kvm_read_guest() will eventually look up in kvm_memslots(), which requires
either to hold the kvm->slots_lock or to be inside a kvm->srcu critical
section.
In contrast to x86 and s390 we don't take the SRCU lock on every guest
exit, so we have to do it individually for each kvm_read_guest() call.
Use the newly introduced wrapper for that.
Cc: Stable <stable@vger.kernel.org> # 4.12+
Reported-by: Jan Glauber <jan.glauber@caviumnetworks.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
kvm_read_guest() will eventually look up in kvm_memslots(), which requires
either to hold the kvm->slots_lock or to be inside a kvm->srcu critical
section.
In contrast to x86 and s390 we don't take the SRCU lock on every guest
exit, so we have to do it individually for each kvm_read_guest() call.
Provide a wrapper which does that and use that everywhere.
Note that ending the SRCU critical section before returning from the
kvm_read_guest() wrapper is safe, because the data has been *copied*, so
we don't need to rely on valid references to the memslot anymore.
Cc: Stable <stable@vger.kernel.org> # 4.8+
Reported-by: Jan Glauber <jan.glauber@caviumnetworks.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Apparently the development of update_affinity() overlapped with the
promotion of irq_lock to be _irqsave, so the patch didn't convert this
lock over. This will make lockdep complain.
Fix this by disabling IRQs around the lock.
Cc: stable@vger.kernel.org
Fixes: 08c9fd0421 ("KVM: arm/arm64: vITS: Add a helper to update the affinity of an LPI")
Reported-by: Jan Glauber <jan.glauber@caviumnetworks.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
As Jan reported [1], lockdep complains about the VGIC not being bullet
proof. This seems to be due to two issues:
- When commit 006df0f349 ("KVM: arm/arm64: Support calling
vgic_update_irq_pending from irq context") promoted irq_lock and
ap_list_lock to _irqsave, we forgot two instances of irq_lock.
lockdeps seems to pick those up.
- If a lock is _irqsave, any other locks we take inside them should be
_irqsafe as well. So the lpi_list_lock needs to be promoted also.
This fixes both issues by simply making the remaining instances of those
locks _irqsave.
One irq_lock is addressed in a separate patch, to simplify backporting.
[1] http://lists.infradead.org/pipermail/linux-arm-kernel/2018-May/575718.html
Cc: stable@vger.kernel.org
Fixes: 006df0f349 ("KVM: arm/arm64: Support calling vgic_update_irq_pending from irq context")
Reported-by: Jan Glauber <jan.glauber@caviumnetworks.com>
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
- Fix proxying of GICv2 CPU interface accesses
- Fix crash when switching to BE
- Track source vcpu git GICv2 SGIs
- Fix an outdated bit of documentation
-----BEGIN PGP SIGNATURE-----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=cO6X
-----END PGP SIGNATURE-----
Merge tag 'kvmarm-fixes-for-4.17-2' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmarm
KVM/arm fixes for 4.17, take #2
- Fix proxying of GICv2 CPU interface accesses
- Fix crash when switching to BE
- Track source vcpu git GICv2 SGIs
- Fix an outdated bit of documentation
One comment still mentioned process_maintenance operations after
commit af0614991a ("KVM: arm/arm64: vgic: Get rid of unnecessary
process_maintenance operation")
Update the comment to point to vgic_fold_lr_state instead, which
is where maintenance interrupts are taken care of.
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Valentin Schneider <valentin.schneider@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
ARM:
- PSCI selection API, a leftover from 4.16 (for stable)
- Kick vcpu on active interrupt affinity change
- Plug a VMID allocation race on oversubscribed systems
- Silence debug messages
- Update Christoffer's email address (linaro -> arm)
x86:
- Expose userspace-relevant bits of a newly added feature
- Fix TLB flushing on VMX with VPID, but without EPT
-----BEGIN PGP SIGNATURE-----
iQEcBAABCAAGBQJa44lQAAoJEED/6hsPKofo1dIH/3n9AZSWvavgL2V3j6agT8Yy
hxF4nHCFEJd5aqDNwbG9QEzivKw88r3o3mdB2XAQESB2MlCYR1jkTONm7yvVJTs/
/P9gj+DEQbCj2AgT//u3BGsAsZDKFhB9JwfmV2Mp4zDIqWFa6oCOGeq/iPVAGDcN
vUpuYeIicuH9SRoxH7de3z+BEXW0O+gCABXQtvA93FKTMz35yFTgmbDVCnvaV0zL
3B+3/4/jdbTRICW8EX6Li43+gEBUMtnVNkdqxLPTuCtDG8iuPUGfgF02gH99/9gj
hliV3Q4VUZKkSABW5AqKPe4+9rbsHCh9eL0LpHFGI9y+6LeUIOXAX4CtohR8gWE=
=W9Vz
-----END PGP SIGNATURE-----
rMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull KVM fixes from Radim Krčmář:
"ARM:
- PSCI selection API, a leftover from 4.16 (for stable)
- Kick vcpu on active interrupt affinity change
- Plug a VMID allocation race on oversubscribed systems
- Silence debug messages
- Update Christoffer's email address (linaro -> arm)
x86:
- Expose userspace-relevant bits of a newly added feature
- Fix TLB flushing on VMX with VPID, but without EPT"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
x86/headers/UAPI: Move DISABLE_EXITS KVM capability bits to the UAPI
kvm: apic: Flush TLB after APIC mode/address change if VPIDs are in use
arm/arm64: KVM: Add PSCI version selection API
KVM: arm/arm64: vgic: Kick new VCPU on interrupt migration
arm64: KVM: Demote SVE and LORegion warnings to debug only
MAINTAINERS: Update e-mail address for Christoffer Dall
KVM: arm/arm64: Close VMID generation race
Now that we make sure we don't inject multiple instances of the
same GICv2 SGI at the same time, we've made another bug more
obvious:
If we exit with an active SGI, we completely lose track of which
vcpu it came from. On the next entry, we restore it with 0 as a
source, and if that wasn't the right one, too bad. While this
doesn't seem to trouble GIC-400, the architectural model gets
offended and doesn't deactivate the interrupt on EOI.
Another connected issue is that we will happilly make pending
an interrupt from another vcpu, overriding the above zero with
something that is just as inconsistent. Don't do that.
The final issue is that we signal a maintenance interrupt when
no pending interrupts are present in the LR. Assuming we've fixed
the two issues above, we end-up in a situation where we keep
exiting as soon as we've reached the active state, and not be
able to inject the following pending.
The fix comes in 3 parts:
- GICv2 SGIs have their source vcpu saved if they are active on
exit, and restored on entry
- Multi-SGIs cannot go via the Pending+Active state, as this would
corrupt the source field
- Multi-SGIs are converted to using MI on EOI instead of NPIE
Fixes: 16ca6a607d ("KVM: arm/arm64: vgic: Don't populate multiple LRs with the same vintid")
Reported-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
It's possible for userspace to control n. Sanitize n when using it as an
array index.
Note that while it appears that n must be bound to the interval [0,3]
due to the way it is extracted from addr, we cannot guarantee that
compiler transformations (and/or future refactoring) will ensure this is
the case, and given this is a slow path it's better to always perform
the masking.
Found by smatch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
Signed-off-by: Will Deacon <will.deacon@arm.com>
It's possible for userspace to control intid. Sanitize intid when using
it as an array index.
At the same time, sort the includes when adding <linux/nospec.h>.
Found by smatch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Cc: kvmarm@lists.cs.columbia.edu
Signed-off-by: Will Deacon <will.deacon@arm.com>
Although we've implemented PSCI 0.1, 0.2 and 1.0, we expose either 0.1
or 1.0 to a guest, defaulting to the latest version of the PSCI
implementation that is compatible with the requested version. This is
no different from doing a firmware upgrade on KVM.
But in order to give a chance to hypothetical badly implemented guests
that would have a fit by discovering something other than PSCI 0.2,
let's provide a new API that allows userspace to pick one particular
version of the API.
This is implemented as a new class of "firmware" registers, where
we expose the PSCI version. This allows the PSCI version to be
save/restored as part of a guest migration, and also set to
any supported version if the guest requires it.
Cc: stable@vger.kernel.org #4.16
Reviewed-by: Christoffer Dall <cdall@kernel.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
When vgic_prune_ap_list() finds an interrupt that needs to be migrated
to a new VCPU, we should notify this VCPU of the pending interrupt,
since it requires immediate action.
Kick this VCPU once we have added the new IRQ to the list, but only
after dropping the locks.
Reported-by: Stefano Stabellini <sstabellini@kernel.org>
Reviewed-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Before entering the guest, we check whether our VMID is still
part of the current generation. In order to avoid taking a lock,
we start with checking that the generation is still current, and
only if not current do we take the lock, recheck, and update the
generation and VMID.
This leaves open a small race: A vcpu can bump up the global
generation number as well as the VM's, but has not updated
the VMID itself yet.
At that point another vcpu from the same VM comes in, checks
the generation (and finds it not needing anything), and jumps
into the guest. At this point, we end-up with two vcpus belonging
to the same VM running with two different VMIDs. Eventually, the
VMID used by the second vcpu will get reassigned, and things will
really go wrong...
A simple solution would be to drop this initial check, and always take
the lock. This is likely to cause performance issues. A middle ground
is to convert the spinlock to a rwlock, and only take the read lock
on the fast path. If the check fails at that point, drop it and
acquire the write lock, rechecking the condition.
This ensures that the above scenario doesn't occur.
Cc: stable@vger.kernel.org
Reported-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Shannon Zhao <zhaoshenglong@huawei.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
vgic_copy_lpi_list() parses the LPI list and picks LPIs targeting
a given vcpu. We allocate the array containing the intids before taking
the lpi_list_lock, which means we can have an array size that is not
equal to the number of LPIs.
This is particularly obvious when looking at the path coming from
vgic_enable_lpis, which is not a command, and thus can run in parallel
with commands:
vcpu 0: vcpu 1:
vgic_enable_lpis
its_sync_lpi_pending_table
vgic_copy_lpi_list
intids = kmalloc_array(irq_count)
MAPI(lpi targeting vcpu 0)
list_for_each_entry(lpi_list_head)
intids[i++] = irq->intid;
At that stage, we will happily overrun the intids array. Boo. An easy
fix is is to break once the array is full. The MAPI command will update
the config anyway, and we won't miss a thing. We also make sure that
lpi_list_count is read exactly once, so that further updates of that
value will not affect the array bound check.
Cc: stable@vger.kernel.org
Fixes: ccb1d791ab ("KVM: arm64: vgic-its: Fix pending table sync")
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
It was recently reported that VFIO mediated devices, and anything
that VFIO exposes as level interrupts, do no strictly follow the
expected logic of such interrupts as it only lowers the input
line when the guest has EOId the interrupt at the GIC level, rather
than when it Acked the interrupt at the device level.
THe GIC's Active+Pending state is fundamentally incompatible with
this behaviour, as it prevents KVM from observing the EOI, and in
turn results in VFIO never dropping the line. This results in an
interrupt storm in the guest, which it really never expected.
As we cannot really change VFIO to follow the strict rules of level
signalling, let's forbid the A+P state altogether, as it is in the
end only an optimization. It ensures that we will transition via
an invalid state, which we can use to notify VFIO of the EOI.
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Shunyong Yang <shunyong.yang@hxt-semitech.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
- Peace of mind locking fix in vgic_mmio_read_pending
- Allow hw-mapped interrupts to be reset when the VM resets
- Fix GICv2 multi-source SGI injection
- Fix MMIO synchronization for GICv2 on v3 emulation
- Remove excess verbosity on the console
-----BEGIN PGP SIGNATURE-----
iQJJBAABCAAzFiEEn9UcU+C1Yxj9lZw9I9DQutE9ekMFAlqqp/cVHG1hcmMuenlu
Z2llckBhcm0uY29tAAoJECPQ0LrRPXpDAGkP/2LMhFN561PKlqgu5V4hFvowJiXb
Gbb/qi095vtDGccbKmJKAZp3jyOM2oJEMUkx5RBYglWjW0mxb3zPAAxhldXiqv/2
CrOGGlS/FwfyIjCt7870pltDOIgRmk8Fv/MyQjjGKF6VAghd6yVHIZiOUjiriUyz
6hNyc2znLm0tBqm4j3HTXKHpD23YseW387pQoeQ03/WiXiZ60O3e3k0yppXO81qE
b7TGT4Bz04mxlAISZVZeTmG7P7P4ej6+NhOH+1kxacseLzHdECPBA0JRcwRpfLkP
5JFodUOX7/KHpvpMLUxRNRnLBei9WUL4o2LAEV0qDaj7nlAud0kKUm22RLaVKDm+
8FSUQ12XKqnZsRrl6IizU1oAb1I1iV3j9HF5iNf3mk9AO27REGk0b8fDyRzDj300
xpySgvIgA+f+EyY+3ve0AmEUa5QKz/WLuik2ZCqpVOuufrO8XpS+zjn1L1tzTlkR
95EahDA7enutw47G0uWtxoPMeU4HTZS/CAiFwUbq8BEK7T3Rct7UySPLwgeYBoji
MUlCRhPyAANCJmtO6rpOS3htkQ3XkkO1DVIGLuWC5Zl00W1T5I5+VRrVL1YI4v3O
d2ui9r5X5Vmg4OUdhr2D9fXgPWWKEbqD90jv40rGLsMl0g/IwrC+o2VxgYxSeu5x
CLUYILwEA5NDZSof
=iyYE
-----END PGP SIGNATURE-----
Merge tag 'kvm-arm-fixes-for-v4.16-2' into HEAD
Resolve conflicts with current mainline
Until now, all EL2 executable mappings were derived from their
EL1 VA. Since we want to decouple the vectors mapping from
the rest of the hypervisor, we need to be able to map some
text somewhere else.
The "idmap" region (for lack of a better name) is ideally suited
for this, as we have a huge range that hardly has anything in it.
Let's extend the IO allocator to also deal with executable mappings,
thus providing the required feature.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
The main idea behind randomising the EL2 VA is that we usually have
a few spare bits between the most significant bit of the VA mask
and the most significant bit of the linear mapping.
Those bits could be a bunch of zeroes, and could be useful
to move things around a bit. Of course, the more memory you have,
the less randomisation you get...
Alternatively, these bits could be the result of KASLR, in which
case they are already random. But it would be nice to have a
*different* randomization, just to make the job of a potential
attacker a bit more difficult.
Inserting these random bits is a bit involved. We don't have a spare
register (short of rewriting all the kern_hyp_va call sites), and
the immediate we want to insert is too random to be used with the
ORR instruction. The best option I could come up with is the following
sequence:
and x0, x0, #va_mask
ror x0, x0, #first_random_bit
add x0, x0, #(random & 0xfff)
add x0, x0, #(random >> 12), lsl #12
ror x0, x0, #(63 - first_random_bit)
making it a fairly long sequence, but one that a decent CPU should
be able to execute without breaking a sweat. It is of course NOPed
out on VHE. The last 4 instructions can also be turned into NOPs
if it appears that there is no free bits to use.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
We so far mapped our HYP IO (which is essentially the GICv2 control
registers) using the same method as for memory. It recently appeared
that is a bit unsafe:
We compute the HYP VA using the kern_hyp_va helper, but that helper
is only designed to deal with kernel VAs coming from the linear map,
and not from the vmalloc region... This could in turn cause some bad
aliasing between the two, amplified by the upcoming VA randomisation.
A solution is to come up with our very own basic VA allocator for
MMIO. Since half of the HYP address space only contains a single
page (the idmap), we have plenty to borrow from. Let's use the idmap
as a base, and allocate downwards from it. GICv2 now lives on the
other side of the great VA barrier.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Unmapping the idmap range using 52bit PA is quite broken, as we
don't take into account the right number of PGD entries, and rely
on PTRS_PER_PGD. The result is that pgd_index() truncates the
address, and we end-up in the weed.
Let's introduce a new unmap_hyp_idmap_range() that knows about this,
together with a kvm_pgd_index() helper, which hides a bit of the
complexity of the issue.
Fixes: 98732d1b18 ("KVM: arm/arm64: fix HYP ID map extension to 52 bits")
Reported-by: James Morse <james.morse@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Although the idmap section of KVM can only be at most 4kB and
must be aligned on a 4kB boundary, the rest of the code expects
it to be page aligned. Things get messy when tearing down the
HYP page tables when PAGE_SIZE is 64K, and the idmap section isn't
64K aligned.
Let's fix this by computing aligned boundaries that the HYP code
will use.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reported-by: James Morse <james.morse@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
As we're about to change the way we map devices at HYP, we need
to move away from kern_hyp_va on an IO address.
One way of achieving this is to store the VAs in kvm_vgic_global_state,
and use that directly from the HYP code. This requires a small change
to create_hyp_io_mappings so that it can also return a HYP VA.
We take this opportunity to nuke the vctrl_base field in the emulated
distributor, as it is not used anymore.
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>
Both HYP io mappings call ioremap, followed by create_hyp_io_mappings.
Let's move the ioremap call into create_hyp_io_mappings itself, which
simplifies the code a bit and allows for further refactoring.
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Displaying the HYP VA information is slightly counterproductive when
using VA randomization. Turn it into a debug feature only, and adjust
the last displayed value to reflect the top of RAM instead of ~0.
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
We can finally get completely rid of any calls to the VGICv3
save/restore functions when the AP lists are empty on VHE systems. This
requires carefully factoring out trap configuration from saving and
restoring state, and carefully choosing what to do on the VHE and
non-VHE path.
One of the challenges is that we cannot save/restore the VMCR lazily
because we can only write the VMCR when ICC_SRE_EL1.SRE is cleared when
emulating a GICv2-on-GICv3, since otherwise all Group-0 interrupts end
up being delivered as FIQ.
To solve this problem, and still provide fast performance in the fast
path of exiting a VM when no interrupts are pending (which also
optimized the latency for actually delivering virtual interrupts coming
from physical interrupts), we orchestrate a dance of only doing the
activate/deactivate traps in vgic load/put for VHE systems (which can
have ICC_SRE_EL1.SRE cleared when running in the host), and doing the
configuration on every round-trip on non-VHE systems.
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
The APRs can only have bits set when the guest acknowledges an interrupt
in the LR and can only have a bit cleared when the guest EOIs an
interrupt in the LR. Therefore, if we have no LRs with any
pending/active interrupts, the APR cannot change value and there is no
need to clear it on every exit from the VM (hint: it will have already
been cleared when we exited the guest the last time with the LRs all
EOIed).
The only case we need to take care of is when we migrate the VCPU away
from a CPU or migrate a new VCPU onto a CPU, or when we return to
userspace to capture the state of the VCPU for migration. To make sure
this works, factor out the APR save/restore functionality into separate
functions called from the VCPU (and by extension VGIC) put/load hooks.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Just like we can program the GICv2 hypervisor control interface directly
from the core vgic code, we can do the same for the GICv3 hypervisor
control interface on VHE systems.
We do this by simply calling the save/restore functions when we have VHE
and we can then get rid of the save/restore function calls from the VHE
world switch function.
One caveat is that we now write GICv3 system register state before the
potential early exit path in the run loop, and because we sync back
state in the early exit path, we have to ensure that we read a
consistent GIC state from the sync path, even though we have never
actually run the guest with the newly written GIC state. We solve this
by inserting an ISB in the early exit path.
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
The vgic-v2-sr.c file now only contains the logic to replay unaligned
accesses to the virtual CPU interface on 16K and 64K page systems, which
is only relevant on 64-bit platforms. Therefore move this file to the
arm64 KVM tree, remove the compile directive from the 32-bit side
makefile, and remove the ifdef in the C file.
Since this file also no longer saves/restores anything, rename the file
to vgic-v2-cpuif-proxy.c to more accurately describe the logic in this
file.
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
We can program the GICv2 hypervisor control interface logic directly
from the core vgic code and can instead do the save/restore directly
from the flush/sync functions, which can lead to a number of future
optimizations.
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
There is really no need to store the vgic_elrsr on the VGIC data
structures as the only need we have for the elrsr is to figure out if an
LR is inactive when we save the VGIC state upon returning from the
guest. We can might as well store this in a temporary local variable.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
SPSR_EL1 is not used by a VHE host kernel and can be deferred, but we
need to rework the accesses to this register to access the latest value
depending on whether or not guest system registers are loaded on the CPU
or only reside in memory.
The handling of accessing the various banked SPSRs for 32-bit VMs is a
bit clunky, but this will be improved in following patches which will
first prepare and subsequently implement deferred save/restore of the
32-bit registers, including the 32-bit SPSRs.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Currently we access the system registers array via the vcpu_sys_reg()
macro. However, we are about to change the behavior to some times
modify the register file directly, so let's change this to two
primitives:
* Accessor macros vcpu_write_sys_reg() and vcpu_read_sys_reg()
* Direct array access macro __vcpu_sys_reg()
The accessor macros should be used in places where the code needs to
access the currently loaded VCPU's state as observed by the guest. For
example, when trapping on cache related registers, a write to a system
register should go directly to the VCPU version of the register.
The direct array access macro can be used in places where the VCPU is
known to never be running (for example userspace access) or for
registers which are never context switched (for example all the PMU
system registers).
This rewrites all users of vcpu_sys_regs to one of the macros described
above.
No functional change.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Christoffer Dall <cdall@cs.columbia.edu>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
The VHE switch function calls __timer_enable_traps and
__timer_disable_traps which don't do anything on VHE systems.
Therefore, simply remove these calls from the VHE switch function and
make the functions non-conditional as they are now only called from the
non-VHE switch path.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
So far this is mostly (see below) a copy of the legacy non-VHE switch
function, but we will start reworking these functions in separate
directions to work on VHE and non-VHE in the most optimal way in later
patches.
The only difference after this patch between the VHE and non-VHE run
functions is that we omit the branch-predictor variant-2 hardening for
QC Falkor CPUs, because this workaround is specific to a series of
non-VHE ARMv8.0 CPUs.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
As we are about to move a bunch of save/restore logic for VHE kernels to
the load and put functions, we need some infrastructure to do this.
Reviewed-by: Andrew Jones <drjones@redhat.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
We currently have a separate read-modify-write of the HCR_EL2 on entry
to the guest for the sole purpose of setting the VF and VI bits, if set.
Since this is most rarely the case (only when using userspace IRQ chip
and interrupts are in flight), let's get rid of this operation and
instead modify the bits in the vcpu->arch.hcr[_el2] directly when
needed.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Moving the call to vcpu_load() in kvm_arch_vcpu_ioctl_run() to after
we've called kvm_vcpu_first_run_init() simplifies some of the vgic and
there is also no need to do vcpu_load() for things such as handling the
immediate_exit flag.
Reviewed-by: Julien Grall <julien.grall@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Calling vcpu_load() registers preempt notifiers for this vcpu and calls
kvm_arch_vcpu_load(). The latter will soon be doing a lot of heavy
lifting on arm/arm64 and will try to do things such as enabling the
virtual timer and setting us up to handle interrupts from the timer
hardware.
Loading state onto hardware registers and enabling hardware to signal
interrupts can be problematic when we're not actually about to run the
VCPU, because it makes it difficult to establish the right context when
handling interrupts from the timer, and it makes the register access
code difficult to reason about.
Luckily, now when we call vcpu_load in each ioctl implementation, we can
simply remove the call from the non-KVM_RUN vcpu ioctls, and our
kvm_arch_vcpu_load() is only used for loading vcpu content to the
physical CPU when we're actually going to run the vcpu.
Reviewed-by: Julien Grall <julien.grall@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
On guest exit, and when using GICv2 on GICv3, we use a dsb(st) to
force synchronization between the memory-mapped guest view and
the system-register view that the hypervisor uses.
This is incorrect, as the spec calls out the need for "a DSB whose
required access type is both loads and stores with any Shareability
attribute", while we're only synchronizing stores.
We also lack an isb after the dsb to ensure that the latter has
actually been executed before we start reading stuff from the sysregs.
The fix is pretty easy: turn dsb(st) into dsb(sy), and slap an isb()
just after.
Cc: stable@vger.kernel.org
Fixes: f68d2b1b73 ("arm64: KVM: Implement vgic-v3 save/restore")
Acked-by: Christoffer Dall <cdall@kernel.org>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
The vgic code is trying to be clever when injecting GICv2 SGIs,
and will happily populate LRs with the same interrupt number if
they come from multiple vcpus (after all, they are distinct
interrupt sources).
Unfortunately, this is against the letter of the architecture,
and the GICv2 architecture spec says "Each valid interrupt stored
in the List registers must have a unique VirtualID for that
virtual CPU interface.". GICv3 has similar (although slightly
ambiguous) restrictions.
This results in guests locking up when using GICv2-on-GICv3, for
example. The obvious fix is to stop trying so hard, and inject
a single vcpu per SGI per guest entry. After all, pending SGIs
with multiple source vcpus are pretty rare, and are mostly seen
in scenario where the physical CPUs are severely overcomitted.
But as we now only inject a single instance of a multi-source SGI per
vcpu entry, we may delay those interrupts for longer than strictly
necessary, and run the risk of injecting lower priority interrupts
in the meantime.
In order to address this, we adopt a three stage strategy:
- If we encounter a multi-source SGI in the AP list while computing
its depth, we force the list to be sorted
- When populating the LRs, we prevent the injection of any interrupt
of lower priority than that of the first multi-source SGI we've
injected.
- Finally, the injection of a multi-source SGI triggers the request
of a maintenance interrupt when there will be no pending interrupt
in the LRs (HCR_NPIE).
At the point where the last pending interrupt in the LRs switches
from Pending to Active, the maintenance interrupt will be delivered,
allowing us to add the remaining SGIs using the same process.
Cc: stable@vger.kernel.org
Fixes: 0919e84c0f ("KVM: arm/arm64: vgic-new: Add IRQ sync/flush framework")
Acked-by: Christoffer Dall <cdall@kernel.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
On my GICv3 system, the following is printed to the kernel log at boot:
kvm [1]: 8-bit VMID
kvm [1]: IDMAP page: d20e35000
kvm [1]: HYP VA range: 800000000000:ffffffffffff
kvm [1]: vgic-v2@2c020000
kvm [1]: GIC system register CPU interface enabled
kvm [1]: vgic interrupt IRQ1
kvm [1]: virtual timer IRQ4
kvm [1]: Hyp mode initialized successfully
The KVM IDMAP is a mapping of a statically allocated kernel structure,
and so printing its physical address leaks the physical placement of
the kernel when physical KASLR in effect. So change the kvm_info() to
kvm_debug() to remove it from the log output.
While at it, trim the output a bit more: IRQ numbers can be found in
/proc/interrupts, and the HYP VA and vgic-v2 lines are not highly
informational either.
Cc: <stable@vger.kernel.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Acked-by: Christoffer Dall <cdall@kernel.org>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
We currently don't allow resetting mapped IRQs from userspace, because
their state is controlled by the hardware. But we do need to reset the
state when the VM is reset, so we provide a function for the 'owner' of
the mapped interrupt to reset the interrupt state.
Currently only the timer uses mapped interrupts, so we call this
function from the timer reset logic.
Cc: stable@vger.kernel.org
Fixes: 4c60e360d6 ("KVM: arm/arm64: Provide a get_input_level for the arch timer")
Signed-off-by: Christoffer Dall <cdall@kernel.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Calling vcpu_load() registers preempt notifiers for this vcpu and calls
kvm_arch_vcpu_load(). The latter will soon be doing a lot of heavy
lifting on arm/arm64 and will try to do things such as enabling the
virtual timer and setting us up to handle interrupts from the timer
hardware.
Loading state onto hardware registers and enabling hardware to signal
interrupts can be problematic when we're not actually about to run the
VCPU, because it makes it difficult to establish the right context when
handling interrupts from the timer, and it makes the register access
code difficult to reason about.
Luckily, now when we call vcpu_load in each ioctl implementation, we can
simply remove the call from the non-KVM_RUN vcpu ioctls, and our
kvm_arch_vcpu_load() is only used for loading vcpu content to the
physical CPU when we're actually going to run the vcpu.
Cc: stable@vger.kernel.org
Fixes: 9b062471e5 ("KVM: Move vcpu_load to arch-specific kvm_arch_vcpu_ioctl")
Reviewed-by: Julien Grall <julien.grall@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Our irq_is_pending() helper function accesses multiple members of the
vgic_irq struct, so we need to hold the lock when calling it.
Add that requirement as a comment to the definition and take the lock
around the call in vgic_mmio_read_pending(), where we were missing it
before.
Fixes: 96b298000d ("KVM: arm/arm64: vgic-new: Add PENDING registers handlers")
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
The loading time of a VM is quite significant with a CPU usage
reaching 100% when loading a VM that its virtio devices use a
large amount of virt-queues (e.g. a virtio-serial device with
max_ports=511). Most of the time is spend in re-sorting the
kvm_io_bus kvm_io_range array when a new eventfd is registered.
The patch replaces the existing method with an insert sort.
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Uri Lublin <ulublin@redhat.com>
Signed-off-by: Gal Hammer <ghammer@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
In AArch64/AArch32, the virtual counter uses a fixed virtual offset
of zero in the following situations as per ARMv8 specifications:
1) HCR_EL2.E2H is 1, and CNTVCT_EL0/CNTVCT are read from EL2.
2) HCR_EL2.{E2H, TGE} is {1, 1}, and either:
— CNTVCT_EL0 is read from Non-secure EL0 or EL2.
— CNTVCT is read from Non-secure EL0.
So, no need to zero CNTVOFF_EL2/CNTVOFF for VHE case.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Shanker Donthineni <shankerd@codeaurora.org>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
When introducing support for irqchip in userspace we needed a way to
mask the timer signal to prevent the guest continuously exiting due to a
screaming timer.
We did this by disabling the corresponding percpu interrupt on the
host interrupt controller, because we cannot rely on the host system
having a GIC, and therefore cannot make any assumptions about having an
active state to hide the timer signal.
Unfortunately, when introducing this feature, it became entirely
possible that a VCPU which belongs to a VM that has a userspace irqchip
can disable the vtimer irq on the host on some physical CPU, and then go
away without ever enabling the vtimer irq on that physical CPU again.
This means that using irqchips in userspace on a system that also
supports running VMs with an in-kernel GIC can prevent forward progress
from in-kernel GIC VMs.
Later on, when we started taking virtual timer interrupts in the arch
timer code, we would also leave this timer state active for userspace
irqchip VMs, because we leave it up to a VGIC-enabled guest to
deactivate the hardware IRQ using the HW bit in the LR.
Both issues are solved by only using the enable/disable trick on systems
that do not have a host GIC which supports the active state, because all
VMs on such systems must use irqchips in userspace. Systems that have a
working GIC with support for an active state use the active state to
mask the timer signal for both userspace and in-kernel irqchips.
Cc: Alexander Graf <agraf@suse.de>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Cc: <stable@vger.kernel.org> # v4.12+
Fixes: d9e1397783 ("KVM: arm/arm64: Support arch timers with a userspace gic")
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
This is the mindless scripted replacement of kernel use of POLL*
variables as described by Al, done by this script:
for V in IN OUT PRI ERR RDNORM RDBAND WRNORM WRBAND HUP RDHUP NVAL MSG; do
L=`git grep -l -w POLL$V | grep -v '^t' | grep -v /um/ | grep -v '^sa' | grep -v '/poll.h$'|grep -v '^D'`
for f in $L; do sed -i "-es/^\([^\"]*\)\(\<POLL$V\>\)/\\1E\\2/" $f; done
done
with de-mangling cleanups yet to come.
NOTE! On almost all architectures, the EPOLL* constants have the same
values as the POLL* constants do. But they keyword here is "almost".
For various bad reasons they aren't the same, and epoll() doesn't
actually work quite correctly in some cases due to this on Sparc et al.
The next patch from Al will sort out the final differences, and we
should be all done.
Scripted-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Andre Przywara
Radim Krčmář
Valentin Schneider
Linus Torvalds
Marc Zyngier
Mark Rutland
Christoffer Dall
Christoffer Dall
Ard Biesheuvel
Christoffer Dall
Gal Hammer
Shanker Donthineni
Wanpeng Li