On Fri, Oct 24, 2025, Ackerley Tng wrote:

Ackerley Tng [off-list ref] writes:

quoted

From: Sean Christopherson <seanjc@google.com>

Accept gmem_flags in vm_mem_add() to be able to create a guest_memfd within
vm_mem_add().

When vm_mem_add() is used to set up a guest_memfd for a memslot, set up the
provided (or created) gmem_fd as the fd for the user memory region. This
makes it available to be mmap()-ed from just like fds from other memory
sources. mmap() from guest_memfd using the provided gmem_flags and
gmem_offset.

Add a kvm_slot_to_fd() helper to provide convenient access to the file
descriptor of a memslot.

Update existing callers of vm_mem_add() to pass 0 for gmem_flags to
preserve existing behavior.

Signed-off-by: Sean Christopherson <seanjc@google.com>
[For guest_memfds, mmap() using gmem_offset instead of 0 all the time.]
Signed-off-by: Ackerley Tng <redacted>
---
 tools/testing/selftests/kvm/include/kvm_util.h |  7 ++++++-
 tools/testing/selftests/kvm/lib/kvm_util.c     | 18 ++++++++++--------
 .../kvm/x86/private_mem_conversions_test.c     |  2 +-
 3 files changed, 17 insertions(+), 10 deletions(-)


[...snip...]

@@ -1050,13 +1049,16 @@ void vm_mem_add(struct kvm_vm *vm, enum vm_mem_backing_src_type src_type,
 	}
 
 	region->fd = -1;
-	if (backing_src_is_shared(src_type))
+	if (flags & KVM_MEM_GUEST_MEMFD && gmem_flags & GUEST_MEMFD_FLAG_MMAP)
+		region->fd = kvm_dup(gmem_fd);
+	else if (backing_src_is_shared(src_type))
 		region->fd = kvm_memfd_alloc(region->mmap_size,
 					     src_type == VM_MEM_SRC_SHARED_HUGETLB);
 

Doing this makes it hard to test the legacy dual-backing case.

It actually broke x86/private_mem_conversions_test for the legacy
dual-backing case because there's no way to mmap or provide a
userspace_address from the memory provider that is not guest_memfd, as
determined by src_type.

Yes there is.  This patch is a giant nop.  The only thing that the core library
doesn't support is mmap() on guest_memfd *and* the other src_type, and IMO that
is big "don't care", because KVM doesn't even support that combination:

	if (kvm_gmem_supports_mmap(inode))
		slot->flags |= KVM_MEMSLOT_GMEM_ONLY;

I mean, we _could_ test that KVM ignores the hva for mapping, but that's a
different and unique test entirely.

I did break x86/private_mem_conversions_test (I could have sworn I tested, *sigh*),
but the bug is in:

  KVM: selftests: Provide function to look up guest_memfd details from gpa

not here.  And it's a trivial /facepalm-style fix:

diff --git a/tools/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
index ee5b63f7cb50..23a8676fee6d 100644
--- a/tools/testing/selftests/kvm/lib/kvm_util.c
+++ b/tools/testing/selftests/kvm/lib/kvm_util.c

@@ -1680,7 +1680,7 @@ int kvm_gpa_to_guest_memfd(struct kvm_vm *vm, vm_paddr_t gpa, off_t *fd_offset,
        gpa_offset = gpa - region->region.guest_phys_addr;
        *fd_offset = region->region.guest_memfd_offset + gpa_offset;
        *nr_bytes = region->region.memory_size - gpa_offset;
-       return region->fd;
+       return region->region.guest_memfd;
 }
 
 /* Create an interrupt controller chip for the specified VM. */