Initialize the number of LVT entries during APIC creation, else the field
will be incorrectly left '0' if userspace never invokes KVM_X86_SETUP_MCE.
Add and use a helper to calculate the number of entries even though
MCG_CMCI_P is not set by default in vcpu->arch.mcg_cap. Relying on that
to always be true is unnecessarily risky, and subtle/confusing as well.
Fixes: 4b903561ec ("KVM: x86: Add Corrected Machine Check Interrupt (CMCI) emulation to lapic.")
Reported-by: Xiaoyao Li <xiaoyao.li@intel.com>
Cc: Jue Wang <juew@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Merge a bug fix and cleanups for {g,s}et_msr_mce() using a base that
predates commit 281b52780b ("KVM: x86: Add emulation for
MSR_IA32_MCx_CTL2 MSRs."), which was written with the intention that it
be applied _after_ the bug fix and cleanups. The bug fix in particular
needs to be sent to stable trees; give them a stable hash to use.
Add helpers to identify CTL (control) and STATUS MCi MSR types instead of
open coding the checks using the offset. Using the offset is perfectly
safe, but unintuitive, as understanding what the code does requires
knowing that the offset calcuation will not affect the lower three bits.
Opportunistically comment the STATUS logic to save readers a trip to
Intel's SDM or AMD's APM to understand the "data != 0" check.
No functional change intended.
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Link: https://lore.kernel.org/r/20220512222716.4112548-4-seanjc@google.com
Use an explicit case statement to grab the full range of MCx bank MSRs
in {g,s}et_msr_mce(), and manually check only the "end" (the number of
banks configured by userspace may be less than the max). The "default"
trick works, but is a bit odd now, and will be quite odd if/when support
for accessing MCx_CTL2 MSRs is added, which has near identical logic.
Hoist "offset" to function scope so as to avoid curly braces for the case
statement, and because MCx_CTL2 support will need the same variables.
Opportunstically clean up the comment about allowing bit 10 to be cleared
from bank 4.
No functional change intended.
Cc: Jue Wang <juew@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Link: https://lore.kernel.org/r/20220512222716.4112548-3-seanjc@google.com
Return '1', not '-1', when handling an illegal WRMSR to a MCi_CTL or
MCi_STATUS MSR. The behavior of "all zeros' or "all ones" for CTL MSRs
is architectural, as is the "only zeros" behavior for STATUS MSRs. I.e.
the intent is to inject a #GP, not exit to userspace due to an unhandled
emulation case. Returning '-1' gets interpreted as -EPERM up the stack
and effecitvely kills the guest.
Fixes: 890ca9aefa ("KVM: Add MCE support")
Fixes: 9ffd986c6e ("KVM: X86: #GP when guest attempts to write MCi_STATUS register w/o 0")
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Link: https://lore.kernel.org/r/20220512222716.4112548-2-seanjc@google.com
Buffer split_desc_cache, the cache used to allcoate rmap list entries,
only by the default cache capacity (currently 40), not by doubling the
minimum (513). Aliasing L2 GPAs to L1 GPAs is uncommon, thus eager page
splitting is unlikely to need 500+ entries. And because each object is a
non-trivial 128 bytes (see struct pte_list_desc), those extra ~500
entries means KVM is in all likelihood wasting ~64kb of memory per VM.
Link: https://lore.kernel.org/all/YrTDcrsn0%2F+alpzf@google.com
Cc: David Matlack <dmatlack@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20220624171808.2845941-4-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use an "unsigned int" for @access parameters instead of a "u32", mostly
to be consistent throughout KVM, but also because "u32" is misleading.
@access can actually squeeze into a u8, i.e. doesn't need 32 bits, but is
as an "unsigned int" because sp->role.access is an unsigned int.
No functional change intended.
Link: https://lore.kernel.org/all/YqyZxEfxXLsHGoZ%2F@google.com
Cc: David Matlack <dmatlack@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20220624171808.2845941-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
complete_emulator_pio_in() only has to be called by
complete_sev_es_emulated_ins() now; therefore, all that the function does
now is adjust sev_pio_count and sev_pio_data. Which is the same for
both IN and OUT.
No functional change intended.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Now all callers except emulator_pio_in_emulated are using
__emulator_pio_in/complete_emulator_pio_in explicitly.
Move the "either copy the result or attempt PIO" logic in
emulator_pio_in_emulated, and rename __emulator_pio_in to
just emulator_pio_in.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use __emulator_pio_in() directly for fast PIO instead of bouncing through
emulator_pio_in() now that __emulator_pio_in() fills "val" when handling
in-kernel PIO. vcpu->arch.pio.count is guaranteed to be '0', so this a
pure nop.
emulator_pio_in_emulated is now the last caller of emulator_pio_in.
No functional change intended.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Make emulator_pio_in_out operate directly on the provided buffer
as long as PIO is handled inside KVM.
For input operations, this means that, in the case of in-kernel
PIO, __emulator_pio_in() does not have to be always followed
by complete_emulator_pio_in(). This affects emulator_pio_in() and
kvm_sev_es_ins(); for the latter, that is why the call moves from
advance_sev_es_emulated_ins() to complete_sev_es_emulated_ins().
For output, it means that vcpu->pio.count is never set unnecessarily
and there is no need to clear it; but also vcpu->pio.size must not
be used in kvm_sev_es_outs(), because it will not be updated for
in-kernel OUT.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
For now, this is basically an excuse to add back the void* argument to
the function, while removing some knowledge of vcpu->arch.pio* from
its callers. The WARN that vcpu->arch.pio.count is zero is also
extended to OUT operations.
The vcpu->arch.pio* fields still need to be filled even when the PIO is
handled in-kernel as __emulator_pio_in() is always followed by
complete_emulator_pio_in(). But after fixing that, it will be possible to
to only populate the vcpu->arch.pio* fields on userspace exits.
No functional change intended.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
KVM protects the device list with SRCU, and therefore different calls
to kvm_io_bus_read()/kvm_io_bus_write() can very well see different
incarnations of kvm->buses. If userspace unregisters a device while
vCPUs are running there is no well-defined result. This patch applies
a safe fallback by returning early from emulator_pio_in_out(). This
corresponds to returning zeroes from IN, and dropping the writes on
the floor for OUT.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The caller of kernel_pio already has arguments for most of what kernel_pio
fishes out of vcpu->arch.pio. This is the first step towards ensuring that
vcpu->arch.pio.* is only used when exiting to userspace.
No functional change intended.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use complete_emulator_pio_in() directly when completing fast PIO, there's
no need to bounce through emulator_pio_in(): the comment about ECX
changing doesn't apply to fast PIO, which isn't used for string I/O.
No functional change intended.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
- Avoid toggling the x2apic msr interception if it is already up to date.
- Avoid touching L0 msr bitmap when AVIC is inhibited on entry to
the guest mode, because in this case the guest usually uses its
own msr bitmap.
Later on VM exit, the 1st optimization will allow KVM to skip
touching the L0 msr bitmap as well.
Reviewed-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Tested-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20220519102709.24125-18-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a tracepoint to track number of doorbells being sent
to signal a running vCPU to process IRQ after being injected.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-17-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
For x2AVIC, the index from incomplete IPI #vmexit info is invalid
for logical cluster mode. Only ICRH/ICRL values can be used
to determine the IPI destination APIC ID.
Since QEMU defines guest physical APIC ID to be the same as
vCPU ID, it can be used to quickly identify the target vCPU to deliver IPI,
and avoid the overhead from searching through all vCPUs to match the target
vCPU.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-16-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When launching a VM with x2APIC and specify more than 255 vCPUs,
the guest kernel can disable x2APIC (e.g. specify nox2apic kernel option).
The VM fallbacks to xAPIC mode, and disable the vCPU ID 255 and greater.
In this case, APICV is deactivated for the disabled vCPUs.
However, the current APICv consistency warning does not account for
this case, which results in a warning.
Therefore, modify warning logic to report only when vCPU APIC mode
is valid.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-15-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Currently, AVIC is inhibited when booting a VM w/ x2APIC support.
because AVIC cannot virtualize x2APIC MSR register accesses.
However, the AVIC doorbell can be used to accelerate interrupt
injection into a running vCPU, while all guest accesses to x2APIC MSRs
will be intercepted and emulated by KVM.
With hybrid-AVIC support, the APICV_INHIBIT_REASON_X2APIC is
no longer enforced.
Suggested-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-14-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Originalliy, this WARN_ON is designed to detect when calling
avic_vcpu_load() on an already running vcpu in AVIC mode (i.e. the AVIC
is_running bit is set).
However, for x2AVIC, the vCPU can switch from xAPIC to x2APIC mode while in
running state, in which the avic_vcpu_load() will be called from
svm_refresh_apicv_exec_ctrl().
Therefore, remove this warning since it is no longer appropriate.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta@amd.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-13-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce logic to (de)activate AVIC, which also allows
switching between AVIC to x2AVIC mode at runtime.
When an AVIC-enabled guest switches from APIC to x2APIC mode,
the SVM driver needs to perform the following steps:
1. Set the x2APIC mode bit for AVIC in VMCB along with the maximum
APIC ID support for each mode accodingly.
2. Disable x2APIC MSRs interception in order to allow the hardware
to virtualize x2APIC MSRs accesses.
Reported-by: kernel test robot <lkp@intel.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-12-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
As a preparation for x2avic, this patch ensures that x2apic msrs
are always intercepted for the nested guest.
Reviewed-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Tested-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20220519102709.24125-11-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
AMD AVIC can support xAPIC and x2APIC virtualization,
which requires changing x2APIC bit VMCB and MSR intercepton
for x2APIC MSRs. Therefore, call avic_refresh_apicv_exec_ctrl()
to refresh configuration accordingly.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-10-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
APICv should be deactivated on vCPU that has APIC disabled.
Therefore, call kvm_vcpu_update_apicv() when changing
APIC mode, and add additional check for APIC disable mode
when determine APICV activation,
Suggested-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-9-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When enabling x2APIC virtualization (x2AVIC), the interception of
x2APIC MSRs must be disabled to let the hardware virtualize guest
MSR accesses.
Current implementation keeps track of list of MSR interception state
in the svm_direct_access_msrs array. Therefore, extends the array to
include x2APIC MSRs.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-8-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In X2APIC mode, the Logical Destination Register is read-only,
which provides a fixed mapping between the logical and physical
APIC IDs. Therefore, there is no Logical APIC ID table in X2AVIC
and the processor uses the X2APIC ID in the backing page to create
a vCPU’s logical ID.
In addition, KVM does not support updating APIC ID in x2APIC mode,
which means AVIC does not need to handle this case.
Therefore, check x2APIC mode when handling physical and logical
APIC ID update, and when invalidating logical APIC ID table.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Suggested-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-7-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
xAVIC and x2AVIC modes can support diffferent number of vcpus.
Update existing logics to support each mode accordingly.
Also, modify the maximum physical APIC ID for AVIC to 255 to reflect
the actual value supported by the architecture.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta@amd.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-5-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add CPUID check for the x2APIC virtualization (x2AVIC) feature.
If available, the SVM driver can support both AVIC and x2AVIC modes
when load the kvm_amd driver with avic=1. The operating mode will be
determined at runtime depending on the guest APIC mode.
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta@amd.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <20220519102709.24125-4-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce a new feature bit for virtualized x2APIC (x2AVIC) in
CPUID_Fn8000000A_EDX [SVM Revision and Feature Identification].
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220519102709.24125-2-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a new debugfs file to expose the pid of each vcpu threads. This
is very helpful for userland tools to get the vcpu pids without
worrying about thread naming conventions of the VMM.
Signed-off-by: Vineeth Pillai (Google) <vineeth@bitbyteword.org>
Message-Id: <20220523190327.2658-1-vineeth@bitbyteword.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rely on try_cmpxchg64 for re-reading the PID on failure, using READ_ONCE
only right before the first iteration.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Hardware would directly write x2APIC ICR register instead of software
emulation in some circumstances, e.g when Intel IPI virtualization is
enabled. This behavior requires normal reserved bits checking to ensure
them input as zero, otherwise it will cause #GP. So we need mask out
those reserved bits from the data written to vICR register.
Remove Delivery Status bit emulation in test case as this flag
is invalid and not needed in x2APIC mode. KVM may ignore clearing
it during interrupt dispatch which will lead to fake test failure.
Opportunistically correct vector number for test sending IPI to
non-existent vCPUs.
Signed-off-by: Zeng Guang <guang.zeng@intel.com>
Message-Id: <20220623094511.26066-1-guang.zeng@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch add a self test that verifies user space can inject
UnCorrectable No Action required (UCNA) memory errors to the guest.
It also verifies that incorrectly configured MSRs for Corrected
Machine Check Interrupt (CMCI) emulation will result in #GP.
Signed-off-by: Jue Wang <juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220610171134.772566-9-juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch enables MCG_CMCI_P by default in kvm_mce_cap_supported. It
reuses ioctl KVM_X86_SET_MCE to implement injection of UnCorrectable
No Action required (UCNA) errors, signaled via Corrected Machine
Check Interrupt (CMCI).
Neither of the CMCI and UCNA emulations depends on hardware.
Signed-off-by: Jue Wang <juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220610171134.772566-8-juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch adds the emulation of IA32_MCi_CTL2 registers to KVM. A
separate mci_ctl2_banks array is used to keep the existing mce_banks
register layout intact.
In Machine Check Architecture, in addition to MCG_CMCI_P, bit 30 of
the per-bank register IA32_MCi_CTL2 controls whether Corrected Machine
Check error reporting is enabled.
Signed-off-by: Jue Wang <juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220610171134.772566-7-juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch updates the allocation of mce_banks with the array allocation
API (kcalloc) as a precedent for the later mci_ctl2_banks to implement
per-bank control of Corrected Machine Check Interrupt (CMCI).
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Jue Wang <juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220610171134.772566-6-juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch calculates the number of lvt entries as part of
KVM_X86_MCE_SETUP conditioned on the presence of MCG_CMCI_P bit in
MCG_CAP and stores result in kvm_lapic. It translats from APIC_LVTx
register to index in lapic_lvt_entry enum. It extends the APIC_LVTx
macro as well as other lapic write/reset handling etc to support
Corrected Machine Check Interrupt.
Signed-off-by: Jue Wang <juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220610171134.772566-5-juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
An APIC_LVTx macro is introduced to calcualte the APIC_LVTx register
offset based on the index in the lapic_lvt_entry enum. Later patches
will extend the APIC_LVTx macro to support the APIC_LVTCMCI register
in order to implement Corrected Machine Check Interrupt signaling.
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Jue Wang <juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220610171134.772566-4-juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The TLB flush before installing the newly-populated lower level
page table is unnecessary if the lower-level page table maps
the huge page identically. KVM knows it is if it did not reuse
an existing shadow page table, tell drop_large_spte() to skip
the flush in that case.
Extracted from a patch by David Matlack.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch defines a lapic_lvt_entry enum used as explicit indices to
the apic_lvt_mask array. In later patches a LVT_CMCI will be added to
implement the Corrected Machine Check Interrupt signaling.
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Jue Wang <juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220610171134.772566-3-juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Refactor APIC_VERSION so that the maximum number of LVT entries is
inserted at runtime rather than compile time. This will be used in a
subsequent commit to expose the LVT CMCI Register to VMs that support
Corrected Machine Check error counting/signaling
(IA32_MCG_CAP.MCG_CMCI_P=1).
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Jue Wang <juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220610171134.772566-2-juew@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add support for Eager Page Splitting pages that are mapped by nested
MMUs. Walk through the rmap first splitting all 1GiB pages to 2MiB
pages, and then splitting all 2MiB pages to 4KiB pages.
Note, Eager Page Splitting is limited to nested MMUs as a policy rather
than due to any technical reason (the sp->role.guest_mode check could
just be deleted and Eager Page Splitting would work correctly for all
shadow MMU pages). There is really no reason to support Eager Page
Splitting for tdp_mmu=N, since such support will eventually be phased
out, and there is no current use case supporting Eager Page Splitting on
hosts where TDP is either disabled or unavailable in hardware.
Furthermore, future improvements to nested MMU scalability may diverge
the code from the legacy shadow paging implementation. These
improvements will be simpler to make if Eager Page Splitting does not
have to worry about legacy shadow paging.
Splitting huge pages mapped by nested MMUs requires dealing with some
extra complexity beyond that of the TDP MMU:
(1) The shadow MMU has a limit on the number of shadow pages that are
allowed to be allocated. So, as a policy, Eager Page Splitting
refuses to split if there are KVM_MIN_FREE_MMU_PAGES or fewer
pages available.
(2) Splitting a huge page may end up re-using an existing lower level
shadow page tables. This is unlike the TDP MMU which always allocates
new shadow page tables when splitting.
(3) When installing the lower level SPTEs, they must be added to the
rmap which may require allocating additional pte_list_desc structs.
Case (2) is especially interesting since it may require a TLB flush,
unlike the TDP MMU which can fully split huge pages without any TLB
flushes. Specifically, an existing lower level page table may point to
even lower level page tables that are not fully populated, effectively
unmapping a portion of the huge page, which requires a flush. As of
this commit, a flush is always done always after dropping the huge page
and before installing the lower level page table.
This TLB flush could instead be delayed until the MMU lock is about to be
dropped, which would batch flushes for multiple splits. However these
flushes should be rare in practice (a huge page must be aliased in
multiple SPTEs and have been split for NX Huge Pages in only some of
them). Flushing immediately is simpler to plumb and also reduces the
chances of tripping over a CPU bug (e.g. see iTLB multihit).
[ This commit is based off of the original implementation of Eager Page
Splitting from Peter in Google's kernel from 2016. ]
Suggested-by: Peter Feiner <pfeiner@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220516232138.1783324-23-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Allow the capacity of the kvm_mmu_memory_cache struct to be chosen at
declaration time rather than being fixed for all declarations. This will
be used in a follow-up commit to declare an cache in x86 with a capacity
of 512+ objects without having to increase the capacity of all caches in
KVM.
This change requires each cache now specify its capacity at runtime,
since the cache struct itself no longer has a fixed capacity known at
compile time. To protect against someone accidentally defining a
kvm_mmu_memory_cache struct directly (without the extra storage), this
commit includes a WARN_ON() in kvm_mmu_topup_memory_cache().
In order to support different capacities, this commit changes the
objects pointer array to be dynamically allocated the first time the
cache is topped-up.
While here, opportunistically clean up the stack-allocated
kvm_mmu_memory_cache structs in riscv and arm64 to use designated
initializers.
No functional change intended.
Reviewed-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220516232138.1783324-22-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Before allocating a child shadow page table, all callers check
whether the parent already points to a huge page and, if so, they
drop that SPTE. This is done by drop_large_spte().
However, dropping the large SPTE is really only necessary before the
sp is installed. While the sp is returned by kvm_mmu_get_child_sp(),
installing it happens later in __link_shadow_page(). Move the call
there instead of having it in each and every caller.
To ensure that the shadow page is not linked twice if it was present,
do _not_ opportunistically make kvm_mmu_get_child_sp() idempotent:
instead, return an error value if the shadow page already existed.
This is a bit more verbose, but clearer than NULL.
Finally, now that the drop_large_spte() name is not taken anymore,
remove the two underscores in front of __drop_large_spte().
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Currently KVM only zaps collapsible 4KiB SPTEs in the shadow MMU. This
is fine for now since KVM never creates intermediate huge pages during
dirty logging. In other words, KVM always replaces 1GiB pages directly
with 4KiB pages, so there is no reason to look for collapsible 2MiB
pages.
However, this will stop being true once the shadow MMU participates in
eager page splitting. During eager page splitting, each 1GiB is first
split into 2MiB pages and then those are split into 4KiB pages. The
intermediate 2MiB pages may be left behind if an error condition causes
eager page splitting to bail early.
No functional change intended.
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220516232138.1783324-20-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Currently make_huge_page_split_spte() assumes execute permissions can be
granted to any 4K SPTE when splitting huge pages. This is true for the
TDP MMU but is not necessarily true for the shadow MMU, since KVM may be
shadowing a non-executable huge page.
To fix this, pass in the role of the child shadow page where the huge
page will be split and derive the execution permission from that. This
is correct because huge pages are always split with direct shadow page
and thus the shadow page role contains the correct access permissions.
No functional change intended.
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220516232138.1783324-19-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Splitting huge pages requires allocating/finding shadow pages to replace
the huge page. Shadow pages are keyed, in part, off the guest access
permissions they are shadowing. For fully direct MMUs, there is no
shadowing so the access bits in the shadow page role are always ACC_ALL.
But during shadow paging, the guest can enforce whatever access
permissions it wants.
In particular, eager page splitting needs to know the permissions to use
for the subpages, but KVM cannot retrieve them from the guest page
tables because eager page splitting does not have a vCPU. Fortunately,
the guest access permissions are easy to cache whenever page faults or
FNAME(sync_page) update the shadow page tables; this is an extension of
the existing cache of the shadowed GFNs in the gfns array of the shadow
page. The access bits only take up 3 bits, which leaves 61 bits left
over for gfns, which is more than enough.
Now that the gfns array caches more information than just GFNs, rename
it to shadowed_translation.
While here, preemptively fix up the WARN_ON() that detects gfn
mismatches in direct SPs. The WARN_ON() was paired with a
pr_err_ratelimited(), which means that users could sometimes see the
WARN without the accompanying error message. Fix this by outputting the
error message as part of the WARN splat, and opportunistically make
them WARN_ONCE() because if these ever fire, they are all but guaranteed
to fire a lot and will bring down the kernel.
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20220516232138.1783324-18-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Sean Christopherson
Paolo Bonzini
Maxim Levitsky
Suravee Suthikulpanit
Vineeth Pillai
Zeng Guang
Jue Wang
David Matlack