platform/kernel/linux-rpi.git
3 years agoKVM: nVMX: nSVM: 'nested_run' should count guest-entry attempts that make it to guest...
Krish Sadhukhan [Wed, 9 Jun 2021 18:03:38 +0000 (14:03 -0400)]
KVM: nVMX: nSVM: 'nested_run' should count guest-entry attempts that make it to guest code

Currently, the 'nested_run' statistic counts all guest-entry attempts,
including those that fail during vmentry checks on Intel and during
consistency checks on AMD. Convert this statistic to count only those
guest-entries that make it past these state checks and make it to guest
code. This will tell us the number of guest-entries that actually executed
or tried to execute guest code.

Signed-off-by: Krish Sadhukhan <Krish.Sadhukhan@oracle.com>
Message-Id: <20210609180340.104248-2-krish.sadhukhan@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Drop "pre_" from enter/leave_smm() helpers
Sean Christopherson [Wed, 9 Jun 2021 18:56:19 +0000 (11:56 -0700)]
KVM: x86: Drop "pre_" from enter/leave_smm() helpers

Now that .post_leave_smm() is gone, drop "pre_" from the remaining
helpers.  The helpers aren't invoked purely before SMI/RSM processing,
e.g. both helpers are invoked after state is snapshotted (from regs or
SMRAM), and the RSM helper is invoked after some amount of register state
has been stuffed.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-10-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Drop .post_leave_smm(), i.e. the manual post-RSM MMU reset
Sean Christopherson [Wed, 9 Jun 2021 18:56:18 +0000 (11:56 -0700)]
KVM: x86: Drop .post_leave_smm(), i.e. the manual post-RSM MMU reset

Drop the .post_leave_smm() emulator callback, which at this point is just
a wrapper to kvm_mmu_reset_context().  The manual context reset is
unnecessary, because unlike enter_smm() which calls vendor MSR/CR helpers
directly, em_rsm() bounces through the KVM helpers, e.g. kvm_set_cr4(),
which are responsible for processing side effects.  em_rsm() is already
subtly relying on this behavior as it doesn't manually do
kvm_update_cpuid_runtime(), e.g. to recognize CR4.OSXSAVE changes.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-9-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Rename SMM tracepoint to make it reflect reality
Sean Christopherson [Wed, 9 Jun 2021 18:56:17 +0000 (11:56 -0700)]
KVM: x86: Rename SMM tracepoint to make it reflect reality

Rename the SMM tracepoint, which handles both entering and exiting SMM,
from kvm_enter_smm to kvm_smm_transition.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-8-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Move "entering SMM" tracepoint into kvm_smm_changed()
Sean Christopherson [Wed, 9 Jun 2021 18:56:16 +0000 (11:56 -0700)]
KVM: x86: Move "entering SMM" tracepoint into kvm_smm_changed()

Invoke the "entering SMM" tracepoint from kvm_smm_changed() instead of
enter_smm(), effectively moving it from before reading vCPU state to
after reading state (but still before writing it to SMRAM!).  The primary
motivation is to consolidate code, but calling the tracepoint from
kvm_smm_changed() also makes its invocation consistent with respect to
SMI and RSM, and with respect to KVM_SET_VCPU_EVENTS (which previously
only invoked the tracepoint when forcing the vCPU out of SMM).

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-7-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Move (most) SMM hflags modifications into kvm_smm_changed()
Sean Christopherson [Wed, 9 Jun 2021 18:56:15 +0000 (11:56 -0700)]
KVM: x86: Move (most) SMM hflags modifications into kvm_smm_changed()

Move the core of SMM hflags modifications into kvm_smm_changed() and use
kvm_smm_changed() in enter_smm().  Clear HF_SMM_INSIDE_NMI_MASK for
leaving SMM but do not set it for entering SMM.  If the vCPU is executing
outside of SMM, the flag should unequivocally be cleared, e.g. this
technically fixes a benign bug where the flag could be left set after
KVM_SET_VCPU_EVENTS, but the reverse is not true as NMI blocking depends
on pre-SMM state or userspace input.

Note, this adds an extra kvm_mmu_reset_context() to enter_smm().  The
extra/early reset isn't strictly necessary, and in a way can never be
necessary since the vCPU/MMU context is in a half-baked state until the
final context reset at the end of the function.  But, enter_smm() is not
a hot path, and exploding on an invalid root_hpa is probably better than
having a stale SMM flag in the MMU role; it's at least no worse.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-6-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Invoke kvm_smm_changed() immediately after clearing SMM flag
Sean Christopherson [Wed, 9 Jun 2021 18:56:14 +0000 (11:56 -0700)]
KVM: x86: Invoke kvm_smm_changed() immediately after clearing SMM flag

Move RSM emulation's call to kvm_smm_changed() from .post_leave_smm() to
.exiting_smm(), leaving behind the MMU context reset.  The primary
motivation is to allow for future cleanup, but this also fixes a bug of
sorts by queueing KVM_REQ_EVENT even if RSM causes shutdown, e.g. to let
an INIT wake the vCPU from shutdown.  Of course, KVM doesn't properly
emulate a shutdown state, e.g. KVM doesn't block SMIs after shutdown, and
immediately exits to userspace, so the event request is a moot point in
practice.

Moving kvm_smm_changed() also moves the RSM tracepoint.  This isn't
strictly necessary, but will allow consolidating the SMI and RSM
tracepoints in a future commit (by also moving the SMI tracepoint).
Invoking the tracepoint before loading SMRAM state also means the SMBASE
that reported in the tracepoint will point that the state that will be
used for RSM, as opposed to the SMBASE _after_ RSM completes, which is
arguably a good thing if the tracepoint is being used to debug a RSM/SMM
issue.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-5-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Replace .set_hflags() with dedicated .exiting_smm() helper
Sean Christopherson [Wed, 9 Jun 2021 18:56:13 +0000 (11:56 -0700)]
KVM: x86: Replace .set_hflags() with dedicated .exiting_smm() helper

Replace the .set_hflags() emulator hook with a dedicated .exiting_smm(),
moving the SMM and SMM_INSIDE_NMI flag handling out of the emulator in
the process.  This is a step towards consolidating much of the logic in
kvm_smm_changed(), including the SMM hflags updates.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-4-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Emulate triple fault shutdown if RSM emulation fails
Sean Christopherson [Wed, 9 Jun 2021 18:56:12 +0000 (11:56 -0700)]
KVM: x86: Emulate triple fault shutdown if RSM emulation fails

Use the recently introduced KVM_REQ_TRIPLE_FAULT to properly emulate
shutdown if RSM from SMM fails.

Note, entering shutdown after clearing the SMM flag and restoring NMI
blocking is architecturally correct with respect to AMD's APM, which KVM
also uses for SMRAM layout and RSM NMI blocking behavior.  The APM says:

  An RSM causes a processor shutdown if an invalid-state condition is
  found in the SMRAM state-save area. Only an external reset, external
  processor-initialization, or non-maskable external interrupt (NMI) can
  cause the processor to leave the shutdown state.

Of note is processor-initialization (INIT) as a valid shutdown wake
event, as INIT is blocked by SMM, implying that entering shutdown also
forces the CPU out of SMM.

For recent Intel CPUs, restoring NMI blocking is technically wrong, but
so is restoring NMI blocking in the first place, and Intel's RSM
"architecture" is such a mess that just about anything is allowed and can
be justified as micro-architectural behavior.

Per the SDM:

  On Pentium 4 and later processors, shutdown will inhibit INTR and A20M
  but will not change any of the other inhibits. On these processors,
  NMIs will be inhibited if no action is taken in the SMI handler to
  uninhibit them (see Section 34.8).

where Section 34.8 says:

  When the processor enters SMM while executing an NMI handler, the
  processor saves the SMRAM state save map but does not save the
  attribute to keep NMI interrupts disabled. Potentially, an NMI could be
  latched (while in SMM or upon exit) and serviced upon exit of SMM even
  though the previous NMI handler has still not completed.

I.e. RSM unconditionally unblocks NMI, but shutdown on RSM does not,
which is in direct contradiction of KVM's behavior.  But, as mentioned
above, KVM follows AMD architecture and restores NMI blocking on RSM, so
that micro-architectural detail is already lost.

And for Pentium era CPUs, SMI# can break shutdown, meaning that at least
some Intel CPUs fully leave SMM when entering shutdown:

  In the shutdown state, Intel processors stop executing instructions
  until a RESET#, INIT# or NMI# is asserted.  While Pentium family
  processors recognize the SMI# signal in shutdown state, P6 family and
  Intel486 processors do not.

In other words, the fact that Intel CPUs have implemented the two
extremes gives KVM carte blanche when it comes to honoring Intel's
architecture for handling shutdown during RSM.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-3-seanjc@google.com>
[Return X86EMUL_CONTINUE after triple fault. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Drop vendor specific functions for APICv/AVIC enablement
Vitaly Kuznetsov [Wed, 9 Jun 2021 15:09:09 +0000 (17:09 +0200)]
KVM: x86: Drop vendor specific functions for APICv/AVIC enablement

Now that APICv/AVIC enablement is kept in common 'enable_apicv' variable,
there's no need to call kvm_apicv_init() from vendor specific code.

No functional change intended.

Reviewed-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20210609150911.1471882-3-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Use common 'enable_apicv' variable for both APICv and AVIC
Vitaly Kuznetsov [Wed, 9 Jun 2021 15:09:08 +0000 (17:09 +0200)]
KVM: x86: Use common 'enable_apicv' variable for both APICv and AVIC

Unify VMX and SVM code by moving APICv/AVIC enablement tracking to common
'enable_apicv' variable. Note: unlike APICv, AVIC is disabled by default.

No functional change intended.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20210609150911.1471882-2-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: x86: implement KVM PM-notifier
Sergey Senozhatsky [Sun, 6 Jun 2021 02:10:45 +0000 (11:10 +0900)]
kvm: x86: implement KVM PM-notifier

Implement PM hibernation/suspend prepare notifiers so that KVM
can reliably set PVCLOCK_GUEST_STOPPED on VCPUs and properly
suspend VMs.

Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Message-Id: <20210606021045.14159-2-senozhatsky@chromium.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: add PM-notifier
Sergey Senozhatsky [Sun, 6 Jun 2021 02:10:44 +0000 (11:10 +0900)]
kvm: add PM-notifier

Add KVM PM-notifier so that architectures can have arch-specific
VM suspend/resume routines. Such architectures need to select
CONFIG_HAVE_KVM_PM_NOTIFIER and implement kvm_arch_pm_notifier().

Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Acked-by: Marc Zyngier <maz@kernel.org>
Message-Id: <20210606021045.14159-1-senozhatsky@chromium.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Introduce x2APIC register manipulation functions
Jim Mattson [Fri, 4 Jun 2021 17:26:09 +0000 (10:26 -0700)]
KVM: selftests: Introduce x2APIC register manipulation functions

Standardize reads and writes of the x2APIC MSRs.

Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20210604172611.281819-11-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Hoist APIC functions out of individual tests
Jim Mattson [Fri, 4 Jun 2021 17:26:08 +0000 (10:26 -0700)]
KVM: selftests: Hoist APIC functions out of individual tests

Move the APIC functions into the library to encourage code reuse and
to avoid unintended deviations.

Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20210604172611.281819-10-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Move APIC definitions into a separate file
Jim Mattson [Fri, 4 Jun 2021 17:26:07 +0000 (10:26 -0700)]
KVM: selftests: Move APIC definitions into a separate file

Processor.h is a hodgepodge of definitions. Though the local APIC is
technically built into the CPU these days, move the APIC definitions
into a new header file: apic.h.

Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20210604172611.281819-9-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: nVMX: Disable vmcs02 posted interrupts if vmcs12 PID isn't mappable
Jim Mattson [Fri, 4 Jun 2021 17:26:06 +0000 (10:26 -0700)]
KVM: nVMX: Disable vmcs02 posted interrupts if vmcs12 PID isn't mappable

Don't allow posted interrupts to modify a stale posted interrupt
descriptor (including the initial value of 0).

Empirical tests on real hardware reveal that a posted interrupt
descriptor referencing an unbacked address has PCI bus error semantics
(reads as all 1's; writes are ignored). However, kvm can't distinguish
unbacked addresses from device-backed (MMIO) addresses, so it should
really ask userspace for an MMIO completion. That's overly
complicated, so just punt with KVM_INTERNAL_ERROR.

Don't return the error until the posted interrupt descriptor is
actually accessed. We don't want to break the existing kvm-unit-tests
that assume they can launch an L2 VM with a posted interrupt
descriptor that references MMIO space in L1.

Fixes: 6beb7bd52e48 ("kvm: nVMX: Refactor nested_get_vmcs12_pages()")
Signed-off-by: Jim Mattson <jmattson@google.com>
Message-Id: <20210604172611.281819-8-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: nVMX: Fail on MMIO completion for nested posted interrupts
Jim Mattson [Fri, 4 Jun 2021 17:26:05 +0000 (10:26 -0700)]
KVM: nVMX: Fail on MMIO completion for nested posted interrupts

When the kernel has no mapping for the vmcs02 virtual APIC page,
userspace MMIO completion is necessary to process nested posted
interrupts. This is not a configuration that KVM supports. Rather than
silently ignoring the problem, try to exit to userspace with
KVM_INTERNAL_ERROR.

Note that the event that triggers this error is consumed as a
side-effect of a call to kvm_check_nested_events. On some paths
(notably through kvm_vcpu_check_block), the error is dropped. In any
case, this is an incremental improvement over always ignoring the
error.

Signed-off-by: Jim Mattson <jmattson@google.com>
Message-Id: <20210604172611.281819-7-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Add a return code to kvm_apic_accept_events
Jim Mattson [Fri, 4 Jun 2021 17:26:04 +0000 (10:26 -0700)]
KVM: x86: Add a return code to kvm_apic_accept_events

No functional change intended. At present, the only negative value
returned by kvm_check_nested_events is -EBUSY.

Signed-off-by: Jim Mattson <jmattson@google.com>
Message-Id: <20210604172611.281819-6-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Add a return code to inject_pending_event
Jim Mattson [Fri, 4 Jun 2021 17:26:03 +0000 (10:26 -0700)]
KVM: x86: Add a return code to inject_pending_event

No functional change intended. At present, 'r' will always be -EBUSY
on a control transfer to the 'out' label.

Signed-off-by: Jim Mattson <jmattson@google.com>
Message-Id: <20210604172611.281819-5-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: nVMX: Add a return code to vmx_complete_nested_posted_interrupt
Jim Mattson [Fri, 4 Jun 2021 17:26:02 +0000 (10:26 -0700)]
KVM: nVMX: Add a return code to vmx_complete_nested_posted_interrupt

No functional change intended.

Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20210604172611.281819-4-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Remove guest mode check from kvm_check_nested_events
Jim Mattson [Fri, 4 Jun 2021 17:26:00 +0000 (10:26 -0700)]
KVM: x86: Remove guest mode check from kvm_check_nested_events

A survey of the callsites reveals that they all ensure the vCPU is in
guest mode before calling kvm_check_nested_events. Remove this dead
code so that the only negative value this function returns (at the
moment) is -EBUSY.

Signed-off-by: Jim Mattson <jmattson@google.com>
Message-Id: <20210604172611.281819-2-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: x86: Add vmx_nested_tsc_scaling_test
Ilias Stamatis [Wed, 26 May 2021 18:44:18 +0000 (19:44 +0100)]
KVM: selftests: x86: Add vmx_nested_tsc_scaling_test

Test that nested TSC scaling works as expected with both L1 and L2
scaled.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-12-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: nVMX: Enable nested TSC scaling
Ilias Stamatis [Wed, 26 May 2021 18:44:17 +0000 (19:44 +0100)]
KVM: nVMX: Enable nested TSC scaling

Calculate the TSC offset and multiplier on nested transitions and expose
the TSC scaling feature to L1.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-11-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Add vendor callbacks for writing the TSC multiplier
Ilias Stamatis [Mon, 7 Jun 2021 10:54:38 +0000 (11:54 +0100)]
KVM: X86: Add vendor callbacks for writing the TSC multiplier

Currently vmx_vcpu_load_vmcs() writes the TSC_MULTIPLIER field of the
VMCS every time the VMCS is loaded. Instead of doing this, set this
field from common code on initialization and whenever the scaling ratio
changes.

Additionally remove vmx->current_tsc_ratio. This field is redundant as
vcpu->arch.tsc_scaling_ratio already tracks the current TSC scaling
ratio. The vmx->current_tsc_ratio field is only used for avoiding
unnecessary writes but it is no longer needed after removing the code
from the VMCS load path.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Message-Id: <20210607105438.16541-1-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Move write_l1_tsc_offset() logic to common code and rename it
Ilias Stamatis [Wed, 26 May 2021 18:44:15 +0000 (19:44 +0100)]
KVM: X86: Move write_l1_tsc_offset() logic to common code and rename it

The write_l1_tsc_offset() callback has a misleading name. It does not
set L1's TSC offset, it rather updates the current TSC offset which
might be different if a nested guest is executing. Additionally, both
the vmx and svm implementations use the same logic for calculating the
current TSC before writing it to hardware.

Rename the function and move the common logic to the caller. The vmx/svm
specific code now merely sets the given offset to the corresponding
hardware structure.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-9-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Add functions that calculate the nested TSC fields
Ilias Stamatis [Wed, 26 May 2021 18:44:14 +0000 (19:44 +0100)]
KVM: X86: Add functions that calculate the nested TSC fields

When L2 is entered we need to "merge" the TSC multiplier and TSC offset
values of 01 and 12 together.

The merging is done using the following equations:
  offset_02 = ((offset_01 * mult_12) >> shift_bits) + offset_12
  mult_02 = (mult_01 * mult_12) >> shift_bits

Where shift_bits is kvm_tsc_scaling_ratio_frac_bits.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-8-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Add functions for retrieving L2 TSC fields from common code
Ilias Stamatis [Wed, 26 May 2021 18:44:13 +0000 (19:44 +0100)]
KVM: X86: Add functions for retrieving L2 TSC fields from common code

In order to implement as much of the nested TSC scaling logic as
possible in common code, we need these vendor callbacks for retrieving
the TSC offset and the TSC multiplier that L1 has set for L2.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-7-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: nVMX: Add a TSC multiplier field in VMCS12
Ilias Stamatis [Wed, 26 May 2021 18:44:12 +0000 (19:44 +0100)]
KVM: nVMX: Add a TSC multiplier field in VMCS12

This is required for supporting nested TSC scaling.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-6-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Add a ratio parameter to kvm_scale_tsc()
Ilias Stamatis [Wed, 26 May 2021 18:44:11 +0000 (19:44 +0100)]
KVM: X86: Add a ratio parameter to kvm_scale_tsc()

Sometimes kvm_scale_tsc() needs to use the current scaling ratio and
other times (like when reading the TSC from user space) it needs to use
L1's scaling ratio. Have the caller specify this by passing the ratio as
a parameter.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-5-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Rename kvm_compute_tsc_offset() to kvm_compute_l1_tsc_offset()
Ilias Stamatis [Wed, 26 May 2021 18:44:10 +0000 (19:44 +0100)]
KVM: X86: Rename kvm_compute_tsc_offset() to kvm_compute_l1_tsc_offset()

All existing code uses kvm_compute_tsc_offset() passing L1 TSC values to
it. Let's document this by renaming it to kvm_compute_l1_tsc_offset().

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-4-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Store L1's TSC scaling ratio in 'struct kvm_vcpu_arch'
Ilias Stamatis [Wed, 26 May 2021 18:44:09 +0000 (19:44 +0100)]
KVM: X86: Store L1's TSC scaling ratio in 'struct kvm_vcpu_arch'

Store L1's scaling ratio in the kvm_vcpu_arch struct like we already do
for L1's TSC offset. This allows for easy save/restore when we enter and
then exit the nested guest.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-3-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agomath64.h: Add mul_s64_u64_shr()
Ilias Stamatis [Wed, 26 May 2021 18:44:08 +0000 (19:44 +0100)]
math64.h: Add mul_s64_u64_shr()

This function is needed for KVM's nested virtualization. The nested TSC
scaling implementation requires multiplying the signed TSC offset with
the unsigned TSC multiplier.

Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210526184418.28881-2-ilstam@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Lazily allocate memslot rmaps
Ben Gardon [Tue, 18 May 2021 17:34:14 +0000 (10:34 -0700)]
KVM: x86/mmu: Lazily allocate memslot rmaps

If the TDP MMU is in use, wait to allocate the rmaps until the shadow
MMU is actually used. (i.e. a nested VM is launched.) This saves memory
equal to 0.2% of guest memory in cases where the TDP MMU is used and
there are no nested guests involved.

Signed-off-by: Ben Gardon <bgardon@google.com>
Message-Id: <20210518173414.450044-8-bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Skip rmap operations if rmaps not allocated
Ben Gardon [Tue, 18 May 2021 17:34:13 +0000 (10:34 -0700)]
KVM: x86/mmu: Skip rmap operations if rmaps not allocated

If only the TDP MMU is being used to manage the memory mappings for a VM,
then many rmap operations can be skipped as they are guaranteed to be
no-ops. This saves some time which would be spent on the rmap operation.
It also avoids acquiring the MMU lock in write mode for many operations.

This makes it safe to run the VM without rmaps allocated, when only
using the TDP MMU and sets the stage for waiting to allocate the rmaps
until they're needed.

Signed-off-by: Ben Gardon <bgardon@google.com>
Message-Id: <20210518173414.450044-7-bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Add a field to control memslot rmap allocation
Ben Gardon [Tue, 18 May 2021 17:34:12 +0000 (10:34 -0700)]
KVM: x86/mmu: Add a field to control memslot rmap allocation

Add a field to control whether new memslots should have rmaps allocated
for them. As of this change, it's not safe to skip allocating rmaps, so
the field is always set to allocate rmaps. Future changes will make it
safe to operate without rmaps, using the TDP MMU. Then further changes
will allow the rmaps to be allocated lazily when needed for nested
oprtation.

No functional change expected.

Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Ben Gardon <bgardon@google.com>
Message-Id: <20210518173414.450044-6-bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: mmu: Add slots_arch_lock for memslot arch fields
Ben Gardon [Tue, 18 May 2021 17:34:11 +0000 (10:34 -0700)]
KVM: mmu: Add slots_arch_lock for memslot arch fields

Add a new lock to protect the arch-specific fields of memslots if they
need to be modified in a kvm->srcu read critical section. A future
commit will use this lock to lazily allocate memslot rmaps for x86.

Signed-off-by: Ben Gardon <bgardon@google.com>
Message-Id: <20210518173414.450044-5-bgardon@google.com>
[Add Documentation/ hunk. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: mmu: Refactor memslot copy
Ben Gardon [Tue, 18 May 2021 17:34:10 +0000 (10:34 -0700)]
KVM: mmu: Refactor memslot copy

Factor out copying kvm_memslots from allocating the memory for new ones
in preparation for adding a new lock to protect the arch-specific fields
of the memslots.

No functional change intended.

Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Ben Gardon <bgardon@google.com>
Message-Id: <20210518173414.450044-4-bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Factor out allocating memslot rmap
Ben Gardon [Tue, 18 May 2021 17:34:09 +0000 (10:34 -0700)]
KVM: x86/mmu: Factor out allocating memslot rmap

Small refactor to facilitate allocating rmaps for all memslots at once.

No functional change expected.

Signed-off-by: Ben Gardon <bgardon@google.com>
Message-Id: <20210518173414.450044-3-bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Deduplicate rmap freeing
Ben Gardon [Tue, 18 May 2021 17:34:08 +0000 (10:34 -0700)]
KVM: x86/mmu: Deduplicate rmap freeing

Small code deduplication. No functional change expected.

Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Ben Gardon <bgardon@google.com>
Message-Id: <20210518173414.450044-2-bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Do not write protect huge page in initially-all-set mode
Keqian Zhu [Thu, 29 Apr 2021 03:41:15 +0000 (11:41 +0800)]
KVM: x86: Do not write protect huge page in initially-all-set mode

Currently, when dirty logging is started in initially-all-set mode,
we write protect huge pages to prepare for splitting them into
4K pages, and leave normal pages untouched as the logging will
be enabled lazily as dirty bits are cleared.

However, enabling dirty logging lazily is also feasible for huge pages.
This not only reduces the time of start dirty logging, but it also
greatly reduces side-effect on guest when there is high dirty rate.

Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Message-Id: <20210429034115.35560-3-zhukeqian1@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Support write protecting only large pages
Keqian Zhu [Thu, 29 Apr 2021 03:41:14 +0000 (11:41 +0800)]
KVM: x86: Support write protecting only large pages

Prepare for write protecting large page lazily during dirty log tracking,
for which we will only need to write protect gfns at large page
granularity.

No functional or performance change expected.

Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Message-Id: <20210429034115.35560-2-zhukeqian1@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: hyper-v: Advertise support for fast XMM hypercalls
Siddharth Chandrasekaran [Wed, 26 May 2021 09:03:56 +0000 (11:03 +0200)]
KVM: hyper-v: Advertise support for fast XMM hypercalls

Now that kvm_hv_flush_tlb() has been patched to support XMM hypercall
inputs, we can start advertising this feature to guests.

Cc: Alexander Graf <graf@amazon.com>
Cc: Evgeny Iakovlev <eyakovl@amazon.de>
Signed-off-by: Siddharth Chandrasekaran <sidcha@amazon.de>
Message-Id: <e63fc1c61dd2efecbefef239f4f0a598bd552750.1622019134.git.sidcha@amazon.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: kvm_hv_flush_tlb use inputs from XMM registers
Siddharth Chandrasekaran [Wed, 26 May 2021 08:56:10 +0000 (10:56 +0200)]
KVM: x86: kvm_hv_flush_tlb use inputs from XMM registers

Hyper-V supports the use of XMM registers to perform fast hypercalls.
This allows guests to take advantage of the improved performance of the
fast hypercall interface even though a hypercall may require more than
(the current maximum of) two input registers.

The XMM fast hypercall interface uses six additional XMM registers (XMM0
to XMM5) to allow the guest to pass an input parameter block of up to
112 bytes.

Add framework to read from XMM registers in kvm_hv_hypercall() and use
the additional hypercall inputs from XMM registers in kvm_hv_flush_tlb()
when possible.

Cc: Alexander Graf <graf@amazon.com>
Co-developed-by: Evgeny Iakovlev <eyakovl@amazon.de>
Signed-off-by: Evgeny Iakovlev <eyakovl@amazon.de>
Signed-off-by: Siddharth Chandrasekaran <sidcha@amazon.de>
Message-Id: <fc62edad33f1920fe5c74dde47d7d0b4275a9012.1622019134.git.sidcha@amazon.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: hyper-v: Collect hypercall params into struct
Siddharth Chandrasekaran [Wed, 26 May 2021 08:56:09 +0000 (10:56 +0200)]
KVM: hyper-v: Collect hypercall params into struct

As of now there are 7 parameters (and flags) that are used in various
hyper-v hypercall handlers. There are 6 more input/output parameters
passed from XMM registers which are to be added in an upcoming patch.

To make passing arguments to the handlers more readable, capture all
these parameters into a single structure.

Cc: Alexander Graf <graf@amazon.com>
Cc: Evgeny Iakovlev <eyakovl@amazon.de>
Signed-off-by: Siddharth Chandrasekaran <sidcha@amazon.de>
Message-Id: <273f7ed510a1f6ba177e61b73a5c7bfbee4a4a87.1622019133.git.sidcha@amazon.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Move FPU register accessors into fpu.h
Siddharth Chandrasekaran [Wed, 26 May 2021 08:56:08 +0000 (10:56 +0200)]
KVM: x86: Move FPU register accessors into fpu.h

Hyper-v XMM fast hypercalls use XMM registers to pass input/output
parameters. To access these, hyperv.c can reuse some FPU register
accessors defined in emulator.c. Move them to a common location so both
can access them.

While at it, reorder the parameters of these accessor methods to make
them more readable.

Cc: Alexander Graf <graf@amazon.com>
Cc: Evgeny Iakovlev <eyakovl@amazon.de>
Signed-off-by: Siddharth Chandrasekaran <sidcha@amazon.de>
Message-Id: <01a85a6560714d4d3637d3d86e5eba65073318fa.1622019133.git.sidcha@amazon.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Make is_nx_huge_page_enabled an inline function
Shaokun Zhang [Thu, 27 May 2021 07:57:51 +0000 (15:57 +0800)]
KVM: x86/mmu: Make is_nx_huge_page_enabled an inline function

Function 'is_nx_huge_page_enabled' is called only by kvm/mmu, so make
it as inline fucntion and remove the unnecessary declaration.

Cc: Ben Gardon <bgardon@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Sean Christopherson <seanjc@google.com>
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Shaokun Zhang <zhangshaokun@hisilicon.com>
Message-Id: <1622102271-63107-1-git-send-email-zhangshaokun@hisilicon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Fix kvm_check_cap() assertion
Fuad Tabba [Tue, 15 Jun 2021 15:04:43 +0000 (16:04 +0100)]
KVM: selftests: Fix kvm_check_cap() assertion

KVM_CHECK_EXTENSION ioctl can return any negative value on error,
and not necessarily -1. Change the assertion to reflect that.

Signed-off-by: Fuad Tabba <tabba@google.com>
Message-Id: <20210615150443.1183365-1-tabba@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Calculate and check "full" mmu_role for nested MMU
Sean Christopherson [Thu, 10 Jun 2021 22:00:26 +0000 (15:00 -0700)]
KVM: x86/mmu: Calculate and check "full" mmu_role for nested MMU

Calculate and check the full mmu_role when initializing the MMU context
for the nested MMU, where "full" means the bits and pieces of the role
that aren't handled by kvm_calc_mmu_role_common().  While the nested MMU
isn't used for shadow paging, things like the number of levels in the
guest's page tables are surprisingly important when walking the guest
page tables.  Failure to reinitialize the nested MMU context if L2's
paging mode changes can result in unexpected and/or missed page faults,
and likely other explosions.

E.g. if an L1 vCPU is running both a 32-bit PAE L2 and a 64-bit L2, the
"common" role calculation will yield the same role for both L2s.  If the
64-bit L2 is run after the 32-bit PAE L2, L0 will fail to reinitialize
the nested MMU context, ultimately resulting in a bad walk of L2's page
tables as the MMU will still have a guest root_level of PT32E_ROOT_LEVEL.

  WARNING: CPU: 4 PID: 167334 at arch/x86/kvm/vmx/vmx.c:3075 ept_save_pdptrs+0x15/0xe0 [kvm_intel]
  Modules linked in: kvm_intel]
  CPU: 4 PID: 167334 Comm: CPU 3/KVM Not tainted 5.13.0-rc1-d849817d5673-reqs #185
  Hardware name: ASUS Q87M-E/Q87M-E, BIOS 1102 03/03/2014
  RIP: 0010:ept_save_pdptrs+0x15/0xe0 [kvm_intel]
  Code: <0f> 0b c3 f6 87 d8 02 00f
  RSP: 0018:ffffbba702dbba00 EFLAGS: 00010202
  RAX: 0000000000000011 RBX: 0000000000000002 RCX: ffffffff810a2c08
  RDX: ffff91d7bc30acc0 RSI: 0000000000000011 RDI: ffff91d7bc30a600
  RBP: ffff91d7bc30a600 R08: 0000000000000010 R09: 0000000000000007
  R10: 0000000000000000 R11: 0000000000000000 R12: ffff91d7bc30a600
  R13: ffff91d7bc30acc0 R14: ffff91d67c123460 R15: 0000000115d7e005
  FS:  00007fe8e9ffb700(0000) GS:ffff91d90fb00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 0000000000000000 CR3: 000000029f15a001 CR4: 00000000001726e0
  Call Trace:
   kvm_pdptr_read+0x3a/0x40 [kvm]
   paging64_walk_addr_generic+0x327/0x6a0 [kvm]
   paging64_gva_to_gpa_nested+0x3f/0xb0 [kvm]
   kvm_fetch_guest_virt+0x4c/0xb0 [kvm]
   __do_insn_fetch_bytes+0x11a/0x1f0 [kvm]
   x86_decode_insn+0x787/0x1490 [kvm]
   x86_decode_emulated_instruction+0x58/0x1e0 [kvm]
   x86_emulate_instruction+0x122/0x4f0 [kvm]
   vmx_handle_exit+0x120/0x660 [kvm_intel]
   kvm_arch_vcpu_ioctl_run+0xe25/0x1cb0 [kvm]
   kvm_vcpu_ioctl+0x211/0x5a0 [kvm]
   __x64_sys_ioctl+0x83/0xb0
   do_syscall_64+0x40/0xb0
   entry_SYSCALL_64_after_hwframe+0x44/0xae

Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: stable@vger.kernel.org
Fixes: bf627a928837 ("x86/kvm/mmu: check if MMU reconfiguration is needed in init_kvm_nested_mmu()")
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210610220026.1364486-1-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Fix x86_emulator slab cache leak
Wanpeng Li [Fri, 11 Jun 2021 04:59:33 +0000 (21:59 -0700)]
KVM: X86: Fix x86_emulator slab cache leak

Commit c9b8b07cded58 (KVM: x86: Dynamically allocate per-vCPU emulation context)
tries to allocate per-vCPU emulation context dynamically, however, the
x86_emulator slab cache is still exiting after the kvm module is unload
as below after destroying the VM and unloading the kvm module.

grep x86_emulator /proc/slabinfo
x86_emulator          36     36   2672   12    8 : tunables    0    0    0 : slabdata      3      3      0

This patch fixes this slab cache leak by destroying the x86_emulator slab cache
when the kvm module is unloaded.

Fixes: c9b8b07cded58 (KVM: x86: Dynamically allocate per-vCPU emulation context)
Cc: stable@vger.kernel.org
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1623387573-5969-1-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: SVM: Call SEV Guest Decommission if ASID binding fails
Alper Gun [Thu, 10 Jun 2021 17:46:04 +0000 (17:46 +0000)]
KVM: SVM: Call SEV Guest Decommission if ASID binding fails

Send SEV_CMD_DECOMMISSION command to PSP firmware if ASID binding
fails. If a failure happens after  a successful LAUNCH_START command,
a decommission command should be executed. Otherwise, guest context
will be unfreed inside the AMD SP. After the firmware will not have
memory to allocate more SEV guest context, LAUNCH_START command will
begin to fail with SEV_RET_RESOURCE_LIMIT error.

The existing code calls decommission inside sev_unbind_asid, but it is
not called if a failure happens before guest activation succeeds. If
sev_bind_asid fails, decommission is never called. PSP firmware has a
limit for the number of guests. If sev_asid_binding fails many times,
PSP firmware will not have resources to create another guest context.

Cc: stable@vger.kernel.org
Fixes: 59414c989220 ("KVM: SVM: Add support for KVM_SEV_LAUNCH_START command")
Reported-by: Peter Gonda <pgonda@google.com>
Signed-off-by: Alper Gun <alpergun@google.com>
Reviewed-by: Marc Orr <marcorr@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210610174604.2554090-1-alpergun@google.com>

3 years agoKVM: x86: Immediately reset the MMU context when the SMM flag is cleared
Sean Christopherson [Wed, 9 Jun 2021 18:56:11 +0000 (11:56 -0700)]
KVM: x86: Immediately reset the MMU context when the SMM flag is cleared

Immediately reset the MMU context when the vCPU's SMM flag is cleared so
that the SMM flag in the MMU role is always synchronized with the vCPU's
flag.  If RSM fails (which isn't correctly emulated), KVM will bail
without calling post_leave_smm() and leave the MMU in a bad state.

The bad MMU role can lead to a NULL pointer dereference when grabbing a
shadow page's rmap for a page fault as the initial lookups for the gfn
will happen with the vCPU's SMM flag (=0), whereas the rmap lookup will
use the shadow page's SMM flag, which comes from the MMU (=1).  SMM has
an entirely different set of memslots, and so the initial lookup can find
a memslot (SMM=0) and then explode on the rmap memslot lookup (SMM=1).

  general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] PREEMPT SMP KASAN
  KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
  CPU: 1 PID: 8410 Comm: syz-executor382 Not tainted 5.13.0-rc5-syzkaller #0
  Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
  RIP: 0010:__gfn_to_rmap arch/x86/kvm/mmu/mmu.c:935 [inline]
  RIP: 0010:gfn_to_rmap+0x2b0/0x4d0 arch/x86/kvm/mmu/mmu.c:947
  Code: <42> 80 3c 20 00 74 08 4c 89 ff e8 f1 79 a9 00 4c 89 fb 4d 8b 37 44
  RSP: 0018:ffffc90000ffef98 EFLAGS: 00010246
  RAX: 0000000000000000 RBX: ffff888015b9f414 RCX: ffff888019669c40
  RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000001
  RBP: 0000000000000001 R08: ffffffff811d9cdb R09: ffffed10065a6002
  R10: ffffed10065a6002 R11: 0000000000000000 R12: dffffc0000000000
  R13: 0000000000000003 R14: 0000000000000001 R15: 0000000000000000
  FS:  000000000124b300(0000) GS:ffff8880b9b00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 0000000000000000 CR3: 0000000028e31000 CR4: 00000000001526e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Call Trace:
   rmap_add arch/x86/kvm/mmu/mmu.c:965 [inline]
   mmu_set_spte+0x862/0xe60 arch/x86/kvm/mmu/mmu.c:2604
   __direct_map arch/x86/kvm/mmu/mmu.c:2862 [inline]
   direct_page_fault+0x1f74/0x2b70 arch/x86/kvm/mmu/mmu.c:3769
   kvm_mmu_do_page_fault arch/x86/kvm/mmu.h:124 [inline]
   kvm_mmu_page_fault+0x199/0x1440 arch/x86/kvm/mmu/mmu.c:5065
   vmx_handle_exit+0x26/0x160 arch/x86/kvm/vmx/vmx.c:6122
   vcpu_enter_guest+0x3bdd/0x9630 arch/x86/kvm/x86.c:9428
   vcpu_run+0x416/0xc20 arch/x86/kvm/x86.c:9494
   kvm_arch_vcpu_ioctl_run+0x4e8/0xa40 arch/x86/kvm/x86.c:9722
   kvm_vcpu_ioctl+0x70f/0xbb0 arch/x86/kvm/../../../virt/kvm/kvm_main.c:3460
   vfs_ioctl fs/ioctl.c:51 [inline]
   __do_sys_ioctl fs/ioctl.c:1069 [inline]
   __se_sys_ioctl+0xfb/0x170 fs/ioctl.c:1055
   do_syscall_64+0x3f/0xb0 arch/x86/entry/common.c:47
   entry_SYSCALL_64_after_hwframe+0x44/0xae
  RIP: 0033:0x440ce9

Cc: stable@vger.kernel.org
Reported-by: syzbot+fb0b6a7e8713aeb0319c@syzkaller.appspotmail.com
Fixes: 9ec19493fb86 ("KVM: x86: clear SMM flags before loading state while leaving SMM")
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210609185619.992058-2-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Fix fall-through warnings for Clang
Gustavo A. R. Silva [Fri, 28 May 2021 20:07:56 +0000 (15:07 -0500)]
KVM: x86: Fix fall-through warnings for Clang

In preparation to enable -Wimplicit-fallthrough for Clang, fix a couple
of warnings by explicitly adding break statements instead of just letting
the code fall through to the next case.

Link: https://github.com/KSPP/linux/issues/115
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Message-Id: <20210528200756.GA39320@embeddedor>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: SVM: fix doc warnings
ChenXiaoSong [Wed, 9 Jun 2021 12:22:17 +0000 (20:22 +0800)]
KVM: SVM: fix doc warnings

Fix kernel-doc warnings:

arch/x86/kvm/svm/avic.c:233: warning: Function parameter or member 'activate' not described in 'avic_update_access_page'
arch/x86/kvm/svm/avic.c:233: warning: Function parameter or member 'kvm' not described in 'avic_update_access_page'
arch/x86/kvm/svm/avic.c:781: warning: Function parameter or member 'e' not described in 'get_pi_vcpu_info'
arch/x86/kvm/svm/avic.c:781: warning: Function parameter or member 'kvm' not described in 'get_pi_vcpu_info'
arch/x86/kvm/svm/avic.c:781: warning: Function parameter or member 'svm' not described in 'get_pi_vcpu_info'
arch/x86/kvm/svm/avic.c:781: warning: Function parameter or member 'vcpu_info' not described in 'get_pi_vcpu_info'
arch/x86/kvm/svm/avic.c:1009: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst

Signed-off-by: ChenXiaoSong <chenxiaosong2@huawei.com>
Message-Id: <20210609122217.2967131-1-chenxiaosong2@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Fix compiling errors when initializing the static structure
Yanan Wang [Thu, 10 Jun 2021 08:54:18 +0000 (16:54 +0800)]
KVM: selftests: Fix compiling errors when initializing the static structure

Errors like below were produced from test_util.c when compiling the KVM
selftests on my local platform.

lib/test_util.c: In function 'vm_mem_backing_src_alias':
lib/test_util.c:177:12: error: initializer element is not constant
    .flag = anon_flags,
            ^~~~~~~~~~
lib/test_util.c:177:12: note: (near initialization for 'aliases[0].flag')

The reason is that we are using non-const expressions to initialize the
static structure, which will probably trigger a compiling error/warning
on stricter GCC versions. Fix it by converting the two const variables
"anon_flags" and "anon_huge_flags" into more stable macros.

Fixes: b3784bc28ccc0 ("KVM: selftests: refactor vm_mem_backing_src_type flags")
Reported-by: Zenghui Yu <yuzenghui@huawei.com>
Signed-off-by: Yanan Wang <wangyanan55@huawei.com>
Message-Id: <20210610085418.35544-1-wangyanan55@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: LAPIC: Restore guard to prevent illegal APIC register access
Jim Mattson [Wed, 2 Jun 2021 20:52:24 +0000 (13:52 -0700)]
kvm: LAPIC: Restore guard to prevent illegal APIC register access

Per the SDM, "any access that touches bytes 4 through 15 of an APIC
register may cause undefined behavior and must not be executed."
Worse, such an access in kvm_lapic_reg_read can result in a leak of
kernel stack contents. Prior to commit 01402cf81051 ("kvm: LAPIC:
write down valid APIC registers"), such an access was explicitly
disallowed. Restore the guard that was removed in that commit.

Fixes: 01402cf81051 ("kvm: LAPIC: write down valid APIC registers")
Signed-off-by: Jim Mattson <jmattson@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Message-Id: <20210602205224.3189316-1-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: fix previous commit for 32-bit builds
Paolo Bonzini [Wed, 9 Jun 2021 05:49:13 +0000 (01:49 -0400)]
kvm: fix previous commit for 32-bit builds

array_index_nospec does not work for uint64_t on 32-bit builds.
However, the size of a memory slot must be less than 20 bits wide
on those system, since the memory slot must fit in the user
address space.  So just store it in an unsigned long.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: avoid speculation-based attacks from out-of-range memslot accesses
Paolo Bonzini [Tue, 8 Jun 2021 19:31:42 +0000 (15:31 -0400)]
kvm: avoid speculation-based attacks from out-of-range memslot accesses

KVM's mechanism for accessing guest memory translates a guest physical
address (gpa) to a host virtual address using the right-shifted gpa
(also known as gfn) and a struct kvm_memory_slot.  The translation is
performed in __gfn_to_hva_memslot using the following formula:

      hva = slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE

It is expected that gfn falls within the boundaries of the guest's
physical memory.  However, a guest can access invalid physical addresses
in such a way that the gfn is invalid.

__gfn_to_hva_memslot is called from kvm_vcpu_gfn_to_hva_prot, which first
retrieves a memslot through __gfn_to_memslot.  While __gfn_to_memslot
does check that the gfn falls within the boundaries of the guest's
physical memory or not, a CPU can speculate the result of the check and
continue execution speculatively using an illegal gfn. The speculation
can result in calculating an out-of-bounds hva.  If the resulting host
virtual address is used to load another guest physical address, this
is effectively a Spectre gadget consisting of two consecutive reads,
the second of which is data dependent on the first.

Right now it's not clear if there are any cases in which this is
exploitable.  One interesting case was reported by the original author
of this patch, and involves visiting guest page tables on x86.  Right
now these are not vulnerable because the hva read goes through get_user(),
which contains an LFENCE speculation barrier.  However, there are
patches in progress for x86 uaccess.h to mask kernel addresses instead of
using LFENCE; once these land, a guest could use speculation to read
from the VMM's ring 3 address space.  Other architectures such as ARM
already use the address masking method, and would be susceptible to
this same kind of data-dependent access gadgets.  Therefore, this patch
proactively protects from these attacks by masking out-of-bounds gfns
in __gfn_to_hva_memslot, which blocks speculation of invalid hvas.

Sean Christopherson noted that this patch does not cover
kvm_read_guest_offset_cached.  This however is limited to a few bytes
past the end of the cache, and therefore it is unlikely to be useful in
the context of building a chain of data dependent accesses.

Reported-by: Artemiy Margaritov <artemiy.margaritov@gmail.com>
Co-developed-by: Artemiy Margaritov <artemiy.margaritov@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Unload MMU on guest TLB flush if TDP disabled to force MMU sync
Lai Jiangshan [Mon, 31 May 2021 17:22:56 +0000 (01:22 +0800)]
KVM: x86: Unload MMU on guest TLB flush if TDP disabled to force MMU sync

When using shadow paging, unload the guest MMU when emulating a guest TLB
flush to ensure all roots are synchronized.  From the guest's perspective,
flushing the TLB ensures any and all modifications to its PTEs will be
recognized by the CPU.

Note, unloading the MMU is overkill, but is done to mirror KVM's existing
handling of INVPCID(all) and ensure the bug is squashed.  Future cleanup
can be done to more precisely synchronize roots when servicing a guest
TLB flush.

If TDP is enabled, synchronizing the MMU is unnecessary even if nested
TDP is in play, as a "legacy" TLB flush from L1 does not invalidate L1's
TDP mappings.  For EPT, an explicit INVEPT is required to invalidate
guest-physical mappings; for NPT, guest mappings are always tagged with
an ASID and thus can only be invalidated via the VMCB's ASID control.

This bug has existed since the introduction of KVM_VCPU_FLUSH_TLB.
It was only recently exposed after Linux guests stopped flushing the
local CPU's TLB prior to flushing remote TLBs (see commit 4ce94eabac16,
"x86/mm/tlb: Flush remote and local TLBs concurrently"), but is also
visible in Windows 10 guests.

Tested-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Fixes: f38a7b75267f ("KVM: X86: support paravirtualized help for TLB shootdowns")
Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
[sean: massaged comment and changelog]
Message-Id: <20210531172256.2908-1-jiangshanlai@gmail.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Ensure liveliness of nested VM-Enter fail tracepoint message
Sean Christopherson [Mon, 7 Jun 2021 17:57:48 +0000 (10:57 -0700)]
KVM: x86: Ensure liveliness of nested VM-Enter fail tracepoint message

Use the __string() machinery provided by the tracing subystem to make a
copy of the string literals consumed by the "nested VM-Enter failed"
tracepoint.  A complete copy is necessary to ensure that the tracepoint
can't outlive the data/memory it consumes and deference stale memory.

Because the tracepoint itself is defined by kvm, if kvm-intel and/or
kvm-amd are built as modules, the memory holding the string literals
defined by the vendor modules will be freed when the module is unloaded,
whereas the tracepoint and its data in the ring buffer will live until
kvm is unloaded (or "indefinitely" if kvm is built-in).

This bug has existed since the tracepoint was added, but was recently
exposed by a new check in tracing to detect exactly this type of bug.

  fmt: '%s%s
  ' current_buffer: ' vmx_dirty_log_t-140127  [003] ....  kvm_nested_vmenter_failed: '
  WARNING: CPU: 3 PID: 140134 at kernel/trace/trace.c:3759 trace_check_vprintf+0x3be/0x3e0
  CPU: 3 PID: 140134 Comm: less Not tainted 5.13.0-rc1-ce2e73ce600a-req #184
  Hardware name: ASUS Q87M-E/Q87M-E, BIOS 1102 03/03/2014
  RIP: 0010:trace_check_vprintf+0x3be/0x3e0
  Code: <0f> 0b 44 8b 4c 24 1c e9 a9 fe ff ff c6 44 02 ff 00 49 8b 97 b0 20
  RSP: 0018:ffffa895cc37bcb0 EFLAGS: 00010282
  RAX: 0000000000000000 RBX: ffffa895cc37bd08 RCX: 0000000000000027
  RDX: 0000000000000027 RSI: 00000000ffffdfff RDI: ffff9766cfad74f8
  RBP: ffffffffc0a041d4 R08: ffff9766cfad74f0 R09: ffffa895cc37bad8
  R10: 0000000000000001 R11: 0000000000000001 R12: ffffffffc0a041d4
  R13: ffffffffc0f4dba8 R14: 0000000000000000 R15: ffff976409f2c000
  FS:  00007f92fa200740(0000) GS:ffff9766cfac0000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 0000559bd11b0000 CR3: 000000019fbaa002 CR4: 00000000001726e0
  Call Trace:
   trace_event_printf+0x5e/0x80
   trace_raw_output_kvm_nested_vmenter_failed+0x3a/0x60 [kvm]
   print_trace_line+0x1dd/0x4e0
   s_show+0x45/0x150
   seq_read_iter+0x2d5/0x4c0
   seq_read+0x106/0x150
   vfs_read+0x98/0x180
   ksys_read+0x5f/0xe0
   do_syscall_64+0x40/0xb0
   entry_SYSCALL_64_after_hwframe+0x44/0xae

Cc: Steven Rostedt <rostedt@goodmis.org>
Fixes: 380e0055bc7e ("KVM: nVMX: trace nested VM-Enter failures detected by H/W")
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Message-Id: <20210607175748.674002-1-seanjc@google.com>

3 years agoselftests: kvm: Add support for customized slot0 memory size
Zhenzhong Duan [Tue, 8 Jun 2021 23:38:16 +0000 (07:38 +0800)]
selftests: kvm: Add support for customized slot0 memory size

Until commit 39fe2fc96694 ("selftests: kvm: make allocation of extra
memory take effect", 2021-05-27), parameter extra_mem_pages was used
only to calculate the page table size for all the memory chunks,
because real memory allocation happened with calls of
vm_userspace_mem_region_add() after vm_create_default().

Commit 39fe2fc96694 however changed the meaning of extra_mem_pages to
the size of memory slot 0.  This makes the memory allocation more
flexible, but makes it harder to account for the number of
pages needed for the page tables.  For example, memslot_perf_test
has a small amount of memory in slot 0 but a lot in other slots,
and adding that memory twice (both in slot 0 and with later
calls to vm_userspace_mem_region_add()) causes an error that
was fixed in commit 000ac4295339 ("selftests: kvm: fix overlapping
addresses in memslot_perf_test", 2021-05-29)

Since both uses are sensible, add a new parameter slot0_mem_pages
to vm_create_with_vcpus() and some comments to clarify the meaning of
slot0_mem_pages and extra_mem_pages.  With this change,
memslot_perf_test can go back to passing the number of memory
pages as extra_mem_pages.

Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Message-Id: <20210608233816.423958-4-zhenzhong.duan@intel.com>
[Squashed in a single patch and rewrote the commit message. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: introduce P47V64 for s390x
Christian Borntraeger [Tue, 8 Jun 2021 12:39:54 +0000 (14:39 +0200)]
KVM: selftests: introduce P47V64 for s390x

s390x can have up to 47bits of physical guest and 64bits of virtual
address  bits. Add a new address mode to avoid errors of testcases
going beyond 47bits.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Message-Id: <20210608123954.10991-1-borntraeger@de.ibm.com>
Fixes: ef4c9f4f6546 ("KVM: selftests: Fix 32-bit truncation of vm_get_max_gfn()")
Cc: stable@vger.kernel.org
Reviewed-by: David Matlack <dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Ensure PV TLB flush tracepoint reflects KVM behavior
Lai Jiangshan [Mon, 31 May 2021 17:46:28 +0000 (01:46 +0800)]
KVM: x86: Ensure PV TLB flush tracepoint reflects KVM behavior

In record_steal_time(), st->preempted is read twice, and
trace_kvm_pv_tlb_flush() might output result inconsistent if
kvm_vcpu_flush_tlb_guest() see a different st->preempted later.

It is a very trivial problem and hardly has actual harm and can be
avoided by reseting and reading st->preempted in atomic way via xchg().

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Message-Id: <20210531174628.10265-1-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: MMU: Use the correct inherited permissions to get shadow page
Lai Jiangshan [Thu, 3 Jun 2021 05:24:55 +0000 (13:24 +0800)]
KVM: X86: MMU: Use the correct inherited permissions to get shadow page

When computing the access permissions of a shadow page, use the effective
permissions of the walk up to that point, i.e. the logic AND of its parents'
permissions.  Two guest PxE entries that point at the same table gfn need to
be shadowed with different shadow pages if their parents' permissions are
different.  KVM currently uses the effective permissions of the last
non-leaf entry for all non-leaf entries.  Because all non-leaf SPTEs have
full ("uwx") permissions, and the effective permissions are recorded only
in role.access and merged into the leaves, this can lead to incorrect
reuse of a shadow page and eventually to a missing guest protection page
fault.

For example, here is a shared pagetable:

   pgd[]   pud[]        pmd[]            virtual address pointers
                     /->pmd1(u--)->pte1(uw-)->page1 <- ptr1 (u--)
        /->pud1(uw-)--->pmd2(uw-)->pte2(uw-)->page2 <- ptr2 (uw-)
   pgd-|           (shared pmd[] as above)
        \->pud2(u--)--->pmd1(u--)->pte1(uw-)->page1 <- ptr3 (u--)
                     \->pmd2(uw-)->pte2(uw-)->page2 <- ptr4 (u--)

  pud1 and pud2 point to the same pmd table, so:
  - ptr1 and ptr3 points to the same page.
  - ptr2 and ptr4 points to the same page.

(pud1 and pud2 here are pud entries, while pmd1 and pmd2 here are pmd entries)

- First, the guest reads from ptr1 first and KVM prepares a shadow
  page table with role.access=u--, from ptr1's pud1 and ptr1's pmd1.
  "u--" comes from the effective permissions of pgd, pud1 and
  pmd1, which are stored in pt->access.  "u--" is used also to get
  the pagetable for pud1, instead of "uw-".

- Then the guest writes to ptr2 and KVM reuses pud1 which is present.
  The hypervisor set up a shadow page for ptr2 with pt->access is "uw-"
  even though the pud1 pmd (because of the incorrect argument to
  kvm_mmu_get_page in the previous step) has role.access="u--".

- Then the guest reads from ptr3.  The hypervisor reuses pud1's
  shadow pmd for pud2, because both use "u--" for their permissions.
  Thus, the shadow pmd already includes entries for both pmd1 and pmd2.

- At last, the guest writes to ptr4.  This causes no vmexit or pagefault,
  because pud1's shadow page structures included an "uw-" page even though
  its role.access was "u--".

Any kind of shared pagetable might have the similar problem when in
virtual machine without TDP enabled if the permissions are different
from different ancestors.

In order to fix the problem, we change pt->access to be an array, and
any access in it will not include permissions ANDed from child ptes.

The test code is: https://lore.kernel.org/kvm/20210603050537.19605-1-jiangshanlai@gmail.com/
Remember to test it with TDP disabled.

The problem had existed long before the commit 41074d07c78b ("KVM: MMU:
Fix inherited permissions for emulated guest pte updates"), and it
is hard to find which is the culprit.  So there is no fixes tag here.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Message-Id: <20210603052455.21023-1-jiangshanlai@gmail.com>
Cc: stable@vger.kernel.org
Fixes: cea0f0e7ea54 ("[PATCH] KVM: MMU: Shadow page table caching")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: LAPIC: Write 0 to TMICT should also cancel vmx-preemption timer
Wanpeng Li [Mon, 7 Jun 2021 07:19:43 +0000 (00:19 -0700)]
KVM: LAPIC: Write 0 to TMICT should also cancel vmx-preemption timer

According to the SDM 10.5.4.1:

  A write of 0 to the initial-count register effectively stops the local
  APIC timer, in both one-shot and periodic mode.

However, the lapic timer oneshot/periodic mode which is emulated by vmx-preemption
timer doesn't stop by writing 0 to TMICT since vmx->hv_deadline_tsc is still
programmed and the guest will receive the spurious timer interrupt later. This
patch fixes it by also cancelling the vmx-preemption timer when writing 0 to
the initial-count register.

Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1623050385-100988-1-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: SVM: Fix SEV SEND_START session length & SEND_UPDATE_DATA query length after...
Ashish Kalra [Mon, 7 Jun 2021 06:15:32 +0000 (06:15 +0000)]
KVM: SVM: Fix SEV SEND_START session length & SEND_UPDATE_DATA query length after commit 238eca821cee

Commit 238eca821cee ("KVM: SVM: Allocate SEV command structures on local stack")
uses the local stack to allocate the structures used to communicate with the PSP,
which were earlier being kzalloced. This breaks SEV live migration for
computing the SEND_START session length and SEND_UPDATE_DATA query length as
session_len and trans_len and hdr_len fields are not zeroed respectively for
the above commands before issuing the SEV Firmware API call, hence the
firmware returns incorrect session length and update data header or trans length.

Also the SEV Firmware API returns SEV_RET_INVALID_LEN firmware error
for these length query API calls, and the return value and the
firmware error needs to be passed to the userspace as it is, so
need to remove the return check in the KVM code.

Signed-off-by: Ashish Kalra <ashish.kalra@amd.com>
Message-Id: <20210607061532.27459-1-Ashish.Kalra@amd.com>
Fixes: 238eca821cee ("KVM: SVM: Allocate SEV command structures on local stack")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoselftests: kvm: fix overlapping addresses in memslot_perf_test
Paolo Bonzini [Fri, 28 May 2021 19:10:58 +0000 (15:10 -0400)]
selftests: kvm: fix overlapping addresses in memslot_perf_test

vm_create allocates memory and maps it close to GPA.  This memory
is separate from what is allocated in subsequent calls to
vm_userspace_mem_region_add, so it is incorrect to pass the
test memory size to vm_create_default.  Just pass a small
fixed amount of memory which can be used later for page table,
otherwise GPAs are already allocated at MEM_GPA and the
test aborts.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoMerge tag 'kvmarm-fixes-5.13-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Fri, 28 May 2021 17:02:03 +0000 (13:02 -0400)]
Merge tag 'kvmarm-fixes-5.13-2' of git://git./linux/kernel/git/kvmarm/kvmarm into HEAD

KVM/arm64 fixes for 5.13, take #2

- Another state update on exit to userspace fix
- Prevent the creation of mixed 32/64 VMs

3 years agoKVM: X86: Kill off ctxt->ud
Wanpeng Li [Fri, 28 May 2021 00:01:37 +0000 (17:01 -0700)]
KVM: X86: Kill off ctxt->ud

ctxt->ud is consumed only by x86_decode_insn(), we can kill it off by
passing emulation_type to x86_decode_insn() and dropping ctxt->ud
altogether. Tracking that info in ctxt for literally one call is silly.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <1622160097-37633-2-git-send-email-wanpengli@tencent.com>

3 years agoKVM: X86: Fix warning caused by stale emulation context
Wanpeng Li [Fri, 28 May 2021 00:01:36 +0000 (17:01 -0700)]
KVM: X86: Fix warning caused by stale emulation context

Reported by syzkaller:

  WARNING: CPU: 7 PID: 10526 at linux/arch/x86/kvm//x86.c:7621 x86_emulate_instruction+0x41b/0x510 [kvm]
  RIP: 0010:x86_emulate_instruction+0x41b/0x510 [kvm]
  Call Trace:
   kvm_mmu_page_fault+0x126/0x8f0 [kvm]
   vmx_handle_exit+0x11e/0x680 [kvm_intel]
   vcpu_enter_guest+0xd95/0x1b40 [kvm]
   kvm_arch_vcpu_ioctl_run+0x377/0x6a0 [kvm]
   kvm_vcpu_ioctl+0x389/0x630 [kvm]
   __x64_sys_ioctl+0x8e/0xd0
   do_syscall_64+0x3c/0xb0
   entry_SYSCALL_64_after_hwframe+0x44/0xae

Commit 4a1e10d5b5d8 ("KVM: x86: handle hardware breakpoints during emulation())
adds hardware breakpoints check before emulation the instruction and parts of
emulation context initialization, actually we don't have the EMULTYPE_NO_DECODE flag
here and the emulation context will not be reused. Commit c8848cee74ff ("KVM: x86:
set ctxt->have_exception in x86_decode_insn()) triggers the warning because it
catches the stale emulation context has #UD, however, it is not during instruction
decoding which should result in EMULATION_FAILED. This patch fixes it by moving
the second part emulation context initialization into init_emulate_ctxt() and
before hardware breakpoints check. The ctxt->ud will be dropped by a follow-up
patch.

syzkaller source: https://syzkaller.appspot.com/x/repro.c?x=134683fdd00000

Reported-by: syzbot+71271244f206d17f6441@syzkaller.appspotmail.com
Fixes: 4a1e10d5b5d8 (KVM: x86: handle hardware breakpoints during emulation)
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <1622160097-37633-1-git-send-email-wanpengli@tencent.com>

3 years agoKVM: X86: Use kvm_get_linear_rip() in single-step and #DB/#BP interception
Yuan Yao [Wed, 26 May 2021 06:38:28 +0000 (14:38 +0800)]
KVM: X86: Use kvm_get_linear_rip() in single-step and #DB/#BP interception

The kvm_get_linear_rip() handles x86/long mode cases well and has
better readability, __kvm_set_rflags() also use the paired
function kvm_is_linear_rip() to check the vcpu->arch.singlestep_rip
set in kvm_arch_vcpu_ioctl_set_guest_debug(), so change the
"CS.BASE + RIP" code in kvm_arch_vcpu_ioctl_set_guest_debug() and
handle_exception_nmi() to this one.

Signed-off-by: Yuan Yao <yuan.yao@intel.com>
Message-Id: <20210526063828.1173-1-yuan.yao@linux.intel.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Fix comment mentioning skip_4k
David Matlack [Wed, 26 May 2021 16:32:27 +0000 (16:32 +0000)]
KVM: x86/mmu: Fix comment mentioning skip_4k

This comment was left over from a previous version of the patch that
introduced wrprot_gfn_range, when skip_4k was passed in instead of
min_level.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210526163227.3113557-1-dmatlack@google.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: update vcpu posted-interrupt descriptor when assigning device
Marcelo Tosatti [Wed, 26 May 2021 17:20:14 +0000 (14:20 -0300)]
KVM: VMX: update vcpu posted-interrupt descriptor when assigning device

For VMX, when a vcpu enters HLT emulation, pi_post_block will:

1) Add vcpu to per-cpu list of blocked vcpus.

2) Program the posted-interrupt descriptor "notification vector"
to POSTED_INTR_WAKEUP_VECTOR

With interrupt remapping, an interrupt will set the PIR bit for the
vector programmed for the device on the CPU, test-and-set the
ON bit on the posted interrupt descriptor, and if the ON bit is clear
generate an interrupt for the notification vector.

This way, the target CPU wakes upon a device interrupt and wakes up
the target vcpu.

Problem is that pi_post_block only programs the notification vector
if kvm_arch_has_assigned_device() is true. Its possible for the
following to happen:

1) vcpu V HLTs on pcpu P, kvm_arch_has_assigned_device is false,
notification vector is not programmed
2) device is assigned to VM
3) device interrupts vcpu V, sets ON bit
(notification vector not programmed, so pcpu P remains in idle)
4) vcpu 0 IPIs vcpu V (in guest), but since pi descriptor ON bit is set,
kvm_vcpu_kick is skipped
5) vcpu 0 busy spins on vcpu V's response for several seconds, until
RCU watchdog NMIs all vCPUs.

To fix this, use the start_assignment kvm_x86_ops callback to kick
vcpus out of the halt loop, so the notification vector is
properly reprogrammed to the wakeup vector.

Reported-by: Pei Zhang <pezhang@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Message-Id: <20210526172014.GA29007@fuller.cnet>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: rename KVM_REQ_PENDING_TIMER to KVM_REQ_UNBLOCK
Marcelo Tosatti [Tue, 25 May 2021 13:41:17 +0000 (10:41 -0300)]
KVM: rename KVM_REQ_PENDING_TIMER to KVM_REQ_UNBLOCK

KVM_REQ_UNBLOCK will be used to exit a vcpu from
its inner vcpu halt emulation loop.

Rename KVM_REQ_PENDING_TIMER to KVM_REQ_UNBLOCK, switch
PowerPC to arch specific request bit.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Message-Id: <20210525134321.303768132@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: add start_assignment hook to kvm_x86_ops
Marcelo Tosatti [Tue, 25 May 2021 13:41:16 +0000 (10:41 -0300)]
KVM: x86: add start_assignment hook to kvm_x86_ops

Add a start_assignment hook to kvm_x86_ops, which is called when
kvm_arch_start_assignment is done.

The hook is required to update the wakeup vector of a sleeping vCPU
when a device is assigned to the guest.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Message-Id: <20210525134321.254128742@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: LAPIC: Narrow the timer latency between wait_lapic_expire and world switch
Wanpeng Li [Tue, 18 May 2021 12:00:35 +0000 (05:00 -0700)]
KVM: LAPIC: Narrow the timer latency between wait_lapic_expire and world switch

Let's treat lapic_timer_advance_ns automatic tuning logic as hypervisor
overhead, move it before wait_lapic_expire instead of between wait_lapic_expire
and the world switch, the wait duration should be calculated by the
up-to-date guest_tsc after the overhead of automatic tuning logic. This
patch reduces ~30+ cycles for kvm-unit-tests/tscdeadline-latency when testing
busy waits.

Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1621339235-11131-5-git-send-email-wanpengli@tencent.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoselftests: kvm: do only 1 memslot_perf_test run by default
Paolo Bonzini [Wed, 26 May 2021 18:36:14 +0000 (14:36 -0400)]
selftests: kvm: do only 1 memslot_perf_test run by default

The test takes a long time with the current implementation of
memslots, so cut the run time a bit.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Use _BITUL() macro in UAPI headers
Joe Richey [Fri, 21 May 2021 08:58:43 +0000 (01:58 -0700)]
KVM: X86: Use _BITUL() macro in UAPI headers

Replace BIT() in KVM's UPAI header with _BITUL(). BIT() is not defined
in the UAPI headers and its usage may cause userspace build errors.

Fixes: fb04a1eddb1a ("KVM: X86: Implement ring-based dirty memory tracking")
Signed-off-by: Joe Richey <joerichey@google.com>
Message-Id: <20210521085849.37676-3-joerichey94@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: add shared hugetlbfs backing source type
Axel Rasmussen [Wed, 19 May 2021 20:03:39 +0000 (13:03 -0700)]
KVM: selftests: add shared hugetlbfs backing source type

This lets us run the demand paging test on top of a shared
hugetlbfs-backed area. The "shared" is key, as this allows us to
exercise userfaultfd minor faults on hugetlbfs.

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-11-axelrasmussen@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: allow using UFFD minor faults for demand paging
Axel Rasmussen [Wed, 19 May 2021 20:03:38 +0000 (13:03 -0700)]
KVM: selftests: allow using UFFD minor faults for demand paging

UFFD handling of MINOR faults is a new feature whose use case is to
speed up demand paging (compared to MISSING faults). So, it's
interesting to let this selftest exercise this new mode.

Modify the demand paging test to have the option of using UFFD minor
faults, as opposed to missing faults. Now, when turning on userfaultfd
with '-u', the desired mode has to be specified ("MISSING" or "MINOR").

If we're in minor mode, before registering, prefault via the *alias*.
This way, the guest will trigger minor faults, instead of missing
faults, and we can UFFDIO_CONTINUE to resolve them.

Modify the page fault handler function to use the right ioctl depending
on the mode we're running in. In MINOR mode, use UFFDIO_CONTINUE.

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-10-axelrasmussen@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: create alias mappings when using shared memory
Axel Rasmussen [Wed, 19 May 2021 20:03:37 +0000 (13:03 -0700)]
KVM: selftests: create alias mappings when using shared memory

When a memory region is added with a src_type specifying that it should
use some kind of shared memory, also create an alias mapping to the same
underlying physical pages.

And, add an API so tests can get access to these alias addresses.
Basically, for a guest physical address, let us look up the analogous
host *alias* address.

In a future commit, we'll modify the demand paging test to take
advantage of this to exercise UFFD minor faults. The idea is, we
pre-fault the underlying pages *via the alias*. When the *guest*
faults, it gets a "minor" fault (PTEs don't exist yet, but a page is
already in the page cache). Then, the userfaultfd theads can handle the
fault: they could potentially modify the underlying memory *via the
alias* if they wanted to, and then they install the PTEs and let the
guest carry on via a UFFDIO_CONTINUE ioctl.

Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-9-axelrasmussen@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: add shmem backing source type
Axel Rasmussen [Wed, 19 May 2021 20:03:36 +0000 (13:03 -0700)]
KVM: selftests: add shmem backing source type

This lets us run the demand paging test on top of a shmem-backed area.
In follow-up commits, we'll 1) leverage this new capability to create an
alias mapping, and then 2) use the alias mapping to exercise UFFD minor
faults.

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-8-axelrasmussen@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: refactor vm_mem_backing_src_type flags
Axel Rasmussen [Wed, 19 May 2021 20:03:35 +0000 (13:03 -0700)]
KVM: selftests: refactor vm_mem_backing_src_type flags

Each struct vm_mem_backing_src_alias has a flags field, which denotes
the flags used to mmap() an area of that type. Previously, this field
never included MAP_PRIVATE | MAP_ANONYMOUS, because
vm_userspace_mem_region_add assumed that *all* types would always use
those flags, and so it hardcoded them.

In a follow-up commit, we'll add a new type: shmem. Areas of this type
must not have MAP_PRIVATE | MAP_ANONYMOUS, and instead they must have
MAP_SHARED.

So, refactor things. Make it so that the flags field of
struct vm_mem_backing_src_alias really is a complete set of flags, and
don't add in any extras in vm_userspace_mem_region_add. This will let us
easily tack on shmem.

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-7-axelrasmussen@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: allow different backing source types
Axel Rasmussen [Wed, 19 May 2021 20:03:34 +0000 (13:03 -0700)]
KVM: selftests: allow different backing source types

Add an argument which lets us specify a different backing memory type
for the test. The default is just to use anonymous, matching existing
behavior.

This is in preparation for testing UFFD minor faults. For that, we'll
need to use a new backing memory type which is setup with MAP_SHARED.

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-6-axelrasmussen@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: compute correct demand paging size
Axel Rasmussen [Wed, 19 May 2021 20:03:33 +0000 (13:03 -0700)]
KVM: selftests: compute correct demand paging size

This is a preparatory commit needed before we can use different kinds of
backing pages for guest memory.

Previously, we used perf_test_args.host_page_size, which is the host's
native page size (commonly 4K). For VM_MEM_SRC_ANONYMOUS this turns out
to be okay, but in a follow-up commit we want to allow using different
kinds of backing memory.

Take VM_MEM_SRC_ANONYMOUS_HUGETLB for example. Without this change, if
we used that backing page type, when we issued a UFFDIO_COPY ioctl we'd
only do so with 4K, rather than the full 2M of a backing hugepage. In
this case, UFFDIO_COPY returns -EINVAL (__mcopy_atomic_hugetlb checks
the size).

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-5-axelrasmussen@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: simplify setup_demand_paging error handling
Axel Rasmussen [Wed, 19 May 2021 20:03:31 +0000 (13:03 -0700)]
KVM: selftests: simplify setup_demand_paging error handling

A small cleanup. Our caller writes:

  r = setup_demand_paging(...);
  if (r < 0) exit(-r);

Since we're just going to exit anyway, instead of returning an error we
can just re-use TEST_ASSERT. This makes the caller simpler, as well as
the function itself - no need to write our branches, etc.

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-3-axelrasmussen@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Print a message if /dev/kvm is missing
David Matlack [Tue, 11 May 2021 20:21:20 +0000 (20:21 +0000)]
KVM: selftests: Print a message if /dev/kvm is missing

If a KVM selftest is run on a machine without /dev/kvm, it will exit
silently. Make it easy to tell what's happening by printing an error
message.

Opportunistically consolidate all codepaths that open /dev/kvm into a
single function so they all print the same message.

This slightly changes the semantics of vm_is_unrestricted_guest() by
changing a TEST_ASSERT() to exit(KSFT_SKIP). However
vm_is_unrestricted_guest() is only called in one place
(x86_64/mmio_warning_test.c) and that is to determine if the test should
be skipped or not.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210511202120.1371800-1-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: trivial comment/logging fixes
Axel Rasmussen [Wed, 19 May 2021 20:03:30 +0000 (13:03 -0700)]
KVM: selftests: trivial comment/logging fixes

Some trivial fixes I found while touching related code in this series,
factored out into a separate commit for easier reviewing:

- s/gor/got/ and add a newline in demand_paging_test.c
- s/backing_src/src_type/ in a comment to be consistent with the real
  function signature in kvm_util.c

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Message-Id: <20210519200339.829146-2-axelrasmussen@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Fix hang in hardware_disable_test
David Matlack [Fri, 14 May 2021 23:05:21 +0000 (23:05 +0000)]
KVM: selftests: Fix hang in hardware_disable_test

If /dev/kvm is not available then hardware_disable_test will hang
indefinitely because the child process exits before posting to the
semaphore for which the parent is waiting.

Fix this by making the parent periodically check if the child has
exited. We have to be careful to forward the child's exit status to
preserve a KSFT_SKIP status.

I considered just checking for /dev/kvm before creating the child
process, but there are so many other reasons why the child could exit
early that it seemed better to handle that as general case.

Tested:

$ ./hardware_disable_test
/dev/kvm not available, skipping test
$ echo $?
4
$ modprobe kvm_intel
$ ./hardware_disable_test
$ echo $?
0

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210514230521.2608768-1-dmatlack@google.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Ignore CPUID.0DH.1H in get_cpuid_test
David Matlack [Wed, 19 May 2021 21:13:45 +0000 (21:13 +0000)]
KVM: selftests: Ignore CPUID.0DH.1H in get_cpuid_test

Similar to CPUID.0DH.0H this entry depends on the vCPU's XCR0 register
and IA32_XSS MSR. Since this test does not control for either before
assigning the vCPU's CPUID, these entries will not necessarily match
the supported CPUID exposed by KVM.

This fixes get_cpuid_test on Cascade Lake CPUs.

Suggested-by: Jim Mattson <jmattson@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210519211345.3944063-1-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Fix 32-bit truncation of vm_get_max_gfn()
David Matlack [Fri, 21 May 2021 17:38:28 +0000 (17:38 +0000)]
KVM: selftests: Fix 32-bit truncation of vm_get_max_gfn()

vm_get_max_gfn() casts vm->max_gfn from a uint64_t to an unsigned int,
which causes the upper 32-bits of the max_gfn to get truncated.

Nobody noticed until now likely because vm_get_max_gfn() is only used
as a mechanism to create a memslot in an unused region of the guest
physical address space (the top), and the top of the 32-bit physical
address space was always good enough.

This fix reveals a bug in memslot_modification_stress_test which was
trying to create a dummy memslot past the end of guest physical memory.
Fix that by moving the dummy memslot lower.

Fixes: 52200d0d944e ("KVM: selftests: Remove duplicate guest mode handling")
Reviewed-by: Venkatesh Srinivas <venkateshs@chromium.org>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210521173828.1180619-1-dmatlack@google.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: add a memslot-related performance benchmark
Maciej S. Szmigiero [Tue, 13 Apr 2021 14:08:28 +0000 (16:08 +0200)]
KVM: selftests: add a memslot-related performance benchmark

This benchmark contains the following tests:
* Map test, where the host unmaps guest memory while the guest writes to
it (maps it).

The test is designed in a way to make the unmap operation on the host
take a negligible amount of time in comparison with the mapping
operation in the guest.

The test area is actually split in two: the first half is being mapped
by the guest while the second half in being unmapped by the host.
Then a guest <-> host sync happens and the areas are reversed.

* Unmap test which is broadly similar to the above map test, but it is
designed in an opposite way: to make the mapping operation in the guest
take a negligible amount of time in comparison with the unmap operation
on the host.
This test is available in two variants: with per-page unmap operation
or a chunked one (using 2 MiB chunk size).

* Move active area test which involves moving the last (highest gfn)
memslot a bit back and forth on the host while the guest is
concurrently writing around the area being moved (including over the
moved memslot).

* Move inactive area test which is similar to the previous move active
area test, but now guest writes all happen outside of the area being
moved.

* Read / write test in which the guest writes to the beginning of each
page of the test area while the host writes to the middle of each such
page.
Then each side checks the values the other side has written.
This particular test is not expected to give different results depending
on particular memslots implementation, it is meant as a rough sanity
check and to provide insight on the spread of test results expected.

Each test performs its operation in a loop until a test period ends
(this is 5 seconds by default, but it is configurable).
Then the total count of loops done is divided by the actual elapsed
time to give the test result.

The tests have a configurable memslot cap with the "-s" test option, by
default the system maximum is used.
Each test is repeated a particular number of times (by default 20
times), the best result achieved is printed.

The test memory area is divided equally between memslots, the reminder
is added to the last memslot.
The test area size does not depend on the number of memslots in use.

The tests also measure the time that it took to add all these memslots.
The best result from the tests that use the whole test area is printed
after all the requested tests are done.

In general, these tests are designed to use as much memory as possible
(within reason) while still doing 100+ loops even on high memslot counts
with the default test length.
Increasing the test runtime makes it increasingly more likely that some
event will happen on the system during the test run, which might lower
the test result.

Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Message-Id: <8d31bb3d92bc8fa33a9756fa802ee14266ab994e.1618253574.git.maciej.szmigiero@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Keep track of memslots more efficiently
Maciej S. Szmigiero [Tue, 13 Apr 2021 14:08:27 +0000 (16:08 +0200)]
KVM: selftests: Keep track of memslots more efficiently

The KVM selftest framework was using a simple list for keeping track of
the memslots currently in use.
This resulted in lookups and adding a single memslot being O(n), the
later due to linear scanning of the existing memslot set to check for
the presence of any conflicting entries.

Before this change, benchmarking high count of memslots was more or less
impossible as pretty much all the benchmark time was spent in the
selftest framework code.

We can simply use a rbtree for keeping track of both of gfn and hva.
We don't need an interval tree for hva here as we can't have overlapping
memslots because we allocate a completely new memory chunk for each new
memslot.

Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Message-Id: <b12749d47ee860468240cf027412c91b76dbe3db.1618253574.git.maciej.szmigiero@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoselftests: kvm: fix potential issue with ELF loading
Paolo Bonzini [Mon, 24 May 2021 12:27:38 +0000 (14:27 +0200)]
selftests: kvm: fix potential issue with ELF loading

vm_vaddr_alloc() sets up GVA to GPA mapping page by page; therefore, GPAs
may not be continuous if same memslot is used for data and page table allocation.

kvm_vm_elf_load() however expects a continuous range of HVAs (and thus GPAs)
because it does not try to read file data page by page.  Fix this mismatch
by allocating memory in one step.

Reported-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoselftests: kvm: make allocation of extra memory take effect
Zhenzhong Duan [Wed, 12 May 2021 04:31:06 +0000 (12:31 +0800)]
selftests: kvm: make allocation of extra memory take effect

The extra memory pages is missed to be allocated during VM creating.
perf_test_util and kvm_page_table_test use it to alloc extra memory
currently.

Fix it by adding extra_mem_pages to the total memory calculation before
allocate.

Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
Message-Id: <20210512043107.30076-1-zhenzhong.duan@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: hyper-v: Task srcu lock when accessing kvm_memslots()
Wanpeng Li [Tue, 18 May 2021 12:00:34 +0000 (05:00 -0700)]
KVM: X86: hyper-v: Task srcu lock when accessing kvm_memslots()

   WARNING: suspicious RCU usage
   5.13.0-rc1 #4 Not tainted
   -----------------------------
   ./include/linux/kvm_host.h:710 suspicious rcu_dereference_check() usage!

  other info that might help us debug this:

  rcu_scheduler_active = 2, debug_locks = 1
   1 lock held by hyperv_clock/8318:
    #0: ffffb6b8cb05a7d8 (&hv->hv_lock){+.+.}-{3:3}, at: kvm_hv_invalidate_tsc_page+0x3e/0xa0 [kvm]

  stack backtrace:
  CPU: 3 PID: 8318 Comm: hyperv_clock Not tainted 5.13.0-rc1 #4
  Call Trace:
   dump_stack+0x87/0xb7
   lockdep_rcu_suspicious+0xce/0xf0
   kvm_write_guest_page+0x1c1/0x1d0 [kvm]
   kvm_write_guest+0x50/0x90 [kvm]
   kvm_hv_invalidate_tsc_page+0x79/0xa0 [kvm]
   kvm_gen_update_masterclock+0x1d/0x110 [kvm]
   kvm_arch_vm_ioctl+0x2a7/0xc50 [kvm]
   kvm_vm_ioctl+0x123/0x11d0 [kvm]
   __x64_sys_ioctl+0x3ed/0x9d0
   do_syscall_64+0x3d/0x80
   entry_SYSCALL_64_after_hwframe+0x44/0xae

kvm_memslots() will be called by kvm_write_guest(), so we should take the srcu lock.

Fixes: e880c6ea5 (KVM: x86: hyper-v: Prevent using not-yet-updated TSC page by secondary CPUs)
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1621339235-11131-4-git-send-email-wanpengli@tencent.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Fix vCPU preempted state from guest's point of view
Wanpeng Li [Tue, 18 May 2021 12:00:33 +0000 (05:00 -0700)]
KVM: X86: Fix vCPU preempted state from guest's point of view

Commit 66570e966dd9 (kvm: x86: only provide PV features if enabled in guest's
CPUID) avoids to access pv tlb shootdown host side logic when this pv feature
is not exposed to guest, however, kvm_steal_time.preempted not only leveraged
by pv tlb shootdown logic but also mitigate the lock holder preemption issue.
From guest's point of view, vCPU is always preempted since we lose the reset
of kvm_steal_time.preempted before vmentry if pv tlb shootdown feature is not
exposed. This patch fixes it by clearing kvm_steal_time.preempted before
vmentry.

Fixes: 66570e966dd9 (kvm: x86: only provide PV features if enabled in guest's CPUID)
Reviewed-by: Sean Christopherson <seanjc@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1621339235-11131-3-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Bail out of direct yield in case of under-committed scenarios
Wanpeng Li [Tue, 18 May 2021 12:00:32 +0000 (05:00 -0700)]
KVM: X86: Bail out of direct yield in case of under-committed scenarios

In case of under-committed scenarios, vCPUs can be scheduled easily;
kvm_vcpu_yield_to adds extra overhead, and it is also common to see
when vcpu->ready is true but yield later failing due to p->state is
TASK_RUNNING.

Let's bail out in such scenarios by checking the length of current cpu
runqueue, which can be treated as a hint of under-committed instead of
guarantee of accuracy. 30%+ of directed-yield attempts can now avoid
the expensive lookups in kvm_sched_yield() in an under-committed scenario.

Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1621339235-11131-2-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: PPC: exit halt polling on need_resched()
Wanpeng Li [Tue, 18 May 2021 12:00:31 +0000 (05:00 -0700)]
KVM: PPC: exit halt polling on need_resched()

This is inspired by commit 262de4102c7bb8 (kvm: exit halt polling on
need_resched() as well). Due to PPC implements an arch specific halt
polling logic, we have to the need_resched() check there as well. This
patch adds a helper function that can be shared between book3s and generic
halt-polling loops.

Reviewed-by: David Matlack <dmatlack@google.com>
Reviewed-by: Venkatesh Srinivas <venkateshs@chromium.org>
Cc: Ben Segall <bsegall@google.com>
Cc: Venkatesh Srinivas <venkateshs@chromium.org>
Cc: Jim Mattson <jmattson@google.com>
Cc: David Matlack <dmatlack@google.com>
Cc: Paul Mackerras <paulus@ozlabs.org>
Cc: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1621339235-11131-1-git-send-email-wanpengli@tencent.com>
[Make the function inline. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: arm64: Prevent mixed-width VM creation
Marc Zyngier [Mon, 24 May 2021 17:07:52 +0000 (18:07 +0100)]
KVM: arm64: Prevent mixed-width VM creation

It looks like we have tolerated creating mixed-width VMs since...
forever. However, that was never the intention, and we'd rather
not have to support that pointless complexity.

Forbid such a setup by making sure all the vcpus have the same
register width.

Reported-by: Steven Price <steven.price@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Cc: stable@vger.kernel.org
Acked-by: Mark Rutland <mark.rutland@arm.com>
Link: https://lore.kernel.org/r/20210524170752.1549797-1-maz@kernel.org