According to the ARM ARM, registers CNT{P,V}_TVAL_EL0 have bits [63:32]
RES0 [1]. When reading the register, the value is truncated to the least
significant 32 bits [2], and on writes, TimerValue is treated as a signed
32-bit integer [1, 2].
When the guest behaves correctly and writes 32-bit values, treating TVAL
as an unsigned 64 bit register works as expected. However, things start
to break down when the guest writes larger values, because
(u64)0x1_ffff_ffff = 8589934591. but (s32)0x1_ffff_ffff = -1, and the
former will cause the timer interrupt to be asserted in the future, but
the latter will cause it to be asserted now. Let's treat TVAL as a
signed 32-bit register on writes, to match the behaviour described in
the architecture, and the behaviour experimentally exhibited by the
virtual timer on a non-vhe host.
[1] Arm DDI 0487E.a, section D13.8.18
[2] Arm DDI 0487E.a, section D11.2.4
Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
[maz: replaced the read-side mask with lower_32_bits]
Signed-off-by: Marc Zyngier <maz@kernel.org>
Fixes: 8fa7616248 ("KVM: arm/arm64: arch_timer: Fix CNTP_TVAL calculation")
Link: https://lore.kernel.org/r/20200127103652.2326-1-alexandru.elisei@arm.com
The timers are canceled from an preempt-notifier which is invoked with
disabled preemption which is not allowed on PREEMPT_RT.
The timer callback is short so in could be invoked in hard-IRQ context
on -RT.
Let the timer expire on hard-IRQ context even on -RT.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Julien Grall <julien.grall@arm.com>
Acked-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20191107095424.16647-1-bigeasy@linutronix.de
Currently, the {read,write}_sysreg_el*() accessors for accessing
particular ELs' sysregs in the presence of VHE rely on some local
hacks and define their system register encodings in a way that is
inconsistent with the core definitions in <asm/sysreg.h>.
As a result, it is necessary to add duplicate definitions for any
system register that already needs a definition in sysreg.h for
other reasons.
This is a bit of a maintenance headache, and the reasons for the
_el*() accessors working the way they do is a bit historical.
This patch gets rid of the shadow sysreg definitions in
<asm/kvm_hyp.h>, converts the _el*() accessors to use the core
__msr_s/__mrs_s interface, and converts all call sites to use the
standard sysreg #define names (i.e., upper case, with SYS_ prefix).
This patch will conflict heavily anyway, so the opportunity
to clean up some bad whitespace in the context of the changes is
taken.
The change exposes a few system registers that have no sysreg.h
definition, due to msr_s/mrs_s being used in place of msr/mrs:
additions are made in order to fill in the gaps.
Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christoffer Dall <christoffer.dall@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Link: https://www.spinics.net/lists/kvm-arm/msg31717.html
[Rebased to v4.21-rc1]
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
[Rebased to v5.2-rc5, changelog updates]
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
for nested state save/restore.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
iQEcBAABAgAGBQJdC7NHAAoJEL/70l94x66DHm0H/R8L80sWe1OJbHHK8caPpwm2
mPt6JNcG/ysbG/uoMuVsdRAjZsg9l8JZB9xfA2m/ZPQQThjSG/WX0rU+gWMMI3X8
8ZbN4BCFoiNpOzOkhmStwzMWnvovKvMfhFW0BAI3HLUfM9A+XyVvNM/JbLOvEMRk
WB2SxYRc38ZvIbi8eXgsoFrVyLFB2Fj/0jps4FbKnkjkl37PTDehYLWQ1pt9KsWS
2KdGoXm7/18ottqf0DPfLe0hiiiDuK3akKz7WQBMsAJHi4Fm5j39NuseeRdlablk
uE4vM/sVaLn4xwM9JfrsBl9TzZ2qHsOTRlMQG4iNWjEAuPKa45lt0Jo7OBs6DSY=
=Lzxe
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull kvm fixes from Paolo Bonzini:
"Fixes for ARM and x86, plus selftest patches and nicer structs for
nested state save/restore"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
KVM: nVMX: reorganize initial steps of vmx_set_nested_state
KVM: arm/arm64: Fix emulated ptimer irq injection
tests: kvm: Check for a kernel warning
kvm: tests: Sort tests in the Makefile alphabetically
KVM: x86/mmu: Allocate PAE root array when using SVM's 32-bit NPT
KVM: x86: Modify struct kvm_nested_state to have explicit fields for data
KVM: fix typo in documentation
KVM: nVMX: use correct clean fields when copying from eVMCS
KVM: arm/arm64: vgic: Fix kvm_device leak in vgic_its_destroy
KVM: arm64: Filter out invalid core register IDs in KVM_GET_REG_LIST
KVM: arm64: Implement vq_present() as a macro
The emulated ptimer needs to track the level changes, otherwise the
the interrupt will never get deasserted, resulting in the guest getting
stuck in an interrupt storm if it enables ptimer interrupts. This was
found with kvm-unit-tests; the ptimer tests hung as soon as interrupts
were enabled. Typical Linux guests don't have a problem as they prefer
using the virtual timer.
Fixes: bee038a674 ("KVM: arm/arm64: Rework the timer code to use a timer_map")
Signed-off-by: Andrew Jones <drjones@redhat.com>
[Simplified the patch to res we only care about emulated timers here]
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Based on 1 normalized pattern(s):
this program is free software you can redistribute it and or modify
it under the terms of the gnu general public license version 2 as
published by the free software foundation this program is
distributed in the hope that it will be useful but without any
warranty without even the implied warranty of merchantability or
fitness for a particular purpose see the gnu general public license
for more details you should have received a copy of the gnu general
public license along with this program if not write to the free
software foundation inc 59 temple place suite 330 boston ma 02111
1307 usa
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-only
has been chosen to replace the boilerplate/reference in 136 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Alexios Zavras <alexios.zavras@intel.com>
Reviewed-by: Allison Randal <allison@lohutok.net>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190530000436.384967451@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When a VCPU never runs before a guest exists, but we set timer registers
up via ioctls, the associated hrtimer might never get cancelled.
Since we moved vcpu_load/put into the arch-specific implementations and
only have load/put for KVM_RUN, we won't ever have a scheduled hrtimer
for emulating a timer when modifying the timer state via an ioctl from
user space. All we need to do is make sure that we pick up the right
state when we load the timer state next time userspace calls KVM_RUN
again.
We also do not need to worry about this interacting with the bg_timer,
because if we were in WFI from the guest, and somehow ended up in a
kvm_arm_timer_set_reg, it means that:
1. the VCPU thread has received a signal,
2. we have called vcpu_load when being scheduled in again,
3. we have called vcpu_put when we returned to userspace for it to issue
another ioctl
And therefore will not have a bg_timer programmed and the event is
treated as a spurious wakeup from WFI if userspace decides to run the
vcpu again even if there are not virtual interrupts.
This fixes stray virtual timer interrupts triggered by an expiring
hrtimer, which happens after a failed live migration, for instance.
Fixes: bee038a674 ("KVM: arm/arm64: Rework the timer code to use a timer_map")
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
Reported-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Recently the generic timer test of kvm-unit-tests failed to complete
(stalled) when a physical timer is being used. This issue is caused
by incorrect update of CNTP_CVAL when CNTP_TVAL is being accessed,
introduced by 'Commit 84135d3d18 ("KVM: arm/arm64: consolidate arch
timer trap handlers")'. According to Arm ARM, the read/write behavior
of accesses to the TVAL registers is expected to be:
* READ: TimerValue = (CompareValue – (Counter - Offset)
* WRITE: CompareValue = ((Counter - Offset) + Sign(TimerValue)
This patch fixes the TVAL read/write code path according to the
specification.
Fixes: 84135d3d18 ("KVM: arm/arm64: consolidate arch timer trap handlers")
Signed-off-by: Wei Huang <wei@redhat.com>
[maz: commit message tidy-up]
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
The 'timer' local variable became unused after commit bee038a674
("KVM: arm/arm64: Rework the timer code to use a timer_map").
Remove it to avoid [-Wunused-but-set-variable] warning.
Cc: Christoffer Dall <christoffer.dall@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Suzuki K Pouloze <suzuki.poulose@arm.com>
Reviewed-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Shaokun Zhang <zhangshaokun@hisilicon.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
When a guest gets scheduled, KVM performs a "load" operation,
which for the timer includes evaluating the virtual "active" state
of the interrupt, and replicating it on the physical side. This
ensures that the deactivation in the guest will also take place
in the physical GIC distributor.
If the interrupt is not yet active, we flag it as inactive on the
physical side. This means that on restoring the timer registers,
if the timer has expired, we'll immediately take an interrupt.
That's absolutely fine, as the interrupt will then be flagged as
active on the physical side. What this assumes though is that we'll
enter the guest right after having taken the interrupt, and that
the guest will quickly ACK the interrupt, making it active at on
the virtual side.
It turns out that quite often, this assumption doesn't really hold.
The guest may be preempted on the back on this interrupt, either
from kernel space or whilst running at EL1 when a host interrupt
fires. When this happens, we repeat the whole sequence on the
next load (interrupt marked as inactive, timer registers restored,
interrupt fires). And if it takes a really long time for a guest
to activate the interrupt (as it does with nested virt), we end-up
with many such events in quick succession, leading to the guest only
making very slow progress.
This can also be seen with the number of virtual timer interrupt on the
host being far greater than the same number in the guest.
An easy way to fix this is to evaluate the timer state when performing
the "load" operation, just like we do when the interrupt actually fires.
If the timer has a pending virtual interrupt at this stage, then we
can safely flag the physical interrupt as being active, which prevents
spurious exits.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
We are currently emulating two timers in two different ways. When we
add support for nested virtualization in the future, we are going to be
emulating either two timers in two diffferent ways, or four timers in a
single way.
We need a unified data structure to keep track of how we map virtual
state to physical state and we need to cleanup some of the timer code to
operate more independently on a struct arch_timer_context instead of
trying to consider the global state of the VCPU and recomputing all
state.
Co-written with Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
VHE systems don't have to emulate the physical timer, we can simply
assign the EL1 physical timer directly to the VM as the host always
uses the EL2 timers.
In order to minimize the amount of cruft, AArch32 gets definitions for
the physical timer too, but is should be generally unused on this
architecture.
Co-written with Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
Prepare for having 4 timer data structures (2 for now).
Move loaded to the cpu data structure and not the individual timer
structure, in preparation for assigning the EL1 phys timer as well.
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
At the moment we have separate system register emulation handlers for
each timer register. Actually they are quite similar, and we rely on
kvm_arm_timer_[gs]et_reg() for the actual emulation anyways, so let's
just merge all of those handlers into one function, which just marshalls
the arguments and then hands off to a set of common accessors.
This makes extending the emulation to include EL2 timers much easier.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
[Fixed 32-bit VM breakage and reduced to reworking existing code]
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
[Fixed 32bit host, general cleanup]
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Instead of calling into kvm_timer_[un]schedule from the main kvm
blocking path, test if the VCPU is on the wait queue from the load/put
path and perform the background timer setup/cancel in this path.
This has the distinct advantage that we no longer race between load/put
and schedule/unschedule and programming and canceling of the bg_timer
always happens when the timer state is not loaded.
Note that we must now remove the checks in kvm_timer_blocking that do
not schedule a background timer if one of the timers can fire, because
we no longer have a guarantee that kvm_vcpu_check_block() will be called
before kvm_timer_blocking.
Reported-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
kvm_timer_vcpu_terminate can only be called in two scenarios:
1. As part of cleanup during a failed VCPU create
2. As part of freeing the whole VM (struct kvm refcount == 0)
In the first case, we cannot have programmed any timers or mapped any
IRQs, and therefore we do not have to cancel anything or unmap anything.
In the second case, the VCPU will have gone through kvm_timer_vcpu_put,
which will have canceled the emulated physical timer's hrtimer, and we
do not need to that here as well. We also do not care if the irq is
recorded as mapped or not in the VGIC data structure, because the whole
VM is going away. That leaves us only with having to ensure that we
cancel the bg_timer if we were blocking the last time we called
kvm_timer_vcpu_put().
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
The use of a work queue in the hrtimer expire function for the bg_timer
is a leftover from the time when we would inject interrupts when the
bg_timer expired.
Since we are no longer doing that, we can instead call
kvm_vcpu_wake_up() directly from the hrtimer function and remove all
workqueue functionality from the arch timer code.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
When the VCPU is blocked (for example from WFI) we don't inject the
physical timer interrupt if it should fire while the CPU is blocked, but
instead we just wake up the VCPU and expect kvm_timer_vcpu_load to take
care of injecting the interrupt.
Unfortunately, kvm_timer_vcpu_load() doesn't actually do that, it only
has support to schedule a soft timer if the emulated phys timer is
expected to fire in the future.
Follow the same pattern as kvm_timer_update_state() and update the irq
state after potentially scheduling a soft timer.
Reported-by: Andre Przywara <andre.przywara@arm.com>
Cc: Stable <stable@vger.kernel.org> # 4.15+
Fixes: bbdd52cfcb ("KVM: arm/arm64: Avoid phys timer emulation in vcpu entry/exit")
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
kvm_timer_update_state() is called when changing the phys timer
configuration registers, either via vcpu reset, as a result of a trap
from the guest, or when userspace programs the registers.
phys_timer_emulate() is in turn called by kvm_timer_update_state() to
either cancel an existing software timer, or program a new software
timer, to emulate the behavior of a real phys timer, based on the change
in configuration registers.
Unfortunately, the interaction between these two functions left a small
race; if the conceptual emulated phys timer should actually fire, but
the soft timer hasn't executed its callback yet, we cancel the timer in
phys_timer_emulate without injecting an irq. This only happens if the
check in kvm_timer_update_state is called before the timer should fire,
which is relatively unlikely, but possible.
The solution is to update the state of the phys timer after calling
phys_timer_emulate, which will pick up the pending timer state and
update the interrupt value.
Note that this leaves the opportunity of raising the interrupt twice,
once in the just-programmed soft timer, and once in
kvm_timer_update_state. Since this always happens synchronously with
the VCPU execution, there is no harm in this, and the guest ever only
sees a single timer interrupt.
Cc: Stable <stable@vger.kernel.org> # 4.15+
Signed-off-by: Christoffer Dall <christoffer.dall@arm.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-----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=iyYE
-----END PGP SIGNATURE-----
Merge tag 'kvm-arm-fixes-for-v4.16-2' into HEAD
Resolve conflicts with current mainline
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>
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>
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>
When I introduced a static key to avoid work in the critical path for
userspace irqchips which is very rarely used, I accidentally messed up
my logic and used && where I should have used ||, because the point was
to short-circuit the evaluation in case userspace irqchips weren't even
in use.
This fixes an issue when running in-kernel irqchip VMs alongside
userspace irqchip VMs.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Fixes: c44c232ee2d3 ("KVM: arm/arm64: Avoid work when userspace iqchips are not used")
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
After the recently introduced support for level-triggered mapped
interrupt, I accidentally left the VCPU thread busily going back and
forward between the guest and the hypervisor whenever the guest was
blocking, because I would always incorrectly report that a timer
interrupt was pending.
This is because the timer->irq.level field is not valid for mapped
interrupts, where we offload the level state to the hardware, and as a
result this field is always true.
Luckily the problem can be relatively easily solved by not checking the
cached signal state of either timer in kvm_timer_should_fire() but
instead compute the timer state on the fly, which we do already if the
cached signal state wasn't high. In fact, the only reason for checking
the cached signal state was a tiny optimization which would only be
potentially faster when the polling loop detects a pending timer
interrupt, which is quite unlikely.
Instead of duplicating the logic from kvm_arch_timer_handler(), we
enlighten kvm_timer_should_fire() to report something valid when the
timer state is loaded onto the hardware. We can then call this from
kvm_arch_timer_handler() as well and avoid the call to
__timer_snapshot_state() in kvm_arch_timer_get_input_level().
Reported-by: Tomasz Nowicki <tn@semihalf.com>
Tested-by: Tomasz Nowicki <tn@semihalf.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
We currently check if the VM has a userspace irqchip in several places
along the critical path, and if so, we do some work which is only
required for having an irqchip in userspace. This is unfortunate, as we
could avoid doing any work entirely, if we didn't have to support
irqchip in userspace.
Realizing the userspace irqchip on ARM is mostly a developer or hobby
feature, and is unlikely to be used in servers or other scenarios where
performance is a priority, we can use a refcounted static key to only
check the irqchip configuration when we have at least one VM that uses
an irqchip in userspace.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The VGIC can now support the life-cycle of mapped level-triggered
interrupts, and we no longer have to read back the timer state on every
exit from the VM if we had an asserted timer interrupt signal, because
the VGIC already knows if we hit the unlikely case where the guest
disables the timer without ACKing the virtual timer interrupt.
This means we rework a bit of the code to factor out the functionality
to snapshot the timer state from vtimer_save_state(), and we can reuse
this functionality in the sync path when we have an irqchip in
userspace, and also to support our implementation of the
get_input_level() function for the timer.
This change also means that we can no longer rely on the timer's view of
the interrupt line to set the active state, because we no longer
maintain this state for mapped interrupts when exiting from the guest.
Instead, we only set the active state if the virtual interrupt is
active, and otherwise we simply let the timer fire again and raise the
virtual interrupt from the ISR.
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The GIC sometimes need to sample the physical line of a mapped
interrupt. As we know this to be notoriously slow, provide a callback
function for devices (such as the timer) which can do this much faster
than talking to the distributor, for example by comparing a few
in-memory values. Fall back to the good old method of poking the
physical GIC if no callback is provided.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The timer logic was designed after a strict idea of modeling an
interrupt line level in software, meaning that only transitions in the
level need to be reported to the VGIC. This works well for the timer,
because the arch timer code is in complete control of the device and can
track the transitions of the line.
However, as we are about to support using the HW bit in the VGIC not
just for the timer, but also for VFIO which cannot track transitions of
the interrupt line, we have to decide on an interface between the GIC
and other subsystems for level triggered mapped interrupts, which both
the timer and VFIO can use.
VFIO only sees an asserting transition of the physical interrupt line,
and tells the VGIC when that happens. That means that part of the
interrupt flow is offloaded to the hardware.
To use the same interface for VFIO devices and the timer, we therefore
have to change the timer (we cannot change VFIO because it doesn't know
the details of the device it is assigning to a VM).
Luckily, changing the timer is simple, we just need to stop 'caching'
the line level, but instead let the VGIC know the state of the timer
every time there is a potential change in the line level, and when the
line level should be asserted from the timer ISR. The VGIC can ignore
extra notifications using its validate mechanism.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
When enabling the timer on the first run, we fail to ever restore the
state and mark it as loaded. That means, that in the initial entry to
the VCPU ioctl, unless we exit to userspace for some reason such as a
pending signal, if the guest programs a timer and blocks, we will wait
forever, because we never read back the hardware state (the loaded flag
is not set), and so we think the timer is disabled, and we never
schedule a background soft timer.
The end result? The VCPU blocks forever, and the only solution is to
kill the thread.
Fixes: 4a2c4da125 ("arm/arm64: KVM: Load the timer state when enabling the timer")
Reported-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
The recent timer rework was assuming that once the timer was disabled,
we should no longer see any interrupts from the timer. This assumption
turns out to not be true, and instead we have to handle the case when
the timer ISR runs even after the timer has been disabled.
This requires a couple of changes:
First, we should never overwrite the cached guest state of the timer
control register when the ISR runs, because KVM may have disabled its
timers when doing vcpu_put(), even though the guest still had the timer
enabled.
Second, we shouldn't assume that the timer is actually firing just
because we see an interrupt, but we should check the actual state of the
timer in the timer control register to understand if the hardware timer
is really firing or not.
We also add an ISB to vtimer_save_state() to ensure the timer is
actually disabled once we enable interrupts, which should clarify the
intention of the implementation, and reduce the risk of unwanted
interrupts.
Fixes: b103cc3f10 ("KVM: arm/arm64: Avoid timer save/restore in vcpu entry/exit")
Reported-by: Marc Zyngier <marc.zyngier@arm.com>
Reported-by: Jia He <hejianet@gmail.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
If we don't have a usable GIC, do not try to set the vcpu affinity
as this is guaranteed to fail.
Reported-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Andre Przywara <andre.przywara@arm.com>
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>
The timer optimization patches inadvertendly changed the logic to always
load the timer state as if we have a vgic, even if we don't have a vgic.
Fix this by doing the usual irqchip_in_kernel() check and call the
appropriate load function.
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
After the timer optimization rework we accidentally end up calling
physical timer enable/disable functions on VHE systems, which is neither
needed nor correct, since the CNTHCTL_EL2 register format is
different when HCR_EL2.E2H is set.
The CNTHCTL_EL2 is initialized when CPUs become online in
kvm_timer_init_vhe() and we don't have to call these functions on VHE
systems, which also allows us to inline the non-VHE functionality.
Reported-by: Jintack Lim <jintack@cs.columbia.edu>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
We want to reuse the core of the map/unmap functions for IRQ
forwarding. Let's move the computation of the hwirq in
kvm_vgic_map_phys_irq and pass the linux IRQ as parameter.
the host_irq is added to struct vgic_irq.
We introduce kvm_vgic_map/unmap_irq which take a struct vgic_irq
handle as a parameter.
Acked-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
After being lazy with saving/restoring the timer state, we defer that
work to vcpu_load and vcpu_put, which ensure that the timer state is
loaded on the hardware timers whenever the VCPU runs.
Unfortunately, we are failing to do that the first time vcpu_load()
runs, because the timer has not yet been enabled at that time. As long
as the initialized timer state matches what happens to be in the
hardware (a disabled timer, because we never leave the timer screaming),
this does not show up as a problem, but is nevertheless incorrect.
The solution is simple; disable preemption while setting the timer to be
enabled, and call the timer load function when first enabling the timer.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
kvm_timer_should_fire() can be called in two different situations from
the kvm_vcpu_block().
The first case is before calling kvm_timer_schedule(), used for wait
polling, and in this case the VCPU thread is running and the timer state
is loaded onto the hardware so all we have to do is check if the virtual
interrupt lines are asserted, becasue the timer interrupt handler
functions will raise those lines as appropriate.
The second case is inside the wait loop of kvm_vcpu_block(), where we
have already called kvm_timer_schedule() and therefore the hardware will
be disabled and the software view of the timer state is up to date
(timer->loaded is false), and so we can simply check if the timer should
fire by looking at the software state.
Signed-off-by: Christoffer Dall <cdall@linaro.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Now when both the vtimer and the ptimer when using both the in-kernel
vgic emulation and a userspace IRQ chip are driven by the timer signals
and at the vcpu load/put boundaries, instead of recomputing the timer
state at every entry/exit to/from the guest, we can get entirely rid of
the flush hwstate function.
Signed-off-by: Christoffer Dall <cdall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
There is no need to schedule and cancel a hrtimer when entering and
exiting the guest, because we know when the physical timer is going to
fire when the guest programs it, and we can simply program the hrtimer
at that point.
Now when the register modifications from the guest go through the
kvm_arm_timer_set/get_reg functions, which always call
kvm_timer_update_state(), we can simply consider the timer state in this
function and schedule and cancel the timers as needed.
This avoids looking at the physical timer emulation state when entering
and exiting the VCPU, allowing for faster servicing of the VM when
needed.
Signed-off-by: Christoffer Dall <cdall@linaro.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
We are about to call phys_timer_emulate() from kvm_timer_update_state()
and modify phys_timer_emulate() at the same time. Moving the function
and modifying it in a single patch makes the diff hard to read, so do
this separately first.
No functional change.
Signed-off-by: Christoffer Dall <cdall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Add suport for the physical timer registers in kvm_arm_timer_set_reg and
kvm_arm_timer_get_reg so that these functions can be reused to interact
with the rest of the system.
Note that this paves part of the way for the physical timer state
save/restore, but we still need to add those registers to
KVM_GET_REG_LIST before we support migrating the physical timer state.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
We don't need to save and restore the hardware timer state and examine
if it generates interrupts on on every entry/exit to the guest. The
timer hardware is perfectly capable of telling us when it has expired
by signaling interrupts.
When taking a vtimer interrupt in the host, we don't want to mess with
the timer configuration, we just want to forward the physical interrupt
to the guest as a virtual interrupt. We can use the split priority drop
and deactivate feature of the GIC to do this, which leaves an EOI'ed
interrupt active on the physical distributor, making sure we don't keep
taking timer interrupts which would prevent the guest from running. We
can then forward the physical interrupt to the VM using the HW bit in
the LR of the GIC, like we do already, which lets the guest directly
deactivate both the physical and virtual timer simultaneously, allowing
the timer hardware to exit the VM and generate a new physical interrupt
when the timer output is again asserted later on.
We do need to capture this state when migrating VCPUs between physical
CPUs, however, which we use the vcpu put/load functions for, which are
called through preempt notifiers whenever the thread is scheduled away
from the CPU or called directly if we return from the ioctl to
userspace.
One caveat is that we have to save and restore the timer state in both
kvm_timer_vcpu_[put/load] and kvm_timer_[schedule/unschedule], because
we can have the following flows:
1. kvm_vcpu_block
2. kvm_timer_schedule
3. schedule
4. kvm_timer_vcpu_put (preempt notifier)
5. schedule (vcpu thread gets scheduled back)
6. kvm_timer_vcpu_load (preempt notifier)
7. kvm_timer_unschedule
And a version where we don't actually call schedule:
1. kvm_vcpu_block
2. kvm_timer_schedule
7. kvm_timer_unschedule
Since kvm_timer_[schedule/unschedule] may not be followed by put/load,
but put/load also may be called independently, we call the timer
save/restore functions from both paths. Since they rely on the loaded
flag to never save/restore when unnecessary, this doesn't cause any
harm, and we ensure that all invokations of either set of functions work
as intended.
An added benefit beyond not having to read and write the timer sysregs
on every entry and exit is that we no longer have to actively write the
active state to the physical distributor, because we configured the
irq for the vtimer to only get a priority drop when handling the
interrupt in the GIC driver (we called irq_set_vcpu_affinity()), and
the interrupt stays active after firing on the host.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
As we are about to take physical interrupts for the virtual timer on the
host but want to leave those active while running the VM (and let the VM
deactivate them), we need to set the vtimer PPI affinity accordingly.
Signed-off-by: Christoffer Dall <cdall@linaro.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
As we are about to be lazy with saving and restoring the timer
registers, we prepare by moving all possible timer configuration logic
out of the hyp code. All virtual timer registers can be programmed from
EL1 and since the arch timer is always a level triggered interrupt we
can safely do this with interrupts disabled in the host kernel on the
way to the guest without taking vtimer interrupts in the host kernel
(yet).
The downside is that the cntvoff register can only be programmed from
hyp mode, so we jump into hyp mode and back to program it. This is also
safe, because the host kernel doesn't use the virtual timer in the KVM
code. It may add a little performance performance penalty, but only
until following commits where we move this operation to vcpu load/put.
Signed-off-by: Christoffer Dall <cdall@linaro.org>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
We were using the same hrtimer for emulating the physical timer and for
making sure a blocking VCPU thread would be eventually woken up. That
worked fine in the previous arch timer design, but as we are about to
actually use the soft timer expire function for the physical timer
emulation, change the logic to use a dedicated hrtimer.
This has the added benefit of not having to cancel any work in the sync
path, which in turn allows us to run the flush and sync with IRQs
disabled.
Note that the hrtimer used to program the host kernel's timer to
generate an exit from the guest when the emulated physical timer fires
never has to inject any work, and to share the soft_timer_cancel()
function with the bg_timer, we change the function to only cancel any
pending work if the pointer to the work struct is not null.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
As we are about to introduce a separate hrtimer for the physical timer,
call this timer bg_timer, because we refer to this timer as the
background timer in the code and comments elsewhere.
Signed-off-by: Christoffer Dall <cdall@linaro.org>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
We are about to add an additional soft timer to the arch timer state for
a VCPU and would like to be able to reuse the functions to program and
cancel a timer, so we make them slightly more generic and rename to make
it more clear that these functions work on soft timers and not the
hardware resource that this code is managing.
The armed flag on the timer state is only used to assert a condition,
and we don't rely on this assertion in any meaningful way, so we can
simply get rid of this flack and slightly reduce complexity.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@linaro.org>
Alexandru Elisei