kvm: selftest: add dirty logging test

Test KVM dirty logging functionality.

The test creates a standalone memory slot to test tracking the dirty
pages since we can't really write to the default memory slot which still
contains the guest ELF image.

We have two threads running during the test:

(1) the vcpu thread continuously dirties random guest pages by writting
    a iteration number to the first 8 bytes of the page

(2) the host thread continuously fetches dirty logs for the testing
    memory region and verify each single bit of the dirty bitmap by
    checking against the values written onto the page

Note that since the guest cannot calls the general userspace APIs like
random(), it depends on the host to provide random numbers for the
page indexes to dirty.
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

tools/testing/selftests/kvm/Makefile

@@ -11,6 +11,7 @@ TEST_GEN_PROGS_x86_64 += sync_regs_test
 TEST_GEN_PROGS_x86_64 += vmx_tsc_adjust_test
 TEST_GEN_PROGS_x86_64 += cr4_cpuid_sync_test
 TEST_GEN_PROGS_x86_64 += state_test
+TEST_GEN_PROGS_x86_64 += dirty_log_test
 
 TEST_GEN_PROGS += $(TEST_GEN_PROGS_$(UNAME_M))
 LIBKVM += $(LIBKVM_$(UNAME_M))
@@ -19,6 +20,7 @@ INSTALL_HDR_PATH = $(top_srcdir)/usr
 LINUX_HDR_PATH = $(INSTALL_HDR_PATH)/include/
 LINUX_TOOL_INCLUDE = $(top_srcdir)tools/include
 CFLAGS += -O2 -g -std=gnu99 -I$(LINUX_TOOL_INCLUDE) -I$(LINUX_HDR_PATH) -Iinclude -I$(<D) -I..
+LDFLAGS += -lpthread
 
 # After inclusion, $(OUTPUT) is defined and
 # $(TEST_GEN_PROGS) starts with $(OUTPUT)/

tools/testing/selftests/kvm/dirty_log_test.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * KVM dirty page logging test
+ *
+ * Copyright (C) 2018, Red Hat, Inc.
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <time.h>
+#include <pthread.h>
+#include <linux/bitmap.h>
+#include <linux/bitops.h>
+
+#include "test_util.h"
+#include "kvm_util.h"
+
+#define  DEBUG                 printf
+
+#define  VCPU_ID                        1
+/* The memory slot index to track dirty pages */
+#define  TEST_MEM_SLOT_INDEX            1
+/*
+ * GPA offset of the testing memory slot. Must be bigger than the
+ * default vm mem slot, which is DEFAULT_GUEST_PHY_PAGES.
+ */
+#define  TEST_MEM_OFFSET                (1ULL << 30) /* 1G */
+/* Size of the testing memory slot */
+#define  TEST_MEM_PAGES                 (1ULL << 18) /* 1G for 4K pages */
+/* How many pages to dirty for each guest loop */
+#define  TEST_PAGES_PER_LOOP            1024
+/* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */
+#define  TEST_HOST_LOOP_N               32
+/* Interval for each host loop (ms) */
+#define  TEST_HOST_LOOP_INTERVAL        10
+
+/*
+ * Guest variables.  We use these variables to share data between host
+ * and guest.  There are two copies of the variables, one in host memory
+ * (which is unused) and one in guest memory.  When the host wants to
+ * access these variables, it needs to call addr_gva2hva() to access the
+ * guest copy.
+ */
+uint64_t guest_random_array[TEST_PAGES_PER_LOOP];
+uint64_t guest_iteration;
+uint64_t guest_page_size;
+
+/*
+ * Writes to the first byte of a random page within the testing memory
+ * region continuously.
+ */
+void guest_code(void)
+{
+	int i = 0;
+	uint64_t volatile *array = guest_random_array;
+	uint64_t volatile *guest_addr;
+
+	while (true) {
+		for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
+			/*
+			 * Write to the first 8 bytes of a random page
+			 * on the testing memory region.
+			 */
+			guest_addr = (uint64_t *)
+			    (TEST_MEM_OFFSET +
+			     (array[i] % TEST_MEM_PAGES) * guest_page_size);
+			*guest_addr = guest_iteration;
+		}
+		/* Tell the host that we need more random numbers */
+		GUEST_SYNC(1);
+	}
+}
+
+/*
+ * Host variables.  These variables should only be used by the host
+ * rather than the guest.
+ */
+bool host_quit;
+
+/* Points to the test VM memory region on which we track dirty logs */
+void *host_test_mem;
+
+/* For statistics only */
+uint64_t host_dirty_count;
+uint64_t host_clear_count;
+uint64_t host_track_next_count;
+
+/*
+ * We use this bitmap to track some pages that should have its dirty
+ * bit set in the _next_ iteration.  For example, if we detected the
+ * page value changed to current iteration but at the same time the
+ * page bit is cleared in the latest bitmap, then the system must
+ * report that write in the next get dirty log call.
+ */
+unsigned long *host_bmap_track;
+
+void generate_random_array(uint64_t *guest_array, uint64_t size)
+{
+	uint64_t i;
+
+	for (i = 0; i < size; i++) {
+		guest_array[i] = random();
+	}
+}
+
+void *vcpu_worker(void *data)
+{
+	int ret;
+	uint64_t loops, *guest_array, pages_count = 0;
+	struct kvm_vm *vm = data;
+	struct kvm_run *run;
+	struct guest_args args;
+
+	run = vcpu_state(vm, VCPU_ID);
+
+	/* Retrieve the guest random array pointer and cache it */
+	guest_array = addr_gva2hva(vm, (vm_vaddr_t)guest_random_array);
+
+	DEBUG("VCPU starts\n");
+
+	generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
+
+	while (!READ_ONCE(host_quit)) {
+		/* Let the guest to dirty these random pages */
+		ret = _vcpu_run(vm, VCPU_ID);
+		guest_args_read(vm, VCPU_ID, &args);
+		if (run->exit_reason == KVM_EXIT_IO &&
+		    args.port == GUEST_PORT_SYNC) {
+			pages_count += TEST_PAGES_PER_LOOP;
+			generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
+		} else {
+			TEST_ASSERT(false,
+				    "Invalid guest sync status: "
+				    "exit_reason=%s\n",
+				    exit_reason_str(run->exit_reason));
+		}
+	}
+
+	DEBUG("VCPU exits, dirtied %"PRIu64" pages\n", pages_count);
+
+	return NULL;
+}
+
+void vm_dirty_log_verify(unsigned long *bmap, uint64_t iteration)
+{
+	uint64_t page;
+	uint64_t volatile *value_ptr;
+
+	for (page = 0; page < TEST_MEM_PAGES; page++) {
+		value_ptr = host_test_mem + page * getpagesize();
+
+		/* If this is a special page that we were tracking... */
+		if (test_and_clear_bit(page, host_bmap_track)) {
+			host_track_next_count++;
+			TEST_ASSERT(test_bit(page, bmap),
+				    "Page %"PRIu64" should have its dirty bit "
+				    "set in this iteration but it is missing",
+				    page);
+		}
+
+		if (test_bit(page, bmap)) {
+			host_dirty_count++;
+			/*
+			 * If the bit is set, the value written onto
+			 * the corresponding page should be either the
+			 * previous iteration number or the current one.
+			 */
+			TEST_ASSERT(*value_ptr == iteration ||
+				    *value_ptr == iteration - 1,
+				    "Set page %"PRIu64" value %"PRIu64
+				    " incorrect (iteration=%"PRIu64")",
+				    page, *value_ptr, iteration);
+		} else {
+			host_clear_count++;
+			/*
+			 * If cleared, the value written can be any
+			 * value smaller or equals to the iteration
+			 * number.  Note that the value can be exactly
+			 * (iteration-1) if that write can happen
+			 * like this:
+			 *
+			 * (1) increase loop count to "iteration-1"
+			 * (2) write to page P happens (with value
+			 *     "iteration-1")
+			 * (3) get dirty log for "iteration-1"; we'll
+			 *     see that page P bit is set (dirtied),
+			 *     and not set the bit in host_bmap_track
+			 * (4) increase loop count to "iteration"
+			 *     (which is current iteration)
+			 * (5) get dirty log for current iteration,
+			 *     we'll see that page P is cleared, with
+			 *     value "iteration-1".
+			 */
+			TEST_ASSERT(*value_ptr <= iteration,
+				    "Clear page %"PRIu64" value %"PRIu64
+				    " incorrect (iteration=%"PRIu64")",
+				    page, *value_ptr, iteration);
+			if (*value_ptr == iteration) {
+				/*
+				 * This page is _just_ modified; it
+				 * should report its dirtyness in the
+				 * next run
+				 */
+				set_bit(page, host_bmap_track);
+			}
+		}
+	}
+}
+
+void help(char *name)
+{
+	puts("");
+	printf("usage: %s [-i iterations] [-I interval] [-h]\n", name);
+	puts("");
+	printf(" -i: specify iteration counts (default: %"PRIu64")\n",
+	       TEST_HOST_LOOP_N);
+	printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
+	       TEST_HOST_LOOP_INTERVAL);
+	puts("");
+	exit(0);
+}
+
+int main(int argc, char *argv[])
+{
+	pthread_t vcpu_thread;
+	struct kvm_vm *vm;
+	uint64_t volatile *psize, *iteration;
+	unsigned long *bmap, iterations = TEST_HOST_LOOP_N,
+	    interval = TEST_HOST_LOOP_INTERVAL;
+	int opt;
+
+	while ((opt = getopt(argc, argv, "hi:I:")) != -1) {
+		switch (opt) {
+		case 'i':
+			iterations = strtol(optarg, NULL, 10);
+			break;
+		case 'I':
+			interval = strtol(optarg, NULL, 10);
+			break;
+		case 'h':
+		default:
+			help(argv[0]);
+			break;
+		}
+	}
+
+	TEST_ASSERT(iterations > 2, "Iteration must be bigger than zero\n");
+	TEST_ASSERT(interval > 0, "Interval must be bigger than zero");
+
+	DEBUG("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
+	      iterations, interval);
+
+	srandom(time(0));
+
+	bmap = bitmap_alloc(TEST_MEM_PAGES);
+	host_bmap_track = bitmap_alloc(TEST_MEM_PAGES);
+
+	vm = vm_create_default(VCPU_ID, TEST_MEM_PAGES, guest_code);
+
+	/* Add an extra memory slot for testing dirty logging */
+	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
+				    TEST_MEM_OFFSET,
+				    TEST_MEM_SLOT_INDEX,
+				    TEST_MEM_PAGES,
+				    KVM_MEM_LOG_DIRTY_PAGES);
+	/* Cache the HVA pointer of the region */
+	host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)TEST_MEM_OFFSET);
+
+	/* Do 1:1 mapping for the dirty track memory slot */
+	virt_map(vm, TEST_MEM_OFFSET, TEST_MEM_OFFSET,
+		 TEST_MEM_PAGES * getpagesize(), 0);
+
+	vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
+
+	/* Tell the guest about the page size on the system */
+	psize = addr_gva2hva(vm, (vm_vaddr_t)&guest_page_size);
+	*psize = getpagesize();
+
+	/* Start the iterations */
+	iteration = addr_gva2hva(vm, (vm_vaddr_t)&guest_iteration);
+	*iteration = 1;
+
+	/* Start dirtying pages */
+	pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
+
+	while (*iteration < iterations) {
+		/* Give the vcpu thread some time to dirty some pages */
+		usleep(interval * 1000);
+		kvm_vm_get_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap);
+		vm_dirty_log_verify(bmap, *iteration);
+		(*iteration)++;
+	}
+
+	/* Tell the vcpu thread to quit */
+	host_quit = true;
+	pthread_join(vcpu_thread, NULL);
+
+	DEBUG("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
+	      "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
+	      host_track_next_count);
+
+	free(bmap);
+	free(host_bmap_track);
+	kvm_vm_free(vm);
+
+	return 0;
+}

tools/testing/selftests/kvm/include/kvm_util.h

@@ -55,6 +55,7 @@ struct kvm_vm *vm_create(enum vm_guest_mode mode, uint64_t phy_pages, int perm);
 void kvm_vm_free(struct kvm_vm *vmp);
 void kvm_vm_restart(struct kvm_vm *vmp, int perm);
 void kvm_vm_release(struct kvm_vm *vmp);
+void kvm_vm_get_dirty_log(struct kvm_vm *vm, int slot, void *log);
 
 int kvm_memcmp_hva_gva(void *hva,
 	struct kvm_vm *vm, const vm_vaddr_t gva, size_t len);
@@ -80,6 +81,8 @@ void vm_mem_region_set_flags(struct kvm_vm *vm, uint32_t slot, uint32_t flags);
 void vm_vcpu_add(struct kvm_vm *vm, uint32_t vcpuid, int pgd_memslot, int gdt_memslot);
 vm_vaddr_t vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min,
 	uint32_t data_memslot, uint32_t pgd_memslot);
+void virt_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
+	      size_t size, uint32_t pgd_memslot);
 void *addr_gpa2hva(struct kvm_vm *vm, vm_paddr_t gpa);
 void *addr_gva2hva(struct kvm_vm *vm, vm_vaddr_t gva);
 vm_paddr_t addr_hva2gpa(struct kvm_vm *vm, void *hva);

tools/testing/selftests/kvm/lib/kvm_util.c

@@ -169,6 +169,16 @@ void kvm_vm_restart(struct kvm_vm *vmp, int perm)
 	}
 }
 
+void kvm_vm_get_dirty_log(struct kvm_vm *vm, int slot, void *log)
+{
+	struct kvm_dirty_log args = { .dirty_bitmap = log, .slot = slot };
+	int ret;
+
+	ret = ioctl(vm->fd, KVM_GET_DIRTY_LOG, &args);
+	TEST_ASSERT(ret == 0, "%s: KVM_GET_DIRTY_LOG failed: %s",
+		    strerror(-ret));
+}
+
 /* Userspace Memory Region Find
  *
  * Input Args:
@@ -924,6 +934,39 @@ vm_vaddr_t vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min,
 	return vaddr_start;
 }
 
+/*
+ * Map a range of VM virtual address to the VM's physical address
+ *
+ * Input Args:
+ *   vm - Virtual Machine
+ *   vaddr - Virtuall address to map
+ *   paddr - VM Physical Address
+ *   size - The size of the range to map
+ *   pgd_memslot - Memory region slot for new virtual translation tables
+ *
+ * Output Args: None
+ *
+ * Return: None
+ *
+ * Within the VM given by vm, creates a virtual translation for the
+ * page range starting at vaddr to the page range starting at paddr.
+ */
+void virt_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
+	      size_t size, uint32_t pgd_memslot)
+{
+	size_t page_size = vm->page_size;
+	size_t npages = size / page_size;
+
+	TEST_ASSERT(vaddr + size > vaddr, "Vaddr overflow");
+	TEST_ASSERT(paddr + size > paddr, "Paddr overflow");
+
+	while (npages--) {
+		virt_pg_map(vm, vaddr, paddr, pgd_memslot);
+		vaddr += page_size;
+		paddr += page_size;
+	}
+}
+
 /* Address VM Physical to Host Virtual
  *
  * Input Args: