It is possible that the guest hypervisor has selected a smaller VL than
the maximum for its nested guest. As such, ZCR_EL2 may be configured for
a different VL when exiting a nested guest.
Set ZCR_EL2 (via the EL1 alias) to the maximum VL for the VM before
saving SVE state as the SVE save area is dimensioned by the max VL.
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20240620164653.1130714-8-oliver.upton@linux.dev
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
KVM (and pKVM) do not support SME guests. Therefore KVM ensures
that the host's SME state is flushed and that SME controls for
enabling access to ZA storage and for streaming are disabled.
pKVM needs to protect against a buggy/malicious host. Ensure that
it wouldn't run a guest when protected mode is enabled should any
of the SME controls be enabled.
Signed-off-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20240603122852.3923848-10-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
When setting/clearing CPACR bits for EL0 and EL1, use the ELx
format of the bits, which covers both. This makes the code
clearer, and reduces the chances of accidentally missing a bit.
No functional change intended.
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20240603122852.3923848-9-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
Expand comment clarifying why the host value representing SVE
vector length being restored for ZCR_EL1 on guest exit isn't the
same as it was on guest entry.
Signed-off-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Mark Brown <broonie@kernel.org>
Acked-by: Oliver Upton <oliver.upton@linux.dev>
Link: https://lore.kernel.org/r/20240423150538.2103045-21-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
pKVM maintains its own state at EL2 for tracking the host fpsimd
state. Therefore, no need to map and share the host's view with
it.
Signed-off-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Mark Brown <broonie@kernel.org>
Acked-by: Oliver Upton <oliver.upton@linux.dev>
Link: https://lore.kernel.org/r/20240423150538.2103045-12-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
To avoid direct comparison against the fp_owner enum, add a new
function that performs the check, host_owns_fp_regs(), to
complement the existing guest_owns_fp_regs().
To check for fpsimd state ownership, use the helpers instead of
directly using the enums.
No functional change intended.
Suggested-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Mark Brown <broonie@kernel.org>
Acked-by: Oliver Upton <oliver.upton@linux.dev>
Link: https://lore.kernel.org/r/20240423150538.2103045-4-tabba@google.com
Signed-off-by: Marc Zyngier <maz@kernel.org>
In retrospect, it is fairly obvious that the FP state ownership
is only meaningful for a given CPU, and that locating this
information in the vcpu was just a mistake.
Move the ownership tracking into the host data structure, and
rename it from fp_state to fp_owner, which is a better description
(name suggested by Mark Brown).
Reviewed-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
As the name of the field indicates, host_fpsimd_state is strictly
a host piece of data, and we reset this pointer on each PID change.
So let's move it where it belongs, and set it at load-time. Although
this is slightly more often, it is a well defined life-cycle which
matches other pieces of data.
Reviewed-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
* Changes to FPU handling came in via the main s390 pull request
* Only deliver to the guest the SCLP events that userspace has
requested.
* More virtual vs physical address fixes (only a cleanup since
virtual and physical address spaces are currently the same).
* Fix selftests undefined behavior.
x86:
* Fix a restriction that the guest can't program a PMU event whose
encoding matches an architectural event that isn't included in the
guest CPUID. The enumeration of an architectural event only says
that if a CPU supports an architectural event, then the event can be
programmed *using the architectural encoding*. The enumeration does
NOT say anything about the encoding when the CPU doesn't report support
the event *in general*. It might support it, and it might support it
using the same encoding that made it into the architectural PMU spec.
* Fix a variety of bugs in KVM's emulation of RDPMC (more details on
individual commits) and add a selftest to verify KVM correctly emulates
RDMPC, counter availability, and a variety of other PMC-related
behaviors that depend on guest CPUID and therefore are easier to
validate with selftests than with custom guests (aka kvm-unit-tests).
* Zero out PMU state on AMD if the virtual PMU is disabled, it does not
cause any bug but it wastes time in various cases where KVM would check
if a PMC event needs to be synthesized.
* Optimize triggering of emulated events, with a nice ~10% performance
improvement in VM-Exit microbenchmarks when a vPMU is exposed to the
guest.
* Tighten the check for "PMI in guest" to reduce false positives if an NMI
arrives in the host while KVM is handling an IRQ VM-Exit.
* Fix a bug where KVM would report stale/bogus exit qualification information
when exiting to userspace with an internal error exit code.
* Add a VMX flag in /proc/cpuinfo to report 5-level EPT support.
* Rework TDP MMU root unload, free, and alloc to run with mmu_lock held for
read, e.g. to avoid serializing vCPUs when userspace deletes a memslot.
* Tear down TDP MMU page tables at 4KiB granularity (used to be 1GiB). KVM
doesn't support yielding in the middle of processing a zap, and 1GiB
granularity resulted in multi-millisecond lags that are quite impolite
for CONFIG_PREEMPT kernels.
* Allocate write-tracking metadata on-demand to avoid the memory overhead when
a kernel is built with i915 virtualization support but the workloads use
neither shadow paging nor i915 virtualization.
* Explicitly initialize a variety of on-stack variables in the emulator that
triggered KMSAN false positives.
* Fix the debugregs ABI for 32-bit KVM.
* Rework the "force immediate exit" code so that vendor code ultimately decides
how and when to force the exit, which allowed some optimization for both
Intel and AMD.
* Fix a long-standing bug where kvm_has_noapic_vcpu could be left elevated if
vCPU creation ultimately failed, causing extra unnecessary work.
* Cleanup the logic for checking if the currently loaded vCPU is in-kernel.
* Harden against underflowing the active mmu_notifier invalidation
count, so that "bad" invalidations (usually due to bugs elsehwere in the
kernel) are detected earlier and are less likely to hang the kernel.
x86 Xen emulation:
* Overlay pages can now be cached based on host virtual address,
instead of guest physical addresses. This removes the need to
reconfigure and invalidate the cache if the guest changes the
gpa but the underlying host virtual address remains the same.
* When possible, use a single host TSC value when computing the deadline for
Xen timers in order to improve the accuracy of the timer emulation.
* Inject pending upcall events when the vCPU software-enables its APIC to fix
a bug where an upcall can be lost (and to follow Xen's behavior).
* Fall back to the slow path instead of warning if "fast" IRQ delivery of Xen
events fails, e.g. if the guest has aliased xAPIC IDs.
RISC-V:
* Support exception and interrupt handling in selftests
* New self test for RISC-V architectural timer (Sstc extension)
* New extension support (Ztso, Zacas)
* Support userspace emulation of random number seed CSRs.
ARM:
* Infrastructure for building KVM's trap configuration based on the
architectural features (or lack thereof) advertised in the VM's ID
registers
* Support for mapping vfio-pci BARs as Normal-NC (vaguely similar to
x86's WC) at stage-2, improving the performance of interacting with
assigned devices that can tolerate it
* Conversion of KVM's representation of LPIs to an xarray, utilized to
address serialization some of the serialization on the LPI injection
path
* Support for _architectural_ VHE-only systems, advertised through the
absence of FEAT_E2H0 in the CPU's ID register
* Miscellaneous cleanups, fixes, and spelling corrections to KVM and
selftests
LoongArch:
* Set reserved bits as zero in CPUCFG.
* Start SW timer only when vcpu is blocking.
* Do not restart SW timer when it is expired.
* Remove unnecessary CSR register saving during enter guest.
* Misc cleanups and fixes as usual.
Generic:
* cleanup Kconfig by removing CONFIG_HAVE_KVM, which was basically always
true on all architectures except MIPS (where Kconfig determines the
available depending on CPU capabilities). It is replaced either by
an architecture-dependent symbol for MIPS, and IS_ENABLED(CONFIG_KVM)
everywhere else.
* Factor common "select" statements in common code instead of requiring
each architecture to specify it
* Remove thoroughly obsolete APIs from the uapi headers.
* Move architecture-dependent stuff to uapi/asm/kvm.h
* Always flush the async page fault workqueue when a work item is being
removed, especially during vCPU destruction, to ensure that there are no
workers running in KVM code when all references to KVM-the-module are gone,
i.e. to prevent a very unlikely use-after-free if kvm.ko is unloaded.
* Grab a reference to the VM's mm_struct in the async #PF worker itself instead
of gifting the worker a reference, so that there's no need to remember
to *conditionally* clean up after the worker.
Selftests:
* Reduce boilerplate especially when utilize selftest TAP infrastructure.
* Add basic smoke tests for SEV and SEV-ES, along with a pile of library
support for handling private/encrypted/protected memory.
* Fix benign bugs where tests neglect to close() guest_memfd files.
-----BEGIN PGP SIGNATURE-----
iQFIBAABCAAyFiEE8TM4V0tmI4mGbHaCv/vSX3jHroMFAmX0iP8UHHBib256aW5p
QHJlZGhhdC5jb20ACgkQv/vSX3jHroND7wf+JZoNvwZ+bmwWe/4jn/YwNoYi/C5z
eypn8M1gsWEccpCpqPBwznVm9T29rF4uOlcMvqLEkHfTpaL1EKUUjP1lXPz/ileP
6a2RdOGxAhyTiFC9fjy+wkkjtLbn1kZf6YsS0hjphP9+w0chNbdn0w81dFVnXryd
j7XYI8R/bFAthNsJOuZXSEjCfIHxvTTG74OrTf1B1FEBB+arPmrgUeJftMVhffQK
Sowgg8L/Ii/x6fgV5NZQVSIyVf1rp8z7c6UaHT4Fwb0+RAMW8p9pYv9Qp1YkKp8y
5j0V9UzOHP7FRaYimZ5BtwQoqiZXYylQ+VuU/Y2f4X85cvlLzSqxaEMAPA==
=mqOV
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull kvm updates from Paolo Bonzini:
"S390:
- Changes to FPU handling came in via the main s390 pull request
- Only deliver to the guest the SCLP events that userspace has
requested
- More virtual vs physical address fixes (only a cleanup since
virtual and physical address spaces are currently the same)
- Fix selftests undefined behavior
x86:
- Fix a restriction that the guest can't program a PMU event whose
encoding matches an architectural event that isn't included in the
guest CPUID. The enumeration of an architectural event only says
that if a CPU supports an architectural event, then the event can
be programmed *using the architectural encoding*. The enumeration
does NOT say anything about the encoding when the CPU doesn't
report support the event *in general*. It might support it, and it
might support it using the same encoding that made it into the
architectural PMU spec
- Fix a variety of bugs in KVM's emulation of RDPMC (more details on
individual commits) and add a selftest to verify KVM correctly
emulates RDMPC, counter availability, and a variety of other
PMC-related behaviors that depend on guest CPUID and therefore are
easier to validate with selftests than with custom guests (aka
kvm-unit-tests)
- Zero out PMU state on AMD if the virtual PMU is disabled, it does
not cause any bug but it wastes time in various cases where KVM
would check if a PMC event needs to be synthesized
- Optimize triggering of emulated events, with a nice ~10%
performance improvement in VM-Exit microbenchmarks when a vPMU is
exposed to the guest
- Tighten the check for "PMI in guest" to reduce false positives if
an NMI arrives in the host while KVM is handling an IRQ VM-Exit
- Fix a bug where KVM would report stale/bogus exit qualification
information when exiting to userspace with an internal error exit
code
- Add a VMX flag in /proc/cpuinfo to report 5-level EPT support
- Rework TDP MMU root unload, free, and alloc to run with mmu_lock
held for read, e.g. to avoid serializing vCPUs when userspace
deletes a memslot
- Tear down TDP MMU page tables at 4KiB granularity (used to be
1GiB). KVM doesn't support yielding in the middle of processing a
zap, and 1GiB granularity resulted in multi-millisecond lags that
are quite impolite for CONFIG_PREEMPT kernels
- Allocate write-tracking metadata on-demand to avoid the memory
overhead when a kernel is built with i915 virtualization support
but the workloads use neither shadow paging nor i915 virtualization
- Explicitly initialize a variety of on-stack variables in the
emulator that triggered KMSAN false positives
- Fix the debugregs ABI for 32-bit KVM
- Rework the "force immediate exit" code so that vendor code
ultimately decides how and when to force the exit, which allowed
some optimization for both Intel and AMD
- Fix a long-standing bug where kvm_has_noapic_vcpu could be left
elevated if vCPU creation ultimately failed, causing extra
unnecessary work
- Cleanup the logic for checking if the currently loaded vCPU is
in-kernel
- Harden against underflowing the active mmu_notifier invalidation
count, so that "bad" invalidations (usually due to bugs elsehwere
in the kernel) are detected earlier and are less likely to hang the
kernel
x86 Xen emulation:
- Overlay pages can now be cached based on host virtual address,
instead of guest physical addresses. This removes the need to
reconfigure and invalidate the cache if the guest changes the gpa
but the underlying host virtual address remains the same
- When possible, use a single host TSC value when computing the
deadline for Xen timers in order to improve the accuracy of the
timer emulation
- Inject pending upcall events when the vCPU software-enables its
APIC to fix a bug where an upcall can be lost (and to follow Xen's
behavior)
- Fall back to the slow path instead of warning if "fast" IRQ
delivery of Xen events fails, e.g. if the guest has aliased xAPIC
IDs
RISC-V:
- Support exception and interrupt handling in selftests
- New self test for RISC-V architectural timer (Sstc extension)
- New extension support (Ztso, Zacas)
- Support userspace emulation of random number seed CSRs
ARM:
- Infrastructure for building KVM's trap configuration based on the
architectural features (or lack thereof) advertised in the VM's ID
registers
- Support for mapping vfio-pci BARs as Normal-NC (vaguely similar to
x86's WC) at stage-2, improving the performance of interacting with
assigned devices that can tolerate it
- Conversion of KVM's representation of LPIs to an xarray, utilized
to address serialization some of the serialization on the LPI
injection path
- Support for _architectural_ VHE-only systems, advertised through
the absence of FEAT_E2H0 in the CPU's ID register
- Miscellaneous cleanups, fixes, and spelling corrections to KVM and
selftests
LoongArch:
- Set reserved bits as zero in CPUCFG
- Start SW timer only when vcpu is blocking
- Do not restart SW timer when it is expired
- Remove unnecessary CSR register saving during enter guest
- Misc cleanups and fixes as usual
Generic:
- Clean up Kconfig by removing CONFIG_HAVE_KVM, which was basically
always true on all architectures except MIPS (where Kconfig
determines the available depending on CPU capabilities). It is
replaced either by an architecture-dependent symbol for MIPS, and
IS_ENABLED(CONFIG_KVM) everywhere else
- Factor common "select" statements in common code instead of
requiring each architecture to specify it
- Remove thoroughly obsolete APIs from the uapi headers
- Move architecture-dependent stuff to uapi/asm/kvm.h
- Always flush the async page fault workqueue when a work item is
being removed, especially during vCPU destruction, to ensure that
there are no workers running in KVM code when all references to
KVM-the-module are gone, i.e. to prevent a very unlikely
use-after-free if kvm.ko is unloaded
- Grab a reference to the VM's mm_struct in the async #PF worker
itself instead of gifting the worker a reference, so that there's
no need to remember to *conditionally* clean up after the worker
Selftests:
- Reduce boilerplate especially when utilize selftest TAP
infrastructure
- Add basic smoke tests for SEV and SEV-ES, along with a pile of
library support for handling private/encrypted/protected memory
- Fix benign bugs where tests neglect to close() guest_memfd files"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (246 commits)
selftests: kvm: remove meaningless assignments in Makefiles
KVM: riscv: selftests: Add Zacas extension to get-reg-list test
RISC-V: KVM: Allow Zacas extension for Guest/VM
KVM: riscv: selftests: Add Ztso extension to get-reg-list test
RISC-V: KVM: Allow Ztso extension for Guest/VM
RISC-V: KVM: Forward SEED CSR access to user space
KVM: riscv: selftests: Add sstc timer test
KVM: riscv: selftests: Change vcpu_has_ext to a common function
KVM: riscv: selftests: Add guest helper to get vcpu id
KVM: riscv: selftests: Add exception handling support
LoongArch: KVM: Remove unnecessary CSR register saving during enter guest
LoongArch: KVM: Do not restart SW timer when it is expired
LoongArch: KVM: Start SW timer only when vcpu is blocking
LoongArch: KVM: Set reserved bits as zero in CPUCFG
KVM: selftests: Explicitly close guest_memfd files in some gmem tests
KVM: x86/xen: fix recursive deadlock in timer injection
KVM: pfncache: simplify locking and make more self-contained
KVM: x86/xen: remove WARN_ON_ONCE() with false positives in evtchn delivery
KVM: x86/xen: inject vCPU upcall vector when local APIC is enabled
KVM: x86/xen: improve accuracy of Xen timers
...
FEAT_FPMR defines a new EL0 accessible register FPMR use to configure the
FP8 related features added to the architecture at the same time. Detect
support for this register and context switch it for EL0 when present.
Due to the sharing of responsibility for saving floating point state
between the host kernel and KVM FP8 support is not yet implemented in KVM
and a stub similar to that used for SVCR is provided for FPMR in order to
avoid bisection issues. To make it easier to share host state with the
hypervisor we store FPMR as a hardened usercopy field in uw (along with
some padding).
Signed-off-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20240306-arm64-2023-dpisa-v5-3-c568edc8ed7f@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Just like we repainted the early arm64 code, we need to update
the CPTR_EL2 accesses that are taking place in the nVHE code
when hVHE is used, making them look as if they were CPACR_EL1
accesses. Just like the VHE code.
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20230609162200.2024064-14-maz@kernel.org
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Normally when running a guest we do not touch the floating point
register state until first use of floating point by the guest, saving
the current state and loading the guest state at that point. This has
been found to offer a performance benefit in common cases. However
currently if SME is active when switching to a guest then we exit
streaming mode, disable ZA and invalidate the floating point register
state prior to starting the guest.
The exit from streaming mode is required for correct guest operation, if
we leave streaming mode enabled then many non-SME operations can
generate SME traps (eg, SVE operations will become streaming SVE
operations). If EL1 leaves CPACR_EL1.SMEN disabled then the host is
unable to intercept these traps. This will mean that a SME unaware guest
will see SME exceptions which will confuse it. Disabling streaming mode
also avoids creating spurious indications of usage of the SME hardware
which could impact system performance, especially with shared SME
implementations. Document the requirement to exit streaming mode
clearly.
There is no issue with guest operation caused by PSTATE.ZA so we can
defer handling for that until first floating point usage, do so if the
register state is not that of the current task and hence has already
been saved. We could also do this for the case where the register state
is that for the current task however this is very unlikely to happen and
would require disproportionate effort so continue to save the state in
that case.
Saving this state on first use would require that we map and unmap
storage for the host version of these registers for use by the
hypervisor, taking care to deal with protected KVM and the fact that the
host can free or reallocate the backing storage. Given that the strong
recommendation is that applications should only keep PSTATE.ZA enabled
when the state it enables is in active use it is difficult to see a case
where a VMM would wish to do this, it would need to not only be using
SME but also running the guest in the middle of SME usage. This can be
revisited in the future if a use case does arises, in the interim such
tasks will work but experience a performance overhead.
This brings our handling of SME more into line with our handling of
other floating point state and documents more clearly the constraints we
have, especially around streaming mode.
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221214-kvm-arm64-sme-context-switch-v2-3-57ba0082e9ff@kernel.org
In kvm_arch_vcpu_load_fp() we unconditionally set the current FP state
to FP_STATE_HOST_OWNED, this will be overridden to FP_STATE_NONE if
TIF_FOREIGN_FPSTATE is set but the check is deferred until
kvm_arch_vcpu_ctxflush_fp() where we are no longer preemptable. Add a
comment to this effect to help avoid people being concerned about the
lack of a check and discover where the check is done.
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221214-kvm-arm64-sme-context-switch-v2-1-57ba0082e9ff@kernel.org
* kvm-arm64/misc:
: Miscellaneous updates
:
: - Convert CPACR_EL1_TTA to the new, generated system register
: definitions.
:
: - Serialize toggling CPACR_EL1.SMEN to avoid unexpected exceptions when
: accessing SVCR in the host.
:
: - Avoid quiescing the guest if a vCPU accesses its own redistributor's
: SGIs/PPIs, eliminating the need to IPI. Largely an optimization for
: nested virtualization, as the L1 accesses the affected registers
: rather often.
:
: - Conversion to kstrtobool()
:
: - Common definition of INVALID_GPA across architectures
:
: - Enable CONFIG_USERFAULTFD for CI runs of KVM selftests
KVM: arm64: Fix non-kerneldoc comments
KVM: selftests: Enable USERFAULTFD
KVM: selftests: Remove redundant setbuf()
arm64/sysreg: clean up some inconsistent indenting
KVM: MMU: Make the definition of 'INVALID_GPA' common
KVM: arm64: vgic-v3: Use kstrtobool() instead of strtobool()
KVM: arm64: vgic-v3: Limit IPI-ing when accessing GICR_{C,S}ACTIVER0
KVM: arm64: Synchronize SMEN on vcpu schedule out
KVM: arm64: Kill CPACR_EL1_TTA definition
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
In preparation for adding support for storage for ZT0 to the thread_struct
rename za_state to sme_state. Since ZT0 is accessible when PSTATE.ZA is
set just like ZA itself we will extend the allocation done for ZA to
cover it, avoiding the need to further expand task_struct for non-SME
tasks.
No functional changes.
Signed-off-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20221208-arm64-sme2-v4-1-f2fa0aef982f@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
If we have VHE and need to reenable SME for host in
kvm_arch_vcpu_put_fp, CPACR.SMEN is modified from 0 to 1. Trap
control for reading SVCR is modified from enable to disable.
Synchronization is needed before reading SVCR later in
fpsimd_save, or it may cause sync exception which can not be
handled by host.
Cc: Marc Zyngier <maz@kernel.org>
Cc: James Morse <james.morse@arm.com>
Cc: Alexandru Elisei <alexandru.elisei@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Oliver Upton <oliver.upton@linux.dev>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Nianyao Tang <tangnianyao@huawei.com>
Reviewed-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20221220105024.13484-1-dongbo4@huawei.com
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
For reasons that are unclear to this reader fpsimd_bind_state_to_cpu()
populates the struct fpsimd_last_state_struct that it uses to store the
active floating point state for KVM guests by passing an argument for
each member of the structure. As the richness of the architecture increases
this is resulting in a function with a rather large number of arguments
which isn't ideal.
Simplify the interface by using the struct directly as the single argument
for the function, renaming it as we lift the definition into the header.
This could be built on further to reduce the work we do adding storage for
new FP state in various places but for now it just simplifies this one
interface.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221115094640.112848-9-broonie@kernel.org
Signed-off-by: Will Deacon <will@kernel.org>
Now that we are explicitly telling the host FP code which register state
it needs to save we can remove the manipulation of TIF_SVE from the KVM
code, simplifying it and allowing us to optimise our handling of normal
tasks. Remove the manipulation of TIF_SVE from KVM and instead rely on
to_save to ensure we save the correct data for it.
There should be no functional or performance impact from this change.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221115094640.112848-5-broonie@kernel.org
Signed-off-by: Will Deacon <will@kernel.org>
In order to avoid needlessly saving and restoring the guest registers KVM
relies on the host FPSMID code to save the guest registers when we context
switch away from the guest. This is done by binding the KVM guest state to
the CPU on top of the task state that was originally there, then carefully
managing the TIF_SVE flag for the task to cause the host to save the full
SVE state when needed regardless of the needs of the host task. This works
well enough but isn't terribly direct about what is going on and makes it
much more complicated to try to optimise what we're doing with the SVE
register state.
Let's instead have KVM pass in the register state it wants saving when it
binds to the CPU. We introduce a new FP_STATE_CURRENT for use
during normal task binding to indicate that we should base our
decisions on the current task. This should not be used when
actually saving. Ideally we might want to use a separate enum for
the type to save but this enum and the enum values would then
need to be named which has problems with clarity and ambiguity.
In order to ease any future debugging that might be required this patch
does not actually update any of the decision making about what to save,
it merely starts tracking the new information and warns if the requested
state is not what we would otherwise have decided to save.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221115094640.112848-4-broonie@kernel.org
Signed-off-by: Will Deacon <will@kernel.org>
When we save the state for the floating point registers this can be done
in the form visible through either the FPSIMD V registers or the SVE Z and
P registers. At present we track which format is currently used based on
TIF_SVE and the SME streaming mode state but particularly in the SVE case
this limits our options for optimising things, especially around syscalls.
Introduce a new enum which we place together with saved floating point
state in both thread_struct and the KVM guest state which explicitly
states which format is active and keep it up to date when we change it.
At present we do not use this state except to verify that it has the
expected value when loading the state, future patches will introduce
functional changes.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221115094640.112848-3-broonie@kernel.org
Signed-off-by: Will Deacon <will@kernel.org>
Since 8383741ab2 (KVM: arm64: Get rid of host SVE tracking/saving)
KVM has not tracked the host SVE state, relying on the fact that we
currently disable SVE whenever we perform a syscall. This may not be true
in future since performance optimisation may result in us keeping SVE
enabled in order to avoid needing to take access traps to reenable it.
Handle this by clearing TIF_SVE and converting the stored task state to
FPSIMD format when preparing to run the guest. This is done with a new
call fpsimd_kvm_prepare() to keep the direct state manipulation
functions internal to fpsimd.c.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20221115094640.112848-2-broonie@kernel.org
Signed-off-by: Will Deacon <will@kernel.org>
We currently start by assuming that the host owns the FP unit
at load time, then check again whether this is the case as
we are about to run. Only at this point do we account for the
fact that there is a (vanishingly small) chance that we're running
on a system without a FPSIMD unit (yes, this is madness).
We can actually move this FPSIMD check as early as load-time,
and drop the check at run time.
No intended change in behaviour.
Suggested-by: Reiji Watanabe <reijiw@google.com>
Reviewed-by: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
The two HOST_{SVE,SME}_ENABLED are only used for the host kernel
to track its own state across a vcpu run so that it can be fully
restored.
Move these flags to the so called state set.
Reviewed-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
The KVM FP code uses a pair of flags to denote three states:
- FP_ENABLED set: the guest owns the FP state
- FP_HOST set: the host owns the FP state
- FP_ENABLED and FP_HOST clear: nobody owns the FP state at all
and both flags set is an illegal state, which nothing ever checks
for...
As it turns out, this isn't really a good match for flags, and
we'd be better off if this was a simpler tristate, each state
having a name that actually reflect the state:
- FP_STATE_FREE
- FP_STATE_HOST_OWNED
- FP_STATE_GUEST_OWNED
Kill the two flags, and move over to an enum encoding these
three states. This results in less confusing code, and less risk of
ending up in the uncharted territory of a 4th state if we forget
to clear one of the two flags.
Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Reiji Watanabe <reijiw@google.com>
The vcpu KVM_ARM64_FP_FOREIGN_FPSTATE flag tracks the thread's own
TIF_FOREIGN_FPSTATE so that we can evaluate just before running
the vcpu whether it the FP regs contain something that is owned
by the vcpu or not by updating the rest of the FP flags.
We do this in the hypervisor code in order to make sure we're
in a context where we are not interruptible. But we already
have a hook in the run loop to generate this flag. We may as
well update the FP flags directly and save the pointless flag
tracking.
Whilst we're at it, rename update_fp_enabled() to guest_owns_fp_regs()
to indicate what the leftover of this helper actually do.
Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Reiji Watanabe <reijiw@google.com>
Reviewed-by: Mark Brown <broonie@kernel.org>
On each vcpu load, we set the KVM_ARM64_HOST_SME_ENABLED
flag if SME is enabled for EL0 on the host. This is used to
restore the correct state on vpcu put.
However, it appears that nothing ever clears this flag. Once
set, it will stick until the vcpu is destroyed, which has the
potential to spuriously enable SME for userspace. As it turns
out, this is due to the SME code being more or less copied from
SVE, and inheriting the same shortcomings.
We never saw the issue because nothing uses SME, and the amount
of testing is probably still pretty low.
Fixes: 861262ab86 ("KVM: arm64: Handle SME host state when running guests")
Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviwed-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20220528113829.1043361-3-maz@kernel.org
On each vcpu load, we set the KVM_ARM64_HOST_SVE_ENABLED
flag if SVE is enabled for EL0 on the host. This is used to restore
the correct state on vpcu put.
However, it appears that nothing ever clears this flag. Once
set, it will stick until the vcpu is destroyed, which has the
potential to spuriously enable SVE for userspace.
We probably never saw the issue because no VMM uses SVE, but
that's still pretty bad. Unconditionally clearing the flag
on vcpu load addresses the issue.
Fixes: 8383741ab2 ("KVM: arm64: Get rid of host SVE tracking/saving")
Signed-off-by: Marc Zyngier <maz@kernel.org>
Cc: stable@vger.kernel.org
Reviewed-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20220528113829.1043361-2-maz@kernel.org
The defines for SVCR call it SVCR_EL0 however the architecture calls the
register SVCR with no _EL0 suffix. In preparation for generating the sysreg
definitions rename to match the architecture, no functional change.
Signed-off-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20220510161208.631259-6-broonie@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
While we don't currently support SME in guests we do currently support it
for the host system so we need to take care of SME's impact, including
the floating point register state, when running guests. Simiarly to SVE
we need to manage the traps in CPACR_RL1, what is new is the handling of
streaming mode and ZA.
Normally we defer any handling of the floating point register state until
the guest first uses it however if the system is in streaming mode FPSIMD
and SVE operations may generate SME traps which we would need to distinguish
from actual attempts by the guest to use SME. Rather than do this for the
time being if we are in streaming mode when entering the guest we force
the floating point state to be saved immediately and exit streaming mode,
meaning that the guest won't generate SME traps for supported operations.
We could handle ZA in the access trap similarly to the FPSIMD/SVE state
without the disruption caused by streaming mode but for simplicity
handle it the same way as streaming mode for now.
This will be revisited when we support SME for guests (hopefully before SME
hardware becomes available), for now it will only incur additional cost on
systems with SME and even there only if streaming mode or ZA are enabled.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20220419112247.711548-27-broonie@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Allocate space for storing ZA on first access to SME and use that to save
and restore ZA state when context switching. We do this by using the vector
form of the LDR and STR ZA instructions, these do not require streaming
mode and have implementation recommendations that they avoid contention
issues in shared SMCU implementations.
Since ZA is architecturally guaranteed to be zeroed when enabled we do not
need to explicitly zero ZA, either we will be restoring from a saved copy
or trapping on first use of SME so we know that ZA must be disabled.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20220419112247.711548-16-broonie@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
When in streaming mode we need to save and restore the streaming mode
SVE register state rather than the regular SVE register state. This uses
the streaming mode vector length and omits FFR but is otherwise identical,
if TIF_SVE is enabled when we are in streaming mode then streaming mode
takes precedence.
This does not handle use of streaming SVE state with KVM, ptrace or
signals. This will be updated in further patches.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20220419112247.711548-15-broonie@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
In SME the use of both streaming SVE mode and ZA are tracked through
PSTATE.SM and PSTATE.ZA, visible through the system register SVCR. In
order to context switch the floating point state for SME we need to
context switch the contents of this register as part of context
switching the floating point state.
Since changing the vector length exits streaming SVE mode and disables
ZA we also make sure we update SVCR appropriately when setting vector
length, and similarly ensure that new threads have streaming SVE mode
and ZA disabled.
Signed-off-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Link: https://lore.kernel.org/r/20220419112247.711548-14-broonie@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Add a little bit of information on where _ctxflush_fp() and _ctxsync_fp()
are called to help people unfamiliar with the code get up to speed.
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20220124155720.3943374-2-broonie@kernel.org
Make use of the newly introduced unshare hypercall during guest teardown
to unmap guest-related data structures from the hyp stage-1.
Signed-off-by: Quentin Perret <qperret@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20211215161232.1480836-15-qperret@google.com
The create_hyp_mappings() function can currently be called at any point
in time. However, its behaviour in protected mode changes widely
depending on when it is being called. Prior to KVM init, it is used to
create the temporary page-table used to bring-up the hypervisor, and
later on it is transparently turned into a 'share' hypercall when the
kernel has lost control over the hypervisor stage-1. In order to prepare
the ground for also unsharing pages with the hypervisor during guest
teardown, introduce a kvm_share_hyp() function to make it clear in which
places a share hypercall should be expected, as we will soon need a
matching unshare hypercall in all those places.
Signed-off-by: Quentin Perret <qperret@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20211215161232.1480836-7-qperret@google.com
* kvm-arm64/fpsimd-tracking:
: .
: Simplify the handling of both the FP/SIMD and SVE state by
: removing the need for mapping the thread at EL2, and by
: dropping the tracking of the host's SVE state which is
: always invalid by construction.
: .
arm64/fpsimd: Document the use of TIF_FOREIGN_FPSTATE by KVM
KVM: arm64: Stop mapping current thread_info at EL2
KVM: arm64: Introduce flag shadowing TIF_FOREIGN_FPSTATE
KVM: arm64: Remove unused __sve_save_state
KVM: arm64: Get rid of host SVE tracking/saving
KVM: arm64: Reorder vcpu flag definitions
Signed-off-by: Marc Zyngier <maz@kernel.org>
We currently map the SVE state to HYP on detection of a PID change.
Although this matches what we do for FPSIMD, this is pretty pointless
for SVE, as the buffer is per-vcpu and has nothing to do with the
thread that is being run.
Move the mapping of the SVE state to finalize-time, which is where
we allocate the state memory, and thus the most logical place to
do this.
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Now that we can track an equivalent of TIF_FOREIGN_FPSTATE, drop
the mapping of current's thread_info at EL2.
Signed-off-by: Marc Zyngier <maz@kernel.org>
We currently have to maintain a mapping the thread_info structure
at EL2 in order to be able to check the TIF_FOREIGN_FPSTATE flag.
In order to eventually get rid of this, start with a vcpu flag that
shadows the thread flag on each entry into the hypervisor.
Reviewed-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
The SVE host tracking in KVM is pretty involved. It relies on a
set of flags tracking the ownership of the SVE register, as well
as that of the EL0 access.
It is also pretty scary: __hyp_sve_save_host() computes
a thread_struct pointer and obtains a sve_state which gets directly
accessed without further ado, even on nVHE. How can this even work?
The answer to that is that it doesn't, and that this is mostly dead
code. Closer examination shows that on executing a syscall, userspace
loses its SVE state entirely. This is part of the ABI. Another
thing to notice is that although the kernel provides helpers such as
kernel_neon_begin()/end(), they only deal with the FP/NEON state,
and not SVE.
Given that you can only execute a guest as the result of a syscall,
and that the kernel cannot use SVE by itself, it becomes pretty
obvious that there is never any host SVE state to save, and that
this code is only there to increase confusion.
Get rid of the TIF_SVE tracking and host save infrastructure altogether.
Reviewed-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Implement the SVE save/restore for nVHE, following a similar
logic to that of the VHE implementation:
- the SVE state is switched on trap from EL1 to EL2
- no further changes to ZCR_EL2 occur as long as the guest isn't
preempted or exit to userspace
- ZCR_EL2 is reset to its default value on the first SVE access from
the host EL1, and ZCR_EL1 restored to the default guest value in
vcpu_put()
Acked-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Make sure the guest's ZCR_EL1 is saved before we save/flush the
state. This will be useful in later patches.
Acked-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
When running on nVHE, and that the vcpu supports SVE, map the
SVE state at EL2 so that KVM can access it.
Acked-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Switch to the unified EL1 accessors for ZCR_EL1, which will make
things easier for nVHE support.
Acked-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
struct kvm_regs is used by userspace to indicate which register gets
accessed by the {GET,SET}_ONE_REG API. But as we're about to refactor
the layout of the in-kernel register structures, we need the kernel to
move away from it.
Let's make kvm_regs userspace only, and let the kernel map it to its own
internal representation.
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Remove includes of asm/kvm_host.h from files that already include
linux/kvm_host.h to make it more obvious that there is no ordering issue
between the two headers. linux/kvm_host.h includes asm/kvm_host.h to
pick up architecture specific settings, and this will never change, i.e.
including asm/kvm_host.h after linux/kvm_host.h may seem problematic,
but in practice is simply redundant.
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The only external user of fpsimd_save() and fpsimd_flush_cpu_state() is
the KVM FPSIMD code.
A following patch will introduce a mechanism to acquire owernship of the
FPSIMD/SVE context for performing context management operations. Rather
than having to export the new helpers to get/put the context, we can just
introduce a new function to combine fpsimd_save() and
fpsimd_flush_cpu_state().
This has also the advantage to remove any external call of fpsimd_save()
and fpsimd_flush_cpu_state(), so they can be turned static.
Lastly, the new function can also be used in the PM notifier.
Reviewed-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Julien Grall <julien.grall@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
