When unbinding a memslot from a guest_memfd instance, remove the bindings
even if the guest_memfd file is dying, i.e. even if its file refcount has
gone to zero. If the memslot is freed before the file is fully released,
nullifying the memslot side of the binding in kvm_gmem_release() will
write to freed memory, as detected by syzbot+KASAN:
==================================================================
BUG: KASAN: slab-use-after-free in kvm_gmem_release+0x176/0x440 virt/kvm/guest_memfd.c:353
Write of size 8 at addr ffff88807befa508 by task syz.0.17/6022
CPU: 0 UID: 0 PID: 6022 Comm: syz.0.17 Not tainted syzkaller #0 PREEMPT(full)
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 10/02/2025
Call Trace:
<TASK>
dump_stack_lvl+0x189/0x250 lib/dump_stack.c:120
print_address_description mm/kasan/report.c:378 [inline]
print_report+0xca/0x240 mm/kasan/report.c:482
kasan_report+0x118/0x150 mm/kasan/report.c:595
kvm_gmem_release+0x176/0x440 virt/kvm/guest_memfd.c:353
__fput+0x44c/0xa70 fs/file_table.c:468
task_work_run+0x1d4/0x260 kernel/task_work.c:227
resume_user_mode_work include/linux/resume_user_mode.h:50 [inline]
exit_to_user_mode_loop+0xe9/0x130 kernel/entry/common.c:43
exit_to_user_mode_prepare include/linux/irq-entry-common.h:225 [inline]
syscall_exit_to_user_mode_work include/linux/entry-common.h:175 [inline]
syscall_exit_to_user_mode include/linux/entry-common.h:210 [inline]
do_syscall_64+0x2bd/0xfa0 arch/x86/entry/syscall_64.c:100
entry_SYSCALL_64_after_hwframe+0x77/0x7f
RIP: 0033:0x7fbeeff8efc9
</TASK>
Allocated by task 6023:
kasan_save_stack mm/kasan/common.c:56 [inline]
kasan_save_track+0x3e/0x80 mm/kasan/common.c:77
poison_kmalloc_redzone mm/kasan/common.c:397 [inline]
__kasan_kmalloc+0x93/0xb0 mm/kasan/common.c:414
kasan_kmalloc include/linux/kasan.h:262 [inline]
__kmalloc_cache_noprof+0x3e2/0x700 mm/slub.c:5758
kmalloc_noprof include/linux/slab.h:957 [inline]
kzalloc_noprof include/linux/slab.h:1094 [inline]
kvm_set_memory_region+0x747/0xb90 virt/kvm/kvm_main.c:2104
kvm_vm_ioctl_set_memory_region+0x6f/0xd0 virt/kvm/kvm_main.c:2154
kvm_vm_ioctl+0x957/0xc60 virt/kvm/kvm_main.c:5201
vfs_ioctl fs/ioctl.c:51 [inline]
__do_sys_ioctl fs/ioctl.c:597 [inline]
__se_sys_ioctl+0xfc/0x170 fs/ioctl.c:583
do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
do_syscall_64+0xfa/0xfa0 arch/x86/entry/syscall_64.c:94
entry_SYSCALL_64_after_hwframe+0x77/0x7f
Freed by task 6023:
kasan_save_stack mm/kasan/common.c:56 [inline]
kasan_save_track+0x3e/0x80 mm/kasan/common.c:77
kasan_save_free_info+0x46/0x50 mm/kasan/generic.c:584
poison_slab_object mm/kasan/common.c:252 [inline]
__kasan_slab_free+0x5c/0x80 mm/kasan/common.c:284
kasan_slab_free include/linux/kasan.h:234 [inline]
slab_free_hook mm/slub.c:2533 [inline]
slab_free mm/slub.c:6622 [inline]
kfree+0x19a/0x6d0 mm/slub.c:6829
kvm_set_memory_region+0x9c4/0xb90 virt/kvm/kvm_main.c:2130
kvm_vm_ioctl_set_memory_region+0x6f/0xd0 virt/kvm/kvm_main.c:2154
kvm_vm_ioctl+0x957/0xc60 virt/kvm/kvm_main.c:5201
vfs_ioctl fs/ioctl.c:51 [inline]
__do_sys_ioctl fs/ioctl.c:597 [inline]
__se_sys_ioctl+0xfc/0x170 fs/ioctl.c:583
do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
do_syscall_64+0xfa/0xfa0 arch/x86/entry/syscall_64.c:94
entry_SYSCALL_64_after_hwframe+0x77/0x7f
Deliberately don't acquire filemap invalid lock when the file is dying as
the lifecycle of f_mapping is outside the purview of KVM. Dereferencing
the mapping is *probably* fine, but there's no need to invalidate anything
as memslot deletion is responsible for zapping SPTEs, and the only code
that can access the dying file is kvm_gmem_release(), whose core code is
mutually exclusive with unbinding.
Note, the mutual exclusivity is also what makes it safe to access the
bindings on a dying gmem instance. Unbinding either runs with slots_lock
held, or after the last reference to the owning "struct kvm" is put, and
kvm_gmem_release() nullifies the slot pointer under slots_lock, and puts
its reference to the VM after that is done.
Reported-by: syzbot+2479e53d0db9b32ae2aa@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/68fa7a22.a70a0220.3bf6c6.008b.GAE@google.com
Tested-by: syzbot+2479e53d0db9b32ae2aa@syzkaller.appspotmail.com
Fixes: a7800aa80e ("KVM: Add KVM_CREATE_GUEST_MEMFD ioctl() for guest-specific backing memory")
Cc: stable@vger.kernel.org
Cc: Hillf Danton <hdanton@sina.com>
Reviewed-By: Vishal Annapurve <vannapurve@google.com>
Link: https://patch.msgid.link/20251104011205.3853541-1-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
KVM x86 fixes for 6.18:
- Expand the KVM_PRE_FAULT_MEMORY selftest to add a regression test for the
bug fixed by commit 3ccbf6f470 ("KVM: x86/mmu: Return -EAGAIN if userspace
deletes/moves memslot during prefault")
- Don't try to get PMU capabbilities from perf when running a CPU with hybrid
CPUs/PMUs, as perf will rightly WARN.
- Rework KVM_CAP_GUEST_MEMFD_MMAP (newly introduced in 6.18) into a more
generic KVM_CAP_GUEST_MEMFD_FLAGS
- Add a guest_memfd INIT_SHARED flag and require userspace to explicitly set
said flag to initialize memory as SHARED, irrespective of MMAP. The
behavior merged in 6.18 is that enabling mmap() implicitly initializes
memory as SHARED, which would result in an ABI collision for x86 CoCo VMs
as their memory is currently always initialized PRIVATE.
- Allow mmap() on guest_memfd for x86 CoCo VMs, i.e. on VMs with private
memory, to enable testing such setups, i.e. to hopefully flush out any
other lurking ABI issues before 6.18 is officially released.
- Add testcases to the guest_memfd selftest to cover guest_memfd without MMAP,
and host userspace accesses to mmap()'d private memory.
Allow mmap() on guest_memfd instances for x86 VMs with private memory as
the need to track private vs. shared state in the guest_memfd instance is
only pertinent to INIT_SHARED. Doing mmap() on private memory isn't
terrible useful (yet!), but it's now possible, and will be desirable when
guest_memfd gains support for other VMA-based syscalls, e.g. mbind() to
set NUMA policy.
Lift the restriction now, before MMAP support is officially released, so
that KVM doesn't need to add another capability to enumerate support for
mmap() on private memory.
Fixes: 3d3a04fad2 ("KVM: Allow and advertise support for host mmap() on guest_memfd files")
Reviewed-by: Ackerley Tng <ackerleytng@google.com>
Tested-by: Ackerley Tng <ackerleytng@google.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/20251003232606.4070510-6-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Add KVM_GENERIC_MMU_NOTIFIER as a dependency for selecting KVM_GUEST_MEMFD,
as guest_memfd relies on kvm_mmu_invalidate_{begin,end}(), which are
defined if and only if the generic mmu_notifier implementation is enabled.
The missing dependency is currently benign as s390 is the only KVM arch
that doesn't utilize the generic mmu_notifier infrastructure, and s390
doesn't currently support guest_memfd.
Fixes: a7800aa80e ("KVM: Add KVM_CREATE_GUEST_MEMFD ioctl() for guest-specific backing memory")
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/20251003232606.4070510-5-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
When invalidating gmem ranges, e.g. in response to PUNCH_HOLE, process all
possible range types (PRIVATE vs. SHARED) for the gmem instance. Since
since guest_memfd doesn't yet support in-place conversions, simply pivot
on INIT_SHARED as a gmem instance can currently only have private or shared
memory, not both.
Failure to mark shared GPAs for invalidation is benign in the current code
base, as only x86's TDX consumes KVM_FILTER_{PRIVATE,SHARED}, and TDX
doesn't yet support INIT_SHARED with guest_memfd. However, invalidating
only private GPAs is conceptually wrong and a lurking bug, e.g. could
result in missed invalidations if ARM starts filtering invalidations based
on attributes.
Fixes: 3d3a04fad2 ("KVM: Allow and advertise support for host mmap() on guest_memfd files")
Reviewed-by: Ackerley Tng <ackerleytng@google.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/20251003232606.4070510-4-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Add a guest_memfd flag to allow userspace to state that the underlying
memory should be configured to be initialized as shared, and reject user
page faults if the guest_memfd instance's memory isn't shared. Because
KVM doesn't yet support in-place private<=>shared conversions, all
guest_memfd memory effectively follows the initial state.
Alternatively, KVM could deduce the initial state based on MMAP, which for
all intents and purposes is what KVM currently does. However, implicitly
deriving the default state based on MMAP will result in a messy ABI when
support for in-place conversions is added.
For x86 CoCo VMs, which don't yet support MMAP, memory is currently private
by default (otherwise the memory would be unusable). If MMAP implies
memory is shared by default, then the default state for CoCo VMs will vary
based on MMAP, and from userspace's perspective, will change when in-place
conversion support is added. I.e. to maintain guest<=>host ABI, userspace
would need to immediately convert all memory from shared=>private, which
is both ugly and inefficient. The inefficiency could be avoided by adding
a flag to state that memory is _private_ by default, irrespective of MMAP,
but that would lead to an equally messy and hard to document ABI.
Bite the bullet and immediately add a flag to control the default state so
that the effective behavior is explicit and straightforward.
Fixes: 3d3a04fad2 ("KVM: Allow and advertise support for host mmap() on guest_memfd files")
Cc: David Hildenbrand <david@redhat.com>
Reviewed-by: Fuad Tabba <tabba@google.com>
Tested-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Ackerley Tng <ackerleytng@google.com>
Tested-by: Ackerley Tng <ackerleytng@google.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/20251003232606.4070510-3-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Rework the not-yet-released KVM_CAP_GUEST_MEMFD_MMAP into a more generic
KVM_CAP_GUEST_MEMFD_FLAGS capability so that adding new flags doesn't
require a new capability, and so that developers aren't tempted to bundle
multiple flags into a single capability.
Note, kvm_vm_ioctl_check_extension_generic() can only return a 32-bit
value, but that limitation can be easily circumvented by adding e.g.
KVM_CAP_GUEST_MEMFD_FLAGS2 in the unlikely event guest_memfd supports more
than 32 flags.
Reviewed-by: Ackerley Tng <ackerleytng@google.com>
Tested-by: Ackerley Tng <ackerleytng@google.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/20251003232606.4070510-2-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Pull hyperv updates from Wei Liu:
- Unify guest entry code for KVM and MSHV (Sean Christopherson)
- Switch Hyper-V MSI domain to use msi_create_parent_irq_domain()
(Nam Cao)
- Add CONFIG_HYPERV_VMBUS and limit the semantics of CONFIG_HYPERV
(Mukesh Rathor)
- Add kexec/kdump support on Azure CVMs (Vitaly Kuznetsov)
- Deprecate hyperv_fb in favor of Hyper-V DRM driver (Prasanna
Kumar T S M)
- Miscellaneous enhancements, fixes and cleanups (Abhishek Tiwari,
Alok Tiwari, Nuno Das Neves, Wei Liu, Roman Kisel, Michael Kelley)
* tag 'hyperv-next-signed-20251006' of git://git.kernel.org/pub/scm/linux/kernel/git/hyperv/linux:
hyperv: Remove the spurious null directive line
MAINTAINERS: Mark hyperv_fb driver Obsolete
fbdev/hyperv_fb: deprecate this in favor of Hyper-V DRM driver
Drivers: hv: Make CONFIG_HYPERV bool
Drivers: hv: Add CONFIG_HYPERV_VMBUS option
Drivers: hv: vmbus: Fix typos in vmbus_drv.c
Drivers: hv: vmbus: Fix sysfs output format for ring buffer index
Drivers: hv: vmbus: Clean up sscanf format specifier in target_cpu_store()
x86/hyperv: Switch to msi_create_parent_irq_domain()
mshv: Use common "entry virt" APIs to do work in root before running guest
entry: Rename "kvm" entry code assets to "virt" to genericize APIs
entry/kvm: KVM: Move KVM details related to signal/-EINTR into KVM proper
mshv: Handle NEED_RESCHED_LAZY before transferring to guest
x86/hyperv: Add kexec/kdump support on Azure CVMs
Drivers: hv: Simplify data structures for VMBus channel close message
Drivers: hv: util: Cosmetic changes for hv_utils_transport.c
mshv: Add support for a new parent partition configuration
clocksource: hyper-v: Skip unnecessary checks for the root partition
hyperv: Add missing field to hv_output_map_device_interrupt
Rename the "kvm" entry code files and Kconfigs to use generic "virt"
nomenclature so that the code can be reused by other hypervisors (or
rather, their root/dom0 partition drivers), without incorrectly suggesting
the code somehow relies on and/or involves KVM.
No functional change intended.
Signed-off-by: Sean Christopherson <seanjc@google.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Joel Fernandes <joelagnelf@nvidia.com>
Signed-off-by: Wei Liu <wei.liu@kernel.org>
Rework the vast majority of KVM's exports to expose symbols only to KVM
submodules, i.e. to x86's kvm-{amd,intel}.ko and PPC's kvm-{pr,hv}.ko.
With few exceptions, KVM's exported APIs are intended (and safe) for KVM-
internal usage only.
Keep kvm_get_kvm(), kvm_get_kvm_safe(), and kvm_put_kvm() as normal
exports, as they are needed by VFIO, and are generally safe for external
usage (though ideally even the get/put APIs would be KVM-internal, and
VFIO would pin a VM by grabbing a reference to its associated file).
Implement a framework in kvm_types.h in anticipation of providing a macro
to restrict KVM-specific kernel exports, i.e. to provide symbol exports
for KVM if and only if KVM is built as one or more modules.
Link: https://lore.kernel.org/r/20250919003303.1355064-3-seanjc@google.com
Cc: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
KVM SVM changes for 6.18
- Require a minimum GHCB version of 2 when starting SEV-SNP guests via
KVM_SEV_INIT2 so that invalid GHCB versions result in immediate errors
instead of latent guest failures.
- Add support for Secure TSC for SEV-SNP guests, which prevents the untrusted
host from tampering with the guest's TSC frequency, while still allowing the
the VMM to configure the guest's TSC frequency prior to launch.
- Mitigate the potential for TOCTOU bugs when accessing GHCB fields by
wrapping all accesses via READ_ONCE().
- Validate the XCR0 provided by the guest (via the GHCB) to avoid tracking a
bogous XCR0 value in KVM's software model.
- Save an SEV guest's policy if and only if LAUNCH_START fully succeeds to
avoid leaving behind stale state (thankfully not consumed in KVM).
- Explicitly reject non-positive effective lengths during SNP's LAUNCH_UPDATE
instead of subtly relying on guest_memfd to do the "heavy" lifting.
- Reload the pre-VMRUN TSC_AUX on #VMEXIT for SEV-ES guests, not the host's
desired TSC_AUX, to fix a bug where KVM could clobber a different vCPU's
TSC_AUX due to hardware not matching the value cached in the user-return MSR
infrastructure.
- Enable AVIC by default for Zen4+ if x2AVIC (and other prereqs) is supported,
and clean up the AVIC initialization code along the way.
KVM x86 MMU changes for 6.18
- Recover possible NX huge pages within the TDP MMU under read lock to
reduce guest jitter when restoring NX huge pages.
- Return -EAGAIN during prefault if userspace concurrently deletes/moves the
relevant memslot to fix an issue where prefaulting could deadlock with the
memslot update.
- Don't retry in TDX's anti-zero-step mitigation if the target memslot is
invalid, i.e. is being deleted or moved, to fix a deadlock scenario similar
to the aforementioned prefaulting case.
KVM/arm64 updates for 6.18
- Add support for FF-A 1.2 as the secure memory conduit for pKVM,
allowing more registers to be used as part of the message payload.
- Change the way pKVM allocates its VM handles, making sure that the
privileged hypervisor is never tricked into using uninitialised
data.
- Speed up MMIO range registration by avoiding unnecessary RCU
synchronisation, which results in VMs starting much quicker.
- Add the dump of the instruction stream when panic-ing in the EL2
payload, just like the rest of the kernel has always done. This will
hopefully help debugging non-VHE setups.
- Add 52bit PA support to the stage-1 page-table walker, and make use
of it to populate the fault level reported to the guest on failing
to translate a stage-1 walk.
- Add NV support to the GICv3-on-GICv5 emulation code, ensuring
feature parity for guests, irrespective of the host platform.
- Fix some really ugly architecture problems when dealing with debug
in a nested VM. This has some bad performance impacts, but is at
least correct.
- Add enough infrastructure to be able to disable EL2 features and
give effective values to the EL2 control registers. This then allows
a bunch of features to be turned off, which helps cross-host
migration.
- Large rework of the selftest infrastructure to allow most tests to
transparently run at EL2. This is the first step towards enabling
NV testing.
- Various fixes and improvements all over the map, including one BE
fix, just in time for the removal of the feature.
Check for an invalid length during LAUNCH_UPDATE at the start of
snp_launch_update() instead of subtly relying on kvm_gmem_populate() to
detect the bad state. Code that directly handles userspace input
absolutely should sanitize those inputs; failure to do so is asking for
bugs where KVM consumes an invalid "npages".
Keep the check in gmem, but wrap it in a WARN to flag any bad usage by
the caller.
Note, this is technically an ABI change as KVM would previously allow a
length of '0'. But allowing a length of '0' is nonsensical and creates
pointless conundrums in KVM. E.g. an empty range is arguably neither
private nor shared, but LAUNCH_UPDATE will fail if the starting gpa can't
be made private. In practice, no known or well-behaved VMM passes a
length of '0'.
Note #2, the PAGE_ALIGNED(params.len) check ensures that lengths between
1 and 4095 (inclusive) are also rejected, i.e. that KVM won't end up with
npages=0 when doing "npages = params.len / PAGE_SIZE".
Cc: Thomas Lendacky <thomas.lendacky@amd.com>
Cc: Michael Roth <michael.roth@amd.com>
Reviewed-by: Tom Lendacky <thomas.lendacky@amd.com>
Link: https://lore.kernel.org/r/20250919211649.1575654-1-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Device MMIO registration may happen quite frequently during VM boot,
and the SRCU synchronization each time has a measurable effect
on VM startup time. In our experiments it can account for around 25%
of a VM's startup time.
Replace the synchronization with a deferred free of the old kvm_io_bus
structure.
Tested-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: Keir Fraser <keirf@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
This ensures that, if a VCPU has "observed" that an IO registration has
occurred, the instruction currently being trapped or emulated will also
observe the IO registration.
At the same time, enforce that kvm_get_bus() is used only on the
update side, ensuring that a long-term reference cannot be obtained by
an SRCU reader.
Signed-off-by: Keir Fraser <keirf@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Avoid local retries within the TDX EPT violation handler if a retry is
triggered by faulting in an invalid memslot, indicating that the memslot
is undergoing a removal process. Faulting in a GPA from an invalid
memslot will never succeed, and holding SRCU prevents memslot deletion
from succeeding, i.e. retrying when the memslot is being actively deleted
will lead to (breakable) deadlock.
Opportunistically export kvm_vcpu_gfn_to_memslot() to allow for a per-vCPU
lookup (which, strictly speaking, is unnecessary since TDX doesn't support
SMM, but aligns the TDX code with the MMU code).
Fixes: b0327bb2e7 ("KVM: TDX: Retry locally in TDX EPT violation handler on RET_PF_RETRY")
Reported-by: Reinette Chatre <reinette.chatre@intel.com>
Closes: https://lore.kernel.org/all/20250519023737.30360-1-yan.y.zhao@intel.com
[Yan: Wrote patch log, comment, fixed a minor error, function export]
Signed-off-by: Yan Zhao <yan.y.zhao@intel.com>
Link: https://lore.kernel.org/r/20250822070523.26495-1-yan.y.zhao@intel.com
[sean: massage changelog, relocate and tweak comment]
Signed-off-by: Sean Christopherson <seanjc@google.com>
Now that all the x86 and arm64 plumbing for mmap() on guest_memfd is in
place, allow userspace to set GUEST_MEMFD_FLAG_MMAP and advertise support
via a new capability, KVM_CAP_GUEST_MEMFD_MMAP.
The availability of this capability is determined per architecture, and
its enablement for a specific guest_memfd instance is controlled by the
GUEST_MEMFD_FLAG_MMAP flag at creation time.
Update the KVM API documentation to detail the KVM_CAP_GUEST_MEMFD_MMAP
capability, the associated GUEST_MEMFD_FLAG_MMAP, and provide essential
information regarding support for mmap in guest_memfd.
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Shivank Garg <shivankg@amd.com>
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20250729225455.670324-22-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a new internal flag, KVM_MEMSLOT_GMEM_ONLY, to the top half of
memslot->flags (which makes it strictly for KVM's internal use). This
flag tracks when a guest_memfd-backed memory slot supports host
userspace mmap operations, which implies that all memory, not just
private memory for CoCo VMs, is consumed through guest_memfd: "gmem
only".
This optimization avoids repeatedly checking the underlying guest_memfd
file for mmap support, which would otherwise require taking and
releasing a reference on the file for each check. By caching this
information directly in the memslot, we reduce overhead and simplify the
logic involved in handling guest_memfd-backed pages for host mappings.
Reviewed-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Shivank Garg <shivankg@amd.com>
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Acked-by: David Hildenbrand <david@redhat.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20250729225455.670324-12-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce the core infrastructure to enable host userspace to mmap()
guest_memfd-backed memory. This is needed for several evolving KVM use
cases:
* Non-CoCo VM backing: Allows VMMs like Firecracker to run guests
entirely backed by guest_memfd, even for non-CoCo VMs [1]. This
provides a unified memory management model and simplifies guest memory
handling.
* Direct map removal for enhanced security: This is an important step
for direct map removal of guest memory [2]. By allowing host userspace
to fault in guest_memfd pages directly, we can avoid maintaining host
kernel direct maps of guest memory. This provides additional hardening
against Spectre-like transient execution attacks by removing a
potential attack surface within the kernel.
* Future guest_memfd features: This also lays the groundwork for future
enhancements to guest_memfd, such as supporting huge pages and
enabling in-place sharing of guest memory with the host for CoCo
platforms that permit it [3].
Enable the basic mmap and fault handling logic within guest_memfd, but
hold off on allow userspace to actually do mmap() until the architecture
support is also in place.
[1] https://github.com/firecracker-microvm/firecracker/tree/feature/secret-hiding
[2] https://lore.kernel.org/linux-mm/cc1bb8e9bc3e1ab637700a4d3defeec95b55060a.camel@amazon.com
[3] https://lore.kernel.org/all/c1c9591d-218a-495c-957b-ba356c8f8e09@redhat.com/T/#u
Reviewed-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Shivank Garg <shivankg@amd.com>
Acked-by: David Hildenbrand <david@redhat.com>
Co-developed-by: Ackerley Tng <ackerleytng@google.com>
Signed-off-by: Ackerley Tng <ackerleytng@google.com>
Signed-off-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20250729225455.670324-11-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Enable KVM_GUEST_MEMFD for all KVM x86 64-bit builds, i.e. for "default"
VM types when running on 64-bit KVM. This will allow using guest_memfd
to back non-private memory for all VM shapes, by supporting mmap() on
guest_memfd.
Opportunistically clean up various conditionals that become tautologies
once x86 selects KVM_GUEST_MEMFD more broadly. Specifically, because
SW protected VMs, SEV, and TDX are all 64-bit only, private memory no
longer needs to take explicit dependencies on KVM_GUEST_MEMFD, because
it is effectively a prerequisite.
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20250729225455.670324-10-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rename kvm_slot_can_be_private() to kvm_slot_has_gmem() to improve
clarity and accurately reflect its purpose.
The function kvm_slot_can_be_private() was previously used to check if a
given kvm_memory_slot is backed by guest_memfd. However, its name
implied that the memory in such a slot was exclusively "private".
As guest_memfd support expands to include non-private memory (e.g.,
shared host mappings), it's important to remove this association. The
new name, kvm_slot_has_gmem(), states that the slot is backed by
guest_memfd without making assumptions about the memory's privacy
attributes.
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Shivank Garg <shivankg@amd.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Co-developed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20250729225455.670324-7-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The original name was vague regarding its functionality. This Kconfig
option specifically enables and gates the kvm_gmem_populate() function,
which is responsible for populating a GPA range with guest data.
The new name, HAVE_KVM_ARCH_GMEM_POPULATE, describes the purpose of the
option: to enable arch-specific guest_memfd population mechanisms. It
also follows the same pattern as the other HAVE_KVM_ARCH_* configuration
options.
This improves clarity for developers and ensures the name accurately
reflects the functionality it controls, especially as guest_memfd
support expands beyond purely "private" memory scenarios.
Temporarily keep KVM_GENERIC_PRIVATE_MEM as an x86-only config so as to
minimize churn, and to hopefully make it easier to see what features
require HAVE_KVM_ARCH_GMEM_POPULATE. On that note, omit GMEM_POPULATE
for KVM_X86_SW_PROTECTED_VM, as regular ol' memset() suffices for
software-protected VMs.
As for KVM_GENERIC_PRIVATE_MEM, a future change will select KVM_GUEST_MEMFD
for all 64-bit KVM builds, at which point the intermediate config will
become obsolete and can/will be dropped.
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Shivank Garg <shivankg@amd.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Co-developed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Co-developed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20250729225455.670324-6-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rename the Kconfig option CONFIG_KVM_PRIVATE_MEM to
CONFIG_KVM_GUEST_MEMFD. The original name implied that the feature only
supported "private" memory. However, CONFIG_KVM_PRIVATE_MEM enables
guest_memfd in general, which is not exclusively for private memory.
Subsequent patches in this series will add guest_memfd support for
non-CoCo VMs, whose memory is not private.
Renaming the Kconfig option to CONFIG_KVM_GUEST_MEMFD more accurately
reflects its broader scope as the main Kconfig option for all
guest_memfd-backed memory. This provides clearer semantics for the
option and avoids confusion as new features are introduced.
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Shivank Garg <shivankg@amd.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Co-developed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20250729225455.670324-2-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
KVM VFIO device assignment cleanups for 6.17
Kill off kvm_arch_{start,end}_assignment() and x86's associated tracking now
that KVM no longer uses assigned_device_count as a bad heuristic for "VM has
an irqbypass producer" or for "VM has access to host MMIO".
KVM Dirty Ring changes for 6.17
Fix issues with dirty ring harvesting where KVM doesn't bound the processing
of entries in any way, which allows userspace to keep KVM in a tight loop
indefinitely. Clean up code and comments along the way.
KVM generic changes for 6.17
- Add a tracepoint for KVM_SET_MEMORY_ATTRIBUTES to help debug issues related
to private <=> shared memory conversions.
- Drop guest_memfd's .getattr() implementation as the VFS layer will call
generic_fillattr() if inode_operations.getattr is NULL.
KVM IRQ changes for 6.17
- Rework irqbypass to track/match producers and consumers via an xarray
instead of a linked list. Using a linked list leads to O(n^2) insertion
times, which is hugely problematic for use cases that create large numbers
of VMs. Such use cases typically don't actually use irqbypass, but
eliminating the pointless registration is a future problem to solve as it
likely requires new uAPI.
- Track irqbypass's "token" as "struct eventfd_ctx *" instead of a "void *",
to avoid making a simple concept unnecessarily difficult to understand.
- Add CONFIG_KVM_IOAPIC for x86 to allow disabling support for I/O APIC, PIC,
and PIT emulation at compile time.
- Drop x86's irq_comm.c, and move a pile of IRQ related code into irq.c.
- Fix a variety of flaws and bugs in the AVIC device posted IRQ code.
- Inhibited AVIC if a vCPU's ID is too big (relative to what hardware
supports) instead of rejecting vCPU creation.
- Extend enable_ipiv module param support to SVM, by simply leaving IsRunning
clear in the vCPU's physical ID table entry.
- Disable IPI virtualization, via enable_ipiv, if the CPU is affected by
erratum #1235, to allow (safely) enabling AVIC on such CPUs.
- Dedup x86's device posted IRQ code, as the vast majority of functionality
can be shared verbatime between SVM and VMX.
- Harden the device posted IRQ code against bugs and runtime errors.
- Use vcpu_idx, not vcpu_id, for GA log tag/metadata, to make lookups O(1)
instead of O(n).
- Generate GA Log interrupts if and only if the target vCPU is blocking, i.e.
only if KVM needs a notification in order to wake the vCPU.
- Decouple device posted IRQs from VFIO device assignment, as binding a VM to
a VFIO group is not a requirement for enabling device posted IRQs.
- Clean up and document/comment the irqfd assignment code.
- Disallow binding multiple irqfds to an eventfd with a priority waiter, i.e.
ensure an eventfd is bound to at most one irqfd through the entire host,
and add a selftest to verify eventfd:irqfd bindings are globally unique.
Remove the redundant kvm_gmem_getattr() implementation that simply calls
generic_fillattr() without any special handling. The VFS layer
(vfs_getattr_nosec()) will call generic_fillattr() by default when no
custom getattr operation is provided in the inode_operations structure.
This is a cleanup with no functional change.
Signed-off-by: Shivank Garg <shivankg@amd.com>
Reviewed-By: Ackerley Tng <ackerleytng@google.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta@amd.com>
Link: https://lore.kernel.org/r/20250602172317.10601-1-shivankg@amd.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Drop kvm_arch_{start,end}_assignment() and all associated code now that
KVM x86 no longer consumes assigned_device_count. Tracking whether or not
a VFIO-assigned device is formally associated with a VM is fundamentally
flawed, as such an association is optional for general usage, i.e. is prone
to false negatives. E.g. prior to commit 2edd9cb79f ("kvm: detect
assigned device via irqbypass manager"), device passthrough via VFIO would
fail to enable IRQ bypass if userspace omitted the formal VFIO<=>KVM
binding.
And device drivers that *need* the VFIO<=>KVM connection, e.g. KVM-GT,
shouldn't be relying on generic x86 tracking infrastructure.
Cc: Jim Mattson <jmattson@google.com>
Link: https://lore.kernel.org/r/20250523011756.3243624-6-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Disallow binding an irqfd to an eventfd that already has a priority waiter,
i.e. to an eventfd that already has an attached irqfd. KVM always
operates in exclusive mode for EPOLL_IN (unconditionally returns '1'),
i.e. only the first waiter will be notified.
KVM already disallows binding multiple irqfds to an eventfd in a single
VM, but doesn't guard against multiple VMs binding to an eventfd. Adding
the extra protection reduces the pain of a userspace VMM bug, e.g. if
userspace fails to de-assign before re-assigning when transferring state
for intra-host migration, then the migration will explicitly fail as
opposed to dropping IRQs on the destination VM.
Temporarily keep KVM's manual check on irqfds.items, but add a WARN, e.g.
to allow sanity checking the waitqueue enforcement.
Cc: Oliver Upton <oliver.upton@linux.dev>
Cc: David Matlack <dmatlack@google.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250522235223.3178519-10-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Drop the setting of WQ_FLAG_EXCLUSIVE from add_wait_queue_priority() and
instead have callers manually add the flag prior to adding their structure
to the queue. Blindly setting WQ_FLAG_EXCLUSIVE is flawed, as the nature
of exclusive, priority waiters means that only the first waiter added will
ever receive notifications.
Pushing the flawed behavior to callers will allow fixing the problem one
hypervisor at a time (KVM added the flawed API, and then KVM's code was
copy+pasted nearly verbatim by Xen and Hyper-V), and will also allow for
adding an API that provides true exclusivity, i.e. that guarantees at most
one priority waiter is in the queue.
Opportunistically add a comment in Hyper-V to call out the mess. Xen
privcmd's irqfd_wakefup() doesn't actually operate in exclusive mode, i.e.
can be "fixed" simply by dropping WQ_FLAG_EXCLUSIVE. And KVM is primed to
switch to the aforementioned fully exclusive API, i.e. won't be carrying
the flawed code for long.
No functional change intended.
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250522235223.3178519-7-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Add an irqfd to its target eventfd's waitqueue while holding irqfds.lock,
which is mildly terrifying but functionally safe. irqfds.lock is taken
inside the waitqueue's lock, but if and only if the eventfd is being
released, i.e. that path is mutually exclusive with registration as KVM
holds a reference to the eventfd (and obviously must do so to avoid UAF).
This will allow using the eventfd's waitqueue to enforce KVM's requirement
that eventfd is assigned to at most one irqfd, without introducing races.
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250522235223.3178519-6-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Add the irqfd structure to KVM's list of irqfds in kvm_irqfd_register(),
i.e. via the vfs_poll() callback. This will allow taking irqfds.lock
across the entire registration sequence (add to waitqueue, add to list),
and more importantly will allow inserting into KVM's list if and only if
adding to the waitqueue succeeds (spoiler alert), without needing to
juggle return codes in weird ways.
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250522235223.3178519-5-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Initialize the irqfd waitqueue callback immediately prior to inserting the
irqfd into the eventfd's waitqueue. Pre-initializing the state in a
completely different context is all kinds of confusing, and incorrectly
suggests that the waitqueue function needs to be initialize prior to
vfs_poll().
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250522235223.3178519-4-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Use a function-local struct for the poll_table passed to vfs_poll(), as
nothing in the vfs_poll() callchain grabs a long-term reference to the
structure, i.e. its lifetime doesn't need to be tied to the irqfd. Using
a local structure will also allow propagating failures out of the polling
callback without further polluting kvm_kernel_irqfd.
Opportunstically rename irqfd_ptable_queue_proc() to kvm_irqfd_register()
to capture what it actually does.
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250522235223.3178519-2-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Don't bother WARNing if updating an IRTE route fails now that vendor code
provides much more precise WARNs. The generic WARN doesn't provide enough
information to actually debug the problem, and has obviously done nothing
to surface the myriad bugs in KVM x86's implementation.
Drop all of the associated return code plumbing that existed just so that
common KVM could WARN.
Link: https://lore.kernel.org/r/20250611224604.313496-34-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
When updating IRTEs in response to a GSI routing or IRQ bypass change,
pass the new/current routing information along with the associated irqfd.
This will allow KVM x86 to harden, simplify, and deduplicate its code.
Since adding/removing a bypass producer is now conveniently protected with
irqfds.lock, i.e. can't run concurrently with kvm_irq_routing_update(),
use the routing information cached in the irqfd instead of looking up
the information in the current GSI routing tables.
Opportunistically convert an existing printk() to pr_info() and put its
string onto a single line (old code that strictly adhered to 80 chars).
Link: https://lore.kernel.org/r/20250611224604.313496-5-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Trigger the I/O APIC route rescan that's performed for a split IRQ chip
after userspace updates IRQ routes in kvm_arch_irq_routing_update(), i.e.
before dropping kvm->irq_lock. Calling kvm_make_all_cpus_request() under
a mutex is perfectly safe, and the smp_wmb()+smp_mb__after_atomic() pair
in __kvm_make_request()+kvm_check_request() ensures the new routing is
visible to vCPUs prior to the request being visible to vCPUs.
In all likelihood, commit b053b2aef2 ("KVM: x86: Add EOI exit bitmap
inference") somewhat arbitrarily made the request outside of irq_lock to
avoid holding irq_lock any longer than is strictly necessary. And then
commit abdb080f7a ("kvm/irqchip: kvm_arch_irq_routing_update renaming
split") took the easy route of adding another arch hook instead of risking
a functional change.
Note, the call to synchronize_srcu_expedited() does NOT provide ordering
guarantees with respect to vCPUs scanning the new routing; as above, the
request infrastructure provides the necessary ordering. I.e. there's no
need to wait for kvm_scan_ioapic_routes() to complete if it's actively
running, because regardless of whether it grabs the old or new table, the
vCPU will have another KVM_REQ_SCAN_IOAPIC pending, i.e. will rescan again
and see the new mappings.
Acked-by: Kai Huang <kai.huang@intel.com>
Link: https://lore.kernel.org/r/20250611213557.294358-2-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Pass in the Linux IRQ associated with an IRQ bypass producer instead of
relying on the caller to set the field prior to registration, as there's
no benefit to relying on callers to do the right thing.
Take care to set producer->irq before __connect(), as KVM expects the IRQ
to be valid as soon as a connection is possible.
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://lore.kernel.org/r/20250516230734.2564775-9-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
