platform/kernel/linux-starfive.git
3 years agoKVM: x86: avoid warning with -Wbitwise-instead-of-logical
Paolo Bonzini [Fri, 15 Oct 2021 08:50:01 +0000 (04:50 -0400)]
KVM: x86: avoid warning with -Wbitwise-instead-of-logical

This is a new warning in clang top-of-tree (will be clang 14):

In file included from arch/x86/kvm/mmu/mmu.c:27:
arch/x86/kvm/mmu/spte.h:318:9: error: use of bitwise '|' with boolean operands [-Werror,-Wbitwise-instead-of-logical]
        return __is_bad_mt_xwr(rsvd_check, spte) |
               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                                                 ||
arch/x86/kvm/mmu/spte.h:318:9: note: cast one or both operands to int to silence this warning

The code is fine, but change it anyway to shut up this clever clogs
of a compiler.

Reported-by: torvic9@mailbox.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoMerge commit 'kvm-pagedata-alloc-fixes' into HEAD
Paolo Bonzini [Mon, 18 Oct 2021 18:13:37 +0000 (14:13 -0400)]
Merge commit 'kvm-pagedata-alloc-fixes' into HEAD

3 years agoKVM: X86: fix lazy allocation of rmaps
Paolo Bonzini [Fri, 15 Oct 2021 17:05:00 +0000 (13:05 -0400)]
KVM: X86: fix lazy allocation of rmaps

If allocation of rmaps fails, but some of the pointers have already been written,
those pointers can be cleaned up when the memslot is freed, or even reused later
for another attempt at allocating the rmaps.  Therefore there is no need to
WARN, as done for example in memslot_rmap_alloc, but the allocation *must* be
skipped lest KVM will overwrite the previous pointer and will indeed leak memory.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: kvm_faultin_pfn has to return false if pfh is returned
Andrei Vagin [Fri, 15 Oct 2021 16:32:21 +0000 (09:32 -0700)]
KVM: x86/mmu: kvm_faultin_pfn has to return false if pfh is returned

This looks like a typo in 8f32d5e563cb. This change didn't intend to do
any functional changes.

The problem was caught by gVisor tests.

Fixes: 8f32d5e563cb ("KVM: x86/mmu: allow kvm_faultin_pfn to return page fault handling code")
Cc: Maxim Levitsky <mlevitsk@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andrei Vagin <avagin@gmail.com>
Message-Id: <20211015163221.472508-1-avagin@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoMerge tag 'kvmarm-fixes-5.15-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Fri, 15 Oct 2021 08:47:55 +0000 (04:47 -0400)]
Merge tag 'kvmarm-fixes-5.15-2' of git://git./linux/kernel/git/kvmarm/kvmarm into HEAD

KVM/arm64 fixes for 5.15, take #2

- Properly refcount pages used as a concatenated stage-2 PGD
- Fix missing unlock when detecting the use of MTE+VM_SHARED

3 years agoKVM: SEV-ES: fix length of string I/O
Paolo Bonzini [Tue, 12 Oct 2021 15:07:59 +0000 (11:07 -0400)]
KVM: SEV-ES: fix length of string I/O

The size of the data in the scratch buffer is not divided by the size of
each port I/O operation, so vcpu->arch.pio.count ends up being larger
than it should be by a factor of size.

Cc: stable@vger.kernel.org
Fixes: 7ed9abfe8e9f ("KVM: SVM: Support string IO operations for an SEV-ES guest")
Acked-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: arm64: Release mmap_lock when using VM_SHARED with MTE
Quentin Perret [Tue, 5 Oct 2021 12:20:31 +0000 (13:20 +0100)]
KVM: arm64: Release mmap_lock when using VM_SHARED with MTE

VM_SHARED mappings are currently forbidden in a memslot with MTE to
prevent two VMs racing to sanitise the same page. However, this check
is performed while holding current->mm's mmap_lock, but fails to release
it. Fix this by releasing the lock when needed.

Fixes: ea7fc1bb1cd1 ("KVM: arm64: Introduce MTE VM feature")
Signed-off-by: Quentin Perret <qperret@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20211005122031.809857-1-qperret@google.com
3 years agoKVM: arm64: Report corrupted refcount at EL2
Quentin Perret [Tue, 5 Oct 2021 09:01:42 +0000 (10:01 +0100)]
KVM: arm64: Report corrupted refcount at EL2

Some of the refcount manipulation helpers used at EL2 are instrumented
to catch a corrupted state, but not all of them are treated equally. Let's
make things more consistent by instrumenting hyp_page_ref_dec_and_test()
as well.

Acked-by: Will Deacon <will@kernel.org>
Suggested-by: Will Deacon <will@kernel.org>
Signed-off-by: Quentin Perret <qperret@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20211005090155.734578-6-qperret@google.com
3 years agoKVM: arm64: Fix host stage-2 PGD refcount
Quentin Perret [Tue, 5 Oct 2021 09:01:41 +0000 (10:01 +0100)]
KVM: arm64: Fix host stage-2 PGD refcount

The KVM page-table library refcounts the pages of concatenated stage-2
PGDs individually. However, when running KVM in protected mode, the
host's stage-2 PGD is currently managed by EL2 as a single high-order
compound page, which can cause the refcount of the tail pages to reach 0
when they shouldn't, hence corrupting the page-table.

Fix this by introducing a new hyp_split_page() helper in the EL2 page
allocator (matching the kernel's split_page() function), and make use of
it from host_s2_zalloc_pages_exact().

Fixes: 1025c8c0c6ac ("KVM: arm64: Wrap the host with a stage 2")
Acked-by: Will Deacon <will@kernel.org>
Suggested-by: Will Deacon <will@kernel.org>
Signed-off-by: Quentin Perret <qperret@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20211005090155.734578-5-qperret@google.com
3 years agoMerge tag 'kvm-riscv-5.16-1' of git://github.com/kvm-riscv/linux into HEAD
Paolo Bonzini [Tue, 5 Oct 2021 08:19:24 +0000 (04:19 -0400)]
Merge tag 'kvm-riscv-5.16-1' of git://github.com/kvm-riscv/linux into HEAD

Initial KVM RISC-V support

Following features are supported by the initial KVM RISC-V support:
1. No RISC-V specific KVM IOCTL
2. Loadable KVM RISC-V module
3. Minimal possible KVM world-switch which touches only GPRs and few CSRs
4. Works on both RV64 and RV32 host
5. Full Guest/VM switch via vcpu_get/vcpu_put infrastructure
6. KVM ONE_REG interface for VCPU register access from KVM user-space
7. Interrupt controller emulation in KVM user-space
8. Timer and IPI emuation in kernel
9. Both Sv39x4 and Sv48x4 supported for RV64 host
10. MMU notifiers supported
11. Generic dirty log supported
12. FP lazy save/restore supported
13. SBI v0.1 emulation for Guest/VM
14. Forward unhandled SBI calls to KVM user-space
15. Hugepage support for Guest/VM
16. IOEVENTFD support for Vhost

3 years agoRISC-V: KVM: Add MAINTAINERS entry
Anup Patel [Mon, 27 Sep 2021 11:40:16 +0000 (17:10 +0530)]
RISC-V: KVM: Add MAINTAINERS entry

Add myself as maintainer for KVM RISC-V and Atish as designated reviewer.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Document RISC-V specific parts of KVM API
Anup Patel [Mon, 27 Sep 2021 11:40:15 +0000 (17:10 +0530)]
RISC-V: KVM: Document RISC-V specific parts of KVM API

Document RISC-V specific parts of the KVM API, such as:
 - The interrupt numbers passed to the KVM_INTERRUPT ioctl.
 - The states supported by the KVM_{GET,SET}_MP_STATE ioctls.
 - The registers supported by the KVM_{GET,SET}_ONE_REG interface
   and the encoding of those register ids.
 - The exit reason KVM_EXIT_RISCV_SBI for SBI calls forwarded to
   userspace tool.

CC: Jonathan Corbet <corbet@lwn.net>
CC: linux-doc@vger.kernel.org
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Add SBI v0.1 support
Atish Patra [Mon, 27 Sep 2021 11:40:14 +0000 (17:10 +0530)]
RISC-V: KVM: Add SBI v0.1 support

The KVM host kernel is running in HS-mode needs so we need to handle
the SBI calls coming from guest kernel running in VS-mode.

This patch adds SBI v0.1 support in KVM RISC-V. Almost all SBI v0.1
calls are implemented in KVM kernel module except GETCHAR and PUTCHART
calls which are forwarded to user space because these calls cannot be
implemented in kernel space. In future, when we implement SBI v0.2 for
Guest, we will forward SBI v0.2 experimental and vendor extension calls
to user space.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Implement ONE REG interface for FP registers
Atish Patra [Mon, 27 Sep 2021 11:40:13 +0000 (17:10 +0530)]
RISC-V: KVM: Implement ONE REG interface for FP registers

Add a KVM_GET_ONE_REG/KVM_SET_ONE_REG ioctl interface for floating
point registers such as F0-F31 and FCSR. This support is added for
both 'F' and 'D' extensions.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: FP lazy save/restore
Atish Patra [Mon, 27 Sep 2021 11:40:12 +0000 (17:10 +0530)]
RISC-V: KVM: FP lazy save/restore

This patch adds floating point (F and D extension) context save/restore
for guest VCPUs. The FP context is saved and restored lazily only when
kernel enter/exits the in-kernel run loop and not during the KVM world
switch. This way FP save/restore has minimal impact on KVM performance.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Add timer functionality
Atish Patra [Mon, 27 Sep 2021 11:40:11 +0000 (17:10 +0530)]
RISC-V: KVM: Add timer functionality

The RISC-V hypervisor specification doesn't have any virtual timer
feature.

Due to this, the guest VCPU timer will be programmed via SBI calls.
The host will use a separate hrtimer event for each guest VCPU to
provide timer functionality. We inject a virtual timer interrupt to
the guest VCPU whenever the guest VCPU hrtimer event expires.

This patch adds guest VCPU timer implementation along with ONE_REG
interface to access VCPU timer state from user space.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Implement MMU notifiers
Anup Patel [Mon, 27 Sep 2021 11:40:10 +0000 (17:10 +0530)]
RISC-V: KVM: Implement MMU notifiers

This patch implements MMU notifiers for KVM RISC-V so that Guest
physical address space is in-sync with Host physical address space.

This will allow swapping, page migration, etc to work transparently
with KVM RISC-V.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Implement stage2 page table programming
Anup Patel [Mon, 27 Sep 2021 11:40:09 +0000 (17:10 +0530)]
RISC-V: KVM: Implement stage2 page table programming

This patch implements all required functions for programming
the stage2 page table for each Guest/VM.

At high-level, the flow of stage2 related functions is similar
from KVM ARM/ARM64 implementation but the stage2 page table
format is quite different for KVM RISC-V.

[jiangyifei: stage2 dirty log support]
Signed-off-by: Yifei Jiang <jiangyifei@huawei.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Implement VMID allocator
Anup Patel [Mon, 27 Sep 2021 11:40:08 +0000 (17:10 +0530)]
RISC-V: KVM: Implement VMID allocator

We implement a simple VMID allocator for Guests/VMs which:
1. Detects number of VMID bits at boot-time
2. Uses atomic number to track VMID version and increments
   VMID version whenever we run-out of VMIDs
3. Flushes Guest TLBs on all host CPUs whenever we run-out
   of VMIDs
4. Force updates HW Stage2 VMID for each Guest VCPU whenever
   VMID changes using VCPU request KVM_REQ_UPDATE_HGATP

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Handle WFI exits for VCPU
Anup Patel [Mon, 27 Sep 2021 11:40:07 +0000 (17:10 +0530)]
RISC-V: KVM: Handle WFI exits for VCPU

We get illegal instruction trap whenever Guest/VM executes WFI
instruction.

This patch handles WFI trap by blocking the trapped VCPU using
kvm_vcpu_block() API. The blocked VCPU will be automatically
resumed whenever a VCPU interrupt is injected from user-space
or from in-kernel IRQCHIP emulation.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Handle MMIO exits for VCPU
Anup Patel [Mon, 27 Sep 2021 11:40:06 +0000 (17:10 +0530)]
RISC-V: KVM: Handle MMIO exits for VCPU

We will get stage2 page faults whenever Guest/VM access SW emulated
MMIO device or unmapped Guest RAM.

This patch implements MMIO read/write emulation by extracting MMIO
details from the trapped load/store instruction and forwarding the
MMIO read/write to user-space. The actual MMIO emulation will happen
in user-space and KVM kernel module will only take care of register
updates before resuming the trapped VCPU.

The handling for stage2 page faults for unmapped Guest RAM will be
implemeted by a separate patch later.

[jiangyifei: ioeventfd and in-kernel mmio device support]
Signed-off-by: Yifei Jiang <jiangyifei@huawei.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Implement VCPU world-switch
Anup Patel [Mon, 27 Sep 2021 11:40:05 +0000 (17:10 +0530)]
RISC-V: KVM: Implement VCPU world-switch

This patch implements the VCPU world-switch for KVM RISC-V.

The KVM RISC-V world-switch (i.e. __kvm_riscv_switch_to()) mostly
switches general purpose registers, SSTATUS, STVEC, SSCRATCH and
HSTATUS CSRs. Other CSRs are switched via vcpu_load() and vcpu_put()
interface in kvm_arch_vcpu_load() and kvm_arch_vcpu_put() functions
respectively.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Implement KVM_GET_ONE_REG/KVM_SET_ONE_REG ioctls
Anup Patel [Mon, 27 Sep 2021 11:40:04 +0000 (17:10 +0530)]
RISC-V: KVM: Implement KVM_GET_ONE_REG/KVM_SET_ONE_REG ioctls

For KVM RISC-V, we use KVM_GET_ONE_REG/KVM_SET_ONE_REG ioctls to access
VCPU config and registers from user-space.

We have three types of VCPU registers:
1. CONFIG - these are VCPU config and capabilities
2. CORE   - these are VCPU general purpose registers
3. CSR    - these are VCPU control and status registers

The CONFIG register available to user-space is ISA. The ISA register is
a read and write register where user-space can only write the desired
VCPU ISA capabilities before running the VCPU.

The CORE registers available to user-space are PC, RA, SP, GP, TP, A0-A7,
T0-T6, S0-S11 and MODE. Most of these are RISC-V general registers except
PC and MODE. The PC register represents program counter whereas the MODE
register represent VCPU privilege mode (i.e. S/U-mode).

The CSRs available to user-space are SSTATUS, SIE, STVEC, SSCRATCH, SEPC,
SCAUSE, STVAL, SIP, and SATP. All of these are read/write registers.

In future, more VCPU register types will be added (such as FP) for the
KVM_GET_ONE_REG/KVM_SET_ONE_REG ioctls.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Implement VCPU interrupts and requests handling
Anup Patel [Mon, 27 Sep 2021 11:40:03 +0000 (17:10 +0530)]
RISC-V: KVM: Implement VCPU interrupts and requests handling

This patch implements VCPU interrupts and requests which are both
asynchronous events.

The VCPU interrupts can be set/unset using KVM_INTERRUPT ioctl from
user-space. In future, the in-kernel IRQCHIP emulation will use
kvm_riscv_vcpu_set_interrupt() and kvm_riscv_vcpu_unset_interrupt()
functions to set/unset VCPU interrupts.

Important VCPU requests implemented by this patch are:
KVM_REQ_SLEEP       - set whenever VCPU itself goes to sleep state
KVM_REQ_VCPU_RESET  - set whenever VCPU reset is requested

The WFI trap-n-emulate (added later) will use KVM_REQ_SLEEP request
and kvm_riscv_vcpu_has_interrupt() function.

The KVM_REQ_VCPU_RESET request will be used by SBI emulation (added
later) to power-up a VCPU in power-off state. The user-space can use
the GET_MPSTATE/SET_MPSTATE ioctls to get/set power state of a VCPU.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: KVM: Implement VCPU create, init and destroy functions
Anup Patel [Mon, 27 Sep 2021 11:40:02 +0000 (17:10 +0530)]
RISC-V: KVM: Implement VCPU create, init and destroy functions

This patch implements VCPU create, init and destroy functions
required by generic KVM module. We don't have much dynamic
resources in struct kvm_vcpu_arch so these functions are quite
simple for KVM RISC-V.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: Add initial skeletal KVM support
Anup Patel [Mon, 27 Sep 2021 11:40:01 +0000 (17:10 +0530)]
RISC-V: Add initial skeletal KVM support

This patch adds initial skeletal KVM RISC-V support which has:
1. A simple implementation of arch specific VM functions
   except kvm_vm_ioctl_get_dirty_log() which will implemeted
   in-future as part of stage2 page loging.
2. Stubs of required arch specific VCPU functions except
   kvm_arch_vcpu_ioctl_run() which is semi-complete and
   extended by subsequent patches.
3. Stubs for required arch specific stage2 MMU functions.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: Add hypervisor extension related CSR defines
Anup Patel [Mon, 27 Sep 2021 11:40:00 +0000 (17:10 +0530)]
RISC-V: Add hypervisor extension related CSR defines

This patch adds asm/kvm_csr.h for RISC-V hypervisor extension
related defines.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alexander Graf <graf@amazon.com>
Message-Id: <20210927114016.1089328-2-anup.patel@wdc.com>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoMerge tag 'kvm-s390-master-5.15-1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Mon, 4 Oct 2021 07:58:25 +0000 (03:58 -0400)]
Merge tag 'kvm-s390-master-5.15-1' of git://git./linux/kernel/git/kvms390/linux into kvm-master

KVM: s390: allow to compile without warning with W=1

3 years agoKVM: x86: only allocate gfn_track when necessary
David Stevens [Wed, 22 Sep 2021 04:58:59 +0000 (13:58 +0900)]
KVM: x86: only allocate gfn_track when necessary

Avoid allocating the gfn_track arrays if nothing needs them. If there
are no external to KVM users of the API (i.e. no GVT-g), then page
tracking is only needed for shadow page tables. This means that when tdp
is enabled and there are no external users, then the gfn_track arrays
can be lazily allocated when the shadow MMU is actually used. This avoid
allocations equal to .05% of guest memory when nested virtualization is
not used, if the kernel is compiled without GVT-g.

Signed-off-by: David Stevens <stevensd@chromium.org>
Message-Id: <20210922045859.2011227-3-stevensd@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: add config for non-kvm users of page tracking
David Stevens [Wed, 22 Sep 2021 04:58:58 +0000 (13:58 +0900)]
KVM: x86: add config for non-kvm users of page tracking

Add a config option that allows kvm to determine whether or not there
are any external users of page tracking.

Signed-off-by: David Stevens <stevensd@chromium.org>
Message-Id: <20210922045859.2011227-2-stevensd@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agonSVM: Check for reserved encodings of TLB_CONTROL in nested VMCB
Krish Sadhukhan [Mon, 20 Sep 2021 23:51:31 +0000 (19:51 -0400)]
nSVM: Check for reserved encodings of TLB_CONTROL in nested VMCB

According to section "TLB Flush" in APM vol 2,

    "Support for TLB_CONTROL commands other than the first two, is
     optional and is indicated by CPUID Fn8000_000A_EDX[FlushByAsid].

     All encodings of TLB_CONTROL not defined in the APM are reserved."

Signed-off-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Message-Id: <20210920235134.101970-3-krish.sadhukhan@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: use kvfree() in kvm_arch_free_vm()
Juergen Gross [Fri, 3 Sep 2021 13:08:05 +0000 (15:08 +0200)]
kvm: use kvfree() in kvm_arch_free_vm()

By switching from kfree() to kvfree() in kvm_arch_free_vm() Arm64 can
use the common variant. This can be accomplished by adding another
macro __KVM_HAVE_ARCH_VM_FREE, which will be used only by x86 for now.

Further simplification can be achieved by adding __kvm_arch_free_vm()
doing the common part.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Message-Id: <20210903130808.30142-5-jgross@suse.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Expose Predictive Store Forwarding Disable
Babu Moger [Fri, 24 Sep 2021 01:15:28 +0000 (20:15 -0500)]
KVM: x86: Expose Predictive Store Forwarding Disable

Predictive Store Forwarding: AMD Zen3 processors feature a new
technology called Predictive Store Forwarding (PSF).

PSF is a hardware-based micro-architectural optimization designed
to improve the performance of code execution by predicting address
dependencies between loads and stores.

How PSF works:

It is very common for a CPU to execute a load instruction to an address
that was recently written by a store. Modern CPUs implement a technique
known as Store-To-Load-Forwarding (STLF) to improve performance in such
cases. With STLF, data from the store is forwarded directly to the load
without having to wait for it to be written to memory. In a typical CPU,
STLF occurs after the address of both the load and store are calculated
and determined to match.

PSF expands on this by speculating on the relationship between loads and
stores without waiting for the address calculation to complete. With PSF,
the CPU learns over time the relationship between loads and stores. If
STLF typically occurs between a particular store and load, the CPU will
remember this.

In typical code, PSF provides a performance benefit by speculating on
the load result and allowing later instructions to begin execution
sooner than they otherwise would be able to.

The details of security analysis of AMD predictive store forwarding is
documented here.
https://www.amd.com/system/files/documents/security-analysis-predictive-store-forwarding.pdf

Predictive Store Forwarding controls:
There are two hardware control bits which influence the PSF feature:
- MSR 48h bit 2 – Speculative Store Bypass (SSBD)
- MSR 48h bit 7 – Predictive Store Forwarding Disable (PSFD)

The PSF feature is disabled if either of these bits are set.  These bits
are controllable on a per-thread basis in an SMT system. By default, both
SSBD and PSFD are 0 meaning that the speculation features are enabled.

While the SSBD bit disables PSF and speculative store bypass, PSFD only
disables PSF.

PSFD may be desirable for software which is concerned with the
speculative behavior of PSF but desires a smaller performance impact than
setting SSBD.

Support for PSFD is indicated in CPUID Fn8000_0008 EBX[28].
All processors that support PSF will also support PSFD.

Linux kernel does not have the interface to enable/disable PSFD yet. Plan
here is to expose the PSFD technology to KVM so that the guest kernel can
make use of it if they wish to.

Signed-off-by: Babu Moger <Babu.Moger@amd.com>
Message-Id: <163244601049.30292.5855870305350227855.stgit@bmoger-ubuntu>
[Keep feature private to KVM, as requested by Borislav Petkov. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Avoid memslot lookup in make_spte and mmu_try_to_unsync_pages
David Matlack [Tue, 17 Aug 2021 12:46:45 +0000 (08:46 -0400)]
KVM: x86/mmu: Avoid memslot lookup in make_spte and mmu_try_to_unsync_pages

mmu_try_to_unsync_pages checks if page tracking is active for the given
gfn, which requires knowing the memslot. We can pass down the memslot
via make_spte to avoid this lookup.

The memslot is also handy for make_spte's marking of the gfn as dirty:
we can test whether dirty page tracking is enabled, and if so ensure that
pages are mapped as writable with 4K granularity.  Apart from the warning,
no functional change is intended.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210813203504.2742757-7-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Avoid memslot lookup in rmap_add
David Matlack [Fri, 13 Aug 2021 20:35:03 +0000 (20:35 +0000)]
KVM: x86/mmu: Avoid memslot lookup in rmap_add

Avoid the memslot lookup in rmap_add, by passing it down from the fault
handling code to mmu_set_spte and then to rmap_add.

No functional change intended.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210813203504.2742757-6-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: pass struct kvm_page_fault to mmu_set_spte
Paolo Bonzini [Tue, 17 Aug 2021 11:49:47 +0000 (07:49 -0400)]
KVM: MMU: pass struct kvm_page_fault to mmu_set_spte

mmu_set_spte is called for either PTE prefetching or page faults.  The
three boolean arguments write_fault, speculative and host_writable are
always respectively false/true/true for prefetching and coming from
a struct kvm_page_fault for page faults.

Let mmu_set_spte distinguish these two situation by accepting a
possibly NULL struct kvm_page_fault argument.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: pass kvm_mmu_page struct to make_spte
Paolo Bonzini [Tue, 17 Aug 2021 11:43:19 +0000 (07:43 -0400)]
KVM: MMU: pass kvm_mmu_page struct to make_spte

The level and A/D bit support of the new SPTE can be found in the role,
which is stored in the kvm_mmu_page struct.  This merges two arguments
into one.

For the TDP MMU, the kvm_mmu_page was not used (kvm_tdp_mmu_map does
not use it if the SPTE is already present) so we fetch it just before
calling make_spte.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: set ad_disabled in TDP MMU role
Paolo Bonzini [Tue, 17 Aug 2021 11:34:04 +0000 (07:34 -0400)]
KVM: MMU: set ad_disabled in TDP MMU role

Prepare for removing the ad_disabled argument of make_spte; instead it can
be found in the role of a struct kvm_mmu_page.  First of all, the TDP MMU
must set the role accurately.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: remove unnecessary argument to mmu_set_spte
Paolo Bonzini [Tue, 17 Aug 2021 11:42:10 +0000 (07:42 -0400)]
KVM: MMU: remove unnecessary argument to mmu_set_spte

The level of the new SPTE can be found in the kvm_mmu_page struct; there
is no need to pass it down.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: clean up make_spte return value
Paolo Bonzini [Tue, 17 Aug 2021 11:32:09 +0000 (07:32 -0400)]
KVM: MMU: clean up make_spte return value

Now that make_spte is called directly by the shadow MMU (rather than
wrapped by set_spte), it only has to return one boolean value.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: inline set_spte in FNAME(sync_page)
Paolo Bonzini [Tue, 17 Aug 2021 11:22:32 +0000 (07:22 -0400)]
KVM: MMU: inline set_spte in FNAME(sync_page)

Since the two callers of set_spte do different things with the results,
inlining it actually makes the code simpler to reason about.  For example,
FNAME(sync_page) already has a struct kvm_mmu_page *, but set_spte had to
fish it back out of sptep's private page data.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: inline set_spte in mmu_set_spte
Paolo Bonzini [Tue, 17 Aug 2021 11:22:32 +0000 (07:22 -0400)]
KVM: MMU: inline set_spte in mmu_set_spte

Since the two callers of set_spte do different things with the results,
inlining it actually makes the code simpler to reason about.  For example,
mmu_set_spte looks quite like tdp_mmu_map_handle_target_level, but the
similarity is hidden by set_spte.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Avoid memslot lookup in page_fault_handle_page_track
David Matlack [Fri, 13 Aug 2021 20:35:02 +0000 (20:35 +0000)]
KVM: x86/mmu: Avoid memslot lookup in page_fault_handle_page_track

Now that kvm_page_fault has a pointer to the memslot it can be passed
down to the page tracking code to avoid a redundant slot lookup.

No functional change intended.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210813203504.2742757-5-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Pass the memslot around via struct kvm_page_fault
David Matlack [Fri, 24 Sep 2021 09:05:26 +0000 (05:05 -0400)]
KVM: x86/mmu: Pass the memslot around via struct kvm_page_fault

The memslot for the faulting gfn is used throughout the page fault
handling code, so capture it in kvm_page_fault as soon as we know the
gfn and use it in the page fault handling code that has direct access
to the kvm_page_fault struct.  Replace various tests using is_noslot_pfn
with more direct tests on fault->slot being NULL.

This, in combination with the subsequent patch, improves "Populate
memory time" in dirty_log_perf_test by 5% when using the legacy MMU.
There is no discerable improvement to the performance of the TDP MMU.

No functional change intended.

Suggested-by: Ben Gardon <bgardon@google.com>
Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210813203504.2742757-4-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: unify tdp_mmu_map_set_spte_atomic and tdp_mmu_set_spte_atomic_no_dirty_log
Paolo Bonzini [Thu, 23 Sep 2021 15:20:48 +0000 (11:20 -0400)]
KVM: MMU: unify tdp_mmu_map_set_spte_atomic and tdp_mmu_set_spte_atomic_no_dirty_log

tdp_mmu_map_set_spte_atomic is not taking care of dirty logging anymore,
the only difference that remains is that it takes a vCPU instead of
the struct kvm.  Merge the two functions.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: mark page dirty in make_spte
Paolo Bonzini [Fri, 24 Sep 2021 08:52:23 +0000 (04:52 -0400)]
KVM: MMU: mark page dirty in make_spte

This simplifies set_spte, which we want to remove, and unifies code
between the shadow MMU and the TDP MMU.  The warning will be added
back later to make_spte as well.

There is a small disadvantage in the TDP MMU; it may unnecessarily mark
a page as dirty twice if two vCPUs end up mapping the same page twice.
However, this is a very small cost for a case that is already rare.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Fold rmap_recycle into rmap_add
David Matlack [Fri, 13 Aug 2021 20:35:00 +0000 (20:35 +0000)]
KVM: x86/mmu: Fold rmap_recycle into rmap_add

Consolidate rmap_recycle and rmap_add into a single function since they
are only ever called together (and only from one place). This has a nice
side effect of eliminating an extra kvm_vcpu_gfn_to_memslot(). In
addition it makes mmu_set_spte(), which is a very long function, a
little shorter.

No functional change intended.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20210813203504.2742757-3-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Verify shadow walk doesn't terminate early in page faults
Sean Christopherson [Mon, 6 Sep 2021 12:25:46 +0000 (20:25 +0800)]
KVM: x86/mmu: Verify shadow walk doesn't terminate early in page faults

WARN and bail if the shadow walk for faulting in a SPTE terminates early,
i.e. doesn't reach the expected level because the walk encountered a
terminal SPTE.  The shadow walks for page faults are subtle in that they
install non-leaf SPTEs (zapping leaf SPTEs if necessary!) in the loop
body, and consume the newly created non-leaf SPTE in the loop control,
e.g. __shadow_walk_next().  In other words, the walks guarantee that the
walk will stop if and only if the target level is reached by installing
non-leaf SPTEs to guarantee the walk remains valid.

Opportunistically use fault->goal-level instead of it.level in
FNAME(fetch) to further clarify that KVM always installs the leaf SPTE at
the target level.

Reviewed-by: Lai Jiangshan <jiangshanlai@gmail.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Message-Id: <20210906122547.263316-1-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change tracepoints arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change tracepoints arguments to kvm_page_fault

Pass struct kvm_page_fault to tracepoints instead of extracting the
arguments from the struct.  This also lets the kvm_mmu_spte_requested
tracepoint pick the gfn directly from fault->gfn, instead of using
the address.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change disallowed_hugepage_adjust() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change disallowed_hugepage_adjust() arguments to kvm_page_fault

Pass struct kvm_page_fault to disallowed_hugepage_adjust() instead of
extracting the arguments from the struct.  Tweak a bit the conditions
to avoid long lines.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change kvm_mmu_hugepage_adjust() arguments to kvm_page_fault
Paolo Bonzini [Sat, 7 Aug 2021 13:21:53 +0000 (09:21 -0400)]
KVM: MMU: change kvm_mmu_hugepage_adjust() arguments to kvm_page_fault

Pass struct kvm_page_fault to kvm_mmu_hugepage_adjust() instead of
extracting the arguments from the struct; the results are also stored
in the struct, so the callers are adjusted consequently.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change fast_page_fault() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change fast_page_fault() arguments to kvm_page_fault

Pass struct kvm_page_fault to fast_page_fault() instead of
extracting the arguments from the struct.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change tdp_mmu_map_handle_target_level() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change tdp_mmu_map_handle_target_level() arguments to kvm_page_fault

Pass struct kvm_page_fault to tdp_mmu_map_handle_target_level() instead of
extracting the arguments from the struct.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change kvm_tdp_mmu_map() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change kvm_tdp_mmu_map() arguments to kvm_page_fault

Pass struct kvm_page_fault to kvm_tdp_mmu_map() instead of
extracting the arguments from the struct.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change FNAME(fetch)() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change FNAME(fetch)() arguments to kvm_page_fault

Pass struct kvm_page_fault to FNAME(fetch)() instead of
extracting the arguments from the struct.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change __direct_map() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change __direct_map() arguments to kvm_page_fault

Pass struct kvm_page_fault to __direct_map() instead of
extracting the arguments from the struct.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change handle_abnormal_pfn() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change handle_abnormal_pfn() arguments to kvm_page_fault

Pass struct kvm_page_fault to handle_abnormal_pfn() instead of
extracting the arguments from the struct.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change kvm_faultin_pfn() arguments to kvm_page_fault
Paolo Bonzini [Sat, 7 Aug 2021 12:57:34 +0000 (08:57 -0400)]
KVM: MMU: change kvm_faultin_pfn() arguments to kvm_page_fault

Add fields to struct kvm_page_fault corresponding to outputs of
kvm_faultin_pfn().  For now they have to be extracted again from struct
kvm_page_fault in the subsequent steps, but this is temporary until
other functions in the chain are switched over as well.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change page_fault_handle_page_track() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:21:58 +0000 (04:21 -0400)]
KVM: MMU: change page_fault_handle_page_track() arguments to kvm_page_fault

Add fields to struct kvm_page_fault corresponding to the arguments
of page_fault_handle_page_track().  The fields are initialized in the
callers, and page_fault_handle_page_track() receives a struct
kvm_page_fault instead of having to extract the arguments out of it.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change direct_page_fault() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:21:58 +0000 (04:21 -0400)]
KVM: MMU: change direct_page_fault() arguments to kvm_page_fault

Add fields to struct kvm_page_fault corresponding to
the arguments of direct_page_fault().  The fields are
initialized in the callers, and direct_page_fault()
receives a struct kvm_page_fault instead of having to
extract the arguments out of it.

Also adjust FNAME(page_fault) to store the max_level in
struct kvm_page_fault, to keep it similar to the direct
map path.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: change mmu->page_fault() arguments to kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 08:35:50 +0000 (04:35 -0400)]
KVM: MMU: change mmu->page_fault() arguments to kvm_page_fault

Pass struct kvm_page_fault to mmu->page_fault() instead of
extracting the arguments from the struct.  FNAME(page_fault) can use
the precomputed bools from the error code.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: Introduce struct kvm_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 07:52:18 +0000 (03:52 -0400)]
KVM: MMU: Introduce struct kvm_page_fault

Create a single structure for arguments that are passed from
kvm_mmu_do_page_fault to the page fault handlers.  Later
the structure will grow to include various output parameters
that are passed back to the next steps in the page fault
handling.

Suggested-by: Isaku Yamahata <isaku.yamahata@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: MMU: pass unadulterated gpa to direct_page_fault
Paolo Bonzini [Fri, 6 Aug 2021 09:21:17 +0000 (05:21 -0400)]
KVM: MMU: pass unadulterated gpa to direct_page_fault

Do not bother removing the low bits of the gpa.  This masking dates back
to the very first commit of KVM but it is unnecessary, as exemplified
by the other call in kvm_tdp_page_fault.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Fix potential race in KVM_GET_CLOCK
Oliver Upton [Thu, 16 Sep 2021 18:15:34 +0000 (18:15 +0000)]
KVM: x86: Fix potential race in KVM_GET_CLOCK

Sean noticed that KVM_GET_CLOCK was checking kvm_arch.use_master_clock
outside of the pvclock sync lock. This is problematic, as the clock
value written to the user may or may not actually correspond to a stable
TSC.

Fix the race by populating the entire kvm_clock_data structure behind
the pvclock_gtod_sync_lock.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Message-Id: <20210916181538.968978-4-oupton@google.com>
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: extract KVM_GET_CLOCK/KVM_SET_CLOCK to separate functions
Paolo Bonzini [Thu, 16 Sep 2021 18:15:33 +0000 (18:15 +0000)]
KVM: x86: extract KVM_GET_CLOCK/KVM_SET_CLOCK to separate functions

No functional change intended.

Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: x86: abstract locking around pvclock_update_vm_gtod_copy
Paolo Bonzini [Thu, 16 Sep 2021 18:15:32 +0000 (18:15 +0000)]
kvm: x86: abstract locking around pvclock_update_vm_gtod_copy

Updates to the kvmclock parameters needs to do a complicated dance of
KVM_REQ_MCLOCK_INPROGRESS and KVM_REQ_CLOCK_UPDATE in addition to taking
pvclock_gtod_sync_lock.  Place that in two functions that can be called
on all of master clock update, KVM_SET_CLOCK, and Hyper-V reenlightenment.

Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Move PTE present check from loop body to __shadow_walk_next()
Lai Jiangshan [Mon, 6 Sep 2021 12:25:47 +0000 (20:25 +0800)]
KVM: X86: Move PTE present check from loop body to __shadow_walk_next()

So far, the loop bodies already ensure the PTE is present before calling
__shadow_walk_next():  Some loop bodies simply exit with a !PRESENT
directly and some other loop bodies, i.e. FNAME(fetch) and __direct_map()
do not currently guard their walks with is_shadow_present_pte, but only
because they install present non-leaf SPTEs in the loop itself.

But checking pte present in __shadow_walk_next() (which is called from
shadow_walk_okay()) is more prudent; walking past a !PRESENT SPTE
would lead to attempting to read a the next level SPTE from a garbage
iter->shadow_addr.  It also allows to remove the is_shadow_present_pte()
checks from the loop bodies.

Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Message-Id: <20210906122547.263316-2-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: nSVM: implement nested TSC scaling
Maxim Levitsky [Tue, 14 Sep 2021 15:48:24 +0000 (18:48 +0300)]
KVM: x86: nSVM: implement nested TSC scaling

This was tested by booting a nested guest with TSC=1Ghz,
observing the clocks, and doing about 100 cycles of migration.

Note that qemu patch is needed to support migration because
of a new MSR that needs to be placed in the migration state.

The patch will be sent to the qemu mailing list soon.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210914154825.104886-14-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: SVM: add module param to control TSC scaling
Maxim Levitsky [Tue, 14 Sep 2021 15:48:23 +0000 (18:48 +0300)]
KVM: x86: SVM: add module param to control TSC scaling

This allows to easily simulate a CPU without this feature.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210914154825.104886-13-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: SVM: don't set VMLOAD/VMSAVE intercepts on vCPU reset
Paolo Bonzini [Thu, 23 Sep 2021 16:46:07 +0000 (12:46 -0400)]
KVM: x86: SVM: don't set VMLOAD/VMSAVE intercepts on vCPU reset

Commit adc2a23734ac ("KVM: nSVM: improve SYSENTER emulation on AMD"),
made init_vmcb set vmload/vmsave intercepts unconditionally,
and relied on svm_vcpu_after_set_cpuid to clear them when possible.

However init_vmcb is also called when the vCPU is reset, and it is
not followed by another call to svm_vcpu_after_set_cpuid because
the CPUID is already set.  This mistake makes the VMSAVE/VMLOAD intercept
to be set when it is not needed, and harms performance of the nested
guest.

Extract the relevant parts of svm_vcpu_after_set_cpuid so that they
can be called again on reset.

Fixes: adc2a23734ac ("KVM: nSVM: improve SYSENTER emulation on AMD")
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: SVM: add module param to control LBR virtualization
Maxim Levitsky [Tue, 14 Sep 2021 15:48:19 +0000 (18:48 +0300)]
KVM: x86: SVM: add module param to control LBR virtualization

This is useful for debug and also makes it consistent with
the rest of the SVM optional features.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210914154825.104886-9-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: nSVM: don't copy pause related settings
Maxim Levitsky [Tue, 14 Sep 2021 15:48:15 +0000 (18:48 +0300)]
KVM: x86: nSVM: don't copy pause related settings

According to the SDM, the CPU never modifies these settings.
It loads them on VM entry and updates an internal copy instead.

Also don't load them from the vmcb12 as we don't expose these
features to the nested guest yet.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20210914154825.104886-5-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: irqfd: avoid update unmodified entries of the routing
Longpeng(Mike) [Fri, 27 Aug 2021 08:00:03 +0000 (16:00 +0800)]
kvm: irqfd: avoid update unmodified entries of the routing

All of the irqfds would to be updated when update the irq
routing, it's too expensive if there're too many irqfds.

However we can reduce the cost by avoid some unnecessary
updates. For irqs of MSI type on X86, the update can be
saved if the msi values are not change.

The vfio migration could receives benefit from this optimi-
zaiton. The test VM has 128 vcpus and 8 VF (with 65 vectors
enabled), so the VM has more than 520 irqfds. We mesure the
cost of the vfio_msix_enable (in QEMU, it would set routing
for each irqfd) for each VF, and we can see the total cost
can be significantly reduced.

                Origin         Apply this Patch
1st             8              4
2nd             15             5
3rd             22             6
4th             24             6
5th             36             7
6th             44             7
7th             51             8
8th             58             8
Total           258ms          51ms

We're also tring to optimize the QEMU part [1], but it's still
worth to optimize the KVM to gain more benefits.

[1] https://lists.gnu.org/archive/html/qemu-devel/2021-08/msg04215.html

Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Message-Id: <20210827080003.2689-1-longpeng2@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Don't check unsync if the original spte is writible
Lai Jiangshan [Sat, 18 Sep 2021 00:56:36 +0000 (08:56 +0800)]
KVM: X86: Don't check unsync if the original spte is writible

If the original spte is writable, the target gfn should not be the
gfn of synchronized shadowpage and can continue to be writable.

When !can_unsync, speculative must be false.  So when the check of
"!can_unsync" is removed, we need to move the label of "out" up.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210918005636.3675-11-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Don't unsync pagetables when speculative
Lai Jiangshan [Sat, 18 Sep 2021 00:56:35 +0000 (08:56 +0800)]
KVM: X86: Don't unsync pagetables when speculative

We'd better only unsync the pagetable when there just was a really
write fault on a level-1 pagetable.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210918005636.3675-10-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Remove FNAME(update_pte)
Lai Jiangshan [Sat, 18 Sep 2021 00:56:34 +0000 (08:56 +0800)]
KVM: X86: Remove FNAME(update_pte)

Its solo caller is changed to use FNAME(prefetch_gpte) directly.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210918005636.3675-9-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Zap the invalid list after remote tlb flushing
Lai Jiangshan [Sat, 18 Sep 2021 00:56:33 +0000 (08:56 +0800)]
KVM: X86: Zap the invalid list after remote tlb flushing

In mmu_sync_children(), it can zap the invalid list after remote tlb flushing.
Emptifying the invalid list ASAP might help reduce a remote tlb flushing
in some cases.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210918005636.3675-8-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Change kvm_sync_page() to return true when remote flush is needed
Lai Jiangshan [Sat, 18 Sep 2021 00:56:32 +0000 (08:56 +0800)]
KVM: X86: Change kvm_sync_page() to return true when remote flush is needed

Currently kvm_sync_page() returns true when there is any present spte.
But the return value is ignored in the callers.

Changing kvm_sync_page() to return true when remote flush is needed and
changing mmu->sync_page() not to directly flush can combine and reduce
remote flush requests.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210918005636.3675-7-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Remove kvm_mmu_flush_or_zap()
Lai Jiangshan [Sat, 18 Sep 2021 00:56:31 +0000 (08:56 +0800)]
KVM: X86: Remove kvm_mmu_flush_or_zap()

Because local_flush is useless, kvm_mmu_flush_or_zap() can be removed
and kvm_mmu_remote_flush_or_zap is used instead.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210918005636.3675-6-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Don't flush current tlb on shadow page modification
Lai Jiangshan [Sat, 18 Sep 2021 00:56:30 +0000 (08:56 +0800)]
KVM: X86: Don't flush current tlb on shadow page modification

After any shadow page modification, flushing tlb only on current VCPU
is weird due to other VCPU's tlb might still be stale.

In other words, if there is any mandatory tlb-flushing after shadow page
modification, SET_SPTE_NEED_REMOTE_TLB_FLUSH or remote_flush should be
set and the tlbs of all VCPUs should be flushed.  There is not point to
only flush current tlb except when the request is from vCPU's or pCPU's
activities.

If there was any bug that mandatory tlb-flushing is required and
SET_SPTE_NEED_REMOTE_TLB_FLUSH/remote_flush is failed to set, this patch
would expose the bug in a more destructive way.  The related code paths
are checked and no missing SET_SPTE_NEED_REMOTE_TLB_FLUSH is found yet.

Currently, there is no optional tlb-flushing after sync page related code
is changed to flush tlb timely.  So we can just remove these local flushing
code.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210918005636.3675-5-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Complete prefetch for trailing SPTEs for direct, legacy MMU
Sean Christopherson [Wed, 18 Aug 2021 23:56:15 +0000 (23:56 +0000)]
KVM: x86/mmu: Complete prefetch for trailing SPTEs for direct, legacy MMU

Make a final call to direct_pte_prefetch_many() if there are "trailing"
SPTEs to prefetch, i.e. SPTEs for GFNs following the faulting GFN.  The
call to direct_pte_prefetch_many() in the loop only handles the case
where there are !PRESENT SPTEs preceding a PRESENT SPTE.

E.g. if the faulting GFN is a multiple of 8 (the prefetch size) and all
SPTEs for the following GFNs are !PRESENT, the loop will terminate with
"start = sptep+1" and not prefetch any SPTEs.

Prefetching trailing SPTEs as intended can drastically reduce the number
of guest page faults, e.g. accessing the first byte of every 4kb page in
a 6gb chunk of virtual memory, in a VM with 8gb of preallocated memory,
the number of pf_fixed events observed in L0 drops from ~1.75M to <0.27M.

Note, this only affects memory that is backed by 4kb pages as KVM doesn't
prefetch when installing hugepages.  Shadow paging prefetching is not
affected as it does not batch the prefetches due to the need to process
the corresponding guest PTE.  The TDP MMU is not affected because it
doesn't have prefetching, yet...

Fixes: 957ed9effd80 ("KVM: MMU: prefetch ptes when intercepted guest #PF")
Cc: Sergey Senozhatsky <senozhatsky@google.com>
Cc: Ben Gardon <bgardon@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210818235615.2047588-1-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: Fix kvm_vm_free() in cr4_cpuid_sync and vmx_tsc_adjust tests
Thomas Huth [Thu, 26 Aug 2021 07:49:28 +0000 (09:49 +0200)]
KVM: selftests: Fix kvm_vm_free() in cr4_cpuid_sync and vmx_tsc_adjust tests

The kvm_vm_free() statement here is currently dead code, since the loop
in front of it can only be left with the "goto done" that jumps right
after the kvm_vm_free(). Fix it by swapping the locations of the "done"
label and the kvm_vm_free().

Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210826074928.240942-1-thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: selftests: Fix spelling mistake "missmatch" -> "mismatch"
Colin Ian King [Thu, 26 Aug 2021 12:07:52 +0000 (13:07 +0100)]
kvm: selftests: Fix spelling mistake "missmatch" -> "mismatch"

There is a spelling mistake in an error message. Fix it.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Message-Id: <20210826120752.12633-1-colin.king@canonical.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Manually retrieve CPUID.0x1 when getting FMS for RESET/INIT
Sean Christopherson [Wed, 29 Sep 2021 22:24:26 +0000 (15:24 -0700)]
KVM: x86: Manually retrieve CPUID.0x1 when getting FMS for RESET/INIT

Manually look for a CPUID.0x1 entry instead of bouncing through
kvm_cpuid() when retrieving the Family-Model-Stepping information for
vCPU RESET/INIT.  This fixes a potential undefined behavior bug due to
kvm_cpuid() using the uninitialized "dummy" param as the ECX _input_,
a.k.a. the index.

A more minimal fix would be to simply zero "dummy", but the extra work in
kvm_cpuid() is wasteful, and KVM should be treating the FMS retrieval as
an out-of-band access, e.g. same as how KVM computes guest.MAXPHYADDR.
Both Intel's SDM and AMD's APM describe the RDX value at RESET/INIT as
holding the CPU's FMS information, not as holding CPUID.0x1.EAX.  KVM's
usage of CPUID entries to get FMS is simply a pragmatic approach to avoid
having yet another way for userspace to provide inconsistent data.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Message-Id: <20210929222426.1855730-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: WARN on non-zero CRs at RESET to detect improper initalization
Sean Christopherson [Tue, 21 Sep 2021 00:03:03 +0000 (17:03 -0700)]
KVM: x86: WARN on non-zero CRs at RESET to detect improper initalization

WARN if CR0, CR3, or CR4 are non-zero at RESET, which given the current
KVM implementation, really means WARN if they're not zeroed at vCPU
creation.  VMX in particular has several ->set_*() flows that read other
registers to handle side effects, and because those flows are common to
RESET and INIT, KVM subtly relies on emulated/virtualized registers to be
zeroed at vCPU creation in order to do the right thing at RESET.

Use CRs as a sentinel because they are most likely to be written as side
effects, and because KVM specifically needs CR0.PG and CR0.PE to be '0'
to correctly reflect the state of the vCPU's MMU.  CRs are also loaded
and stored from/to the VMCS, and so adds some level of coverage to verify
that KVM doesn't conflate zero-allocating the VMCS with properly
initializing the VMCS with VMWRITEs.

Note, '0' is somewhat arbitrary, vCPU creation can technically stuff any
value for a register so long as it's coherent with respect to the current
vCPU state.  In practice, '0' works for all registers and is convenient.

Suggested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210921000303.400537-11-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: SVM: Move RESET emulation to svm_vcpu_reset()
Sean Christopherson [Tue, 21 Sep 2021 00:03:02 +0000 (17:03 -0700)]
KVM: SVM: Move RESET emulation to svm_vcpu_reset()

Move RESET emulation for SVM vCPUs to svm_vcpu_reset(), and drop an extra
init_vmcb() from svm_create_vcpu() in the process.  Hopefully KVM will
someday expose a dedicated RESET ioctl(), and in the meantime separating
"create" from "RESET" is a nice cleanup.

Keep the call to svm_switch_vmcb() so that misuse of svm->vmcb at worst
breaks the guest, e.g. premature accesses doesn't cause a NULL pointer
dereference.

Cc: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210921000303.400537-10-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: Move RESET emulation to vmx_vcpu_reset()
Sean Christopherson [Tue, 21 Sep 2021 00:03:01 +0000 (17:03 -0700)]
KVM: VMX: Move RESET emulation to vmx_vcpu_reset()

Move vCPU RESET emulation, including initializating of select VMCS state,
to vmx_vcpu_reset().  Drop the open coded "vCPU load" sequence, as
->vcpu_reset() is invoked while the vCPU is properly loaded (which is
kind of the point of ->vcpu_reset()...).  Hopefully KVM will someday
expose a dedicated RESET ioctl(), and in the meantime separating "create"
from "RESET" is a nice cleanup.

Deferring VMCS initialization is effectively a nop as it's impossible to
safely access the VMCS between the current call site and its new home, as
both the vCPU and the pCPU are put immediately after init_vmcs(), i.e.
the VMCS isn't guaranteed to be loaded.

Note, task preemption is not a problem as vmx_sched_in() _can't_ touch
the VMCS as ->sched_in() is invoked before the vCPU, and thus VMCS, is
reloaded.  I.e. the preemption path also can't consume VMCS state.

Cc: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210921000303.400537-9-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: Drop explicit zeroing of MSR guest values at vCPU creation
Sean Christopherson [Tue, 21 Sep 2021 00:03:00 +0000 (17:03 -0700)]
KVM: VMX: Drop explicit zeroing of MSR guest values at vCPU creation

Don't zero out user return and nested MSRs during vCPU creation, and
instead rely on vcpu_vmx being zero-allocated.  Explicitly zeroing MSRs
is not wrong, and is in fact necessary if KVM ever emulates vCPU RESET
outside of vCPU creation, but zeroing only a subset of MSRs is confusing.

Poking directly into KVM's backing is also undesirable in that it doesn't
scale and is error prone.  Ideally KVM would have a common RESET path for
all MSRs, e.g. by expanding kvm_set_msr(), which would obviate the need
for this out-of-bad code (to support standalone RESET).

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210921000303.400537-8-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Fold fx_init() into kvm_arch_vcpu_create()
Sean Christopherson [Tue, 21 Sep 2021 00:02:59 +0000 (17:02 -0700)]
KVM: x86: Fold fx_init() into kvm_arch_vcpu_create()

Move the few bits of relevant fx_init() code into kvm_arch_vcpu_create(),
dropping the superfluous check on vcpu->arch.guest_fpu that was blindly
and wrongly added by commit ed02b213098a ("KVM: SVM: Guest FPU state
save/restore not needed for SEV-ES guest").

Note, KVM currently allocates and then frees FPU state for SEV-ES guests,
rather than avoid the allocation in the first place.  While that approach
is inarguably inefficient and unnecessary, it's a cleanup for the future.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210921000303.400537-7-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Remove defunct setting of XCR0 for guest during vCPU create
Sean Christopherson [Tue, 21 Sep 2021 00:02:58 +0000 (17:02 -0700)]
KVM: x86: Remove defunct setting of XCR0 for guest during vCPU create

Drop code to initialize XCR0 during fx_init(), a.k.a. vCPU creation, as
XCR0 has been initialized during kvm_vcpu_reset() (for RESET) since
commit a554d207dc46 ("KVM: X86: Processor States following Reset or INIT").

Back when XCR0 support was added by commit 2acf923e38fb ("KVM: VMX:
Enable XSAVE/XRSTOR for guest"), KVM didn't differentiate between RESET
and INIT.  Ignoring the fact that calling fx_init() for INIT is obviously
wrong, e.g. FPU state after INIT is not the same as after RESET, setting
XCR0 in fx_init() was correct.

Eventually fx_init() got moved to kvm_arch_vcpu_init(), a.k.a. vCPU
creation (ignore the terrible name) by commit 0ee6a5172573 ("x86/fpu,
kvm: Simplify fx_init()").  Finally, commit 95a0d01eef7a ("KVM: x86: Move
all vcpu init code into kvm_arch_vcpu_create()") killed off
kvm_arch_vcpu_init(), leaving behind the oddity of redundant setting of
guest state during vCPU creation.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210921000303.400537-6-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Remove defunct setting of CR0.ET for guests during vCPU create
Sean Christopherson [Tue, 21 Sep 2021 00:02:57 +0000 (17:02 -0700)]
KVM: x86: Remove defunct setting of CR0.ET for guests during vCPU create

Drop code to set CR0.ET for the guest during initialization of the guest
FPU.  The code was added as a misguided bug fix by commit 380102c8e431
("KVM Set the ET flag in CR0 after initializing FX") to resolve an issue
where vcpu->cr0 (now vcpu->arch.cr0) was not correctly initialized on SVM
systems.  While init_vmcb() did set CR0.ET, it only did so in the VMCB,
and subtly did not update vcpu->cr0.  Stuffing CR0.ET worked around the
immediate problem, but did not fix the real bug of vcpu->cr0 and the VMCB
being out of sync.  That underlying bug was eventually remedied by commit
18fa000ae453 ("KVM: SVM: Reset cr0 properly on vcpu reset").

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210921000303.400537-5-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Do not mark all registers as avail/dirty during RESET/INIT
Sean Christopherson [Tue, 21 Sep 2021 00:02:56 +0000 (17:02 -0700)]
KVM: x86: Do not mark all registers as avail/dirty during RESET/INIT

Do not blindly mark all registers as available+dirty at RESET/INIT, and
instead rely on writes to registers to go through the proper mutators or
to explicitly mark registers as dirty.  INIT in particular does not blindly
overwrite all registers, e.g. select bits in CR0 are preserved across INIT,
thus marking registers available+dirty without first reading the register
from hardware is incorrect.

In practice this is a benign bug as KVM doesn't let the guest control CR0
bits that are preserved across INIT, and all other true registers are
explicitly written during the RESET/INIT flows.  The PDPTRs and EX_INFO
"registers" are not explicitly written, but accessing those values during
RESET/INIT is nonsensical and would be a KVM bug regardless of register
caching.

Fixes: 66f7b72e1171 ("KVM: x86: Make register state after reset conform to specification")
[sean: !!! NOT FOR STABLE !!!]
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210921000303.400537-4-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Simplify retrieving the page offset when loading PDTPRs
Sean Christopherson [Tue, 31 Aug 2021 16:42:24 +0000 (09:42 -0700)]
KVM: x86: Simplify retrieving the page offset when loading PDTPRs

Replace impressively complex "logic" for computing the page offset from
CR3 when loading PDPTRs.  Unlike other paging modes, the address held in
CR3 for PAE paging is 32-byte aligned, i.e. occupies bits 31:5, thus bits
11:5 need to be used as the offset from the gfn when reading PDPTRs.

The existing calculation originated in commit 1342d3536d6a ("[PATCH] KVM:
MMU: Load the pae pdptrs on cr3 change like the processor does"), which
read the PDPTRs from guest memory as individual 8-byte loads.  At the
time, the so called "offset" was the base index of PDPTR0 as a _u64_, not
a byte offset.  Naming aside, the computation was useful and arguably
simplified the overall flow.

Unfortunately, when commit 195aefde9cc2 ("KVM: Add general accessors to
read and write guest memory") added accessors with offsets at byte
granularity, the cleverness of the original code was lost and KVM was
left with convoluted code for a simple operation.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210831164224.1119728-4-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Subsume nested GPA read helper into load_pdptrs()
Sean Christopherson [Tue, 31 Aug 2021 16:42:23 +0000 (09:42 -0700)]
KVM: x86: Subsume nested GPA read helper into load_pdptrs()

Open code the call to mmu->translate_gpa() when loading nested PDPTRs and
kill off the existing helper, kvm_read_guest_page_mmu(), to discourage
incorrect use.  Reading guest memory straight from an L2 GPA is extremely
rare (as evidenced by the lack of users), as very few constructs in x86
specify physical addresses, even fewer are virtualized by KVM, and even
fewer yet require emulation of L2 by L0 KVM.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210831164224.1119728-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: rename KVM_MAX_VCPU_ID to KVM_MAX_VCPU_IDS
Juergen Gross [Mon, 13 Sep 2021 13:57:44 +0000 (15:57 +0200)]
kvm: rename KVM_MAX_VCPU_ID to KVM_MAX_VCPU_IDS

KVM_MAX_VCPU_ID is not specifying the highest allowed vcpu-id, but the
number of allowed vcpu-ids. This has already led to confusion, so
rename KVM_MAX_VCPU_ID to KVM_MAX_VCPU_IDS to make its semantics more
clear

Suggested-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210913135745.13944-3-jgross@suse.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoRevert "x86/kvm: fix vcpu-id indexed array sizes"
Juergen Gross [Mon, 13 Sep 2021 13:57:43 +0000 (15:57 +0200)]
Revert "x86/kvm: fix vcpu-id indexed array sizes"

This reverts commit 76b4f357d0e7d8f6f0013c733e6cba1773c266d3.

The commit has the wrong reasoning, as KVM_MAX_VCPU_ID is not defining the
maximum allowed vcpu-id as its name suggests, but the number of vcpu-ids.
So revert this patch again.

Suggested-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210913135745.13944-2-jgross@suse.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: Make kvm_make_vcpus_request_mask() use pre-allocated cpu_kick_mask
Vitaly Kuznetsov [Fri, 3 Sep 2021 07:51:41 +0000 (09:51 +0200)]
KVM: Make kvm_make_vcpus_request_mask() use pre-allocated cpu_kick_mask

kvm_make_vcpus_request_mask() already disables preemption so just like
kvm_make_all_cpus_request_except() it can be switched to using
pre-allocated per-cpu cpumasks. This allows for improvements for both
users of the function: in Hyper-V emulation code 'tlb_flush' can now be
dropped from 'struct kvm_vcpu_hv' and kvm_make_scan_ioapic_request_mask()
gets rid of dynamic allocation.

cpumask_available() checks in kvm_make_vcpu_request() and
kvm_kick_many_cpus() can now be dropped as they checks for an impossible
condition: kvm_init() makes sure per-cpu masks are allocated.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210903075141.403071-9-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: Pre-allocate cpumasks for kvm_make_all_cpus_request_except()
Vitaly Kuznetsov [Fri, 3 Sep 2021 07:51:40 +0000 (09:51 +0200)]
KVM: Pre-allocate cpumasks for kvm_make_all_cpus_request_except()

Allocating cpumask dynamically in zalloc_cpumask_var() is not ideal.
Allocation is somewhat slow and can (in theory and when CPUMASK_OFFSTACK)
fail. kvm_make_all_cpus_request_except() already disables preemption so
we can use pre-allocated per-cpu cpumasks instead.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210903075141.403071-8-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: Drop 'except' parameter from kvm_make_vcpus_request_mask()
Vitaly Kuznetsov [Fri, 3 Sep 2021 07:51:38 +0000 (09:51 +0200)]
KVM: Drop 'except' parameter from kvm_make_vcpus_request_mask()

Both remaining callers of kvm_make_vcpus_request_mask() pass 'NULL' for
'except' parameter so it can just be dropped.

No functional change intended ©.

Suggested-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210903075141.403071-6-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: Optimize kvm_make_vcpus_request_mask() a bit
Vitaly Kuznetsov [Fri, 3 Sep 2021 07:51:37 +0000 (09:51 +0200)]
KVM: Optimize kvm_make_vcpus_request_mask() a bit

Iterating over set bits in 'vcpu_bitmap' should be faster than going
through all vCPUs, especially when just a few bits are set.

Drop kvm_make_vcpus_request_mask() call from kvm_make_all_cpus_request_except()
to avoid handling the special case when 'vcpu_bitmap' is NULL, move the
code to kvm_make_all_cpus_request_except() itself.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210903075141.403071-5-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>