platform/kernel/linux-starfive.git
2 years agoKVM: x86: Fix uninitialized eoi_exit_bitmap usage in vcpu_load_eoi_exitmap()
黄乐 [Mon, 15 Nov 2021 14:08:29 +0000 (14:08 +0000)]
KVM: x86: Fix uninitialized eoi_exit_bitmap usage in vcpu_load_eoi_exitmap()

In vcpu_load_eoi_exitmap(), currently the eoi_exit_bitmap[4] array is
initialized only when Hyper-V context is available, in other path it is
just passed to kvm_x86_ops.load_eoi_exitmap() directly from on the stack,
which would cause unexpected interrupt delivery/handling issues, e.g. an
*old* linux kernel that relies on PIT to do clock calibration on KVM might
randomly fail to boot.

Fix it by passing ioapic_handled_vectors to load_eoi_exitmap() when Hyper-V
context is not available.

Fixes: f2bc14b69c38 ("KVM: x86: hyper-v: Prepare to meet unallocated Hyper-V context")
Cc: stable@vger.kernel.org
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Huang Le <huangle1@jd.com>
Message-Id: <62115b277dab49ea97da5633f8522daf@jd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Drop arbitrary KVM_SOFT_MAX_VCPUS
Vitaly Kuznetsov [Thu, 11 Nov 2021 13:47:33 +0000 (14:47 +0100)]
KVM: x86: Drop arbitrary KVM_SOFT_MAX_VCPUS

KVM_CAP_NR_VCPUS is used to get the "recommended" maximum number of
VCPUs and arm64/mips/riscv report num_online_cpus(). Powerpc reports
either num_online_cpus() or num_present_cpus(), s390 has multiple
constants depending on hardware features. On x86, KVM reports an
arbitrary value of '710' which is supposed to be the maximum tested
value but it's possible to test all KVM_MAX_VCPUS even when there are
less physical CPUs available.

Drop the arbitrary '710' value and return num_online_cpus() on x86 as
well. The recommendation will match other architectures and will mean
'no CPU overcommit'.

For reference, QEMU only queries KVM_CAP_NR_VCPUS to print a warning
when the requested vCPU number exceeds it. The static limit of '710'
is quite weird as smaller systems with just a few physical CPUs should
certainly "recommend" less.

Suggested-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20211111134733.86601-1-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: Move INVPCID type check from vmx and svm to the common kvm_handle_invpcid()
Vipin Sharma [Tue, 9 Nov 2021 17:44:26 +0000 (17:44 +0000)]
KVM: Move INVPCID type check from vmx and svm to the common kvm_handle_invpcid()

Handle #GP on INVPCID due to an invalid type in the common switch
statement instead of relying on the callers (VMX and SVM) to manually
validate the type.

Unlike INVVPID and INVEPT, INVPCID is not explicitly documented to check
the type before reading the operand from memory, so deferring the
type validity check until after that point is architecturally allowed.

Signed-off-by: Vipin Sharma <vipinsh@google.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211109174426.2350547-3-vipinsh@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: VMX: Add a helper function to retrieve the GPR index for INVPCID, INVVPID, and...
Vipin Sharma [Tue, 9 Nov 2021 17:44:25 +0000 (17:44 +0000)]
KVM: VMX: Add a helper function to retrieve the GPR index for INVPCID, INVVPID, and INVEPT

handle_invept(), handle_invvpid(), handle_invpcid() read the same reg2
field in vmcs.VMX_INSTRUCTION_INFO to get the index of the GPR that
holds the invalidation type. Add a helper to retrieve reg2 from VMX
instruction info to consolidate and document the shift+mask magic.

Signed-off-by: Vipin Sharma <vipinsh@google.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211109174426.2350547-2-vipinsh@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: nVMX: Clean up x2APIC MSR handling for L2
Sean Christopherson [Tue, 9 Nov 2021 01:30:47 +0000 (01:30 +0000)]
KVM: nVMX: Clean up x2APIC MSR handling for L2

Clean up the x2APIC MSR bitmap intereption code for L2, which is the last
holdout of open coded bitmap manipulations.  Freshen up the SDM/PRM
comment, rename the function to make it abundantly clear the funky
behavior is x2APIC specific, and explain _why_ vmcs01's bitmap is ignored
(the previous comment was flat out wrong for x2APIC behavior).

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211109013047.2041518-5-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: VMX: Macrofy the MSR bitmap getters and setters
Sean Christopherson [Tue, 9 Nov 2021 01:30:46 +0000 (01:30 +0000)]
KVM: VMX: Macrofy the MSR bitmap getters and setters

Add builder macros to generate the MSR bitmap helpers to reduce the
amount of copy-paste code, especially with respect to all the magic
numbers needed to calc the correct bit location.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211109013047.2041518-4-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: nVMX: Handle dynamic MSR intercept toggling
Sean Christopherson [Tue, 9 Nov 2021 01:30:45 +0000 (01:30 +0000)]
KVM: nVMX: Handle dynamic MSR intercept toggling

Always check vmcs01's MSR bitmap when merging L0 and L1 bitmaps for L2,
and always update the relevant bits in vmcs02.  This fixes two distinct,
but intertwined bugs related to dynamic MSR bitmap modifications.

The first issue is that KVM fails to enable MSR interception in vmcs02
for the FS/GS base MSRs if L1 first runs L2 with interception disabled,
and later enables interception.

The second issue is that KVM fails to honor userspace MSR filtering when
preparing vmcs02.

Fix both issues simultaneous as fixing only one of the issues (doesn't
matter which) would create a mess that no one should have to bisect.
Fixing only the first bug would exacerbate the MSR filtering issue as
userspace would see inconsistent behavior depending on the whims of L1.
Fixing only the second bug (MSR filtering) effectively requires fixing
the first, as the nVMX code only knows how to transition vmcs02's
bitmap from 1->0.

Move the various accessor/mutators that are currently buried in vmx.c
into vmx.h so that they can be shared by the nested code.

Fixes: 1a155254ff93 ("KVM: x86: Introduce MSR filtering")
Fixes: d69129b4e46a ("KVM: nVMX: Disable intercept for FS/GS base MSRs in vmcs02 when possible")
Cc: stable@vger.kernel.org
Cc: Alexander Graf <graf@amazon.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211109013047.2041518-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: nVMX: Query current VMCS when determining if MSR bitmaps are in use
Sean Christopherson [Tue, 9 Nov 2021 01:30:44 +0000 (01:30 +0000)]
KVM: nVMX: Query current VMCS when determining if MSR bitmaps are in use

Check the current VMCS controls to determine if an MSR write will be
intercepted due to MSR bitmaps being disabled.  In the nested VMX case,
KVM will disable MSR bitmaps in vmcs02 if they're disabled in vmcs12 or
if KVM can't map L1's bitmaps for whatever reason.

Note, the bad behavior is relatively benign in the current code base as
KVM sets all bits in vmcs02's MSR bitmap by default, clears bits if and
only if L0 KVM also disables interception of an MSR, and only uses the
buggy helper for MSR_IA32_SPEC_CTRL.  Because KVM explicitly tests WRMSR
before disabling interception of MSR_IA32_SPEC_CTRL, the flawed check
will only result in KVM reading MSR_IA32_SPEC_CTRL from hardware when it
isn't strictly necessary.

Tag the fix for stable in case a future fix wants to use
msr_write_intercepted(), in which case a buggy implementation in older
kernels could prove subtly problematic.

Fixes: d28b387fb74d ("KVM/VMX: Allow direct access to MSR_IA32_SPEC_CTRL")
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211109013047.2041518-2-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Don't update vcpu->arch.pv_eoi.msr_val when a bogus value was written to...
Vitaly Kuznetsov [Mon, 8 Nov 2021 15:28:19 +0000 (16:28 +0100)]
KVM: x86: Don't update vcpu->arch.pv_eoi.msr_val when a bogus value was written to MSR_KVM_PV_EOI_EN

When kvm_gfn_to_hva_cache_init() call from kvm_lapic_set_pv_eoi() fails,
MSR write to MSR_KVM_PV_EOI_EN results in #GP so it is reasonable to
expect that the value we keep internally in KVM wasn't updated.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20211108152819.12485-3-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Rename kvm_lapic_enable_pv_eoi()
Vitaly Kuznetsov [Mon, 8 Nov 2021 15:28:18 +0000 (16:28 +0100)]
KVM: x86: Rename kvm_lapic_enable_pv_eoi()

kvm_lapic_enable_pv_eoi() is a misnomer as the function is also
used to disable PV EOI. Rename it to kvm_lapic_set_pv_eoi().

No functional change intended.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20211108152819.12485-2-vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Make sure KVM_CPUID_FEATURES really are KVM_CPUID_FEATURES
Paul Durrant [Fri, 5 Nov 2021 09:51:01 +0000 (09:51 +0000)]
KVM: x86: Make sure KVM_CPUID_FEATURES really are KVM_CPUID_FEATURES

Currently when kvm_update_cpuid_runtime() runs, it assumes that the
KVM_CPUID_FEATURES leaf is located at 0x40000001. This is not true,
however, if Hyper-V support is enabled. In this case the KVM leaves will
be offset.

This patch introdues as new 'kvm_cpuid_base' field into struct
kvm_vcpu_arch to track the location of the KVM leaves and function
kvm_update_kvm_cpuid_base() (called from kvm_set_cpuid()) to locate the
leaves using the 'KVMKVMKVM\0\0\0' signature (which is now given a
definition in kvm_para.h). Adjustment of KVM_CPUID_FEATURES will hence now
target the correct leaf.

NOTE: A new for_each_possible_hypervisor_cpuid_base() macro is intoduced
      into processor.h to avoid having duplicate code for the iteration
      over possible hypervisor base leaves.

Signed-off-by: Paul Durrant <pdurrant@amazon.com>
Message-Id: <20211105095101.5384-3-pdurrant@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Add helper to consolidate core logic of SET_CPUID{2} flows
Sean Christopherson [Fri, 5 Nov 2021 09:51:00 +0000 (09:51 +0000)]
KVM: x86: Add helper to consolidate core logic of SET_CPUID{2} flows

Move the core logic of SET_CPUID and SET_CPUID2 to a common helper, the
only difference between the two ioctls() is the format of the userspace
struct.  A future fix will add yet more code to the core logic.

No functional change intended.

Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211105095101.5384-2-pdurrant@amazon.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agokvm: mmu: Use fast PF path for access tracking of huge pages when possible
Junaid Shahid [Thu, 4 Nov 2021 00:33:59 +0000 (17:33 -0700)]
kvm: mmu: Use fast PF path for access tracking of huge pages when possible

The fast page fault path bails out on write faults to huge pages in
order to accommodate dirty logging. This change adds a check to do that
only when dirty logging is actually enabled, so that access tracking for
huge pages can still use the fast path for write faults in the common
case.

Signed-off-by: Junaid Shahid <junaids@google.com>
Reviewed-by: Ben Gardon <bgardon@google.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211104003359.2201967-1-junaids@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86/mmu: Properly dereference rcu-protected TDP MMU sptep iterator
Sean Christopherson [Wed, 3 Nov 2021 16:18:33 +0000 (09:18 -0700)]
KVM: x86/mmu: Properly dereference rcu-protected TDP MMU sptep iterator

Wrap the read of iter->sptep in tdp_mmu_map_handle_target_level() with
rcu_dereference().  Shadow pages in the TDP MMU, and thus their SPTEs,
are protected by rcu.

This fixes a Sparse warning at tdp_mmu.c:900:51:
  warning: incorrect type in argument 1 (different address spaces)
  expected unsigned long long [usertype] *sptep
  got unsigned long long [noderef] [usertype] __rcu *[usertype] sptep

Fixes: 7158bee4b475 ("KVM: MMU: pass kvm_mmu_page struct to make_spte")
Cc: Ben Gardon <bgardon@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211103161833.3769487-1-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: inhibit APICv when KVM_GUESTDBG_BLOCKIRQ active
Maxim Levitsky [Mon, 8 Nov 2021 09:02:45 +0000 (11:02 +0200)]
KVM: x86: inhibit APICv when KVM_GUESTDBG_BLOCKIRQ active

KVM_GUESTDBG_BLOCKIRQ relies on interrupts being injected using
standard kvm's inject_pending_event, and not via APICv/AVIC.

Since this is a debug feature, just inhibit APICv/AVIC while
KVM_GUESTDBG_BLOCKIRQ is in use on at least one vCPU.

Fixes: 61e5f69ef0837 ("KVM: x86: implement KVM_GUESTDBG_BLOCKIRQ")

Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Tested-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211108090245.166408-1-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agokvm: x86: Convert return type of *is_valid_rdpmc_ecx() to bool
Jim Mattson [Fri, 5 Nov 2021 20:20:58 +0000 (13:20 -0700)]
kvm: x86: Convert return type of *is_valid_rdpmc_ecx() to bool

These function names sound like predicates, and they have siblings,
*is_valid_msr(), which _are_ predicates. Moreover, there are comments
that essentially warn that these functions behave unexpectedly.

Flip the polarity of the return values, so that they become
predicates, and convert the boolean result to a success/failure code
at the outer call site.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211105202058.1048757-1-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKVM: x86: Fix recording of guest steal time / preempted status
David Woodhouse [Tue, 2 Nov 2021 17:36:39 +0000 (17:36 +0000)]
KVM: x86: Fix recording of guest steal time / preempted status

In commit b043138246a4 ("x86/KVM: Make sure KVM_VCPU_FLUSH_TLB flag is
not missed") we switched to using a gfn_to_pfn_cache for accessing the
guest steal time structure in order to allow for an atomic xchg of the
preempted field. This has a couple of problems.

Firstly, kvm_map_gfn() doesn't work at all for IOMEM pages when the
atomic flag is set, which it is in kvm_steal_time_set_preempted(). So a
guest vCPU using an IOMEM page for its steal time would never have its
preempted field set.

Secondly, the gfn_to_pfn_cache is not invalidated in all cases where it
should have been. There are two stages to the GFN->PFN conversion;
first the GFN is converted to a userspace HVA, and then that HVA is
looked up in the process page tables to find the underlying host PFN.
Correct invalidation of the latter would require being hooked up to the
MMU notifiers, but that doesn't happen---so it just keeps mapping and
unmapping the *wrong* PFN after the userspace page tables change.

In the !IOMEM case at least the stale page *is* pinned all the time it's
cached, so it won't be freed and reused by anyone else while still
receiving the steal time updates. The map/unmap dance only takes care
of the KVM administrivia such as marking the page dirty.

Until the gfn_to_pfn cache handles the remapping automatically by
integrating with the MMU notifiers, we might as well not get a
kernel mapping of it, and use the perfectly serviceable userspace HVA
that we already have.  We just need to implement the atomic xchg on
the userspace address with appropriate exception handling, which is
fairly trivial.

Cc: stable@vger.kernel.org
Fixes: b043138246a4 ("x86/KVM: Make sure KVM_VCPU_FLUSH_TLB flag is not missed")
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <3645b9b889dac6438394194bb5586a46b68d581f.camel@infradead.org>
[I didn't entirely agree with David's assessment of the
 usefulness of the gfn_to_pfn cache, and integrated the outcome
 of the discussion in the above commit message. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoMerge tag 'kvm-riscv-5.16-2' of https://github.com/kvm-riscv/linux into HEAD
Paolo Bonzini [Tue, 2 Nov 2021 13:15:31 +0000 (09:15 -0400)]
Merge tag 'kvm-riscv-5.16-2' of https://github.com/kvm-riscv/linux into HEAD

Minor cocci warning fixes:
1) Bool return warning fix
2) Unnedded semicolon warning fix

3 years agoRISC-V: KVM: fix boolreturn.cocci warnings
Bixuan Cui [Wed, 27 Oct 2021 07:20:53 +0000 (15:20 +0800)]
RISC-V: KVM: fix boolreturn.cocci warnings

Fix boolreturn.cocci warnings:
./arch/riscv/kvm/mmu.c:603:9-10: WARNING: return of 0/1 in function
'kvm_age_gfn' with return type bool
./arch/riscv/kvm/mmu.c:582:9-10: WARNING: return of 0/1 in function
'kvm_set_spte_gfn' with return type bool
./arch/riscv/kvm/mmu.c:621:9-10: WARNING: return of 0/1 in function
'kvm_test_age_gfn' with return type bool
./arch/riscv/kvm/mmu.c:568:9-10: WARNING: return of 0/1 in function
'kvm_unmap_gfn_range' with return type bool

Signed-off-by: Bixuan Cui <cuibixuan@linux.alibaba.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
3 years agoRISC-V: KVM: remove unneeded semicolon
ran jianping [Thu, 21 Oct 2021 11:57:06 +0000 (11:57 +0000)]
RISC-V: KVM: remove unneeded semicolon

 Elimate the following coccinelle check warning:
 ./arch/riscv/kvm/vcpu_sbi.c:169:2-3: Unneeded semicolon
 ./arch/riscv/kvm/vcpu_exit.c:397:2-3: Unneeded semicolon
 ./arch/riscv/kvm/vcpu_exit.c:687:2-3: Unneeded semicolon
 ./arch/riscv/kvm/vcpu_exit.c:645:2-3: Unneeded semicolon
 ./arch/riscv/kvm/vcpu.c:247:2-3: Unneeded semicolon
 ./arch/riscv/kvm/vcpu.c:284:2-3: Unneeded semicolon
 ./arch/riscv/kvm/vcpu_timer.c:123:2-3: Unneeded semicolon
 ./arch/riscv/kvm/vcpu_timer.c:170:2-3: Unneeded semicolon

Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: ran jianping <ran.jianping@zte.com.cn>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
3 years agoMerge tag 'kvm-s390-next-5.16-1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Sun, 31 Oct 2021 13:01:25 +0000 (09:01 -0400)]
Merge tag 'kvm-s390-next-5.16-1' of git://git./linux/kernel/git/kvms390/linux into HEAD

KVM: s390: Fixes and Features for 5.16

- SIGP Fixes
- initial preparations for lazy destroy of secure VMs
- storage key improvements/fixes
- Log the guest CPNC

3 years agoRISC-V: KVM: Fix GPA passed to __kvm_riscv_hfence_gvma_xyz() functions
Anup Patel [Tue, 26 Oct 2021 17:01:36 +0000 (22:31 +0530)]
RISC-V: KVM: Fix GPA passed to __kvm_riscv_hfence_gvma_xyz() functions

The parameter passed to HFENCE.GVMA instruction in rs1 register
is guest physical address right shifted by 2 (i.e. divided by 4).

Unfortunately, we overlooked the semantics of rs1 registers for
HFENCE.GVMA instruction and never right shifted guest physical
address by 2. This issue did not manifest for hypervisors till
now because:
  1) Currently, only __kvm_riscv_hfence_gvma_all() and SBI
     HFENCE calls are used to invalidate TLB.
  2) All H-extension implementations (such as QEMU, Spike,
     Rocket Core FPGA, etc) that we tried till now were
     conservatively flushing everything upon any HFENCE.GVMA
     instruction.

This patch fixes GPA passed to __kvm_riscv_hfence_gvma_vmid_gpa()
and __kvm_riscv_hfence_gvma_gpa() functions.

Fixes: fd7bb4a251df ("RISC-V: KVM: Implement VMID allocator")
Reported-by: Ian Huang <ihuang@ventanamicro.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Message-Id: <20211026170136.2147619-4-anup.patel@wdc.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoRISC-V: KVM: Factor-out FP virtualization into separate sources
Anup Patel [Tue, 26 Oct 2021 17:01:35 +0000 (22:31 +0530)]
RISC-V: KVM: Factor-out FP virtualization into separate sources

The timer and SBI virtualization is already in separate sources.
In future, we will have vector and AIA virtualization also added
as separate sources.

To align with above described modularity, we factor-out FP
virtualization into separate sources.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Message-Id: <20211026170136.2147619-3-anup.patel@wdc.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoMerge tag 'kvmarm-5.16' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmar...
Paolo Bonzini [Sun, 31 Oct 2021 06:28:48 +0000 (02:28 -0400)]
Merge tag 'kvmarm-5.16' of git://git./linux/kernel/git/kvmarm/kvmarm into HEAD

KVM/arm64 updates for Linux 5.16

- More progress on the protected VM front, now with the full
  fixed feature set as well as the limitation of some hypercalls
  after initialisation.

- Cleanup of the RAZ/WI sysreg handling, which was pointlessly
  complicated

- Fixes for the vgic placement in the IPA space, together with a
  bunch of selftests

- More memcg accounting of the memory allocated on behalf of a guest

- Timer and vgic selftests

- Workarounds for the Apple M1 broken vgic implementation

- KConfig cleanups

- New kvmarm.mode=none option, for those who really dislike us

3 years agoKVM: s390: add debug statement for diag 318 CPNC data
Collin Walling [Wed, 27 Oct 2021 02:54:51 +0000 (22:54 -0400)]
KVM: s390: add debug statement for diag 318 CPNC data

The diag 318 data contains values that denote information regarding the
guest's environment. Currently, it is unecessarily difficult to observe
this value (either manually-inserted debug statements, gdb stepping, mem
dumping etc). It's useful to observe this information to obtain an
at-a-glance view of the guest's environment, so lets add a simple VCPU
event that prints the CPNC to the s390dbf logs.

Signed-off-by: Collin Walling <walling@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Link: https://lore.kernel.org/r/20211027025451.290124-1-walling@linux.ibm.com
[borntraeger@de.ibm.com]: change debug level to 3
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: s390: pv: properly handle page flags for protected guests
Claudio Imbrenda [Mon, 20 Sep 2021 13:24:54 +0000 (15:24 +0200)]
KVM: s390: pv: properly handle page flags for protected guests

Introduce variants of the convert and destroy page functions that also
clear the PG_arch_1 bit used to mark them as secure pages.

The PG_arch_1 flag is always allowed to overindicate; using the new
functions introduced here allows to reduce the extent of overindication
and thus improve performance.

These new functions can only be called on pages for which a reference
is already being held.

Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Link: https://lore.kernel.org/r/20210920132502.36111-7-imbrenda@linux.ibm.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: s390: Fix handle_sske page fault handling
Janis Schoetterl-Glausch [Fri, 22 Oct 2021 15:26:48 +0000 (17:26 +0200)]
KVM: s390: Fix handle_sske page fault handling

If handle_sske cannot set the storage key, because there is no
page table entry or no present large page entry, it calls
fixup_user_fault.
However, currently, if the call succeeds, handle_sske returns
-EAGAIN, without having set the storage key.
Instead, retry by continue'ing the loop without incrementing the
address.
The same issue in handle_pfmf was fixed by
a11bdb1a6b78 ("KVM: s390: Fix pfmf and conditional skey emulation").

Fixes: bd096f644319 ("KVM: s390: Add skey emulation fault handling")
Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Link: https://lore.kernel.org/r/20211022152648.26536-1-scgl@linux.ibm.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoMerge branch 'kvm-pvclock-raw-spinlock' into HEAD
Paolo Bonzini [Mon, 25 Oct 2021 12:39:43 +0000 (08:39 -0400)]
Merge branch 'kvm-pvclock-raw-spinlock' into HEAD

pvclock_gtod_sync_lock is completely gone in Linux 5.16.  Include this
fix into the kvm/next history to record that the syzkaller report is
not valid there.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: switch pvclock_gtod_sync_lock to a raw spinlock
David Woodhouse [Sat, 23 Oct 2021 20:29:22 +0000 (21:29 +0100)]
KVM: x86: switch pvclock_gtod_sync_lock to a raw spinlock

On the preemption path when updating a Xen guest's runstate times, this
lock is taken inside the scheduler rq->lock, which is a raw spinlock.
This was shown in a lockdep warning:

[   89.138354] =============================
[   89.138356] [ BUG: Invalid wait context ]
[   89.138358] 5.15.0-rc5+ #834 Tainted: G S        I E
[   89.138360] -----------------------------
[   89.138361] xen_shinfo_test/2575 is trying to lock:
[   89.138363] ffffa34a0364efd8 (&kvm->arch.pvclock_gtod_sync_lock){....}-{3:3}, at: get_kvmclock_ns+0x1f/0x130 [kvm]
[   89.138442] other info that might help us debug this:
[   89.138444] context-{5:5}
[   89.138445] 4 locks held by xen_shinfo_test/2575:
[   89.138447]  #0: ffff972bdc3b8108 (&vcpu->mutex){+.+.}-{4:4}, at: kvm_vcpu_ioctl+0x77/0x6f0 [kvm]
[   89.138483]  #1: ffffa34a03662e90 (&kvm->srcu){....}-{0:0}, at: kvm_arch_vcpu_ioctl_run+0xdc/0x8b0 [kvm]
[   89.138526]  #2: ffff97331fdbac98 (&rq->__lock){-.-.}-{2:2}, at: __schedule+0xff/0xbd0
[   89.138534]  #3: ffffa34a03662e90 (&kvm->srcu){....}-{0:0}, at: kvm_arch_vcpu_put+0x26/0x170 [kvm]
...
[   89.138695]  get_kvmclock_ns+0x1f/0x130 [kvm]
[   89.138734]  kvm_xen_update_runstate+0x14/0x90 [kvm]
[   89.138783]  kvm_xen_update_runstate_guest+0x15/0xd0 [kvm]
[   89.138830]  kvm_arch_vcpu_put+0xe6/0x170 [kvm]
[   89.138870]  kvm_sched_out+0x2f/0x40 [kvm]
[   89.138900]  __schedule+0x5de/0xbd0

Cc: stable@vger.kernel.org
Reported-by: syzbot+b282b65c2c68492df769@syzkaller.appspotmail.com
Fixes: 30b5c851af79 ("KVM: x86/xen: Add support for vCPU runstate information")
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <1b02a06421c17993df337493a68ba923f3bd5c0f.camel@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: SGX must obey the KVM_INTERNAL_ERROR_EMULATION protocol
David Edmondson [Mon, 20 Sep 2021 10:37:37 +0000 (11:37 +0100)]
KVM: x86: SGX must obey the KVM_INTERNAL_ERROR_EMULATION protocol

When passing the failing address and size out to user space, SGX must
ensure not to trample on the earlier fields of the emulation_failure
sub-union of struct kvm_run.

Signed-off-by: David Edmondson <david.edmondson@oracle.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210920103737.2696756-5-david.edmondson@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: On emulation failure, convey the exit reason, etc. to userspace
David Edmondson [Mon, 20 Sep 2021 10:37:36 +0000 (11:37 +0100)]
KVM: x86: On emulation failure, convey the exit reason, etc. to userspace

Should instruction emulation fail, include the VM exit reason, etc. in
the emulation_failure data passed to userspace, in order that the VMM
can report it as a debugging aid when describing the failure.

Suggested-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: David Edmondson <david.edmondson@oracle.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210920103737.2696756-4-david.edmondson@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Get exit_reason as part of kvm_x86_ops.get_exit_info
David Edmondson [Mon, 20 Sep 2021 10:37:35 +0000 (11:37 +0100)]
KVM: x86: Get exit_reason as part of kvm_x86_ops.get_exit_info

Extend the get_exit_info static call to provide the reason for the VM
exit. Modify relevant trace points to use this rather than extracting
the reason in the caller.

Signed-off-by: David Edmondson <david.edmondson@oracle.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210920103737.2696756-3-david.edmondson@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Clarify the kvm_run.emulation_failure structure layout
David Edmondson [Mon, 20 Sep 2021 10:37:34 +0000 (11:37 +0100)]
KVM: x86: Clarify the kvm_run.emulation_failure structure layout

Until more flags for kvm_run.emulation_failure flags are defined, it
is undetermined whether new payload elements corresponding to those
flags will be additive or alternative. As a hint to userspace that an
alternative is possible, wrap the current payload elements in a union.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: David Edmondson <david.edmondson@oracle.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210920103737.2696756-2-david.edmondson@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: s390: Add a routine for setting userspace CPU state
Eric Farman [Fri, 8 Oct 2021 20:31:12 +0000 (22:31 +0200)]
KVM: s390: Add a routine for setting userspace CPU state

This capability exists, but we don't record anything when userspace
enables it. Let's refactor that code so that a note can be made in
the debug logs that it was enabled.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/20211008203112.1979843-7-farman@linux.ibm.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: s390: Simplify SIGP Set Arch handling
Eric Farman [Fri, 8 Oct 2021 20:31:07 +0000 (22:31 +0200)]
KVM: s390: Simplify SIGP Set Arch handling

The Principles of Operations describe the various reasons that
each individual SIGP orders might be rejected, and the status
bit that are set for each condition.

For example, for the Set Architecture order, it states:

  "If it is not true that all other CPUs in the configu-
   ration are in the stopped or check-stop state, ...
   bit 54 (incorrect state) ... is set to one."

However, it also states:

  "... if the CZAM facility is installed, ...
   bit 55 (invalid parameter) ... is set to one."

Since the Configuration-z/Architecture-Architectural Mode (CZAM)
facility is unconditionally presented, there is no need to examine
each VCPU to determine if it is started/stopped. It can simply be
rejected outright with the Invalid Parameter bit.

Fixes: b697e435aeee ("KVM: s390: Support Configuration z/Architecture Mode")
Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Link: https://lore.kernel.org/r/20211008203112.1979843-2-farman@linux.ibm.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: s390: pv: avoid stalls when making pages secure
Claudio Imbrenda [Mon, 20 Sep 2021 13:24:52 +0000 (15:24 +0200)]
KVM: s390: pv: avoid stalls when making pages secure

Improve make_secure_pte to avoid stalls when the system is heavily
overcommitted. This was especially problematic in kvm_s390_pv_unpack,
because of the loop over all pages that needed unpacking.

Due to the locks being held, it was not possible to simply replace
uv_call with uv_call_sched. A more complex approach was
needed, in which uv_call is replaced with __uv_call, which does not
loop. When the UVC needs to be executed again, -EAGAIN is returned, and
the caller (or its caller) will try again.

When -EAGAIN is returned, the path is the same as when the page is in
writeback (and the writeback check is also performed, which is
harmless).

Fixes: 214d9bbcd3a672 ("s390/mm: provide memory management functions for protected KVM guests")
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Link: https://lore.kernel.org/r/20210920132502.36111-5-imbrenda@linux.ibm.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: s390: pv: avoid stalls for kvm_s390_pv_init_vm
Claudio Imbrenda [Mon, 20 Sep 2021 13:24:51 +0000 (15:24 +0200)]
KVM: s390: pv: avoid stalls for kvm_s390_pv_init_vm

When the system is heavily overcommitted, kvm_s390_pv_init_vm might
generate stall notifications.

Fix this by using uv_call_sched instead of just uv_call. This is ok because
we are not holding spinlocks.

Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Fixes: 214d9bbcd3a672 ("s390/mm: provide memory management functions for protected KVM guests")
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
Message-Id: <20210920132502.36111-4-imbrenda@linux.ibm.com>
Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: s390: pv: avoid double free of sida page
Claudio Imbrenda [Mon, 20 Sep 2021 13:24:50 +0000 (15:24 +0200)]
KVM: s390: pv: avoid double free of sida page

If kvm_s390_pv_destroy_cpu is called more than once, we risk calling
free_page on a random page, since the sidad field is aliased with the
gbea, which is not guaranteed to be zero.

This can happen, for example, if userspace calls the KVM_PV_DISABLE
IOCTL, and it fails, and then userspace calls the same IOCTL again.
This scenario is only possible if KVM has some serious bug or if the
hardware is broken.

The solution is to simply return successfully immediately if the vCPU
was already non secure.

Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Fixes: 19e1227768863a1469797c13ef8fea1af7beac2c ("KVM: S390: protvirt: Introduce instruction data area bounce buffer")
Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Message-Id: <20210920132502.36111-3-imbrenda@linux.ibm.com>
Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: s390: pv: add macros for UVC CC values
Claudio Imbrenda [Mon, 20 Sep 2021 13:24:49 +0000 (15:24 +0200)]
KVM: s390: pv: add macros for UVC CC values

Add macros to describe the 4 possible CC values returned by the UVC
instruction.

Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Janosch Frank <frankja@linux.ibm.com>
Message-Id: <20210920132502.36111-2-imbrenda@linux.ibm.com>
Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agos390/mm: optimize reset_guest_reference_bit()
David Hildenbrand [Thu, 9 Sep 2021 16:22:48 +0000 (18:22 +0200)]
s390/mm: optimize reset_guest_reference_bit()

We already optimize get_guest_storage_key() to assume that if we don't have
a PTE table and don't have a huge page mapped that the storage key is 0.

Similarly, optimize reset_guest_reference_bit() to simply do nothing if
there is no PTE table and no huge page mapped.

Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20210909162248.14969-10-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agos390/mm: optimize set_guest_storage_key()
David Hildenbrand [Thu, 9 Sep 2021 16:22:47 +0000 (18:22 +0200)]
s390/mm: optimize set_guest_storage_key()

We already optimize get_guest_storage_key() to assume that if we don't have
a PTE table and don't have a huge page mapped that the storage key is 0.

Similarly, optimize set_guest_storage_key() to simply do nothing in case
the key to set is 0.

Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20210909162248.14969-9-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agos390/mm: no need for pte_alloc_map_lock() if we know the pmd is present
David Hildenbrand [Thu, 9 Sep 2021 16:22:46 +0000 (18:22 +0200)]
s390/mm: no need for pte_alloc_map_lock() if we know the pmd is present

pte_map_lock() is sufficient.

Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20210909162248.14969-8-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agos390/uv: fully validate the VMA before calling follow_page()
David Hildenbrand [Thu, 9 Sep 2021 16:22:44 +0000 (18:22 +0200)]
s390/uv: fully validate the VMA before calling follow_page()

We should not walk/touch page tables outside of VMA boundaries when
holding only the mmap sem in read mode. Evil user space can modify the
VMA layout just before this function runs and e.g., trigger races with
page table removal code since commit dd2283f2605e ("mm: mmap: zap pages
with read mmap_sem in munmap").

find_vma() does not check if the address is >= the VMA start address;
use vma_lookup() instead.

Fixes: 214d9bbcd3a6 ("s390/mm: provide memory management functions for protected KVM guests")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Link: https://lore.kernel.org/r/20210909162248.14969-6-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agos390/mm: fix VMA and page table handling code in storage key handling functions
David Hildenbrand [Thu, 9 Sep 2021 16:22:43 +0000 (18:22 +0200)]
s390/mm: fix VMA and page table handling code in storage key handling functions

There are multiple things broken about our storage key handling
functions:

1. We should not walk/touch page tables outside of VMA boundaries when
   holding only the mmap sem in read mode. Evil user space can modify the
   VMA layout just before this function runs and e.g., trigger races with
   page table removal code since commit dd2283f2605e ("mm: mmap: zap pages
   with read mmap_sem in munmap"). gfn_to_hva() will only translate using
   KVM memory regions, but won't validate the VMA.

2. We should not allocate page tables outside of VMA boundaries: if
   evil user space decides to map hugetlbfs to these ranges, bad things
   will happen because we suddenly have PTE or PMD page tables where we
   shouldn't have them.

3. We don't handle large PUDs that might suddenly appeared inside our page
   table hierarchy.

Don't manually allocate page tables, properly validate that we have VMA and
bail out on pud_large().

All callers of page table handling functions, except
get_guest_storage_key(), call fixup_user_fault() in case they
receive an -EFAULT and retry; this will allocate the necessary page tables
if required.

To keep get_guest_storage_key() working as expected and not requiring
kvm_s390_get_skeys() to call fixup_user_fault() distinguish between
"there is simply no page table or huge page yet and the key is assumed
to be 0" and "this is a fault to be reported".

Although commit 637ff9efe5ea ("s390/mm: Add huge pmd storage key handling")
introduced most of the affected code, it was actually already broken
before when using get_locked_pte() without any VMA checks.

Note: Ever since commit 637ff9efe5ea ("s390/mm: Add huge pmd storage key
handling") we can no longer set a guest storage key (for example from
QEMU during VM live migration) without actually resolving a fault.
Although we would have created most page tables, we would choke on the
!pmd_present(), requiring a call to fixup_user_fault(). I would
have thought that this is problematic in combination with postcopy life
migration ... but nobody noticed and this patch doesn't change the
situation. So maybe it's just fine.

Fixes: 9fcf93b5de06 ("KVM: S390: Create helper function get_guest_storage_key")
Fixes: 24d5dd0208ed ("s390/kvm: Provide function for setting the guest storage key")
Fixes: a7e19ab55ffd ("KVM: s390: handle missing storage-key facility")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20210909162248.14969-5-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agos390/mm: validate VMA in PGSTE manipulation functions
David Hildenbrand [Thu, 9 Sep 2021 16:22:42 +0000 (18:22 +0200)]
s390/mm: validate VMA in PGSTE manipulation functions

We should not walk/touch page tables outside of VMA boundaries when
holding only the mmap sem in read mode. Evil user space can modify the
VMA layout just before this function runs and e.g., trigger races with
page table removal code since commit dd2283f2605e ("mm: mmap: zap pages
with read mmap_sem in munmap"). gfn_to_hva() will only translate using
KVM memory regions, but won't validate the VMA.

Further, we should not allocate page tables outside of VMA boundaries: if
evil user space decides to map hugetlbfs to these ranges, bad things will
happen because we suddenly have PTE or PMD page tables where we
shouldn't have them.

Similarly, we have to check if we suddenly find a hugetlbfs VMA, before
calling get_locked_pte().

Fixes: 2d42f9477320 ("s390/kvm: Add PGSTE manipulation functions")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20210909162248.14969-4-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agos390/gmap: don't unconditionally call pte_unmap_unlock() in __gmap_zap()
David Hildenbrand [Thu, 9 Sep 2021 16:22:41 +0000 (18:22 +0200)]
s390/gmap: don't unconditionally call pte_unmap_unlock() in __gmap_zap()

... otherwise we will try unlocking a spinlock that was never locked via a
garbage pointer.

At the time we reach this code path, we usually successfully looked up
a PGSTE already; however, evil user space could have manipulated the VMA
layout in the meantime and triggered removal of the page table.

Fixes: 1e133ab296f3 ("s390/mm: split arch/s390/mm/pgtable.c")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20210909162248.14969-3-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agos390/gmap: validate VMA in __gmap_zap()
David Hildenbrand [Thu, 9 Sep 2021 16:22:40 +0000 (18:22 +0200)]
s390/gmap: validate VMA in __gmap_zap()

We should not walk/touch page tables outside of VMA boundaries when
holding only the mmap sem in read mode. Evil user space can modify the
VMA layout just before this function runs and e.g., trigger races with
page table removal code since commit dd2283f2605e ("mm: mmap: zap pages
with read mmap_sem in munmap"). The pure prescence in our guest_to_host
radix tree does not imply that there is a VMA.

Further, we should not allocate page tables (via get_locked_pte()) outside
of VMA boundaries: if evil user space decides to map hugetlbfs to these
ranges, bad things will happen because we suddenly have PTE or PMD page
tables where we shouldn't have them.

Similarly, we have to check if we suddenly find a hugetlbfs VMA, before
calling get_locked_pte().

Note that gmap_discard() is different:
zap_page_range()->unmap_single_vma() makes sure to stay within VMA
boundaries.

Fixes: b31288fa83b2 ("s390/kvm: support collaborative memory management")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20210909162248.14969-2-david@redhat.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: selftests: Fix nested SVM tests when built with clang
Jim Mattson [Thu, 30 Sep 2021 00:36:49 +0000 (17:36 -0700)]
KVM: selftests: Fix nested SVM tests when built with clang

Though gcc conveniently compiles a simple memset to "rep stos," clang
prefers to call the libc version of memset. If a test is dynamically
linked, the libc memset isn't available in L1 (nor is the PLT or the
GOT, for that matter). Even if the test is statically linked, the libc
memset may choose to use some CPU features, like AVX, which may not be
enabled in L1. Note that __builtin_memset doesn't solve the problem,
because (a) the compiler is free to call memset anyway, and (b)
__builtin_memset may also choose to use features like AVX, which may
not be available in L1.

To avoid a myriad of problems, use an explicit "rep stos" to clear the
VMCB in generic_svm_setup(), which is called both from L0 and L1.

Reported-by: Ricardo Koller <ricarkol@google.com>
Signed-off-by: Jim Mattson <jmattson@google.com>
Fixes: 20ba262f8631a ("selftests: KVM: AMD Nested test infrastructure")
Message-Id: <20210930003649.4026553-1-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: x86: Remove stale declaration of kvm_no_apic_vcpu
Jim Mattson [Thu, 21 Oct 2021 18:54:49 +0000 (11:54 -0700)]
kvm: x86: Remove stale declaration of kvm_no_apic_vcpu

This variable was renamed to kvm_has_noapic_vcpu in commit
6e4e3b4df4e3 ("KVM: Stop using deprecated jump label APIs").

Signed-off-by: Jim Mattson <jmattson@google.com>
Message-Id: <20211021185449.3471763-1-jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: Unregister posted interrupt wakeup handler on hardware unsetup
Sean Christopherson [Sat, 9 Oct 2021 00:11:05 +0000 (17:11 -0700)]
KVM: VMX: Unregister posted interrupt wakeup handler on hardware unsetup

Unregister KVM's posted interrupt wakeup handler during unsetup so that a
spurious interrupt that arrives after kvm_intel.ko is unloaded doesn't
call into freed memory.

Fixes: bf9f6ac8d749 ("KVM: Update Posted-Interrupts Descriptor when vCPU is blocked")
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211009001107.3936588-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agox86/irq: Ensure PI wakeup handler is unregistered before module unload
Sean Christopherson [Sat, 9 Oct 2021 00:11:04 +0000 (17:11 -0700)]
x86/irq: Ensure PI wakeup handler is unregistered before module unload

Add a synchronize_rcu() after clearing the posted interrupt wakeup handler
to ensure all readers, i.e. in-flight IRQ handlers, see the new handler
before returning to the caller.  If the caller is an exiting module and
is unregistering its handler, failure to wait could result in the IRQ
handler jumping into an unloaded module.

The registration path doesn't require synchronization, as it's the
caller's responsibility to not generate interrupts it cares about until
after its handler is registered.

Fixes: f6b3c72c2366 ("x86/irq: Define a global vector for VT-d Posted-Interrupts")
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211009001107.3936588-2-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Use rw_semaphore for APICv lock to allow vCPU parallelism
Sean Christopherson [Fri, 22 Oct 2021 00:49:27 +0000 (17:49 -0700)]
KVM: x86: Use rw_semaphore for APICv lock to allow vCPU parallelism

Use a rw_semaphore instead of a mutex to coordinate APICv updates so that
vCPUs responding to requests can take the lock for read and run in
parallel.  Using a mutex forces serialization of vCPUs even though
kvm_vcpu_update_apicv() only touches data local to that vCPU or is
protected by a different lock, e.g. SVM's ir_list_lock.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211022004927.1448382-5-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Move SVM's APICv sanity check to common x86
Sean Christopherson [Fri, 22 Oct 2021 00:49:25 +0000 (17:49 -0700)]
KVM: x86: Move SVM's APICv sanity check to common x86

Move SVM's assertion that vCPU's APICv state is consistent with its VM's
state out of svm_vcpu_run() and into x86's common inner run loop.  The
assertion and underlying logic is not unique to SVM, it's just that SVM
has more inhibiting conditions and thus is more likely to run headfirst
into any KVM bugs.

Add relevant comments to document exactly why the update path has unusual
ordering between the update the kick, why said ordering is safe, and also
the basic rules behind the assertion in the run loop.

Cc: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211022004927.1448382-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoriscv: do not select non-existing config ANON_INODES
Lukas Bulwahn [Fri, 22 Oct 2021 06:15:14 +0000 (08:15 +0200)]
riscv: do not select non-existing config ANON_INODES

Commit 99cdc6c18c2d ("RISC-V: Add initial skeletal KVM support") selects
the config ANON_INODES in config KVM, but the config ANON_INODES is removed
since commit 5dd50aaeb185 ("Make anon_inodes unconditional") in 2018.

Hence, ./scripts/checkkconfigsymbols.py warns on non-existing symbols:

  ANON_INODES
  Referencing files: arch/riscv/kvm/Kconfig

Remove selecting the non-existing config ANON_INODES.

Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Message-Id: <20211022061514.25946-1-lukas.bulwahn@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Extract zapping of rmaps for gfn range to separate helper
Sean Christopherson [Fri, 22 Oct 2021 01:00:05 +0000 (18:00 -0700)]
KVM: x86/mmu: Extract zapping of rmaps for gfn range to separate helper

Extract the zapping of rmaps, a.k.a. legacy MMU, for a gfn range to a
separate helper to clean up the unholy mess that kvm_zap_gfn_range() has
become.  In addition to deep nesting, the rmaps zapping spreads out the
declaration of several variables and is generally a mess.  Clean up the
mess now so that future work to improve the memslots implementation
doesn't need to deal with it.

Cc: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211022010005.1454978-4-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Drop a redundant remote TLB flush in kvm_zap_gfn_range()
Sean Christopherson [Fri, 22 Oct 2021 01:00:04 +0000 (18:00 -0700)]
KVM: x86/mmu: Drop a redundant remote TLB flush in kvm_zap_gfn_range()

Remove an unnecessary remote TLB flush in kvm_zap_gfn_range() now that
said function holds mmu_lock for write for its entire duration.  The
flush was added by the now-reverted commit to allow TDP MMU to flush while
holding mmu_lock for read, as the transition from write=>read required
dropping the lock and thus a pending flush needed to be serviced.

Fixes: 5a324c24b638 ("Revert "KVM: x86/mmu: Allow zap gfn range to operate under the mmu read lock"")
Cc: Maxim Levitsky <mlevitsk@redhat.com>
Cc: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Cc: Ben Gardon <bgardon@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211022010005.1454978-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Drop a redundant, broken remote TLB flush
Sean Christopherson [Fri, 22 Oct 2021 01:00:03 +0000 (18:00 -0700)]
KVM: x86/mmu: Drop a redundant, broken remote TLB flush

A recent commit to fix the calls to kvm_flush_remote_tlbs_with_address()
in kvm_zap_gfn_range() inadvertantly added yet another flush instead of
fixing the existing flush.  Drop the redundant flush, and fix the params
for the existing flush.

Cc: stable@vger.kernel.org
Fixes: 2822da446640 ("KVM: x86/mmu: fix parameters to kvm_flush_remote_tlbs_with_address")
Cc: Maxim Levitsky <mlevitsk@redhat.com>
Cc: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211022010005.1454978-2-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Don't unload MMU in kvm_vcpu_flush_tlb_guest()
Lai Jiangshan [Tue, 19 Oct 2021 11:01:54 +0000 (19:01 +0800)]
KVM: X86: Don't unload MMU in kvm_vcpu_flush_tlb_guest()

kvm_mmu_unload() destroys all the PGD caches.  Use the lighter
kvm_mmu_sync_roots() and kvm_mmu_sync_prev_roots() instead.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Message-Id: <20211019110154.4091-5-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: pair smp_wmb() of mmu_try_to_unsync_pages() with smp_rmb()
Lai Jiangshan [Tue, 19 Oct 2021 11:01:53 +0000 (19:01 +0800)]
KVM: X86: pair smp_wmb() of mmu_try_to_unsync_pages() with smp_rmb()

The commit 578e1c4db2213 ("kvm: x86: Avoid taking MMU lock
in kvm_mmu_sync_roots if no sync is needed") added smp_wmb() in
mmu_try_to_unsync_pages(), but the corresponding smp_load_acquire() isn't
used on the load of SPTE.W.  smp_load_acquire() orders _subsequent_
loads after sp->is_unsync; it does not order _earlier_ loads before
the load of sp->is_unsync.

This has no functional change; smp_rmb() is a NOP on x86, and no
compiler barrier is required because there is a VMEXIT between the
load of SPTE.W and kvm_mmu_snc_roots.

Cc: Junaid Shahid <junaids@google.com>
Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Message-Id: <20211019110154.4091-4-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Cache CR3 in prev_roots when PCID is disabled
Lai Jiangshan [Tue, 19 Oct 2021 11:01:52 +0000 (19:01 +0800)]
KVM: X86: Cache CR3 in prev_roots when PCID is disabled

The commit 21823fbda5522 ("KVM: x86: Invalidate all PGDs for the
current PCID on MOV CR3 w/ flush") invalidates all PGDs for the specific
PCID and in the case of PCID is disabled, it includes all PGDs in the
prev_roots and the commit made prev_roots totally unused in this case.

Not using prev_roots fixes a problem when CR4.PCIDE is changed 0 -> 1
before the said commit:

(CR4.PCIDE=0, CR4.PGE=1; CR3=cr3_a; the page for the guest
 RIP is global; cr3_b is cached in prev_roots)

modify page tables under cr3_b
the shadow root of cr3_b is unsync in kvm
INVPCID single context
the guest expects the TLB is clean for PCID=0
change CR4.PCIDE 0 -> 1
switch to cr3_b with PCID=0,NOFLUSH=1
No sync in kvm, cr3_b is still unsync in kvm
jump to the page that was modified in step 1
shadow page tables point to the wrong page

It is a very unlikely case, but it shows that stale prev_roots can be
a problem after CR4.PCIDE changes from 0 to 1.  However, to fix this
case, the commit disabled caching CR3 in prev_roots altogether when PCID
is disabled.  Not all CPUs have PCID; especially the PCID support
for AMD CPUs is kind of recent.  To restore the prev_roots optimization
for CR4.PCIDE=0, flush the whole MMU (including all prev_roots) when
CR4.PCIDE changes.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Message-Id: <20211019110154.4091-3-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Fix tlb flush for tdp in kvm_invalidate_pcid()
Lai Jiangshan [Tue, 19 Oct 2021 11:01:51 +0000 (19:01 +0800)]
KVM: X86: Fix tlb flush for tdp in kvm_invalidate_pcid()

The KVM doesn't know whether any TLB for a specific pcid is cached in
the CPU when tdp is enabled.  So it is better to flush all the guest
TLB when invalidating any single PCID context.

The case is very rare or even impossible since KVM generally doesn't
intercept CR3 write or INVPCID instructions when tdp is enabled, so the
fix is mostly for the sake of overall robustness.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Message-Id: <20211019110154.4091-2-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Don't reset mmu context when toggling X86_CR4_PGE
Lai Jiangshan [Sun, 19 Sep 2021 02:42:46 +0000 (10:42 +0800)]
KVM: X86: Don't reset mmu context when toggling X86_CR4_PGE

X86_CR4_PGE doesn't participate in kvm_mmu_role, so the mmu context
doesn't need to be reset.  It is only required to flush all the guest
tlb.

It is also inconsistent that X86_CR4_PGE is in KVM_MMU_CR4_ROLE_BITS
while kvm_mmu_role doesn't use X86_CR4_PGE.  So X86_CR4_PGE is also
removed from KVM_MMU_CR4_ROLE_BITS.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210919024246.89230-3-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: X86: Don't reset mmu context when X86_CR4_PCIDE 1->0
Lai Jiangshan [Sun, 19 Sep 2021 02:42:45 +0000 (10:42 +0800)]
KVM: X86: Don't reset mmu context when X86_CR4_PCIDE 1->0

X86_CR4_PCIDE doesn't participate in kvm_mmu_role, so the mmu context
doesn't need to be reset.  It is only required to flush all the guest
tlb.

Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Reviewed-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210919024246.89230-2-jiangshanlai@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: selftests: set CPUID before setting sregs in vcpu creation
Michael Roth [Wed, 6 Oct 2021 20:36:17 +0000 (15:36 -0500)]
KVM: selftests: set CPUID before setting sregs in vcpu creation

Recent kernels have checks to ensure the GPA values in special-purpose
registers like CR3 are within the maximum physical address range and
don't overlap with anything in the upper/reserved range. In the case of
SEV kselftest guests booting directly into 64-bit mode, CR3 needs to be
initialized to the GPA of the page table root, with the encryption bit
set. The kernel accounts for this encryption bit by removing it from
reserved bit range when the guest advertises the bit position via
KVM_SET_CPUID*, but kselftests currently call KVM_SET_SREGS as part of
vm_vcpu_add_default(), before KVM_SET_CPUID*.

As a result, KVM_SET_SREGS will return an error in these cases.
Address this by moving vcpu_set_cpuid() (which calls KVM_SET_CPUID*)
ahead of vcpu_setup() (which calls KVM_SET_SREGS).

While there, address a typo in the assertion that triggers when
KVM_SET_SREGS fails.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Michael Roth <michael.roth@amd.com>
Message-Id: <20211006203617.13045-1-michael.roth@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Nathan Tempelman <natet@google.com>
3 years agoKVM: emulate: Comment on difference between RDPMC implementation and manual
Wanpeng Li [Wed, 20 Oct 2021 10:13:56 +0000 (03:13 -0700)]
KVM: emulate: Comment on difference between RDPMC implementation and manual

SDM mentioned that, RDPMC:

  IF (((CR4.PCE = 1) or (CPL = 0) or (CR0.PE = 0)) and (ECX indicates a supported counter))
      THEN
          EAX := counter[31:0];
          EDX := ZeroExtend(counter[MSCB:32]);
      ELSE (* ECX is not valid or CR4.PCE is 0 and CPL is 1, 2, or 3 and CR0.PE is 1 *)
          #GP(0);
  FI;

Let's add a comment why CR0.PE isn't tested since it's impossible for CPL to be >0 if
CR0.PE=0.

Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1634724836-73721-1-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Add vendor name to kvm_x86_ops, use it for error messages
Sean Christopherson [Mon, 18 Oct 2021 18:39:28 +0000 (11:39 -0700)]
KVM: x86: Add vendor name to kvm_x86_ops, use it for error messages

Paul pointed out the error messages when KVM fails to load are unhelpful
in understanding exactly what went wrong if userspace probes the "wrong"
module.

Add a mandatory kvm_x86_ops field to track vendor module names, kvm_intel
and kvm_amd, and use the name for relevant error message when KVM fails
to load so that the user knows which module failed to load.

Opportunistically tweak the "disabled by bios" error message to clarify
that _support_ was disabled, not that the module itself was magically
disabled by BIOS.

Suggested-by: Paul Menzel <pmenzel@molgen.mpg.de>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20211018183929.897461-2-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: x86: mmu: Make NX huge page recovery period configurable
Junaid Shahid [Wed, 20 Oct 2021 01:06:27 +0000 (18:06 -0700)]
kvm: x86: mmu: Make NX huge page recovery period configurable

Currently, the NX huge page recovery thread wakes up every minute and
zaps 1/nx_huge_pages_recovery_ratio of the total number of split NX
huge pages at a time. This is intended to ensure that only a
relatively small number of pages get zapped at a time. But for very
large VMs (or more specifically, VMs with a large number of
executable pages), a period of 1 minute could still result in this
number being too high (unless the ratio is changed significantly,
but that can result in split pages lingering on for too long).

This change makes the period configurable instead of fixing it at
1 minute. Users of large VMs can then adjust the period and/or the
ratio to reduce the number of pages zapped at one time while still
maintaining the same overall duration for cycling through the
entire list. By default, KVM derives a period from the ratio such
that a page will remain on the list for 1 hour on average.

Signed-off-by: Junaid Shahid <junaids@google.com>
Message-Id: <20211020010627.305925-1-junaids@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: vPMU: Fill get_msr MSR_CORE_PERF_GLOBAL_OVF_CTRL w/ 0
Wanpeng Li [Tue, 19 Oct 2021 08:12:39 +0000 (01:12 -0700)]
KVM: vPMU: Fill get_msr MSR_CORE_PERF_GLOBAL_OVF_CTRL w/ 0

SDM section 18.2.3 mentioned that:

  "IA32_PERF_GLOBAL_OVF_CTL MSR allows software to clear overflow indicator(s) of
   any general-purpose or fixed-function counters via a single WRMSR."

It is R/W mentioned by SDM, we read this msr on bare-metal during perf testing,
the value is always 0 for ICX/SKX boxes on hands. Let's fill get_msr
MSR_CORE_PERF_GLOBAL_OVF_CTRL w/ 0 as hardware behavior and drop
global_ovf_ctrl variable.

Tested-by: Like Xu <likexu@tencent.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Message-Id: <1634631160-67276-2-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Rename slot_handle_leaf to slot_handle_level_4k
David Matlack [Tue, 19 Oct 2021 16:22:23 +0000 (16:22 +0000)]
KVM: x86/mmu: Rename slot_handle_leaf to slot_handle_level_4k

slot_handle_leaf is a misnomer because it only operates on 4K SPTEs
whereas "leaf" is used to describe any valid terminal SPTE (4K or
large page). Rename slot_handle_leaf to slot_handle_level_4k to
avoid confusion.

Making this change makes it more obvious there is a benign discrepency
between the legacy MMU and the TDP MMU when it comes to dirty logging.
The legacy MMU only iterates through 4K SPTEs when zapping for
collapsing and when clearing D-bits. The TDP MMU, on the other hand,
iterates through SPTEs on all levels.

The TDP MMU behavior of zapping SPTEs at all levels is technically
overkill for its current dirty logging implementation, which always
demotes to 4k SPTES, but both the TDP MMU and legacy MMU zap if and only
if the SPTE can be replaced by a larger page, i.e. will not spuriously
zap 2m (or larger) SPTEs. Opportunistically add comments to explain this
discrepency in the code.

Signed-off-by: David Matlack <dmatlack@google.com>
Message-Id: <20211019162223.3935109-1-dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: RTIT_CTL_BRANCH_EN has no dependency on other CPUID bit
Xiaoyao Li [Fri, 27 Aug 2021 07:02:46 +0000 (15:02 +0800)]
KVM: VMX: RTIT_CTL_BRANCH_EN has no dependency on other CPUID bit

Per Intel SDM, RTIT_CTL_BRANCH_EN bit has no dependency on any CPUID
leaf 0x14.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-Id: <20210827070249.924633-5-xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: Rename pt_desc.addr_range to pt_desc.num_address_ranges
Xiaoyao Li [Fri, 27 Aug 2021 07:02:45 +0000 (15:02 +0800)]
KVM: VMX: Rename pt_desc.addr_range to pt_desc.num_address_ranges

To better self explain the meaning of this field and match the
PT_CAP_num_address_ranges constatn.

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-Id: <20210827070249.924633-4-xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: Use precomputed vmx->pt_desc.addr_range
Xiaoyao Li [Fri, 27 Aug 2021 07:02:44 +0000 (15:02 +0800)]
KVM: VMX: Use precomputed vmx->pt_desc.addr_range

The number of valid PT ADDR MSRs for the guest is precomputed in
vmx->pt_desc.addr_range. Use it instead of calculating again.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-Id: <20210827070249.924633-3-xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: VMX: Restore host's MSR_IA32_RTIT_CTL when it's not zero
Xiaoyao Li [Fri, 27 Aug 2021 07:02:43 +0000 (15:02 +0800)]
KVM: VMX: Restore host's MSR_IA32_RTIT_CTL when it's not zero

A minor optimization to WRMSR MSR_IA32_RTIT_CTL when necessary.

Opportunistically refine the comment to call out that KVM requires
VM_EXIT_CLEAR_IA32_RTIT_CTL to expose PT to the guest.

Reviewed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-Id: <20210827070249.924633-2-xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: clean up prefetch/prefault/speculative naming
Paolo Bonzini [Wed, 29 Sep 2021 13:19:32 +0000 (09:19 -0400)]
KVM: x86/mmu: clean up prefetch/prefault/speculative naming

"prefetch", "prefault" and "speculative" are used throughout KVM to mean
the same thing.  Use a single name, standardizing on "prefetch" which
is already used by various functions such as direct_pte_prefetch,
FNAME(prefetch_gpte), FNAME(pte_prefetch), etc.

Suggested-by: David Matlack <dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: cleanup allocation of rmaps and page tracking data
David Stevens [Fri, 15 Oct 2021 16:30:21 +0000 (12:30 -0400)]
KVM: cleanup allocation of rmaps and page tracking data

Unify the flags for rmaps and page tracking data, using a
single flag in struct kvm_arch and a single loop to go
over all the address spaces and memslots.  This avoids
code duplication between alloc_all_memslots_rmaps and
kvm_page_track_enable_mmu_write_tracking.

Signed-off-by: David Stevens <stevensd@chromium.org>
[This patch is the delta between David's v2 and v3, with conflicts
 fixed and my own commit message. - Paolo]
Co-developed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoMerge branch kvm/selftests/memslot into kvmarm-master/next
Marc Zyngier [Thu, 21 Oct 2021 10:40:03 +0000 (11:40 +0100)]
Merge branch kvm/selftests/memslot into kvmarm-master/next

* kvm/selftests/memslot:
  : .
  : Enable KVM memslot selftests on arm64, making them less
  : x86 specific.
  : .
  KVM: selftests: Build the memslot tests for arm64
  KVM: selftests: Make memslot_perf_test arch independent

Signed-off-by: Marc Zyngier <maz@kernel.org>
3 years agoKVM: selftests: Build the memslot tests for arm64
Ricardo Koller [Tue, 7 Sep 2021 18:09:57 +0000 (11:09 -0700)]
KVM: selftests: Build the memslot tests for arm64

Add memslot_perf_test and memslot_modification_stress_test to the list
of aarch64 selftests.

Signed-off-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20210907180957.609966-3-ricarkol@google.com
3 years agoKVM: selftests: Make memslot_perf_test arch independent
Ricardo Koller [Tue, 7 Sep 2021 18:09:56 +0000 (11:09 -0700)]
KVM: selftests: Make memslot_perf_test arch independent

memslot_perf_test uses ucalls for synchronization between guest and
host. Ucalls API is architecture independent: tests do not need to know
details like what kind of exit they generate on a specific arch.  More
specifically, there is no need to check whether an exit is KVM_EXIT_IO
in x86 for the host to know that the exit is ucall related, as
get_ucall() already makes that check.

Change memslot_perf_test to not require specifying what exit does a
ucall generate. Also add a missing ucall_init.

Signed-off-by: Ricardo Koller <ricarkol@google.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20210907180957.609966-2-ricarkol@google.com
3 years agoKVM: s390: preserve deliverable_mask in __airqs_kick_single_vcpu
Halil Pasic [Tue, 19 Oct 2021 17:54:00 +0000 (19:54 +0200)]
KVM: s390: preserve deliverable_mask in __airqs_kick_single_vcpu

Changing the deliverable mask in __airqs_kick_single_vcpu() is a bug. If
one idle vcpu can't take the interrupts we want to deliver, we should
look for another vcpu that can, instead of saying that we don't want
to deliver these interrupts by clearing the bits from the
deliverable_mask.

Fixes: 9f30f6216378 ("KVM: s390: add gib_alert_irq_handler()")
Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Michael Mueller <mimu@linux.ibm.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Link: https://lore.kernel.org/r/20211019175401.3757927-3-pasic@linux.ibm.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoKVM: s390: clear kicked_mask before sleeping again
Halil Pasic [Tue, 19 Oct 2021 17:53:59 +0000 (19:53 +0200)]
KVM: s390: clear kicked_mask before sleeping again

The idea behind kicked mask is that we should not re-kick a vcpu that
is already in the "kick" process, i.e. that was kicked and is
is about to be dispatched if certain conditions are met.

The problem with the current implementation is, that it assumes the
kicked vcpu is going to enter SIE shortly. But under certain
circumstances, the vcpu we just kicked will be deemed non-runnable and
will remain in wait state. This can happen, if the interrupt(s) this
vcpu got kicked to deal with got already cleared (because the interrupts
got delivered to another vcpu). In this case kvm_arch_vcpu_runnable()
would return false, and the vcpu would remain in kvm_vcpu_block(),
but this time with its kicked_mask bit set. So next time around we
wouldn't kick the vcpu form __airqs_kick_single_vcpu(), but would assume
that we just kicked it.

Let us make sure the kicked_mask is cleared before we give up on
re-dispatching the vcpu.

Fixes: 9f30f6216378 ("KVM: s390: add gib_alert_irq_handler()")
Reported-by: Matthew Rosato <mjrosato@linux.ibm.com>
Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Michael Mueller <mimu@linux.ibm.com>
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Link: https://lore.kernel.org/r/20211019175401.3757927-2-pasic@linux.ibm.com
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
3 years agoselftests: KVM: Introduce system counter offset test
Oliver Upton [Thu, 16 Sep 2021 18:15:51 +0000 (18:15 +0000)]
selftests: KVM: Introduce system counter offset test

Introduce a KVM selftest to verify that userspace manipulation of the
TSC (via the new vCPU attribute) results in the correct behavior within
the guest.

Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Message-Id: <20210916181555.973085-6-oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoselftests: KVM: Add helpers for vCPU device attributes
Oliver Upton [Thu, 16 Sep 2021 18:15:50 +0000 (18:15 +0000)]
selftests: KVM: Add helpers for vCPU device attributes

vCPU file descriptors are abstracted away from test code in KVM
selftests, meaning that tests cannot directly access a vCPU's device
attributes. Add helpers that tests can use to get at vCPU device
attributes.

Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Message-Id: <20210916181555.973085-5-oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoselftests: KVM: Fix kvm device helper ioctl assertions
Oliver Upton [Thu, 16 Sep 2021 18:15:49 +0000 (18:15 +0000)]
selftests: KVM: Fix kvm device helper ioctl assertions

The KVM_CREATE_DEVICE and KVM_{GET,SET}_DEVICE_ATTR ioctls are defined
to return a value of zero on success. As such, tighten the assertions in
the helper functions to only pass if the return code is zero.

Suggested-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Message-Id: <20210916181555.973085-4-oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoselftests: KVM: Add test for KVM_{GET,SET}_CLOCK
Oliver Upton [Thu, 16 Sep 2021 18:15:48 +0000 (18:15 +0000)]
selftests: KVM: Add test for KVM_{GET,SET}_CLOCK

Add a selftest for the new KVM clock UAPI that was introduced. Ensure
that the KVM clock is consistent between userspace and the guest, and
that the difference in realtime will only ever cause the KVM clock to
advance forward.

Cc: Andrew Jones <drjones@redhat.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Message-Id: <20210916181555.973085-3-oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agotools: arch: x86: pull in pvclock headers
Oliver Upton [Thu, 16 Sep 2021 18:15:47 +0000 (18:15 +0000)]
tools: arch: x86: pull in pvclock headers

Copy over approximately clean versions of the pvclock headers into
tools. Reconcile headers/symbols missing in tools that are unneeded.

Signed-off-by: Oliver Upton <oupton@google.com>
Message-Id: <20210916181555.973085-2-oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Expose TSC offset controls to userspace
Oliver Upton [Thu, 16 Sep 2021 18:15:38 +0000 (18:15 +0000)]
KVM: x86: Expose TSC offset controls to userspace

To date, VMM-directed TSC synchronization and migration has been a bit
messy. KVM has some baked-in heuristics around TSC writes to infer if
the VMM is attempting to synchronize. This is problematic, as it depends
on host userspace writing to the guest's TSC within 1 second of the last
write.

A much cleaner approach to configuring the guest's views of the TSC is to
simply migrate the TSC offset for every vCPU. Offsets are idempotent,
and thus not subject to change depending on when the VMM actually
reads/writes values from/to KVM. The VMM can then read the TSC once with
KVM_GET_CLOCK to capture a (realtime, host_tsc) pair at the instant when
the guest is paused.

Cc: David Matlack <dmatlack@google.com>
Cc: Sean Christopherson <seanjc@google.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210916181538.968978-8-oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Refactor tsc synchronization code
Oliver Upton [Thu, 16 Sep 2021 18:15:37 +0000 (18:15 +0000)]
KVM: x86: Refactor tsc synchronization code

Refactor kvm_synchronize_tsc to make a new function that allows callers
to specify TSC parameters (offset, value, nanoseconds, etc.) explicitly
for the sake of participating in TSC synchronization.

Signed-off-by: Oliver Upton <oupton@google.com>
Message-Id: <20210916181538.968978-7-oupton@google.com>
[Make sure kvm->arch.cur_tsc_generation and vcpu->arch.this_tsc_generation are
 equal at the end of __kvm_synchronize_tsc, if matched is false. Reported by
 Maxim Levitsky. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agokvm: x86: protect masterclock with a seqcount
Paolo Bonzini [Thu, 16 Sep 2021 18:15:36 +0000 (18:15 +0000)]
kvm: x86: protect masterclock with a seqcount

Protect the reference point for kvmclock with a seqcount, so that
kvmclock updates for all vCPUs can proceed in parallel.  Xen runstate
updates will also run in parallel and not bounce the kvmclock cacheline.

Of the variables that were protected by pvclock_gtod_sync_lock,
nr_vcpus_matched_tsc is different because it is updated outside
pvclock_update_vm_gtod_copy and read inside it.  Therefore, we
need to keep it protected by a spinlock.  In fact it must now
be a raw spinlock, because pvclock_update_vm_gtod_copy, being the
write-side of a seqcount, is non-preemptible.  Since we already
have tsc_write_lock which is a raw spinlock, we can just use
tsc_write_lock as the lock that protects the write-side of the
seqcount.

Co-developed-by: Oliver Upton <oupton@google.com>
Message-Id: <20210916181538.968978-6-oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86: Report host tsc and realtime values in KVM_GET_CLOCK
Oliver Upton [Thu, 16 Sep 2021 18:15:35 +0000 (18:15 +0000)]
KVM: x86: Report host tsc and realtime values in KVM_GET_CLOCK

Handling the migration of TSCs correctly is difficult, in part because
Linux does not provide userspace with the ability to retrieve a (TSC,
realtime) clock pair for a single instant in time. In lieu of a more
convenient facility, KVM can report similar information in the kvm_clock
structure.

Provide userspace with a host TSC & realtime pair iff the realtime clock
is based on the TSC. If userspace provides KVM_SET_CLOCK with a valid
realtime value, advance the KVM clock by the amount of elapsed time. Do
not step the KVM clock backwards, though, as it is a monotonic
oscillator.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210916181538.968978-5-oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
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 agoMerge branch kvm-arm64/pkvm/fixed-features into kvmarm-master/next
Marc Zyngier [Mon, 18 Oct 2021 16:20:50 +0000 (17:20 +0100)]
Merge branch kvm-arm64/pkvm/fixed-features into kvmarm-master/next

* kvm-arm64/pkvm/fixed-features: (22 commits)
  : .
  : Add the pKVM fixed feature that allows a bunch of exceptions
  : to either be forbidden or be easily handled at EL2.
  : .
  KVM: arm64: pkvm: Give priority to standard traps over pvm handling
  KVM: arm64: pkvm: Pass vpcu instead of kvm to kvm_get_exit_handler_array()
  KVM: arm64: pkvm: Move kvm_handle_pvm_restricted around
  KVM: arm64: pkvm: Consolidate include files
  KVM: arm64: pkvm: Preserve pending SError on exit from AArch32
  KVM: arm64: pkvm: Handle GICv3 traps as required
  KVM: arm64: pkvm: Drop sysregs that should never be routed to the host
  KVM: arm64: pkvm: Drop AArch32-specific registers
  KVM: arm64: pkvm: Make the ERR/ERX*_EL1 registers RAZ/WI
  KVM: arm64: pkvm: Use a single function to expose all id-regs
  KVM: arm64: Fix early exit ptrauth handling
  KVM: arm64: Handle protected guests at 32 bits
  KVM: arm64: Trap access to pVM restricted features
  KVM: arm64: Move sanitized copies of CPU features
  KVM: arm64: Initialize trap registers for protected VMs
  KVM: arm64: Add handlers for protected VM System Registers
  KVM: arm64: Simplify masking out MTE in feature id reg
  KVM: arm64: Add missing field descriptor for MDCR_EL2
  KVM: arm64: Pass struct kvm to per-EC handlers
  KVM: arm64: Move early handlers to per-EC handlers
  ...

Signed-off-by: Marc Zyngier <maz@kernel.org>
3 years agoKVM: arm64: pkvm: Give priority to standard traps over pvm handling
Marc Zyngier [Wed, 13 Oct 2021 12:03:46 +0000 (13:03 +0100)]
KVM: arm64: pkvm: Give priority to standard traps over pvm handling

Checking for pvm handling first means that we cannot handle ptrauth
traps or apply any of the workarounds (GICv3 or TX2 #219).

Flip the order around.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20211013120346.2926621-12-maz@kernel.org
3 years agoKVM: arm64: pkvm: Pass vpcu instead of kvm to kvm_get_exit_handler_array()
Marc Zyngier [Wed, 13 Oct 2021 12:03:45 +0000 (13:03 +0100)]
KVM: arm64: pkvm: Pass vpcu instead of kvm to kvm_get_exit_handler_array()

Passing a VM pointer around is odd, and results in extra work on
VHE. Follow the rest of the design that uses the vcpu instead, and
let the nVHE code look into the struct kvm as required.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20211013120346.2926621-11-maz@kernel.org
3 years agoKVM: arm64: pkvm: Move kvm_handle_pvm_restricted around
Marc Zyngier [Wed, 13 Oct 2021 12:03:44 +0000 (13:03 +0100)]
KVM: arm64: pkvm: Move kvm_handle_pvm_restricted around

Place kvm_handle_pvm_restricted() next to its little friends such
as kvm_handle_pvm_sysreg().

This allows to make inject_undef64() static.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20211013120346.2926621-10-maz@kernel.org
3 years agoKVM: arm64: pkvm: Consolidate include files
Marc Zyngier [Wed, 13 Oct 2021 12:03:43 +0000 (13:03 +0100)]
KVM: arm64: pkvm: Consolidate include files

kvm_fixed_config.h is pkvm specific, and would be better placed
near its users. At the same time, include/nvhe/sys_regs.h is now
almost empty.

Merge the two into arch/arm64/kvm/hyp/include/nvhe/fixed_config.h.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20211013120346.2926621-9-maz@kernel.org
3 years agoKVM: arm64: pkvm: Preserve pending SError on exit from AArch32
Marc Zyngier [Wed, 13 Oct 2021 12:03:42 +0000 (13:03 +0100)]
KVM: arm64: pkvm: Preserve pending SError on exit from AArch32

Don't drop a potential SError when a guest gets caught red-handed
running AArch32 code.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20211013120346.2926621-8-maz@kernel.org
3 years agoKVM: arm64: pkvm: Handle GICv3 traps as required
Marc Zyngier [Wed, 13 Oct 2021 12:03:41 +0000 (13:03 +0100)]
KVM: arm64: pkvm: Handle GICv3 traps as required

Forward accesses to the ICV_*SGI*_EL1 registers to EL1, and
emulate ICV_SRE_EL1 by returning a fixed value.

This should be enough to support GICv3 in a protected guest.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20211013120346.2926621-7-maz@kernel.org
3 years agoKVM: arm64: pkvm: Drop sysregs that should never be routed to the host
Marc Zyngier [Wed, 13 Oct 2021 12:03:40 +0000 (13:03 +0100)]
KVM: arm64: pkvm: Drop sysregs that should never be routed to the host

A bunch of system registers (most of them MM related) should never
trap to the host under any circumstance. Keep them close to our chest.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com>
Link: https://lore.kernel.org/r/20211013120346.2926621-6-maz@kernel.org