linux-2.6-microblaze.git
3 years agoMerge branch 'kvm-amd-fixes' into HEAD
Paolo Bonzini [Wed, 13 May 2020 16:14:05 +0000 (12:14 -0400)]
Merge branch 'kvm-amd-fixes' into HEAD

3 years agoKVM: x86: Fix pkru save/restore when guest CR4.PKE=0, move it to x86.c
Babu Moger [Tue, 12 May 2020 23:59:06 +0000 (18:59 -0500)]
KVM: x86: Fix pkru save/restore when guest CR4.PKE=0, move it to x86.c

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>
3 years agoKVM: SVM: Disable AVIC before setting V_IRQ
Suravee Suthikulpanit [Thu, 7 May 2020 02:35:39 +0000 (21:35 -0500)]
KVM: SVM: Disable AVIC before setting V_IRQ

The commit 64b5bd270426 ("KVM: nSVM: ignore L1 interrupt window
while running L2 with V_INTR_MASKING=1") introduced a WARN_ON,
which checks if AVIC is enabled when trying to set V_IRQ
in the VMCB for enabling irq window.

The following warning is triggered because the requesting vcpu
(to deactivate AVIC) does not get to process APICv update request
for itself until the next #vmexit.

WARNING: CPU: 0 PID: 118232 at arch/x86/kvm/svm/svm.c:1372 enable_irq_window+0x6a/0xa0 [kvm_amd]
 RIP: 0010:enable_irq_window+0x6a/0xa0 [kvm_amd]
 Call Trace:
  kvm_arch_vcpu_ioctl_run+0x6e3/0x1b50 [kvm]
  ? kvm_vm_ioctl_irq_line+0x27/0x40 [kvm]
  ? _copy_to_user+0x26/0x30
  ? kvm_vm_ioctl+0xb3e/0xd90 [kvm]
  ? set_next_entity+0x78/0xc0
  kvm_vcpu_ioctl+0x236/0x610 [kvm]
  ksys_ioctl+0x8a/0xc0
  __x64_sys_ioctl+0x1a/0x20
  do_syscall_64+0x58/0x210
  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes by sending APICV update request to all other vcpus, and
immediately update APIC for itself.

Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lkml.org/lkml/2020/5/2/167
Fixes: 64b5bd270426 ("KVM: nSVM: ignore L1 interrupt window while running L2 with V_INTR_MASKING=1")
Message-Id: <1588818939-54264-1-git-send-email-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: Introduce kvm_make_all_cpus_request_except()
Suravee Suthikulpanit [Wed, 6 May 2020 13:17:53 +0000 (08:17 -0500)]
KVM: Introduce kvm_make_all_cpus_request_except()

This allows making request to all other vcpus except the one
specified in the parameter.

Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Message-Id: <1588771076-73790-2-git-send-email-suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: pass correct DR6 for GD userspace exit
Paolo Bonzini [Wed, 6 May 2020 09:59:39 +0000 (05:59 -0400)]
KVM: VMX: pass correct DR6 for GD userspace exit

When KVM_EXIT_DEBUG is raised for the disabled-breakpoints case (DR7.GD),
DR6 was incorrectly copied from the value in the VM.  Instead,
DR6.BD should be set in order to catch this case.

On AMD this does not need any special code because the processor triggers
a #DB exception that is intercepted.  However, the testcase would fail
without the previous patch because both DR6.BS and DR6.BD would be set.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86, SVM: isolate vcpu->arch.dr6 from vmcb->save.dr6
Paolo Bonzini [Wed, 6 May 2020 10:40:04 +0000 (06:40 -0400)]
KVM: x86, SVM: isolate vcpu->arch.dr6 from vmcb->save.dr6

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>
3 years agoKVM: SVM: keep DR6 synchronized with vcpu->arch.dr6
Paolo Bonzini [Mon, 4 May 2020 15:28:25 +0000 (11:28 -0400)]
KVM: SVM: keep DR6 synchronized with vcpu->arch.dr6

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>
3 years agoKVM: nSVM: trap #DB and #BP to userspace if guest debugging is on
Paolo Bonzini [Thu, 7 May 2020 11:20:27 +0000 (07:20 -0400)]
KVM: nSVM: trap #DB and #BP to userspace if guest debugging is on

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Add KVM_SET_GUEST_DEBUG test
Peter Xu [Tue, 5 May 2020 20:50:00 +0000 (16:50 -0400)]
KVM: selftests: Add KVM_SET_GUEST_DEBUG test

Covers fundamental tests for KVM_SET_GUEST_DEBUG. It is very close to the debug
test in kvm-unit-test, but doing it from outside the guest.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200505205000.188252-4-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Fix single-step with KVM_SET_GUEST_DEBUG
Peter Xu [Tue, 5 May 2020 20:49:59 +0000 (16:49 -0400)]
KVM: X86: Fix single-step with KVM_SET_GUEST_DEBUG

When single-step triggered with KVM_SET_GUEST_DEBUG, we should fill in the pc
value with current linear RIP rather than the cached singlestep address.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200505205000.188252-3-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Set RTM for DB_VECTOR too for KVM_EXIT_DEBUG
Peter Xu [Tue, 5 May 2020 20:49:58 +0000 (16:49 -0400)]
KVM: X86: Set RTM for DB_VECTOR too for KVM_EXIT_DEBUG

RTM should always been set even with KVM_EXIT_DEBUG on #DB.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200505205000.188252-2-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: fix DR6 delivery for various cases of #DB injection
Paolo Bonzini [Tue, 5 May 2020 11:33:20 +0000 (07:33 -0400)]
KVM: x86: fix DR6 delivery for various cases of #DB injection

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>
3 years agoKVM: X86: Declare KVM_CAP_SET_GUEST_DEBUG properly
Peter Xu [Tue, 5 May 2020 15:47:50 +0000 (11:47 -0400)]
KVM: X86: Declare KVM_CAP_SET_GUEST_DEBUG properly

KVM_CAP_SET_GUEST_DEBUG should be supported for x86 however it's not declared
as supported.  My wild guess is that userspaces like QEMU are using "#ifdef
KVM_CAP_SET_GUEST_DEBUG" to check for the capability instead, but that could be
wrong because the compilation host may not be the runtime host.

The userspace might still want to keep the old "#ifdef" though to not break the
guest debug on old kernels.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200505154750.126300-1-peterx@redhat.com>
[Do the same for PPC and s390. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Fix build for evmcs.h
Peter Xu [Mon, 4 May 2020 22:06:07 +0000 (18:06 -0400)]
KVM: selftests: Fix build for evmcs.h

I got this error when building kvm selftests:

/usr/bin/ld: /home/xz/git/linux/tools/testing/selftests/kvm/libkvm.a(vmx.o):/home/xz/git/linux/tools/testing/selftests/kvm/include/evmcs.h:222: multiple definition of `current_evmcs'; /tmp/cco1G48P.o:/home/xz/git/linux/tools/testing/selftests/kvm/include/evmcs.h:222: first defined here
/usr/bin/ld: /home/xz/git/linux/tools/testing/selftests/kvm/libkvm.a(vmx.o):/home/xz/git/linux/tools/testing/selftests/kvm/include/evmcs.h:223: multiple definition of `current_vp_assist'; /tmp/cco1G48P.o:/home/xz/git/linux/tools/testing/selftests/kvm/include/evmcs.h:223: first defined here

I think it's because evmcs.h is included both in a test file and a lib file so
the structs have multiple declarations when linking.  After all it's not a good
habit to declare structs in the header files.

Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200504220607.99627-1-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: x86: Use KVM CPU capabilities to determine CR4 reserved bits
Paolo Bonzini [Tue, 5 May 2020 13:40:46 +0000 (09:40 -0400)]
kvm: x86: Use KVM CPU capabilities to determine CR4 reserved bits

Using CPUID data can be useful for the processor compatibility
check, but that's it.  Using it to compute guest-reserved bits
can have both false positives (such as LA57 and UMIP which we
are already handling) and false negatives: in particular, with
this patch we don't allow anymore a KVM guest to set CR4.PKE
when CR4.PKE is clear on the host.

Fixes: b9dd21e104bc ("KVM: x86: simplify handling of PKRU")
Reported-by: Jim Mattson <jmattson@google.com>
Tested-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: Explicitly clear RFLAGS.CF and RFLAGS.ZF in VM-Exit RSB path
Sean Christopherson [Wed, 6 May 2020 03:53:55 +0000 (20:53 -0700)]
KVM: VMX: Explicitly clear RFLAGS.CF and RFLAGS.ZF in VM-Exit RSB path

Clear CF and ZF in the VM-Exit path after doing __FILL_RETURN_BUFFER so
that KVM doesn't interpret clobbered RFLAGS as a VM-Fail.  Filling the
RSB has always clobbered RFLAGS, its current incarnation just happens
clear CF and ZF in the processs.  Relying on the macro to clear CF and
ZF is extremely fragile, e.g. commit 089dd8e53126e ("x86/speculation:
Change FILL_RETURN_BUFFER to work with objtool") tweaks the loop such
that the ZF flag is always set.

Reported-by: Qian Cai <cai@lca.pw>
Cc: Rick Edgecombe <rick.p.edgecombe@intel.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: stable@vger.kernel.org
Fixes: f2fde6a5bcfcf ("KVM: VMX: Move RSB stuffing to before the first RET after VM-Exit")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200506035355.2242-1-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agodocs/virt/kvm: Document configuring and running nested guests
Kashyap Chamarthy [Tue, 5 May 2020 11:28:39 +0000 (13:28 +0200)]
docs/virt/kvm: Document configuring and running nested guests

This is a rewrite of this[1] Wiki page with further enhancements.  The
doc also includes a section on debugging problems in nested
environments, among other improvements.

[1] https://www.linux-kvm.org/page/Nested_Guests

Signed-off-by: Kashyap Chamarthy <kchamart@redhat.com>
Message-Id: <20200505112839.30534-1-kchamart@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: ioapic: Restrict lazy EOI update to edge-triggered interrupts
Paolo Bonzini [Mon, 4 May 2020 16:19:45 +0000 (12:19 -0400)]
kvm: ioapic: Restrict lazy EOI update to edge-triggered interrupts

Commit f458d039db7e ("kvm: ioapic: Lazy update IOAPIC EOI") introduces
the following infinite loop:

BUG: stack guard page was hit at 000000008f595917 \
(stack is 00000000bdefe5a4..00000000ae2b06f5)
kernel stack overflow (double-fault): 0000 [#1] SMP NOPTI
RIP: 0010:kvm_set_irq+0x51/0x160 [kvm]
Call Trace:
 irqfd_resampler_ack+0x32/0x90 [kvm]
 kvm_notify_acked_irq+0x62/0xd0 [kvm]
 kvm_ioapic_update_eoi_one.isra.0+0x30/0x120 [kvm]
 ioapic_set_irq+0x20e/0x240 [kvm]
 kvm_ioapic_set_irq+0x5c/0x80 [kvm]
 kvm_set_irq+0xbb/0x160 [kvm]
 ? kvm_hv_set_sint+0x20/0x20 [kvm]
 irqfd_resampler_ack+0x32/0x90 [kvm]
 kvm_notify_acked_irq+0x62/0xd0 [kvm]
 kvm_ioapic_update_eoi_one.isra.0+0x30/0x120 [kvm]
 ioapic_set_irq+0x20e/0x240 [kvm]
 kvm_ioapic_set_irq+0x5c/0x80 [kvm]
 kvm_set_irq+0xbb/0x160 [kvm]
 ? kvm_hv_set_sint+0x20/0x20 [kvm]
....

The re-entrancy happens because the irq state is the OR of
the interrupt state and the resamplefd state.  That is, we don't
want to show the state as 0 until we've had a chance to set the
resamplefd.  But if the interrupt has _not_ gone low then
ioapic_set_irq is invoked again, causing an infinite loop.

This can only happen for a level-triggered interrupt, otherwise
irqfd_inject would immediately set the KVM_USERSPACE_IRQ_SOURCE_ID high
and then low.  Fortunately, in the case of level-triggered interrupts the VMEXIT already happens because
TMR is set.  Thus, fix the bug by restricting the lazy invocation
of the ack notifier to edge-triggered interrupts, the only ones that
need it.

Tested-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Reported-by: borisvk@bstnet.org
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Link: https://www.spinics.net/lists/kvm/msg213512.html
Fixes: f458d039db7e ("kvm: ioapic: Lazy update IOAPIC EOI")
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=207489
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Fixes posted interrupt check for IRQs delivery modes
Suravee Suthikulpanit [Tue, 7 Apr 2020 06:13:09 +0000 (01:13 -0500)]
KVM: x86: Fixes posted interrupt check for IRQs delivery modes

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: (fdcf75621375 '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>
3 years agoMerge tag 'kvmarm-fixes-5.7-2' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmar...
Paolo Bonzini [Mon, 4 May 2020 16:01:37 +0000 (12:01 -0400)]
Merge tag 'kvmarm-fixes-5.7-2' of git://git./linux/kernel/git/kvmarm/kvmarm into kvm-master

KVM/arm fixes for Linux 5.7, take #2

- Fix compilation with Clang
- Correctly initialize GICv4.1 in the absence of a virtual ITS
- Move SP_EL0 save/restore to the guest entry/exit code
- Handle PC wrap around on 32bit guests, and narrow all 32bit
  registers on userspace access

3 years agoMerge tag 'kvmarm-fixes-5.7-1' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmar...
Paolo Bonzini [Mon, 4 May 2020 16:01:12 +0000 (12:01 -0400)]
Merge tag 'kvmarm-fixes-5.7-1' of git://git./linux/kernel/git/kvmarm/kvmarm into kvm-master

KVM/arm fixes for Linux 5.7, take #1

- Prevent the userspace API from interacting directly with the HW
  stage of the virtual GIC
- Fix a couple of vGIC memory leaks
- Tighten the rules around the use of the 32bit PSCI functions
  for 64bit guest, as well as the opposite situation (matches the
  specification)

3 years agoKVM: SVM: fill in kvm_run->debug.arch.dr[67]
Paolo Bonzini [Mon, 4 May 2020 13:34:10 +0000 (09:34 -0400)]
KVM: SVM: fill in kvm_run->debug.arch.dr[67]

The corresponding code was added for VMX in commit 42dbaa5a057
("KVM: x86: Virtualize debug registers, 2008-12-15) but never for AMD.
Fix this.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: nVMX: Replace a BUG_ON(1) with BUG() to squash clang warning
Sean Christopherson [Mon, 4 May 2020 15:35:06 +0000 (08:35 -0700)]
KVM: nVMX: Replace a BUG_ON(1) with BUG() to squash clang warning

Use BUG() in the impossible-to-hit default case when switching on the
scope of INVEPT to squash a warning with clang 11 due to clang treating
the BUG_ON() as conditional.

  >> arch/x86/kvm/vmx/nested.c:5246:3: warning: variable 'roots_to_free'
     is used uninitialized whenever 'if' condition is false
     [-Wsometimes-uninitialized]
                   BUG_ON(1);

Reported-by: kbuild test robot <lkp@intel.com>
Fixes: ce8fe7b77bd8 ("KVM: nVMX: Free only the affected contexts when emulating INVEPT")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200504153506.28898-1-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: arm64: Fix 32bit PC wrap-around
Marc Zyngier [Wed, 29 Apr 2020 10:21:55 +0000 (11:21 +0100)]
KVM: arm64: Fix 32bit PC wrap-around

In the unlikely event that a 32bit vcpu traps into the hypervisor
on an instruction that is located right at the end of the 32bit
range, the emulation of that instruction is going to increment
PC past the 32bit range. This isn't great, as userspace can then
observe this value and get a bit confused.

Conversly, userspace can do things like (in the context of a 64bit
guest that is capable of 32bit EL0) setting PSTATE to AArch64-EL0,
set PC to a 64bit value, change PSTATE to AArch32-USR, and observe
that PC hasn't been truncated. More confusion.

Fix both by:
- truncating PC increments for 32bit guests
- sanitizing all 32bit regs every time a core reg is changed by
  userspace, and that PSTATE indicates a 32bit mode.

Cc: stable@vger.kernel.org
Acked-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
3 years agoKVM: arm64: vgic-v4: Initialize GICv4.1 even in the absence of a virtual ITS
Marc Zyngier [Fri, 24 Apr 2020 14:30:30 +0000 (15:30 +0100)]
KVM: arm64: vgic-v4: Initialize GICv4.1 even in the absence of a virtual ITS

KVM now expects to be able to use HW-accelerated delivery of vSGIs
as soon as the guest has enabled thm. Unfortunately, we only
initialize the GICv4 context if we have a virtual ITS exposed to
the guest.

Fix it by always initializing the GICv4.1 context if it is
available on the host.

Fixes: 2291ff2f2a56 ("KVM: arm64: GICv4.1: Plumb SGI implementation selection in the distributor")
Reviewed-by: Zenghui Yu <yuzenghui@huawei.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
3 years agoKVM: arm64: Save/restore sp_el0 as part of __guest_enter
Marc Zyngier [Fri, 24 Apr 2020 13:24:34 +0000 (14:24 +0100)]
KVM: arm64: Save/restore sp_el0 as part of __guest_enter

We currently save/restore sp_el0 in C code. This is a bit unsafe,
as a lot of the C code expects 'current' to be accessible from
there (and the opportunity to run kernel code in HYP is specially
great with VHE).

Instead, let's move the save/restore of sp_el0 to the assembly
code (in __guest_enter), making sure that sp_el0 is correct
very early on when we exit the guest, and is preserved as long
as possible to its host value when we enter the guest.

Reviewed-by: Andrew Jones <drjones@redhat.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
3 years agoKVM: arm64: Delete duplicated label in invalid_vector
Fangrui Song [Mon, 13 Apr 2020 23:10:16 +0000 (16:10 -0700)]
KVM: arm64: Delete duplicated label in invalid_vector

SYM_CODE_START defines \label , so it is redundant to define \label again.
A redefinition at the same place is accepted by GNU as
(https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=159fbb6088f17a341bcaaac960623cab881b4981)
but rejected by the clang integrated assembler.

Fixes: 617a2f392c92 ("arm64: kvm: Annotate assembly using modern annoations")
Signed-off-by: Fangrui Song <maskray@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Link: https://github.com/ClangBuiltLinux/linux/issues/988
Link: https://lore.kernel.org/r/20200413231016.250737-1-maskray@google.com
4 years agoKVM: SVM: do not allow VMRUN inside SMM
Paolo Bonzini [Thu, 23 Apr 2020 14:52:48 +0000 (10:52 -0400)]
KVM: SVM: do not allow VMRUN inside SMM

VMRUN is not supported inside the SMM handler and the behavior is undefined.
Just raise a #UD.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agokvm: add capability for halt polling
David Matlack [Fri, 17 Apr 2020 22:14:46 +0000 (15:14 -0700)]
kvm: add capability for halt polling

KVM_CAP_HALT_POLL is a per-VM capability that lets userspace
control the halt-polling time, allowing halt-polling to be tuned or
disabled on particular VMs.

With dynamic halt-polling, a VM's VCPUs can poll from anywhere from
[0, halt_poll_ns] on each halt. KVM_CAP_HALT_POLL sets the
upper limit on the poll time.

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: <20200417221446.108733-1-jcargill@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Store vmcs.EXIT_QUALIFICATION as an unsigned long, not u32
Sean Christopherson [Thu, 23 Apr 2020 00:11:27 +0000 (17:11 -0700)]
KVM: nVMX: Store vmcs.EXIT_QUALIFICATION as an unsigned long, not u32

Use an unsigned long for 'exit_qual' in nested_vmx_reflect_vmexit(), the
EXIT_QUALIFICATION field is naturally sized, not a 32-bit field.

The bug is most easily observed by doing VMXON (or any VMX instruction)
in L2 with a negative displacement, in which case dropping the upper
bits on nested VM-Exit results in L1 calculating the wrong virtual
address for the memory operand, e.g. "vmxon -0x8(%rbp)" yields:

  Unhandled cpu exception 14 #PF at ip 0000000000400553
  rbp=0000000000537000 cr2=0000000100536ff8

Fixes: fbdd50250396d ("KVM: nVMX: Move VM-Fail check out of nested_vmx_exit_reflected()")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200423001127.13490-1-sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Drop a redundant call to vmx_get_intr_info()
Sean Christopherson [Tue, 21 Apr 2020 07:53:27 +0000 (00:53 -0700)]
KVM: nVMX: Drop a redundant call to vmx_get_intr_info()

Drop nested_vmx_l1_wants_exit()'s initialization of intr_info from
vmx_get_intr_info() that was inadvertantly introduced along with the
caching mechanism.  EXIT_REASON_EXCEPTION_NMI, the only consumer of
intr_info, populates the variable before using it.

Fixes: bb53120d67cd ("KVM: VMX: Cache vmcs.EXIT_INTR_INFO using arch avail_reg flags")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200421075328.14458-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoMerge branch 'kvm-arm64/vgic-fixes-5.7' into kvmarm-master/master
Marc Zyngier [Thu, 23 Apr 2020 15:27:33 +0000 (16:27 +0100)]
Merge branch 'kvm-arm64/vgic-fixes-5.7' into kvmarm-master/master

4 years agoMerge branch 'kvm-arm64/psci-fixes-5.7' into kvmarm-master/master
Marc Zyngier [Thu, 23 Apr 2020 15:27:26 +0000 (16:27 +0100)]
Merge branch 'kvm-arm64/psci-fixes-5.7' into kvmarm-master/master

4 years agoKVM: arm64: vgic-its: Fix memory leak on the error path of vgic_add_lpi()
Zenghui Yu [Tue, 14 Apr 2020 03:03:48 +0000 (11:03 +0800)]
KVM: arm64: vgic-its: Fix memory leak on the error path of vgic_add_lpi()

If we're going to fail out the vgic_add_lpi(), let's make sure the
allocated vgic_irq memory is also freed. Though it seems that both
cases are unlikely to fail.

Signed-off-by: Zenghui Yu <yuzenghui@huawei.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20200414030349.625-3-yuzenghui@huawei.com
4 years agoKVM: arm64: vgic-v3: Retire all pending LPIs on vcpu destroy
Zenghui Yu [Tue, 14 Apr 2020 03:03:47 +0000 (11:03 +0800)]
KVM: arm64: vgic-v3: Retire all pending LPIs on vcpu destroy

It's likely that the vcpu fails to handle all virtual interrupts if
userspace decides to destroy it, leaving the pending ones stay in the
ap_list. If the un-handled one is a LPI, its vgic_irq structure will
be eventually leaked because of an extra refcount increment in
vgic_queue_irq_unlock().

This was detected by kmemleak on almost every guest destroy, the
backtrace is as follows:

unreferenced object 0xffff80725aed5500 (size 128):
comm "CPU 5/KVM", pid 40711, jiffies 4298024754 (age 166366.512s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 08 01 a9 73 6d 80 ff ff ...........sm...
c8 61 ee a9 00 20 ff ff 28 1e 55 81 6c 80 ff ff .a... ..(.U.l...
backtrace:
[<000000004bcaa122>] kmem_cache_alloc_trace+0x2dc/0x418
[<0000000069c7dabb>] vgic_add_lpi+0x88/0x418
[<00000000bfefd5c5>] vgic_its_cmd_handle_mapi+0x4dc/0x588
[<00000000cf993975>] vgic_its_process_commands.part.5+0x484/0x1198
[<000000004bd3f8e3>] vgic_its_process_commands+0x50/0x80
[<00000000b9a65b2b>] vgic_mmio_write_its_cwriter+0xac/0x108
[<0000000009641ebb>] dispatch_mmio_write+0xd0/0x188
[<000000008f79d288>] __kvm_io_bus_write+0x134/0x240
[<00000000882f39ac>] kvm_io_bus_write+0xe0/0x150
[<0000000078197602>] io_mem_abort+0x484/0x7b8
[<0000000060954e3c>] kvm_handle_guest_abort+0x4cc/0xa58
[<00000000e0d0cd65>] handle_exit+0x24c/0x770
[<00000000b44a7fad>] kvm_arch_vcpu_ioctl_run+0x460/0x1988
[<0000000025fb897c>] kvm_vcpu_ioctl+0x4f8/0xee0
[<000000003271e317>] do_vfs_ioctl+0x160/0xcd8
[<00000000e7f39607>] ksys_ioctl+0x98/0xd8

Fix it by retiring all pending LPIs in the ap_list on the destroy path.

p.s. I can also reproduce it on a normal guest shutdown. It is because
userspace still send LPIs to vcpu (through KVM_SIGNAL_MSI ioctl) while
the guest is being shutdown and unable to handle it. A little strange
though and haven't dig further...

Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Zenghui Yu <yuzenghui@huawei.com>
[maz: moved the distributor deallocation down to avoid an UAF splat]
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20200414030349.625-2-yuzenghui@huawei.com
4 years agoKVM: arm: vgic-v2: Only use the virtual state when userspace accesses pending bits
Marc Zyngier [Thu, 9 Apr 2020 12:05:26 +0000 (13:05 +0100)]
KVM: arm: vgic-v2: Only use the virtual state when userspace accesses pending bits

There is no point in accessing the HW when writing to any of the
ISPENDR/ICPENDR registers from userspace, as only the guest should
be allowed to change the HW state.

Introduce new userspace-specific accessors that deal solely with
the virtual state. Note that the API differs from that of GICv3,
where userspace exclusively uses ISPENDR to set the state. Too
bad we can't reuse it.

Fixes: 82e40f558de56 ("KVM: arm/arm64: vgic-v2: Handle SGI bits in GICD_I{S,C}PENDR0 as WI")
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
4 years agoKVM: x86: move nested-related kvm_x86_ops to a separate struct
Paolo Bonzini [Fri, 17 Apr 2020 14:24:18 +0000 (10:24 -0400)]
KVM: x86: move nested-related kvm_x86_ops to a separate struct

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>
4 years agoKVM: eVMCS: check if nesting is enabled
Paolo Bonzini [Fri, 17 Apr 2020 14:34:29 +0000 (10:34 -0400)]
KVM: eVMCS: check if nesting is enabled

In the next patch nested_get_evmcs_version will be always set in kvm_x86_ops for
VMX, even if nesting is disabled.  Therefore, check whether VMX (aka nesting)
is available in the function, the caller will not do the check anymore.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86: check_nested_events is never NULL
Paolo Bonzini [Fri, 17 Apr 2020 14:32:53 +0000 (10:32 -0400)]
KVM: x86: check_nested_events is never NULL

Both Intel and AMD now implement it, so there is no need to check if the
callback is implemented.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: arm: vgic: Only use the virtual state when userspace accesses enable bits
Marc Zyngier [Thu, 9 Apr 2020 12:05:26 +0000 (13:05 +0100)]
KVM: arm: vgic: Only use the virtual state when userspace accesses enable bits

There is no point in accessing the HW when writing to any of the
ISENABLER/ICENABLER registers from userspace, as only the guest
should be allowed to change the HW state.

Introduce new userspace-specific accessors that deal solely with
the virtual state.

Reported-by: James Morse <james.morse@arm.com>
Tested-by: James Morse <james.morse@arm.com>
Reviewed-by: James Morse <james.morse@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
4 years agoKVM: arm: vgic: Synchronize the whole guest on GIC{D,R}_I{S,C}ACTIVER read
Marc Zyngier [Mon, 6 Apr 2020 15:21:20 +0000 (16:21 +0100)]
KVM: arm: vgic: Synchronize the whole guest on GIC{D,R}_I{S,C}ACTIVER read

When a guest tries to read the active state of its interrupts,
we currently just return whatever state we have in memory. This
means that if such an interrupt lives in a List Register on another
CPU, we fail to obsertve the latest active state for this interrupt.

In order to remedy this, stop all the other vcpus so that they exit
and we can observe the most recent value for the state. This is
similar to what we are doing for the write side of the same
registers, and results in new MMIO handlers for userspace (which
do not need to stop the guest, as it is supposed to be stopped
already).

Reported-by: Julien Grall <julien@xen.org>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
4 years agoMerge tag 'kvm-ppc-fixes-5.7-1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Tue, 21 Apr 2020 13:39:55 +0000 (09:39 -0400)]
Merge tag 'kvm-ppc-fixes-5.7-1' of git://git./linux/kernel/git/paulus/powerpc into kvm-master

PPC KVM fix for 5.7

- Fix a regression introduced in the last merge window, which results
  in guests in HPT mode dying randomly.

4 years agoMerge tag 'kvm-s390-master-5.7-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Tue, 21 Apr 2020 13:37:13 +0000 (09:37 -0400)]
Merge tag 'kvm-s390-master-5.7-2' of git://git./linux/kernel/git/kvms390/linux into kvm-master

KVM: s390: Fix for 5.7 and maintainer update

- Silence false positive lockdep warning
- add Claudio as reviewer

4 years agoselftests: kvm/set_memory_region_test: do not check RIP if the guest shuts down
Paolo Bonzini [Fri, 17 Apr 2020 16:35:38 +0000 (12:35 -0400)]
selftests: kvm/set_memory_region_test: do not check RIP if the guest shuts down

On AMD, the state of the VMCB is undefined after a shutdown VMEXIT.  KVM
takes a very conservative approach to that and resets the guest altogether
when that happens.  This causes the set_memory_region_test to fail
because the RIP is 0xfff0 (the reset vector).  Restrict the RIP test
to KVM_EXIT_INTERNAL_ERROR in order to fix this.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: SVM: avoid infinite loop on NPF from bad address
Paolo Bonzini [Fri, 17 Apr 2020 16:21:06 +0000 (12:21 -0400)]
KVM: SVM: avoid infinite loop on NPF from bad address

When a nested page fault is taken from an address that does not have
a memslot associated to it, kvm_mmu_do_page_fault returns RET_PF_EMULATE
(via mmu_set_spte) and kvm_mmu_page_fault then invokes svm_need_emulation_on_page_fault.

The default answer there is to return false, but in this case this just
causes the page fault to be retried ad libitum.  Since this is not a
fast path, and the only other case where it is taken is an erratum,
just stick a kvm_vcpu_gfn_to_memslot check in there to detect the
common case where the erratum is not happening.

This fixes an infinite loop in the new set_memory_region_test.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agotools/kvm_stat: add sample systemd unit file
Stefan Raspl [Thu, 2 Apr 2020 08:57:05 +0000 (10:57 +0200)]
tools/kvm_stat: add sample systemd unit file

Add a sample unit file as a basis for systemd integration of kvm_stat
logs.

Signed-off-by: Stefan Raspl <raspl@de.ibm.com>
Message-Id: <20200402085705.61155-4-raspl@linux.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agotools/kvm_stat: Add command line switch '-L' to log to file
Stefan Raspl [Thu, 2 Apr 2020 08:57:04 +0000 (10:57 +0200)]
tools/kvm_stat: Add command line switch '-L' to log to file

To integrate with logrotate, we have a signal handler that will re-open
the logfile.
Assuming we have a systemd unit file with
     ExecStart=kvm_stat -dtc -s 10 -L /var/log/kvm_stat.csv
     ExecReload=/bin/kill -HUP $MAINPID
and a logrotate config featuring
     postrotate
        /bin/systemctl reload kvm_stat.service
     endscript
Then the overall flow will look like this:
(1) systemd starts kvm_stat, logging to A.
(2) At some point, logrotate runs, moving A to B.
    kvm_stat continues to write to B at this point.
(3) After rotating, logrotate restarts the kvm_stat unit via systemctl.
(4) The kvm_stat unit sends a SIGHUP to kvm_stat, finally making it
    switch over to writing to A again.
Note that in order to keep the structure of the cvs output in tact, we
make sure to, in contrast to the standard log format, only write the
header once at the beginning of a file. This implies that the header is
suppressed when appending to an existing file. Unlike with the standard
format, where we append to an existing file by starting out with a
header.

Signed-off-by: Stefan Raspl <raspl@de.ibm.com>
Message-Id: <20200402085705.61155-3-raspl@linux.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agotools/kvm_stat: add command line switch '-z' to skip zero records
Stefan Raspl [Thu, 2 Apr 2020 08:57:03 +0000 (10:57 +0200)]
tools/kvm_stat: add command line switch '-z' to skip zero records

When running in logging mode, skip records with all zeros (=empty records)
to preserve space when logging to files.

Signed-off-by: Stefan Raspl <raspl@de.ibm.com>
Message-Id: <20200402085705.61155-2-raspl@linux.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: Remove redundant argument to kvm_arch_vcpu_ioctl_run
Tianjia Zhang [Thu, 16 Apr 2020 05:10:57 +0000 (13:10 +0800)]
KVM: Remove redundant argument to kvm_arch_vcpu_ioctl_run

In earlier versions of kvm, 'kvm_run' was an independent structure
and was not included in the vcpu structure. At present, 'kvm_run'
is already included in the vcpu structure, so the parameter
'kvm_run' is redundant.

This patch simplifies the function definition, removes the extra
'kvm_run' parameter, and extracts it from the 'kvm_vcpu' structure
if necessary.

Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Message-Id: <20200416051057.26526-1-tianjia.zhang@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nSVM: Check for CR0.CD and CR0.NW on VMRUN of nested guests
Krish Sadhukhan [Thu, 9 Apr 2020 20:50:33 +0000 (16:50 -0400)]
KVM: nSVM: Check for CR0.CD and CR0.NW on VMRUN of nested guests

According to section "Canonicalization and Consistency Checks" in APM vol. 2,
the following guest state combination is illegal:

"CR0.CD is zero and CR0.NW is set"

Signed-off-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Message-Id: <20200409205035.16830-2-krish.sadhukhan@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: X86: Improve latency for single target IPI fastpath
Wanpeng Li [Fri, 10 Apr 2020 17:47:03 +0000 (10:47 -0700)]
KVM: X86: Improve latency for single target IPI fastpath

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>
4 years agoKVM: VMX: Optimize handling of VM-Entry failures in vmx_vcpu_run()
Sean Christopherson [Fri, 10 Apr 2020 17:47:02 +0000 (10:47 -0700)]
KVM: VMX: Optimize handling of VM-Entry failures in vmx_vcpu_run()

Mark the VM-Fail, VM-Exit on VM-Enter, and #MC on VM-Enter paths as
'unlikely' so as to improve code generation so that it favors successful
VM-Enter.  The performance of successful VM-Enter is for more important,
irrespective of whether or not success is actually likely.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200410174703.1138-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Remove non-functional "support" for CR3 target values
Sean Christopherson [Thu, 16 Apr 2020 00:07:39 +0000 (17:07 -0700)]
KVM: nVMX: Remove non-functional "support" for CR3 target values

Remove all references to cr3_target_value[0-3] and replace the fields
in vmcs12 with "dead_space" to preserve the vmcs12 layout.  KVM doesn't
support emulating CR3-target values, despite a variety of code that
implies otherwise, as KVM unconditionally reports '0' for the number of
supported CR3-target values.

This technically fixes a bug where KVM would incorrectly allow VMREAD
and VMWRITE to nonexistent fields, i.e. cr3_target_value[0-3].  Per
Intel's SDM, the number of supported CR3-target values reported in
VMX_MISC also enumerates the existence of the associated VMCS fields:

  If a future implementation supports more than 4 CR3-target values, they
  will be encoded consecutively following the 4 encodings given here.

Alternatively, the "bug" could be fixed by actually advertisting support
for 4 CR3-target values, but that'd likely just enable kvm-unit-tests
given that no one has complained about lack of support for going on ten
years, e.g. KVM, Xen and HyperV don't use CR3-target values.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200416000739.9012-1-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86/mmu: Avoid an extra memslot lookup in try_async_pf() for L2
Paolo Bonzini [Thu, 16 Apr 2020 13:48:07 +0000 (09:48 -0400)]
KVM: x86/mmu: Avoid an extra memslot lookup in try_async_pf() for L2

Create a new function kvm_is_visible_memslot() and use it from
kvm_is_visible_gfn(); use the new function in try_async_pf() too,
to avoid an extra memslot lookup.

Opportunistically squish a multi-line comment into a single-line comment.

Note, the end result, KVM_PFN_NOSLOT, is unchanged.

Cc: Jim Mattson <jmattson@google.com>
Cc: Rick Edgecombe <rick.p.edgecombe@intel.com>
Suggested-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86/mmu: Set @writable to false for non-visible accesses by L2
Sean Christopherson [Wed, 15 Apr 2020 21:44:13 +0000 (14:44 -0700)]
KVM: x86/mmu: Set @writable to false for non-visible accesses by L2

Explicitly set @writable to false in try_async_pf() if the GFN->PFN
translation is short-circuited due to the requested GFN not being
visible to L2.

Leaving @writable ('map_writable' in the callers) uninitialized is ok
in that it's never actually consumed, but one has to track it all the
way through set_spte() being short-circuited by set_mmio_spte() to
understand that the uninitialized variable is benign, and relying on
@writable being ignored is an unnecessary risk.  Explicitly setting
@writable also aligns try_async_pf() with __gfn_to_pfn_memslot().

Jim Mattson <jmattson@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415214414.10194-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: VMX: Cache vmcs.EXIT_INTR_INFO using arch avail_reg flags
Sean Christopherson [Wed, 15 Apr 2020 20:34:54 +0000 (13:34 -0700)]
KVM: VMX: Cache vmcs.EXIT_INTR_INFO using arch avail_reg flags

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>
4 years agoKVM: VMX: Cache vmcs.EXIT_QUALIFICATION using arch avail_reg flags
Sean Christopherson [Wed, 15 Apr 2020 20:34:53 +0000 (13:34 -0700)]
KVM: VMX: Cache vmcs.EXIT_QUALIFICATION using arch avail_reg flags

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>
4 years agoKVM: nVMX: Drop manual clearing of segment cache on nested VMCS switch
Sean Christopherson [Wed, 15 Apr 2020 20:34:52 +0000 (13:34 -0700)]
KVM: nVMX: Drop manual clearing of segment cache on nested VMCS switch

Drop the call to vmx_segment_cache_clear() in vmx_switch_vmcs() now that
the entire register cache is reset when switching the active VMCS, e.g.
vmx_segment_cache_test_set() will reset the segment cache due to
VCPU_EXREG_SEGMENTS being unavailable.

Move vmx_segment_cache_clear() to vmx.c now that it's no longer invoked
by the nested code.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415203454.8296-4-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Reset register cache (available and dirty masks) on VMCS switch
Sean Christopherson [Wed, 15 Apr 2020 20:34:51 +0000 (13:34 -0700)]
KVM: nVMX: Reset register cache (available and dirty masks) on VMCS switch

Reset the per-vCPU available and dirty register masks when switching
between vmcs01 and vmcs02, as the masks track state relative to the
current VMCS.  The stale masks don't cause problems in the current code
base because the registers are either unconditionally written on nested
transitions or, in the case of segment registers, have an additional
tracker that is manually reset.

Note, by dropping (previously implicitly, now explicitly) the dirty mask
when switching the active VMCS, KVM is technically losing writes to the
associated fields.  But, the only regs that can be dirtied (RIP, RSP and
PDPTRs) are unconditionally written on nested transitions, e.g. explicit
writeback is a waste of cycles, and a WARN_ON would be rather pointless.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415203454.8296-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Invoke ept_save_pdptrs() if and only if PAE paging is enabled
Sean Christopherson [Wed, 15 Apr 2020 20:34:50 +0000 (13:34 -0700)]
KVM: nVMX: Invoke ept_save_pdptrs() if and only if PAE paging is enabled

Invoke ept_save_pdptrs() when restoring L1's host state on a "late"
VM-Fail if and only if PAE paging is enabled.  This saves a CALL in the
common case where L1 is a 64-bit host, and avoids incorrectly marking
the PDPTRs as dirty.

WARN if ept_save_pdptrs() is called with PAE disabled now that the
nested usage pre-checks is_pae_paging().  Barring a bug in KVM's MMU,
attempting to read the PDPTRs with PAE disabled is now impossible.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415203454.8296-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Rename exit_reason to vm_exit_reason for nested VM-Exit
Sean Christopherson [Wed, 15 Apr 2020 17:55:18 +0000 (10:55 -0700)]
KVM: nVMX: Rename exit_reason to vm_exit_reason for nested VM-Exit

Use "vm_exit_reason" for code related to injecting a nested VM-Exit to
VM-Exits to make it clear that nested_vmx_vmexit() expects the full exit
eason, not just the basic exit reason.  The basic exit reason (bits 15:0
of vmcs.VM_EXIT_REASON) is colloquially referred to as simply "exit
reason".

Note, other flows, e.g. vmx_handle_exit(), are intentionally left as is.
A future patch will convert vmx->exit_reason to a union + bit-field, and
the exempted flows will interact with the unionized of "exit_reason".

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-10-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Cast exit_reason to u16 to check for nested EXTERNAL_INTERRUPT
Sean Christopherson [Wed, 15 Apr 2020 17:55:17 +0000 (10:55 -0700)]
KVM: nVMX: Cast exit_reason to u16 to check for nested EXTERNAL_INTERRUPT

Explicitly check only the basic exit reason when emulating an external
interrupt VM-Exit in nested_vmx_vmexit().  Checking the full exit reason
doesn't currently cause problems, but only because the only exit reason
modifier support by KVM is FAILED_VMENTRY, which is mutually exclusive
with EXTERNAL_INTERRUPT.  Future modifiers, e.g. ENCLAVE_MODE, will
coexist with EXTERNAL_INTERRUPT.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-9-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Pull exit_reason from vcpu_vmx in nested_vmx_reflect_vmexit()
Sean Christopherson [Wed, 15 Apr 2020 17:55:16 +0000 (10:55 -0700)]
KVM: nVMX: Pull exit_reason from vcpu_vmx in nested_vmx_reflect_vmexit()

Grab the exit reason from the vcpu struct in nested_vmx_reflect_vmexit()
instead of having the exit reason explicitly passed from the caller.
This fixes a discrepancy between VM-Fail and VM-Exit handling, as the
VM-Fail case is already handled by checking vcpu_vmx, e.g. the exit
reason previously passed on the stack is bogus if vmx->fail is set.

Not taking the exit reason on the stack also avoids having to document
that nested_vmx_reflect_vmexit() requires the full exit reason, as
opposed to just the basic exit reason, which is not at all obvious since
the only usages of the full exit reason are for tracing and way down in
prepare_vmcs12() where it's propagated to vmcs12.

No functional change intended.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-8-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Drop a superfluous WARN on reflecting EXTERNAL_INTERRUPT
Sean Christopherson [Wed, 15 Apr 2020 17:55:15 +0000 (10:55 -0700)]
KVM: nVMX: Drop a superfluous WARN on reflecting EXTERNAL_INTERRUPT

Drop the WARN in nested_vmx_reflect_vmexit() that fires if KVM attempts
to reflect an external interrupt.  The WARN is blatantly impossible to
hit now that nested_vmx_l0_wants_exit() is called from
nested_vmx_reflect_vmexit() unconditionally returns true for
EXTERNAL_INTERRUPT.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-7-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Split VM-Exit reflection logic into L0 vs. L1 wants
Sean Christopherson [Wed, 15 Apr 2020 17:55:14 +0000 (10:55 -0700)]
KVM: nVMX: Split VM-Exit reflection logic into L0 vs. L1 wants

Split the logic that determines whether a nested VM-Exit is reflected
into L1 into "L0 wants" and "L1 wants" to document the core control flow
at a high level.  If L0 wants the VM-Exit, e.g. because the exit is due
to a hardware event that isn't passed through to L1, then KVM should
handle the exit in L0 without considering L1's configuration.  Then, if
L0 doesn't want the exit, KVM needs to query L1's wants to determine
whether or not L1 "caused" the exit, e.g. by setting an exiting control,
versus the exit occurring due to an L0 setting, e.g. when L0 intercepts
an action that L1 chose to pass-through.

Note, this adds an extra read on vmcs.VM_EXIT_INTR_INFO for exception.
This will be addressed in a future patch via a VMX-wide enhancement,
rather than pile on another case where vmx->exit_intr_info is
conditionally available.

Suggested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-6-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Move nested VM-Exit tracepoint into nested_vmx_reflect_vmexit()
Sean Christopherson [Wed, 15 Apr 2020 17:55:13 +0000 (10:55 -0700)]
KVM: nVMX: Move nested VM-Exit tracepoint into nested_vmx_reflect_vmexit()

Move the tracepoint for nested VM-Exits in preparation of splitting the
reflection logic into L1 wants the exit vs. L0 always handles the exit.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-5-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Move VM-Fail check out of nested_vmx_exit_reflected()
Sean Christopherson [Wed, 15 Apr 2020 17:55:12 +0000 (10:55 -0700)]
KVM: nVMX: Move VM-Fail check out of nested_vmx_exit_reflected()

Check for VM-Fail on nested VM-Enter in nested_vmx_reflect_vmexit() in
preparation for separating nested_vmx_exit_reflected() into separate "L0
wants exit exit" and "L1 wants the exit" helpers.

Explicitly set exit_intr_info and exit_qual to zero instead of reading
them from vmcs02, as they are invalid on VM-Fail (and thankfully ignored
by nested_vmx_vmexit() for nested VM-Fail).

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-4-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Uninline nested_vmx_reflect_vmexit(), i.e. move it to nested.c
Sean Christopherson [Wed, 15 Apr 2020 17:55:11 +0000 (10:55 -0700)]
KVM: nVMX: Uninline nested_vmx_reflect_vmexit(), i.e. move it to nested.c

Uninline nested_vmx_reflect_vmexit() in preparation of refactoring
nested_vmx_exit_reflected() to split up the reflection logic into more
consumable chunks, e.g. VM-Fail vs. L1 wants the exit vs. L0 always
handles the exit.

No functional change intended.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-3-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Move reflection check into nested_vmx_reflect_vmexit()
Sean Christopherson [Wed, 15 Apr 2020 17:55:10 +0000 (10:55 -0700)]
KVM: nVMX: Move reflection check into nested_vmx_reflect_vmexit()

Move the call to nested_vmx_exit_reflected() from vmx_handle_exit() into
nested_vmx_reflect_vmexit() and change the semantics of the return value
for nested_vmx_reflect_vmexit() to indicate whether or not the exit was
reflected into L1.  nested_vmx_exit_reflected() and
nested_vmx_reflect_vmexit() are intrinsically tied together, calling one
without simultaneously calling the other makes little sense.

No functional change intended.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200415175519.14230-2-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agokvm_host: unify VM_STAT and VCPU_STAT definitions in a single place
Emanuele Giuseppe Esposito [Tue, 14 Apr 2020 15:56:25 +0000 (17:56 +0200)]
kvm_host: unify VM_STAT and VCPU_STAT definitions in a single place

The macros VM_STAT and VCPU_STAT are redundantly implemented in multiple
files, each used by a different architecure to initialize the debugfs
entries for statistics. Since they all have the same purpose, they can be
unified in a single common definition in include/linux/kvm_host.h

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Message-Id: <20200414155625.20559-1-eesposit@redhat.com>
Acked-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86: move kvm_create_vcpu_debugfs after last failure point
Paolo Bonzini [Tue, 31 Mar 2020 22:42:22 +0000 (00:42 +0200)]
KVM: x86: move kvm_create_vcpu_debugfs after last failure point

The placement of kvm_create_vcpu_debugfs is more or less irrelevant, since
it cannot fail and userspace should not care about the debugfs entries until
it knows the vcpu has been created.  Moving it after the last failure
point removes the need to remove the directory when unwinding the creation.

Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Message-Id: <20200331224222.393439-1-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: SVM: Use do_machine_check to pass MCE to the host
Uros Bizjak [Sat, 11 Apr 2020 15:36:27 +0000 (17:36 +0200)]
KVM: SVM: Use do_machine_check to pass MCE to the host

Use do_machine_check instead of INT $12 to pass MCE to the host,
the same approach VMX uses.

On a related note, there is no reason to limit the use of do_machine_check
to 64 bit targets, as is currently done for VMX. MCE handling works
for both target families.

The patch is only compile tested, for both, 64 and 32 bit targets,
someone should test the passing of the exception by injecting
some MCEs into the guest.

For future non-RFC patch, kvm_machine_check should be moved to some
appropriate header file.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Message-Id: <20200411153627.3474710-1-ubizjak@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: VMX: Clean cr3/pgd handling in vmx_load_mmu_pgd()
Sean Christopherson [Fri, 20 Mar 2020 21:28:33 +0000 (14:28 -0700)]
KVM: VMX: Clean cr3/pgd handling in vmx_load_mmu_pgd()

Rename @cr3 to @pgd in vmx_load_mmu_pgd() to reflect that it will be
loaded into vmcs.EPT_POINTER and not vmcs.GUEST_CR3 when EPT is enabled.
Similarly, load guest_cr3 with @pgd if and only if EPT is disabled.

This fixes one of the last, if not _the_ last, cases in KVM where a
variable that is not strictly a cr3 value uses "cr3" instead of "pgd".

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-38-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86: Replace "cr3" with "pgd" in "new cr3/pgd" related code
Sean Christopherson [Fri, 20 Mar 2020 21:28:32 +0000 (14:28 -0700)]
KVM: x86: Replace "cr3" with "pgd" in "new cr3/pgd" related code

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
727a7e27cf88a ("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>
4 years agoKVM: nVMX: Free only the affected contexts when emulating INVEPT
Sean Christopherson [Fri, 20 Mar 2020 21:28:31 +0000 (14:28 -0700)]
KVM: nVMX: Free only the affected contexts when emulating INVEPT

Add logic to handle_invept() to free only those roots that match the
target EPT context when emulating a single-context INVEPT.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-36-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Don't flush TLB on nested VMX transition
Sean Christopherson [Fri, 20 Mar 2020 21:28:30 +0000 (14:28 -0700)]
KVM: nVMX: Don't flush TLB on nested VMX transition

Unconditionally skip the TLB flush triggered when reusing a root for a
nested transition as nested_vmx_transition_tlb_flush() ensures the TLB
is flushed when needed, regardless of whether the MMU can reuse a cached
root (or the last root).

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-35-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Skip MMU sync on nested VMX transition when possible
Sean Christopherson [Fri, 20 Mar 2020 21:28:29 +0000 (14:28 -0700)]
KVM: nVMX: Skip MMU sync on nested VMX transition when possible

Skip the MMU sync when reusing a cached root if EPT is enabled or L1
enabled VPID for L2.

If EPT is enabled, guest-physical mappings aren't flushed even if VPID
is disabled, i.e. L1 can't expect stale TLB entries to be flushed if it
has enabled EPT and L0 isn't shadowing PTEs (for L1 or L2) if L1 has
EPT disabled.

If VPID is enabled (and EPT is disabled), then L1 can't expect stale TLB
entries to be flushed (for itself or L2).

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-34-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86/mmu: Add module param to force TLB flush on root reuse
Sean Christopherson [Fri, 20 Mar 2020 21:28:28 +0000 (14:28 -0700)]
KVM: x86/mmu: Add module param to force TLB flush on root reuse

Add a module param, flush_on_reuse, to override skip_tlb_flush and
skip_mmu_sync when performing a so called "fast cr3 switch", i.e. when
reusing a cached root.  The primary motiviation for the control is to
provide a fallback mechanism in the event that TLB flushing and/or MMU
sync bugs are exposed/introduced by upcoming changes to stop
unconditionally flushing on nested VMX transitions.

Suggested-by: Jim Mattson <jmattson@google.com>
Suggested-by: Junaid Shahid <junaids@google.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-33-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86/mmu: Add separate override for MMU sync during fast CR3 switch
Sean Christopherson [Fri, 20 Mar 2020 21:28:27 +0000 (14:28 -0700)]
KVM: x86/mmu: Add separate override for MMU sync during fast CR3 switch

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>
4 years agoKVM: x86/mmu: Move fast_cr3_switch() side effects to __kvm_mmu_new_cr3()
Sean Christopherson [Fri, 20 Mar 2020 21:28:26 +0000 (14:28 -0700)]
KVM: x86/mmu: Move fast_cr3_switch() side effects to __kvm_mmu_new_cr3()

Handle the side effects of a fast CR3 (PGD) switch up a level in
__kvm_mmu_new_cr3(), which is the only caller of fast_cr3_switch().

This consolidates handling all side effects in __kvm_mmu_new_cr3()
(where freeing the current root when KVM can't do a fast switch is
already handled), and ameliorates the pain of adding a second boolean in
a future patch to provide a separate "skip" override for the MMU sync.

Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-31-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: VMX: Don't reload APIC access page if its control is disabled
Sean Christopherson [Fri, 20 Mar 2020 21:28:25 +0000 (14:28 -0700)]
KVM: VMX: Don't reload APIC access page if its control is disabled

Don't reload the APIC access page if its control is disabled, e.g. if
the guest is running with x2APIC (likely) or with the local APIC
disabled (unlikely), to avoid unnecessary TLB flushes and VMWRITEs.
Unconditionally reload the APIC access page and flush the TLB when
the guest's virtual APIC transitions to "xAPIC enabled", as any
changes to the APIC access page's mapping will not be recorded while
the guest's virtual APIC is disabled.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-30-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: VMX: Retrieve APIC access page HPA only when necessary
Sean Christopherson [Fri, 20 Mar 2020 21:28:24 +0000 (14:28 -0700)]
KVM: VMX: Retrieve APIC access page HPA only when necessary

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>
4 years agoKVM: nVMX: Reload APIC access page on nested VM-Exit only if necessary
Sean Christopherson [Fri, 20 Mar 2020 21:28:23 +0000 (14:28 -0700)]
KVM: nVMX: Reload APIC access page on nested VM-Exit only if necessary

Defer reloading L1's APIC page by logging the need for a reload and
processing it during nested VM-Exit instead of unconditionally reloading
the APIC page on nested VM-Exit.  This eliminates a TLB flush on the
majority of VM-Exits as the APIC page rarely needs to be reloaded.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-28-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Selectively use TLB_FLUSH_CURRENT for nested VM-Enter/VM-Exit
Sean Christopherson [Fri, 20 Mar 2020 21:28:22 +0000 (14:28 -0700)]
KVM: nVMX: Selectively use TLB_FLUSH_CURRENT for nested VM-Enter/VM-Exit

Flush only the current context, as opposed to all contexts, when
requesting a TLB flush to handle the scenario where a L1 does not expect
a TLB flush, but one is required because L1 and L2 shared an ASID.  This
occurs if EPT is disabled (no per-EPTP tag), VPID is enabled (hardware
doesn't flush unconditionally) and vmcs02 does not have its own VPID due
to exhaustion of available VPIDs.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-27-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86/mmu: Use KVM_REQ_TLB_FLUSH_CURRENT for MMU specific flushes
Sean Christopherson [Fri, 20 Mar 2020 21:28:21 +0000 (14:28 -0700)]
KVM: x86/mmu: Use KVM_REQ_TLB_FLUSH_CURRENT for MMU specific flushes

Flush only the current ASID/context when requesting a TLB flush due to a
change in the current vCPU's MMU to avoid blasting away TLB entries
associated with other ASIDs/contexts, e.g. entries cached for L1 when
a change in L2's MMU requires a flush.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-26-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86: Introduce KVM_REQ_TLB_FLUSH_CURRENT to flush current ASID
Sean Christopherson [Fri, 20 Mar 2020 21:28:20 +0000 (14:28 -0700)]
KVM: x86: Introduce KVM_REQ_TLB_FLUSH_CURRENT to flush current ASID

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>
4 years agoKVM: nVMX: Add helper to handle TLB flushes on nested VM-Enter/VM-Exit
Sean Christopherson [Fri, 20 Mar 2020 21:28:19 +0000 (14:28 -0700)]
KVM: nVMX: Add helper to handle TLB flushes on nested VM-Enter/VM-Exit

Add a helper to determine whether or not a full TLB flush needs to be
performed on nested VM-Enter/VM-Exit, as the logic is identical for both
flows and needs a fairly beefy comment to boot.  This also provides a
common point to make future adjustments to the logic.

Handle vpid12 changes the new helper as well even though it is specific
to VM-Enter.  The vpid12 logic is an extension of the flushing logic,
and it's worth the extra bool parameter to provide a single location for
the flushing logic.

Cc: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-24-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86: Rename ->tlb_flush() to ->tlb_flush_all()
Sean Christopherson [Fri, 20 Mar 2020 21:28:18 +0000 (14:28 -0700)]
KVM: x86: Rename ->tlb_flush() to ->tlb_flush_all()

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>
4 years agoKVM: SVM: Document the ASID logic in svm_flush_tlb()
Sean Christopherson [Fri, 20 Mar 2020 21:28:17 +0000 (14:28 -0700)]
KVM: SVM: Document the ASID logic in svm_flush_tlb()

Add a comment in svm_flush_tlb() to document why it flushes only the
current ASID, even when it is invoked when flushing remote TLBs.

Cc: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-22-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: VMX: Introduce vmx_flush_tlb_current()
Sean Christopherson [Fri, 20 Mar 2020 21:28:16 +0000 (14:28 -0700)]
KVM: VMX: Introduce vmx_flush_tlb_current()

Add a helper to flush TLB entries only for the current EPTP/VPID context
and use it for the existing direct invocations of vmx_flush_tlb().  TLB
flushes that are specific to the current vCPU state do not need to flush
other contexts.

Note, both converted call sites happen to be related to the APIC access
page, this is purely coincidental.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-21-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: nVMX: Move nested_get_vpid02() to vmx/nested.h
Sean Christopherson [Fri, 20 Mar 2020 21:28:15 +0000 (14:28 -0700)]
KVM: nVMX: Move nested_get_vpid02() to vmx/nested.h

Move nested_get_vpid02() to vmx/nested.h so that a future patch can
reference it from vmx.c to implement context-specific TLB flushing.

No functional change intended.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-20-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: VMX: Move vmx_flush_tlb() to vmx.c
Sean Christopherson [Fri, 20 Mar 2020 21:28:14 +0000 (14:28 -0700)]
KVM: VMX: Move vmx_flush_tlb() to vmx.c

Move vmx_flush_tlb() to vmx.c and make it non-inline static now that all
its callers live in vmx.c.

No functional change intended.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-19-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: SVM: Wire up ->tlb_flush_guest() directly to svm_flush_tlb()
Sean Christopherson [Fri, 20 Mar 2020 21:28:13 +0000 (14:28 -0700)]
KVM: SVM: Wire up ->tlb_flush_guest() directly to svm_flush_tlb()

Use svm_flush_tlb() directly for kvm_x86_ops->tlb_flush_guest() now that
the @invalidate_gpa param to ->tlb_flush() is gone, i.e. the wrapper for
->tlb_flush_guest() is no longer necessary.

No functional change intended.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-18-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86: Drop @invalidate_gpa param from kvm_x86_ops' tlb_flush()
Sean Christopherson [Fri, 20 Mar 2020 21:28:12 +0000 (14:28 -0700)]
KVM: x86: Drop @invalidate_gpa param from kvm_x86_ops' tlb_flush()

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>
4 years agoKVM: VMX: Clean up vmx_flush_tlb_gva()
Sean Christopherson [Fri, 20 Mar 2020 21:28:11 +0000 (14:28 -0700)]
KVM: VMX: Clean up vmx_flush_tlb_gva()

Refactor vmx_flush_tlb_gva() to remove a superfluous local variable and
clean up its comment, which is oddly located below the code it is
commenting.

No functional change intended.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-16-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86: make Hyper-V PV TLB flush use tlb_flush_guest()
Vitaly Kuznetsov [Wed, 25 Mar 2020 16:28:24 +0000 (12:28 -0400)]
KVM: x86: make Hyper-V PV TLB flush use tlb_flush_guest()

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>
4 years agoKVM: PPC: Book3S HV: Handle non-present PTEs in page fault functions
Paul Mackerras [Thu, 16 Apr 2020 04:23:43 +0000 (14:23 +1000)]
KVM: PPC: Book3S HV: Handle non-present PTEs in page fault functions

Since cd758a9b57ee "KVM: PPC: Book3S HV: Use __gfn_to_pfn_memslot in HPT
page fault handler", it's been possible in fairly rare circumstances to
load a non-present PTE in kvmppc_book3s_hv_page_fault() when running a
guest on a POWER8 host.

Because that case wasn't checked for, we could misinterpret the non-present
PTE as being a cache-inhibited PTE.  That could mismatch with the
corresponding hash PTE, which would cause the function to fail with -EFAULT
a little further down.  That would propagate up to the KVM_RUN ioctl()
generally causing the KVM userspace (usually qemu) to fall over.

This addresses the problem by catching that case and returning to the guest
instead.

For completeness, this fixes the radix page fault handler in the same
way.  For radix this didn't cause any obvious misbehaviour, because we
ended up putting the non-present PTE into the guest's partition-scoped
page tables, leading immediately to another hypervisor data/instruction
storage interrupt, which would go through the page fault path again
and fix things up.

Fixes: cd758a9b57ee "KVM: PPC: Book3S HV: Use __gfn_to_pfn_memslot in HPT page fault handler"
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1820402
Reported-by: David Gibson <david@gibson.dropbear.id.au>
Tested-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years agoKVM: x86: Move "flush guest's TLB" logic to separate kvm_x86_ops hook
Sean Christopherson [Fri, 20 Mar 2020 21:28:10 +0000 (14:28 -0700)]
KVM: x86: Move "flush guest's TLB" logic to separate kvm_x86_ops hook

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>
4 years agoKVM: nVMX: Use vpid_sync_vcpu_addr() to emulate INVVPID with address
Sean Christopherson [Fri, 20 Mar 2020 21:28:09 +0000 (14:28 -0700)]
KVM: nVMX: Use vpid_sync_vcpu_addr() to emulate INVVPID with address

Use vpid_sync_vcpu_addr() to emulate the "individual address" variant of
INVVPID now that said function handles the fallback case of the (host)
CPU not supporting "individual address".

Note, the "vpid == 0" checks in the vpid_sync_*() helpers aren't
actually redundant with the "!operand.vpid" check in handle_invvpid(),
as the vpid passed to vpid_sync_vcpu_addr() is a KVM (host) controlled
value, i.e. vpid02 can be zero even if operand.vpid is non-zero.

No functional change intended.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-14-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: VMX: Drop redundant capability checks in low level INVVPID helpers
Sean Christopherson [Fri, 20 Mar 2020 21:28:08 +0000 (14:28 -0700)]
KVM: VMX: Drop redundant capability checks in low level INVVPID helpers

Remove the INVVPID capabilities checks from vpid_sync_vcpu_single() and
vpid_sync_vcpu_global() now that all callers ensure the INVVPID variant
is supported.  Note, in some cases the guarantee is provided in concert
with hardware_setup(), which enables VPID if and only if at least of
invvpid_single() or invvpid_global() is supported.

Drop the WARN_ON_ONCE() from vmx_flush_tlb() as vpid_sync_vcpu_single()
will trigger a WARN() on INVVPID failure, i.e. if SINGLE_CONTEXT isn't
supported.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200320212833.3507-13-sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>