Play nice with a NULL shadow page when checking for an obsolete root in
the page fault handler by flagging the page fault as stale if there's no
shadow page associated with the root and KVM_REQ_MMU_RELOAD is pending.
Invalidating memslots, which is the only case where _all_ roots need to
be reloaded, requests all vCPUs to reload their MMUs while holding
mmu_lock for lock.
The "special" roots, e.g. pae_root when KVM uses PAE paging, are not
backed by a shadow page. Running with TDP disabled or with nested NPT
explodes spectaculary due to dereferencing a NULL shadow page pointer.
Skip the KVM_REQ_MMU_RELOAD check if there is a valid shadow page for the
root. Zapping shadow pages in response to guest activity, e.g. when the
guest frees a PGD, can trigger KVM_REQ_MMU_RELOAD even if the current
vCPU isn't using the affected root. I.e. KVM_REQ_MMU_RELOAD can be seen
with a completely valid root shadow page. This is a bit of a moot point
as KVM currently unloads all roots on KVM_REQ_MMU_RELOAD, but that will
be cleaned up in the future.
Fixes: a955cad84c ("KVM: x86/mmu: Retry page fault if root is invalidated by memslot update")
Cc: [email protected]
Cc: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Add an x86 selftest to verify that KVM doesn't WARN or otherwise explode
if userspace modifies RCX during a userspace exit to handle string I/O.
This is a regression test for a user-triggerable WARN introduced by
commit 3b27de2718 ("KVM: x86: split the two parts of emulator_pio_in").
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Replace a WARN with a comment to call out that userspace can modify RCX
during an exit to userspace to handle string I/O. KVM doesn't actually
support changing the rep count during an exit, i.e. the scenario can be
ignored, but the WARN needs to go as it's trivial to trigger from
userspace.
Cc: [email protected]
Fixes: 3b27de2718 ("KVM: x86: split the two parts of emulator_pio_in")
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Do not bail early if there are no bits set in the sparse banks for a
non-sparse, a.k.a. "all CPUs", IPI request. Per the Hyper-V spec, it is
legal to have a variable length of '0', e.g. VP_SET's BankContents in
this case, if the request can be serviced without the extra info.
It is possible that for a given invocation of a hypercall that does
accept variable sized input headers that all the header input fits
entirely within the fixed size header. In such cases the variable sized
input header is zero-sized and the corresponding bits in the hypercall
input should be set to zero.
Bailing early results in KVM failing to send IPIs to all CPUs as expected
by the guest.
Fixes: 214ff83d44 ("KVM: x86: hyperv: implement PV IPI send hypercalls")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Use kvzalloc() to allocate KVM's buffer for SEV-ES's GHCB scratch area so
that KVM falls back to __vmalloc() if physically contiguous memory isn't
available. The buffer is purely a KVM software construct, i.e. there's
no need for it to be physically contiguous.
Cc: Tom Lendacky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Return appropriate error codes if setting up the GHCB scratch area for an
SEV-ES guest fails. In particular, returning -EINVAL instead of -ENOMEM
when allocating the kernel buffer could be confusing as userspace would
likely suspect a guest issue.
Fixes: 8f423a80d2 ("KVM: SVM: Support MMIO for an SEV-ES guest")
Cc: Tom Lendacky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Bail from the page fault handler if the root shadow page was obsoleted by
a memslot update. Do the check _after_ acuiring mmu_lock, as the TDP MMU
doesn't rely on the memslot/MMU generation, and instead relies on the
root being explicit marked invalid by kvm_mmu_zap_all_fast(), which takes
mmu_lock for write.
For the TDP MMU, inserting a SPTE into an obsolete root can leak a SP if
kvm_tdp_mmu_zap_invalidated_roots() has already zapped the SP, i.e. has
moved past the gfn associated with the SP.
For other MMUs, the resulting behavior is far more convoluted, though
unlikely to be truly problematic. Installing SPs/SPTEs into the obsolete
root isn't directly problematic, as the obsolete root will be unloaded
and dropped before the vCPU re-enters the guest. But because the legacy
MMU tracks shadow pages by their role, any SP created by the fault can
can be reused in the new post-reload root. Again, that _shouldn't_ be
problematic as any leaf child SPTEs will be created for the current/valid
memslot generation, and kvm_mmu_get_page() will not reuse child SPs from
the old generation as they will be flagged as obsolete. But, given that
continuing with the fault is pointess (the root will be unloaded), apply
the check to all MMUs.
Fixes: b7cccd397f ("KVM: x86/mmu: Fast invalidation for TDP MMU")
Cc: [email protected]
Cc: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Account for the '0' being a default, "let KVM choose" period, when
determining whether or not the recovery worker needs to be awakened in
response to userspace reducing the period. Failure to do so results in
the worker not being awakened properly, e.g. when changing the period
from '0' to any small-ish value.
Fixes: 4dfe4f40d8 ("kvm: x86: mmu: Make NX huge page recovery period configurable")
Cc: [email protected]
Cc: Junaid Shahid <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Drop the "flush" param and return values to/from the TDP MMU's helper for
zapping collapsible SPTEs. Because the helper runs with mmu_lock held
for read, not write, it uses tdp_mmu_zap_spte_atomic(), and the atomic
zap handles the necessary remote TLB flush.
Similarly, because mmu_lock is dropped and re-acquired between zapping
legacy MMUs and zapping TDP MMUs, kvm_mmu_zap_collapsible_sptes() must
handle remote TLB flushes from the legacy MMU before calling into the TDP
MMU.
Fixes: e2209710cc ("KVM: x86/mmu: Skip rmap operations if rmaps not allocated")
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Use the yield-safe variant of the TDP MMU iterator when handling an
unmapping event from the MMU notifier, as most occurences of the event
allow yielding.
Fixes: e1eed5847b ("KVM: x86/mmu: Allow yielding during MMU notifier unmap/zap, if possible")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Fully emulate a guest TLB flush on nested VM-Enter which changes vpid12,
i.e. L2's VPID, instead of simply doing INVVPID to flush real hardware's
TLB entries for vpid02. From L1's perspective, changing L2's VPID is
effectively a TLB flush unless "hardware" has previously cached entries
for the new vpid12. Because KVM tracks only a single vpid12, KVM doesn't
know if the new vpid12 has been used in the past and so must treat it as
a brand new, never been used VPID, i.e. must assume that the new vpid12
represents a TLB flush from L1's perspective.
For example, if L1 and L2 share a CR3, the first VM-Enter to L2 (with a
VPID) is effectively a TLB flush as hardware/KVM has never seen vpid12
and thus can't have cached entries in the TLB for vpid12.
Reported-by: Lai Jiangshan <[email protected]>
Fixes: 5c614b3583 ("KVM: nVMX: nested VPID emulation")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Like KVM_REQ_TLB_FLUSH_CURRENT, the GUEST variant needs to be serviced at
nested transitions, as KVM doesn't track requests for L1 vs L2. E.g. if
there's a pending flush when a nested VM-Exit occurs, then the flush was
requested in the context of L2 and needs to be handled before switching
to L1, otherwise the flush for L2 would effectiely be lost.
Opportunistically add a helper to handle CURRENT and GUEST as a pair, the
logic for when they need to be serviced is identical as both requests are
tied to L1 vs. L2, the only difference is the scope of the flush.
Reported-by: Lai Jiangshan <[email protected]>
Fixes: 07ffaf343e ("KVM: nVMX: Sync all PGDs on nested transition with shadow paging")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Flush the current VPID when handling KVM_REQ_TLB_FLUSH_GUEST instead of
always flushing vpid01. Any TLB flush that is triggered when L2 is
active is scoped to L2's VPID (if it has one), e.g. if L2 toggles CR4.PGE
and L1 doesn't intercept PGE writes, then KVM's emulation of the TLB
flush needs to be applied to L2's VPID.
Reported-by: Lai Jiangshan <[email protected]>
Fixes: 07ffaf343e ("KVM: nVMX: Sync all PGDs on nested transition with shadow paging")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Unmap stage2 page tables when a memslot is being deleted or moved. It's
the architectures' responsibility to ensure existing mappings are removed
when kvm_arch_flush_shadow_memslot() returns.
Fixes: 9d05c1fee8 ("RISC-V: KVM: Implement stage2 page table programming")
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Anup Patel <[email protected]>
Reject userspace memslots whose size exceeds the storage capacity of an
"unsigned long". KVM's uAPI takes the size as u64 to support large slots
on 64-bit hosts, but does not account for the size being truncated on
32-bit hosts in various flows. The access_ok() check on the userspace
virtual address in particular casts the size to "unsigned long" and will
check the wrong number of bytes.
KVM doesn't actually support slots whose size doesn't fit in an "unsigned
long", e.g. KVM's internal kvm_memory_slot.npages is an "unsigned long",
not a "u64", and misc arch specific code follows that behavior.
Fixes: fa3d315a4c ("KVM: Validate userspace_addr of memslot when registered")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Maciej S. Szmigiero <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
When modifying memslots, snapshot the "old" memslot and copy it to the
"new" memslot's arch data after (re)acquiring slots_arch_lock. x86 can
change a memslot's arch data while memslot updates are in-progress so
long as it holds slots_arch_lock, thus snapshotting a memslot without
holding the lock can result in the consumption of stale data.
Fixes: b10a038e84 ("KVM: mmu: Add slots_arch_lock for memslot arch fields")
Cc: [email protected]
Cc: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Rename cmd_allowed_from_miror() to is_cmd_allowed_from_mirror(), fixing
a typo and making it obvious that the result is a boolean where
false means "not allowed".
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Remove a fully redundant write to sev->asid during SEV/SEV-ES guest
initialization. The ASID is set a few lines earlier prior to the call to
sev_platform_init(), which doesn't take "sev" as a param, i.e. can't
muck with the ASID barring some truly magical behind-the-scenes code.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
WARN if the VM is tagged as SEV-ES but not SEV. KVM relies on SEV and
SEV-ES being set atomically, and guards common flows with "is SEV", i.e.
observing SEV-ES without SEV means KVM has a fatal bug.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Set sev_info.active during SEV/SEV-ES activation before calling any code
that can potentially consume sev_info.es_active, e.g. set "active" and
"es_active" as a pair immediately after the initial sanity checks. KVM
generally expects that es_active can be true if and only if active is
true, e.g. sev_asid_new() deliberately avoids sev_es_guest() so that it
doesn't get a false negative. This will allow WARNing in sev_es_guest()
if the VM is tagged as SEV-ES but not SEV.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Copy perf_test_args to the guest during VM creation instead of relying on
the caller to do so at their leisure. Ideally, tests wouldn't even be
able to modify perf_test_args, i.e. they would have no motivation to do
the sync, but enforcing that is arguably a net negative for readability.
No functional change intended.
[Set wr_fract=1 by default and add helper to override it since the new
access_tracking_perf_test needs to set it dynamically.]
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Fill the per-vCPU args when creating the perf_test VM instead of having
the caller do so. This helps ensure that any adjustments to the number
of pages (and thus vcpu_memory_bytes) are reflected in the per-VM args.
Automatically filling the per-vCPU args will also allow a future patch
to do the sync to the guest during creation.
Signed-off-by: Sean Christopherson <[email protected]>
[Updated access_tracking_perf_test as well.]
Signed-off-by: David Matlack <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Use the already computed guest_num_pages when creating the so called
extra VM pages for a perf test, and add a comment explaining why the
pages are allocated as extra pages.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Remove perf_test_args.host_page_size and instead use getpagesize() so
that it's somewhat obvious that, for tests that care about the host page
size, they care about the system page size, not the hardware page size,
e.g. that the logic is unchanged if hugepages are in play.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Move the per-VM GPA into perf_test_args instead of storing it as a
separate global variable. It's not obvious that guest_test_phys_mem
holds a GPA, nor that it's connected/coupled with per_vcpu->gpa.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Grab the per-vCPU GPA and number of pages from perf_util in the demand
paging test instead of duplicating perf_util's calculations.
Note, this may or may not result in a functional change. It's not clear
that the test's calculations are guaranteed to yield the same value as
perf_util, e.g. if guest_percpu_mem_size != vcpu_args->pages.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Capture the per-vCPU GPA in perf_test_vcpu_args so that tests can get
the GPA without having to calculate the GPA on their own.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Use 'pta' as a local pointer to the global perf_tests_args in order to
shorten line lengths and make the code borderline readable.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Assert that the GPA for a memslot backed by a hugepage is aligned to
the hugepage size and fix perf_test_util accordingly. Lack of GPA
alignment prevents KVM from backing the guest with hugepages, e.g. x86's
write-protection of hugepages when dirty logging is activated is
otherwise not exercised.
Add a comment explaining that guest_page_size is for non-huge pages to
try and avoid confusion about what it actually tracks.
Cc: Ben Gardon <[email protected]>
Cc: Yanan Wang <[email protected]>
Cc: Andrew Jones <[email protected]>
Cc: Peter Xu <[email protected]>
Cc: Aaron Lewis <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
[Used get_backing_src_pagesz() to determine alignment dynamically.]
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Manually padding and aligning the mmap region is only needed when using
THP. When using HugeTLB, mmap will always return an address aligned to
the HugeTLB page size. Add a comment to clarify this and assert the mmap
behavior for HugeTLB.
[Removed requirement that HugeTLB mmaps must be padded per Yanan's
feedback and added assertion that mmap returns aligned addresses
when using HugeTLB.]
Cc: Ben Gardon <[email protected]>
Cc: Yanan Wang <[email protected]>
Cc: Andrew Jones <[email protected]>
Cc: Peter Xu <[email protected]>
Cc: Aaron Lewis <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Refactor align() to work with non-pointers and split into separate
helpers for aligning up vs. down. Add align_ptr_up() for use with
pointers. Expose all helpers so that they can be used by tests and/or
other utilities. The align_down() helper in particular will be used to
ensure gpa alignment for hugepages.
No functional change intended.
[Added sepearate up/down helpers and replaced open-coded alignment
bit math throughout the KVM selftests.]
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: David Matlack <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Explicitly state the indices when populating vm_guest_mode_params to
make it marginally easier to visualize what's going on.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
[Added indices for new guest modes.]
Signed-off-by: David Matlack <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Explicitly check for MSR_HYPERCALL and MSR_VP_INDEX support when probing
for running as a Hyper-V guest instead of waiting until hyperv_init() to
detect the bogus configuration. Add messages to give the admin a heads
up that they are likely running on a broken virtual machine setup.
At best, silently disabling Hyper-V is confusing and difficult to debug,
e.g. the kernel _says_ it's using all these fancy Hyper-V features, but
always falls back to the native versions. At worst, the half baked setup
will crash/hang the kernel.
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Wei Liu <[email protected]>
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 <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: 1a155254ff ("KVM: x86: Introduce MSR filtering")
Fixes: d69129b4e4 ("KVM: nVMX: Disable intercept for FS/GS base MSRs in vmcs02 when possible")
Cc: [email protected]
Cc: Alexander Graf <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: d28b387fb7 ("KVM/VMX: Allow direct access to MSR_IA32_SPEC_CTRL")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: 7158bee4b4 ("KVM: MMU: pass kvm_mmu_page struct to make_spte")
Cc: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: bf9f6ac8d7 ("KVM: Update Posted-Interrupts Descriptor when vCPU is blocked")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: f6b3c72c23 ("x86/irq: Define a global vector for VT-d Posted-Interrupts")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: 5a324c24b6 ("Revert "KVM: x86/mmu: Allow zap gfn range to operate under the mmu read lock"")
Cc: Maxim Levitsky <[email protected]>
Cc: Maciej S. Szmigiero <[email protected]>
Cc: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: [email protected]
Fixes: 2822da4466 ("KVM: x86/mmu: fix parameters to kvm_flush_remote_tlbs_with_address")
Cc: Maxim Levitsky <[email protected]>
Cc: Maciej S. Szmigiero <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
WARN if the static keys used to track if any vCPU has disabled its APIC
are left elevated at module exit. Unlike the underflow case, nothing in
the static key infrastructure will complain if a key is left elevated,
and because an elevated key only affects performance, nothing in KVM will
fail if either key is improperly incremented.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Revert a change to open code bits of kvm_lapic_set_base() when emulating
APIC RESET to fix an apic_hw_disabled underflow bug due to arch.apic_base
and apic_hw_disabled being unsyncrhonized when the APIC is created. If
kvm_arch_vcpu_create() fails after creating the APIC, kvm_free_lapic()
will see the initialized-to-zero vcpu->arch.apic_base and decrement
apic_hw_disabled without KVM ever having incremented apic_hw_disabled.
Using kvm_lapic_set_base() in kvm_lapic_reset() is also desirable for a
potential future where KVM supports RESET outside of vCPU creation, in
which case all the side effects of kvm_lapic_set_base() are needed, e.g.
to handle the transition from x2APIC => xAPIC.
Alternatively, KVM could temporarily increment apic_hw_disabled (and call
kvm_lapic_set_base() at RESET), but that's a waste of cycles and would
impact the performance of other vCPUs and VMs. The other subtle side
effect is that updating the xAPIC ID needs to be done at RESET regardless
of whether the APIC was previously enabled, i.e. kvm_lapic_reset() needs
an explicit call to kvm_apic_set_xapic_id() regardless of whether or not
kvm_lapic_set_base() also performs the update. That makes stuffing the
enable bit at vCPU creation slightly more palatable, as doing so affects
only the apic_hw_disabled key.
Opportunistically tweak the comment to explicitly call out the connection
between vcpu->arch.apic_base and apic_hw_disabled, and add a comment to
call out the need to always do kvm_apic_set_xapic_id() at RESET.
Underflow scenario:
kvm_vm_ioctl() {
kvm_vm_ioctl_create_vcpu() {
kvm_arch_vcpu_create() {
if (something_went_wrong)
goto fail_free_lapic;
/* vcpu->arch.apic_base is initialized when something_went_wrong is false. */
kvm_vcpu_reset() {
kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event) {
vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
}
}
return 0;
fail_free_lapic:
kvm_free_lapic() {
/* vcpu->arch.apic_base is not yet initialized when something_went_wrong is true. */
if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
static_branch_slow_dec_deferred(&apic_hw_disabled); // <= underflow bug.
}
return r;
}
}
}
This (mostly) reverts commit 421221234a.
Fixes: 421221234a ("KVM: x86: Open code necessary bits of kvm_lapic_set_base() at vCPU RESET")
Reported-by: [email protected]
Debugged-by: Tetsuo Handa <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Select HAVE_KVM at all times on arm64, as the OF requirement is
always there (even in the case of an ACPI system, we still depend
on some of the OF infrastructure), and won't fo away.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Acked-by: Will Deacon <[email protected]>
[maz: Drop the "HAVE_KVM if OF" dependency, as OF is always there on arm64,
new commit message]
Signed-off-by: Marc Zyngier <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Unconditionally "source" the generic KVM Kconfig instead of wrapping it
with KVM=y. A future patch will select HAVE_KVM so that referencing
HAVE_KVM in common kernel code doesn't break, and because KVM=y and
HAVE_KVM=n is weird. Source the generic KVM Kconfig unconditionally so
that HAVE_KVM and KVM don't end up with a circular dependency.
Note, all but one of generic KVM's "configs" are of the HAVE_XYZ nature,
and the one outlier correctly takes a dependency on CONFIG_KVM, i.e. the
generic Kconfig is intended to be included unconditionally.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
[maz: made NVHE_EL2_DEBUG depend on KVM]
Signed-off-by: Marc Zyngier <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
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 <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Lai Jiangshan <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: 957ed9effd ("KVM: MMU: prefetch ptes when intercepted guest #PF")
Cc: Sergey Senozhatsky <[email protected]>
Cc: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Reviewed-by: Jim Mattson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 ed02b21309 ("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 <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 a554d207dc ("KVM: X86: Processor States following Reset or INIT").
Back when XCR0 support was added by commit 2acf923e38 ("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 0ee6a51725 ("x86/fpu,
kvm: Simplify fx_init()"). Finally, commit 95a0d01eef ("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 <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 380102c8e4
("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
18fa000ae4 ("KVM: SVM: Reset cr0 properly on vcpu reset").
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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: 66f7b72e11 ("KVM: x86: Make register state after reset conform to specification")
[sean: !!! NOT FOR STABLE !!!]
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 1342d3536d ("[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 195aefde9c ("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 <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
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 <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Rework the CPU selection in the migration worker to ensure the specified
number of migrations are performed when the test iteslf is affined to a
subset of CPUs. The existing logic skips iterations if the target CPU is
not in the original set of possible CPUs, which causes the test to fail
if too many iterations are skipped.
==== Test Assertion Failure ====
rseq_test.c:228: i > (NR_TASK_MIGRATIONS / 2)
pid=10127 tid=10127 errno=4 - Interrupted system call
1 0x00000000004018e5: main at rseq_test.c:227
2 0x00007fcc8fc66bf6: ?? ??:0
3 0x0000000000401959: _start at ??:?
Only performed 4 KVM_RUNs, task stalled too much?
Calculate the min/max possible CPUs as a cheap "best effort" to avoid
high runtimes when the test is affined to a small percentage of CPUs.
Alternatively, a list or xarray of the possible CPUs could be used, but
even in a horrendously inefficient setup, such optimizations are not
needed because the runtime is completely dominated by the cost of
migrating the task, and the absolute runtime is well under a minute in
even truly absurd setups, e.g. running on a subset of vCPUs in a VM that
is heavily overcommited (16 vCPUs per pCPU).
Fixes: 61e52f1630 ("KVM: selftests: Add a test for KVM_RUN+rseq to detect task migration bugs")
Reported-by: Dongli Zhang <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Check whether a CPUID entry's index is significant before checking for a
matching index to hack-a-fix an undefined behavior bug due to consuming
uninitialized data. RESET/INIT emulation uses kvm_cpuid() to retrieve
CPUID.0x1, which does _not_ have a significant index, and fails to
initialize the dummy variable that doubles as EBX/ECX/EDX output _and_
ECX, a.k.a. index, input.
Practically speaking, it's _extremely_ unlikely any compiler will yield
code that causes problems, as the compiler would need to inline the
kvm_cpuid() call to detect the uninitialized data, and intentionally hose
the kernel, e.g. insert ud2, instead of simply ignoring the result of
the index comparison.
Although the sketchy "dummy" pattern was introduced in SVM by commit
66f7b72e11 ("KVM: x86: Make register state after reset conform to
specification"), it wasn't actually broken until commit 7ff6c03503
("KVM: x86: Remove stateful CPUID handling") arbitrarily swapped the
order of operations such that "index" was checked before the significant
flag.
Avoid consuming uninitialized data by reverting to checking the flag
before the index purely so that the fix can be easily backported; the
offending RESET/INIT code has been refactored, moved, and consolidated
from vendor code to common x86 since the bug was introduced. A future
patch will directly address the bad RESET/INIT behavior.
The undefined behavior was detected by syzbot + KernelMemorySanitizer.
BUG: KMSAN: uninit-value in cpuid_entry2_find arch/x86/kvm/cpuid.c:68
BUG: KMSAN: uninit-value in kvm_find_cpuid_entry arch/x86/kvm/cpuid.c:1103
BUG: KMSAN: uninit-value in kvm_cpuid+0x456/0x28f0 arch/x86/kvm/cpuid.c:1183
cpuid_entry2_find arch/x86/kvm/cpuid.c:68 [inline]
kvm_find_cpuid_entry arch/x86/kvm/cpuid.c:1103 [inline]
kvm_cpuid+0x456/0x28f0 arch/x86/kvm/cpuid.c:1183
kvm_vcpu_reset+0x13fb/0x1c20 arch/x86/kvm/x86.c:10885
kvm_apic_accept_events+0x58f/0x8c0 arch/x86/kvm/lapic.c:2923
vcpu_enter_guest+0xfd2/0x6d80 arch/x86/kvm/x86.c:9534
vcpu_run+0x7f5/0x18d0 arch/x86/kvm/x86.c:9788
kvm_arch_vcpu_ioctl_run+0x245b/0x2d10 arch/x86/kvm/x86.c:10020
Local variable ----dummy@kvm_vcpu_reset created at:
kvm_vcpu_reset+0x1fb/0x1c20 arch/x86/kvm/x86.c:10812
kvm_apic_accept_events+0x58f/0x8c0 arch/x86/kvm/lapic.c:2923
Reported-by: [email protected]
Reported-by: Alexander Potapenko <[email protected]>
Fixes: 2a24be79b6 ("KVM: VMX: Set EDX at INIT with CPUID.0x1, Family-Model-Stepping")
Fixes: 7ff6c03503 ("KVM: x86: Remove stateful CPUID handling")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Jim Mattson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Check for a NULL cpumask_var_t when kicking multiple vCPUs via
cpumask_available(), which performs a !NULL check if and only if cpumasks
are configured to be allocated off-stack. This is a meaningless
optimization, e.g. avoids a TEST+Jcc and TEST+CMOV on x86, but more
importantly helps document that the NULL check is necessary even though
all callers pass in a local variable.
No functional change intended.
Cc: Lai Jiangshan <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Vitaly Kuznetsov <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Fix a benign data race reported by syzbot+KCSAN[*] by ensuring vcpu->cpu
is read exactly once, and by ensuring the vCPU is booted from guest mode
if kvm_arch_vcpu_should_kick() returns true. Fix a similar race in
kvm_make_vcpus_request_mask() by ensuring the vCPU is interrupted if
kvm_request_needs_ipi() returns true.
Reading vcpu->cpu before vcpu->mode (via kvm_arch_vcpu_should_kick() or
kvm_request_needs_ipi()) means the target vCPU could get migrated (change
vcpu->cpu) and enter !OUTSIDE_GUEST_MODE between reading vcpu->cpud and
reading vcpu->mode. If that happens, the kick/IPI will be sent to the
old pCPU, not the new pCPU that is now running the vCPU or reading SPTEs.
Although failing to kick the vCPU is not exactly ideal, practically
speaking it cannot cause a functional issue unless there is also a bug in
the caller, and any such bug would exist regardless of kvm_vcpu_kick()'s
behavior.
The purpose of sending an IPI is purely to get a vCPU into the host (or
out of reading SPTEs) so that the vCPU can recognize a change in state,
e.g. a KVM_REQ_* request. If vCPU's handling of the state change is
required for correctness, KVM must ensure either the vCPU sees the change
before entering the guest, or that the sender sees the vCPU as running in
guest mode. All architectures handle this by (a) sending the request
before calling kvm_vcpu_kick() and (b) checking for requests _after_
setting vcpu->mode.
x86's READING_SHADOW_PAGE_TABLES has similar requirements; KVM needs to
ensure it kicks and waits for vCPUs that started reading SPTEs _before_
MMU changes were finalized, but any vCPU that starts reading after MMU
changes were finalized will see the new state and can continue on
uninterrupted.
For uses of kvm_vcpu_kick() that are not paired with a KVM_REQ_*, e.g.
x86's kvm_arch_sync_dirty_log(), the order of the kick must not be relied
upon for functional correctness, e.g. in the dirty log case, userspace
cannot assume it has a 100% complete log if vCPUs are still running.
All that said, eliminate the benign race since the cost of doing so is an
"extra" atomic cmpxchg() in the case where the target vCPU is loaded by
the current pCPU or is not loaded at all. I.e. the kick will be skipped
due to kvm_vcpu_exiting_guest_mode() seeing a compatible vcpu->mode as
opposed to the kick being skipped because of the cpu checks.
Keep the "cpu != me" checks even though they appear useless/impossible at
first glance. x86 processes guest IPI writes in a fast path that runs in
IN_GUEST_MODE, i.e. can call kvm_vcpu_kick() from IN_GUEST_MODE. And
calling kvm_vm_bugged()->kvm_make_vcpus_request_mask() from IN_GUEST or
READING_SHADOW_PAGE_TABLES is perfectly reasonable.
Note, a race with the cpu_online() check in kvm_vcpu_kick() likely
persists, e.g. the vCPU could exit guest mode and get offlined between
the cpu_online() check and the sending of smp_send_reschedule(). But,
the online check appears to exist only to avoid a WARN in x86's
native_smp_send_reschedule() that fires if the target CPU is not online.
The reschedule WARN exists because CPU offlining takes the CPU out of the
scheduling pool, i.e. the WARN is intended to detect the case where the
kernel attempts to schedule a task on an offline CPU. The actual sending
of the IPI is a non-issue as at worst it will simpy be dropped on the
floor. In other words, KVM's usurping of the reschedule IPI could
theoretically trigger a WARN if the stars align, but there will be no
loss of functionality.
[*] https://syzkaller.appspot.com/bug?extid=cd4154e502f43f10808a
Cc: Venkatesh Srinivas <[email protected]>
Cc: Vitaly Kuznetsov <[email protected]>
Fixes: 97222cc831 ("KVM: Emulate local APIC in kernel")
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Vitaly Kuznetsov <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Use vcpu_idx to identify vCPU0 when updating HyperV's TSC page, which is
shared by all vCPUs and "owned" by vCPU0 (because vCPU0 is the only vCPU
that's guaranteed to exist). Using kvm_get_vcpu() to find vCPU works,
but it's a rather odd and suboptimal method to check the index of a given
vCPU.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Jim Mattson <[email protected]>
Reviewed-by: Maxim Levitsky <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Read vcpu->vcpu_idx directly instead of bouncing through the one-line
wrapper, kvm_vcpu_get_idx(), and drop the wrapper. The wrapper is a
remnant of the original implementation and serves no purpose; remove it
before it gains more users.
Back when kvm_vcpu_get_idx() was added by commit 497d72d80a ("KVM: Add
kvm_vcpu_get_idx to get vcpu index in kvm->vcpus"), the implementation
was more than just a simple wrapper as vcpu->vcpu_idx did not exist and
retrieving the index meant walking over the vCPU array to find the given
vCPU.
When vcpu_idx was introduced by commit 8750e72a79 ("KVM: remember
position in kvm->vcpus array"), the helper was left behind, likely to
avoid extra thrash (but even then there were only two users, the original
arm usage having been removed at some point in the past).
No functional change intended.
Suggested-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Maxim Levitsky <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Remove vcpu_vmx.nr_active_uret_msrs and its associated comment, which are
both defunct now that KVM keeps the list constant and instead explicitly
tracks which entries need to be loaded into hardware.
No functional change intended.
Fixes: ee9d22e08d ("KVM: VMX: Use flag to indicate "active" uret MSRs instead of sorting list")
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Explicitly zero the guest's CR3 and mark it available+dirty at RESET/INIT.
Per Intel's SDM and AMD's APM, CR3 is zeroed at both RESET and INIT. For
RESET, this is a nop as vcpu is zero-allocated. For INIT, the bug has
likely escaped notice because no firmware/kernel puts its page tables root
at PA=0, let alone relies on INIT to get the desired CR3 for such page
tables.
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Mark all registers as available and dirty at vCPU creation, as the vCPU has
obviously not been loaded into hardware, let alone been given the chance to
be modified in hardware. On SVM, reading from "uninitialized" hardware is
a non-issue as VMCBs are zero allocated (thus not truly uninitialized) and
hardware does not allow for arbitrary field encoding schemes.
On VMX, backing memory for VMCSes is also zero allocated, but true
initialization of the VMCS _technically_ requires VMWRITEs, as the VMX
architectural specification technically allows CPU implementations to
encode fields with arbitrary schemes. E.g. a CPU could theoretically store
the inverted value of every field, which would result in VMREAD to a
zero-allocated field returns all ones.
In practice, only the AR_BYTES fields are known to be manipulated by
hardware during VMREAD/VMREAD; no known hardware or VMM (for nested VMX)
does fancy encoding of cacheable field values (CR0, CR3, CR4, etc...). In
other words, this is technically a bug fix, but practically speakings it's
a glorified nop.
Failure to mark registers as available has been a lurking bug for quite
some time. The original register caching supported only GPRs (+RIP, which
is kinda sorta a GPR), with the masks initialized at ->vcpu_reset(). That
worked because the two cacheable registers, RIP and RSP, are generally
speaking not read as side effects in other flows.
Arguably, commit aff48baa34 ("KVM: Fetch guest cr3 from hardware on
demand") was the first instance of failure to mark regs available. While
_just_ marking CR3 available during vCPU creation wouldn't have fixed the
VMREAD from an uninitialized VMCS bug because ept_update_paging_mode_cr0()
unconditionally read vmcs.GUEST_CR3, marking CR3 _and_ intentionally not
reading GUEST_CR3 when it's available would have avoided VMREAD to a
technically-uninitialized VMCS.
Fixes: aff48baa34 ("KVM: Fetch guest cr3 from hardware on demand")
Fixes: 6de4f3ada4 ("KVM: Cache pdptrs")
Fixes: 6de12732c4 ("KVM: VMX: Optimize vmx_get_rflags()")
Fixes: 2fb92db1ec ("KVM: VMX: Cache vmcs segment fields")
Fixes: bd31fe495d ("KVM: VMX: Add proper cache tracking for CR0")
Fixes: f98c1e7712 ("KVM: VMX: Add proper cache tracking for CR4")
Fixes: 5addc23519 ("KVM: VMX: Cache vmcs.EXIT_QUALIFICATION using arch avail_reg flags")
Fixes: 8791585837 ("KVM: VMX: Cache vmcs.EXIT_INTR_INFO using arch avail_reg flags")
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Revert the __NR_userfaultfd syscall fallback added for KVM selftests now
that x86's unistd_{32,63}.h overrides are under uapi/ and thus not in
KVM selftests' search path, i.e. now that KVM gets x86 syscall numbers
from the installed kernel headers.
No functional change intended.
Reviewed-by: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Add a test to verify an rseq's CPU ID is updated correctly if the task is
migrated while the kernel is handling KVM_RUN. This is a regression test
for a bug introduced by commit 72c3c0fe54 ("x86/kvm: Use generic xfer
to guest work function"), where TIF_NOTIFY_RESUME would be cleared by KVM
without updating rseq, leading to a stale CPU ID and other badness.
Signed-off-by: Sean Christopherson <[email protected]>
Acked-by: Mathieu Desnoyers <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Move unistd_{32,64}.h from x86/include/asm to x86/include/uapi/asm so
that tools/selftests that install kernel headers, e.g. KVM selftests, can
include non-uapi tools headers, e.g. to get 'struct list_head', without
effectively overriding the installed non-tool uapi headers.
Swapping KVM's search order, e.g. to search the kernel headers before
tool headers, is not a viable option as doing results in linux/type.h and
other core headers getting pulled from the kernel headers, which do not
have the kernel-internal typedefs that are used through tools, including
many files outside of selftests/kvm's control.
Prior to commit cec07f53c3 ("perf tools: Move syscall number fallbacks
from perf-sys.h to tools/arch/x86/include/asm/"), the handcoded numbers
were actual fallbacks, i.e. overriding unistd_{32,64}.h from the kernel
headers was unintentional.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Invoke rseq_handle_notify_resume() from tracehook_notify_resume() now
that the two function are always called back-to-back by architectures
that have rseq. The rseq helper is stubbed out for architectures that
don't support rseq, i.e. this is a nop across the board.
Note, tracehook_notify_resume() is horribly named and arguably does not
belong in tracehook.h as literally every line of code in it has nothing
to do with tracing. But, that's been true since commit a42c6ded82
("move key_repace_session_keyring() into tracehook_notify_resume()")
first usurped tracehook_notify_resume() back in 2012. Punt cleaning that
mess up to future patches.
No functional change intended.
Acked-by: Mathieu Desnoyers <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Invoke rseq's NOTIFY_RESUME handler when processing the flag prior to
transferring to a KVM guest, which is roughly equivalent to an exit to
userspace and processes many of the same pending actions. While the task
cannot be in an rseq critical section as the KVM path is reachable only
by via ioctl(KVM_RUN), the side effects that apply to rseq outside of a
critical section still apply, e.g. the current CPU needs to be updated if
the task is migrated.
Clearing TIF_NOTIFY_RESUME without informing rseq can lead to segfaults
and other badness in userspace VMMs that use rseq in combination with KVM,
e.g. due to the CPU ID being stale after task migration.
Fixes: 72c3c0fe54 ("x86/kvm: Use generic xfer to guest work function")
Reported-by: Peter Foley <[email protected]>
Bisected-by: Doug Evans <[email protected]>
Acked-by: Mathieu Desnoyers <[email protected]>
Cc: Shakeel Butt <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Don't export KVM's MMU notifier count helpers, under no circumstance
should any downstream module, including x86's vendor code, have a
legitimate reason to piggyback KVM's MMU notifier logic. E.g in the x86
case, only KVM's MMU should be elevating the notifier count, and that
code is always built into the core kvm.ko module.
Fixes: edb298c663 ("KVM: x86/mmu: bump mmu notifier count in kvm_zap_gfn_range")
Cc: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Reviewed-by: Maxim Levitsky <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Move "lpage_disallowed_link" out of the first 64 bytes, i.e. out of the
first cache line, of kvm_mmu_page so that "spt" and to a lesser extent
"gfns" land in the first cache line. "lpage_disallowed_link" is accessed
relatively infrequently compared to "spt", which is accessed any time KVM
is walking and/or manipulating the shadow page tables.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Move "tdp_mmu_page" into the 1-byte void left by the recently removed
"mmio_cached" so that it resides in the first 64 bytes of kvm_mmu_page,
i.e. in the same cache line as the most commonly accessed fields.
Don't bother wrapping tdp_mmu_page in CONFIG_X86_64, including the field in
32-bit builds doesn't affect the size of kvm_mmu_page, and a future patch
can always wrap the field in the unlikely event KVM gains a 1-byte flag
that is 32-bit specific.
Note, the size of kvm_mmu_page is also unchanged on CONFIG_X86_64=y due
to it previously sharing an 8-byte chunk with write_flooding_count.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Revert a misguided illegal GPA check when "translating" a non-nested GPA.
The check is woefully incomplete as it does not fill in @exception as
expected by all callers, which leads to KVM attempting to inject a bogus
exception, potentially exposing kernel stack information in the process.
WARNING: CPU: 0 PID: 8469 at arch/x86/kvm/x86.c:525 exception_type+0x98/0xb0 arch/x86/kvm/x86.c:525
CPU: 1 PID: 8469 Comm: syz-executor531 Not tainted 5.14.0-rc7-syzkaller #0
RIP: 0010:exception_type+0x98/0xb0 arch/x86/kvm/x86.c:525
Call Trace:
x86_emulate_instruction+0xef6/0x1460 arch/x86/kvm/x86.c:7853
kvm_mmu_page_fault+0x2f0/0x1810 arch/x86/kvm/mmu/mmu.c:5199
handle_ept_misconfig+0xdf/0x3e0 arch/x86/kvm/vmx/vmx.c:5336
__vmx_handle_exit arch/x86/kvm/vmx/vmx.c:6021 [inline]
vmx_handle_exit+0x336/0x1800 arch/x86/kvm/vmx/vmx.c:6038
vcpu_enter_guest+0x2a1c/0x4430 arch/x86/kvm/x86.c:9712
vcpu_run arch/x86/kvm/x86.c:9779 [inline]
kvm_arch_vcpu_ioctl_run+0x47d/0x1b20 arch/x86/kvm/x86.c:10010
kvm_vcpu_ioctl+0x49e/0xe50 arch/x86/kvm/../../../virt/kvm/kvm_main.c:3652
The bug has escaped notice because practically speaking the GPA check is
useless. The GPA check in question only comes into play when KVM is
walking guest page tables (or "translating" CR3), and KVM already handles
illegal GPA checks by setting reserved bits in rsvd_bits_mask for each
PxE, or in the case of CR3 for loading PTDPTRs, manually checks for an
illegal CR3. This particular failure doesn't hit the existing reserved
bits checks because syzbot sets guest.MAXPHYADDR=1, and IA32 architecture
simply doesn't allow for such an absurd MAXPHYADDR, e.g. 32-bit paging
doesn't define any reserved PA bits checks, which KVM emulates by only
incorporating the reserved PA bits into the "high" bits, i.e. bits 63:32.
Simply remove the bogus check. There is zero meaningful value and no
architectural justification for supporting guest.MAXPHYADDR < 32, and
properly filling the exception would introduce non-trivial complexity.
This reverts commit ec7771ab47.
Fixes: ec7771ab47 ("KVM: x86: mmu: Add guest physical address check in translate_gpa()")
Cc: [email protected]
Reported-by: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Reviewed-by: Vitaly Kuznetsov <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Include pml5_root in the set of special roots if and only if the host,
and thus NPT, is using 5-level paging. mmu_alloc_special_roots() expects
special roots to be allocated as a bundle, i.e. they're either all valid
or all NULL. But for pml5_root, that expectation only holds true if the
host uses 5-level paging, which causes KVM to WARN about pml5_root being
NULL when the other special roots are valid.
The silver lining of 4-level vs. 5-level NPT being tied to the host
kernel's paging level is that KVM's shadow root level is constant; unlike
VMX's EPT, KVM can't choose 4-level NPT based on guest.MAXPHYADDR. That
means KVM can still expect pml5_root to be bundled with the other special
roots, it just needs to be conditioned on the shadow root level.
Fixes: cb0f722aff ("KVM: x86/mmu: Support shadowing NPT when 5-level paging is enabled in host")
Reported-by: Maxim Levitsky <[email protected]>
Reviewed-by: Maxim Levitsky <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Drop @shared from tdp_mmu_link_page() and hardcode it to work for
mmu_lock being held for read. The helper has exactly one caller and
in all likelihood will only ever have exactly one caller. Even if KVM
adds a path to install translations without an initiating page fault,
odds are very, very good that the path will just be a wrapper to the
"page fault" handler (both SNP and TDX RFCs propose patches to do
exactly that).
No functional change intended.
Cc: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Factor in whether or not the old/new SPTEs are shadow-present when
adjusting the large page stats in the TDP MMU. A modified MMIO SPTE can
toggle the page size bit, as bit 7 is used to store the MMIO generation,
i.e. is_large_pte() can get a false positive when called on a MMIO SPTE.
Ditto for nuking SPTEs with REMOVED_SPTE, which sets bit 7 in its magic
value.
Opportunistically move the logic below the check to verify at least one
of the old/new SPTEs is shadow present.
Use is/was_leaf even though is/was_present would suffice. The code
generation is roughly equivalent since all flags need to be computed
prior to the code in question, and using the *_leaf flags will minimize
the diff in a future enhancement to account all pages, i.e. will change
the check to "is_leaf != was_leaf".
Reviewed-by: David Matlack <[email protected]>
Reviewed-by: Ben Gardon <[email protected]>
Fixes: 1699f65c8b ("kvm/x86: Fix 'lpages' kvm stat for TDM MMU")
Cc: [email protected]
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Mingwei Zhang <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
This together with the next patch will fix a future race between
kvm_zap_gfn_range and the page fault handler, which will happen
when AVIC memslot is going to be only partially disabled.
The performance impact is minimal since kvm_zap_gfn_range is only
called by users, update_mtrr() and kvm_post_set_cr0().
Both only use it if the guest has non-coherent DMA, in order to
honor the guest's UC memtype.
MTRR and CD setup only happens at boot, and generally in an area
where the page tables should be small (for CD) or should not
include the affected GFNs at all (for MTRRs).
This is based on a patch suggested by Sean Christopherson:
https://lkml.org/lkml/2021/7/22/1025
Signed-off-by: Sean Christopherson <[email protected]>
Signed-off-by: Maxim Levitsky <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Clear nested.pi_pending on nested VM-Enter even if L2 will run without
posted interrupts enabled. If nested.pi_pending is left set from a
previous L2, vmx_complete_nested_posted_interrupt() will pick up the
stale flag and exit to userspace with an "internal emulation error" due
the new L2 not having a valid nested.pi_desc.
Arguably, vmx_complete_nested_posted_interrupt() should first check for
posted interrupts being enabled, but it's also completely reasonable that
KVM wouldn't screw up a fundamental flag. Not to mention that the mere
existence of nested.pi_pending is a long-standing bug as KVM shouldn't
move the posted interrupt out of the IRR until it's actually processed,
e.g. KVM effectively drops an interrupt when it performs a nested VM-Exit
with a "pending" posted interrupt. Fixing the mess is a future problem.
Prior to vmx_complete_nested_posted_interrupt() interpreting a null PI
descriptor as an error, this was a benign bug as the null PI descriptor
effectively served as a check on PI not being enabled. Even then, the
new flow did not become problematic until KVM started checking the result
of kvm_check_nested_events().
Fixes: 705699a139 ("KVM: nVMX: Enable nested posted interrupt processing")
Fixes: 966eefb896 ("KVM: nVMX: Disable vmcs02 posted interrupts if vmcs12 PID isn't mappable")
Fixes: 47d3530f86c0 ("KVM: x86: Exit to userspace when kvm_check_nested_events fails")
Cc: [email protected]
Cc: Jim Mattson <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Remove the __kvm_handle_fault_on_reboot() and __ex() macros now that all
VMX and SVM instructions use asm goto to handle the fault (or in the
case of VMREAD, completely custom logic). Drop kvm_spurious_fault()'s
asmlinkage annotation as __kvm_handle_fault_on_reboot() was the only
flow that invoked it from assembly code.
Cc: Uros Bizjak <[email protected]>
Cc: Like Xu <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Now that nested VMX pulls KVM's desired VMCS controls from vmcs01 instead
of re-calculating on the fly, bury the helpers that do the calcluations
in vmx.c.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Remove the secondary execution controls cache now that it's effectively
dead code; it is only read immediately after it is written.
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
When preparing controls for vmcs02, grab KVM's desired controls from
vmcs01's shadow state instead of recalculating the controls from scratch,
or in the secondary execution controls, instead of using the dedicated
cache. Calculating secondary exec controls is eye-poppingly expensive
due to the guest CPUID checks, hence the dedicated cache, but the other
calculations aren't exactly free either.
Explicitly clear several bits (x2APIC, DESC exiting, and load EFER on
exit) as appropriate as they may be set in vmcs01, whereas the previous
implementation relied on dynamic bits being cleared in the calculator.
Intentionally propagate VM_{ENTRY,EXIT}_LOAD_IA32_PERF_GLOBAL_CTRL from
vmcs01 to vmcs02. Whether or not PERF_GLOBAL_CTRL is loaded depends on
whether or not perf itself is active, so unless perf stops between the
exit from L1 and entry to L2, vmcs01 will hold the desired value. This
is purely an optimization as atomic_switch_perf_msrs() will set/clear
the control as needed at VM-Enter, i.e. it avoids two extra VMWRITEs in
the case where perf is active (versus starting with the bits clear in
vmcs02, which was the previous behavior).
Cc: Zeng Guang <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Add yet another spinlock for the TDP MMU and take it when marking indirect
shadow pages unsync. When using the TDP MMU and L1 is running L2(s) with
nested TDP, KVM may encounter shadow pages for the TDP entries managed by
L1 (controlling L2) when handling a TDP MMU page fault. The unsync logic
is not thread safe, e.g. the kvm_mmu_page fields are not atomic, and
misbehaves when a shadow page is marked unsync via a TDP MMU page fault,
which runs with mmu_lock held for read, not write.
Lack of a critical section manifests most visibly as an underflow of
unsync_children in clear_unsync_child_bit() due to unsync_children being
corrupted when multiple CPUs write it without a critical section and
without atomic operations. But underflow is the best case scenario. The
worst case scenario is that unsync_children prematurely hits '0' and
leads to guest memory corruption due to KVM neglecting to properly sync
shadow pages.
Use an entirely new spinlock even though piggybacking tdp_mmu_pages_lock
would functionally be ok. Usurping the lock could degrade performance when
building upper level page tables on different vCPUs, especially since the
unsync flow could hold the lock for a comparatively long time depending on
the number of indirect shadow pages and the depth of the paging tree.
For simplicity, take the lock for all MMUs, even though KVM could fairly
easily know that mmu_lock is held for write. If mmu_lock is held for
write, there cannot be contention for the inner spinlock, and marking
shadow pages unsync across multiple vCPUs will be slow enough that
bouncing the kvm_arch cacheline should be in the noise.
Note, even though L2 could theoretically be given access to its own EPT
entries, a nested MMU must hold mmu_lock for write and thus cannot race
against a TDP MMU page fault. I.e. the additional spinlock only _needs_ to
be taken by the TDP MMU, as opposed to being taken by any MMU for a VM
that is running with the TDP MMU enabled. Holding mmu_lock for read also
prevents the indirect shadow page from being freed. But as above, keep
it simple and always take the lock.
Alternative #1, the TDP MMU could simply pass "false" for can_unsync and
effectively disable unsync behavior for nested TDP. Write protecting leaf
shadow pages is unlikely to noticeably impact traditional L1 VMMs, as such
VMMs typically don't modify TDP entries, but the same may not hold true for
non-standard use cases and/or VMMs that are migrating physical pages (from
L1's perspective).
Alternative #2, the unsync logic could be made thread safe. In theory,
simply converting all relevant kvm_mmu_page fields to atomics and using
atomic bitops for the bitmap would suffice. However, (a) an in-depth audit
would be required, (b) the code churn would be substantial, and (c) legacy
shadow paging would incur additional atomic operations in performance
sensitive paths for no benefit (to legacy shadow paging).
Fixes: a2855afc7e ("KVM: x86/mmu: Allow parallel page faults for the TDP MMU")
Cc: [email protected]
Cc: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Set the min_level for the TDP iterator at the root level when zapping all
SPTEs to optimize the iterator's try_step_down(). Zapping a non-leaf
SPTE will recursively zap all its children, thus there is no need for the
iterator to attempt to step down. This avoids rereading the top-level
SPTEs after they are zapped by causing try_step_down() to short-circuit.
In most cases, optimizing try_step_down() will be in the noise as the cost
of zapping SPTEs completely dominates the overall time. The optimization
is however helpful if the zap occurs with relatively few SPTEs, e.g. if KVM
is zapping in response to multiple memslot updates when userspace is adding
and removing read-only memslots for option ROMs. In that case, the task
doing the zapping likely isn't a vCPU thread, but it still holds mmu_lock
for read and thus can be a noisy neighbor of sorts.
Reviewed-by: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Pass "all ones" as the end GFN to signal "zap all" for the TDP MMU and
really zap all SPTEs in this case. As is, zap_gfn_range() skips non-leaf
SPTEs whose range exceeds the range to be zapped. If shadow_phys_bits is
not aligned to the range size of top-level SPTEs, e.g. 512gb with 4-level
paging, the "zap all" flows will skip top-level SPTEs whose range extends
beyond shadow_phys_bits and leak their SPs when the VM is destroyed.
Use the current upper bound (based on host.MAXPHYADDR) to detect that the
caller wants to zap all SPTEs, e.g. instead of using the max theoretical
gfn, 1 << (52 - 12). The more precise upper bound allows the TDP iterator
to terminate its walk earlier when running on hosts with MAXPHYADDR < 52.
Add a WARN on kmv->arch.tdp_mmu_pages when the TDP MMU is destroyed to
help future debuggers should KVM decide to leak SPTEs again.
The bug is most easily reproduced by running (and unloading!) KVM in a
VM whose host.MAXPHYADDR < 39, as the SPTE for gfn=0 will be skipped.
=============================================================================
BUG kvm_mmu_page_header (Not tainted): Objects remaining in kvm_mmu_page_header on __kmem_cache_shutdown()
-----------------------------------------------------------------------------
Slab 0x000000004d8f7af1 objects=22 used=2 fp=0x00000000624d29ac flags=0x4000000000000200(slab|zone=1)
CPU: 0 PID: 1582 Comm: rmmod Not tainted 5.14.0-rc2+ #420
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
Call Trace:
dump_stack_lvl+0x45/0x59
slab_err+0x95/0xc9
__kmem_cache_shutdown.cold+0x3c/0x158
kmem_cache_destroy+0x3d/0xf0
kvm_mmu_module_exit+0xa/0x30 [kvm]
kvm_arch_exit+0x5d/0x90 [kvm]
kvm_exit+0x78/0x90 [kvm]
vmx_exit+0x1a/0x50 [kvm_intel]
__x64_sys_delete_module+0x13f/0x220
do_syscall_64+0x3b/0xc0
entry_SYSCALL_64_after_hwframe+0x44/0xae
Fixes: faaf05b00a ("kvm: x86/mmu: Support zapping SPTEs in the TDP MMU")
Cc: [email protected]
Cc: Ben Gardon <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Use vmx_need_pf_intercept() when determining if L0 wants to handle a #PF
in L2 or if the VM-Exit should be forwarded to L1. The current logic fails
to account for the case where #PF is intercepted to handle
guest.MAXPHYADDR < host.MAXPHYADDR and ends up reflecting all #PFs into
L1. At best, L1 will complain and inject the #PF back into L2. At
worst, L1 will eat the unexpected fault and cause L2 to hang on infinite
page faults.
Note, while the bug was technically introduced by the commit that added
support for the MAXPHYADDR madness, the shame is all on commit
a0c134347b ("KVM: VMX: introduce vmx_need_pf_intercept").
Fixes: 1dbf5d68af ("KVM: VMX: Add guest physical address check in EPT violation and misconfig")
Cc: [email protected]
Cc: Peter Shier <[email protected]>
Cc: Oliver Upton <[email protected]>
Cc: Jim Mattson <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Remove an ancient restriction that disallowed exposing EFER.NX to the
guest if EFER.NX=0 on the host, even if NX is fully supported by the CPU.
The motivation of the check, added by commit 2cc51560ae ("KVM: VMX:
Avoid saving and restoring msr_efer on lightweight vmexit"), was to rule
out the case of host.EFER.NX=0 and guest.EFER.NX=1 so that KVM could run
the guest with the host's EFER.NX and thus avoid context switching EFER
if the only divergence was the NX bit.
Fast forward to today, and KVM has long since stopped running the guest
with the host's EFER.NX. Not only does KVM context switch EFER if
host.EFER.NX=1 && guest.EFER.NX=0, KVM also forces host.EFER.NX=0 &&
guest.EFER.NX=1 when using shadow paging (to emulate SMEP). Furthermore,
the entire motivation for the restriction was made obsolete over a decade
ago when Intel added dedicated host and guest EFER fields in the VMCS
(Nehalem timeframe), which reduced the overhead of context switching EFER
from 400+ cycles (2 * WRMSR + 1 * RDMSR) to a mere ~2 cycles.
In practice, the removed restriction only affects non-PAE 32-bit kernels,
as EFER.NX is set during boot if NX is supported and the kernel will use
PAE paging (32-bit or 64-bit), regardless of whether or not the kernel
will actually use NX itself (mark PTEs non-executable).
Alternatively and/or complementarily, startup_32_smp() in head_32.S could
be modified to set EFER.NX=1 regardless of paging mode, thus eliminating
the scenario where NX is supported but not enabled. However, that runs
the risk of breaking non-KVM non-PAE kernels (though the risk is very,
very low as there are no known EFER.NX errata), and also eliminates an
easy-to-use mechanism for stressing KVM's handling of guest vs. host EFER
across nested virtualization transitions.
Suggested-by: Paolo Bonzini <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>
Sean Christopherson