Two new stats for exposing halt-polling cpu usage:
halt_poll_success_ns
halt_poll_fail_ns
Thus sum of these 2 stats is the total cpu time spent polling. "success"
means the VCPU polled until a virtual interrupt was delivered. "fail"
means the VCPU had to schedule out (either because the maximum poll time
was reached or it needed to yield the CPU).
To avoid touching every arch's kvm_vcpu_stat struct, only update and
export halt-polling cpu usage stats if we're on x86.
Exporting cpu usage as a u64 and in nanoseconds means we will overflow at
~500 years, which seems reasonably large.
Signed-off-by: David Matlack <dmatlack@google.com>
Signed-off-by: Jon Cargille <jcargill@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Message-Id: <20200508182240.68440-1-jcargill@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The hrtimer used to emulate the VMX-preemption timer must be pinned to
the same logical processor as the vCPU thread to be interrupted if we
want to have any hope of adhering to the architectural specification
of the VMX-preemption timer. Even with this change, the emulated
VMX-preemption timer VM-exit occasionally arrives too late.
Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Peter Shier <pshier@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Message-Id: <20200508203643.85477-4-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Adds a fastpath_t typedef since enum lines are a bit long, and replace
EXIT_FASTPATH_SKIP_EMUL_INS with two new exit_fastpath_completion enum values.
- EXIT_FASTPATH_EXIT_HANDLED kvm will still go through it's full run loop,
but it would skip invoking the exit handler.
- EXIT_FASTPATH_REENTER_GUEST complete fastpath, guest can be re-entered
without invoking the exit handler or going
back to vcpu_run
Tested-by: Haiwei Li <lihaiwei@tencent.com>
Cc: Haiwei Li <lihaiwei@tencent.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1588055009-12677-4-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Forcing the ASYNC_PF_PER_VCPU to be power of two is much easier to be
used rather than calling roundup_pow_of_two() from time to time. Do
this by adding a BUILD_BUG_ON() inside the hash function.
Another point is that generally async pf does not allow concurrency
over ASYNC_PF_PER_VCPU after all (see kvm_setup_async_pf()), so it
does not make much sense either to have it not a power of two or some
of the entries will definitely be wasted.
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200416155859.267366-1-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Replace KVM's PT_PAGE_TABLE_LEVEL, PT_DIRECTORY_LEVEL and PT_PDPE_LEVEL
with the kernel's PG_LEVEL_4K, PG_LEVEL_2M and PG_LEVEL_1G. KVM's
enums are borderline impossible to remember and result in code that is
visually difficult to audit, e.g.
if (!enable_ept)
ept_lpage_level = 0;
else if (cpu_has_vmx_ept_1g_page())
ept_lpage_level = PT_PDPE_LEVEL;
else if (cpu_has_vmx_ept_2m_page())
ept_lpage_level = PT_DIRECTORY_LEVEL;
else
ept_lpage_level = PT_PAGE_TABLE_LEVEL;
versus
if (!enable_ept)
ept_lpage_level = 0;
else if (cpu_has_vmx_ept_1g_page())
ept_lpage_level = PG_LEVEL_1G;
else if (cpu_has_vmx_ept_2m_page())
ept_lpage_level = PG_LEVEL_2M;
else
ept_lpage_level = PG_LEVEL_4K;
No functional change intended.
Suggested-by: Barret Rhoden <brho@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200428005422.4235-4-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rename PT_MAX_HUGEPAGE_LEVEL to KVM_MAX_HUGEPAGE_LEVEL and make it a
separate define in anticipation of dropping KVM's PT_*_LEVEL enums in
favor of the kernel's PG_LEVEL_* enums.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200428005422.4235-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
vcpu->arch.guest_xstate_size lost its only user since commit df1daba7d1
("KVM: x86: support XSAVES usage in the host"), so clean it up.
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-Id: <20200429154312.1411-1-xiaoyao.li@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Snapshot the TDP level now that it's invariant (SVM) or dependent only
on host capabilities and guest CPUID (VMX). This avoids having to call
kvm_x86_ops.get_tdp_level() when initializing a TDP MMU and/or
calculating the page role, and thus avoids the associated retpoline.
Drop the WARN in vmx_get_tdp_level() as updating CPUID while L2 is
active is legal, if dodgy.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-11-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move CR0 caching into the standard register caching mechanism in order
to take advantage of the availability checks provided by regs_avail.
This avoids multiple VMREADs in the (uncommon) case where kvm_read_cr0()
is called multiple times in a single VM-Exit, and more importantly
eliminates a kvm_x86_ops hook, saves a retpoline on SVM when reading
CR0, and squashes the confusing naming discrepancy of "cache_reg" vs.
"decache_cr0_guest_bits".
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-8-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move CR4 caching into the standard register caching mechanism in order
to take advantage of the availability checks provided by regs_avail.
This avoids multiple VMREADs and retpolines (when configured) during
nested VMX transitions as kvm_read_cr4_bits() is invoked multiple times
on each transition, e.g. when stuffing CR0 and CR3.
As an added bonus, this eliminates a kvm_x86_ops hook, saves a retpoline
on SVM when reading CR4, and squashes the confusing naming discrepancy
of "cache_reg" vs. "decache_cr4_guest_bits".
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-7-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Save L1's TSC offset in 'struct kvm_vcpu_arch' and drop the kvm_x86_ops
hook read_l1_tsc_offset(). This avoids a retpoline (when configured)
when reading L1's effective TSC, which is done at least once on every
VM-Exit.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200502043234.12481-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add an argument to interrupt_allowed and nmi_allowed, to checking if
interrupt injection is blocked. Use the hook to handle the case where
an interrupt arrives between check_nested_events() and the injection
logic. Drop the retry of check_nested_events() that hack-a-fixed the
same condition.
Blocking injection is also a bit of a hack, e.g. KVM should do exiting
and non-exiting interrupt processing in a single pass, but it's a more
precise hack. The old comment is also misleading, e.g. KVM_REQ_EVENT is
purely an optimization, setting it on every run loop (which KVM doesn't
do) should not affect functionality, only performance.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-13-sean.j.christopherson@intel.com>
[Extend to SVM, add SMI and NMI. Even though NMI and SMI cannot come
asynchronously right now, making the fix generic is easy and removes a
special case. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Return an actual bool for kvm_x86_ops' {interrupt_nmi}_allowed() hook to
better reflect the return semantics, and to avoid creating an even
bigger mess when the related VMX code is refactored in upcoming patches.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-5-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a kvm_x86_ops hook to detect a nested pending "hypervisor timer" and
use it to effectively open a window for servicing the expired timer.
Like pending SMIs on VMX, opening a window simply means requesting an
immediate exit.
This fixes a bug where an expired VMX preemption timer (for L2) will be
delayed and/or lost if a pending exception is injected into L2. The
pending exception is rightly prioritized by vmx_check_nested_events()
and injected into L2, with the preemption timer left pending. Because
no window opened, L2 is free to run uninterrupted.
Fixes: f4124500c2 ("KVM: nVMX: Fully emulate preemption timer")
Reported-by: Jim Mattson <jmattson@google.com>
Cc: Oliver Upton <oupton@google.com>
Cc: Peter Shier <pshier@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423022550.15113-3-sean.j.christopherson@intel.com>
[Check it in kvm_vcpu_has_events too, to ensure that the preemption
timer is serviced promptly even if the vCPU is halted and L1 is not
intercepting HLT. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Though rdpkru and wrpkru are contingent upon CR4.PKE, the PKRU
resource isn't. It can be read with XSAVE and written with XRSTOR.
So, if we don't set the guest PKRU value here(kvm_load_guest_xsave_state),
the guest can read the host value.
In case of kvm_load_host_xsave_state, guest with CR4.PKE clear could
potentially use XRSTOR to change the host PKRU value.
While at it, move pkru state save/restore to common code and the
host_pkru field to kvm_vcpu_arch. This will let SVM support protection keys.
Cc: stable@vger.kernel.org
Reported-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Babu Moger <babu.moger@amd.com>
Message-Id: <158932794619.44260.14508381096663848853.stgit@naples-babu.amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
There are two issues with KVM_EXIT_DEBUG on AMD, whose root cause is the
different handling of DR6 on intercepted #DB exceptions on Intel and AMD.
On Intel, #DB exceptions transmit the DR6 value via the exit qualification
field of the VMCS, and the exit qualification only contains the description
of the precise event that caused a vmexit.
On AMD, instead the DR6 field of the VMCB is filled in as if the #DB exception
was to be injected into the guest. This has two effects when guest debugging
is in use:
* the guest DR6 is clobbered
* the kvm_run->debug.arch.dr6 field can accumulate more debug events, rather
than just the last one that happened (the testcase in the next patch covers
this issue).
This patch fixes both issues by emulating, so to speak, the Intel behavior
on AMD processors. The important observation is that (after the previous
patches) the VMCB value of DR6 is only ever observable from the guest is
KVM_DEBUGREG_WONT_EXIT is set. Therefore we can actually set vmcb->save.dr6
to any value we want as long as KVM_DEBUGREG_WONT_EXIT is clear, which it
will be if guest debugging is enabled.
Therefore it is possible to enter the guest with an all-zero DR6,
reconstruct the #DB payload from the DR6 we get at exit time, and let
kvm_deliver_exception_payload move the newly set bits into vcpu->arch.dr6.
Some extra bits may be included in the payload if KVM_DEBUGREG_WONT_EXIT
is set, but this is harmless.
This may not be the most optimized way to deal with this, but it is
simple and, being confined within SVM code, it gets rid of the set_dr6
callback and kvm_update_dr6.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
kvm_x86_ops.set_dr6 is only ever called with vcpu->arch.dr6 as the
second argument. Ensure that the VMCB value is synchronized to
vcpu->arch.dr6 on #DB (both "normal" and nested) and nested vmentry, so
that the current value of DR6 is always available in vcpu->arch.dr6.
The get_dr6 callback can just access vcpu->arch.dr6 and becomes redundant.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Go through kvm_queue_exception_p so that the payload is correctly delivered
through the exit qualification, and add a kvm_update_dr6 call to
kvm_deliver_exception_payload that is needed on AMD.
Reported-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Current logic incorrectly uses the enum ioapic_irq_destination_types
to check the posted interrupt destination types. However, the value was
set using APIC_DM_XXX macros, which are left-shifted by 8 bits.
Fixes by using the APIC_DM_FIXED and APIC_DM_LOWEST instead.
Fixes: (fdcf756213 'KVM: x86: Disable posted interrupts for non-standard IRQs delivery modes')
Cc: Alexander Graf <graf@amazon.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <1586239989-58305-1-git-send-email-suravee.suthikulpanit@amd.com>
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Tested-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Clean up some of the patching of kvm_x86_ops, by moving kvm_x86_ops related to
nested virtualization into a separate struct.
As a result, these ops will always be non-NULL on VMX. This is not a problem:
* check_nested_events is only called if is_guest_mode(vcpu) returns true
* get_nested_state treats VMXOFF state the same as nested being disabled
* set_nested_state fails if you attempt to set nested state while
nesting is disabled
* nested_enable_evmcs could already be called on a CPU without VMX enabled
in CPUID.
* nested_get_evmcs_version was fixed in the previous patch
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
IPI and Timer cause the main MSRs write vmexits in cloud environment
observation, let's optimize virtual IPI latency more aggressively to
inject target IPI as soon as possible.
Running kvm-unit-tests/vmexit.flat IPI testing on SKX server, disable
adaptive advance lapic timer and adaptive halt-polling to avoid the
interference, this patch can give another 7% improvement.
w/o fastpath -> x86.c fastpath 4238 -> 3543 16.4%
x86.c fastpath -> vmx.c fastpath 3543 -> 3293 7%
w/o fastpath -> vmx.c fastpath 4238 -> 3293 22.3%
Cc: Haiwei Li <lihaiwei@tencent.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200410174703.1138-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce a new "extended register" type, EXIT_INFO_2 (to pair with the
nomenclature in .get_exit_info()), and use it to cache VMX's
vmcs.EXIT_INTR_INFO. Drop a comment in vmx_recover_nmi_blocking() that
is obsoleted by the generic caching mechanism.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415203454.8296-6-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce a new "extended register" type, EXIT_INFO_1 (to pair with the
nomenclature in .get_exit_info()), and use it to cache VMX's
vmcs.EXIT_QUALIFICATION.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415203454.8296-5-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rename functions and variables in kvm_mmu_new_cr3() and related code to
replace "cr3" with "pgd", i.e. continue the work started by commit
727a7e27cf ("KVM: x86: rename set_cr3 callback and related flags to
load_mmu_pgd"). kvm_mmu_new_cr3() and company are not always loading a
new CR3, e.g. when nested EPT is enabled "cr3" is actually an EPTP.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-37-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a separate "skip" override for MMU sync, a future change to avoid
TLB flushes on nested VMX transitions may need to sync the MMU even if
the TLB flush is unnecessary.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-32-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the retrieval of the HPA associated with L1's APIC access page into
VMX code to avoid unnecessarily calling gfn_to_page(), e.g. when the
vCPU is in guest mode (L2). Alternatively, the optimization logic in
VMX could be mirrored into the common x86 code, but that will get ugly
fast when further optimizations are introduced.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-29-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add KVM_REQ_TLB_FLUSH_CURRENT to allow optimized TLB flushing of VMX's
EPTP/VPID contexts[*] from the KVM MMU and/or in a deferred manner, e.g.
to flush L2's context during nested VM-Enter.
Convert KVM_REQ_TLB_FLUSH to KVM_REQ_TLB_FLUSH_CURRENT in flows where
the flush is directly associated with vCPU-scoped instruction emulation,
i.e. MOV CR3 and INVPCID.
Add a comment in vmx_vcpu_load_vmcs() above its KVM_REQ_TLB_FLUSH to
make it clear that it deliberately requests a flush of all contexts.
Service any pending flush request on nested VM-Exit as it's possible a
nested VM-Exit could occur after requesting a flush for L2. Add the
same logic for nested VM-Enter even though it's _extremely_ unlikely
for flush to be pending on nested VM-Enter, but theoretically possible
(in the future) due to RSM (SMM) emulation.
[*] Intel also has an Address Space Identifier (ASID) concept, e.g.
EPTP+VPID+PCID == ASID, it's just not documented in the SDM because
the rules of invalidation are different based on which piece of the
ASID is being changed, i.e. whether the EPTP, VPID, or PCID context
must be invalidated.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-25-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rename ->tlb_flush() to ->tlb_flush_all() in preparation for adding a
new hook to flush only the current ASID/context.
Opportunstically replace the comment in vmx_flush_tlb() that explains
why it flushes all EPTP/VPID contexts with a comment explaining why it
unconditionally uses INVEPT when EPT is enabled. I.e. rely on the "all"
part of the name to clarify why it does global INVEPT/INVVPID.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-23-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Drop @invalidate_gpa from ->tlb_flush() and kvm_vcpu_flush_tlb() now
that all callers pass %true for said param, or ignore the param (SVM has
an internal call to svm_flush_tlb() in svm_flush_tlb_guest that somewhat
arbitrarily passes %false).
Remove __vmx_flush_tlb() as it is no longer used.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-17-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Hyper-V PV TLB flush mechanism does TLB flush on behalf of the guest
so doing tlb_flush_all() is an overkill, switch to using tlb_flush_guest()
(just like KVM PV TLB flush mechanism) instead. Introduce
KVM_REQ_HV_TLB_FLUSH to support the change.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a dedicated hook to handle flushing TLB entries on behalf of the
guest, i.e. for a paravirtualized TLB flush, and use it directly instead
of bouncing through kvm_vcpu_flush_tlb().
For VMX, change the effective implementation implementation to never do
INVEPT and flush only the current context, i.e. to always flush via
INVVPID(SINGLE_CONTEXT). The INVEPT performed by __vmx_flush_tlb() when
@invalidate_gpa=false and enable_vpid=0 is unnecessary, as it will only
flush guest-physical mappings; linear and combined mappings are flushed
by VM-Enter when VPID is disabled, and changes in the guest pages tables
do not affect guest-physical mappings.
When EPT and VPID are enabled, doing INVVPID is not required (by Intel's
architecture) to invalidate guest-physical mappings, i.e. TLB entries
that cache guest-physical mappings can live across INVVPID as the
mappings are associated with an EPTP, not a VPID. The intent of
@invalidate_gpa is to inform vmx_flush_tlb() that it must "invalidate
gpa mappings", i.e. do INVEPT and not simply INVVPID. Other than nested
VPID handling, which now calls vpid_sync_context() directly, the only
scenario where KVM can safely do INVVPID instead of INVEPT (when EPT is
enabled) is if KVM is flushing TLB entries from the guest's perspective,
i.e. is only required to invalidate linear mappings.
For SVM, flushing TLB entries from the guest's perspective can be done
by flushing the current ASID, as changes to the guest's page tables are
associated only with the current ASID.
Adding a dedicated ->tlb_flush_guest() paves the way toward removing
@invalidate_gpa, which is a potentially dangerous control flag as its
meaning is not exactly crystal clear, even for those who are familiar
with the subtleties of what mappings Intel CPUs are/aren't allowed to
keep across various invalidation scenarios.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-15-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Wrap the combination of mmu->invlpg and kvm_x86_ops->tlb_flush_gva
into a new function. This function also lets us specify the host PGD to
invalidate and also the MMU, both of which will be useful in fixing and
simplifying kvm_inject_emulated_page_fault.
A nested guest's MMU however has g_context->invlpg == NULL. Instead of
setting it to nonpaging_invlpg, make kvm_mmu_invalidate_gva the only
entry point to mmu->invlpg and make a NULL invlpg pointer equivalent
to nonpaging_invlpg, saving a retpoline.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Export the page fault propagation helper so that VMX can use it to
correctly emulate TLB invalidation on page faults in an upcoming patch.
In the (hopefully) not-too-distant future, SGX virtualization will also
want access to the helper for injecting page faults to the correct level
(L1 vs. L2) when emulating ENCLS instructions.
Rename the function to kvm_inject_emulated_page_fault() to clarify that
it is (a) injecting a fault and (b) only for page faults. WARN if it's
invoked with an exception other than PF_VECTOR.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-6-sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Remove the __exit annotation from VMX hardware_unsetup(), the hook
can be reached during kvm_init() by way of kvm_arch_hardware_unsetup()
if failure occurs at various points during initialization.
Removing the annotation also lets us annotate vmx_x86_ops and svm_x86_ops
with __initdata; otherwise, objtool complains because it doesn't
understand that the vendor specific __initdata is being copied by value
to a non-__initdata instance.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200321202603.19355-8-sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Replace the kvm_x86_ops pointer in common x86 with an instance of the
struct to save one pointer dereference when invoking functions. Copy the
struct by value to set the ops during kvm_init().
Arbitrarily use kvm_x86_ops.hardware_enable to track whether or not the
ops have been initialized, i.e. a vendor KVM module has been loaded.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200321202603.19355-7-sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the kvm_x86_ops functions that are used only within the scope of
kvm_init() into a separate struct, kvm_x86_init_ops. In addition to
identifying the init-only functions without restorting to code comments,
this also sets the stage for waiting until after ->hardware_setup() to
set kvm_x86_ops. Setting kvm_x86_ops after ->hardware_setup() is
desirable as many of the hooks are not usable until ->hardware_setup()
completes.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200321202603.19355-3-sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The set_cr3 callback is not setting the guest CR3, it is setting the
root of the guest page tables, either shadow or two-dimensional.
To make this clearer as well as to indicate that the MMU calls it
via kvm_mmu_load_cr3, rename it to load_mmu_pgd.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Similar to what kvm-intel.ko is doing, provide a single callback that
merges svm_set_cr3, set_tdp_cr3 and nested_svm_set_tdp_cr3.
This lets us unify the set_cr3 and set_tdp_cr3 entries in kvm_x86_ops.
I'm doing that in this same patch because splitting it adds quite a bit
of churn due to the need for forward declarations. For the same reason
the assignment to vcpu->arch.mmu->set_cr3 is moved to kvm_init_shadow_mmu
from init_kvm_softmmu and nested_svm_init_mmu_context.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Handle CPUID 0x8000000A in the main switch in __do_cpuid_func() and drop
->set_supported_cpuid() now that both VMX and SVM implementations are
empty. Like leaf 0x14 (Intel PT) and leaf 0x8000001F (SEV), leaf
0x8000000A is is (obviously) vendor specific but can be queried in
common code while respecting SVM's wishes by querying kvm_cpu_cap_has().
Suggested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move host_efer to common x86 code and use it for CPUID's is_efer_nx() to
avoid constantly re-reading the MSR.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Configure the max page level during hardware setup to avoid a retpoline
in the page fault handler. Drop ->get_lpage_level() as the page fault
handler was the last user.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Combine kvm_enable_tdp() and kvm_disable_tdp() into a single function,
kvm_configure_mmu(), in preparation for doing additional configuration
during hardware setup. And because having separate helpers is silly.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use vmx_pt_mode_is_host_guest() in intel_pmu_refresh() instead of
bouncing through kvm_x86_ops->pt_supported, and remove ->pt_supported()
as the PMU code was the last remaining user.
Opportunistically clean up the wording of a comment that referenced
kvm_x86_ops->pt_supported().
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Check for MSR_TSC_AUX virtualization via kvm_cpu_cap_has() and drop
->rdtscp_supported().
Note, vmx_rdtscp_supported() needs to hang around a tiny bit longer due
other usage in VMX code.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Set UMIP in kvm_cpu_caps when it is emulated by VMX, even though the
bit will effectively be dropped by do_host_cpuid(). This allows
checking for UMIP emulation via kvm_cpu_caps instead of a dedicated
kvm_x86_ops callback.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the clearing of the XSAVES CPUID bit into VMX, which has a separate
VMCS control to enable XSAVES in non-root, to eliminate the last ugly
renmant of the undesirable "unsigned f_* = *_supported ? F(*) : 0"
pattern in the common CPUID handling code.
Drop ->xsaves_supported(), CPUID adjustment was the only user.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the setting of the PKU CPUID bit into VMX to eliminate an instance
of the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in
the common CPUID handling code. Drop ->pku_supported(), CPUID
adjustment was the only user.
Note, some AMD CPUs now support PKU, but SVM doesn't yet support
exposing it to a guest.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the INVPCID CPUID adjustments into VMX to eliminate an instance of
the undesirable "unsigned f_* = *_supported ? F(*) : 0" pattern in the
common CPUID handling code. Drop ->invpcid_supported(), CPUID
adjustment was the only user.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Drop the explicit @func param from ->set_supported_cpuid() and instead
pull the CPUID function from the relevant entry. This sets the stage
for hardening guest CPUID updates in future patches, e.g. allows adding
run-time assertions that the CPUID feature being changed is actually
a bit in the referenced CPUID entry.
No functional change intended.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Query supported_xcr0 when checking for MPX support instead of invoking
->mpx_supported() and drop ->mpx_supported() as kvm_mpx_supported() was
its last user. Rename vmx_mpx_supported() to cpu_has_vmx_mpx() to
better align with VMX/VMCS nomenclature.
Modify VMX's adjustment of xcr0 to call cpus_has_vmx_mpx() (renamed from
vmx_mpx_supported()) directly to avoid reading supported_xcr0 before
it's fully configured.
No functional change intended.
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
[Test that *all* bits are set. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Now that the emulation context is dynamically allocated and not embedded
in struct kvm_vcpu, move its header, kvm_emulate.h, out of the public
asm directory and into KVM's private x86 directory.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Allocate the emulation context instead of embedding it in struct
kvm_vcpu_arch.
Dynamic allocation provides several benefits:
- Shrinks the size x86 vcpus by ~2.5k bytes, dropping them back below
the PAGE_ALLOC_COSTLY_ORDER threshold.
- Allows for dropping the include of kvm_emulate.h from asm/kvm_host.h
and moving kvm_emulate.h into KVM's private directory.
- Allows a reducing KVM's attack surface by shrinking the amount of
vCPU data that is exposed to usercopy.
- Allows a future patch to disable the emulator entirely, which may or
may not be a realistic endeavor.
Mark the entire struct as valid for usercopy to maintain existing
behavior with respect to hardened usercopy. Future patches can shrink
the usercopy range to cover only what is necessary.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Explicitly pass an exception struct when checking for intercept from
the emulator, which eliminates the last reference to arch.emulate_ctxt
in vendor specific code.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rename kvm_mmu->get_cr3() to call out that it is retrieving a guest
value, as opposed to kvm_mmu->set_cr3(), which sets a host value, and to
note that it will return something other than CR3 when nested EPT is in
use. Hopefully the new name will also make it more obvious that L1's
nested_cr3 is returned in SVM's nested NPT case.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Drop kvm_mmu_extended_role.cr4_la57 now that mmu_role doesn't mask off
level, which already incorporates the guest's CR4.LA57 for a shadow MMU
by querying is_la57_mode().
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Return true for vmx_interrupt_allowed() if the vCPU is in L2 and L1 has
external interrupt exiting enabled. IRQs are never blocked in hardware
if the CPU is in the guest (L2 from L1's perspective) when IRQs trigger
VM-Exit.
The new check percolates up to kvm_vcpu_ready_for_interrupt_injection()
and thus vcpu_run(), and so KVM will exit to userspace if userspace has
requested an interrupt window (to inject an IRQ into L1).
Remove the @external_intr param from vmx_check_nested_events(), which is
actually an indicator that userspace wants an interrupt window, e.g.
it's named @req_int_win further up the stack. Injecting a VM-Exit into
L1 to try and bounce out to L0 userspace is all kinds of broken and is
no longer necessary.
Remove the hack in nested_vmx_vmexit() that attempted to workaround the
breakage in vmx_check_nested_events() by only filling interrupt info if
there's an actual interrupt pending. The hack actually made things
worse because it caused KVM to _never_ fill interrupt info when the
LAPIC resides in userspace (kvm_cpu_has_interrupt() queries
interrupt.injected, which is always cleared by prepare_vmcs12() before
reaching the hack in nested_vmx_vmexit()).
Fixes: 6550c4df7e ("KVM: nVMX: Fix interrupt window request with "Acknowledge interrupt on exit"")
Cc: stable@vger.kernel.org
Cc: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In the vCPU reset and set APIC_BASE MSR path, the apic map will be recalculated
several times, each time it will consume 10+ us observed by ftrace in my
non-overcommit environment since the expensive memory allocate/mutex/rcu etc
operations. This patch optimizes it by recaluating apic map in batch, I hope
this can benefit the serverless scenario which can frequently create/destroy
VMs.
Before patch:
kvm_lapic_reset ~27us
After patch:
kvm_lapic_reset ~14us
Observed by ftrace, improve ~48%.
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It could take kvm->mmu_lock for an extended period of time when
enabling dirty log for the first time. The main cost is to clear
all the D-bits of last level SPTEs. This situation can benefit from
manual dirty log protect as well, which can reduce the mmu_lock
time taken. The sequence is like this:
1. Initialize all the bits of the dirty bitmap to 1 when enabling
dirty log for the first time
2. Only write protect the huge pages
3. KVM_GET_DIRTY_LOG returns the dirty bitmap info
4. KVM_CLEAR_DIRTY_LOG will clear D-bit for each of the leaf level
SPTEs gradually in small chunks
Under the Intel(R) Xeon(R) Gold 6152 CPU @ 2.10GHz environment,
I did some tests with a 128G windows VM and counted the time taken
of memory_global_dirty_log_start, here is the numbers:
VM Size Before After optimization
128G 460ms 10ms
Signed-off-by: Jay Zhou <jianjay.zhou@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The AVIC does not support guest use of the x2APIC interface. Currently,
KVM simply chooses to squash the x2APIC feature in the guest's CPUID
If the AVIC is enabled. Doing so prevents KVM from running a guest
with greater than 255 vCPUs, as such a guest necessitates the use
of the x2APIC interface.
Instead, inhibit AVIC enablement on a per-VM basis whenever the x2APIC
feature is set in the guest's CPUID.
Signed-off-by: Oliver Upton <oupton@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the VM allocation and free code to common x86 as the logic is
more or less identical across SVM and VMX.
Note, although hyperv.hv_pa_pg is part of the common kvm->arch, it's
(currently) only allocated by VMX VMs. But, since kfree() plays nice
when passed a NULL pointer, the superfluous call for SVM is harmless
and avoids future churn if SVM gains support for HyperV's direct TLB
flush.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
[Make vm_size a field instead of a function. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the GPA tracking into the emulator context now that the context is
guaranteed to be initialized via __init_emulate_ctxt() prior to
dereferencing gpa_{available,val}, i.e. now that seeing a stale
gpa_available will also trigger a WARN due to an invalid context.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a new emulation type flag to explicitly mark emulation related to a
page fault. Move the propation of the GPA into the emulator from the
page fault handler into x86_emulate_instruction, using EMULTYPE_PF as an
indicator that cr2 is valid. Similarly, don't propagate cr2 into the
exception.address when it's *not* valid.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Since commit 5f3d45e7f2 ("kvm/x86: add support for
MONITOR_TRAP_FLAG"), KVM has allowed an L1 guest to use the monitor trap
flag processor-based execution control for its L2 guest. KVM simply
forwards any MTF VM-exits to the L1 guest, which works for normal
instruction execution.
However, when KVM needs to emulate an instruction on the behalf of an L2
guest, the monitor trap flag is not emulated. Add the necessary logic to
kvm_skip_emulated_instruction() to synthesize an MTF VM-exit to L1 upon
instruction emulation for L2.
Fixes: 5f3d45e7f2 ("kvm/x86: add support for MONITOR_TRAP_FLAG")
Signed-off-by: Oliver Upton <oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Even when APICv is disabled for L1 it can (and, actually, is) still
available for L2, this means we need to always call
vmx_deliver_nested_posted_interrupt() when attempting an interrupt
delivery.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Fix some typos in the comments. Also fix coding style.
[Sean Christopherson rewrites the comment of write_fault_to_shadow_pgtable
field in struct kvm_vcpu_arch.]
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
AMD SVM AVIC accelerates EOI write and does not trap. This causes
in-kernel PIT re-injection mode to fail since it relies on irq-ack
notifier mechanism. So, APICv is activated only when in-kernel PIT
is in discard mode e.g. w/ qemu option:
-global kvm-pit.lost_tick_policy=discard
Also, introduce APICV_INHIBIT_REASON_PIT_REINJ bit to be used for this
reason.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
AMD AVIC does not support ExtINT. Therefore, AVIC must be temporary
deactivated and fall back to using legacy interrupt injection via vINTR
and interrupt window.
Also, introduce APICV_INHIBIT_REASON_IRQWIN to be used for this reason.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
[Rename svm_request_update_avic to svm_toggle_avic_for_extint. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Since AVIC does not currently work w/ nested virtualization,
deactivate AVIC for the guest if setting CPUID Fn80000001_ECX[SVM]
(i.e. indicate support for SVM, which is needed for nested virtualization).
Also, introduce a new APICV_INHIBIT_REASON_NESTED bit to be used for
this reason.
Suggested-by: Alexander Graf <graf@amazon.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Since disabling APICv has to be done for all vcpus on AMD-based
system, adopt the newly introduced kvm_request_apicv_update()
interface, and introduce a new APICV_INHIBIT_REASON_HYPERV.
Also, remove the kvm_vcpu_deactivate_apicv() since no longer used.
Cc: Roman Kagan <rkagan@virtuozzo.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
AMD SVM AVIC needs to update APIC backing page mapping before changing
APICv mode. Introduce struct kvm_x86_ops.pre_update_apicv_exec_ctrl
function hook to be called prior KVM APICv update request to each vcpu.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Inibit reason bits are used to determine if APICv deactivation is
applicable for a particular hardware virtualization architecture.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Certain runtime conditions require APICv to be temporary deactivated
during runtime. The current implementation only support run-time
deactivation of APICv when Hyper-V SynIC is enabled, which is not
temporary.
In addition, for AMD, when APICv is (de)activated at runtime,
all vcpus in the VM have to operate in the same mode. Thus the
requesting vcpu must notify the others.
So, introduce the following:
* A new KVM_REQ_APICV_UPDATE request bit
* Interfaces to request all vcpus to update APICv status
* A new interface to update APICV-related parameters for each vcpu
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
There are several reasons in which a VM needs to deactivate APICv
e.g. disable APICv via parameter during module loading, or when
enable Hyper-V SynIC support. Additional inhibit reasons will be
introduced later on when dynamic APICv is supported,
Introduce KVM APICv inhibit reason bits along with a new variable,
apicv_inhibit_reasons, to help keep track of APICv state for each VM,
Initially, the APICV_INHIBIT_REASON_DISABLE bit is used to indicate
the case where APICv is disabled during KVM module load.
(e.g. insmod kvm_amd avic=0 or insmod kvm_intel enable_apicv=0).
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
[Do not use get_enable_apicv; consider irqchip_split in svm.c. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
From Boris Ostrovsky:
The KVM hypervisor may provide a guest with ability to defer remote TLB
flush when the remote VCPU is not running. When this feature is used,
the TLB flush will happen only when the remote VPCU is scheduled to run
again. This will avoid unnecessary (and expensive) IPIs.
Under certain circumstances, when a guest initiates such deferred action,
the hypervisor may miss the request. It is also possible that the guest
may mistakenly assume that it has already marked remote VCPU as needing
a flush when in fact that request had already been processed by the
hypervisor. In both cases this will result in an invalid translation
being present in a vCPU, potentially allowing accesses to memory locations
in that guest's address space that should not be accessible.
Note that only intra-guest memory is vulnerable.
The five patches address both of these problems:
1. The first patch makes sure the hypervisor doesn't accidentally clear
a guest's remote flush request
2. The rest of the patches prevent the race between hypervisor
acknowledging a remote flush request and guest issuing a new one.
Conflicts:
arch/x86/kvm/x86.c [move from kvm_arch_vcpu_free to kvm_arch_vcpu_destroy]
Now that we are mapping kvm_steal_time from the guest directly we
don't need keep a copy of it in kvm_vcpu_arch.st. The same is true
for the stime field.
This is part of CVE-2019-3016.
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
__kvm_map_gfn()'s call to gfn_to_pfn_memslot() is
* relatively expensive
* in certain cases (such as when done from atomic context) cannot be called
Stashing gfn-to-pfn mapping should help with both cases.
This is part of CVE-2019-3016.
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The helper x86_set_memory_region() is only used in vmx_set_tss_addr()
and kvm_arch_destroy_vm(). Push the lock upper in both cases. With
that, drop x86_set_memory_region().
This prepares to allow __x86_set_memory_region() to return a HVA
mapped, because the HVA will need to be protected by the lock too even
after __x86_set_memory_region() returns.
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Current SVM implementation does not have support for handling PKU. Guests
running on a host with future AMD cpus that support the feature will read
garbage from the PKRU register and will hit segmentation faults on boot as
memory is getting marked as protected that should not be. Ensure that cpuid
from SVM does not advertise the feature.
Signed-off-by: John Allen <john.allen@amd.com>
Cc: stable@vger.kernel.org
Fixes: 0556cbdc2f ("x86/pkeys: Don't check if PKRU is zero before writing it")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the kvm_cpu_{un}init() calls to common x86 code as an intermediate
step to removing kvm_cpu_{un}init() altogether.
Note, VMX'x alloc_apic_access_page() and init_rmode_identity_map() are
per-VM allocations and are intentionally kept if vCPU creation fails.
They are freed by kvm_arch_destroy_vm().
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move allocation of VMX and SVM vcpus to common x86. Although the struct
being allocated is technically a VMX/SVM struct, it can be interpreted
directly as a 'struct kvm_vcpu' because of the pre-existing requirement
that 'struct kvm_vcpu' be located at offset zero of the arch/vendor vcpu
struct.
Remove the message from the build-time assertions regarding placement of
the struct, as compatibility with the arch usercopy region is no longer
the sole dependent on 'struct kvm_vcpu' being at offset zero.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
ICR and TSCDEADLINE MSRs write cause the main MSRs write vmexits in our
product observation, multicast IPIs are not as common as unicast IPI like
RESCHEDULE_VECTOR and CALL_FUNCTION_SINGLE_VECTOR etc.
This patch introduce a mechanism to handle certain performance-critical
WRMSRs in a very early stage of KVM VMExit handler.
This mechanism is specifically used for accelerating writes to x2APIC ICR
that attempt to send a virtual IPI with physical destination-mode, fixed
delivery-mode and single target. Which was found as one of the main causes
of VMExits for Linux workloads.
The reason this mechanism significantly reduce the latency of such virtual
IPIs is by sending the physical IPI to the target vCPU in a very early stage
of KVM VMExit handler, before host interrupts are enabled and before expensive
operations such as reacquiring KVM’s SRCU lock.
Latency is reduced even more when KVM is able to use APICv posted-interrupt
mechanism (which allows to deliver the virtual IPI directly to target vCPU
without the need to kick it to host).
Testing on Xeon Skylake server:
The virtual IPI latency from sender send to receiver receive reduces
more than 200+ cpu cycles.
Reviewed-by: Liran Alon <liran.alon@oracle.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Convert a plethora of parameters and variables in the MMU and page fault
flows from type gva_t to gpa_t to properly handle TDP on 32-bit KVM.
Thanks to PSE and PAE paging, 32-bit kernels can access 64-bit physical
addresses. When TDP is enabled, the fault address is a guest physical
address and thus can be a 64-bit value, even when both KVM and its guest
are using 32-bit virtual addressing, e.g. VMX's VMCS.GUEST_PHYSICAL is a
64-bit field, not a natural width field.
Using a gva_t for the fault address means KVM will incorrectly drop the
upper 32-bits of the GPA. Ditto for gva_to_gpa() when it is used to
translate L2 GPAs to L1 GPAs.
Opportunistically rename variables and parameters to better reflect the
dual address modes, e.g. use "cr2_or_gpa" for fault addresses and plain
"addr" instead of "vaddr" when the address may be either a GVA or an L2
GPA. Similarly, use "gpa" in the nonpaging_page_fault() flows to avoid
a confusing "gpa_t gva" declaration; this also sets the stage for a
future patch to combing nonpaging_page_fault() and tdp_page_fault() with
minimal churn.
Sprinkle in a few comments to document flows where an address is known
to be a GVA and thus can be safely truncated to a 32-bit value. Add
WARNs in kvm_handle_page_fault() and FNAME(gva_to_gpa_nested)() to help
document such cases and detect bugs.
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We were using either APIC_DEST_PHYSICAL|APIC_DEST_LOGICAL or 0|1 to
fill in kvm_lapic_irq.dest_mode. It's fine only because in most cases
when we check against dest_mode it's against APIC_DEST_PHYSICAL (which
equals to 0). However, that's not consistent. We'll have problem
when we want to start checking against APIC_DEST_LOGICAL, which does
not equals to 1.
This patch firstly introduces kvm_lapic_irq_dest_mode() helper to take
any boolean of destination mode and return the APIC_DEST_* macro.
Then, it replaces the 0|1 settings of irq.dest_mode with the helper.
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Because KVM always emulates CPUID, the CPUID clear bit
(bit 1) of MSR_IA32_TSX_CTRL must be emulated "manually"
by the hypervisor when performing said emulation.
Right now neither kvm-intel.ko nor kvm-amd.ko implement
MSR_IA32_TSX_CTRL but this will change in the next patch.
Reviewed-by: Jim Mattson <jmattson@google.com>
Tested-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In IOAPIC fixed delivery mode instead of flushing the scan
requests to all vCPUs, we should only send the requests to
vCPUs specified within the destination field.
This patch introduces kvm_get_dest_vcpus_mask() API which
retrieves an array of target vCPUs by using
kvm_apic_map_get_dest_lapic() and then based on the
vcpus_idx, it sets the bit in a bitmap. However, if the above
fails kvm_get_dest_vcpus_mask() finds the target vCPUs by
traversing all available vCPUs. Followed by setting the
bits in the bitmap.
If we had different vCPUs in the previous request for the
same redirection table entry then bits corresponding to
these vCPUs are also set. This to done to keep
ioapic_handled_vectors synchronized.
This bitmap is then eventually passed on to
kvm_make_vcpus_request_mask() to generate a masked request
only for the target vCPUs.
This would enable us to reduce the latency overhead on isolated
vCPUs caused by the IPI to process due to KVM_REQ_IOAPIC_SCAN.
Suggested-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Nitesh Narayan Lal <nitesh@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Currently, a host perf_event is created for a vPMC functionality emulation.
It’s unpredictable to determine if a disabled perf_event will be reused.
If they are disabled and are not reused for a considerable period of time,
those obsolete perf_events would increase host context switch overhead that
could have been avoided.
If the guest doesn't WRMSR any of the vPMC's MSRs during an entire vcpu
sched time slice, and its independent enable bit of the vPMC isn't set,
we can predict that the guest has finished the use of this vPMC, and then
do request KVM_REQ_PMU in kvm_arch_sched_in and release those perf_events
in the first call of kvm_pmu_handle_event() after the vcpu is scheduled in.
This lazy mechanism delays the event release time to the beginning of the
next scheduled time slice if vPMC's MSRs aren't changed during this time
slice. If guest comes back to use this vPMC in next time slice, a new perf
event would be re-created via perf_event_create_kernel_counter() as usual.
Suggested-by: Wei Wang <wei.w.wang@intel.com>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The perf_event_create_kernel_counter() in the pmc_reprogram_counter() is
a heavyweight and high-frequency operation, especially when host disables
the watchdog (maximum 21000000 ns) which leads to an unacceptable latency
of the guest NMI handler. It limits the use of vPMUs in the guest.
When a vPMC is fully enabled, the legacy reprogram_*_counter() would stop
and release its existing perf_event (if any) every time EVEN in most cases
almost the same requested perf_event will be created and configured again.
For each vPMC, if the reuqested config ('u64 eventsel' for gp and 'u8 ctrl'
for fixed) is the same as its current config AND a new sample period based
on pmc->counter is accepted by host perf interface, the current event could
be reused safely as a new created one does. Otherwise, do release the
undesirable perf_event and reprogram a new one as usual.
It's light-weight to call pmc_pause_counter (disable, read and reset event)
and pmc_resume_counter (recalibrate period and re-enable event) as guest
expects instead of release-and-create again on any condition. Compared to
use the filterable event->attr or hw.config, a new 'u64 current_config'
field is added to save the last original programed config for each vPMC.
Based on this implementation, the number of calls to pmc_reprogram_counter
is reduced by ~82.5% for a gp sampling event and ~99.9% for a fixed event.
In the usage of multiplexing perf sampling mode, the average latency of the
guest NMI handler is reduced from 104923 ns to 48393 ns (~2.16x speed up).
If host disables watchdog, the minimum latecy of guest NMI handler could be
speed up at ~3413x (from 20407603 to 5979 ns) and at ~786x in the average.
Suggested-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The page table pages corresponding to broken down large pages are zapped in
FIFO order, so that the large page can potentially be recovered, if it is
not longer being used for execution. This removes the performance penalty
for walking deeper EPT page tables.
By default, one large page will last about one hour once the guest
reaches a steady state.
Signed-off-by: Junaid Shahid <junaids@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
With some Intel processors, putting the same virtual address in the TLB
as both a 4 KiB and 2 MiB page can confuse the instruction fetch unit
and cause the processor to issue a machine check resulting in a CPU lockup.
Unfortunately when EPT page tables use huge pages, it is possible for a
malicious guest to cause this situation.
Add a knob to mark huge pages as non-executable. When the nx_huge_pages
parameter is enabled (and we are using EPT), all huge pages are marked as
NX. If the guest attempts to execute in one of those pages, the page is
broken down into 4K pages, which are then marked executable.
This is not an issue for shadow paging (except nested EPT), because then
the host is in control of TLB flushes and the problematic situation cannot
happen. With nested EPT, again the nested guest can cause problems shadow
and direct EPT is treated in the same way.
[ tglx: Fixup default to auto and massage wording a bit ]
Originally-by: Junaid Shahid <junaids@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
If the "virtualize APIC accesses" VM-execution control is set in the
VMCS, the APIC virtualization hardware is triggered when a page walk
in VMX non-root mode terminates at a PTE wherein the address of the 4k
page frame matches the APIC-access address specified in the VMCS. On
hardware, the APIC-access address may be any valid 4k-aligned physical
address.
KVM's nVMX implementation enforces the additional constraint that the
APIC-access address specified in the vmcs12 must be backed by
a "struct page" in L1. If not, L0 will simply clear the "virtualize
APIC accesses" VM-execution control in the vmcs02.
The problem with this approach is that the L1 guest has arranged the
vmcs12 EPT tables--or shadow page tables, if the "enable EPT"
VM-execution control is clear in the vmcs12--so that the L2 guest
physical address(es)--or L2 guest linear address(es)--that reference
the L2 APIC map to the APIC-access address specified in the
vmcs12. Without the "virtualize APIC accesses" VM-execution control in
the vmcs02, the APIC accesses in the L2 guest will directly access the
APIC-access page in L1.
When there is no mapping whatsoever for the APIC-access address in L1,
the L2 VM just loses the intended APIC virtualization. However, when
the APIC-access address is mapped to an MMIO region in L1, the L2
guest gets direct access to the L1 MMIO device. For example, if the
APIC-access address specified in the vmcs12 is 0xfee00000, then L2
gets direct access to L1's APIC.
Since this vmcs12 configuration is something that KVM cannot
faithfully emulate, the appropriate response is to exit to userspace
with KVM_INTERNAL_ERROR_EMULATION.
Fixes: fe3ef05c75 ("KVM: nVMX: Prepare vmcs02 from vmcs01 and vmcs12")
Reported-by: Dan Cross <dcross@google.com>
Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Peter Shier <pshier@google.com>
Reviewed-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cache whether XSAVES is enabled in the guest by adding xsaves_enabled to
vcpu->arch.
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: If4638e0901c28a4494dad2e103e2c075e8ab5d68
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Replace the explicit declaration of "u64 reprogram_pmi" with the generic
macro DECLARE_BITMAP for all possible appropriate number of bits.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Generally, APICv for all vcpus in the VM are enable/disable in the same
manner. So, get_enable_apicv() should represent APICv status of the VM
instead of each VCPU.
Modify kvm_x86_ops.get_enable_apicv() to take struct kvm as parameter
instead of struct kvm_vcpu.
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Handle caching CR3 (from VMX's VMCS) into struct kvm_vcpu via the common
cache_reg() callback and drop the dedicated decache_cr3(). The name
decache_cr3() is somewhat confusing as the caching behavior of CR3
follows that of GPRs, RFLAGS and PDPTRs, (handled via cache_reg()), and
has nothing in common with the caching behavior of CR0/CR4 (whose
decache_cr{0,4}_guest_bits() likely provided the 'decache' verbiage).
This would effectivel adds a BUG() if KVM attempts to cache CR3 on SVM.
Change it to a WARN_ON_ONCE() -- if the cache never requires filling,
the value is already in the right place -- and opportunistically add one
in VMX to provide an equivalent check.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Now that indexing into arch.regs is either protected by WARN_ON_ONCE or
done with hardcoded enums, combine all definitions for registers that
are tracked by regs_avail and regs_dirty into 'enum kvm_reg'. Having a
single enum type will simplify additional cleanup related to regs_avail
and regs_dirty.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Currently, we are overloading SPTE_SPECIAL_MASK to mean both
"A/D bits unavailable" and MMIO, where the difference between the
two is determined by mio_mask and mmio_value.
However, the next patch will need two bits to distinguish
availability of A/D bits from write protection. So, while at
it give MMIO its own bit pattern, and move the two bits from
bit 62 to bits 52..53 since Intel is allocating EPT page table
bits from the top.
Reviewed-by: Junaid Shahid <junaids@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
