1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * KVM dirty page logging test
4  *
5  * Copyright (C) 2018, Red Hat, Inc.
6  */
7 
8 #define _GNU_SOURCE /* for program_invocation_name */
9 
10 #include <stdio.h>
11 #include <stdlib.h>
12 #include <pthread.h>
13 #include <semaphore.h>
14 #include <sys/types.h>
15 #include <signal.h>
16 #include <errno.h>
17 #include <linux/bitmap.h>
18 #include <linux/bitops.h>
19 #include <linux/atomic.h>
20 
21 #include "kvm_util.h"
22 #include "test_util.h"
23 #include "guest_modes.h"
24 #include "processor.h"
25 
26 #define VCPU_ID				1
27 
28 /* The memory slot index to track dirty pages */
29 #define TEST_MEM_SLOT_INDEX		1
30 
31 /* Default guest test virtual memory offset */
32 #define DEFAULT_GUEST_TEST_MEM		0xc0000000
33 
34 /* How many pages to dirty for each guest loop */
35 #define TEST_PAGES_PER_LOOP		1024
36 
37 /* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */
38 #define TEST_HOST_LOOP_N		32UL
39 
40 /* Interval for each host loop (ms) */
41 #define TEST_HOST_LOOP_INTERVAL		10UL
42 
43 /* Dirty bitmaps are always little endian, so we need to swap on big endian */
44 #if defined(__s390x__)
45 # define BITOP_LE_SWIZZLE	((BITS_PER_LONG-1) & ~0x7)
46 # define test_bit_le(nr, addr) \
47 	test_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
48 # define set_bit_le(nr, addr) \
49 	set_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
50 # define clear_bit_le(nr, addr) \
51 	clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
52 # define test_and_set_bit_le(nr, addr) \
53 	test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
54 # define test_and_clear_bit_le(nr, addr) \
55 	test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
56 #else
57 # define test_bit_le		test_bit
58 # define set_bit_le		set_bit
59 # define clear_bit_le		clear_bit
60 # define test_and_set_bit_le	test_and_set_bit
61 # define test_and_clear_bit_le	test_and_clear_bit
62 #endif
63 
64 #define TEST_DIRTY_RING_COUNT		65536
65 
66 #define SIG_IPI SIGUSR1
67 
68 /*
69  * Guest/Host shared variables. Ensure addr_gva2hva() and/or
70  * sync_global_to/from_guest() are used when accessing from
71  * the host. READ/WRITE_ONCE() should also be used with anything
72  * that may change.
73  */
74 static uint64_t host_page_size;
75 static uint64_t guest_page_size;
76 static uint64_t guest_num_pages;
77 static uint64_t random_array[TEST_PAGES_PER_LOOP];
78 static uint64_t iteration;
79 
80 /*
81  * Guest physical memory offset of the testing memory slot.
82  * This will be set to the topmost valid physical address minus
83  * the test memory size.
84  */
85 static uint64_t guest_test_phys_mem;
86 
87 /*
88  * Guest virtual memory offset of the testing memory slot.
89  * Must not conflict with identity mapped test code.
90  */
91 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
92 
93 /*
94  * Continuously write to the first 8 bytes of a random pages within
95  * the testing memory region.
96  */
97 static void guest_code(void)
98 {
99 	uint64_t addr;
100 	int i;
101 
102 	/*
103 	 * On s390x, all pages of a 1M segment are initially marked as dirty
104 	 * when a page of the segment is written to for the very first time.
105 	 * To compensate this specialty in this test, we need to touch all
106 	 * pages during the first iteration.
107 	 */
108 	for (i = 0; i < guest_num_pages; i++) {
109 		addr = guest_test_virt_mem + i * guest_page_size;
110 		*(uint64_t *)addr = READ_ONCE(iteration);
111 	}
112 
113 	while (true) {
114 		for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
115 			addr = guest_test_virt_mem;
116 			addr += (READ_ONCE(random_array[i]) % guest_num_pages)
117 				* guest_page_size;
118 			addr &= ~(host_page_size - 1);
119 			*(uint64_t *)addr = READ_ONCE(iteration);
120 		}
121 
122 		/* Tell the host that we need more random numbers */
123 		GUEST_SYNC(1);
124 	}
125 }
126 
127 /* Host variables */
128 static bool host_quit;
129 
130 /* Points to the test VM memory region on which we track dirty logs */
131 static void *host_test_mem;
132 static uint64_t host_num_pages;
133 
134 /* For statistics only */
135 static uint64_t host_dirty_count;
136 static uint64_t host_clear_count;
137 static uint64_t host_track_next_count;
138 
139 /* Whether dirty ring reset is requested, or finished */
140 static sem_t sem_vcpu_stop;
141 static sem_t sem_vcpu_cont;
142 /*
143  * This is only set by main thread, and only cleared by vcpu thread.  It is
144  * used to request vcpu thread to stop at the next GUEST_SYNC, since GUEST_SYNC
145  * is the only place that we'll guarantee both "dirty bit" and "dirty data"
146  * will match.  E.g., SIG_IPI won't guarantee that if the vcpu is interrupted
147  * after setting dirty bit but before the data is written.
148  */
149 static atomic_t vcpu_sync_stop_requested;
150 /*
151  * This is updated by the vcpu thread to tell the host whether it's a
152  * ring-full event.  It should only be read until a sem_wait() of
153  * sem_vcpu_stop and before vcpu continues to run.
154  */
155 static bool dirty_ring_vcpu_ring_full;
156 /*
157  * This is only used for verifying the dirty pages.  Dirty ring has a very
158  * tricky case when the ring just got full, kvm will do userspace exit due to
159  * ring full.  When that happens, the very last PFN is set but actually the
160  * data is not changed (the guest WRITE is not really applied yet), because
161  * we found that the dirty ring is full, refused to continue the vcpu, and
162  * recorded the dirty gfn with the old contents.
163  *
164  * For this specific case, it's safe to skip checking this pfn for this
165  * bit, because it's a redundant bit, and when the write happens later the bit
166  * will be set again.  We use this variable to always keep track of the latest
167  * dirty gfn we've collected, so that if a mismatch of data found later in the
168  * verifying process, we let it pass.
169  */
170 static uint64_t dirty_ring_last_page;
171 
172 enum log_mode_t {
173 	/* Only use KVM_GET_DIRTY_LOG for logging */
174 	LOG_MODE_DIRTY_LOG = 0,
175 
176 	/* Use both KVM_[GET|CLEAR]_DIRTY_LOG for logging */
177 	LOG_MODE_CLEAR_LOG = 1,
178 
179 	/* Use dirty ring for logging */
180 	LOG_MODE_DIRTY_RING = 2,
181 
182 	LOG_MODE_NUM,
183 
184 	/* Run all supported modes */
185 	LOG_MODE_ALL = LOG_MODE_NUM,
186 };
187 
188 /* Mode of logging to test.  Default is to run all supported modes */
189 static enum log_mode_t host_log_mode_option = LOG_MODE_ALL;
190 /* Logging mode for current run */
191 static enum log_mode_t host_log_mode;
192 static pthread_t vcpu_thread;
193 static uint32_t test_dirty_ring_count = TEST_DIRTY_RING_COUNT;
194 
195 static void vcpu_kick(void)
196 {
197 	pthread_kill(vcpu_thread, SIG_IPI);
198 }
199 
200 /*
201  * In our test we do signal tricks, let's use a better version of
202  * sem_wait to avoid signal interrupts
203  */
204 static void sem_wait_until(sem_t *sem)
205 {
206 	int ret;
207 
208 	do
209 		ret = sem_wait(sem);
210 	while (ret == -1 && errno == EINTR);
211 }
212 
213 static bool clear_log_supported(void)
214 {
215 	return kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
216 }
217 
218 static void clear_log_create_vm_done(struct kvm_vm *vm)
219 {
220 	struct kvm_enable_cap cap = {};
221 	u64 manual_caps;
222 
223 	manual_caps = kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
224 	TEST_ASSERT(manual_caps, "MANUAL_CAPS is zero!");
225 	manual_caps &= (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE |
226 			KVM_DIRTY_LOG_INITIALLY_SET);
227 	cap.cap = KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2;
228 	cap.args[0] = manual_caps;
229 	vm_enable_cap(vm, &cap);
230 }
231 
232 static void dirty_log_collect_dirty_pages(struct kvm_vm *vm, int slot,
233 					  void *bitmap, uint32_t num_pages)
234 {
235 	kvm_vm_get_dirty_log(vm, slot, bitmap);
236 }
237 
238 static void clear_log_collect_dirty_pages(struct kvm_vm *vm, int slot,
239 					  void *bitmap, uint32_t num_pages)
240 {
241 	kvm_vm_get_dirty_log(vm, slot, bitmap);
242 	kvm_vm_clear_dirty_log(vm, slot, bitmap, 0, num_pages);
243 }
244 
245 /* Should only be called after a GUEST_SYNC */
246 static void vcpu_handle_sync_stop(void)
247 {
248 	if (atomic_read(&vcpu_sync_stop_requested)) {
249 		/* It means main thread is sleeping waiting */
250 		atomic_set(&vcpu_sync_stop_requested, false);
251 		sem_post(&sem_vcpu_stop);
252 		sem_wait_until(&sem_vcpu_cont);
253 	}
254 }
255 
256 static void default_after_vcpu_run(struct kvm_vm *vm, int ret, int err)
257 {
258 	struct kvm_run *run = vcpu_state(vm, VCPU_ID);
259 
260 	TEST_ASSERT(ret == 0 || (ret == -1 && err == EINTR),
261 		    "vcpu run failed: errno=%d", err);
262 
263 	TEST_ASSERT(get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC,
264 		    "Invalid guest sync status: exit_reason=%s\n",
265 		    exit_reason_str(run->exit_reason));
266 
267 	vcpu_handle_sync_stop();
268 }
269 
270 static bool dirty_ring_supported(void)
271 {
272 	return kvm_check_cap(KVM_CAP_DIRTY_LOG_RING);
273 }
274 
275 static void dirty_ring_create_vm_done(struct kvm_vm *vm)
276 {
277 	/*
278 	 * Switch to dirty ring mode after VM creation but before any
279 	 * of the vcpu creation.
280 	 */
281 	vm_enable_dirty_ring(vm, test_dirty_ring_count *
282 			     sizeof(struct kvm_dirty_gfn));
283 }
284 
285 static inline bool dirty_gfn_is_dirtied(struct kvm_dirty_gfn *gfn)
286 {
287 	return gfn->flags == KVM_DIRTY_GFN_F_DIRTY;
288 }
289 
290 static inline void dirty_gfn_set_collected(struct kvm_dirty_gfn *gfn)
291 {
292 	gfn->flags = KVM_DIRTY_GFN_F_RESET;
293 }
294 
295 static uint32_t dirty_ring_collect_one(struct kvm_dirty_gfn *dirty_gfns,
296 				       int slot, void *bitmap,
297 				       uint32_t num_pages, uint32_t *fetch_index)
298 {
299 	struct kvm_dirty_gfn *cur;
300 	uint32_t count = 0;
301 
302 	while (true) {
303 		cur = &dirty_gfns[*fetch_index % test_dirty_ring_count];
304 		if (!dirty_gfn_is_dirtied(cur))
305 			break;
306 		TEST_ASSERT(cur->slot == slot, "Slot number didn't match: "
307 			    "%u != %u", cur->slot, slot);
308 		TEST_ASSERT(cur->offset < num_pages, "Offset overflow: "
309 			    "0x%llx >= 0x%x", cur->offset, num_pages);
310 		//pr_info("fetch 0x%x page %llu\n", *fetch_index, cur->offset);
311 		set_bit_le(cur->offset, bitmap);
312 		dirty_ring_last_page = cur->offset;
313 		dirty_gfn_set_collected(cur);
314 		(*fetch_index)++;
315 		count++;
316 	}
317 
318 	return count;
319 }
320 
321 static void dirty_ring_wait_vcpu(void)
322 {
323 	/* This makes sure that hardware PML cache flushed */
324 	vcpu_kick();
325 	sem_wait_until(&sem_vcpu_stop);
326 }
327 
328 static void dirty_ring_continue_vcpu(void)
329 {
330 	pr_info("Notifying vcpu to continue\n");
331 	sem_post(&sem_vcpu_cont);
332 }
333 
334 static void dirty_ring_collect_dirty_pages(struct kvm_vm *vm, int slot,
335 					   void *bitmap, uint32_t num_pages)
336 {
337 	/* We only have one vcpu */
338 	static uint32_t fetch_index = 0;
339 	uint32_t count = 0, cleared;
340 	bool continued_vcpu = false;
341 
342 	dirty_ring_wait_vcpu();
343 
344 	if (!dirty_ring_vcpu_ring_full) {
345 		/*
346 		 * This is not a ring-full event, it's safe to allow
347 		 * vcpu to continue
348 		 */
349 		dirty_ring_continue_vcpu();
350 		continued_vcpu = true;
351 	}
352 
353 	/* Only have one vcpu */
354 	count = dirty_ring_collect_one(vcpu_map_dirty_ring(vm, VCPU_ID),
355 				       slot, bitmap, num_pages, &fetch_index);
356 
357 	cleared = kvm_vm_reset_dirty_ring(vm);
358 
359 	/* Cleared pages should be the same as collected */
360 	TEST_ASSERT(cleared == count, "Reset dirty pages (%u) mismatch "
361 		    "with collected (%u)", cleared, count);
362 
363 	if (!continued_vcpu) {
364 		TEST_ASSERT(dirty_ring_vcpu_ring_full,
365 			    "Didn't continue vcpu even without ring full");
366 		dirty_ring_continue_vcpu();
367 	}
368 
369 	pr_info("Iteration %ld collected %u pages\n", iteration, count);
370 }
371 
372 static void dirty_ring_after_vcpu_run(struct kvm_vm *vm, int ret, int err)
373 {
374 	struct kvm_run *run = vcpu_state(vm, VCPU_ID);
375 
376 	/* A ucall-sync or ring-full event is allowed */
377 	if (get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC) {
378 		/* We should allow this to continue */
379 		;
380 	} else if (run->exit_reason == KVM_EXIT_DIRTY_RING_FULL ||
381 		   (ret == -1 && err == EINTR)) {
382 		/* Update the flag first before pause */
383 		WRITE_ONCE(dirty_ring_vcpu_ring_full,
384 			   run->exit_reason == KVM_EXIT_DIRTY_RING_FULL);
385 		sem_post(&sem_vcpu_stop);
386 		pr_info("vcpu stops because %s...\n",
387 			dirty_ring_vcpu_ring_full ?
388 			"dirty ring is full" : "vcpu is kicked out");
389 		sem_wait_until(&sem_vcpu_cont);
390 		pr_info("vcpu continues now.\n");
391 	} else {
392 		TEST_ASSERT(false, "Invalid guest sync status: "
393 			    "exit_reason=%s\n",
394 			    exit_reason_str(run->exit_reason));
395 	}
396 }
397 
398 static void dirty_ring_before_vcpu_join(void)
399 {
400 	/* Kick another round of vcpu just to make sure it will quit */
401 	sem_post(&sem_vcpu_cont);
402 }
403 
404 struct log_mode {
405 	const char *name;
406 	/* Return true if this mode is supported, otherwise false */
407 	bool (*supported)(void);
408 	/* Hook when the vm creation is done (before vcpu creation) */
409 	void (*create_vm_done)(struct kvm_vm *vm);
410 	/* Hook to collect the dirty pages into the bitmap provided */
411 	void (*collect_dirty_pages) (struct kvm_vm *vm, int slot,
412 				     void *bitmap, uint32_t num_pages);
413 	/* Hook to call when after each vcpu run */
414 	void (*after_vcpu_run)(struct kvm_vm *vm, int ret, int err);
415 	void (*before_vcpu_join) (void);
416 } log_modes[LOG_MODE_NUM] = {
417 	{
418 		.name = "dirty-log",
419 		.collect_dirty_pages = dirty_log_collect_dirty_pages,
420 		.after_vcpu_run = default_after_vcpu_run,
421 	},
422 	{
423 		.name = "clear-log",
424 		.supported = clear_log_supported,
425 		.create_vm_done = clear_log_create_vm_done,
426 		.collect_dirty_pages = clear_log_collect_dirty_pages,
427 		.after_vcpu_run = default_after_vcpu_run,
428 	},
429 	{
430 		.name = "dirty-ring",
431 		.supported = dirty_ring_supported,
432 		.create_vm_done = dirty_ring_create_vm_done,
433 		.collect_dirty_pages = dirty_ring_collect_dirty_pages,
434 		.before_vcpu_join = dirty_ring_before_vcpu_join,
435 		.after_vcpu_run = dirty_ring_after_vcpu_run,
436 	},
437 };
438 
439 /*
440  * We use this bitmap to track some pages that should have its dirty
441  * bit set in the _next_ iteration.  For example, if we detected the
442  * page value changed to current iteration but at the same time the
443  * page bit is cleared in the latest bitmap, then the system must
444  * report that write in the next get dirty log call.
445  */
446 static unsigned long *host_bmap_track;
447 
448 static void log_modes_dump(void)
449 {
450 	int i;
451 
452 	printf("all");
453 	for (i = 0; i < LOG_MODE_NUM; i++)
454 		printf(", %s", log_modes[i].name);
455 	printf("\n");
456 }
457 
458 static bool log_mode_supported(void)
459 {
460 	struct log_mode *mode = &log_modes[host_log_mode];
461 
462 	if (mode->supported)
463 		return mode->supported();
464 
465 	return true;
466 }
467 
468 static void log_mode_create_vm_done(struct kvm_vm *vm)
469 {
470 	struct log_mode *mode = &log_modes[host_log_mode];
471 
472 	if (mode->create_vm_done)
473 		mode->create_vm_done(vm);
474 }
475 
476 static void log_mode_collect_dirty_pages(struct kvm_vm *vm, int slot,
477 					 void *bitmap, uint32_t num_pages)
478 {
479 	struct log_mode *mode = &log_modes[host_log_mode];
480 
481 	TEST_ASSERT(mode->collect_dirty_pages != NULL,
482 		    "collect_dirty_pages() is required for any log mode!");
483 	mode->collect_dirty_pages(vm, slot, bitmap, num_pages);
484 }
485 
486 static void log_mode_after_vcpu_run(struct kvm_vm *vm, int ret, int err)
487 {
488 	struct log_mode *mode = &log_modes[host_log_mode];
489 
490 	if (mode->after_vcpu_run)
491 		mode->after_vcpu_run(vm, ret, err);
492 }
493 
494 static void log_mode_before_vcpu_join(void)
495 {
496 	struct log_mode *mode = &log_modes[host_log_mode];
497 
498 	if (mode->before_vcpu_join)
499 		mode->before_vcpu_join();
500 }
501 
502 static void generate_random_array(uint64_t *guest_array, uint64_t size)
503 {
504 	uint64_t i;
505 
506 	for (i = 0; i < size; i++)
507 		guest_array[i] = random();
508 }
509 
510 static void *vcpu_worker(void *data)
511 {
512 	int ret, vcpu_fd;
513 	struct kvm_vm *vm = data;
514 	uint64_t *guest_array;
515 	uint64_t pages_count = 0;
516 	struct kvm_signal_mask *sigmask = alloca(offsetof(struct kvm_signal_mask, sigset)
517 						 + sizeof(sigset_t));
518 	sigset_t *sigset = (sigset_t *) &sigmask->sigset;
519 
520 	vcpu_fd = vcpu_get_fd(vm, VCPU_ID);
521 
522 	/*
523 	 * SIG_IPI is unblocked atomically while in KVM_RUN.  It causes the
524 	 * ioctl to return with -EINTR, but it is still pending and we need
525 	 * to accept it with the sigwait.
526 	 */
527 	sigmask->len = 8;
528 	pthread_sigmask(0, NULL, sigset);
529 	sigdelset(sigset, SIG_IPI);
530 	vcpu_ioctl(vm, VCPU_ID, KVM_SET_SIGNAL_MASK, sigmask);
531 
532 	sigemptyset(sigset);
533 	sigaddset(sigset, SIG_IPI);
534 
535 	guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
536 
537 	while (!READ_ONCE(host_quit)) {
538 		/* Clear any existing kick signals */
539 		generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
540 		pages_count += TEST_PAGES_PER_LOOP;
541 		/* Let the guest dirty the random pages */
542 		ret = ioctl(vcpu_fd, KVM_RUN, NULL);
543 		if (ret == -1 && errno == EINTR) {
544 			int sig = -1;
545 			sigwait(sigset, &sig);
546 			assert(sig == SIG_IPI);
547 		}
548 		log_mode_after_vcpu_run(vm, ret, errno);
549 	}
550 
551 	pr_info("Dirtied %"PRIu64" pages\n", pages_count);
552 
553 	return NULL;
554 }
555 
556 static void vm_dirty_log_verify(enum vm_guest_mode mode, unsigned long *bmap)
557 {
558 	uint64_t step = vm_num_host_pages(mode, 1);
559 	uint64_t page;
560 	uint64_t *value_ptr;
561 	uint64_t min_iter = 0;
562 
563 	for (page = 0; page < host_num_pages; page += step) {
564 		value_ptr = host_test_mem + page * host_page_size;
565 
566 		/* If this is a special page that we were tracking... */
567 		if (test_and_clear_bit_le(page, host_bmap_track)) {
568 			host_track_next_count++;
569 			TEST_ASSERT(test_bit_le(page, bmap),
570 				    "Page %"PRIu64" should have its dirty bit "
571 				    "set in this iteration but it is missing",
572 				    page);
573 		}
574 
575 		if (test_and_clear_bit_le(page, bmap)) {
576 			bool matched;
577 
578 			host_dirty_count++;
579 
580 			/*
581 			 * If the bit is set, the value written onto
582 			 * the corresponding page should be either the
583 			 * previous iteration number or the current one.
584 			 */
585 			matched = (*value_ptr == iteration ||
586 				   *value_ptr == iteration - 1);
587 
588 			if (host_log_mode == LOG_MODE_DIRTY_RING && !matched) {
589 				if (*value_ptr == iteration - 2 && min_iter <= iteration - 2) {
590 					/*
591 					 * Short answer: this case is special
592 					 * only for dirty ring test where the
593 					 * page is the last page before a kvm
594 					 * dirty ring full in iteration N-2.
595 					 *
596 					 * Long answer: Assuming ring size R,
597 					 * one possible condition is:
598 					 *
599 					 *      main thr       vcpu thr
600 					 *      --------       --------
601 					 *    iter=1
602 					 *                   write 1 to page 0~(R-1)
603 					 *                   full, vmexit
604 					 *    collect 0~(R-1)
605 					 *    kick vcpu
606 					 *                   write 1 to (R-1)~(2R-2)
607 					 *                   full, vmexit
608 					 *    iter=2
609 					 *    collect (R-1)~(2R-2)
610 					 *    kick vcpu
611 					 *                   write 1 to (2R-2)
612 					 *                   (NOTE!!! "1" cached in cpu reg)
613 					 *                   write 2 to (2R-1)~(3R-3)
614 					 *                   full, vmexit
615 					 *    iter=3
616 					 *    collect (2R-2)~(3R-3)
617 					 *    (here if we read value on page
618 					 *     "2R-2" is 1, while iter=3!!!)
619 					 *
620 					 * This however can only happen once per iteration.
621 					 */
622 					min_iter = iteration - 1;
623 					continue;
624 				} else if (page == dirty_ring_last_page) {
625 					/*
626 					 * Please refer to comments in
627 					 * dirty_ring_last_page.
628 					 */
629 					continue;
630 				}
631 			}
632 
633 			TEST_ASSERT(matched,
634 				    "Set page %"PRIu64" value %"PRIu64
635 				    " incorrect (iteration=%"PRIu64")",
636 				    page, *value_ptr, iteration);
637 		} else {
638 			host_clear_count++;
639 			/*
640 			 * If cleared, the value written can be any
641 			 * value smaller or equals to the iteration
642 			 * number.  Note that the value can be exactly
643 			 * (iteration-1) if that write can happen
644 			 * like this:
645 			 *
646 			 * (1) increase loop count to "iteration-1"
647 			 * (2) write to page P happens (with value
648 			 *     "iteration-1")
649 			 * (3) get dirty log for "iteration-1"; we'll
650 			 *     see that page P bit is set (dirtied),
651 			 *     and not set the bit in host_bmap_track
652 			 * (4) increase loop count to "iteration"
653 			 *     (which is current iteration)
654 			 * (5) get dirty log for current iteration,
655 			 *     we'll see that page P is cleared, with
656 			 *     value "iteration-1".
657 			 */
658 			TEST_ASSERT(*value_ptr <= iteration,
659 				    "Clear page %"PRIu64" value %"PRIu64
660 				    " incorrect (iteration=%"PRIu64")",
661 				    page, *value_ptr, iteration);
662 			if (*value_ptr == iteration) {
663 				/*
664 				 * This page is _just_ modified; it
665 				 * should report its dirtyness in the
666 				 * next run
667 				 */
668 				set_bit_le(page, host_bmap_track);
669 			}
670 		}
671 	}
672 }
673 
674 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid,
675 				uint64_t extra_mem_pages, void *guest_code)
676 {
677 	struct kvm_vm *vm;
678 	uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
679 
680 	pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
681 
682 	vm = vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages, O_RDWR);
683 	kvm_vm_elf_load(vm, program_invocation_name);
684 #ifdef __x86_64__
685 	vm_create_irqchip(vm);
686 #endif
687 	log_mode_create_vm_done(vm);
688 	vm_vcpu_add_default(vm, vcpuid, guest_code);
689 	return vm;
690 }
691 
692 #define DIRTY_MEM_BITS 30 /* 1G */
693 #define PAGE_SHIFT_4K  12
694 
695 struct test_params {
696 	unsigned long iterations;
697 	unsigned long interval;
698 	uint64_t phys_offset;
699 };
700 
701 static void run_test(enum vm_guest_mode mode, void *arg)
702 {
703 	struct test_params *p = arg;
704 	struct kvm_vm *vm;
705 	unsigned long *bmap;
706 
707 	if (!log_mode_supported()) {
708 		print_skip("Log mode '%s' not supported",
709 			   log_modes[host_log_mode].name);
710 		return;
711 	}
712 
713 	/*
714 	 * We reserve page table for 2 times of extra dirty mem which
715 	 * will definitely cover the original (1G+) test range.  Here
716 	 * we do the calculation with 4K page size which is the
717 	 * smallest so the page number will be enough for all archs
718 	 * (e.g., 64K page size guest will need even less memory for
719 	 * page tables).
720 	 */
721 	vm = create_vm(mode, VCPU_ID,
722 		       2ul << (DIRTY_MEM_BITS - PAGE_SHIFT_4K),
723 		       guest_code);
724 
725 	guest_page_size = vm_get_page_size(vm);
726 	/*
727 	 * A little more than 1G of guest page sized pages.  Cover the
728 	 * case where the size is not aligned to 64 pages.
729 	 */
730 	guest_num_pages = (1ul << (DIRTY_MEM_BITS -
731 				   vm_get_page_shift(vm))) + 3;
732 	guest_num_pages = vm_adjust_num_guest_pages(mode, guest_num_pages);
733 
734 	host_page_size = getpagesize();
735 	host_num_pages = vm_num_host_pages(mode, guest_num_pages);
736 
737 	if (!p->phys_offset) {
738 		guest_test_phys_mem = (vm_get_max_gfn(vm) -
739 				       guest_num_pages) * guest_page_size;
740 		guest_test_phys_mem &= ~(host_page_size - 1);
741 	} else {
742 		guest_test_phys_mem = p->phys_offset;
743 	}
744 
745 #ifdef __s390x__
746 	/* Align to 1M (segment size) */
747 	guest_test_phys_mem &= ~((1 << 20) - 1);
748 #endif
749 
750 	pr_info("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
751 
752 	bmap = bitmap_alloc(host_num_pages);
753 	host_bmap_track = bitmap_alloc(host_num_pages);
754 
755 	/* Add an extra memory slot for testing dirty logging */
756 	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
757 				    guest_test_phys_mem,
758 				    TEST_MEM_SLOT_INDEX,
759 				    guest_num_pages,
760 				    KVM_MEM_LOG_DIRTY_PAGES);
761 
762 	/* Do mapping for the dirty track memory slot */
763 	virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages);
764 
765 	/* Cache the HVA pointer of the region */
766 	host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
767 
768 	ucall_init(vm, NULL);
769 
770 	/* Export the shared variables to the guest */
771 	sync_global_to_guest(vm, host_page_size);
772 	sync_global_to_guest(vm, guest_page_size);
773 	sync_global_to_guest(vm, guest_test_virt_mem);
774 	sync_global_to_guest(vm, guest_num_pages);
775 
776 	/* Start the iterations */
777 	iteration = 1;
778 	sync_global_to_guest(vm, iteration);
779 	host_quit = false;
780 	host_dirty_count = 0;
781 	host_clear_count = 0;
782 	host_track_next_count = 0;
783 
784 	pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
785 
786 	while (iteration < p->iterations) {
787 		/* Give the vcpu thread some time to dirty some pages */
788 		usleep(p->interval * 1000);
789 		log_mode_collect_dirty_pages(vm, TEST_MEM_SLOT_INDEX,
790 					     bmap, host_num_pages);
791 
792 		/*
793 		 * See vcpu_sync_stop_requested definition for details on why
794 		 * we need to stop vcpu when verify data.
795 		 */
796 		atomic_set(&vcpu_sync_stop_requested, true);
797 		sem_wait_until(&sem_vcpu_stop);
798 		/*
799 		 * NOTE: for dirty ring, it's possible that we didn't stop at
800 		 * GUEST_SYNC but instead we stopped because ring is full;
801 		 * that's okay too because ring full means we're only missing
802 		 * the flush of the last page, and since we handle the last
803 		 * page specially verification will succeed anyway.
804 		 */
805 		assert(host_log_mode == LOG_MODE_DIRTY_RING ||
806 		       atomic_read(&vcpu_sync_stop_requested) == false);
807 		vm_dirty_log_verify(mode, bmap);
808 		sem_post(&sem_vcpu_cont);
809 
810 		iteration++;
811 		sync_global_to_guest(vm, iteration);
812 	}
813 
814 	/* Tell the vcpu thread to quit */
815 	host_quit = true;
816 	log_mode_before_vcpu_join();
817 	pthread_join(vcpu_thread, NULL);
818 
819 	pr_info("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
820 		"track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
821 		host_track_next_count);
822 
823 	free(bmap);
824 	free(host_bmap_track);
825 	ucall_uninit(vm);
826 	kvm_vm_free(vm);
827 }
828 
829 static void help(char *name)
830 {
831 	puts("");
832 	printf("usage: %s [-h] [-i iterations] [-I interval] "
833 	       "[-p offset] [-m mode]\n", name);
834 	puts("");
835 	printf(" -c: specify dirty ring size, in number of entries\n");
836 	printf("     (only useful for dirty-ring test; default: %"PRIu32")\n",
837 	       TEST_DIRTY_RING_COUNT);
838 	printf(" -i: specify iteration counts (default: %"PRIu64")\n",
839 	       TEST_HOST_LOOP_N);
840 	printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
841 	       TEST_HOST_LOOP_INTERVAL);
842 	printf(" -p: specify guest physical test memory offset\n"
843 	       "     Warning: a low offset can conflict with the loaded test code.\n");
844 	printf(" -M: specify the host logging mode "
845 	       "(default: run all log modes).  Supported modes: \n\t");
846 	log_modes_dump();
847 	guest_modes_help();
848 	puts("");
849 	exit(0);
850 }
851 
852 int main(int argc, char *argv[])
853 {
854 	struct test_params p = {
855 		.iterations = TEST_HOST_LOOP_N,
856 		.interval = TEST_HOST_LOOP_INTERVAL,
857 	};
858 	int opt, i;
859 	sigset_t sigset;
860 
861 	sem_init(&sem_vcpu_stop, 0, 0);
862 	sem_init(&sem_vcpu_cont, 0, 0);
863 
864 	guest_modes_append_default();
865 
866 	while ((opt = getopt(argc, argv, "c:hi:I:p:m:M:")) != -1) {
867 		switch (opt) {
868 		case 'c':
869 			test_dirty_ring_count = strtol(optarg, NULL, 10);
870 			break;
871 		case 'i':
872 			p.iterations = strtol(optarg, NULL, 10);
873 			break;
874 		case 'I':
875 			p.interval = strtol(optarg, NULL, 10);
876 			break;
877 		case 'p':
878 			p.phys_offset = strtoull(optarg, NULL, 0);
879 			break;
880 		case 'm':
881 			guest_modes_cmdline(optarg);
882 			break;
883 		case 'M':
884 			if (!strcmp(optarg, "all")) {
885 				host_log_mode_option = LOG_MODE_ALL;
886 				break;
887 			}
888 			for (i = 0; i < LOG_MODE_NUM; i++) {
889 				if (!strcmp(optarg, log_modes[i].name)) {
890 					pr_info("Setting log mode to: '%s'\n",
891 						optarg);
892 					host_log_mode_option = i;
893 					break;
894 				}
895 			}
896 			if (i == LOG_MODE_NUM) {
897 				printf("Log mode '%s' invalid. Please choose "
898 				       "from: ", optarg);
899 				log_modes_dump();
900 				exit(1);
901 			}
902 			break;
903 		case 'h':
904 		default:
905 			help(argv[0]);
906 			break;
907 		}
908 	}
909 
910 	TEST_ASSERT(p.iterations > 2, "Iterations must be greater than two");
911 	TEST_ASSERT(p.interval > 0, "Interval must be greater than zero");
912 
913 	pr_info("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
914 		p.iterations, p.interval);
915 
916 	srandom(time(0));
917 
918 	/* Ensure that vCPU threads start with SIG_IPI blocked.  */
919 	sigemptyset(&sigset);
920 	sigaddset(&sigset, SIG_IPI);
921 	pthread_sigmask(SIG_BLOCK, &sigset, NULL);
922 
923 	if (host_log_mode_option == LOG_MODE_ALL) {
924 		/* Run each log mode */
925 		for (i = 0; i < LOG_MODE_NUM; i++) {
926 			pr_info("Testing Log Mode '%s'\n", log_modes[i].name);
927 			host_log_mode = i;
928 			for_each_guest_mode(run_test, &p);
929 		}
930 	} else {
931 		host_log_mode = host_log_mode_option;
932 		for_each_guest_mode(run_test, &p);
933 	}
934 
935 	return 0;
936 }
937