]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
KVM: selftests: Add guest_memfd testcase to fault-in on !mmap()'d memory
authorSean Christopherson <seanjc@google.com>
Tue, 29 Jul 2025 22:54:55 +0000 (15:54 -0700)
committerPaolo Bonzini <pbonzini@redhat.com>
Wed, 27 Aug 2025 08:41:34 +0000 (04:41 -0400)
Add a guest_memfd testcase to verify that a vCPU can fault-in guest_memfd
memory that supports mmap(), but that is not currently mapped into host
userspace and/or has a userspace address (in the memslot) that points at
something other than the target guest_memfd range.  Mapping guest_memfd
memory into the guest is supposed to operate completely independently from
any userspace mappings.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20250729225455.670324-25-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
tools/testing/selftests/kvm/guest_memfd_test.c

index b88f580b2fd69c06e5e78ca64475f615fc397891..b3ca6737f3044f961415ad3dab040bec74d27bb0 100644 (file)
@@ -13,6 +13,7 @@
 
 #include <linux/bitmap.h>
 #include <linux/falloc.h>
+#include <linux/sizes.h>
 #include <setjmp.h>
 #include <signal.h>
 #include <sys/mman.h>
@@ -21,6 +22,7 @@
 
 #include "kvm_util.h"
 #include "test_util.h"
+#include "ucall_common.h"
 
 static void test_file_read_write(int fd)
 {
@@ -298,6 +300,66 @@ static void test_guest_memfd(unsigned long vm_type)
        kvm_vm_free(vm);
 }
 
+static void guest_code(uint8_t *mem, uint64_t size)
+{
+       size_t i;
+
+       for (i = 0; i < size; i++)
+               __GUEST_ASSERT(mem[i] == 0xaa,
+                              "Guest expected 0xaa at offset %lu, got 0x%x", i, mem[i]);
+
+       memset(mem, 0xff, size);
+       GUEST_DONE();
+}
+
+static void test_guest_memfd_guest(void)
+{
+       /*
+        * Skip the first 4gb and slot0.  slot0 maps <1gb and is used to back
+        * the guest's code, stack, and page tables, and low memory contains
+        * the PCI hole and other MMIO regions that need to be avoided.
+        */
+       const uint64_t gpa = SZ_4G;
+       const int slot = 1;
+
+       struct kvm_vcpu *vcpu;
+       struct kvm_vm *vm;
+       uint8_t *mem;
+       size_t size;
+       int fd, i;
+
+       if (!kvm_has_cap(KVM_CAP_GUEST_MEMFD_MMAP))
+               return;
+
+       vm = __vm_create_shape_with_one_vcpu(VM_SHAPE_DEFAULT, &vcpu, 1, guest_code);
+
+       TEST_ASSERT(vm_check_cap(vm, KVM_CAP_GUEST_MEMFD_MMAP),
+                   "Default VM type should always support guest_memfd mmap()");
+
+       size = vm->page_size;
+       fd = vm_create_guest_memfd(vm, size, GUEST_MEMFD_FLAG_MMAP);
+       vm_set_user_memory_region2(vm, slot, KVM_MEM_GUEST_MEMFD, gpa, size, NULL, fd, 0);
+
+       mem = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+       TEST_ASSERT(mem != MAP_FAILED, "mmap() on guest_memfd failed");
+       memset(mem, 0xaa, size);
+       munmap(mem, size);
+
+       virt_pg_map(vm, gpa, gpa);
+       vcpu_args_set(vcpu, 2, gpa, size);
+       vcpu_run(vcpu);
+
+       TEST_ASSERT_EQ(get_ucall(vcpu, NULL), UCALL_DONE);
+
+       mem = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+       TEST_ASSERT(mem != MAP_FAILED, "mmap() on guest_memfd failed");
+       for (i = 0; i < size; i++)
+               TEST_ASSERT_EQ(mem[i], 0xff);
+
+       close(fd);
+       kvm_vm_free(vm);
+}
+
 int main(int argc, char *argv[])
 {
        unsigned long vm_types, vm_type;
@@ -314,4 +376,6 @@ int main(int argc, char *argv[])
 
        for_each_set_bit(vm_type, &vm_types, BITS_PER_TYPE(vm_types))
                test_guest_memfd(vm_type);
+
+       test_guest_memfd_guest();
 }