[RFC PATCH v2 3/3] KVM: selftests: Add set_memory_region_io to test memslots for MMIO BARs

[Yan Zhao]

Added a selftest set_memory_region_io to test memslots for MMIO BARs.
The MMIO BARs' backends are compound/non-compound huge pages serving as
device resources allocated by a mock device driver.

This selftest will assert and report "errno=14 - Bad address" in vcpu_run()
if any failure is met to add such MMIO BAR memslots.
After MMIO memslots removal, page reference counts of the device resources
are also checked.

As this selftest will interacts with a mock device "/dev/kvm_mock_device",
it depends on test driver test_kvm_mock_device.ko in the kernel.
CONFIG_TEST_KVM_MOCK_DEVICE=m must be enabled in the kernel.

Currently, this selftest is only compiled for __x86_64__.

Signed-off-by: Yan Zhao <yan.y.zhao@xxxxxxxxx>
---
 tools/testing/selftests/kvm/Makefile          |   1 +
 .../selftests/kvm/set_memory_region_io.c      | 188 ++++++++++++++++++
 2 files changed, 189 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/set_memory_region_io.c

diff --git a/tools/testing/selftests/kvm/Makefile b/tools/testing/selftests/kvm/Makefile
index 4412b42d95de..9d39514b6403 100644
--- a/tools/testing/selftests/kvm/Makefile
+++ b/tools/testing/selftests/kvm/Makefile
@@ -144,6 +144,7 @@ TEST_GEN_PROGS_x86_64 += memslot_modification_stress_test
 TEST_GEN_PROGS_x86_64 += memslot_perf_test
 TEST_GEN_PROGS_x86_64 += rseq_test
 TEST_GEN_PROGS_x86_64 += set_memory_region_test
+TEST_GEN_PROGS_x86_64 += set_memory_region_io
 TEST_GEN_PROGS_x86_64 += steal_time
 TEST_GEN_PROGS_x86_64 += kvm_binary_stats_test
 TEST_GEN_PROGS_x86_64 += system_counter_offset_test
diff --git a/tools/testing/selftests/kvm/set_memory_region_io.c b/tools/testing/selftests/kvm/set_memory_region_io.c
new file mode 100644
index 000000000000..e221103091f4
--- /dev/null
+++ b/tools/testing/selftests/kvm/set_memory_region_io.c
@@ -0,0 +1,188 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE /* for program_invocation_short_name */
+#include <fcntl.h>
+#include <pthread.h>
+#include <sched.h>
+#include <semaphore.h>
+#include <signal.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+
+#include <linux/compiler.h>
+
+#include <test_util.h>
+#include <kvm_util.h>
+#include <processor.h>
+
+#include <../../../../lib/test_kvm_mock_device_uapi.h>
+
+/*
+ * Somewhat arbitrary location and slot, intended to not overlap anything.
+ */
+#define MEM_REGION_GPA_BASE	0xc0000000
+#define RANDOM_OFFSET		0x1000
+#define MEM_REGION_GPA_RANDOM	(MEM_REGION_GPA_BASE + RANDOM_OFFSET)
+#define MEM_REGION_SLOT_ID	10
+
+static const bool non_compound_supported;
+
+static const uint64_t BASE_VAL = 0x1111;
+static const uint64_t RANDOM_VAL = 0x2222;
+
+static unsigned long bar_size;
+
+static sem_t vcpu_ready;
+
+static void guest_code_read_bar(void)
+{
+	uint64_t val;
+
+	GUEST_SYNC(0);
+
+	val = READ_ONCE(*((uint64_t *)MEM_REGION_GPA_RANDOM));
+	GUEST_ASSERT_EQ(val, RANDOM_VAL);
+
+	val = READ_ONCE(*((uint64_t *)MEM_REGION_GPA_BASE));
+	GUEST_ASSERT_EQ(val, BASE_VAL);
+
+	GUEST_DONE();
+}
+
+static void *vcpu_worker(void *data)
+{
+	struct kvm_vcpu *vcpu = data;
+	struct kvm_run *run = vcpu->run;
+	struct ucall uc;
+	uint64_t cmd;
+
+	/*
+	 * Loop until the guest is done.  Re-enter the guest on all MMIO exits,
+	 * which will occur if the guest attempts to access a memslot after it
+	 * has been deleted or while it is being moved .
+	 */
+	while (1) {
+		vcpu_run(vcpu);
+
+		if (run->exit_reason == KVM_EXIT_IO) {
+			cmd = get_ucall(vcpu, &uc);
+			if (cmd != UCALL_SYNC)
+				break;
+
+			sem_post(&vcpu_ready);
+			continue;
+		}
+
+		if (run->exit_reason != KVM_EXIT_MMIO)
+			break;
+
+		TEST_ASSERT(!run->mmio.is_write, "Unexpected exit mmio write");
+		TEST_ASSERT(run->mmio.len == 8,
+			    "Unexpected exit mmio size = %u", run->mmio.len);
+
+		TEST_ASSERT(run->mmio.phys_addr < MEM_REGION_GPA_BASE ||
+			    run->mmio.phys_addr >= MEM_REGION_GPA_BASE + bar_size,
+			    "Unexpected exit mmio address = 0x%llx",
+			    run->mmio.phys_addr);
+	}
+
+	if (run->exit_reason == KVM_EXIT_IO && cmd == UCALL_ABORT)
+		REPORT_GUEST_ASSERT(uc);
+
+	return NULL;
+}
+
+static void wait_for_vcpu(void)
+{
+	struct timespec ts;
+
+	TEST_ASSERT(!clock_gettime(CLOCK_REALTIME, &ts),
+		    "clock_gettime() failed: %d\n", errno);
+
+	ts.tv_sec += 2;
+	TEST_ASSERT(!sem_timedwait(&vcpu_ready, &ts),
+		    "sem_timedwait() failed: %d\n", errno);
+
+	/* Wait for the vCPU thread to reenter the guest. */
+	usleep(100000);
+}
+
+static void test_kvm_mock_device_bar(bool compound)
+{
+	struct kvm_vm *vm;
+	void *mem;
+	struct kvm_vcpu *vcpu;
+	pthread_t vcpu_thread;
+	int fd, ret;
+	u32 param_compound = compound;
+	u32 inequal = 0;
+
+	fd = open("/dev/kvm_mock_device", O_RDWR);
+	if (fd < 0) {
+		pr_info("Please ensure \"CONFIG_TEST_KVM_MOCK_DEVICE=m\" is enabled in the kernel");
+		pr_info(", and execute\n\"modprobe test_kvm_mock_device\n");
+	}
+	TEST_ASSERT(fd >= 0, "Failed to open kvm mock device.");
+
+	ret = ioctl(fd, KVM_MOCK_DEVICE_GET_BAR_SIZE, &bar_size);
+	TEST_ASSERT(ret == 0, "Failed to get bar size of kvm mock device");
+
+	ret = ioctl(fd, KVM_MOCK_DEVICE_PREPARE_RESOURCE, &param_compound);
+	TEST_ASSERT(ret == 0, "Failed to prepare resource of kvm mock device");
+
+	mem = mmap(NULL, (size_t)bar_size, PROT_READ | PROT_WRITE, MAP_SHARED,
+		   fd, 0);
+	TEST_ASSERT(mem != MAP_FAILED, "Failed to mmap() kvm mock device bar");
+
+	*(u64 *)mem = BASE_VAL;
+	*(u64 *)(mem + RANDOM_OFFSET) = RANDOM_VAL;
+
+	vm = vm_create_with_one_vcpu(&vcpu, guest_code_read_bar);
+
+	vm_set_user_memory_region(vm, MEM_REGION_SLOT_ID, 0, MEM_REGION_GPA_BASE,
+				  bar_size, mem);
+
+	virt_map(vm, MEM_REGION_GPA_BASE, MEM_REGION_GPA_BASE,
+		 (RANDOM_OFFSET / getpagesize()) + 1);
+
+	pthread_create(&vcpu_thread, NULL, vcpu_worker, vcpu);
+
+	/* Ensure the guest thread is spun up. */
+	wait_for_vcpu();
+
+	pthread_join(vcpu_thread, NULL);
+
+	vm_set_user_memory_region(vm, MEM_REGION_SLOT_ID, 0, 0, 0, 0);
+	kvm_vm_free(vm);
+
+	ret = ioctl(fd, KVM_MOCK_DEVICE_CHECK_BACKEND_REF, &inequal);
+	TEST_ASSERT(ret == 0 && inequal == 0, "Incorrect resource ref of KVM device");
+
+	munmap(mem, bar_size);
+	close(fd);
+}
+
+static void test_non_compound_backend(void)
+{
+	pr_info("Testing non-compound huge page backend for mem slot\n");
+	test_kvm_mock_device_bar(false);
+}
+
+static void test_compound_backend(void)
+{
+	pr_info("Testing compound huge page backend for mem slot\n");
+	test_kvm_mock_device_bar(true);
+}
+
+int main(int argc, char *argv[])
+{
+#ifdef __x86_64__
+	test_compound_backend();
+	if (non_compound_supported)
+		test_non_compound_backend();
+#endif
+
+	return 0;
+}
-- 
2.17.1