Drop KVM's manipulation of guest's CPUID.0x12.1 ECX and EDX, i.e. the
allowed XFRM of SGX enclaves, now that KVM explicitly checks the guest's
allowed XCR0 when emulating ECREATE.
Note, this could theoretically break a setup where userspace advertises
a "bad" XFRM and relies on KVM to provide a sane CPUID model, but QEMU
is the only known user of KVM SGX, and QEMU explicitly sets the SGX CPUID
XFRM subleaf based on the guest's XCR0.
Reviewed-by: Kai Huang <[email protected]>
Tested-by: Kai Huang <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Explicitly check the vCPU's supported XCR0 when determining whether or not
the XFRM for ECREATE is valid. Checking CPUID works because KVM updates
guest CPUID.0x12.1 to restrict the leaf to a subset of the guest's allowed
XCR0, but that is rather subtle and KVM should not modify guest CPUID
except for modeling true runtime behavior (allowed XFRM is most definitely
not "runtime" behavior).
Reviewed-by: Kai Huang <[email protected]>
Tested-by: Kai Huang <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Reject hardware enabling, i.e. VM creation, if a restart/shutdown has
been initiated to avoid re-enabling hardware between kvm_reboot() and
machine_{halt,power_off,restart}(). The restart case is especially
problematic (for x86) as enabling VMX (or clearing GIF in KVM_RUN on
SVM) blocks INIT, which results in the restart/reboot hanging as BIOS
is unable to wake and rendezvous with APs.
Note, this bug, and the original issue that motivated the addition of
kvm_reboot(), is effectively limited to a forced reboot, e.g. `reboot -f`.
In a "normal" reboot, userspace will gracefully teardown userspace before
triggering the kernel reboot (modulo bugs, errors, etc), i.e. any process
that might do ioctl(KVM_CREATE_VM) is long gone.
Fixes: 8e1c18157d ("KVM: VMX: Disable VMX when system shutdown")
Signed-off-by: Sean Christopherson <[email protected]>
Acked-by: Marc Zyngier <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Use syscore_ops.shutdown to disable hardware virtualization during a
reboot instead of using the dedicated reboot_notifier so that KVM disables
virtualization _after_ system_state has been updated. This will allow
fixing a race in KVM's handling of a forced reboot where KVM can end up
enabling hardware virtualization between kernel_restart_prepare() and
machine_restart().
Rename KVM's hook to match the syscore op to avoid any possible confusion
from wiring up a "reboot" helper to a "shutdown" hook (neither "shutdown
nor "reboot" is completely accurate as the hook handles both).
Opportunistically rewrite kvm_shutdown()'s comment to make it less VMX
specific, and to explain why kvm_rebooting exists.
Cc: Marc Zyngier <[email protected]>
Cc: Oliver Upton <[email protected]>
Cc: James Morse <[email protected]>
Cc: Suzuki K Poulose <[email protected]>
Cc: Zenghui Yu <[email protected]>
Cc: [email protected]
Cc: Huacai Chen <[email protected]>
Cc: Aleksandar Markovic <[email protected]>
Cc: Anup Patel <[email protected]>
Cc: Atish Patra <[email protected]>
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Acked-by: Marc Zyngier <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Preserve TDP MMU roots until they are explicitly invalidated by gifting
the TDP MMU itself a reference to a root when it is allocated. Keeping a
reference in the TDP MMU fixes a flaw where the TDP MMU exhibits terrible
performance, and can potentially even soft-hang a vCPU, if a vCPU
frequently unloads its roots, e.g. when KVM is emulating SMI+RSM.
When KVM emulates something that invalidates _all_ TLB entries, e.g. SMI
and RSM, KVM unloads all of the vCPUs roots (KVM keeps a small per-vCPU
cache of previous roots). Unloading roots is a simple way to ensure KVM
flushes and synchronizes all roots for the vCPU, as KVM flushes and syncs
when allocating a "new" root (from the vCPU's perspective).
In the shadow MMU, KVM keeps track of all shadow pages, roots included, in
a per-VM hash table. Unloading a shadow MMU root just wipes it from the
per-vCPU cache; the root is still tracked in the per-VM hash table. When
KVM loads a "new" root for the vCPU, KVM will find the old, unloaded root
in the per-VM hash table.
Unlike the shadow MMU, the TDP MMU doesn't track "inactive" roots in a
per-VM structure, where "active" in this case means a root is either
in-use or cached as a previous root by at least one vCPU. When a TDP MMU
root becomes inactive, i.e. the last vCPU reference to the root is put,
KVM immediately frees the root (asterisk on "immediately" as the actual
freeing may be done by a worker, but for all intents and purposes the root
is gone).
The TDP MMU behavior is especially problematic for 1-vCPU setups, as
unloading all roots effectively frees all roots. The issue is mitigated
to some degree in multi-vCPU setups as a different vCPU usually holds a
reference to an unloaded root and thus keeps the root alive, allowing the
vCPU to reuse its old root after unloading (with a flush+sync).
The TDP MMU flaw has been known for some time, as until very recently,
KVM's handling of CR0.WP also triggered unloading of all roots. The
CR0.WP toggling scenario was eventually addressed by not unloading roots
when _only_ CR0.WP is toggled, but such an approach doesn't Just Work
for emulating SMM as KVM must emulate a full TLB flush on entry and exit
to/from SMM. Given that the shadow MMU plays nice with unloading roots
at will, teaching the TDP MMU to do the same is far less complex than
modifying KVM to track which roots need to be flushed before reuse.
Note, preserving all possible TDP MMU roots is not a concern with respect
to memory consumption. Now that the role for direct MMUs doesn't include
information about the guest, e.g. CR0.PG, CR0.WP, CR4.SMEP, etc., there
are _at most_ six possible roots (where "guest_mode" here means L2):
1. 4-level !SMM !guest_mode
2. 4-level SMM !guest_mode
3. 5-level !SMM !guest_mode
4. 5-level SMM !guest_mode
5. 4-level !SMM guest_mode
6. 5-level !SMM guest_mode
And because each vCPU can track 4 valid roots, a VM can already have all
6 root combinations live at any given time. Not to mention that, in
practice, no sane VMM will advertise different guest.MAXPHYADDR values
across vCPUs, i.e. KVM won't ever use both 4-level and 5-level roots for
a single VM. Furthermore, the vast majority of modern hypervisors will
utilize EPT/NPT when available, thus the guest_mode=%true cases are also
unlikely to be utilized.
Reported-by: Jeremi Piotrowski <[email protected]>
Link: https://lore.kernel.org/all/[email protected]
Link: https://lkml.kernel.org/r/20220209170020.1775368-1-pbonzini%40redhat.com
Link: https://lore.kernel.org/all/[email protected]
Link: https://lore.kernel.org/all/[email protected]
Link: https://lore.kernel.org/all/[email protected]
Cc: Ben Gardon <[email protected]>
Cc: David Matlack <[email protected]>
Cc: [email protected]
Tested-by: Jeremi Piotrowski <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Use a single struct to track all PMC event counts in the PMU filter test,
and copy the full struct to/from the guest when running and measuring each
guest workload. Using a common struct avoids naming conflicts, e.g. the
loads/stores testcase has claimed "perf_counter", and eliminates the
unnecessary truncation of the counter values when they are propagated from
the guest MSRs to the host structs.
Zero the struct before running the guest workload to ensure that the test
doesn't get a false pass due to consuming data from a previous run.
Link: https://lore.kernel.org/r/[email protected]
Reviewed by: Aaron Lewis <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Use '0' to signal success and '-errno' to signal failure in the PMU event
filter test so that the values are slightly less magical/arbitrary. Using
'0' in the error paths is especially confusing as understanding it's an
error value requires following the breadcrumbs to the host code that
ultimately consumes the value.
Arguably there should also be a #define for "success", but 0/-errno is a
common enough pattern that defining another macro on top would likely do
more harm than good.
Link: https://lore.kernel.org/r/[email protected]
Reviewed by: Aaron Lewis <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Take the XFeature mask in __vm_xsave_require_permission() instead of the
bit so that there's no need to define macros for both the bit and the
mask. Asserting that only a single bit is set and retrieving said bit
is easy enough via log2 helpers.
Opportunistically clean up the error message for the
ARCH_REQ_XCOMP_GUEST_PERM sanity check.
Reviewed-by: Aaron Lewis <[email protected]>
Tested-by: Aaron Lewis <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Filter out XTILE_CFG from the supported XCR0 reported to userspace if the
current process doesn't have access to XTILE_DATA. Attempting to set
XTILE_CFG in XCR0 will #GP if XTILE_DATA is also not set, and so keeping
XTILE_CFG as supported results in explosions if userspace feeds
KVM_GET_SUPPORTED_CPUID back into KVM and the guest doesn't sanity check
CPUID.
Fixes: 445ecdf79b ("kvm: x86: Exclude unpermitted xfeatures at KVM_GET_SUPPORTED_CPUID")
Reported-by: Aaron Lewis <[email protected]>
Reviewed-by: Aaron Lewis <[email protected]>
Tested-by: Aaron Lewis <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Extend VMX's nested intercept logic for emulated instructions to handle
"pause" interception, in quotes because KVM's emulator doesn't filter out
NOPs when checking for nested intercepts. Failure to allow emulation of
NOPs results in KVM injecting a #UD into L2 on any NOP that collides with
the emulator's definition of PAUSE, i.e. on all single-byte NOPs.
For PAUSE itself, honor L1's PAUSE-exiting control, but ignore PLE to
avoid unnecessarily injecting a #UD into L2. Per the SDM, the first
execution of PAUSE after VM-Entry is treated as the beginning of a new
loop, i.e. will never trigger a PLE VM-Exit, and so L1 can't expect any
given execution of PAUSE to deterministically exit.
... the processor considers this execution to be the first execution of
PAUSE in a loop. (It also does so for the first execution of PAUSE at
CPL 0 after VM entry.)
All that said, the PLE side of things is currently a moot point, as KVM
doesn't expose PLE to L1.
Note, vmx_check_intercept() is still wildly broken when L1 wants to
intercept an instruction, as KVM injects a #UD instead of synthesizing a
nested VM-Exit. That issue extends far beyond NOP/PAUSE and needs far
more effort to fix, i.e. is a problem for the future.
Fixes: 07721feee4 ("KVM: nVMX: Don't emulate instructions in guest mode")
Cc: Mathias Krause <[email protected]>
Cc: [email protected]
Reviewed-by: Paolo Bonzini <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Refresh the MMU's snapshot of the vCPU's CR0.WP prior to checking for
permission faults when emulating a guest memory access and CR0.WP may be
guest owned. If the guest toggles only CR0.WP and triggers emulation of
a supervisor write, e.g. when KVM is emulating UMIP, KVM may consume a
stale CR0.WP, i.e. use stale protection bits metadata.
Note, KVM passes through CR0.WP if and only if EPT is enabled as CR0.WP
is part of the MMU role for legacy shadow paging, and SVM (NPT) doesn't
support per-bit interception controls for CR0. Don't bother checking for
EPT vs. NPT as the "old == new" check will always be true under NPT, i.e.
the only cost is the read of vcpu->arch.cr4 (SVM unconditionally grabs CR0
from the VMCB on VM-Exit).
Reported-by: Mathias Krause <[email protected]>
Link: https://lkml.kernel.org/r/677169b4-051f-fcae-756b-9a3e1bb9f8fe%40grsecurity.net
Fixes: fb509f76ac ("KVM: VMX: Make CR0.WP a guest owned bit")
Tested-by: Mathias Krause <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Refactor Hyper-V's range-based TLB flushing API to take a gfn+nr_pages
pair instead of a struct, and bury said struct in Hyper-V specific code.
Passing along two params generates much better code for the common case
where KVM is _not_ running on Hyper-V, as forwarding the flush on to
Hyper-V's hv_flush_remote_tlbs_range() from kvm_flush_remote_tlbs_range()
becomes a tail call.
Cc: David Matlack <[email protected]>
Reviewed-by: David Matlack <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Rename the Hyper-V hooks for TLB flushing to match the naming scheme used
by all the other TLB flushing hooks, e.g. in kvm_x86_ops, vendor code,
arch hooks from common code, etc.
Reviewed-by: David Matlack <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Invert the flows in intel_pmu_{g,s}et_msr()'s case statements so that
they follow the kernel's preferred style of:
if (<not valid>)
return <error>
<commit change>
return <success>
which is also the style used by every other {g,s}et_msr() helper (except
AMD's PMU variant, which doesn't use a switch statement).
Modify the "set" paths with costly side effects, i.e. that reprogram
counters, to skip only the side effects, i.e. to perform reserved bits
checks even if the value is unchanged. None of the reserved bits checks
are expensive, so there's no strong justification for skipping them, and
guarding only the side effect makes it slightly more obvious what is being
skipped and why.
No functional change intended (assuming no reserved bit bugs).
Link: https://lkml.kernel.org/r/Y%2B6cfen%2FCpO3%2FdLO%40google.com
Signed-off-by: Sean Christopherson <[email protected]>
Verify that disabling the guest's vPMU via CPUID also disables LBRs.
KVM has had at least one bug where LBRs would remain enabled even though
the intent was to disable everything PMU related.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Expand the immutable features sub-test for PERF_CAPABILITIES to verify
KVM rejects any attempt to use a PEBS format other than the host's.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Rework the LBR format test to use the bitfield instead of a separate
mask macro, mainly so that adding a nearly-identical PEBS format test
doesn't have to copy-paste-tweak the macro too.
No functional change intended.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Now that KVM disallows changing PERF_CAPABILITIES after KVM_RUN, expand
the host side checks to verify KVM rejects any attempts to change bits
from userspace.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Test that the guest can't write 0 to PERF_CAPABILITIES, can't write the
current value, and can't toggle _any_ bits. There is no reason to special
case the LBR format.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Add negative testing of all immutable bits in PERF_CAPABILITIES, i.e.
single bits that are reserved-0 or are effectively reserved-1 by KVM.
Omit LBR and PEBS format bits from the test as it's easier to test them
manually than it is to add safeguards to the comment path, e.g. toggling
a single bit can yield a format of '0', which is legal as a "disable"
value.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Now that vcpu_set_msr() verifies the expected "read what was wrote"
semantics of all durable MSRs, including PERF_CAPABILITIES, drop the
now-redundant manual checks in the VMX PMU caps test.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Assert that KVM provides "read what you wrote" semantics for all "durable"
MSRs (for lack of a better name). The extra coverage is cheap from a
runtime performance perspective, and verifying the behavior in the common
helper avoids gratuitous copy+paste in individual tests.
Note, this affects all tests that set MSRs from userspace!
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Reimplement vcpu_set_msr() as a macro and pretty print the failing MSR
(when possible) and the value if KVM_SET_MSRS fails instead of using the
using the standard KVM_IOCTL_ERROR(). KVM_SET_MSRS is somewhat odd in
that it returns the index of the last successful write, i.e. will be
'0' on failure barring an entirely different KVM bug. And for writing
MSRs, the MSR being written and the value being written are almost always
relevant to the failure, i.e. just saying "failed!" doesn't help debug.
Place the string goo in a separate macro in anticipation of using it to
further expand MSR testing.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
KVM emulates full-width PMC writes in software, assert that KVM reports
full-width writes as supported if PERF_CAPABILITIES is supported.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Use a separate sub-test to verify userspace can clear PERF_CAPABILITIES
and restore it to the KVM-supported value, as the testcase isn't unique
to the LBR format.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Zero out the LBR capabilities during PMU refresh to avoid exposing LBRs
to the guest against userspace's wishes. If userspace modifies the
guest's CPUID model or invokes KVM_CAP_PMU_CAPABILITY to disable vPMU
after an initial KVM_SET_CPUID2, but before the first KVM_RUN, KVM will
retain the previous LBR info due to bailing before refreshing the LBR
descriptor.
Note, this is a very theoretical bug, there is no known use case where a
VMM would deliberately enable the vPMU via KVM_SET_CPUID2, and then later
disable the vPMU.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Now that KVM disallows changing feature MSRs, i.e. PERF_CAPABILITIES,
after running a vCPU, WARN and bug the VM if the PMU is refreshed after
the vCPU has run.
Note, KVM has disallowed CPUID updates after running a vCPU since commit
feb627e8d6 ("KVM: x86: Forbid KVM_SET_CPUID{,2} after KVM_RUN"), i.e.
PERF_CAPABILITIES was the only remaining way to trigger a PMU refresh
after KVM_RUN.
Cc: Like Xu <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Disallow writes to feature MSRs after KVM_RUN to prevent userspace from
changing the vCPU model after running the vCPU. Similar to guest CPUID,
KVM uses feature MSRs to configure intercepts, determine what operations
are/aren't allowed, etc. Changing the capabilities while the vCPU is
active will at best yield unpredictable guest behavior, and at worst
could be dangerous to KVM.
Allow writing the current value, e.g. so that userspace can blindly set
all MSRs when emulating RESET, and unconditionally allow writes to
MSR_IA32_UCODE_REV so that userspace can emulate patch loads.
Special case the VMX MSRs to keep the generic list small, i.e. so that
KVM can do a linear walk of the generic list without incurring meaningful
overhead.
Cc: Like Xu <[email protected]>
Cc: Yu Zhang <[email protected]>
Reviewed-by: Xiaoyao Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Split the PERF_CAPABILITIES subtests into two parts so that the LBR format
testcases don't execute after KVM_RUN. Similar to the guest CPUID model,
KVM will soon disallow changing PERF_CAPABILITIES after KVM_RUN, at which
point attempting to set the MSR after KVM_RUN will yield false positives
and/or false negatives depending on what the test is trying to do.
Land the LBR format test in a more generic "immutable features" test in
anticipation of expanding its scope to other immutable features.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Add VMX MSRs to the runtime list of feature MSRs by iterating over the
range of emulated MSRs instead of manually defining each MSR in the "all"
list. Using the range definition reduces the cost of emulating a new VMX
MSR, e.g. prevents forgetting to add an MSR to the list.
Extracting the VMX MSRs from the "all" list, which is a compile-time
constant, also shrinks the list to the point where the compiler can
heavily optimize code that iterates over the list.
No functional change intended.
Reviewed-by: Xiaoyao Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Add macros to track the range of VMX feature MSRs that are emulated by
KVM to reduce the maintenance cost of extending the set of emulated MSRs.
Note, KVM doesn't necessarily emulate all known/consumed VMX MSRs, e.g.
PROCBASED_CTLS3 is consumed by KVM to enable IPI virtualization, but is
not emulated as KVM doesn't emulate/virtualize IPI virtualization for
nested guests.
No functional change intended.
Reviewed-by: Xiaoyao Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Rename kvm_init_msr_list() to kvm_init_msr_lists() to clarify that it
initializes multiple lists: MSRs to save, emulated MSRs, and feature MSRs.
No functional change intended.
Reviewed-by: Xiaoyao Li <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Rename "r" to "ret" and actually return it from svm_set_msr() to reduce
the probability of repeating the mistake of commit 723d5fb0ff ("kvm:
svm: Add IA32_FLUSH_CMD guest support"), which set "r" thinking that it
would be propagated to the caller.
Alternatively, the declaration of "r" could be moved into the handling of
MSR_TSC_AUX, but that risks variable shadowing in the future. A wrapper
for kvm_set_user_return_msr() would allow eliding a local variable, but
that feels like delaying the inevitable.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Virtualize FLUSH_L1D so that the guest can use the performant L1D flush
if one of the many mitigations might require a flush in the guest, e.g.
Linux provides an option to flush the L1D when switching mms.
Passthrough MSR_IA32_FLUSH_CMD for write when it's supported in hardware
and exposed to the guest, i.e. always let the guest write it directly if
FLUSH_L1D is fully supported.
Forward writes to hardware in host context on the off chance that KVM
ends up emulating a WRMSR, or in the really unlikely scenario where
userspace wants to force a flush. Restrict these forwarded WRMSRs to
the known command out of an abundance of caution. Passing through the
MSR means the guest can throw any and all values at hardware, but doing
so in host context is arguably a bit more dangerous.
Link: https://lkml.kernel.org/r/CALMp9eTt3xzAEoQ038bJQ9LN0ZOXrSWsN7xnNUD%2B0SS%3DWwF7Pg%40mail.gmail.com
Link: https://lore.kernel.org/all/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Dedup the handling of MSR_IA32_PRED_CMD across VMX and SVM by moving the
logic to kvm_set_msr_common(). Now that the MSR interception toggling is
handled as part of setting guest CPUID, the VMX and SVM paths are
identical.
Opportunistically massage the code to make it a wee bit denser.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Xiaoyao Li <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Passthrough MSR_IA32_PRED_CMD based purely on whether or not the MSR is
supported and enabled, i.e. don't wait until the first write. There's no
benefit to deferred passthrough, and the extra logic only adds complexity.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Passthrough MSR_IA32_PRED_CMD based purely on whether or not the MSR is
supported and enabled, i.e. don't wait until the first write. There's no
benefit to deferred passthrough, and the extra logic only adds complexity.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Xiaoyao Li <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Revert the recently added virtualizing of MSR_IA32_FLUSH_CMD, as both
the VMX and SVM are fatally buggy to guests that use MSR_IA32_FLUSH_CMD or
MSR_IA32_PRED_CMD, and because the entire foundation of the logic is
flawed.
The most immediate problem is an inverted check on @cmd that results in
rejecting legal values. SVM doubles down on bugs and drops the error,
i.e. silently breaks all guest mitigations based on the command MSRs.
The next issue is that neither VMX nor SVM was updated to mark
MSR_IA32_FLUSH_CMD as being a possible passthrough MSR,
which isn't hugely problematic, but does break MSR filtering and triggers
a WARN on VMX designed to catch this exact bug.
The foundational issues stem from the MSR_IA32_FLUSH_CMD code reusing
logic from MSR_IA32_PRED_CMD, which in turn was likely copied from KVM's
support for MSR_IA32_SPEC_CTRL. The copy+paste from MSR_IA32_SPEC_CTRL
was misguided as MSR_IA32_PRED_CMD (and MSR_IA32_FLUSH_CMD) is a
write-only MSR, i.e. doesn't need the same "deferred passthrough"
shenanigans as MSR_IA32_SPEC_CTRL.
Revert all MSR_IA32_FLUSH_CMD enabling in one fell swoop so that there is
no point where KVM advertises, but does not support, L1D_FLUSH.
This reverts commits 45cf86f261,
723d5fb0ff, and
a807b78ad0.
Reported-by: Nathan Chancellor <[email protected]>
Link: https://lkml.kernel.org/r/20230317190432.GA863767%40dev-arch.thelio-3990X
Cc: Emanuele Giuseppe Esposito <[email protected]>
Cc: Pawan Gupta <[email protected]>
Cc: Jim Mattson <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Tested-by: Mathias Krause <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Disallow enabling LBR support if the CPU supports architectural LBRs.
Traditional LBR support is absent on CPU models that have architectural
LBRs, and KVM doesn't yet support arch LBRs, i.e. KVM will pass through
non-existent MSRs if userspace enables LBRs for the guest.
Cc: [email protected]
Cc: Yang Weijiang <[email protected]>
Cc: Like Xu <[email protected]>
Reported-by: Paolo Bonzini <[email protected]>
Fixes: be635e34c2 ("KVM: vmx/pmu: Expose LBR_FMT in the MSR_IA32_PERF_CAPABILITIES")
Tested-by: Like Xu <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Don't report an error code to L1 when synthesizing a nested VM-Exit and
L2 is in Real Mode. Per Intel's SDM, regarding the error code valid bit:
This bit is always 0 if the VM exit occurred while the logical processor
was in real-address mode (CR0.PE=0).
The bug was introduced by a recent fix for AMD's Paged Real Mode, which
moved the error code suppression from the common "queue exception" path
to the "inject exception" path, but missed VMX's "synthesize VM-Exit"
path.
Fixes: b97f074583 ("KVM: x86: determine if an exception has an error code only when injecting it.")
Cc: [email protected]
Cc: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
When injecting an exception into a vCPU in Real Mode, suppress the error
code by clearing the flag that tracks whether the error code is valid, not
by clearing the error code itself. The "typo" was introduced by recent
fix for SVM's funky Paged Real Mode.
Opportunistically hoist the logic above the tracepoint so that the trace
is coherent with respect to what is actually injected (this was also the
behavior prior to the buggy commit).
Fixes: b97f074583 ("KVM: x86: determine if an exception has an error code only when injecting it.")
Cc: [email protected]
Cc: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Clear vcpu->mmio_needed when injecting an exception from the emulator to
squash a (legitimate) warning about vcpu->mmio_needed being true at the
start of KVM_RUN without a callback being registered to complete the
userspace MMIO exit. Suppressing the MMIO write exit is inarguably wrong
from an architectural perspective, but it is the least awful hack-a-fix
due to shortcomings in KVM's uAPI, not to mention that KVM already
suppresses MMIO writes in this scenario.
Outside of REP string instructions, KVM doesn't provide a way to resume
an instruction at the exact point where it was "interrupted" if said
instruction partially completed before encountering an MMIO access. For
MMIO reads, KVM immediately exits to userspace upon detecting MMIO as
userspace provides the to-be-read value in a buffer, and so KVM can safely
(more or less) restart the instruction from the beginning. When the
emulator re-encounters the MMIO read, KVM will service the MMIO by getting
the value from the buffer instead of exiting to userspace, i.e. KVM won't
put the vCPU into an infinite loop.
On an emulated MMIO write, KVM finishes the instruction before exiting to
userspace, as exiting immediately would ultimately hang the vCPU due to
the aforementioned shortcoming of KVM not being able to resume emulation
in the middle of an instruction.
For the vast majority of _emulated_ instructions, deferring the userspace
exit doesn't cause problems as very few x86 instructions (again ignoring
string operations) generate multiple writes. But for instructions that
generate multiple writes, e.g. PUSHA (multiple pushes onto the stack),
deferring the exit effectively results in only the final write triggering
an exit to userspace. KVM does support multiple MMIO "fragments", but
only for page splits; if an instruction performs multiple distinct MMIO
writes, the number of fragments gets reset when the next MMIO write comes
along and any previous MMIO writes are dropped.
Circling back to the warning, if a deferred MMIO write coincides with an
exception, e.g. in this case a #SS due to PUSHA underflowing the stack
after queueing a write to an MMIO page on a previous push, KVM injects
the exceptions and leaves the deferred MMIO pending without registering a
callback, thus triggering the splat.
Sweep the problem under the proverbial rug as dropping MMIO writes is not
unique to the exception scenario (see above), i.e. instructions like PUSHA
are fundamentally broken with respect to MMIO, and have been since KVM's
inception.
Reported-by: zhangjianguo <[email protected]>
Reported-by: [email protected]
Reported-by: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Yell loudly if KVM attempts to load CS outside of Real Mode without an
accompanying control transfer type, i.e. on X86_TRANSFER_NONE. KVM uses
X86_TRANSFER_NONE when emulating IRET and exceptions/interrupts for Real
Mode, but IRET emulation for Protected Mode is non-existent. WARN instead
of trying to pass in a less-wrong type, e.g. X86_TRANSFER_RET, as
emulating IRET goes even beyond emulating FAR RET (which KVM also doesn't
fully support).
Reported-by: Hou Wenlong <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Use the asynchronous NMI queue to handle pending NMIs coming in from
userspace during KVM_SET_VCPU_EVENTS so that all of KVM's logic for
handling multiple NMIs goes through process_nmi(). This will simplify
supporting SVM's upcoming "virtual NMI" functionality, which will need
changes KVM manages pending NMIs.
Signed-off-by: Sean Christopherson <[email protected]>
The existing X86_FEATURE_VNMI is a synthetic feature flag that exists
purely to maintain /proc/cpuinfo's ABI, the "real" Intel vNMI feature flag
is tracked as VMX_FEATURE_VIRTUAL_NMIS, as the feature is enumerated
through VMX MSRs, not CPUID.
AMD is also gaining virtual NMI support, but in true VMX vs. SVM form,
enumerates support through CPUID, i.e. wants to add real feature flag for
vNMI.
Redefine the syntheic X86_FEATURE_VNMI to AMD's real CPUID bit to avoid
having both X86_FEATURE_VNMI and e.g. X86_FEATURE_AMD_VNMI.
Signed-off-by: Sean Christopherson <[email protected]>
Save all pending NMIs in KVM_GET_VCPU_EVENTS, and queue KVM_REQ_NMI if one
or more NMIs are pending after KVM_SET_VCPU_EVENTS in order to re-evaluate
pending NMIs with respect to NMI blocking.
KVM allows multiple NMIs to be pending in order to faithfully emulate bare
metal handling of simultaneous NMIs (on bare metal, truly simultaneous
NMIs are impossible, i.e. one will always arrive first and be consumed).
Support for simultaneous NMIs botched the save/restore though. KVM only
saves one pending NMI, but allows userspace to restore 255 pending NMIs
as kvm_vcpu_events.nmi.pending is a u8, and KVM's internal state is stored
in an unsigned int.
Fixes: 7460fb4a34 ("KVM: Fix simultaneous NMIs")
Signed-off-by: Santosh Shukla <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Tweak the code and comment that deals with concurrent NMIs to explicitly
call out that x86 allows exactly one pending NMI, but that KVM needs to
temporarily allow two pending NMIs in order to workaround the fact that
the target vCPU cannot immediately recognize an incoming NMI, unlike bare
metal.
No functional change intended.
Signed-off-by: Santosh Shukla <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Don't raise KVM_REQ_EVENT if no NMIs are pending at the end of
process_nmi(). Finishing process_nmi() without a pending NMI will become
much more likely when KVM gains support for AMD's vNMI, which allows
pending vNMIs in hardware, i.e. doesn't require explicit injection.
Signed-off-by: Santosh Shukla <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Explicitly convert the return from is_paging() to a bool when comparing
against old_paging, which is also a boolean. is_paging() sneakily uses
kvm_read_cr0_bits() and returns an int, i.e. returns X86_CR0_PG or 0, not
1 or 0.
Luckily, the bug is benign as it only results in a false positive, not a
false negative, i.e. only causes a spurious refresh of CR4 when paging is
enabled in both the old and new.
Cc: Maxim Levitsky <[email protected]>
Fixes: c53bbe2145 ("KVM: x86: SVM: don't passthrough SMAP/SMEP/PKE bits in !NPT && !gCR0.PG case")
Signed-off-by: Sean Christopherson <[email protected]>
Adjust a variety of functions in mmu.c to put the function return type on
the same line as the function declaration. As stated in the Linus
specification:
But the "on their own line" is complete garbage to begin with. That
will NEVER be a kernel rule. We should never have a rule that assumes
things are so long that they need to be on multiple lines.
We don't put function return types on their own lines either, even if
some other projects have that rule (just to get function names at the
beginning of lines or some other odd reason).
Leave the functions generated by BUILD_MMU_ROLE_REGS_ACCESSOR() as-is,
that code is basically illegible no matter how it's formatted.
No functional change intended.
Link: https://lore.kernel.org/mm-commits/CAHk-=wjS-Jg7sGMwUPpDsjv392nDOOs0CtUtVkp=S6Q7JzFJRw@mail.gmail.com
Signed-off-by: Ben Gardon <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Tweak KVM_MMU_ROOTS_ALL to precisely cover all current+previous root
flags, and add a sanity in kvm_mmu_free_roots() to verify that the set
of roots to free doesn't stray outside KVM_MMU_ROOTS_ALL.
Signed-off-by: Lai Jiangshan <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Rename enable_evmcs to __kvm_is_using_evmcs to match its wrapper, and to
avoid confusion with enabling eVMCS for nested virtualization, i.e. have
"enable eVMCS" be reserved for "enable eVMCS support for L1".
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Wrap enable_evmcs in a helper and stub it out when CONFIG_HYPERV=n in
order to eliminate the static branch nop placeholders. clang-14 is clever
enough to elide the nop, but gcc-12 is not. Stubbing out the key reduces
the size of kvm-intel.ko by ~7.5% (200KiB) when compiled with gcc-12
(there are a _lot_ of VMCS accesses throughout KVM).
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Move the macros that define the set of VMCS controls that are supported
by eVMCS1 from hyperv.h to hyperv.c, i.e. make them "private". The
macros should never be consumed directly by KVM at-large since the "final"
set of supported controls depends on guest CPUID.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Use a new EMULTYPE flag, EMULTYPE_WRITE_PF_TO_SP, to track page faults
on self-changing writes to shadowed page tables instead of propagating
that information to the emulator via a semi-persistent vCPU flag. Using
a flag in "struct kvm_vcpu_arch" is confusing, especially as implemented,
as it's not at all obvious that clearing the flag only when emulation
actually occurs is correct.
E.g. if KVM sets the flag and then retries the fault without ever getting
to the emulator, the flag will be left set for future calls into the
emulator. But because the flag is consumed if and only if both
EMULTYPE_PF and EMULTYPE_ALLOW_RETRY_PF are set, and because
EMULTYPE_ALLOW_RETRY_PF is deliberately not set for direct MMUs, emulated
MMIO, or while L2 is active, KVM avoids false positives on a stale flag
since FNAME(page_fault) is guaranteed to be run and refresh the flag
before it's ultimately consumed by the tail end of reexecute_instruction().
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Add and use a macro to generate the KVM exit reason strings array
instead of relying on developers to correctly copy+paste+edit each
string.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Add wrappers to do hypercalls using VMCALL/VMMCALL and Xen's register ABI
(as opposed to full Xen-style hypercalls through a hypervisor provided
page). Using the common helpers dedups a pile of code, and uses the
native hypercall instruction when running on AMD.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Extract the guts of kvm_hypercall() to a macro so that Xen hypercalls,
which have a different register ABI, can reuse the VMCALL vs. VMMCALL
logic.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
WARN if generating a GATag given a VM ID and vCPU ID doesn't yield the
same IDs when pulling the IDs back out of the tag. Don't bother adding
error handling to callers, this is very much a paranoid sanity check as
KVM fully controls the VM ID and is supposed to reject too-big vCPU IDs.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Suravee Suthikulpanit <[email protected]>
Tested-by: Suravee Suthikulpanit <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Define the "physical table max index mask" as bits 8:0, not 9:0. x2AVIC
currently supports a max of 512 entries, i.e. the max index is 511, and
the inputs to GENMASK_ULL() are inclusive. The bug is benign as bit 9 is
reserved and never set by KVM, i.e. KVM is just clearing bits that are
guaranteed to be zero.
Note, as of this writing, APM "Rev. 3.39-October 2022" incorrectly states
that bits 11:8 are reserved in Table B-1. VMCB Layout, Control Area. I.e.
that table wasn't updated when x2AVIC support was added.
Opportunistically fix the comment for the max AVIC ID to align with the
code, and clean up comment formatting too.
Fixes: 4d1d7942e3 ("KVM: SVM: Introduce logic to (de)activate x2AVIC mode")
Cc: [email protected]
Cc: Alejandro Jimenez <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Suravee Suthikulpanit <[email protected]>
Tested-by: Suravee Suthikulpanit <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Disable KVM support for virtualizing PMUs on hosts with hybrid PMUs until
KVM gains a sane way to enumeration the hybrid vPMU to userspace and/or
gains a mechanism to let userspace opt-in to the dangers of exposing a
hybrid vPMU to KVM guests. Virtualizing a hybrid PMU, or at least part of
a hybrid PMU, is possible, but it requires careful, deliberate
configuration from userspace.
E.g. to expose full functionality, vCPUs need to be pinned to pCPUs to
prevent migrating a vCPU between a big core and a little core, userspace
must enumerate a reasonable topology to the guest, and guest CPUID must be
curated per vCPU to enumerate accurate vPMU capabilities.
The last point is especially problematic, as KVM doesn't control which
pCPU it runs on when enumerating KVM's vPMU capabilities to userspace,
i.e. userspace can't rely on KVM_GET_SUPPORTED_CPUID in it's current form.
Alternatively, userspace could enable vPMU support by enumerating the
set of features that are common and coherent across all cores, e.g. by
filtering PMU events and restricting guest capabilities. But again, that
requires userspace to take action far beyond reflecting KVM's supported
feature set into the guest.
For now, simply disable vPMU support on hybrid CPUs to avoid inducing
seemingly random #GPs in guests, and punt support for hybrid CPUs to a
future enabling effort.
Reported-by: Jianfeng Gao <[email protected]>
Cc: [email protected]
Cc: Andrew Cooper <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Kan Liang <[email protected]>
Cc: Andi Kleen <[email protected]>
Link: https://lore.kernel.org/all/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Provide "error" semantics (read zeros, drop writes) for userspace accesses
to MSRs that are ultimately unsupported for whatever reason, but for which
KVM told userspace to save and restore the MSR, i.e. for MSRs that KVM
included in KVM_GET_MSR_INDEX_LIST.
Previously, KVM special cased a few PMU MSRs that were problematic at one
point or another. Extend the treatment to all PMU MSRs, e.g. to avoid
spurious unsupported accesses.
Note, the logic can also be used for non-PMU MSRs, but as of today only
PMU MSRs can end up being unsupported after KVM told userspace to save and
restore them.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Omit all PMU MSRs from the "MSRs to save" list if the PMU is disabled so
that userspace doesn't waste time saving and restoring dummy values. KVM
provides "error" semantics (read zeros, drop writes) for such known-but-
unsupported MSRs, i.e. has fudged around this issue for quite some time.
Keep the "error" semantics as-is for now, the logic will be cleaned up in
a separate patch.
Cc: Aaron Lewis <[email protected]>
Cc: Weijiang Yang <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Move all potential to-be-saved PMU MSRs into a separate array so that a
future patch can easily omit all PMU MSRs from the list when the PMU is
disabled.
No functional change intended.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Add helpers to print unimplemented MSR accesses and condition all such
prints on report_ignored_msrs, i.e. honor userspace's request to not
print unimplemented MSRs. Even though vcpu_unimpl() is ratelimited,
printing can still be problematic, e.g. if a print gets stalled when host
userspace is writing MSRs during live migration, an effective stall can
result in very noticeable disruption in the guest.
E.g. the profile below was taken while calling KVM_SET_MSRS on the PMU
counters while the PMU was disabled in KVM.
- 99.75% 0.00% [.] __ioctl
- __ioctl
- 99.74% entry_SYSCALL_64_after_hwframe
do_syscall_64
sys_ioctl
- do_vfs_ioctl
- 92.48% kvm_vcpu_ioctl
- kvm_arch_vcpu_ioctl
- 85.12% kvm_set_msr_ignored_check
svm_set_msr
kvm_set_msr_common
printk
vprintk_func
vprintk_default
vprintk_emit
console_unlock
call_console_drivers
univ8250_console_write
serial8250_console_write
uart_console_write
Reported-by: Aaron Lewis <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Limit kvm_pmu_cap.num_counters_gp during kvm_init_pmu_capability() based
on the vendor PMU capabilities so that consuming num_counters_gp naturally
does the right thing. This fixes a mostly theoretical bug where KVM could
over-report its PMU support in KVM_GET_SUPPORTED_CPUID for leaf 0xA, e.g.
if the number of counters reported by perf is greater than KVM's
hardcoded internal limit. Incorporating input from the AMD PMU also
avoids over-reporting MSRs to save when running on AMD.
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Move VMX's handling of NMI VM-Exits into vmx_vcpu_enter_exit() so that
the NMI is handled prior to leaving the safety of noinstr. Handling the
NMI after leaving noinstr exposes the kernel to potential ordering
problems as an instrumentation-induced fault, e.g. #DB, #BP, #PF, etc.
will unblock NMIs when IRETing back to the faulting instruction.
Reported-by: Peter Zijlstra <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Split the asm subroutines for handling NMIs versus IRQs that occur in the
guest so that the NMI handler can be called from a noinstr section. As a
bonus, the NMI path doesn't need an indirect branch.
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Use a dedicated entry for invoking the NMI handler from KVM VMX's VM-Exit
path for 32-bit even though using a dedicated entry for 32-bit isn't
strictly necessary. Exposing a single symbol will allow KVM to reference
the entry point in assembly code without having to resort to more #ifdefs
(or #defines). identry.h is intended to be included from asm files only
once, and so simply including idtentry.h in KVM assembly isn't an option.
Bypassing the ESP fixup and CR3 switching in the standard NMI entry code
is safe as KVM always handles NMIs that occur in the guest on a kernel
stack, with a kernel CR3.
Cc: Andy Lutomirski <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Reviewed-by: Lai Jiangshan <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Tag to_vmx() and to_kvm_vmx() __always_inline as they both just reflect
the passed in pointer (the embedded struct is the first field in the
container), and drop the @vmx param from vmx_vcpu_enter_exit(), which
likely existed purely to make noinstr validation happy.
Amusingly, when the compiler decides to not inline the helpers, e.g. for
KASAN builds, to_vmx() and to_kvm_vmx() may end up pointing at the same
symbol, which generates very confusing objtool warnings. E.g. the use of
to_vmx() in a future patch led to objtool complaining about to_kvm_vmx(),
and only once all use of to_kvm_vmx() was commented out did to_vmx() pop
up in the obj tool report.
vmlinux.o: warning: objtool: vmx_vcpu_enter_exit+0x160: call to to_kvm_vmx()
leaves .noinstr.text section
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Tag all evmcs_{read,write}() helpers __always_inline so that they can be
freely used in noinstr sections, e.g. to get the VM-Exit reason in
vcpu_vmx_enter_exit() (in a future patch). For consistency and to avoid
more spot fixes in the future, e.g. see commit 010050a863 ("x86/kvm:
Always inline evmcs_write64()"), tag all accessors even though
evmcs_read32() is the only anticipated use case in the near future. In
practice, non-KASAN builds are all but guaranteed to inline the helpers
anyways.
vmlinux.o: warning: objtool: vmx_vcpu_enter_exit+0x107: call to evmcs_read32()
leaves .noinstr.text section
Reported-by: kernel test robot <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Allow instrumentation in the VM-Fail path of __vmcs_readl() so that the
helper can be used in noinstr functions, e.g. to get the exit reason in
vmx_vcpu_enter_exit() in order to handle NMI VM-Exits in the noinstr
section. While allowing instrumentation isn't technically safe, KVM has
much bigger problems if VMREAD fails in a noinstr section.
Note, all other VMX instructions also allow instrumentation in their
VM-Fail paths for similar reasons, VMREAD was simply omitted by commit
3ebccdf373 ("x86/kvm/vmx: Move guest enter/exit into .noinstr.text")
because VMREAD wasn't used in a noinstr section at the time.
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Add an extra special noinstr-friendly helper to test+mark a "register"
available and use it when caching vmcs.EXIT_QUALIFICATION and
vmcs.VM_EXIT_INTR_INFO. Make the caching helpers __always_inline too so
that they can be used in noinstr functions.
A future fix will move VMX's handling of NMI exits into the noinstr
vmx_vcpu_enter_exit() so that the NMI is processed before any kind of
instrumentation can trigger a fault and thus IRET, i.e. so that KVM
doesn't invoke the NMI handler with NMIs enabled.
Cc: Peter Zijlstra <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Access @flags using 32-bit operands when saving and testing @flags for
VMX_RUN_VMRESUME, as using 8-bit operands is unnecessarily fragile due
to relying on VMX_RUN_VMRESUME being in bits 0-7. The behavior of
treating @flags a single byte is a holdover from when the param was
"bool launched", i.e. is not deliberate.
Cc: Alexey Dobriyan <[email protected]>
Reviewed-by: Jim Mattson <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Disable SVM and more importantly force GIF=1 when halting a CPU or
rebooting the machine. Similar to VMX, SVM allows software to block
INITs via CLGI, and thus can be problematic for a crash/reboot. The
window for failure is smaller with SVM as INIT is only blocked while
GIF=0, i.e. between CLGI and STGI, but the window does exist.
Fixes: fba4f472b3 ("x86/reboot: Turn off KVM when halting a CPU")
Cc: [email protected]
Reviewed-by: Thomas Gleixner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Disable SVM on all CPUs via NMI shootdown during an emergency reboot.
Like VMX, SVM can block INIT, e.g. if the emergency reboot is triggered
between CLGI and STGI, and thus can prevent bringing up other CPUs via
INIT-SIPI-SIPI.
Cc: [email protected]
Reviewed-by: Thomas Gleixner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Set GIF=1 prior to disabling SVM to ensure that INIT is recognized if the
kernel is disabling SVM in an emergency, e.g. if the kernel is about to
jump into a crash kernel or may reboot without doing a full CPU RESET.
If GIF is left cleared, the new kernel (or firmware) will be unabled to
awaken APs. Eat faults on STGI (due to EFER.SVME=0) as it's possible
that SVM could be disabled via NMI shootdown between reading EFER.SVME
and executing STGI.
Link: https://lore.kernel.org/all/[email protected]
Cc: [email protected]
Cc: Andrew Cooper <[email protected]>
Cc: Tom Lendacky <[email protected]>
Reviewed-by: Thomas Gleixner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Disable virtualization in crash_nmi_callback() and rework the
emergency_vmx_disable_all() path to do an NMI shootdown if and only if a
shootdown has not already occurred. NMI crash shootdown fundamentally
can't support multiple invocations as responding CPUs are deliberately
put into halt state without unblocking NMIs. But, the emergency reboot
path doesn't have any work of its own, it simply cares about disabling
virtualization, i.e. so long as a shootdown occurred, emergency reboot
doesn't care who initiated the shootdown, or when.
If "crash_kexec_post_notifiers" is specified on the kernel command line,
panic() will invoke crash_smp_send_stop() and result in a second call to
nmi_shootdown_cpus() during native_machine_emergency_restart().
Invoke the callback _before_ disabling virtualization, as the current
VMCS needs to be cleared before doing VMXOFF. Note, this results in a
subtle change in ordering between disabling virtualization and stopping
Intel PT on the responding CPUs. While VMX and Intel PT do interact,
VMXOFF and writes to MSR_IA32_RTIT_CTL do not induce faults between one
another, which is all that matters when panicking.
Harden nmi_shootdown_cpus() against multiple invocations to try and
capture any such kernel bugs via a WARN instead of hanging the system
during a crash/dump, e.g. prior to the recent hardening of
register_nmi_handler(), re-registering the NMI handler would trigger a
double list_add() and hang the system if CONFIG_BUG_ON_DATA_CORRUPTION=y.
list_add double add: new=ffffffff82220800, prev=ffffffff8221cfe8, next=ffffffff82220800.
WARNING: CPU: 2 PID: 1319 at lib/list_debug.c:29 __list_add_valid+0x67/0x70
Call Trace:
__register_nmi_handler+0xcf/0x130
nmi_shootdown_cpus+0x39/0x90
native_machine_emergency_restart+0x1c9/0x1d0
panic+0x237/0x29b
Extract the disabling logic to a common helper to deduplicate code, and
to prepare for doing the shootdown in the emergency reboot path if SVM
is supported.
Note, prior to commit ed72736183 ("x86/reboot: Force all cpus to exit
VMX root if VMX is supported"), nmi_shootdown_cpus() was subtly protected
against a second invocation by a cpu_vmx_enabled() check as the kdump
handler would disable VMX if it ran first.
Fixes: ed72736183 ("x86/reboot: Force all cpus to exit VMX root if VMX is supported")
Cc: [email protected]
Reported-by: Guilherme G. Piccoli <[email protected]>
Cc: Vitaly Kuznetsov <[email protected]>
Cc: Paolo Bonzini <[email protected]>
Link: https://lore.kernel.org/all/[email protected]
Tested-by: Guilherme G. Piccoli <[email protected]>
Reviewed-by: Thomas Gleixner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Intercept reads to invalid (non-existent) and write-only x2APIC registers
when configuring VMX's MSR bitmaps for x2APIC+APICv. When APICv is fully
enabled, Intel hardware doesn't validate the registers on RDMSR and
instead blindly retrieves data from the vAPIC page, i.e. it's software's
responsibility to intercept reads to non-existent and write-only MSRs.
Fixes: 8d14695f95 ("x86, apicv: add virtual x2apic support")
Reviewed-by: Maxim Levitsky <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Don't clear the "read" bits for x2APIC registers above SELF_IPI (APIC regs
0x400 - 0xff0, MSRs 0x840 - 0x8ff). KVM doesn't emulate registers in that
space (there are a smattering of AMD-only extensions) and so should
intercept reads in order to inject #GP. When APICv is fully enabled,
Intel hardware doesn't validate the registers on RDMSR and instead blindly
retrieves data from the vAPIC page, i.e. it's software's responsibility to
intercept reads to non-existent MSRs.
Fixes: 8d14695f95 ("x86, apicv: add virtual x2apic support")
Reviewed-by: Maxim Levitsky <[email protected]>
Reviewed-by: Jim Mattson <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Move the generation of the readable APIC regs bitmask to a standalone
helper so that VMX can use the mask for its MSR interception bitmaps.
No functional change intended.
Reviewed-by: Maxim Levitsky <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Mark APIC_DFR as being invalid/non-existent in x2APIC mode instead of
handling it as a one-off check in kvm_x2apic_msr_read(). This will allow
reusing "valid_reg_mask" to generate VMX's interception bitmaps for
x2APIC. Handling DFR in the common read path may also fix the Hyper-V
PV MSR interface, if that can coexist with x2APIC.
Reviewed-by: Maxim Levitsky <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Reject attempts to set bits 63:32 for 32-bit x2APIC registers, i.e. all
x2APIC registers except ICR. Per Intel's SDM:
Non-zero writes (by WRMSR instruction) to reserved bits to these
registers will raise a general protection fault exception
Opportunistically fix a typo in a nearby comment.
Reported-by: Marc Orr <[email protected]>
Cc: [email protected]
Reviewed-by: Maxim Levitsky <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Fix a build error due to a mixup during a recent refactoring. The error
was reported during code review, but the fixed up patch didn't make it
into the final commit.
Fixes: 474856bad921 ("KVM: PPC: Move processor compatibility check to module init")
Link: https://lore.kernel.org/all/[email protected]
Cc: Michael Ellerman <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Move the guts of kvm_recalculate_apic_map()'s main loop to two separate
helpers to handle recalculating the physical and logical pieces of the
optimized map. Having 100+ lines of code in the for-loop makes it hard
to understand what is being calculated where.
No functional change intended.
Suggested-by: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Track the per-vendor required APICv inhibits with a variable instead of
calling into vendor code every time KVM wants to query the set of
required inhibits. The required inhibits are a property of the vendor's
virtualization architecture, i.e. are 100% static.
Using a variable allows the compiler to inline the check, e.g. generate
a single-uop TEST+Jcc, and thus eliminates any desire to avoid checking
inhibits for performance reasons.
No functional change intended.
Reviewed-by: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Turns out that some warnings exist for good reasons. Restore the warning
in avic_vcpu_load() that guards against calling avic_vcpu_load() on a
running vCPU now that KVM avoids doing so when switching between x2APIC
and xAPIC. The entire point of the WARN is to highlight that KVM should
not be reloading an AVIC.
Opportunistically convert the WARN_ON() to WARN_ON_ONCE() to avoid
spamming the kernel if it does fire.
This reverts commit c0caeee65a.
Reviewed-by: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Drop writes to APIC_RRR, a.k.a. Remote Read Data Register, on AVIC
unaccelerated write traps. The register is read-only and isn't emulated
by KVM. Sending the register through kvm_apic_write_nodecode() will
result in screaming when x2APIC is enabled due to the unexpected failure
to retrieve the MSR (KVM expects that only "legal" accesses will trap).
Fixes: 4d1d7942e3 ("KVM: SVM: Introduce logic to (de)activate x2AVIC mode")
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Maxim Levitsky <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Iterate over all target logical IDs in the AVIC kick fastpath instead of
bailing if there is more than one target. Now that KVM inhibits AVIC if
vCPUs aren't mapped 1:1 with logical IDs, each bit in the destination is
guaranteed to match to at most one vCPU, i.e. iterating over the bitmap
is guaranteed to kick each valid target exactly once.
Reviewed-by: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Sean Christopherson