switch code.

Just turn __kvm_hyp_code_{start,end} into #defines mapping the
linker-emited symbols.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 45451914c875bba44903ce4f1445e047b7992bf7)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: revisit implementation of irqchip_in_kernel

So far, irqchip_in_kernel() was implemented by testing the value of
vctrl_base, which worked fine with GICv2.

With GICv3, this field is useless, as we're using system registers
instead of a emmory mapped interface. To solve this, add a boolean
flag indicating if the we're using a vgic or not.

Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit f982cf4e9c37b19478c7bc6e0484a43a7e78cf57)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: split GICv2 backend from the main vgic code

Brutally hack the innocent vgic code, and move the GICv2 specific code
to its own file, using vgic_ops and vgic_params as a way to pass
information between the two blocks.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 8f186d522c69bb18dd9b93a634da4953228c67d4)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: introduce vgic_params structure

Move all the data specific to a given GIC implementation into its own
little structure.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit ca85f623e37d096206e092ef037a145a60fa7f85)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: introduce vgic_enable

Move the code dealing with enabling the VGIC on to vgic_ops.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit da8dafd1777cdd93091207952297d221a88e6479)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: abstract VMCR access

Instead of directly messing with with the GICH_VMCR bits for the CPU
interface save/restore code, add accessors that encode/decode the
entire set of registers exposed by VMCR.

Not the most efficient thing, but given that this code is only used
by the save/restore code, performance is far from being critical.

Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit beee38b9d0c0ea6cf2a7f35c3108f7d8281d4545)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: move underflow handling to vgic_ops

Move the code dealing with LR underflow handling to its own functions,
and make them accessible through vgic_ops.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 909d9b5025f149af6cfc304a76ad6218e6622cc0)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: abstract MISR decoding

Instead of directly dealing with the GICH_MISR bits, move the code to
its own function and use a couple of public flags to represent the
actual state.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 495dd859f304689a7cd5ef413c439cb090dc25e6)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: abstract EISR bitmap access

Move the GICH_EISR access to its own function.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 8d6a0313c125c3c7b208b75695fe6ab00afab4c5)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: abstract access to the ELRSR bitmap

Move the GICH_ELRSR access to its own functions, and add them to
the vgic_ops structure.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 69bb2c9fbc11d9d4358fbb798db15c9092eb4d8c)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: introduce vgic_ops and LR manipulation primitives

In order to split the various register manipulation from the main vgic
code, introduce a vgic_ops structure, and start by abstracting the
LR manipulation code with a couple of accessors.

Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit 8d5c6b06a5d5f8ebcf40558e566781d572920740)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: arm/arm64: vgic: move GICv2 registers to their own structure

In order to make way for the GICv3 registers, move the v2-specific
registers to their own structure.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit eede821dbfd58df89edb072da64e006321eaef58)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: allow export and import of generic timer regs

For correct guest suspend/resume behaviour we need to ensure we include
the generic timer registers for 64 bit guests. As CONFIG_KVM_ARM_TIMER is
always set for arm64 we don't need to worry about null implementations.
However I have re-jigged the kvm_arm_timer_set/get_reg declarations to
be in the common include/kvm/arm_arch_timer.h headers.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 1df08ba0aa95f1a8832b7162eec51069bd9be7ae)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: export demux regids as KVM_REG_ARM64

I suspect this is a -ECUTPASTE fault from the initial implementation. If
we don't declare the register ID to be KVM_REG_ARM64 the KVM_GET_ONE_REG
implementation kvm_arm_get_reg() returns -EINVAL and hilarity ensues.

The kvm/api.txt document describes all arm64 registers as starting with
0x60xx... (i.e KVM_REG_ARM64).

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit efd48ceacea78e4d4656aa0a6bf4c5b92ed22130)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: user_mem_abort: support stage 2 MMIO page mapping

A userspace process can map device MMIO memory via VFIO or /dev/mem,
e.g., for platform device passthrough support in QEMU.

During early development, we found the PAGE_S2 memory type being used
for MMIO mappings.  This patch corrects that by using the more strongly
ordered memory type for device MMIO mappings: PAGE_S2_DEVICE.

Signed-off-by: Kim Phillips <kim.phillips@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
(cherry picked from commit b88657674d39fc2127d62d0de9ca142e166443c8)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: Unmap IPA on memslot delete/move

Currently when a KVM region is deleted or moved after
KVM_SET_USER_MEMORY_REGION ioctl, the corresponding
intermediate physical memory is not unmapped.

This patch corrects this and unmaps the region's IPA range
in kvm_arch_commit_memory_region using unmap_stage2_range.

Signed-off-by: Eric Auger <eric.auger@linaro.org>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit df6ce24f2ee485c4f9a5cb610063a5eb60da8267)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm/arm64: KVM: Fix and refactor unmap_range

unmap_range() was utterly broken, to quote Marc, and broke in all sorts
of situations.  It was also quite complicated to follow and didn't
follow the usual scheme of having a separate iterating function for each
level of page tables.

Address this by refactoring the code and introduce a pgd_clear()
function.

Reviewed-by: Jungseok Lee <jays.lee@samsung.com>
Reviewed-by: Mario Smarduch <m.smarduch@samsung.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 4f853a714bf16338ff5261128e6c7ae2569e9505)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: prepare for KVM_(S|G)ET_MP_STATE on other architectures

Highlight the aspects of the ioctls that are actually specific to x86
and ia64. As defined restrictions (irqchip) and mp states may not apply
to other architectures, these parts are flagged to belong to x86 and ia64.

In preparation for the use of KVM_(S|G)ET_MP_STATE by s390.
Fix a spelling error (KVM_SET_MP_STATE vs. KVM_SET_MPSTATE) on the way.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
(cherry picked from commit 0b4820d6d8b6448bc9f7fac1bb1a801a53b425e1)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: add missing cleanup_srcu_struct

Reported-by: hrg <hrgstephen@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 820b3fcdeb80d30410f4427d2cbf9161c35fdeef)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: Enable minimalistic support for Cortex-A53

In order to allow KVM to run on Cortex-A53 implementations, wire the
minimal support required.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 1252b3313642c3d0dff5b951b625468bf0dcd059)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: kvm: use inner-shareable barriers for inner-shareable maintenance

In order to ensure completion of inner-shareable maintenance instructions
(cache and TLB) on AArch64, we can use the -ish suffix to the dsb
instruction.

This patch relaxes our dsb sy instructions to dsb ish where possible.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit ee9e101c11478680d579bd20bb38a4d3e2514fe3)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: barriers: make use of barrier options with explicit barriers

When calling our low-level barrier macros directly, we can often suffice
with more relaxed behaviour than the default "all accesses, full system"
option.

This patch updates the users of dsb() to specify the option which they
actually require.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit 98f7685ee69f871ba991089cb9685f0da07517ea)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

kvm/irqchip: Speed up KVM_SET_GSI_ROUTING

When starting lots of dataplane devices the bootup takes very long on
Christian's s390 with irqfd patches. With larger setups he is even
able to trigger some timeouts in some components.  Turns out that the
KVM_SET_GSI_ROUTING ioctl takes very long (strace claims up to 0.1 sec)
when having multiple CPUs.  This is caused by the  synchronize_rcu and
the HZ=100 of s390.  By changing the code to use a private srcu we can
speed things up.  This patch reduces the boot time till mounting root
from 8 to 2 seconds on my s390 guest with 100 disks.

Uses of hlist_for_each_entry_rcu, hlist_add_head_rcu, hlist_del_init_rcu
are fine because they do not have lockdep checks (hlist_for_each_entry_rcu
uses rcu_dereference_raw rather than rcu_dereference, and write-sides
do not do rcu lockdep at all).

Note that we're hardly relying on the "sleepable" part of srcu.  We just
want SRCU's faster detection of grace periods.

Testing was done by Andrew Theurer using netperf tests STREAM, MAERTS
and RR.  The difference between results "before" and "after" the patch
has mean -0.2% and standard deviation 0.6%.  Using a paired t-test on the
data points says that there is a 2.5% probability that the patch is the
cause of the performance difference (rather than a random fluctuation).

(Restricting the t-test to RR, which is the most likely to be affected,
changes the numbers to respectively -0.3% mean, 0.7% stdev, and 8%
probability that the numbers actually say something about the patch.
The probability increases mostly because there are fewer data points).

Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # s390
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 719d93cd5f5c5c8775b7a38192069e8e1d1ac46e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Advertise KVM_CAP_ARM_PSCI_0_2 to user space

We have PSCI v0.2 emulation available in KVM ARM/ARM64
hence advertise this to user space (i.e. QEMU or KVMTOOL)
via KVM_CHECK_EXTENSION ioctl.

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 4447a208f7fc2e2dff8c6a8df2a1fd6dd72fb3e2)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Emulate PSCI v0.2 CPU_SUSPEND

This patch adds emulation of PSCI v0.2 CPU_SUSPEND function call for
KVM ARM/ARM64. This is a CPU-level function call which can suspend
current CPU or current CPU cluster. We don't have VCPU clusters in
KVM so we only suspend the current VCPU.

The CPU_SUSPEND emulation is not tested much because currently there
is no CPUIDLE driver in Linux kernel that uses PSCI CPU_SUSPEND. The
PSCI CPU_SUSPEND implementation in ARM64 kernel was tested using a
Simple CPUIDLE driver which is not published due to unstable DT-bindings
for PSCI.
(For more info, http://lwn.net/Articles/574950/)

For simplicity, we implement CPU_SUSPEND emulation similar to WFI
(Wait-for-interrupt) emulation and we also treat power-down request
to be same as stand-by request. This is consistent with section
5.4.1 and section 5.4.2 of PSCI v0.2 specification.

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit b376d02b53b87f8684f91f13ba4ee43331850fcd)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Fix CPU_ON emulation for PSCI v0.2

As-per PSCI v0.2, the source CPU provides physical address of
"entry point" and "context id" for starting a target CPU. Also,
if target CPU is already running then we should return ALREADY_ON.

Current emulation of CPU_ON function does not consider physical
address of "context id" and returns INVALID_PARAMETERS if target
CPU is already running.

This patch updates kvm_psci_vcpu_on() such that it works for both
PSCI v0.1 and PSCI v0.2.

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit aa8aeefe5e567637bbec7d7a3031cc057e3af303)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Emulate PSCI v0.2 MIGRATE_INFO_TYPE and related functions

This patch adds emulation of PSCI v0.2 MIGRATE, MIGRATE_INFO_TYPE, and
MIGRATE_INFO_UP_CPU function calls for KVM ARM/ARM64.

KVM ARM/ARM64 being a hypervisor (and not a Trusted OS), we cannot provide
this functions hence we emulate these functions in following way:
1. MIGRATE - Returns "Not Supported"
2. MIGRATE_INFO_TYPE - Return 2 i.e. Trusted OS is not present
3. MIGRATE_INFO_UP_CPU - Returns "Not Supported"

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit bab0b43012a8ad64877fa46134370a7f5c6ce861)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Emulate PSCI v0.2 AFFINITY_INFO

This patch adds emulation of PSCI v0.2 AFFINITY_INFO function call
for KVM ARM/ARM64. This is a VCPU-level function call which will be
used to determine current state of given affinity level.

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit e6bc13c8a70eabc6a39098ccedf6129c734e3db3)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Emulate PSCI v0.2 SYSTEM_OFF and SYSTEM_RESET

The PSCI v0.2 SYSTEM_OFF and SYSTEM_RESET functions are system-level
functions hence cannot be fully emulated by in-kernel PSCI emulation code.

To tackle this, we forward PSCI v0.2 SYSTEM_OFF and SYSTEM_RESET function
calls from vcpu to user space (i.e. QEMU or KVMTOOL) via kvm_run structure
using KVM_EXIT_SYSTEM_EVENT exit reasons.

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 4b1238269ed340d59ef829fd9c30a39cfb2923a8)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Add KVM_EXIT_SYSTEM_EVENT to user space API header

Currently, we don't have an exit reason to notify user space about
a system-level event (for e.g. system reset or shutdown) triggered
by the VCPU. This patch adds exit reason KVM_EXIT_SYSTEM_EVENT for
this purpose. We can also inform user space about the 'type' and
architecture specific 'flags' of a system-level event using the
kvm_run structure.

This newly added KVM_EXIT_SYSTEM_EVENT will be used by KVM ARM/ARM64
in-kernel PSCI v0.2 support to reset/shutdown VMs.

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 8ad6b634928a25971dc42dce101808b1491f87ec)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Make kvm_psci_call() return convention more flexible

Currently, the kvm_psci_call() returns 'true' or 'false' based on whether
the PSCI function call was handled successfully or not. This does not help
us emulate system-level PSCI functions where the actual emulation work will
be done by user space (QEMU or KVMTOOL). Examples of such system-level PSCI
functions are: PSCI v0.2 SYSTEM_OFF and SYSTEM_RESET.

This patch updates kvm_psci_call() to return three types of values:
1) > 0 (success)
2) = 0 (success but exit to user space)
3) < 0 (errors)

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit e8e7fcc5e2710b31ef842ee799db99c07986c364)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Add base for PSCI v0.2 emulation

Currently, the in-kernel PSCI emulation provides PSCI v0.1 interface to
VCPUs. This patch extends current in-kernel PSCI emulation to provide
PSCI v0.2 interface to VCPUs.

By default, ARM/ARM64 KVM will always provide PSCI v0.1 interface for
keeping the ABI backward-compatible.

To select PSCI v0.2 interface for VCPUs, the user space (i.e. QEMU or
KVMTOOL) will have to set KVM_ARM_VCPU_PSCI_0_2 feature when doing VCPU
init using KVM_ARM_VCPU_INIT ioctl.

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 7d0f84aae9e231930985eaff63ac91b61aaa15d6)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM/ARM64: KVM: Add common header for PSCI related defines

We need a common place to share PSCI related defines among ARM kernel,
ARM64 kernel, KVM ARM/ARM64 PSCI emulation, and user space.

We introduce uapi/linux/psci.h for this purpose. This newly added
header will be first used by KVM ARM/ARM64 in-kernel PSCI emulation
and user space (i.e. QEMU or KVMTOOL).

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Signed-off-by: Ashwin Chaugule <ashwin.chaugule@linaro.org>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit e546eea74ec66698e29c583639cf6e2a11e46490)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: Add capability to advertise PSCI v0.2 support

User space (i.e. QEMU or KVMTOOL) should be able to check whether KVM
ARM/ARM64 supports in-kernel PSCI v0.2 emulation. For this purpose, we
define KVM_CAP_ARM_PSCI_0_2 in KVM user space interface header.

Signed-off-by: Anup Patel <anup.patel@linaro.org>
Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar@linaro.org>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 717abd208dff75b343243aa5ed688f62190dda5e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: Fix the overlap check action about setting the GICD & GICC base address.

Currently below check in vgic_ioaddr_overlap will always succeed,
because the vgic dist base and vgic cpu base are still kept UNDEF
after initialization. The code as follows will be return forever.

if (IS_VGIC_ADDR_UNDEF(dist) || IS_VGIC_ADDR_UNDEF(cpu))
                return 0;

So, before invoking the vgic_ioaddr_overlap, it needs to set the
corresponding base address firstly.

Signed-off-by: Haibin Wang <wanghaibin.wang@huawei.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 30c2117085bc4e05d091cee6eba79f069b41a9cd)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: async_pf: change async_pf_execute() to use get_user_pages(tsk => NULL)

async_pf_execute() passes tsk == current to gup(), this is doesn't
hurt but unnecessary and misleading. "tsk" is only used to account
the number of faults and current is the random workqueue thread.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Suggested-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit e9545b9f8aeb63e05818e4b3250057260bc072aa)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: async_pf: kill the unnecessary use_mm/unuse_mm async_pf_execute()

async_pf_execute() has no reasons to adopt apf->mm, gup(current, mm)
should work just fine even if current has another or NULL ->mm.

Recently kvm_async_page_present_sync() was added insedie the "use_mm"
section, but it seems that it doesn't need current->mm too.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit d72d946d0b649b79709b99b9d5cb7269fff8afaa)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: arm/arm64: vgic: fix GICD_ICFGR register accesses

Since KVM internally represents the ICFGR registers by stuffing two
of them into one word, the offset for accessing the internal
representation and the one for the MMIO based access are different.
So keep the original offset around, but adjust the internal array
offset by one bit.

Reported-by: Haibin Wang <wanghaibin.wang@huawei.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit f2ae85b2ab3776b9e4e42e5b6fa090f40d396794)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: async_pf: mm->mm_users can not pin apf->mm

get_user_pages(mm) is simply wrong if mm->mm_users == 0 and exit_mmap/etc
was already called (or is in progress), mm->mm_count can only pin mm->pgd
and mm_struct itself.

Change kvm_setup_async_pf/async_pf_execute to inc/dec mm->mm_users.

kvm_create_vm/kvm_destroy_vm play with ->mm_count too but this case looks
fine at first glance, it seems that this ->mm is only used to verify that
current->mm == kvm->mm.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 41c22f626254b9dc0376928cae009e73d1b6a49a)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: ARM: vgic: Fix sgi dispatch problem

When dispatch SGI(mode == 0), that is the vcpu of VM should send
sgi to the cpu which the target_cpus list.
So, there must add the "break" to branch of case 0.

Cc: <stable@vger.kernel.org> # 3.10+
Signed-off-by: Haibin Wang <wanghaibin.wang@huawei.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 91021a6c8ffdc55804dab5acdfc7de4f278b9ac3)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm: KVM: fix possible misalignment of PGDs and bounce page

The kvm/mmu code shared by arm and arm64 uses kalloc() to allocate
a bounce page (if hypervisor init code crosses page boundary) and
hypervisor PGDs. The problem is that kalloc() does not guarantee
the proper alignment. In the case of the bounce page, the page sized
buffer allocated may also cross a page boundary negating the purpose
and leading to a hang during kvm initialization. Likewise the PGDs
allocated may not meet the minimum alignment requirements of the
underlying MMU. This patch uses __get_free_page() to guarantee the
worst case alignment needs of the bounce page and PGDs on both arm
and arm64.

Cc: <stable@vger.kernel.org> # 3.10+
Signed-off-by: Mark Salter <msalter@redhat.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 5d4e08c45a6cf8f1ab3c7fa375007635ac569165)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: disable KVM in Kconfig on big-endian systems

KVM currently crashes and burns on big-endian hosts, so don't allow it
to be selected until we've got that fixed.

Cc: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 4e4468fac4381b92eb333d94256e7fb8350f3de3)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

KVM: add kvm_is_error_gpa() helper

It's quite common (in the s390 guest access code) to test if a guest
physical address points to a valid guest memory area or not.
So add a simple helper function in common code, since this might be
of interest for other architectures as well.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
(cherry picked from commit dfeec843fb237d73947e818f961e8d6f0df22b01)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm, kvm: fix double lock on cpu_add_remove_lock

Commit 8146875de7d4 (arm, kvm: Fix CPU hotplug callback registration)
holds the lock before calling the two functions:

kvm_vgic_hyp_init()
kvm_timer_hyp_init()

and both the two functions are calling register_cpu_notifier()
to register cpu notifier, so cause double lock on cpu_add_remove_lock.

Considered that both two functions are only called inside
kvm_arch_init() with holding cpu_add_remove_lock, so simply use
__register_cpu_notifier() to fix the problem.

Fixes: 8146875de7d4 (arm, kvm: Fix CPU hotplug callback registration)
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
(cherry picked from commit 553f809e23f00976caea7a1ebdabaa58a6383e7d)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: Add boot time configuration of Intermediate Physical Address size

ARMv8 supports a range of physical address bit sizes. The PARange bits
from ID_AA64MMFR0_EL1 register are read during boot-time and the
intermediate physical address size bits are written in the translation
control registers (TCR_EL1 and VTCR_EL2).

There is no change in the VA bits and levels of translation.

Signed-off-by: Radha Mohan Chintakuntla <rchintakuntla@cavium.com>
Reviewed-by: Will Deacon <Will.deacon@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit 87366d8cf7b3f6dc34633938aa8766e5a390ce33)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: fix non-VGIC compilation

Add a stub for kvm_vgic_addr when compiling without
CONFIG_KVM_ARM_VGIC. The usefulness of this configurarion is extremely
doubtful, but let's fix it anyway (until we decide that we'll always
support a VGIC).

Reported-by: Michele Paolino <m.paolino@virtualopensystems.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 6cbde8253a8143ada18ec0d1711230747a7c1934)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: fix warning in mmu.c

Compiling with THP enabled leads to the following warning:

arch/arm/kvm/mmu.c: In function ‘unmap_range’:
arch/arm/kvm/mmu.c:177:39: warning: ‘pte’ may be used uninitialized in this function [-Wmaybe-uninitialized]
   if (kvm_pmd_huge(*pmd) || page_empty(pte)) {
                                        ^
Code inspection reveals that these two cases are mutually exclusive,
so GCC is a bit overzealous here. Silence it anyway by initializing
pte to NULL and testing it later on.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 56041bf920d2937b7cadcb30cb206f0372eee814)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: trap VM system registers until MMU and caches are ON

In order to be able to detect the point where the guest enables
its MMU and caches, trap all the VM related system registers.

Once we see the guest enabling both the MMU and the caches, we
can go back to a saner mode of operation, which is to leave these
registers in complete control of the guest.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 8034699a42d68043b495c7e0cfafccd920707ec8)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: add world-switch for AMAIR{0,1}

HCR.TVM traps (among other things) accesses to AMAIR0 and AMAIR1.
In order to minimise the amount of surprise a guest could generate by
trying to access these registers with caches off, add them to the
list of registers we switch/handle.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit af20814ee927ed888288d98917a766b4179c4fe0)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: introduce per-vcpu HYP Configuration Register

So far, KVM/ARM used a fixed HCR configuration per guest, except for
the VI/VF/VA bits to control the interrupt in absence of VGIC.

With the upcoming need to dynamically reconfigure trapping, it becomes
necessary to allow the HCR to be changed on a per-vcpu basis.

The fix here is to mimic what KVM/arm64 already does: a per vcpu HCR
field, initialized at setup time.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit ac30a11e8e92a03dbe236b285c5cbae0bf563141)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: fix ordering of 64bit coprocessor accesses

Commit 240e99cbd00a (ARM: KVM: Fix 64-bit coprocessor handling)
added an ordering dependency for the 64bit registers.

The order described is: CRn, CRm, Op1, Op2, 64bit-first.

Unfortunately, the implementation is: CRn, 64bit-first, CRm...

Move the 64bit test to be last in order to match the documentation.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit 547f781378a22b65c2ab468f235c23001b5924da)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: fix handling of trapped 64bit coprocessor accesses

Commit 240e99cbd00a (ARM: KVM: Fix 64-bit coprocessor handling)
changed the way we match the 64bit coprocessor access from
user space, but didn't update the trap handler for the same
set of registers.

The effect is that a trapped 64bit access is never matched, leading
to a fault being injected into the guest. This went unnoticed as we
didn't really trap any 64bit register so far.

Placing the CRm field of the access into the CRn field of the matching
structure fixes the problem. Also update the debug feature to emit the
expected string in case of failing match.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit 46c214dd595381c880794413facadfa07fba5c95)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: force cache clean on page fault when caches are off

In order for a guest with caches disabled to observe data written
contained in a given page, we need to make sure that page is
committed to memory, and not just hanging in the cache (as guest
accesses are completely bypassing the cache until it decides to
enable it).

For this purpose, hook into the coherent_cache_guest_page
function and flush the region if the guest SCTLR
register doesn't show the MMU and caches as being enabled.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
(cherry picked from commit 159793001d7d85af17855630c94f0a176848e16b)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arm64: KVM: flush VM pages before letting the guest enable caches

When the guest runs with caches disabled (like in an early boot
sequence, for example), all the writes are diectly going to RAM,
bypassing the caches altogether.

Once the MMU and caches are enabled, whatever sits in the cache
becomes suddenly visible, which isn't what the guest expects.

A way to avoid this potential disaster is to invalidate the cache
when the MMU is being turned on. For this, we hook into the SCTLR_EL1
trapping code, and scan the stage-2 page tables, invalidating the
pages/sections that have already been mapped in.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit 9d218a1fcf4c6b759d442ef702842fae92e1ea61)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

ARM: KVM: introduce kvm_p*d_addr_end

The use of p*d_addr_end with stage-2 translation is slightly dodgy,
as the IPA is 40bits, while all the p*d_addr_end helpers are
taking an unsigned long (arm64 is fine with that as unligned long
is 64bit).

The fix is to introduce 64bit clean versions of the same helpers,
and use them in the stage-2 page table code.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
(cherry picked from commit a3c8bd31af260a17d626514f636849ee1cd1f63e)
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>